blob: 28d49b301d5516bbdfb5bf7651f3c51b8f370f31 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/quotaops.h>
23#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/smp_lock.h>
26#include <linux/personality.h>
27#include <linux/security.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
34#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/namei.h>
36#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400162 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
170
171/**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184{
185 umode_t mode = inode->i_mode;
186
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
202 /*
203 * If the DACs are ok we don't need any capability check.
204 */
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
207
208 check_capabilities:
209 /*
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
212 */
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
217
218 /*
219 * Searching includes executable on directories, else just read.
220 */
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
224
225 return -EACCES;
226}
227
228int permission(struct inode *inode, int mask, struct nameidata *nd)
229{
Trond Myklebusta343bb72006-08-22 20:06:03 -0400230 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 int retval, submask;
232
233 if (mask & MAY_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235 /*
236 * Nobody gets write access to a read-only fs.
237 */
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
241
242 /*
243 * Nobody gets write access to an immutable file.
244 */
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
247 }
248
249
Trond Myklebusta343bb72006-08-22 20:06:03 -0400250 /*
251 * MAY_EXEC on regular files requires special handling: We override
252 * filesystem execute permissions if the mode bits aren't set.
253 */
254 if ((mask & MAY_EXEC) && S_ISREG(mode) && !(mode & S_IXUGO))
255 return -EACCES;
256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 /* Ordinary permission routines do not understand MAY_APPEND. */
258 submask = mask & ~MAY_APPEND;
259 if (inode->i_op && inode->i_op->permission)
260 retval = inode->i_op->permission(inode, submask, nd);
261 else
262 retval = generic_permission(inode, submask, NULL);
263 if (retval)
264 return retval;
265
266 return security_inode_permission(inode, mask, nd);
267}
268
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800269/**
270 * vfs_permission - check for access rights to a given path
271 * @nd: lookup result that describes the path
272 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
273 *
274 * Used to check for read/write/execute permissions on a path.
275 * We use "fsuid" for this, letting us set arbitrary permissions
276 * for filesystem access without changing the "normal" uids which
277 * are used for other things.
278 */
279int vfs_permission(struct nameidata *nd, int mask)
280{
281 return permission(nd->dentry->d_inode, mask, nd);
282}
283
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800284/**
285 * file_permission - check for additional access rights to a given file
286 * @file: file to check access rights for
287 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
288 *
289 * Used to check for read/write/execute permissions on an already opened
290 * file.
291 *
292 * Note:
293 * Do not use this function in new code. All access checks should
294 * be done using vfs_permission().
295 */
296int file_permission(struct file *file, int mask)
297{
298 return permission(file->f_dentry->d_inode, mask, NULL);
299}
300
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301/*
302 * get_write_access() gets write permission for a file.
303 * put_write_access() releases this write permission.
304 * This is used for regular files.
305 * We cannot support write (and maybe mmap read-write shared) accesses and
306 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
307 * can have the following values:
308 * 0: no writers, no VM_DENYWRITE mappings
309 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
310 * > 0: (i_writecount) users are writing to the file.
311 *
312 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
313 * except for the cases where we don't hold i_writecount yet. Then we need to
314 * use {get,deny}_write_access() - these functions check the sign and refuse
315 * to do the change if sign is wrong. Exclusion between them is provided by
316 * the inode->i_lock spinlock.
317 */
318
319int get_write_access(struct inode * inode)
320{
321 spin_lock(&inode->i_lock);
322 if (atomic_read(&inode->i_writecount) < 0) {
323 spin_unlock(&inode->i_lock);
324 return -ETXTBSY;
325 }
326 atomic_inc(&inode->i_writecount);
327 spin_unlock(&inode->i_lock);
328
329 return 0;
330}
331
332int deny_write_access(struct file * file)
333{
334 struct inode *inode = file->f_dentry->d_inode;
335
336 spin_lock(&inode->i_lock);
337 if (atomic_read(&inode->i_writecount) > 0) {
338 spin_unlock(&inode->i_lock);
339 return -ETXTBSY;
340 }
341 atomic_dec(&inode->i_writecount);
342 spin_unlock(&inode->i_lock);
343
344 return 0;
345}
346
347void path_release(struct nameidata *nd)
348{
349 dput(nd->dentry);
350 mntput(nd->mnt);
351}
352
353/*
354 * umount() mustn't call path_release()/mntput() as that would clear
355 * mnt_expiry_mark
356 */
357void path_release_on_umount(struct nameidata *nd)
358{
359 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700360 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361}
362
Trond Myklebust834f2a42005-10-18 14:20:16 -0700363/**
364 * release_open_intent - free up open intent resources
365 * @nd: pointer to nameidata
366 */
367void release_open_intent(struct nameidata *nd)
368{
369 if (nd->intent.open.file->f_dentry == NULL)
370 put_filp(nd->intent.open.file);
371 else
372 fput(nd->intent.open.file);
373}
374
Ian Kentbcdc5e02006-09-27 01:50:44 -0700375static inline struct dentry *
376do_revalidate(struct dentry *dentry, struct nameidata *nd)
377{
378 int status = dentry->d_op->d_revalidate(dentry, nd);
379 if (unlikely(status <= 0)) {
380 /*
381 * The dentry failed validation.
382 * If d_revalidate returned 0 attempt to invalidate
383 * the dentry otherwise d_revalidate is asking us
384 * to return a fail status.
385 */
386 if (!status) {
387 if (!d_invalidate(dentry)) {
388 dput(dentry);
389 dentry = NULL;
390 }
391 } else {
392 dput(dentry);
393 dentry = ERR_PTR(status);
394 }
395 }
396 return dentry;
397}
398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399/*
400 * Internal lookup() using the new generic dcache.
401 * SMP-safe
402 */
403static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
404{
405 struct dentry * dentry = __d_lookup(parent, name);
406
407 /* lockess __d_lookup may fail due to concurrent d_move()
408 * in some unrelated directory, so try with d_lookup
409 */
410 if (!dentry)
411 dentry = d_lookup(parent, name);
412
Ian Kentbcdc5e02006-09-27 01:50:44 -0700413 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
414 dentry = do_revalidate(dentry, nd);
415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 return dentry;
417}
418
419/*
420 * Short-cut version of permission(), for calling by
421 * path_walk(), when dcache lock is held. Combines parts
422 * of permission() and generic_permission(), and tests ONLY for
423 * MAY_EXEC permission.
424 *
425 * If appropriate, check DAC only. If not appropriate, or
426 * short-cut DAC fails, then call permission() to do more
427 * complete permission check.
428 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800429static int exec_permission_lite(struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 struct nameidata *nd)
431{
432 umode_t mode = inode->i_mode;
433
434 if (inode->i_op && inode->i_op->permission)
435 return -EAGAIN;
436
437 if (current->fsuid == inode->i_uid)
438 mode >>= 6;
439 else if (in_group_p(inode->i_gid))
440 mode >>= 3;
441
442 if (mode & MAY_EXEC)
443 goto ok;
444
445 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
446 goto ok;
447
448 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
449 goto ok;
450
451 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
452 goto ok;
453
454 return -EACCES;
455ok:
456 return security_inode_permission(inode, MAY_EXEC, nd);
457}
458
459/*
460 * This is called when everything else fails, and we actually have
461 * to go to the low-level filesystem to find out what we should do..
462 *
463 * We get the directory semaphore, and after getting that we also
464 * make sure that nobody added the entry to the dcache in the meantime..
465 * SMP-safe
466 */
467static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
468{
469 struct dentry * result;
470 struct inode *dir = parent->d_inode;
471
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800472 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 /*
474 * First re-do the cached lookup just in case it was created
475 * while we waited for the directory semaphore..
476 *
477 * FIXME! This could use version numbering or similar to
478 * avoid unnecessary cache lookups.
479 *
480 * The "dcache_lock" is purely to protect the RCU list walker
481 * from concurrent renames at this point (we mustn't get false
482 * negatives from the RCU list walk here, unlike the optimistic
483 * fast walk).
484 *
485 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
486 */
487 result = d_lookup(parent, name);
488 if (!result) {
489 struct dentry * dentry = d_alloc(parent, name);
490 result = ERR_PTR(-ENOMEM);
491 if (dentry) {
492 result = dir->i_op->lookup(dir, dentry, nd);
493 if (result)
494 dput(dentry);
495 else
496 result = dentry;
497 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800498 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 return result;
500 }
501
502 /*
503 * Uhhuh! Nasty case: the cache was re-populated while
504 * we waited on the semaphore. Need to revalidate.
505 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800506 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700508 result = do_revalidate(result, nd);
509 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 }
512 return result;
513}
514
515static int __emul_lookup_dentry(const char *, struct nameidata *);
516
517/* SMP-safe */
Arjan van de Venf1662352006-01-14 13:21:31 -0800518static __always_inline int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519walk_init_root(const char *name, struct nameidata *nd)
520{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700521 struct fs_struct *fs = current->fs;
522
523 read_lock(&fs->lock);
524 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
525 nd->mnt = mntget(fs->altrootmnt);
526 nd->dentry = dget(fs->altroot);
527 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 if (__emul_lookup_dentry(name,nd))
529 return 0;
Andreas Mohre518ddb2006-09-29 02:01:22 -0700530 read_lock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700532 nd->mnt = mntget(fs->rootmnt);
533 nd->dentry = dget(fs->root);
534 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 return 1;
536}
537
Arjan van de Venf1662352006-01-14 13:21:31 -0800538static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539{
540 int res = 0;
541 char *name;
542 if (IS_ERR(link))
543 goto fail;
544
545 if (*link == '/') {
546 path_release(nd);
547 if (!walk_init_root(link, nd))
548 /* weird __emul_prefix() stuff did it */
549 goto out;
550 }
551 res = link_path_walk(link, nd);
552out:
553 if (nd->depth || res || nd->last_type!=LAST_NORM)
554 return res;
555 /*
556 * If it is an iterative symlinks resolution in open_namei() we
557 * have to copy the last component. And all that crap because of
558 * bloody create() on broken symlinks. Furrfu...
559 */
560 name = __getname();
561 if (unlikely(!name)) {
562 path_release(nd);
563 return -ENOMEM;
564 }
565 strcpy(name, nd->last.name);
566 nd->last.name = name;
567 return 0;
568fail:
569 path_release(nd);
570 return PTR_ERR(link);
571}
572
Al Viro90ebe562005-06-06 13:35:58 -0700573struct path {
574 struct vfsmount *mnt;
575 struct dentry *dentry;
576};
577
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700578static inline void dput_path(struct path *path, struct nameidata *nd)
579{
580 dput(path->dentry);
581 if (path->mnt != nd->mnt)
582 mntput(path->mnt);
583}
584
585static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
586{
587 dput(nd->dentry);
588 if (nd->mnt != path->mnt)
589 mntput(nd->mnt);
590 nd->mnt = path->mnt;
591 nd->dentry = path->dentry;
592}
593
Ian Kent051d3812006-03-27 01:14:53 -0800594static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
595{
596 int error;
597 void *cookie;
598 struct dentry *dentry = path->dentry;
599
600 touch_atime(path->mnt, dentry);
601 nd_set_link(nd, NULL);
602
603 if (path->mnt != nd->mnt) {
604 path_to_nameidata(path, nd);
605 dget(dentry);
606 }
607 mntget(path->mnt);
608 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
609 error = PTR_ERR(cookie);
610 if (!IS_ERR(cookie)) {
611 char *s = nd_get_link(nd);
612 error = 0;
613 if (s)
614 error = __vfs_follow_link(nd, s);
615 if (dentry->d_inode->i_op->put_link)
616 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
617 }
618 dput(dentry);
619 mntput(path->mnt);
620
621 return error;
622}
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624/*
625 * This limits recursive symlink follows to 8, while
626 * limiting consecutive symlinks to 40.
627 *
628 * Without that kind of total limit, nasty chains of consecutive
629 * symlinks can cause almost arbitrarily long lookups.
630 */
Al Viro90ebe562005-06-06 13:35:58 -0700631static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
633 int err = -ELOOP;
634 if (current->link_count >= MAX_NESTED_LINKS)
635 goto loop;
636 if (current->total_link_count >= 40)
637 goto loop;
638 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
639 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700640 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 if (err)
642 goto loop;
643 current->link_count++;
644 current->total_link_count++;
645 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700646 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 current->link_count--;
648 nd->depth--;
649 return err;
650loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700651 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700652 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 return err;
654}
655
656int follow_up(struct vfsmount **mnt, struct dentry **dentry)
657{
658 struct vfsmount *parent;
659 struct dentry *mountpoint;
660 spin_lock(&vfsmount_lock);
661 parent=(*mnt)->mnt_parent;
662 if (parent == *mnt) {
663 spin_unlock(&vfsmount_lock);
664 return 0;
665 }
666 mntget(parent);
667 mountpoint=dget((*mnt)->mnt_mountpoint);
668 spin_unlock(&vfsmount_lock);
669 dput(*dentry);
670 *dentry = mountpoint;
671 mntput(*mnt);
672 *mnt = parent;
673 return 1;
674}
675
676/* no need for dcache_lock, as serialization is taken care in
677 * namespace.c
678 */
Al Viro463ffb22005-06-06 13:36:05 -0700679static int __follow_mount(struct path *path)
680{
681 int res = 0;
682 while (d_mountpoint(path->dentry)) {
683 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
684 if (!mounted)
685 break;
686 dput(path->dentry);
687 if (res)
688 mntput(path->mnt);
689 path->mnt = mounted;
690 path->dentry = dget(mounted->mnt_root);
691 res = 1;
692 }
693 return res;
694}
695
Al Viro58c465e2005-06-06 13:36:13 -0700696static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 while (d_mountpoint(*dentry)) {
699 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
700 if (!mounted)
701 break;
Al Viro58c465e2005-06-06 13:36:13 -0700702 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 mntput(*mnt);
704 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707}
708
709/* no need for dcache_lock, as serialization is taken care in
710 * namespace.c
711 */
Al Viroe13b2102005-06-06 13:36:06 -0700712int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
714 struct vfsmount *mounted;
715
716 mounted = lookup_mnt(*mnt, *dentry);
717 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700718 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 mntput(*mnt);
720 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 *dentry = dget(mounted->mnt_root);
722 return 1;
723 }
724 return 0;
725}
726
Arjan van de Venf1662352006-01-14 13:21:31 -0800727static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700729 struct fs_struct *fs = current->fs;
730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 while(1) {
732 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700733 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Andreas Mohre518ddb2006-09-29 02:01:22 -0700735 read_lock(&fs->lock);
736 if (nd->dentry == fs->root &&
737 nd->mnt == fs->rootmnt) {
738 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 break;
740 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700741 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700743 if (nd->dentry != nd->mnt->mnt_root) {
744 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 spin_unlock(&dcache_lock);
746 dput(old);
747 break;
748 }
749 spin_unlock(&dcache_lock);
750 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700751 parent = nd->mnt->mnt_parent;
752 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 spin_unlock(&vfsmount_lock);
754 break;
755 }
756 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700757 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 spin_unlock(&vfsmount_lock);
759 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700760 mntput(nd->mnt);
761 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 }
Al Viro58c465e2005-06-06 13:36:13 -0700763 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764}
765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766/*
767 * It's more convoluted than I'd like it to be, but... it's still fairly
768 * small and for now I'd prefer to have fast path as straight as possible.
769 * It _is_ time-critical.
770 */
771static int do_lookup(struct nameidata *nd, struct qstr *name,
772 struct path *path)
773{
774 struct vfsmount *mnt = nd->mnt;
775 struct dentry *dentry = __d_lookup(nd->dentry, name);
776
777 if (!dentry)
778 goto need_lookup;
779 if (dentry->d_op && dentry->d_op->d_revalidate)
780 goto need_revalidate;
781done:
782 path->mnt = mnt;
783 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700784 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 return 0;
786
787need_lookup:
788 dentry = real_lookup(nd->dentry, name, nd);
789 if (IS_ERR(dentry))
790 goto fail;
791 goto done;
792
793need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700794 dentry = do_revalidate(dentry, nd);
795 if (!dentry)
796 goto need_lookup;
797 if (IS_ERR(dentry))
798 goto fail;
799 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
801fail:
802 return PTR_ERR(dentry);
803}
804
805/*
806 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100807 * This is the basic name resolution function, turning a pathname into
808 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100810 * Returns 0 and nd will have valid dentry and mnt on success.
811 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 */
813static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
814{
815 struct path next;
816 struct inode *inode;
817 int err;
818 unsigned int lookup_flags = nd->flags;
819
820 while (*name=='/')
821 name++;
822 if (!*name)
823 goto return_reval;
824
825 inode = nd->dentry->d_inode;
826 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800827 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
829 /* At this point we know we have a real path component. */
830 for(;;) {
831 unsigned long hash;
832 struct qstr this;
833 unsigned int c;
834
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700835 nd->flags |= LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 err = exec_permission_lite(inode, nd);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800837 if (err == -EAGAIN)
838 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 if (err)
840 break;
841
842 this.name = name;
843 c = *(const unsigned char *)name;
844
845 hash = init_name_hash();
846 do {
847 name++;
848 hash = partial_name_hash(c, hash);
849 c = *(const unsigned char *)name;
850 } while (c && (c != '/'));
851 this.len = name - (const char *) this.name;
852 this.hash = end_name_hash(hash);
853
854 /* remove trailing slashes? */
855 if (!c)
856 goto last_component;
857 while (*++name == '/');
858 if (!*name)
859 goto last_with_slashes;
860
861 /*
862 * "." and ".." are special - ".." especially so because it has
863 * to be able to know about the current root directory and
864 * parent relationships.
865 */
866 if (this.name[0] == '.') switch (this.len) {
867 default:
868 break;
869 case 2:
870 if (this.name[1] != '.')
871 break;
Al Viro58c465e2005-06-06 13:36:13 -0700872 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 inode = nd->dentry->d_inode;
874 /* fallthrough */
875 case 1:
876 continue;
877 }
878 /*
879 * See if the low-level filesystem might want
880 * to use its own hash..
881 */
882 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
883 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
884 if (err < 0)
885 break;
886 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 /* This does the actual lookups.. */
888 err = do_lookup(nd, &this, &next);
889 if (err)
890 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 err = -ENOENT;
893 inode = next.dentry->d_inode;
894 if (!inode)
895 goto out_dput;
896 err = -ENOTDIR;
897 if (!inode->i_op)
898 goto out_dput;
899
900 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700901 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 if (err)
903 goto return_err;
904 err = -ENOENT;
905 inode = nd->dentry->d_inode;
906 if (!inode)
907 break;
908 err = -ENOTDIR;
909 if (!inode->i_op)
910 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700911 } else
912 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 err = -ENOTDIR;
914 if (!inode->i_op->lookup)
915 break;
916 continue;
917 /* here ends the main loop */
918
919last_with_slashes:
920 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
921last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800922 /* Clear LOOKUP_CONTINUE iff it was previously unset */
923 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 if (lookup_flags & LOOKUP_PARENT)
925 goto lookup_parent;
926 if (this.name[0] == '.') switch (this.len) {
927 default:
928 break;
929 case 2:
930 if (this.name[1] != '.')
931 break;
Al Viro58c465e2005-06-06 13:36:13 -0700932 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 inode = nd->dentry->d_inode;
934 /* fallthrough */
935 case 1:
936 goto return_reval;
937 }
938 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
939 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
940 if (err < 0)
941 break;
942 }
943 err = do_lookup(nd, &this, &next);
944 if (err)
945 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 inode = next.dentry->d_inode;
947 if ((lookup_flags & LOOKUP_FOLLOW)
948 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700949 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 if (err)
951 goto return_err;
952 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700953 } else
954 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 err = -ENOENT;
956 if (!inode)
957 break;
958 if (lookup_flags & LOOKUP_DIRECTORY) {
959 err = -ENOTDIR;
960 if (!inode->i_op || !inode->i_op->lookup)
961 break;
962 }
963 goto return_base;
964lookup_parent:
965 nd->last = this;
966 nd->last_type = LAST_NORM;
967 if (this.name[0] != '.')
968 goto return_base;
969 if (this.len == 1)
970 nd->last_type = LAST_DOT;
971 else if (this.len == 2 && this.name[1] == '.')
972 nd->last_type = LAST_DOTDOT;
973 else
974 goto return_base;
975return_reval:
976 /*
977 * We bypassed the ordinary revalidation routines.
978 * We may need to check the cached dentry for staleness.
979 */
980 if (nd->dentry && nd->dentry->d_sb &&
981 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
982 err = -ESTALE;
983 /* Note: we do not d_invalidate() */
984 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
985 break;
986 }
987return_base:
988 return 0;
989out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700990 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 break;
992 }
993 path_release(nd);
994return_err:
995 return err;
996}
997
998/*
999 * Wrapper to retry pathname resolution whenever the underlying
1000 * file system returns an ESTALE.
1001 *
1002 * Retry the whole path once, forcing real lookup requests
1003 * instead of relying on the dcache.
1004 */
1005int fastcall link_path_walk(const char *name, struct nameidata *nd)
1006{
1007 struct nameidata save = *nd;
1008 int result;
1009
1010 /* make sure the stuff we saved doesn't go away */
1011 dget(save.dentry);
1012 mntget(save.mnt);
1013
1014 result = __link_path_walk(name, nd);
1015 if (result == -ESTALE) {
1016 *nd = save;
1017 dget(nd->dentry);
1018 mntget(nd->mnt);
1019 nd->flags |= LOOKUP_REVAL;
1020 result = __link_path_walk(name, nd);
1021 }
1022
1023 dput(save.dentry);
1024 mntput(save.mnt);
1025
1026 return result;
1027}
1028
1029int fastcall path_walk(const char * name, struct nameidata *nd)
1030{
1031 current->total_link_count = 0;
1032 return link_path_walk(name, nd);
1033}
1034
Prasanna Medaea3834d2005-04-29 16:00:17 +01001035/*
1036 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1037 * everything is done. Returns 0 and drops input nd, if lookup failed;
1038 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1040{
1041 if (path_walk(name, nd))
1042 return 0; /* something went wrong... */
1043
1044 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1045 struct dentry *old_dentry = nd->dentry;
1046 struct vfsmount *old_mnt = nd->mnt;
1047 struct qstr last = nd->last;
1048 int last_type = nd->last_type;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001049 struct fs_struct *fs = current->fs;
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 /*
Andreas Mohre518ddb2006-09-29 02:01:22 -07001052 * NAME was not found in alternate root or it's a directory.
1053 * Try to find it in the normal root:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 */
1055 nd->last_type = LAST_ROOT;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001056 read_lock(&fs->lock);
1057 nd->mnt = mntget(fs->rootmnt);
1058 nd->dentry = dget(fs->root);
1059 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 if (path_walk(name, nd) == 0) {
1061 if (nd->dentry->d_inode) {
1062 dput(old_dentry);
1063 mntput(old_mnt);
1064 return 1;
1065 }
1066 path_release(nd);
1067 }
1068 nd->dentry = old_dentry;
1069 nd->mnt = old_mnt;
1070 nd->last = last;
1071 nd->last_type = last_type;
1072 }
1073 return 1;
1074}
1075
1076void set_fs_altroot(void)
1077{
1078 char *emul = __emul_prefix();
1079 struct nameidata nd;
1080 struct vfsmount *mnt = NULL, *oldmnt;
1081 struct dentry *dentry = NULL, *olddentry;
1082 int err;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001083 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
1085 if (!emul)
1086 goto set_it;
1087 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1088 if (!err) {
1089 mnt = nd.mnt;
1090 dentry = nd.dentry;
1091 }
1092set_it:
Andreas Mohre518ddb2006-09-29 02:01:22 -07001093 write_lock(&fs->lock);
1094 oldmnt = fs->altrootmnt;
1095 olddentry = fs->altroot;
1096 fs->altrootmnt = mnt;
1097 fs->altroot = dentry;
1098 write_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 if (olddentry) {
1100 dput(olddentry);
1101 mntput(oldmnt);
1102 }
1103}
1104
Prasanna Medaea3834d2005-04-29 16:00:17 +01001105/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001106static int fastcall do_path_lookup(int dfd, const char *name,
1107 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001109 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001110 int fput_needed;
1111 struct file *file;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001112 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1115 nd->flags = flags;
1116 nd->depth = 0;
1117
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 if (*name=='/') {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001119 read_lock(&fs->lock);
1120 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1121 nd->mnt = mntget(fs->altrootmnt);
1122 nd->dentry = dget(fs->altroot);
1123 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001125 goto out; /* found in altroot */
Andreas Mohre518ddb2006-09-29 02:01:22 -07001126 read_lock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
Andreas Mohre518ddb2006-09-29 02:01:22 -07001128 nd->mnt = mntget(fs->rootmnt);
1129 nd->dentry = dget(fs->root);
1130 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001131 } else if (dfd == AT_FDCWD) {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001132 read_lock(&fs->lock);
1133 nd->mnt = mntget(fs->pwdmnt);
1134 nd->dentry = dget(fs->pwd);
1135 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001136 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001137 struct dentry *dentry;
1138
1139 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001140 retval = -EBADF;
1141 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001142 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001143
1144 dentry = file->f_dentry;
1145
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001146 retval = -ENOTDIR;
1147 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001148 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001149
1150 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001151 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001152 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001153
1154 nd->mnt = mntget(file->f_vfsmnt);
1155 nd->dentry = dget(dentry);
1156
1157 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 current->total_link_count = 0;
1160 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001161out:
Suzuki3bc8414b2006-02-07 12:58:36 -08001162 if (likely(retval == 0)) {
Al Viro5ac3a9c2006-07-16 06:38:45 -04001163 if (unlikely(!audit_dummy_context() && nd && nd->dentry &&
Suzuki3bc8414b2006-02-07 12:58:36 -08001164 nd->dentry->d_inode))
Amy Griffis9c937dc2006-06-08 23:19:31 -04001165 audit_inode(name, nd->dentry->d_inode);
Suzuki3bc8414b2006-02-07 12:58:36 -08001166 }
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001167out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001168 return retval;
1169
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001170fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001171 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001172 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001175int fastcall path_lookup(const char *name, unsigned int flags,
1176 struct nameidata *nd)
1177{
1178 return do_path_lookup(AT_FDCWD, name, flags, nd);
1179}
1180
1181static int __path_lookup_intent_open(int dfd, const char *name,
1182 unsigned int lookup_flags, struct nameidata *nd,
1183 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001184{
1185 struct file *filp = get_empty_filp();
1186 int err;
1187
1188 if (filp == NULL)
1189 return -ENFILE;
1190 nd->intent.open.file = filp;
1191 nd->intent.open.flags = open_flags;
1192 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001193 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001194 if (IS_ERR(nd->intent.open.file)) {
1195 if (err == 0) {
1196 err = PTR_ERR(nd->intent.open.file);
1197 path_release(nd);
1198 }
1199 } else if (err != 0)
1200 release_open_intent(nd);
1201 return err;
1202}
1203
1204/**
1205 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001206 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001207 * @name: pointer to file name
1208 * @lookup_flags: lookup intent flags
1209 * @nd: pointer to nameidata
1210 * @open_flags: open intent flags
1211 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001212int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001213 struct nameidata *nd, int open_flags)
1214{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001215 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001216 open_flags, 0);
1217}
1218
1219/**
1220 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001221 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001222 * @name: pointer to file name
1223 * @lookup_flags: lookup intent flags
1224 * @nd: pointer to nameidata
1225 * @open_flags: open intent flags
1226 * @create_mode: create intent flags
1227 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001228static int path_lookup_create(int dfd, const char *name,
1229 unsigned int lookup_flags, struct nameidata *nd,
1230 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001231{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001232 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1233 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001234}
1235
1236int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1237 struct nameidata *nd, int open_flags)
1238{
1239 char *tmp = getname(name);
1240 int err = PTR_ERR(tmp);
1241
1242 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001243 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001244 putname(tmp);
1245 }
1246 return err;
1247}
1248
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249/*
1250 * Restricted form of lookup. Doesn't follow links, single-component only,
1251 * needs parent already locked. Doesn't follow mounts.
1252 * SMP-safe.
1253 */
1254static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1255{
1256 struct dentry * dentry;
1257 struct inode *inode;
1258 int err;
1259
1260 inode = base->d_inode;
1261 err = permission(inode, MAY_EXEC, nd);
1262 dentry = ERR_PTR(err);
1263 if (err)
1264 goto out;
1265
1266 /*
1267 * See if the low-level filesystem might want
1268 * to use its own hash..
1269 */
1270 if (base->d_op && base->d_op->d_hash) {
1271 err = base->d_op->d_hash(base, name);
1272 dentry = ERR_PTR(err);
1273 if (err < 0)
1274 goto out;
1275 }
1276
1277 dentry = cached_lookup(base, name, nd);
1278 if (!dentry) {
1279 struct dentry *new = d_alloc(base, name);
1280 dentry = ERR_PTR(-ENOMEM);
1281 if (!new)
1282 goto out;
1283 dentry = inode->i_op->lookup(inode, new, nd);
1284 if (!dentry)
1285 dentry = new;
1286 else
1287 dput(new);
1288 }
1289out:
1290 return dentry;
1291}
1292
Adrian Bunka244e162006-03-31 02:32:11 -08001293static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001295 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296}
1297
1298/* SMP-safe */
1299struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1300{
1301 unsigned long hash;
1302 struct qstr this;
1303 unsigned int c;
1304
1305 this.name = name;
1306 this.len = len;
1307 if (!len)
1308 goto access;
1309
1310 hash = init_name_hash();
1311 while (len--) {
1312 c = *(const unsigned char *)name++;
1313 if (c == '/' || c == '\0')
1314 goto access;
1315 hash = partial_name_hash(c, hash);
1316 }
1317 this.hash = end_name_hash(hash);
1318
Christoph Hellwig49705b72005-11-08 21:35:06 -08001319 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320access:
1321 return ERR_PTR(-EACCES);
1322}
1323
1324/*
1325 * namei()
1326 *
1327 * is used by most simple commands to get the inode of a specified name.
1328 * Open, link etc use their own routines, but this is enough for things
1329 * like 'chmod' etc.
1330 *
1331 * namei exists in two versions: namei/lnamei. The only difference is
1332 * that namei follows links, while lnamei does not.
1333 * SMP-safe
1334 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001335int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1336 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
1338 char *tmp = getname(name);
1339 int err = PTR_ERR(tmp);
1340
1341 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001342 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 putname(tmp);
1344 }
1345 return err;
1346}
1347
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001348int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1349{
1350 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1351}
1352
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353/*
1354 * It's inline, so penalty for filesystems that don't use sticky bit is
1355 * minimal.
1356 */
1357static inline int check_sticky(struct inode *dir, struct inode *inode)
1358{
1359 if (!(dir->i_mode & S_ISVTX))
1360 return 0;
1361 if (inode->i_uid == current->fsuid)
1362 return 0;
1363 if (dir->i_uid == current->fsuid)
1364 return 0;
1365 return !capable(CAP_FOWNER);
1366}
1367
1368/*
1369 * Check whether we can remove a link victim from directory dir, check
1370 * whether the type of victim is right.
1371 * 1. We can't do it if dir is read-only (done in permission())
1372 * 2. We should have write and exec permissions on dir
1373 * 3. We can't remove anything from append-only dir
1374 * 4. We can't do anything with immutable dir (done in permission())
1375 * 5. If the sticky bit on dir is set we should either
1376 * a. be owner of dir, or
1377 * b. be owner of victim, or
1378 * c. have CAP_FOWNER capability
1379 * 6. If the victim is append-only or immutable we can't do antyhing with
1380 * links pointing to it.
1381 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1382 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1383 * 9. We can't remove a root or mountpoint.
1384 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1385 * nfs_async_unlink().
1386 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001387static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
1389 int error;
1390
1391 if (!victim->d_inode)
1392 return -ENOENT;
1393
1394 BUG_ON(victim->d_parent->d_inode != dir);
Amy Griffis73d3ec52006-07-13 13:16:39 -04001395 audit_inode_child(victim->d_name.name, victim->d_inode, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396
1397 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1398 if (error)
1399 return error;
1400 if (IS_APPEND(dir))
1401 return -EPERM;
1402 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1403 IS_IMMUTABLE(victim->d_inode))
1404 return -EPERM;
1405 if (isdir) {
1406 if (!S_ISDIR(victim->d_inode->i_mode))
1407 return -ENOTDIR;
1408 if (IS_ROOT(victim))
1409 return -EBUSY;
1410 } else if (S_ISDIR(victim->d_inode->i_mode))
1411 return -EISDIR;
1412 if (IS_DEADDIR(dir))
1413 return -ENOENT;
1414 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1415 return -EBUSY;
1416 return 0;
1417}
1418
1419/* Check whether we can create an object with dentry child in directory
1420 * dir.
1421 * 1. We can't do it if child already exists (open has special treatment for
1422 * this case, but since we are inlined it's OK)
1423 * 2. We can't do it if dir is read-only (done in permission())
1424 * 3. We should have write and exec permissions on dir
1425 * 4. We can't do it if dir is immutable (done in permission())
1426 */
1427static inline int may_create(struct inode *dir, struct dentry *child,
1428 struct nameidata *nd)
1429{
1430 if (child->d_inode)
1431 return -EEXIST;
1432 if (IS_DEADDIR(dir))
1433 return -ENOENT;
1434 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1435}
1436
1437/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 * O_DIRECTORY translates into forcing a directory lookup.
1439 */
1440static inline int lookup_flags(unsigned int f)
1441{
1442 unsigned long retval = LOOKUP_FOLLOW;
1443
1444 if (f & O_NOFOLLOW)
1445 retval &= ~LOOKUP_FOLLOW;
1446
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 if (f & O_DIRECTORY)
1448 retval |= LOOKUP_DIRECTORY;
1449
1450 return retval;
1451}
1452
1453/*
1454 * p1 and p2 should be directories on the same fs.
1455 */
1456struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1457{
1458 struct dentry *p;
1459
1460 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001461 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462 return NULL;
1463 }
1464
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001465 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
1467 for (p = p1; p->d_parent != p; p = p->d_parent) {
1468 if (p->d_parent == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001469 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1470 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 return p;
1472 }
1473 }
1474
1475 for (p = p2; p->d_parent != p; p = p->d_parent) {
1476 if (p->d_parent == p1) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001477 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1478 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 return p;
1480 }
1481 }
1482
Ingo Molnarf2eace22006-07-03 00:25:05 -07001483 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1484 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 return NULL;
1486}
1487
1488void unlock_rename(struct dentry *p1, struct dentry *p2)
1489{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001490 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001492 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001493 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 }
1495}
1496
1497int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1498 struct nameidata *nd)
1499{
1500 int error = may_create(dir, dentry, nd);
1501
1502 if (error)
1503 return error;
1504
1505 if (!dir->i_op || !dir->i_op->create)
1506 return -EACCES; /* shouldn't it be ENOSYS? */
1507 mode &= S_IALLUGO;
1508 mode |= S_IFREG;
1509 error = security_inode_create(dir, dentry, mode);
1510 if (error)
1511 return error;
1512 DQUOT_INIT(dir);
1513 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001514 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001515 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 return error;
1517}
1518
1519int may_open(struct nameidata *nd, int acc_mode, int flag)
1520{
1521 struct dentry *dentry = nd->dentry;
1522 struct inode *inode = dentry->d_inode;
1523 int error;
1524
1525 if (!inode)
1526 return -ENOENT;
1527
1528 if (S_ISLNK(inode->i_mode))
1529 return -ELOOP;
1530
1531 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1532 return -EISDIR;
1533
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001534 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 if (error)
1536 return error;
1537
1538 /*
1539 * FIFO's, sockets and device files are special: they don't
1540 * actually live on the filesystem itself, and as such you
1541 * can write to them even if the filesystem is read-only.
1542 */
1543 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1544 flag &= ~O_TRUNC;
1545 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1546 if (nd->mnt->mnt_flags & MNT_NODEV)
1547 return -EACCES;
1548
1549 flag &= ~O_TRUNC;
1550 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1551 return -EROFS;
1552 /*
1553 * An append-only file must be opened in append mode for writing.
1554 */
1555 if (IS_APPEND(inode)) {
1556 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1557 return -EPERM;
1558 if (flag & O_TRUNC)
1559 return -EPERM;
1560 }
1561
1562 /* O_NOATIME can only be set by the owner or superuser */
1563 if (flag & O_NOATIME)
1564 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1565 return -EPERM;
1566
1567 /*
1568 * Ensure there are no outstanding leases on the file.
1569 */
1570 error = break_lease(inode, flag);
1571 if (error)
1572 return error;
1573
1574 if (flag & O_TRUNC) {
1575 error = get_write_access(inode);
1576 if (error)
1577 return error;
1578
1579 /*
1580 * Refuse to truncate files with mandatory locks held on them.
1581 */
1582 error = locks_verify_locked(inode);
1583 if (!error) {
1584 DQUOT_INIT(inode);
1585
NeilBrown4a301312006-01-08 01:02:39 -08001586 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 }
1588 put_write_access(inode);
1589 if (error)
1590 return error;
1591 } else
1592 if (flag & FMODE_WRITE)
1593 DQUOT_INIT(inode);
1594
1595 return 0;
1596}
1597
Dave Hansenaab520e2006-09-30 23:29:02 -07001598static int open_namei_create(struct nameidata *nd, struct path *path,
1599 int flag, int mode)
1600{
1601 int error;
1602 struct dentry *dir = nd->dentry;
1603
1604 if (!IS_POSIXACL(dir->d_inode))
1605 mode &= ~current->fs->umask;
1606 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1607 mutex_unlock(&dir->d_inode->i_mutex);
1608 dput(nd->dentry);
1609 nd->dentry = path->dentry;
1610 if (error)
1611 return error;
1612 /* Don't check for write permission, don't truncate */
1613 return may_open(nd, 0, flag & ~O_TRUNC);
1614}
1615
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616/*
1617 * open_namei()
1618 *
1619 * namei for open - this is in fact almost the whole open-routine.
1620 *
1621 * Note that the low bits of "flag" aren't the same as in the open
1622 * system call - they are 00 - no permissions needed
1623 * 01 - read permission needed
1624 * 10 - write permission needed
1625 * 11 - read/write permissions needed
1626 * which is a lot more logical, and also allows the "no perm" needed
1627 * for symlinks (where the permissions are checked later).
1628 * SMP-safe
1629 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001630int open_namei(int dfd, const char *pathname, int flag,
1631 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001633 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001634 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 struct dentry *dir;
1636 int count = 0;
1637
1638 acc_mode = ACC_MODE(flag);
1639
Trond Myklebust834f2a42005-10-18 14:20:16 -07001640 /* O_TRUNC implies we need access checks for write permissions */
1641 if (flag & O_TRUNC)
1642 acc_mode |= MAY_WRITE;
1643
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 /* Allow the LSM permission hook to distinguish append
1645 access from general write access. */
1646 if (flag & O_APPEND)
1647 acc_mode |= MAY_APPEND;
1648
Linus Torvalds1da177e2005-04-16 15:20:36 -07001649 /*
1650 * The simplest case - just a plain lookup.
1651 */
1652 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001653 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1654 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 if (error)
1656 return error;
1657 goto ok;
1658 }
1659
1660 /*
1661 * Create - we need to know the parent.
1662 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001663 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 if (error)
1665 return error;
1666
1667 /*
1668 * We have the parent and last component. First of all, check
1669 * that we are not asked to creat(2) an obvious directory - that
1670 * will not do.
1671 */
1672 error = -EISDIR;
1673 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1674 goto exit;
1675
1676 dir = nd->dentry;
1677 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001678 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001679 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001680 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681
1682do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001683 error = PTR_ERR(path.dentry);
1684 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001685 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 goto exit;
1687 }
1688
Oleg Drokin4af4c522006-03-25 03:06:54 -08001689 if (IS_ERR(nd->intent.open.file)) {
1690 mutex_unlock(&dir->d_inode->i_mutex);
1691 error = PTR_ERR(nd->intent.open.file);
1692 goto exit_dput;
1693 }
1694
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001696 if (!path.dentry->d_inode) {
Dave Hansenaab520e2006-09-30 23:29:02 -07001697 error = open_namei_create(nd, &path, flag, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 if (error)
1699 goto exit;
Dave Hansenaab520e2006-09-30 23:29:02 -07001700 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 }
1702
1703 /*
1704 * It already exists.
1705 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001706 mutex_unlock(&dir->d_inode->i_mutex);
Amy Griffis3e2efce2006-07-13 13:16:02 -04001707 audit_inode_update(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708
1709 error = -EEXIST;
1710 if (flag & O_EXCL)
1711 goto exit_dput;
1712
Al Viroe13b2102005-06-06 13:36:06 -07001713 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001715 if (flag & O_NOFOLLOW)
1716 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001718
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001720 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001722 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 goto do_link;
1724
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001725 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001727 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728 goto exit;
1729ok:
1730 error = may_open(nd, acc_mode, flag);
1731 if (error)
1732 goto exit;
1733 return 0;
1734
1735exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001736 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001738 if (!IS_ERR(nd->intent.open.file))
1739 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 path_release(nd);
1741 return error;
1742
1743do_link:
1744 error = -ELOOP;
1745 if (flag & O_NOFOLLOW)
1746 goto exit_dput;
1747 /*
1748 * This is subtle. Instead of calling do_follow_link() we do the
1749 * thing by hands. The reason is that this way we have zero link_count
1750 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1751 * After that we have the parent and last component, i.e.
1752 * we are in the same situation as after the first path_walk().
1753 * Well, almost - if the last component is normal we get its copy
1754 * stored in nd->last.name and we will have to putname() it when we
1755 * are done. Procfs-like symlinks just set LAST_BIND.
1756 */
1757 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001758 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (error)
1760 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001761 error = __do_follow_link(&path, nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001762 if (error) {
1763 /* Does someone understand code flow here? Or it is only
1764 * me so stupid? Anathema to whoever designed this non-sense
1765 * with "intent.open".
1766 */
1767 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 return error;
Kirill Korotaevde459212006-07-14 00:23:49 -07001769 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001771 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 error = -EISDIR;
1774 if (nd->last_type != LAST_NORM)
1775 goto exit;
1776 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001777 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 goto exit;
1779 }
1780 error = -ELOOP;
1781 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001782 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 goto exit;
1784 }
1785 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001786 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001787 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001788 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001789 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 goto do_last;
1791}
1792
1793/**
1794 * lookup_create - lookup a dentry, creating it if it doesn't exist
1795 * @nd: nameidata info
1796 * @is_dir: directory flag
1797 *
1798 * Simple function to lookup and return a dentry and create it
1799 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001800 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001801 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 */
1803struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1804{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001805 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
Ingo Molnarf2eace22006-07-03 00:25:05 -07001807 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001808 /*
1809 * Yucky last component or no last component at all?
1810 * (foo/., foo/.., /////)
1811 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 if (nd->last_type != LAST_NORM)
1813 goto fail;
1814 nd->flags &= ~LOOKUP_PARENT;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001815 nd->flags |= LOOKUP_CREATE;
1816 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001817
1818 /*
1819 * Do the final lookup.
1820 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001821 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 if (IS_ERR(dentry))
1823 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001824
1825 /*
1826 * Special case - lookup gave negative, but... we had foo/bar/
1827 * From the vfs_mknod() POV we just have a negative dentry -
1828 * all is fine. Let's be bastards - you had / on the end, you've
1829 * been asking for (non-existent) directory. -ENOENT for you.
1830 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1832 goto enoent;
1833 return dentry;
1834enoent:
1835 dput(dentry);
1836 dentry = ERR_PTR(-ENOENT);
1837fail:
1838 return dentry;
1839}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001840EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841
1842int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1843{
1844 int error = may_create(dir, dentry, NULL);
1845
1846 if (error)
1847 return error;
1848
1849 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1850 return -EPERM;
1851
1852 if (!dir->i_op || !dir->i_op->mknod)
1853 return -EPERM;
1854
1855 error = security_inode_mknod(dir, dentry, mode, dev);
1856 if (error)
1857 return error;
1858
1859 DQUOT_INIT(dir);
1860 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001861 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001862 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 return error;
1864}
1865
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001866asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1867 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868{
1869 int error = 0;
1870 char * tmp;
1871 struct dentry * dentry;
1872 struct nameidata nd;
1873
1874 if (S_ISDIR(mode))
1875 return -EPERM;
1876 tmp = getname(filename);
1877 if (IS_ERR(tmp))
1878 return PTR_ERR(tmp);
1879
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001880 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 if (error)
1882 goto out;
1883 dentry = lookup_create(&nd, 0);
1884 error = PTR_ERR(dentry);
1885
1886 if (!IS_POSIXACL(nd.dentry->d_inode))
1887 mode &= ~current->fs->umask;
1888 if (!IS_ERR(dentry)) {
1889 switch (mode & S_IFMT) {
1890 case 0: case S_IFREG:
1891 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1892 break;
1893 case S_IFCHR: case S_IFBLK:
1894 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1895 new_decode_dev(dev));
1896 break;
1897 case S_IFIFO: case S_IFSOCK:
1898 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1899 break;
1900 case S_IFDIR:
1901 error = -EPERM;
1902 break;
1903 default:
1904 error = -EINVAL;
1905 }
1906 dput(dentry);
1907 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001908 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 path_release(&nd);
1910out:
1911 putname(tmp);
1912
1913 return error;
1914}
1915
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001916asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1917{
1918 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1919}
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1922{
1923 int error = may_create(dir, dentry, NULL);
1924
1925 if (error)
1926 return error;
1927
1928 if (!dir->i_op || !dir->i_op->mkdir)
1929 return -EPERM;
1930
1931 mode &= (S_IRWXUGO|S_ISVTX);
1932 error = security_inode_mkdir(dir, dentry, mode);
1933 if (error)
1934 return error;
1935
1936 DQUOT_INIT(dir);
1937 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001938 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001939 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001940 return error;
1941}
1942
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001943asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944{
1945 int error = 0;
1946 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07001947 struct dentry *dentry;
1948 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949
1950 tmp = getname(pathname);
1951 error = PTR_ERR(tmp);
Dave Hansen6902d922006-09-30 23:29:01 -07001952 if (IS_ERR(tmp))
1953 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
Dave Hansen6902d922006-09-30 23:29:01 -07001955 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1956 if (error)
1957 goto out;
1958 dentry = lookup_create(&nd, 1);
1959 error = PTR_ERR(dentry);
1960 if (IS_ERR(dentry))
1961 goto out_unlock;
1962
1963 if (!IS_POSIXACL(nd.dentry->d_inode))
1964 mode &= ~current->fs->umask;
1965 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1966 dput(dentry);
1967out_unlock:
1968 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1969 path_release(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970out:
Dave Hansen6902d922006-09-30 23:29:01 -07001971 putname(tmp);
1972out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 return error;
1974}
1975
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001976asmlinkage long sys_mkdir(const char __user *pathname, int mode)
1977{
1978 return sys_mkdirat(AT_FDCWD, pathname, mode);
1979}
1980
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981/*
1982 * We try to drop the dentry early: we should have
1983 * a usage count of 2 if we're the only user of this
1984 * dentry, and if that is true (possibly after pruning
1985 * the dcache), then we drop the dentry now.
1986 *
1987 * A low-level filesystem can, if it choses, legally
1988 * do a
1989 *
1990 * if (!d_unhashed(dentry))
1991 * return -EBUSY;
1992 *
1993 * if it cannot handle the case of removing a directory
1994 * that is still in use by something else..
1995 */
1996void dentry_unhash(struct dentry *dentry)
1997{
1998 dget(dentry);
1999 if (atomic_read(&dentry->d_count))
2000 shrink_dcache_parent(dentry);
2001 spin_lock(&dcache_lock);
2002 spin_lock(&dentry->d_lock);
2003 if (atomic_read(&dentry->d_count) == 2)
2004 __d_drop(dentry);
2005 spin_unlock(&dentry->d_lock);
2006 spin_unlock(&dcache_lock);
2007}
2008
2009int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2010{
2011 int error = may_delete(dir, dentry, 1);
2012
2013 if (error)
2014 return error;
2015
2016 if (!dir->i_op || !dir->i_op->rmdir)
2017 return -EPERM;
2018
2019 DQUOT_INIT(dir);
2020
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002021 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 dentry_unhash(dentry);
2023 if (d_mountpoint(dentry))
2024 error = -EBUSY;
2025 else {
2026 error = security_inode_rmdir(dir, dentry);
2027 if (!error) {
2028 error = dir->i_op->rmdir(dir, dentry);
2029 if (!error)
2030 dentry->d_inode->i_flags |= S_DEAD;
2031 }
2032 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002033 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 d_delete(dentry);
2036 }
2037 dput(dentry);
2038
2039 return error;
2040}
2041
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002042static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043{
2044 int error = 0;
2045 char * name;
2046 struct dentry *dentry;
2047 struct nameidata nd;
2048
2049 name = getname(pathname);
2050 if(IS_ERR(name))
2051 return PTR_ERR(name);
2052
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002053 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 if (error)
2055 goto exit;
2056
2057 switch(nd.last_type) {
2058 case LAST_DOTDOT:
2059 error = -ENOTEMPTY;
2060 goto exit1;
2061 case LAST_DOT:
2062 error = -EINVAL;
2063 goto exit1;
2064 case LAST_ROOT:
2065 error = -EBUSY;
2066 goto exit1;
2067 }
Ingo Molnarf2eace22006-07-03 00:25:05 -07002068 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002069 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002071 if (IS_ERR(dentry))
2072 goto exit2;
2073 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2074 dput(dentry);
2075exit2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002076 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077exit1:
2078 path_release(&nd);
2079exit:
2080 putname(name);
2081 return error;
2082}
2083
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002084asmlinkage long sys_rmdir(const char __user *pathname)
2085{
2086 return do_rmdir(AT_FDCWD, pathname);
2087}
2088
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089int vfs_unlink(struct inode *dir, struct dentry *dentry)
2090{
2091 int error = may_delete(dir, dentry, 0);
2092
2093 if (error)
2094 return error;
2095
2096 if (!dir->i_op || !dir->i_op->unlink)
2097 return -EPERM;
2098
2099 DQUOT_INIT(dir);
2100
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002101 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 if (d_mountpoint(dentry))
2103 error = -EBUSY;
2104 else {
2105 error = security_inode_unlink(dir, dentry);
2106 if (!error)
2107 error = dir->i_op->unlink(dir, dentry);
2108 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002109 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
2111 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2112 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2113 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 }
Robert Love0eeca282005-07-12 17:06:03 -04002115
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 return error;
2117}
2118
2119/*
2120 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002121 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 * writeout happening, and we don't want to prevent access to the directory
2123 * while waiting on the I/O.
2124 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002125static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126{
2127 int error = 0;
2128 char * name;
2129 struct dentry *dentry;
2130 struct nameidata nd;
2131 struct inode *inode = NULL;
2132
2133 name = getname(pathname);
2134 if(IS_ERR(name))
2135 return PTR_ERR(name);
2136
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002137 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 if (error)
2139 goto exit;
2140 error = -EISDIR;
2141 if (nd.last_type != LAST_NORM)
2142 goto exit1;
Ingo Molnarf2eace22006-07-03 00:25:05 -07002143 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002144 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 error = PTR_ERR(dentry);
2146 if (!IS_ERR(dentry)) {
2147 /* Why not before? Because we want correct error value */
2148 if (nd.last.name[nd.last.len])
2149 goto slashes;
2150 inode = dentry->d_inode;
2151 if (inode)
2152 atomic_inc(&inode->i_count);
2153 error = vfs_unlink(nd.dentry->d_inode, dentry);
2154 exit2:
2155 dput(dentry);
2156 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002157 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 if (inode)
2159 iput(inode); /* truncate the inode here */
2160exit1:
2161 path_release(&nd);
2162exit:
2163 putname(name);
2164 return error;
2165
2166slashes:
2167 error = !dentry->d_inode ? -ENOENT :
2168 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2169 goto exit2;
2170}
2171
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002172asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2173{
2174 if ((flag & ~AT_REMOVEDIR) != 0)
2175 return -EINVAL;
2176
2177 if (flag & AT_REMOVEDIR)
2178 return do_rmdir(dfd, pathname);
2179
2180 return do_unlinkat(dfd, pathname);
2181}
2182
2183asmlinkage long sys_unlink(const char __user *pathname)
2184{
2185 return do_unlinkat(AT_FDCWD, pathname);
2186}
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2189{
2190 int error = may_create(dir, dentry, NULL);
2191
2192 if (error)
2193 return error;
2194
2195 if (!dir->i_op || !dir->i_op->symlink)
2196 return -EPERM;
2197
2198 error = security_inode_symlink(dir, dentry, oldname);
2199 if (error)
2200 return error;
2201
2202 DQUOT_INIT(dir);
2203 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002204 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002205 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 return error;
2207}
2208
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002209asmlinkage long sys_symlinkat(const char __user *oldname,
2210 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211{
2212 int error = 0;
2213 char * from;
2214 char * to;
Dave Hansen6902d922006-09-30 23:29:01 -07002215 struct dentry *dentry;
2216 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
2218 from = getname(oldname);
2219 if(IS_ERR(from))
2220 return PTR_ERR(from);
2221 to = getname(newname);
2222 error = PTR_ERR(to);
Dave Hansen6902d922006-09-30 23:29:01 -07002223 if (IS_ERR(to))
2224 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
Dave Hansen6902d922006-09-30 23:29:01 -07002226 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2227 if (error)
2228 goto out;
2229 dentry = lookup_create(&nd, 0);
2230 error = PTR_ERR(dentry);
2231 if (IS_ERR(dentry))
2232 goto out_unlock;
2233
2234 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2235 dput(dentry);
2236out_unlock:
2237 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2238 path_release(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239out:
Dave Hansen6902d922006-09-30 23:29:01 -07002240 putname(to);
2241out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 putname(from);
2243 return error;
2244}
2245
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002246asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2247{
2248 return sys_symlinkat(oldname, AT_FDCWD, newname);
2249}
2250
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2252{
2253 struct inode *inode = old_dentry->d_inode;
2254 int error;
2255
2256 if (!inode)
2257 return -ENOENT;
2258
2259 error = may_create(dir, new_dentry, NULL);
2260 if (error)
2261 return error;
2262
2263 if (dir->i_sb != inode->i_sb)
2264 return -EXDEV;
2265
2266 /*
2267 * A link to an append-only or immutable file cannot be created.
2268 */
2269 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2270 return -EPERM;
2271 if (!dir->i_op || !dir->i_op->link)
2272 return -EPERM;
2273 if (S_ISDIR(old_dentry->d_inode->i_mode))
2274 return -EPERM;
2275
2276 error = security_inode_link(old_dentry, dir, new_dentry);
2277 if (error)
2278 return error;
2279
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002280 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 DQUOT_INIT(dir);
2282 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002283 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002284 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002285 fsnotify_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 return error;
2287}
2288
2289/*
2290 * Hardlinks are often used in delicate situations. We avoid
2291 * security-related surprises by not following symlinks on the
2292 * newname. --KAB
2293 *
2294 * We don't follow them on the oldname either to be compatible
2295 * with linux 2.0, and to avoid hard-linking to directories
2296 * and other special files. --ADM
2297 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002298asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002299 int newdfd, const char __user *newname,
2300 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301{
2302 struct dentry *new_dentry;
2303 struct nameidata nd, old_nd;
2304 int error;
2305 char * to;
2306
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002307 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002308 return -EINVAL;
2309
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 to = getname(newname);
2311 if (IS_ERR(to))
2312 return PTR_ERR(to);
2313
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002314 error = __user_walk_fd(olddfd, oldname,
2315 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2316 &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 if (error)
2318 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002319 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 if (error)
2321 goto out;
2322 error = -EXDEV;
2323 if (old_nd.mnt != nd.mnt)
2324 goto out_release;
2325 new_dentry = lookup_create(&nd, 0);
2326 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002327 if (IS_ERR(new_dentry))
2328 goto out_unlock;
2329 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2330 dput(new_dentry);
2331out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002332 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333out_release:
2334 path_release(&nd);
2335out:
2336 path_release(&old_nd);
2337exit:
2338 putname(to);
2339
2340 return error;
2341}
2342
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002343asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2344{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002345 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002346}
2347
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348/*
2349 * The worst of all namespace operations - renaming directory. "Perverted"
2350 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2351 * Problems:
2352 * a) we can get into loop creation. Check is done in is_subdir().
2353 * b) race potential - two innocent renames can create a loop together.
2354 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002355 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 * story.
2357 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002358 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 * whether the target exists). Solution: try to be smart with locking
2360 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002361 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 * move will be locked. Thus we can rank directories by the tree
2363 * (ancestors first) and rank all non-directories after them.
2364 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002365 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 * HOWEVER, it relies on the assumption that any object with ->lookup()
2367 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2368 * we'd better make sure that there's no link(2) for them.
2369 * d) some filesystems don't support opened-but-unlinked directories,
2370 * either because of layout or because they are not ready to deal with
2371 * all cases correctly. The latter will be fixed (taking this sort of
2372 * stuff into VFS), but the former is not going away. Solution: the same
2373 * trick as in rmdir().
2374 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002375 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002377 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 * locking].
2379 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002380static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2381 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382{
2383 int error = 0;
2384 struct inode *target;
2385
2386 /*
2387 * If we are going to change the parent - check write permissions,
2388 * we'll need to flip '..'.
2389 */
2390 if (new_dir != old_dir) {
2391 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2392 if (error)
2393 return error;
2394 }
2395
2396 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2397 if (error)
2398 return error;
2399
2400 target = new_dentry->d_inode;
2401 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002402 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002403 dentry_unhash(new_dentry);
2404 }
2405 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2406 error = -EBUSY;
2407 else
2408 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2409 if (target) {
2410 if (!error)
2411 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002412 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 if (d_unhashed(new_dentry))
2414 d_rehash(new_dentry);
2415 dput(new_dentry);
2416 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002417 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002418 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2419 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002420 return error;
2421}
2422
Adrian Bunk75c96f82005-05-05 16:16:09 -07002423static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2424 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425{
2426 struct inode *target;
2427 int error;
2428
2429 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2430 if (error)
2431 return error;
2432
2433 dget(new_dentry);
2434 target = new_dentry->d_inode;
2435 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002436 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2438 error = -EBUSY;
2439 else
2440 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2441 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002442 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 }
2445 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002446 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447 dput(new_dentry);
2448 return error;
2449}
2450
2451int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2452 struct inode *new_dir, struct dentry *new_dentry)
2453{
2454 int error;
2455 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002456 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457
2458 if (old_dentry->d_inode == new_dentry->d_inode)
2459 return 0;
2460
2461 error = may_delete(old_dir, old_dentry, is_dir);
2462 if (error)
2463 return error;
2464
2465 if (!new_dentry->d_inode)
2466 error = may_create(new_dir, new_dentry, NULL);
2467 else
2468 error = may_delete(new_dir, new_dentry, is_dir);
2469 if (error)
2470 return error;
2471
2472 if (!old_dir->i_op || !old_dir->i_op->rename)
2473 return -EPERM;
2474
2475 DQUOT_INIT(old_dir);
2476 DQUOT_INIT(new_dir);
2477
Robert Love0eeca282005-07-12 17:06:03 -04002478 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2479
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 if (is_dir)
2481 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2482 else
2483 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2484 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002485 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002486 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2487 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 }
Robert Love0eeca282005-07-12 17:06:03 -04002489 fsnotify_oldname_free(old_name);
2490
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 return error;
2492}
2493
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002494static int do_rename(int olddfd, const char *oldname,
2495 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496{
2497 int error = 0;
2498 struct dentry * old_dir, * new_dir;
2499 struct dentry * old_dentry, *new_dentry;
2500 struct dentry * trap;
2501 struct nameidata oldnd, newnd;
2502
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002503 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 if (error)
2505 goto exit;
2506
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002507 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 if (error)
2509 goto exit1;
2510
2511 error = -EXDEV;
2512 if (oldnd.mnt != newnd.mnt)
2513 goto exit2;
2514
2515 old_dir = oldnd.dentry;
2516 error = -EBUSY;
2517 if (oldnd.last_type != LAST_NORM)
2518 goto exit2;
2519
2520 new_dir = newnd.dentry;
2521 if (newnd.last_type != LAST_NORM)
2522 goto exit2;
2523
2524 trap = lock_rename(new_dir, old_dir);
2525
Christoph Hellwig49705b72005-11-08 21:35:06 -08002526 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527 error = PTR_ERR(old_dentry);
2528 if (IS_ERR(old_dentry))
2529 goto exit3;
2530 /* source must exist */
2531 error = -ENOENT;
2532 if (!old_dentry->d_inode)
2533 goto exit4;
2534 /* unless the source is a directory trailing slashes give -ENOTDIR */
2535 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2536 error = -ENOTDIR;
2537 if (oldnd.last.name[oldnd.last.len])
2538 goto exit4;
2539 if (newnd.last.name[newnd.last.len])
2540 goto exit4;
2541 }
2542 /* source should not be ancestor of target */
2543 error = -EINVAL;
2544 if (old_dentry == trap)
2545 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002546 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 error = PTR_ERR(new_dentry);
2548 if (IS_ERR(new_dentry))
2549 goto exit4;
2550 /* target should not be an ancestor of source */
2551 error = -ENOTEMPTY;
2552 if (new_dentry == trap)
2553 goto exit5;
2554
2555 error = vfs_rename(old_dir->d_inode, old_dentry,
2556 new_dir->d_inode, new_dentry);
2557exit5:
2558 dput(new_dentry);
2559exit4:
2560 dput(old_dentry);
2561exit3:
2562 unlock_rename(new_dir, old_dir);
2563exit2:
2564 path_release(&newnd);
2565exit1:
2566 path_release(&oldnd);
2567exit:
2568 return error;
2569}
2570
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002571asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2572 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573{
2574 int error;
2575 char * from;
2576 char * to;
2577
2578 from = getname(oldname);
2579 if(IS_ERR(from))
2580 return PTR_ERR(from);
2581 to = getname(newname);
2582 error = PTR_ERR(to);
2583 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002584 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 putname(to);
2586 }
2587 putname(from);
2588 return error;
2589}
2590
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002591asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2592{
2593 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2594}
2595
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2597{
2598 int len;
2599
2600 len = PTR_ERR(link);
2601 if (IS_ERR(link))
2602 goto out;
2603
2604 len = strlen(link);
2605 if (len > (unsigned) buflen)
2606 len = buflen;
2607 if (copy_to_user(buffer, link, len))
2608 len = -EFAULT;
2609out:
2610 return len;
2611}
2612
2613/*
2614 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2615 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2616 * using) it for any given inode is up to filesystem.
2617 */
2618int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2619{
2620 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002621 void *cookie;
2622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002624 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2625 if (!IS_ERR(cookie)) {
2626 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002628 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2629 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002631 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632}
2633
2634int vfs_follow_link(struct nameidata *nd, const char *link)
2635{
2636 return __vfs_follow_link(nd, link);
2637}
2638
2639/* get the link contents into pagecache */
2640static char *page_getlink(struct dentry * dentry, struct page **ppage)
2641{
2642 struct page * page;
2643 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002644 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 if (IS_ERR(page))
2646 goto sync_fail;
2647 wait_on_page_locked(page);
2648 if (!PageUptodate(page))
2649 goto async_fail;
2650 *ppage = page;
2651 return kmap(page);
2652
2653async_fail:
2654 page_cache_release(page);
2655 return ERR_PTR(-EIO);
2656
2657sync_fail:
2658 return (char*)page;
2659}
2660
2661int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2662{
2663 struct page *page = NULL;
2664 char *s = page_getlink(dentry, &page);
2665 int res = vfs_readlink(dentry,buffer,buflen,s);
2666 if (page) {
2667 kunmap(page);
2668 page_cache_release(page);
2669 }
2670 return res;
2671}
2672
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002673void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002675 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002677 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678}
2679
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002680void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002682 struct page *page = cookie;
2683
2684 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 kunmap(page);
2686 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 }
2688}
2689
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002690int __page_symlink(struct inode *inode, const char *symname, int len,
2691 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692{
2693 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002694 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 int err = -ENOMEM;
2696 char *kaddr;
2697
NeilBrown7e53cac2006-03-25 03:07:57 -08002698retry:
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002699 page = find_or_create_page(mapping, 0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 if (!page)
2701 goto fail;
2702 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
NeilBrown7e53cac2006-03-25 03:07:57 -08002703 if (err == AOP_TRUNCATED_PAGE) {
2704 page_cache_release(page);
2705 goto retry;
2706 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707 if (err)
2708 goto fail_map;
2709 kaddr = kmap_atomic(page, KM_USER0);
2710 memcpy(kaddr, symname, len-1);
2711 kunmap_atomic(kaddr, KM_USER0);
NeilBrown7e53cac2006-03-25 03:07:57 -08002712 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2713 if (err == AOP_TRUNCATED_PAGE) {
2714 page_cache_release(page);
2715 goto retry;
2716 }
2717 if (err)
2718 goto fail_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 /*
2720 * Notice that we are _not_ going to block here - end of page is
2721 * unmapped, so this will only try to map the rest of page, see
2722 * that it is unmapped (typically even will not look into inode -
2723 * ->i_size will be enough for everything) and zero it out.
2724 * OTOH it's obviously correct and should make the page up-to-date.
2725 */
2726 if (!PageUptodate(page)) {
2727 err = mapping->a_ops->readpage(NULL, page);
NeilBrown7e53cac2006-03-25 03:07:57 -08002728 if (err != AOP_TRUNCATED_PAGE)
2729 wait_on_page_locked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 } else {
2731 unlock_page(page);
2732 }
2733 page_cache_release(page);
2734 if (err < 0)
2735 goto fail;
2736 mark_inode_dirty(inode);
2737 return 0;
2738fail_map:
2739 unlock_page(page);
2740 page_cache_release(page);
2741fail:
2742 return err;
2743}
2744
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002745int page_symlink(struct inode *inode, const char *symname, int len)
2746{
2747 return __page_symlink(inode, symname, len,
2748 mapping_gfp_mask(inode->i_mapping));
2749}
2750
Linus Torvalds1da177e2005-04-16 15:20:36 -07002751struct inode_operations page_symlink_inode_operations = {
2752 .readlink = generic_readlink,
2753 .follow_link = page_follow_link_light,
2754 .put_link = page_put_link,
2755};
2756
2757EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002758EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759EXPORT_SYMBOL(follow_down);
2760EXPORT_SYMBOL(follow_up);
2761EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2762EXPORT_SYMBOL(getname);
2763EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764EXPORT_SYMBOL(lookup_one_len);
2765EXPORT_SYMBOL(page_follow_link_light);
2766EXPORT_SYMBOL(page_put_link);
2767EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002768EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002769EXPORT_SYMBOL(page_symlink);
2770EXPORT_SYMBOL(page_symlink_inode_operations);
2771EXPORT_SYMBOL(path_lookup);
2772EXPORT_SYMBOL(path_release);
2773EXPORT_SYMBOL(path_walk);
2774EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002775EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002776EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777EXPORT_SYMBOL(unlock_rename);
2778EXPORT_SYMBOL(vfs_create);
2779EXPORT_SYMBOL(vfs_follow_link);
2780EXPORT_SYMBOL(vfs_link);
2781EXPORT_SYMBOL(vfs_mkdir);
2782EXPORT_SYMBOL(vfs_mknod);
2783EXPORT_SYMBOL(generic_permission);
2784EXPORT_SYMBOL(vfs_readlink);
2785EXPORT_SYMBOL(vfs_rename);
2786EXPORT_SYMBOL(vfs_rmdir);
2787EXPORT_SYMBOL(vfs_symlink);
2788EXPORT_SYMBOL(vfs_unlink);
2789EXPORT_SYMBOL(dentry_unhash);
2790EXPORT_SYMBOL(generic_readlink);