Version:  2.0.40 2.2.26 2.4.37 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 3.19

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 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 struct nameidata {
491         struct path     path;
492         struct qstr     last;
493         struct path     root;
494         struct inode    *inode; /* path.dentry.d_inode */
495         unsigned int    flags;
496         unsigned        seq, m_seq;
497         int             last_type;
498         unsigned        depth;
499         struct file     *base;
500         char *saved_names[MAX_NESTED_LINKS + 1];
501 };
502 
503 /*
504  * Path walking has 2 modes, rcu-walk and ref-walk (see
505  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
506  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
507  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
508  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
509  * got stuck, so ref-walk may continue from there. If this is not successful
510  * (eg. a seqcount has changed), then failure is returned and it's up to caller
511  * to restart the path walk from the beginning in ref-walk mode.
512  */
513 
514 /**
515  * unlazy_walk - try to switch to ref-walk mode.
516  * @nd: nameidata pathwalk data
517  * @dentry: child of nd->path.dentry or NULL
518  * Returns: 0 on success, -ECHILD on failure
519  *
520  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
521  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
522  * @nd or NULL.  Must be called from rcu-walk context.
523  */
524 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
525 {
526         struct fs_struct *fs = current->fs;
527         struct dentry *parent = nd->path.dentry;
528 
529         BUG_ON(!(nd->flags & LOOKUP_RCU));
530 
531         /*
532          * After legitimizing the bastards, terminate_walk()
533          * will do the right thing for non-RCU mode, and all our
534          * subsequent exit cases should rcu_read_unlock()
535          * before returning.  Do vfsmount first; if dentry
536          * can't be legitimized, just set nd->path.dentry to NULL
537          * and rely on dput(NULL) being a no-op.
538          */
539         if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
540                 return -ECHILD;
541         nd->flags &= ~LOOKUP_RCU;
542 
543         if (!lockref_get_not_dead(&parent->d_lockref)) {
544                 nd->path.dentry = NULL; 
545                 goto out;
546         }
547 
548         /*
549          * For a negative lookup, the lookup sequence point is the parents
550          * sequence point, and it only needs to revalidate the parent dentry.
551          *
552          * For a positive lookup, we need to move both the parent and the
553          * dentry from the RCU domain to be properly refcounted. And the
554          * sequence number in the dentry validates *both* dentry counters,
555          * since we checked the sequence number of the parent after we got
556          * the child sequence number. So we know the parent must still
557          * be valid if the child sequence number is still valid.
558          */
559         if (!dentry) {
560                 if (read_seqcount_retry(&parent->d_seq, nd->seq))
561                         goto out;
562                 BUG_ON(nd->inode != parent->d_inode);
563         } else {
564                 if (!lockref_get_not_dead(&dentry->d_lockref))
565                         goto out;
566                 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
567                         goto drop_dentry;
568         }
569 
570         /*
571          * Sequence counts matched. Now make sure that the root is
572          * still valid and get it if required.
573          */
574         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
575                 spin_lock(&fs->lock);
576                 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
577                         goto unlock_and_drop_dentry;
578                 path_get(&nd->root);
579                 spin_unlock(&fs->lock);
580         }
581 
582         rcu_read_unlock();
583         return 0;
584 
585 unlock_and_drop_dentry:
586         spin_unlock(&fs->lock);
587 drop_dentry:
588         rcu_read_unlock();
589         dput(dentry);
590         goto drop_root_mnt;
591 out:
592         rcu_read_unlock();
593 drop_root_mnt:
594         if (!(nd->flags & LOOKUP_ROOT))
595                 nd->root.mnt = NULL;
596         return -ECHILD;
597 }
598 
599 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
600 {
601         return dentry->d_op->d_revalidate(dentry, flags);
602 }
603 
604 /**
605  * complete_walk - successful completion of path walk
606  * @nd:  pointer nameidata
607  *
608  * If we had been in RCU mode, drop out of it and legitimize nd->path.
609  * Revalidate the final result, unless we'd already done that during
610  * the path walk or the filesystem doesn't ask for it.  Return 0 on
611  * success, -error on failure.  In case of failure caller does not
612  * need to drop nd->path.
613  */
614 static int complete_walk(struct nameidata *nd)
615 {
616         struct dentry *dentry = nd->path.dentry;
617         int status;
618 
619         if (nd->flags & LOOKUP_RCU) {
620                 nd->flags &= ~LOOKUP_RCU;
621                 if (!(nd->flags & LOOKUP_ROOT))
622                         nd->root.mnt = NULL;
623 
624                 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
625                         rcu_read_unlock();
626                         return -ECHILD;
627                 }
628                 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
629                         rcu_read_unlock();
630                         mntput(nd->path.mnt);
631                         return -ECHILD;
632                 }
633                 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
634                         rcu_read_unlock();
635                         dput(dentry);
636                         mntput(nd->path.mnt);
637                         return -ECHILD;
638                 }
639                 rcu_read_unlock();
640         }
641 
642         if (likely(!(nd->flags & LOOKUP_JUMPED)))
643                 return 0;
644 
645         if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
646                 return 0;
647 
648         status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
649         if (status > 0)
650                 return 0;
651 
652         if (!status)
653                 status = -ESTALE;
654 
655         path_put(&nd->path);
656         return status;
657 }
658 
659 static __always_inline void set_root(struct nameidata *nd)
660 {
661         get_fs_root(current->fs, &nd->root);
662 }
663 
664 static int link_path_walk(const char *, struct nameidata *);
665 
666 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
667 {
668         struct fs_struct *fs = current->fs;
669         unsigned seq, res;
670 
671         do {
672                 seq = read_seqcount_begin(&fs->seq);
673                 nd->root = fs->root;
674                 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
675         } while (read_seqcount_retry(&fs->seq, seq));
676         return res;
677 }
678 
679 static void path_put_conditional(struct path *path, struct nameidata *nd)
680 {
681         dput(path->dentry);
682         if (path->mnt != nd->path.mnt)
683                 mntput(path->mnt);
684 }
685 
686 static inline void path_to_nameidata(const struct path *path,
687                                         struct nameidata *nd)
688 {
689         if (!(nd->flags & LOOKUP_RCU)) {
690                 dput(nd->path.dentry);
691                 if (nd->path.mnt != path->mnt)
692                         mntput(nd->path.mnt);
693         }
694         nd->path.mnt = path->mnt;
695         nd->path.dentry = path->dentry;
696 }
697 
698 /*
699  * Helper to directly jump to a known parsed path from ->follow_link,
700  * caller must have taken a reference to path beforehand.
701  */
702 void nd_jump_link(struct nameidata *nd, struct path *path)
703 {
704         path_put(&nd->path);
705 
706         nd->path = *path;
707         nd->inode = nd->path.dentry->d_inode;
708         nd->flags |= LOOKUP_JUMPED;
709 }
710 
711 void nd_set_link(struct nameidata *nd, char *path)
712 {
713         nd->saved_names[nd->depth] = path;
714 }
715 EXPORT_SYMBOL(nd_set_link);
716 
717 char *nd_get_link(struct nameidata *nd)
718 {
719         return nd->saved_names[nd->depth];
720 }
721 EXPORT_SYMBOL(nd_get_link);
722 
723 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
724 {
725         struct inode *inode = link->dentry->d_inode;
726         if (inode->i_op->put_link)
727                 inode->i_op->put_link(link->dentry, nd, cookie);
728         path_put(link);
729 }
730 
731 int sysctl_protected_symlinks __read_mostly = 0;
732 int sysctl_protected_hardlinks __read_mostly = 0;
733 
734 /**
735  * may_follow_link - Check symlink following for unsafe situations
736  * @link: The path of the symlink
737  * @nd: nameidata pathwalk data
738  *
739  * In the case of the sysctl_protected_symlinks sysctl being enabled,
740  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
741  * in a sticky world-writable directory. This is to protect privileged
742  * processes from failing races against path names that may change out
743  * from under them by way of other users creating malicious symlinks.
744  * It will permit symlinks to be followed only when outside a sticky
745  * world-writable directory, or when the uid of the symlink and follower
746  * match, or when the directory owner matches the symlink's owner.
747  *
748  * Returns 0 if following the symlink is allowed, -ve on error.
749  */
750 static inline int may_follow_link(struct path *link, struct nameidata *nd)
751 {
752         const struct inode *inode;
753         const struct inode *parent;
754 
755         if (!sysctl_protected_symlinks)
756                 return 0;
757 
758         /* Allowed if owner and follower match. */
759         inode = link->dentry->d_inode;
760         if (uid_eq(current_cred()->fsuid, inode->i_uid))
761                 return 0;
762 
763         /* Allowed if parent directory not sticky and world-writable. */
764         parent = nd->path.dentry->d_inode;
765         if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
766                 return 0;
767 
768         /* Allowed if parent directory and link owner match. */
769         if (uid_eq(parent->i_uid, inode->i_uid))
770                 return 0;
771 
772         audit_log_link_denied("follow_link", link);
773         path_put_conditional(link, nd);
774         path_put(&nd->path);
775         return -EACCES;
776 }
777 
778 /**
779  * safe_hardlink_source - Check for safe hardlink conditions
780  * @inode: the source inode to hardlink from
781  *
782  * Return false if at least one of the following conditions:
783  *    - inode is not a regular file
784  *    - inode is setuid
785  *    - inode is setgid and group-exec
786  *    - access failure for read and write
787  *
788  * Otherwise returns true.
789  */
790 static bool safe_hardlink_source(struct inode *inode)
791 {
792         umode_t mode = inode->i_mode;
793 
794         /* Special files should not get pinned to the filesystem. */
795         if (!S_ISREG(mode))
796                 return false;
797 
798         /* Setuid files should not get pinned to the filesystem. */
799         if (mode & S_ISUID)
800                 return false;
801 
802         /* Executable setgid files should not get pinned to the filesystem. */
803         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
804                 return false;
805 
806         /* Hardlinking to unreadable or unwritable sources is dangerous. */
807         if (inode_permission(inode, MAY_READ | MAY_WRITE))
808                 return false;
809 
810         return true;
811 }
812 
813 /**
814  * may_linkat - Check permissions for creating a hardlink
815  * @link: the source to hardlink from
816  *
817  * Block hardlink when all of:
818  *  - sysctl_protected_hardlinks enabled
819  *  - fsuid does not match inode
820  *  - hardlink source is unsafe (see safe_hardlink_source() above)
821  *  - not CAP_FOWNER
822  *
823  * Returns 0 if successful, -ve on error.
824  */
825 static int may_linkat(struct path *link)
826 {
827         const struct cred *cred;
828         struct inode *inode;
829 
830         if (!sysctl_protected_hardlinks)
831                 return 0;
832 
833         cred = current_cred();
834         inode = link->dentry->d_inode;
835 
836         /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
837          * otherwise, it must be a safe source.
838          */
839         if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
840             capable(CAP_FOWNER))
841                 return 0;
842 
843         audit_log_link_denied("linkat", link);
844         return -EPERM;
845 }
846 
847 static __always_inline int
848 follow_link(struct path *link, struct nameidata *nd, void **p)
849 {
850         struct dentry *dentry = link->dentry;
851         int error;
852         char *s;
853 
854         BUG_ON(nd->flags & LOOKUP_RCU);
855 
856         if (link->mnt == nd->path.mnt)
857                 mntget(link->mnt);
858 
859         error = -ELOOP;
860         if (unlikely(current->total_link_count >= 40))
861                 goto out_put_nd_path;
862 
863         cond_resched();
864         current->total_link_count++;
865 
866         touch_atime(link);
867         nd_set_link(nd, NULL);
868 
869         error = security_inode_follow_link(link->dentry, nd);
870         if (error)
871                 goto out_put_nd_path;
872 
873         nd->last_type = LAST_BIND;
874         *p = dentry->d_inode->i_op->follow_link(dentry, nd);
875         error = PTR_ERR(*p);
876         if (IS_ERR(*p))
877                 goto out_put_nd_path;
878 
879         error = 0;
880         s = nd_get_link(nd);
881         if (s) {
882                 if (unlikely(IS_ERR(s))) {
883                         path_put(&nd->path);
884                         put_link(nd, link, *p);
885                         return PTR_ERR(s);
886                 }
887                 if (*s == '/') {
888                         if (!nd->root.mnt)
889                                 set_root(nd);
890                         path_put(&nd->path);
891                         nd->path = nd->root;
892                         path_get(&nd->root);
893                         nd->flags |= LOOKUP_JUMPED;
894                 }
895                 nd->inode = nd->path.dentry->d_inode;
896                 error = link_path_walk(s, nd);
897                 if (unlikely(error))
898                         put_link(nd, link, *p);
899         }
900 
901         return error;
902 
903 out_put_nd_path:
904         *p = NULL;
905         path_put(&nd->path);
906         path_put(link);
907         return error;
908 }
909 
910 static int follow_up_rcu(struct path *path)
911 {
912         struct mount *mnt = real_mount(path->mnt);
913         struct mount *parent;
914         struct dentry *mountpoint;
915 
916         parent = mnt->mnt_parent;
917         if (&parent->mnt == path->mnt)
918                 return 0;
919         mountpoint = mnt->mnt_mountpoint;
920         path->dentry = mountpoint;
921         path->mnt = &parent->mnt;
922         return 1;
923 }
924 
925 /*
926  * follow_up - Find the mountpoint of path's vfsmount
927  *
928  * Given a path, find the mountpoint of its source file system.
929  * Replace @path with the path of the mountpoint in the parent mount.
930  * Up is towards /.
931  *
932  * Return 1 if we went up a level and 0 if we were already at the
933  * root.
934  */
935 int follow_up(struct path *path)
936 {
937         struct mount *mnt = real_mount(path->mnt);
938         struct mount *parent;
939         struct dentry *mountpoint;
940 
941         read_seqlock_excl(&mount_lock);
942         parent = mnt->mnt_parent;
943         if (parent == mnt) {
944                 read_sequnlock_excl(&mount_lock);
945                 return 0;
946         }
947         mntget(&parent->mnt);
948         mountpoint = dget(mnt->mnt_mountpoint);
949         read_sequnlock_excl(&mount_lock);
950         dput(path->dentry);
951         path->dentry = mountpoint;
952         mntput(path->mnt);
953         path->mnt = &parent->mnt;
954         return 1;
955 }
956 EXPORT_SYMBOL(follow_up);
957 
958 /*
959  * Perform an automount
960  * - return -EISDIR to tell follow_managed() to stop and return the path we
961  *   were called with.
962  */
963 static int follow_automount(struct path *path, unsigned flags,
964                             bool *need_mntput)
965 {
966         struct vfsmount *mnt;
967         int err;
968 
969         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
970                 return -EREMOTE;
971 
972         /* We don't want to mount if someone's just doing a stat -
973          * unless they're stat'ing a directory and appended a '/' to
974          * the name.
975          *
976          * We do, however, want to mount if someone wants to open or
977          * create a file of any type under the mountpoint, wants to
978          * traverse through the mountpoint or wants to open the
979          * mounted directory.  Also, autofs may mark negative dentries
980          * as being automount points.  These will need the attentions
981          * of the daemon to instantiate them before they can be used.
982          */
983         if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
984                      LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
985             path->dentry->d_inode)
986                 return -EISDIR;
987 
988         current->total_link_count++;
989         if (current->total_link_count >= 40)
990                 return -ELOOP;
991 
992         mnt = path->dentry->d_op->d_automount(path);
993         if (IS_ERR(mnt)) {
994                 /*
995                  * The filesystem is allowed to return -EISDIR here to indicate
996                  * it doesn't want to automount.  For instance, autofs would do
997                  * this so that its userspace daemon can mount on this dentry.
998                  *
999                  * However, we can only permit this if it's a terminal point in
1000                  * the path being looked up; if it wasn't then the remainder of
1001                  * the path is inaccessible and we should say so.
1002                  */
1003                 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1004                         return -EREMOTE;
1005                 return PTR_ERR(mnt);
1006         }
1007 
1008         if (!mnt) /* mount collision */
1009                 return 0;
1010 
1011         if (!*need_mntput) {
1012                 /* lock_mount() may release path->mnt on error */
1013                 mntget(path->mnt);
1014                 *need_mntput = true;
1015         }
1016         err = finish_automount(mnt, path);
1017 
1018         switch (err) {
1019         case -EBUSY:
1020                 /* Someone else made a mount here whilst we were busy */
1021                 return 0;
1022         case 0:
1023                 path_put(path);
1024                 path->mnt = mnt;
1025                 path->dentry = dget(mnt->mnt_root);
1026                 return 0;
1027         default:
1028                 return err;
1029         }
1030 
1031 }
1032 
1033 /*
1034  * Handle a dentry that is managed in some way.
1035  * - Flagged for transit management (autofs)
1036  * - Flagged as mountpoint
1037  * - Flagged as automount point
1038  *
1039  * This may only be called in refwalk mode.
1040  *
1041  * Serialization is taken care of in namespace.c
1042  */
1043 static int follow_managed(struct path *path, unsigned flags)
1044 {
1045         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1046         unsigned managed;
1047         bool need_mntput = false;
1048         int ret = 0;
1049 
1050         /* Given that we're not holding a lock here, we retain the value in a
1051          * local variable for each dentry as we look at it so that we don't see
1052          * the components of that value change under us */
1053         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1054                managed &= DCACHE_MANAGED_DENTRY,
1055                unlikely(managed != 0)) {
1056                 /* Allow the filesystem to manage the transit without i_mutex
1057                  * being held. */
1058                 if (managed & DCACHE_MANAGE_TRANSIT) {
1059                         BUG_ON(!path->dentry->d_op);
1060                         BUG_ON(!path->dentry->d_op->d_manage);
1061                         ret = path->dentry->d_op->d_manage(path->dentry, false);
1062                         if (ret < 0)
1063                                 break;
1064                 }
1065 
1066                 /* Transit to a mounted filesystem. */
1067                 if (managed & DCACHE_MOUNTED) {
1068                         struct vfsmount *mounted = lookup_mnt(path);
1069                         if (mounted) {
1070                                 dput(path->dentry);
1071                                 if (need_mntput)
1072                                         mntput(path->mnt);
1073                                 path->mnt = mounted;
1074                                 path->dentry = dget(mounted->mnt_root);
1075                                 need_mntput = true;
1076                                 continue;
1077                         }
1078 
1079                         /* Something is mounted on this dentry in another
1080                          * namespace and/or whatever was mounted there in this
1081                          * namespace got unmounted before lookup_mnt() could
1082                          * get it */
1083                 }
1084 
1085                 /* Handle an automount point */
1086                 if (managed & DCACHE_NEED_AUTOMOUNT) {
1087                         ret = follow_automount(path, flags, &need_mntput);
1088                         if (ret < 0)
1089                                 break;
1090                         continue;
1091                 }
1092 
1093                 /* We didn't change the current path point */
1094                 break;
1095         }
1096 
1097         if (need_mntput && path->mnt == mnt)
1098                 mntput(path->mnt);
1099         if (ret == -EISDIR)
1100                 ret = 0;
1101         return ret < 0 ? ret : need_mntput;
1102 }
1103 
1104 int follow_down_one(struct path *path)
1105 {
1106         struct vfsmount *mounted;
1107 
1108         mounted = lookup_mnt(path);
1109         if (mounted) {
1110                 dput(path->dentry);
1111                 mntput(path->mnt);
1112                 path->mnt = mounted;
1113                 path->dentry = dget(mounted->mnt_root);
1114                 return 1;
1115         }
1116         return 0;
1117 }
1118 EXPORT_SYMBOL(follow_down_one);
1119 
1120 static inline int managed_dentry_rcu(struct dentry *dentry)
1121 {
1122         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1123                 dentry->d_op->d_manage(dentry, true) : 0;
1124 }
1125 
1126 /*
1127  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1128  * we meet a managed dentry that would need blocking.
1129  */
1130 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1131                                struct inode **inode)
1132 {
1133         for (;;) {
1134                 struct mount *mounted;
1135                 /*
1136                  * Don't forget we might have a non-mountpoint managed dentry
1137                  * that wants to block transit.
1138                  */
1139                 switch (managed_dentry_rcu(path->dentry)) {
1140                 case -ECHILD:
1141                 default:
1142                         return false;
1143                 case -EISDIR:
1144                         return true;
1145                 case 0:
1146                         break;
1147                 }
1148 
1149                 if (!d_mountpoint(path->dentry))
1150                         return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1151 
1152                 mounted = __lookup_mnt(path->mnt, path->dentry);
1153                 if (!mounted)
1154                         break;
1155                 path->mnt = &mounted->mnt;
1156                 path->dentry = mounted->mnt.mnt_root;
1157                 nd->flags |= LOOKUP_JUMPED;
1158                 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1159                 /*
1160                  * Update the inode too. We don't need to re-check the
1161                  * dentry sequence number here after this d_inode read,
1162                  * because a mount-point is always pinned.
1163                  */
1164                 *inode = path->dentry->d_inode;
1165         }
1166         return !read_seqretry(&mount_lock, nd->m_seq) &&
1167                 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1168 }
1169 
1170 static int follow_dotdot_rcu(struct nameidata *nd)
1171 {
1172         struct inode *inode = nd->inode;
1173         if (!nd->root.mnt)
1174                 set_root_rcu(nd);
1175 
1176         while (1) {
1177                 if (nd->path.dentry == nd->root.dentry &&
1178                     nd->path.mnt == nd->root.mnt) {
1179                         break;
1180                 }
1181                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1182                         struct dentry *old = nd->path.dentry;
1183                         struct dentry *parent = old->d_parent;
1184                         unsigned seq;
1185 
1186                         inode = parent->d_inode;
1187                         seq = read_seqcount_begin(&parent->d_seq);
1188                         if (read_seqcount_retry(&old->d_seq, nd->seq))
1189                                 goto failed;
1190                         nd->path.dentry = parent;
1191                         nd->seq = seq;
1192                         break;
1193                 }
1194                 if (!follow_up_rcu(&nd->path))
1195                         break;
1196                 inode = nd->path.dentry->d_inode;
1197                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1198         }
1199         while (d_mountpoint(nd->path.dentry)) {
1200                 struct mount *mounted;
1201                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1202                 if (!mounted)
1203                         break;
1204                 nd->path.mnt = &mounted->mnt;
1205                 nd->path.dentry = mounted->mnt.mnt_root;
1206                 inode = nd->path.dentry->d_inode;
1207                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1208                 if (read_seqretry(&mount_lock, nd->m_seq))
1209                         goto failed;
1210         }
1211         nd->inode = inode;
1212         return 0;
1213 
1214 failed:
1215         nd->flags &= ~LOOKUP_RCU;
1216         if (!(nd->flags & LOOKUP_ROOT))
1217                 nd->root.mnt = NULL;
1218         rcu_read_unlock();
1219         return -ECHILD;
1220 }
1221 
1222 /*
1223  * Follow down to the covering mount currently visible to userspace.  At each
1224  * point, the filesystem owning that dentry may be queried as to whether the
1225  * caller is permitted to proceed or not.
1226  */
1227 int follow_down(struct path *path)
1228 {
1229         unsigned managed;
1230         int ret;
1231 
1232         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1233                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1234                 /* Allow the filesystem to manage the transit without i_mutex
1235                  * being held.
1236                  *
1237                  * We indicate to the filesystem if someone is trying to mount
1238                  * something here.  This gives autofs the chance to deny anyone
1239                  * other than its daemon the right to mount on its
1240                  * superstructure.
1241                  *
1242                  * The filesystem may sleep at this point.
1243                  */
1244                 if (managed & DCACHE_MANAGE_TRANSIT) {
1245                         BUG_ON(!path->dentry->d_op);
1246                         BUG_ON(!path->dentry->d_op->d_manage);
1247                         ret = path->dentry->d_op->d_manage(
1248                                 path->dentry, false);
1249                         if (ret < 0)
1250                                 return ret == -EISDIR ? 0 : ret;
1251                 }
1252 
1253                 /* Transit to a mounted filesystem. */
1254                 if (managed & DCACHE_MOUNTED) {
1255                         struct vfsmount *mounted = lookup_mnt(path);
1256                         if (!mounted)
1257                                 break;
1258                         dput(path->dentry);
1259                         mntput(path->mnt);
1260                         path->mnt = mounted;
1261                         path->dentry = dget(mounted->mnt_root);
1262                         continue;
1263                 }
1264 
1265                 /* Don't handle automount points here */
1266                 break;
1267         }
1268         return 0;
1269 }
1270 EXPORT_SYMBOL(follow_down);
1271 
1272 /*
1273  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1274  */
1275 static void follow_mount(struct path *path)
1276 {
1277         while (d_mountpoint(path->dentry)) {
1278                 struct vfsmount *mounted = lookup_mnt(path);
1279                 if (!mounted)
1280                         break;
1281                 dput(path->dentry);
1282                 mntput(path->mnt);
1283                 path->mnt = mounted;
1284                 path->dentry = dget(mounted->mnt_root);
1285         }
1286 }
1287 
1288 static void follow_dotdot(struct nameidata *nd)
1289 {
1290         if (!nd->root.mnt)
1291                 set_root(nd);
1292 
1293         while(1) {
1294                 struct dentry *old = nd->path.dentry;
1295 
1296                 if (nd->path.dentry == nd->root.dentry &&
1297                     nd->path.mnt == nd->root.mnt) {
1298                         break;
1299                 }
1300                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1301                         /* rare case of legitimate dget_parent()... */
1302                         nd->path.dentry = dget_parent(nd->path.dentry);
1303                         dput(old);
1304                         break;
1305                 }
1306                 if (!follow_up(&nd->path))
1307                         break;
1308         }
1309         follow_mount(&nd->path);
1310         nd->inode = nd->path.dentry->d_inode;
1311 }
1312 
1313 /*
1314  * This looks up the name in dcache, possibly revalidates the old dentry and
1315  * allocates a new one if not found or not valid.  In the need_lookup argument
1316  * returns whether i_op->lookup is necessary.
1317  *
1318  * dir->d_inode->i_mutex must be held
1319  */
1320 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1321                                     unsigned int flags, bool *need_lookup)
1322 {
1323         struct dentry *dentry;
1324         int error;
1325 
1326         *need_lookup = false;
1327         dentry = d_lookup(dir, name);
1328         if (dentry) {
1329                 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1330                         error = d_revalidate(dentry, flags);
1331                         if (unlikely(error <= 0)) {
1332                                 if (error < 0) {
1333                                         dput(dentry);
1334                                         return ERR_PTR(error);
1335                                 } else {
1336                                         d_invalidate(dentry);
1337                                         dput(dentry);
1338                                         dentry = NULL;
1339                                 }
1340                         }
1341                 }
1342         }
1343 
1344         if (!dentry) {
1345                 dentry = d_alloc(dir, name);
1346                 if (unlikely(!dentry))
1347                         return ERR_PTR(-ENOMEM);
1348 
1349                 *need_lookup = true;
1350         }
1351         return dentry;
1352 }
1353 
1354 /*
1355  * Call i_op->lookup on the dentry.  The dentry must be negative and
1356  * unhashed.
1357  *
1358  * dir->d_inode->i_mutex must be held
1359  */
1360 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1361                                   unsigned int flags)
1362 {
1363         struct dentry *old;
1364 
1365         /* Don't create child dentry for a dead directory. */
1366         if (unlikely(IS_DEADDIR(dir))) {
1367                 dput(dentry);
1368                 return ERR_PTR(-ENOENT);
1369         }
1370 
1371         old = dir->i_op->lookup(dir, dentry, flags);
1372         if (unlikely(old)) {
1373                 dput(dentry);
1374                 dentry = old;
1375         }
1376         return dentry;
1377 }
1378 
1379 static struct dentry *__lookup_hash(struct qstr *name,
1380                 struct dentry *base, unsigned int flags)
1381 {
1382         bool need_lookup;
1383         struct dentry *dentry;
1384 
1385         dentry = lookup_dcache(name, base, flags, &need_lookup);
1386         if (!need_lookup)
1387                 return dentry;
1388 
1389         return lookup_real(base->d_inode, dentry, flags);
1390 }
1391 
1392 /*
1393  *  It's more convoluted than I'd like it to be, but... it's still fairly
1394  *  small and for now I'd prefer to have fast path as straight as possible.
1395  *  It _is_ time-critical.
1396  */
1397 static int lookup_fast(struct nameidata *nd,
1398                        struct path *path, struct inode **inode)
1399 {
1400         struct vfsmount *mnt = nd->path.mnt;
1401         struct dentry *dentry, *parent = nd->path.dentry;
1402         int need_reval = 1;
1403         int status = 1;
1404         int err;
1405 
1406         /*
1407          * Rename seqlock is not required here because in the off chance
1408          * of a false negative due to a concurrent rename, we're going to
1409          * do the non-racy lookup, below.
1410          */
1411         if (nd->flags & LOOKUP_RCU) {
1412                 unsigned seq;
1413                 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1414                 if (!dentry)
1415                         goto unlazy;
1416 
1417                 /*
1418                  * This sequence count validates that the inode matches
1419                  * the dentry name information from lookup.
1420                  */
1421                 *inode = dentry->d_inode;
1422                 if (read_seqcount_retry(&dentry->d_seq, seq))
1423                         return -ECHILD;
1424 
1425                 /*
1426                  * This sequence count validates that the parent had no
1427                  * changes while we did the lookup of the dentry above.
1428                  *
1429                  * The memory barrier in read_seqcount_begin of child is
1430                  *  enough, we can use __read_seqcount_retry here.
1431                  */
1432                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1433                         return -ECHILD;
1434                 nd->seq = seq;
1435 
1436                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1437                         status = d_revalidate(dentry, nd->flags);
1438                         if (unlikely(status <= 0)) {
1439                                 if (status != -ECHILD)
1440                                         need_reval = 0;
1441                                 goto unlazy;
1442                         }
1443                 }
1444                 path->mnt = mnt;
1445                 path->dentry = dentry;
1446                 if (likely(__follow_mount_rcu(nd, path, inode)))
1447                         return 0;
1448 unlazy:
1449                 if (unlazy_walk(nd, dentry))
1450                         return -ECHILD;
1451         } else {
1452                 dentry = __d_lookup(parent, &nd->last);
1453         }
1454 
1455         if (unlikely(!dentry))
1456                 goto need_lookup;
1457 
1458         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1459                 status = d_revalidate(dentry, nd->flags);
1460         if (unlikely(status <= 0)) {
1461                 if (status < 0) {
1462                         dput(dentry);
1463                         return status;
1464                 }
1465                 d_invalidate(dentry);
1466                 dput(dentry);
1467                 goto need_lookup;
1468         }
1469 
1470         path->mnt = mnt;
1471         path->dentry = dentry;
1472         err = follow_managed(path, nd->flags);
1473         if (unlikely(err < 0)) {
1474                 path_put_conditional(path, nd);
1475                 return err;
1476         }
1477         if (err)
1478                 nd->flags |= LOOKUP_JUMPED;
1479         *inode = path->dentry->d_inode;
1480         return 0;
1481 
1482 need_lookup:
1483         return 1;
1484 }
1485 
1486 /* Fast lookup failed, do it the slow way */
1487 static int lookup_slow(struct nameidata *nd, struct path *path)
1488 {
1489         struct dentry *dentry, *parent;
1490         int err;
1491 
1492         parent = nd->path.dentry;
1493         BUG_ON(nd->inode != parent->d_inode);
1494 
1495         mutex_lock(&parent->d_inode->i_mutex);
1496         dentry = __lookup_hash(&nd->last, parent, nd->flags);
1497         mutex_unlock(&parent->d_inode->i_mutex);
1498         if (IS_ERR(dentry))
1499                 return PTR_ERR(dentry);
1500         path->mnt = nd->path.mnt;
1501         path->dentry = dentry;
1502         err = follow_managed(path, nd->flags);
1503         if (unlikely(err < 0)) {
1504                 path_put_conditional(path, nd);
1505                 return err;
1506         }
1507         if (err)
1508                 nd->flags |= LOOKUP_JUMPED;
1509         return 0;
1510 }
1511 
1512 static inline int may_lookup(struct nameidata *nd)
1513 {
1514         if (nd->flags & LOOKUP_RCU) {
1515                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1516                 if (err != -ECHILD)
1517                         return err;
1518                 if (unlazy_walk(nd, NULL))
1519                         return -ECHILD;
1520         }
1521         return inode_permission(nd->inode, MAY_EXEC);
1522 }
1523 
1524 static inline int handle_dots(struct nameidata *nd, int type)
1525 {
1526         if (type == LAST_DOTDOT) {
1527                 if (nd->flags & LOOKUP_RCU) {
1528                         if (follow_dotdot_rcu(nd))
1529                                 return -ECHILD;
1530                 } else
1531                         follow_dotdot(nd);
1532         }
1533         return 0;
1534 }
1535 
1536 static void terminate_walk(struct nameidata *nd)
1537 {
1538         if (!(nd->flags & LOOKUP_RCU)) {
1539                 path_put(&nd->path);
1540         } else {
1541                 nd->flags &= ~LOOKUP_RCU;
1542                 if (!(nd->flags & LOOKUP_ROOT))
1543                         nd->root.mnt = NULL;
1544                 rcu_read_unlock();
1545         }
1546 }
1547 
1548 /*
1549  * Do we need to follow links? We _really_ want to be able
1550  * to do this check without having to look at inode->i_op,
1551  * so we keep a cache of "no, this doesn't need follow_link"
1552  * for the common case.
1553  */
1554 static inline int should_follow_link(struct dentry *dentry, int follow)
1555 {
1556         return unlikely(d_is_symlink(dentry)) ? follow : 0;
1557 }
1558 
1559 static inline int walk_component(struct nameidata *nd, struct path *path,
1560                 int follow)
1561 {
1562         struct inode *inode;
1563         int err;
1564         /*
1565          * "." and ".." are special - ".." especially so because it has
1566          * to be able to know about the current root directory and
1567          * parent relationships.
1568          */
1569         if (unlikely(nd->last_type != LAST_NORM))
1570                 return handle_dots(nd, nd->last_type);
1571         err = lookup_fast(nd, path, &inode);
1572         if (unlikely(err)) {
1573                 if (err < 0)
1574                         goto out_err;
1575 
1576                 err = lookup_slow(nd, path);
1577                 if (err < 0)
1578                         goto out_err;
1579 
1580                 inode = path->dentry->d_inode;
1581         }
1582         err = -ENOENT;
1583         if (!inode || d_is_negative(path->dentry))
1584                 goto out_path_put;
1585 
1586         if (should_follow_link(path->dentry, follow)) {
1587                 if (nd->flags & LOOKUP_RCU) {
1588                         if (unlikely(unlazy_walk(nd, path->dentry))) {
1589                                 err = -ECHILD;
1590                                 goto out_err;
1591                         }
1592                 }
1593                 BUG_ON(inode != path->dentry->d_inode);
1594                 return 1;
1595         }
1596         path_to_nameidata(path, nd);
1597         nd->inode = inode;
1598         return 0;
1599 
1600 out_path_put:
1601         path_to_nameidata(path, nd);
1602 out_err:
1603         terminate_walk(nd);
1604         return err;
1605 }
1606 
1607 /*
1608  * This limits recursive symlink follows to 8, while
1609  * limiting consecutive symlinks to 40.
1610  *
1611  * Without that kind of total limit, nasty chains of consecutive
1612  * symlinks can cause almost arbitrarily long lookups.
1613  */
1614 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1615 {
1616         int res;
1617 
1618         if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1619                 path_put_conditional(path, nd);
1620                 path_put(&nd->path);
1621                 return -ELOOP;
1622         }
1623         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1624 
1625         nd->depth++;
1626         current->link_count++;
1627 
1628         do {
1629                 struct path link = *path;
1630                 void *cookie;
1631 
1632                 res = follow_link(&link, nd, &cookie);
1633                 if (res)
1634                         break;
1635                 res = walk_component(nd, path, LOOKUP_FOLLOW);
1636                 put_link(nd, &link, cookie);
1637         } while (res > 0);
1638 
1639         current->link_count--;
1640         nd->depth--;
1641         return res;
1642 }
1643 
1644 /*
1645  * We can do the critical dentry name comparison and hashing
1646  * operations one word at a time, but we are limited to:
1647  *
1648  * - Architectures with fast unaligned word accesses. We could
1649  *   do a "get_unaligned()" if this helps and is sufficiently
1650  *   fast.
1651  *
1652  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1653  *   do not trap on the (extremely unlikely) case of a page
1654  *   crossing operation.
1655  *
1656  * - Furthermore, we need an efficient 64-bit compile for the
1657  *   64-bit case in order to generate the "number of bytes in
1658  *   the final mask". Again, that could be replaced with a
1659  *   efficient population count instruction or similar.
1660  */
1661 #ifdef CONFIG_DCACHE_WORD_ACCESS
1662 
1663 #include <asm/word-at-a-time.h>
1664 
1665 #ifdef CONFIG_64BIT
1666 
1667 static inline unsigned int fold_hash(unsigned long hash)
1668 {
1669         return hash_64(hash, 32);
1670 }
1671 
1672 #else   /* 32-bit case */
1673 
1674 #define fold_hash(x) (x)
1675 
1676 #endif
1677 
1678 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1679 {
1680         unsigned long a, mask;
1681         unsigned long hash = 0;
1682 
1683         for (;;) {
1684                 a = load_unaligned_zeropad(name);
1685                 if (len < sizeof(unsigned long))
1686                         break;
1687                 hash += a;
1688                 hash *= 9;
1689                 name += sizeof(unsigned long);
1690                 len -= sizeof(unsigned long);
1691                 if (!len)
1692                         goto done;
1693         }
1694         mask = bytemask_from_count(len);
1695         hash += mask & a;
1696 done:
1697         return fold_hash(hash);
1698 }
1699 EXPORT_SYMBOL(full_name_hash);
1700 
1701 /*
1702  * Calculate the length and hash of the path component, and
1703  * return the "hash_len" as the result.
1704  */
1705 static inline u64 hash_name(const char *name)
1706 {
1707         unsigned long a, b, adata, bdata, mask, hash, len;
1708         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1709 
1710         hash = a = 0;
1711         len = -sizeof(unsigned long);
1712         do {
1713                 hash = (hash + a) * 9;
1714                 len += sizeof(unsigned long);
1715                 a = load_unaligned_zeropad(name+len);
1716                 b = a ^ REPEAT_BYTE('/');
1717         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1718 
1719         adata = prep_zero_mask(a, adata, &constants);
1720         bdata = prep_zero_mask(b, bdata, &constants);
1721 
1722         mask = create_zero_mask(adata | bdata);
1723 
1724         hash += a & zero_bytemask(mask);
1725         len += find_zero(mask);
1726         return hashlen_create(fold_hash(hash), len);
1727 }
1728 
1729 #else
1730 
1731 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1732 {
1733         unsigned long hash = init_name_hash();
1734         while (len--)
1735                 hash = partial_name_hash(*name++, hash);
1736         return end_name_hash(hash);
1737 }
1738 EXPORT_SYMBOL(full_name_hash);
1739 
1740 /*
1741  * We know there's a real path component here of at least
1742  * one character.
1743  */
1744 static inline u64 hash_name(const char *name)
1745 {
1746         unsigned long hash = init_name_hash();
1747         unsigned long len = 0, c;
1748 
1749         c = (unsigned char)*name;
1750         do {
1751                 len++;
1752                 hash = partial_name_hash(c, hash);
1753                 c = (unsigned char)name[len];
1754         } while (c && c != '/');
1755         return hashlen_create(end_name_hash(hash), len);
1756 }
1757 
1758 #endif
1759 
1760 /*
1761  * Name resolution.
1762  * This is the basic name resolution function, turning a pathname into
1763  * the final dentry. We expect 'base' to be positive and a directory.
1764  *
1765  * Returns 0 and nd will have valid dentry and mnt on success.
1766  * Returns error and drops reference to input namei data on failure.
1767  */
1768 static int link_path_walk(const char *name, struct nameidata *nd)
1769 {
1770         struct path next;
1771         int err;
1772         
1773         while (*name=='/')
1774                 name++;
1775         if (!*name)
1776                 return 0;
1777 
1778         /* At this point we know we have a real path component. */
1779         for(;;) {
1780                 u64 hash_len;
1781                 int type;
1782 
1783                 err = may_lookup(nd);
1784                 if (err)
1785                         break;
1786 
1787                 hash_len = hash_name(name);
1788 
1789                 type = LAST_NORM;
1790                 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1791                         case 2:
1792                                 if (name[1] == '.') {
1793                                         type = LAST_DOTDOT;
1794                                         nd->flags |= LOOKUP_JUMPED;
1795                                 }
1796                                 break;
1797                         case 1:
1798                                 type = LAST_DOT;
1799                 }
1800                 if (likely(type == LAST_NORM)) {
1801                         struct dentry *parent = nd->path.dentry;
1802                         nd->flags &= ~LOOKUP_JUMPED;
1803                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1804                                 struct qstr this = { { .hash_len = hash_len }, .name = name };
1805                                 err = parent->d_op->d_hash(parent, &this);
1806                                 if (err < 0)
1807                                         break;
1808                                 hash_len = this.hash_len;
1809                                 name = this.name;
1810                         }
1811                 }
1812 
1813                 nd->last.hash_len = hash_len;
1814                 nd->last.name = name;
1815                 nd->last_type = type;
1816 
1817                 name += hashlen_len(hash_len);
1818                 if (!*name)
1819                         return 0;
1820                 /*
1821                  * If it wasn't NUL, we know it was '/'. Skip that
1822                  * slash, and continue until no more slashes.
1823                  */
1824                 do {
1825                         name++;
1826                 } while (unlikely(*name == '/'));
1827                 if (!*name)
1828                         return 0;
1829 
1830                 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1831                 if (err < 0)
1832                         return err;
1833 
1834                 if (err) {
1835                         err = nested_symlink(&next, nd);
1836                         if (err)
1837                                 return err;
1838                 }
1839                 if (!d_can_lookup(nd->path.dentry)) {
1840                         err = -ENOTDIR; 
1841                         break;
1842                 }
1843         }
1844         terminate_walk(nd);
1845         return err;
1846 }
1847 
1848 static int path_init(int dfd, const char *name, unsigned int flags,
1849                      struct nameidata *nd)
1850 {
1851         int retval = 0;
1852 
1853         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1854         nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1855         nd->depth = 0;
1856         nd->base = NULL;
1857         if (flags & LOOKUP_ROOT) {
1858                 struct dentry *root = nd->root.dentry;
1859                 struct inode *inode = root->d_inode;
1860                 if (*name) {
1861                         if (!d_can_lookup(root))
1862                                 return -ENOTDIR;
1863                         retval = inode_permission(inode, MAY_EXEC);
1864                         if (retval)
1865                                 return retval;
1866                 }
1867                 nd->path = nd->root;
1868                 nd->inode = inode;
1869                 if (flags & LOOKUP_RCU) {
1870                         rcu_read_lock();
1871                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1872                         nd->m_seq = read_seqbegin(&mount_lock);
1873                 } else {
1874                         path_get(&nd->path);
1875                 }
1876                 goto done;
1877         }
1878 
1879         nd->root.mnt = NULL;
1880 
1881         nd->m_seq = read_seqbegin(&mount_lock);
1882         if (*name=='/') {
1883                 if (flags & LOOKUP_RCU) {
1884                         rcu_read_lock();
1885                         nd->seq = set_root_rcu(nd);
1886                 } else {
1887                         set_root(nd);
1888                         path_get(&nd->root);
1889                 }
1890                 nd->path = nd->root;
1891         } else if (dfd == AT_FDCWD) {
1892                 if (flags & LOOKUP_RCU) {
1893                         struct fs_struct *fs = current->fs;
1894                         unsigned seq;
1895 
1896                         rcu_read_lock();
1897 
1898                         do {
1899                                 seq = read_seqcount_begin(&fs->seq);
1900                                 nd->path = fs->pwd;
1901                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1902                         } while (read_seqcount_retry(&fs->seq, seq));
1903                 } else {
1904                         get_fs_pwd(current->fs, &nd->path);
1905                 }
1906         } else {
1907                 /* Caller must check execute permissions on the starting path component */
1908                 struct fd f = fdget_raw(dfd);
1909                 struct dentry *dentry;
1910 
1911                 if (!f.file)
1912                         return -EBADF;
1913 
1914                 dentry = f.file->f_path.dentry;
1915 
1916                 if (*name) {
1917                         if (!d_can_lookup(dentry)) {
1918                                 fdput(f);
1919                                 return -ENOTDIR;
1920                         }
1921                 }
1922 
1923                 nd->path = f.file->f_path;
1924                 if (flags & LOOKUP_RCU) {
1925                         if (f.flags & FDPUT_FPUT)
1926                                 nd->base = f.file;
1927                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1928                         rcu_read_lock();
1929                 } else {
1930                         path_get(&nd->path);
1931                         fdput(f);
1932                 }
1933         }
1934 
1935         nd->inode = nd->path.dentry->d_inode;
1936         if (!(flags & LOOKUP_RCU))
1937                 goto done;
1938         if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1939                 goto done;
1940         if (!(nd->flags & LOOKUP_ROOT))
1941                 nd->root.mnt = NULL;
1942         rcu_read_unlock();
1943         return -ECHILD;
1944 done:
1945         current->total_link_count = 0;
1946         return link_path_walk(name, nd);
1947 }
1948 
1949 static void path_cleanup(struct nameidata *nd)
1950 {
1951         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1952                 path_put(&nd->root);
1953                 nd->root.mnt = NULL;
1954         }
1955         if (unlikely(nd->base))
1956                 fput(nd->base);
1957 }
1958 
1959 static inline int lookup_last(struct nameidata *nd, struct path *path)
1960 {
1961         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1962                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1963 
1964         nd->flags &= ~LOOKUP_PARENT;
1965         return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1966 }
1967 
1968 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1969 static int path_lookupat(int dfd, const char *name,
1970                                 unsigned int flags, struct nameidata *nd)
1971 {
1972         struct path path;
1973         int err;
1974 
1975         /*
1976          * Path walking is largely split up into 2 different synchronisation
1977          * schemes, rcu-walk and ref-walk (explained in
1978          * Documentation/filesystems/path-lookup.txt). These share much of the
1979          * path walk code, but some things particularly setup, cleanup, and
1980          * following mounts are sufficiently divergent that functions are
1981          * duplicated. Typically there is a function foo(), and its RCU
1982          * analogue, foo_rcu().
1983          *
1984          * -ECHILD is the error number of choice (just to avoid clashes) that
1985          * is returned if some aspect of an rcu-walk fails. Such an error must
1986          * be handled by restarting a traditional ref-walk (which will always
1987          * be able to complete).
1988          */
1989         err = path_init(dfd, name, flags, nd);
1990         if (!err && !(flags & LOOKUP_PARENT)) {
1991                 err = lookup_last(nd, &path);
1992                 while (err > 0) {
1993                         void *cookie;
1994                         struct path link = path;
1995                         err = may_follow_link(&link, nd);
1996                         if (unlikely(err))
1997                                 break;
1998                         nd->flags |= LOOKUP_PARENT;
1999                         err = follow_link(&link, nd, &cookie);
2000                         if (err)
2001                                 break;
2002                         err = lookup_last(nd, &path);
2003                         put_link(nd, &link, cookie);
2004                 }
2005         }
2006 
2007         if (!err)
2008                 err = complete_walk(nd);
2009 
2010         if (!err && nd->flags & LOOKUP_DIRECTORY) {
2011                 if (!d_can_lookup(nd->path.dentry)) {
2012                         path_put(&nd->path);
2013                         err = -ENOTDIR;
2014                 }
2015         }
2016 
2017         path_cleanup(nd);
2018         return err;
2019 }
2020 
2021 static int filename_lookup(int dfd, struct filename *name,
2022                                 unsigned int flags, struct nameidata *nd)
2023 {
2024         int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2025         if (unlikely(retval == -ECHILD))
2026                 retval = path_lookupat(dfd, name->name, flags, nd);
2027         if (unlikely(retval == -ESTALE))
2028                 retval = path_lookupat(dfd, name->name,
2029                                                 flags | LOOKUP_REVAL, nd);
2030 
2031         if (likely(!retval))
2032                 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2033         return retval;
2034 }
2035 
2036 static int do_path_lookup(int dfd, const char *name,
2037                                 unsigned int flags, struct nameidata *nd)
2038 {
2039         struct filename filename = { .name = name };
2040 
2041         return filename_lookup(dfd, &filename, flags, nd);
2042 }
2043 
2044 /* does lookup, returns the object with parent locked */
2045 struct dentry *kern_path_locked(const char *name, struct path *path)
2046 {
2047         struct nameidata nd;
2048         struct dentry *d;
2049         int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2050         if (err)
2051                 return ERR_PTR(err);
2052         if (nd.last_type != LAST_NORM) {
2053                 path_put(&nd.path);
2054                 return ERR_PTR(-EINVAL);
2055         }
2056         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2057         d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2058         if (IS_ERR(d)) {
2059                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2060                 path_put(&nd.path);
2061                 return d;
2062         }
2063         *path = nd.path;
2064         return d;
2065 }
2066 
2067 int kern_path(const char *name, unsigned int flags, struct path *path)
2068 {
2069         struct nameidata nd;
2070         int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2071         if (!res)
2072                 *path = nd.path;
2073         return res;
2074 }
2075 EXPORT_SYMBOL(kern_path);
2076 
2077 /**
2078  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2079  * @dentry:  pointer to dentry of the base directory
2080  * @mnt: pointer to vfs mount of the base directory
2081  * @name: pointer to file name
2082  * @flags: lookup flags
2083  * @path: pointer to struct path to fill
2084  */
2085 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2086                     const char *name, unsigned int flags,
2087                     struct path *path)
2088 {
2089         struct nameidata nd;
2090         int err;
2091         nd.root.dentry = dentry;
2092         nd.root.mnt = mnt;
2093         BUG_ON(flags & LOOKUP_PARENT);
2094         /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2095         err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2096         if (!err)
2097                 *path = nd.path;
2098         return err;
2099 }
2100 EXPORT_SYMBOL(vfs_path_lookup);
2101 
2102 /*
2103  * Restricted form of lookup. Doesn't follow links, single-component only,
2104  * needs parent already locked. Doesn't follow mounts.
2105  * SMP-safe.
2106  */
2107 static struct dentry *lookup_hash(struct nameidata *nd)
2108 {
2109         return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2110 }
2111 
2112 /**
2113  * lookup_one_len - filesystem helper to lookup single pathname component
2114  * @name:       pathname component to lookup
2115  * @base:       base directory to lookup from
2116  * @len:        maximum length @len should be interpreted to
2117  *
2118  * Note that this routine is purely a helper for filesystem usage and should
2119  * not be called by generic code.  Also note that by using this function the
2120  * nameidata argument is passed to the filesystem methods and a filesystem
2121  * using this helper needs to be prepared for that.
2122  */
2123 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2124 {
2125         struct qstr this;
2126         unsigned int c;
2127         int err;
2128 
2129         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2130 
2131         this.name = name;
2132         this.len = len;
2133         this.hash = full_name_hash(name, len);
2134         if (!len)
2135                 return ERR_PTR(-EACCES);
2136 
2137         if (unlikely(name[0] == '.')) {
2138                 if (len < 2 || (len == 2 && name[1] == '.'))
2139                         return ERR_PTR(-EACCES);
2140         }
2141 
2142         while (len--) {
2143                 c = *(const unsigned char *)name++;
2144                 if (c == '/' || c == '\0')
2145                         return ERR_PTR(-EACCES);
2146         }
2147         /*
2148          * See if the low-level filesystem might want
2149          * to use its own hash..
2150          */
2151         if (base->d_flags & DCACHE_OP_HASH) {
2152                 int err = base->d_op->d_hash(base, &this);
2153                 if (err < 0)
2154                         return ERR_PTR(err);
2155         }
2156 
2157         err = inode_permission(base->d_inode, MAY_EXEC);
2158         if (err)
2159                 return ERR_PTR(err);
2160 
2161         return __lookup_hash(&this, base, 0);
2162 }
2163 EXPORT_SYMBOL(lookup_one_len);
2164 
2165 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2166                  struct path *path, int *empty)
2167 {
2168         struct nameidata nd;
2169         struct filename *tmp = getname_flags(name, flags, empty);
2170         int err = PTR_ERR(tmp);
2171         if (!IS_ERR(tmp)) {
2172 
2173                 BUG_ON(flags & LOOKUP_PARENT);
2174 
2175                 err = filename_lookup(dfd, tmp, flags, &nd);
2176                 putname(tmp);
2177                 if (!err)
2178                         *path = nd.path;
2179         }
2180         return err;
2181 }
2182 
2183 int user_path_at(int dfd, const char __user *name, unsigned flags,
2184                  struct path *path)
2185 {
2186         return user_path_at_empty(dfd, name, flags, path, NULL);
2187 }
2188 EXPORT_SYMBOL(user_path_at);
2189 
2190 /*
2191  * NB: most callers don't do anything directly with the reference to the
2192  *     to struct filename, but the nd->last pointer points into the name string
2193  *     allocated by getname. So we must hold the reference to it until all
2194  *     path-walking is complete.
2195  */
2196 static struct filename *
2197 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2198                  unsigned int flags)
2199 {
2200         struct filename *s = getname(path);
2201         int error;
2202 
2203         /* only LOOKUP_REVAL is allowed in extra flags */
2204         flags &= LOOKUP_REVAL;
2205 
2206         if (IS_ERR(s))
2207                 return s;
2208 
2209         error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2210         if (error) {
2211                 putname(s);
2212                 return ERR_PTR(error);
2213         }
2214 
2215         return s;
2216 }
2217 
2218 /**
2219  * mountpoint_last - look up last component for umount
2220  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2221  * @path: pointer to container for result
2222  *
2223  * This is a special lookup_last function just for umount. In this case, we
2224  * need to resolve the path without doing any revalidation.
2225  *
2226  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2227  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2228  * in almost all cases, this lookup will be served out of the dcache. The only
2229  * cases where it won't are if nd->last refers to a symlink or the path is
2230  * bogus and it doesn't exist.
2231  *
2232  * Returns:
2233  * -error: if there was an error during lookup. This includes -ENOENT if the
2234  *         lookup found a negative dentry. The nd->path reference will also be
2235  *         put in this case.
2236  *
2237  * 0:      if we successfully resolved nd->path and found it to not to be a
2238  *         symlink that needs to be followed. "path" will also be populated.
2239  *         The nd->path reference will also be put.
2240  *
2241  * 1:      if we successfully resolved nd->last and found it to be a symlink
2242  *         that needs to be followed. "path" will be populated with the path
2243  *         to the link, and nd->path will *not* be put.
2244  */
2245 static int
2246 mountpoint_last(struct nameidata *nd, struct path *path)
2247 {
2248         int error = 0;
2249         struct dentry *dentry;
2250         struct dentry *dir = nd->path.dentry;
2251 
2252         /* If we're in rcuwalk, drop out of it to handle last component */
2253         if (nd->flags & LOOKUP_RCU) {
2254                 if (unlazy_walk(nd, NULL)) {
2255                         error = -ECHILD;
2256                         goto out;
2257                 }
2258         }
2259 
2260         nd->flags &= ~LOOKUP_PARENT;
2261 
2262         if (unlikely(nd->last_type != LAST_NORM)) {
2263                 error = handle_dots(nd, nd->last_type);
2264                 if (error)
2265                         goto out;
2266                 dentry = dget(nd->path.dentry);
2267                 goto done;
2268         }
2269 
2270         mutex_lock(&dir->d_inode->i_mutex);
2271         dentry = d_lookup(dir, &nd->last);
2272         if (!dentry) {
2273                 /*
2274                  * No cached dentry. Mounted dentries are pinned in the cache,
2275                  * so that means that this dentry is probably a symlink or the
2276                  * path doesn't actually point to a mounted dentry.
2277                  */
2278                 dentry = d_alloc(dir, &nd->last);
2279                 if (!dentry) {
2280                         error = -ENOMEM;
2281                         mutex_unlock(&dir->d_inode->i_mutex);
2282                         goto out;
2283                 }
2284                 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2285                 error = PTR_ERR(dentry);
2286                 if (IS_ERR(dentry)) {
2287                         mutex_unlock(&dir->d_inode->i_mutex);
2288                         goto out;
2289                 }
2290         }
2291         mutex_unlock(&dir->d_inode->i_mutex);
2292 
2293 done:
2294         if (!dentry->d_inode || d_is_negative(dentry)) {
2295                 error = -ENOENT;
2296                 dput(dentry);
2297                 goto out;
2298         }
2299         path->dentry = dentry;
2300         path->mnt = nd->path.mnt;
2301         if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2302                 return 1;
2303         mntget(path->mnt);
2304         follow_mount(path);
2305         error = 0;
2306 out:
2307         terminate_walk(nd);
2308         return error;
2309 }
2310 
2311 /**
2312  * path_mountpoint - look up a path to be umounted
2313  * @dfd:        directory file descriptor to start walk from
2314  * @name:       full pathname to walk
2315  * @path:       pointer to container for result
2316  * @flags:      lookup flags
2317  *
2318  * Look up the given name, but don't attempt to revalidate the last component.
2319  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2320  */
2321 static int
2322 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2323 {
2324         struct nameidata nd;
2325         int err;
2326 
2327         err = path_init(dfd, name, flags, &nd);
2328         if (unlikely(err))
2329                 goto out;
2330 
2331         err = mountpoint_last(&nd, path);
2332         while (err > 0) {
2333                 void *cookie;
2334                 struct path link = *path;
2335                 err = may_follow_link(&link, &nd);
2336                 if (unlikely(err))
2337                         break;
2338                 nd.flags |= LOOKUP_PARENT;
2339                 err = follow_link(&link, &nd, &cookie);
2340                 if (err)
2341                         break;
2342                 err = mountpoint_last(&nd, path);
2343                 put_link(&nd, &link, cookie);
2344         }
2345 out:
2346         path_cleanup(&nd);
2347         return err;
2348 }
2349 
2350 static int
2351 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2352                         unsigned int flags)
2353 {
2354         int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2355         if (unlikely(error == -ECHILD))
2356                 error = path_mountpoint(dfd, s->name, path, flags);
2357         if (unlikely(error == -ESTALE))
2358                 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2359         if (likely(!error))
2360                 audit_inode(s, path->dentry, 0);
2361         return error;
2362 }
2363 
2364 /**
2365  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2366  * @dfd:        directory file descriptor
2367  * @name:       pathname from userland
2368  * @flags:      lookup flags
2369  * @path:       pointer to container to hold result
2370  *
2371  * A umount is a special case for path walking. We're not actually interested
2372  * in the inode in this situation, and ESTALE errors can be a problem. We
2373  * simply want track down the dentry and vfsmount attached at the mountpoint
2374  * and avoid revalidating the last component.
2375  *
2376  * Returns 0 and populates "path" on success.
2377  */
2378 int
2379 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2380                         struct path *path)
2381 {
2382         struct filename *s = getname(name);
2383         int error;
2384         if (IS_ERR(s))
2385                 return PTR_ERR(s);
2386         error = filename_mountpoint(dfd, s, path, flags);
2387         putname(s);
2388         return error;
2389 }
2390 
2391 int
2392 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2393                         unsigned int flags)
2394 {
2395         struct filename s = {.name = name};
2396         return filename_mountpoint(dfd, &s, path, flags);
2397 }
2398 EXPORT_SYMBOL(kern_path_mountpoint);
2399 
2400 int __check_sticky(struct inode *dir, struct inode *inode)
2401 {
2402         kuid_t fsuid = current_fsuid();
2403 
2404         if (uid_eq(inode->i_uid, fsuid))
2405                 return 0;
2406         if (uid_eq(dir->i_uid, fsuid))
2407                 return 0;
2408         return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2409 }
2410 EXPORT_SYMBOL(__check_sticky);
2411 
2412 /*
2413  *      Check whether we can remove a link victim from directory dir, check
2414  *  whether the type of victim is right.
2415  *  1. We can't do it if dir is read-only (done in permission())
2416  *  2. We should have write and exec permissions on dir
2417  *  3. We can't remove anything from append-only dir
2418  *  4. We can't do anything with immutable dir (done in permission())
2419  *  5. If the sticky bit on dir is set we should either
2420  *      a. be owner of dir, or
2421  *      b. be owner of victim, or
2422  *      c. have CAP_FOWNER capability
2423  *  6. If the victim is append-only or immutable we can't do antyhing with
2424  *     links pointing to it.
2425  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2426  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2427  *  9. We can't remove a root or mountpoint.
2428  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2429  *     nfs_async_unlink().
2430  */
2431 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2432 {
2433         struct inode *inode = victim->d_inode;
2434         int error;
2435 
2436         if (d_is_negative(victim))
2437                 return -ENOENT;
2438         BUG_ON(!inode);
2439 
2440         BUG_ON(victim->d_parent->d_inode != dir);
2441         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2442 
2443         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2444         if (error)
2445                 return error;
2446         if (IS_APPEND(dir))
2447                 return -EPERM;
2448 
2449         if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2450             IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2451                 return -EPERM;
2452         if (isdir) {
2453                 if (!d_is_dir(victim))
2454                         return -ENOTDIR;
2455                 if (IS_ROOT(victim))
2456                         return -EBUSY;
2457         } else if (d_is_dir(victim))
2458                 return -EISDIR;
2459         if (IS_DEADDIR(dir))
2460                 return -ENOENT;
2461         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2462                 return -EBUSY;
2463         return 0;
2464 }
2465 
2466 /*      Check whether we can create an object with dentry child in directory
2467  *  dir.
2468  *  1. We can't do it if child already exists (open has special treatment for
2469  *     this case, but since we are inlined it's OK)
2470  *  2. We can't do it if dir is read-only (done in permission())
2471  *  3. We should have write and exec permissions on dir
2472  *  4. We can't do it if dir is immutable (done in permission())
2473  */
2474 static inline int may_create(struct inode *dir, struct dentry *child)
2475 {
2476         audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2477         if (child->d_inode)
2478                 return -EEXIST;
2479         if (IS_DEADDIR(dir))
2480                 return -ENOENT;
2481         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2482 }
2483 
2484 /*
2485  * p1 and p2 should be directories on the same fs.
2486  */
2487 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2488 {
2489         struct dentry *p;
2490 
2491         if (p1 == p2) {
2492                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2493                 return NULL;
2494         }
2495 
2496         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2497 
2498         p = d_ancestor(p2, p1);
2499         if (p) {
2500                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2501                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2502                 return p;
2503         }
2504 
2505         p = d_ancestor(p1, p2);
2506         if (p) {
2507                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2508                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2509                 return p;
2510         }
2511 
2512         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2513         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2514         return NULL;
2515 }
2516 EXPORT_SYMBOL(lock_rename);
2517 
2518 void unlock_rename(struct dentry *p1, struct dentry *p2)
2519 {
2520         mutex_unlock(&p1->d_inode->i_mutex);
2521         if (p1 != p2) {
2522                 mutex_unlock(&p2->d_inode->i_mutex);
2523                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2524         }
2525 }
2526 EXPORT_SYMBOL(unlock_rename);
2527 
2528 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2529                 bool want_excl)
2530 {
2531         int error = may_create(dir, dentry);
2532         if (error)
2533                 return error;
2534 
2535         if (!dir->i_op->create)
2536                 return -EACCES; /* shouldn't it be ENOSYS? */
2537         mode &= S_IALLUGO;
2538         mode |= S_IFREG;
2539         error = security_inode_create(dir, dentry, mode);
2540         if (error)
2541                 return error;
2542         error = dir->i_op->create(dir, dentry, mode, want_excl);
2543         if (!error)
2544                 fsnotify_create(dir, dentry);
2545         return error;
2546 }
2547 EXPORT_SYMBOL(vfs_create);
2548 
2549 static int may_open(struct path *path, int acc_mode, int flag)
2550 {
2551         struct dentry *dentry = path->dentry;
2552         struct inode *inode = dentry->d_inode;
2553         int error;
2554 
2555         /* O_PATH? */
2556         if (!acc_mode)
2557                 return 0;
2558 
2559         if (!inode)
2560                 return -ENOENT;
2561 
2562         switch (inode->i_mode & S_IFMT) {
2563         case S_IFLNK:
2564                 return -ELOOP;
2565         case S_IFDIR:
2566                 if (acc_mode & MAY_WRITE)
2567                         return -EISDIR;
2568                 break;
2569         case S_IFBLK:
2570         case S_IFCHR:
2571                 if (path->mnt->mnt_flags & MNT_NODEV)
2572                         return -EACCES;
2573                 /*FALLTHRU*/
2574         case S_IFIFO:
2575         case S_IFSOCK:
2576                 flag &= ~O_TRUNC;
2577                 break;
2578         }
2579 
2580         error = inode_permission(inode, acc_mode);
2581         if (error)
2582                 return error;
2583 
2584         /*
2585          * An append-only file must be opened in append mode for writing.
2586          */
2587         if (IS_APPEND(inode)) {
2588                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2589                         return -EPERM;
2590                 if (flag & O_TRUNC)
2591                         return -EPERM;
2592         }
2593 
2594         /* O_NOATIME can only be set by the owner or superuser */
2595         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2596                 return -EPERM;
2597 
2598         return 0;
2599 }
2600 
2601 static int handle_truncate(struct file *filp)
2602 {
2603         struct path *path = &filp->f_path;
2604         struct inode *inode = path->dentry->d_inode;
2605         int error = get_write_access(inode);
2606         if (error)
2607                 return error;
2608         /*
2609          * Refuse to truncate files with mandatory locks held on them.
2610          */
2611         error = locks_verify_locked(filp);
2612         if (!error)
2613                 error = security_path_truncate(path);
2614         if (!error) {
2615                 error = do_truncate(path->dentry, 0,
2616                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2617                                     filp);
2618         }
2619         put_write_access(inode);
2620         return error;
2621 }
2622 
2623 static inline int open_to_namei_flags(int flag)
2624 {
2625         if ((flag & O_ACCMODE) == 3)
2626                 flag--;
2627         return flag;
2628 }
2629 
2630 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2631 {
2632         int error = security_path_mknod(dir, dentry, mode, 0);
2633         if (error)
2634                 return error;
2635 
2636         error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2637         if (error)
2638                 return error;
2639 
2640         return security_inode_create(dir->dentry->d_inode, dentry, mode);
2641 }
2642 
2643 /*
2644  * Attempt to atomically look up, create and open a file from a negative
2645  * dentry.
2646  *
2647  * Returns 0 if successful.  The file will have been created and attached to
2648  * @file by the filesystem calling finish_open().
2649  *
2650  * Returns 1 if the file was looked up only or didn't need creating.  The
2651  * caller will need to perform the open themselves.  @path will have been
2652  * updated to point to the new dentry.  This may be negative.
2653  *
2654  * Returns an error code otherwise.
2655  */
2656 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2657                         struct path *path, struct file *file,
2658                         const struct open_flags *op,
2659                         bool got_write, bool need_lookup,
2660                         int *opened)
2661 {
2662         struct inode *dir =  nd->path.dentry->d_inode;
2663         unsigned open_flag = open_to_namei_flags(op->open_flag);
2664         umode_t mode;
2665         int error;
2666         int acc_mode;
2667         int create_error = 0;
2668         struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2669         bool excl;
2670 
2671         BUG_ON(dentry->d_inode);
2672 
2673         /* Don't create child dentry for a dead directory. */
2674         if (unlikely(IS_DEADDIR(dir))) {
2675                 error = -ENOENT;
2676                 goto out;
2677         }
2678 
2679         mode = op->mode;
2680         if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2681                 mode &= ~current_umask();
2682 
2683         excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2684         if (excl)
2685                 open_flag &= ~O_TRUNC;
2686 
2687         /*
2688          * Checking write permission is tricky, bacuse we don't know if we are
2689          * going to actually need it: O_CREAT opens should work as long as the
2690          * file exists.  But checking existence breaks atomicity.  The trick is
2691          * to check access and if not granted clear O_CREAT from the flags.
2692          *
2693          * Another problem is returing the "right" error value (e.g. for an
2694          * O_EXCL open we want to return EEXIST not EROFS).
2695          */
2696         if (((open_flag & (O_CREAT | O_TRUNC)) ||
2697             (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2698                 if (!(open_flag & O_CREAT)) {
2699                         /*
2700                          * No O_CREATE -> atomicity not a requirement -> fall
2701                          * back to lookup + open
2702                          */
2703                         goto no_open;
2704                 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2705                         /* Fall back and fail with the right error */
2706                         create_error = -EROFS;
2707                         goto no_open;
2708                 } else {
2709                         /* No side effects, safe to clear O_CREAT */
2710                         create_error = -EROFS;
2711                         open_flag &= ~O_CREAT;
2712                 }
2713         }
2714 
2715         if (open_flag & O_CREAT) {
2716                 error = may_o_create(&nd->path, dentry, mode);
2717                 if (error) {
2718                         create_error = error;
2719                         if (open_flag & O_EXCL)
2720                                 goto no_open;
2721                         open_flag &= ~O_CREAT;
2722                 }
2723         }
2724 
2725         if (nd->flags & LOOKUP_DIRECTORY)
2726                 open_flag |= O_DIRECTORY;
2727 
2728         file->f_path.dentry = DENTRY_NOT_SET;
2729         file->f_path.mnt = nd->path.mnt;
2730         error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2731                                       opened);
2732         if (error < 0) {
2733                 if (create_error && error == -ENOENT)
2734                         error = create_error;
2735                 goto out;
2736         }
2737 
2738         if (error) {    /* returned 1, that is */
2739                 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2740                         error = -EIO;
2741                         goto out;
2742                 }
2743                 if (file->f_path.dentry) {
2744                         dput(dentry);
2745                         dentry = file->f_path.dentry;
2746                 }
2747                 if (*opened & FILE_CREATED)
2748                         fsnotify_create(dir, dentry);
2749                 if (!dentry->d_inode) {
2750                         WARN_ON(*opened & FILE_CREATED);
2751                         if (create_error) {
2752                                 error = create_error;
2753                                 goto out;
2754                         }
2755                 } else {
2756                         if (excl && !(*opened & FILE_CREATED)) {
2757                                 error = -EEXIST;
2758                                 goto out;
2759                         }
2760                 }
2761                 goto looked_up;
2762         }
2763 
2764         /*
2765          * We didn't have the inode before the open, so check open permission
2766          * here.
2767          */
2768         acc_mode = op->acc_mode;
2769         if (*opened & FILE_CREATED) {
2770                 WARN_ON(!(open_flag & O_CREAT));
2771                 fsnotify_create(dir, dentry);
2772                 acc_mode = MAY_OPEN;
2773         }
2774         error = may_open(&file->f_path, acc_mode, open_flag);
2775         if (error)
2776                 fput(file);
2777 
2778 out:
2779         dput(dentry);
2780         return error;
2781 
2782 no_open:
2783         if (need_lookup) {
2784                 dentry = lookup_real(dir, dentry, nd->flags);
2785                 if (IS_ERR(dentry))
2786                         return PTR_ERR(dentry);
2787 
2788                 if (create_error) {
2789                         int open_flag = op->open_flag;
2790 
2791                         error = create_error;
2792                         if ((open_flag & O_EXCL)) {
2793                                 if (!dentry->d_inode)
2794                                         goto out;
2795                         } else if (!dentry->d_inode) {
2796                                 goto out;
2797                         } else if ((open_flag & O_TRUNC) &&
2798                                    S_ISREG(dentry->d_inode->i_mode)) {
2799                                 goto out;
2800                         }
2801                         /* will fail later, go on to get the right error */
2802                 }
2803         }
2804 looked_up:
2805         path->dentry = dentry;
2806         path->mnt = nd->path.mnt;
2807         return 1;
2808 }
2809 
2810 /*
2811  * Look up and maybe create and open the last component.
2812  *
2813  * Must be called with i_mutex held on parent.
2814  *
2815  * Returns 0 if the file was successfully atomically created (if necessary) and
2816  * opened.  In this case the file will be returned attached to @file.
2817  *
2818  * Returns 1 if the file was not completely opened at this time, though lookups
2819  * and creations will have been performed and the dentry returned in @path will
2820  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2821  * specified then a negative dentry may be returned.
2822  *
2823  * An error code is returned otherwise.
2824  *
2825  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2826  * cleared otherwise prior to returning.
2827  */
2828 static int lookup_open(struct nameidata *nd, struct path *path,
2829                         struct file *file,
2830                         const struct open_flags *op,
2831                         bool got_write, int *opened)
2832 {
2833         struct dentry *dir = nd->path.dentry;
2834         struct inode *dir_inode = dir->d_inode;
2835         struct dentry *dentry;
2836         int error;
2837         bool need_lookup;
2838 
2839         *opened &= ~FILE_CREATED;
2840         dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2841         if (IS_ERR(dentry))
2842                 return PTR_ERR(dentry);
2843 
2844         /* Cached positive dentry: will open in f_op->open */
2845         if (!need_lookup && dentry->d_inode)
2846                 goto out_no_open;
2847 
2848         if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2849                 return atomic_open(nd, dentry, path, file, op, got_write,
2850                                    need_lookup, opened);
2851         }
2852 
2853         if (need_lookup) {
2854                 BUG_ON(dentry->d_inode);
2855 
2856                 dentry = lookup_real(dir_inode, dentry, nd->flags);
2857                 if (IS_ERR(dentry))
2858                         return PTR_ERR(dentry);
2859         }
2860 
2861         /* Negative dentry, just create the file */
2862         if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2863                 umode_t mode = op->mode;
2864                 if (!IS_POSIXACL(dir->d_inode))
2865                         mode &= ~current_umask();
2866                 /*
2867                  * This write is needed to ensure that a
2868                  * rw->ro transition does not occur between
2869                  * the time when the file is created and when
2870                  * a permanent write count is taken through
2871                  * the 'struct file' in finish_open().
2872                  */
2873                 if (!got_write) {
2874                         error = -EROFS;
2875                         goto out_dput;
2876                 }
2877                 *opened |= FILE_CREATED;
2878                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2879                 if (error)
2880                         goto out_dput;
2881                 error = vfs_create(dir->d_inode, dentry, mode,
2882                                    nd->flags & LOOKUP_EXCL);
2883                 if (error)
2884                         goto out_dput;
2885         }
2886 out_no_open:
2887         path->dentry = dentry;
2888         path->mnt = nd->path.mnt;
2889         return 1;
2890 
2891 out_dput:
2892         dput(dentry);
2893         return error;
2894 }
2895 
2896 /*
2897  * Handle the last step of open()
2898  */
2899 static int do_last(struct nameidata *nd, struct path *path,
2900                    struct file *file, const struct open_flags *op,
2901                    int *opened, struct filename *name)
2902 {
2903         struct dentry *dir = nd->path.dentry;
2904         int open_flag = op->open_flag;
2905         bool will_truncate = (open_flag & O_TRUNC) != 0;
2906         bool got_write = false;
2907         int acc_mode = op->acc_mode;
2908         struct inode *inode;
2909         bool symlink_ok = false;
2910         struct path save_parent = { .dentry = NULL, .mnt = NULL };
2911         bool retried = false;
2912         int error;
2913 
2914         nd->flags &= ~LOOKUP_PARENT;
2915         nd->flags |= op->intent;
2916 
2917         if (nd->last_type != LAST_NORM) {
2918                 error = handle_dots(nd, nd->last_type);
2919                 if (error)
2920                         return error;
2921                 goto finish_open;
2922         }
2923 
2924         if (!(open_flag & O_CREAT)) {
2925                 if (nd->last.name[nd->last.len])
2926                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2927                 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2928                         symlink_ok = true;
2929                 /* we _can_ be in RCU mode here */
2930                 error = lookup_fast(nd, path, &inode);
2931                 if (likely(!error))
2932                         goto finish_lookup;
2933 
2934                 if (error < 0)
2935                         goto out;
2936 
2937                 BUG_ON(nd->inode != dir->d_inode);
2938         } else {
2939                 /* create side of things */
2940                 /*
2941                  * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2942                  * has been cleared when we got to the last component we are
2943                  * about to look up
2944                  */
2945                 error = complete_walk(nd);
2946                 if (error)
2947                         return error;
2948 
2949                 audit_inode(name, dir, LOOKUP_PARENT);
2950                 error = -EISDIR;
2951                 /* trailing slashes? */
2952                 if (nd->last.name[nd->last.len])
2953                         goto out;
2954         }
2955 
2956 retry_lookup:
2957         if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2958                 error = mnt_want_write(nd->path.mnt);
2959                 if (!error)
2960                         got_write = true;
2961                 /*
2962                  * do _not_ fail yet - we might not need that or fail with
2963                  * a different error; let lookup_open() decide; we'll be
2964                  * dropping this one anyway.
2965                  */
2966         }
2967         mutex_lock(&dir->d_inode->i_mutex);
2968         error = lookup_open(nd, path, file, op, got_write, opened);
2969         mutex_unlock(&dir->d_inode->i_mutex);
2970 
2971         if (error <= 0) {
2972                 if (error)
2973                         goto out;
2974 
2975                 if ((*opened & FILE_CREATED) ||
2976                     !S_ISREG(file_inode(file)->i_mode))
2977                         will_truncate = false;
2978 
2979                 audit_inode(name, file->f_path.dentry, 0);
2980                 goto opened;
2981         }
2982 
2983         if (*opened & FILE_CREATED) {
2984                 /* Don't check for write permission, don't truncate */
2985                 open_flag &= ~O_TRUNC;
2986                 will_truncate = false;
2987                 acc_mode = MAY_OPEN;
2988                 path_to_nameidata(path, nd);
2989                 goto finish_open_created;
2990         }
2991 
2992         /*
2993          * create/update audit record if it already exists.
2994          */
2995         if (d_is_positive(path->dentry))
2996                 audit_inode(name, path->dentry, 0);
2997 
2998         /*
2999          * If atomic_open() acquired write access it is dropped now due to
3000          * possible mount and symlink following (this might be optimized away if
3001          * necessary...)
3002          */
3003         if (got_write) {
3004                 mnt_drop_write(nd->path.mnt);
3005                 got_write = false;
3006         }
3007 
3008         error = -EEXIST;
3009         if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3010                 goto exit_dput;
3011 
3012         error = follow_managed(path, nd->flags);
3013         if (error < 0)
3014                 goto exit_dput;
3015 
3016         if (error)
3017                 nd->flags |= LOOKUP_JUMPED;
3018 
3019         BUG_ON(nd->flags & LOOKUP_RCU);
3020         inode = path->dentry->d_inode;
3021 finish_lookup:
3022         /* we _can_ be in RCU mode here */
3023         error = -ENOENT;
3024         if (!inode || d_is_negative(path->dentry)) {
3025                 path_to_nameidata(path, nd);
3026                 goto out;
3027         }
3028 
3029         if (should_follow_link(path->dentry, !symlink_ok)) {
3030                 if (nd->flags & LOOKUP_RCU) {
3031                         if (unlikely(unlazy_walk(nd, path->dentry))) {
3032                                 error = -ECHILD;
3033                                 goto out;
3034                         }
3035                 }
3036                 BUG_ON(inode != path->dentry->d_inode);
3037                 return 1;
3038         }
3039 
3040         if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3041                 path_to_nameidata(path, nd);
3042         } else {
3043                 save_parent.dentry = nd->path.dentry;
3044                 save_parent.mnt = mntget(path->mnt);
3045                 nd->path.dentry = path->dentry;
3046 
3047         }
3048         nd->inode = inode;
3049         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3050 finish_open:
3051         error = complete_walk(nd);
3052         if (error) {
3053                 path_put(&save_parent);
3054                 return error;
3055         }
3056         audit_inode(name, nd->path.dentry, 0);
3057         error = -EISDIR;
3058         if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3059                 goto out;
3060         error = -ENOTDIR;
3061         if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3062                 goto out;
3063         if (!S_ISREG(nd->inode->i_mode))
3064                 will_truncate = false;
3065 
3066         if (will_truncate) {
3067                 error = mnt_want_write(nd->path.mnt);
3068                 if (error)
3069                         goto out;
3070                 got_write = true;
3071         }
3072 finish_open_created:
3073         error = may_open(&nd->path, acc_mode, open_flag);
3074         if (error)
3075                 goto out;
3076 
3077         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3078         error = vfs_open(&nd->path, file, current_cred());
3079         if (!error) {
3080                 *opened |= FILE_OPENED;
3081         } else {
3082                 if (error == -EOPENSTALE)
3083                         goto stale_open;
3084                 goto out;
3085         }
3086 opened:
3087         error = open_check_o_direct(file);
3088         if (error)
3089                 goto exit_fput;
3090         error = ima_file_check(file, op->acc_mode, *opened);
3091         if (error)
3092                 goto exit_fput;
3093 
3094         if (will_truncate) {
3095                 error = handle_truncate(file);
3096                 if (error)
3097                         goto exit_fput;
3098         }
3099 out:
3100         if (got_write)
3101                 mnt_drop_write(nd->path.mnt);
3102         path_put(&save_parent);
3103         terminate_walk(nd);
3104         return error;
3105 
3106 exit_dput:
3107         path_put_conditional(path, nd);
3108         goto out;
3109 exit_fput:
3110         fput(file);
3111         goto out;
3112 
3113 stale_open:
3114         /* If no saved parent or already retried then can't retry */
3115         if (!save_parent.dentry || retried)
3116                 goto out;
3117 
3118         BUG_ON(save_parent.dentry != dir);
3119         path_put(&nd->path);
3120         nd->path = save_parent;
3121         nd->inode = dir->d_inode;
3122         save_parent.mnt = NULL;
3123         save_parent.dentry = NULL;
3124         if (got_write) {
3125                 mnt_drop_write(nd->path.mnt);
3126                 got_write = false;
3127         }
3128         retried = true;
3129         goto retry_lookup;
3130 }
3131 
3132 static int do_tmpfile(int dfd, struct filename *pathname,
3133                 struct nameidata *nd, int flags,
3134                 const struct open_flags *op,
3135                 struct file *file, int *opened)
3136 {
3137         static const struct qstr name = QSTR_INIT("/", 1);
3138         struct dentry *dentry, *child;
3139         struct inode *dir;
3140         int error = path_lookupat(dfd, pathname->name,
3141                                   flags | LOOKUP_DIRECTORY, nd);
3142         if (unlikely(error))
3143                 return error;
3144         error = mnt_want_write(nd->path.mnt);
3145         if (unlikely(error))
3146                 goto out;
3147         /* we want directory to be writable */
3148         error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3149         if (error)
3150                 goto out2;
3151         dentry = nd->path.dentry;
3152         dir = dentry->d_inode;
3153         if (!dir->i_op->tmpfile) {
3154                 error = -EOPNOTSUPP;
3155                 goto out2;
3156         }
3157         child = d_alloc(dentry, &name);
3158         if (unlikely(!child)) {
3159                 error = -ENOMEM;
3160                 goto out2;
3161         }
3162         nd->flags &= ~LOOKUP_DIRECTORY;
3163         nd->flags |= op->intent;
3164         dput(nd->path.dentry);
3165         nd->path.dentry = child;
3166         error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3167         if (error)
3168                 goto out2;
3169         audit_inode(pathname, nd->path.dentry, 0);
3170         /* Don't check for other permissions, the inode was just created */
3171         error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3172         if (error)
3173                 goto out2;
3174         file->f_path.mnt = nd->path.mnt;
3175         error = finish_open(file, nd->path.dentry, NULL, opened);
3176         if (error)
3177                 goto out2;
3178         error = open_check_o_direct(file);
3179         if (error) {
3180                 fput(file);
3181         } else if (!(op->open_flag & O_EXCL)) {
3182                 struct inode *inode = file_inode(file);
3183                 spin_lock(&inode->i_lock);
3184                 inode->i_state |= I_LINKABLE;
3185                 spin_unlock(&inode->i_lock);
3186         }
3187 out2:
3188         mnt_drop_write(nd->path.mnt);
3189 out:
3190         path_put(&nd->path);
3191         return error;
3192 }
3193 
3194 static struct file *path_openat(int dfd, struct filename *pathname,
3195                 struct nameidata *nd, const struct open_flags *op, int flags)
3196 {
3197         struct file *file;
3198         struct path path;
3199         int opened = 0;
3200         int error;
3201 
3202         file = get_empty_filp();
3203         if (IS_ERR(file))
3204                 return file;
3205 
3206         file->f_flags = op->open_flag;
3207 
3208         if (unlikely(file->f_flags & __O_TMPFILE)) {
3209                 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3210                 goto out;
3211         }
3212 
3213         error = path_init(dfd, pathname->name, flags, nd);
3214         if (unlikely(error))
3215                 goto out;
3216 
3217         error = do_last(nd, &path, file, op, &opened, pathname);
3218         while (unlikely(error > 0)) { /* trailing symlink */
3219                 struct path link = path;
3220                 void *cookie;
3221                 if (!(nd->flags & LOOKUP_FOLLOW)) {
3222                         path_put_conditional(&path, nd);
3223                         path_put(&nd->path);
3224                         error = -ELOOP;
3225                         break;
3226                 }
3227                 error = may_follow_link(&link, nd);
3228                 if (unlikely(error))
3229                         break;
3230                 nd->flags |= LOOKUP_PARENT;
3231                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3232                 error = follow_link(&link, nd, &cookie);
3233                 if (unlikely(error))
3234                         break;
3235                 error = do_last(nd, &path, file, op, &opened, pathname);
3236                 put_link(nd, &link, cookie);
3237         }
3238 out:
3239         path_cleanup(nd);
3240         if (!(opened & FILE_OPENED)) {
3241                 BUG_ON(!error);
3242                 put_filp(file);
3243         }
3244         if (unlikely(error)) {
3245                 if (error == -EOPENSTALE) {
3246                         if (flags & LOOKUP_RCU)
3247                                 error = -ECHILD;
3248                         else
3249                                 error = -ESTALE;
3250                 }
3251                 file = ERR_PTR(error);
3252         }
3253         return file;
3254 }
3255 
3256 struct file *do_filp_open(int dfd, struct filename *pathname,
3257                 const struct open_flags *op)
3258 {
3259         struct nameidata nd;
3260         int flags = op->lookup_flags;
3261         struct file *filp;
3262 
3263         filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3264         if (unlikely(filp == ERR_PTR(-ECHILD)))
3265                 filp = path_openat(dfd, pathname, &nd, op, flags);
3266         if (unlikely(filp == ERR_PTR(-ESTALE)))
3267                 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3268         return filp;
3269 }
3270 
3271 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3272                 const char *name, const struct open_flags *op)
3273 {
3274         struct nameidata nd;
3275         struct file *file;
3276         struct filename filename = { .name = name };
3277         int flags = op->lookup_flags | LOOKUP_ROOT;
3278 
3279         nd.root.mnt = mnt;
3280         nd.root.dentry = dentry;
3281 
3282         if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3283                 return ERR_PTR(-ELOOP);
3284 
3285         file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3286         if (unlikely(file == ERR_PTR(-ECHILD)))
3287                 file = path_openat(-1, &filename, &nd, op, flags);
3288         if (unlikely(file == ERR_PTR(-ESTALE)))
3289                 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3290         return file;
3291 }
3292 
3293 struct dentry *kern_path_create(int dfd, const char *pathname,
3294                                 struct path *path, unsigned int lookup_flags)
3295 {
3296         struct dentry *dentry = ERR_PTR(-EEXIST);
3297         struct nameidata nd;
3298         int err2;
3299         int error;
3300         bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3301 
3302         /*
3303          * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3304          * other flags passed in are ignored!
3305          */
3306         lookup_flags &= LOOKUP_REVAL;
3307 
3308         error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3309         if (error)
3310                 return ERR_PTR(error);
3311 
3312         /*
3313          * Yucky last component or no last component at all?
3314          * (foo/., foo/.., /////)
3315          */
3316         if (nd.last_type != LAST_NORM)
3317                 goto out;
3318         nd.flags &= ~LOOKUP_PARENT;
3319         nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3320 
3321         /* don't fail immediately if it's r/o, at least try to report other errors */
3322         err2 = mnt_want_write(nd.path.mnt);
3323         /*
3324          * Do the final lookup.
3325          */
3326         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3327         dentry = lookup_hash(&nd);
3328         if (IS_ERR(dentry))
3329                 goto unlock;
3330 
3331         error = -EEXIST;
3332         if (d_is_positive(dentry))
3333                 goto fail;
3334 
3335         /*
3336          * Special case - lookup gave negative, but... we had foo/bar/
3337          * From the vfs_mknod() POV we just have a negative dentry -
3338          * all is fine. Let's be bastards - you had / on the end, you've
3339          * been asking for (non-existent) directory. -ENOENT for you.
3340          */
3341         if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3342                 error = -ENOENT;
3343                 goto fail;
3344         }
3345         if (unlikely(err2)) {
3346                 error = err2;
3347                 goto fail;
3348         }
3349         *path = nd.path;
3350         return dentry;
3351 fail:
3352         dput(dentry);
3353         dentry = ERR_PTR(error);
3354 unlock:
3355         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3356         if (!err2)
3357                 mnt_drop_write(nd.path.mnt);
3358 out:
3359         path_put(&nd.path);
3360         return dentry;
3361 }
3362 EXPORT_SYMBOL(kern_path_create);
3363 
3364 void done_path_create(struct path *path, struct dentry *dentry)
3365 {
3366         dput(dentry);
3367         mutex_unlock(&path->dentry->d_inode->i_mutex);
3368         mnt_drop_write(path->mnt);
3369         path_put(path);
3370 }
3371 EXPORT_SYMBOL(done_path_create);
3372 
3373 struct dentry *user_path_create(int dfd, const char __user *pathname,
3374                                 struct path *path, unsigned int lookup_flags)
3375 {
3376         struct filename *tmp = getname(pathname);
3377         struct dentry *res;
3378         if (IS_ERR(tmp))
3379                 return ERR_CAST(tmp);
3380         res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3381         putname(tmp);
3382         return res;
3383 }
3384 EXPORT_SYMBOL(user_path_create);
3385 
3386 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3387 {
3388         int error = may_create(dir, dentry);
3389 
3390         if (error)
3391                 return error;
3392 
3393         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3394                 return -EPERM;
3395 
3396         if (!dir->i_op->mknod)
3397                 return -EPERM;
3398 
3399         error = devcgroup_inode_mknod(mode, dev);
3400         if (error)
3401                 return error;
3402 
3403         error = security_inode_mknod(dir, dentry, mode, dev);
3404         if (error)
3405                 return error;
3406 
3407         error = dir->i_op->mknod(dir, dentry, mode, dev);
3408         if (!error)
3409                 fsnotify_create(dir, dentry);
3410         return error;
3411 }
3412 EXPORT_SYMBOL(vfs_mknod);
3413 
3414 static int may_mknod(umode_t mode)
3415 {
3416         switch (mode & S_IFMT) {
3417         case S_IFREG:
3418         case S_IFCHR:
3419         case S_IFBLK:
3420         case S_IFIFO:
3421         case S_IFSOCK:
3422         case 0: /* zero mode translates to S_IFREG */
3423                 return 0;
3424         case S_IFDIR:
3425                 return -EPERM;
3426         default:
3427                 return -EINVAL;
3428         }
3429 }
3430 
3431 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3432                 unsigned, dev)
3433 {
3434         struct dentry *dentry;
3435         struct path path;
3436         int error;
3437         unsigned int lookup_flags = 0;
3438 
3439         error = may_mknod(mode);
3440         if (error)
3441                 return error;
3442 retry:
3443         dentry = user_path_create(dfd, filename, &path, lookup_flags);
3444         if (IS_ERR(dentry))
3445                 return PTR_ERR(dentry);
3446 
3447         if (!IS_POSIXACL(path.dentry->d_inode))
3448                 mode &= ~current_umask();
3449         error = security_path_mknod(&path, dentry, mode, dev);
3450         if (error)
3451                 goto out;
3452         switch (mode & S_IFMT) {
3453                 case 0: case S_IFREG:
3454                         error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3455                         break;
3456                 case S_IFCHR: case S_IFBLK:
3457                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3458                                         new_decode_dev(dev));
3459                         break;
3460                 case S_IFIFO: case S_IFSOCK:
3461                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3462                         break;
3463         }
3464 out:
3465         done_path_create(&path, dentry);
3466         if (retry_estale(error, lookup_flags)) {
3467                 lookup_flags |= LOOKUP_REVAL;
3468                 goto retry;
3469         }
3470         return error;
3471 }
3472 
3473 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3474 {
3475         return sys_mknodat(AT_FDCWD, filename, mode, dev);
3476 }
3477 
3478 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3479 {
3480         int error = may_create(dir, dentry);
3481         unsigned max_links = dir->i_sb->s_max_links;
3482 
3483         if (error)
3484                 return error;
3485 
3486         if (!dir->i_op->mkdir)
3487                 return -EPERM;
3488 
3489         mode &= (S_IRWXUGO|S_ISVTX);
3490         error = security_inode_mkdir(dir, dentry, mode);
3491         if (error)
3492                 return error;
3493 
3494         if (max_links && dir->i_nlink >= max_links)
3495                 return -EMLINK;
3496 
3497         error = dir->i_op->mkdir(dir, dentry, mode);
3498         if (!error)
3499                 fsnotify_mkdir(dir, dentry);
3500         return error;
3501 }
3502 EXPORT_SYMBOL(vfs_mkdir);
3503 
3504 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3505 {
3506         struct dentry *dentry;
3507         struct path path;
3508         int error;
3509         unsigned int lookup_flags = LOOKUP_DIRECTORY;
3510 
3511 retry:
3512         dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3513         if (IS_ERR(dentry))
3514                 return PTR_ERR(dentry);
3515 
3516         if (!IS_POSIXACL(path.dentry->d_inode))
3517                 mode &= ~current_umask();
3518         error = security_path_mkdir(&path, dentry, mode);
3519         if (!error)
3520                 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3521         done_path_create(&path, dentry);
3522         if (retry_estale(error, lookup_flags)) {
3523                 lookup_flags |= LOOKUP_REVAL;
3524                 goto retry;
3525         }
3526         return error;
3527 }
3528 
3529 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3530 {
3531         return sys_mkdirat(AT_FDCWD, pathname, mode);
3532 }
3533 
3534 /*
3535  * The dentry_unhash() helper will try to drop the dentry early: we
3536  * should have a usage count of 1 if we're the only user of this
3537  * dentry, and if that is true (possibly after pruning the dcache),
3538  * then we drop the dentry now.
3539  *
3540  * A low-level filesystem can, if it choses, legally
3541  * do a
3542  *
3543  *      if (!d_unhashed(dentry))
3544  *              return -EBUSY;
3545  *
3546  * if it cannot handle the case of removing a directory
3547  * that is still in use by something else..
3548  */
3549 void dentry_unhash(struct dentry *dentry)
3550 {
3551         shrink_dcache_parent(dentry);
3552         spin_lock(&dentry->d_lock);
3553         if (dentry->d_lockref.count == 1)
3554                 __d_drop(dentry);
3555         spin_unlock(&dentry->d_lock);
3556 }
3557 EXPORT_SYMBOL(dentry_unhash);
3558 
3559 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3560 {
3561         int error = may_delete(dir, dentry, 1);
3562 
3563         if (error)
3564                 return error;
3565 
3566         if (!dir->i_op->rmdir)
3567                 return -EPERM;
3568 
3569         dget(dentry);
3570         mutex_lock(&dentry->d_inode->i_mutex);
3571 
3572         error = -EBUSY;
3573         if (is_local_mountpoint(dentry))
3574                 goto out;
3575 
3576         error = security_inode_rmdir(dir, dentry);
3577         if (error)
3578                 goto out;
3579 
3580         shrink_dcache_parent(dentry);
3581         error = dir->i_op->rmdir(dir, dentry);
3582         if (error)
3583                 goto out;
3584 
3585         dentry->d_inode->i_flags |= S_DEAD;
3586         dont_mount(dentry);
3587         detach_mounts(dentry);
3588 
3589 out:
3590         mutex_unlock(&dentry->d_inode->i_mutex);
3591         dput(dentry);
3592         if (!error)
3593                 d_delete(dentry);
3594         return error;
3595 }
3596 EXPORT_SYMBOL(vfs_rmdir);
3597 
3598 static long do_rmdir(int dfd, const char __user *pathname)
3599 {
3600         int error = 0;
3601         struct filename *name;
3602         struct dentry *dentry;
3603         struct nameidata nd;
3604         unsigned int lookup_flags = 0;
3605 retry:
3606         name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3607         if (IS_ERR(name))
3608                 return PTR_ERR(name);
3609 
3610         switch(nd.last_type) {
3611         case LAST_DOTDOT:
3612                 error = -ENOTEMPTY;
3613                 goto exit1;
3614         case LAST_DOT:
3615                 error = -EINVAL;
3616                 goto exit1;
3617         case LAST_ROOT:
3618                 error = -EBUSY;
3619                 goto exit1;
3620         }
3621 
3622         nd.flags &= ~LOOKUP_PARENT;
3623         error = mnt_want_write(nd.path.mnt);
3624         if (error)
3625                 goto exit1;
3626 
3627         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3628         dentry = lookup_hash(&nd);
3629         error = PTR_ERR(dentry);
3630         if (IS_ERR(dentry))
3631                 goto exit2;
3632         if (!dentry->d_inode) {
3633                 error = -ENOENT;
3634                 goto exit3;
3635         }
3636         error = security_path_rmdir(&nd.path, dentry);
3637         if (error)
3638                 goto exit3;
3639         error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3640 exit3:
3641         dput(dentry);
3642 exit2:
3643         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3644         mnt_drop_write(nd.path.mnt);
3645 exit1:
3646         path_put(&nd.path);
3647         putname(name);
3648         if (retry_estale(error, lookup_flags)) {
3649                 lookup_flags |= LOOKUP_REVAL;
3650                 goto retry;
3651         }
3652         return error;
3653 }
3654 
3655 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3656 {
3657         return do_rmdir(AT_FDCWD, pathname);
3658 }
3659 
3660 /**
3661  * vfs_unlink - unlink a filesystem object
3662  * @dir:        parent directory
3663  * @dentry:     victim
3664  * @delegated_inode: returns victim inode, if the inode is delegated.
3665  *
3666  * The caller must hold dir->i_mutex.
3667  *
3668  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3669  * return a reference to the inode in delegated_inode.  The caller
3670  * should then break the delegation on that inode and retry.  Because
3671  * breaking a delegation may take a long time, the caller should drop
3672  * dir->i_mutex before doing so.
3673  *
3674  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3675  * be appropriate for callers that expect the underlying filesystem not
3676  * to be NFS exported.
3677  */
3678 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3679 {
3680         struct inode *target = dentry->d_inode;
3681         int error = may_delete(dir, dentry, 0);
3682 
3683         if (error)
3684                 return error;
3685 
3686         if (!dir->i_op->unlink)
3687                 return -EPERM;
3688 
3689         mutex_lock(&target->i_mutex);
3690         if (is_local_mountpoint(dentry))
3691                 error = -EBUSY;
3692         else {
3693                 error = security_inode_unlink(dir, dentry);
3694                 if (!error) {
3695                         error = try_break_deleg(target, delegated_inode);
3696                         if (error)
3697                                 goto out;
3698                         error = dir->i_op->unlink(dir, dentry);
3699                         if (!error) {
3700                                 dont_mount(dentry);
3701                                 detach_mounts(dentry);
3702                         }
3703                 }
3704         }
3705 out:
3706         mutex_unlock(&target->i_mutex);
3707 
3708         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3709         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3710                 fsnotify_link_count(target);
3711                 d_delete(dentry);
3712         }
3713 
3714         return error;
3715 }
3716 EXPORT_SYMBOL(vfs_unlink);
3717 
3718 /*
3719  * Make sure that the actual truncation of the file will occur outside its
3720  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3721  * writeout happening, and we don't want to prevent access to the directory
3722  * while waiting on the I/O.
3723  */
3724 static long do_unlinkat(int dfd, const char __user *pathname)
3725 {
3726         int error;
3727         struct filename *name;
3728         struct dentry *dentry;
3729         struct nameidata nd;
3730         struct inode *inode = NULL;
3731         struct inode *delegated_inode = NULL;
3732         unsigned int lookup_flags = 0;
3733 retry:
3734         name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3735         if (IS_ERR(name))
3736                 return PTR_ERR(name);
3737 
3738         error = -EISDIR;
3739         if (nd.last_type != LAST_NORM)
3740                 goto exit1;
3741 
3742         nd.flags &= ~LOOKUP_PARENT;
3743         error = mnt_want_write(nd.path.mnt);
3744         if (error)
3745                 goto exit1;
3746 retry_deleg:
3747         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3748         dentry = lookup_hash(&nd);
3749         error = PTR_ERR(dentry);
3750         if (!IS_ERR(dentry)) {
3751                 /* Why not before? Because we want correct error value */
3752                 if (nd.last.name[nd.last.len])
3753                         goto slashes;
3754                 inode = dentry->d_inode;
3755                 if (d_is_negative(dentry))
3756                         goto slashes;
3757                 ihold(inode);
3758                 error = security_path_unlink(&nd.path, dentry);
3759                 if (error)
3760                         goto exit2;
3761                 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3762 exit2:
3763                 dput(dentry);
3764         }
3765         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3766         if (inode)
3767                 iput(inode);    /* truncate the inode here */
3768         inode = NULL;
3769         if (delegated_inode) {
3770                 error = break_deleg_wait(&delegated_inode);
3771                 if (!error)
3772                         goto retry_deleg;
3773         }
3774         mnt_drop_write(nd.path.mnt);
3775 exit1:
3776         path_put(&nd.path);
3777         putname(name);
3778         if (retry_estale(error, lookup_flags)) {
3779                 lookup_flags |= LOOKUP_REVAL;
3780                 inode = NULL;
3781                 goto retry;
3782         }
3783         return error;
3784 
3785 slashes:
3786         if (d_is_negative(dentry))
3787                 error = -ENOENT;
3788         else if (d_is_dir(dentry))
3789                 error = -EISDIR;
3790         else
3791                 error = -ENOTDIR;
3792         goto exit2;
3793 }
3794 
3795 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3796 {
3797         if ((flag & ~AT_REMOVEDIR) != 0)
3798                 return -EINVAL;
3799 
3800         if (flag & AT_REMOVEDIR)
3801                 return do_rmdir(dfd, pathname);
3802 
3803         return do_unlinkat(dfd, pathname);
3804 }
3805 
3806 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3807 {
3808         return do_unlinkat(AT_FDCWD, pathname);
3809 }
3810 
3811 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3812 {
3813         int error = may_create(dir, dentry);
3814 
3815         if (error)
3816                 return error;
3817 
3818         if (!dir->i_op->symlink)
3819                 return -EPERM;
3820 
3821         error = security_inode_symlink(dir, dentry, oldname);
3822         if (error)
3823                 return error;
3824 
3825         error = dir->i_op->symlink(dir, dentry, oldname);
3826         if (!error)
3827                 fsnotify_create(dir, dentry);
3828         return error;
3829 }
3830 EXPORT_SYMBOL(vfs_symlink);
3831 
3832 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3833                 int, newdfd, const char __user *, newname)
3834 {
3835         int error;
3836         struct filename *from;
3837         struct dentry *dentry;
3838         struct path path;
3839         unsigned int lookup_flags = 0;
3840 
3841         from = getname(oldname);
3842         if (IS_ERR(from))
3843                 return PTR_ERR(from);
3844 retry:
3845         dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3846         error = PTR_ERR(dentry);
3847         if (IS_ERR(dentry))
3848                 goto out_putname;
3849 
3850         error = security_path_symlink(&path, dentry, from->name);
3851         if (!error)
3852                 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3853         done_path_create(&path, dentry);
3854         if (retry_estale(error, lookup_flags)) {
3855                 lookup_flags |= LOOKUP_REVAL;
3856                 goto retry;
3857         }
3858 out_putname:
3859         putname(from);
3860         return error;
3861 }
3862 
3863 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3864 {
3865         return sys_symlinkat(oldname, AT_FDCWD, newname);
3866 }
3867 
3868 /**
3869  * vfs_link - create a new link
3870  * @old_dentry: object to be linked
3871  * @dir:        new parent
3872  * @new_dentry: where to create the new link
3873  * @delegated_inode: returns inode needing a delegation break
3874  *
3875  * The caller must hold dir->i_mutex
3876  *
3877  * If vfs_link discovers a delegation on the to-be-linked file in need
3878  * of breaking, it will return -EWOULDBLOCK and return a reference to the
3879  * inode in delegated_inode.  The caller should then break the delegation
3880  * and retry.  Because breaking a delegation may take a long time, the
3881  * caller should drop the i_mutex before doing so.
3882  *
3883  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3884  * be appropriate for callers that expect the underlying filesystem not
3885  * to be NFS exported.
3886  */
3887 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3888 {
3889         struct inode *inode = old_dentry->d_inode;
3890         unsigned max_links = dir->i_sb->s_max_links;
3891         int error;
3892 
3893         if (!inode)
3894                 return -ENOENT;
3895 
3896         error = may_create(dir, new_dentry);
3897         if (error)
3898                 return error;
3899 
3900         if (dir->i_sb != inode->i_sb)
3901                 return -EXDEV;
3902 
3903         /*
3904          * A link to an append-only or immutable file cannot be created.
3905          */
3906         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3907                 return -EPERM;
3908         if (!dir->i_op->link)
3909                 return -EPERM;
3910         if (S_ISDIR(inode->i_mode))
3911                 return -EPERM;
3912 
3913         error = security_inode_link(old_dentry, dir, new_dentry);
3914         if (error)
3915                 return error;
3916 
3917         mutex_lock(&inode->i_mutex);
3918         /* Make sure we don't allow creating hardlink to an unlinked file */
3919         if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3920                 error =  -ENOENT;
3921         else if (max_links && inode->i_nlink >= max_links)
3922                 error = -EMLINK;
3923         else {
3924                 error = try_break_deleg(inode, delegated_inode);
3925                 if (!error)
3926                         error = dir->i_op->link(old_dentry, dir, new_dentry);
3927         }
3928 
3929         if (!error && (inode->i_state & I_LINKABLE)) {
3930                 spin_lock(&inode->i_lock);
3931                 inode->i_state &= ~I_LINKABLE;
3932                 spin_unlock(&inode->i_lock);
3933         }
3934         mutex_unlock(&inode->i_mutex);
3935         if (!error)
3936                 fsnotify_link(dir, inode, new_dentry);
3937         return error;
3938 }
3939 EXPORT_SYMBOL(vfs_link);
3940 
3941 /*
3942  * Hardlinks are often used in delicate situations.  We avoid
3943  * security-related surprises by not following symlinks on the
3944  * newname.  --KAB
3945  *
3946  * We don't follow them on the oldname either to be compatible
3947  * with linux 2.0, and to avoid hard-linking to directories
3948  * and other special files.  --ADM
3949  */
3950 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3951                 int, newdfd, const char __user *, newname, int, flags)
3952 {
3953         struct dentry *new_dentry;
3954         struct path old_path, new_path;
3955         struct inode *delegated_inode = NULL;
3956         int how = 0;
3957         int error;
3958 
3959         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3960                 return -EINVAL;
3961         /*
3962          * To use null names we require CAP_DAC_READ_SEARCH
3963          * This ensures that not everyone will be able to create
3964          * handlink using the passed filedescriptor.
3965          */
3966         if (flags & AT_EMPTY_PATH) {
3967                 if (!capable(CAP_DAC_READ_SEARCH))
3968                         return -ENOENT;
3969                 how = LOOKUP_EMPTY;
3970         }
3971 
3972         if (flags & AT_SYMLINK_FOLLOW)
3973                 how |= LOOKUP_FOLLOW;
3974 retry:
3975         error = user_path_at(olddfd, oldname, how, &old_path);
3976         if (error)
3977                 return error;
3978 
3979         new_dentry = user_path_create(newdfd, newname, &new_path,
3980                                         (how & LOOKUP_REVAL));
3981         error = PTR_ERR(new_dentry);
3982         if (IS_ERR(new_dentry))
3983                 goto out;
3984 
3985         error = -EXDEV;
3986         if (old_path.mnt != new_path.mnt)
3987                 goto out_dput;
3988         error = may_linkat(&old_path);
3989         if (unlikely(error))
3990                 goto out_dput;
3991         error = security_path_link(old_path.dentry, &new_path, new_dentry);
3992         if (error)
3993                 goto out_dput;
3994         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3995 out_dput:
3996         done_path_create(&new_path, new_dentry);
3997         if (delegated_inode) {
3998                 error = break_deleg_wait(&delegated_inode);
3999                 if (!error) {
4000                         path_put(&old_path);
4001                         goto retry;
4002                 }
4003         }
4004         if (retry_estale(error, how)) {
4005                 path_put(&old_path);
4006                 how |= LOOKUP_REVAL;
4007                 goto retry;
4008         }
4009 out:
4010         path_put(&old_path);
4011 
4012         return error;
4013 }
4014 
4015 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4016 {
4017         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4018 }
4019 
4020 /**
4021  * vfs_rename - rename a filesystem object
4022  * @old_dir:    parent of source
4023  * @old_dentry: source
4024  * @new_dir:    parent of destination
4025  * @new_dentry: destination
4026  * @delegated_inode: returns an inode needing a delegation break
4027  * @flags:      rename flags
4028  *
4029  * The caller must hold multiple mutexes--see lock_rename()).
4030  *
4031  * If vfs_rename discovers a delegation in need of breaking at either
4032  * the source or destination, it will return -EWOULDBLOCK and return a
4033  * reference to the inode in delegated_inode.  The caller should then
4034  * break the delegation and retry.  Because breaking a delegation may
4035  * take a long time, the caller should drop all locks before doing
4036  * so.
4037  *
4038  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4039  * be appropriate for callers that expect the underlying filesystem not
4040  * to be NFS exported.
4041  *
4042  * The worst of all namespace operations - renaming directory. "Perverted"
4043  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4044  * Problems:
4045  *      a) we can get into loop creation.
4046  *      b) race potential - two innocent renames can create a loop together.
4047  *         That's where 4.4 screws up. Current fix: serialization on
4048  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4049  *         story.
4050  *      c) we have to lock _four_ objects - parents and victim (if it exists),
4051  *         and source (if it is not a directory).
4052  *         And that - after we got ->i_mutex on parents (until then we don't know
4053  *         whether the target exists).  Solution: try to be smart with locking
4054  *         order for inodes.  We rely on the fact that tree topology may change
4055  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4056  *         move will be locked.  Thus we can rank directories by the tree
4057  *         (ancestors first) and rank all non-directories after them.
4058  *         That works since everybody except rename does "lock parent, lookup,
4059  *         lock child" and rename is under ->s_vfs_rename_mutex.
4060  *         HOWEVER, it relies on the assumption that any object with ->lookup()
4061  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4062  *         we'd better make sure that there's no link(2) for them.
4063  *      d) conversion from fhandle to dentry may come in the wrong moment - when
4064  *         we are removing the target. Solution: we will have to grab ->i_mutex
4065  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4066  *         ->i_mutex on parents, which works but leads to some truly excessive
4067  *         locking].
4068  */
4069 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4070                struct inode *new_dir, struct dentry *new_dentry,
4071                struct inode **delegated_inode, unsigned int flags)
4072 {
4073         int error;
4074         bool is_dir = d_is_dir(old_dentry);
4075         const unsigned char *old_name;
4076         struct inode *source = old_dentry->d_inode;
4077         struct inode *target = new_dentry->d_inode;
4078         bool new_is_dir = false;
4079         unsigned max_links = new_dir->i_sb->s_max_links;
4080 
4081         if (source == target)
4082                 return 0;
4083 
4084         error = may_delete(old_dir, old_dentry, is_dir);
4085         if (error)
4086                 return error;
4087 
4088         if (!target) {
4089                 error = may_create(new_dir, new_dentry);
4090         } else {
4091                 new_is_dir = d_is_dir(new_dentry);
4092 
4093                 if (!(flags & RENAME_EXCHANGE))
4094                         error = may_delete(new_dir, new_dentry, is_dir);
4095                 else
4096                         error = may_delete(new_dir, new_dentry, new_is_dir);
4097         }
4098         if (error)
4099                 return error;
4100 
4101         if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4102                 return -EPERM;
4103 
4104         if (flags && !old_dir->i_op->rename2)
4105                 return -EINVAL;
4106 
4107         /*
4108          * If we are going to change the parent - check write permissions,
4109          * we'll need to flip '..'.
4110          */
4111         if (new_dir != old_dir) {
4112                 if (is_dir) {
4113                         error = inode_permission(source, MAY_WRITE);
4114                         if (error)
4115                                 return error;
4116                 }
4117                 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4118                         error = inode_permission(target, MAY_WRITE);
4119                         if (error)
4120                                 return error;
4121                 }
4122         }
4123 
4124         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4125                                       flags);
4126         if (error)
4127                 return error;
4128 
4129         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4130         dget(new_dentry);
4131         if (!is_dir || (flags & RENAME_EXCHANGE))
4132                 lock_two_nondirectories(source, target);
4133         else if (target)
4134                 mutex_lock(&target->i_mutex);
4135 
4136         error = -EBUSY;
4137         if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4138                 goto out;
4139 
4140         if (max_links && new_dir != old_dir) {
4141                 error = -EMLINK;
4142                 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4143                         goto out;
4144                 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4145                     old_dir->i_nlink >= max_links)
4146                         goto out;
4147         }
4148         if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4149                 shrink_dcache_parent(new_dentry);
4150         if (!is_dir) {
4151                 error = try_break_deleg(source, delegated_inode);
4152                 if (error)
4153                         goto out;
4154         }
4155         if (target && !new_is_dir) {
4156                 error = try_break_deleg(target, delegated_inode);
4157                 if (error)
4158                         goto out;
4159         }
4160         if (!old_dir->i_op->rename2) {
4161                 error = old_dir->i_op->rename(old_dir, old_dentry,
4162                                               new_dir, new_dentry);
4163         } else {
4164                 WARN_ON(old_dir->i_op->rename != NULL);
4165                 error = old_dir->i_op->rename2(old_dir, old_dentry,
4166                                                new_dir, new_dentry, flags);
4167         }
4168         if (error)
4169                 goto out;
4170 
4171         if (!(flags & RENAME_EXCHANGE) && target) {
4172                 if (is_dir)
4173                         target->i_flags |= S_DEAD;
4174                 dont_mount(new_dentry);
4175                 detach_mounts(new_dentry);
4176         }
4177         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4178                 if (!(flags & RENAME_EXCHANGE))
4179                         d_move(old_dentry, new_dentry);
4180                 else
4181                         d_exchange(old_dentry, new_dentry);
4182         }
4183 out:
4184         if (!is_dir || (flags & RENAME_EXCHANGE))
4185                 unlock_two_nondirectories(source, target);
4186         else if (target)
4187                 mutex_unlock(&target->i_mutex);
4188         dput(new_dentry);
4189         if (!error) {
4190                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4191                               !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4192                 if (flags & RENAME_EXCHANGE) {
4193                         fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4194                                       new_is_dir, NULL, new_dentry);
4195                 }
4196         }
4197         fsnotify_oldname_free(old_name);
4198 
4199         return error;
4200 }
4201 EXPORT_SYMBOL(vfs_rename);
4202 
4203 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4204                 int, newdfd, const char __user *, newname, unsigned int, flags)
4205 {
4206         struct dentry *old_dir, *new_dir;
4207         struct dentry *old_dentry, *new_dentry;
4208         struct dentry *trap;
4209         struct nameidata oldnd, newnd;
4210         struct inode *delegated_inode = NULL;
4211         struct filename *from;
4212         struct filename *to;
4213         unsigned int lookup_flags = 0;
4214         bool should_retry = false;
4215         int error;
4216 
4217         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4218                 return -EINVAL;
4219 
4220         if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4221             (flags & RENAME_EXCHANGE))
4222                 return -EINVAL;
4223 
4224         if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4225                 return -EPERM;
4226 
4227 retry:
4228         from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4229         if (IS_ERR(from)) {
4230                 error = PTR_ERR(from);
4231                 goto exit;
4232         }
4233 
4234         to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4235         if (IS_ERR(to)) {
4236                 error = PTR_ERR(to);
4237                 goto exit1;
4238         }
4239 
4240         error = -EXDEV;
4241         if (oldnd.path.mnt != newnd.path.mnt)
4242                 goto exit2;
4243 
4244         old_dir = oldnd.path.dentry;
4245         error = -EBUSY;
4246         if (oldnd.last_type != LAST_NORM)
4247                 goto exit2;
4248 
4249         new_dir = newnd.path.dentry;
4250         if (flags & RENAME_NOREPLACE)
4251                 error = -EEXIST;
4252         if (newnd.last_type != LAST_NORM)
4253                 goto exit2;
4254 
4255         error = mnt_want_write(oldnd.path.mnt);
4256         if (error)
4257                 goto exit2;
4258 
4259         oldnd.flags &= ~LOOKUP_PARENT;
4260         newnd.flags &= ~LOOKUP_PARENT;
4261         if (!(flags & RENAME_EXCHANGE))
4262                 newnd.flags |= LOOKUP_RENAME_TARGET;
4263 
4264 retry_deleg:
4265         trap = lock_rename(new_dir, old_dir);
4266 
4267         old_dentry = lookup_hash(&oldnd);
4268         error = PTR_ERR(old_dentry);
4269         if (IS_ERR(old_dentry))
4270                 goto exit3;
4271         /* source must exist */
4272         error = -ENOENT;
4273         if (d_is_negative(old_dentry))
4274                 goto exit4;
4275         new_dentry = lookup_hash(&newnd);
4276         error = PTR_ERR(new_dentry);
4277         if (IS_ERR(new_dentry))
4278                 goto exit4;
4279         error = -EEXIST;
4280         if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4281                 goto exit5;
4282         if (flags & RENAME_EXCHANGE) {
4283                 error = -ENOENT;
4284                 if (d_is_negative(new_dentry))
4285                         goto exit5;
4286 
4287                 if (!d_is_dir(new_dentry)) {
4288                         error = -ENOTDIR;
4289                         if (newnd.last.name[newnd.last.len])
4290                                 goto exit5;
4291                 }
4292         }
4293         /* unless the source is a directory trailing slashes give -ENOTDIR */
4294         if (!d_is_dir(old_dentry)) {
4295                 error = -ENOTDIR;
4296                 if (oldnd.last.name[oldnd.last.len])
4297                         goto exit5;
4298                 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4299                         goto exit5;
4300         }
4301         /* source should not be ancestor of target */
4302         error = -EINVAL;
4303         if (old_dentry == trap)
4304                 goto exit5;
4305         /* target should not be an ancestor of source */
4306         if (!(flags & RENAME_EXCHANGE))
4307                 error = -ENOTEMPTY;
4308         if (new_dentry == trap)
4309                 goto exit5;
4310 
4311         error = security_path_rename(&oldnd.path, old_dentry,
4312                                      &newnd.path, new_dentry, flags);
4313         if (error)
4314                 goto exit5;
4315         error = vfs_rename(old_dir->d_inode, old_dentry,
4316                            new_dir->d_inode, new_dentry,
4317                            &delegated_inode, flags);
4318 exit5:
4319         dput(new_dentry);
4320 exit4:
4321         dput(old_dentry);
4322 exit3:
4323         unlock_rename(new_dir, old_dir);
4324         if (delegated_inode) {
4325                 error = break_deleg_wait(&delegated_inode);
4326                 if (!error)
4327                         goto retry_deleg;
4328         }
4329         mnt_drop_write(oldnd.path.mnt);
4330 exit2:
4331         if (retry_estale(error, lookup_flags))
4332                 should_retry = true;
4333         path_put(&newnd.path);
4334         putname(to);
4335 exit1:
4336         path_put(&oldnd.path);
4337         putname(from);
4338         if (should_retry) {
4339                 should_retry = false;
4340                 lookup_flags |= LOOKUP_REVAL;
4341                 goto retry;
4342         }
4343 exit:
4344         return error;
4345 }
4346 
4347 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4348                 int, newdfd, const char __user *, newname)
4349 {
4350         return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4351 }
4352 
4353 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4354 {
4355         return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4356 }
4357 
4358 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4359 {
4360         int error = may_create(dir, dentry);
4361         if (error)
4362                 return error;
4363 
4364         if (!dir->i_op->mknod)
4365                 return -EPERM;
4366 
4367         return dir->i_op->mknod(dir, dentry,
4368                                 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4369 }
4370 EXPORT_SYMBOL(vfs_whiteout);
4371 
4372 int readlink_copy(char __user *buffer, int buflen, const char *link)
4373 {
4374         int len = PTR_ERR(link);
4375         if (IS_ERR(link))
4376                 goto out;
4377 
4378         len = strlen(link);
4379         if (len > (unsigned) buflen)
4380                 len = buflen;
4381         if (copy_to_user(buffer, link, len))
4382                 len = -EFAULT;
4383 out:
4384         return len;
4385 }
4386 EXPORT_SYMBOL(readlink_copy);
4387 
4388 /*
4389  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4390  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4391  * using) it for any given inode is up to filesystem.
4392  */
4393 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4394 {
4395         struct nameidata nd;
4396         void *cookie;
4397         int res;
4398 
4399         nd.depth = 0;
4400         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4401         if (IS_ERR(cookie))
4402                 return PTR_ERR(cookie);
4403 
4404         res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4405         if (dentry->d_inode->i_op->put_link)
4406                 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4407         return res;
4408 }
4409 EXPORT_SYMBOL(generic_readlink);
4410 
4411 /* get the link contents into pagecache */
4412 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4413 {
4414         char *kaddr;
4415         struct page *page;
4416         struct address_space *mapping = dentry->d_inode->i_mapping;
4417         page = read_mapping_page(mapping, 0, NULL);
4418         if (IS_ERR(page))
4419                 return (char*)page;
4420         *ppage = page;
4421         kaddr = kmap(page);
4422         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4423         return kaddr;
4424 }
4425 
4426 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4427 {
4428         struct page *page = NULL;
4429         int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4430         if (page) {
4431                 kunmap(page);
4432                 page_cache_release(page);
4433         }
4434         return res;
4435 }
4436 EXPORT_SYMBOL(page_readlink);
4437 
4438 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4439 {
4440         struct page *page = NULL;
4441         nd_set_link(nd, page_getlink(dentry, &page));
4442         return page;
4443 }
4444 EXPORT_SYMBOL(page_follow_link_light);
4445 
4446 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4447 {
4448         struct page *page = cookie;
4449 
4450         if (page) {
4451                 kunmap(page);
4452                 page_cache_release(page);
4453         }
4454 }
4455 EXPORT_SYMBOL(page_put_link);
4456 
4457 /*
4458  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4459  */
4460 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4461 {
4462         struct address_space *mapping = inode->i_mapping;
4463         struct page *page;
4464         void *fsdata;
4465         int err;
4466         char *kaddr;
4467         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4468         if (nofs)
4469                 flags |= AOP_FLAG_NOFS;
4470 
4471 retry:
4472         err = pagecache_write_begin(NULL, mapping, 0, len-1,
4473                                 flags, &page, &fsdata);
4474         if (err)
4475                 goto fail;
4476 
4477         kaddr = kmap_atomic(page);
4478         memcpy(kaddr, symname, len-1);
4479         kunmap_atomic(kaddr);
4480 
4481         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4482                                                         page, fsdata);
4483         if (err < 0)
4484                 goto fail;
4485         if (err < len-1)
4486                 goto retry;
4487 
4488         mark_inode_dirty(inode);
4489         return 0;
4490 fail:
4491         return err;
4492 }
4493 EXPORT_SYMBOL(__page_symlink);
4494 
4495 int page_symlink(struct inode *inode, const char *symname, int len)
4496 {
4497         return __page_symlink(inode, symname, len,
4498                         !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4499 }
4500 EXPORT_SYMBOL(page_symlink);
4501 
4502 const struct inode_operations page_symlink_inode_operations = {
4503         .readlink       = generic_readlink,
4504         .follow_link    = page_follow_link_light,
4505         .put_link       = page_put_link,
4506 };
4507 EXPORT_SYMBOL(page_symlink_inode_operations);
4508 

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