Version:  2.0.40 2.2.26 2.4.37 3.8 3.9 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

Linux/fs/namei.c

  1 /*
  2  *  linux/fs/namei.c
  3  *
  4  *  Copyright (C) 1991, 1992  Linus Torvalds
  5  */
  6 
  7 /*
  8  * Some corrections by tytso.
  9  */
 10 
 11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
 12  * lookup logic.
 13  */
 14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
 15  */
 16 
 17 #include <linux/init.h>
 18 #include <linux/export.h>
 19 #include <linux/kernel.h>
 20 #include <linux/slab.h>
 21 #include <linux/fs.h>
 22 #include <linux/namei.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/fsnotify.h>
 25 #include <linux/personality.h>
 26 #include <linux/security.h>
 27 #include <linux/ima.h>
 28 #include <linux/syscalls.h>
 29 #include <linux/mount.h>
 30 #include <linux/audit.h>
 31 #include <linux/capability.h>
 32 #include <linux/file.h>
 33 #include <linux/fcntl.h>
 34 #include <linux/device_cgroup.h>
 35 #include <linux/fs_struct.h>
 36 #include <linux/posix_acl.h>
 37 #include <linux/hash.h>
 38 #include <asm/uaccess.h>
 39 
 40 #include "internal.h"
 41 #include "mount.h"
 42 
 43 /* [Feb-1997 T. Schoebel-Theuer]
 44  * Fundamental changes in the pathname lookup mechanisms (namei)
 45  * were necessary because of omirr.  The reason is that omirr needs
 46  * to know the _real_ pathname, not the user-supplied one, in case
 47  * of symlinks (and also when transname replacements occur).
 48  *
 49  * The new code replaces the old recursive symlink resolution with
 50  * an iterative one (in case of non-nested symlink chains).  It does
 51  * this with calls to <fs>_follow_link().
 52  * As a side effect, dir_namei(), _namei() and follow_link() are now 
 53  * replaced with a single function lookup_dentry() that can handle all 
 54  * the special cases of the former code.
 55  *
 56  * With the new dcache, the pathname is stored at each inode, at least as
 57  * long as the refcount of the inode is positive.  As a side effect, the
 58  * size of the dcache depends on the inode cache and thus is dynamic.
 59  *
 60  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
 61  * resolution to correspond with current state of the code.
 62  *
 63  * Note that the symlink resolution is not *completely* iterative.
 64  * There is still a significant amount of tail- and mid- recursion in
 65  * the algorithm.  Also, note that <fs>_readlink() is not used in
 66  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
 67  * may return different results than <fs>_follow_link().  Many virtual
 68  * filesystems (including /proc) exhibit this behavior.
 69  */
 70 
 71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
 72  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
 73  * and the name already exists in form of a symlink, try to create the new
 74  * name indicated by the symlink. The old code always complained that the
 75  * name already exists, due to not following the symlink even if its target
 76  * is nonexistent.  The new semantics affects also mknod() and link() when
 77  * the name is a symlink pointing to a non-existent name.
 78  *
 79  * I don't know which semantics is the right one, since I have no access
 80  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
 81  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
 82  * "old" one. Personally, I think the new semantics is much more logical.
 83  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
 84  * file does succeed in both HP-UX and SunOs, but not in Solaris
 85  * and in the old Linux semantics.
 86  */
 87 
 88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
 89  * semantics.  See the comments in "open_namei" and "do_link" below.
 90  *
 91  * [10-Sep-98 Alan Modra] Another symlink change.
 92  */
 93 
 94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
 95  *      inside the path - always follow.
 96  *      in the last component in creation/removal/renaming - never follow.
 97  *      if LOOKUP_FOLLOW passed - follow.
 98  *      if the pathname has trailing slashes - follow.
 99  *      otherwise - don't follow.
100  * (applied in that order).
101  *
102  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104  * During the 2.4 we need to fix the userland stuff depending on it -
105  * hopefully we will be able to get rid of that wart in 2.5. So far only
106  * XEmacs seems to be relying on it...
107  */
108 /*
109  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
111  * any extra contention...
112  */
113 
114 /* In order to reduce some races, while at the same time doing additional
115  * checking and hopefully speeding things up, we copy filenames to the
116  * kernel data space before using them..
117  *
118  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119  * PATH_MAX includes the nul terminator --RR.
120  */
121 
122 #define EMBEDDED_NAME_MAX       (PATH_MAX - offsetof(struct filename, iname))
123 
124 struct filename *
125 getname_flags(const char __user *filename, int flags, int *empty)
126 {
127         struct filename *result;
128         char *kname;
129         int len;
130 
131         result = audit_reusename(filename);
132         if (result)
133                 return result;
134 
135         result = __getname();
136         if (unlikely(!result))
137                 return ERR_PTR(-ENOMEM);
138 
139         /*
140          * First, try to embed the struct filename inside the names_cache
141          * allocation
142          */
143         kname = (char *)result->iname;
144         result->name = kname;
145 
146         len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147         if (unlikely(len < 0)) {
148                 __putname(result);
149                 return ERR_PTR(len);
150         }
151 
152         /*
153          * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154          * separate struct filename so we can dedicate the entire
155          * names_cache allocation for the pathname, and re-do the copy from
156          * userland.
157          */
158         if (unlikely(len == EMBEDDED_NAME_MAX)) {
159                 const size_t size = offsetof(struct filename, iname[1]);
160                 kname = (char *)result;
161 
162                 /*
163                  * size is chosen that way we to guarantee that
164                  * result->iname[0] is within the same object and that
165                  * kname can't be equal to result->iname, no matter what.
166                  */
167                 result = kzalloc(size, GFP_KERNEL);
168                 if (unlikely(!result)) {
169                         __putname(kname);
170                         return ERR_PTR(-ENOMEM);
171                 }
172                 result->name = kname;
173                 len = strncpy_from_user(kname, filename, PATH_MAX);
174                 if (unlikely(len < 0)) {
175                         __putname(kname);
176                         kfree(result);
177                         return ERR_PTR(len);
178                 }
179                 if (unlikely(len == PATH_MAX)) {
180                         __putname(kname);
181                         kfree(result);
182                         return ERR_PTR(-ENAMETOOLONG);
183                 }
184         }
185 
186         result->refcnt = 1;
187         /* The empty path is special. */
188         if (unlikely(!len)) {
189                 if (empty)
190                         *empty = 1;
191                 if (!(flags & LOOKUP_EMPTY)) {
192                         putname(result);
193                         return ERR_PTR(-ENOENT);
194                 }
195         }
196 
197         result->uptr = filename;
198         result->aname = NULL;
199         audit_getname(result);
200         return result;
201 }
202 
203 struct filename *
204 getname(const char __user * filename)
205 {
206         return getname_flags(filename, 0, NULL);
207 }
208 
209 struct filename *
210 getname_kernel(const char * filename)
211 {
212         struct filename *result;
213         int len = strlen(filename) + 1;
214 
215         result = __getname();
216         if (unlikely(!result))
217                 return ERR_PTR(-ENOMEM);
218 
219         if (len <= EMBEDDED_NAME_MAX) {
220                 result->name = (char *)result->iname;
221         } else if (len <= PATH_MAX) {
222                 struct filename *tmp;
223 
224                 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225                 if (unlikely(!tmp)) {
226                         __putname(result);
227                         return ERR_PTR(-ENOMEM);
228                 }
229                 tmp->name = (char *)result;
230                 result = tmp;
231         } else {
232                 __putname(result);
233                 return ERR_PTR(-ENAMETOOLONG);
234         }
235         memcpy((char *)result->name, filename, len);
236         result->uptr = NULL;
237         result->aname = NULL;
238         result->refcnt = 1;
239         audit_getname(result);
240 
241         return result;
242 }
243 
244 void putname(struct filename *name)
245 {
246         BUG_ON(name->refcnt <= 0);
247 
248         if (--name->refcnt > 0)
249                 return;
250 
251         if (name->name != name->iname) {
252                 __putname(name->name);
253                 kfree(name);
254         } else
255                 __putname(name);
256 }
257 
258 static int check_acl(struct inode *inode, int mask)
259 {
260 #ifdef CONFIG_FS_POSIX_ACL
261         struct posix_acl *acl;
262 
263         if (mask & MAY_NOT_BLOCK) {
264                 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
265                 if (!acl)
266                         return -EAGAIN;
267                 /* no ->get_acl() calls in RCU mode... */
268                 if (acl == ACL_NOT_CACHED)
269                         return -ECHILD;
270                 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
271         }
272 
273         acl = get_acl(inode, ACL_TYPE_ACCESS);
274         if (IS_ERR(acl))
275                 return PTR_ERR(acl);
276         if (acl) {
277                 int error = posix_acl_permission(inode, acl, mask);
278                 posix_acl_release(acl);
279                 return error;
280         }
281 #endif
282 
283         return -EAGAIN;
284 }
285 
286 /*
287  * This does the basic permission checking
288  */
289 static int acl_permission_check(struct inode *inode, int mask)
290 {
291         unsigned int mode = inode->i_mode;
292 
293         if (likely(uid_eq(current_fsuid(), inode->i_uid)))
294                 mode >>= 6;
295         else {
296                 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297                         int error = check_acl(inode, mask);
298                         if (error != -EAGAIN)
299                                 return error;
300                 }
301 
302                 if (in_group_p(inode->i_gid))
303                         mode >>= 3;
304         }
305 
306         /*
307          * If the DACs are ok we don't need any capability check.
308          */
309         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
310                 return 0;
311         return -EACCES;
312 }
313 
314 /**
315  * generic_permission -  check for access rights on a Posix-like filesystem
316  * @inode:      inode to check access rights for
317  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
318  *
319  * Used to check for read/write/execute permissions on a file.
320  * We use "fsuid" for this, letting us set arbitrary permissions
321  * for filesystem access without changing the "normal" uids which
322  * are used for other things.
323  *
324  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325  * request cannot be satisfied (eg. requires blocking or too much complexity).
326  * It would then be called again in ref-walk mode.
327  */
328 int generic_permission(struct inode *inode, int mask)
329 {
330         int ret;
331 
332         /*
333          * Do the basic permission checks.
334          */
335         ret = acl_permission_check(inode, mask);
336         if (ret != -EACCES)
337                 return ret;
338 
339         if (S_ISDIR(inode->i_mode)) {
340                 /* DACs are overridable for directories */
341                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
342                         return 0;
343                 if (!(mask & MAY_WRITE))
344                         if (capable_wrt_inode_uidgid(inode,
345                                                      CAP_DAC_READ_SEARCH))
346                                 return 0;
347                 return -EACCES;
348         }
349         /*
350          * Read/write DACs are always overridable.
351          * Executable DACs are overridable when there is
352          * at least one exec bit set.
353          */
354         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
356                         return 0;
357 
358         /*
359          * Searching includes executable on directories, else just read.
360          */
361         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362         if (mask == MAY_READ)
363                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
364                         return 0;
365 
366         return -EACCES;
367 }
368 EXPORT_SYMBOL(generic_permission);
369 
370 /*
371  * We _really_ want to just do "generic_permission()" without
372  * even looking at the inode->i_op values. So we keep a cache
373  * flag in inode->i_opflags, that says "this has not special
374  * permission function, use the fast case".
375  */
376 static inline int do_inode_permission(struct inode *inode, int mask)
377 {
378         if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379                 if (likely(inode->i_op->permission))
380                         return inode->i_op->permission(inode, mask);
381 
382                 /* This gets set once for the inode lifetime */
383                 spin_lock(&inode->i_lock);
384                 inode->i_opflags |= IOP_FASTPERM;
385                 spin_unlock(&inode->i_lock);
386         }
387         return generic_permission(inode, mask);
388 }
389 
390 /**
391  * __inode_permission - Check for access rights to a given inode
392  * @inode: Inode to check permission on
393  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
394  *
395  * Check for read/write/execute permissions on an inode.
396  *
397  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
398  *
399  * This does not check for a read-only file system.  You probably want
400  * inode_permission().
401  */
402 int __inode_permission(struct inode *inode, int mask)
403 {
404         int retval;
405 
406         if (unlikely(mask & MAY_WRITE)) {
407                 /*
408                  * Nobody gets write access to an immutable file.
409                  */
410                 if (IS_IMMUTABLE(inode))
411                         return -EACCES;
412         }
413 
414         retval = do_inode_permission(inode, mask);
415         if (retval)
416                 return retval;
417 
418         retval = devcgroup_inode_permission(inode, mask);
419         if (retval)
420                 return retval;
421 
422         return security_inode_permission(inode, mask);
423 }
424 EXPORT_SYMBOL(__inode_permission);
425 
426 /**
427  * sb_permission - Check superblock-level permissions
428  * @sb: Superblock of inode to check permission on
429  * @inode: Inode to check permission on
430  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
431  *
432  * Separate out file-system wide checks from inode-specific permission checks.
433  */
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
435 {
436         if (unlikely(mask & MAY_WRITE)) {
437                 umode_t mode = inode->i_mode;
438 
439                 /* Nobody gets write access to a read-only fs. */
440                 if ((sb->s_flags & MS_RDONLY) &&
441                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
442                         return -EROFS;
443         }
444         return 0;
445 }
446 
447 /**
448  * inode_permission - Check for access rights to a given inode
449  * @inode: Inode to check permission on
450  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
451  *
452  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
453  * this, letting us set arbitrary permissions for filesystem access without
454  * changing the "normal" UIDs which are used for other things.
455  *
456  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
457  */
458 int inode_permission(struct inode *inode, int mask)
459 {
460         int retval;
461 
462         retval = sb_permission(inode->i_sb, inode, mask);
463         if (retval)
464                 return retval;
465         return __inode_permission(inode, mask);
466 }
467 EXPORT_SYMBOL(inode_permission);
468 
469 /**
470  * path_get - get a reference to a path
471  * @path: path to get the reference to
472  *
473  * Given a path increment the reference count to the dentry and the vfsmount.
474  */
475 void path_get(const struct path *path)
476 {
477         mntget(path->mnt);
478         dget(path->dentry);
479 }
480 EXPORT_SYMBOL(path_get);
481 
482 /**
483  * path_put - put a reference to a path
484  * @path: path to put the reference to
485  *
486  * Given a path decrement the reference count to the dentry and the vfsmount.
487  */
488 void path_put(const struct path *path)
489 {
490         dput(path->dentry);
491         mntput(path->mnt);
492 }
493 EXPORT_SYMBOL(path_put);
494 
495 #define EMBEDDED_LEVELS 2
496 struct nameidata {
497         struct path     path;
498         struct qstr     last;
499         struct path     root;
500         struct inode    *inode; /* path.dentry.d_inode */
501         unsigned int    flags;
502         unsigned        seq, m_seq;
503         int             last_type;
504         unsigned        depth;
505         int             total_link_count;
506         struct saved {
507                 struct path link;
508                 void *cookie;
509                 const char *name;
510                 struct inode *inode;
511                 unsigned seq;
512         } *stack, internal[EMBEDDED_LEVELS];
513         struct filename *name;
514         struct nameidata *saved;
515         unsigned        root_seq;
516         int             dfd;
517 };
518 
519 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
520 {
521         struct nameidata *old = current->nameidata;
522         p->stack = p->internal;
523         p->dfd = dfd;
524         p->name = name;
525         p->total_link_count = old ? old->total_link_count : 0;
526         p->saved = old;
527         current->nameidata = p;
528 }
529 
530 static void restore_nameidata(void)
531 {
532         struct nameidata *now = current->nameidata, *old = now->saved;
533 
534         current->nameidata = old;
535         if (old)
536                 old->total_link_count = now->total_link_count;
537         if (now->stack != now->internal) {
538                 kfree(now->stack);
539                 now->stack = now->internal;
540         }
541 }
542 
543 static int __nd_alloc_stack(struct nameidata *nd)
544 {
545         struct saved *p;
546 
547         if (nd->flags & LOOKUP_RCU) {
548                 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
549                                   GFP_ATOMIC);
550                 if (unlikely(!p))
551                         return -ECHILD;
552         } else {
553                 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
554                                   GFP_KERNEL);
555                 if (unlikely(!p))
556                         return -ENOMEM;
557         }
558         memcpy(p, nd->internal, sizeof(nd->internal));
559         nd->stack = p;
560         return 0;
561 }
562 
563 /**
564  * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
565  * @path: nameidate to verify
566  *
567  * Rename can sometimes move a file or directory outside of a bind
568  * mount, path_connected allows those cases to be detected.
569  */
570 static bool path_connected(const struct path *path)
571 {
572         struct vfsmount *mnt = path->mnt;
573 
574         /* Only bind mounts can have disconnected paths */
575         if (mnt->mnt_root == mnt->mnt_sb->s_root)
576                 return true;
577 
578         return is_subdir(path->dentry, mnt->mnt_root);
579 }
580 
581 static inline int nd_alloc_stack(struct nameidata *nd)
582 {
583         if (likely(nd->depth != EMBEDDED_LEVELS))
584                 return 0;
585         if (likely(nd->stack != nd->internal))
586                 return 0;
587         return __nd_alloc_stack(nd);
588 }
589 
590 static void drop_links(struct nameidata *nd)
591 {
592         int i = nd->depth;
593         while (i--) {
594                 struct saved *last = nd->stack + i;
595                 struct inode *inode = last->inode;
596                 if (last->cookie && inode->i_op->put_link) {
597                         inode->i_op->put_link(inode, last->cookie);
598                         last->cookie = NULL;
599                 }
600         }
601 }
602 
603 static void terminate_walk(struct nameidata *nd)
604 {
605         drop_links(nd);
606         if (!(nd->flags & LOOKUP_RCU)) {
607                 int i;
608                 path_put(&nd->path);
609                 for (i = 0; i < nd->depth; i++)
610                         path_put(&nd->stack[i].link);
611                 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
612                         path_put(&nd->root);
613                         nd->root.mnt = NULL;
614                 }
615         } else {
616                 nd->flags &= ~LOOKUP_RCU;
617                 if (!(nd->flags & LOOKUP_ROOT))
618                         nd->root.mnt = NULL;
619                 rcu_read_unlock();
620         }
621         nd->depth = 0;
622 }
623 
624 /* path_put is needed afterwards regardless of success or failure */
625 static bool legitimize_path(struct nameidata *nd,
626                             struct path *path, unsigned seq)
627 {
628         int res = __legitimize_mnt(path->mnt, nd->m_seq);
629         if (unlikely(res)) {
630                 if (res > 0)
631                         path->mnt = NULL;
632                 path->dentry = NULL;
633                 return false;
634         }
635         if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
636                 path->dentry = NULL;
637                 return false;
638         }
639         return !read_seqcount_retry(&path->dentry->d_seq, seq);
640 }
641 
642 static bool legitimize_links(struct nameidata *nd)
643 {
644         int i;
645         for (i = 0; i < nd->depth; i++) {
646                 struct saved *last = nd->stack + i;
647                 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
648                         drop_links(nd);
649                         nd->depth = i + 1;
650                         return false;
651                 }
652         }
653         return true;
654 }
655 
656 /*
657  * Path walking has 2 modes, rcu-walk and ref-walk (see
658  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
659  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
660  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
661  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
662  * got stuck, so ref-walk may continue from there. If this is not successful
663  * (eg. a seqcount has changed), then failure is returned and it's up to caller
664  * to restart the path walk from the beginning in ref-walk mode.
665  */
666 
667 /**
668  * unlazy_walk - try to switch to ref-walk mode.
669  * @nd: nameidata pathwalk data
670  * @dentry: child of nd->path.dentry or NULL
671  * @seq: seq number to check dentry against
672  * Returns: 0 on success, -ECHILD on failure
673  *
674  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
675  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
676  * @nd or NULL.  Must be called from rcu-walk context.
677  * Nothing should touch nameidata between unlazy_walk() failure and
678  * terminate_walk().
679  */
680 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
681 {
682         struct dentry *parent = nd->path.dentry;
683 
684         BUG_ON(!(nd->flags & LOOKUP_RCU));
685 
686         nd->flags &= ~LOOKUP_RCU;
687         if (unlikely(!legitimize_links(nd)))
688                 goto out2;
689         if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
690                 goto out2;
691         if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
692                 goto out1;
693 
694         /*
695          * For a negative lookup, the lookup sequence point is the parents
696          * sequence point, and it only needs to revalidate the parent dentry.
697          *
698          * For a positive lookup, we need to move both the parent and the
699          * dentry from the RCU domain to be properly refcounted. And the
700          * sequence number in the dentry validates *both* dentry counters,
701          * since we checked the sequence number of the parent after we got
702          * the child sequence number. So we know the parent must still
703          * be valid if the child sequence number is still valid.
704          */
705         if (!dentry) {
706                 if (read_seqcount_retry(&parent->d_seq, nd->seq))
707                         goto out;
708                 BUG_ON(nd->inode != parent->d_inode);
709         } else {
710                 if (!lockref_get_not_dead(&dentry->d_lockref))
711                         goto out;
712                 if (read_seqcount_retry(&dentry->d_seq, seq))
713                         goto drop_dentry;
714         }
715 
716         /*
717          * Sequence counts matched. Now make sure that the root is
718          * still valid and get it if required.
719          */
720         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
721                 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
722                         rcu_read_unlock();
723                         dput(dentry);
724                         return -ECHILD;
725                 }
726         }
727 
728         rcu_read_unlock();
729         return 0;
730 
731 drop_dentry:
732         rcu_read_unlock();
733         dput(dentry);
734         goto drop_root_mnt;
735 out2:
736         nd->path.mnt = NULL;
737 out1:
738         nd->path.dentry = NULL;
739 out:
740         rcu_read_unlock();
741 drop_root_mnt:
742         if (!(nd->flags & LOOKUP_ROOT))
743                 nd->root.mnt = NULL;
744         return -ECHILD;
745 }
746 
747 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
748 {
749         if (unlikely(!legitimize_path(nd, link, seq))) {
750                 drop_links(nd);
751                 nd->depth = 0;
752                 nd->flags &= ~LOOKUP_RCU;
753                 nd->path.mnt = NULL;
754                 nd->path.dentry = NULL;
755                 if (!(nd->flags & LOOKUP_ROOT))
756                         nd->root.mnt = NULL;
757                 rcu_read_unlock();
758         } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
759                 return 0;
760         }
761         path_put(link);
762         return -ECHILD;
763 }
764 
765 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
766 {
767         return dentry->d_op->d_revalidate(dentry, flags);
768 }
769 
770 /**
771  * complete_walk - successful completion of path walk
772  * @nd:  pointer nameidata
773  *
774  * If we had been in RCU mode, drop out of it and legitimize nd->path.
775  * Revalidate the final result, unless we'd already done that during
776  * the path walk or the filesystem doesn't ask for it.  Return 0 on
777  * success, -error on failure.  In case of failure caller does not
778  * need to drop nd->path.
779  */
780 static int complete_walk(struct nameidata *nd)
781 {
782         struct dentry *dentry = nd->path.dentry;
783         int status;
784 
785         if (nd->flags & LOOKUP_RCU) {
786                 if (!(nd->flags & LOOKUP_ROOT))
787                         nd->root.mnt = NULL;
788                 if (unlikely(unlazy_walk(nd, NULL, 0)))
789                         return -ECHILD;
790         }
791 
792         if (likely(!(nd->flags & LOOKUP_JUMPED)))
793                 return 0;
794 
795         if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
796                 return 0;
797 
798         status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
799         if (status > 0)
800                 return 0;
801 
802         if (!status)
803                 status = -ESTALE;
804 
805         return status;
806 }
807 
808 static void set_root(struct nameidata *nd)
809 {
810         get_fs_root(current->fs, &nd->root);
811 }
812 
813 static void set_root_rcu(struct nameidata *nd)
814 {
815         struct fs_struct *fs = current->fs;
816         unsigned seq;
817 
818         do {
819                 seq = read_seqcount_begin(&fs->seq);
820                 nd->root = fs->root;
821                 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
822         } while (read_seqcount_retry(&fs->seq, seq));
823 }
824 
825 static void path_put_conditional(struct path *path, struct nameidata *nd)
826 {
827         dput(path->dentry);
828         if (path->mnt != nd->path.mnt)
829                 mntput(path->mnt);
830 }
831 
832 static inline void path_to_nameidata(const struct path *path,
833                                         struct nameidata *nd)
834 {
835         if (!(nd->flags & LOOKUP_RCU)) {
836                 dput(nd->path.dentry);
837                 if (nd->path.mnt != path->mnt)
838                         mntput(nd->path.mnt);
839         }
840         nd->path.mnt = path->mnt;
841         nd->path.dentry = path->dentry;
842 }
843 
844 /*
845  * Helper to directly jump to a known parsed path from ->follow_link,
846  * caller must have taken a reference to path beforehand.
847  */
848 void nd_jump_link(struct path *path)
849 {
850         struct nameidata *nd = current->nameidata;
851         path_put(&nd->path);
852 
853         nd->path = *path;
854         nd->inode = nd->path.dentry->d_inode;
855         nd->flags |= LOOKUP_JUMPED;
856 }
857 
858 static inline void put_link(struct nameidata *nd)
859 {
860         struct saved *last = nd->stack + --nd->depth;
861         struct inode *inode = last->inode;
862         if (last->cookie && inode->i_op->put_link)
863                 inode->i_op->put_link(inode, last->cookie);
864         if (!(nd->flags & LOOKUP_RCU))
865                 path_put(&last->link);
866 }
867 
868 int sysctl_protected_symlinks __read_mostly = 0;
869 int sysctl_protected_hardlinks __read_mostly = 0;
870 
871 /**
872  * may_follow_link - Check symlink following for unsafe situations
873  * @nd: nameidata pathwalk data
874  *
875  * In the case of the sysctl_protected_symlinks sysctl being enabled,
876  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
877  * in a sticky world-writable directory. This is to protect privileged
878  * processes from failing races against path names that may change out
879  * from under them by way of other users creating malicious symlinks.
880  * It will permit symlinks to be followed only when outside a sticky
881  * world-writable directory, or when the uid of the symlink and follower
882  * match, or when the directory owner matches the symlink's owner.
883  *
884  * Returns 0 if following the symlink is allowed, -ve on error.
885  */
886 static inline int may_follow_link(struct nameidata *nd)
887 {
888         const struct inode *inode;
889         const struct inode *parent;
890 
891         if (!sysctl_protected_symlinks)
892                 return 0;
893 
894         /* Allowed if owner and follower match. */
895         inode = nd->stack[0].inode;
896         if (uid_eq(current_cred()->fsuid, inode->i_uid))
897                 return 0;
898 
899         /* Allowed if parent directory not sticky and world-writable. */
900         parent = nd->inode;
901         if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
902                 return 0;
903 
904         /* Allowed if parent directory and link owner match. */
905         if (uid_eq(parent->i_uid, inode->i_uid))
906                 return 0;
907 
908         if (nd->flags & LOOKUP_RCU)
909                 return -ECHILD;
910 
911         audit_log_link_denied("follow_link", &nd->stack[0].link);
912         return -EACCES;
913 }
914 
915 /**
916  * safe_hardlink_source - Check for safe hardlink conditions
917  * @inode: the source inode to hardlink from
918  *
919  * Return false if at least one of the following conditions:
920  *    - inode is not a regular file
921  *    - inode is setuid
922  *    - inode is setgid and group-exec
923  *    - access failure for read and write
924  *
925  * Otherwise returns true.
926  */
927 static bool safe_hardlink_source(struct inode *inode)
928 {
929         umode_t mode = inode->i_mode;
930 
931         /* Special files should not get pinned to the filesystem. */
932         if (!S_ISREG(mode))
933                 return false;
934 
935         /* Setuid files should not get pinned to the filesystem. */
936         if (mode & S_ISUID)
937                 return false;
938 
939         /* Executable setgid files should not get pinned to the filesystem. */
940         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
941                 return false;
942 
943         /* Hardlinking to unreadable or unwritable sources is dangerous. */
944         if (inode_permission(inode, MAY_READ | MAY_WRITE))
945                 return false;
946 
947         return true;
948 }
949 
950 /**
951  * may_linkat - Check permissions for creating a hardlink
952  * @link: the source to hardlink from
953  *
954  * Block hardlink when all of:
955  *  - sysctl_protected_hardlinks enabled
956  *  - fsuid does not match inode
957  *  - hardlink source is unsafe (see safe_hardlink_source() above)
958  *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
959  *
960  * Returns 0 if successful, -ve on error.
961  */
962 static int may_linkat(struct path *link)
963 {
964         struct inode *inode;
965 
966         if (!sysctl_protected_hardlinks)
967                 return 0;
968 
969         inode = link->dentry->d_inode;
970 
971         /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
972          * otherwise, it must be a safe source.
973          */
974         if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
975                 return 0;
976 
977         audit_log_link_denied("linkat", link);
978         return -EPERM;
979 }
980 
981 static __always_inline
982 const char *get_link(struct nameidata *nd)
983 {
984         struct saved *last = nd->stack + nd->depth - 1;
985         struct dentry *dentry = last->link.dentry;
986         struct inode *inode = last->inode;
987         int error;
988         const char *res;
989 
990         if (!(nd->flags & LOOKUP_RCU)) {
991                 touch_atime(&last->link);
992                 cond_resched();
993         } else if (atime_needs_update(&last->link, inode)) {
994                 if (unlikely(unlazy_walk(nd, NULL, 0)))
995                         return ERR_PTR(-ECHILD);
996                 touch_atime(&last->link);
997         }
998 
999         error = security_inode_follow_link(dentry, inode,
1000                                            nd->flags & LOOKUP_RCU);
1001         if (unlikely(error))
1002                 return ERR_PTR(error);
1003 
1004         nd->last_type = LAST_BIND;
1005         res = inode->i_link;
1006         if (!res) {
1007                 if (nd->flags & LOOKUP_RCU) {
1008                         if (unlikely(unlazy_walk(nd, NULL, 0)))
1009                                 return ERR_PTR(-ECHILD);
1010                 }
1011                 res = inode->i_op->follow_link(dentry, &last->cookie);
1012                 if (IS_ERR_OR_NULL(res)) {
1013                         last->cookie = NULL;
1014                         return res;
1015                 }
1016         }
1017         if (*res == '/') {
1018                 if (nd->flags & LOOKUP_RCU) {
1019                         struct dentry *d;
1020                         if (!nd->root.mnt)
1021                                 set_root_rcu(nd);
1022                         nd->path = nd->root;
1023                         d = nd->path.dentry;
1024                         nd->inode = d->d_inode;
1025                         nd->seq = nd->root_seq;
1026                         if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
1027                                 return ERR_PTR(-ECHILD);
1028                 } else {
1029                         if (!nd->root.mnt)
1030                                 set_root(nd);
1031                         path_put(&nd->path);
1032                         nd->path = nd->root;
1033                         path_get(&nd->root);
1034                         nd->inode = nd->path.dentry->d_inode;
1035                 }
1036                 nd->flags |= LOOKUP_JUMPED;
1037                 while (unlikely(*++res == '/'))
1038                         ;
1039         }
1040         if (!*res)
1041                 res = NULL;
1042         return res;
1043 }
1044 
1045 /*
1046  * follow_up - Find the mountpoint of path's vfsmount
1047  *
1048  * Given a path, find the mountpoint of its source file system.
1049  * Replace @path with the path of the mountpoint in the parent mount.
1050  * Up is towards /.
1051  *
1052  * Return 1 if we went up a level and 0 if we were already at the
1053  * root.
1054  */
1055 int follow_up(struct path *path)
1056 {
1057         struct mount *mnt = real_mount(path->mnt);
1058         struct mount *parent;
1059         struct dentry *mountpoint;
1060 
1061         read_seqlock_excl(&mount_lock);
1062         parent = mnt->mnt_parent;
1063         if (parent == mnt) {
1064                 read_sequnlock_excl(&mount_lock);
1065                 return 0;
1066         }
1067         mntget(&parent->mnt);
1068         mountpoint = dget(mnt->mnt_mountpoint);
1069         read_sequnlock_excl(&mount_lock);
1070         dput(path->dentry);
1071         path->dentry = mountpoint;
1072         mntput(path->mnt);
1073         path->mnt = &parent->mnt;
1074         return 1;
1075 }
1076 EXPORT_SYMBOL(follow_up);
1077 
1078 /*
1079  * Perform an automount
1080  * - return -EISDIR to tell follow_managed() to stop and return the path we
1081  *   were called with.
1082  */
1083 static int follow_automount(struct path *path, struct nameidata *nd,
1084                             bool *need_mntput)
1085 {
1086         struct vfsmount *mnt;
1087         int err;
1088 
1089         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1090                 return -EREMOTE;
1091 
1092         /* We don't want to mount if someone's just doing a stat -
1093          * unless they're stat'ing a directory and appended a '/' to
1094          * the name.
1095          *
1096          * We do, however, want to mount if someone wants to open or
1097          * create a file of any type under the mountpoint, wants to
1098          * traverse through the mountpoint or wants to open the
1099          * mounted directory.  Also, autofs may mark negative dentries
1100          * as being automount points.  These will need the attentions
1101          * of the daemon to instantiate them before they can be used.
1102          */
1103         if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1104                            LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1105             path->dentry->d_inode)
1106                 return -EISDIR;
1107 
1108         nd->total_link_count++;
1109         if (nd->total_link_count >= 40)
1110                 return -ELOOP;
1111 
1112         mnt = path->dentry->d_op->d_automount(path);
1113         if (IS_ERR(mnt)) {
1114                 /*
1115                  * The filesystem is allowed to return -EISDIR here to indicate
1116                  * it doesn't want to automount.  For instance, autofs would do
1117                  * this so that its userspace daemon can mount on this dentry.
1118                  *
1119                  * However, we can only permit this if it's a terminal point in
1120                  * the path being looked up; if it wasn't then the remainder of
1121                  * the path is inaccessible and we should say so.
1122                  */
1123                 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1124                         return -EREMOTE;
1125                 return PTR_ERR(mnt);
1126         }
1127 
1128         if (!mnt) /* mount collision */
1129                 return 0;
1130 
1131         if (!*need_mntput) {
1132                 /* lock_mount() may release path->mnt on error */
1133                 mntget(path->mnt);
1134                 *need_mntput = true;
1135         }
1136         err = finish_automount(mnt, path);
1137 
1138         switch (err) {
1139         case -EBUSY:
1140                 /* Someone else made a mount here whilst we were busy */
1141                 return 0;
1142         case 0:
1143                 path_put(path);
1144                 path->mnt = mnt;
1145                 path->dentry = dget(mnt->mnt_root);
1146                 return 0;
1147         default:
1148                 return err;
1149         }
1150 
1151 }
1152 
1153 /*
1154  * Handle a dentry that is managed in some way.
1155  * - Flagged for transit management (autofs)
1156  * - Flagged as mountpoint
1157  * - Flagged as automount point
1158  *
1159  * This may only be called in refwalk mode.
1160  *
1161  * Serialization is taken care of in namespace.c
1162  */
1163 static int follow_managed(struct path *path, struct nameidata *nd)
1164 {
1165         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1166         unsigned managed;
1167         bool need_mntput = false;
1168         int ret = 0;
1169 
1170         /* Given that we're not holding a lock here, we retain the value in a
1171          * local variable for each dentry as we look at it so that we don't see
1172          * the components of that value change under us */
1173         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1174                managed &= DCACHE_MANAGED_DENTRY,
1175                unlikely(managed != 0)) {
1176                 /* Allow the filesystem to manage the transit without i_mutex
1177                  * being held. */
1178                 if (managed & DCACHE_MANAGE_TRANSIT) {
1179                         BUG_ON(!path->dentry->d_op);
1180                         BUG_ON(!path->dentry->d_op->d_manage);
1181                         ret = path->dentry->d_op->d_manage(path->dentry, false);
1182                         if (ret < 0)
1183                                 break;
1184                 }
1185 
1186                 /* Transit to a mounted filesystem. */
1187                 if (managed & DCACHE_MOUNTED) {
1188                         struct vfsmount *mounted = lookup_mnt(path);
1189                         if (mounted) {
1190                                 dput(path->dentry);
1191                                 if (need_mntput)
1192                                         mntput(path->mnt);
1193                                 path->mnt = mounted;
1194                                 path->dentry = dget(mounted->mnt_root);
1195                                 need_mntput = true;
1196                                 continue;
1197                         }
1198 
1199                         /* Something is mounted on this dentry in another
1200                          * namespace and/or whatever was mounted there in this
1201                          * namespace got unmounted before lookup_mnt() could
1202                          * get it */
1203                 }
1204 
1205                 /* Handle an automount point */
1206                 if (managed & DCACHE_NEED_AUTOMOUNT) {
1207                         ret = follow_automount(path, nd, &need_mntput);
1208                         if (ret < 0)
1209                                 break;
1210                         continue;
1211                 }
1212 
1213                 /* We didn't change the current path point */
1214                 break;
1215         }
1216 
1217         if (need_mntput && path->mnt == mnt)
1218                 mntput(path->mnt);
1219         if (ret == -EISDIR)
1220                 ret = 0;
1221         if (need_mntput)
1222                 nd->flags |= LOOKUP_JUMPED;
1223         if (unlikely(ret < 0))
1224                 path_put_conditional(path, nd);
1225         return ret;
1226 }
1227 
1228 int follow_down_one(struct path *path)
1229 {
1230         struct vfsmount *mounted;
1231 
1232         mounted = lookup_mnt(path);
1233         if (mounted) {
1234                 dput(path->dentry);
1235                 mntput(path->mnt);
1236                 path->mnt = mounted;
1237                 path->dentry = dget(mounted->mnt_root);
1238                 return 1;
1239         }
1240         return 0;
1241 }
1242 EXPORT_SYMBOL(follow_down_one);
1243 
1244 static inline int managed_dentry_rcu(struct dentry *dentry)
1245 {
1246         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1247                 dentry->d_op->d_manage(dentry, true) : 0;
1248 }
1249 
1250 /*
1251  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1252  * we meet a managed dentry that would need blocking.
1253  */
1254 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1255                                struct inode **inode, unsigned *seqp)
1256 {
1257         for (;;) {
1258                 struct mount *mounted;
1259                 /*
1260                  * Don't forget we might have a non-mountpoint managed dentry
1261                  * that wants to block transit.
1262                  */
1263                 switch (managed_dentry_rcu(path->dentry)) {
1264                 case -ECHILD:
1265                 default:
1266                         return false;
1267                 case -EISDIR:
1268                         return true;
1269                 case 0:
1270                         break;
1271                 }
1272 
1273                 if (!d_mountpoint(path->dentry))
1274                         return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1275 
1276                 mounted = __lookup_mnt(path->mnt, path->dentry);
1277                 if (!mounted)
1278                         break;
1279                 path->mnt = &mounted->mnt;
1280                 path->dentry = mounted->mnt.mnt_root;
1281                 nd->flags |= LOOKUP_JUMPED;
1282                 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1283                 /*
1284                  * Update the inode too. We don't need to re-check the
1285                  * dentry sequence number here after this d_inode read,
1286                  * because a mount-point is always pinned.
1287                  */
1288                 *inode = path->dentry->d_inode;
1289         }
1290         return !read_seqretry(&mount_lock, nd->m_seq) &&
1291                 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1292 }
1293 
1294 static int follow_dotdot_rcu(struct nameidata *nd)
1295 {
1296         struct inode *inode = nd->inode;
1297         if (!nd->root.mnt)
1298                 set_root_rcu(nd);
1299 
1300         while (1) {
1301                 if (path_equal(&nd->path, &nd->root))
1302                         break;
1303                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1304                         struct dentry *old = nd->path.dentry;
1305                         struct dentry *parent = old->d_parent;
1306                         unsigned seq;
1307 
1308                         inode = parent->d_inode;
1309                         seq = read_seqcount_begin(&parent->d_seq);
1310                         if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1311                                 return -ECHILD;
1312                         nd->path.dentry = parent;
1313                         nd->seq = seq;
1314                         if (unlikely(!path_connected(&nd->path)))
1315                                 return -ENOENT;
1316                         break;
1317                 } else {
1318                         struct mount *mnt = real_mount(nd->path.mnt);
1319                         struct mount *mparent = mnt->mnt_parent;
1320                         struct dentry *mountpoint = mnt->mnt_mountpoint;
1321                         struct inode *inode2 = mountpoint->d_inode;
1322                         unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1323                         if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1324                                 return -ECHILD;
1325                         if (&mparent->mnt == nd->path.mnt)
1326                                 break;
1327                         /* we know that mountpoint was pinned */
1328                         nd->path.dentry = mountpoint;
1329                         nd->path.mnt = &mparent->mnt;
1330                         inode = inode2;
1331                         nd->seq = seq;
1332                 }
1333         }
1334         while (unlikely(d_mountpoint(nd->path.dentry))) {
1335                 struct mount *mounted;
1336                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1337                 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1338                         return -ECHILD;
1339                 if (!mounted)
1340                         break;
1341                 nd->path.mnt = &mounted->mnt;
1342                 nd->path.dentry = mounted->mnt.mnt_root;
1343                 inode = nd->path.dentry->d_inode;
1344                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1345         }
1346         nd->inode = inode;
1347         return 0;
1348 }
1349 
1350 /*
1351  * Follow down to the covering mount currently visible to userspace.  At each
1352  * point, the filesystem owning that dentry may be queried as to whether the
1353  * caller is permitted to proceed or not.
1354  */
1355 int follow_down(struct path *path)
1356 {
1357         unsigned managed;
1358         int ret;
1359 
1360         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1361                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1362                 /* Allow the filesystem to manage the transit without i_mutex
1363                  * being held.
1364                  *
1365                  * We indicate to the filesystem if someone is trying to mount
1366                  * something here.  This gives autofs the chance to deny anyone
1367                  * other than its daemon the right to mount on its
1368                  * superstructure.
1369                  *
1370                  * The filesystem may sleep at this point.
1371                  */
1372                 if (managed & DCACHE_MANAGE_TRANSIT) {
1373                         BUG_ON(!path->dentry->d_op);
1374                         BUG_ON(!path->dentry->d_op->d_manage);
1375                         ret = path->dentry->d_op->d_manage(
1376                                 path->dentry, false);
1377                         if (ret < 0)
1378                                 return ret == -EISDIR ? 0 : ret;
1379                 }
1380 
1381                 /* Transit to a mounted filesystem. */
1382                 if (managed & DCACHE_MOUNTED) {
1383                         struct vfsmount *mounted = lookup_mnt(path);
1384                         if (!mounted)
1385                                 break;
1386                         dput(path->dentry);
1387                         mntput(path->mnt);
1388                         path->mnt = mounted;
1389                         path->dentry = dget(mounted->mnt_root);
1390                         continue;
1391                 }
1392 
1393                 /* Don't handle automount points here */
1394                 break;
1395         }
1396         return 0;
1397 }
1398 EXPORT_SYMBOL(follow_down);
1399 
1400 /*
1401  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1402  */
1403 static void follow_mount(struct path *path)
1404 {
1405         while (d_mountpoint(path->dentry)) {
1406                 struct vfsmount *mounted = lookup_mnt(path);
1407                 if (!mounted)
1408                         break;
1409                 dput(path->dentry);
1410                 mntput(path->mnt);
1411                 path->mnt = mounted;
1412                 path->dentry = dget(mounted->mnt_root);
1413         }
1414 }
1415 
1416 static int follow_dotdot(struct nameidata *nd)
1417 {
1418         if (!nd->root.mnt)
1419                 set_root(nd);
1420 
1421         while(1) {
1422                 struct dentry *old = nd->path.dentry;
1423 
1424                 if (nd->path.dentry == nd->root.dentry &&
1425                     nd->path.mnt == nd->root.mnt) {
1426                         break;
1427                 }
1428                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1429                         /* rare case of legitimate dget_parent()... */
1430                         nd->path.dentry = dget_parent(nd->path.dentry);
1431                         dput(old);
1432                         if (unlikely(!path_connected(&nd->path)))
1433                                 return -ENOENT;
1434                         break;
1435                 }
1436                 if (!follow_up(&nd->path))
1437                         break;
1438         }
1439         follow_mount(&nd->path);
1440         nd->inode = nd->path.dentry->d_inode;
1441         return 0;
1442 }
1443 
1444 /*
1445  * This looks up the name in dcache, possibly revalidates the old dentry and
1446  * allocates a new one if not found or not valid.  In the need_lookup argument
1447  * returns whether i_op->lookup is necessary.
1448  *
1449  * dir->d_inode->i_mutex must be held
1450  */
1451 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1452                                     unsigned int flags, bool *need_lookup)
1453 {
1454         struct dentry *dentry;
1455         int error;
1456 
1457         *need_lookup = false;
1458         dentry = d_lookup(dir, name);
1459         if (dentry) {
1460                 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1461                         error = d_revalidate(dentry, flags);
1462                         if (unlikely(error <= 0)) {
1463                                 if (error < 0) {
1464                                         dput(dentry);
1465                                         return ERR_PTR(error);
1466                                 } else {
1467                                         d_invalidate(dentry);
1468                                         dput(dentry);
1469                                         dentry = NULL;
1470                                 }
1471                         }
1472                 }
1473         }
1474 
1475         if (!dentry) {
1476                 dentry = d_alloc(dir, name);
1477                 if (unlikely(!dentry))
1478                         return ERR_PTR(-ENOMEM);
1479 
1480                 *need_lookup = true;
1481         }
1482         return dentry;
1483 }
1484 
1485 /*
1486  * Call i_op->lookup on the dentry.  The dentry must be negative and
1487  * unhashed.
1488  *
1489  * dir->d_inode->i_mutex must be held
1490  */
1491 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1492                                   unsigned int flags)
1493 {
1494         struct dentry *old;
1495 
1496         /* Don't create child dentry for a dead directory. */
1497         if (unlikely(IS_DEADDIR(dir))) {
1498                 dput(dentry);
1499                 return ERR_PTR(-ENOENT);
1500         }
1501 
1502         old = dir->i_op->lookup(dir, dentry, flags);
1503         if (unlikely(old)) {
1504                 dput(dentry);
1505                 dentry = old;
1506         }
1507         return dentry;
1508 }
1509 
1510 static struct dentry *__lookup_hash(struct qstr *name,
1511                 struct dentry *base, unsigned int flags)
1512 {
1513         bool need_lookup;
1514         struct dentry *dentry;
1515 
1516         dentry = lookup_dcache(name, base, flags, &need_lookup);
1517         if (!need_lookup)
1518                 return dentry;
1519 
1520         return lookup_real(base->d_inode, dentry, flags);
1521 }
1522 
1523 /*
1524  *  It's more convoluted than I'd like it to be, but... it's still fairly
1525  *  small and for now I'd prefer to have fast path as straight as possible.
1526  *  It _is_ time-critical.
1527  */
1528 static int lookup_fast(struct nameidata *nd,
1529                        struct path *path, struct inode **inode,
1530                        unsigned *seqp)
1531 {
1532         struct vfsmount *mnt = nd->path.mnt;
1533         struct dentry *dentry, *parent = nd->path.dentry;
1534         int need_reval = 1;
1535         int status = 1;
1536         int err;
1537 
1538         /*
1539          * Rename seqlock is not required here because in the off chance
1540          * of a false negative due to a concurrent rename, we're going to
1541          * do the non-racy lookup, below.
1542          */
1543         if (nd->flags & LOOKUP_RCU) {
1544                 unsigned seq;
1545                 bool negative;
1546                 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1547                 if (!dentry)
1548                         goto unlazy;
1549 
1550                 /*
1551                  * This sequence count validates that the inode matches
1552                  * the dentry name information from lookup.
1553                  */
1554                 *inode = d_backing_inode(dentry);
1555                 negative = d_is_negative(dentry);
1556                 if (read_seqcount_retry(&dentry->d_seq, seq))
1557                         return -ECHILD;
1558 
1559                 /*
1560                  * This sequence count validates that the parent had no
1561                  * changes while we did the lookup of the dentry above.
1562                  *
1563                  * The memory barrier in read_seqcount_begin of child is
1564                  *  enough, we can use __read_seqcount_retry here.
1565                  */
1566                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1567                         return -ECHILD;
1568 
1569                 *seqp = seq;
1570                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1571                         status = d_revalidate(dentry, nd->flags);
1572                         if (unlikely(status <= 0)) {
1573                                 if (status != -ECHILD)
1574                                         need_reval = 0;
1575                                 goto unlazy;
1576                         }
1577                 }
1578                 /*
1579                  * Note: do negative dentry check after revalidation in
1580                  * case that drops it.
1581                  */
1582                 if (negative)
1583                         return -ENOENT;
1584                 path->mnt = mnt;
1585                 path->dentry = dentry;
1586                 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1587                         return 0;
1588 unlazy:
1589                 if (unlazy_walk(nd, dentry, seq))
1590                         return -ECHILD;
1591         } else {
1592                 dentry = __d_lookup(parent, &nd->last);
1593         }
1594 
1595         if (unlikely(!dentry))
1596                 goto need_lookup;
1597 
1598         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1599                 status = d_revalidate(dentry, nd->flags);
1600         if (unlikely(status <= 0)) {
1601                 if (status < 0) {
1602                         dput(dentry);
1603                         return status;
1604                 }
1605                 d_invalidate(dentry);
1606                 dput(dentry);
1607                 goto need_lookup;
1608         }
1609 
1610         if (unlikely(d_is_negative(dentry))) {
1611                 dput(dentry);
1612                 return -ENOENT;
1613         }
1614         path->mnt = mnt;
1615         path->dentry = dentry;
1616         err = follow_managed(path, nd);
1617         if (likely(!err))
1618                 *inode = d_backing_inode(path->dentry);
1619         return err;
1620 
1621 need_lookup:
1622         return 1;
1623 }
1624 
1625 /* Fast lookup failed, do it the slow way */
1626 static int lookup_slow(struct nameidata *nd, struct path *path)
1627 {
1628         struct dentry *dentry, *parent;
1629 
1630         parent = nd->path.dentry;
1631         BUG_ON(nd->inode != parent->d_inode);
1632 
1633         mutex_lock(&parent->d_inode->i_mutex);
1634         dentry = __lookup_hash(&nd->last, parent, nd->flags);
1635         mutex_unlock(&parent->d_inode->i_mutex);
1636         if (IS_ERR(dentry))
1637                 return PTR_ERR(dentry);
1638         path->mnt = nd->path.mnt;
1639         path->dentry = dentry;
1640         return follow_managed(path, nd);
1641 }
1642 
1643 static inline int may_lookup(struct nameidata *nd)
1644 {
1645         if (nd->flags & LOOKUP_RCU) {
1646                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1647                 if (err != -ECHILD)
1648                         return err;
1649                 if (unlazy_walk(nd, NULL, 0))
1650                         return -ECHILD;
1651         }
1652         return inode_permission(nd->inode, MAY_EXEC);
1653 }
1654 
1655 static inline int handle_dots(struct nameidata *nd, int type)
1656 {
1657         if (type == LAST_DOTDOT) {
1658                 if (nd->flags & LOOKUP_RCU) {
1659                         return follow_dotdot_rcu(nd);
1660                 } else
1661                         return follow_dotdot(nd);
1662         }
1663         return 0;
1664 }
1665 
1666 static int pick_link(struct nameidata *nd, struct path *link,
1667                      struct inode *inode, unsigned seq)
1668 {
1669         int error;
1670         struct saved *last;
1671         if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1672                 path_to_nameidata(link, nd);
1673                 return -ELOOP;
1674         }
1675         if (!(nd->flags & LOOKUP_RCU)) {
1676                 if (link->mnt == nd->path.mnt)
1677                         mntget(link->mnt);
1678         }
1679         error = nd_alloc_stack(nd);
1680         if (unlikely(error)) {
1681                 if (error == -ECHILD) {
1682                         if (unlikely(unlazy_link(nd, link, seq)))
1683                                 return -ECHILD;
1684                         error = nd_alloc_stack(nd);
1685                 }
1686                 if (error) {
1687                         path_put(link);
1688                         return error;
1689                 }
1690         }
1691 
1692         last = nd->stack + nd->depth++;
1693         last->link = *link;
1694         last->cookie = NULL;
1695         last->inode = inode;
1696         last->seq = seq;
1697         return 1;
1698 }
1699 
1700 /*
1701  * Do we need to follow links? We _really_ want to be able
1702  * to do this check without having to look at inode->i_op,
1703  * so we keep a cache of "no, this doesn't need follow_link"
1704  * for the common case.
1705  */
1706 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1707                                      int follow,
1708                                      struct inode *inode, unsigned seq)
1709 {
1710         if (likely(!d_is_symlink(link->dentry)))
1711                 return 0;
1712         if (!follow)
1713                 return 0;
1714         return pick_link(nd, link, inode, seq);
1715 }
1716 
1717 enum {WALK_GET = 1, WALK_PUT = 2};
1718 
1719 static int walk_component(struct nameidata *nd, int flags)
1720 {
1721         struct path path;
1722         struct inode *inode;
1723         unsigned seq;
1724         int err;
1725         /*
1726          * "." and ".." are special - ".." especially so because it has
1727          * to be able to know about the current root directory and
1728          * parent relationships.
1729          */
1730         if (unlikely(nd->last_type != LAST_NORM)) {
1731                 err = handle_dots(nd, nd->last_type);
1732                 if (flags & WALK_PUT)
1733                         put_link(nd);
1734                 return err;
1735         }
1736         err = lookup_fast(nd, &path, &inode, &seq);
1737         if (unlikely(err)) {
1738                 if (err < 0)
1739                         return err;
1740 
1741                 err = lookup_slow(nd, &path);
1742                 if (err < 0)
1743                         return err;
1744 
1745                 inode = d_backing_inode(path.dentry);
1746                 seq = 0;        /* we are already out of RCU mode */
1747                 err = -ENOENT;
1748                 if (d_is_negative(path.dentry))
1749                         goto out_path_put;
1750         }
1751 
1752         if (flags & WALK_PUT)
1753                 put_link(nd);
1754         err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1755         if (unlikely(err))
1756                 return err;
1757         path_to_nameidata(&path, nd);
1758         nd->inode = inode;
1759         nd->seq = seq;
1760         return 0;
1761 
1762 out_path_put:
1763         path_to_nameidata(&path, nd);
1764         return err;
1765 }
1766 
1767 /*
1768  * We can do the critical dentry name comparison and hashing
1769  * operations one word at a time, but we are limited to:
1770  *
1771  * - Architectures with fast unaligned word accesses. We could
1772  *   do a "get_unaligned()" if this helps and is sufficiently
1773  *   fast.
1774  *
1775  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1776  *   do not trap on the (extremely unlikely) case of a page
1777  *   crossing operation.
1778  *
1779  * - Furthermore, we need an efficient 64-bit compile for the
1780  *   64-bit case in order to generate the "number of bytes in
1781  *   the final mask". Again, that could be replaced with a
1782  *   efficient population count instruction or similar.
1783  */
1784 #ifdef CONFIG_DCACHE_WORD_ACCESS
1785 
1786 #include <asm/word-at-a-time.h>
1787 
1788 #ifdef CONFIG_64BIT
1789 
1790 static inline unsigned int fold_hash(unsigned long hash)
1791 {
1792         return hash_64(hash, 32);
1793 }
1794 
1795 #else   /* 32-bit case */
1796 
1797 #define fold_hash(x) (x)
1798 
1799 #endif
1800 
1801 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1802 {
1803         unsigned long a, mask;
1804         unsigned long hash = 0;
1805 
1806         for (;;) {
1807                 a = load_unaligned_zeropad(name);
1808                 if (len < sizeof(unsigned long))
1809                         break;
1810                 hash += a;
1811                 hash *= 9;
1812                 name += sizeof(unsigned long);
1813                 len -= sizeof(unsigned long);
1814                 if (!len)
1815                         goto done;
1816         }
1817         mask = bytemask_from_count(len);
1818         hash += mask & a;
1819 done:
1820         return fold_hash(hash);
1821 }
1822 EXPORT_SYMBOL(full_name_hash);
1823 
1824 /*
1825  * Calculate the length and hash of the path component, and
1826  * return the "hash_len" as the result.
1827  */
1828 static inline u64 hash_name(const char *name)
1829 {
1830         unsigned long a, b, adata, bdata, mask, hash, len;
1831         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1832 
1833         hash = a = 0;
1834         len = -sizeof(unsigned long);
1835         do {
1836                 hash = (hash + a) * 9;
1837                 len += sizeof(unsigned long);
1838                 a = load_unaligned_zeropad(name+len);
1839                 b = a ^ REPEAT_BYTE('/');
1840         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1841 
1842         adata = prep_zero_mask(a, adata, &constants);
1843         bdata = prep_zero_mask(b, bdata, &constants);
1844 
1845         mask = create_zero_mask(adata | bdata);
1846 
1847         hash += a & zero_bytemask(mask);
1848         len += find_zero(mask);
1849         return hashlen_create(fold_hash(hash), len);
1850 }
1851 
1852 #else
1853 
1854 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1855 {
1856         unsigned long hash = init_name_hash();
1857         while (len--)
1858                 hash = partial_name_hash(*name++, hash);
1859         return end_name_hash(hash);
1860 }
1861 EXPORT_SYMBOL(full_name_hash);
1862 
1863 /*
1864  * We know there's a real path component here of at least
1865  * one character.
1866  */
1867 static inline u64 hash_name(const char *name)
1868 {
1869         unsigned long hash = init_name_hash();
1870         unsigned long len = 0, c;
1871 
1872         c = (unsigned char)*name;
1873         do {
1874                 len++;
1875                 hash = partial_name_hash(c, hash);
1876                 c = (unsigned char)name[len];
1877         } while (c && c != '/');
1878         return hashlen_create(end_name_hash(hash), len);
1879 }
1880 
1881 #endif
1882 
1883 /*
1884  * Name resolution.
1885  * This is the basic name resolution function, turning a pathname into
1886  * the final dentry. We expect 'base' to be positive and a directory.
1887  *
1888  * Returns 0 and nd will have valid dentry and mnt on success.
1889  * Returns error and drops reference to input namei data on failure.
1890  */
1891 static int link_path_walk(const char *name, struct nameidata *nd)
1892 {
1893         int err;
1894 
1895         while (*name=='/')
1896                 name++;
1897         if (!*name)
1898                 return 0;
1899 
1900         /* At this point we know we have a real path component. */
1901         for(;;) {
1902                 u64 hash_len;
1903                 int type;
1904 
1905                 err = may_lookup(nd);
1906                 if (err)
1907                         return err;
1908 
1909                 hash_len = hash_name(name);
1910 
1911                 type = LAST_NORM;
1912                 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1913                         case 2:
1914                                 if (name[1] == '.') {
1915                                         type = LAST_DOTDOT;
1916                                         nd->flags |= LOOKUP_JUMPED;
1917                                 }
1918                                 break;
1919                         case 1:
1920                                 type = LAST_DOT;
1921                 }
1922                 if (likely(type == LAST_NORM)) {
1923                         struct dentry *parent = nd->path.dentry;
1924                         nd->flags &= ~LOOKUP_JUMPED;
1925                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1926                                 struct qstr this = { { .hash_len = hash_len }, .name = name };
1927                                 err = parent->d_op->d_hash(parent, &this);
1928                                 if (err < 0)
1929                                         return err;
1930                                 hash_len = this.hash_len;
1931                                 name = this.name;
1932                         }
1933                 }
1934 
1935                 nd->last.hash_len = hash_len;
1936                 nd->last.name = name;
1937                 nd->last_type = type;
1938 
1939                 name += hashlen_len(hash_len);
1940                 if (!*name)
1941                         goto OK;
1942                 /*
1943                  * If it wasn't NUL, we know it was '/'. Skip that
1944                  * slash, and continue until no more slashes.
1945                  */
1946                 do {
1947                         name++;
1948                 } while (unlikely(*name == '/'));
1949                 if (unlikely(!*name)) {
1950 OK:
1951                         /* pathname body, done */
1952                         if (!nd->depth)
1953                                 return 0;
1954                         name = nd->stack[nd->depth - 1].name;
1955                         /* trailing symlink, done */
1956                         if (!name)
1957                                 return 0;
1958                         /* last component of nested symlink */
1959                         err = walk_component(nd, WALK_GET | WALK_PUT);
1960                 } else {
1961                         err = walk_component(nd, WALK_GET);
1962                 }
1963                 if (err < 0)
1964                         return err;
1965 
1966                 if (err) {
1967                         const char *s = get_link(nd);
1968 
1969                         if (IS_ERR(s))
1970                                 return PTR_ERR(s);
1971                         err = 0;
1972                         if (unlikely(!s)) {
1973                                 /* jumped */
1974                                 put_link(nd);
1975                         } else {
1976                                 nd->stack[nd->depth - 1].name = name;
1977                                 name = s;
1978                                 continue;
1979                         }
1980                 }
1981                 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1982                         if (nd->flags & LOOKUP_RCU) {
1983                                 if (unlazy_walk(nd, NULL, 0))
1984                                         return -ECHILD;
1985                         }
1986                         return -ENOTDIR;
1987                 }
1988         }
1989 }
1990 
1991 static const char *path_init(struct nameidata *nd, unsigned flags)
1992 {
1993         int retval = 0;
1994         const char *s = nd->name->name;
1995 
1996         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1997         nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1998         nd->depth = 0;
1999         if (flags & LOOKUP_ROOT) {
2000                 struct dentry *root = nd->root.dentry;
2001                 struct inode *inode = root->d_inode;
2002                 if (*s) {
2003                         if (!d_can_lookup(root))
2004                                 return ERR_PTR(-ENOTDIR);
2005                         retval = inode_permission(inode, MAY_EXEC);
2006                         if (retval)
2007                                 return ERR_PTR(retval);
2008                 }
2009                 nd->path = nd->root;
2010                 nd->inode = inode;
2011                 if (flags & LOOKUP_RCU) {
2012                         rcu_read_lock();
2013                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2014                         nd->root_seq = nd->seq;
2015                         nd->m_seq = read_seqbegin(&mount_lock);
2016                 } else {
2017                         path_get(&nd->path);
2018                 }
2019                 return s;
2020         }
2021 
2022         nd->root.mnt = NULL;
2023 
2024         nd->m_seq = read_seqbegin(&mount_lock);
2025         if (*s == '/') {
2026                 if (flags & LOOKUP_RCU) {
2027                         rcu_read_lock();
2028                         set_root_rcu(nd);
2029                         nd->seq = nd->root_seq;
2030                 } else {
2031                         set_root(nd);
2032                         path_get(&nd->root);
2033                 }
2034                 nd->path = nd->root;
2035         } else if (nd->dfd == AT_FDCWD) {
2036                 if (flags & LOOKUP_RCU) {
2037                         struct fs_struct *fs = current->fs;
2038                         unsigned seq;
2039 
2040                         rcu_read_lock();
2041 
2042                         do {
2043                                 seq = read_seqcount_begin(&fs->seq);
2044                                 nd->path = fs->pwd;
2045                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2046                         } while (read_seqcount_retry(&fs->seq, seq));
2047                 } else {
2048                         get_fs_pwd(current->fs, &nd->path);
2049                 }
2050         } else {
2051                 /* Caller must check execute permissions on the starting path component */
2052                 struct fd f = fdget_raw(nd->dfd);
2053                 struct dentry *dentry;
2054 
2055                 if (!f.file)
2056                         return ERR_PTR(-EBADF);
2057 
2058                 dentry = f.file->f_path.dentry;
2059 
2060                 if (*s) {
2061                         if (!d_can_lookup(dentry)) {
2062                                 fdput(f);
2063                                 return ERR_PTR(-ENOTDIR);
2064                         }
2065                 }
2066 
2067                 nd->path = f.file->f_path;
2068                 if (flags & LOOKUP_RCU) {
2069                         rcu_read_lock();
2070                         nd->inode = nd->path.dentry->d_inode;
2071                         nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2072                 } else {
2073                         path_get(&nd->path);
2074                         nd->inode = nd->path.dentry->d_inode;
2075                 }
2076                 fdput(f);
2077                 return s;
2078         }
2079 
2080         nd->inode = nd->path.dentry->d_inode;
2081         if (!(flags & LOOKUP_RCU))
2082                 return s;
2083         if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2084                 return s;
2085         if (!(nd->flags & LOOKUP_ROOT))
2086                 nd->root.mnt = NULL;
2087         rcu_read_unlock();
2088         return ERR_PTR(-ECHILD);
2089 }
2090 
2091 static const char *trailing_symlink(struct nameidata *nd)
2092 {
2093         const char *s;
2094         int error = may_follow_link(nd);
2095         if (unlikely(error))
2096                 return ERR_PTR(error);
2097         nd->flags |= LOOKUP_PARENT;
2098         nd->stack[0].name = NULL;
2099         s = get_link(nd);
2100         return s ? s : "";
2101 }
2102 
2103 static inline int lookup_last(struct nameidata *nd)
2104 {
2105         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2106                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2107 
2108         nd->flags &= ~LOOKUP_PARENT;
2109         return walk_component(nd,
2110                         nd->flags & LOOKUP_FOLLOW
2111                                 ? nd->depth
2112                                         ? WALK_PUT | WALK_GET
2113                                         : WALK_GET
2114                                 : 0);
2115 }
2116 
2117 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2118 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2119 {
2120         const char *s = path_init(nd, flags);
2121         int err;
2122 
2123         if (IS_ERR(s))
2124                 return PTR_ERR(s);
2125         while (!(err = link_path_walk(s, nd))
2126                 && ((err = lookup_last(nd)) > 0)) {
2127                 s = trailing_symlink(nd);
2128                 if (IS_ERR(s)) {
2129                         err = PTR_ERR(s);
2130                         break;
2131                 }
2132         }
2133         if (!err)
2134                 err = complete_walk(nd);
2135 
2136         if (!err && nd->flags & LOOKUP_DIRECTORY)
2137                 if (!d_can_lookup(nd->path.dentry))
2138                         err = -ENOTDIR;
2139         if (!err) {
2140                 *path = nd->path;
2141                 nd->path.mnt = NULL;
2142                 nd->path.dentry = NULL;
2143         }
2144         terminate_walk(nd);
2145         return err;
2146 }
2147 
2148 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2149                            struct path *path, struct path *root)
2150 {
2151         int retval;
2152         struct nameidata nd;
2153         if (IS_ERR(name))
2154                 return PTR_ERR(name);
2155         if (unlikely(root)) {
2156                 nd.root = *root;
2157                 flags |= LOOKUP_ROOT;
2158         }
2159         set_nameidata(&nd, dfd, name);
2160         retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2161         if (unlikely(retval == -ECHILD))
2162                 retval = path_lookupat(&nd, flags, path);
2163         if (unlikely(retval == -ESTALE))
2164                 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2165 
2166         if (likely(!retval))
2167                 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2168         restore_nameidata();
2169         putname(name);
2170         return retval;
2171 }
2172 
2173 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2174 static int path_parentat(struct nameidata *nd, unsigned flags,
2175                                 struct path *parent)
2176 {
2177         const char *s = path_init(nd, flags);
2178         int err;
2179         if (IS_ERR(s))
2180                 return PTR_ERR(s);
2181         err = link_path_walk(s, nd);
2182         if (!err)
2183                 err = complete_walk(nd);
2184         if (!err) {
2185                 *parent = nd->path;
2186                 nd->path.mnt = NULL;
2187                 nd->path.dentry = NULL;
2188         }
2189         terminate_walk(nd);
2190         return err;
2191 }
2192 
2193 static struct filename *filename_parentat(int dfd, struct filename *name,
2194                                 unsigned int flags, struct path *parent,
2195                                 struct qstr *last, int *type)
2196 {
2197         int retval;
2198         struct nameidata nd;
2199 
2200         if (IS_ERR(name))
2201                 return name;
2202         set_nameidata(&nd, dfd, name);
2203         retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2204         if (unlikely(retval == -ECHILD))
2205                 retval = path_parentat(&nd, flags, parent);
2206         if (unlikely(retval == -ESTALE))
2207                 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2208         if (likely(!retval)) {
2209                 *last = nd.last;
2210                 *type = nd.last_type;
2211                 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2212         } else {
2213                 putname(name);
2214                 name = ERR_PTR(retval);
2215         }
2216         restore_nameidata();
2217         return name;
2218 }
2219 
2220 /* does lookup, returns the object with parent locked */
2221 struct dentry *kern_path_locked(const char *name, struct path *path)
2222 {
2223         struct filename *filename;
2224         struct dentry *d;
2225         struct qstr last;
2226         int type;
2227 
2228         filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2229                                     &last, &type);
2230         if (IS_ERR(filename))
2231                 return ERR_CAST(filename);
2232         if (unlikely(type != LAST_NORM)) {
2233                 path_put(path);
2234                 putname(filename);
2235                 return ERR_PTR(-EINVAL);
2236         }
2237         mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2238         d = __lookup_hash(&last, path->dentry, 0);
2239         if (IS_ERR(d)) {
2240                 mutex_unlock(&path->dentry->d_inode->i_mutex);
2241                 path_put(path);
2242         }
2243         putname(filename);
2244         return d;
2245 }
2246 
2247 int kern_path(const char *name, unsigned int flags, struct path *path)
2248 {
2249         return filename_lookup(AT_FDCWD, getname_kernel(name),
2250                                flags, path, NULL);
2251 }
2252 EXPORT_SYMBOL(kern_path);
2253 
2254 /**
2255  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2256  * @dentry:  pointer to dentry of the base directory
2257  * @mnt: pointer to vfs mount of the base directory
2258  * @name: pointer to file name
2259  * @flags: lookup flags
2260  * @path: pointer to struct path to fill
2261  */
2262 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2263                     const char *name, unsigned int flags,
2264                     struct path *path)
2265 {
2266         struct path root = {.mnt = mnt, .dentry = dentry};
2267         /* the first argument of filename_lookup() is ignored with root */
2268         return filename_lookup(AT_FDCWD, getname_kernel(name),
2269                                flags , path, &root);
2270 }
2271 EXPORT_SYMBOL(vfs_path_lookup);
2272 
2273 /**
2274  * lookup_one_len - filesystem helper to lookup single pathname component
2275  * @name:       pathname component to lookup
2276  * @base:       base directory to lookup from
2277  * @len:        maximum length @len should be interpreted to
2278  *
2279  * Note that this routine is purely a helper for filesystem usage and should
2280  * not be called by generic code.
2281  */
2282 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2283 {
2284         struct qstr this;
2285         unsigned int c;
2286         int err;
2287 
2288         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2289 
2290         this.name = name;
2291         this.len = len;
2292         this.hash = full_name_hash(name, len);
2293         if (!len)
2294                 return ERR_PTR(-EACCES);
2295 
2296         if (unlikely(name[0] == '.')) {
2297                 if (len < 2 || (len == 2 && name[1] == '.'))
2298                         return ERR_PTR(-EACCES);
2299         }
2300 
2301         while (len--) {
2302                 c = *(const unsigned char *)name++;
2303                 if (c == '/' || c == '\0')
2304                         return ERR_PTR(-EACCES);
2305         }
2306         /*
2307          * See if the low-level filesystem might want
2308          * to use its own hash..
2309          */
2310         if (base->d_flags & DCACHE_OP_HASH) {
2311                 int err = base->d_op->d_hash(base, &this);
2312                 if (err < 0)
2313                         return ERR_PTR(err);
2314         }
2315 
2316         err = inode_permission(base->d_inode, MAY_EXEC);
2317         if (err)
2318                 return ERR_PTR(err);
2319 
2320         return __lookup_hash(&this, base, 0);
2321 }
2322 EXPORT_SYMBOL(lookup_one_len);
2323 
2324 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2325                  struct path *path, int *empty)
2326 {
2327         return filename_lookup(dfd, getname_flags(name, flags, empty),
2328                                flags, path, NULL);
2329 }
2330 EXPORT_SYMBOL(user_path_at_empty);
2331 
2332 /*
2333  * NB: most callers don't do anything directly with the reference to the
2334  *     to struct filename, but the nd->last pointer points into the name string
2335  *     allocated by getname. So we must hold the reference to it until all
2336  *     path-walking is complete.
2337  */
2338 static inline struct filename *
2339 user_path_parent(int dfd, const char __user *path,
2340                  struct path *parent,
2341                  struct qstr *last,
2342                  int *type,
2343                  unsigned int flags)
2344 {
2345         /* only LOOKUP_REVAL is allowed in extra flags */
2346         return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2347                                  parent, last, type);
2348 }
2349 
2350 /**
2351  * mountpoint_last - look up last component for umount
2352  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2353  * @path: pointer to container for result
2354  *
2355  * This is a special lookup_last function just for umount. In this case, we
2356  * need to resolve the path without doing any revalidation.
2357  *
2358  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2359  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2360  * in almost all cases, this lookup will be served out of the dcache. The only
2361  * cases where it won't are if nd->last refers to a symlink or the path is
2362  * bogus and it doesn't exist.
2363  *
2364  * Returns:
2365  * -error: if there was an error during lookup. This includes -ENOENT if the
2366  *         lookup found a negative dentry. The nd->path reference will also be
2367  *         put in this case.
2368  *
2369  * 0:      if we successfully resolved nd->path and found it to not to be a
2370  *         symlink that needs to be followed. "path" will also be populated.
2371  *         The nd->path reference will also be put.
2372  *
2373  * 1:      if we successfully resolved nd->last and found it to be a symlink
2374  *         that needs to be followed. "path" will be populated with the path
2375  *         to the link, and nd->path will *not* be put.
2376  */
2377 static int
2378 mountpoint_last(struct nameidata *nd, struct path *path)
2379 {
2380         int error = 0;
2381         struct dentry *dentry;
2382         struct dentry *dir = nd->path.dentry;
2383 
2384         /* If we're in rcuwalk, drop out of it to handle last component */
2385         if (nd->flags & LOOKUP_RCU) {
2386                 if (unlazy_walk(nd, NULL, 0))
2387                         return -ECHILD;
2388         }
2389 
2390         nd->flags &= ~LOOKUP_PARENT;
2391 
2392         if (unlikely(nd->last_type != LAST_NORM)) {
2393                 error = handle_dots(nd, nd->last_type);
2394                 if (error)
2395                         return error;
2396                 dentry = dget(nd->path.dentry);
2397                 goto done;
2398         }
2399 
2400         mutex_lock(&dir->d_inode->i_mutex);
2401         dentry = d_lookup(dir, &nd->last);
2402         if (!dentry) {
2403                 /*
2404                  * No cached dentry. Mounted dentries are pinned in the cache,
2405                  * so that means that this dentry is probably a symlink or the
2406                  * path doesn't actually point to a mounted dentry.
2407                  */
2408                 dentry = d_alloc(dir, &nd->last);
2409                 if (!dentry) {
2410                         mutex_unlock(&dir->d_inode->i_mutex);
2411                         return -ENOMEM;
2412                 }
2413                 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2414                 if (IS_ERR(dentry)) {
2415                         mutex_unlock(&dir->d_inode->i_mutex);
2416                         return PTR_ERR(dentry);
2417                 }
2418         }
2419         mutex_unlock(&dir->d_inode->i_mutex);
2420 
2421 done:
2422         if (d_is_negative(dentry)) {
2423                 dput(dentry);
2424                 return -ENOENT;
2425         }
2426         if (nd->depth)
2427                 put_link(nd);
2428         path->dentry = dentry;
2429         path->mnt = nd->path.mnt;
2430         error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2431                                    d_backing_inode(dentry), 0);
2432         if (unlikely(error))
2433                 return error;
2434         mntget(path->mnt);
2435         follow_mount(path);
2436         return 0;
2437 }
2438 
2439 /**
2440  * path_mountpoint - look up a path to be umounted
2441  * @nd:         lookup context
2442  * @flags:      lookup flags
2443  * @path:       pointer to container for result
2444  *
2445  * Look up the given name, but don't attempt to revalidate the last component.
2446  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2447  */
2448 static int
2449 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2450 {
2451         const char *s = path_init(nd, flags);
2452         int err;
2453         if (IS_ERR(s))
2454                 return PTR_ERR(s);
2455         while (!(err = link_path_walk(s, nd)) &&
2456                 (err = mountpoint_last(nd, path)) > 0) {
2457                 s = trailing_symlink(nd);
2458                 if (IS_ERR(s)) {
2459                         err = PTR_ERR(s);
2460                         break;
2461                 }
2462         }
2463         terminate_walk(nd);
2464         return err;
2465 }
2466 
2467 static int
2468 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2469                         unsigned int flags)
2470 {
2471         struct nameidata nd;
2472         int error;
2473         if (IS_ERR(name))
2474                 return PTR_ERR(name);
2475         set_nameidata(&nd, dfd, name);
2476         error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2477         if (unlikely(error == -ECHILD))
2478                 error = path_mountpoint(&nd, flags, path);
2479         if (unlikely(error == -ESTALE))
2480                 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2481         if (likely(!error))
2482                 audit_inode(name, path->dentry, 0);
2483         restore_nameidata();
2484         putname(name);
2485         return error;
2486 }
2487 
2488 /**
2489  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2490  * @dfd:        directory file descriptor
2491  * @name:       pathname from userland
2492  * @flags:      lookup flags
2493  * @path:       pointer to container to hold result
2494  *
2495  * A umount is a special case for path walking. We're not actually interested
2496  * in the inode in this situation, and ESTALE errors can be a problem. We
2497  * simply want track down the dentry and vfsmount attached at the mountpoint
2498  * and avoid revalidating the last component.
2499  *
2500  * Returns 0 and populates "path" on success.
2501  */
2502 int
2503 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2504                         struct path *path)
2505 {
2506         return filename_mountpoint(dfd, getname(name), path, flags);
2507 }
2508 
2509 int
2510 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2511                         unsigned int flags)
2512 {
2513         return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2514 }
2515 EXPORT_SYMBOL(kern_path_mountpoint);
2516 
2517 int __check_sticky(struct inode *dir, struct inode *inode)
2518 {
2519         kuid_t fsuid = current_fsuid();
2520 
2521         if (uid_eq(inode->i_uid, fsuid))
2522                 return 0;
2523         if (uid_eq(dir->i_uid, fsuid))
2524                 return 0;
2525         return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2526 }
2527 EXPORT_SYMBOL(__check_sticky);
2528 
2529 /*
2530  *      Check whether we can remove a link victim from directory dir, check
2531  *  whether the type of victim is right.
2532  *  1. We can't do it if dir is read-only (done in permission())
2533  *  2. We should have write and exec permissions on dir
2534  *  3. We can't remove anything from append-only dir
2535  *  4. We can't do anything with immutable dir (done in permission())
2536  *  5. If the sticky bit on dir is set we should either
2537  *      a. be owner of dir, or
2538  *      b. be owner of victim, or
2539  *      c. have CAP_FOWNER capability
2540  *  6. If the victim is append-only or immutable we can't do antyhing with
2541  *     links pointing to it.
2542  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2543  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2544  *  9. We can't remove a root or mountpoint.
2545  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2546  *     nfs_async_unlink().
2547  */
2548 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2549 {
2550         struct inode *inode = d_backing_inode(victim);
2551         int error;
2552 
2553         if (d_is_negative(victim))
2554                 return -ENOENT;
2555         BUG_ON(!inode);
2556 
2557         BUG_ON(victim->d_parent->d_inode != dir);
2558         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2559 
2560         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2561         if (error)
2562                 return error;
2563         if (IS_APPEND(dir))
2564                 return -EPERM;
2565 
2566         if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2567             IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2568                 return -EPERM;
2569         if (isdir) {
2570                 if (!d_is_dir(victim))
2571                         return -ENOTDIR;
2572                 if (IS_ROOT(victim))
2573                         return -EBUSY;
2574         } else if (d_is_dir(victim))
2575                 return -EISDIR;
2576         if (IS_DEADDIR(dir))
2577                 return -ENOENT;
2578         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2579                 return -EBUSY;
2580         return 0;
2581 }
2582 
2583 /*      Check whether we can create an object with dentry child in directory
2584  *  dir.
2585  *  1. We can't do it if child already exists (open has special treatment for
2586  *     this case, but since we are inlined it's OK)
2587  *  2. We can't do it if dir is read-only (done in permission())
2588  *  3. We should have write and exec permissions on dir
2589  *  4. We can't do it if dir is immutable (done in permission())
2590  */
2591 static inline int may_create(struct inode *dir, struct dentry *child)
2592 {
2593         audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2594         if (child->d_inode)
2595                 return -EEXIST;
2596         if (IS_DEADDIR(dir))
2597                 return -ENOENT;
2598         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2599 }
2600 
2601 /*
2602  * p1 and p2 should be directories on the same fs.
2603  */
2604 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2605 {
2606         struct dentry *p;
2607 
2608         if (p1 == p2) {
2609                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2610                 return NULL;
2611         }
2612 
2613         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2614 
2615         p = d_ancestor(p2, p1);
2616         if (p) {
2617                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2618                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2619                 return p;
2620         }
2621 
2622         p = d_ancestor(p1, p2);
2623         if (p) {
2624                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2625                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2626                 return p;
2627         }
2628 
2629         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2630         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2631         return NULL;
2632 }
2633 EXPORT_SYMBOL(lock_rename);
2634 
2635 void unlock_rename(struct dentry *p1, struct dentry *p2)
2636 {
2637         mutex_unlock(&p1->d_inode->i_mutex);
2638         if (p1 != p2) {
2639                 mutex_unlock(&p2->d_inode->i_mutex);
2640                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2641         }
2642 }
2643 EXPORT_SYMBOL(unlock_rename);
2644 
2645 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2646                 bool want_excl)
2647 {
2648         int error = may_create(dir, dentry);
2649         if (error)
2650                 return error;
2651 
2652         if (!dir->i_op->create)
2653                 return -EACCES; /* shouldn't it be ENOSYS? */
2654         mode &= S_IALLUGO;
2655         mode |= S_IFREG;
2656         error = security_inode_create(dir, dentry, mode);
2657         if (error)
2658                 return error;
2659         error = dir->i_op->create(dir, dentry, mode, want_excl);
2660         if (!error)
2661                 fsnotify_create(dir, dentry);
2662         return error;
2663 }
2664 EXPORT_SYMBOL(vfs_create);
2665 
2666 static int may_open(struct path *path, int acc_mode, int flag)
2667 {
2668         struct dentry *dentry = path->dentry;
2669         struct inode *inode = dentry->d_inode;
2670         int error;
2671 
2672         /* O_PATH? */
2673         if (!acc_mode)
2674                 return 0;
2675 
2676         if (!inode)
2677                 return -ENOENT;
2678 
2679         switch (inode->i_mode & S_IFMT) {
2680         case S_IFLNK:
2681                 return -ELOOP;
2682         case S_IFDIR:
2683                 if (acc_mode & MAY_WRITE)
2684                         return -EISDIR;
2685                 break;
2686         case S_IFBLK:
2687         case S_IFCHR:
2688                 if (path->mnt->mnt_flags & MNT_NODEV)
2689                         return -EACCES;
2690                 /*FALLTHRU*/
2691         case S_IFIFO:
2692         case S_IFSOCK:
2693                 flag &= ~O_TRUNC;
2694                 break;
2695         }
2696 
2697         error = inode_permission(inode, acc_mode);
2698         if (error)
2699                 return error;
2700 
2701         /*
2702          * An append-only file must be opened in append mode for writing.
2703          */
2704         if (IS_APPEND(inode)) {
2705                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2706                         return -EPERM;
2707                 if (flag & O_TRUNC)
2708                         return -EPERM;
2709         }
2710 
2711         /* O_NOATIME can only be set by the owner or superuser */
2712         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2713                 return -EPERM;
2714 
2715         return 0;
2716 }
2717 
2718 static int handle_truncate(struct file *filp)
2719 {
2720         struct path *path = &filp->f_path;
2721         struct inode *inode = path->dentry->d_inode;
2722         int error = get_write_access(inode);
2723         if (error)
2724                 return error;
2725         /*
2726          * Refuse to truncate files with mandatory locks held on them.
2727          */
2728         error = locks_verify_locked(filp);
2729         if (!error)
2730                 error = security_path_truncate(path);
2731         if (!error) {
2732                 error = do_truncate(path->dentry, 0,
2733                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2734                                     filp);
2735         }
2736         put_write_access(inode);
2737         return error;
2738 }
2739 
2740 static inline int open_to_namei_flags(int flag)
2741 {
2742         if ((flag & O_ACCMODE) == 3)
2743                 flag--;
2744         return flag;
2745 }
2746 
2747 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2748 {
2749         int error = security_path_mknod(dir, dentry, mode, 0);
2750         if (error)
2751                 return error;
2752 
2753         error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2754         if (error)
2755                 return error;
2756 
2757         return security_inode_create(dir->dentry->d_inode, dentry, mode);
2758 }
2759 
2760 /*
2761  * Attempt to atomically look up, create and open a file from a negative
2762  * dentry.
2763  *
2764  * Returns 0 if successful.  The file will have been created and attached to
2765  * @file by the filesystem calling finish_open().
2766  *
2767  * Returns 1 if the file was looked up only or didn't need creating.  The
2768  * caller will need to perform the open themselves.  @path will have been
2769  * updated to point to the new dentry.  This may be negative.
2770  *
2771  * Returns an error code otherwise.
2772  */
2773 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2774                         struct path *path, struct file *file,
2775                         const struct open_flags *op,
2776                         bool got_write, bool need_lookup,
2777                         int *opened)
2778 {
2779         struct inode *dir =  nd->path.dentry->d_inode;
2780         unsigned open_flag = open_to_namei_flags(op->open_flag);
2781         umode_t mode;
2782         int error;
2783         int acc_mode;
2784         int create_error = 0;
2785         struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2786         bool excl;
2787 
2788         BUG_ON(dentry->d_inode);
2789 
2790         /* Don't create child dentry for a dead directory. */
2791         if (unlikely(IS_DEADDIR(dir))) {
2792                 error = -ENOENT;
2793                 goto out;
2794         }
2795 
2796         mode = op->mode;
2797         if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2798                 mode &= ~current_umask();
2799 
2800         excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2801         if (excl)
2802                 open_flag &= ~O_TRUNC;
2803 
2804         /*
2805          * Checking write permission is tricky, bacuse we don't know if we are
2806          * going to actually need it: O_CREAT opens should work as long as the
2807          * file exists.  But checking existence breaks atomicity.  The trick is
2808          * to check access and if not granted clear O_CREAT from the flags.
2809          *
2810          * Another problem is returing the "right" error value (e.g. for an
2811          * O_EXCL open we want to return EEXIST not EROFS).
2812          */
2813         if (((open_flag & (O_CREAT | O_TRUNC)) ||
2814             (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2815                 if (!(open_flag & O_CREAT)) {
2816                         /*
2817                          * No O_CREATE -> atomicity not a requirement -> fall
2818                          * back to lookup + open
2819                          */
2820                         goto no_open;
2821                 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2822                         /* Fall back and fail with the right error */
2823                         create_error = -EROFS;
2824                         goto no_open;
2825                 } else {
2826                         /* No side effects, safe to clear O_CREAT */
2827                         create_error = -EROFS;
2828                         open_flag &= ~O_CREAT;
2829                 }
2830         }
2831 
2832         if (open_flag & O_CREAT) {
2833                 error = may_o_create(&nd->path, dentry, mode);
2834                 if (error) {
2835                         create_error = error;
2836                         if (open_flag & O_EXCL)
2837                                 goto no_open;
2838                         open_flag &= ~O_CREAT;
2839                 }
2840         }
2841 
2842         if (nd->flags & LOOKUP_DIRECTORY)
2843                 open_flag |= O_DIRECTORY;
2844 
2845         file->f_path.dentry = DENTRY_NOT_SET;
2846         file->f_path.mnt = nd->path.mnt;
2847         error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2848                                       opened);
2849         if (error < 0) {
2850                 if (create_error && error == -ENOENT)
2851                         error = create_error;
2852                 goto out;
2853         }
2854 
2855         if (error) {    /* returned 1, that is */
2856                 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2857                         error = -EIO;
2858                         goto out;
2859                 }
2860                 if (file->f_path.dentry) {
2861                         dput(dentry);
2862                         dentry = file->f_path.dentry;
2863                 }
2864                 if (*opened & FILE_CREATED)
2865                         fsnotify_create(dir, dentry);
2866                 if (!dentry->d_inode) {
2867                         WARN_ON(*opened & FILE_CREATED);
2868                         if (create_error) {
2869                                 error = create_error;
2870                                 goto out;
2871                         }
2872                 } else {
2873                         if (excl && !(*opened & FILE_CREATED)) {
2874                                 error = -EEXIST;
2875                                 goto out;
2876                         }
2877                 }
2878                 goto looked_up;
2879         }
2880 
2881         /*
2882          * We didn't have the inode before the open, so check open permission
2883          * here.
2884          */
2885         acc_mode = op->acc_mode;
2886         if (*opened & FILE_CREATED) {
2887                 WARN_ON(!(open_flag & O_CREAT));
2888                 fsnotify_create(dir, dentry);
2889                 acc_mode = MAY_OPEN;
2890         }
2891         error = may_open(&file->f_path, acc_mode, open_flag);
2892         if (error)
2893                 fput(file);
2894 
2895 out:
2896         dput(dentry);
2897         return error;
2898 
2899 no_open:
2900         if (need_lookup) {
2901                 dentry = lookup_real(dir, dentry, nd->flags);
2902                 if (IS_ERR(dentry))
2903                         return PTR_ERR(dentry);
2904 
2905                 if (create_error) {
2906                         int open_flag = op->open_flag;
2907 
2908                         error = create_error;
2909                         if ((open_flag & O_EXCL)) {
2910                                 if (!dentry->d_inode)
2911                                         goto out;
2912                         } else if (!dentry->d_inode) {
2913                                 goto out;
2914                         } else if ((open_flag & O_TRUNC) &&
2915                                    d_is_reg(dentry)) {
2916                                 goto out;
2917                         }
2918                         /* will fail later, go on to get the right error */
2919                 }
2920         }
2921 looked_up:
2922         path->dentry = dentry;
2923         path->mnt = nd->path.mnt;
2924         return 1;
2925 }
2926 
2927 /*
2928  * Look up and maybe create and open the last component.
2929  *
2930  * Must be called with i_mutex held on parent.
2931  *
2932  * Returns 0 if the file was successfully atomically created (if necessary) and
2933  * opened.  In this case the file will be returned attached to @file.
2934  *
2935  * Returns 1 if the file was not completely opened at this time, though lookups
2936  * and creations will have been performed and the dentry returned in @path will
2937  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2938  * specified then a negative dentry may be returned.
2939  *
2940  * An error code is returned otherwise.
2941  *
2942  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2943  * cleared otherwise prior to returning.
2944  */
2945 static int lookup_open(struct nameidata *nd, struct path *path,
2946                         struct file *file,
2947                         const struct open_flags *op,
2948                         bool got_write, int *opened)
2949 {
2950         struct dentry *dir = nd->path.dentry;
2951         struct inode *dir_inode = dir->d_inode;
2952         struct dentry *dentry;
2953         int error;
2954         bool need_lookup;
2955 
2956         *opened &= ~FILE_CREATED;
2957         dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2958         if (IS_ERR(dentry))
2959                 return PTR_ERR(dentry);
2960 
2961         /* Cached positive dentry: will open in f_op->open */
2962         if (!need_lookup && dentry->d_inode)
2963                 goto out_no_open;
2964 
2965         if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2966                 return atomic_open(nd, dentry, path, file, op, got_write,
2967                                    need_lookup, opened);
2968         }
2969 
2970         if (need_lookup) {
2971                 BUG_ON(dentry->d_inode);
2972 
2973                 dentry = lookup_real(dir_inode, dentry, nd->flags);
2974                 if (IS_ERR(dentry))
2975                         return PTR_ERR(dentry);
2976         }
2977 
2978         /* Negative dentry, just create the file */
2979         if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2980                 umode_t mode = op->mode;
2981                 if (!IS_POSIXACL(dir->d_inode))
2982                         mode &= ~current_umask();
2983                 /*
2984                  * This write is needed to ensure that a
2985                  * rw->ro transition does not occur between
2986                  * the time when the file is created and when
2987                  * a permanent write count is taken through
2988                  * the 'struct file' in finish_open().
2989                  */
2990                 if (!got_write) {
2991                         error = -EROFS;
2992                         goto out_dput;
2993                 }
2994                 *opened |= FILE_CREATED;
2995                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2996                 if (error)
2997                         goto out_dput;
2998                 error = vfs_create(dir->d_inode, dentry, mode,
2999                                    nd->flags & LOOKUP_EXCL);
3000                 if (error)
3001                         goto out_dput;
3002         }
3003 out_no_open:
3004         path->dentry = dentry;
3005         path->mnt = nd->path.mnt;
3006         return 1;
3007 
3008 out_dput:
3009         dput(dentry);
3010         return error;
3011 }
3012 
3013 /*
3014  * Handle the last step of open()
3015  */
3016 static int do_last(struct nameidata *nd,
3017                    struct file *file, const struct open_flags *op,
3018                    int *opened)
3019 {
3020         struct dentry *dir = nd->path.dentry;
3021         int open_flag = op->open_flag;
3022         bool will_truncate = (open_flag & O_TRUNC) != 0;
3023         bool got_write = false;
3024         int acc_mode = op->acc_mode;
3025         unsigned seq;
3026         struct inode *inode;
3027         struct path save_parent = { .dentry = NULL, .mnt = NULL };
3028         struct path path;
3029         bool retried = false;
3030         int error;
3031 
3032         nd->flags &= ~LOOKUP_PARENT;
3033         nd->flags |= op->intent;
3034 
3035         if (nd->last_type != LAST_NORM) {
3036                 error = handle_dots(nd, nd->last_type);
3037                 if (unlikely(error))
3038                         return error;
3039                 goto finish_open;
3040         }
3041 
3042         if (!(open_flag & O_CREAT)) {
3043                 if (nd->last.name[nd->last.len])
3044                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3045                 /* we _can_ be in RCU mode here */
3046                 error = lookup_fast(nd, &path, &inode, &seq);
3047                 if (likely(!error))
3048                         goto finish_lookup;
3049 
3050                 if (error < 0)
3051                         return error;
3052 
3053                 BUG_ON(nd->inode != dir->d_inode);
3054         } else {
3055                 /* create side of things */
3056                 /*
3057                  * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3058                  * has been cleared when we got to the last component we are
3059                  * about to look up
3060                  */
3061                 error = complete_walk(nd);
3062                 if (error)
3063                         return error;
3064 
3065                 audit_inode(nd->name, dir, LOOKUP_PARENT);
3066                 /* trailing slashes? */
3067                 if (unlikely(nd->last.name[nd->last.len]))
3068                         return -EISDIR;
3069         }
3070 
3071 retry_lookup:
3072         if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3073                 error = mnt_want_write(nd->path.mnt);
3074                 if (!error)
3075                         got_write = true;
3076                 /*
3077                  * do _not_ fail yet - we might not need that or fail with
3078                  * a different error; let lookup_open() decide; we'll be
3079                  * dropping this one anyway.
3080                  */
3081         }
3082         mutex_lock(&dir->d_inode->i_mutex);
3083         error = lookup_open(nd, &path, file, op, got_write, opened);
3084         mutex_unlock(&dir->d_inode->i_mutex);
3085 
3086         if (error <= 0) {
3087                 if (error)
3088                         goto out;
3089 
3090                 if ((*opened & FILE_CREATED) ||
3091                     !S_ISREG(file_inode(file)->i_mode))
3092                         will_truncate = false;
3093 
3094                 audit_inode(nd->name, file->f_path.dentry, 0);
3095                 goto opened;
3096         }
3097 
3098         if (*opened & FILE_CREATED) {
3099                 /* Don't check for write permission, don't truncate */
3100                 open_flag &= ~O_TRUNC;
3101                 will_truncate = false;
3102                 acc_mode = MAY_OPEN;
3103                 path_to_nameidata(&path, nd);
3104                 goto finish_open_created;
3105         }
3106 
3107         /*
3108          * create/update audit record if it already exists.
3109          */
3110         if (d_is_positive(path.dentry))
3111                 audit_inode(nd->name, path.dentry, 0);
3112 
3113         /*
3114          * If atomic_open() acquired write access it is dropped now due to
3115          * possible mount and symlink following (this might be optimized away if
3116          * necessary...)
3117          */
3118         if (got_write) {
3119                 mnt_drop_write(nd->path.mnt);
3120                 got_write = false;
3121         }
3122 
3123         if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3124                 path_to_nameidata(&path, nd);
3125                 return -EEXIST;
3126         }
3127 
3128         error = follow_managed(&path, nd);
3129         if (unlikely(error < 0))
3130                 return error;
3131 
3132         BUG_ON(nd->flags & LOOKUP_RCU);
3133         inode = d_backing_inode(path.dentry);
3134         seq = 0;        /* out of RCU mode, so the value doesn't matter */
3135         if (unlikely(d_is_negative(path.dentry))) {
3136                 path_to_nameidata(&path, nd);
3137                 return -ENOENT;
3138         }
3139 finish_lookup:
3140         if (nd->depth)
3141                 put_link(nd);
3142         error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3143                                    inode, seq);
3144         if (unlikely(error))
3145                 return error;
3146 
3147         if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3148                 path_to_nameidata(&path, nd);
3149                 return -ELOOP;
3150         }
3151 
3152         if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3153                 path_to_nameidata(&path, nd);
3154         } else {
3155                 save_parent.dentry = nd->path.dentry;
3156                 save_parent.mnt = mntget(path.mnt);
3157                 nd->path.dentry = path.dentry;
3158 
3159         }
3160         nd->inode = inode;
3161         nd->seq = seq;
3162         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3163 finish_open:
3164         error = complete_walk(nd);
3165         if (error) {
3166                 path_put(&save_parent);
3167                 return error;
3168         }
3169         audit_inode(nd->name, nd->path.dentry, 0);
3170         error = -EISDIR;
3171         if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3172                 goto out;
3173         error = -ENOTDIR;
3174         if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3175                 goto out;
3176         if (!d_is_reg(nd->path.dentry))
3177                 will_truncate = false;
3178 
3179         if (will_truncate) {
3180                 error = mnt_want_write(nd->path.mnt);
3181                 if (error)
3182                         goto out;
3183                 got_write = true;
3184         }
3185 finish_open_created:
3186         error = may_open(&nd->path, acc_mode, open_flag);
3187         if (error)
3188                 goto out;
3189 
3190         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3191         error = vfs_open(&nd->path, file, current_cred());
3192         if (!error) {
3193                 *opened |= FILE_OPENED;
3194         } else {
3195                 if (error == -EOPENSTALE)
3196                         goto stale_open;
3197                 goto out;
3198         }
3199 opened:
3200         error = open_check_o_direct(file);
3201         if (error)
3202                 goto exit_fput;
3203         error = ima_file_check(file, op->acc_mode, *opened);
3204         if (error)
3205                 goto exit_fput;
3206 
3207         if (will_truncate) {
3208                 error = handle_truncate(file);
3209                 if (error)
3210                         goto exit_fput;
3211         }
3212 out:
3213         if (got_write)
3214                 mnt_drop_write(nd->path.mnt);
3215         path_put(&save_parent);
3216         return error;
3217 
3218 exit_fput:
3219         fput(file);
3220         goto out;
3221 
3222 stale_open:
3223         /* If no saved parent or already retried then can't retry */
3224         if (!save_parent.dentry || retried)
3225                 goto out;
3226 
3227         BUG_ON(save_parent.dentry != dir);
3228         path_put(&nd->path);
3229         nd->path = save_parent;
3230         nd->inode = dir->d_inode;
3231         save_parent.mnt = NULL;
3232         save_parent.dentry = NULL;
3233         if (got_write) {
3234                 mnt_drop_write(nd->path.mnt);
3235                 got_write = false;
3236         }
3237         retried = true;
3238         goto retry_lookup;
3239 }
3240 
3241 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3242                 const struct open_flags *op,
3243                 struct file *file, int *opened)
3244 {
3245         static const struct qstr name = QSTR_INIT("/", 1);
3246         struct dentry *child;
3247         struct inode *dir;
3248         struct path path;
3249         int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3250         if (unlikely(error))
3251                 return error;
3252         error = mnt_want_write(path.mnt);
3253         if (unlikely(error))
3254                 goto out;
3255         dir = path.dentry->d_inode;
3256         /* we want directory to be writable */
3257         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3258         if (error)
3259                 goto out2;
3260         if (!dir->i_op->tmpfile) {
3261                 error = -EOPNOTSUPP;
3262                 goto out2;
3263         }
3264         child = d_alloc(path.dentry, &name);
3265         if (unlikely(!child)) {
3266                 error = -ENOMEM;
3267                 goto out2;
3268         }
3269         dput(path.dentry);
3270         path.dentry = child;
3271         error = dir->i_op->tmpfile(dir, child, op->mode);
3272         if (error)
3273                 goto out2;
3274         audit_inode(nd->name, child, 0);
3275         /* Don't check for other permissions, the inode was just created */
3276         error = may_open(&path, MAY_OPEN, op->open_flag);
3277         if (error)
3278                 goto out2;
3279         file->f_path.mnt = path.mnt;
3280         error = finish_open(file, child, NULL, opened);
3281         if (error)
3282                 goto out2;
3283         error = open_check_o_direct(file);
3284         if (error) {
3285                 fput(file);
3286         } else if (!(op->open_flag & O_EXCL)) {
3287                 struct inode *inode = file_inode(file);
3288                 spin_lock(&inode->i_lock);
3289                 inode->i_state |= I_LINKABLE;
3290                 spin_unlock(&inode->i_lock);
3291         }
3292 out2:
3293         mnt_drop_write(path.mnt);
3294 out:
3295         path_put(&path);
3296         return error;
3297 }
3298 
3299 static struct file *path_openat(struct nameidata *nd,
3300                         const struct open_flags *op, unsigned flags)
3301 {
3302         const char *s;
3303         struct file *file;
3304         int opened = 0;
3305         int error;
3306 
3307         file = get_empty_filp();
3308         if (IS_ERR(file))
3309                 return file;
3310 
3311         file->f_flags = op->open_flag;
3312 
3313         if (unlikely(file->f_flags & __O_TMPFILE)) {
3314                 error = do_tmpfile(nd, flags, op, file, &opened);
3315                 goto out2;
3316         }
3317 
3318         s = path_init(nd, flags);
3319         if (IS_ERR(s)) {
3320                 put_filp(file);
3321                 return ERR_CAST(s);
3322         }
3323         while (!(error = link_path_walk(s, nd)) &&
3324                 (error = do_last(nd, file, op, &opened)) > 0) {
3325                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3326                 s = trailing_symlink(nd);
3327                 if (IS_ERR(s)) {
3328                         error = PTR_ERR(s);
3329                         break;
3330                 }
3331         }
3332         terminate_walk(nd);
3333 out2:
3334         if (!(opened & FILE_OPENED)) {
3335                 BUG_ON(!error);
3336                 put_filp(file);
3337         }
3338         if (unlikely(error)) {
3339                 if (error == -EOPENSTALE) {
3340                         if (flags & LOOKUP_RCU)
3341                                 error = -ECHILD;
3342                         else
3343                                 error = -ESTALE;
3344                 }
3345                 file = ERR_PTR(error);
3346         }
3347         return file;
3348 }
3349 
3350 struct file *do_filp_open(int dfd, struct filename *pathname,
3351                 const struct open_flags *op)
3352 {
3353         struct nameidata nd;
3354         int flags = op->lookup_flags;
3355         struct file *filp;
3356 
3357         set_nameidata(&nd, dfd, pathname);
3358         filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3359         if (unlikely(filp == ERR_PTR(-ECHILD)))
3360                 filp = path_openat(&nd, op, flags);
3361         if (unlikely(filp == ERR_PTR(-ESTALE)))
3362                 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3363         restore_nameidata();
3364         return filp;
3365 }
3366 
3367 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3368                 const char *name, const struct open_flags *op)
3369 {
3370         struct nameidata nd;
3371         struct file *file;
3372         struct filename *filename;
3373         int flags = op->lookup_flags | LOOKUP_ROOT;
3374 
3375         nd.root.mnt = mnt;
3376         nd.root.dentry = dentry;
3377 
3378         if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3379                 return ERR_PTR(-ELOOP);
3380 
3381         filename = getname_kernel(name);
3382         if (IS_ERR(filename))
3383                 return ERR_CAST(filename);
3384 
3385         set_nameidata(&nd, -1, filename);
3386         file = path_openat(&nd, op, flags | LOOKUP_RCU);
3387         if (unlikely(file == ERR_PTR(-ECHILD)))
3388                 file = path_openat(&nd, op, flags);
3389         if (unlikely(file == ERR_PTR(-ESTALE)))
3390                 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3391         restore_nameidata();
3392         putname(filename);
3393         return file;
3394 }
3395 
3396 static struct dentry *filename_create(int dfd, struct filename *name,
3397                                 struct path *path, unsigned int lookup_flags)
3398 {
3399         struct dentry *dentry = ERR_PTR(-EEXIST);
3400         struct qstr last;
3401         int type;
3402         int err2;
3403         int error;
3404         bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3405 
3406         /*
3407          * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3408          * other flags passed in are ignored!
3409          */
3410         lookup_flags &= LOOKUP_REVAL;
3411 
3412         name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3413         if (IS_ERR(name))
3414                 return ERR_CAST(name);
3415 
3416         /*
3417          * Yucky last component or no last component at all?
3418          * (foo/., foo/.., /////)
3419          */
3420         if (unlikely(type != LAST_NORM))
3421                 goto out;
3422 
3423         /* don't fail immediately if it's r/o, at least try to report other errors */
3424         err2 = mnt_want_write(path->mnt);
3425         /*
3426          * Do the final lookup.
3427          */
3428         lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3429         mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3430         dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3431         if (IS_ERR(dentry))
3432                 goto unlock;
3433 
3434         error = -EEXIST;
3435         if (d_is_positive(dentry))
3436                 goto fail;
3437 
3438         /*
3439          * Special case - lookup gave negative, but... we had foo/bar/
3440          * From the vfs_mknod() POV we just have a negative dentry -
3441          * all is fine. Let's be bastards - you had / on the end, you've
3442          * been asking for (non-existent) directory. -ENOENT for you.
3443          */
3444         if (unlikely(!is_dir && last.name[last.len])) {
3445                 error = -ENOENT;
3446                 goto fail;
3447         }
3448         if (unlikely(err2)) {
3449                 error = err2;
3450                 goto fail;
3451         }
3452         putname(name);
3453         return dentry;
3454 fail:
3455         dput(dentry);
3456         dentry = ERR_PTR(error);
3457 unlock:
3458         mutex_unlock(&path->dentry->d_inode->i_mutex);
3459         if (!err2)
3460                 mnt_drop_write(path->mnt);
3461 out:
3462         path_put(path);
3463         putname(name);
3464         return dentry;
3465 }
3466 
3467 struct dentry *kern_path_create(int dfd, const char *pathname,
3468                                 struct path *path, unsigned int lookup_flags)
3469 {
3470         return filename_create(dfd, getname_kernel(pathname),
3471                                 path, lookup_flags);
3472 }
3473 EXPORT_SYMBOL(kern_path_create);
3474 
3475 void done_path_create(struct path *path, struct dentry *dentry)
3476 {
3477         dput(dentry);
3478         mutex_unlock(&path->dentry->d_inode->i_mutex);
3479         mnt_drop_write(path->mnt);
3480         path_put(path);
3481 }
3482 EXPORT_SYMBOL(done_path_create);
3483 
3484 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3485                                 struct path *path, unsigned int lookup_flags)
3486 {
3487         return filename_create(dfd, getname(pathname), path, lookup_flags);
3488 }
3489 EXPORT_SYMBOL(user_path_create);
3490 
3491 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3492 {
3493         int error = may_create(dir, dentry);
3494 
3495         if (error)
3496                 return error;
3497 
3498         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3499                 return -EPERM;
3500 
3501         if (!dir->i_op->mknod)
3502                 return -EPERM;
3503 
3504         error = devcgroup_inode_mknod(mode, dev);
3505         if (error)
3506                 return error;
3507 
3508         error = security_inode_mknod(dir, dentry, mode, dev);
3509         if (error)
3510                 return error;
3511 
3512         error = dir->i_op->mknod(dir, dentry, mode, dev);
3513         if (!error)
3514                 fsnotify_create(dir, dentry);
3515         return error;
3516 }
3517 EXPORT_SYMBOL(vfs_mknod);
3518 
3519 static int may_mknod(umode_t mode)
3520 {
3521         switch (mode & S_IFMT) {
3522         case S_IFREG:
3523         case S_IFCHR:
3524         case S_IFBLK:
3525         case S_IFIFO:
3526         case S_IFSOCK:
3527         case 0: /* zero mode translates to S_IFREG */
3528                 return 0;
3529         case S_IFDIR:
3530                 return -EPERM;
3531         default:
3532                 return -EINVAL;
3533         }
3534 }
3535 
3536 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3537                 unsigned, dev)
3538 {
3539         struct dentry *dentry;
3540         struct path path;
3541         int error;
3542         unsigned int lookup_flags = 0;
3543 
3544         error = may_mknod(mode);
3545         if (error)
3546                 return error;
3547 retry:
3548         dentry = user_path_create(dfd, filename, &path, lookup_flags);
3549         if (IS_ERR(dentry))
3550                 return PTR_ERR(dentry);
3551 
3552         if (!IS_POSIXACL(path.dentry->d_inode))
3553                 mode &= ~current_umask();
3554         error = security_path_mknod(&path, dentry, mode, dev);
3555         if (error)
3556                 goto out;
3557         switch (mode & S_IFMT) {
3558                 case 0: case S_IFREG:
3559                         error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3560                         break;
3561                 case S_IFCHR: case S_IFBLK:
3562                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3563                                         new_decode_dev(dev));
3564                         break;
3565                 case S_IFIFO: case S_IFSOCK:
3566                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3567                         break;
3568         }
3569 out:
3570         done_path_create(&path, dentry);
3571         if (retry_estale(error, lookup_flags)) {
3572                 lookup_flags |= LOOKUP_REVAL;
3573                 goto retry;
3574         }
3575         return error;
3576 }
3577 
3578 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3579 {
3580         return sys_mknodat(AT_FDCWD, filename, mode, dev);
3581 }
3582 
3583 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3584 {
3585         int error = may_create(dir, dentry);
3586         unsigned max_links = dir->i_sb->s_max_links;
3587 
3588         if (error)
3589                 return error;
3590 
3591         if (!dir->i_op->mkdir)
3592                 return -EPERM;
3593 
3594         mode &= (S_IRWXUGO|S_ISVTX);
3595         error = security_inode_mkdir(dir, dentry, mode);
3596         if (error)
3597                 return error;
3598 
3599         if (max_links && dir->i_nlink >= max_links)
3600                 return -EMLINK;
3601 
3602         error = dir->i_op->mkdir(dir, dentry, mode);
3603         if (!error)
3604                 fsnotify_mkdir(dir, dentry);
3605         return error;
3606 }
3607 EXPORT_SYMBOL(vfs_mkdir);
3608 
3609 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3610 {
3611         struct dentry *dentry;
3612         struct path path;
3613         int error;
3614         unsigned int lookup_flags = LOOKUP_DIRECTORY;
3615 
3616 retry:
3617         dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3618         if (IS_ERR(dentry))
3619                 return PTR_ERR(dentry);
3620 
3621         if (!IS_POSIXACL(path.dentry->d_inode))
3622                 mode &= ~current_umask();
3623         error = security_path_mkdir(&path, dentry, mode);
3624         if (!error)
3625                 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3626         done_path_create(&path, dentry);
3627         if (retry_estale(error, lookup_flags)) {
3628                 lookup_flags |= LOOKUP_REVAL;
3629                 goto retry;
3630         }
3631         return error;
3632 }
3633 
3634 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3635 {
3636         return sys_mkdirat(AT_FDCWD, pathname, mode);
3637 }
3638 
3639 /*
3640  * The dentry_unhash() helper will try to drop the dentry early: we
3641  * should have a usage count of 1 if we're the only user of this
3642  * dentry, and if that is true (possibly after pruning the dcache),
3643  * then we drop the dentry now.
3644  *
3645  * A low-level filesystem can, if it choses, legally
3646  * do a
3647  *
3648  *      if (!d_unhashed(dentry))
3649  *              return -EBUSY;
3650  *
3651  * if it cannot handle the case of removing a directory
3652  * that is still in use by something else..
3653  */
3654 void dentry_unhash(struct dentry *dentry)
3655 {
3656         shrink_dcache_parent(dentry);
3657         spin_lock(&dentry->d_lock);
3658         if (dentry->d_lockref.count == 1)
3659                 __d_drop(dentry);
3660         spin_unlock(&dentry->d_lock);
3661 }
3662 EXPORT_SYMBOL(dentry_unhash);
3663 
3664 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3665 {
3666         int error = may_delete(dir, dentry, 1);
3667 
3668         if (error)
3669                 return error;
3670 
3671         if (!dir->i_op->rmdir)
3672                 return -EPERM;
3673 
3674         dget(dentry);
3675         mutex_lock(&dentry->d_inode->i_mutex);
3676 
3677         error = -EBUSY;
3678         if (is_local_mountpoint(dentry))
3679                 goto out;
3680 
3681         error = security_inode_rmdir(dir, dentry);
3682         if (error)
3683                 goto out;
3684 
3685         shrink_dcache_parent(dentry);
3686         error = dir->i_op->rmdir(dir, dentry);
3687         if (error)
3688                 goto out;
3689 
3690         dentry->d_inode->i_flags |= S_DEAD;
3691         dont_mount(dentry);
3692         detach_mounts(dentry);
3693 
3694 out:
3695         mutex_unlock(&dentry->d_inode->i_mutex);
3696         dput(dentry);
3697         if (!error)
3698                 d_delete(dentry);
3699         return error;
3700 }
3701 EXPORT_SYMBOL(vfs_rmdir);
3702 
3703 static long do_rmdir(int dfd, const char __user *pathname)
3704 {
3705         int error = 0;
3706         struct filename *name;
3707         struct dentry *dentry;
3708         struct path path;
3709         struct qstr last;
3710         int type;
3711         unsigned int lookup_flags = 0;
3712 retry:
3713         name = user_path_parent(dfd, pathname,
3714                                 &path, &last, &type, lookup_flags);
3715         if (IS_ERR(name))
3716                 return PTR_ERR(name);
3717 
3718         switch (type) {
3719         case LAST_DOTDOT:
3720                 error = -ENOTEMPTY;
3721                 goto exit1;
3722         case LAST_DOT:
3723                 error = -EINVAL;
3724                 goto exit1;
3725         case LAST_ROOT:
3726                 error = -EBUSY;
3727                 goto exit1;
3728         }
3729 
3730         error = mnt_want_write(path.mnt);
3731         if (error)
3732                 goto exit1;
3733 
3734         mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3735         dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3736         error = PTR_ERR(dentry);
3737         if (IS_ERR(dentry))
3738                 goto exit2;
3739         if (!dentry->d_inode) {
3740                 error = -ENOENT;
3741                 goto exit3;
3742         }
3743         error = security_path_rmdir(&path, dentry);
3744         if (error)
3745                 goto exit3;
3746         error = vfs_rmdir(path.dentry->d_inode, dentry);
3747 exit3:
3748         dput(dentry);
3749 exit2:
3750         mutex_unlock(&path.dentry->d_inode->i_mutex);
3751         mnt_drop_write(path.mnt);
3752 exit1:
3753         path_put(&path);
3754         putname(name);
3755         if (retry_estale(error, lookup_flags)) {
3756                 lookup_flags |= LOOKUP_REVAL;
3757                 goto retry;
3758         }
3759         return error;
3760 }
3761 
3762 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3763 {
3764         return do_rmdir(AT_FDCWD, pathname);
3765 }
3766 
3767 /**
3768  * vfs_unlink - unlink a filesystem object
3769  * @dir:        parent directory
3770  * @dentry:     victim
3771  * @delegated_inode: returns victim inode, if the inode is delegated.
3772  *
3773  * The caller must hold dir->i_mutex.
3774  *
3775  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3776  * return a reference to the inode in delegated_inode.  The caller
3777  * should then break the delegation on that inode and retry.  Because
3778  * breaking a delegation may take a long time, the caller should drop
3779  * dir->i_mutex before doing so.
3780  *
3781  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3782  * be appropriate for callers that expect the underlying filesystem not
3783  * to be NFS exported.
3784  */
3785 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3786 {
3787         struct inode *target = dentry->d_inode;
3788         int error = may_delete(dir, dentry, 0);
3789 
3790         if (error)
3791                 return error;
3792 
3793         if (!dir->i_op->unlink)
3794                 return -EPERM;
3795 
3796         mutex_lock(&target->i_mutex);
3797         if (is_local_mountpoint(dentry))
3798                 error = -EBUSY;
3799         else {
3800                 error = security_inode_unlink(dir, dentry);
3801                 if (!error) {
3802                         error = try_break_deleg(target, delegated_inode);
3803                         if (error)
3804                                 goto out;
3805                         error = dir->i_op->unlink(dir, dentry);
3806                         if (!error) {
3807                                 dont_mount(dentry);
3808                                 detach_mounts(dentry);
3809                         }
3810                 }
3811         }
3812 out:
3813         mutex_unlock(&target->i_mutex);
3814 
3815         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3816         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3817                 fsnotify_link_count(target);
3818                 d_delete(dentry);
3819         }
3820 
3821         return error;
3822 }
3823 EXPORT_SYMBOL(vfs_unlink);
3824 
3825 /*
3826  * Make sure that the actual truncation of the file will occur outside its
3827  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3828  * writeout happening, and we don't want to prevent access to the directory
3829  * while waiting on the I/O.
3830  */
3831 static long do_unlinkat(int dfd, const char __user *pathname)
3832 {
3833         int error;
3834         struct filename *name;
3835         struct dentry *dentry;
3836         struct path path;
3837         struct qstr last;
3838         int type;
3839         struct inode *inode = NULL;
3840         struct inode *delegated_inode = NULL;
3841         unsigned int lookup_flags = 0;
3842 retry:
3843         name = user_path_parent(dfd, pathname,
3844                                 &path, &last, &type, lookup_flags);
3845         if (IS_ERR(name))
3846                 return PTR_ERR(name);
3847 
3848         error = -EISDIR;
3849         if (type != LAST_NORM)
3850                 goto exit1;
3851 
3852         error = mnt_want_write(path.mnt);
3853         if (error)
3854                 goto exit1;
3855 retry_deleg:
3856         mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3857         dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3858         error = PTR_ERR(dentry);
3859         if (!IS_ERR(dentry)) {
3860                 /* Why not before? Because we want correct error value */
3861                 if (last.name[last.len])
3862                         goto slashes;
3863                 inode = dentry->d_inode;
3864                 if (d_is_negative(dentry))
3865                         goto slashes;
3866                 ihold(inode);
3867                 error = security_path_unlink(&path, dentry);
3868                 if (error)
3869                         goto exit2;
3870                 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3871 exit2:
3872                 dput(dentry);
3873         }
3874         mutex_unlock(&path.dentry->d_inode->i_mutex);
3875         if (inode)
3876                 iput(inode);    /* truncate the inode here */
3877         inode = NULL;
3878         if (delegated_inode) {
3879                 error = break_deleg_wait(&delegated_inode);
3880                 if (!error)
3881                         goto retry_deleg;
3882         }
3883         mnt_drop_write(path.mnt);
3884 exit1:
3885         path_put(&path);
3886         putname(name);
3887         if (retry_estale(error, lookup_flags)) {
3888                 lookup_flags |= LOOKUP_REVAL;
3889                 inode = NULL;
3890                 goto retry;
3891         }
3892         return error;
3893 
3894 slashes:
3895         if (d_is_negative(dentry))
3896                 error = -ENOENT;
3897         else if (d_is_dir(dentry))
3898                 error = -EISDIR;
3899         else
3900                 error = -ENOTDIR;
3901         goto exit2;
3902 }
3903 
3904 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3905 {
3906         if ((flag & ~AT_REMOVEDIR) != 0)
3907                 return -EINVAL;
3908 
3909         if (flag & AT_REMOVEDIR)
3910                 return do_rmdir(dfd, pathname);
3911 
3912         return do_unlinkat(dfd, pathname);
3913 }
3914 
3915 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3916 {
3917         return do_unlinkat(AT_FDCWD, pathname);
3918 }
3919 
3920 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3921 {
3922         int error = may_create(dir, dentry);
3923 
3924         if (error)
3925                 return error;
3926 
3927         if (!dir->i_op->symlink)
3928                 return -EPERM;
3929 
3930         error = security_inode_symlink(dir, dentry, oldname);
3931         if (error)
3932                 return error;
3933 
3934         error = dir->i_op->symlink(dir, dentry, oldname);
3935         if (!error)
3936                 fsnotify_create(dir, dentry);
3937         return error;
3938 }
3939 EXPORT_SYMBOL(vfs_symlink);
3940 
3941 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3942                 int, newdfd, const char __user *, newname)
3943 {
3944         int error;
3945         struct filename *from;
3946         struct dentry *dentry;
3947         struct path path;
3948         unsigned int lookup_flags = 0;
3949 
3950         from = getname(oldname);
3951         if (IS_ERR(from))
3952                 return PTR_ERR(from);
3953 retry:
3954         dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3955         error = PTR_ERR(dentry);
3956         if (IS_ERR(dentry))
3957                 goto out_putname;
3958 
3959         error = security_path_symlink(&path, dentry, from->name);
3960         if (!error)
3961                 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3962         done_path_create(&path, dentry);
3963         if (retry_estale(error, lookup_flags)) {
3964                 lookup_flags |= LOOKUP_REVAL;
3965                 goto retry;
3966         }
3967 out_putname:
3968         putname(from);
3969         return error;
3970 }
3971 
3972 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3973 {
3974         return sys_symlinkat(oldname, AT_FDCWD, newname);
3975 }
3976 
3977 /**
3978  * vfs_link - create a new link
3979  * @old_dentry: object to be linked
3980  * @dir:        new parent
3981  * @new_dentry: where to create the new link
3982  * @delegated_inode: returns inode needing a delegation break
3983  *
3984  * The caller must hold dir->i_mutex
3985  *
3986  * If vfs_link discovers a delegation on the to-be-linked file in need
3987  * of breaking, it will return -EWOULDBLOCK and return a reference to the
3988  * inode in delegated_inode.  The caller should then break the delegation
3989  * and retry.  Because breaking a delegation may take a long time, the
3990  * caller should drop the i_mutex before doing so.
3991  *
3992  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3993  * be appropriate for callers that expect the underlying filesystem not
3994  * to be NFS exported.
3995  */
3996 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3997 {
3998         struct inode *inode = old_dentry->d_inode;
3999         unsigned max_links = dir->i_sb->s_max_links;
4000         int error;
4001 
4002         if (!inode)
4003                 return -ENOENT;
4004 
4005         error = may_create(dir, new_dentry);
4006         if (error)
4007                 return error;
4008 
4009         if (dir->i_sb != inode->i_sb)
4010                 return -EXDEV;
4011 
4012         /*
4013          * A link to an append-only or immutable file cannot be created.
4014          */
4015         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4016                 return -EPERM;
4017         if (!dir->i_op->link)
4018                 return -EPERM;
4019         if (S_ISDIR(inode->i_mode))
4020                 return -EPERM;
4021 
4022         error = security_inode_link(old_dentry, dir, new_dentry);
4023         if (error)
4024                 return error;
4025 
4026         mutex_lock(&inode->i_mutex);
4027         /* Make sure we don't allow creating hardlink to an unlinked file */
4028         if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4029                 error =  -ENOENT;
4030         else if (max_links && inode->i_nlink >= max_links)
4031                 error = -EMLINK;
4032         else {
4033                 error = try_break_deleg(inode, delegated_inode);
4034                 if (!error)
4035                         error = dir->i_op->link(old_dentry, dir, new_dentry);
4036         }
4037 
4038         if (!error && (inode->i_state & I_LINKABLE)) {
4039                 spin_lock(&inode->i_lock);
4040                 inode->i_state &= ~I_LINKABLE;
4041                 spin_unlock(&inode->i_lock);
4042         }
4043         mutex_unlock(&inode->i_mutex);
4044         if (!error)
4045                 fsnotify_link(dir, inode, new_dentry);
4046         return error;
4047 }
4048 EXPORT_SYMBOL(vfs_link);
4049 
4050 /*
4051  * Hardlinks are often used in delicate situations.  We avoid
4052  * security-related surprises by not following symlinks on the
4053  * newname.  --KAB
4054  *
4055  * We don't follow them on the oldname either to be compatible
4056  * with linux 2.0, and to avoid hard-linking to directories
4057  * and other special files.  --ADM
4058  */
4059 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4060                 int, newdfd, const char __user *, newname, int, flags)
4061 {
4062         struct dentry *new_dentry;
4063         struct path old_path, new_path;
4064         struct inode *delegated_inode = NULL;
4065         int how = 0;
4066         int error;
4067 
4068         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4069                 return -EINVAL;
4070         /*
4071          * To use null names we require CAP_DAC_READ_SEARCH
4072          * This ensures that not everyone will be able to create
4073          * handlink using the passed filedescriptor.
4074          */
4075         if (flags & AT_EMPTY_PATH) {
4076                 if (!capable(CAP_DAC_READ_SEARCH))
4077                         return -ENOENT;
4078                 how = LOOKUP_EMPTY;
4079         }
4080 
4081         if (flags & AT_SYMLINK_FOLLOW)
4082                 how |= LOOKUP_FOLLOW;
4083 retry:
4084         error = user_path_at(olddfd, oldname, how, &old_path);
4085         if (error)
4086                 return error;
4087 
4088         new_dentry = user_path_create(newdfd, newname, &new_path,
4089                                         (how & LOOKUP_REVAL));
4090         error = PTR_ERR(new_dentry);
4091         if (IS_ERR(new_dentry))
4092                 goto out;
4093 
4094         error = -EXDEV;
4095         if (old_path.mnt != new_path.mnt)
4096                 goto out_dput;
4097         error = may_linkat(&old_path);
4098         if (unlikely(error))
4099                 goto out_dput;
4100         error = security_path_link(old_path.dentry, &new_path, new_dentry);
4101         if (error)
4102                 goto out_dput;
4103         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4104 out_dput:
4105         done_path_create(&new_path, new_dentry);
4106         if (delegated_inode) {
4107                 error = break_deleg_wait(&delegated_inode);
4108                 if (!error) {
4109                         path_put(&old_path);
4110                         goto retry;
4111                 }
4112         }
4113         if (retry_estale(error, how)) {
4114                 path_put(&old_path);
4115                 how |= LOOKUP_REVAL;
4116                 goto retry;
4117         }
4118 out:
4119         path_put(&old_path);
4120 
4121         return error;
4122 }
4123 
4124 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4125 {
4126         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4127 }
4128 
4129 /**
4130  * vfs_rename - rename a filesystem object
4131  * @old_dir:    parent of source
4132  * @old_dentry: source
4133  * @new_dir:    parent of destination
4134  * @new_dentry: destination
4135  * @delegated_inode: returns an inode needing a delegation break
4136  * @flags:      rename flags
4137  *
4138  * The caller must hold multiple mutexes--see lock_rename()).
4139  *
4140  * If vfs_rename discovers a delegation in need of breaking at either
4141  * the source or destination, it will return -EWOULDBLOCK and return a
4142  * reference to the inode in delegated_inode.  The caller should then
4143  * break the delegation and retry.  Because breaking a delegation may
4144  * take a long time, the caller should drop all locks before doing
4145  * so.
4146  *
4147  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4148  * be appropriate for callers that expect the underlying filesystem not
4149  * to be NFS exported.
4150  *
4151  * The worst of all namespace operations - renaming directory. "Perverted"
4152  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4153  * Problems:
4154  *      a) we can get into loop creation.
4155  *      b) race potential - two innocent renames can create a loop together.
4156  *         That's where 4.4 screws up. Current fix: serialization on
4157  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4158  *         story.
4159  *      c) we have to lock _four_ objects - parents and victim (if it exists),
4160  *         and source (if it is not a directory).
4161  *         And that - after we got ->i_mutex on parents (until then we don't know
4162  *         whether the target exists).  Solution: try to be smart with locking
4163  *         order for inodes.  We rely on the fact that tree topology may change
4164  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4165  *         move will be locked.  Thus we can rank directories by the tree
4166  *         (ancestors first) and rank all non-directories after them.
4167  *         That works since everybody except rename does "lock parent, lookup,
4168  *         lock child" and rename is under ->s_vfs_rename_mutex.
4169  *         HOWEVER, it relies on the assumption that any object with ->lookup()
4170  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4171  *         we'd better make sure that there's no link(2) for them.
4172  *      d) conversion from fhandle to dentry may come in the wrong moment - when
4173  *         we are removing the target. Solution: we will have to grab ->i_mutex
4174  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4175  *         ->i_mutex on parents, which works but leads to some truly excessive
4176  *         locking].
4177  */
4178 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4179                struct inode *new_dir, struct dentry *new_dentry,
4180                struct inode **delegated_inode, unsigned int flags)
4181 {
4182         int error;
4183         bool is_dir = d_is_dir(old_dentry);
4184         const unsigned char *old_name;
4185         struct inode *source = old_dentry->d_inode;
4186         struct inode *target = new_dentry->d_inode;
4187         bool new_is_dir = false;
4188         unsigned max_links = new_dir->i_sb->s_max_links;
4189 
4190         if (source == target)
4191                 return 0;
4192 
4193         error = may_delete(old_dir, old_dentry, is_dir);
4194         if (error)
4195                 return error;
4196 
4197         if (!target) {
4198                 error = may_create(new_dir, new_dentry);
4199         } else {
4200                 new_is_dir = d_is_dir(new_dentry);
4201 
4202                 if (!(flags & RENAME_EXCHANGE))
4203                         error = may_delete(new_dir, new_dentry, is_dir);
4204                 else
4205                         error = may_delete(new_dir, new_dentry, new_is_dir);
4206         }
4207         if (error)
4208                 return error;
4209 
4210         if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4211                 return -EPERM;
4212 
4213         if (flags && !old_dir->i_op->rename2)
4214                 return -EINVAL;
4215 
4216         /*
4217          * If we are going to change the parent - check write permissions,
4218          * we'll need to flip '..'.
4219          */
4220         if (new_dir != old_dir) {
4221                 if (is_dir) {
4222                         error = inode_permission(source, MAY_WRITE);
4223                         if (error)
4224                                 return error;
4225                 }
4226                 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4227                         error = inode_permission(target, MAY_WRITE);
4228                         if (error)
4229                                 return error;
4230                 }
4231         }
4232 
4233         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4234                                       flags);
4235         if (error)
4236                 return error;
4237 
4238         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4239         dget(new_dentry);
4240         if (!is_dir || (flags & RENAME_EXCHANGE))
4241                 lock_two_nondirectories(source, target);
4242         else if (target)
4243                 mutex_lock(&target->i_mutex);
4244 
4245         error = -EBUSY;
4246         if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4247                 goto out;
4248 
4249         if (max_links && new_dir != old_dir) {
4250                 error = -EMLINK;
4251                 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4252                         goto out;
4253                 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4254                     old_dir->i_nlink >= max_links)
4255                         goto out;
4256         }
4257         if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4258                 shrink_dcache_parent(new_dentry);
4259         if (!is_dir) {
4260                 error = try_break_deleg(source, delegated_inode);
4261                 if (error)
4262                         goto out;
4263         }
4264         if (target && !new_is_dir) {
4265                 error = try_break_deleg(target, delegated_inode);
4266                 if (error)
4267                         goto out;
4268         }
4269         if (!old_dir->i_op->rename2) {
4270                 error = old_dir->i_op->rename(old_dir, old_dentry,
4271                                               new_dir, new_dentry);
4272         } else {
4273                 WARN_ON(old_dir->i_op->rename != NULL);
4274                 error = old_dir->i_op->rename2(old_dir, old_dentry,
4275                                                new_dir, new_dentry, flags);
4276         }
4277         if (error)
4278                 goto out;
4279 
4280         if (!(flags & RENAME_EXCHANGE) && target) {
4281                 if (is_dir)
4282                         target->i_flags |= S_DEAD;
4283                 dont_mount(new_dentry);
4284                 detach_mounts(new_dentry);
4285         }
4286         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4287                 if (!(flags & RENAME_EXCHANGE))
4288                         d_move(old_dentry, new_dentry);
4289                 else
4290                         d_exchange(old_dentry, new_dentry);
4291         }
4292 out:
4293         if (!is_dir || (flags & RENAME_EXCHANGE))
4294                 unlock_two_nondirectories(source, target);
4295         else if (target)
4296                 mutex_unlock(&target->i_mutex);
4297         dput(new_dentry);
4298         if (!error) {
4299                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4300                               !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4301                 if (flags & RENAME_EXCHANGE) {
4302                         fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4303                                       new_is_dir, NULL, new_dentry);
4304                 }
4305         }
4306         fsnotify_oldname_free(old_name);
4307 
4308         return error;
4309 }
4310 EXPORT_SYMBOL(vfs_rename);
4311 
4312 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4313                 int, newdfd, const char __user *, newname, unsigned int, flags)
4314 {
4315         struct dentry *old_dentry, *new_dentry;
4316         struct dentry *trap;
4317         struct path old_path, new_path;
4318         struct qstr old_last, new_last;
4319         int old_type, new_type;
4320         struct inode *delegated_inode = NULL;
4321         struct filename *from;
4322         struct filename *to;
4323         unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4324         bool should_retry = false;
4325         int error;
4326 
4327         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4328                 return -EINVAL;
4329 
4330         if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4331             (flags & RENAME_EXCHANGE))
4332                 return -EINVAL;
4333 
4334         if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4335                 return -EPERM;
4336 
4337         if (flags & RENAME_EXCHANGE)
4338                 target_flags = 0;
4339 
4340 retry:
4341         from = user_path_parent(olddfd, oldname,
4342                                 &old_path, &old_last, &old_type, lookup_flags);
4343         if (IS_ERR(from)) {
4344                 error = PTR_ERR(from);
4345                 goto exit;
4346         }
4347 
4348         to = user_path_parent(newdfd, newname,
4349                                 &new_path, &new_last, &new_type, lookup_flags);
4350         if (IS_ERR(to)) {
4351                 error = PTR_ERR(to);
4352                 goto exit1;
4353         }
4354 
4355         error = -EXDEV;
4356         if (old_path.mnt != new_path.mnt)
4357                 goto exit2;
4358 
4359         error = -EBUSY;
4360         if (old_type != LAST_NORM)
4361                 goto exit2;
4362 
4363         if (flags & RENAME_NOREPLACE)
4364                 error = -EEXIST;
4365         if (new_type != LAST_NORM)
4366                 goto exit2;
4367 
4368         error = mnt_want_write(old_path.mnt);
4369         if (error)
4370                 goto exit2;
4371 
4372 retry_deleg:
4373         trap = lock_rename(new_path.dentry, old_path.dentry);
4374 
4375         old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4376         error = PTR_ERR(old_dentry);
4377         if (IS_ERR(old_dentry))
4378                 goto exit3;
4379         /* source must exist */
4380         error = -ENOENT;
4381         if (d_is_negative(old_dentry))
4382                 goto exit4;
4383         new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4384         error = PTR_ERR(new_dentry);
4385         if (IS_ERR(new_dentry))
4386                 goto exit4;
4387         error = -EEXIST;
4388         if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4389                 goto exit5;
4390         if (flags & RENAME_EXCHANGE) {
4391                 error = -ENOENT;
4392                 if (d_is_negative(new_dentry))
4393                         goto exit5;
4394 
4395                 if (!d_is_dir(new_dentry)) {
4396                         error = -ENOTDIR;
4397                         if (new_last.name[new_last.len])
4398                                 goto exit5;
4399                 }
4400         }
4401         /* unless the source is a directory trailing slashes give -ENOTDIR */
4402         if (!d_is_dir(old_dentry)) {
4403                 error = -ENOTDIR;
4404                 if (old_last.name[old_last.len])
4405                         goto exit5;
4406                 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4407                         goto exit5;
4408         }
4409         /* source should not be ancestor of target */
4410         error = -EINVAL;
4411         if (old_dentry == trap)
4412                 goto exit5;
4413         /* target should not be an ancestor of source */
4414         if (!(flags & RENAME_EXCHANGE))
4415                 error = -ENOTEMPTY;
4416         if (new_dentry == trap)
4417                 goto exit5;
4418 
4419         error = security_path_rename(&old_path, old_dentry,
4420                                      &new_path, new_dentry, flags);
4421         if (error)
4422                 goto exit5;
4423         error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4424                            new_path.dentry->d_inode, new_dentry,
4425                            &delegated_inode, flags);
4426 exit5:
4427         dput(new_dentry);
4428 exit4:
4429         dput(old_dentry);
4430 exit3:
4431         unlock_rename(new_path.dentry, old_path.dentry);
4432         if (delegated_inode) {
4433                 error = break_deleg_wait(&delegated_inode);
4434                 if (!error)
4435                         goto retry_deleg;
4436         }
4437         mnt_drop_write(old_path.mnt);
4438 exit2:
4439         if (retry_estale(error, lookup_flags))
4440                 should_retry = true;
4441         path_put(&new_path);
4442         putname(to);
4443 exit1:
4444         path_put(&old_path);
4445         putname(from);
4446         if (should_retry) {
4447                 should_retry = false;
4448                 lookup_flags |= LOOKUP_REVAL;
4449                 goto retry;
4450         }
4451 exit:
4452         return error;
4453 }
4454 
4455 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4456                 int, newdfd, const char __user *, newname)
4457 {
4458         return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4459 }
4460 
4461 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4462 {
4463         return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4464 }
4465 
4466 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4467 {
4468         int error = may_create(dir, dentry);
4469         if (error)
4470                 return error;
4471 
4472         if (!dir->i_op->mknod)
4473                 return -EPERM;
4474 
4475         return dir->i_op->mknod(dir, dentry,
4476                                 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4477 }
4478 EXPORT_SYMBOL(vfs_whiteout);
4479 
4480 int readlink_copy(char __user *buffer, int buflen, const char *link)
4481 {
4482         int len = PTR_ERR(link);
4483         if (IS_ERR(link))
4484                 goto out;
4485 
4486         len = strlen(link);
4487         if (len > (unsigned) buflen)
4488                 len = buflen;
4489         if (copy_to_user(buffer, link, len))
4490                 len = -EFAULT;
4491 out:
4492         return len;
4493 }
4494 EXPORT_SYMBOL(readlink_copy);
4495 
4496 /*
4497  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4498  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4499  * using) it for any given inode is up to filesystem.
4500  */
4501 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4502 {
4503         void *cookie;
4504         struct inode *inode = d_inode(dentry);
4505         const char *link = inode->i_link;
4506         int res;
4507 
4508         if (!link) {
4509                 link = inode->i_op->follow_link(dentry, &cookie);
4510                 if (IS_ERR(link))
4511                         return PTR_ERR(link);
4512         }
4513         res = readlink_copy(buffer, buflen, link);
4514         if (inode->i_op->put_link)
4515                 inode->i_op->put_link(inode, cookie);
4516         return res;
4517 }
4518 EXPORT_SYMBOL(generic_readlink);
4519 
4520 /* get the link contents into pagecache */
4521 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4522 {
4523         char *kaddr;
4524         struct page *page;
4525         struct address_space *mapping = dentry->d_inode->i_mapping;
4526         page = read_mapping_page(mapping, 0, NULL);
4527         if (IS_ERR(page))
4528                 return (char*)page;
4529         *ppage = page;
4530         kaddr = kmap(page);
4531         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4532         return kaddr;
4533 }
4534 
4535 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4536 {
4537         struct page *page = NULL;
4538         int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4539         if (page) {
4540                 kunmap(page);
4541                 page_cache_release(page);
4542         }
4543         return res;
4544 }
4545 EXPORT_SYMBOL(page_readlink);
4546 
4547 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4548 {
4549         struct page *page = NULL;
4550         char *res = page_getlink(dentry, &page);
4551         if (!IS_ERR(res))
4552                 *cookie = page;
4553         return res;
4554 }
4555 EXPORT_SYMBOL(page_follow_link_light);
4556 
4557 void page_put_link(struct inode *unused, void *cookie)
4558 {
4559         struct page *page = cookie;
4560         kunmap(page);
4561         page_cache_release(page);
4562 }
4563 EXPORT_SYMBOL(page_put_link);
4564 
4565 /*
4566  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4567  */
4568 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4569 {
4570         struct address_space *mapping = inode->i_mapping;
4571         struct page *page;
4572         void *fsdata;
4573         int err;
4574         char *kaddr;
4575         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4576         if (nofs)
4577                 flags |= AOP_FLAG_NOFS;
4578 
4579 retry:
4580         err = pagecache_write_begin(NULL, mapping, 0, len-1,
4581                                 flags, &page, &fsdata);
4582         if (err)
4583                 goto fail;
4584 
4585         kaddr = kmap_atomic(page);
4586         memcpy(kaddr, symname, len-1);
4587         kunmap_atomic(kaddr);
4588 
4589         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4590                                                         page, fsdata);
4591         if (err < 0)
4592                 goto fail;
4593         if (err < len-1)
4594                 goto retry;
4595 
4596         mark_inode_dirty(inode);
4597         return 0;
4598 fail:
4599         return err;
4600 }
4601 EXPORT_SYMBOL(__page_symlink);
4602 
4603 int page_symlink(struct inode *inode, const char *symname, int len)
4604 {
4605         return __page_symlink(inode, symname, len,
4606                         !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4607 }
4608 EXPORT_SYMBOL(page_symlink);
4609 
4610 const struct inode_operations page_symlink_inode_operations = {
4611         .readlink       = generic_readlink,
4612         .follow_link    = page_follow_link_light,
4613         .put_link       = page_put_link,
4614 };
4615 EXPORT_SYMBOL(page_symlink_inode_operations);
4616 

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