Version:  2.0.40 2.2.26 2.4.37 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7

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 <linux/bitops.h>
 39 #include <asm/uaccess.h>
 40 
 41 #include "internal.h"
 42 #include "mount.h"
 43 
 44 /* [Feb-1997 T. Schoebel-Theuer]
 45  * Fundamental changes in the pathname lookup mechanisms (namei)
 46  * were necessary because of omirr.  The reason is that omirr needs
 47  * to know the _real_ pathname, not the user-supplied one, in case
 48  * of symlinks (and also when transname replacements occur).
 49  *
 50  * The new code replaces the old recursive symlink resolution with
 51  * an iterative one (in case of non-nested symlink chains).  It does
 52  * this with calls to <fs>_follow_link().
 53  * As a side effect, dir_namei(), _namei() and follow_link() are now 
 54  * replaced with a single function lookup_dentry() that can handle all 
 55  * the special cases of the former code.
 56  *
 57  * With the new dcache, the pathname is stored at each inode, at least as
 58  * long as the refcount of the inode is positive.  As a side effect, the
 59  * size of the dcache depends on the inode cache and thus is dynamic.
 60  *
 61  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
 62  * resolution to correspond with current state of the code.
 63  *
 64  * Note that the symlink resolution is not *completely* iterative.
 65  * There is still a significant amount of tail- and mid- recursion in
 66  * the algorithm.  Also, note that <fs>_readlink() is not used in
 67  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
 68  * may return different results than <fs>_follow_link().  Many virtual
 69  * filesystems (including /proc) exhibit this behavior.
 70  */
 71 
 72 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
 73  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
 74  * and the name already exists in form of a symlink, try to create the new
 75  * name indicated by the symlink. The old code always complained that the
 76  * name already exists, due to not following the symlink even if its target
 77  * is nonexistent.  The new semantics affects also mknod() and link() when
 78  * the name is a symlink pointing to a non-existent name.
 79  *
 80  * I don't know which semantics is the right one, since I have no access
 81  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
 82  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
 83  * "old" one. Personally, I think the new semantics is much more logical.
 84  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
 85  * file does succeed in both HP-UX and SunOs, but not in Solaris
 86  * and in the old Linux semantics.
 87  */
 88 
 89 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
 90  * semantics.  See the comments in "open_namei" and "do_link" below.
 91  *
 92  * [10-Sep-98 Alan Modra] Another symlink change.
 93  */
 94 
 95 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
 96  *      inside the path - always follow.
 97  *      in the last component in creation/removal/renaming - never follow.
 98  *      if LOOKUP_FOLLOW passed - follow.
 99  *      if the pathname has trailing slashes - follow.
100  *      otherwise - don't follow.
101  * (applied in that order).
102  *
103  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
104  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
105  * During the 2.4 we need to fix the userland stuff depending on it -
106  * hopefully we will be able to get rid of that wart in 2.5. So far only
107  * XEmacs seems to be relying on it...
108  */
109 /*
110  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
111  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
112  * any extra contention...
113  */
114 
115 /* In order to reduce some races, while at the same time doing additional
116  * checking and hopefully speeding things up, we copy filenames to the
117  * kernel data space before using them..
118  *
119  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
120  * PATH_MAX includes the nul terminator --RR.
121  */
122 
123 #define EMBEDDED_NAME_MAX       (PATH_MAX - offsetof(struct filename, iname))
124 
125 struct filename *
126 getname_flags(const char __user *filename, int flags, int *empty)
127 {
128         struct filename *result;
129         char *kname;
130         int len;
131 
132         result = audit_reusename(filename);
133         if (result)
134                 return result;
135 
136         result = __getname();
137         if (unlikely(!result))
138                 return ERR_PTR(-ENOMEM);
139 
140         /*
141          * First, try to embed the struct filename inside the names_cache
142          * allocation
143          */
144         kname = (char *)result->iname;
145         result->name = kname;
146 
147         len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
148         if (unlikely(len < 0)) {
149                 __putname(result);
150                 return ERR_PTR(len);
151         }
152 
153         /*
154          * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
155          * separate struct filename so we can dedicate the entire
156          * names_cache allocation for the pathname, and re-do the copy from
157          * userland.
158          */
159         if (unlikely(len == EMBEDDED_NAME_MAX)) {
160                 const size_t size = offsetof(struct filename, iname[1]);
161                 kname = (char *)result;
162 
163                 /*
164                  * size is chosen that way we to guarantee that
165                  * result->iname[0] is within the same object and that
166                  * kname can't be equal to result->iname, no matter what.
167                  */
168                 result = kzalloc(size, GFP_KERNEL);
169                 if (unlikely(!result)) {
170                         __putname(kname);
171                         return ERR_PTR(-ENOMEM);
172                 }
173                 result->name = kname;
174                 len = strncpy_from_user(kname, filename, PATH_MAX);
175                 if (unlikely(len < 0)) {
176                         __putname(kname);
177                         kfree(result);
178                         return ERR_PTR(len);
179                 }
180                 if (unlikely(len == PATH_MAX)) {
181                         __putname(kname);
182                         kfree(result);
183                         return ERR_PTR(-ENAMETOOLONG);
184                 }
185         }
186 
187         result->refcnt = 1;
188         /* The empty path is special. */
189         if (unlikely(!len)) {
190                 if (empty)
191                         *empty = 1;
192                 if (!(flags & LOOKUP_EMPTY)) {
193                         putname(result);
194                         return ERR_PTR(-ENOENT);
195                 }
196         }
197 
198         result->uptr = filename;
199         result->aname = NULL;
200         audit_getname(result);
201         return result;
202 }
203 
204 struct filename *
205 getname(const char __user * filename)
206 {
207         return getname_flags(filename, 0, NULL);
208 }
209 
210 struct filename *
211 getname_kernel(const char * filename)
212 {
213         struct filename *result;
214         int len = strlen(filename) + 1;
215 
216         result = __getname();
217         if (unlikely(!result))
218                 return ERR_PTR(-ENOMEM);
219 
220         if (len <= EMBEDDED_NAME_MAX) {
221                 result->name = (char *)result->iname;
222         } else if (len <= PATH_MAX) {
223                 struct filename *tmp;
224 
225                 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
226                 if (unlikely(!tmp)) {
227                         __putname(result);
228                         return ERR_PTR(-ENOMEM);
229                 }
230                 tmp->name = (char *)result;
231                 result = tmp;
232         } else {
233                 __putname(result);
234                 return ERR_PTR(-ENAMETOOLONG);
235         }
236         memcpy((char *)result->name, filename, len);
237         result->uptr = NULL;
238         result->aname = NULL;
239         result->refcnt = 1;
240         audit_getname(result);
241 
242         return result;
243 }
244 
245 void putname(struct filename *name)
246 {
247         BUG_ON(name->refcnt <= 0);
248 
249         if (--name->refcnt > 0)
250                 return;
251 
252         if (name->name != name->iname) {
253                 __putname(name->name);
254                 kfree(name);
255         } else
256                 __putname(name);
257 }
258 
259 static int check_acl(struct inode *inode, int mask)
260 {
261 #ifdef CONFIG_FS_POSIX_ACL
262         struct posix_acl *acl;
263 
264         if (mask & MAY_NOT_BLOCK) {
265                 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
266                 if (!acl)
267                         return -EAGAIN;
268                 /* no ->get_acl() calls in RCU mode... */
269                 if (is_uncached_acl(acl))
270                         return -ECHILD;
271                 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
272         }
273 
274         acl = get_acl(inode, ACL_TYPE_ACCESS);
275         if (IS_ERR(acl))
276                 return PTR_ERR(acl);
277         if (acl) {
278                 int error = posix_acl_permission(inode, acl, mask);
279                 posix_acl_release(acl);
280                 return error;
281         }
282 #endif
283 
284         return -EAGAIN;
285 }
286 
287 /*
288  * This does the basic permission checking
289  */
290 static int acl_permission_check(struct inode *inode, int mask)
291 {
292         unsigned int mode = inode->i_mode;
293 
294         if (likely(uid_eq(current_fsuid(), inode->i_uid)))
295                 mode >>= 6;
296         else {
297                 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
298                         int error = check_acl(inode, mask);
299                         if (error != -EAGAIN)
300                                 return error;
301                 }
302 
303                 if (in_group_p(inode->i_gid))
304                         mode >>= 3;
305         }
306 
307         /*
308          * If the DACs are ok we don't need any capability check.
309          */
310         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
311                 return 0;
312         return -EACCES;
313 }
314 
315 /**
316  * generic_permission -  check for access rights on a Posix-like filesystem
317  * @inode:      inode to check access rights for
318  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319  *
320  * Used to check for read/write/execute permissions on a file.
321  * We use "fsuid" for this, letting us set arbitrary permissions
322  * for filesystem access without changing the "normal" uids which
323  * are used for other things.
324  *
325  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
326  * request cannot be satisfied (eg. requires blocking or too much complexity).
327  * It would then be called again in ref-walk mode.
328  */
329 int generic_permission(struct inode *inode, int mask)
330 {
331         int ret;
332 
333         /*
334          * Do the basic permission checks.
335          */
336         ret = acl_permission_check(inode, mask);
337         if (ret != -EACCES)
338                 return ret;
339 
340         if (S_ISDIR(inode->i_mode)) {
341                 /* DACs are overridable for directories */
342                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343                         return 0;
344                 if (!(mask & MAY_WRITE))
345                         if (capable_wrt_inode_uidgid(inode,
346                                                      CAP_DAC_READ_SEARCH))
347                                 return 0;
348                 return -EACCES;
349         }
350         /*
351          * Read/write DACs are always overridable.
352          * Executable DACs are overridable when there is
353          * at least one exec bit set.
354          */
355         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
356                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
357                         return 0;
358 
359         /*
360          * Searching includes executable on directories, else just read.
361          */
362         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
363         if (mask == MAY_READ)
364                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
365                         return 0;
366 
367         return -EACCES;
368 }
369 EXPORT_SYMBOL(generic_permission);
370 
371 /*
372  * We _really_ want to just do "generic_permission()" without
373  * even looking at the inode->i_op values. So we keep a cache
374  * flag in inode->i_opflags, that says "this has not special
375  * permission function, use the fast case".
376  */
377 static inline int do_inode_permission(struct inode *inode, int mask)
378 {
379         if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
380                 if (likely(inode->i_op->permission))
381                         return inode->i_op->permission(inode, mask);
382 
383                 /* This gets set once for the inode lifetime */
384                 spin_lock(&inode->i_lock);
385                 inode->i_opflags |= IOP_FASTPERM;
386                 spin_unlock(&inode->i_lock);
387         }
388         return generic_permission(inode, mask);
389 }
390 
391 /**
392  * __inode_permission - Check for access rights to a given inode
393  * @inode: Inode to check permission on
394  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395  *
396  * Check for read/write/execute permissions on an inode.
397  *
398  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399  *
400  * This does not check for a read-only file system.  You probably want
401  * inode_permission().
402  */
403 int __inode_permission(struct inode *inode, int mask)
404 {
405         int retval;
406 
407         if (unlikely(mask & MAY_WRITE)) {
408                 /*
409                  * Nobody gets write access to an immutable file.
410                  */
411                 if (IS_IMMUTABLE(inode))
412                         return -EACCES;
413         }
414 
415         retval = do_inode_permission(inode, mask);
416         if (retval)
417                 return retval;
418 
419         retval = devcgroup_inode_permission(inode, mask);
420         if (retval)
421                 return retval;
422 
423         return security_inode_permission(inode, mask);
424 }
425 EXPORT_SYMBOL(__inode_permission);
426 
427 /**
428  * sb_permission - Check superblock-level permissions
429  * @sb: Superblock of inode to check permission on
430  * @inode: Inode to check permission on
431  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432  *
433  * Separate out file-system wide checks from inode-specific permission checks.
434  */
435 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 {
437         if (unlikely(mask & MAY_WRITE)) {
438                 umode_t mode = inode->i_mode;
439 
440                 /* Nobody gets write access to a read-only fs. */
441                 if ((sb->s_flags & MS_RDONLY) &&
442                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
443                         return -EROFS;
444         }
445         return 0;
446 }
447 
448 /**
449  * inode_permission - Check for access rights to a given inode
450  * @inode: Inode to check permission on
451  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452  *
453  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
454  * this, letting us set arbitrary permissions for filesystem access without
455  * changing the "normal" UIDs which are used for other things.
456  *
457  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458  */
459 int inode_permission(struct inode *inode, int mask)
460 {
461         int retval;
462 
463         retval = sb_permission(inode->i_sb, inode, mask);
464         if (retval)
465                 return retval;
466         return __inode_permission(inode, mask);
467 }
468 EXPORT_SYMBOL(inode_permission);
469 
470 /**
471  * path_get - get a reference to a path
472  * @path: path to get the reference to
473  *
474  * Given a path increment the reference count to the dentry and the vfsmount.
475  */
476 void path_get(const struct path *path)
477 {
478         mntget(path->mnt);
479         dget(path->dentry);
480 }
481 EXPORT_SYMBOL(path_get);
482 
483 /**
484  * path_put - put a reference to a path
485  * @path: path to put the reference to
486  *
487  * Given a path decrement the reference count to the dentry and the vfsmount.
488  */
489 void path_put(const struct path *path)
490 {
491         dput(path->dentry);
492         mntput(path->mnt);
493 }
494 EXPORT_SYMBOL(path_put);
495 
496 #define EMBEDDED_LEVELS 2
497 struct nameidata {
498         struct path     path;
499         struct qstr     last;
500         struct path     root;
501         struct inode    *inode; /* path.dentry.d_inode */
502         unsigned int    flags;
503         unsigned        seq, m_seq;
504         int             last_type;
505         unsigned        depth;
506         int             total_link_count;
507         struct saved {
508                 struct path link;
509                 struct delayed_call done;
510                 const char *name;
511                 unsigned seq;
512         } *stack, internal[EMBEDDED_LEVELS];
513         struct filename *name;
514         struct nameidata *saved;
515         struct inode    *link_inode;
516         unsigned        root_seq;
517         int             dfd;
518 };
519 
520 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
521 {
522         struct nameidata *old = current->nameidata;
523         p->stack = p->internal;
524         p->dfd = dfd;
525         p->name = name;
526         p->total_link_count = old ? old->total_link_count : 0;
527         p->saved = old;
528         current->nameidata = p;
529 }
530 
531 static void restore_nameidata(void)
532 {
533         struct nameidata *now = current->nameidata, *old = now->saved;
534 
535         current->nameidata = old;
536         if (old)
537                 old->total_link_count = now->total_link_count;
538         if (now->stack != now->internal)
539                 kfree(now->stack);
540 }
541 
542 static int __nd_alloc_stack(struct nameidata *nd)
543 {
544         struct saved *p;
545 
546         if (nd->flags & LOOKUP_RCU) {
547                 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
548                                   GFP_ATOMIC);
549                 if (unlikely(!p))
550                         return -ECHILD;
551         } else {
552                 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
553                                   GFP_KERNEL);
554                 if (unlikely(!p))
555                         return -ENOMEM;
556         }
557         memcpy(p, nd->internal, sizeof(nd->internal));
558         nd->stack = p;
559         return 0;
560 }
561 
562 /**
563  * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
564  * @path: nameidate to verify
565  *
566  * Rename can sometimes move a file or directory outside of a bind
567  * mount, path_connected allows those cases to be detected.
568  */
569 static bool path_connected(const struct path *path)
570 {
571         struct vfsmount *mnt = path->mnt;
572 
573         /* Only bind mounts can have disconnected paths */
574         if (mnt->mnt_root == mnt->mnt_sb->s_root)
575                 return true;
576 
577         return is_subdir(path->dentry, mnt->mnt_root);
578 }
579 
580 static inline int nd_alloc_stack(struct nameidata *nd)
581 {
582         if (likely(nd->depth != EMBEDDED_LEVELS))
583                 return 0;
584         if (likely(nd->stack != nd->internal))
585                 return 0;
586         return __nd_alloc_stack(nd);
587 }
588 
589 static void drop_links(struct nameidata *nd)
590 {
591         int i = nd->depth;
592         while (i--) {
593                 struct saved *last = nd->stack + i;
594                 do_delayed_call(&last->done);
595                 clear_delayed_call(&last->done);
596         }
597 }
598 
599 static void terminate_walk(struct nameidata *nd)
600 {
601         drop_links(nd);
602         if (!(nd->flags & LOOKUP_RCU)) {
603                 int i;
604                 path_put(&nd->path);
605                 for (i = 0; i < nd->depth; i++)
606                         path_put(&nd->stack[i].link);
607                 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
608                         path_put(&nd->root);
609                         nd->root.mnt = NULL;
610                 }
611         } else {
612                 nd->flags &= ~LOOKUP_RCU;
613                 if (!(nd->flags & LOOKUP_ROOT))
614                         nd->root.mnt = NULL;
615                 rcu_read_unlock();
616         }
617         nd->depth = 0;
618 }
619 
620 /* path_put is needed afterwards regardless of success or failure */
621 static bool legitimize_path(struct nameidata *nd,
622                             struct path *path, unsigned seq)
623 {
624         int res = __legitimize_mnt(path->mnt, nd->m_seq);
625         if (unlikely(res)) {
626                 if (res > 0)
627                         path->mnt = NULL;
628                 path->dentry = NULL;
629                 return false;
630         }
631         if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
632                 path->dentry = NULL;
633                 return false;
634         }
635         return !read_seqcount_retry(&path->dentry->d_seq, seq);
636 }
637 
638 static bool legitimize_links(struct nameidata *nd)
639 {
640         int i;
641         for (i = 0; i < nd->depth; i++) {
642                 struct saved *last = nd->stack + i;
643                 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
644                         drop_links(nd);
645                         nd->depth = i + 1;
646                         return false;
647                 }
648         }
649         return true;
650 }
651 
652 /*
653  * Path walking has 2 modes, rcu-walk and ref-walk (see
654  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
655  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
656  * normal reference counts on dentries and vfsmounts to transition to ref-walk
657  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
658  * got stuck, so ref-walk may continue from there. If this is not successful
659  * (eg. a seqcount has changed), then failure is returned and it's up to caller
660  * to restart the path walk from the beginning in ref-walk mode.
661  */
662 
663 /**
664  * unlazy_walk - try to switch to ref-walk mode.
665  * @nd: nameidata pathwalk data
666  * @dentry: child of nd->path.dentry or NULL
667  * @seq: seq number to check dentry against
668  * Returns: 0 on success, -ECHILD on failure
669  *
670  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
671  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
672  * @nd or NULL.  Must be called from rcu-walk context.
673  * Nothing should touch nameidata between unlazy_walk() failure and
674  * terminate_walk().
675  */
676 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
677 {
678         struct dentry *parent = nd->path.dentry;
679 
680         BUG_ON(!(nd->flags & LOOKUP_RCU));
681 
682         nd->flags &= ~LOOKUP_RCU;
683         if (unlikely(!legitimize_links(nd)))
684                 goto out2;
685         if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
686                 goto out2;
687         if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
688                 goto out1;
689 
690         /*
691          * For a negative lookup, the lookup sequence point is the parents
692          * sequence point, and it only needs to revalidate the parent dentry.
693          *
694          * For a positive lookup, we need to move both the parent and the
695          * dentry from the RCU domain to be properly refcounted. And the
696          * sequence number in the dentry validates *both* dentry counters,
697          * since we checked the sequence number of the parent after we got
698          * the child sequence number. So we know the parent must still
699          * be valid if the child sequence number is still valid.
700          */
701         if (!dentry) {
702                 if (read_seqcount_retry(&parent->d_seq, nd->seq))
703                         goto out;
704                 BUG_ON(nd->inode != parent->d_inode);
705         } else {
706                 if (!lockref_get_not_dead(&dentry->d_lockref))
707                         goto out;
708                 if (read_seqcount_retry(&dentry->d_seq, seq))
709                         goto drop_dentry;
710         }
711 
712         /*
713          * Sequence counts matched. Now make sure that the root is
714          * still valid and get it if required.
715          */
716         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
717                 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
718                         rcu_read_unlock();
719                         dput(dentry);
720                         return -ECHILD;
721                 }
722         }
723 
724         rcu_read_unlock();
725         return 0;
726 
727 drop_dentry:
728         rcu_read_unlock();
729         dput(dentry);
730         goto drop_root_mnt;
731 out2:
732         nd->path.mnt = NULL;
733 out1:
734         nd->path.dentry = NULL;
735 out:
736         rcu_read_unlock();
737 drop_root_mnt:
738         if (!(nd->flags & LOOKUP_ROOT))
739                 nd->root.mnt = NULL;
740         return -ECHILD;
741 }
742 
743 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
744 {
745         if (unlikely(!legitimize_path(nd, link, seq))) {
746                 drop_links(nd);
747                 nd->depth = 0;
748                 nd->flags &= ~LOOKUP_RCU;
749                 nd->path.mnt = NULL;
750                 nd->path.dentry = NULL;
751                 if (!(nd->flags & LOOKUP_ROOT))
752                         nd->root.mnt = NULL;
753                 rcu_read_unlock();
754         } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
755                 return 0;
756         }
757         path_put(link);
758         return -ECHILD;
759 }
760 
761 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
762 {
763         return dentry->d_op->d_revalidate(dentry, flags);
764 }
765 
766 /**
767  * complete_walk - successful completion of path walk
768  * @nd:  pointer nameidata
769  *
770  * If we had been in RCU mode, drop out of it and legitimize nd->path.
771  * Revalidate the final result, unless we'd already done that during
772  * the path walk or the filesystem doesn't ask for it.  Return 0 on
773  * success, -error on failure.  In case of failure caller does not
774  * need to drop nd->path.
775  */
776 static int complete_walk(struct nameidata *nd)
777 {
778         struct dentry *dentry = nd->path.dentry;
779         int status;
780 
781         if (nd->flags & LOOKUP_RCU) {
782                 if (!(nd->flags & LOOKUP_ROOT))
783                         nd->root.mnt = NULL;
784                 if (unlikely(unlazy_walk(nd, NULL, 0)))
785                         return -ECHILD;
786         }
787 
788         if (likely(!(nd->flags & LOOKUP_JUMPED)))
789                 return 0;
790 
791         if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
792                 return 0;
793 
794         status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
795         if (status > 0)
796                 return 0;
797 
798         if (!status)
799                 status = -ESTALE;
800 
801         return status;
802 }
803 
804 static void set_root(struct nameidata *nd)
805 {
806         struct fs_struct *fs = current->fs;
807 
808         if (nd->flags & LOOKUP_RCU) {
809                 unsigned seq;
810 
811                 do {
812                         seq = read_seqcount_begin(&fs->seq);
813                         nd->root = fs->root;
814                         nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
815                 } while (read_seqcount_retry(&fs->seq, seq));
816         } else {
817                 get_fs_root(fs, &nd->root);
818         }
819 }
820 
821 static void path_put_conditional(struct path *path, struct nameidata *nd)
822 {
823         dput(path->dentry);
824         if (path->mnt != nd->path.mnt)
825                 mntput(path->mnt);
826 }
827 
828 static inline void path_to_nameidata(const struct path *path,
829                                         struct nameidata *nd)
830 {
831         if (!(nd->flags & LOOKUP_RCU)) {
832                 dput(nd->path.dentry);
833                 if (nd->path.mnt != path->mnt)
834                         mntput(nd->path.mnt);
835         }
836         nd->path.mnt = path->mnt;
837         nd->path.dentry = path->dentry;
838 }
839 
840 static int nd_jump_root(struct nameidata *nd)
841 {
842         if (nd->flags & LOOKUP_RCU) {
843                 struct dentry *d;
844                 nd->path = nd->root;
845                 d = nd->path.dentry;
846                 nd->inode = d->d_inode;
847                 nd->seq = nd->root_seq;
848                 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
849                         return -ECHILD;
850         } else {
851                 path_put(&nd->path);
852                 nd->path = nd->root;
853                 path_get(&nd->path);
854                 nd->inode = nd->path.dentry->d_inode;
855         }
856         nd->flags |= LOOKUP_JUMPED;
857         return 0;
858 }
859 
860 /*
861  * Helper to directly jump to a known parsed path from ->get_link,
862  * caller must have taken a reference to path beforehand.
863  */
864 void nd_jump_link(struct path *path)
865 {
866         struct nameidata *nd = current->nameidata;
867         path_put(&nd->path);
868 
869         nd->path = *path;
870         nd->inode = nd->path.dentry->d_inode;
871         nd->flags |= LOOKUP_JUMPED;
872 }
873 
874 static inline void put_link(struct nameidata *nd)
875 {
876         struct saved *last = nd->stack + --nd->depth;
877         do_delayed_call(&last->done);
878         if (!(nd->flags & LOOKUP_RCU))
879                 path_put(&last->link);
880 }
881 
882 int sysctl_protected_symlinks __read_mostly = 0;
883 int sysctl_protected_hardlinks __read_mostly = 0;
884 
885 /**
886  * may_follow_link - Check symlink following for unsafe situations
887  * @nd: nameidata pathwalk data
888  *
889  * In the case of the sysctl_protected_symlinks sysctl being enabled,
890  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891  * in a sticky world-writable directory. This is to protect privileged
892  * processes from failing races against path names that may change out
893  * from under them by way of other users creating malicious symlinks.
894  * It will permit symlinks to be followed only when outside a sticky
895  * world-writable directory, or when the uid of the symlink and follower
896  * match, or when the directory owner matches the symlink's owner.
897  *
898  * Returns 0 if following the symlink is allowed, -ve on error.
899  */
900 static inline int may_follow_link(struct nameidata *nd)
901 {
902         const struct inode *inode;
903         const struct inode *parent;
904 
905         if (!sysctl_protected_symlinks)
906                 return 0;
907 
908         /* Allowed if owner and follower match. */
909         inode = nd->link_inode;
910         if (uid_eq(current_cred()->fsuid, inode->i_uid))
911                 return 0;
912 
913         /* Allowed if parent directory not sticky and world-writable. */
914         parent = nd->inode;
915         if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
916                 return 0;
917 
918         /* Allowed if parent directory and link owner match. */
919         if (uid_eq(parent->i_uid, inode->i_uid))
920                 return 0;
921 
922         if (nd->flags & LOOKUP_RCU)
923                 return -ECHILD;
924 
925         audit_log_link_denied("follow_link", &nd->stack[0].link);
926         return -EACCES;
927 }
928 
929 /**
930  * safe_hardlink_source - Check for safe hardlink conditions
931  * @inode: the source inode to hardlink from
932  *
933  * Return false if at least one of the following conditions:
934  *    - inode is not a regular file
935  *    - inode is setuid
936  *    - inode is setgid and group-exec
937  *    - access failure for read and write
938  *
939  * Otherwise returns true.
940  */
941 static bool safe_hardlink_source(struct inode *inode)
942 {
943         umode_t mode = inode->i_mode;
944 
945         /* Special files should not get pinned to the filesystem. */
946         if (!S_ISREG(mode))
947                 return false;
948 
949         /* Setuid files should not get pinned to the filesystem. */
950         if (mode & S_ISUID)
951                 return false;
952 
953         /* Executable setgid files should not get pinned to the filesystem. */
954         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
955                 return false;
956 
957         /* Hardlinking to unreadable or unwritable sources is dangerous. */
958         if (inode_permission(inode, MAY_READ | MAY_WRITE))
959                 return false;
960 
961         return true;
962 }
963 
964 /**
965  * may_linkat - Check permissions for creating a hardlink
966  * @link: the source to hardlink from
967  *
968  * Block hardlink when all of:
969  *  - sysctl_protected_hardlinks enabled
970  *  - fsuid does not match inode
971  *  - hardlink source is unsafe (see safe_hardlink_source() above)
972  *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
973  *
974  * Returns 0 if successful, -ve on error.
975  */
976 static int may_linkat(struct path *link)
977 {
978         struct inode *inode;
979 
980         if (!sysctl_protected_hardlinks)
981                 return 0;
982 
983         inode = link->dentry->d_inode;
984 
985         /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
986          * otherwise, it must be a safe source.
987          */
988         if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
989                 return 0;
990 
991         audit_log_link_denied("linkat", link);
992         return -EPERM;
993 }
994 
995 static __always_inline
996 const char *get_link(struct nameidata *nd)
997 {
998         struct saved *last = nd->stack + nd->depth - 1;
999         struct dentry *dentry = last->link.dentry;
1000         struct inode *inode = nd->link_inode;
1001         int error;
1002         const char *res;
1003 
1004         if (!(nd->flags & LOOKUP_RCU)) {
1005                 touch_atime(&last->link);
1006                 cond_resched();
1007         } else if (atime_needs_update(&last->link, inode)) {
1008                 if (unlikely(unlazy_walk(nd, NULL, 0)))
1009                         return ERR_PTR(-ECHILD);
1010                 touch_atime(&last->link);
1011         }
1012 
1013         error = security_inode_follow_link(dentry, inode,
1014                                            nd->flags & LOOKUP_RCU);
1015         if (unlikely(error))
1016                 return ERR_PTR(error);
1017 
1018         nd->last_type = LAST_BIND;
1019         res = inode->i_link;
1020         if (!res) {
1021                 const char * (*get)(struct dentry *, struct inode *,
1022                                 struct delayed_call *);
1023                 get = inode->i_op->get_link;
1024                 if (nd->flags & LOOKUP_RCU) {
1025                         res = get(NULL, inode, &last->done);
1026                         if (res == ERR_PTR(-ECHILD)) {
1027                                 if (unlikely(unlazy_walk(nd, NULL, 0)))
1028                                         return ERR_PTR(-ECHILD);
1029                                 res = get(dentry, inode, &last->done);
1030                         }
1031                 } else {
1032                         res = get(dentry, inode, &last->done);
1033                 }
1034                 if (IS_ERR_OR_NULL(res))
1035                         return res;
1036         }
1037         if (*res == '/') {
1038                 if (!nd->root.mnt)
1039                         set_root(nd);
1040                 if (unlikely(nd_jump_root(nd)))
1041                         return ERR_PTR(-ECHILD);
1042                 while (unlikely(*++res == '/'))
1043                         ;
1044         }
1045         if (!*res)
1046                 res = NULL;
1047         return res;
1048 }
1049 
1050 /*
1051  * follow_up - Find the mountpoint of path's vfsmount
1052  *
1053  * Given a path, find the mountpoint of its source file system.
1054  * Replace @path with the path of the mountpoint in the parent mount.
1055  * Up is towards /.
1056  *
1057  * Return 1 if we went up a level and 0 if we were already at the
1058  * root.
1059  */
1060 int follow_up(struct path *path)
1061 {
1062         struct mount *mnt = real_mount(path->mnt);
1063         struct mount *parent;
1064         struct dentry *mountpoint;
1065 
1066         read_seqlock_excl(&mount_lock);
1067         parent = mnt->mnt_parent;
1068         if (parent == mnt) {
1069                 read_sequnlock_excl(&mount_lock);
1070                 return 0;
1071         }
1072         mntget(&parent->mnt);
1073         mountpoint = dget(mnt->mnt_mountpoint);
1074         read_sequnlock_excl(&mount_lock);
1075         dput(path->dentry);
1076         path->dentry = mountpoint;
1077         mntput(path->mnt);
1078         path->mnt = &parent->mnt;
1079         return 1;
1080 }
1081 EXPORT_SYMBOL(follow_up);
1082 
1083 /*
1084  * Perform an automount
1085  * - return -EISDIR to tell follow_managed() to stop and return the path we
1086  *   were called with.
1087  */
1088 static int follow_automount(struct path *path, struct nameidata *nd,
1089                             bool *need_mntput)
1090 {
1091         struct vfsmount *mnt;
1092         int err;
1093 
1094         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1095                 return -EREMOTE;
1096 
1097         /* We don't want to mount if someone's just doing a stat -
1098          * unless they're stat'ing a directory and appended a '/' to
1099          * the name.
1100          *
1101          * We do, however, want to mount if someone wants to open or
1102          * create a file of any type under the mountpoint, wants to
1103          * traverse through the mountpoint or wants to open the
1104          * mounted directory.  Also, autofs may mark negative dentries
1105          * as being automount points.  These will need the attentions
1106          * of the daemon to instantiate them before they can be used.
1107          */
1108         if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1109                            LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1110             path->dentry->d_inode)
1111                 return -EISDIR;
1112 
1113         nd->total_link_count++;
1114         if (nd->total_link_count >= 40)
1115                 return -ELOOP;
1116 
1117         mnt = path->dentry->d_op->d_automount(path);
1118         if (IS_ERR(mnt)) {
1119                 /*
1120                  * The filesystem is allowed to return -EISDIR here to indicate
1121                  * it doesn't want to automount.  For instance, autofs would do
1122                  * this so that its userspace daemon can mount on this dentry.
1123                  *
1124                  * However, we can only permit this if it's a terminal point in
1125                  * the path being looked up; if it wasn't then the remainder of
1126                  * the path is inaccessible and we should say so.
1127                  */
1128                 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1129                         return -EREMOTE;
1130                 return PTR_ERR(mnt);
1131         }
1132 
1133         if (!mnt) /* mount collision */
1134                 return 0;
1135 
1136         if (!*need_mntput) {
1137                 /* lock_mount() may release path->mnt on error */
1138                 mntget(path->mnt);
1139                 *need_mntput = true;
1140         }
1141         err = finish_automount(mnt, path);
1142 
1143         switch (err) {
1144         case -EBUSY:
1145                 /* Someone else made a mount here whilst we were busy */
1146                 return 0;
1147         case 0:
1148                 path_put(path);
1149                 path->mnt = mnt;
1150                 path->dentry = dget(mnt->mnt_root);
1151                 return 0;
1152         default:
1153                 return err;
1154         }
1155 
1156 }
1157 
1158 /*
1159  * Handle a dentry that is managed in some way.
1160  * - Flagged for transit management (autofs)
1161  * - Flagged as mountpoint
1162  * - Flagged as automount point
1163  *
1164  * This may only be called in refwalk mode.
1165  *
1166  * Serialization is taken care of in namespace.c
1167  */
1168 static int follow_managed(struct path *path, struct nameidata *nd)
1169 {
1170         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1171         unsigned managed;
1172         bool need_mntput = false;
1173         int ret = 0;
1174 
1175         /* Given that we're not holding a lock here, we retain the value in a
1176          * local variable for each dentry as we look at it so that we don't see
1177          * the components of that value change under us */
1178         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1179                managed &= DCACHE_MANAGED_DENTRY,
1180                unlikely(managed != 0)) {
1181                 /* Allow the filesystem to manage the transit without i_mutex
1182                  * being held. */
1183                 if (managed & DCACHE_MANAGE_TRANSIT) {
1184                         BUG_ON(!path->dentry->d_op);
1185                         BUG_ON(!path->dentry->d_op->d_manage);
1186                         ret = path->dentry->d_op->d_manage(path->dentry, false);
1187                         if (ret < 0)
1188                                 break;
1189                 }
1190 
1191                 /* Transit to a mounted filesystem. */
1192                 if (managed & DCACHE_MOUNTED) {
1193                         struct vfsmount *mounted = lookup_mnt(path);
1194                         if (mounted) {
1195                                 dput(path->dentry);
1196                                 if (need_mntput)
1197                                         mntput(path->mnt);
1198                                 path->mnt = mounted;
1199                                 path->dentry = dget(mounted->mnt_root);
1200                                 need_mntput = true;
1201                                 continue;
1202                         }
1203 
1204                         /* Something is mounted on this dentry in another
1205                          * namespace and/or whatever was mounted there in this
1206                          * namespace got unmounted before lookup_mnt() could
1207                          * get it */
1208                 }
1209 
1210                 /* Handle an automount point */
1211                 if (managed & DCACHE_NEED_AUTOMOUNT) {
1212                         ret = follow_automount(path, nd, &need_mntput);
1213                         if (ret < 0)
1214                                 break;
1215                         continue;
1216                 }
1217 
1218                 /* We didn't change the current path point */
1219                 break;
1220         }
1221 
1222         if (need_mntput && path->mnt == mnt)
1223                 mntput(path->mnt);
1224         if (ret == -EISDIR || !ret)
1225                 ret = 1;
1226         if (need_mntput)
1227                 nd->flags |= LOOKUP_JUMPED;
1228         if (unlikely(ret < 0))
1229                 path_put_conditional(path, nd);
1230         return ret;
1231 }
1232 
1233 int follow_down_one(struct path *path)
1234 {
1235         struct vfsmount *mounted;
1236 
1237         mounted = lookup_mnt(path);
1238         if (mounted) {
1239                 dput(path->dentry);
1240                 mntput(path->mnt);
1241                 path->mnt = mounted;
1242                 path->dentry = dget(mounted->mnt_root);
1243                 return 1;
1244         }
1245         return 0;
1246 }
1247 EXPORT_SYMBOL(follow_down_one);
1248 
1249 static inline int managed_dentry_rcu(struct dentry *dentry)
1250 {
1251         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1252                 dentry->d_op->d_manage(dentry, true) : 0;
1253 }
1254 
1255 /*
1256  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1257  * we meet a managed dentry that would need blocking.
1258  */
1259 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1260                                struct inode **inode, unsigned *seqp)
1261 {
1262         for (;;) {
1263                 struct mount *mounted;
1264                 /*
1265                  * Don't forget we might have a non-mountpoint managed dentry
1266                  * that wants to block transit.
1267                  */
1268                 switch (managed_dentry_rcu(path->dentry)) {
1269                 case -ECHILD:
1270                 default:
1271                         return false;
1272                 case -EISDIR:
1273                         return true;
1274                 case 0:
1275                         break;
1276                 }
1277 
1278                 if (!d_mountpoint(path->dentry))
1279                         return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1280 
1281                 mounted = __lookup_mnt(path->mnt, path->dentry);
1282                 if (!mounted)
1283                         break;
1284                 path->mnt = &mounted->mnt;
1285                 path->dentry = mounted->mnt.mnt_root;
1286                 nd->flags |= LOOKUP_JUMPED;
1287                 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1288                 /*
1289                  * Update the inode too. We don't need to re-check the
1290                  * dentry sequence number here after this d_inode read,
1291                  * because a mount-point is always pinned.
1292                  */
1293                 *inode = path->dentry->d_inode;
1294         }
1295         return !read_seqretry(&mount_lock, nd->m_seq) &&
1296                 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1297 }
1298 
1299 static int follow_dotdot_rcu(struct nameidata *nd)
1300 {
1301         struct inode *inode = nd->inode;
1302 
1303         while (1) {
1304                 if (path_equal(&nd->path, &nd->root))
1305                         break;
1306                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1307                         struct dentry *old = nd->path.dentry;
1308                         struct dentry *parent = old->d_parent;
1309                         unsigned seq;
1310 
1311                         inode = parent->d_inode;
1312                         seq = read_seqcount_begin(&parent->d_seq);
1313                         if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1314                                 return -ECHILD;
1315                         nd->path.dentry = parent;
1316                         nd->seq = seq;
1317                         if (unlikely(!path_connected(&nd->path)))
1318                                 return -ENOENT;
1319                         break;
1320                 } else {
1321                         struct mount *mnt = real_mount(nd->path.mnt);
1322                         struct mount *mparent = mnt->mnt_parent;
1323                         struct dentry *mountpoint = mnt->mnt_mountpoint;
1324                         struct inode *inode2 = mountpoint->d_inode;
1325                         unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1326                         if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1327                                 return -ECHILD;
1328                         if (&mparent->mnt == nd->path.mnt)
1329                                 break;
1330                         /* we know that mountpoint was pinned */
1331                         nd->path.dentry = mountpoint;
1332                         nd->path.mnt = &mparent->mnt;
1333                         inode = inode2;
1334                         nd->seq = seq;
1335                 }
1336         }
1337         while (unlikely(d_mountpoint(nd->path.dentry))) {
1338                 struct mount *mounted;
1339                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1340                 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1341                         return -ECHILD;
1342                 if (!mounted)
1343                         break;
1344                 nd->path.mnt = &mounted->mnt;
1345                 nd->path.dentry = mounted->mnt.mnt_root;
1346                 inode = nd->path.dentry->d_inode;
1347                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1348         }
1349         nd->inode = inode;
1350         return 0;
1351 }
1352 
1353 /*
1354  * Follow down to the covering mount currently visible to userspace.  At each
1355  * point, the filesystem owning that dentry may be queried as to whether the
1356  * caller is permitted to proceed or not.
1357  */
1358 int follow_down(struct path *path)
1359 {
1360         unsigned managed;
1361         int ret;
1362 
1363         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1364                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1365                 /* Allow the filesystem to manage the transit without i_mutex
1366                  * being held.
1367                  *
1368                  * We indicate to the filesystem if someone is trying to mount
1369                  * something here.  This gives autofs the chance to deny anyone
1370                  * other than its daemon the right to mount on its
1371                  * superstructure.
1372                  *
1373                  * The filesystem may sleep at this point.
1374                  */
1375                 if (managed & DCACHE_MANAGE_TRANSIT) {
1376                         BUG_ON(!path->dentry->d_op);
1377                         BUG_ON(!path->dentry->d_op->d_manage);
1378                         ret = path->dentry->d_op->d_manage(
1379                                 path->dentry, false);
1380                         if (ret < 0)
1381                                 return ret == -EISDIR ? 0 : ret;
1382                 }
1383 
1384                 /* Transit to a mounted filesystem. */
1385                 if (managed & DCACHE_MOUNTED) {
1386                         struct vfsmount *mounted = lookup_mnt(path);
1387                         if (!mounted)
1388                                 break;
1389                         dput(path->dentry);
1390                         mntput(path->mnt);
1391                         path->mnt = mounted;
1392                         path->dentry = dget(mounted->mnt_root);
1393                         continue;
1394                 }
1395 
1396                 /* Don't handle automount points here */
1397                 break;
1398         }
1399         return 0;
1400 }
1401 EXPORT_SYMBOL(follow_down);
1402 
1403 /*
1404  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1405  */
1406 static void follow_mount(struct path *path)
1407 {
1408         while (d_mountpoint(path->dentry)) {
1409                 struct vfsmount *mounted = lookup_mnt(path);
1410                 if (!mounted)
1411                         break;
1412                 dput(path->dentry);
1413                 mntput(path->mnt);
1414                 path->mnt = mounted;
1415                 path->dentry = dget(mounted->mnt_root);
1416         }
1417 }
1418 
1419 static int path_parent_directory(struct path *path)
1420 {
1421         struct dentry *old = path->dentry;
1422         /* rare case of legitimate dget_parent()... */
1423         path->dentry = dget_parent(path->dentry);
1424         dput(old);
1425         if (unlikely(!path_connected(path)))
1426                 return -ENOENT;
1427         return 0;
1428 }
1429 
1430 static int follow_dotdot(struct nameidata *nd)
1431 {
1432         while(1) {
1433                 if (nd->path.dentry == nd->root.dentry &&
1434                     nd->path.mnt == nd->root.mnt) {
1435                         break;
1436                 }
1437                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1438                         int ret = path_parent_directory(&nd->path);
1439                         if (ret)
1440                                 return ret;
1441                         break;
1442                 }
1443                 if (!follow_up(&nd->path))
1444                         break;
1445         }
1446         follow_mount(&nd->path);
1447         nd->inode = nd->path.dentry->d_inode;
1448         return 0;
1449 }
1450 
1451 /*
1452  * This looks up the name in dcache, possibly revalidates the old dentry and
1453  * allocates a new one if not found or not valid.  In the need_lookup argument
1454  * returns whether i_op->lookup is necessary.
1455  */
1456 static struct dentry *lookup_dcache(const struct qstr *name,
1457                                     struct dentry *dir,
1458                                     unsigned int flags)
1459 {
1460         struct dentry *dentry;
1461         int error;
1462 
1463         dentry = d_lookup(dir, name);
1464         if (dentry) {
1465                 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1466                         error = d_revalidate(dentry, flags);
1467                         if (unlikely(error <= 0)) {
1468                                 if (!error)
1469                                         d_invalidate(dentry);
1470                                 dput(dentry);
1471                                 return ERR_PTR(error);
1472                         }
1473                 }
1474         }
1475         return dentry;
1476 }
1477 
1478 /*
1479  * Call i_op->lookup on the dentry.  The dentry must be negative and
1480  * unhashed.
1481  *
1482  * dir->d_inode->i_mutex must be held
1483  */
1484 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1485                                   unsigned int flags)
1486 {
1487         struct dentry *old;
1488 
1489         /* Don't create child dentry for a dead directory. */
1490         if (unlikely(IS_DEADDIR(dir))) {
1491                 dput(dentry);
1492                 return ERR_PTR(-ENOENT);
1493         }
1494 
1495         old = dir->i_op->lookup(dir, dentry, flags);
1496         if (unlikely(old)) {
1497                 dput(dentry);
1498                 dentry = old;
1499         }
1500         return dentry;
1501 }
1502 
1503 static struct dentry *__lookup_hash(const struct qstr *name,
1504                 struct dentry *base, unsigned int flags)
1505 {
1506         struct dentry *dentry = lookup_dcache(name, base, flags);
1507 
1508         if (dentry)
1509                 return dentry;
1510 
1511         dentry = d_alloc(base, name);
1512         if (unlikely(!dentry))
1513                 return ERR_PTR(-ENOMEM);
1514 
1515         return lookup_real(base->d_inode, dentry, flags);
1516 }
1517 
1518 static int lookup_fast(struct nameidata *nd,
1519                        struct path *path, struct inode **inode,
1520                        unsigned *seqp)
1521 {
1522         struct vfsmount *mnt = nd->path.mnt;
1523         struct dentry *dentry, *parent = nd->path.dentry;
1524         int status = 1;
1525         int err;
1526 
1527         /*
1528          * Rename seqlock is not required here because in the off chance
1529          * of a false negative due to a concurrent rename, the caller is
1530          * going to fall back to non-racy lookup.
1531          */
1532         if (nd->flags & LOOKUP_RCU) {
1533                 unsigned seq;
1534                 bool negative;
1535                 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1536                 if (unlikely(!dentry)) {
1537                         if (unlazy_walk(nd, NULL, 0))
1538                                 return -ECHILD;
1539                         return 0;
1540                 }
1541 
1542                 /*
1543                  * This sequence count validates that the inode matches
1544                  * the dentry name information from lookup.
1545                  */
1546                 *inode = d_backing_inode(dentry);
1547                 negative = d_is_negative(dentry);
1548                 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1549                         return -ECHILD;
1550 
1551                 /*
1552                  * This sequence count validates that the parent had no
1553                  * changes while we did the lookup of the dentry above.
1554                  *
1555                  * The memory barrier in read_seqcount_begin of child is
1556                  *  enough, we can use __read_seqcount_retry here.
1557                  */
1558                 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1559                         return -ECHILD;
1560 
1561                 *seqp = seq;
1562                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1563                         status = d_revalidate(dentry, nd->flags);
1564                 if (unlikely(status <= 0)) {
1565                         if (unlazy_walk(nd, dentry, seq))
1566                                 return -ECHILD;
1567                         if (status == -ECHILD)
1568                                 status = d_revalidate(dentry, nd->flags);
1569                 } else {
1570                         /*
1571                          * Note: do negative dentry check after revalidation in
1572                          * case that drops it.
1573                          */
1574                         if (unlikely(negative))
1575                                 return -ENOENT;
1576                         path->mnt = mnt;
1577                         path->dentry = dentry;
1578                         if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1579                                 return 1;
1580                         if (unlazy_walk(nd, dentry, seq))
1581                                 return -ECHILD;
1582                 }
1583         } else {
1584                 dentry = __d_lookup(parent, &nd->last);
1585                 if (unlikely(!dentry))
1586                         return 0;
1587                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1588                         status = d_revalidate(dentry, nd->flags);
1589         }
1590         if (unlikely(status <= 0)) {
1591                 if (!status)
1592                         d_invalidate(dentry);
1593                 dput(dentry);
1594                 return status;
1595         }
1596         if (unlikely(d_is_negative(dentry))) {
1597                 dput(dentry);
1598                 return -ENOENT;
1599         }
1600 
1601         path->mnt = mnt;
1602         path->dentry = dentry;
1603         err = follow_managed(path, nd);
1604         if (likely(err > 0))
1605                 *inode = d_backing_inode(path->dentry);
1606         return err;
1607 }
1608 
1609 /* Fast lookup failed, do it the slow way */
1610 static struct dentry *lookup_slow(const struct qstr *name,
1611                                   struct dentry *dir,
1612                                   unsigned int flags)
1613 {
1614         struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1615         struct inode *inode = dir->d_inode;
1616         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1617 
1618         inode_lock_shared(inode);
1619         /* Don't go there if it's already dead */
1620         if (unlikely(IS_DEADDIR(inode)))
1621                 goto out;
1622 again:
1623         dentry = d_alloc_parallel(dir, name, &wq);
1624         if (IS_ERR(dentry))
1625                 goto out;
1626         if (unlikely(!d_in_lookup(dentry))) {
1627                 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1628                     !(flags & LOOKUP_NO_REVAL)) {
1629                         int error = d_revalidate(dentry, flags);
1630                         if (unlikely(error <= 0)) {
1631                                 if (!error) {
1632                                         d_invalidate(dentry);
1633                                         dput(dentry);
1634                                         goto again;
1635                                 }
1636                                 dput(dentry);
1637                                 dentry = ERR_PTR(error);
1638                         }
1639                 }
1640         } else {
1641                 old = inode->i_op->lookup(inode, dentry, flags);
1642                 d_lookup_done(dentry);
1643                 if (unlikely(old)) {
1644                         dput(dentry);
1645                         dentry = old;
1646                 }
1647         }
1648 out:
1649         inode_unlock_shared(inode);
1650         return dentry;
1651 }
1652 
1653 static inline int may_lookup(struct nameidata *nd)
1654 {
1655         if (nd->flags & LOOKUP_RCU) {
1656                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1657                 if (err != -ECHILD)
1658                         return err;
1659                 if (unlazy_walk(nd, NULL, 0))
1660                         return -ECHILD;
1661         }
1662         return inode_permission(nd->inode, MAY_EXEC);
1663 }
1664 
1665 static inline int handle_dots(struct nameidata *nd, int type)
1666 {
1667         if (type == LAST_DOTDOT) {
1668                 if (!nd->root.mnt)
1669                         set_root(nd);
1670                 if (nd->flags & LOOKUP_RCU) {
1671                         return follow_dotdot_rcu(nd);
1672                 } else
1673                         return follow_dotdot(nd);
1674         }
1675         return 0;
1676 }
1677 
1678 static int pick_link(struct nameidata *nd, struct path *link,
1679                      struct inode *inode, unsigned seq)
1680 {
1681         int error;
1682         struct saved *last;
1683         if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1684                 path_to_nameidata(link, nd);
1685                 return -ELOOP;
1686         }
1687         if (!(nd->flags & LOOKUP_RCU)) {
1688                 if (link->mnt == nd->path.mnt)
1689                         mntget(link->mnt);
1690         }
1691         error = nd_alloc_stack(nd);
1692         if (unlikely(error)) {
1693                 if (error == -ECHILD) {
1694                         if (unlikely(unlazy_link(nd, link, seq)))
1695                                 return -ECHILD;
1696                         error = nd_alloc_stack(nd);
1697                 }
1698                 if (error) {
1699                         path_put(link);
1700                         return error;
1701                 }
1702         }
1703 
1704         last = nd->stack + nd->depth++;
1705         last->link = *link;
1706         clear_delayed_call(&last->done);
1707         nd->link_inode = inode;
1708         last->seq = seq;
1709         return 1;
1710 }
1711 
1712 /*
1713  * Do we need to follow links? We _really_ want to be able
1714  * to do this check without having to look at inode->i_op,
1715  * so we keep a cache of "no, this doesn't need follow_link"
1716  * for the common case.
1717  */
1718 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1719                                      int follow,
1720                                      struct inode *inode, unsigned seq)
1721 {
1722         if (likely(!d_is_symlink(link->dentry)))
1723                 return 0;
1724         if (!follow)
1725                 return 0;
1726         /* make sure that d_is_symlink above matches inode */
1727         if (nd->flags & LOOKUP_RCU) {
1728                 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1729                         return -ECHILD;
1730         }
1731         return pick_link(nd, link, inode, seq);
1732 }
1733 
1734 enum {WALK_GET = 1, WALK_PUT = 2};
1735 
1736 static int walk_component(struct nameidata *nd, int flags)
1737 {
1738         struct path path;
1739         struct inode *inode;
1740         unsigned seq;
1741         int err;
1742         /*
1743          * "." and ".." are special - ".." especially so because it has
1744          * to be able to know about the current root directory and
1745          * parent relationships.
1746          */
1747         if (unlikely(nd->last_type != LAST_NORM)) {
1748                 err = handle_dots(nd, nd->last_type);
1749                 if (flags & WALK_PUT)
1750                         put_link(nd);
1751                 return err;
1752         }
1753         err = lookup_fast(nd, &path, &inode, &seq);
1754         if (unlikely(err <= 0)) {
1755                 if (err < 0)
1756                         return err;
1757                 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1758                                           nd->flags);
1759                 if (IS_ERR(path.dentry))
1760                         return PTR_ERR(path.dentry);
1761 
1762                 path.mnt = nd->path.mnt;
1763                 err = follow_managed(&path, nd);
1764                 if (unlikely(err < 0))
1765                         return err;
1766 
1767                 if (unlikely(d_is_negative(path.dentry))) {
1768                         path_to_nameidata(&path, nd);
1769                         return -ENOENT;
1770                 }
1771 
1772                 seq = 0;        /* we are already out of RCU mode */
1773                 inode = d_backing_inode(path.dentry);
1774         }
1775 
1776         if (flags & WALK_PUT)
1777                 put_link(nd);
1778         err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1779         if (unlikely(err))
1780                 return err;
1781         path_to_nameidata(&path, nd);
1782         nd->inode = inode;
1783         nd->seq = seq;
1784         return 0;
1785 }
1786 
1787 /*
1788  * We can do the critical dentry name comparison and hashing
1789  * operations one word at a time, but we are limited to:
1790  *
1791  * - Architectures with fast unaligned word accesses. We could
1792  *   do a "get_unaligned()" if this helps and is sufficiently
1793  *   fast.
1794  *
1795  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1796  *   do not trap on the (extremely unlikely) case of a page
1797  *   crossing operation.
1798  *
1799  * - Furthermore, we need an efficient 64-bit compile for the
1800  *   64-bit case in order to generate the "number of bytes in
1801  *   the final mask". Again, that could be replaced with a
1802  *   efficient population count instruction or similar.
1803  */
1804 #ifdef CONFIG_DCACHE_WORD_ACCESS
1805 
1806 #include <asm/word-at-a-time.h>
1807 
1808 #ifdef HASH_MIX
1809 
1810 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1811 
1812 #elif defined(CONFIG_64BIT)
1813 /*
1814  * Register pressure in the mixing function is an issue, particularly
1815  * on 32-bit x86, but almost any function requires one state value and
1816  * one temporary.  Instead, use a function designed for two state values
1817  * and no temporaries.
1818  *
1819  * This function cannot create a collision in only two iterations, so
1820  * we have two iterations to achieve avalanche.  In those two iterations,
1821  * we have six layers of mixing, which is enough to spread one bit's
1822  * influence out to 2^6 = 64 state bits.
1823  *
1824  * Rotate constants are scored by considering either 64 one-bit input
1825  * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1826  * probability of that delta causing a change to each of the 128 output
1827  * bits, using a sample of random initial states.
1828  *
1829  * The Shannon entropy of the computed probabilities is then summed
1830  * to produce a score.  Ideally, any input change has a 50% chance of
1831  * toggling any given output bit.
1832  *
1833  * Mixing scores (in bits) for (12,45):
1834  * Input delta: 1-bit      2-bit
1835  * 1 round:     713.3    42542.6
1836  * 2 rounds:   2753.7   140389.8
1837  * 3 rounds:   5954.1   233458.2
1838  * 4 rounds:   7862.6   256672.2
1839  * Perfect:    8192     258048
1840  *            (64*128) (64*63/2 * 128)
1841  */
1842 #define HASH_MIX(x, y, a)       \
1843         (       x ^= (a),       \
1844         y ^= x, x = rol64(x,12),\
1845         x += y, y = rol64(y,45),\
1846         y *= 9                  )
1847 
1848 /*
1849  * Fold two longs into one 32-bit hash value.  This must be fast, but
1850  * latency isn't quite as critical, as there is a fair bit of additional
1851  * work done before the hash value is used.
1852  */
1853 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1854 {
1855         y ^= x * GOLDEN_RATIO_64;
1856         y *= GOLDEN_RATIO_64;
1857         return y >> 32;
1858 }
1859 
1860 #else   /* 32-bit case */
1861 
1862 /*
1863  * Mixing scores (in bits) for (7,20):
1864  * Input delta: 1-bit      2-bit
1865  * 1 round:     330.3     9201.6
1866  * 2 rounds:   1246.4    25475.4
1867  * 3 rounds:   1907.1    31295.1
1868  * 4 rounds:   2042.3    31718.6
1869  * Perfect:    2048      31744
1870  *            (32*64)   (32*31/2 * 64)
1871  */
1872 #define HASH_MIX(x, y, a)       \
1873         (       x ^= (a),       \
1874         y ^= x, x = rol32(x, 7),\
1875         x += y, y = rol32(y,20),\
1876         y *= 9                  )
1877 
1878 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1879 {
1880         /* Use arch-optimized multiply if one exists */
1881         return __hash_32(y ^ __hash_32(x));
1882 }
1883 
1884 #endif
1885 
1886 /*
1887  * Return the hash of a string of known length.  This is carfully
1888  * designed to match hash_name(), which is the more critical function.
1889  * In particular, we must end by hashing a final word containing 0..7
1890  * payload bytes, to match the way that hash_name() iterates until it
1891  * finds the delimiter after the name.
1892  */
1893 unsigned int full_name_hash(const char *name, unsigned int len)
1894 {
1895         unsigned long a, x = 0, y = 0;
1896 
1897         for (;;) {
1898                 if (!len)
1899                         goto done;
1900                 a = load_unaligned_zeropad(name);
1901                 if (len < sizeof(unsigned long))
1902                         break;
1903                 HASH_MIX(x, y, a);
1904                 name += sizeof(unsigned long);
1905                 len -= sizeof(unsigned long);
1906         }
1907         x ^= a & bytemask_from_count(len);
1908 done:
1909         return fold_hash(x, y);
1910 }
1911 EXPORT_SYMBOL(full_name_hash);
1912 
1913 /* Return the "hash_len" (hash and length) of a null-terminated string */
1914 u64 hashlen_string(const char *name)
1915 {
1916         unsigned long a = 0, x = 0, y = 0, adata, mask, len;
1917         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1918 
1919         len = -sizeof(unsigned long);
1920         do {
1921                 HASH_MIX(x, y, a);
1922                 len += sizeof(unsigned long);
1923                 a = load_unaligned_zeropad(name+len);
1924         } while (!has_zero(a, &adata, &constants));
1925 
1926         adata = prep_zero_mask(a, adata, &constants);
1927         mask = create_zero_mask(adata);
1928         x ^= a & zero_bytemask(mask);
1929 
1930         return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1931 }
1932 EXPORT_SYMBOL(hashlen_string);
1933 
1934 /*
1935  * Calculate the length and hash of the path component, and
1936  * return the "hash_len" as the result.
1937  */
1938 static inline u64 hash_name(const char *name)
1939 {
1940         unsigned long a = 0, b, x = 0, y = 0, adata, bdata, mask, len;
1941         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1942 
1943         len = -sizeof(unsigned long);
1944         do {
1945                 HASH_MIX(x, y, a);
1946                 len += sizeof(unsigned long);
1947                 a = load_unaligned_zeropad(name+len);
1948                 b = a ^ REPEAT_BYTE('/');
1949         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1950 
1951         adata = prep_zero_mask(a, adata, &constants);
1952         bdata = prep_zero_mask(b, bdata, &constants);
1953         mask = create_zero_mask(adata | bdata);
1954         x ^= a & zero_bytemask(mask);
1955 
1956         return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1957 }
1958 
1959 #else   /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1960 
1961 /* Return the hash of a string of known length */
1962 unsigned int full_name_hash(const char *name, unsigned int len)
1963 {
1964         unsigned long hash = init_name_hash();
1965         while (len--)
1966                 hash = partial_name_hash((unsigned char)*name++, hash);
1967         return end_name_hash(hash);
1968 }
1969 EXPORT_SYMBOL(full_name_hash);
1970 
1971 /* Return the "hash_len" (hash and length) of a null-terminated string */
1972 u64 hashlen_string(const char *name)
1973 {
1974         unsigned long hash = init_name_hash();
1975         unsigned long len = 0, c;
1976 
1977         c = (unsigned char)*name;
1978         while (c) {
1979                 len++;
1980                 hash = partial_name_hash(c, hash);
1981                 c = (unsigned char)name[len];
1982         }
1983         return hashlen_create(end_name_hash(hash), len);
1984 }
1985 EXPORT_SYMBOL(hashlen_string);
1986 
1987 /*
1988  * We know there's a real path component here of at least
1989  * one character.
1990  */
1991 static inline u64 hash_name(const char *name)
1992 {
1993         unsigned long hash = init_name_hash();
1994         unsigned long len = 0, c;
1995 
1996         c = (unsigned char)*name;
1997         do {
1998                 len++;
1999                 hash = partial_name_hash(c, hash);
2000                 c = (unsigned char)name[len];
2001         } while (c && c != '/');
2002         return hashlen_create(end_name_hash(hash), len);
2003 }
2004 
2005 #endif
2006 
2007 /*
2008  * Name resolution.
2009  * This is the basic name resolution function, turning a pathname into
2010  * the final dentry. We expect 'base' to be positive and a directory.
2011  *
2012  * Returns 0 and nd will have valid dentry and mnt on success.
2013  * Returns error and drops reference to input namei data on failure.
2014  */
2015 static int link_path_walk(const char *name, struct nameidata *nd)
2016 {
2017         int err;
2018 
2019         while (*name=='/')
2020                 name++;
2021         if (!*name)
2022                 return 0;
2023 
2024         /* At this point we know we have a real path component. */
2025         for(;;) {
2026                 u64 hash_len;
2027                 int type;
2028 
2029                 err = may_lookup(nd);
2030                 if (err)
2031                         return err;
2032 
2033                 hash_len = hash_name(name);
2034 
2035                 type = LAST_NORM;
2036                 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2037                         case 2:
2038                                 if (name[1] == '.') {
2039                                         type = LAST_DOTDOT;
2040                                         nd->flags |= LOOKUP_JUMPED;
2041                                 }
2042                                 break;
2043                         case 1:
2044                                 type = LAST_DOT;
2045                 }
2046                 if (likely(type == LAST_NORM)) {
2047                         struct dentry *parent = nd->path.dentry;
2048                         nd->flags &= ~LOOKUP_JUMPED;
2049                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2050                                 struct qstr this = { { .hash_len = hash_len }, .name = name };
2051                                 err = parent->d_op->d_hash(parent, &this);
2052                                 if (err < 0)
2053                                         return err;
2054                                 hash_len = this.hash_len;
2055                                 name = this.name;
2056                         }
2057                 }
2058 
2059                 nd->last.hash_len = hash_len;
2060                 nd->last.name = name;
2061                 nd->last_type = type;
2062 
2063                 name += hashlen_len(hash_len);
2064                 if (!*name)
2065                         goto OK;
2066                 /*
2067                  * If it wasn't NUL, we know it was '/'. Skip that
2068                  * slash, and continue until no more slashes.
2069                  */
2070                 do {
2071                         name++;
2072                 } while (unlikely(*name == '/'));
2073                 if (unlikely(!*name)) {
2074 OK:
2075                         /* pathname body, done */
2076                         if (!nd->depth)
2077                                 return 0;
2078                         name = nd->stack[nd->depth - 1].name;
2079                         /* trailing symlink, done */
2080                         if (!name)
2081                                 return 0;
2082                         /* last component of nested symlink */
2083                         err = walk_component(nd, WALK_GET | WALK_PUT);
2084                 } else {
2085                         err = walk_component(nd, WALK_GET);
2086                 }
2087                 if (err < 0)
2088                         return err;
2089 
2090                 if (err) {
2091                         const char *s = get_link(nd);
2092 
2093                         if (IS_ERR(s))
2094                                 return PTR_ERR(s);
2095                         err = 0;
2096                         if (unlikely(!s)) {
2097                                 /* jumped */
2098                                 put_link(nd);
2099                         } else {
2100                                 nd->stack[nd->depth - 1].name = name;
2101                                 name = s;
2102                                 continue;
2103                         }
2104                 }
2105                 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2106                         if (nd->flags & LOOKUP_RCU) {
2107                                 if (unlazy_walk(nd, NULL, 0))
2108                                         return -ECHILD;
2109                         }
2110                         return -ENOTDIR;
2111                 }
2112         }
2113 }
2114 
2115 static const char *path_init(struct nameidata *nd, unsigned flags)
2116 {
2117         int retval = 0;
2118         const char *s = nd->name->name;
2119 
2120         nd->last_type = LAST_ROOT; /* if there are only slashes... */
2121         nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2122         nd->depth = 0;
2123         if (flags & LOOKUP_ROOT) {
2124                 struct dentry *root = nd->root.dentry;
2125                 struct inode *inode = root->d_inode;
2126                 if (*s) {
2127                         if (!d_can_lookup(root))
2128                                 return ERR_PTR(-ENOTDIR);
2129                         retval = inode_permission(inode, MAY_EXEC);
2130                         if (retval)
2131                                 return ERR_PTR(retval);
2132                 }
2133                 nd->path = nd->root;
2134                 nd->inode = inode;
2135                 if (flags & LOOKUP_RCU) {
2136                         rcu_read_lock();
2137                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2138                         nd->root_seq = nd->seq;
2139                         nd->m_seq = read_seqbegin(&mount_lock);
2140                 } else {
2141                         path_get(&nd->path);
2142                 }
2143                 return s;
2144         }
2145 
2146         nd->root.mnt = NULL;
2147         nd->path.mnt = NULL;
2148         nd->path.dentry = NULL;
2149 
2150         nd->m_seq = read_seqbegin(&mount_lock);
2151         if (*s == '/') {
2152                 if (flags & LOOKUP_RCU)
2153                         rcu_read_lock();
2154                 set_root(nd);
2155                 if (likely(!nd_jump_root(nd)))
2156                         return s;
2157                 nd->root.mnt = NULL;
2158                 rcu_read_unlock();
2159                 return ERR_PTR(-ECHILD);
2160         } else if (nd->dfd == AT_FDCWD) {
2161                 if (flags & LOOKUP_RCU) {
2162                         struct fs_struct *fs = current->fs;
2163                         unsigned seq;
2164 
2165                         rcu_read_lock();
2166 
2167                         do {
2168                                 seq = read_seqcount_begin(&fs->seq);
2169                                 nd->path = fs->pwd;
2170                                 nd->inode = nd->path.dentry->d_inode;
2171                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2172                         } while (read_seqcount_retry(&fs->seq, seq));
2173                 } else {
2174                         get_fs_pwd(current->fs, &nd->path);
2175                         nd->inode = nd->path.dentry->d_inode;
2176                 }
2177                 return s;
2178         } else {
2179                 /* Caller must check execute permissions on the starting path component */
2180                 struct fd f = fdget_raw(nd->dfd);
2181                 struct dentry *dentry;
2182 
2183                 if (!f.file)
2184                         return ERR_PTR(-EBADF);
2185 
2186                 dentry = f.file->f_path.dentry;
2187 
2188                 if (*s) {
2189                         if (!d_can_lookup(dentry)) {
2190                                 fdput(f);
2191                                 return ERR_PTR(-ENOTDIR);
2192                         }
2193                 }
2194 
2195                 nd->path = f.file->f_path;
2196                 if (flags & LOOKUP_RCU) {
2197                         rcu_read_lock();
2198                         nd->inode = nd->path.dentry->d_inode;
2199                         nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2200                 } else {
2201                         path_get(&nd->path);
2202                         nd->inode = nd->path.dentry->d_inode;
2203                 }
2204                 fdput(f);
2205                 return s;
2206         }
2207 }
2208 
2209 static const char *trailing_symlink(struct nameidata *nd)
2210 {
2211         const char *s;
2212         int error = may_follow_link(nd);
2213         if (unlikely(error))
2214                 return ERR_PTR(error);
2215         nd->flags |= LOOKUP_PARENT;
2216         nd->stack[0].name = NULL;
2217         s = get_link(nd);
2218         return s ? s : "";
2219 }
2220 
2221 static inline int lookup_last(struct nameidata *nd)
2222 {
2223         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2224                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2225 
2226         nd->flags &= ~LOOKUP_PARENT;
2227         return walk_component(nd,
2228                         nd->flags & LOOKUP_FOLLOW
2229                                 ? nd->depth
2230                                         ? WALK_PUT | WALK_GET
2231                                         : WALK_GET
2232                                 : 0);
2233 }
2234 
2235 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2236 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2237 {
2238         const char *s = path_init(nd, flags);
2239         int err;
2240 
2241         if (IS_ERR(s))
2242                 return PTR_ERR(s);
2243         while (!(err = link_path_walk(s, nd))
2244                 && ((err = lookup_last(nd)) > 0)) {
2245                 s = trailing_symlink(nd);
2246                 if (IS_ERR(s)) {
2247                         err = PTR_ERR(s);
2248                         break;
2249                 }
2250         }
2251         if (!err)
2252                 err = complete_walk(nd);
2253 
2254         if (!err && nd->flags & LOOKUP_DIRECTORY)
2255                 if (!d_can_lookup(nd->path.dentry))
2256                         err = -ENOTDIR;
2257         if (!err) {
2258                 *path = nd->path;
2259                 nd->path.mnt = NULL;
2260                 nd->path.dentry = NULL;
2261         }
2262         terminate_walk(nd);
2263         return err;
2264 }
2265 
2266 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2267                            struct path *path, struct path *root)
2268 {
2269         int retval;
2270         struct nameidata nd;
2271         if (IS_ERR(name))
2272                 return PTR_ERR(name);
2273         if (unlikely(root)) {
2274                 nd.root = *root;
2275                 flags |= LOOKUP_ROOT;
2276         }
2277         set_nameidata(&nd, dfd, name);
2278         retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2279         if (unlikely(retval == -ECHILD))
2280                 retval = path_lookupat(&nd, flags, path);
2281         if (unlikely(retval == -ESTALE))
2282                 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2283 
2284         if (likely(!retval))
2285                 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2286         restore_nameidata();
2287         putname(name);
2288         return retval;
2289 }
2290 
2291 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2292 static int path_parentat(struct nameidata *nd, unsigned flags,
2293                                 struct path *parent)
2294 {
2295         const char *s = path_init(nd, flags);
2296         int err;
2297         if (IS_ERR(s))
2298                 return PTR_ERR(s);
2299         err = link_path_walk(s, nd);
2300         if (!err)
2301                 err = complete_walk(nd);
2302         if (!err) {
2303                 *parent = nd->path;
2304                 nd->path.mnt = NULL;
2305                 nd->path.dentry = NULL;
2306         }
2307         terminate_walk(nd);
2308         return err;
2309 }
2310 
2311 static struct filename *filename_parentat(int dfd, struct filename *name,
2312                                 unsigned int flags, struct path *parent,
2313                                 struct qstr *last, int *type)
2314 {
2315         int retval;
2316         struct nameidata nd;
2317 
2318         if (IS_ERR(name))
2319                 return name;
2320         set_nameidata(&nd, dfd, name);
2321         retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2322         if (unlikely(retval == -ECHILD))
2323                 retval = path_parentat(&nd, flags, parent);
2324         if (unlikely(retval == -ESTALE))
2325                 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2326         if (likely(!retval)) {
2327                 *last = nd.last;
2328                 *type = nd.last_type;
2329                 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2330         } else {
2331                 putname(name);
2332                 name = ERR_PTR(retval);
2333         }
2334         restore_nameidata();
2335         return name;
2336 }
2337 
2338 /* does lookup, returns the object with parent locked */
2339 struct dentry *kern_path_locked(const char *name, struct path *path)
2340 {
2341         struct filename *filename;
2342         struct dentry *d;
2343         struct qstr last;
2344         int type;
2345 
2346         filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2347                                     &last, &type);
2348         if (IS_ERR(filename))
2349                 return ERR_CAST(filename);
2350         if (unlikely(type != LAST_NORM)) {
2351                 path_put(path);
2352                 putname(filename);
2353                 return ERR_PTR(-EINVAL);
2354         }
2355         inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2356         d = __lookup_hash(&last, path->dentry, 0);
2357         if (IS_ERR(d)) {
2358                 inode_unlock(path->dentry->d_inode);
2359                 path_put(path);
2360         }
2361         putname(filename);
2362         return d;
2363 }
2364 
2365 int kern_path(const char *name, unsigned int flags, struct path *path)
2366 {
2367         return filename_lookup(AT_FDCWD, getname_kernel(name),
2368                                flags, path, NULL);
2369 }
2370 EXPORT_SYMBOL(kern_path);
2371 
2372 /**
2373  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2374  * @dentry:  pointer to dentry of the base directory
2375  * @mnt: pointer to vfs mount of the base directory
2376  * @name: pointer to file name
2377  * @flags: lookup flags
2378  * @path: pointer to struct path to fill
2379  */
2380 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2381                     const char *name, unsigned int flags,
2382                     struct path *path)
2383 {
2384         struct path root = {.mnt = mnt, .dentry = dentry};
2385         /* the first argument of filename_lookup() is ignored with root */
2386         return filename_lookup(AT_FDCWD, getname_kernel(name),
2387                                flags , path, &root);
2388 }
2389 EXPORT_SYMBOL(vfs_path_lookup);
2390 
2391 /**
2392  * lookup_hash - lookup single pathname component on already hashed name
2393  * @name:       name and hash to lookup
2394  * @base:       base directory to lookup from
2395  *
2396  * The name must have been verified and hashed (see lookup_one_len()).  Using
2397  * this after just full_name_hash() is unsafe.
2398  *
2399  * This function also doesn't check for search permission on base directory.
2400  *
2401  * Use lookup_one_len_unlocked() instead, unless you really know what you are
2402  * doing.
2403  *
2404  * Do not hold i_mutex; this helper takes i_mutex if necessary.
2405  */
2406 struct dentry *lookup_hash(const struct qstr *name, struct dentry *base)
2407 {
2408         struct dentry *ret;
2409 
2410         ret = lookup_dcache(name, base, 0);
2411         if (!ret)
2412                 ret = lookup_slow(name, base, 0);
2413 
2414         return ret;
2415 }
2416 EXPORT_SYMBOL(lookup_hash);
2417 
2418 /**
2419  * lookup_one_len - filesystem helper to lookup single pathname component
2420  * @name:       pathname component to lookup
2421  * @base:       base directory to lookup from
2422  * @len:        maximum length @len should be interpreted to
2423  *
2424  * Note that this routine is purely a helper for filesystem usage and should
2425  * not be called by generic code.
2426  *
2427  * The caller must hold base->i_mutex.
2428  */
2429 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2430 {
2431         struct qstr this;
2432         unsigned int c;
2433         int err;
2434 
2435         WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2436 
2437         this.name = name;
2438         this.len = len;
2439         this.hash = full_name_hash(name, len);
2440         if (!len)
2441                 return ERR_PTR(-EACCES);
2442 
2443         if (unlikely(name[0] == '.')) {
2444                 if (len < 2 || (len == 2 && name[1] == '.'))
2445                         return ERR_PTR(-EACCES);
2446         }
2447 
2448         while (len--) {
2449                 c = *(const unsigned char *)name++;
2450                 if (c == '/' || c == '\0')
2451                         return ERR_PTR(-EACCES);
2452         }
2453         /*
2454          * See if the low-level filesystem might want
2455          * to use its own hash..
2456          */
2457         if (base->d_flags & DCACHE_OP_HASH) {
2458                 int err = base->d_op->d_hash(base, &this);
2459                 if (err < 0)
2460                         return ERR_PTR(err);
2461         }
2462 
2463         err = inode_permission(base->d_inode, MAY_EXEC);
2464         if (err)
2465                 return ERR_PTR(err);
2466 
2467         return __lookup_hash(&this, base, 0);
2468 }
2469 EXPORT_SYMBOL(lookup_one_len);
2470 
2471 /**
2472  * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2473  * @name:       pathname component to lookup
2474  * @base:       base directory to lookup from
2475  * @len:        maximum length @len should be interpreted to
2476  *
2477  * Note that this routine is purely a helper for filesystem usage and should
2478  * not be called by generic code.
2479  *
2480  * Unlike lookup_one_len, it should be called without the parent
2481  * i_mutex held, and will take the i_mutex itself if necessary.
2482  */
2483 struct dentry *lookup_one_len_unlocked(const char *name,
2484                                        struct dentry *base, int len)
2485 {
2486         struct qstr this;
2487         unsigned int c;
2488         int err;
2489 
2490         this.name = name;
2491         this.len = len;
2492         this.hash = full_name_hash(name, len);
2493         if (!len)
2494                 return ERR_PTR(-EACCES);
2495 
2496         if (unlikely(name[0] == '.')) {
2497                 if (len < 2 || (len == 2 && name[1] == '.'))
2498                         return ERR_PTR(-EACCES);
2499         }
2500 
2501         while (len--) {
2502                 c = *(const unsigned char *)name++;
2503                 if (c == '/' || c == '\0')
2504                         return ERR_PTR(-EACCES);
2505         }
2506         /*
2507          * See if the low-level filesystem might want
2508          * to use its own hash..
2509          */
2510         if (base->d_flags & DCACHE_OP_HASH) {
2511                 int err = base->d_op->d_hash(base, &this);
2512                 if (err < 0)
2513                         return ERR_PTR(err);
2514         }
2515 
2516         err = inode_permission(base->d_inode, MAY_EXEC);
2517         if (err)
2518                 return ERR_PTR(err);
2519 
2520         return lookup_hash(&this, base);
2521 }
2522 EXPORT_SYMBOL(lookup_one_len_unlocked);
2523 
2524 #ifdef CONFIG_UNIX98_PTYS
2525 int path_pts(struct path *path)
2526 {
2527         /* Find something mounted on "pts" in the same directory as
2528          * the input path.
2529          */
2530         struct dentry *child, *parent;
2531         struct qstr this;
2532         int ret;
2533 
2534         ret = path_parent_directory(path);
2535         if (ret)
2536                 return ret;
2537 
2538         parent = path->dentry;
2539         this.name = "pts";
2540         this.len = 3;
2541         child = d_hash_and_lookup(parent, &this);
2542         if (!child)
2543                 return -ENOENT;
2544 
2545         path->dentry = child;
2546         dput(parent);
2547         follow_mount(path);
2548         return 0;
2549 }
2550 #endif
2551 
2552 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2553                  struct path *path, int *empty)
2554 {
2555         return filename_lookup(dfd, getname_flags(name, flags, empty),
2556                                flags, path, NULL);
2557 }
2558 EXPORT_SYMBOL(user_path_at_empty);
2559 
2560 /*
2561  * NB: most callers don't do anything directly with the reference to the
2562  *     to struct filename, but the nd->last pointer points into the name string
2563  *     allocated by getname. So we must hold the reference to it until all
2564  *     path-walking is complete.
2565  */
2566 static inline struct filename *
2567 user_path_parent(int dfd, const char __user *path,
2568                  struct path *parent,
2569                  struct qstr *last,
2570                  int *type,
2571                  unsigned int flags)
2572 {
2573         /* only LOOKUP_REVAL is allowed in extra flags */
2574         return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2575                                  parent, last, type);
2576 }
2577 
2578 /**
2579  * mountpoint_last - look up last component for umount
2580  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2581  * @path: pointer to container for result
2582  *
2583  * This is a special lookup_last function just for umount. In this case, we
2584  * need to resolve the path without doing any revalidation.
2585  *
2586  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2587  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2588  * in almost all cases, this lookup will be served out of the dcache. The only
2589  * cases where it won't are if nd->last refers to a symlink or the path is
2590  * bogus and it doesn't exist.
2591  *
2592  * Returns:
2593  * -error: if there was an error during lookup. This includes -ENOENT if the
2594  *         lookup found a negative dentry. The nd->path reference will also be
2595  *         put in this case.
2596  *
2597  * 0:      if we successfully resolved nd->path and found it to not to be a
2598  *         symlink that needs to be followed. "path" will also be populated.
2599  *         The nd->path reference will also be put.
2600  *
2601  * 1:      if we successfully resolved nd->last and found it to be a symlink
2602  *         that needs to be followed. "path" will be populated with the path
2603  *         to the link, and nd->path will *not* be put.
2604  */
2605 static int
2606 mountpoint_last(struct nameidata *nd, struct path *path)
2607 {
2608         int error = 0;
2609         struct dentry *dentry;
2610         struct dentry *dir = nd->path.dentry;
2611 
2612         /* If we're in rcuwalk, drop out of it to handle last component */
2613         if (nd->flags & LOOKUP_RCU) {
2614                 if (unlazy_walk(nd, NULL, 0))
2615                         return -ECHILD;
2616         }
2617 
2618         nd->flags &= ~LOOKUP_PARENT;
2619 
2620         if (unlikely(nd->last_type != LAST_NORM)) {
2621                 error = handle_dots(nd, nd->last_type);
2622                 if (error)
2623                         return error;
2624                 dentry = dget(nd->path.dentry);
2625         } else {
2626                 dentry = d_lookup(dir, &nd->last);
2627                 if (!dentry) {
2628                         /*
2629                          * No cached dentry. Mounted dentries are pinned in the
2630                          * cache, so that means that this dentry is probably
2631                          * a symlink or the path doesn't actually point
2632                          * to a mounted dentry.
2633                          */
2634                         dentry = lookup_slow(&nd->last, dir,
2635                                              nd->flags | LOOKUP_NO_REVAL);
2636                         if (IS_ERR(dentry))
2637                                 return PTR_ERR(dentry);
2638                 }
2639         }
2640         if (d_is_negative(dentry)) {
2641                 dput(dentry);
2642                 return -ENOENT;
2643         }
2644         if (nd->depth)
2645                 put_link(nd);
2646         path->dentry = dentry;
2647         path->mnt = nd->path.mnt;
2648         error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2649                                    d_backing_inode(dentry), 0);
2650         if (unlikely(error))
2651                 return error;
2652         mntget(path->mnt);
2653         follow_mount(path);
2654         return 0;
2655 }
2656 
2657 /**
2658  * path_mountpoint - look up a path to be umounted
2659  * @nd:         lookup context
2660  * @flags:      lookup flags
2661  * @path:       pointer to container for result
2662  *
2663  * Look up the given name, but don't attempt to revalidate the last component.
2664  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2665  */
2666 static int
2667 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2668 {
2669         const char *s = path_init(nd, flags);
2670         int err;
2671         if (IS_ERR(s))
2672                 return PTR_ERR(s);
2673         while (!(err = link_path_walk(s, nd)) &&
2674                 (err = mountpoint_last(nd, path)) > 0) {
2675                 s = trailing_symlink(nd);
2676                 if (IS_ERR(s)) {
2677                         err = PTR_ERR(s);
2678                         break;
2679                 }
2680         }
2681         terminate_walk(nd);
2682         return err;
2683 }
2684 
2685 static int
2686 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2687                         unsigned int flags)
2688 {
2689         struct nameidata nd;
2690         int error;
2691         if (IS_ERR(name))
2692                 return PTR_ERR(name);
2693         set_nameidata(&nd, dfd, name);
2694         error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2695         if (unlikely(error == -ECHILD))
2696                 error = path_mountpoint(&nd, flags, path);
2697         if (unlikely(error == -ESTALE))
2698                 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2699         if (likely(!error))
2700                 audit_inode(name, path->dentry, 0);
2701         restore_nameidata();
2702         putname(name);
2703         return error;
2704 }
2705 
2706 /**
2707  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2708  * @dfd:        directory file descriptor
2709  * @name:       pathname from userland
2710  * @flags:      lookup flags
2711  * @path:       pointer to container to hold result
2712  *
2713  * A umount is a special case for path walking. We're not actually interested
2714  * in the inode in this situation, and ESTALE errors can be a problem. We
2715  * simply want track down the dentry and vfsmount attached at the mountpoint
2716  * and avoid revalidating the last component.
2717  *
2718  * Returns 0 and populates "path" on success.
2719  */
2720 int
2721 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2722                         struct path *path)
2723 {
2724         return filename_mountpoint(dfd, getname(name), path, flags);
2725 }
2726 
2727 int
2728 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2729                         unsigned int flags)
2730 {
2731         return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2732 }
2733 EXPORT_SYMBOL(kern_path_mountpoint);
2734 
2735 int __check_sticky(struct inode *dir, struct inode *inode)
2736 {
2737         kuid_t fsuid = current_fsuid();
2738 
2739         if (uid_eq(inode->i_uid, fsuid))
2740                 return 0;
2741         if (uid_eq(dir->i_uid, fsuid))
2742                 return 0;
2743         return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2744 }
2745 EXPORT_SYMBOL(__check_sticky);
2746 
2747 /*
2748  *      Check whether we can remove a link victim from directory dir, check
2749  *  whether the type of victim is right.
2750  *  1. We can't do it if dir is read-only (done in permission())
2751  *  2. We should have write and exec permissions on dir
2752  *  3. We can't remove anything from append-only dir
2753  *  4. We can't do anything with immutable dir (done in permission())
2754  *  5. If the sticky bit on dir is set we should either
2755  *      a. be owner of dir, or
2756  *      b. be owner of victim, or
2757  *      c. have CAP_FOWNER capability
2758  *  6. If the victim is append-only or immutable we can't do antyhing with
2759  *     links pointing to it.
2760  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2761  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2762  *  9. We can't remove a root or mountpoint.
2763  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2764  *     nfs_async_unlink().
2765  */
2766 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2767 {
2768         struct inode *inode = d_backing_inode(victim);
2769         int error;
2770 
2771         if (d_is_negative(victim))
2772                 return -ENOENT;
2773         BUG_ON(!inode);
2774 
2775         BUG_ON(victim->d_parent->d_inode != dir);
2776         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2777 
2778         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2779         if (error)
2780                 return error;
2781         if (IS_APPEND(dir))
2782                 return -EPERM;
2783 
2784         if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2785             IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2786                 return -EPERM;
2787         if (isdir) {
2788                 if (!d_is_dir(victim))
2789                         return -ENOTDIR;
2790                 if (IS_ROOT(victim))
2791                         return -EBUSY;
2792         } else if (d_is_dir(victim))
2793                 return -EISDIR;
2794         if (IS_DEADDIR(dir))
2795                 return -ENOENT;
2796         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2797                 return -EBUSY;
2798         return 0;
2799 }
2800 
2801 /*      Check whether we can create an object with dentry child in directory
2802  *  dir.
2803  *  1. We can't do it if child already exists (open has special treatment for
2804  *     this case, but since we are inlined it's OK)
2805  *  2. We can't do it if dir is read-only (done in permission())
2806  *  3. We should have write and exec permissions on dir
2807  *  4. We can't do it if dir is immutable (done in permission())
2808  */
2809 static inline int may_create(struct inode *dir, struct dentry *child)
2810 {
2811         audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2812         if (child->d_inode)
2813                 return -EEXIST;
2814         if (IS_DEADDIR(dir))
2815                 return -ENOENT;
2816         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2817 }
2818 
2819 /*
2820  * p1 and p2 should be directories on the same fs.
2821  */
2822 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2823 {
2824         struct dentry *p;
2825 
2826         if (p1 == p2) {
2827                 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2828                 return NULL;
2829         }
2830 
2831         mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2832 
2833         p = d_ancestor(p2, p1);
2834         if (p) {
2835                 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2836                 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2837                 return p;
2838         }
2839 
2840         p = d_ancestor(p1, p2);
2841         if (p) {
2842                 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2843                 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2844                 return p;
2845         }
2846 
2847         inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2848         inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2849         return NULL;
2850 }
2851 EXPORT_SYMBOL(lock_rename);
2852 
2853 void unlock_rename(struct dentry *p1, struct dentry *p2)
2854 {
2855         inode_unlock(p1->d_inode);
2856         if (p1 != p2) {
2857                 inode_unlock(p2->d_inode);
2858                 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2859         }
2860 }
2861 EXPORT_SYMBOL(unlock_rename);
2862 
2863 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2864                 bool want_excl)
2865 {
2866         int error = may_create(dir, dentry);
2867         if (error)
2868                 return error;
2869 
2870         if (!dir->i_op->create)
2871                 return -EACCES; /* shouldn't it be ENOSYS? */
2872         mode &= S_IALLUGO;
2873         mode |= S_IFREG;
2874         error = security_inode_create(dir, dentry, mode);
2875         if (error)
2876                 return error;
2877         error = dir->i_op->create(dir, dentry, mode, want_excl);
2878         if (!error)
2879                 fsnotify_create(dir, dentry);
2880         return error;
2881 }
2882 EXPORT_SYMBOL(vfs_create);
2883 
2884 static int may_open(struct path *path, int acc_mode, int flag)
2885 {
2886         struct dentry *dentry = path->dentry;
2887         struct inode *inode = dentry->d_inode;
2888         int error;
2889 
2890         if (!inode)
2891                 return -ENOENT;
2892 
2893         switch (inode->i_mode & S_IFMT) {
2894         case S_IFLNK:
2895                 return -ELOOP;
2896         case S_IFDIR:
2897                 if (acc_mode & MAY_WRITE)
2898                         return -EISDIR;
2899                 break;
2900         case S_IFBLK:
2901         case S_IFCHR:
2902                 if (path->mnt->mnt_flags & MNT_NODEV)
2903                         return -EACCES;
2904                 /*FALLTHRU*/
2905         case S_IFIFO:
2906         case S_IFSOCK:
2907                 flag &= ~O_TRUNC;
2908                 break;
2909         }
2910 
2911         error = inode_permission(inode, MAY_OPEN | acc_mode);
2912         if (error)
2913                 return error;
2914 
2915         /*
2916          * An append-only file must be opened in append mode for writing.
2917          */
2918         if (IS_APPEND(inode)) {
2919                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2920                         return -EPERM;
2921                 if (flag & O_TRUNC)
2922                         return -EPERM;
2923         }
2924 
2925         /* O_NOATIME can only be set by the owner or superuser */
2926         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2927                 return -EPERM;
2928 
2929         return 0;
2930 }
2931 
2932 static int handle_truncate(struct file *filp)
2933 {
2934         struct path *path = &filp->f_path;
2935         struct inode *inode = path->dentry->d_inode;
2936         int error = get_write_access(inode);
2937         if (error)
2938                 return error;
2939         /*
2940          * Refuse to truncate files with mandatory locks held on them.
2941          */
2942         error = locks_verify_locked(filp);
2943         if (!error)
2944                 error = security_path_truncate(path);
2945         if (!error) {
2946                 error = do_truncate(path->dentry, 0,
2947                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2948                                     filp);
2949         }
2950         put_write_access(inode);
2951         return error;
2952 }
2953 
2954 static inline int open_to_namei_flags(int flag)
2955 {
2956         if ((flag & O_ACCMODE) == 3)
2957                 flag--;
2958         return flag;
2959 }
2960 
2961 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2962 {
2963         int error = security_path_mknod(dir, dentry, mode, 0);
2964         if (error)
2965                 return error;
2966 
2967         error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2968         if (error)
2969                 return error;
2970 
2971         return security_inode_create(dir->dentry->d_inode, dentry, mode);
2972 }
2973 
2974 /*
2975  * Attempt to atomically look up, create and open a file from a negative
2976  * dentry.
2977  *
2978  * Returns 0 if successful.  The file will have been created and attached to
2979  * @file by the filesystem calling finish_open().
2980  *
2981  * Returns 1 if the file was looked up only or didn't need creating.  The
2982  * caller will need to perform the open themselves.  @path will have been
2983  * updated to point to the new dentry.  This may be negative.
2984  *
2985  * Returns an error code otherwise.
2986  */
2987 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2988                         struct path *path, struct file *file,
2989                         const struct open_flags *op,
2990                         int open_flag, umode_t mode,
2991                         int *opened)
2992 {
2993         struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2994         struct inode *dir =  nd->path.dentry->d_inode;
2995         int error;
2996 
2997         if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
2998                 open_flag &= ~O_TRUNC;
2999 
3000         if (nd->flags & LOOKUP_DIRECTORY)
3001                 open_flag |= O_DIRECTORY;
3002 
3003         file->f_path.dentry = DENTRY_NOT_SET;
3004         file->f_path.mnt = nd->path.mnt;
3005         error = dir->i_op->atomic_open(dir, dentry, file,
3006                                        open_to_namei_flags(open_flag),
3007                                        mode, opened);
3008         d_lookup_done(dentry);
3009         if (!error) {
3010                 /*
3011                  * We didn't have the inode before the open, so check open
3012                  * permission here.
3013                  */
3014                 int acc_mode = op->acc_mode;
3015                 if (*opened & FILE_CREATED) {
3016                         WARN_ON(!(open_flag & O_CREAT));
3017                         fsnotify_create(dir, dentry);
3018                         acc_mode = 0;
3019                 }
3020                 error = may_open(&file->f_path, acc_mode, open_flag);
3021                 if (WARN_ON(error > 0))
3022                         error = -EINVAL;
3023         } else if (error > 0) {
3024                 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3025                         error = -EIO;
3026                 } else {
3027                         if (file->f_path.dentry) {
3028                                 dput(dentry);
3029                                 dentry = file->f_path.dentry;
3030                         }
3031                         if (*opened & FILE_CREATED)
3032                                 fsnotify_create(dir, dentry);
3033                         if (unlikely(d_is_negative(dentry))) {
3034                                 error = -ENOENT;
3035                         } else {
3036                                 path->dentry = dentry;
3037                                 path->mnt = nd->path.mnt;
3038                                 return 1;
3039                         }
3040                 }
3041         }
3042         dput(dentry);
3043         return error;
3044 }
3045 
3046 /*
3047  * Look up and maybe create and open the last component.
3048  *
3049  * Must be called with i_mutex held on parent.
3050  *
3051  * Returns 0 if the file was successfully atomically created (if necessary) and
3052  * opened.  In this case the file will be returned attached to @file.
3053  *
3054  * Returns 1 if the file was not completely opened at this time, though lookups
3055  * and creations will have been performed and the dentry returned in @path will
3056  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
3057  * specified then a negative dentry may be returned.
3058  *
3059  * An error code is returned otherwise.
3060  *
3061  * FILE_CREATE will be set in @*opened if the dentry was created and will be
3062  * cleared otherwise prior to returning.
3063  */
3064 static int lookup_open(struct nameidata *nd, struct path *path,
3065                         struct file *file,
3066                         const struct open_flags *op,
3067                         bool got_write, int *opened)
3068 {
3069         struct dentry *dir = nd->path.dentry;
3070         struct inode *dir_inode = dir->d_inode;
3071         int open_flag = op->open_flag;
3072         struct dentry *dentry;
3073         int error, create_error = 0;
3074         umode_t mode = op->mode;
3075         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3076 
3077         if (unlikely(IS_DEADDIR(dir_inode)))
3078                 return -ENOENT;
3079 
3080         *opened &= ~FILE_CREATED;
3081         dentry = d_lookup(dir, &nd->last);
3082         for (;;) {
3083                 if (!dentry) {
3084                         dentry = d_alloc_parallel(dir, &nd->last, &wq);
3085                         if (IS_ERR(dentry))
3086                                 return PTR_ERR(dentry);
3087                 }
3088                 if (d_in_lookup(dentry))
3089                         break;
3090 
3091                 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3092                         break;
3093 
3094                 error = d_revalidate(dentry, nd->flags);
3095                 if (likely(error > 0))
3096                         break;
3097                 if (error)
3098                         goto out_dput;
3099                 d_invalidate(dentry);
3100                 dput(dentry);
3101                 dentry = NULL;
3102         }
3103         if (dentry->d_inode) {
3104                 /* Cached positive dentry: will open in f_op->open */
3105                 goto out_no_open;
3106         }
3107 
3108         /*
3109          * Checking write permission is tricky, bacuse we don't know if we are
3110          * going to actually need it: O_CREAT opens should work as long as the
3111          * file exists.  But checking existence breaks atomicity.  The trick is
3112          * to check access and if not granted clear O_CREAT from the flags.
3113          *
3114          * Another problem is returing the "right" error value (e.g. for an
3115          * O_EXCL open we want to return EEXIST not EROFS).
3116          */
3117         if (open_flag & O_CREAT) {
3118                 if (!IS_POSIXACL(dir->d_inode))
3119                         mode &= ~current_umask();
3120                 if (unlikely(!got_write)) {
3121                         create_error = -EROFS;
3122                         open_flag &= ~O_CREAT;
3123                         if (open_flag & (O_EXCL | O_TRUNC))
3124                                 goto no_open;
3125                         /* No side effects, safe to clear O_CREAT */
3126                 } else {
3127                         create_error = may_o_create(&nd->path, dentry, mode);
3128                         if (create_error) {
3129                                 open_flag &= ~O_CREAT;
3130                                 if (open_flag & O_EXCL)
3131                                         goto no_open;
3132                         }
3133                 }
3134         } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3135                    unlikely(!got_write)) {
3136                 /*
3137                  * No O_CREATE -> atomicity not a requirement -> fall
3138                  * back to lookup + open
3139                  */
3140                 goto no_open;
3141         }
3142 
3143         if (dir_inode->i_op->atomic_open) {
3144                 error = atomic_open(nd, dentry, path, file, op, open_flag,
3145                                     mode, opened);
3146                 if (unlikely(error == -ENOENT) && create_error)
3147                         error = create_error;
3148                 return error;
3149         }
3150 
3151 no_open:
3152         if (d_in_lookup(dentry)) {
3153                 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3154                                                              nd->flags);
3155                 d_lookup_done(dentry);
3156                 if (unlikely(res)) {
3157                         if (IS_ERR(res)) {
3158                                 error = PTR_ERR(res);
3159                                 goto out_dput;
3160                         }
3161                         dput(dentry);
3162                         dentry = res;
3163                 }
3164         }
3165 
3166         /* Negative dentry, just create the file */
3167         if (!dentry->d_inode && (open_flag & O_CREAT)) {
3168                 *opened |= FILE_CREATED;
3169                 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3170                 if (!dir_inode->i_op->create) {
3171                         error = -EACCES;
3172                         goto out_dput;
3173                 }
3174                 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3175                                                 open_flag & O_EXCL);
3176                 if (error)
3177                         goto out_dput;
3178                 fsnotify_create(dir_inode, dentry);
3179         }
3180         if (unlikely(create_error) && !dentry->d_inode) {
3181                 error = create_error;
3182                 goto out_dput;
3183         }
3184 out_no_open:
3185         path->dentry = dentry;
3186         path->mnt = nd->path.mnt;
3187         return 1;
3188 
3189 out_dput:
3190         dput(dentry);
3191         return error;
3192 }
3193 
3194 /*
3195  * Handle the last step of open()
3196  */
3197 static int do_last(struct nameidata *nd,
3198                    struct file *file, const struct open_flags *op,
3199                    int *opened)
3200 {
3201         struct dentry *dir = nd->path.dentry;
3202         int open_flag = op->open_flag;
3203         bool will_truncate = (open_flag & O_TRUNC) != 0;
3204         bool got_write = false;
3205         int acc_mode = op->acc_mode;
3206         unsigned seq;
3207         struct inode *inode;
3208         struct path path;
3209         int error;
3210 
3211         nd->flags &= ~LOOKUP_PARENT;
3212         nd->flags |= op->intent;
3213 
3214         if (nd->last_type != LAST_NORM) {
3215                 error = handle_dots(nd, nd->last_type);
3216                 if (unlikely(error))
3217                         return error;
3218                 goto finish_open;
3219         }
3220 
3221         if (!(open_flag & O_CREAT)) {
3222                 if (nd->last.name[nd->last.len])
3223                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3224                 /* we _can_ be in RCU mode here */
3225                 error = lookup_fast(nd, &path, &inode, &seq);
3226                 if (likely(error > 0))
3227                         goto finish_lookup;
3228 
3229                 if (error < 0)
3230                         return error;
3231 
3232                 BUG_ON(nd->inode != dir->d_inode);
3233                 BUG_ON(nd->flags & LOOKUP_RCU);
3234         } else {
3235                 /* create side of things */
3236                 /*
3237                  * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3238                  * has been cleared when we got to the last component we are
3239                  * about to look up
3240                  */
3241                 error = complete_walk(nd);
3242                 if (error)
3243                         return error;
3244 
3245                 audit_inode(nd->name, dir, LOOKUP_PARENT);
3246                 /* trailing slashes? */
3247                 if (unlikely(nd->last.name[nd->last.len]))
3248                         return -EISDIR;
3249         }
3250 
3251         if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3252                 error = mnt_want_write(nd->path.mnt);
3253                 if (!error)
3254                         got_write = true;
3255                 /*
3256                  * do _not_ fail yet - we might not need that or fail with
3257                  * a different error; let lookup_open() decide; we'll be
3258                  * dropping this one anyway.
3259                  */
3260         }
3261         if (open_flag & O_CREAT)
3262                 inode_lock(dir->d_inode);
3263         else
3264                 inode_lock_shared(dir->d_inode);
3265         error = lookup_open(nd, &path, file, op, got_write, opened);
3266         if (open_flag & O_CREAT)
3267                 inode_unlock(dir->d_inode);
3268         else
3269                 inode_unlock_shared(dir->d_inode);
3270 
3271         if (error <= 0) {
3272                 if (error)
3273                         goto out;
3274 
3275                 if ((*opened & FILE_CREATED) ||
3276                     !S_ISREG(file_inode(file)->i_mode))
3277                         will_truncate = false;
3278 
3279                 audit_inode(nd->name, file->f_path.dentry, 0);
3280                 goto opened;
3281         }
3282 
3283         if (*opened & FILE_CREATED) {
3284                 /* Don't check for write permission, don't truncate */
3285                 open_flag &= ~O_TRUNC;
3286                 will_truncate = false;
3287                 acc_mode = 0;
3288                 path_to_nameidata(&path, nd);
3289                 goto finish_open_created;
3290         }
3291 
3292         /*
3293          * If atomic_open() acquired write access it is dropped now due to
3294          * possible mount and symlink following (this might be optimized away if
3295          * necessary...)
3296          */
3297         if (got_write) {
3298                 mnt_drop_write(nd->path.mnt);
3299                 got_write = false;
3300         }
3301 
3302         error = follow_managed(&path, nd);
3303         if (unlikely(error < 0))
3304                 return error;
3305 
3306         if (unlikely(d_is_negative(path.dentry))) {
3307                 path_to_nameidata(&path, nd);
3308                 return -ENOENT;
3309         }
3310 
3311         /*
3312          * create/update audit record if it already exists.
3313          */
3314         audit_inode(nd->name, path.dentry, 0);
3315 
3316         if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3317                 path_to_nameidata(&path, nd);
3318                 return -EEXIST;
3319         }
3320 
3321         seq = 0;        /* out of RCU mode, so the value doesn't matter */
3322         inode = d_backing_inode(path.dentry);
3323 finish_lookup:
3324         if (nd->depth)
3325                 put_link(nd);
3326         error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3327                                    inode, seq);
3328         if (unlikely(error))
3329                 return error;
3330 
3331         path_to_nameidata(&path, nd);
3332         nd->inode = inode;
3333         nd->seq = seq;
3334         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3335 finish_open:
3336         error = complete_walk(nd);
3337         if (error)
3338                 return error;
3339         audit_inode(nd->name, nd->path.dentry, 0);
3340         error = -EISDIR;
3341         if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3342                 goto out;
3343         error = -ENOTDIR;
3344         if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3345                 goto out;
3346         if (!d_is_reg(nd->path.dentry))
3347                 will_truncate = false;
3348 
3349         if (will_truncate) {
3350                 error = mnt_want_write(nd->path.mnt);
3351                 if (error)
3352                         goto out;
3353                 got_write = true;
3354         }
3355 finish_open_created:
3356         error = may_open(&nd->path, acc_mode, open_flag);
3357         if (error)
3358                 goto out;
3359         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3360         error = vfs_open(&nd->path, file, current_cred());
3361         if (error)
3362                 goto out;
3363         *opened |= FILE_OPENED;
3364 opened:
3365         error = open_check_o_direct(file);
3366         if (!error)
3367                 error = ima_file_check(file, op->acc_mode, *opened);
3368         if (!error && will_truncate)
3369                 error = handle_truncate(file);
3370 out:
3371         if (unlikely(error) && (*opened & FILE_OPENED))
3372                 fput(file);
3373         if (unlikely(error > 0)) {
3374                 WARN_ON(1);
3375                 error = -EINVAL;
3376         }
3377         if (got_write)
3378                 mnt_drop_write(nd->path.mnt);
3379         return error;
3380 }
3381 
3382 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3383                 const struct open_flags *op,
3384                 struct file *file, int *opened)
3385 {
3386         static const struct qstr name = QSTR_INIT("/", 1);
3387         struct dentry *child;
3388         struct inode *dir;
3389         struct path path;
3390         int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3391         if (unlikely(error))
3392                 return error;
3393         error = mnt_want_write(path.mnt);
3394         if (unlikely(error))
3395                 goto out;
3396         dir = path.dentry->d_inode;
3397         /* we want directory to be writable */
3398         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3399         if (error)
3400                 goto out2;
3401         if (!dir->i_op->tmpfile) {
3402                 error = -EOPNOTSUPP;
3403                 goto out2;
3404         }
3405         child = d_alloc(path.dentry, &name);
3406         if (unlikely(!child)) {
3407                 error = -ENOMEM;
3408                 goto out2;
3409         }
3410         dput(path.dentry);
3411         path.dentry = child;
3412         error = dir->i_op->tmpfile(dir, child, op->mode);
3413         if (error)
3414                 goto out2;
3415         audit_inode(nd->name, child, 0);
3416         /* Don't check for other permissions, the inode was just created */
3417         error = may_open(&path, 0, op->open_flag);
3418         if (error)
3419                 goto out2;
3420         file->f_path.mnt = path.mnt;
3421         error = finish_open(file, child, NULL, opened);
3422         if (error)
3423                 goto out2;
3424         error = open_check_o_direct(file);
3425         if (error) {
3426                 fput(file);
3427         } else if (!(op->open_flag & O_EXCL)) {
3428                 struct inode *inode = file_inode(file);
3429                 spin_lock(&inode->i_lock);
3430                 inode->i_state |= I_LINKABLE;
3431                 spin_unlock(&inode->i_lock);
3432         }
3433 out2:
3434         mnt_drop_write(path.mnt);
3435 out:
3436         path_put(&path);
3437         return error;
3438 }
3439 
3440 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3441 {
3442         struct path path;
3443         int error = path_lookupat(nd, flags, &path);
3444         if (!error) {
3445                 audit_inode(nd->name, path.dentry, 0);
3446                 error = vfs_open(&path, file, current_cred());
3447                 path_put(&path);
3448         }
3449         return error;
3450 }
3451 
3452 static struct file *path_openat(struct nameidata *nd,
3453                         const struct open_flags *op, unsigned flags)
3454 {
3455         const char *s;
3456         struct file *file;
3457         int opened = 0;
3458         int error;
3459 
3460         file = get_empty_filp();
3461         if (IS_ERR(file))
3462                 return file;
3463 
3464         file->f_flags = op->open_flag;
3465 
3466         if (unlikely(file->f_flags & __O_TMPFILE)) {
3467                 error = do_tmpfile(nd, flags, op, file, &opened);
3468                 goto out2;
3469         }
3470 
3471         if (unlikely(file->f_flags & O_PATH)) {
3472                 error = do_o_path(nd, flags, file);
3473                 if (!error)
3474                         opened |= FILE_OPENED;
3475                 goto out2;
3476         }
3477 
3478         s = path_init(nd, flags);
3479         if (IS_ERR(s)) {
3480                 put_filp(file);
3481                 return ERR_CAST(s);
3482         }
3483         while (!(error = link_path_walk(s, nd)) &&
3484                 (error = do_last(nd, file, op, &opened)) > 0) {
3485                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3486                 s = trailing_symlink(nd);
3487                 if (IS_ERR(s)) {
3488                         error = PTR_ERR(s);
3489                         break;
3490                 }
3491         }
3492         terminate_walk(nd);
3493 out2:
3494         if (!(opened & FILE_OPENED)) {
3495                 BUG_ON(!error);
3496                 put_filp(file);
3497         }
3498         if (unlikely(error)) {
3499                 if (error == -EOPENSTALE) {
3500                         if (flags & LOOKUP_RCU)
3501                                 error = -ECHILD;
3502                         else
3503                                 error = -ESTALE;
3504                 }
3505                 file = ERR_PTR(error);
3506         }
3507         return file;
3508 }
3509 
3510 struct file *do_filp_open(int dfd, struct filename *pathname,
3511                 const struct open_flags *op)
3512 {
3513         struct nameidata nd;
3514         int flags = op->lookup_flags;
3515         struct file *filp;
3516 
3517         set_nameidata(&nd, dfd, pathname);
3518         filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3519         if (unlikely(filp == ERR_PTR(-ECHILD)))
3520                 filp = path_openat(&nd, op, flags);
3521         if (unlikely(filp == ERR_PTR(-ESTALE)))
3522                 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3523         restore_nameidata();
3524         return filp;
3525 }
3526 
3527 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3528                 const char *name, const struct open_flags *op)
3529 {
3530         struct nameidata nd;
3531         struct file *file;
3532         struct filename *filename;
3533         int flags = op->lookup_flags | LOOKUP_ROOT;
3534 
3535         nd.root.mnt = mnt;
3536         nd.root.dentry = dentry;
3537 
3538         if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3539                 return ERR_PTR(-ELOOP);
3540 
3541         filename = getname_kernel(name);
3542         if (IS_ERR(filename))
3543                 return ERR_CAST(filename);
3544 
3545         set_nameidata(&nd, -1, filename);
3546         file = path_openat(&nd, op, flags | LOOKUP_RCU);
3547         if (unlikely(file == ERR_PTR(-ECHILD)))
3548                 file = path_openat(&nd, op, flags);
3549         if (unlikely(file == ERR_PTR(-ESTALE)))
3550                 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3551         restore_nameidata();
3552         putname(filename);
3553         return file;
3554 }
3555 
3556 static struct dentry *filename_create(int dfd, struct filename *name,
3557                                 struct path *path, unsigned int lookup_flags)
3558 {
3559         struct dentry *dentry = ERR_PTR(-EEXIST);
3560         struct qstr last;
3561         int type;
3562         int err2;
3563         int error;
3564         bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3565 
3566         /*
3567          * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3568          * other flags passed in are ignored!
3569          */
3570         lookup_flags &= LOOKUP_REVAL;
3571 
3572         name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3573         if (IS_ERR(name))
3574                 return ERR_CAST(name);
3575 
3576         /*
3577          * Yucky last component or no last component at all?
3578          * (foo/., foo/.., /////)
3579          */
3580         if (unlikely(type != LAST_NORM))
3581                 goto out;
3582 
3583         /* don't fail immediately if it's r/o, at least try to report other errors */
3584         err2 = mnt_want_write(path->mnt);
3585         /*
3586          * Do the final lookup.
3587          */
3588         lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3589         inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3590         dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3591         if (IS_ERR(dentry))
3592                 goto unlock;
3593 
3594         error = -EEXIST;
3595         if (d_is_positive(dentry))
3596                 goto fail;
3597 
3598         /*
3599          * Special case - lookup gave negative, but... we had foo/bar/
3600          * From the vfs_mknod() POV we just have a negative dentry -
3601          * all is fine. Let's be bastards - you had / on the end, you've
3602          * been asking for (non-existent) directory. -ENOENT for you.
3603          */
3604         if (unlikely(!is_dir && last.name[last.len])) {
3605                 error = -ENOENT;
3606                 goto fail;
3607         }
3608         if (unlikely(err2)) {
3609                 error = err2;
3610                 goto fail;
3611         }
3612         putname(name);
3613         return dentry;
3614 fail:
3615         dput(dentry);
3616         dentry = ERR_PTR(error);
3617 unlock:
3618         inode_unlock(path->dentry->d_inode);
3619         if (!err2)
3620                 mnt_drop_write(path->mnt);
3621 out:
3622         path_put(path);
3623         putname(name);
3624         return dentry;
3625 }
3626 
3627 struct dentry *kern_path_create(int dfd, const char *pathname,
3628                                 struct path *path, unsigned int lookup_flags)
3629 {
3630         return filename_create(dfd, getname_kernel(pathname),
3631                                 path, lookup_flags);
3632 }
3633 EXPORT_SYMBOL(kern_path_create);
3634 
3635 void done_path_create(struct path *path, struct dentry *dentry)
3636 {
3637         dput(dentry);
3638         inode_unlock(path->dentry->d_inode);
3639         mnt_drop_write(path->mnt);
3640         path_put(path);
3641 }
3642 EXPORT_SYMBOL(done_path_create);
3643 
3644 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3645                                 struct path *path, unsigned int lookup_flags)
3646 {
3647         return filename_create(dfd, getname(pathname), path, lookup_flags);
3648 }
3649 EXPORT_SYMBOL(user_path_create);
3650 
3651 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3652 {
3653         int error = may_create(dir, dentry);
3654 
3655         if (error)
3656                 return error;
3657 
3658         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3659                 return -EPERM;
3660 
3661         if (!dir->i_op->mknod)
3662                 return -EPERM;
3663 
3664         error = devcgroup_inode_mknod(mode, dev);
3665         if (error)
3666                 return error;
3667 
3668         error = security_inode_mknod(dir, dentry, mode, dev);
3669         if (error)
3670                 return error;
3671 
3672         error = dir->i_op->mknod(dir, dentry, mode, dev);
3673         if (!error)
3674                 fsnotify_create(dir, dentry);
3675         return error;
3676 }
3677 EXPORT_SYMBOL(vfs_mknod);
3678 
3679 static int may_mknod(umode_t mode)
3680 {
3681         switch (mode & S_IFMT) {
3682         case S_IFREG:
3683         case S_IFCHR:
3684         case S_IFBLK:
3685         case S_IFIFO:
3686         case S_IFSOCK:
3687         case 0: /* zero mode translates to S_IFREG */
3688                 return 0;
3689         case S_IFDIR:
3690                 return -EPERM;
3691         default:
3692                 return -EINVAL;
3693         }
3694 }
3695 
3696 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3697                 unsigned, dev)
3698 {
3699         struct dentry *dentry;
3700         struct path path;
3701         int error;
3702         unsigned int lookup_flags = 0;
3703 
3704         error = may_mknod(mode);
3705         if (error)
3706                 return error;
3707 retry:
3708         dentry = user_path_create(dfd, filename, &path, lookup_flags);
3709         if (IS_ERR(dentry))
3710                 return PTR_ERR(dentry);
3711 
3712         if (!IS_POSIXACL(path.dentry->d_inode))
3713                 mode &= ~current_umask();
3714         error = security_path_mknod(&path, dentry, mode, dev);
3715         if (error)
3716                 goto out;
3717         switch (mode & S_IFMT) {
3718                 case 0: case S_IFREG:
3719                         error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3720                         if (!error)
3721                                 ima_post_path_mknod(dentry);
3722                         break;
3723                 case S_IFCHR: case S_IFBLK:
3724                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3725                                         new_decode_dev(dev));
3726                         break;
3727                 case S_IFIFO: case S_IFSOCK:
3728                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3729                         break;
3730         }
3731 out:
3732         done_path_create(&path, dentry);
3733         if (retry_estale(error, lookup_flags)) {
3734                 lookup_flags |= LOOKUP_REVAL;
3735                 goto retry;
3736         }
3737         return error;
3738 }
3739 
3740 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3741 {
3742         return sys_mknodat(AT_FDCWD, filename, mode, dev);
3743 }
3744 
3745 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3746 {
3747         int error = may_create(dir, dentry);
3748         unsigned max_links = dir->i_sb->s_max_links;
3749 
3750         if (error)
3751                 return error;
3752 
3753         if (!dir->i_op->mkdir)
3754                 return -EPERM;
3755 
3756         mode &= (S_IRWXUGO|S_ISVTX);
3757         error = security_inode_mkdir(dir, dentry, mode);
3758         if (error)
3759                 return error;
3760 
3761         if (max_links && dir->i_nlink >= max_links)
3762                 return -EMLINK;
3763 
3764         error = dir->i_op->mkdir(dir, dentry, mode);
3765         if (!error)
3766                 fsnotify_mkdir(dir, dentry);
3767         return error;
3768 }
3769 EXPORT_SYMBOL(vfs_mkdir);
3770 
3771 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3772 {
3773         struct dentry *dentry;
3774         struct path path;
3775         int error;
3776         unsigned int lookup_flags = LOOKUP_DIRECTORY;
3777 
3778 retry:
3779         dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3780         if (IS_ERR(dentry))
3781                 return PTR_ERR(dentry);
3782 
3783         if (!IS_POSIXACL(path.dentry->d_inode))
3784                 mode &= ~current_umask();
3785         error = security_path_mkdir(&path, dentry, mode);
3786         if (!error)
3787                 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3788         done_path_create(&path, dentry);
3789         if (retry_estale(error, lookup_flags)) {
3790                 lookup_flags |= LOOKUP_REVAL;
3791                 goto retry;
3792         }
3793         return error;
3794 }
3795 
3796 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3797 {
3798         return sys_mkdirat(AT_FDCWD, pathname, mode);
3799 }
3800 
3801 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3802 {
3803         int error = may_delete(dir, dentry, 1);
3804 
3805         if (error)
3806                 return error;
3807 
3808         if (!dir->i_op->rmdir)
3809                 return -EPERM;
3810 
3811         dget(dentry);
3812         inode_lock(dentry->d_inode);
3813 
3814         error = -EBUSY;
3815         if (is_local_mountpoint(dentry))
3816                 goto out;
3817 
3818         error = security_inode_rmdir(dir, dentry);
3819         if (error)
3820                 goto out;
3821 
3822         shrink_dcache_parent(dentry);
3823         error = dir->i_op->rmdir(dir, dentry);
3824         if (error)
3825                 goto out;
3826 
3827         dentry->d_inode->i_flags |= S_DEAD;
3828         dont_mount(dentry);
3829         detach_mounts(dentry);
3830 
3831 out:
3832         inode_unlock(dentry->d_inode);
3833         dput(dentry);
3834         if (!error)
3835                 d_delete(dentry);
3836         return error;
3837 }
3838 EXPORT_SYMBOL(vfs_rmdir);
3839 
3840 static long do_rmdir(int dfd, const char __user *pathname)
3841 {
3842         int error = 0;
3843         struct filename *name;
3844         struct dentry *dentry;
3845         struct path path;
3846         struct qstr last;
3847         int type;
3848         unsigned int lookup_flags = 0;
3849 retry:
3850         name = user_path_parent(dfd, pathname,
3851                                 &path, &last, &type, lookup_flags);
3852         if (IS_ERR(name))
3853                 return PTR_ERR(name);
3854 
3855         switch (type) {
3856         case LAST_DOTDOT:
3857                 error = -ENOTEMPTY;
3858                 goto exit1;
3859         case LAST_DOT:
3860                 error = -EINVAL;
3861                 goto exit1;
3862         case LAST_ROOT:
3863                 error = -EBUSY;
3864                 goto exit1;
3865         }
3866 
3867         error = mnt_want_write(path.mnt);
3868         if (error)
3869                 goto exit1;
3870 
3871         inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3872         dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3873         error = PTR_ERR(dentry);
3874         if (IS_ERR(dentry))
3875                 goto exit2;
3876         if (!dentry->d_inode) {
3877                 error = -ENOENT;
3878                 goto exit3;
3879         }
3880         error = security_path_rmdir(&path, dentry);
3881         if (error)
3882                 goto exit3;
3883         error = vfs_rmdir(path.dentry->d_inode, dentry);
3884 exit3:
3885         dput(dentry);
3886 exit2:
3887         inode_unlock(path.dentry->d_inode);
3888         mnt_drop_write(path.mnt);
3889 exit1:
3890         path_put(&path);
3891         putname(name);
3892         if (retry_estale(error, lookup_flags)) {
3893                 lookup_flags |= LOOKUP_REVAL;
3894                 goto retry;
3895         }
3896         return error;
3897 }
3898 
3899 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3900 {
3901         return do_rmdir(AT_FDCWD, pathname);
3902 }
3903 
3904 /**
3905  * vfs_unlink - unlink a filesystem object
3906  * @dir:        parent directory
3907  * @dentry:     victim
3908  * @delegated_inode: returns victim inode, if the inode is delegated.
3909  *
3910  * The caller must hold dir->i_mutex.
3911  *
3912  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3913  * return a reference to the inode in delegated_inode.  The caller
3914  * should then break the delegation on that inode and retry.  Because
3915  * breaking a delegation may take a long time, the caller should drop
3916  * dir->i_mutex before doing so.
3917  *
3918  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3919  * be appropriate for callers that expect the underlying filesystem not
3920  * to be NFS exported.
3921  */
3922 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3923 {
3924         struct inode *target = dentry->d_inode;
3925         int error = may_delete(dir, dentry, 0);
3926 
3927         if (error)
3928                 return error;
3929 
3930         if (!dir->i_op->unlink)
3931                 return -EPERM;
3932 
3933         inode_lock(target);
3934         if (is_local_mountpoint(dentry))
3935                 error = -EBUSY;
3936         else {
3937                 error = security_inode_unlink(dir, dentry);
3938                 if (!error) {
3939                         error = try_break_deleg(target, delegated_inode);
3940                         if (error)
3941                                 goto out;
3942                         error = dir->i_op->unlink(dir, dentry);
3943                         if (!error) {
3944                                 dont_mount(dentry);
3945                                 detach_mounts(dentry);
3946                         }
3947                 }
3948         }
3949 out:
3950         inode_unlock(target);
3951 
3952         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3953         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3954                 fsnotify_link_count(target);
3955                 d_delete(dentry);
3956         }
3957 
3958         return error;
3959 }
3960 EXPORT_SYMBOL(vfs_unlink);
3961 
3962 /*
3963  * Make sure that the actual truncation of the file will occur outside its
3964  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3965  * writeout happening, and we don't want to prevent access to the directory
3966  * while waiting on the I/O.
3967  */
3968 static long do_unlinkat(int dfd, const char __user *pathname)
3969 {
3970         int error;
3971         struct filename *name;
3972         struct dentry *dentry;
3973         struct path path;
3974         struct qstr last;
3975         int type;
3976         struct inode *inode = NULL;
3977         struct inode *delegated_inode = NULL;
3978         unsigned int lookup_flags = 0;
3979 retry:
3980         name = user_path_parent(dfd, pathname,
3981                                 &path, &last, &type, lookup_flags);
3982         if (IS_ERR(name))
3983                 return PTR_ERR(name);
3984 
3985         error = -EISDIR;
3986         if (type != LAST_NORM)
3987                 goto exit1;
3988 
3989         error = mnt_want_write(path.mnt);
3990         if (error)
3991                 goto exit1;
3992 retry_deleg:
3993         inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3994         dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3995         error = PTR_ERR(dentry);
3996         if (!IS_ERR(dentry)) {
3997                 /* Why not before? Because we want correct error value */
3998                 if (last.name[last.len])
3999                         goto slashes;
4000                 inode = dentry->d_inode;
4001                 if (d_is_negative(dentry))
4002                         goto slashes;
4003                 ihold(inode);
4004                 error = security_path_unlink(&path, dentry);
4005                 if (error)
4006                         goto exit2;
4007                 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4008 exit2:
4009                 dput(dentry);
4010         }
4011         inode_unlock(path.dentry->d_inode);
4012         if (inode)
4013                 iput(inode);    /* truncate the inode here */
4014         inode = NULL;
4015         if (delegated_inode) {
4016                 error = break_deleg_wait(&delegated_inode);
4017                 if (!error)
4018                         goto retry_deleg;
4019         }
4020         mnt_drop_write(path.mnt);
4021 exit1:
4022         path_put(&path);
4023         putname(name);
4024         if (retry_estale(error, lookup_flags)) {
4025                 lookup_flags |= LOOKUP_REVAL;
4026                 inode = NULL;
4027                 goto retry;
4028         }
4029         return error;
4030 
4031 slashes:
4032         if (d_is_negative(dentry))
4033                 error = -ENOENT;
4034         else if (d_is_dir(dentry))
4035                 error = -EISDIR;
4036         else
4037                 error = -ENOTDIR;
4038         goto exit2;
4039 }
4040 
4041 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4042 {
4043         if ((flag & ~AT_REMOVEDIR) != 0)
4044                 return -EINVAL;
4045 
4046         if (flag & AT_REMOVEDIR)
4047                 return do_rmdir(dfd, pathname);
4048 
4049         return do_unlinkat(dfd, pathname);
4050 }
4051 
4052 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4053 {
4054         return do_unlinkat(AT_FDCWD, pathname);
4055 }
4056 
4057 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4058 {
4059         int error = may_create(dir, dentry);
4060 
4061         if (error)
4062                 return error;
4063 
4064         if (!dir->i_op->symlink)
4065                 return -EPERM;
4066 
4067         error = security_inode_symlink(dir, dentry, oldname);
4068         if (error)
4069                 return error;
4070 
4071         error = dir->i_op->symlink(dir, dentry, oldname);
4072         if (!error)
4073                 fsnotify_create(dir, dentry);
4074         return error;
4075 }
4076 EXPORT_SYMBOL(vfs_symlink);
4077 
4078 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4079                 int, newdfd, const char __user *, newname)
4080 {
4081         int error;
4082         struct filename *from;
4083         struct dentry *dentry;
4084         struct path path;
4085         unsigned int lookup_flags = 0;
4086 
4087         from = getname(oldname);
4088         if (IS_ERR(from))
4089                 return PTR_ERR(from);
4090 retry:
4091         dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4092         error = PTR_ERR(dentry);
4093         if (IS_ERR(dentry))
4094                 goto out_putname;
4095 
4096         error = security_path_symlink(&path, dentry, from->name);
4097         if (!error)
4098                 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4099         done_path_create(&path, dentry);
4100         if (retry_estale(error, lookup_flags)) {
4101                 lookup_flags |= LOOKUP_REVAL;
4102                 goto retry;
4103         }
4104 out_putname:
4105         putname(from);
4106         return error;
4107 }
4108 
4109 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4110 {
4111         return sys_symlinkat(oldname, AT_FDCWD, newname);
4112 }
4113 
4114 /**
4115  * vfs_link - create a new link
4116  * @old_dentry: object to be linked
4117  * @dir:        new parent
4118  * @new_dentry: where to create the new link
4119  * @delegated_inode: returns inode needing a delegation break
4120  *
4121  * The caller must hold dir->i_mutex
4122  *
4123  * If vfs_link discovers a delegation on the to-be-linked file in need
4124  * of breaking, it will return -EWOULDBLOCK and return a reference to the
4125  * inode in delegated_inode.  The caller should then break the delegation
4126  * and retry.  Because breaking a delegation may take a long time, the
4127  * caller should drop the i_mutex before doing so.
4128  *
4129  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4130  * be appropriate for callers that expect the underlying filesystem not
4131  * to be NFS exported.
4132  */
4133 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4134 {
4135         struct inode *inode = old_dentry->d_inode;
4136         unsigned max_links = dir->i_sb->s_max_links;
4137         int error;
4138 
4139         if (!inode)
4140                 return -ENOENT;
4141 
4142         error = may_create(dir, new_dentry);
4143         if (error)
4144                 return error;
4145 
4146         if (dir->i_sb != inode->i_sb)
4147                 return -EXDEV;
4148 
4149         /*
4150          * A link to an append-only or immutable file cannot be created.
4151          */
4152         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4153                 return -EPERM;
4154         if (!dir->i_op->link)
4155                 return -EPERM;
4156         if (S_ISDIR(inode->i_mode))
4157                 return -EPERM;
4158 
4159         error = security_inode_link(old_dentry, dir, new_dentry);
4160         if (error)
4161                 return error;
4162 
4163         inode_lock(inode);
4164         /* Make sure we don't allow creating hardlink to an unlinked file */
4165         if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4166                 error =  -ENOENT;
4167         else if (max_links && inode->i_nlink >= max_links)
4168                 error = -EMLINK;
4169         else {
4170                 error = try_break_deleg(inode, delegated_inode);
4171                 if (!error)
4172                         error = dir->i_op->link(old_dentry, dir, new_dentry);
4173         }
4174 
4175         if (!error && (inode->i_state & I_LINKABLE)) {
4176                 spin_lock(&inode->i_lock);
4177                 inode->i_state &= ~I_LINKABLE;
4178                 spin_unlock(&inode->i_lock);
4179         }
4180         inode_unlock(inode);
4181         if (!error)
4182                 fsnotify_link(dir, inode, new_dentry);
4183         return error;
4184 }
4185 EXPORT_SYMBOL(vfs_link);
4186 
4187 /*
4188  * Hardlinks are often used in delicate situations.  We avoid
4189  * security-related surprises by not following symlinks on the
4190  * newname.  --KAB
4191  *
4192  * We don't follow them on the oldname either to be compatible
4193  * with linux 2.0, and to avoid hard-linking to directories
4194  * and other special files.  --ADM
4195  */
4196 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4197                 int, newdfd, const char __user *, newname, int, flags)
4198 {
4199         struct dentry *new_dentry;
4200         struct path old_path, new_path;
4201         struct inode *delegated_inode = NULL;
4202         int how = 0;
4203         int error;
4204 
4205         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4206                 return -EINVAL;
4207         /*
4208          * To use null names we require CAP_DAC_READ_SEARCH
4209          * This ensures that not everyone will be able to create
4210          * handlink using the passed filedescriptor.
4211          */
4212         if (flags & AT_EMPTY_PATH) {
4213                 if (!capable(CAP_DAC_READ_SEARCH))
4214                         return -ENOENT;
4215                 how = LOOKUP_EMPTY;
4216         }
4217 
4218         if (flags & AT_SYMLINK_FOLLOW)
4219                 how |= LOOKUP_FOLLOW;
4220 retry:
4221         error = user_path_at(olddfd, oldname, how, &old_path);
4222         if (error)
4223                 return error;
4224 
4225         new_dentry = user_path_create(newdfd, newname, &new_path,
4226                                         (how & LOOKUP_REVAL));
4227         error = PTR_ERR(new_dentry);
4228         if (IS_ERR(new_dentry))
4229                 goto out;
4230 
4231         error = -EXDEV;
4232         if (old_path.mnt != new_path.mnt)
4233                 goto out_dput;
4234         error = may_linkat(&old_path);
4235         if (unlikely(error))
4236                 goto out_dput;
4237         error = security_path_link(old_path.dentry, &new_path, new_dentry);
4238         if (error)
4239                 goto out_dput;
4240         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4241 out_dput:
4242         done_path_create(&new_path, new_dentry);
4243         if (delegated_inode) {
4244                 error = break_deleg_wait(&delegated_inode);
4245                 if (!error) {
4246                         path_put(&old_path);
4247                         goto retry;
4248                 }
4249         }
4250         if (retry_estale(error, how)) {
4251                 path_put(&old_path);
4252                 how |= LOOKUP_REVAL;
4253                 goto retry;
4254         }
4255 out:
4256         path_put(&old_path);
4257 
4258         return error;
4259 }
4260 
4261 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4262 {
4263         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4264 }
4265 
4266 /**
4267  * vfs_rename - rename a filesystem object
4268  * @old_dir:    parent of source
4269  * @old_dentry: source
4270  * @new_dir:    parent of destination
4271  * @new_dentry: destination
4272  * @delegated_inode: returns an inode needing a delegation break
4273  * @flags:      rename flags
4274  *
4275  * The caller must hold multiple mutexes--see lock_rename()).
4276  *
4277  * If vfs_rename discovers a delegation in need of breaking at either
4278  * the source or destination, it will return -EWOULDBLOCK and return a
4279  * reference to the inode in delegated_inode.  The caller should then
4280  * break the delegation and retry.  Because breaking a delegation may
4281  * take a long time, the caller should drop all locks before doing
4282  * so.
4283  *
4284  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4285  * be appropriate for callers that expect the underlying filesystem not
4286  * to be NFS exported.
4287  *
4288  * The worst of all namespace operations - renaming directory. "Perverted"
4289  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4290  * Problems:
4291  *      a) we can get into loop creation.
4292  *      b) race potential - two innocent renames can create a loop together.
4293  *         That's where 4.4 screws up. Current fix: serialization on
4294  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4295  *         story.
4296  *      c) we have to lock _four_ objects - parents and victim (if it exists),
4297  *         and source (if it is not a directory).
4298  *         And that - after we got ->i_mutex on parents (until then we don't know
4299  *         whether the target exists).  Solution: try to be smart with locking
4300  *         order for inodes.  We rely on the fact that tree topology may change
4301  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4302  *         move will be locked.  Thus we can rank directories by the tree
4303  *         (ancestors first) and rank all non-directories after them.
4304  *         That works since everybody except rename does "lock parent, lookup,
4305  *         lock child" and rename is under ->s_vfs_rename_mutex.
4306  *         HOWEVER, it relies on the assumption that any object with ->lookup()
4307  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4308  *         we'd better make sure that there's no link(2) for them.
4309  *      d) conversion from fhandle to dentry may come in the wrong moment - when
4310  *         we are removing the target. Solution: we will have to grab ->i_mutex
4311  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4312  *         ->i_mutex on parents, which works but leads to some truly excessive
4313  *         locking].
4314  */
4315 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4316                struct inode *new_dir, struct dentry *new_dentry,
4317                struct inode **delegated_inode, unsigned int flags)
4318 {
4319         int error;
4320         bool is_dir = d_is_dir(old_dentry);
4321         const unsigned char *old_name;
4322         struct inode *source = old_dentry->d_inode;
4323         struct inode *target = new_dentry->d_inode;
4324         bool new_is_dir = false;
4325         unsigned max_links = new_dir->i_sb->s_max_links;
4326 
4327         /*
4328          * Check source == target.
4329          * On overlayfs need to look at underlying inodes.
4330          */
4331         if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0))
4332                 return 0;
4333 
4334         error = may_delete(old_dir, old_dentry, is_dir);
4335         if (error)
4336                 return error;
4337 
4338         if (!target) {
4339                 error = may_create(new_dir, new_dentry);
4340         } else {
4341                 new_is_dir = d_is_dir(new_dentry);
4342 
4343                 if (!(flags & RENAME_EXCHANGE))
4344                         error = may_delete(new_dir, new_dentry, is_dir);
4345                 else
4346                         error = may_delete(new_dir, new_dentry, new_is_dir);
4347         }
4348         if (error)
4349                 return error;
4350 
4351         if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4352                 return -EPERM;
4353 
4354         if (flags && !old_dir->i_op->rename2)
4355                 return -EINVAL;
4356 
4357         /*
4358          * If we are going to change the parent - check write permissions,
4359          * we'll need to flip '..'.
4360          */
4361         if (new_dir != old_dir) {
4362                 if (is_dir) {
4363                         error = inode_permission(source, MAY_WRITE);
4364                         if (error)
4365                                 return error;
4366                 }
4367                 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4368                         error = inode_permission(target, MAY_WRITE);
4369                         if (error)
4370                                 return error;
4371                 }
4372         }
4373 
4374         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4375                                       flags);
4376         if (error)
4377                 return error;
4378 
4379         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4380         dget(new_dentry);
4381         if (!is_dir || (flags & RENAME_EXCHANGE))
4382                 lock_two_nondirectories(source, target);
4383         else if (target)
4384                 inode_lock(target);
4385 
4386         error = -EBUSY;
4387         if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4388                 goto out;
4389 
4390         if (max_links && new_dir != old_dir) {
4391                 error = -EMLINK;
4392                 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4393                         goto out;
4394                 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4395                     old_dir->i_nlink >= max_links)
4396                         goto out;
4397         }
4398         if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4399                 shrink_dcache_parent(new_dentry);
4400         if (!is_dir) {
4401                 error = try_break_deleg(source, delegated_inode);
4402                 if (error)
4403                         goto out;
4404         }
4405         if (target && !new_is_dir) {
4406                 error = try_break_deleg(target, delegated_inode);
4407                 if (error)
4408                         goto out;
4409         }
4410         if (!old_dir->i_op->rename2) {
4411                 error = old_dir->i_op->rename(old_dir, old_dentry,
4412                                               new_dir, new_dentry);
4413         } else {
4414                 WARN_ON(old_dir->i_op->rename != NULL);
4415                 error = old_dir->i_op->rename2(old_dir, old_dentry,
4416                                                new_dir, new_dentry, flags);
4417         }
4418         if (error)
4419                 goto out;
4420 
4421         if (!(flags & RENAME_EXCHANGE) && target) {
4422                 if (is_dir)
4423                         target->i_flags |= S_DEAD;
4424                 dont_mount(new_dentry);
4425                 detach_mounts(new_dentry);
4426         }
4427         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4428                 if (!(flags & RENAME_EXCHANGE))
4429                         d_move(old_dentry, new_dentry);
4430                 else
4431                         d_exchange(old_dentry, new_dentry);
4432         }
4433 out:
4434         if (!is_dir || (flags & RENAME_EXCHANGE))
4435                 unlock_two_nondirectories(source, target);
4436         else if (target)
4437                 inode_unlock(target);
4438         dput(new_dentry);
4439         if (!error) {
4440                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4441                               !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4442                 if (flags & RENAME_EXCHANGE) {
4443                         fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4444                                       new_is_dir, NULL, new_dentry);
4445                 }
4446         }
4447         fsnotify_oldname_free(old_name);
4448 
4449         return error;
4450 }
4451 EXPORT_SYMBOL(vfs_rename);
4452 
4453 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4454                 int, newdfd, const char __user *, newname, unsigned int, flags)
4455 {
4456         struct dentry *old_dentry, *new_dentry;
4457         struct dentry *trap;
4458         struct path old_path, new_path;
4459         struct qstr old_last, new_last;
4460         int old_type, new_type;
4461         struct inode *delegated_inode = NULL;
4462         struct filename *from;
4463         struct filename *to;
4464         unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4465         bool should_retry = false;
4466         int error;
4467 
4468         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4469                 return -EINVAL;
4470 
4471         if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4472             (flags & RENAME_EXCHANGE))
4473                 return -EINVAL;
4474 
4475         if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4476                 return -EPERM;
4477 
4478         if (flags & RENAME_EXCHANGE)
4479                 target_flags = 0;
4480 
4481 retry:
4482         from = user_path_parent(olddfd, oldname,
4483                                 &old_path, &old_last, &old_type, lookup_flags);
4484         if (IS_ERR(from)) {
4485                 error = PTR_ERR(from);
4486                 goto exit;
4487         }
4488 
4489         to = user_path_parent(newdfd, newname,
4490                                 &new_path, &new_last, &new_type, lookup_flags);
4491         if (IS_ERR(to)) {
4492                 error = PTR_ERR(to);
4493                 goto exit1;
4494         }
4495 
4496         error = -EXDEV;
4497         if (old_path.mnt != new_path.mnt)
4498                 goto exit2;
4499 
4500         error = -EBUSY;
4501         if (old_type != LAST_NORM)
4502                 goto exit2;
4503 
4504         if (flags & RENAME_NOREPLACE)
4505                 error = -EEXIST;
4506         if (new_type != LAST_NORM)
4507                 goto exit2;
4508 
4509         error = mnt_want_write(old_path.mnt);
4510         if (error)
4511                 goto exit2;
4512 
4513 retry_deleg:
4514         trap = lock_rename(new_path.dentry, old_path.dentry);
4515 
4516         old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4517         error = PTR_ERR(old_dentry);
4518         if (IS_ERR(old_dentry))
4519                 goto exit3;
4520         /* source must exist */
4521         error = -ENOENT;
4522         if (d_is_negative(old_dentry))
4523                 goto exit4;
4524         new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4525         error = PTR_ERR(new_dentry);
4526         if (IS_ERR(new_dentry))
4527                 goto exit4;
4528         error = -EEXIST;
4529         if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4530                 goto exit5;
4531         if (flags & RENAME_EXCHANGE) {
4532                 error = -ENOENT;
4533                 if (d_is_negative(new_dentry))
4534                         goto exit5;
4535 
4536                 if (!d_is_dir(new_dentry)) {
4537                         error = -ENOTDIR;
4538                         if (new_last.name[new_last.len])
4539                                 goto exit5;
4540                 }
4541         }
4542         /* unless the source is a directory trailing slashes give -ENOTDIR */
4543         if (!d_is_dir(old_dentry)) {
4544                 error = -ENOTDIR;
4545                 if (old_last.name[old_last.len])
4546                         goto exit5;
4547                 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4548                         goto exit5;
4549         }
4550         /* source should not be ancestor of target */
4551         error = -EINVAL;
4552         if (old_dentry == trap)
4553                 goto exit5;
4554         /* target should not be an ancestor of source */
4555         if (!(flags & RENAME_EXCHANGE))
4556                 error = -ENOTEMPTY;
4557         if (new_dentry == trap)
4558                 goto exit5;
4559 
4560         error = security_path_rename(&old_path, old_dentry,
4561                                      &new_path, new_dentry, flags);
4562         if (error)
4563                 goto exit5;
4564         error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4565                            new_path.dentry->d_inode, new_dentry,
4566                            &delegated_inode, flags);
4567 exit5:
4568         dput(new_dentry);
4569 exit4:
4570         dput(old_dentry);
4571 exit3:
4572         unlock_rename(new_path.dentry, old_path.dentry);
4573         if (delegated_inode) {
4574                 error = break_deleg_wait(&delegated_inode);
4575                 if (!error)
4576                         goto retry_deleg;
4577         }
4578         mnt_drop_write(old_path.mnt);
4579 exit2:
4580         if (retry_estale(error, lookup_flags))
4581                 should_retry = true;
4582         path_put(&new_path);
4583         putname(to);
4584 exit1:
4585         path_put(&old_path);
4586         putname(from);
4587         if (should_retry) {
4588                 should_retry = false;
4589                 lookup_flags |= LOOKUP_REVAL;
4590                 goto retry;
4591         }
4592 exit:
4593         return error;
4594 }
4595 
4596 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4597                 int, newdfd, const char __user *, newname)
4598 {
4599         return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4600 }
4601 
4602 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4603 {
4604         return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4605 }
4606 
4607 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4608 {
4609         int error = may_create(dir, dentry);
4610         if (error)
4611                 return error;
4612 
4613         if (!dir->i_op->mknod)
4614                 return -EPERM;
4615 
4616         return dir->i_op->mknod(dir, dentry,
4617                                 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4618 }
4619 EXPORT_SYMBOL(vfs_whiteout);
4620 
4621 int readlink_copy(char __user *buffer, int buflen, const char *link)
4622 {
4623         int len = PTR_ERR(link);
4624         if (IS_ERR(link))
4625                 goto out;
4626 
4627         len = strlen(link);
4628         if (len > (unsigned) buflen)
4629                 len = buflen;
4630         if (copy_to_user(buffer, link, len))
4631                 len = -EFAULT;
4632 out:
4633         return len;
4634 }
4635 
4636 /*
4637  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4638  * have ->get_link() not calling nd_jump_link().  Using (or not using) it
4639  * for any given inode is up to filesystem.
4640  */
4641 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4642 {
4643         DEFINE_DELAYED_CALL(done);
4644         struct inode *inode = d_inode(dentry);
4645         const char *link = inode->i_link;
4646         int res;
4647 
4648         if (!link) {
4649                 link = inode->i_op->get_link(dentry, inode, &done);
4650                 if (IS_ERR(link))
4651                         return PTR_ERR(link);
4652         }
4653         res = readlink_copy(buffer, buflen, link);
4654         do_delayed_call(&done);
4655         return res;
4656 }
4657 EXPORT_SYMBOL(generic_readlink);
4658 
4659 /* get the link contents into pagecache */
4660 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4661                           struct delayed_call *callback)
4662 {
4663         char *kaddr;
4664         struct page *page;
4665         struct address_space *mapping = inode->i_mapping;
4666 
4667         if (!dentry) {
4668                 page = find_get_page(mapping, 0);
4669                 if (!page)
4670                         return ERR_PTR(-ECHILD);
4671                 if (!PageUptodate(page)) {
4672                         put_page(page);
4673                         return ERR_PTR(-ECHILD);
4674                 }
4675         } else {
4676                 page = read_mapping_page(mapping, 0, NULL);
4677                 if (IS_ERR(page))
4678                         return (char*)page;
4679         }
4680         set_delayed_call(callback, page_put_link, page);
4681         BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4682         kaddr = page_address(page);
4683         nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4684         return kaddr;
4685 }
4686 
4687 EXPORT_SYMBOL(page_get_link);
4688 
4689 void page_put_link(void *arg)
4690 {
4691         put_page(arg);
4692 }
4693 EXPORT_SYMBOL(page_put_link);
4694 
4695 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4696 {
4697         DEFINE_DELAYED_CALL(done);
4698         int res = readlink_copy(buffer, buflen,
4699                                 page_get_link(dentry, d_inode(dentry),
4700                                               &done));
4701         do_delayed_call(&done);
4702         return res;
4703 }
4704 EXPORT_SYMBOL(page_readlink);
4705 
4706 /*
4707  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4708  */
4709 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4710 {
4711         struct address_space *mapping = inode->i_mapping;
4712         struct page *page;
4713         void *fsdata;
4714         int err;
4715         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4716         if (nofs)
4717                 flags |= AOP_FLAG_NOFS;
4718 
4719 retry:
4720         err = pagecache_write_begin(NULL, mapping, 0, len-1,
4721                                 flags, &page, &fsdata);
4722         if (err)
4723                 goto fail;
4724 
4725         memcpy(page_address(page), symname, len-1);
4726 
4727         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4728                                                         page, fsdata);
4729         if (err < 0)
4730                 goto fail;
4731         if (err < len-1)
4732                 goto retry;
4733 
4734         mark_inode_dirty(inode);
4735         return 0;
4736 fail:
4737         return err;
4738 }
4739 EXPORT_SYMBOL(__page_symlink);
4740 
4741 int page_symlink(struct inode *inode, const char *symname, int len)
4742 {
4743         return __page_symlink(inode, symname, len,
4744                         !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4745 }
4746 EXPORT_SYMBOL(page_symlink);
4747 
4748 const struct inode_operations page_symlink_inode_operations = {
4749         .readlink       = generic_readlink,
4750         .get_link       = page_get_link,
4751 };
4752 EXPORT_SYMBOL(page_symlink_inode_operations);
4753 

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