blob: 6dbbd42d8b95fb933ed7a4f3e0183baff73fe3fb [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>
Trond Myklebust834f2a42005-10-18 14:20:16 -070031#include <linux/file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <asm/namei.h>
33#include <asm/uaccess.h>
34
35#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
36
37/* [Feb-1997 T. Schoebel-Theuer]
38 * Fundamental changes in the pathname lookup mechanisms (namei)
39 * were necessary because of omirr. The reason is that omirr needs
40 * to know the _real_ pathname, not the user-supplied one, in case
41 * of symlinks (and also when transname replacements occur).
42 *
43 * The new code replaces the old recursive symlink resolution with
44 * an iterative one (in case of non-nested symlink chains). It does
45 * this with calls to <fs>_follow_link().
46 * As a side effect, dir_namei(), _namei() and follow_link() are now
47 * replaced with a single function lookup_dentry() that can handle all
48 * the special cases of the former code.
49 *
50 * With the new dcache, the pathname is stored at each inode, at least as
51 * long as the refcount of the inode is positive. As a side effect, the
52 * size of the dcache depends on the inode cache and thus is dynamic.
53 *
54 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
55 * resolution to correspond with current state of the code.
56 *
57 * Note that the symlink resolution is not *completely* iterative.
58 * There is still a significant amount of tail- and mid- recursion in
59 * the algorithm. Also, note that <fs>_readlink() is not used in
60 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
61 * may return different results than <fs>_follow_link(). Many virtual
62 * filesystems (including /proc) exhibit this behavior.
63 */
64
65/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
66 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
67 * and the name already exists in form of a symlink, try to create the new
68 * name indicated by the symlink. The old code always complained that the
69 * name already exists, due to not following the symlink even if its target
70 * is nonexistent. The new semantics affects also mknod() and link() when
71 * the name is a symlink pointing to a non-existant name.
72 *
73 * I don't know which semantics is the right one, since I have no access
74 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
75 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
76 * "old" one. Personally, I think the new semantics is much more logical.
77 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
78 * file does succeed in both HP-UX and SunOs, but not in Solaris
79 * and in the old Linux semantics.
80 */
81
82/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
83 * semantics. See the comments in "open_namei" and "do_link" below.
84 *
85 * [10-Sep-98 Alan Modra] Another symlink change.
86 */
87
88/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
89 * inside the path - always follow.
90 * in the last component in creation/removal/renaming - never follow.
91 * if LOOKUP_FOLLOW passed - follow.
92 * if the pathname has trailing slashes - follow.
93 * otherwise - don't follow.
94 * (applied in that order).
95 *
96 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
97 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
98 * During the 2.4 we need to fix the userland stuff depending on it -
99 * hopefully we will be able to get rid of that wart in 2.5. So far only
100 * XEmacs seems to be relying on it...
101 */
102/*
103 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
104 * implemented. Let's see if raised priority of ->s_vfs_rename_sem gives
105 * any extra contention...
106 */
107
108/* In order to reduce some races, while at the same time doing additional
109 * checking and hopefully speeding things up, we copy filenames to the
110 * kernel data space before using them..
111 *
112 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
113 * PATH_MAX includes the nul terminator --RR.
114 */
115static inline int do_getname(const char __user *filename, char *page)
116{
117 int retval;
118 unsigned long len = PATH_MAX;
119
120 if (!segment_eq(get_fs(), KERNEL_DS)) {
121 if ((unsigned long) filename >= TASK_SIZE)
122 return -EFAULT;
123 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
124 len = TASK_SIZE - (unsigned long) filename;
125 }
126
127 retval = strncpy_from_user(page, filename, len);
128 if (retval > 0) {
129 if (retval < len)
130 return 0;
131 return -ENAMETOOLONG;
132 } else if (!retval)
133 retval = -ENOENT;
134 return retval;
135}
136
137char * getname(const char __user * filename)
138{
139 char *tmp, *result;
140
141 result = ERR_PTR(-ENOMEM);
142 tmp = __getname();
143 if (tmp) {
144 int retval = do_getname(filename, tmp);
145
146 result = tmp;
147 if (retval < 0) {
148 __putname(tmp);
149 result = ERR_PTR(retval);
150 }
151 }
152 audit_getname(result);
153 return result;
154}
155
156#ifdef CONFIG_AUDITSYSCALL
157void putname(const char *name)
158{
159 if (unlikely(current->audit_context))
160 audit_putname(name);
161 else
162 __putname(name);
163}
164EXPORT_SYMBOL(putname);
165#endif
166
167
168/**
169 * generic_permission - check for access rights on a Posix-like filesystem
170 * @inode: inode to check access rights for
171 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
172 * @check_acl: optional callback to check for Posix ACLs
173 *
174 * Used to check for read/write/execute permissions on a file.
175 * We use "fsuid" for this, letting us set arbitrary permissions
176 * for filesystem access without changing the "normal" uids which
177 * are used for other things..
178 */
179int generic_permission(struct inode *inode, int mask,
180 int (*check_acl)(struct inode *inode, int mask))
181{
182 umode_t mode = inode->i_mode;
183
184 if (current->fsuid == inode->i_uid)
185 mode >>= 6;
186 else {
187 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
188 int error = check_acl(inode, mask);
189 if (error == -EACCES)
190 goto check_capabilities;
191 else if (error != -EAGAIN)
192 return error;
193 }
194
195 if (in_group_p(inode->i_gid))
196 mode >>= 3;
197 }
198
199 /*
200 * If the DACs are ok we don't need any capability check.
201 */
202 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
203 return 0;
204
205 check_capabilities:
206 /*
207 * Read/write DACs are always overridable.
208 * Executable DACs are overridable if at least one exec bit is set.
209 */
210 if (!(mask & MAY_EXEC) ||
211 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
212 if (capable(CAP_DAC_OVERRIDE))
213 return 0;
214
215 /*
216 * Searching includes executable on directories, else just read.
217 */
218 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
219 if (capable(CAP_DAC_READ_SEARCH))
220 return 0;
221
222 return -EACCES;
223}
224
225int permission(struct inode *inode, int mask, struct nameidata *nd)
226{
227 int retval, submask;
228
229 if (mask & MAY_WRITE) {
230 umode_t mode = inode->i_mode;
231
232 /*
233 * Nobody gets write access to a read-only fs.
234 */
235 if (IS_RDONLY(inode) &&
236 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
237 return -EROFS;
238
239 /*
240 * Nobody gets write access to an immutable file.
241 */
242 if (IS_IMMUTABLE(inode))
243 return -EACCES;
244 }
245
246
247 /* Ordinary permission routines do not understand MAY_APPEND. */
248 submask = mask & ~MAY_APPEND;
249 if (inode->i_op && inode->i_op->permission)
250 retval = inode->i_op->permission(inode, submask, nd);
251 else
252 retval = generic_permission(inode, submask, NULL);
253 if (retval)
254 return retval;
255
256 return security_inode_permission(inode, mask, nd);
257}
258
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800259/**
260 * vfs_permission - check for access rights to a given path
261 * @nd: lookup result that describes the path
262 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
263 *
264 * Used to check for read/write/execute permissions on a path.
265 * We use "fsuid" for this, letting us set arbitrary permissions
266 * for filesystem access without changing the "normal" uids which
267 * are used for other things.
268 */
269int vfs_permission(struct nameidata *nd, int mask)
270{
271 return permission(nd->dentry->d_inode, mask, nd);
272}
273
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800274/**
275 * file_permission - check for additional access rights to a given file
276 * @file: file to check access rights for
277 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
278 *
279 * Used to check for read/write/execute permissions on an already opened
280 * file.
281 *
282 * Note:
283 * Do not use this function in new code. All access checks should
284 * be done using vfs_permission().
285 */
286int file_permission(struct file *file, int mask)
287{
288 return permission(file->f_dentry->d_inode, mask, NULL);
289}
290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291/*
292 * get_write_access() gets write permission for a file.
293 * put_write_access() releases this write permission.
294 * This is used for regular files.
295 * We cannot support write (and maybe mmap read-write shared) accesses and
296 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
297 * can have the following values:
298 * 0: no writers, no VM_DENYWRITE mappings
299 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
300 * > 0: (i_writecount) users are writing to the file.
301 *
302 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
303 * except for the cases where we don't hold i_writecount yet. Then we need to
304 * use {get,deny}_write_access() - these functions check the sign and refuse
305 * to do the change if sign is wrong. Exclusion between them is provided by
306 * the inode->i_lock spinlock.
307 */
308
309int get_write_access(struct inode * inode)
310{
311 spin_lock(&inode->i_lock);
312 if (atomic_read(&inode->i_writecount) < 0) {
313 spin_unlock(&inode->i_lock);
314 return -ETXTBSY;
315 }
316 atomic_inc(&inode->i_writecount);
317 spin_unlock(&inode->i_lock);
318
319 return 0;
320}
321
322int deny_write_access(struct file * file)
323{
324 struct inode *inode = file->f_dentry->d_inode;
325
326 spin_lock(&inode->i_lock);
327 if (atomic_read(&inode->i_writecount) > 0) {
328 spin_unlock(&inode->i_lock);
329 return -ETXTBSY;
330 }
331 atomic_dec(&inode->i_writecount);
332 spin_unlock(&inode->i_lock);
333
334 return 0;
335}
336
337void path_release(struct nameidata *nd)
338{
339 dput(nd->dentry);
340 mntput(nd->mnt);
341}
342
343/*
344 * umount() mustn't call path_release()/mntput() as that would clear
345 * mnt_expiry_mark
346 */
347void path_release_on_umount(struct nameidata *nd)
348{
349 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700350 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351}
352
Trond Myklebust834f2a42005-10-18 14:20:16 -0700353/**
354 * release_open_intent - free up open intent resources
355 * @nd: pointer to nameidata
356 */
357void release_open_intent(struct nameidata *nd)
358{
359 if (nd->intent.open.file->f_dentry == NULL)
360 put_filp(nd->intent.open.file);
361 else
362 fput(nd->intent.open.file);
363}
364
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365/*
366 * Internal lookup() using the new generic dcache.
367 * SMP-safe
368 */
369static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
370{
371 struct dentry * dentry = __d_lookup(parent, name);
372
373 /* lockess __d_lookup may fail due to concurrent d_move()
374 * in some unrelated directory, so try with d_lookup
375 */
376 if (!dentry)
377 dentry = d_lookup(parent, name);
378
379 if (dentry && dentry->d_op && dentry->d_op->d_revalidate) {
380 if (!dentry->d_op->d_revalidate(dentry, nd) && !d_invalidate(dentry)) {
381 dput(dentry);
382 dentry = NULL;
383 }
384 }
385 return dentry;
386}
387
388/*
389 * Short-cut version of permission(), for calling by
390 * path_walk(), when dcache lock is held. Combines parts
391 * of permission() and generic_permission(), and tests ONLY for
392 * MAY_EXEC permission.
393 *
394 * If appropriate, check DAC only. If not appropriate, or
395 * short-cut DAC fails, then call permission() to do more
396 * complete permission check.
397 */
398static inline int exec_permission_lite(struct inode *inode,
399 struct nameidata *nd)
400{
401 umode_t mode = inode->i_mode;
402
403 if (inode->i_op && inode->i_op->permission)
404 return -EAGAIN;
405
406 if (current->fsuid == inode->i_uid)
407 mode >>= 6;
408 else if (in_group_p(inode->i_gid))
409 mode >>= 3;
410
411 if (mode & MAY_EXEC)
412 goto ok;
413
414 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
415 goto ok;
416
417 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
418 goto ok;
419
420 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
421 goto ok;
422
423 return -EACCES;
424ok:
425 return security_inode_permission(inode, MAY_EXEC, nd);
426}
427
428/*
429 * This is called when everything else fails, and we actually have
430 * to go to the low-level filesystem to find out what we should do..
431 *
432 * We get the directory semaphore, and after getting that we also
433 * make sure that nobody added the entry to the dcache in the meantime..
434 * SMP-safe
435 */
436static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
437{
438 struct dentry * result;
439 struct inode *dir = parent->d_inode;
440
441 down(&dir->i_sem);
442 /*
443 * First re-do the cached lookup just in case it was created
444 * while we waited for the directory semaphore..
445 *
446 * FIXME! This could use version numbering or similar to
447 * avoid unnecessary cache lookups.
448 *
449 * The "dcache_lock" is purely to protect the RCU list walker
450 * from concurrent renames at this point (we mustn't get false
451 * negatives from the RCU list walk here, unlike the optimistic
452 * fast walk).
453 *
454 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
455 */
456 result = d_lookup(parent, name);
457 if (!result) {
458 struct dentry * dentry = d_alloc(parent, name);
459 result = ERR_PTR(-ENOMEM);
460 if (dentry) {
461 result = dir->i_op->lookup(dir, dentry, nd);
462 if (result)
463 dput(dentry);
464 else
465 result = dentry;
466 }
467 up(&dir->i_sem);
468 return result;
469 }
470
471 /*
472 * Uhhuh! Nasty case: the cache was re-populated while
473 * we waited on the semaphore. Need to revalidate.
474 */
475 up(&dir->i_sem);
476 if (result->d_op && result->d_op->d_revalidate) {
477 if (!result->d_op->d_revalidate(result, nd) && !d_invalidate(result)) {
478 dput(result);
479 result = ERR_PTR(-ENOENT);
480 }
481 }
482 return result;
483}
484
485static int __emul_lookup_dentry(const char *, struct nameidata *);
486
487/* SMP-safe */
488static inline int
489walk_init_root(const char *name, struct nameidata *nd)
490{
491 read_lock(&current->fs->lock);
492 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
493 nd->mnt = mntget(current->fs->altrootmnt);
494 nd->dentry = dget(current->fs->altroot);
495 read_unlock(&current->fs->lock);
496 if (__emul_lookup_dentry(name,nd))
497 return 0;
498 read_lock(&current->fs->lock);
499 }
500 nd->mnt = mntget(current->fs->rootmnt);
501 nd->dentry = dget(current->fs->root);
502 read_unlock(&current->fs->lock);
503 return 1;
504}
505
506static inline int __vfs_follow_link(struct nameidata *nd, const char *link)
507{
508 int res = 0;
509 char *name;
510 if (IS_ERR(link))
511 goto fail;
512
513 if (*link == '/') {
514 path_release(nd);
515 if (!walk_init_root(link, nd))
516 /* weird __emul_prefix() stuff did it */
517 goto out;
518 }
519 res = link_path_walk(link, nd);
520out:
521 if (nd->depth || res || nd->last_type!=LAST_NORM)
522 return res;
523 /*
524 * If it is an iterative symlinks resolution in open_namei() we
525 * have to copy the last component. And all that crap because of
526 * bloody create() on broken symlinks. Furrfu...
527 */
528 name = __getname();
529 if (unlikely(!name)) {
530 path_release(nd);
531 return -ENOMEM;
532 }
533 strcpy(name, nd->last.name);
534 nd->last.name = name;
535 return 0;
536fail:
537 path_release(nd);
538 return PTR_ERR(link);
539}
540
Al Viro90ebe562005-06-06 13:35:58 -0700541struct path {
542 struct vfsmount *mnt;
543 struct dentry *dentry;
544};
545
Al Virocd4e91d2005-06-06 13:36:03 -0700546static inline int __do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547{
548 int error;
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700549 void *cookie;
Al Virocd4e91d2005-06-06 13:36:03 -0700550 struct dentry *dentry = path->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
Al Virod671a1c2005-06-06 13:36:14 -0700552 touch_atime(path->mnt, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 nd_set_link(nd, NULL);
Al Virocd4e91d2005-06-06 13:36:03 -0700554
Al Viro39ca6d42005-06-06 13:36:12 -0700555 if (path->mnt == nd->mnt)
556 mntget(path->mnt);
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700557 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
558 error = PTR_ERR(cookie);
559 if (!IS_ERR(cookie)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 char *s = nd_get_link(nd);
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700561 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 if (s)
563 error = __vfs_follow_link(nd, s);
564 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -0700565 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 }
Al Virocd4e91d2005-06-06 13:36:03 -0700567 dput(dentry);
568 mntput(path->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
570 return error;
571}
572
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700573static inline void dput_path(struct path *path, struct nameidata *nd)
574{
575 dput(path->dentry);
576 if (path->mnt != nd->mnt)
577 mntput(path->mnt);
578}
579
580static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
581{
582 dput(nd->dentry);
583 if (nd->mnt != path->mnt)
584 mntput(nd->mnt);
585 nd->mnt = path->mnt;
586 nd->dentry = path->dentry;
587}
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589/*
590 * This limits recursive symlink follows to 8, while
591 * limiting consecutive symlinks to 40.
592 *
593 * Without that kind of total limit, nasty chains of consecutive
594 * symlinks can cause almost arbitrarily long lookups.
595 */
Al Viro90ebe562005-06-06 13:35:58 -0700596static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
598 int err = -ELOOP;
599 if (current->link_count >= MAX_NESTED_LINKS)
600 goto loop;
601 if (current->total_link_count >= 40)
602 goto loop;
603 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
604 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700605 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 if (err)
607 goto loop;
608 current->link_count++;
609 current->total_link_count++;
610 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700611 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 current->link_count--;
613 nd->depth--;
614 return err;
615loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700616 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700617 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 return err;
619}
620
621int follow_up(struct vfsmount **mnt, struct dentry **dentry)
622{
623 struct vfsmount *parent;
624 struct dentry *mountpoint;
625 spin_lock(&vfsmount_lock);
626 parent=(*mnt)->mnt_parent;
627 if (parent == *mnt) {
628 spin_unlock(&vfsmount_lock);
629 return 0;
630 }
631 mntget(parent);
632 mountpoint=dget((*mnt)->mnt_mountpoint);
633 spin_unlock(&vfsmount_lock);
634 dput(*dentry);
635 *dentry = mountpoint;
636 mntput(*mnt);
637 *mnt = parent;
638 return 1;
639}
640
641/* no need for dcache_lock, as serialization is taken care in
642 * namespace.c
643 */
Al Viro463ffb22005-06-06 13:36:05 -0700644static int __follow_mount(struct path *path)
645{
646 int res = 0;
647 while (d_mountpoint(path->dentry)) {
648 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
649 if (!mounted)
650 break;
651 dput(path->dentry);
652 if (res)
653 mntput(path->mnt);
654 path->mnt = mounted;
655 path->dentry = dget(mounted->mnt_root);
656 res = 1;
657 }
658 return res;
659}
660
Al Viro58c465e2005-06-06 13:36:13 -0700661static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 while (d_mountpoint(*dentry)) {
664 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
665 if (!mounted)
666 break;
Al Viro58c465e2005-06-06 13:36:13 -0700667 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 mntput(*mnt);
669 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
673
674/* no need for dcache_lock, as serialization is taken care in
675 * namespace.c
676 */
Al Viroe13b2102005-06-06 13:36:06 -0700677int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678{
679 struct vfsmount *mounted;
680
681 mounted = lookup_mnt(*mnt, *dentry);
682 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700683 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 mntput(*mnt);
685 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 *dentry = dget(mounted->mnt_root);
687 return 1;
688 }
689 return 0;
690}
691
Al Viro58c465e2005-06-06 13:36:13 -0700692static inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
694 while(1) {
695 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700696 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 read_lock(&current->fs->lock);
Al Viro58c465e2005-06-06 13:36:13 -0700699 if (nd->dentry == current->fs->root &&
700 nd->mnt == current->fs->rootmnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 read_unlock(&current->fs->lock);
702 break;
703 }
704 read_unlock(&current->fs->lock);
705 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700706 if (nd->dentry != nd->mnt->mnt_root) {
707 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 spin_unlock(&dcache_lock);
709 dput(old);
710 break;
711 }
712 spin_unlock(&dcache_lock);
713 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700714 parent = nd->mnt->mnt_parent;
715 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 spin_unlock(&vfsmount_lock);
717 break;
718 }
719 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700720 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 spin_unlock(&vfsmount_lock);
722 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700723 mntput(nd->mnt);
724 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 }
Al Viro58c465e2005-06-06 13:36:13 -0700726 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727}
728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729/*
730 * It's more convoluted than I'd like it to be, but... it's still fairly
731 * small and for now I'd prefer to have fast path as straight as possible.
732 * It _is_ time-critical.
733 */
734static int do_lookup(struct nameidata *nd, struct qstr *name,
735 struct path *path)
736{
737 struct vfsmount *mnt = nd->mnt;
738 struct dentry *dentry = __d_lookup(nd->dentry, name);
739
740 if (!dentry)
741 goto need_lookup;
742 if (dentry->d_op && dentry->d_op->d_revalidate)
743 goto need_revalidate;
744done:
745 path->mnt = mnt;
746 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700747 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 return 0;
749
750need_lookup:
751 dentry = real_lookup(nd->dentry, name, nd);
752 if (IS_ERR(dentry))
753 goto fail;
754 goto done;
755
756need_revalidate:
757 if (dentry->d_op->d_revalidate(dentry, nd))
758 goto done;
759 if (d_invalidate(dentry))
760 goto done;
761 dput(dentry);
762 goto need_lookup;
763
764fail:
765 return PTR_ERR(dentry);
766}
767
768/*
769 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100770 * This is the basic name resolution function, turning a pathname into
771 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100773 * Returns 0 and nd will have valid dentry and mnt on success.
774 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 */
776static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
777{
778 struct path next;
779 struct inode *inode;
780 int err;
781 unsigned int lookup_flags = nd->flags;
782
783 while (*name=='/')
784 name++;
785 if (!*name)
786 goto return_reval;
787
788 inode = nd->dentry->d_inode;
789 if (nd->depth)
790 lookup_flags = LOOKUP_FOLLOW;
791
792 /* At this point we know we have a real path component. */
793 for(;;) {
794 unsigned long hash;
795 struct qstr this;
796 unsigned int c;
797
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700798 nd->flags |= LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 err = exec_permission_lite(inode, nd);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800800 if (err == -EAGAIN)
801 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 if (err)
803 break;
804
805 this.name = name;
806 c = *(const unsigned char *)name;
807
808 hash = init_name_hash();
809 do {
810 name++;
811 hash = partial_name_hash(c, hash);
812 c = *(const unsigned char *)name;
813 } while (c && (c != '/'));
814 this.len = name - (const char *) this.name;
815 this.hash = end_name_hash(hash);
816
817 /* remove trailing slashes? */
818 if (!c)
819 goto last_component;
820 while (*++name == '/');
821 if (!*name)
822 goto last_with_slashes;
823
824 /*
825 * "." and ".." are special - ".." especially so because it has
826 * to be able to know about the current root directory and
827 * parent relationships.
828 */
829 if (this.name[0] == '.') switch (this.len) {
830 default:
831 break;
832 case 2:
833 if (this.name[1] != '.')
834 break;
Al Viro58c465e2005-06-06 13:36:13 -0700835 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 inode = nd->dentry->d_inode;
837 /* fallthrough */
838 case 1:
839 continue;
840 }
841 /*
842 * See if the low-level filesystem might want
843 * to use its own hash..
844 */
845 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
846 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
847 if (err < 0)
848 break;
849 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 /* This does the actual lookups.. */
851 err = do_lookup(nd, &this, &next);
852 if (err)
853 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
855 err = -ENOENT;
856 inode = next.dentry->d_inode;
857 if (!inode)
858 goto out_dput;
859 err = -ENOTDIR;
860 if (!inode->i_op)
861 goto out_dput;
862
863 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700864 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 if (err)
866 goto return_err;
867 err = -ENOENT;
868 inode = nd->dentry->d_inode;
869 if (!inode)
870 break;
871 err = -ENOTDIR;
872 if (!inode->i_op)
873 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700874 } else
875 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 err = -ENOTDIR;
877 if (!inode->i_op->lookup)
878 break;
879 continue;
880 /* here ends the main loop */
881
882last_with_slashes:
883 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
884last_component:
885 nd->flags &= ~LOOKUP_CONTINUE;
886 if (lookup_flags & LOOKUP_PARENT)
887 goto lookup_parent;
888 if (this.name[0] == '.') switch (this.len) {
889 default:
890 break;
891 case 2:
892 if (this.name[1] != '.')
893 break;
Al Viro58c465e2005-06-06 13:36:13 -0700894 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 inode = nd->dentry->d_inode;
896 /* fallthrough */
897 case 1:
898 goto return_reval;
899 }
900 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
901 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
902 if (err < 0)
903 break;
904 }
905 err = do_lookup(nd, &this, &next);
906 if (err)
907 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 inode = next.dentry->d_inode;
909 if ((lookup_flags & LOOKUP_FOLLOW)
910 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700911 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912 if (err)
913 goto return_err;
914 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700915 } else
916 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 err = -ENOENT;
918 if (!inode)
919 break;
920 if (lookup_flags & LOOKUP_DIRECTORY) {
921 err = -ENOTDIR;
922 if (!inode->i_op || !inode->i_op->lookup)
923 break;
924 }
925 goto return_base;
926lookup_parent:
927 nd->last = this;
928 nd->last_type = LAST_NORM;
929 if (this.name[0] != '.')
930 goto return_base;
931 if (this.len == 1)
932 nd->last_type = LAST_DOT;
933 else if (this.len == 2 && this.name[1] == '.')
934 nd->last_type = LAST_DOTDOT;
935 else
936 goto return_base;
937return_reval:
938 /*
939 * We bypassed the ordinary revalidation routines.
940 * We may need to check the cached dentry for staleness.
941 */
942 if (nd->dentry && nd->dentry->d_sb &&
943 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
944 err = -ESTALE;
945 /* Note: we do not d_invalidate() */
946 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
947 break;
948 }
949return_base:
950 return 0;
951out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700952 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 break;
954 }
955 path_release(nd);
956return_err:
957 return err;
958}
959
960/*
961 * Wrapper to retry pathname resolution whenever the underlying
962 * file system returns an ESTALE.
963 *
964 * Retry the whole path once, forcing real lookup requests
965 * instead of relying on the dcache.
966 */
967int fastcall link_path_walk(const char *name, struct nameidata *nd)
968{
969 struct nameidata save = *nd;
970 int result;
971
972 /* make sure the stuff we saved doesn't go away */
973 dget(save.dentry);
974 mntget(save.mnt);
975
976 result = __link_path_walk(name, nd);
977 if (result == -ESTALE) {
978 *nd = save;
979 dget(nd->dentry);
980 mntget(nd->mnt);
981 nd->flags |= LOOKUP_REVAL;
982 result = __link_path_walk(name, nd);
983 }
984
985 dput(save.dentry);
986 mntput(save.mnt);
987
988 return result;
989}
990
991int fastcall path_walk(const char * name, struct nameidata *nd)
992{
993 current->total_link_count = 0;
994 return link_path_walk(name, nd);
995}
996
Prasanna Medaea3834d2005-04-29 16:00:17 +0100997/*
998 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
999 * everything is done. Returns 0 and drops input nd, if lookup failed;
1000 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1002{
1003 if (path_walk(name, nd))
1004 return 0; /* something went wrong... */
1005
1006 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1007 struct dentry *old_dentry = nd->dentry;
1008 struct vfsmount *old_mnt = nd->mnt;
1009 struct qstr last = nd->last;
1010 int last_type = nd->last_type;
1011 /*
1012 * NAME was not found in alternate root or it's a directory. Try to find
1013 * it in the normal root:
1014 */
1015 nd->last_type = LAST_ROOT;
1016 read_lock(&current->fs->lock);
1017 nd->mnt = mntget(current->fs->rootmnt);
1018 nd->dentry = dget(current->fs->root);
1019 read_unlock(&current->fs->lock);
1020 if (path_walk(name, nd) == 0) {
1021 if (nd->dentry->d_inode) {
1022 dput(old_dentry);
1023 mntput(old_mnt);
1024 return 1;
1025 }
1026 path_release(nd);
1027 }
1028 nd->dentry = old_dentry;
1029 nd->mnt = old_mnt;
1030 nd->last = last;
1031 nd->last_type = last_type;
1032 }
1033 return 1;
1034}
1035
1036void set_fs_altroot(void)
1037{
1038 char *emul = __emul_prefix();
1039 struct nameidata nd;
1040 struct vfsmount *mnt = NULL, *oldmnt;
1041 struct dentry *dentry = NULL, *olddentry;
1042 int err;
1043
1044 if (!emul)
1045 goto set_it;
1046 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1047 if (!err) {
1048 mnt = nd.mnt;
1049 dentry = nd.dentry;
1050 }
1051set_it:
1052 write_lock(&current->fs->lock);
1053 oldmnt = current->fs->altrootmnt;
1054 olddentry = current->fs->altroot;
1055 current->fs->altrootmnt = mnt;
1056 current->fs->altroot = dentry;
1057 write_unlock(&current->fs->lock);
1058 if (olddentry) {
1059 dput(olddentry);
1060 mntput(oldmnt);
1061 }
1062}
1063
Prasanna Medaea3834d2005-04-29 16:00:17 +01001064/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065int fastcall path_lookup(const char *name, unsigned int flags, struct nameidata *nd)
1066{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001067 int retval = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068
1069 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1070 nd->flags = flags;
1071 nd->depth = 0;
1072
1073 read_lock(&current->fs->lock);
1074 if (*name=='/') {
1075 if (current->fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1076 nd->mnt = mntget(current->fs->altrootmnt);
1077 nd->dentry = dget(current->fs->altroot);
1078 read_unlock(&current->fs->lock);
1079 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001080 goto out; /* found in altroot */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 read_lock(&current->fs->lock);
1082 }
1083 nd->mnt = mntget(current->fs->rootmnt);
1084 nd->dentry = dget(current->fs->root);
1085 } else {
1086 nd->mnt = mntget(current->fs->pwdmnt);
1087 nd->dentry = dget(current->fs->pwd);
1088 }
1089 read_unlock(&current->fs->lock);
1090 current->total_link_count = 0;
1091 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001092out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 if (unlikely(current->audit_context
1094 && nd && nd->dentry && nd->dentry->d_inode))
David Woodhouseae7b9612005-06-20 16:11:05 +01001095 audit_inode(name, nd->dentry->d_inode, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 return retval;
1097}
1098
Trond Myklebust834f2a42005-10-18 14:20:16 -07001099static int __path_lookup_intent_open(const char *name, unsigned int lookup_flags,
1100 struct nameidata *nd, int open_flags, int create_mode)
1101{
1102 struct file *filp = get_empty_filp();
1103 int err;
1104
1105 if (filp == NULL)
1106 return -ENFILE;
1107 nd->intent.open.file = filp;
1108 nd->intent.open.flags = open_flags;
1109 nd->intent.open.create_mode = create_mode;
1110 err = path_lookup(name, lookup_flags|LOOKUP_OPEN, nd);
1111 if (IS_ERR(nd->intent.open.file)) {
1112 if (err == 0) {
1113 err = PTR_ERR(nd->intent.open.file);
1114 path_release(nd);
1115 }
1116 } else if (err != 0)
1117 release_open_intent(nd);
1118 return err;
1119}
1120
1121/**
1122 * path_lookup_open - lookup a file path with open intent
1123 * @name: pointer to file name
1124 * @lookup_flags: lookup intent flags
1125 * @nd: pointer to nameidata
1126 * @open_flags: open intent flags
1127 */
1128int path_lookup_open(const char *name, unsigned int lookup_flags,
1129 struct nameidata *nd, int open_flags)
1130{
1131 return __path_lookup_intent_open(name, lookup_flags, nd,
1132 open_flags, 0);
1133}
1134
1135/**
1136 * path_lookup_create - lookup a file path with open + create intent
1137 * @name: pointer to file name
1138 * @lookup_flags: lookup intent flags
1139 * @nd: pointer to nameidata
1140 * @open_flags: open intent flags
1141 * @create_mode: create intent flags
1142 */
Adrian Bunk0ce6e622005-11-08 21:39:46 -08001143static int path_lookup_create(const char *name, unsigned int lookup_flags,
1144 struct nameidata *nd, int open_flags,
1145 int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001146{
1147 return __path_lookup_intent_open(name, lookup_flags|LOOKUP_CREATE, nd,
1148 open_flags, create_mode);
1149}
1150
1151int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1152 struct nameidata *nd, int open_flags)
1153{
1154 char *tmp = getname(name);
1155 int err = PTR_ERR(tmp);
1156
1157 if (!IS_ERR(tmp)) {
1158 err = __path_lookup_intent_open(tmp, lookup_flags, nd, open_flags, 0);
1159 putname(tmp);
1160 }
1161 return err;
1162}
1163
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164/*
1165 * Restricted form of lookup. Doesn't follow links, single-component only,
1166 * needs parent already locked. Doesn't follow mounts.
1167 * SMP-safe.
1168 */
1169static struct dentry * __lookup_hash(struct qstr *name, struct dentry * base, struct nameidata *nd)
1170{
1171 struct dentry * dentry;
1172 struct inode *inode;
1173 int err;
1174
1175 inode = base->d_inode;
1176 err = permission(inode, MAY_EXEC, nd);
1177 dentry = ERR_PTR(err);
1178 if (err)
1179 goto out;
1180
1181 /*
1182 * See if the low-level filesystem might want
1183 * to use its own hash..
1184 */
1185 if (base->d_op && base->d_op->d_hash) {
1186 err = base->d_op->d_hash(base, name);
1187 dentry = ERR_PTR(err);
1188 if (err < 0)
1189 goto out;
1190 }
1191
1192 dentry = cached_lookup(base, name, nd);
1193 if (!dentry) {
1194 struct dentry *new = d_alloc(base, name);
1195 dentry = ERR_PTR(-ENOMEM);
1196 if (!new)
1197 goto out;
1198 dentry = inode->i_op->lookup(inode, new, nd);
1199 if (!dentry)
1200 dentry = new;
1201 else
1202 dput(new);
1203 }
1204out:
1205 return dentry;
1206}
1207
Christoph Hellwig49705b72005-11-08 21:35:06 -08001208struct dentry * lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001210 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211}
1212
1213/* SMP-safe */
1214struct dentry * lookup_one_len(const char * name, struct dentry * base, int len)
1215{
1216 unsigned long hash;
1217 struct qstr this;
1218 unsigned int c;
1219
1220 this.name = name;
1221 this.len = len;
1222 if (!len)
1223 goto access;
1224
1225 hash = init_name_hash();
1226 while (len--) {
1227 c = *(const unsigned char *)name++;
1228 if (c == '/' || c == '\0')
1229 goto access;
1230 hash = partial_name_hash(c, hash);
1231 }
1232 this.hash = end_name_hash(hash);
1233
Christoph Hellwig49705b72005-11-08 21:35:06 -08001234 return __lookup_hash(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235access:
1236 return ERR_PTR(-EACCES);
1237}
1238
1239/*
1240 * namei()
1241 *
1242 * is used by most simple commands to get the inode of a specified name.
1243 * Open, link etc use their own routines, but this is enough for things
1244 * like 'chmod' etc.
1245 *
1246 * namei exists in two versions: namei/lnamei. The only difference is
1247 * that namei follows links, while lnamei does not.
1248 * SMP-safe
1249 */
1250int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1251{
1252 char *tmp = getname(name);
1253 int err = PTR_ERR(tmp);
1254
1255 if (!IS_ERR(tmp)) {
1256 err = path_lookup(tmp, flags, nd);
1257 putname(tmp);
1258 }
1259 return err;
1260}
1261
1262/*
1263 * It's inline, so penalty for filesystems that don't use sticky bit is
1264 * minimal.
1265 */
1266static inline int check_sticky(struct inode *dir, struct inode *inode)
1267{
1268 if (!(dir->i_mode & S_ISVTX))
1269 return 0;
1270 if (inode->i_uid == current->fsuid)
1271 return 0;
1272 if (dir->i_uid == current->fsuid)
1273 return 0;
1274 return !capable(CAP_FOWNER);
1275}
1276
1277/*
1278 * Check whether we can remove a link victim from directory dir, check
1279 * whether the type of victim is right.
1280 * 1. We can't do it if dir is read-only (done in permission())
1281 * 2. We should have write and exec permissions on dir
1282 * 3. We can't remove anything from append-only dir
1283 * 4. We can't do anything with immutable dir (done in permission())
1284 * 5. If the sticky bit on dir is set we should either
1285 * a. be owner of dir, or
1286 * b. be owner of victim, or
1287 * c. have CAP_FOWNER capability
1288 * 6. If the victim is append-only or immutable we can't do antyhing with
1289 * links pointing to it.
1290 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1291 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1292 * 9. We can't remove a root or mountpoint.
1293 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1294 * nfs_async_unlink().
1295 */
1296static inline int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1297{
1298 int error;
1299
1300 if (!victim->d_inode)
1301 return -ENOENT;
1302
1303 BUG_ON(victim->d_parent->d_inode != dir);
1304
1305 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1306 if (error)
1307 return error;
1308 if (IS_APPEND(dir))
1309 return -EPERM;
1310 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1311 IS_IMMUTABLE(victim->d_inode))
1312 return -EPERM;
1313 if (isdir) {
1314 if (!S_ISDIR(victim->d_inode->i_mode))
1315 return -ENOTDIR;
1316 if (IS_ROOT(victim))
1317 return -EBUSY;
1318 } else if (S_ISDIR(victim->d_inode->i_mode))
1319 return -EISDIR;
1320 if (IS_DEADDIR(dir))
1321 return -ENOENT;
1322 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1323 return -EBUSY;
1324 return 0;
1325}
1326
1327/* Check whether we can create an object with dentry child in directory
1328 * dir.
1329 * 1. We can't do it if child already exists (open has special treatment for
1330 * this case, but since we are inlined it's OK)
1331 * 2. We can't do it if dir is read-only (done in permission())
1332 * 3. We should have write and exec permissions on dir
1333 * 4. We can't do it if dir is immutable (done in permission())
1334 */
1335static inline int may_create(struct inode *dir, struct dentry *child,
1336 struct nameidata *nd)
1337{
1338 if (child->d_inode)
1339 return -EEXIST;
1340 if (IS_DEADDIR(dir))
1341 return -ENOENT;
1342 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1343}
1344
1345/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 * O_DIRECTORY translates into forcing a directory lookup.
1347 */
1348static inline int lookup_flags(unsigned int f)
1349{
1350 unsigned long retval = LOOKUP_FOLLOW;
1351
1352 if (f & O_NOFOLLOW)
1353 retval &= ~LOOKUP_FOLLOW;
1354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 if (f & O_DIRECTORY)
1356 retval |= LOOKUP_DIRECTORY;
1357
1358 return retval;
1359}
1360
1361/*
1362 * p1 and p2 should be directories on the same fs.
1363 */
1364struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1365{
1366 struct dentry *p;
1367
1368 if (p1 == p2) {
1369 down(&p1->d_inode->i_sem);
1370 return NULL;
1371 }
1372
1373 down(&p1->d_inode->i_sb->s_vfs_rename_sem);
1374
1375 for (p = p1; p->d_parent != p; p = p->d_parent) {
1376 if (p->d_parent == p2) {
1377 down(&p2->d_inode->i_sem);
1378 down(&p1->d_inode->i_sem);
1379 return p;
1380 }
1381 }
1382
1383 for (p = p2; p->d_parent != p; p = p->d_parent) {
1384 if (p->d_parent == p1) {
1385 down(&p1->d_inode->i_sem);
1386 down(&p2->d_inode->i_sem);
1387 return p;
1388 }
1389 }
1390
1391 down(&p1->d_inode->i_sem);
1392 down(&p2->d_inode->i_sem);
1393 return NULL;
1394}
1395
1396void unlock_rename(struct dentry *p1, struct dentry *p2)
1397{
1398 up(&p1->d_inode->i_sem);
1399 if (p1 != p2) {
1400 up(&p2->d_inode->i_sem);
1401 up(&p1->d_inode->i_sb->s_vfs_rename_sem);
1402 }
1403}
1404
1405int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1406 struct nameidata *nd)
1407{
1408 int error = may_create(dir, dentry, nd);
1409
1410 if (error)
1411 return error;
1412
1413 if (!dir->i_op || !dir->i_op->create)
1414 return -EACCES; /* shouldn't it be ENOSYS? */
1415 mode &= S_IALLUGO;
1416 mode |= S_IFREG;
1417 error = security_inode_create(dir, dentry, mode);
1418 if (error)
1419 return error;
1420 DQUOT_INIT(dir);
1421 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001422 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001423 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 return error;
1425}
1426
1427int may_open(struct nameidata *nd, int acc_mode, int flag)
1428{
1429 struct dentry *dentry = nd->dentry;
1430 struct inode *inode = dentry->d_inode;
1431 int error;
1432
1433 if (!inode)
1434 return -ENOENT;
1435
1436 if (S_ISLNK(inode->i_mode))
1437 return -ELOOP;
1438
1439 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1440 return -EISDIR;
1441
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001442 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 if (error)
1444 return error;
1445
1446 /*
1447 * FIFO's, sockets and device files are special: they don't
1448 * actually live on the filesystem itself, and as such you
1449 * can write to them even if the filesystem is read-only.
1450 */
1451 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1452 flag &= ~O_TRUNC;
1453 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1454 if (nd->mnt->mnt_flags & MNT_NODEV)
1455 return -EACCES;
1456
1457 flag &= ~O_TRUNC;
1458 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1459 return -EROFS;
1460 /*
1461 * An append-only file must be opened in append mode for writing.
1462 */
1463 if (IS_APPEND(inode)) {
1464 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1465 return -EPERM;
1466 if (flag & O_TRUNC)
1467 return -EPERM;
1468 }
1469
1470 /* O_NOATIME can only be set by the owner or superuser */
1471 if (flag & O_NOATIME)
1472 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1473 return -EPERM;
1474
1475 /*
1476 * Ensure there are no outstanding leases on the file.
1477 */
1478 error = break_lease(inode, flag);
1479 if (error)
1480 return error;
1481
1482 if (flag & O_TRUNC) {
1483 error = get_write_access(inode);
1484 if (error)
1485 return error;
1486
1487 /*
1488 * Refuse to truncate files with mandatory locks held on them.
1489 */
1490 error = locks_verify_locked(inode);
1491 if (!error) {
1492 DQUOT_INIT(inode);
1493
Miklos Szeredicc4e69d2005-11-07 00:59:49 -08001494 error = do_truncate(dentry, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 }
1496 put_write_access(inode);
1497 if (error)
1498 return error;
1499 } else
1500 if (flag & FMODE_WRITE)
1501 DQUOT_INIT(inode);
1502
1503 return 0;
1504}
1505
1506/*
1507 * open_namei()
1508 *
1509 * namei for open - this is in fact almost the whole open-routine.
1510 *
1511 * Note that the low bits of "flag" aren't the same as in the open
1512 * system call - they are 00 - no permissions needed
1513 * 01 - read permission needed
1514 * 10 - write permission needed
1515 * 11 - read/write permissions needed
1516 * which is a lot more logical, and also allows the "no perm" needed
1517 * for symlinks (where the permissions are checked later).
1518 * SMP-safe
1519 */
1520int open_namei(const char * pathname, int flag, int mode, struct nameidata *nd)
1521{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001522 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001523 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 struct dentry *dir;
1525 int count = 0;
1526
1527 acc_mode = ACC_MODE(flag);
1528
Trond Myklebust834f2a42005-10-18 14:20:16 -07001529 /* O_TRUNC implies we need access checks for write permissions */
1530 if (flag & O_TRUNC)
1531 acc_mode |= MAY_WRITE;
1532
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 /* Allow the LSM permission hook to distinguish append
1534 access from general write access. */
1535 if (flag & O_APPEND)
1536 acc_mode |= MAY_APPEND;
1537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 /*
1539 * The simplest case - just a plain lookup.
1540 */
1541 if (!(flag & O_CREAT)) {
Trond Myklebust834f2a42005-10-18 14:20:16 -07001542 error = path_lookup_open(pathname, lookup_flags(flag), nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 if (error)
1544 return error;
1545 goto ok;
1546 }
1547
1548 /*
1549 * Create - we need to know the parent.
1550 */
Trond Myklebust834f2a42005-10-18 14:20:16 -07001551 error = path_lookup_create(pathname, LOOKUP_PARENT, nd, flag, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 if (error)
1553 return error;
1554
1555 /*
1556 * We have the parent and last component. First of all, check
1557 * that we are not asked to creat(2) an obvious directory - that
1558 * will not do.
1559 */
1560 error = -EISDIR;
1561 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1562 goto exit;
1563
1564 dir = nd->dentry;
1565 nd->flags &= ~LOOKUP_PARENT;
1566 down(&dir->d_inode->i_sem);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001567 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001568 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569
1570do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001571 error = PTR_ERR(path.dentry);
1572 if (IS_ERR(path.dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 up(&dir->d_inode->i_sem);
1574 goto exit;
1575 }
1576
1577 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001578 if (!path.dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 if (!IS_POSIXACL(dir->d_inode))
1580 mode &= ~current->fs->umask;
Al Viro4e7506e2005-06-06 13:36:00 -07001581 error = vfs_create(dir->d_inode, path.dentry, mode, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 up(&dir->d_inode->i_sem);
1583 dput(nd->dentry);
Al Viro4e7506e2005-06-06 13:36:00 -07001584 nd->dentry = path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 if (error)
1586 goto exit;
1587 /* Don't check for write permission, don't truncate */
1588 acc_mode = 0;
1589 flag &= ~O_TRUNC;
1590 goto ok;
1591 }
1592
1593 /*
1594 * It already exists.
1595 */
1596 up(&dir->d_inode->i_sem);
1597
1598 error = -EEXIST;
1599 if (flag & O_EXCL)
1600 goto exit_dput;
1601
Al Viroe13b2102005-06-06 13:36:06 -07001602 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001604 if (flag & O_NOFOLLOW)
1605 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 }
1607 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001608 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001610 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 goto do_link;
1612
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001613 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001615 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 goto exit;
1617ok:
1618 error = may_open(nd, acc_mode, flag);
1619 if (error)
1620 goto exit;
1621 return 0;
1622
1623exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001624 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001626 if (!IS_ERR(nd->intent.open.file))
1627 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 path_release(nd);
1629 return error;
1630
1631do_link:
1632 error = -ELOOP;
1633 if (flag & O_NOFOLLOW)
1634 goto exit_dput;
1635 /*
1636 * This is subtle. Instead of calling do_follow_link() we do the
1637 * thing by hands. The reason is that this way we have zero link_count
1638 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1639 * After that we have the parent and last component, i.e.
1640 * we are in the same situation as after the first path_walk().
1641 * Well, almost - if the last component is normal we get its copy
1642 * stored in nd->last.name and we will have to putname() it when we
1643 * are done. Procfs-like symlinks just set LAST_BIND.
1644 */
1645 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001646 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 if (error)
1648 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001649 error = __do_follow_link(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650 if (error)
1651 return error;
1652 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001653 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 error = -EISDIR;
1656 if (nd->last_type != LAST_NORM)
1657 goto exit;
1658 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001659 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 goto exit;
1661 }
1662 error = -ELOOP;
1663 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001664 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 goto exit;
1666 }
1667 dir = nd->dentry;
1668 down(&dir->d_inode->i_sem);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001669 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001670 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001671 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 goto do_last;
1673}
1674
1675/**
1676 * lookup_create - lookup a dentry, creating it if it doesn't exist
1677 * @nd: nameidata info
1678 * @is_dir: directory flag
1679 *
1680 * Simple function to lookup and return a dentry and create it
1681 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001682 *
1683 * Returns with nd->dentry->d_inode->i_sem locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 */
1685struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1686{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001687 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688
1689 down(&nd->dentry->d_inode->i_sem);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001690 /*
1691 * Yucky last component or no last component at all?
1692 * (foo/., foo/.., /////)
1693 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 if (nd->last_type != LAST_NORM)
1695 goto fail;
1696 nd->flags &= ~LOOKUP_PARENT;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001697
1698 /*
1699 * Do the final lookup.
1700 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001701 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 if (IS_ERR(dentry))
1703 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001704
1705 /*
1706 * Special case - lookup gave negative, but... we had foo/bar/
1707 * From the vfs_mknod() POV we just have a negative dentry -
1708 * all is fine. Let's be bastards - you had / on the end, you've
1709 * been asking for (non-existent) directory. -ENOENT for you.
1710 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1712 goto enoent;
1713 return dentry;
1714enoent:
1715 dput(dentry);
1716 dentry = ERR_PTR(-ENOENT);
1717fail:
1718 return dentry;
1719}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001720EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
1722int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1723{
1724 int error = may_create(dir, dentry, NULL);
1725
1726 if (error)
1727 return error;
1728
1729 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1730 return -EPERM;
1731
1732 if (!dir->i_op || !dir->i_op->mknod)
1733 return -EPERM;
1734
1735 error = security_inode_mknod(dir, dentry, mode, dev);
1736 if (error)
1737 return error;
1738
1739 DQUOT_INIT(dir);
1740 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001741 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001742 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 return error;
1744}
1745
1746asmlinkage long sys_mknod(const char __user * filename, int mode, unsigned dev)
1747{
1748 int error = 0;
1749 char * tmp;
1750 struct dentry * dentry;
1751 struct nameidata nd;
1752
1753 if (S_ISDIR(mode))
1754 return -EPERM;
1755 tmp = getname(filename);
1756 if (IS_ERR(tmp))
1757 return PTR_ERR(tmp);
1758
1759 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1760 if (error)
1761 goto out;
1762 dentry = lookup_create(&nd, 0);
1763 error = PTR_ERR(dentry);
1764
1765 if (!IS_POSIXACL(nd.dentry->d_inode))
1766 mode &= ~current->fs->umask;
1767 if (!IS_ERR(dentry)) {
1768 switch (mode & S_IFMT) {
1769 case 0: case S_IFREG:
1770 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1771 break;
1772 case S_IFCHR: case S_IFBLK:
1773 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1774 new_decode_dev(dev));
1775 break;
1776 case S_IFIFO: case S_IFSOCK:
1777 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1778 break;
1779 case S_IFDIR:
1780 error = -EPERM;
1781 break;
1782 default:
1783 error = -EINVAL;
1784 }
1785 dput(dentry);
1786 }
1787 up(&nd.dentry->d_inode->i_sem);
1788 path_release(&nd);
1789out:
1790 putname(tmp);
1791
1792 return error;
1793}
1794
1795int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1796{
1797 int error = may_create(dir, dentry, NULL);
1798
1799 if (error)
1800 return error;
1801
1802 if (!dir->i_op || !dir->i_op->mkdir)
1803 return -EPERM;
1804
1805 mode &= (S_IRWXUGO|S_ISVTX);
1806 error = security_inode_mkdir(dir, dentry, mode);
1807 if (error)
1808 return error;
1809
1810 DQUOT_INIT(dir);
1811 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001812 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04001813 fsnotify_mkdir(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 return error;
1815}
1816
1817asmlinkage long sys_mkdir(const char __user * pathname, int mode)
1818{
1819 int error = 0;
1820 char * tmp;
1821
1822 tmp = getname(pathname);
1823 error = PTR_ERR(tmp);
1824 if (!IS_ERR(tmp)) {
1825 struct dentry *dentry;
1826 struct nameidata nd;
1827
1828 error = path_lookup(tmp, LOOKUP_PARENT, &nd);
1829 if (error)
1830 goto out;
1831 dentry = lookup_create(&nd, 1);
1832 error = PTR_ERR(dentry);
1833 if (!IS_ERR(dentry)) {
1834 if (!IS_POSIXACL(nd.dentry->d_inode))
1835 mode &= ~current->fs->umask;
1836 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1837 dput(dentry);
1838 }
1839 up(&nd.dentry->d_inode->i_sem);
1840 path_release(&nd);
1841out:
1842 putname(tmp);
1843 }
1844
1845 return error;
1846}
1847
1848/*
1849 * We try to drop the dentry early: we should have
1850 * a usage count of 2 if we're the only user of this
1851 * dentry, and if that is true (possibly after pruning
1852 * the dcache), then we drop the dentry now.
1853 *
1854 * A low-level filesystem can, if it choses, legally
1855 * do a
1856 *
1857 * if (!d_unhashed(dentry))
1858 * return -EBUSY;
1859 *
1860 * if it cannot handle the case of removing a directory
1861 * that is still in use by something else..
1862 */
1863void dentry_unhash(struct dentry *dentry)
1864{
1865 dget(dentry);
1866 if (atomic_read(&dentry->d_count))
1867 shrink_dcache_parent(dentry);
1868 spin_lock(&dcache_lock);
1869 spin_lock(&dentry->d_lock);
1870 if (atomic_read(&dentry->d_count) == 2)
1871 __d_drop(dentry);
1872 spin_unlock(&dentry->d_lock);
1873 spin_unlock(&dcache_lock);
1874}
1875
1876int vfs_rmdir(struct inode *dir, struct dentry *dentry)
1877{
1878 int error = may_delete(dir, dentry, 1);
1879
1880 if (error)
1881 return error;
1882
1883 if (!dir->i_op || !dir->i_op->rmdir)
1884 return -EPERM;
1885
1886 DQUOT_INIT(dir);
1887
1888 down(&dentry->d_inode->i_sem);
1889 dentry_unhash(dentry);
1890 if (d_mountpoint(dentry))
1891 error = -EBUSY;
1892 else {
1893 error = security_inode_rmdir(dir, dentry);
1894 if (!error) {
1895 error = dir->i_op->rmdir(dir, dentry);
1896 if (!error)
1897 dentry->d_inode->i_flags |= S_DEAD;
1898 }
1899 }
1900 up(&dentry->d_inode->i_sem);
1901 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 d_delete(dentry);
1903 }
1904 dput(dentry);
1905
1906 return error;
1907}
1908
1909asmlinkage long sys_rmdir(const char __user * pathname)
1910{
1911 int error = 0;
1912 char * name;
1913 struct dentry *dentry;
1914 struct nameidata nd;
1915
1916 name = getname(pathname);
1917 if(IS_ERR(name))
1918 return PTR_ERR(name);
1919
1920 error = path_lookup(name, LOOKUP_PARENT, &nd);
1921 if (error)
1922 goto exit;
1923
1924 switch(nd.last_type) {
1925 case LAST_DOTDOT:
1926 error = -ENOTEMPTY;
1927 goto exit1;
1928 case LAST_DOT:
1929 error = -EINVAL;
1930 goto exit1;
1931 case LAST_ROOT:
1932 error = -EBUSY;
1933 goto exit1;
1934 }
1935 down(&nd.dentry->d_inode->i_sem);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001936 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 error = PTR_ERR(dentry);
1938 if (!IS_ERR(dentry)) {
1939 error = vfs_rmdir(nd.dentry->d_inode, dentry);
1940 dput(dentry);
1941 }
1942 up(&nd.dentry->d_inode->i_sem);
1943exit1:
1944 path_release(&nd);
1945exit:
1946 putname(name);
1947 return error;
1948}
1949
1950int vfs_unlink(struct inode *dir, struct dentry *dentry)
1951{
1952 int error = may_delete(dir, dentry, 0);
1953
1954 if (error)
1955 return error;
1956
1957 if (!dir->i_op || !dir->i_op->unlink)
1958 return -EPERM;
1959
1960 DQUOT_INIT(dir);
1961
1962 down(&dentry->d_inode->i_sem);
1963 if (d_mountpoint(dentry))
1964 error = -EBUSY;
1965 else {
1966 error = security_inode_unlink(dir, dentry);
1967 if (!error)
1968 error = dir->i_op->unlink(dir, dentry);
1969 }
1970 up(&dentry->d_inode->i_sem);
1971
1972 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
1973 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
1974 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 }
Robert Love0eeca282005-07-12 17:06:03 -04001976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 return error;
1978}
1979
1980/*
1981 * Make sure that the actual truncation of the file will occur outside its
1982 * directory's i_sem. Truncate can take a long time if there is a lot of
1983 * writeout happening, and we don't want to prevent access to the directory
1984 * while waiting on the I/O.
1985 */
1986asmlinkage long sys_unlink(const char __user * pathname)
1987{
1988 int error = 0;
1989 char * name;
1990 struct dentry *dentry;
1991 struct nameidata nd;
1992 struct inode *inode = NULL;
1993
1994 name = getname(pathname);
1995 if(IS_ERR(name))
1996 return PTR_ERR(name);
1997
1998 error = path_lookup(name, LOOKUP_PARENT, &nd);
1999 if (error)
2000 goto exit;
2001 error = -EISDIR;
2002 if (nd.last_type != LAST_NORM)
2003 goto exit1;
2004 down(&nd.dentry->d_inode->i_sem);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002005 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 error = PTR_ERR(dentry);
2007 if (!IS_ERR(dentry)) {
2008 /* Why not before? Because we want correct error value */
2009 if (nd.last.name[nd.last.len])
2010 goto slashes;
2011 inode = dentry->d_inode;
2012 if (inode)
2013 atomic_inc(&inode->i_count);
2014 error = vfs_unlink(nd.dentry->d_inode, dentry);
2015 exit2:
2016 dput(dentry);
2017 }
2018 up(&nd.dentry->d_inode->i_sem);
2019 if (inode)
2020 iput(inode); /* truncate the inode here */
2021exit1:
2022 path_release(&nd);
2023exit:
2024 putname(name);
2025 return error;
2026
2027slashes:
2028 error = !dentry->d_inode ? -ENOENT :
2029 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2030 goto exit2;
2031}
2032
2033int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2034{
2035 int error = may_create(dir, dentry, NULL);
2036
2037 if (error)
2038 return error;
2039
2040 if (!dir->i_op || !dir->i_op->symlink)
2041 return -EPERM;
2042
2043 error = security_inode_symlink(dir, dentry, oldname);
2044 if (error)
2045 return error;
2046
2047 DQUOT_INIT(dir);
2048 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002049 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04002050 fsnotify_create(dir, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 return error;
2052}
2053
2054asmlinkage long sys_symlink(const char __user * oldname, const char __user * newname)
2055{
2056 int error = 0;
2057 char * from;
2058 char * to;
2059
2060 from = getname(oldname);
2061 if(IS_ERR(from))
2062 return PTR_ERR(from);
2063 to = getname(newname);
2064 error = PTR_ERR(to);
2065 if (!IS_ERR(to)) {
2066 struct dentry *dentry;
2067 struct nameidata nd;
2068
2069 error = path_lookup(to, LOOKUP_PARENT, &nd);
2070 if (error)
2071 goto out;
2072 dentry = lookup_create(&nd, 0);
2073 error = PTR_ERR(dentry);
2074 if (!IS_ERR(dentry)) {
2075 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2076 dput(dentry);
2077 }
2078 up(&nd.dentry->d_inode->i_sem);
2079 path_release(&nd);
2080out:
2081 putname(to);
2082 }
2083 putname(from);
2084 return error;
2085}
2086
2087int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2088{
2089 struct inode *inode = old_dentry->d_inode;
2090 int error;
2091
2092 if (!inode)
2093 return -ENOENT;
2094
2095 error = may_create(dir, new_dentry, NULL);
2096 if (error)
2097 return error;
2098
2099 if (dir->i_sb != inode->i_sb)
2100 return -EXDEV;
2101
2102 /*
2103 * A link to an append-only or immutable file cannot be created.
2104 */
2105 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2106 return -EPERM;
2107 if (!dir->i_op || !dir->i_op->link)
2108 return -EPERM;
2109 if (S_ISDIR(old_dentry->d_inode->i_mode))
2110 return -EPERM;
2111
2112 error = security_inode_link(old_dentry, dir, new_dentry);
2113 if (error)
2114 return error;
2115
2116 down(&old_dentry->d_inode->i_sem);
2117 DQUOT_INIT(dir);
2118 error = dir->i_op->link(old_dentry, dir, new_dentry);
2119 up(&old_dentry->d_inode->i_sem);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002120 if (!error)
Robert Love0eeca282005-07-12 17:06:03 -04002121 fsnotify_create(dir, new_dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 return error;
2123}
2124
2125/*
2126 * Hardlinks are often used in delicate situations. We avoid
2127 * security-related surprises by not following symlinks on the
2128 * newname. --KAB
2129 *
2130 * We don't follow them on the oldname either to be compatible
2131 * with linux 2.0, and to avoid hard-linking to directories
2132 * and other special files. --ADM
2133 */
2134asmlinkage long sys_link(const char __user * oldname, const char __user * newname)
2135{
2136 struct dentry *new_dentry;
2137 struct nameidata nd, old_nd;
2138 int error;
2139 char * to;
2140
2141 to = getname(newname);
2142 if (IS_ERR(to))
2143 return PTR_ERR(to);
2144
2145 error = __user_walk(oldname, 0, &old_nd);
2146 if (error)
2147 goto exit;
2148 error = path_lookup(to, LOOKUP_PARENT, &nd);
2149 if (error)
2150 goto out;
2151 error = -EXDEV;
2152 if (old_nd.mnt != nd.mnt)
2153 goto out_release;
2154 new_dentry = lookup_create(&nd, 0);
2155 error = PTR_ERR(new_dentry);
2156 if (!IS_ERR(new_dentry)) {
2157 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2158 dput(new_dentry);
2159 }
2160 up(&nd.dentry->d_inode->i_sem);
2161out_release:
2162 path_release(&nd);
2163out:
2164 path_release(&old_nd);
2165exit:
2166 putname(to);
2167
2168 return error;
2169}
2170
2171/*
2172 * The worst of all namespace operations - renaming directory. "Perverted"
2173 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2174 * Problems:
2175 * a) we can get into loop creation. Check is done in is_subdir().
2176 * b) race potential - two innocent renames can create a loop together.
2177 * That's where 4.4 screws up. Current fix: serialization on
2178 * sb->s_vfs_rename_sem. We might be more accurate, but that's another
2179 * story.
2180 * c) we have to lock _three_ objects - parents and victim (if it exists).
2181 * And that - after we got ->i_sem on parents (until then we don't know
2182 * whether the target exists). Solution: try to be smart with locking
2183 * order for inodes. We rely on the fact that tree topology may change
2184 * only under ->s_vfs_rename_sem _and_ that parent of the object we
2185 * move will be locked. Thus we can rank directories by the tree
2186 * (ancestors first) and rank all non-directories after them.
2187 * That works since everybody except rename does "lock parent, lookup,
2188 * lock child" and rename is under ->s_vfs_rename_sem.
2189 * HOWEVER, it relies on the assumption that any object with ->lookup()
2190 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2191 * we'd better make sure that there's no link(2) for them.
2192 * d) some filesystems don't support opened-but-unlinked directories,
2193 * either because of layout or because they are not ready to deal with
2194 * all cases correctly. The latter will be fixed (taking this sort of
2195 * stuff into VFS), but the former is not going away. Solution: the same
2196 * trick as in rmdir().
2197 * e) conversion from fhandle to dentry may come in the wrong moment - when
2198 * we are removing the target. Solution: we will have to grab ->i_sem
2199 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2200 * ->i_sem on parents, which works but leads to some truely excessive
2201 * locking].
2202 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002203static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2204 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205{
2206 int error = 0;
2207 struct inode *target;
2208
2209 /*
2210 * If we are going to change the parent - check write permissions,
2211 * we'll need to flip '..'.
2212 */
2213 if (new_dir != old_dir) {
2214 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2215 if (error)
2216 return error;
2217 }
2218
2219 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2220 if (error)
2221 return error;
2222
2223 target = new_dentry->d_inode;
2224 if (target) {
2225 down(&target->i_sem);
2226 dentry_unhash(new_dentry);
2227 }
2228 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2229 error = -EBUSY;
2230 else
2231 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2232 if (target) {
2233 if (!error)
2234 target->i_flags |= S_DEAD;
2235 up(&target->i_sem);
2236 if (d_unhashed(new_dentry))
2237 d_rehash(new_dentry);
2238 dput(new_dentry);
2239 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002240 if (!error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 return error;
2243}
2244
Adrian Bunk75c96f82005-05-05 16:16:09 -07002245static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2246 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247{
2248 struct inode *target;
2249 int error;
2250
2251 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2252 if (error)
2253 return error;
2254
2255 dget(new_dentry);
2256 target = new_dentry->d_inode;
2257 if (target)
2258 down(&target->i_sem);
2259 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2260 error = -EBUSY;
2261 else
2262 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2263 if (!error) {
2264 /* The following d_move() should become unconditional */
2265 if (!(old_dir->i_sb->s_type->fs_flags & FS_ODD_RENAME))
2266 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 }
2268 if (target)
2269 up(&target->i_sem);
2270 dput(new_dentry);
2271 return error;
2272}
2273
2274int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2275 struct inode *new_dir, struct dentry *new_dentry)
2276{
2277 int error;
2278 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002279 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280
2281 if (old_dentry->d_inode == new_dentry->d_inode)
2282 return 0;
2283
2284 error = may_delete(old_dir, old_dentry, is_dir);
2285 if (error)
2286 return error;
2287
2288 if (!new_dentry->d_inode)
2289 error = may_create(new_dir, new_dentry, NULL);
2290 else
2291 error = may_delete(new_dir, new_dentry, is_dir);
2292 if (error)
2293 return error;
2294
2295 if (!old_dir->i_op || !old_dir->i_op->rename)
2296 return -EPERM;
2297
2298 DQUOT_INIT(old_dir);
2299 DQUOT_INIT(new_dir);
2300
Robert Love0eeca282005-07-12 17:06:03 -04002301 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2302
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303 if (is_dir)
2304 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2305 else
2306 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2307 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002308 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002309 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2310 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311 }
Robert Love0eeca282005-07-12 17:06:03 -04002312 fsnotify_oldname_free(old_name);
2313
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 return error;
2315}
2316
2317static inline int do_rename(const char * oldname, const char * newname)
2318{
2319 int error = 0;
2320 struct dentry * old_dir, * new_dir;
2321 struct dentry * old_dentry, *new_dentry;
2322 struct dentry * trap;
2323 struct nameidata oldnd, newnd;
2324
2325 error = path_lookup(oldname, LOOKUP_PARENT, &oldnd);
2326 if (error)
2327 goto exit;
2328
2329 error = path_lookup(newname, LOOKUP_PARENT, &newnd);
2330 if (error)
2331 goto exit1;
2332
2333 error = -EXDEV;
2334 if (oldnd.mnt != newnd.mnt)
2335 goto exit2;
2336
2337 old_dir = oldnd.dentry;
2338 error = -EBUSY;
2339 if (oldnd.last_type != LAST_NORM)
2340 goto exit2;
2341
2342 new_dir = newnd.dentry;
2343 if (newnd.last_type != LAST_NORM)
2344 goto exit2;
2345
2346 trap = lock_rename(new_dir, old_dir);
2347
Christoph Hellwig49705b72005-11-08 21:35:06 -08002348 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 error = PTR_ERR(old_dentry);
2350 if (IS_ERR(old_dentry))
2351 goto exit3;
2352 /* source must exist */
2353 error = -ENOENT;
2354 if (!old_dentry->d_inode)
2355 goto exit4;
2356 /* unless the source is a directory trailing slashes give -ENOTDIR */
2357 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2358 error = -ENOTDIR;
2359 if (oldnd.last.name[oldnd.last.len])
2360 goto exit4;
2361 if (newnd.last.name[newnd.last.len])
2362 goto exit4;
2363 }
2364 /* source should not be ancestor of target */
2365 error = -EINVAL;
2366 if (old_dentry == trap)
2367 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002368 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 error = PTR_ERR(new_dentry);
2370 if (IS_ERR(new_dentry))
2371 goto exit4;
2372 /* target should not be an ancestor of source */
2373 error = -ENOTEMPTY;
2374 if (new_dentry == trap)
2375 goto exit5;
2376
2377 error = vfs_rename(old_dir->d_inode, old_dentry,
2378 new_dir->d_inode, new_dentry);
2379exit5:
2380 dput(new_dentry);
2381exit4:
2382 dput(old_dentry);
2383exit3:
2384 unlock_rename(new_dir, old_dir);
2385exit2:
2386 path_release(&newnd);
2387exit1:
2388 path_release(&oldnd);
2389exit:
2390 return error;
2391}
2392
2393asmlinkage long sys_rename(const char __user * oldname, const char __user * newname)
2394{
2395 int error;
2396 char * from;
2397 char * to;
2398
2399 from = getname(oldname);
2400 if(IS_ERR(from))
2401 return PTR_ERR(from);
2402 to = getname(newname);
2403 error = PTR_ERR(to);
2404 if (!IS_ERR(to)) {
2405 error = do_rename(from,to);
2406 putname(to);
2407 }
2408 putname(from);
2409 return error;
2410}
2411
2412int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2413{
2414 int len;
2415
2416 len = PTR_ERR(link);
2417 if (IS_ERR(link))
2418 goto out;
2419
2420 len = strlen(link);
2421 if (len > (unsigned) buflen)
2422 len = buflen;
2423 if (copy_to_user(buffer, link, len))
2424 len = -EFAULT;
2425out:
2426 return len;
2427}
2428
2429/*
2430 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2431 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2432 * using) it for any given inode is up to filesystem.
2433 */
2434int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2435{
2436 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002437 void *cookie;
2438
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002440 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2441 if (!IS_ERR(cookie)) {
2442 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002444 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2445 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002447 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448}
2449
2450int vfs_follow_link(struct nameidata *nd, const char *link)
2451{
2452 return __vfs_follow_link(nd, link);
2453}
2454
2455/* get the link contents into pagecache */
2456static char *page_getlink(struct dentry * dentry, struct page **ppage)
2457{
2458 struct page * page;
2459 struct address_space *mapping = dentry->d_inode->i_mapping;
2460 page = read_cache_page(mapping, 0, (filler_t *)mapping->a_ops->readpage,
2461 NULL);
2462 if (IS_ERR(page))
2463 goto sync_fail;
2464 wait_on_page_locked(page);
2465 if (!PageUptodate(page))
2466 goto async_fail;
2467 *ppage = page;
2468 return kmap(page);
2469
2470async_fail:
2471 page_cache_release(page);
2472 return ERR_PTR(-EIO);
2473
2474sync_fail:
2475 return (char*)page;
2476}
2477
2478int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2479{
2480 struct page *page = NULL;
2481 char *s = page_getlink(dentry, &page);
2482 int res = vfs_readlink(dentry,buffer,buflen,s);
2483 if (page) {
2484 kunmap(page);
2485 page_cache_release(page);
2486 }
2487 return res;
2488}
2489
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002490void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002492 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002494 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495}
2496
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002497void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002499 struct page *page = cookie;
2500
2501 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 kunmap(page);
2503 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 }
2505}
2506
2507int page_symlink(struct inode *inode, const char *symname, int len)
2508{
2509 struct address_space *mapping = inode->i_mapping;
2510 struct page *page = grab_cache_page(mapping, 0);
2511 int err = -ENOMEM;
2512 char *kaddr;
2513
2514 if (!page)
2515 goto fail;
2516 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
2517 if (err)
2518 goto fail_map;
2519 kaddr = kmap_atomic(page, KM_USER0);
2520 memcpy(kaddr, symname, len-1);
2521 kunmap_atomic(kaddr, KM_USER0);
2522 mapping->a_ops->commit_write(NULL, page, 0, len-1);
2523 /*
2524 * Notice that we are _not_ going to block here - end of page is
2525 * unmapped, so this will only try to map the rest of page, see
2526 * that it is unmapped (typically even will not look into inode -
2527 * ->i_size will be enough for everything) and zero it out.
2528 * OTOH it's obviously correct and should make the page up-to-date.
2529 */
2530 if (!PageUptodate(page)) {
2531 err = mapping->a_ops->readpage(NULL, page);
2532 wait_on_page_locked(page);
2533 } else {
2534 unlock_page(page);
2535 }
2536 page_cache_release(page);
2537 if (err < 0)
2538 goto fail;
2539 mark_inode_dirty(inode);
2540 return 0;
2541fail_map:
2542 unlock_page(page);
2543 page_cache_release(page);
2544fail:
2545 return err;
2546}
2547
2548struct inode_operations page_symlink_inode_operations = {
2549 .readlink = generic_readlink,
2550 .follow_link = page_follow_link_light,
2551 .put_link = page_put_link,
2552};
2553
2554EXPORT_SYMBOL(__user_walk);
2555EXPORT_SYMBOL(follow_down);
2556EXPORT_SYMBOL(follow_up);
2557EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2558EXPORT_SYMBOL(getname);
2559EXPORT_SYMBOL(lock_rename);
2560EXPORT_SYMBOL(lookup_hash);
2561EXPORT_SYMBOL(lookup_one_len);
2562EXPORT_SYMBOL(page_follow_link_light);
2563EXPORT_SYMBOL(page_put_link);
2564EXPORT_SYMBOL(page_readlink);
2565EXPORT_SYMBOL(page_symlink);
2566EXPORT_SYMBOL(page_symlink_inode_operations);
2567EXPORT_SYMBOL(path_lookup);
2568EXPORT_SYMBOL(path_release);
2569EXPORT_SYMBOL(path_walk);
2570EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002571EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002572EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573EXPORT_SYMBOL(unlock_rename);
2574EXPORT_SYMBOL(vfs_create);
2575EXPORT_SYMBOL(vfs_follow_link);
2576EXPORT_SYMBOL(vfs_link);
2577EXPORT_SYMBOL(vfs_mkdir);
2578EXPORT_SYMBOL(vfs_mknod);
2579EXPORT_SYMBOL(generic_permission);
2580EXPORT_SYMBOL(vfs_readlink);
2581EXPORT_SYMBOL(vfs_rename);
2582EXPORT_SYMBOL(vfs_rmdir);
2583EXPORT_SYMBOL(vfs_symlink);
2584EXPORT_SYMBOL(vfs_unlink);
2585EXPORT_SYMBOL(dentry_unhash);
2586EXPORT_SYMBOL(generic_readlink);