Version:  2.0.40 2.2.26 2.4.37 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18

Linux/fs/namei.c

  1 /*
  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/export.h>
 19 #include <linux/kernel.h>
 20 #include <linux/slab.h>
 21 #include <linux/fs.h>
 22 #include <linux/namei.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/fsnotify.h>
 25 #include <linux/personality.h>
 26 #include <linux/security.h>
 27 #include <linux/ima.h>
 28 #include <linux/syscalls.h>
 29 #include <linux/mount.h>
 30 #include <linux/audit.h>
 31 #include <linux/capability.h>
 32 #include <linux/file.h>
 33 #include <linux/fcntl.h>
 34 #include <linux/device_cgroup.h>
 35 #include <linux/fs_struct.h>
 36 #include <linux/posix_acl.h>
 37 #include <linux/hash.h>
 38 #include <asm/uaccess.h>
 39 
 40 #include "internal.h"
 41 #include "mount.h"
 42 
 43 /* [Feb-1997 T. Schoebel-Theuer]
 44  * Fundamental changes in the pathname lookup mechanisms (namei)
 45  * were necessary because of omirr.  The reason is that omirr needs
 46  * to know the _real_ pathname, not the user-supplied one, in case
 47  * of symlinks (and also when transname replacements occur).
 48  *
 49  * The new code replaces the old recursive symlink resolution with
 50  * an iterative one (in case of non-nested symlink chains).  It does
 51  * this with calls to <fs>_follow_link().
 52  * As a side effect, dir_namei(), _namei() and follow_link() are now 
 53  * replaced with a single function lookup_dentry() that can handle all 
 54  * the special cases of the former code.
 55  *
 56  * With the new dcache, the pathname is stored at each inode, at least as
 57  * long as the refcount of the inode is positive.  As a side effect, the
 58  * size of the dcache depends on the inode cache and thus is dynamic.
 59  *
 60  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
 61  * resolution to correspond with current state of the code.
 62  *
 63  * Note that the symlink resolution is not *completely* iterative.
 64  * There is still a significant amount of tail- and mid- recursion in
 65  * the algorithm.  Also, note that <fs>_readlink() is not used in
 66  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
 67  * may return different results than <fs>_follow_link().  Many virtual
 68  * filesystems (including /proc) exhibit this behavior.
 69  */
 70 
 71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
 72  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
 73  * and the name already exists in form of a symlink, try to create the new
 74  * name indicated by the symlink. The old code always complained that the
 75  * name already exists, due to not following the symlink even if its target
 76  * is nonexistent.  The new semantics affects also mknod() and link() when
 77  * the name is a symlink pointing to a non-existent name.
 78  *
 79  * I don't know which semantics is the right one, since I have no access
 80  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
 81  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
 82  * "old" one. Personally, I think the new semantics is much more logical.
 83  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
 84  * file does succeed in both HP-UX and SunOs, but not in Solaris
 85  * and in the old Linux semantics.
 86  */
 87 
 88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
 89  * semantics.  See the comments in "open_namei" and "do_link" below.
 90  *
 91  * [10-Sep-98 Alan Modra] Another symlink change.
 92  */
 93 
 94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
 95  *      inside the path - always follow.
 96  *      in the last component in creation/removal/renaming - never follow.
 97  *      if LOOKUP_FOLLOW passed - follow.
 98  *      if the pathname has trailing slashes - follow.
 99  *      otherwise - don't follow.
100  * (applied in that order).
101  *
102  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104  * During the 2.4 we need to fix the userland stuff depending on it -
105  * hopefully we will be able to get rid of that wart in 2.5. So far only
106  * XEmacs seems to be relying on it...
107  */
108 /*
109  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
111  * any extra contention...
112  */
113 
114 /* In order to reduce some races, while at the same time doing additional
115  * checking and hopefully speeding things up, we copy filenames to the
116  * kernel data space before using them..
117  *
118  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119  * PATH_MAX includes the nul terminator --RR.
120  */
121 void final_putname(struct filename *name)
122 {
123         if (name->separate) {
124                 __putname(name->name);
125                 kfree(name);
126         } else {
127                 __putname(name);
128         }
129 }
130 
131 #define EMBEDDED_NAME_MAX       (PATH_MAX - sizeof(struct filename))
132 
133 static struct filename *
134 getname_flags(const char __user *filename, int flags, int *empty)
135 {
136         struct filename *result, *err;
137         int len;
138         long max;
139         char *kname;
140 
141         result = audit_reusename(filename);
142         if (result)
143                 return result;
144 
145         result = __getname();
146         if (unlikely(!result))
147                 return ERR_PTR(-ENOMEM);
148 
149         /*
150          * First, try to embed the struct filename inside the names_cache
151          * allocation
152          */
153         kname = (char *)result + sizeof(*result);
154         result->name = kname;
155         result->separate = false;
156         max = EMBEDDED_NAME_MAX;
157 
158 recopy:
159         len = strncpy_from_user(kname, filename, max);
160         if (unlikely(len < 0)) {
161                 err = ERR_PTR(len);
162                 goto error;
163         }
164 
165         /*
166          * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
167          * separate struct filename so we can dedicate the entire
168          * names_cache allocation for the pathname, and re-do the copy from
169          * userland.
170          */
171         if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
172                 kname = (char *)result;
173 
174                 result = kzalloc(sizeof(*result), GFP_KERNEL);
175                 if (!result) {
176                         err = ERR_PTR(-ENOMEM);
177                         result = (struct filename *)kname;
178                         goto error;
179                 }
180                 result->name = kname;
181                 result->separate = true;
182                 max = PATH_MAX;
183                 goto recopy;
184         }
185 
186         /* The empty path is special. */
187         if (unlikely(!len)) {
188                 if (empty)
189                         *empty = 1;
190                 err = ERR_PTR(-ENOENT);
191                 if (!(flags & LOOKUP_EMPTY))
192                         goto error;
193         }
194 
195         err = ERR_PTR(-ENAMETOOLONG);
196         if (unlikely(len >= PATH_MAX))
197                 goto error;
198 
199         result->uptr = filename;
200         result->aname = NULL;
201         audit_getname(result);
202         return result;
203 
204 error:
205         final_putname(result);
206         return err;
207 }
208 
209 struct filename *
210 getname(const char __user * filename)
211 {
212         return getname_flags(filename, 0, NULL);
213 }
214 
215 /*
216  * The "getname_kernel()" interface doesn't do pathnames longer
217  * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
218  */
219 struct filename *
220 getname_kernel(const char * filename)
221 {
222         struct filename *result;
223         char *kname;
224         int len;
225 
226         len = strlen(filename);
227         if (len >= EMBEDDED_NAME_MAX)
228                 return ERR_PTR(-ENAMETOOLONG);
229 
230         result = __getname();
231         if (unlikely(!result))
232                 return ERR_PTR(-ENOMEM);
233 
234         kname = (char *)result + sizeof(*result);
235         result->name = kname;
236         result->uptr = NULL;
237         result->aname = NULL;
238         result->separate = false;
239 
240         strlcpy(kname, filename, EMBEDDED_NAME_MAX);
241         return result;
242 }
243 
244 #ifdef CONFIG_AUDITSYSCALL
245 void putname(struct filename *name)
246 {
247         if (unlikely(!audit_dummy_context()))
248                 return audit_putname(name);
249         final_putname(name);
250 }
251 #endif
252 
253 static int check_acl(struct inode *inode, int mask)
254 {
255 #ifdef CONFIG_FS_POSIX_ACL
256         struct posix_acl *acl;
257 
258         if (mask & MAY_NOT_BLOCK) {
259                 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
260                 if (!acl)
261                         return -EAGAIN;
262                 /* no ->get_acl() calls in RCU mode... */
263                 if (acl == ACL_NOT_CACHED)
264                         return -ECHILD;
265                 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
266         }
267 
268         acl = get_acl(inode, ACL_TYPE_ACCESS);
269         if (IS_ERR(acl))
270                 return PTR_ERR(acl);
271         if (acl) {
272                 int error = posix_acl_permission(inode, acl, mask);
273                 posix_acl_release(acl);
274                 return error;
275         }
276 #endif
277 
278         return -EAGAIN;
279 }
280 
281 /*
282  * This does the basic permission checking
283  */
284 static int acl_permission_check(struct inode *inode, int mask)
285 {
286         unsigned int mode = inode->i_mode;
287 
288         if (likely(uid_eq(current_fsuid(), inode->i_uid)))
289                 mode >>= 6;
290         else {
291                 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
292                         int error = check_acl(inode, mask);
293                         if (error != -EAGAIN)
294                                 return error;
295                 }
296 
297                 if (in_group_p(inode->i_gid))
298                         mode >>= 3;
299         }
300 
301         /*
302          * If the DACs are ok we don't need any capability check.
303          */
304         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
305                 return 0;
306         return -EACCES;
307 }
308 
309 /**
310  * generic_permission -  check for access rights on a Posix-like filesystem
311  * @inode:      inode to check access rights for
312  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
313  *
314  * Used to check for read/write/execute permissions on a file.
315  * We use "fsuid" for this, letting us set arbitrary permissions
316  * for filesystem access without changing the "normal" uids which
317  * are used for other things.
318  *
319  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
320  * request cannot be satisfied (eg. requires blocking or too much complexity).
321  * It would then be called again in ref-walk mode.
322  */
323 int generic_permission(struct inode *inode, int mask)
324 {
325         int ret;
326 
327         /*
328          * Do the basic permission checks.
329          */
330         ret = acl_permission_check(inode, mask);
331         if (ret != -EACCES)
332                 return ret;
333 
334         if (S_ISDIR(inode->i_mode)) {
335                 /* DACs are overridable for directories */
336                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
337                         return 0;
338                 if (!(mask & MAY_WRITE))
339                         if (capable_wrt_inode_uidgid(inode,
340                                                      CAP_DAC_READ_SEARCH))
341                                 return 0;
342                 return -EACCES;
343         }
344         /*
345          * Read/write DACs are always overridable.
346          * Executable DACs are overridable when there is
347          * at least one exec bit set.
348          */
349         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
350                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
351                         return 0;
352 
353         /*
354          * Searching includes executable on directories, else just read.
355          */
356         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
357         if (mask == MAY_READ)
358                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
359                         return 0;
360 
361         return -EACCES;
362 }
363 EXPORT_SYMBOL(generic_permission);
364 
365 /*
366  * We _really_ want to just do "generic_permission()" without
367  * even looking at the inode->i_op values. So we keep a cache
368  * flag in inode->i_opflags, that says "this has not special
369  * permission function, use the fast case".
370  */
371 static inline int do_inode_permission(struct inode *inode, int mask)
372 {
373         if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
374                 if (likely(inode->i_op->permission))
375                         return inode->i_op->permission(inode, mask);
376 
377                 /* This gets set once for the inode lifetime */
378                 spin_lock(&inode->i_lock);
379                 inode->i_opflags |= IOP_FASTPERM;
380                 spin_unlock(&inode->i_lock);
381         }
382         return generic_permission(inode, mask);
383 }
384 
385 /**
386  * __inode_permission - Check for access rights to a given inode
387  * @inode: Inode to check permission on
388  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
389  *
390  * Check for read/write/execute permissions on an inode.
391  *
392  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
393  *
394  * This does not check for a read-only file system.  You probably want
395  * inode_permission().
396  */
397 int __inode_permission(struct inode *inode, int mask)
398 {
399         int retval;
400 
401         if (unlikely(mask & MAY_WRITE)) {
402                 /*
403                  * Nobody gets write access to an immutable file.
404                  */
405                 if (IS_IMMUTABLE(inode))
406                         return -EACCES;
407         }
408 
409         retval = do_inode_permission(inode, mask);
410         if (retval)
411                 return retval;
412 
413         retval = devcgroup_inode_permission(inode, mask);
414         if (retval)
415                 return retval;
416 
417         return security_inode_permission(inode, mask);
418 }
419 EXPORT_SYMBOL(__inode_permission);
420 
421 /**
422  * sb_permission - Check superblock-level permissions
423  * @sb: Superblock of inode to check permission on
424  * @inode: Inode to check permission on
425  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
426  *
427  * Separate out file-system wide checks from inode-specific permission checks.
428  */
429 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
430 {
431         if (unlikely(mask & MAY_WRITE)) {
432                 umode_t mode = inode->i_mode;
433 
434                 /* Nobody gets write access to a read-only fs. */
435                 if ((sb->s_flags & MS_RDONLY) &&
436                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
437                         return -EROFS;
438         }
439         return 0;
440 }
441 
442 /**
443  * inode_permission - Check for access rights to a given inode
444  * @inode: Inode to check permission on
445  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
446  *
447  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
448  * this, letting us set arbitrary permissions for filesystem access without
449  * changing the "normal" UIDs which are used for other things.
450  *
451  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
452  */
453 int inode_permission(struct inode *inode, int mask)
454 {
455         int retval;
456 
457         retval = sb_permission(inode->i_sb, inode, mask);
458         if (retval)
459                 return retval;
460         return __inode_permission(inode, mask);
461 }
462 EXPORT_SYMBOL(inode_permission);
463 
464 /**
465  * path_get - get a reference to a path
466  * @path: path to get the reference to
467  *
468  * Given a path increment the reference count to the dentry and the vfsmount.
469  */
470 void path_get(const struct path *path)
471 {
472         mntget(path->mnt);
473         dget(path->dentry);
474 }
475 EXPORT_SYMBOL(path_get);
476 
477 /**
478  * path_put - put a reference to a path
479  * @path: path to put the reference to
480  *
481  * Given a path decrement the reference count to the dentry and the vfsmount.
482  */
483 void path_put(const struct path *path)
484 {
485         dput(path->dentry);
486         mntput(path->mnt);
487 }
488 EXPORT_SYMBOL(path_put);
489 
490 /*
491  * Path walking has 2 modes, rcu-walk and ref-walk (see
492  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
493  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
494  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
495  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
496  * got stuck, so ref-walk may continue from there. If this is not successful
497  * (eg. a seqcount has changed), then failure is returned and it's up to caller
498  * to restart the path walk from the beginning in ref-walk mode.
499  */
500 
501 /**
502  * unlazy_walk - try to switch to ref-walk mode.
503  * @nd: nameidata pathwalk data
504  * @dentry: child of nd->path.dentry or NULL
505  * Returns: 0 on success, -ECHILD on failure
506  *
507  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
508  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
509  * @nd or NULL.  Must be called from rcu-walk context.
510  */
511 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
512 {
513         struct fs_struct *fs = current->fs;
514         struct dentry *parent = nd->path.dentry;
515 
516         BUG_ON(!(nd->flags & LOOKUP_RCU));
517 
518         /*
519          * After legitimizing the bastards, terminate_walk()
520          * will do the right thing for non-RCU mode, and all our
521          * subsequent exit cases should rcu_read_unlock()
522          * before returning.  Do vfsmount first; if dentry
523          * can't be legitimized, just set nd->path.dentry to NULL
524          * and rely on dput(NULL) being a no-op.
525          */
526         if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
527                 return -ECHILD;
528         nd->flags &= ~LOOKUP_RCU;
529 
530         if (!lockref_get_not_dead(&parent->d_lockref)) {
531                 nd->path.dentry = NULL; 
532                 goto out;
533         }
534 
535         /*
536          * For a negative lookup, the lookup sequence point is the parents
537          * sequence point, and it only needs to revalidate the parent dentry.
538          *
539          * For a positive lookup, we need to move both the parent and the
540          * dentry from the RCU domain to be properly refcounted. And the
541          * sequence number in the dentry validates *both* dentry counters,
542          * since we checked the sequence number of the parent after we got
543          * the child sequence number. So we know the parent must still
544          * be valid if the child sequence number is still valid.
545          */
546         if (!dentry) {
547                 if (read_seqcount_retry(&parent->d_seq, nd->seq))
548                         goto out;
549                 BUG_ON(nd->inode != parent->d_inode);
550         } else {
551                 if (!lockref_get_not_dead(&dentry->d_lockref))
552                         goto out;
553                 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
554                         goto drop_dentry;
555         }
556 
557         /*
558          * Sequence counts matched. Now make sure that the root is
559          * still valid and get it if required.
560          */
561         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
562                 spin_lock(&fs->lock);
563                 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
564                         goto unlock_and_drop_dentry;
565                 path_get(&nd->root);
566                 spin_unlock(&fs->lock);
567         }
568 
569         rcu_read_unlock();
570         return 0;
571 
572 unlock_and_drop_dentry:
573         spin_unlock(&fs->lock);
574 drop_dentry:
575         rcu_read_unlock();
576         dput(dentry);
577         goto drop_root_mnt;
578 out:
579         rcu_read_unlock();
580 drop_root_mnt:
581         if (!(nd->flags & LOOKUP_ROOT))
582                 nd->root.mnt = NULL;
583         return -ECHILD;
584 }
585 
586 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
587 {
588         return dentry->d_op->d_revalidate(dentry, flags);
589 }
590 
591 /**
592  * complete_walk - successful completion of path walk
593  * @nd:  pointer nameidata
594  *
595  * If we had been in RCU mode, drop out of it and legitimize nd->path.
596  * Revalidate the final result, unless we'd already done that during
597  * the path walk or the filesystem doesn't ask for it.  Return 0 on
598  * success, -error on failure.  In case of failure caller does not
599  * need to drop nd->path.
600  */
601 static int complete_walk(struct nameidata *nd)
602 {
603         struct dentry *dentry = nd->path.dentry;
604         int status;
605 
606         if (nd->flags & LOOKUP_RCU) {
607                 nd->flags &= ~LOOKUP_RCU;
608                 if (!(nd->flags & LOOKUP_ROOT))
609                         nd->root.mnt = NULL;
610 
611                 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
612                         rcu_read_unlock();
613                         return -ECHILD;
614                 }
615                 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
616                         rcu_read_unlock();
617                         mntput(nd->path.mnt);
618                         return -ECHILD;
619                 }
620                 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
621                         rcu_read_unlock();
622                         dput(dentry);
623                         mntput(nd->path.mnt);
624                         return -ECHILD;
625                 }
626                 rcu_read_unlock();
627         }
628 
629         if (likely(!(nd->flags & LOOKUP_JUMPED)))
630                 return 0;
631 
632         if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
633                 return 0;
634 
635         status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
636         if (status > 0)
637                 return 0;
638 
639         if (!status)
640                 status = -ESTALE;
641 
642         path_put(&nd->path);
643         return status;
644 }
645 
646 static __always_inline void set_root(struct nameidata *nd)
647 {
648         get_fs_root(current->fs, &nd->root);
649 }
650 
651 static int link_path_walk(const char *, struct nameidata *);
652 
653 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
654 {
655         struct fs_struct *fs = current->fs;
656         unsigned seq, res;
657 
658         do {
659                 seq = read_seqcount_begin(&fs->seq);
660                 nd->root = fs->root;
661                 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
662         } while (read_seqcount_retry(&fs->seq, seq));
663         return res;
664 }
665 
666 static void path_put_conditional(struct path *path, struct nameidata *nd)
667 {
668         dput(path->dentry);
669         if (path->mnt != nd->path.mnt)
670                 mntput(path->mnt);
671 }
672 
673 static inline void path_to_nameidata(const struct path *path,
674                                         struct nameidata *nd)
675 {
676         if (!(nd->flags & LOOKUP_RCU)) {
677                 dput(nd->path.dentry);
678                 if (nd->path.mnt != path->mnt)
679                         mntput(nd->path.mnt);
680         }
681         nd->path.mnt = path->mnt;
682         nd->path.dentry = path->dentry;
683 }
684 
685 /*
686  * Helper to directly jump to a known parsed path from ->follow_link,
687  * caller must have taken a reference to path beforehand.
688  */
689 void nd_jump_link(struct nameidata *nd, struct path *path)
690 {
691         path_put(&nd->path);
692 
693         nd->path = *path;
694         nd->inode = nd->path.dentry->d_inode;
695         nd->flags |= LOOKUP_JUMPED;
696 }
697 
698 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
699 {
700         struct inode *inode = link->dentry->d_inode;
701         if (inode->i_op->put_link)
702                 inode->i_op->put_link(link->dentry, nd, cookie);
703         path_put(link);
704 }
705 
706 int sysctl_protected_symlinks __read_mostly = 0;
707 int sysctl_protected_hardlinks __read_mostly = 0;
708 
709 /**
710  * may_follow_link - Check symlink following for unsafe situations
711  * @link: The path of the symlink
712  * @nd: nameidata pathwalk data
713  *
714  * In the case of the sysctl_protected_symlinks sysctl being enabled,
715  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
716  * in a sticky world-writable directory. This is to protect privileged
717  * processes from failing races against path names that may change out
718  * from under them by way of other users creating malicious symlinks.
719  * It will permit symlinks to be followed only when outside a sticky
720  * world-writable directory, or when the uid of the symlink and follower
721  * match, or when the directory owner matches the symlink's owner.
722  *
723  * Returns 0 if following the symlink is allowed, -ve on error.
724  */
725 static inline int may_follow_link(struct path *link, struct nameidata *nd)
726 {
727         const struct inode *inode;
728         const struct inode *parent;
729 
730         if (!sysctl_protected_symlinks)
731                 return 0;
732 
733         /* Allowed if owner and follower match. */
734         inode = link->dentry->d_inode;
735         if (uid_eq(current_cred()->fsuid, inode->i_uid))
736                 return 0;
737 
738         /* Allowed if parent directory not sticky and world-writable. */
739         parent = nd->path.dentry->d_inode;
740         if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
741                 return 0;
742 
743         /* Allowed if parent directory and link owner match. */
744         if (uid_eq(parent->i_uid, inode->i_uid))
745                 return 0;
746 
747         audit_log_link_denied("follow_link", link);
748         path_put_conditional(link, nd);
749         path_put(&nd->path);
750         return -EACCES;
751 }
752 
753 /**
754  * safe_hardlink_source - Check for safe hardlink conditions
755  * @inode: the source inode to hardlink from
756  *
757  * Return false if at least one of the following conditions:
758  *    - inode is not a regular file
759  *    - inode is setuid
760  *    - inode is setgid and group-exec
761  *    - access failure for read and write
762  *
763  * Otherwise returns true.
764  */
765 static bool safe_hardlink_source(struct inode *inode)
766 {
767         umode_t mode = inode->i_mode;
768 
769         /* Special files should not get pinned to the filesystem. */
770         if (!S_ISREG(mode))
771                 return false;
772 
773         /* Setuid files should not get pinned to the filesystem. */
774         if (mode & S_ISUID)
775                 return false;
776 
777         /* Executable setgid files should not get pinned to the filesystem. */
778         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
779                 return false;
780 
781         /* Hardlinking to unreadable or unwritable sources is dangerous. */
782         if (inode_permission(inode, MAY_READ | MAY_WRITE))
783                 return false;
784 
785         return true;
786 }
787 
788 /**
789  * may_linkat - Check permissions for creating a hardlink
790  * @link: the source to hardlink from
791  *
792  * Block hardlink when all of:
793  *  - sysctl_protected_hardlinks enabled
794  *  - fsuid does not match inode
795  *  - hardlink source is unsafe (see safe_hardlink_source() above)
796  *  - not CAP_FOWNER
797  *
798  * Returns 0 if successful, -ve on error.
799  */
800 static int may_linkat(struct path *link)
801 {
802         const struct cred *cred;
803         struct inode *inode;
804 
805         if (!sysctl_protected_hardlinks)
806                 return 0;
807 
808         cred = current_cred();
809         inode = link->dentry->d_inode;
810 
811         /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
812          * otherwise, it must be a safe source.
813          */
814         if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
815             capable(CAP_FOWNER))
816                 return 0;
817 
818         audit_log_link_denied("linkat", link);
819         return -EPERM;
820 }
821 
822 static __always_inline int
823 follow_link(struct path *link, struct nameidata *nd, void **p)
824 {
825         struct dentry *dentry = link->dentry;
826         int error;
827         char *s;
828 
829         BUG_ON(nd->flags & LOOKUP_RCU);
830 
831         if (link->mnt == nd->path.mnt)
832                 mntget(link->mnt);
833 
834         error = -ELOOP;
835         if (unlikely(current->total_link_count >= 40))
836                 goto out_put_nd_path;
837 
838         cond_resched();
839         current->total_link_count++;
840 
841         touch_atime(link);
842         nd_set_link(nd, NULL);
843 
844         error = security_inode_follow_link(link->dentry, nd);
845         if (error)
846                 goto out_put_nd_path;
847 
848         nd->last_type = LAST_BIND;
849         *p = dentry->d_inode->i_op->follow_link(dentry, nd);
850         error = PTR_ERR(*p);
851         if (IS_ERR(*p))
852                 goto out_put_nd_path;
853 
854         error = 0;
855         s = nd_get_link(nd);
856         if (s) {
857                 if (unlikely(IS_ERR(s))) {
858                         path_put(&nd->path);
859                         put_link(nd, link, *p);
860                         return PTR_ERR(s);
861                 }
862                 if (*s == '/') {
863                         if (!nd->root.mnt)
864                                 set_root(nd);
865                         path_put(&nd->path);
866                         nd->path = nd->root;
867                         path_get(&nd->root);
868                         nd->flags |= LOOKUP_JUMPED;
869                 }
870                 nd->inode = nd->path.dentry->d_inode;
871                 error = link_path_walk(s, nd);
872                 if (unlikely(error))
873                         put_link(nd, link, *p);
874         }
875 
876         return error;
877 
878 out_put_nd_path:
879         *p = NULL;
880         path_put(&nd->path);
881         path_put(link);
882         return error;
883 }
884 
885 static int follow_up_rcu(struct path *path)
886 {
887         struct mount *mnt = real_mount(path->mnt);
888         struct mount *parent;
889         struct dentry *mountpoint;
890 
891         parent = mnt->mnt_parent;
892         if (&parent->mnt == path->mnt)
893                 return 0;
894         mountpoint = mnt->mnt_mountpoint;
895         path->dentry = mountpoint;
896         path->mnt = &parent->mnt;
897         return 1;
898 }
899 
900 /*
901  * follow_up - Find the mountpoint of path's vfsmount
902  *
903  * Given a path, find the mountpoint of its source file system.
904  * Replace @path with the path of the mountpoint in the parent mount.
905  * Up is towards /.
906  *
907  * Return 1 if we went up a level and 0 if we were already at the
908  * root.
909  */
910 int follow_up(struct path *path)
911 {
912         struct mount *mnt = real_mount(path->mnt);
913         struct mount *parent;
914         struct dentry *mountpoint;
915 
916         read_seqlock_excl(&mount_lock);
917         parent = mnt->mnt_parent;
918         if (parent == mnt) {
919                 read_sequnlock_excl(&mount_lock);
920                 return 0;
921         }
922         mntget(&parent->mnt);
923         mountpoint = dget(mnt->mnt_mountpoint);
924         read_sequnlock_excl(&mount_lock);
925         dput(path->dentry);
926         path->dentry = mountpoint;
927         mntput(path->mnt);
928         path->mnt = &parent->mnt;
929         return 1;
930 }
931 EXPORT_SYMBOL(follow_up);
932 
933 /*
934  * Perform an automount
935  * - return -EISDIR to tell follow_managed() to stop and return the path we
936  *   were called with.
937  */
938 static int follow_automount(struct path *path, unsigned flags,
939                             bool *need_mntput)
940 {
941         struct vfsmount *mnt;
942         int err;
943 
944         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
945                 return -EREMOTE;
946 
947         /* We don't want to mount if someone's just doing a stat -
948          * unless they're stat'ing a directory and appended a '/' to
949          * the name.
950          *
951          * We do, however, want to mount if someone wants to open or
952          * create a file of any type under the mountpoint, wants to
953          * traverse through the mountpoint or wants to open the
954          * mounted directory.  Also, autofs may mark negative dentries
955          * as being automount points.  These will need the attentions
956          * of the daemon to instantiate them before they can be used.
957          */
958         if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
959                      LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
960             path->dentry->d_inode)
961                 return -EISDIR;
962 
963         current->total_link_count++;
964         if (current->total_link_count >= 40)
965                 return -ELOOP;
966 
967         mnt = path->dentry->d_op->d_automount(path);
968         if (IS_ERR(mnt)) {
969                 /*
970                  * The filesystem is allowed to return -EISDIR here to indicate
971                  * it doesn't want to automount.  For instance, autofs would do
972                  * this so that its userspace daemon can mount on this dentry.
973                  *
974                  * However, we can only permit this if it's a terminal point in
975                  * the path being looked up; if it wasn't then the remainder of
976                  * the path is inaccessible and we should say so.
977                  */
978                 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
979                         return -EREMOTE;
980                 return PTR_ERR(mnt);
981         }
982 
983         if (!mnt) /* mount collision */
984                 return 0;
985 
986         if (!*need_mntput) {
987                 /* lock_mount() may release path->mnt on error */
988                 mntget(path->mnt);
989                 *need_mntput = true;
990         }
991         err = finish_automount(mnt, path);
992 
993         switch (err) {
994         case -EBUSY:
995                 /* Someone else made a mount here whilst we were busy */
996                 return 0;
997         case 0:
998                 path_put(path);
999                 path->mnt = mnt;
1000                 path->dentry = dget(mnt->mnt_root);
1001                 return 0;
1002         default:
1003                 return err;
1004         }
1005 
1006 }
1007 
1008 /*
1009  * Handle a dentry that is managed in some way.
1010  * - Flagged for transit management (autofs)
1011  * - Flagged as mountpoint
1012  * - Flagged as automount point
1013  *
1014  * This may only be called in refwalk mode.
1015  *
1016  * Serialization is taken care of in namespace.c
1017  */
1018 static int follow_managed(struct path *path, unsigned flags)
1019 {
1020         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1021         unsigned managed;
1022         bool need_mntput = false;
1023         int ret = 0;
1024 
1025         /* Given that we're not holding a lock here, we retain the value in a
1026          * local variable for each dentry as we look at it so that we don't see
1027          * the components of that value change under us */
1028         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1029                managed &= DCACHE_MANAGED_DENTRY,
1030                unlikely(managed != 0)) {
1031                 /* Allow the filesystem to manage the transit without i_mutex
1032                  * being held. */
1033                 if (managed & DCACHE_MANAGE_TRANSIT) {
1034                         BUG_ON(!path->dentry->d_op);
1035                         BUG_ON(!path->dentry->d_op->d_manage);
1036                         ret = path->dentry->d_op->d_manage(path->dentry, false);
1037                         if (ret < 0)
1038                                 break;
1039                 }
1040 
1041                 /* Transit to a mounted filesystem. */
1042                 if (managed & DCACHE_MOUNTED) {
1043                         struct vfsmount *mounted = lookup_mnt(path);
1044                         if (mounted) {
1045                                 dput(path->dentry);
1046                                 if (need_mntput)
1047                                         mntput(path->mnt);
1048                                 path->mnt = mounted;
1049                                 path->dentry = dget(mounted->mnt_root);
1050                                 need_mntput = true;
1051                                 continue;
1052                         }
1053 
1054                         /* Something is mounted on this dentry in another
1055                          * namespace and/or whatever was mounted there in this
1056                          * namespace got unmounted before lookup_mnt() could
1057                          * get it */
1058                 }
1059 
1060                 /* Handle an automount point */
1061                 if (managed & DCACHE_NEED_AUTOMOUNT) {
1062                         ret = follow_automount(path, flags, &need_mntput);
1063                         if (ret < 0)
1064                                 break;
1065                         continue;
1066                 }
1067 
1068                 /* We didn't change the current path point */
1069                 break;
1070         }
1071 
1072         if (need_mntput && path->mnt == mnt)
1073                 mntput(path->mnt);
1074         if (ret == -EISDIR)
1075                 ret = 0;
1076         return ret < 0 ? ret : need_mntput;
1077 }
1078 
1079 int follow_down_one(struct path *path)
1080 {
1081         struct vfsmount *mounted;
1082 
1083         mounted = lookup_mnt(path);
1084         if (mounted) {
1085                 dput(path->dentry);
1086                 mntput(path->mnt);
1087                 path->mnt = mounted;
1088                 path->dentry = dget(mounted->mnt_root);
1089                 return 1;
1090         }
1091         return 0;
1092 }
1093 EXPORT_SYMBOL(follow_down_one);
1094 
1095 static inline int managed_dentry_rcu(struct dentry *dentry)
1096 {
1097         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1098                 dentry->d_op->d_manage(dentry, true) : 0;
1099 }
1100 
1101 /*
1102  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1103  * we meet a managed dentry that would need blocking.
1104  */
1105 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1106                                struct inode **inode)
1107 {
1108         for (;;) {
1109                 struct mount *mounted;
1110                 /*
1111                  * Don't forget we might have a non-mountpoint managed dentry
1112                  * that wants to block transit.
1113                  */
1114                 switch (managed_dentry_rcu(path->dentry)) {
1115                 case -ECHILD:
1116                 default:
1117                         return false;
1118                 case -EISDIR:
1119                         return true;
1120                 case 0:
1121                         break;
1122                 }
1123 
1124                 if (!d_mountpoint(path->dentry))
1125                         return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1126 
1127                 mounted = __lookup_mnt(path->mnt, path->dentry);
1128                 if (!mounted)
1129                         break;
1130                 path->mnt = &mounted->mnt;
1131                 path->dentry = mounted->mnt.mnt_root;
1132                 nd->flags |= LOOKUP_JUMPED;
1133                 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1134                 /*
1135                  * Update the inode too. We don't need to re-check the
1136                  * dentry sequence number here after this d_inode read,
1137                  * because a mount-point is always pinned.
1138                  */
1139                 *inode = path->dentry->d_inode;
1140         }
1141         return !read_seqretry(&mount_lock, nd->m_seq) &&
1142                 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1143 }
1144 
1145 static int follow_dotdot_rcu(struct nameidata *nd)
1146 {
1147         struct inode *inode = nd->inode;
1148         if (!nd->root.mnt)
1149                 set_root_rcu(nd);
1150 
1151         while (1) {
1152                 if (nd->path.dentry == nd->root.dentry &&
1153                     nd->path.mnt == nd->root.mnt) {
1154                         break;
1155                 }
1156                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1157                         struct dentry *old = nd->path.dentry;
1158                         struct dentry *parent = old->d_parent;
1159                         unsigned seq;
1160 
1161                         inode = parent->d_inode;
1162                         seq = read_seqcount_begin(&parent->d_seq);
1163                         if (read_seqcount_retry(&old->d_seq, nd->seq))
1164                                 goto failed;
1165                         nd->path.dentry = parent;
1166                         nd->seq = seq;
1167                         break;
1168                 }
1169                 if (!follow_up_rcu(&nd->path))
1170                         break;
1171                 inode = nd->path.dentry->d_inode;
1172                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1173         }
1174         while (d_mountpoint(nd->path.dentry)) {
1175                 struct mount *mounted;
1176                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1177                 if (!mounted)
1178                         break;
1179                 nd->path.mnt = &mounted->mnt;
1180                 nd->path.dentry = mounted->mnt.mnt_root;
1181                 inode = nd->path.dentry->d_inode;
1182                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1183                 if (read_seqretry(&mount_lock, nd->m_seq))
1184                         goto failed;
1185         }
1186         nd->inode = inode;
1187         return 0;
1188 
1189 failed:
1190         nd->flags &= ~LOOKUP_RCU;
1191         if (!(nd->flags & LOOKUP_ROOT))
1192                 nd->root.mnt = NULL;
1193         rcu_read_unlock();
1194         return -ECHILD;
1195 }
1196 
1197 /*
1198  * Follow down to the covering mount currently visible to userspace.  At each
1199  * point, the filesystem owning that dentry may be queried as to whether the
1200  * caller is permitted to proceed or not.
1201  */
1202 int follow_down(struct path *path)
1203 {
1204         unsigned managed;
1205         int ret;
1206 
1207         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1208                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1209                 /* Allow the filesystem to manage the transit without i_mutex
1210                  * being held.
1211                  *
1212                  * We indicate to the filesystem if someone is trying to mount
1213                  * something here.  This gives autofs the chance to deny anyone
1214                  * other than its daemon the right to mount on its
1215                  * superstructure.
1216                  *
1217                  * The filesystem may sleep at this point.
1218                  */
1219                 if (managed & DCACHE_MANAGE_TRANSIT) {
1220                         BUG_ON(!path->dentry->d_op);
1221                         BUG_ON(!path->dentry->d_op->d_manage);
1222                         ret = path->dentry->d_op->d_manage(
1223                                 path->dentry, false);
1224                         if (ret < 0)
1225                                 return ret == -EISDIR ? 0 : ret;
1226                 }
1227 
1228                 /* Transit to a mounted filesystem. */
1229                 if (managed & DCACHE_MOUNTED) {
1230                         struct vfsmount *mounted = lookup_mnt(path);
1231                         if (!mounted)
1232                                 break;
1233                         dput(path->dentry);
1234                         mntput(path->mnt);
1235                         path->mnt = mounted;
1236                         path->dentry = dget(mounted->mnt_root);
1237                         continue;
1238                 }
1239 
1240                 /* Don't handle automount points here */
1241                 break;
1242         }
1243         return 0;
1244 }
1245 EXPORT_SYMBOL(follow_down);
1246 
1247 /*
1248  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1249  */
1250 static void follow_mount(struct path *path)
1251 {
1252         while (d_mountpoint(path->dentry)) {
1253                 struct vfsmount *mounted = lookup_mnt(path);
1254                 if (!mounted)
1255                         break;
1256                 dput(path->dentry);
1257                 mntput(path->mnt);
1258                 path->mnt = mounted;
1259                 path->dentry = dget(mounted->mnt_root);
1260         }
1261 }
1262 
1263 static void follow_dotdot(struct nameidata *nd)
1264 {
1265         if (!nd->root.mnt)
1266                 set_root(nd);
1267 
1268         while(1) {
1269                 struct dentry *old = nd->path.dentry;
1270 
1271                 if (nd->path.dentry == nd->root.dentry &&
1272                     nd->path.mnt == nd->root.mnt) {
1273                         break;
1274                 }
1275                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1276                         /* rare case of legitimate dget_parent()... */
1277                         nd->path.dentry = dget_parent(nd->path.dentry);
1278                         dput(old);
1279                         break;
1280                 }
1281                 if (!follow_up(&nd->path))
1282                         break;
1283         }
1284         follow_mount(&nd->path);
1285         nd->inode = nd->path.dentry->d_inode;
1286 }
1287 
1288 /*
1289  * This looks up the name in dcache, possibly revalidates the old dentry and
1290  * allocates a new one if not found or not valid.  In the need_lookup argument
1291  * returns whether i_op->lookup is necessary.
1292  *
1293  * dir->d_inode->i_mutex must be held
1294  */
1295 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1296                                     unsigned int flags, bool *need_lookup)
1297 {
1298         struct dentry *dentry;
1299         int error;
1300 
1301         *need_lookup = false;
1302         dentry = d_lookup(dir, name);
1303         if (dentry) {
1304                 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1305                         error = d_revalidate(dentry, flags);
1306                         if (unlikely(error <= 0)) {
1307                                 if (error < 0) {
1308                                         dput(dentry);
1309                                         return ERR_PTR(error);
1310                                 } else {
1311                                         d_invalidate(dentry);
1312                                         dput(dentry);
1313                                         dentry = NULL;
1314                                 }
1315                         }
1316                 }
1317         }
1318 
1319         if (!dentry) {
1320                 dentry = d_alloc(dir, name);
1321                 if (unlikely(!dentry))
1322                         return ERR_PTR(-ENOMEM);
1323 
1324                 *need_lookup = true;
1325         }
1326         return dentry;
1327 }
1328 
1329 /*
1330  * Call i_op->lookup on the dentry.  The dentry must be negative and
1331  * unhashed.
1332  *
1333  * dir->d_inode->i_mutex must be held
1334  */
1335 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1336                                   unsigned int flags)
1337 {
1338         struct dentry *old;
1339 
1340         /* Don't create child dentry for a dead directory. */
1341         if (unlikely(IS_DEADDIR(dir))) {
1342                 dput(dentry);
1343                 return ERR_PTR(-ENOENT);
1344         }
1345 
1346         old = dir->i_op->lookup(dir, dentry, flags);
1347         if (unlikely(old)) {
1348                 dput(dentry);
1349                 dentry = old;
1350         }
1351         return dentry;
1352 }
1353 
1354 static struct dentry *__lookup_hash(struct qstr *name,
1355                 struct dentry *base, unsigned int flags)
1356 {
1357         bool need_lookup;
1358         struct dentry *dentry;
1359 
1360         dentry = lookup_dcache(name, base, flags, &need_lookup);
1361         if (!need_lookup)
1362                 return dentry;
1363 
1364         return lookup_real(base->d_inode, dentry, flags);
1365 }
1366 
1367 /*
1368  *  It's more convoluted than I'd like it to be, but... it's still fairly
1369  *  small and for now I'd prefer to have fast path as straight as possible.
1370  *  It _is_ time-critical.
1371  */
1372 static int lookup_fast(struct nameidata *nd,
1373                        struct path *path, struct inode **inode)
1374 {
1375         struct vfsmount *mnt = nd->path.mnt;
1376         struct dentry *dentry, *parent = nd->path.dentry;
1377         int need_reval = 1;
1378         int status = 1;
1379         int err;
1380 
1381         /*
1382          * Rename seqlock is not required here because in the off chance
1383          * of a false negative due to a concurrent rename, we're going to
1384          * do the non-racy lookup, below.
1385          */
1386         if (nd->flags & LOOKUP_RCU) {
1387                 unsigned seq;
1388                 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1389                 if (!dentry)
1390                         goto unlazy;
1391 
1392                 /*
1393                  * This sequence count validates that the inode matches
1394                  * the dentry name information from lookup.
1395                  */
1396                 *inode = dentry->d_inode;
1397                 if (read_seqcount_retry(&dentry->d_seq, seq))
1398                         return -ECHILD;
1399 
1400                 /*
1401                  * This sequence count validates that the parent had no
1402                  * changes while we did the lookup of the dentry above.
1403                  *
1404                  * The memory barrier in read_seqcount_begin of child is
1405                  *  enough, we can use __read_seqcount_retry here.
1406                  */
1407                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1408                         return -ECHILD;
1409                 nd->seq = seq;
1410 
1411                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1412                         status = d_revalidate(dentry, nd->flags);
1413                         if (unlikely(status <= 0)) {
1414                                 if (status != -ECHILD)
1415                                         need_reval = 0;
1416                                 goto unlazy;
1417                         }
1418                 }
1419                 path->mnt = mnt;
1420                 path->dentry = dentry;
1421                 if (likely(__follow_mount_rcu(nd, path, inode)))
1422                         return 0;
1423 unlazy:
1424                 if (unlazy_walk(nd, dentry))
1425                         return -ECHILD;
1426         } else {
1427                 dentry = __d_lookup(parent, &nd->last);
1428         }
1429 
1430         if (unlikely(!dentry))
1431                 goto need_lookup;
1432 
1433         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1434                 status = d_revalidate(dentry, nd->flags);
1435         if (unlikely(status <= 0)) {
1436                 if (status < 0) {
1437                         dput(dentry);
1438                         return status;
1439                 }
1440                 d_invalidate(dentry);
1441                 dput(dentry);
1442                 goto need_lookup;
1443         }
1444 
1445         path->mnt = mnt;
1446         path->dentry = dentry;
1447         err = follow_managed(path, nd->flags);
1448         if (unlikely(err < 0)) {
1449                 path_put_conditional(path, nd);
1450                 return err;
1451         }
1452         if (err)
1453                 nd->flags |= LOOKUP_JUMPED;
1454         *inode = path->dentry->d_inode;
1455         return 0;
1456 
1457 need_lookup:
1458         return 1;
1459 }
1460 
1461 /* Fast lookup failed, do it the slow way */
1462 static int lookup_slow(struct nameidata *nd, struct path *path)
1463 {
1464         struct dentry *dentry, *parent;
1465         int err;
1466 
1467         parent = nd->path.dentry;
1468         BUG_ON(nd->inode != parent->d_inode);
1469 
1470         mutex_lock(&parent->d_inode->i_mutex);
1471         dentry = __lookup_hash(&nd->last, parent, nd->flags);
1472         mutex_unlock(&parent->d_inode->i_mutex);
1473         if (IS_ERR(dentry))
1474                 return PTR_ERR(dentry);
1475         path->mnt = nd->path.mnt;
1476         path->dentry = dentry;
1477         err = follow_managed(path, nd->flags);
1478         if (unlikely(err < 0)) {
1479                 path_put_conditional(path, nd);
1480                 return err;
1481         }
1482         if (err)
1483                 nd->flags |= LOOKUP_JUMPED;
1484         return 0;
1485 }
1486 
1487 static inline int may_lookup(struct nameidata *nd)
1488 {
1489         if (nd->flags & LOOKUP_RCU) {
1490                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1491                 if (err != -ECHILD)
1492                         return err;
1493                 if (unlazy_walk(nd, NULL))
1494                         return -ECHILD;
1495         }
1496         return inode_permission(nd->inode, MAY_EXEC);
1497 }
1498 
1499 static inline int handle_dots(struct nameidata *nd, int type)
1500 {
1501         if (type == LAST_DOTDOT) {
1502                 if (nd->flags & LOOKUP_RCU) {
1503                         if (follow_dotdot_rcu(nd))
1504                                 return -ECHILD;
1505                 } else
1506                         follow_dotdot(nd);
1507         }
1508         return 0;
1509 }
1510 
1511 static void terminate_walk(struct nameidata *nd)
1512 {
1513         if (!(nd->flags & LOOKUP_RCU)) {
1514                 path_put(&nd->path);
1515         } else {
1516                 nd->flags &= ~LOOKUP_RCU;
1517                 if (!(nd->flags & LOOKUP_ROOT))
1518                         nd->root.mnt = NULL;
1519                 rcu_read_unlock();
1520         }
1521 }
1522 
1523 /*
1524  * Do we need to follow links? We _really_ want to be able
1525  * to do this check without having to look at inode->i_op,
1526  * so we keep a cache of "no, this doesn't need follow_link"
1527  * for the common case.
1528  */
1529 static inline int should_follow_link(struct dentry *dentry, int follow)
1530 {
1531         return unlikely(d_is_symlink(dentry)) ? follow : 0;
1532 }
1533 
1534 static inline int walk_component(struct nameidata *nd, struct path *path,
1535                 int follow)
1536 {
1537         struct inode *inode;
1538         int err;
1539         /*
1540          * "." and ".." are special - ".." especially so because it has
1541          * to be able to know about the current root directory and
1542          * parent relationships.
1543          */
1544         if (unlikely(nd->last_type != LAST_NORM))
1545                 return handle_dots(nd, nd->last_type);
1546         err = lookup_fast(nd, path, &inode);
1547         if (unlikely(err)) {
1548                 if (err < 0)
1549                         goto out_err;
1550 
1551                 err = lookup_slow(nd, path);
1552                 if (err < 0)
1553                         goto out_err;
1554 
1555                 inode = path->dentry->d_inode;
1556         }
1557         err = -ENOENT;
1558         if (!inode || d_is_negative(path->dentry))
1559                 goto out_path_put;
1560 
1561         if (should_follow_link(path->dentry, follow)) {
1562                 if (nd->flags & LOOKUP_RCU) {
1563                         if (unlikely(unlazy_walk(nd, path->dentry))) {
1564                                 err = -ECHILD;
1565                                 goto out_err;
1566                         }
1567                 }
1568                 BUG_ON(inode != path->dentry->d_inode);
1569                 return 1;
1570         }
1571         path_to_nameidata(path, nd);
1572         nd->inode = inode;
1573         return 0;
1574 
1575 out_path_put:
1576         path_to_nameidata(path, nd);
1577 out_err:
1578         terminate_walk(nd);
1579         return err;
1580 }
1581 
1582 /*
1583  * This limits recursive symlink follows to 8, while
1584  * limiting consecutive symlinks to 40.
1585  *
1586  * Without that kind of total limit, nasty chains of consecutive
1587  * symlinks can cause almost arbitrarily long lookups.
1588  */
1589 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1590 {
1591         int res;
1592 
1593         if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1594                 path_put_conditional(path, nd);
1595                 path_put(&nd->path);
1596                 return -ELOOP;
1597         }
1598         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1599 
1600         nd->depth++;
1601         current->link_count++;
1602 
1603         do {
1604                 struct path link = *path;
1605                 void *cookie;
1606 
1607                 res = follow_link(&link, nd, &cookie);
1608                 if (res)
1609                         break;
1610                 res = walk_component(nd, path, LOOKUP_FOLLOW);
1611                 put_link(nd, &link, cookie);
1612         } while (res > 0);
1613 
1614         current->link_count--;
1615         nd->depth--;
1616         return res;
1617 }
1618 
1619 /*
1620  * We can do the critical dentry name comparison and hashing
1621  * operations one word at a time, but we are limited to:
1622  *
1623  * - Architectures with fast unaligned word accesses. We could
1624  *   do a "get_unaligned()" if this helps and is sufficiently
1625  *   fast.
1626  *
1627  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1628  *   do not trap on the (extremely unlikely) case of a page
1629  *   crossing operation.
1630  *
1631  * - Furthermore, we need an efficient 64-bit compile for the
1632  *   64-bit case in order to generate the "number of bytes in
1633  *   the final mask". Again, that could be replaced with a
1634  *   efficient population count instruction or similar.
1635  */
1636 #ifdef CONFIG_DCACHE_WORD_ACCESS
1637 
1638 #include <asm/word-at-a-time.h>
1639 
1640 #ifdef CONFIG_64BIT
1641 
1642 static inline unsigned int fold_hash(unsigned long hash)
1643 {
1644         return hash_64(hash, 32);
1645 }
1646 
1647 #else   /* 32-bit case */
1648 
1649 #define fold_hash(x) (x)
1650 
1651 #endif
1652 
1653 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1654 {
1655         unsigned long a, mask;
1656         unsigned long hash = 0;
1657 
1658         for (;;) {
1659                 a = load_unaligned_zeropad(name);
1660                 if (len < sizeof(unsigned long))
1661                         break;
1662                 hash += a;
1663                 hash *= 9;
1664                 name += sizeof(unsigned long);
1665                 len -= sizeof(unsigned long);
1666                 if (!len)
1667                         goto done;
1668         }
1669         mask = bytemask_from_count(len);
1670         hash += mask & a;
1671 done:
1672         return fold_hash(hash);
1673 }
1674 EXPORT_SYMBOL(full_name_hash);
1675 
1676 /*
1677  * Calculate the length and hash of the path component, and
1678  * return the "hash_len" as the result.
1679  */
1680 static inline u64 hash_name(const char *name)
1681 {
1682         unsigned long a, b, adata, bdata, mask, hash, len;
1683         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1684 
1685         hash = a = 0;
1686         len = -sizeof(unsigned long);
1687         do {
1688                 hash = (hash + a) * 9;
1689                 len += sizeof(unsigned long);
1690                 a = load_unaligned_zeropad(name+len);
1691                 b = a ^ REPEAT_BYTE('/');
1692         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1693 
1694         adata = prep_zero_mask(a, adata, &constants);
1695         bdata = prep_zero_mask(b, bdata, &constants);
1696 
1697         mask = create_zero_mask(adata | bdata);
1698 
1699         hash += a & zero_bytemask(mask);
1700         len += find_zero(mask);
1701         return hashlen_create(fold_hash(hash), len);
1702 }
1703 
1704 #else
1705 
1706 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1707 {
1708         unsigned long hash = init_name_hash();
1709         while (len--)
1710                 hash = partial_name_hash(*name++, hash);
1711         return end_name_hash(hash);
1712 }
1713 EXPORT_SYMBOL(full_name_hash);
1714 
1715 /*
1716  * We know there's a real path component here of at least
1717  * one character.
1718  */
1719 static inline u64 hash_name(const char *name)
1720 {
1721         unsigned long hash = init_name_hash();
1722         unsigned long len = 0, c;
1723 
1724         c = (unsigned char)*name;
1725         do {
1726                 len++;
1727                 hash = partial_name_hash(c, hash);
1728                 c = (unsigned char)name[len];
1729         } while (c && c != '/');
1730         return hashlen_create(end_name_hash(hash), len);
1731 }
1732 
1733 #endif
1734 
1735 /*
1736  * Name resolution.
1737  * This is the basic name resolution function, turning a pathname into
1738  * the final dentry. We expect 'base' to be positive and a directory.
1739  *
1740  * Returns 0 and nd will have valid dentry and mnt on success.
1741  * Returns error and drops reference to input namei data on failure.
1742  */
1743 static int link_path_walk(const char *name, struct nameidata *nd)
1744 {
1745         struct path next;
1746         int err;
1747         
1748         while (*name=='/')
1749                 name++;
1750         if (!*name)
1751                 return 0;
1752 
1753         /* At this point we know we have a real path component. */
1754         for(;;) {
1755                 u64 hash_len;
1756                 int type;
1757 
1758                 err = may_lookup(nd);
1759                 if (err)
1760                         break;
1761 
1762                 hash_len = hash_name(name);
1763 
1764                 type = LAST_NORM;
1765                 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1766                         case 2:
1767                                 if (name[1] == '.') {
1768                                         type = LAST_DOTDOT;
1769                                         nd->flags |= LOOKUP_JUMPED;
1770                                 }
1771                                 break;
1772                         case 1:
1773                                 type = LAST_DOT;
1774                 }
1775                 if (likely(type == LAST_NORM)) {
1776                         struct dentry *parent = nd->path.dentry;
1777                         nd->flags &= ~LOOKUP_JUMPED;
1778                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1779                                 struct qstr this = { { .hash_len = hash_len }, .name = name };
1780                                 err = parent->d_op->d_hash(parent, &this);
1781                                 if (err < 0)
1782                                         break;
1783                                 hash_len = this.hash_len;
1784                                 name = this.name;
1785                         }
1786                 }
1787 
1788                 nd->last.hash_len = hash_len;
1789                 nd->last.name = name;
1790                 nd->last_type = type;
1791 
1792                 name += hashlen_len(hash_len);
1793                 if (!*name)
1794                         return 0;
1795                 /*
1796                  * If it wasn't NUL, we know it was '/'. Skip that
1797                  * slash, and continue until no more slashes.
1798                  */
1799                 do {
1800                         name++;
1801                 } while (unlikely(*name == '/'));
1802                 if (!*name)
1803                         return 0;
1804 
1805                 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1806                 if (err < 0)
1807                         return err;
1808 
1809                 if (err) {
1810                         err = nested_symlink(&next, nd);
1811                         if (err)
1812                                 return err;
1813                 }
1814                 if (!d_can_lookup(nd->path.dentry)) {
1815                         err = -ENOTDIR; 
1816                         break;
1817                 }
1818         }
1819         terminate_walk(nd);
1820         return err;
1821 }
1822 
1823 static int path_init(int dfd, const char *name, unsigned int flags,
1824                      struct nameidata *nd, struct file **fp)
1825 {
1826         int retval = 0;
1827 
1828         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1829         nd->flags = flags | LOOKUP_JUMPED;
1830         nd->depth = 0;
1831         if (flags & LOOKUP_ROOT) {
1832                 struct dentry *root = nd->root.dentry;
1833                 struct inode *inode = root->d_inode;
1834                 if (*name) {
1835                         if (!d_can_lookup(root))
1836                                 return -ENOTDIR;
1837                         retval = inode_permission(inode, MAY_EXEC);
1838                         if (retval)
1839                                 return retval;
1840                 }
1841                 nd->path = nd->root;
1842                 nd->inode = inode;
1843                 if (flags & LOOKUP_RCU) {
1844                         rcu_read_lock();
1845                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1846                         nd->m_seq = read_seqbegin(&mount_lock);
1847                 } else {
1848                         path_get(&nd->path);
1849                 }
1850                 return 0;
1851         }
1852 
1853         nd->root.mnt = NULL;
1854 
1855         nd->m_seq = read_seqbegin(&mount_lock);
1856         if (*name=='/') {
1857                 if (flags & LOOKUP_RCU) {
1858                         rcu_read_lock();
1859                         nd->seq = set_root_rcu(nd);
1860                 } else {
1861                         set_root(nd);
1862                         path_get(&nd->root);
1863                 }
1864                 nd->path = nd->root;
1865         } else if (dfd == AT_FDCWD) {
1866                 if (flags & LOOKUP_RCU) {
1867                         struct fs_struct *fs = current->fs;
1868                         unsigned seq;
1869 
1870                         rcu_read_lock();
1871 
1872                         do {
1873                                 seq = read_seqcount_begin(&fs->seq);
1874                                 nd->path = fs->pwd;
1875                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1876                         } while (read_seqcount_retry(&fs->seq, seq));
1877                 } else {
1878                         get_fs_pwd(current->fs, &nd->path);
1879                 }
1880         } else {
1881                 /* Caller must check execute permissions on the starting path component */
1882                 struct fd f = fdget_raw(dfd);
1883                 struct dentry *dentry;
1884 
1885                 if (!f.file)
1886                         return -EBADF;
1887 
1888                 dentry = f.file->f_path.dentry;
1889 
1890                 if (*name) {
1891                         if (!d_can_lookup(dentry)) {
1892                                 fdput(f);
1893                                 return -ENOTDIR;
1894                         }
1895                 }
1896 
1897                 nd->path = f.file->f_path;
1898                 if (flags & LOOKUP_RCU) {
1899                         if (f.flags & FDPUT_FPUT)
1900                                 *fp = f.file;
1901                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1902                         rcu_read_lock();
1903                 } else {
1904                         path_get(&nd->path);
1905                         fdput(f);
1906                 }
1907         }
1908 
1909         nd->inode = nd->path.dentry->d_inode;
1910         if (!(flags & LOOKUP_RCU))
1911                 return 0;
1912         if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1913                 return 0;
1914         if (!(nd->flags & LOOKUP_ROOT))
1915                 nd->root.mnt = NULL;
1916         rcu_read_unlock();
1917         return -ECHILD;
1918 }
1919 
1920 static inline int lookup_last(struct nameidata *nd, struct path *path)
1921 {
1922         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1923                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1924 
1925         nd->flags &= ~LOOKUP_PARENT;
1926         return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1927 }
1928 
1929 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1930 static int path_lookupat(int dfd, const char *name,
1931                                 unsigned int flags, struct nameidata *nd)
1932 {
1933         struct file *base = NULL;
1934         struct path path;
1935         int err;
1936 
1937         /*
1938          * Path walking is largely split up into 2 different synchronisation
1939          * schemes, rcu-walk and ref-walk (explained in
1940          * Documentation/filesystems/path-lookup.txt). These share much of the
1941          * path walk code, but some things particularly setup, cleanup, and
1942          * following mounts are sufficiently divergent that functions are
1943          * duplicated. Typically there is a function foo(), and its RCU
1944          * analogue, foo_rcu().
1945          *
1946          * -ECHILD is the error number of choice (just to avoid clashes) that
1947          * is returned if some aspect of an rcu-walk fails. Such an error must
1948          * be handled by restarting a traditional ref-walk (which will always
1949          * be able to complete).
1950          */
1951         err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1952 
1953         if (unlikely(err))
1954                 goto out;
1955 
1956         current->total_link_count = 0;
1957         err = link_path_walk(name, nd);
1958 
1959         if (!err && !(flags & LOOKUP_PARENT)) {
1960                 err = lookup_last(nd, &path);
1961                 while (err > 0) {
1962                         void *cookie;
1963                         struct path link = path;
1964                         err = may_follow_link(&link, nd);
1965                         if (unlikely(err))
1966                                 break;
1967                         nd->flags |= LOOKUP_PARENT;
1968                         err = follow_link(&link, nd, &cookie);
1969                         if (err)
1970                                 break;
1971                         err = lookup_last(nd, &path);
1972                         put_link(nd, &link, cookie);
1973                 }
1974         }
1975 
1976         if (!err)
1977                 err = complete_walk(nd);
1978 
1979         if (!err && nd->flags & LOOKUP_DIRECTORY) {
1980                 if (!d_can_lookup(nd->path.dentry)) {
1981                         path_put(&nd->path);
1982                         err = -ENOTDIR;
1983                 }
1984         }
1985 
1986 out:
1987         if (base)
1988                 fput(base);
1989 
1990         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1991                 path_put(&nd->root);
1992                 nd->root.mnt = NULL;
1993         }
1994         return err;
1995 }
1996 
1997 static int filename_lookup(int dfd, struct filename *name,
1998                                 unsigned int flags, struct nameidata *nd)
1999 {
2000         int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2001         if (unlikely(retval == -ECHILD))
2002                 retval = path_lookupat(dfd, name->name, flags, nd);
2003         if (unlikely(retval == -ESTALE))
2004                 retval = path_lookupat(dfd, name->name,
2005                                                 flags | LOOKUP_REVAL, nd);
2006 
2007         if (likely(!retval))
2008                 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2009         return retval;
2010 }
2011 
2012 static int do_path_lookup(int dfd, const char *name,
2013                                 unsigned int flags, struct nameidata *nd)
2014 {
2015         struct filename filename = { .name = name };
2016 
2017         return filename_lookup(dfd, &filename, flags, nd);
2018 }
2019 
2020 /* does lookup, returns the object with parent locked */
2021 struct dentry *kern_path_locked(const char *name, struct path *path)
2022 {
2023         struct nameidata nd;
2024         struct dentry *d;
2025         int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2026         if (err)
2027                 return ERR_PTR(err);
2028         if (nd.last_type != LAST_NORM) {
2029                 path_put(&nd.path);
2030                 return ERR_PTR(-EINVAL);
2031         }
2032         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2033         d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2034         if (IS_ERR(d)) {
2035                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2036                 path_put(&nd.path);
2037                 return d;
2038         }
2039         *path = nd.path;
2040         return d;
2041 }
2042 
2043 int kern_path(const char *name, unsigned int flags, struct path *path)
2044 {
2045         struct nameidata nd;
2046         int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2047         if (!res)
2048                 *path = nd.path;
2049         return res;
2050 }
2051 EXPORT_SYMBOL(kern_path);
2052 
2053 /**
2054  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2055  * @dentry:  pointer to dentry of the base directory
2056  * @mnt: pointer to vfs mount of the base directory
2057  * @name: pointer to file name
2058  * @flags: lookup flags
2059  * @path: pointer to struct path to fill
2060  */
2061 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2062                     const char *name, unsigned int flags,
2063                     struct path *path)
2064 {
2065         struct nameidata nd;
2066         int err;
2067         nd.root.dentry = dentry;
2068         nd.root.mnt = mnt;
2069         BUG_ON(flags & LOOKUP_PARENT);
2070         /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2071         err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2072         if (!err)
2073                 *path = nd.path;
2074         return err;
2075 }
2076 EXPORT_SYMBOL(vfs_path_lookup);
2077 
2078 /*
2079  * Restricted form of lookup. Doesn't follow links, single-component only,
2080  * needs parent already locked. Doesn't follow mounts.
2081  * SMP-safe.
2082  */
2083 static struct dentry *lookup_hash(struct nameidata *nd)
2084 {
2085         return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2086 }
2087 
2088 /**
2089  * lookup_one_len - filesystem helper to lookup single pathname component
2090  * @name:       pathname component to lookup
2091  * @base:       base directory to lookup from
2092  * @len:        maximum length @len should be interpreted to
2093  *
2094  * Note that this routine is purely a helper for filesystem usage and should
2095  * not be called by generic code.  Also note that by using this function the
2096  * nameidata argument is passed to the filesystem methods and a filesystem
2097  * using this helper needs to be prepared for that.
2098  */
2099 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2100 {
2101         struct qstr this;
2102         unsigned int c;
2103         int err;
2104 
2105         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2106 
2107         this.name = name;
2108         this.len = len;
2109         this.hash = full_name_hash(name, len);
2110         if (!len)
2111                 return ERR_PTR(-EACCES);
2112 
2113         if (unlikely(name[0] == '.')) {
2114                 if (len < 2 || (len == 2 && name[1] == '.'))
2115                         return ERR_PTR(-EACCES);
2116         }
2117 
2118         while (len--) {
2119                 c = *(const unsigned char *)name++;
2120                 if (c == '/' || c == '\0')
2121                         return ERR_PTR(-EACCES);
2122         }
2123         /*
2124          * See if the low-level filesystem might want
2125          * to use its own hash..
2126          */
2127         if (base->d_flags & DCACHE_OP_HASH) {
2128                 int err = base->d_op->d_hash(base, &this);
2129                 if (err < 0)
2130                         return ERR_PTR(err);
2131         }
2132 
2133         err = inode_permission(base->d_inode, MAY_EXEC);
2134         if (err)
2135                 return ERR_PTR(err);
2136 
2137         return __lookup_hash(&this, base, 0);
2138 }
2139 EXPORT_SYMBOL(lookup_one_len);
2140 
2141 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2142                  struct path *path, int *empty)
2143 {
2144         struct nameidata nd;
2145         struct filename *tmp = getname_flags(name, flags, empty);
2146         int err = PTR_ERR(tmp);
2147         if (!IS_ERR(tmp)) {
2148 
2149                 BUG_ON(flags & LOOKUP_PARENT);
2150 
2151                 err = filename_lookup(dfd, tmp, flags, &nd);
2152                 putname(tmp);
2153                 if (!err)
2154                         *path = nd.path;
2155         }
2156         return err;
2157 }
2158 
2159 int user_path_at(int dfd, const char __user *name, unsigned flags,
2160                  struct path *path)
2161 {
2162         return user_path_at_empty(dfd, name, flags, path, NULL);
2163 }
2164 EXPORT_SYMBOL(user_path_at);
2165 
2166 /*
2167  * NB: most callers don't do anything directly with the reference to the
2168  *     to struct filename, but the nd->last pointer points into the name string
2169  *     allocated by getname. So we must hold the reference to it until all
2170  *     path-walking is complete.
2171  */
2172 static struct filename *
2173 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2174                  unsigned int flags)
2175 {
2176         struct filename *s = getname(path);
2177         int error;
2178 
2179         /* only LOOKUP_REVAL is allowed in extra flags */
2180         flags &= LOOKUP_REVAL;
2181 
2182         if (IS_ERR(s))
2183                 return s;
2184 
2185         error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2186         if (error) {
2187                 putname(s);
2188                 return ERR_PTR(error);
2189         }
2190 
2191         return s;
2192 }
2193 
2194 /**
2195  * mountpoint_last - look up last component for umount
2196  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2197  * @path: pointer to container for result
2198  *
2199  * This is a special lookup_last function just for umount. In this case, we
2200  * need to resolve the path without doing any revalidation.
2201  *
2202  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2203  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2204  * in almost all cases, this lookup will be served out of the dcache. The only
2205  * cases where it won't are if nd->last refers to a symlink or the path is
2206  * bogus and it doesn't exist.
2207  *
2208  * Returns:
2209  * -error: if there was an error during lookup. This includes -ENOENT if the
2210  *         lookup found a negative dentry. The nd->path reference will also be
2211  *         put in this case.
2212  *
2213  * 0:      if we successfully resolved nd->path and found it to not to be a
2214  *         symlink that needs to be followed. "path" will also be populated.
2215  *         The nd->path reference will also be put.
2216  *
2217  * 1:      if we successfully resolved nd->last and found it to be a symlink
2218  *         that needs to be followed. "path" will be populated with the path
2219  *         to the link, and nd->path will *not* be put.
2220  */
2221 static int
2222 mountpoint_last(struct nameidata *nd, struct path *path)
2223 {
2224         int error = 0;
2225         struct dentry *dentry;
2226         struct dentry *dir = nd->path.dentry;
2227 
2228         /* If we're in rcuwalk, drop out of it to handle last component */
2229         if (nd->flags & LOOKUP_RCU) {
2230                 if (unlazy_walk(nd, NULL)) {
2231                         error = -ECHILD;
2232                         goto out;
2233                 }
2234         }
2235 
2236         nd->flags &= ~LOOKUP_PARENT;
2237 
2238         if (unlikely(nd->last_type != LAST_NORM)) {
2239                 error = handle_dots(nd, nd->last_type);
2240                 if (error)
2241                         goto out;
2242                 dentry = dget(nd->path.dentry);
2243                 goto done;
2244         }
2245 
2246         mutex_lock(&dir->d_inode->i_mutex);
2247         dentry = d_lookup(dir, &nd->last);
2248         if (!dentry) {
2249                 /*
2250                  * No cached dentry. Mounted dentries are pinned in the cache,
2251                  * so that means that this dentry is probably a symlink or the
2252                  * path doesn't actually point to a mounted dentry.
2253                  */
2254                 dentry = d_alloc(dir, &nd->last);
2255                 if (!dentry) {
2256                         error = -ENOMEM;
2257                         mutex_unlock(&dir->d_inode->i_mutex);
2258                         goto out;
2259                 }
2260                 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2261                 error = PTR_ERR(dentry);
2262                 if (IS_ERR(dentry)) {
2263                         mutex_unlock(&dir->d_inode->i_mutex);
2264                         goto out;
2265                 }
2266         }
2267         mutex_unlock(&dir->d_inode->i_mutex);
2268 
2269 done:
2270         if (!dentry->d_inode || d_is_negative(dentry)) {
2271                 error = -ENOENT;
2272                 dput(dentry);
2273                 goto out;
2274         }
2275         path->dentry = dentry;
2276         path->mnt = nd->path.mnt;
2277         if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2278                 return 1;
2279         mntget(path->mnt);
2280         follow_mount(path);
2281         error = 0;
2282 out:
2283         terminate_walk(nd);
2284         return error;
2285 }
2286 
2287 /**
2288  * path_mountpoint - look up a path to be umounted
2289  * @dfd:        directory file descriptor to start walk from
2290  * @name:       full pathname to walk
2291  * @path:       pointer to container for result
2292  * @flags:      lookup flags
2293  *
2294  * Look up the given name, but don't attempt to revalidate the last component.
2295  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2296  */
2297 static int
2298 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2299 {
2300         struct file *base = NULL;
2301         struct nameidata nd;
2302         int err;
2303 
2304         err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2305         if (unlikely(err))
2306                 goto out;
2307 
2308         current->total_link_count = 0;
2309         err = link_path_walk(name, &nd);
2310         if (err)
2311                 goto out;
2312 
2313         err = mountpoint_last(&nd, path);
2314         while (err > 0) {
2315                 void *cookie;
2316                 struct path link = *path;
2317                 err = may_follow_link(&link, &nd);
2318                 if (unlikely(err))
2319                         break;
2320                 nd.flags |= LOOKUP_PARENT;
2321                 err = follow_link(&link, &nd, &cookie);
2322                 if (err)
2323                         break;
2324                 err = mountpoint_last(&nd, path);
2325                 put_link(&nd, &link, cookie);
2326         }
2327 out:
2328         if (base)
2329                 fput(base);
2330 
2331         if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2332                 path_put(&nd.root);
2333 
2334         return err;
2335 }
2336 
2337 static int
2338 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2339                         unsigned int flags)
2340 {
2341         int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2342         if (unlikely(error == -ECHILD))
2343                 error = path_mountpoint(dfd, s->name, path, flags);
2344         if (unlikely(error == -ESTALE))
2345                 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2346         if (likely(!error))
2347                 audit_inode(s, path->dentry, 0);
2348         return error;
2349 }
2350 
2351 /**
2352  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2353  * @dfd:        directory file descriptor
2354  * @name:       pathname from userland
2355  * @flags:      lookup flags
2356  * @path:       pointer to container to hold result
2357  *
2358  * A umount is a special case for path walking. We're not actually interested
2359  * in the inode in this situation, and ESTALE errors can be a problem. We
2360  * simply want track down the dentry and vfsmount attached at the mountpoint
2361  * and avoid revalidating the last component.
2362  *
2363  * Returns 0 and populates "path" on success.
2364  */
2365 int
2366 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2367                         struct path *path)
2368 {
2369         struct filename *s = getname(name);
2370         int error;
2371         if (IS_ERR(s))
2372                 return PTR_ERR(s);
2373         error = filename_mountpoint(dfd, s, path, flags);
2374         putname(s);
2375         return error;
2376 }
2377 
2378 int
2379 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2380                         unsigned int flags)
2381 {
2382         struct filename s = {.name = name};
2383         return filename_mountpoint(dfd, &s, path, flags);
2384 }
2385 EXPORT_SYMBOL(kern_path_mountpoint);
2386 
2387 int __check_sticky(struct inode *dir, struct inode *inode)
2388 {
2389         kuid_t fsuid = current_fsuid();
2390 
2391         if (uid_eq(inode->i_uid, fsuid))
2392                 return 0;
2393         if (uid_eq(dir->i_uid, fsuid))
2394                 return 0;
2395         return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2396 }
2397 EXPORT_SYMBOL(__check_sticky);
2398 
2399 /*
2400  *      Check whether we can remove a link victim from directory dir, check
2401  *  whether the type of victim is right.
2402  *  1. We can't do it if dir is read-only (done in permission())
2403  *  2. We should have write and exec permissions on dir
2404  *  3. We can't remove anything from append-only dir
2405  *  4. We can't do anything with immutable dir (done in permission())
2406  *  5. If the sticky bit on dir is set we should either
2407  *      a. be owner of dir, or
2408  *      b. be owner of victim, or
2409  *      c. have CAP_FOWNER capability
2410  *  6. If the victim is append-only or immutable we can't do antyhing with
2411  *     links pointing to it.
2412  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2413  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2414  *  9. We can't remove a root or mountpoint.
2415  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2416  *     nfs_async_unlink().
2417  */
2418 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2419 {
2420         struct inode *inode = victim->d_inode;
2421         int error;
2422 
2423         if (d_is_negative(victim))
2424                 return -ENOENT;
2425         BUG_ON(!inode);
2426 
2427         BUG_ON(victim->d_parent->d_inode != dir);
2428         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2429 
2430         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2431         if (error)
2432                 return error;
2433         if (IS_APPEND(dir))
2434                 return -EPERM;
2435 
2436         if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2437             IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2438                 return -EPERM;
2439         if (isdir) {
2440                 if (!d_is_dir(victim))
2441                         return -ENOTDIR;
2442                 if (IS_ROOT(victim))
2443                         return -EBUSY;
2444         } else if (d_is_dir(victim))
2445                 return -EISDIR;
2446         if (IS_DEADDIR(dir))
2447                 return -ENOENT;
2448         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2449                 return -EBUSY;
2450         return 0;
2451 }
2452 
2453 /*      Check whether we can create an object with dentry child in directory
2454  *  dir.
2455  *  1. We can't do it if child already exists (open has special treatment for
2456  *     this case, but since we are inlined it's OK)
2457  *  2. We can't do it if dir is read-only (done in permission())
2458  *  3. We should have write and exec permissions on dir
2459  *  4. We can't do it if dir is immutable (done in permission())
2460  */
2461 static inline int may_create(struct inode *dir, struct dentry *child)
2462 {
2463         audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2464         if (child->d_inode)
2465                 return -EEXIST;
2466         if (IS_DEADDIR(dir))
2467                 return -ENOENT;
2468         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2469 }
2470 
2471 /*
2472  * p1 and p2 should be directories on the same fs.
2473  */
2474 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2475 {
2476         struct dentry *p;
2477 
2478         if (p1 == p2) {
2479                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2480                 return NULL;
2481         }
2482 
2483         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2484 
2485         p = d_ancestor(p2, p1);
2486         if (p) {
2487                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2488                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2489                 return p;
2490         }
2491 
2492         p = d_ancestor(p1, p2);
2493         if (p) {
2494                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2495                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2496                 return p;
2497         }
2498 
2499         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2500         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2501         return NULL;
2502 }
2503 EXPORT_SYMBOL(lock_rename);
2504 
2505 void unlock_rename(struct dentry *p1, struct dentry *p2)
2506 {
2507         mutex_unlock(&p1->d_inode->i_mutex);
2508         if (p1 != p2) {
2509                 mutex_unlock(&p2->d_inode->i_mutex);
2510                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2511         }
2512 }
2513 EXPORT_SYMBOL(unlock_rename);
2514 
2515 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2516                 bool want_excl)
2517 {
2518         int error = may_create(dir, dentry);
2519         if (error)
2520                 return error;
2521 
2522         if (!dir->i_op->create)
2523                 return -EACCES; /* shouldn't it be ENOSYS? */
2524         mode &= S_IALLUGO;
2525         mode |= S_IFREG;
2526         error = security_inode_create(dir, dentry, mode);
2527         if (error)
2528                 return error;
2529         error = dir->i_op->create(dir, dentry, mode, want_excl);
2530         if (!error)
2531                 fsnotify_create(dir, dentry);
2532         return error;
2533 }
2534 EXPORT_SYMBOL(vfs_create);
2535 
2536 static int may_open(struct path *path, int acc_mode, int flag)
2537 {
2538         struct dentry *dentry = path->dentry;
2539         struct inode *inode = dentry->d_inode;
2540         int error;
2541 
2542         /* O_PATH? */
2543         if (!acc_mode)
2544                 return 0;
2545 
2546         if (!inode)
2547                 return -ENOENT;
2548 
2549         switch (inode->i_mode & S_IFMT) {
2550         case S_IFLNK:
2551                 return -ELOOP;
2552         case S_IFDIR:
2553                 if (acc_mode & MAY_WRITE)
2554                         return -EISDIR;
2555                 break;
2556         case S_IFBLK:
2557         case S_IFCHR:
2558                 if (path->mnt->mnt_flags & MNT_NODEV)
2559                         return -EACCES;
2560                 /*FALLTHRU*/
2561         case S_IFIFO:
2562         case S_IFSOCK:
2563                 flag &= ~O_TRUNC;
2564                 break;
2565         }
2566 
2567         error = inode_permission(inode, acc_mode);
2568         if (error)
2569                 return error;
2570 
2571         /*
2572          * An append-only file must be opened in append mode for writing.
2573          */
2574         if (IS_APPEND(inode)) {
2575                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2576                         return -EPERM;
2577                 if (flag & O_TRUNC)
2578                         return -EPERM;
2579         }
2580 
2581         /* O_NOATIME can only be set by the owner or superuser */
2582         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2583                 return -EPERM;
2584 
2585         return 0;
2586 }
2587 
2588 static int handle_truncate(struct file *filp)
2589 {
2590         struct path *path = &filp->f_path;
2591         struct inode *inode = path->dentry->d_inode;
2592         int error = get_write_access(inode);
2593         if (error)
2594                 return error;
2595         /*
2596          * Refuse to truncate files with mandatory locks held on them.
2597          */
2598         error = locks_verify_locked(filp);
2599         if (!error)
2600                 error = security_path_truncate(path);
2601         if (!error) {
2602                 error = do_truncate(path->dentry, 0,
2603                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2604                                     filp);
2605         }
2606         put_write_access(inode);
2607         return error;
2608 }
2609 
2610 static inline int open_to_namei_flags(int flag)
2611 {
2612         if ((flag & O_ACCMODE) == 3)
2613                 flag--;
2614         return flag;
2615 }
2616 
2617 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2618 {
2619         int error = security_path_mknod(dir, dentry, mode, 0);
2620         if (error)
2621                 return error;
2622 
2623         error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2624         if (error)
2625                 return error;
2626 
2627         return security_inode_create(dir->dentry->d_inode, dentry, mode);
2628 }
2629 
2630 /*
2631  * Attempt to atomically look up, create and open a file from a negative
2632  * dentry.
2633  *
2634  * Returns 0 if successful.  The file will have been created and attached to
2635  * @file by the filesystem calling finish_open().
2636  *
2637  * Returns 1 if the file was looked up only or didn't need creating.  The
2638  * caller will need to perform the open themselves.  @path will have been
2639  * updated to point to the new dentry.  This may be negative.
2640  *
2641  * Returns an error code otherwise.
2642  */
2643 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2644                         struct path *path, struct file *file,
2645                         const struct open_flags *op,
2646                         bool got_write, bool need_lookup,
2647                         int *opened)
2648 {
2649         struct inode *dir =  nd->path.dentry->d_inode;
2650         unsigned open_flag = open_to_namei_flags(op->open_flag);
2651         umode_t mode;
2652         int error;
2653         int acc_mode;
2654         int create_error = 0;
2655         struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2656         bool excl;
2657 
2658         BUG_ON(dentry->d_inode);
2659 
2660         /* Don't create child dentry for a dead directory. */
2661         if (unlikely(IS_DEADDIR(dir))) {
2662                 error = -ENOENT;
2663                 goto out;
2664         }
2665 
2666         mode = op->mode;
2667         if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2668                 mode &= ~current_umask();
2669 
2670         excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2671         if (excl)
2672                 open_flag &= ~O_TRUNC;
2673 
2674         /*
2675          * Checking write permission is tricky, bacuse we don't know if we are
2676          * going to actually need it: O_CREAT opens should work as long as the
2677          * file exists.  But checking existence breaks atomicity.  The trick is
2678          * to check access and if not granted clear O_CREAT from the flags.
2679          *
2680          * Another problem is returing the "right" error value (e.g. for an
2681          * O_EXCL open we want to return EEXIST not EROFS).
2682          */
2683         if (((open_flag & (O_CREAT | O_TRUNC)) ||
2684             (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2685                 if (!(open_flag & O_CREAT)) {
2686                         /*
2687                          * No O_CREATE -> atomicity not a requirement -> fall
2688                          * back to lookup + open
2689                          */
2690                         goto no_open;
2691                 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2692                         /* Fall back and fail with the right error */
2693                         create_error = -EROFS;
2694                         goto no_open;
2695                 } else {
2696                         /* No side effects, safe to clear O_CREAT */
2697                         create_error = -EROFS;
2698                         open_flag &= ~O_CREAT;
2699                 }
2700         }
2701 
2702         if (open_flag & O_CREAT) {
2703                 error = may_o_create(&nd->path, dentry, mode);
2704                 if (error) {
2705                         create_error = error;
2706                         if (open_flag & O_EXCL)
2707                                 goto no_open;
2708                         open_flag &= ~O_CREAT;
2709                 }
2710         }
2711 
2712         if (nd->flags & LOOKUP_DIRECTORY)
2713                 open_flag |= O_DIRECTORY;
2714 
2715         file->f_path.dentry = DENTRY_NOT_SET;
2716         file->f_path.mnt = nd->path.mnt;
2717         error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2718                                       opened);
2719         if (error < 0) {
2720                 if (create_error && error == -ENOENT)
2721                         error = create_error;
2722                 goto out;
2723         }
2724 
2725         if (error) {    /* returned 1, that is */
2726                 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2727                         error = -EIO;
2728                         goto out;
2729                 }
2730                 if (file->f_path.dentry) {
2731                         dput(dentry);
2732                         dentry = file->f_path.dentry;
2733                 }
2734                 if (*opened & FILE_CREATED)
2735                         fsnotify_create(dir, dentry);
2736                 if (!dentry->d_inode) {
2737                         WARN_ON(*opened & FILE_CREATED);
2738                         if (create_error) {
2739                                 error = create_error;
2740                                 goto out;
2741                         }
2742                 } else {
2743                         if (excl && !(*opened & FILE_CREATED)) {
2744                                 error = -EEXIST;
2745                                 goto out;
2746                         }
2747                 }
2748                 goto looked_up;
2749         }
2750 
2751         /*
2752          * We didn't have the inode before the open, so check open permission
2753          * here.
2754          */
2755         acc_mode = op->acc_mode;
2756         if (*opened & FILE_CREATED) {
2757                 WARN_ON(!(open_flag & O_CREAT));
2758                 fsnotify_create(dir, dentry);
2759                 acc_mode = MAY_OPEN;
2760         }
2761         error = may_open(&file->f_path, acc_mode, open_flag);
2762         if (error)
2763                 fput(file);
2764 
2765 out:
2766         dput(dentry);
2767         return error;
2768 
2769 no_open:
2770         if (need_lookup) {
2771                 dentry = lookup_real(dir, dentry, nd->flags);
2772                 if (IS_ERR(dentry))
2773                         return PTR_ERR(dentry);
2774 
2775                 if (create_error) {
2776                         int open_flag = op->open_flag;
2777 
2778                         error = create_error;
2779                         if ((open_flag & O_EXCL)) {
2780                                 if (!dentry->d_inode)
2781                                         goto out;
2782                         } else if (!dentry->d_inode) {
2783                                 goto out;
2784                         } else if ((open_flag & O_TRUNC) &&
2785                                    S_ISREG(dentry->d_inode->i_mode)) {
2786                                 goto out;
2787                         }
2788                         /* will fail later, go on to get the right error */
2789                 }
2790         }
2791 looked_up:
2792         path->dentry = dentry;
2793         path->mnt = nd->path.mnt;
2794         return 1;
2795 }
2796 
2797 /*
2798  * Look up and maybe create and open the last component.
2799  *
2800  * Must be called with i_mutex held on parent.
2801  *
2802  * Returns 0 if the file was successfully atomically created (if necessary) and
2803  * opened.  In this case the file will be returned attached to @file.
2804  *
2805  * Returns 1 if the file was not completely opened at this time, though lookups
2806  * and creations will have been performed and the dentry returned in @path will
2807  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2808  * specified then a negative dentry may be returned.
2809  *
2810  * An error code is returned otherwise.
2811  *
2812  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2813  * cleared otherwise prior to returning.
2814  */
2815 static int lookup_open(struct nameidata *nd, struct path *path,
2816                         struct file *file,
2817                         const struct open_flags *op,
2818                         bool got_write, int *opened)
2819 {
2820         struct dentry *dir = nd->path.dentry;
2821         struct inode *dir_inode = dir->d_inode;
2822         struct dentry *dentry;
2823         int error;
2824         bool need_lookup;
2825 
2826         *opened &= ~FILE_CREATED;
2827         dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2828         if (IS_ERR(dentry))
2829                 return PTR_ERR(dentry);
2830 
2831         /* Cached positive dentry: will open in f_op->open */
2832         if (!need_lookup && dentry->d_inode)
2833                 goto out_no_open;
2834 
2835         if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2836                 return atomic_open(nd, dentry, path, file, op, got_write,
2837                                    need_lookup, opened);
2838         }
2839 
2840         if (need_lookup) {
2841                 BUG_ON(dentry->d_inode);
2842 
2843                 dentry = lookup_real(dir_inode, dentry, nd->flags);
2844                 if (IS_ERR(dentry))
2845                         return PTR_ERR(dentry);
2846         }
2847 
2848         /* Negative dentry, just create the file */
2849         if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2850                 umode_t mode = op->mode;
2851                 if (!IS_POSIXACL(dir->d_inode))
2852                         mode &= ~current_umask();
2853                 /*
2854                  * This write is needed to ensure that a
2855                  * rw->ro transition does not occur between
2856                  * the time when the file is created and when
2857                  * a permanent write count is taken through
2858                  * the 'struct file' in finish_open().
2859                  */
2860                 if (!got_write) {
2861                         error = -EROFS;
2862                         goto out_dput;
2863                 }
2864                 *opened |= FILE_CREATED;
2865                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2866                 if (error)
2867                         goto out_dput;
2868                 error = vfs_create(dir->d_inode, dentry, mode,
2869                                    nd->flags & LOOKUP_EXCL);
2870                 if (error)
2871                         goto out_dput;
2872         }
2873 out_no_open:
2874         path->dentry = dentry;
2875         path->mnt = nd->path.mnt;
2876         return 1;
2877 
2878 out_dput:
2879         dput(dentry);
2880         return error;
2881 }
2882 
2883 /*
2884  * Handle the last step of open()
2885  */
2886 static int do_last(struct nameidata *nd, struct path *path,
2887                    struct file *file, const struct open_flags *op,
2888                    int *opened, struct filename *name)
2889 {
2890         struct dentry *dir = nd->path.dentry;
2891         int open_flag = op->open_flag;
2892         bool will_truncate = (open_flag & O_TRUNC) != 0;
2893         bool got_write = false;
2894         int acc_mode = op->acc_mode;
2895         struct inode *inode;
2896         bool symlink_ok = false;
2897         struct path save_parent = { .dentry = NULL, .mnt = NULL };
2898         bool retried = false;
2899         int error;
2900 
2901         nd->flags &= ~LOOKUP_PARENT;
2902         nd->flags |= op->intent;
2903 
2904         if (nd->last_type != LAST_NORM) {
2905                 error = handle_dots(nd, nd->last_type);
2906                 if (error)
2907                         return error;
2908                 goto finish_open;
2909         }
2910 
2911         if (!(open_flag & O_CREAT)) {
2912                 if (nd->last.name[nd->last.len])
2913                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2914                 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2915                         symlink_ok = true;
2916                 /* we _can_ be in RCU mode here */
2917                 error = lookup_fast(nd, path, &inode);
2918                 if (likely(!error))
2919                         goto finish_lookup;
2920 
2921                 if (error < 0)
2922                         goto out;
2923 
2924                 BUG_ON(nd->inode != dir->d_inode);
2925         } else {
2926                 /* create side of things */
2927                 /*
2928                  * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2929                  * has been cleared when we got to the last component we are
2930                  * about to look up
2931                  */
2932                 error = complete_walk(nd);
2933                 if (error)
2934                         return error;
2935 
2936                 audit_inode(name, dir, LOOKUP_PARENT);
2937                 error = -EISDIR;
2938                 /* trailing slashes? */
2939                 if (nd->last.name[nd->last.len])
2940                         goto out;
2941         }
2942 
2943 retry_lookup:
2944         if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2945                 error = mnt_want_write(nd->path.mnt);
2946                 if (!error)
2947                         got_write = true;
2948                 /*
2949                  * do _not_ fail yet - we might not need that or fail with
2950                  * a different error; let lookup_open() decide; we'll be
2951                  * dropping this one anyway.
2952                  */
2953         }
2954         mutex_lock(&dir->d_inode->i_mutex);
2955         error = lookup_open(nd, path, file, op, got_write, opened);
2956         mutex_unlock(&dir->d_inode->i_mutex);
2957 
2958         if (error <= 0) {
2959                 if (error)
2960                         goto out;
2961 
2962                 if ((*opened & FILE_CREATED) ||
2963                     !S_ISREG(file_inode(file)->i_mode))
2964                         will_truncate = false;
2965 
2966                 audit_inode(name, file->f_path.dentry, 0);
2967                 goto opened;
2968         }
2969 
2970         if (*opened & FILE_CREATED) {
2971                 /* Don't check for write permission, don't truncate */
2972                 open_flag &= ~O_TRUNC;
2973                 will_truncate = false;
2974                 acc_mode = MAY_OPEN;
2975                 path_to_nameidata(path, nd);
2976                 goto finish_open_created;
2977         }
2978 
2979         /*
2980          * create/update audit record if it already exists.
2981          */
2982         if (d_is_positive(path->dentry))
2983                 audit_inode(name, path->dentry, 0);
2984 
2985         /*
2986          * If atomic_open() acquired write access it is dropped now due to
2987          * possible mount and symlink following (this might be optimized away if
2988          * necessary...)
2989          */
2990         if (got_write) {
2991                 mnt_drop_write(nd->path.mnt);
2992                 got_write = false;
2993         }
2994 
2995         error = -EEXIST;
2996         if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2997                 goto exit_dput;
2998 
2999         error = follow_managed(path, nd->flags);
3000         if (error < 0)
3001                 goto exit_dput;
3002 
3003         if (error)
3004                 nd->flags |= LOOKUP_JUMPED;
3005 
3006         BUG_ON(nd->flags & LOOKUP_RCU);
3007         inode = path->dentry->d_inode;
3008 finish_lookup:
3009         /* we _can_ be in RCU mode here */
3010         error = -ENOENT;
3011         if (!inode || d_is_negative(path->dentry)) {
3012                 path_to_nameidata(path, nd);
3013                 goto out;
3014         }
3015 
3016         if (should_follow_link(path->dentry, !symlink_ok)) {
3017                 if (nd->flags & LOOKUP_RCU) {
3018                         if (unlikely(unlazy_walk(nd, path->dentry))) {
3019                                 error = -ECHILD;
3020                                 goto out;
3021                         }
3022                 }
3023                 BUG_ON(inode != path->dentry->d_inode);
3024                 return 1;
3025         }
3026 
3027         if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3028                 path_to_nameidata(path, nd);
3029         } else {
3030                 save_parent.dentry = nd->path.dentry;
3031                 save_parent.mnt = mntget(path->mnt);
3032                 nd->path.dentry = path->dentry;
3033 
3034         }
3035         nd->inode = inode;
3036         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3037 finish_open:
3038         error = complete_walk(nd);
3039         if (error) {
3040                 path_put(&save_parent);
3041                 return error;
3042         }
3043         audit_inode(name, nd->path.dentry, 0);
3044         error = -EISDIR;
3045         if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3046                 goto out;
3047         error = -ENOTDIR;
3048         if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3049                 goto out;
3050         if (!S_ISREG(nd->inode->i_mode))
3051                 will_truncate = false;
3052 
3053         if (will_truncate) {
3054                 error = mnt_want_write(nd->path.mnt);
3055                 if (error)
3056                         goto out;
3057                 got_write = true;
3058         }
3059 finish_open_created:
3060         error = may_open(&nd->path, acc_mode, open_flag);
3061         if (error)
3062                 goto out;
3063 
3064         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3065         error = vfs_open(&nd->path, file, current_cred());
3066         if (!error) {
3067                 *opened |= FILE_OPENED;
3068         } else {
3069                 if (error == -EOPENSTALE)
3070                         goto stale_open;
3071                 goto out;
3072         }
3073 opened:
3074         error = open_check_o_direct(file);
3075         if (error)
3076                 goto exit_fput;
3077         error = ima_file_check(file, op->acc_mode, *opened);
3078         if (error)
3079                 goto exit_fput;
3080 
3081         if (will_truncate) {
3082                 error = handle_truncate(file);
3083                 if (error)
3084                         goto exit_fput;
3085         }
3086 out:
3087         if (got_write)
3088                 mnt_drop_write(nd->path.mnt);
3089         path_put(&save_parent);
3090         terminate_walk(nd);
3091         return error;
3092 
3093 exit_dput:
3094         path_put_conditional(path, nd);
3095         goto out;
3096 exit_fput:
3097         fput(file);
3098         goto out;
3099 
3100 stale_open:
3101         /* If no saved parent or already retried then can't retry */
3102         if (!save_parent.dentry || retried)
3103                 goto out;
3104 
3105         BUG_ON(save_parent.dentry != dir);
3106         path_put(&nd->path);
3107         nd->path = save_parent;
3108         nd->inode = dir->d_inode;
3109         save_parent.mnt = NULL;
3110         save_parent.dentry = NULL;
3111         if (got_write) {
3112                 mnt_drop_write(nd->path.mnt);
3113                 got_write = false;
3114         }
3115         retried = true;
3116         goto retry_lookup;
3117 }
3118 
3119 static int do_tmpfile(int dfd, struct filename *pathname,
3120                 struct nameidata *nd, int flags,
3121                 const struct open_flags *op,
3122                 struct file *file, int *opened)
3123 {
3124         static const struct qstr name = QSTR_INIT("/", 1);
3125         struct dentry *dentry, *child;
3126         struct inode *dir;
3127         int error = path_lookupat(dfd, pathname->name,
3128                                   flags | LOOKUP_DIRECTORY, nd);
3129         if (unlikely(error))
3130                 return error;
3131         error = mnt_want_write(nd->path.mnt);
3132         if (unlikely(error))
3133                 goto out;
3134         /* we want directory to be writable */
3135         error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3136         if (error)
3137                 goto out2;
3138         dentry = nd->path.dentry;
3139         dir = dentry->d_inode;
3140         if (!dir->i_op->tmpfile) {
3141                 error = -EOPNOTSUPP;
3142                 goto out2;
3143         }
3144         child = d_alloc(dentry, &name);
3145         if (unlikely(!child)) {
3146                 error = -ENOMEM;
3147                 goto out2;
3148         }
3149         nd->flags &= ~LOOKUP_DIRECTORY;
3150         nd->flags |= op->intent;
3151         dput(nd->path.dentry);
3152         nd->path.dentry = child;
3153         error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3154         if (error)
3155                 goto out2;
3156         audit_inode(pathname, nd->path.dentry, 0);
3157         /* Don't check for other permissions, the inode was just created */
3158         error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3159         if (error)
3160                 goto out2;
3161         file->f_path.mnt = nd->path.mnt;
3162         error = finish_open(file, nd->path.dentry, NULL, opened);
3163         if (error)
3164                 goto out2;
3165         error = open_check_o_direct(file);
3166         if (error) {
3167                 fput(file);
3168         } else if (!(op->open_flag & O_EXCL)) {
3169                 struct inode *inode = file_inode(file);
3170                 spin_lock(&inode->i_lock);
3171                 inode->i_state |= I_LINKABLE;
3172                 spin_unlock(&inode->i_lock);
3173         }
3174 out2:
3175         mnt_drop_write(nd->path.mnt);
3176 out:
3177         path_put(&nd->path);
3178         return error;
3179 }
3180 
3181 static struct file *path_openat(int dfd, struct filename *pathname,
3182                 struct nameidata *nd, const struct open_flags *op, int flags)
3183 {
3184         struct file *base = NULL;
3185         struct file *file;
3186         struct path path;
3187         int opened = 0;
3188         int error;
3189 
3190         file = get_empty_filp();
3191         if (IS_ERR(file))
3192                 return file;
3193 
3194         file->f_flags = op->open_flag;
3195 
3196         if (unlikely(file->f_flags & __O_TMPFILE)) {
3197                 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3198                 goto out;
3199         }
3200 
3201         error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3202         if (unlikely(error))
3203                 goto out;
3204 
3205         current->total_link_count = 0;
3206         error = link_path_walk(pathname->name, nd);
3207         if (unlikely(error))
3208                 goto out;
3209 
3210         error = do_last(nd, &path, file, op, &opened, pathname);
3211         while (unlikely(error > 0)) { /* trailing symlink */
3212                 struct path link = path;
3213                 void *cookie;
3214                 if (!(nd->flags & LOOKUP_FOLLOW)) {
3215                         path_put_conditional(&path, nd);
3216                         path_put(&nd->path);
3217                         error = -ELOOP;
3218                         break;
3219                 }
3220                 error = may_follow_link(&link, nd);
3221                 if (unlikely(error))
3222                         break;
3223                 nd->flags |= LOOKUP_PARENT;
3224                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3225                 error = follow_link(&link, nd, &cookie);
3226                 if (unlikely(error))
3227                         break;
3228                 error = do_last(nd, &path, file, op, &opened, pathname);
3229                 put_link(nd, &link, cookie);
3230         }
3231 out:
3232         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3233                 path_put(&nd->root);
3234         if (base)
3235                 fput(base);
3236         if (!(opened & FILE_OPENED)) {
3237                 BUG_ON(!error);
3238                 put_filp(file);
3239         }
3240         if (unlikely(error)) {
3241                 if (error == -EOPENSTALE) {
3242                         if (flags & LOOKUP_RCU)
3243                                 error = -ECHILD;
3244                         else
3245                                 error = -ESTALE;
3246                 }
3247                 file = ERR_PTR(error);
3248         }
3249         return file;
3250 }
3251 
3252 struct file *do_filp_open(int dfd, struct filename *pathname,
3253                 const struct open_flags *op)
3254 {
3255         struct nameidata nd;
3256         int flags = op->lookup_flags;
3257         struct file *filp;
3258 
3259         filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3260         if (unlikely(filp == ERR_PTR(-ECHILD)))
3261                 filp = path_openat(dfd, pathname, &nd, op, flags);
3262         if (unlikely(filp == ERR_PTR(-ESTALE)))
3263                 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3264         return filp;
3265 }
3266 
3267 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3268                 const char *name, const struct open_flags *op)
3269 {
3270         struct nameidata nd;
3271         struct file *file;
3272         struct filename filename = { .name = name };
3273         int flags = op->lookup_flags | LOOKUP_ROOT;
3274 
3275         nd.root.mnt = mnt;
3276         nd.root.dentry = dentry;
3277 
3278         if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3279                 return ERR_PTR(-ELOOP);
3280 
3281         file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3282         if (unlikely(file == ERR_PTR(-ECHILD)))
3283                 file = path_openat(-1, &filename, &nd, op, flags);
3284         if (unlikely(file == ERR_PTR(-ESTALE)))
3285                 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3286         return file;
3287 }
3288 
3289 struct dentry *kern_path_create(int dfd, const char *pathname,
3290                                 struct path *path, unsigned int lookup_flags)
3291 {
3292         struct dentry *dentry = ERR_PTR(-EEXIST);
3293         struct nameidata nd;
3294         int err2;
3295         int error;
3296         bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3297 
3298         /*
3299          * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3300          * other flags passed in are ignored!
3301          */
3302         lookup_flags &= LOOKUP_REVAL;
3303 
3304         error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3305         if (error)
3306                 return ERR_PTR(error);
3307 
3308         /*
3309          * Yucky last component or no last component at all?
3310          * (foo/., foo/.., /////)
3311          */
3312         if (nd.last_type != LAST_NORM)
3313                 goto out;
3314         nd.flags &= ~LOOKUP_PARENT;
3315         nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3316 
3317         /* don't fail immediately if it's r/o, at least try to report other errors */
3318         err2 = mnt_want_write(nd.path.mnt);
3319         /*
3320          * Do the final lookup.
3321          */
3322         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3323         dentry = lookup_hash(&nd);
3324         if (IS_ERR(dentry))
3325                 goto unlock;
3326 
3327         error = -EEXIST;
3328         if (d_is_positive(dentry))
3329                 goto fail;
3330 
3331         /*
3332          * Special case - lookup gave negative, but... we had foo/bar/
3333          * From the vfs_mknod() POV we just have a negative dentry -
3334          * all is fine. Let's be bastards - you had / on the end, you've
3335          * been asking for (non-existent) directory. -ENOENT for you.
3336          */
3337         if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3338                 error = -ENOENT;
3339                 goto fail;
3340         }
3341         if (unlikely(err2)) {
3342                 error = err2;
3343                 goto fail;
3344         }
3345         *path = nd.path;
3346         return dentry;
3347 fail:
3348         dput(dentry);
3349         dentry = ERR_PTR(error);
3350 unlock:
3351         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3352         if (!err2)
3353                 mnt_drop_write(nd.path.mnt);
3354 out:
3355         path_put(&nd.path);
3356         return dentry;
3357 }
3358 EXPORT_SYMBOL(kern_path_create);
3359 
3360 void done_path_create(struct path *path, struct dentry *dentry)
3361 {
3362         dput(dentry);
3363         mutex_unlock(&path->dentry->d_inode->i_mutex);
3364         mnt_drop_write(path->mnt);
3365         path_put(path);
3366 }
3367 EXPORT_SYMBOL(done_path_create);
3368 
3369 struct dentry *user_path_create(int dfd, const char __user *pathname,
3370                                 struct path *path, unsigned int lookup_flags)
3371 {
3372         struct filename *tmp = getname(pathname);
3373         struct dentry *res;
3374         if (IS_ERR(tmp))
3375                 return ERR_CAST(tmp);
3376         res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3377         putname(tmp);
3378         return res;
3379 }
3380 EXPORT_SYMBOL(user_path_create);
3381 
3382 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3383 {
3384         int error = may_create(dir, dentry);
3385 
3386         if (error)
3387                 return error;
3388 
3389         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3390                 return -EPERM;
3391 
3392         if (!dir->i_op->mknod)
3393                 return -EPERM;
3394 
3395         error = devcgroup_inode_mknod(mode, dev);
3396         if (error)
3397                 return error;
3398 
3399         error = security_inode_mknod(dir, dentry, mode, dev);
3400         if (error)
3401                 return error;
3402 
3403         error = dir->i_op->mknod(dir, dentry, mode, dev);
3404         if (!error)
3405                 fsnotify_create(dir, dentry);
3406         return error;
3407 }
3408 EXPORT_SYMBOL(vfs_mknod);
3409 
3410 static int may_mknod(umode_t mode)
3411 {
3412         switch (mode & S_IFMT) {
3413         case S_IFREG:
3414         case S_IFCHR:
3415         case S_IFBLK:
3416         case S_IFIFO:
3417         case S_IFSOCK:
3418         case 0: /* zero mode translates to S_IFREG */
3419                 return 0;
3420         case S_IFDIR:
3421                 return -EPERM;
3422         default:
3423                 return -EINVAL;
3424         }
3425 }
3426 
3427 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3428                 unsigned, dev)
3429 {
3430         struct dentry *dentry;
3431         struct path path;
3432         int error;
3433         unsigned int lookup_flags = 0;
3434 
3435         error = may_mknod(mode);
3436         if (error)
3437                 return error;
3438 retry:
3439         dentry = user_path_create(dfd, filename, &path, lookup_flags);
3440         if (IS_ERR(dentry))
3441                 return PTR_ERR(dentry);
3442 
3443         if (!IS_POSIXACL(path.dentry->d_inode))
3444                 mode &= ~current_umask();
3445         error = security_path_mknod(&path, dentry, mode, dev);
3446         if (error)
3447                 goto out;
3448         switch (mode & S_IFMT) {
3449                 case 0: case S_IFREG:
3450                         error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3451                         break;
3452                 case S_IFCHR: case S_IFBLK:
3453                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3454                                         new_decode_dev(dev));
3455                         break;
3456                 case S_IFIFO: case S_IFSOCK:
3457                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3458                         break;
3459         }
3460 out:
3461         done_path_create(&path, dentry);
3462         if (retry_estale(error, lookup_flags)) {
3463                 lookup_flags |= LOOKUP_REVAL;
3464                 goto retry;
3465         }
3466         return error;
3467 }
3468 
3469 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3470 {
3471         return sys_mknodat(AT_FDCWD, filename, mode, dev);
3472 }
3473 
3474 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3475 {
3476         int error = may_create(dir, dentry);
3477         unsigned max_links = dir->i_sb->s_max_links;
3478 
3479         if (error)
3480                 return error;
3481 
3482         if (!dir->i_op->mkdir)
3483                 return -EPERM;
3484 
3485         mode &= (S_IRWXUGO|S_ISVTX);
3486         error = security_inode_mkdir(dir, dentry, mode);
3487         if (error)
3488                 return error;
3489 
3490         if (max_links && dir->i_nlink >= max_links)
3491                 return -EMLINK;
3492 
3493         error = dir->i_op->mkdir(dir, dentry, mode);
3494         if (!error)
3495                 fsnotify_mkdir(dir, dentry);
3496         return error;
3497 }
3498 EXPORT_SYMBOL(vfs_mkdir);
3499 
3500 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3501 {
3502         struct dentry *dentry;
3503         struct path path;
3504         int error;
3505         unsigned int lookup_flags = LOOKUP_DIRECTORY;
3506 
3507 retry:
3508         dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3509         if (IS_ERR(dentry))
3510                 return PTR_ERR(dentry);
3511 
3512         if (!IS_POSIXACL(path.dentry->d_inode))
3513                 mode &= ~current_umask();
3514         error = security_path_mkdir(&path, dentry, mode);
3515         if (!error)
3516                 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3517         done_path_create(&path, dentry);
3518         if (retry_estale(error, lookup_flags)) {
3519                 lookup_flags |= LOOKUP_REVAL;
3520                 goto retry;
3521         }
3522         return error;
3523 }
3524 
3525 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3526 {
3527         return sys_mkdirat(AT_FDCWD, pathname, mode);
3528 }
3529 
3530 /*
3531  * The dentry_unhash() helper will try to drop the dentry early: we
3532  * should have a usage count of 1 if we're the only user of this
3533  * dentry, and if that is true (possibly after pruning the dcache),
3534  * then we drop the dentry now.
3535  *
3536  * A low-level filesystem can, if it choses, legally
3537  * do a
3538  *
3539  *      if (!d_unhashed(dentry))
3540  *              return -EBUSY;
3541  *
3542  * if it cannot handle the case of removing a directory
3543  * that is still in use by something else..
3544  */
3545 void dentry_unhash(struct dentry *dentry)
3546 {
3547         shrink_dcache_parent(dentry);
3548         spin_lock(&dentry->d_lock);
3549         if (dentry->d_lockref.count == 1)
3550                 __d_drop(dentry);
3551         spin_unlock(&dentry->d_lock);
3552 }
3553 EXPORT_SYMBOL(dentry_unhash);
3554 
3555 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3556 {
3557         int error = may_delete(dir, dentry, 1);
3558 
3559         if (error)
3560                 return error;
3561 
3562         if (!dir->i_op->rmdir)
3563                 return -EPERM;
3564 
3565         dget(dentry);
3566         mutex_lock(&dentry->d_inode->i_mutex);
3567 
3568         error = -EBUSY;
3569         if (is_local_mountpoint(dentry))
3570                 goto out;
3571 
3572         error = security_inode_rmdir(dir, dentry);
3573         if (error)
3574                 goto out;
3575 
3576         shrink_dcache_parent(dentry);
3577         error = dir->i_op->rmdir(dir, dentry);
3578         if (error)
3579                 goto out;
3580 
3581         dentry->d_inode->i_flags |= S_DEAD;
3582         dont_mount(dentry);
3583         detach_mounts(dentry);
3584 
3585 out:
3586         mutex_unlock(&dentry->d_inode->i_mutex);
3587         dput(dentry);
3588         if (!error)
3589                 d_delete(dentry);
3590         return error;
3591 }
3592 EXPORT_SYMBOL(vfs_rmdir);
3593 
3594 static long do_rmdir(int dfd, const char __user *pathname)
3595 {
3596         int error = 0;
3597         struct filename *name;
3598         struct dentry *dentry;
3599         struct nameidata nd;
3600         unsigned int lookup_flags = 0;
3601 retry:
3602         name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3603         if (IS_ERR(name))
3604                 return PTR_ERR(name);
3605 
3606         switch(nd.last_type) {
3607         case LAST_DOTDOT:
3608                 error = -ENOTEMPTY;
3609                 goto exit1;
3610         case LAST_DOT:
3611                 error = -EINVAL;
3612                 goto exit1;
3613         case LAST_ROOT:
3614                 error = -EBUSY;
3615                 goto exit1;
3616         }
3617 
3618         nd.flags &= ~LOOKUP_PARENT;
3619         error = mnt_want_write(nd.path.mnt);
3620         if (error)
3621                 goto exit1;
3622 
3623         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3624         dentry = lookup_hash(&nd);
3625         error = PTR_ERR(dentry);
3626         if (IS_ERR(dentry))
3627                 goto exit2;
3628         if (!dentry->d_inode) {
3629                 error = -ENOENT;
3630                 goto exit3;
3631         }
3632         error = security_path_rmdir(&nd.path, dentry);
3633         if (error)
3634                 goto exit3;
3635         error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3636 exit3:
3637         dput(dentry);
3638 exit2:
3639         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3640         mnt_drop_write(nd.path.mnt);
3641 exit1:
3642         path_put(&nd.path);
3643         putname(name);
3644         if (retry_estale(error, lookup_flags)) {
3645                 lookup_flags |= LOOKUP_REVAL;
3646                 goto retry;
3647         }
3648         return error;
3649 }
3650 
3651 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3652 {
3653         return do_rmdir(AT_FDCWD, pathname);
3654 }
3655 
3656 /**
3657  * vfs_unlink - unlink a filesystem object
3658  * @dir:        parent directory
3659  * @dentry:     victim
3660  * @delegated_inode: returns victim inode, if the inode is delegated.
3661  *
3662  * The caller must hold dir->i_mutex.
3663  *
3664  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3665  * return a reference to the inode in delegated_inode.  The caller
3666  * should then break the delegation on that inode and retry.  Because
3667  * breaking a delegation may take a long time, the caller should drop
3668  * dir->i_mutex before doing so.
3669  *
3670  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3671  * be appropriate for callers that expect the underlying filesystem not
3672  * to be NFS exported.
3673  */
3674 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3675 {
3676         struct inode *target = dentry->d_inode;
3677         int error = may_delete(dir, dentry, 0);
3678 
3679         if (error)
3680                 return error;
3681 
3682         if (!dir->i_op->unlink)
3683                 return -EPERM;
3684 
3685         mutex_lock(&target->i_mutex);
3686         if (is_local_mountpoint(dentry))
3687                 error = -EBUSY;
3688         else {
3689                 error = security_inode_unlink(dir, dentry);
3690                 if (!error) {
3691                         error = try_break_deleg(target, delegated_inode);
3692                         if (error)
3693                                 goto out;
3694                         error = dir->i_op->unlink(dir, dentry);
3695                         if (!error) {
3696                                 dont_mount(dentry);
3697                                 detach_mounts(dentry);
3698                         }
3699                 }
3700         }
3701 out:
3702         mutex_unlock(&target->i_mutex);
3703 
3704         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3705         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3706                 fsnotify_link_count(target);
3707                 d_delete(dentry);
3708         }
3709 
3710         return error;
3711 }
3712 EXPORT_SYMBOL(vfs_unlink);
3713 
3714 /*
3715  * Make sure that the actual truncation of the file will occur outside its
3716  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3717  * writeout happening, and we don't want to prevent access to the directory
3718  * while waiting on the I/O.
3719  */
3720 static long do_unlinkat(int dfd, const char __user *pathname)
3721 {
3722         int error;
3723         struct filename *name;
3724         struct dentry *dentry;
3725         struct nameidata nd;
3726         struct inode *inode = NULL;
3727         struct inode *delegated_inode = NULL;
3728         unsigned int lookup_flags = 0;
3729 retry:
3730         name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3731         if (IS_ERR(name))
3732                 return PTR_ERR(name);
3733 
3734         error = -EISDIR;
3735         if (nd.last_type != LAST_NORM)
3736                 goto exit1;
3737 
3738         nd.flags &= ~LOOKUP_PARENT;
3739         error = mnt_want_write(nd.path.mnt);
3740         if (error)
3741                 goto exit1;
3742 retry_deleg:
3743         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3744         dentry = lookup_hash(&nd);
3745         error = PTR_ERR(dentry);
3746         if (!IS_ERR(dentry)) {
3747                 /* Why not before? Because we want correct error value */
3748                 if (nd.last.name[nd.last.len])
3749                         goto slashes;
3750                 inode = dentry->d_inode;
3751                 if (d_is_negative(dentry))
3752                         goto slashes;
3753                 ihold(inode);
3754                 error = security_path_unlink(&nd.path, dentry);
3755                 if (error)
3756                         goto exit2;
3757                 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3758 exit2:
3759                 dput(dentry);
3760         }
3761         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3762         if (inode)
3763                 iput(inode);    /* truncate the inode here */
3764         inode = NULL;
3765         if (delegated_inode) {
3766                 error = break_deleg_wait(&delegated_inode);
3767                 if (!error)
3768                         goto retry_deleg;
3769         }
3770         mnt_drop_write(nd.path.mnt);
3771 exit1:
3772         path_put(&nd.path);
3773         putname(name);
3774         if (retry_estale(error, lookup_flags)) {
3775                 lookup_flags |= LOOKUP_REVAL;
3776                 inode = NULL;
3777                 goto retry;
3778         }
3779         return error;
3780 
3781 slashes:
3782         if (d_is_negative(dentry))
3783                 error = -ENOENT;
3784         else if (d_is_dir(dentry))
3785                 error = -EISDIR;
3786         else
3787                 error = -ENOTDIR;
3788         goto exit2;
3789 }
3790 
3791 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3792 {
3793         if ((flag & ~AT_REMOVEDIR) != 0)
3794                 return -EINVAL;
3795 
3796         if (flag & AT_REMOVEDIR)
3797                 return do_rmdir(dfd, pathname);
3798 
3799         return do_unlinkat(dfd, pathname);
3800 }
3801 
3802 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3803 {
3804         return do_unlinkat(AT_FDCWD, pathname);
3805 }
3806 
3807 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3808 {
3809         int error = may_create(dir, dentry);
3810 
3811         if (error)
3812                 return error;
3813 
3814         if (!dir->i_op->symlink)
3815                 return -EPERM;
3816 
3817         error = security_inode_symlink(dir, dentry, oldname);
3818         if (error)
3819                 return error;
3820 
3821         error = dir->i_op->symlink(dir, dentry, oldname);
3822         if (!error)
3823                 fsnotify_create(dir, dentry);
3824         return error;
3825 }
3826 EXPORT_SYMBOL(vfs_symlink);
3827 
3828 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3829                 int, newdfd, const char __user *, newname)
3830 {
3831         int error;
3832         struct filename *from;
3833         struct dentry *dentry;
3834         struct path path;
3835         unsigned int lookup_flags = 0;
3836 
3837         from = getname(oldname);
3838         if (IS_ERR(from))
3839                 return PTR_ERR(from);
3840 retry:
3841         dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3842         error = PTR_ERR(dentry);
3843         if (IS_ERR(dentry))
3844                 goto out_putname;
3845 
3846         error = security_path_symlink(&path, dentry, from->name);
3847         if (!error)
3848                 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3849         done_path_create(&path, dentry);
3850         if (retry_estale(error, lookup_flags)) {
3851                 lookup_flags |= LOOKUP_REVAL;
3852                 goto retry;
3853         }
3854 out_putname:
3855         putname(from);
3856         return error;
3857 }
3858 
3859 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3860 {
3861         return sys_symlinkat(oldname, AT_FDCWD, newname);
3862 }
3863 
3864 /**
3865  * vfs_link - create a new link
3866  * @old_dentry: object to be linked
3867  * @dir:        new parent
3868  * @new_dentry: where to create the new link
3869  * @delegated_inode: returns inode needing a delegation break
3870  *
3871  * The caller must hold dir->i_mutex
3872  *
3873  * If vfs_link discovers a delegation on the to-be-linked file in need
3874  * of breaking, it will return -EWOULDBLOCK and return a reference to the
3875  * inode in delegated_inode.  The caller should then break the delegation
3876  * and retry.  Because breaking a delegation may take a long time, the
3877  * caller should drop the i_mutex before doing so.
3878  *
3879  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3880  * be appropriate for callers that expect the underlying filesystem not
3881  * to be NFS exported.
3882  */
3883 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3884 {
3885         struct inode *inode = old_dentry->d_inode;
3886         unsigned max_links = dir->i_sb->s_max_links;
3887         int error;
3888 
3889         if (!inode)
3890                 return -ENOENT;
3891 
3892         error = may_create(dir, new_dentry);
3893         if (error)
3894                 return error;
3895 
3896         if (dir->i_sb != inode->i_sb)
3897                 return -EXDEV;
3898 
3899         /*
3900          * A link to an append-only or immutable file cannot be created.
3901          */
3902         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3903                 return -EPERM;
3904         if (!dir->i_op->link)
3905                 return -EPERM;
3906         if (S_ISDIR(inode->i_mode))
3907                 return -EPERM;
3908 
3909         error = security_inode_link(old_dentry, dir, new_dentry);
3910         if (error)
3911                 return error;
3912 
3913         mutex_lock(&inode->i_mutex);
3914         /* Make sure we don't allow creating hardlink to an unlinked file */
3915         if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3916                 error =  -ENOENT;
3917         else if (max_links && inode->i_nlink >= max_links)
3918                 error = -EMLINK;
3919         else {
3920                 error = try_break_deleg(inode, delegated_inode);
3921                 if (!error)
3922                         error = dir->i_op->link(old_dentry, dir, new_dentry);
3923         }
3924 
3925         if (!error && (inode->i_state & I_LINKABLE)) {
3926                 spin_lock(&inode->i_lock);
3927                 inode->i_state &= ~I_LINKABLE;
3928                 spin_unlock(&inode->i_lock);
3929         }
3930         mutex_unlock(&inode->i_mutex);
3931         if (!error)
3932                 fsnotify_link(dir, inode, new_dentry);
3933         return error;
3934 }
3935 EXPORT_SYMBOL(vfs_link);
3936 
3937 /*
3938  * Hardlinks are often used in delicate situations.  We avoid
3939  * security-related surprises by not following symlinks on the
3940  * newname.  --KAB
3941  *
3942  * We don't follow them on the oldname either to be compatible
3943  * with linux 2.0, and to avoid hard-linking to directories
3944  * and other special files.  --ADM
3945  */
3946 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3947                 int, newdfd, const char __user *, newname, int, flags)
3948 {
3949         struct dentry *new_dentry;
3950         struct path old_path, new_path;
3951         struct inode *delegated_inode = NULL;
3952         int how = 0;
3953         int error;
3954 
3955         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3956                 return -EINVAL;
3957         /*
3958          * To use null names we require CAP_DAC_READ_SEARCH
3959          * This ensures that not everyone will be able to create
3960          * handlink using the passed filedescriptor.
3961          */
3962         if (flags & AT_EMPTY_PATH) {
3963                 if (!capable(CAP_DAC_READ_SEARCH))
3964                         return -ENOENT;
3965                 how = LOOKUP_EMPTY;
3966         }
3967 
3968         if (flags & AT_SYMLINK_FOLLOW)
3969                 how |= LOOKUP_FOLLOW;
3970 retry:
3971         error = user_path_at(olddfd, oldname, how, &old_path);
3972         if (error)
3973                 return error;
3974 
3975         new_dentry = user_path_create(newdfd, newname, &new_path,
3976                                         (how & LOOKUP_REVAL));
3977         error = PTR_ERR(new_dentry);
3978         if (IS_ERR(new_dentry))
3979                 goto out;
3980 
3981         error = -EXDEV;
3982         if (old_path.mnt != new_path.mnt)
3983                 goto out_dput;
3984         error = may_linkat(&old_path);
3985         if (unlikely(error))
3986                 goto out_dput;
3987         error = security_path_link(old_path.dentry, &new_path, new_dentry);
3988         if (error)
3989                 goto out_dput;
3990         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3991 out_dput:
3992         done_path_create(&new_path, new_dentry);
3993         if (delegated_inode) {
3994                 error = break_deleg_wait(&delegated_inode);
3995                 if (!error) {
3996                         path_put(&old_path);
3997                         goto retry;
3998                 }
3999         }
4000         if (retry_estale(error, how)) {
4001                 path_put(&old_path);
4002                 how |= LOOKUP_REVAL;
4003                 goto retry;
4004         }
4005 out:
4006         path_put(&old_path);
4007 
4008         return error;
4009 }
4010 
4011 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4012 {
4013         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4014 }
4015 
4016 /**
4017  * vfs_rename - rename a filesystem object
4018  * @old_dir:    parent of source
4019  * @old_dentry: source
4020  * @new_dir:    parent of destination
4021  * @new_dentry: destination
4022  * @delegated_inode: returns an inode needing a delegation break
4023  * @flags:      rename flags
4024  *
4025  * The caller must hold multiple mutexes--see lock_rename()).
4026  *
4027  * If vfs_rename discovers a delegation in need of breaking at either
4028  * the source or destination, it will return -EWOULDBLOCK and return a
4029  * reference to the inode in delegated_inode.  The caller should then
4030  * break the delegation and retry.  Because breaking a delegation may
4031  * take a long time, the caller should drop all locks before doing
4032  * so.
4033  *
4034  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4035  * be appropriate for callers that expect the underlying filesystem not
4036  * to be NFS exported.
4037  *
4038  * The worst of all namespace operations - renaming directory. "Perverted"
4039  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4040  * Problems:
4041  *      a) we can get into loop creation.
4042  *      b) race potential - two innocent renames can create a loop together.
4043  *         That's where 4.4 screws up. Current fix: serialization on
4044  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4045  *         story.
4046  *      c) we have to lock _four_ objects - parents and victim (if it exists),
4047  *         and source (if it is not a directory).
4048  *         And that - after we got ->i_mutex on parents (until then we don't know
4049  *         whether the target exists).  Solution: try to be smart with locking
4050  *         order for inodes.  We rely on the fact that tree topology may change
4051  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4052  *         move will be locked.  Thus we can rank directories by the tree
4053  *         (ancestors first) and rank all non-directories after them.
4054  *         That works since everybody except rename does "lock parent, lookup,
4055  *         lock child" and rename is under ->s_vfs_rename_mutex.
4056  *         HOWEVER, it relies on the assumption that any object with ->lookup()
4057  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4058  *         we'd better make sure that there's no link(2) for them.
4059  *      d) conversion from fhandle to dentry may come in the wrong moment - when
4060  *         we are removing the target. Solution: we will have to grab ->i_mutex
4061  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4062  *         ->i_mutex on parents, which works but leads to some truly excessive
4063  *         locking].
4064  */
4065 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4066                struct inode *new_dir, struct dentry *new_dentry,
4067                struct inode **delegated_inode, unsigned int flags)
4068 {
4069         int error;
4070         bool is_dir = d_is_dir(old_dentry);
4071         const unsigned char *old_name;
4072         struct inode *source = old_dentry->d_inode;
4073         struct inode *target = new_dentry->d_inode;
4074         bool new_is_dir = false;
4075         unsigned max_links = new_dir->i_sb->s_max_links;
4076 
4077         if (source == target)
4078                 return 0;
4079 
4080         error = may_delete(old_dir, old_dentry, is_dir);
4081         if (error)
4082                 return error;
4083 
4084         if (!target) {
4085                 error = may_create(new_dir, new_dentry);
4086         } else {
4087                 new_is_dir = d_is_dir(new_dentry);
4088 
4089                 if (!(flags & RENAME_EXCHANGE))
4090                         error = may_delete(new_dir, new_dentry, is_dir);
4091                 else
4092                         error = may_delete(new_dir, new_dentry, new_is_dir);
4093         }
4094         if (error)
4095                 return error;
4096 
4097         if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4098                 return -EPERM;
4099 
4100         if (flags && !old_dir->i_op->rename2)
4101                 return -EINVAL;
4102 
4103         /*
4104          * If we are going to change the parent - check write permissions,
4105          * we'll need to flip '..'.
4106          */
4107         if (new_dir != old_dir) {
4108                 if (is_dir) {
4109                         error = inode_permission(source, MAY_WRITE);
4110                         if (error)
4111                                 return error;
4112                 }
4113                 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4114                         error = inode_permission(target, MAY_WRITE);
4115                         if (error)
4116                                 return error;
4117                 }
4118         }
4119 
4120         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4121                                       flags);
4122         if (error)
4123                 return error;
4124 
4125         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4126         dget(new_dentry);
4127         if (!is_dir || (flags & RENAME_EXCHANGE))
4128                 lock_two_nondirectories(source, target);
4129         else if (target)
4130                 mutex_lock(&target->i_mutex);
4131 
4132         error = -EBUSY;
4133         if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4134                 goto out;
4135 
4136         if (max_links && new_dir != old_dir) {
4137                 error = -EMLINK;
4138                 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4139                         goto out;
4140                 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4141                     old_dir->i_nlink >= max_links)
4142                         goto out;
4143         }
4144         if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4145                 shrink_dcache_parent(new_dentry);
4146         if (!is_dir) {
4147                 error = try_break_deleg(source, delegated_inode);
4148                 if (error)
4149                         goto out;
4150         }
4151         if (target && !new_is_dir) {
4152                 error = try_break_deleg(target, delegated_inode);
4153                 if (error)
4154                         goto out;
4155         }
4156         if (!old_dir->i_op->rename2) {
4157                 error = old_dir->i_op->rename(old_dir, old_dentry,
4158                                               new_dir, new_dentry);
4159         } else {
4160                 WARN_ON(old_dir->i_op->rename != NULL);
4161                 error = old_dir->i_op->rename2(old_dir, old_dentry,
4162                                                new_dir, new_dentry, flags);
4163         }
4164         if (error)
4165                 goto out;
4166 
4167         if (!(flags & RENAME_EXCHANGE) && target) {
4168                 if (is_dir)
4169                         target->i_flags |= S_DEAD;
4170                 dont_mount(new_dentry);
4171                 detach_mounts(new_dentry);
4172         }
4173         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4174                 if (!(flags & RENAME_EXCHANGE))
4175                         d_move(old_dentry, new_dentry);
4176                 else
4177                         d_exchange(old_dentry, new_dentry);
4178         }
4179 out:
4180         if (!is_dir || (flags & RENAME_EXCHANGE))
4181                 unlock_two_nondirectories(source, target);
4182         else if (target)
4183                 mutex_unlock(&target->i_mutex);
4184         dput(new_dentry);
4185         if (!error) {
4186                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4187                               !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4188                 if (flags & RENAME_EXCHANGE) {
4189                         fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4190                                       new_is_dir, NULL, new_dentry);
4191                 }
4192         }
4193         fsnotify_oldname_free(old_name);
4194 
4195         return error;
4196 }
4197 EXPORT_SYMBOL(vfs_rename);
4198 
4199 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4200                 int, newdfd, const char __user *, newname, unsigned int, flags)
4201 {
4202         struct dentry *old_dir, *new_dir;
4203         struct dentry *old_dentry, *new_dentry;
4204         struct dentry *trap;
4205         struct nameidata oldnd, newnd;
4206         struct inode *delegated_inode = NULL;
4207         struct filename *from;
4208         struct filename *to;
4209         unsigned int lookup_flags = 0;
4210         bool should_retry = false;
4211         int error;
4212 
4213         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4214                 return -EINVAL;
4215 
4216         if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4217             (flags & RENAME_EXCHANGE))
4218                 return -EINVAL;
4219 
4220         if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4221                 return -EPERM;
4222 
4223 retry:
4224         from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4225         if (IS_ERR(from)) {
4226                 error = PTR_ERR(from);
4227                 goto exit;
4228         }
4229 
4230         to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4231         if (IS_ERR(to)) {
4232                 error = PTR_ERR(to);
4233                 goto exit1;
4234         }
4235 
4236         error = -EXDEV;
4237         if (oldnd.path.mnt != newnd.path.mnt)
4238                 goto exit2;
4239 
4240         old_dir = oldnd.path.dentry;
4241         error = -EBUSY;
4242         if (oldnd.last_type != LAST_NORM)
4243                 goto exit2;
4244 
4245         new_dir = newnd.path.dentry;
4246         if (flags & RENAME_NOREPLACE)
4247                 error = -EEXIST;
4248         if (newnd.last_type != LAST_NORM)
4249                 goto exit2;
4250 
4251         error = mnt_want_write(oldnd.path.mnt);
4252         if (error)
4253                 goto exit2;
4254 
4255         oldnd.flags &= ~LOOKUP_PARENT;
4256         newnd.flags &= ~LOOKUP_PARENT;
4257         if (!(flags & RENAME_EXCHANGE))
4258                 newnd.flags |= LOOKUP_RENAME_TARGET;
4259 
4260 retry_deleg:
4261         trap = lock_rename(new_dir, old_dir);
4262 
4263         old_dentry = lookup_hash(&oldnd);
4264         error = PTR_ERR(old_dentry);
4265         if (IS_ERR(old_dentry))
4266                 goto exit3;
4267         /* source must exist */
4268         error = -ENOENT;
4269         if (d_is_negative(old_dentry))
4270                 goto exit4;
4271         new_dentry = lookup_hash(&newnd);
4272         error = PTR_ERR(new_dentry);
4273         if (IS_ERR(new_dentry))
4274                 goto exit4;
4275         error = -EEXIST;
4276         if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4277                 goto exit5;
4278         if (flags & RENAME_EXCHANGE) {
4279                 error = -ENOENT;
4280                 if (d_is_negative(new_dentry))
4281                         goto exit5;
4282 
4283                 if (!d_is_dir(new_dentry)) {
4284                         error = -ENOTDIR;
4285                         if (newnd.last.name[newnd.last.len])
4286                                 goto exit5;
4287                 }
4288         }
4289         /* unless the source is a directory trailing slashes give -ENOTDIR */
4290         if (!d_is_dir(old_dentry)) {
4291                 error = -ENOTDIR;
4292                 if (oldnd.last.name[oldnd.last.len])
4293                         goto exit5;
4294                 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4295                         goto exit5;
4296         }
4297         /* source should not be ancestor of target */
4298         error = -EINVAL;
4299         if (old_dentry == trap)
4300                 goto exit5;
4301         /* target should not be an ancestor of source */
4302         if (!(flags & RENAME_EXCHANGE))
4303                 error = -ENOTEMPTY;
4304         if (new_dentry == trap)
4305                 goto exit5;
4306 
4307         error = security_path_rename(&oldnd.path, old_dentry,
4308                                      &newnd.path, new_dentry, flags);
4309         if (error)
4310                 goto exit5;
4311         error = vfs_rename(old_dir->d_inode, old_dentry,
4312                            new_dir->d_inode, new_dentry,
4313                            &delegated_inode, flags);
4314 exit5:
4315         dput(new_dentry);
4316 exit4:
4317         dput(old_dentry);
4318 exit3:
4319         unlock_rename(new_dir, old_dir);
4320         if (delegated_inode) {
4321                 error = break_deleg_wait(&delegated_inode);
4322                 if (!error)
4323                         goto retry_deleg;
4324         }
4325         mnt_drop_write(oldnd.path.mnt);
4326 exit2:
4327         if (retry_estale(error, lookup_flags))
4328                 should_retry = true;
4329         path_put(&newnd.path);
4330         putname(to);
4331 exit1:
4332         path_put(&oldnd.path);
4333         putname(from);
4334         if (should_retry) {
4335                 should_retry = false;
4336                 lookup_flags |= LOOKUP_REVAL;
4337                 goto retry;
4338         }
4339 exit:
4340         return error;
4341 }
4342 
4343 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4344                 int, newdfd, const char __user *, newname)
4345 {
4346         return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4347 }
4348 
4349 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4350 {
4351         return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4352 }
4353 
4354 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4355 {
4356         int error = may_create(dir, dentry);
4357         if (error)
4358                 return error;
4359 
4360         if (!dir->i_op->mknod)
4361                 return -EPERM;
4362 
4363         return dir->i_op->mknod(dir, dentry,
4364                                 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4365 }
4366 EXPORT_SYMBOL(vfs_whiteout);
4367 
4368 int readlink_copy(char __user *buffer, int buflen, const char *link)
4369 {
4370         int len = PTR_ERR(link);
4371         if (IS_ERR(link))
4372                 goto out;
4373 
4374         len = strlen(link);
4375         if (len > (unsigned) buflen)
4376                 len = buflen;
4377         if (copy_to_user(buffer, link, len))
4378                 len = -EFAULT;
4379 out:
4380         return len;
4381 }
4382 EXPORT_SYMBOL(readlink_copy);
4383 
4384 /*
4385  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4386  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4387  * using) it for any given inode is up to filesystem.
4388  */
4389 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4390 {
4391         struct nameidata nd;
4392         void *cookie;
4393         int res;
4394 
4395         nd.depth = 0;
4396         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4397         if (IS_ERR(cookie))
4398                 return PTR_ERR(cookie);
4399 
4400         res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4401         if (dentry->d_inode->i_op->put_link)
4402                 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4403         return res;
4404 }
4405 EXPORT_SYMBOL(generic_readlink);
4406 
4407 /* get the link contents into pagecache */
4408 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4409 {
4410         char *kaddr;
4411         struct page *page;
4412         struct address_space *mapping = dentry->d_inode->i_mapping;
4413         page = read_mapping_page(mapping, 0, NULL);
4414         if (IS_ERR(page))
4415                 return (char*)page;
4416         *ppage = page;
4417         kaddr = kmap(page);
4418         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4419         return kaddr;
4420 }
4421 
4422 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4423 {
4424         struct page *page = NULL;
4425         int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4426         if (page) {
4427                 kunmap(page);
4428                 page_cache_release(page);
4429         }
4430         return res;
4431 }
4432 EXPORT_SYMBOL(page_readlink);
4433 
4434 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4435 {
4436         struct page *page = NULL;
4437         nd_set_link(nd, page_getlink(dentry, &page));
4438         return page;
4439 }
4440 EXPORT_SYMBOL(page_follow_link_light);
4441 
4442 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4443 {
4444         struct page *page = cookie;
4445 
4446         if (page) {
4447                 kunmap(page);
4448                 page_cache_release(page);
4449         }
4450 }
4451 EXPORT_SYMBOL(page_put_link);
4452 
4453 /*
4454  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4455  */
4456 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4457 {
4458         struct address_space *mapping = inode->i_mapping;
4459         struct page *page;
4460         void *fsdata;
4461         int err;
4462         char *kaddr;
4463         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4464         if (nofs)
4465                 flags |= AOP_FLAG_NOFS;
4466 
4467 retry:
4468         err = pagecache_write_begin(NULL, mapping, 0, len-1,
4469                                 flags, &page, &fsdata);
4470         if (err)
4471                 goto fail;
4472 
4473         kaddr = kmap_atomic(page);
4474         memcpy(kaddr, symname, len-1);
4475         kunmap_atomic(kaddr);
4476 
4477         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4478                                                         page, fsdata);
4479         if (err < 0)
4480                 goto fail;
4481         if (err < len-1)
4482                 goto retry;
4483 
4484         mark_inode_dirty(inode);
4485         return 0;
4486 fail:
4487         return err;
4488 }
4489 EXPORT_SYMBOL(__page_symlink);
4490 
4491 int page_symlink(struct inode *inode, const char *symname, int len)
4492 {
4493         return __page_symlink(inode, symname, len,
4494                         !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4495 }
4496 EXPORT_SYMBOL(page_symlink);
4497 
4498 const struct inode_operations page_symlink_inode_operations = {
4499         .readlink       = generic_readlink,
4500         .follow_link    = page_follow_link_light,
4501         .put_link       = page_put_link,
4502 };
4503 EXPORT_SYMBOL(page_symlink_inode_operations);
4504 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us