blob: a83160acd7487328a0b86d318ac4e3427df7e329 [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/personality.h>
26#include <linux/security.h>
27#include <linux/syscalls.h>
28#include <linux/mount.h>
29#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080032#include <linux/fcntl.h>
33#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <asm/namei.h>
35#include <asm/uaccess.h>
36
37#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
38
39/* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
44 *
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
51 *
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
55 *
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
58 *
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
65 */
66
67/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
74 *
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
82 */
83
84/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
86 *
87 * [10-Sep-98 Alan Modra] Another symlink change.
88 */
89
90/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
97 *
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
103 */
104/*
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * any extra contention...
108 */
109
Josef 'Jeff' Sipekc4a78082007-07-19 01:48:22 -0700110static int fastcall link_path_walk(const char *name, struct nameidata *nd);
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112/* In order to reduce some races, while at the same time doing additional
113 * checking and hopefully speeding things up, we copy filenames to the
114 * kernel data space before using them..
115 *
116 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
117 * PATH_MAX includes the nul terminator --RR.
118 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800119static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120{
121 int retval;
122 unsigned long len = PATH_MAX;
123
124 if (!segment_eq(get_fs(), KERNEL_DS)) {
125 if ((unsigned long) filename >= TASK_SIZE)
126 return -EFAULT;
127 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
128 len = TASK_SIZE - (unsigned long) filename;
129 }
130
131 retval = strncpy_from_user(page, filename, len);
132 if (retval > 0) {
133 if (retval < len)
134 return 0;
135 return -ENAMETOOLONG;
136 } else if (!retval)
137 retval = -ENOENT;
138 return retval;
139}
140
141char * getname(const char __user * filename)
142{
143 char *tmp, *result;
144
145 result = ERR_PTR(-ENOMEM);
146 tmp = __getname();
147 if (tmp) {
148 int retval = do_getname(filename, tmp);
149
150 result = tmp;
151 if (retval < 0) {
152 __putname(tmp);
153 result = ERR_PTR(retval);
154 }
155 }
156 audit_getname(result);
157 return result;
158}
159
160#ifdef CONFIG_AUDITSYSCALL
161void putname(const char *name)
162{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400163 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 audit_putname(name);
165 else
166 __putname(name);
167}
168EXPORT_SYMBOL(putname);
169#endif
170
171
172/**
173 * generic_permission - check for access rights on a Posix-like filesystem
174 * @inode: inode to check access rights for
175 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
176 * @check_acl: optional callback to check for Posix ACLs
177 *
178 * Used to check for read/write/execute permissions on a file.
179 * We use "fsuid" for this, letting us set arbitrary permissions
180 * for filesystem access without changing the "normal" uids which
181 * are used for other things..
182 */
183int generic_permission(struct inode *inode, int mask,
184 int (*check_acl)(struct inode *inode, int mask))
185{
186 umode_t mode = inode->i_mode;
187
188 if (current->fsuid == inode->i_uid)
189 mode >>= 6;
190 else {
191 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
192 int error = check_acl(inode, mask);
193 if (error == -EACCES)
194 goto check_capabilities;
195 else if (error != -EAGAIN)
196 return error;
197 }
198
199 if (in_group_p(inode->i_gid))
200 mode >>= 3;
201 }
202
203 /*
204 * If the DACs are ok we don't need any capability check.
205 */
206 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
207 return 0;
208
209 check_capabilities:
210 /*
211 * Read/write DACs are always overridable.
212 * Executable DACs are overridable if at least one exec bit is set.
213 */
214 if (!(mask & MAY_EXEC) ||
215 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
216 if (capable(CAP_DAC_OVERRIDE))
217 return 0;
218
219 /*
220 * Searching includes executable on directories, else just read.
221 */
222 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
223 if (capable(CAP_DAC_READ_SEARCH))
224 return 0;
225
226 return -EACCES;
227}
228
229int permission(struct inode *inode, int mask, struct nameidata *nd)
230{
Trond Myklebusta343bb72006-08-22 20:06:03 -0400231 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 int retval, submask;
233
234 if (mask & MAY_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
236 /*
237 * Nobody gets write access to a read-only fs.
238 */
239 if (IS_RDONLY(inode) &&
240 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
241 return -EROFS;
242
243 /*
244 * Nobody gets write access to an immutable file.
245 */
246 if (IS_IMMUTABLE(inode))
247 return -EACCES;
248 }
249
250
Trond Myklebusta343bb72006-08-22 20:06:03 -0400251 /*
252 * MAY_EXEC on regular files requires special handling: We override
Stas Sergeev317a40a2006-12-06 20:35:25 -0800253 * filesystem execute permissions if the mode bits aren't set or
254 * the fs is mounted with the "noexec" flag.
Trond Myklebusta343bb72006-08-22 20:06:03 -0400255 */
Stas Sergeev317a40a2006-12-06 20:35:25 -0800256 if ((mask & MAY_EXEC) && S_ISREG(mode) && (!(mode & S_IXUGO) ||
257 (nd && nd->mnt && (nd->mnt->mnt_flags & MNT_NOEXEC))))
Trond Myklebusta343bb72006-08-22 20:06:03 -0400258 return -EACCES;
259
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 /* Ordinary permission routines do not understand MAY_APPEND. */
261 submask = mask & ~MAY_APPEND;
262 if (inode->i_op && inode->i_op->permission)
263 retval = inode->i_op->permission(inode, submask, nd);
264 else
265 retval = generic_permission(inode, submask, NULL);
266 if (retval)
267 return retval;
268
269 return security_inode_permission(inode, mask, nd);
270}
271
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800272/**
273 * vfs_permission - check for access rights to a given path
274 * @nd: lookup result that describes the path
275 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
276 *
277 * Used to check for read/write/execute permissions on a path.
278 * We use "fsuid" for this, letting us set arbitrary permissions
279 * for filesystem access without changing the "normal" uids which
280 * are used for other things.
281 */
282int vfs_permission(struct nameidata *nd, int mask)
283{
284 return permission(nd->dentry->d_inode, mask, nd);
285}
286
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800287/**
288 * file_permission - check for additional access rights to a given file
289 * @file: file to check access rights for
290 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
291 *
292 * Used to check for read/write/execute permissions on an already opened
293 * file.
294 *
295 * Note:
296 * Do not use this function in new code. All access checks should
297 * be done using vfs_permission().
298 */
299int file_permission(struct file *file, int mask)
300{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800301 return permission(file->f_path.dentry->d_inode, mask, NULL);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800302}
303
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304/*
305 * get_write_access() gets write permission for a file.
306 * put_write_access() releases this write permission.
307 * This is used for regular files.
308 * We cannot support write (and maybe mmap read-write shared) accesses and
309 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
310 * can have the following values:
311 * 0: no writers, no VM_DENYWRITE mappings
312 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
313 * > 0: (i_writecount) users are writing to the file.
314 *
315 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
316 * except for the cases where we don't hold i_writecount yet. Then we need to
317 * use {get,deny}_write_access() - these functions check the sign and refuse
318 * to do the change if sign is wrong. Exclusion between them is provided by
319 * the inode->i_lock spinlock.
320 */
321
322int get_write_access(struct inode * inode)
323{
324 spin_lock(&inode->i_lock);
325 if (atomic_read(&inode->i_writecount) < 0) {
326 spin_unlock(&inode->i_lock);
327 return -ETXTBSY;
328 }
329 atomic_inc(&inode->i_writecount);
330 spin_unlock(&inode->i_lock);
331
332 return 0;
333}
334
335int deny_write_access(struct file * file)
336{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800337 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338
339 spin_lock(&inode->i_lock);
340 if (atomic_read(&inode->i_writecount) > 0) {
341 spin_unlock(&inode->i_lock);
342 return -ETXTBSY;
343 }
344 atomic_dec(&inode->i_writecount);
345 spin_unlock(&inode->i_lock);
346
347 return 0;
348}
349
350void path_release(struct nameidata *nd)
351{
352 dput(nd->dentry);
353 mntput(nd->mnt);
354}
355
356/*
357 * umount() mustn't call path_release()/mntput() as that would clear
358 * mnt_expiry_mark
359 */
360void path_release_on_umount(struct nameidata *nd)
361{
362 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700363 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364}
365
Trond Myklebust834f2a42005-10-18 14:20:16 -0700366/**
367 * release_open_intent - free up open intent resources
368 * @nd: pointer to nameidata
369 */
370void release_open_intent(struct nameidata *nd)
371{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800372 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700373 put_filp(nd->intent.open.file);
374 else
375 fput(nd->intent.open.file);
376}
377
Ian Kentbcdc5e02006-09-27 01:50:44 -0700378static inline struct dentry *
379do_revalidate(struct dentry *dentry, struct nameidata *nd)
380{
381 int status = dentry->d_op->d_revalidate(dentry, nd);
382 if (unlikely(status <= 0)) {
383 /*
384 * The dentry failed validation.
385 * If d_revalidate returned 0 attempt to invalidate
386 * the dentry otherwise d_revalidate is asking us
387 * to return a fail status.
388 */
389 if (!status) {
390 if (!d_invalidate(dentry)) {
391 dput(dentry);
392 dentry = NULL;
393 }
394 } else {
395 dput(dentry);
396 dentry = ERR_PTR(status);
397 }
398 }
399 return dentry;
400}
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402/*
403 * Internal lookup() using the new generic dcache.
404 * SMP-safe
405 */
406static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
407{
408 struct dentry * dentry = __d_lookup(parent, name);
409
410 /* lockess __d_lookup may fail due to concurrent d_move()
411 * in some unrelated directory, so try with d_lookup
412 */
413 if (!dentry)
414 dentry = d_lookup(parent, name);
415
Ian Kentbcdc5e02006-09-27 01:50:44 -0700416 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
417 dentry = do_revalidate(dentry, nd);
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 return dentry;
420}
421
422/*
423 * Short-cut version of permission(), for calling by
424 * path_walk(), when dcache lock is held. Combines parts
425 * of permission() and generic_permission(), and tests ONLY for
426 * MAY_EXEC permission.
427 *
428 * If appropriate, check DAC only. If not appropriate, or
429 * short-cut DAC fails, then call permission() to do more
430 * complete permission check.
431 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800432static int exec_permission_lite(struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 struct nameidata *nd)
434{
435 umode_t mode = inode->i_mode;
436
437 if (inode->i_op && inode->i_op->permission)
438 return -EAGAIN;
439
440 if (current->fsuid == inode->i_uid)
441 mode >>= 6;
442 else if (in_group_p(inode->i_gid))
443 mode >>= 3;
444
445 if (mode & MAY_EXEC)
446 goto ok;
447
448 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
449 goto ok;
450
451 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
452 goto ok;
453
454 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
455 goto ok;
456
457 return -EACCES;
458ok:
459 return security_inode_permission(inode, MAY_EXEC, nd);
460}
461
462/*
463 * This is called when everything else fails, and we actually have
464 * to go to the low-level filesystem to find out what we should do..
465 *
466 * We get the directory semaphore, and after getting that we also
467 * make sure that nobody added the entry to the dcache in the meantime..
468 * SMP-safe
469 */
470static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
471{
472 struct dentry * result;
473 struct inode *dir = parent->d_inode;
474
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800475 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 /*
477 * First re-do the cached lookup just in case it was created
478 * while we waited for the directory semaphore..
479 *
480 * FIXME! This could use version numbering or similar to
481 * avoid unnecessary cache lookups.
482 *
483 * The "dcache_lock" is purely to protect the RCU list walker
484 * from concurrent renames at this point (we mustn't get false
485 * negatives from the RCU list walk here, unlike the optimistic
486 * fast walk).
487 *
488 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
489 */
490 result = d_lookup(parent, name);
491 if (!result) {
492 struct dentry * dentry = d_alloc(parent, name);
493 result = ERR_PTR(-ENOMEM);
494 if (dentry) {
495 result = dir->i_op->lookup(dir, dentry, nd);
496 if (result)
497 dput(dentry);
498 else
499 result = dentry;
500 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800501 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 return result;
503 }
504
505 /*
506 * Uhhuh! Nasty case: the cache was re-populated while
507 * we waited on the semaphore. Need to revalidate.
508 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800509 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700511 result = do_revalidate(result, nd);
512 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 }
515 return result;
516}
517
518static int __emul_lookup_dentry(const char *, struct nameidata *);
519
520/* SMP-safe */
Arjan van de Venf1662352006-01-14 13:21:31 -0800521static __always_inline int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522walk_init_root(const char *name, struct nameidata *nd)
523{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700524 struct fs_struct *fs = current->fs;
525
526 read_lock(&fs->lock);
527 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
528 nd->mnt = mntget(fs->altrootmnt);
529 nd->dentry = dget(fs->altroot);
530 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 if (__emul_lookup_dentry(name,nd))
532 return 0;
Andreas Mohre518ddb2006-09-29 02:01:22 -0700533 read_lock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700535 nd->mnt = mntget(fs->rootmnt);
536 nd->dentry = dget(fs->root);
537 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 return 1;
539}
540
Arjan van de Venf1662352006-01-14 13:21:31 -0800541static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542{
543 int res = 0;
544 char *name;
545 if (IS_ERR(link))
546 goto fail;
547
548 if (*link == '/') {
549 path_release(nd);
550 if (!walk_init_root(link, nd))
551 /* weird __emul_prefix() stuff did it */
552 goto out;
553 }
554 res = link_path_walk(link, nd);
555out:
556 if (nd->depth || res || nd->last_type!=LAST_NORM)
557 return res;
558 /*
559 * If it is an iterative symlinks resolution in open_namei() we
560 * have to copy the last component. And all that crap because of
561 * bloody create() on broken symlinks. Furrfu...
562 */
563 name = __getname();
564 if (unlikely(!name)) {
565 path_release(nd);
566 return -ENOMEM;
567 }
568 strcpy(name, nd->last.name);
569 nd->last.name = name;
570 return 0;
571fail:
572 path_release(nd);
573 return PTR_ERR(link);
574}
575
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700576static inline void dput_path(struct path *path, struct nameidata *nd)
577{
578 dput(path->dentry);
579 if (path->mnt != nd->mnt)
580 mntput(path->mnt);
581}
582
583static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
584{
585 dput(nd->dentry);
586 if (nd->mnt != path->mnt)
587 mntput(nd->mnt);
588 nd->mnt = path->mnt;
589 nd->dentry = path->dentry;
590}
591
Ian Kent051d3812006-03-27 01:14:53 -0800592static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
593{
594 int error;
595 void *cookie;
596 struct dentry *dentry = path->dentry;
597
598 touch_atime(path->mnt, dentry);
599 nd_set_link(nd, NULL);
600
601 if (path->mnt != nd->mnt) {
602 path_to_nameidata(path, nd);
603 dget(dentry);
604 }
605 mntget(path->mnt);
606 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
607 error = PTR_ERR(cookie);
608 if (!IS_ERR(cookie)) {
609 char *s = nd_get_link(nd);
610 error = 0;
611 if (s)
612 error = __vfs_follow_link(nd, s);
613 if (dentry->d_inode->i_op->put_link)
614 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
615 }
616 dput(dentry);
617 mntput(path->mnt);
618
619 return error;
620}
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622/*
623 * This limits recursive symlink follows to 8, while
624 * limiting consecutive symlinks to 40.
625 *
626 * Without that kind of total limit, nasty chains of consecutive
627 * symlinks can cause almost arbitrarily long lookups.
628 */
Al Viro90ebe562005-06-06 13:35:58 -0700629static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630{
631 int err = -ELOOP;
632 if (current->link_count >= MAX_NESTED_LINKS)
633 goto loop;
634 if (current->total_link_count >= 40)
635 goto loop;
636 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
637 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700638 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 if (err)
640 goto loop;
641 current->link_count++;
642 current->total_link_count++;
643 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700644 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 current->link_count--;
646 nd->depth--;
647 return err;
648loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700649 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700650 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 return err;
652}
653
654int follow_up(struct vfsmount **mnt, struct dentry **dentry)
655{
656 struct vfsmount *parent;
657 struct dentry *mountpoint;
658 spin_lock(&vfsmount_lock);
659 parent=(*mnt)->mnt_parent;
660 if (parent == *mnt) {
661 spin_unlock(&vfsmount_lock);
662 return 0;
663 }
664 mntget(parent);
665 mountpoint=dget((*mnt)->mnt_mountpoint);
666 spin_unlock(&vfsmount_lock);
667 dput(*dentry);
668 *dentry = mountpoint;
669 mntput(*mnt);
670 *mnt = parent;
671 return 1;
672}
673
674/* no need for dcache_lock, as serialization is taken care in
675 * namespace.c
676 */
Al Viro463ffb22005-06-06 13:36:05 -0700677static int __follow_mount(struct path *path)
678{
679 int res = 0;
680 while (d_mountpoint(path->dentry)) {
681 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
682 if (!mounted)
683 break;
684 dput(path->dentry);
685 if (res)
686 mntput(path->mnt);
687 path->mnt = mounted;
688 path->dentry = dget(mounted->mnt_root);
689 res = 1;
690 }
691 return res;
692}
693
Al Viro58c465e2005-06-06 13:36:13 -0700694static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 while (d_mountpoint(*dentry)) {
697 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
698 if (!mounted)
699 break;
Al Viro58c465e2005-06-06 13:36:13 -0700700 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 mntput(*mnt);
702 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705}
706
707/* no need for dcache_lock, as serialization is taken care in
708 * namespace.c
709 */
Al Viroe13b2102005-06-06 13:36:06 -0700710int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
712 struct vfsmount *mounted;
713
714 mounted = lookup_mnt(*mnt, *dentry);
715 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700716 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 mntput(*mnt);
718 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 *dentry = dget(mounted->mnt_root);
720 return 1;
721 }
722 return 0;
723}
724
Arjan van de Venf1662352006-01-14 13:21:31 -0800725static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700727 struct fs_struct *fs = current->fs;
728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 while(1) {
730 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700731 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
Andreas Mohre518ddb2006-09-29 02:01:22 -0700733 read_lock(&fs->lock);
734 if (nd->dentry == fs->root &&
735 nd->mnt == fs->rootmnt) {
736 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 break;
738 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700739 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700741 if (nd->dentry != nd->mnt->mnt_root) {
742 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 spin_unlock(&dcache_lock);
744 dput(old);
745 break;
746 }
747 spin_unlock(&dcache_lock);
748 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700749 parent = nd->mnt->mnt_parent;
750 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 spin_unlock(&vfsmount_lock);
752 break;
753 }
754 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700755 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 spin_unlock(&vfsmount_lock);
757 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700758 mntput(nd->mnt);
759 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
Al Viro58c465e2005-06-06 13:36:13 -0700761 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762}
763
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764/*
765 * It's more convoluted than I'd like it to be, but... it's still fairly
766 * small and for now I'd prefer to have fast path as straight as possible.
767 * It _is_ time-critical.
768 */
769static int do_lookup(struct nameidata *nd, struct qstr *name,
770 struct path *path)
771{
772 struct vfsmount *mnt = nd->mnt;
773 struct dentry *dentry = __d_lookup(nd->dentry, name);
774
775 if (!dentry)
776 goto need_lookup;
777 if (dentry->d_op && dentry->d_op->d_revalidate)
778 goto need_revalidate;
779done:
780 path->mnt = mnt;
781 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700782 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 return 0;
784
785need_lookup:
786 dentry = real_lookup(nd->dentry, name, nd);
787 if (IS_ERR(dentry))
788 goto fail;
789 goto done;
790
791need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700792 dentry = do_revalidate(dentry, nd);
793 if (!dentry)
794 goto need_lookup;
795 if (IS_ERR(dentry))
796 goto fail;
797 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
799fail:
800 return PTR_ERR(dentry);
801}
802
803/*
804 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100805 * This is the basic name resolution function, turning a pathname into
806 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100808 * Returns 0 and nd will have valid dentry and mnt on success.
809 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 */
811static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
812{
813 struct path next;
814 struct inode *inode;
815 int err;
816 unsigned int lookup_flags = nd->flags;
817
818 while (*name=='/')
819 name++;
820 if (!*name)
821 goto return_reval;
822
823 inode = nd->dentry->d_inode;
824 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800825 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
827 /* At this point we know we have a real path component. */
828 for(;;) {
829 unsigned long hash;
830 struct qstr this;
831 unsigned int c;
832
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700833 nd->flags |= LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 err = exec_permission_lite(inode, nd);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800835 if (err == -EAGAIN)
836 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 if (err)
838 break;
839
840 this.name = name;
841 c = *(const unsigned char *)name;
842
843 hash = init_name_hash();
844 do {
845 name++;
846 hash = partial_name_hash(c, hash);
847 c = *(const unsigned char *)name;
848 } while (c && (c != '/'));
849 this.len = name - (const char *) this.name;
850 this.hash = end_name_hash(hash);
851
852 /* remove trailing slashes? */
853 if (!c)
854 goto last_component;
855 while (*++name == '/');
856 if (!*name)
857 goto last_with_slashes;
858
859 /*
860 * "." and ".." are special - ".." especially so because it has
861 * to be able to know about the current root directory and
862 * parent relationships.
863 */
864 if (this.name[0] == '.') switch (this.len) {
865 default:
866 break;
867 case 2:
868 if (this.name[1] != '.')
869 break;
Al Viro58c465e2005-06-06 13:36:13 -0700870 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 inode = nd->dentry->d_inode;
872 /* fallthrough */
873 case 1:
874 continue;
875 }
876 /*
877 * See if the low-level filesystem might want
878 * to use its own hash..
879 */
880 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
881 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
882 if (err < 0)
883 break;
884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 /* This does the actual lookups.. */
886 err = do_lookup(nd, &this, &next);
887 if (err)
888 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
890 err = -ENOENT;
891 inode = next.dentry->d_inode;
892 if (!inode)
893 goto out_dput;
894 err = -ENOTDIR;
895 if (!inode->i_op)
896 goto out_dput;
897
898 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700899 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 if (err)
901 goto return_err;
902 err = -ENOENT;
903 inode = nd->dentry->d_inode;
904 if (!inode)
905 break;
906 err = -ENOTDIR;
907 if (!inode->i_op)
908 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700909 } else
910 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 err = -ENOTDIR;
912 if (!inode->i_op->lookup)
913 break;
914 continue;
915 /* here ends the main loop */
916
917last_with_slashes:
918 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
919last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800920 /* Clear LOOKUP_CONTINUE iff it was previously unset */
921 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 if (lookup_flags & LOOKUP_PARENT)
923 goto lookup_parent;
924 if (this.name[0] == '.') switch (this.len) {
925 default:
926 break;
927 case 2:
928 if (this.name[1] != '.')
929 break;
Al Viro58c465e2005-06-06 13:36:13 -0700930 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 inode = nd->dentry->d_inode;
932 /* fallthrough */
933 case 1:
934 goto return_reval;
935 }
936 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
937 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
938 if (err < 0)
939 break;
940 }
941 err = do_lookup(nd, &this, &next);
942 if (err)
943 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 inode = next.dentry->d_inode;
945 if ((lookup_flags & LOOKUP_FOLLOW)
946 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700947 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 if (err)
949 goto return_err;
950 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700951 } else
952 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 err = -ENOENT;
954 if (!inode)
955 break;
956 if (lookup_flags & LOOKUP_DIRECTORY) {
957 err = -ENOTDIR;
958 if (!inode->i_op || !inode->i_op->lookup)
959 break;
960 }
961 goto return_base;
962lookup_parent:
963 nd->last = this;
964 nd->last_type = LAST_NORM;
965 if (this.name[0] != '.')
966 goto return_base;
967 if (this.len == 1)
968 nd->last_type = LAST_DOT;
969 else if (this.len == 2 && this.name[1] == '.')
970 nd->last_type = LAST_DOTDOT;
971 else
972 goto return_base;
973return_reval:
974 /*
975 * We bypassed the ordinary revalidation routines.
976 * We may need to check the cached dentry for staleness.
977 */
978 if (nd->dentry && nd->dentry->d_sb &&
979 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
980 err = -ESTALE;
981 /* Note: we do not d_invalidate() */
982 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
983 break;
984 }
985return_base:
986 return 0;
987out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700988 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 break;
990 }
991 path_release(nd);
992return_err:
993 return err;
994}
995
996/*
997 * Wrapper to retry pathname resolution whenever the underlying
998 * file system returns an ESTALE.
999 *
1000 * Retry the whole path once, forcing real lookup requests
1001 * instead of relying on the dcache.
1002 */
Josef 'Jeff' Sipekc4a78082007-07-19 01:48:22 -07001003static int fastcall link_path_walk(const char *name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004{
1005 struct nameidata save = *nd;
1006 int result;
1007
1008 /* make sure the stuff we saved doesn't go away */
1009 dget(save.dentry);
1010 mntget(save.mnt);
1011
1012 result = __link_path_walk(name, nd);
1013 if (result == -ESTALE) {
1014 *nd = save;
1015 dget(nd->dentry);
1016 mntget(nd->mnt);
1017 nd->flags |= LOOKUP_REVAL;
1018 result = __link_path_walk(name, nd);
1019 }
1020
1021 dput(save.dentry);
1022 mntput(save.mnt);
1023
1024 return result;
1025}
1026
Josef 'Jeff' Sipekf79c20f2007-07-19 01:48:22 -07001027static int fastcall path_walk(const char * name, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028{
1029 current->total_link_count = 0;
1030 return link_path_walk(name, nd);
1031}
1032
Prasanna Medaea3834d2005-04-29 16:00:17 +01001033/*
1034 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1035 * everything is done. Returns 0 and drops input nd, if lookup failed;
1036 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1038{
1039 if (path_walk(name, nd))
1040 return 0; /* something went wrong... */
1041
1042 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1043 struct dentry *old_dentry = nd->dentry;
1044 struct vfsmount *old_mnt = nd->mnt;
1045 struct qstr last = nd->last;
1046 int last_type = nd->last_type;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001047 struct fs_struct *fs = current->fs;
1048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 /*
Andreas Mohre518ddb2006-09-29 02:01:22 -07001050 * NAME was not found in alternate root or it's a directory.
1051 * Try to find it in the normal root:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 */
1053 nd->last_type = LAST_ROOT;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001054 read_lock(&fs->lock);
1055 nd->mnt = mntget(fs->rootmnt);
1056 nd->dentry = dget(fs->root);
1057 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 if (path_walk(name, nd) == 0) {
1059 if (nd->dentry->d_inode) {
1060 dput(old_dentry);
1061 mntput(old_mnt);
1062 return 1;
1063 }
1064 path_release(nd);
1065 }
1066 nd->dentry = old_dentry;
1067 nd->mnt = old_mnt;
1068 nd->last = last;
1069 nd->last_type = last_type;
1070 }
1071 return 1;
1072}
1073
1074void set_fs_altroot(void)
1075{
1076 char *emul = __emul_prefix();
1077 struct nameidata nd;
1078 struct vfsmount *mnt = NULL, *oldmnt;
1079 struct dentry *dentry = NULL, *olddentry;
1080 int err;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001081 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
1083 if (!emul)
1084 goto set_it;
1085 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1086 if (!err) {
1087 mnt = nd.mnt;
1088 dentry = nd.dentry;
1089 }
1090set_it:
Andreas Mohre518ddb2006-09-29 02:01:22 -07001091 write_lock(&fs->lock);
1092 oldmnt = fs->altrootmnt;
1093 olddentry = fs->altroot;
1094 fs->altrootmnt = mnt;
1095 fs->altroot = dentry;
1096 write_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 if (olddentry) {
1098 dput(olddentry);
1099 mntput(oldmnt);
1100 }
1101}
1102
Prasanna Medaea3834d2005-04-29 16:00:17 +01001103/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001104static int fastcall do_path_lookup(int dfd, const char *name,
1105 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001107 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001108 int fput_needed;
1109 struct file *file;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001110 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
1112 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1113 nd->flags = flags;
1114 nd->depth = 0;
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 if (*name=='/') {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001117 read_lock(&fs->lock);
1118 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1119 nd->mnt = mntget(fs->altrootmnt);
1120 nd->dentry = dget(fs->altroot);
1121 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001123 goto out; /* found in altroot */
Andreas Mohre518ddb2006-09-29 02:01:22 -07001124 read_lock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
Andreas Mohre518ddb2006-09-29 02:01:22 -07001126 nd->mnt = mntget(fs->rootmnt);
1127 nd->dentry = dget(fs->root);
1128 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001129 } else if (dfd == AT_FDCWD) {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001130 read_lock(&fs->lock);
1131 nd->mnt = mntget(fs->pwdmnt);
1132 nd->dentry = dget(fs->pwd);
1133 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001134 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001135 struct dentry *dentry;
1136
1137 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001138 retval = -EBADF;
1139 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001140 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001141
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001142 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001143
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001144 retval = -ENOTDIR;
1145 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001146 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001147
1148 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001149 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001150 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001151
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001152 nd->mnt = mntget(file->f_path.mnt);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001153 nd->dentry = dget(dentry);
1154
1155 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 }
Josef 'Jeff' Sipek2dfdd262007-05-09 02:33:41 -07001157
1158 retval = path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001159out:
Josef 'Jeff' Sipek62ce39c2007-05-09 02:33:41 -07001160 if (unlikely(!retval && !audit_dummy_context() && nd->dentry &&
Suzuki3bc8414b2006-02-07 12:58:36 -08001161 nd->dentry->d_inode))
Amy Griffis9c937dc2006-06-08 23:19:31 -04001162 audit_inode(name, nd->dentry->d_inode);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001163out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001164 return retval;
1165
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001166fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001167 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001168 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169}
1170
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001171int fastcall path_lookup(const char *name, unsigned int flags,
1172 struct nameidata *nd)
1173{
1174 return do_path_lookup(AT_FDCWD, name, flags, nd);
1175}
1176
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07001177/**
1178 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1179 * @dentry: pointer to dentry of the base directory
1180 * @mnt: pointer to vfs mount of the base directory
1181 * @name: pointer to file name
1182 * @flags: lookup flags
1183 * @nd: pointer to nameidata
1184 */
1185int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1186 const char *name, unsigned int flags,
1187 struct nameidata *nd)
1188{
1189 int retval;
1190
1191 /* same as do_path_lookup */
1192 nd->last_type = LAST_ROOT;
1193 nd->flags = flags;
1194 nd->depth = 0;
1195
1196 nd->mnt = mntget(mnt);
1197 nd->dentry = dget(dentry);
1198
1199 retval = path_walk(name, nd);
1200 if (unlikely(!retval && !audit_dummy_context() && nd->dentry &&
1201 nd->dentry->d_inode))
1202 audit_inode(name, nd->dentry->d_inode);
1203
1204 return retval;
1205
1206}
1207
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001208static int __path_lookup_intent_open(int dfd, const char *name,
1209 unsigned int lookup_flags, struct nameidata *nd,
1210 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001211{
1212 struct file *filp = get_empty_filp();
1213 int err;
1214
1215 if (filp == NULL)
1216 return -ENFILE;
1217 nd->intent.open.file = filp;
1218 nd->intent.open.flags = open_flags;
1219 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001220 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001221 if (IS_ERR(nd->intent.open.file)) {
1222 if (err == 0) {
1223 err = PTR_ERR(nd->intent.open.file);
1224 path_release(nd);
1225 }
1226 } else if (err != 0)
1227 release_open_intent(nd);
1228 return err;
1229}
1230
1231/**
1232 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001233 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001234 * @name: pointer to file name
1235 * @lookup_flags: lookup intent flags
1236 * @nd: pointer to nameidata
1237 * @open_flags: open intent flags
1238 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001239int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001240 struct nameidata *nd, int open_flags)
1241{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001242 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001243 open_flags, 0);
1244}
1245
1246/**
1247 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001248 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001249 * @name: pointer to file name
1250 * @lookup_flags: lookup intent flags
1251 * @nd: pointer to nameidata
1252 * @open_flags: open intent flags
1253 * @create_mode: create intent flags
1254 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001255static int path_lookup_create(int dfd, const char *name,
1256 unsigned int lookup_flags, struct nameidata *nd,
1257 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001258{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001259 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1260 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001261}
1262
1263int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1264 struct nameidata *nd, int open_flags)
1265{
1266 char *tmp = getname(name);
1267 int err = PTR_ERR(tmp);
1268
1269 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001270 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001271 putname(tmp);
1272 }
1273 return err;
1274}
1275
James Morris057f6c02007-04-26 00:12:05 -07001276static inline struct dentry *__lookup_hash_kern(struct qstr *name, struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277{
James Morris057f6c02007-04-26 00:12:05 -07001278 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 struct inode *inode;
1280 int err;
1281
1282 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
1284 /*
1285 * See if the low-level filesystem might want
1286 * to use its own hash..
1287 */
1288 if (base->d_op && base->d_op->d_hash) {
1289 err = base->d_op->d_hash(base, name);
1290 dentry = ERR_PTR(err);
1291 if (err < 0)
1292 goto out;
1293 }
1294
1295 dentry = cached_lookup(base, name, nd);
1296 if (!dentry) {
1297 struct dentry *new = d_alloc(base, name);
1298 dentry = ERR_PTR(-ENOMEM);
1299 if (!new)
1300 goto out;
1301 dentry = inode->i_op->lookup(inode, new, nd);
1302 if (!dentry)
1303 dentry = new;
1304 else
1305 dput(new);
1306 }
1307out:
1308 return dentry;
1309}
1310
James Morris057f6c02007-04-26 00:12:05 -07001311/*
1312 * Restricted form of lookup. Doesn't follow links, single-component only,
1313 * needs parent already locked. Doesn't follow mounts.
1314 * SMP-safe.
1315 */
1316static inline struct dentry * __lookup_hash(struct qstr *name, struct dentry *base, struct nameidata *nd)
1317{
1318 struct dentry *dentry;
1319 struct inode *inode;
1320 int err;
1321
1322 inode = base->d_inode;
1323
1324 err = permission(inode, MAY_EXEC, nd);
1325 dentry = ERR_PTR(err);
1326 if (err)
1327 goto out;
1328
1329 dentry = __lookup_hash_kern(name, base, nd);
1330out:
1331 return dentry;
1332}
1333
Adrian Bunka244e162006-03-31 02:32:11 -08001334static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001336 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337}
1338
1339/* SMP-safe */
James Morris057f6c02007-04-26 00:12:05 -07001340static inline int __lookup_one_len(const char *name, struct qstr *this, struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341{
1342 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 unsigned int c;
1344
James Morris057f6c02007-04-26 00:12:05 -07001345 this->name = name;
1346 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001348 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349
1350 hash = init_name_hash();
1351 while (len--) {
1352 c = *(const unsigned char *)name++;
1353 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001354 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 hash = partial_name_hash(c, hash);
1356 }
James Morris057f6c02007-04-26 00:12:05 -07001357 this->hash = end_name_hash(hash);
1358 return 0;
1359}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
James Morris057f6c02007-04-26 00:12:05 -07001361struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1362{
1363 int err;
1364 struct qstr this;
1365
1366 err = __lookup_one_len(name, &this, base, len);
1367 if (err)
1368 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001369 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001370}
1371
1372struct dentry *lookup_one_len_kern(const char *name, struct dentry *base, int len)
1373{
1374 int err;
1375 struct qstr this;
1376
1377 err = __lookup_one_len(name, &this, base, len);
1378 if (err)
1379 return ERR_PTR(err);
1380 return __lookup_hash_kern(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381}
1382
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001383int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1384 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385{
1386 char *tmp = getname(name);
1387 int err = PTR_ERR(tmp);
1388
1389 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001390 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 putname(tmp);
1392 }
1393 return err;
1394}
1395
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001396int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1397{
1398 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1399}
1400
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401/*
1402 * It's inline, so penalty for filesystems that don't use sticky bit is
1403 * minimal.
1404 */
1405static inline int check_sticky(struct inode *dir, struct inode *inode)
1406{
1407 if (!(dir->i_mode & S_ISVTX))
1408 return 0;
1409 if (inode->i_uid == current->fsuid)
1410 return 0;
1411 if (dir->i_uid == current->fsuid)
1412 return 0;
1413 return !capable(CAP_FOWNER);
1414}
1415
1416/*
1417 * Check whether we can remove a link victim from directory dir, check
1418 * whether the type of victim is right.
1419 * 1. We can't do it if dir is read-only (done in permission())
1420 * 2. We should have write and exec permissions on dir
1421 * 3. We can't remove anything from append-only dir
1422 * 4. We can't do anything with immutable dir (done in permission())
1423 * 5. If the sticky bit on dir is set we should either
1424 * a. be owner of dir, or
1425 * b. be owner of victim, or
1426 * c. have CAP_FOWNER capability
1427 * 6. If the victim is append-only or immutable we can't do antyhing with
1428 * links pointing to it.
1429 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1430 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1431 * 9. We can't remove a root or mountpoint.
1432 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1433 * nfs_async_unlink().
1434 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001435static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
1437 int error;
1438
1439 if (!victim->d_inode)
1440 return -ENOENT;
1441
1442 BUG_ON(victim->d_parent->d_inode != dir);
Amy Griffis73d3ec52006-07-13 13:16:39 -04001443 audit_inode_child(victim->d_name.name, victim->d_inode, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444
1445 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1446 if (error)
1447 return error;
1448 if (IS_APPEND(dir))
1449 return -EPERM;
1450 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1451 IS_IMMUTABLE(victim->d_inode))
1452 return -EPERM;
1453 if (isdir) {
1454 if (!S_ISDIR(victim->d_inode->i_mode))
1455 return -ENOTDIR;
1456 if (IS_ROOT(victim))
1457 return -EBUSY;
1458 } else if (S_ISDIR(victim->d_inode->i_mode))
1459 return -EISDIR;
1460 if (IS_DEADDIR(dir))
1461 return -ENOENT;
1462 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1463 return -EBUSY;
1464 return 0;
1465}
1466
1467/* Check whether we can create an object with dentry child in directory
1468 * dir.
1469 * 1. We can't do it if child already exists (open has special treatment for
1470 * this case, but since we are inlined it's OK)
1471 * 2. We can't do it if dir is read-only (done in permission())
1472 * 3. We should have write and exec permissions on dir
1473 * 4. We can't do it if dir is immutable (done in permission())
1474 */
1475static inline int may_create(struct inode *dir, struct dentry *child,
1476 struct nameidata *nd)
1477{
1478 if (child->d_inode)
1479 return -EEXIST;
1480 if (IS_DEADDIR(dir))
1481 return -ENOENT;
1482 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1483}
1484
1485/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 * O_DIRECTORY translates into forcing a directory lookup.
1487 */
1488static inline int lookup_flags(unsigned int f)
1489{
1490 unsigned long retval = LOOKUP_FOLLOW;
1491
1492 if (f & O_NOFOLLOW)
1493 retval &= ~LOOKUP_FOLLOW;
1494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 if (f & O_DIRECTORY)
1496 retval |= LOOKUP_DIRECTORY;
1497
1498 return retval;
1499}
1500
1501/*
1502 * p1 and p2 should be directories on the same fs.
1503 */
1504struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1505{
1506 struct dentry *p;
1507
1508 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001509 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 return NULL;
1511 }
1512
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001513 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
1515 for (p = p1; p->d_parent != p; p = p->d_parent) {
1516 if (p->d_parent == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001517 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1518 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 return p;
1520 }
1521 }
1522
1523 for (p = p2; p->d_parent != p; p = p->d_parent) {
1524 if (p->d_parent == p1) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001525 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1526 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 return p;
1528 }
1529 }
1530
Ingo Molnarf2eace22006-07-03 00:25:05 -07001531 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1532 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 return NULL;
1534}
1535
1536void unlock_rename(struct dentry *p1, struct dentry *p2)
1537{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001538 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001540 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001541 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 }
1543}
1544
1545int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1546 struct nameidata *nd)
1547{
1548 int error = may_create(dir, dentry, nd);
1549
1550 if (error)
1551 return error;
1552
1553 if (!dir->i_op || !dir->i_op->create)
1554 return -EACCES; /* shouldn't it be ENOSYS? */
1555 mode &= S_IALLUGO;
1556 mode |= S_IFREG;
1557 error = security_inode_create(dir, dentry, mode);
1558 if (error)
1559 return error;
1560 DQUOT_INIT(dir);
1561 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001562 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001563 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 return error;
1565}
1566
1567int may_open(struct nameidata *nd, int acc_mode, int flag)
1568{
1569 struct dentry *dentry = nd->dentry;
1570 struct inode *inode = dentry->d_inode;
1571 int error;
1572
1573 if (!inode)
1574 return -ENOENT;
1575
1576 if (S_ISLNK(inode->i_mode))
1577 return -ELOOP;
1578
1579 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1580 return -EISDIR;
1581
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001582 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 if (error)
1584 return error;
1585
1586 /*
1587 * FIFO's, sockets and device files are special: they don't
1588 * actually live on the filesystem itself, and as such you
1589 * can write to them even if the filesystem is read-only.
1590 */
1591 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1592 flag &= ~O_TRUNC;
1593 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1594 if (nd->mnt->mnt_flags & MNT_NODEV)
1595 return -EACCES;
1596
1597 flag &= ~O_TRUNC;
1598 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1599 return -EROFS;
1600 /*
1601 * An append-only file must be opened in append mode for writing.
1602 */
1603 if (IS_APPEND(inode)) {
1604 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1605 return -EPERM;
1606 if (flag & O_TRUNC)
1607 return -EPERM;
1608 }
1609
1610 /* O_NOATIME can only be set by the owner or superuser */
1611 if (flag & O_NOATIME)
Satyam Sharma3bd858a2007-07-17 15:00:08 +05301612 if (!is_owner_or_cap(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 return -EPERM;
1614
1615 /*
1616 * Ensure there are no outstanding leases on the file.
1617 */
1618 error = break_lease(inode, flag);
1619 if (error)
1620 return error;
1621
1622 if (flag & O_TRUNC) {
1623 error = get_write_access(inode);
1624 if (error)
1625 return error;
1626
1627 /*
1628 * Refuse to truncate files with mandatory locks held on them.
1629 */
1630 error = locks_verify_locked(inode);
1631 if (!error) {
1632 DQUOT_INIT(inode);
1633
NeilBrown4a301312006-01-08 01:02:39 -08001634 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 }
1636 put_write_access(inode);
1637 if (error)
1638 return error;
1639 } else
1640 if (flag & FMODE_WRITE)
1641 DQUOT_INIT(inode);
1642
1643 return 0;
1644}
1645
Dave Hansenaab520e2006-09-30 23:29:02 -07001646static int open_namei_create(struct nameidata *nd, struct path *path,
1647 int flag, int mode)
1648{
1649 int error;
1650 struct dentry *dir = nd->dentry;
1651
1652 if (!IS_POSIXACL(dir->d_inode))
1653 mode &= ~current->fs->umask;
1654 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1655 mutex_unlock(&dir->d_inode->i_mutex);
1656 dput(nd->dentry);
1657 nd->dentry = path->dentry;
1658 if (error)
1659 return error;
1660 /* Don't check for write permission, don't truncate */
1661 return may_open(nd, 0, flag & ~O_TRUNC);
1662}
1663
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664/*
1665 * open_namei()
1666 *
1667 * namei for open - this is in fact almost the whole open-routine.
1668 *
1669 * Note that the low bits of "flag" aren't the same as in the open
1670 * system call - they are 00 - no permissions needed
1671 * 01 - read permission needed
1672 * 10 - write permission needed
1673 * 11 - read/write permissions needed
1674 * which is a lot more logical, and also allows the "no perm" needed
1675 * for symlinks (where the permissions are checked later).
1676 * SMP-safe
1677 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001678int open_namei(int dfd, const char *pathname, int flag,
1679 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001681 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001682 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 struct dentry *dir;
1684 int count = 0;
1685
1686 acc_mode = ACC_MODE(flag);
1687
Trond Myklebust834f2a42005-10-18 14:20:16 -07001688 /* O_TRUNC implies we need access checks for write permissions */
1689 if (flag & O_TRUNC)
1690 acc_mode |= MAY_WRITE;
1691
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 /* Allow the LSM permission hook to distinguish append
1693 access from general write access. */
1694 if (flag & O_APPEND)
1695 acc_mode |= MAY_APPEND;
1696
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 /*
1698 * The simplest case - just a plain lookup.
1699 */
1700 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001701 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1702 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 if (error)
1704 return error;
1705 goto ok;
1706 }
1707
1708 /*
1709 * Create - we need to know the parent.
1710 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001711 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 if (error)
1713 return error;
1714
1715 /*
1716 * We have the parent and last component. First of all, check
1717 * that we are not asked to creat(2) an obvious directory - that
1718 * will not do.
1719 */
1720 error = -EISDIR;
1721 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1722 goto exit;
1723
1724 dir = nd->dentry;
1725 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001726 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001727 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001728 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
1730do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001731 error = PTR_ERR(path.dentry);
1732 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001733 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 goto exit;
1735 }
1736
Oleg Drokin4af4c522006-03-25 03:06:54 -08001737 if (IS_ERR(nd->intent.open.file)) {
1738 mutex_unlock(&dir->d_inode->i_mutex);
1739 error = PTR_ERR(nd->intent.open.file);
1740 goto exit_dput;
1741 }
1742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001744 if (!path.dentry->d_inode) {
Dave Hansenaab520e2006-09-30 23:29:02 -07001745 error = open_namei_create(nd, &path, flag, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 if (error)
1747 goto exit;
Dave Hansenaab520e2006-09-30 23:29:02 -07001748 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 }
1750
1751 /*
1752 * It already exists.
1753 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001754 mutex_unlock(&dir->d_inode->i_mutex);
Amy Griffis4fc03b92007-02-13 14:15:01 -05001755 audit_inode(pathname, path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756
1757 error = -EEXIST;
1758 if (flag & O_EXCL)
1759 goto exit_dput;
1760
Al Viroe13b2102005-06-06 13:36:06 -07001761 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001763 if (flag & O_NOFOLLOW)
1764 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001766
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001768 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001770 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 goto do_link;
1772
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001773 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001775 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 goto exit;
1777ok:
1778 error = may_open(nd, acc_mode, flag);
1779 if (error)
1780 goto exit;
1781 return 0;
1782
1783exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001784 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001786 if (!IS_ERR(nd->intent.open.file))
1787 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 path_release(nd);
1789 return error;
1790
1791do_link:
1792 error = -ELOOP;
1793 if (flag & O_NOFOLLOW)
1794 goto exit_dput;
1795 /*
1796 * This is subtle. Instead of calling do_follow_link() we do the
1797 * thing by hands. The reason is that this way we have zero link_count
1798 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1799 * After that we have the parent and last component, i.e.
1800 * we are in the same situation as after the first path_walk().
1801 * Well, almost - if the last component is normal we get its copy
1802 * stored in nd->last.name and we will have to putname() it when we
1803 * are done. Procfs-like symlinks just set LAST_BIND.
1804 */
1805 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001806 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 if (error)
1808 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001809 error = __do_follow_link(&path, nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001810 if (error) {
1811 /* Does someone understand code flow here? Or it is only
1812 * me so stupid? Anathema to whoever designed this non-sense
1813 * with "intent.open".
1814 */
1815 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 return error;
Kirill Korotaevde459212006-07-14 00:23:49 -07001817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001819 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 error = -EISDIR;
1822 if (nd->last_type != LAST_NORM)
1823 goto exit;
1824 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001825 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 goto exit;
1827 }
1828 error = -ELOOP;
1829 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001830 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 goto exit;
1832 }
1833 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001834 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001835 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001836 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001837 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 goto do_last;
1839}
1840
1841/**
1842 * lookup_create - lookup a dentry, creating it if it doesn't exist
1843 * @nd: nameidata info
1844 * @is_dir: directory flag
1845 *
1846 * Simple function to lookup and return a dentry and create it
1847 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001848 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001849 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 */
1851struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1852{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001853 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854
Ingo Molnarf2eace22006-07-03 00:25:05 -07001855 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001856 /*
1857 * Yucky last component or no last component at all?
1858 * (foo/., foo/.., /////)
1859 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 if (nd->last_type != LAST_NORM)
1861 goto fail;
1862 nd->flags &= ~LOOKUP_PARENT;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001863 nd->flags |= LOOKUP_CREATE;
1864 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001865
1866 /*
1867 * Do the final lookup.
1868 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001869 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 if (IS_ERR(dentry))
1871 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001872
1873 /*
1874 * Special case - lookup gave negative, but... we had foo/bar/
1875 * From the vfs_mknod() POV we just have a negative dentry -
1876 * all is fine. Let's be bastards - you had / on the end, you've
1877 * been asking for (non-existent) directory. -ENOENT for you.
1878 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1880 goto enoent;
1881 return dentry;
1882enoent:
1883 dput(dentry);
1884 dentry = ERR_PTR(-ENOENT);
1885fail:
1886 return dentry;
1887}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001888EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
1890int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1891{
1892 int error = may_create(dir, dentry, NULL);
1893
1894 if (error)
1895 return error;
1896
1897 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1898 return -EPERM;
1899
1900 if (!dir->i_op || !dir->i_op->mknod)
1901 return -EPERM;
1902
1903 error = security_inode_mknod(dir, dentry, mode, dev);
1904 if (error)
1905 return error;
1906
1907 DQUOT_INIT(dir);
1908 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001909 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001910 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 return error;
1912}
1913
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001914asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1915 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916{
1917 int error = 0;
1918 char * tmp;
1919 struct dentry * dentry;
1920 struct nameidata nd;
1921
1922 if (S_ISDIR(mode))
1923 return -EPERM;
1924 tmp = getname(filename);
1925 if (IS_ERR(tmp))
1926 return PTR_ERR(tmp);
1927
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001928 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 if (error)
1930 goto out;
1931 dentry = lookup_create(&nd, 0);
1932 error = PTR_ERR(dentry);
1933
1934 if (!IS_POSIXACL(nd.dentry->d_inode))
1935 mode &= ~current->fs->umask;
1936 if (!IS_ERR(dentry)) {
1937 switch (mode & S_IFMT) {
1938 case 0: case S_IFREG:
1939 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1940 break;
1941 case S_IFCHR: case S_IFBLK:
1942 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1943 new_decode_dev(dev));
1944 break;
1945 case S_IFIFO: case S_IFSOCK:
1946 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1947 break;
1948 case S_IFDIR:
1949 error = -EPERM;
1950 break;
1951 default:
1952 error = -EINVAL;
1953 }
1954 dput(dentry);
1955 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001956 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 path_release(&nd);
1958out:
1959 putname(tmp);
1960
1961 return error;
1962}
1963
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001964asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1965{
1966 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1967}
1968
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1970{
1971 int error = may_create(dir, dentry, NULL);
1972
1973 if (error)
1974 return error;
1975
1976 if (!dir->i_op || !dir->i_op->mkdir)
1977 return -EPERM;
1978
1979 mode &= (S_IRWXUGO|S_ISVTX);
1980 error = security_inode_mkdir(dir, dentry, mode);
1981 if (error)
1982 return error;
1983
1984 DQUOT_INIT(dir);
1985 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001986 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001987 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 return error;
1989}
1990
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001991asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992{
1993 int error = 0;
1994 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07001995 struct dentry *dentry;
1996 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
1998 tmp = getname(pathname);
1999 error = PTR_ERR(tmp);
Dave Hansen6902d922006-09-30 23:29:01 -07002000 if (IS_ERR(tmp))
2001 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
Dave Hansen6902d922006-09-30 23:29:01 -07002003 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
2004 if (error)
2005 goto out;
2006 dentry = lookup_create(&nd, 1);
2007 error = PTR_ERR(dentry);
2008 if (IS_ERR(dentry))
2009 goto out_unlock;
2010
2011 if (!IS_POSIXACL(nd.dentry->d_inode))
2012 mode &= ~current->fs->umask;
2013 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
2014 dput(dentry);
2015out_unlock:
2016 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2017 path_release(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018out:
Dave Hansen6902d922006-09-30 23:29:01 -07002019 putname(tmp);
2020out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 return error;
2022}
2023
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002024asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2025{
2026 return sys_mkdirat(AT_FDCWD, pathname, mode);
2027}
2028
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029/*
2030 * We try to drop the dentry early: we should have
2031 * a usage count of 2 if we're the only user of this
2032 * dentry, and if that is true (possibly after pruning
2033 * the dcache), then we drop the dentry now.
2034 *
2035 * A low-level filesystem can, if it choses, legally
2036 * do a
2037 *
2038 * if (!d_unhashed(dentry))
2039 * return -EBUSY;
2040 *
2041 * if it cannot handle the case of removing a directory
2042 * that is still in use by something else..
2043 */
2044void dentry_unhash(struct dentry *dentry)
2045{
2046 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002047 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 spin_lock(&dcache_lock);
2049 spin_lock(&dentry->d_lock);
2050 if (atomic_read(&dentry->d_count) == 2)
2051 __d_drop(dentry);
2052 spin_unlock(&dentry->d_lock);
2053 spin_unlock(&dcache_lock);
2054}
2055
2056int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2057{
2058 int error = may_delete(dir, dentry, 1);
2059
2060 if (error)
2061 return error;
2062
2063 if (!dir->i_op || !dir->i_op->rmdir)
2064 return -EPERM;
2065
2066 DQUOT_INIT(dir);
2067
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002068 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 dentry_unhash(dentry);
2070 if (d_mountpoint(dentry))
2071 error = -EBUSY;
2072 else {
2073 error = security_inode_rmdir(dir, dentry);
2074 if (!error) {
2075 error = dir->i_op->rmdir(dir, dentry);
2076 if (!error)
2077 dentry->d_inode->i_flags |= S_DEAD;
2078 }
2079 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002080 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 d_delete(dentry);
2083 }
2084 dput(dentry);
2085
2086 return error;
2087}
2088
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002089static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090{
2091 int error = 0;
2092 char * name;
2093 struct dentry *dentry;
2094 struct nameidata nd;
2095
2096 name = getname(pathname);
2097 if(IS_ERR(name))
2098 return PTR_ERR(name);
2099
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002100 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 if (error)
2102 goto exit;
2103
2104 switch(nd.last_type) {
2105 case LAST_DOTDOT:
2106 error = -ENOTEMPTY;
2107 goto exit1;
2108 case LAST_DOT:
2109 error = -EINVAL;
2110 goto exit1;
2111 case LAST_ROOT:
2112 error = -EBUSY;
2113 goto exit1;
2114 }
Ingo Molnarf2eace22006-07-03 00:25:05 -07002115 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002116 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002118 if (IS_ERR(dentry))
2119 goto exit2;
2120 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2121 dput(dentry);
2122exit2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002123 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124exit1:
2125 path_release(&nd);
2126exit:
2127 putname(name);
2128 return error;
2129}
2130
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002131asmlinkage long sys_rmdir(const char __user *pathname)
2132{
2133 return do_rmdir(AT_FDCWD, pathname);
2134}
2135
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136int vfs_unlink(struct inode *dir, struct dentry *dentry)
2137{
2138 int error = may_delete(dir, dentry, 0);
2139
2140 if (error)
2141 return error;
2142
2143 if (!dir->i_op || !dir->i_op->unlink)
2144 return -EPERM;
2145
2146 DQUOT_INIT(dir);
2147
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002148 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 if (d_mountpoint(dentry))
2150 error = -EBUSY;
2151 else {
2152 error = security_inode_unlink(dir, dentry);
2153 if (!error)
2154 error = dir->i_op->unlink(dir, dentry);
2155 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002156 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157
2158 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2159 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2160 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 }
Robert Love0eeca282005-07-12 17:06:03 -04002162
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 return error;
2164}
2165
2166/*
2167 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002168 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 * writeout happening, and we don't want to prevent access to the directory
2170 * while waiting on the I/O.
2171 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002172static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173{
2174 int error = 0;
2175 char * name;
2176 struct dentry *dentry;
2177 struct nameidata nd;
2178 struct inode *inode = NULL;
2179
2180 name = getname(pathname);
2181 if(IS_ERR(name))
2182 return PTR_ERR(name);
2183
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002184 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 if (error)
2186 goto exit;
2187 error = -EISDIR;
2188 if (nd.last_type != LAST_NORM)
2189 goto exit1;
Ingo Molnarf2eace22006-07-03 00:25:05 -07002190 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002191 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 error = PTR_ERR(dentry);
2193 if (!IS_ERR(dentry)) {
2194 /* Why not before? Because we want correct error value */
2195 if (nd.last.name[nd.last.len])
2196 goto slashes;
2197 inode = dentry->d_inode;
2198 if (inode)
2199 atomic_inc(&inode->i_count);
2200 error = vfs_unlink(nd.dentry->d_inode, dentry);
2201 exit2:
2202 dput(dentry);
2203 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002204 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 if (inode)
2206 iput(inode); /* truncate the inode here */
2207exit1:
2208 path_release(&nd);
2209exit:
2210 putname(name);
2211 return error;
2212
2213slashes:
2214 error = !dentry->d_inode ? -ENOENT :
2215 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2216 goto exit2;
2217}
2218
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002219asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2220{
2221 if ((flag & ~AT_REMOVEDIR) != 0)
2222 return -EINVAL;
2223
2224 if (flag & AT_REMOVEDIR)
2225 return do_rmdir(dfd, pathname);
2226
2227 return do_unlinkat(dfd, pathname);
2228}
2229
2230asmlinkage long sys_unlink(const char __user *pathname)
2231{
2232 return do_unlinkat(AT_FDCWD, pathname);
2233}
2234
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2236{
2237 int error = may_create(dir, dentry, NULL);
2238
2239 if (error)
2240 return error;
2241
2242 if (!dir->i_op || !dir->i_op->symlink)
2243 return -EPERM;
2244
2245 error = security_inode_symlink(dir, dentry, oldname);
2246 if (error)
2247 return error;
2248
2249 DQUOT_INIT(dir);
2250 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002251 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002252 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 return error;
2254}
2255
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002256asmlinkage long sys_symlinkat(const char __user *oldname,
2257 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258{
2259 int error = 0;
2260 char * from;
2261 char * to;
Dave Hansen6902d922006-09-30 23:29:01 -07002262 struct dentry *dentry;
2263 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
2265 from = getname(oldname);
2266 if(IS_ERR(from))
2267 return PTR_ERR(from);
2268 to = getname(newname);
2269 error = PTR_ERR(to);
Dave Hansen6902d922006-09-30 23:29:01 -07002270 if (IS_ERR(to))
2271 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
Dave Hansen6902d922006-09-30 23:29:01 -07002273 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2274 if (error)
2275 goto out;
2276 dentry = lookup_create(&nd, 0);
2277 error = PTR_ERR(dentry);
2278 if (IS_ERR(dentry))
2279 goto out_unlock;
2280
2281 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2282 dput(dentry);
2283out_unlock:
2284 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2285 path_release(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286out:
Dave Hansen6902d922006-09-30 23:29:01 -07002287 putname(to);
2288out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 putname(from);
2290 return error;
2291}
2292
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002293asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2294{
2295 return sys_symlinkat(oldname, AT_FDCWD, newname);
2296}
2297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2299{
2300 struct inode *inode = old_dentry->d_inode;
2301 int error;
2302
2303 if (!inode)
2304 return -ENOENT;
2305
2306 error = may_create(dir, new_dentry, NULL);
2307 if (error)
2308 return error;
2309
2310 if (dir->i_sb != inode->i_sb)
2311 return -EXDEV;
2312
2313 /*
2314 * A link to an append-only or immutable file cannot be created.
2315 */
2316 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2317 return -EPERM;
2318 if (!dir->i_op || !dir->i_op->link)
2319 return -EPERM;
2320 if (S_ISDIR(old_dentry->d_inode->i_mode))
2321 return -EPERM;
2322
2323 error = security_inode_link(old_dentry, dir, new_dentry);
2324 if (error)
2325 return error;
2326
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002327 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 DQUOT_INIT(dir);
2329 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002330 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002331 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002332 fsnotify_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 return error;
2334}
2335
2336/*
2337 * Hardlinks are often used in delicate situations. We avoid
2338 * security-related surprises by not following symlinks on the
2339 * newname. --KAB
2340 *
2341 * We don't follow them on the oldname either to be compatible
2342 * with linux 2.0, and to avoid hard-linking to directories
2343 * and other special files. --ADM
2344 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002345asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002346 int newdfd, const char __user *newname,
2347 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348{
2349 struct dentry *new_dentry;
2350 struct nameidata nd, old_nd;
2351 int error;
2352 char * to;
2353
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002354 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002355 return -EINVAL;
2356
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 to = getname(newname);
2358 if (IS_ERR(to))
2359 return PTR_ERR(to);
2360
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002361 error = __user_walk_fd(olddfd, oldname,
2362 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2363 &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 if (error)
2365 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002366 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 if (error)
2368 goto out;
2369 error = -EXDEV;
2370 if (old_nd.mnt != nd.mnt)
2371 goto out_release;
2372 new_dentry = lookup_create(&nd, 0);
2373 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002374 if (IS_ERR(new_dentry))
2375 goto out_unlock;
2376 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2377 dput(new_dentry);
2378out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002379 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380out_release:
2381 path_release(&nd);
2382out:
2383 path_release(&old_nd);
2384exit:
2385 putname(to);
2386
2387 return error;
2388}
2389
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002390asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2391{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002392 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002393}
2394
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395/*
2396 * The worst of all namespace operations - renaming directory. "Perverted"
2397 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2398 * Problems:
2399 * a) we can get into loop creation. Check is done in is_subdir().
2400 * b) race potential - two innocent renames can create a loop together.
2401 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002402 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 * story.
2404 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002405 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 * whether the target exists). Solution: try to be smart with locking
2407 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002408 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 * move will be locked. Thus we can rank directories by the tree
2410 * (ancestors first) and rank all non-directories after them.
2411 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002412 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 * HOWEVER, it relies on the assumption that any object with ->lookup()
2414 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2415 * we'd better make sure that there's no link(2) for them.
2416 * d) some filesystems don't support opened-but-unlinked directories,
2417 * either because of layout or because they are not ready to deal with
2418 * all cases correctly. The latter will be fixed (taking this sort of
2419 * stuff into VFS), but the former is not going away. Solution: the same
2420 * trick as in rmdir().
2421 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002422 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002424 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425 * locking].
2426 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002427static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2428 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429{
2430 int error = 0;
2431 struct inode *target;
2432
2433 /*
2434 * If we are going to change the parent - check write permissions,
2435 * we'll need to flip '..'.
2436 */
2437 if (new_dir != old_dir) {
2438 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2439 if (error)
2440 return error;
2441 }
2442
2443 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2444 if (error)
2445 return error;
2446
2447 target = new_dentry->d_inode;
2448 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002449 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 dentry_unhash(new_dentry);
2451 }
2452 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2453 error = -EBUSY;
2454 else
2455 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2456 if (target) {
2457 if (!error)
2458 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002459 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002460 if (d_unhashed(new_dentry))
2461 d_rehash(new_dentry);
2462 dput(new_dentry);
2463 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002464 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002465 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2466 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 return error;
2468}
2469
Adrian Bunk75c96f82005-05-05 16:16:09 -07002470static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2471 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472{
2473 struct inode *target;
2474 int error;
2475
2476 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2477 if (error)
2478 return error;
2479
2480 dget(new_dentry);
2481 target = new_dentry->d_inode;
2482 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002483 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2485 error = -EBUSY;
2486 else
2487 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2488 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002489 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 }
2492 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002493 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 dput(new_dentry);
2495 return error;
2496}
2497
2498int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2499 struct inode *new_dir, struct dentry *new_dentry)
2500{
2501 int error;
2502 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002503 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504
2505 if (old_dentry->d_inode == new_dentry->d_inode)
2506 return 0;
2507
2508 error = may_delete(old_dir, old_dentry, is_dir);
2509 if (error)
2510 return error;
2511
2512 if (!new_dentry->d_inode)
2513 error = may_create(new_dir, new_dentry, NULL);
2514 else
2515 error = may_delete(new_dir, new_dentry, is_dir);
2516 if (error)
2517 return error;
2518
2519 if (!old_dir->i_op || !old_dir->i_op->rename)
2520 return -EPERM;
2521
2522 DQUOT_INIT(old_dir);
2523 DQUOT_INIT(new_dir);
2524
Robert Love0eeca282005-07-12 17:06:03 -04002525 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2526
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 if (is_dir)
2528 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2529 else
2530 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2531 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002532 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002533 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2534 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 }
Robert Love0eeca282005-07-12 17:06:03 -04002536 fsnotify_oldname_free(old_name);
2537
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 return error;
2539}
2540
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002541static int do_rename(int olddfd, const char *oldname,
2542 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543{
2544 int error = 0;
2545 struct dentry * old_dir, * new_dir;
2546 struct dentry * old_dentry, *new_dentry;
2547 struct dentry * trap;
2548 struct nameidata oldnd, newnd;
2549
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002550 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 if (error)
2552 goto exit;
2553
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002554 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 if (error)
2556 goto exit1;
2557
2558 error = -EXDEV;
2559 if (oldnd.mnt != newnd.mnt)
2560 goto exit2;
2561
2562 old_dir = oldnd.dentry;
2563 error = -EBUSY;
2564 if (oldnd.last_type != LAST_NORM)
2565 goto exit2;
2566
2567 new_dir = newnd.dentry;
2568 if (newnd.last_type != LAST_NORM)
2569 goto exit2;
2570
2571 trap = lock_rename(new_dir, old_dir);
2572
Christoph Hellwig49705b72005-11-08 21:35:06 -08002573 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 error = PTR_ERR(old_dentry);
2575 if (IS_ERR(old_dentry))
2576 goto exit3;
2577 /* source must exist */
2578 error = -ENOENT;
2579 if (!old_dentry->d_inode)
2580 goto exit4;
2581 /* unless the source is a directory trailing slashes give -ENOTDIR */
2582 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2583 error = -ENOTDIR;
2584 if (oldnd.last.name[oldnd.last.len])
2585 goto exit4;
2586 if (newnd.last.name[newnd.last.len])
2587 goto exit4;
2588 }
2589 /* source should not be ancestor of target */
2590 error = -EINVAL;
2591 if (old_dentry == trap)
2592 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002593 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 error = PTR_ERR(new_dentry);
2595 if (IS_ERR(new_dentry))
2596 goto exit4;
2597 /* target should not be an ancestor of source */
2598 error = -ENOTEMPTY;
2599 if (new_dentry == trap)
2600 goto exit5;
2601
2602 error = vfs_rename(old_dir->d_inode, old_dentry,
2603 new_dir->d_inode, new_dentry);
2604exit5:
2605 dput(new_dentry);
2606exit4:
2607 dput(old_dentry);
2608exit3:
2609 unlock_rename(new_dir, old_dir);
2610exit2:
2611 path_release(&newnd);
2612exit1:
2613 path_release(&oldnd);
2614exit:
2615 return error;
2616}
2617
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002618asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2619 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620{
2621 int error;
2622 char * from;
2623 char * to;
2624
2625 from = getname(oldname);
2626 if(IS_ERR(from))
2627 return PTR_ERR(from);
2628 to = getname(newname);
2629 error = PTR_ERR(to);
2630 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002631 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 putname(to);
2633 }
2634 putname(from);
2635 return error;
2636}
2637
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002638asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2639{
2640 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2641}
2642
Linus Torvalds1da177e2005-04-16 15:20:36 -07002643int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2644{
2645 int len;
2646
2647 len = PTR_ERR(link);
2648 if (IS_ERR(link))
2649 goto out;
2650
2651 len = strlen(link);
2652 if (len > (unsigned) buflen)
2653 len = buflen;
2654 if (copy_to_user(buffer, link, len))
2655 len = -EFAULT;
2656out:
2657 return len;
2658}
2659
2660/*
2661 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2662 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2663 * using) it for any given inode is up to filesystem.
2664 */
2665int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2666{
2667 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002668 void *cookie;
2669
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002671 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2672 if (!IS_ERR(cookie)) {
2673 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002675 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2676 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002678 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679}
2680
2681int vfs_follow_link(struct nameidata *nd, const char *link)
2682{
2683 return __vfs_follow_link(nd, link);
2684}
2685
2686/* get the link contents into pagecache */
2687static char *page_getlink(struct dentry * dentry, struct page **ppage)
2688{
2689 struct page * page;
2690 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002691 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 if (IS_ERR(page))
Nick Piggin6fe69002007-05-06 14:49:04 -07002693 return (char*)page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002694 *ppage = page;
2695 return kmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696}
2697
2698int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2699{
2700 struct page *page = NULL;
2701 char *s = page_getlink(dentry, &page);
2702 int res = vfs_readlink(dentry,buffer,buflen,s);
2703 if (page) {
2704 kunmap(page);
2705 page_cache_release(page);
2706 }
2707 return res;
2708}
2709
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002710void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002712 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002714 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715}
2716
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002717void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002719 struct page *page = cookie;
2720
2721 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 kunmap(page);
2723 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 }
2725}
2726
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002727int __page_symlink(struct inode *inode, const char *symname, int len,
2728 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729{
2730 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002731 struct page *page;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002732 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733 char *kaddr;
2734
NeilBrown7e53cac2006-03-25 03:07:57 -08002735retry:
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002736 err = -ENOMEM;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002737 page = find_or_create_page(mapping, 0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738 if (!page)
2739 goto fail;
2740 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
NeilBrown7e53cac2006-03-25 03:07:57 -08002741 if (err == AOP_TRUNCATED_PAGE) {
2742 page_cache_release(page);
2743 goto retry;
2744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 if (err)
2746 goto fail_map;
2747 kaddr = kmap_atomic(page, KM_USER0);
2748 memcpy(kaddr, symname, len-1);
2749 kunmap_atomic(kaddr, KM_USER0);
NeilBrown7e53cac2006-03-25 03:07:57 -08002750 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2751 if (err == AOP_TRUNCATED_PAGE) {
2752 page_cache_release(page);
2753 goto retry;
2754 }
2755 if (err)
2756 goto fail_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 /*
2758 * Notice that we are _not_ going to block here - end of page is
2759 * unmapped, so this will only try to map the rest of page, see
2760 * that it is unmapped (typically even will not look into inode -
2761 * ->i_size will be enough for everything) and zero it out.
2762 * OTOH it's obviously correct and should make the page up-to-date.
2763 */
2764 if (!PageUptodate(page)) {
2765 err = mapping->a_ops->readpage(NULL, page);
NeilBrown7e53cac2006-03-25 03:07:57 -08002766 if (err != AOP_TRUNCATED_PAGE)
2767 wait_on_page_locked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 } else {
2769 unlock_page(page);
2770 }
2771 page_cache_release(page);
2772 if (err < 0)
2773 goto fail;
2774 mark_inode_dirty(inode);
2775 return 0;
2776fail_map:
2777 unlock_page(page);
2778 page_cache_release(page);
2779fail:
2780 return err;
2781}
2782
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002783int page_symlink(struct inode *inode, const char *symname, int len)
2784{
2785 return __page_symlink(inode, symname, len,
2786 mapping_gfp_mask(inode->i_mapping));
2787}
2788
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002789const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 .readlink = generic_readlink,
2791 .follow_link = page_follow_link_light,
2792 .put_link = page_put_link,
2793};
2794
2795EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002796EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797EXPORT_SYMBOL(follow_down);
2798EXPORT_SYMBOL(follow_up);
2799EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2800EXPORT_SYMBOL(getname);
2801EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802EXPORT_SYMBOL(lookup_one_len);
2803EXPORT_SYMBOL(page_follow_link_light);
2804EXPORT_SYMBOL(page_put_link);
2805EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002806EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807EXPORT_SYMBOL(page_symlink);
2808EXPORT_SYMBOL(page_symlink_inode_operations);
2809EXPORT_SYMBOL(path_lookup);
Josef 'Jeff' Sipek16f18202007-07-19 01:48:18 -07002810EXPORT_SYMBOL(vfs_path_lookup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811EXPORT_SYMBOL(path_release);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002813EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002814EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002815EXPORT_SYMBOL(unlock_rename);
2816EXPORT_SYMBOL(vfs_create);
2817EXPORT_SYMBOL(vfs_follow_link);
2818EXPORT_SYMBOL(vfs_link);
2819EXPORT_SYMBOL(vfs_mkdir);
2820EXPORT_SYMBOL(vfs_mknod);
2821EXPORT_SYMBOL(generic_permission);
2822EXPORT_SYMBOL(vfs_readlink);
2823EXPORT_SYMBOL(vfs_rename);
2824EXPORT_SYMBOL(vfs_rmdir);
2825EXPORT_SYMBOL(vfs_symlink);
2826EXPORT_SYMBOL(vfs_unlink);
2827EXPORT_SYMBOL(dentry_unhash);
2828EXPORT_SYMBOL(generic_readlink);