Version:  2.0.40 2.2.26 2.4.37 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8

Linux/fs/namei.c

  1 /*
  2  *  linux/fs/namei.c
  3  *
  4  *  Copyright (C) 1991, 1992  Linus Torvalds
  5  */
  6 
  7 /*
  8  * Some corrections by tytso.
  9  */
 10 
 11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
 12  * lookup logic.
 13  */
 14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
 15  */
 16 
 17 #include <linux/init.h>
 18 #include <linux/export.h>
 19 #include <linux/kernel.h>
 20 #include <linux/slab.h>
 21 #include <linux/fs.h>
 22 #include <linux/namei.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/fsnotify.h>
 25 #include <linux/personality.h>
 26 #include <linux/security.h>
 27 #include <linux/ima.h>
 28 #include <linux/syscalls.h>
 29 #include <linux/mount.h>
 30 #include <linux/audit.h>
 31 #include <linux/capability.h>
 32 #include <linux/file.h>
 33 #include <linux/fcntl.h>
 34 #include <linux/device_cgroup.h>
 35 #include <linux/fs_struct.h>
 36 #include <linux/posix_acl.h>
 37 #include <linux/hash.h>
 38 #include <linux/bitops.h>
 39 #include <linux/init_task.h>
 40 #include <asm/uaccess.h>
 41 
 42 #include "internal.h"
 43 #include "mount.h"
 44 
 45 /* [Feb-1997 T. Schoebel-Theuer]
 46  * Fundamental changes in the pathname lookup mechanisms (namei)
 47  * were necessary because of omirr.  The reason is that omirr needs
 48  * to know the _real_ pathname, not the user-supplied one, in case
 49  * of symlinks (and also when transname replacements occur).
 50  *
 51  * The new code replaces the old recursive symlink resolution with
 52  * an iterative one (in case of non-nested symlink chains).  It does
 53  * this with calls to <fs>_follow_link().
 54  * As a side effect, dir_namei(), _namei() and follow_link() are now 
 55  * replaced with a single function lookup_dentry() that can handle all 
 56  * the special cases of the former code.
 57  *
 58  * With the new dcache, the pathname is stored at each inode, at least as
 59  * long as the refcount of the inode is positive.  As a side effect, the
 60  * size of the dcache depends on the inode cache and thus is dynamic.
 61  *
 62  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
 63  * resolution to correspond with current state of the code.
 64  *
 65  * Note that the symlink resolution is not *completely* iterative.
 66  * There is still a significant amount of tail- and mid- recursion in
 67  * the algorithm.  Also, note that <fs>_readlink() is not used in
 68  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
 69  * may return different results than <fs>_follow_link().  Many virtual
 70  * filesystems (including /proc) exhibit this behavior.
 71  */
 72 
 73 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
 74  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
 75  * and the name already exists in form of a symlink, try to create the new
 76  * name indicated by the symlink. The old code always complained that the
 77  * name already exists, due to not following the symlink even if its target
 78  * is nonexistent.  The new semantics affects also mknod() and link() when
 79  * the name is a symlink pointing to a non-existent name.
 80  *
 81  * I don't know which semantics is the right one, since I have no access
 82  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
 83  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
 84  * "old" one. Personally, I think the new semantics is much more logical.
 85  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
 86  * file does succeed in both HP-UX and SunOs, but not in Solaris
 87  * and in the old Linux semantics.
 88  */
 89 
 90 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
 91  * semantics.  See the comments in "open_namei" and "do_link" below.
 92  *
 93  * [10-Sep-98 Alan Modra] Another symlink change.
 94  */
 95 
 96 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
 97  *      inside the path - always follow.
 98  *      in the last component in creation/removal/renaming - never follow.
 99  *      if LOOKUP_FOLLOW passed - follow.
100  *      if the pathname has trailing slashes - follow.
101  *      otherwise - don't follow.
102  * (applied in that order).
103  *
104  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106  * During the 2.4 we need to fix the userland stuff depending on it -
107  * hopefully we will be able to get rid of that wart in 2.5. So far only
108  * XEmacs seems to be relying on it...
109  */
110 /*
111  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
112  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
113  * any extra contention...
114  */
115 
116 /* In order to reduce some races, while at the same time doing additional
117  * checking and hopefully speeding things up, we copy filenames to the
118  * kernel data space before using them..
119  *
120  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121  * PATH_MAX includes the nul terminator --RR.
122  */
123 
124 #define EMBEDDED_NAME_MAX       (PATH_MAX - offsetof(struct filename, iname))
125 
126 struct filename *
127 getname_flags(const char __user *filename, int flags, int *empty)
128 {
129         struct filename *result;
130         char *kname;
131         int len;
132 
133         result = audit_reusename(filename);
134         if (result)
135                 return result;
136 
137         result = __getname();
138         if (unlikely(!result))
139                 return ERR_PTR(-ENOMEM);
140 
141         /*
142          * First, try to embed the struct filename inside the names_cache
143          * allocation
144          */
145         kname = (char *)result->iname;
146         result->name = kname;
147 
148         len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
149         if (unlikely(len < 0)) {
150                 __putname(result);
151                 return ERR_PTR(len);
152         }
153 
154         /*
155          * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156          * separate struct filename so we can dedicate the entire
157          * names_cache allocation for the pathname, and re-do the copy from
158          * userland.
159          */
160         if (unlikely(len == EMBEDDED_NAME_MAX)) {
161                 const size_t size = offsetof(struct filename, iname[1]);
162                 kname = (char *)result;
163 
164                 /*
165                  * size is chosen that way we to guarantee that
166                  * result->iname[0] is within the same object and that
167                  * kname can't be equal to result->iname, no matter what.
168                  */
169                 result = kzalloc(size, GFP_KERNEL);
170                 if (unlikely(!result)) {
171                         __putname(kname);
172                         return ERR_PTR(-ENOMEM);
173                 }
174                 result->name = kname;
175                 len = strncpy_from_user(kname, filename, PATH_MAX);
176                 if (unlikely(len < 0)) {
177                         __putname(kname);
178                         kfree(result);
179                         return ERR_PTR(len);
180                 }
181                 if (unlikely(len == PATH_MAX)) {
182                         __putname(kname);
183                         kfree(result);
184                         return ERR_PTR(-ENAMETOOLONG);
185                 }
186         }
187 
188         result->refcnt = 1;
189         /* The empty path is special. */
190         if (unlikely(!len)) {
191                 if (empty)
192                         *empty = 1;
193                 if (!(flags & LOOKUP_EMPTY)) {
194                         putname(result);
195                         return ERR_PTR(-ENOENT);
196                 }
197         }
198 
199         result->uptr = filename;
200         result->aname = NULL;
201         audit_getname(result);
202         return result;
203 }
204 
205 struct filename *
206 getname(const char __user * filename)
207 {
208         return getname_flags(filename, 0, NULL);
209 }
210 
211 struct filename *
212 getname_kernel(const char * filename)
213 {
214         struct filename *result;
215         int len = strlen(filename) + 1;
216 
217         result = __getname();
218         if (unlikely(!result))
219                 return ERR_PTR(-ENOMEM);
220 
221         if (len <= EMBEDDED_NAME_MAX) {
222                 result->name = (char *)result->iname;
223         } else if (len <= PATH_MAX) {
224                 struct filename *tmp;
225 
226                 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227                 if (unlikely(!tmp)) {
228                         __putname(result);
229                         return ERR_PTR(-ENOMEM);
230                 }
231                 tmp->name = (char *)result;
232                 result = tmp;
233         } else {
234                 __putname(result);
235                 return ERR_PTR(-ENAMETOOLONG);
236         }
237         memcpy((char *)result->name, filename, len);
238         result->uptr = NULL;
239         result->aname = NULL;
240         result->refcnt = 1;
241         audit_getname(result);
242 
243         return result;
244 }
245 
246 void putname(struct filename *name)
247 {
248         BUG_ON(name->refcnt <= 0);
249 
250         if (--name->refcnt > 0)
251                 return;
252 
253         if (name->name != name->iname) {
254                 __putname(name->name);
255                 kfree(name);
256         } else
257                 __putname(name);
258 }
259 
260 static int check_acl(struct inode *inode, int mask)
261 {
262 #ifdef CONFIG_FS_POSIX_ACL
263         struct posix_acl *acl;
264 
265         if (mask & MAY_NOT_BLOCK) {
266                 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267                 if (!acl)
268                         return -EAGAIN;
269                 /* no ->get_acl() calls in RCU mode... */
270                 if (is_uncached_acl(acl))
271                         return -ECHILD;
272                 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273         }
274 
275         acl = get_acl(inode, ACL_TYPE_ACCESS);
276         if (IS_ERR(acl))
277                 return PTR_ERR(acl);
278         if (acl) {
279                 int error = posix_acl_permission(inode, acl, mask);
280                 posix_acl_release(acl);
281                 return error;
282         }
283 #endif
284 
285         return -EAGAIN;
286 }
287 
288 /*
289  * This does the basic permission checking
290  */
291 static int acl_permission_check(struct inode *inode, int mask)
292 {
293         unsigned int mode = inode->i_mode;
294 
295         if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296                 mode >>= 6;
297         else {
298                 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
299                         int error = check_acl(inode, mask);
300                         if (error != -EAGAIN)
301                                 return error;
302                 }
303 
304                 if (in_group_p(inode->i_gid))
305                         mode >>= 3;
306         }
307 
308         /*
309          * If the DACs are ok we don't need any capability check.
310          */
311         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
312                 return 0;
313         return -EACCES;
314 }
315 
316 /**
317  * generic_permission -  check for access rights on a Posix-like filesystem
318  * @inode:      inode to check access rights for
319  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
320  *
321  * Used to check for read/write/execute permissions on a file.
322  * We use "fsuid" for this, letting us set arbitrary permissions
323  * for filesystem access without changing the "normal" uids which
324  * are used for other things.
325  *
326  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327  * request cannot be satisfied (eg. requires blocking or too much complexity).
328  * It would then be called again in ref-walk mode.
329  */
330 int generic_permission(struct inode *inode, int mask)
331 {
332         int ret;
333 
334         /*
335          * Do the basic permission checks.
336          */
337         ret = acl_permission_check(inode, mask);
338         if (ret != -EACCES)
339                 return ret;
340 
341         if (S_ISDIR(inode->i_mode)) {
342                 /* DACs are overridable for directories */
343                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
344                         return 0;
345                 if (!(mask & MAY_WRITE))
346                         if (capable_wrt_inode_uidgid(inode,
347                                                      CAP_DAC_READ_SEARCH))
348                                 return 0;
349                 return -EACCES;
350         }
351         /*
352          * Read/write DACs are always overridable.
353          * Executable DACs are overridable when there is
354          * at least one exec bit set.
355          */
356         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
357                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
358                         return 0;
359 
360         /*
361          * Searching includes executable on directories, else just read.
362          */
363         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
364         if (mask == MAY_READ)
365                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
366                         return 0;
367 
368         return -EACCES;
369 }
370 EXPORT_SYMBOL(generic_permission);
371 
372 /*
373  * We _really_ want to just do "generic_permission()" without
374  * even looking at the inode->i_op values. So we keep a cache
375  * flag in inode->i_opflags, that says "this has not special
376  * permission function, use the fast case".
377  */
378 static inline int do_inode_permission(struct inode *inode, int mask)
379 {
380         if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381                 if (likely(inode->i_op->permission))
382                         return inode->i_op->permission(inode, mask);
383 
384                 /* This gets set once for the inode lifetime */
385                 spin_lock(&inode->i_lock);
386                 inode->i_opflags |= IOP_FASTPERM;
387                 spin_unlock(&inode->i_lock);
388         }
389         return generic_permission(inode, mask);
390 }
391 
392 /**
393  * __inode_permission - Check for access rights to a given inode
394  * @inode: Inode to check permission on
395  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
396  *
397  * Check for read/write/execute permissions on an inode.
398  *
399  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
400  *
401  * This does not check for a read-only file system.  You probably want
402  * inode_permission().
403  */
404 int __inode_permission(struct inode *inode, int mask)
405 {
406         int retval;
407 
408         if (unlikely(mask & MAY_WRITE)) {
409                 /*
410                  * Nobody gets write access to an immutable file.
411                  */
412                 if (IS_IMMUTABLE(inode))
413                         return -EPERM;
414 
415                 /*
416                  * Updating mtime will likely cause i_uid and i_gid to be
417                  * written back improperly if their true value is unknown
418                  * to the vfs.
419                  */
420                 if (HAS_UNMAPPED_ID(inode))
421                         return -EACCES;
422         }
423 
424         retval = do_inode_permission(inode, mask);
425         if (retval)
426                 return retval;
427 
428         retval = devcgroup_inode_permission(inode, mask);
429         if (retval)
430                 return retval;
431 
432         return security_inode_permission(inode, mask);
433 }
434 EXPORT_SYMBOL(__inode_permission);
435 
436 /**
437  * sb_permission - Check superblock-level permissions
438  * @sb: Superblock of inode to check permission on
439  * @inode: Inode to check permission on
440  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
441  *
442  * Separate out file-system wide checks from inode-specific permission checks.
443  */
444 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
445 {
446         if (unlikely(mask & MAY_WRITE)) {
447                 umode_t mode = inode->i_mode;
448 
449                 /* Nobody gets write access to a read-only fs. */
450                 if ((sb->s_flags & MS_RDONLY) &&
451                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
452                         return -EROFS;
453         }
454         return 0;
455 }
456 
457 /**
458  * inode_permission - Check for access rights to a given inode
459  * @inode: Inode to check permission on
460  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461  *
462  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
463  * this, letting us set arbitrary permissions for filesystem access without
464  * changing the "normal" UIDs which are used for other things.
465  *
466  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
467  */
468 int inode_permission(struct inode *inode, int mask)
469 {
470         int retval;
471 
472         retval = sb_permission(inode->i_sb, inode, mask);
473         if (retval)
474                 return retval;
475         return __inode_permission(inode, mask);
476 }
477 EXPORT_SYMBOL(inode_permission);
478 
479 /**
480  * path_get - get a reference to a path
481  * @path: path to get the reference to
482  *
483  * Given a path increment the reference count to the dentry and the vfsmount.
484  */
485 void path_get(const struct path *path)
486 {
487         mntget(path->mnt);
488         dget(path->dentry);
489 }
490 EXPORT_SYMBOL(path_get);
491 
492 /**
493  * path_put - put a reference to a path
494  * @path: path to put the reference to
495  *
496  * Given a path decrement the reference count to the dentry and the vfsmount.
497  */
498 void path_put(const struct path *path)
499 {
500         dput(path->dentry);
501         mntput(path->mnt);
502 }
503 EXPORT_SYMBOL(path_put);
504 
505 #define EMBEDDED_LEVELS 2
506 struct nameidata {
507         struct path     path;
508         struct qstr     last;
509         struct path     root;
510         struct inode    *inode; /* path.dentry.d_inode */
511         unsigned int    flags;
512         unsigned        seq, m_seq;
513         int             last_type;
514         unsigned        depth;
515         int             total_link_count;
516         struct saved {
517                 struct path link;
518                 struct delayed_call done;
519                 const char *name;
520                 unsigned seq;
521         } *stack, internal[EMBEDDED_LEVELS];
522         struct filename *name;
523         struct nameidata *saved;
524         struct inode    *link_inode;
525         unsigned        root_seq;
526         int             dfd;
527 };
528 
529 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
530 {
531         struct nameidata *old = current->nameidata;
532         p->stack = p->internal;
533         p->dfd = dfd;
534         p->name = name;
535         p->total_link_count = old ? old->total_link_count : 0;
536         p->saved = old;
537         current->nameidata = p;
538 }
539 
540 static void restore_nameidata(void)
541 {
542         struct nameidata *now = current->nameidata, *old = now->saved;
543 
544         current->nameidata = old;
545         if (old)
546                 old->total_link_count = now->total_link_count;
547         if (now->stack != now->internal)
548                 kfree(now->stack);
549 }
550 
551 static int __nd_alloc_stack(struct nameidata *nd)
552 {
553         struct saved *p;
554 
555         if (nd->flags & LOOKUP_RCU) {
556                 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
557                                   GFP_ATOMIC);
558                 if (unlikely(!p))
559                         return -ECHILD;
560         } else {
561                 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
562                                   GFP_KERNEL);
563                 if (unlikely(!p))
564                         return -ENOMEM;
565         }
566         memcpy(p, nd->internal, sizeof(nd->internal));
567         nd->stack = p;
568         return 0;
569 }
570 
571 /**
572  * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573  * @path: nameidate to verify
574  *
575  * Rename can sometimes move a file or directory outside of a bind
576  * mount, path_connected allows those cases to be detected.
577  */
578 static bool path_connected(const struct path *path)
579 {
580         struct vfsmount *mnt = path->mnt;
581 
582         /* Only bind mounts can have disconnected paths */
583         if (mnt->mnt_root == mnt->mnt_sb->s_root)
584                 return true;
585 
586         return is_subdir(path->dentry, mnt->mnt_root);
587 }
588 
589 static inline int nd_alloc_stack(struct nameidata *nd)
590 {
591         if (likely(nd->depth != EMBEDDED_LEVELS))
592                 return 0;
593         if (likely(nd->stack != nd->internal))
594                 return 0;
595         return __nd_alloc_stack(nd);
596 }
597 
598 static void drop_links(struct nameidata *nd)
599 {
600         int i = nd->depth;
601         while (i--) {
602                 struct saved *last = nd->stack + i;
603                 do_delayed_call(&last->done);
604                 clear_delayed_call(&last->done);
605         }
606 }
607 
608 static void terminate_walk(struct nameidata *nd)
609 {
610         drop_links(nd);
611         if (!(nd->flags & LOOKUP_RCU)) {
612                 int i;
613                 path_put(&nd->path);
614                 for (i = 0; i < nd->depth; i++)
615                         path_put(&nd->stack[i].link);
616                 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
617                         path_put(&nd->root);
618                         nd->root.mnt = NULL;
619                 }
620         } else {
621                 nd->flags &= ~LOOKUP_RCU;
622                 if (!(nd->flags & LOOKUP_ROOT))
623                         nd->root.mnt = NULL;
624                 rcu_read_unlock();
625         }
626         nd->depth = 0;
627 }
628 
629 /* path_put is needed afterwards regardless of success or failure */
630 static bool legitimize_path(struct nameidata *nd,
631                             struct path *path, unsigned seq)
632 {
633         int res = __legitimize_mnt(path->mnt, nd->m_seq);
634         if (unlikely(res)) {
635                 if (res > 0)
636                         path->mnt = NULL;
637                 path->dentry = NULL;
638                 return false;
639         }
640         if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
641                 path->dentry = NULL;
642                 return false;
643         }
644         return !read_seqcount_retry(&path->dentry->d_seq, seq);
645 }
646 
647 static bool legitimize_links(struct nameidata *nd)
648 {
649         int i;
650         for (i = 0; i < nd->depth; i++) {
651                 struct saved *last = nd->stack + i;
652                 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653                         drop_links(nd);
654                         nd->depth = i + 1;
655                         return false;
656                 }
657         }
658         return true;
659 }
660 
661 /*
662  * Path walking has 2 modes, rcu-walk and ref-walk (see
663  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
664  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
665  * normal reference counts on dentries and vfsmounts to transition to ref-walk
666  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
667  * got stuck, so ref-walk may continue from there. If this is not successful
668  * (eg. a seqcount has changed), then failure is returned and it's up to caller
669  * to restart the path walk from the beginning in ref-walk mode.
670  */
671 
672 /**
673  * unlazy_walk - try to switch to ref-walk mode.
674  * @nd: nameidata pathwalk data
675  * @dentry: child of nd->path.dentry or NULL
676  * @seq: seq number to check dentry against
677  * Returns: 0 on success, -ECHILD on failure
678  *
679  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
680  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
681  * @nd or NULL.  Must be called from rcu-walk context.
682  * Nothing should touch nameidata between unlazy_walk() failure and
683  * terminate_walk().
684  */
685 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
686 {
687         struct dentry *parent = nd->path.dentry;
688 
689         BUG_ON(!(nd->flags & LOOKUP_RCU));
690 
691         nd->flags &= ~LOOKUP_RCU;
692         if (unlikely(!legitimize_links(nd)))
693                 goto out2;
694         if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
695                 goto out2;
696         if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
697                 goto out1;
698 
699         /*
700          * For a negative lookup, the lookup sequence point is the parents
701          * sequence point, and it only needs to revalidate the parent dentry.
702          *
703          * For a positive lookup, we need to move both the parent and the
704          * dentry from the RCU domain to be properly refcounted. And the
705          * sequence number in the dentry validates *both* dentry counters,
706          * since we checked the sequence number of the parent after we got
707          * the child sequence number. So we know the parent must still
708          * be valid if the child sequence number is still valid.
709          */
710         if (!dentry) {
711                 if (read_seqcount_retry(&parent->d_seq, nd->seq))
712                         goto out;
713                 BUG_ON(nd->inode != parent->d_inode);
714         } else {
715                 if (!lockref_get_not_dead(&dentry->d_lockref))
716                         goto out;
717                 if (read_seqcount_retry(&dentry->d_seq, seq))
718                         goto drop_dentry;
719         }
720 
721         /*
722          * Sequence counts matched. Now make sure that the root is
723          * still valid and get it if required.
724          */
725         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
726                 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
727                         rcu_read_unlock();
728                         dput(dentry);
729                         return -ECHILD;
730                 }
731         }
732 
733         rcu_read_unlock();
734         return 0;
735 
736 drop_dentry:
737         rcu_read_unlock();
738         dput(dentry);
739         goto drop_root_mnt;
740 out2:
741         nd->path.mnt = NULL;
742 out1:
743         nd->path.dentry = NULL;
744 out:
745         rcu_read_unlock();
746 drop_root_mnt:
747         if (!(nd->flags & LOOKUP_ROOT))
748                 nd->root.mnt = NULL;
749         return -ECHILD;
750 }
751 
752 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
753 {
754         if (unlikely(!legitimize_path(nd, link, seq))) {
755                 drop_links(nd);
756                 nd->depth = 0;
757                 nd->flags &= ~LOOKUP_RCU;
758                 nd->path.mnt = NULL;
759                 nd->path.dentry = NULL;
760                 if (!(nd->flags & LOOKUP_ROOT))
761                         nd->root.mnt = NULL;
762                 rcu_read_unlock();
763         } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
764                 return 0;
765         }
766         path_put(link);
767         return -ECHILD;
768 }
769 
770 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
771 {
772         return dentry->d_op->d_revalidate(dentry, flags);
773 }
774 
775 /**
776  * complete_walk - successful completion of path walk
777  * @nd:  pointer nameidata
778  *
779  * If we had been in RCU mode, drop out of it and legitimize nd->path.
780  * Revalidate the final result, unless we'd already done that during
781  * the path walk or the filesystem doesn't ask for it.  Return 0 on
782  * success, -error on failure.  In case of failure caller does not
783  * need to drop nd->path.
784  */
785 static int complete_walk(struct nameidata *nd)
786 {
787         struct dentry *dentry = nd->path.dentry;
788         int status;
789 
790         if (nd->flags & LOOKUP_RCU) {
791                 if (!(nd->flags & LOOKUP_ROOT))
792                         nd->root.mnt = NULL;
793                 if (unlikely(unlazy_walk(nd, NULL, 0)))
794                         return -ECHILD;
795         }
796 
797         if (likely(!(nd->flags & LOOKUP_JUMPED)))
798                 return 0;
799 
800         if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
801                 return 0;
802 
803         status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
804         if (status > 0)
805                 return 0;
806 
807         if (!status)
808                 status = -ESTALE;
809 
810         return status;
811 }
812 
813 static void set_root(struct nameidata *nd)
814 {
815         struct fs_struct *fs = current->fs;
816 
817         if (nd->flags & LOOKUP_RCU) {
818                 unsigned seq;
819 
820                 do {
821                         seq = read_seqcount_begin(&fs->seq);
822                         nd->root = fs->root;
823                         nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
824                 } while (read_seqcount_retry(&fs->seq, seq));
825         } else {
826                 get_fs_root(fs, &nd->root);
827         }
828 }
829 
830 static void path_put_conditional(struct path *path, struct nameidata *nd)
831 {
832         dput(path->dentry);
833         if (path->mnt != nd->path.mnt)
834                 mntput(path->mnt);
835 }
836 
837 static inline void path_to_nameidata(const struct path *path,
838                                         struct nameidata *nd)
839 {
840         if (!(nd->flags & LOOKUP_RCU)) {
841                 dput(nd->path.dentry);
842                 if (nd->path.mnt != path->mnt)
843                         mntput(nd->path.mnt);
844         }
845         nd->path.mnt = path->mnt;
846         nd->path.dentry = path->dentry;
847 }
848 
849 static int nd_jump_root(struct nameidata *nd)
850 {
851         if (nd->flags & LOOKUP_RCU) {
852                 struct dentry *d;
853                 nd->path = nd->root;
854                 d = nd->path.dentry;
855                 nd->inode = d->d_inode;
856                 nd->seq = nd->root_seq;
857                 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
858                         return -ECHILD;
859         } else {
860                 path_put(&nd->path);
861                 nd->path = nd->root;
862                 path_get(&nd->path);
863                 nd->inode = nd->path.dentry->d_inode;
864         }
865         nd->flags |= LOOKUP_JUMPED;
866         return 0;
867 }
868 
869 /*
870  * Helper to directly jump to a known parsed path from ->get_link,
871  * caller must have taken a reference to path beforehand.
872  */
873 void nd_jump_link(struct path *path)
874 {
875         struct nameidata *nd = current->nameidata;
876         path_put(&nd->path);
877 
878         nd->path = *path;
879         nd->inode = nd->path.dentry->d_inode;
880         nd->flags |= LOOKUP_JUMPED;
881 }
882 
883 static inline void put_link(struct nameidata *nd)
884 {
885         struct saved *last = nd->stack + --nd->depth;
886         do_delayed_call(&last->done);
887         if (!(nd->flags & LOOKUP_RCU))
888                 path_put(&last->link);
889 }
890 
891 int sysctl_protected_symlinks __read_mostly = 0;
892 int sysctl_protected_hardlinks __read_mostly = 0;
893 
894 /**
895  * may_follow_link - Check symlink following for unsafe situations
896  * @nd: nameidata pathwalk data
897  *
898  * In the case of the sysctl_protected_symlinks sysctl being enabled,
899  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
900  * in a sticky world-writable directory. This is to protect privileged
901  * processes from failing races against path names that may change out
902  * from under them by way of other users creating malicious symlinks.
903  * It will permit symlinks to be followed only when outside a sticky
904  * world-writable directory, or when the uid of the symlink and follower
905  * match, or when the directory owner matches the symlink's owner.
906  *
907  * Returns 0 if following the symlink is allowed, -ve on error.
908  */
909 static inline int may_follow_link(struct nameidata *nd)
910 {
911         const struct inode *inode;
912         const struct inode *parent;
913         kuid_t puid;
914 
915         if (!sysctl_protected_symlinks)
916                 return 0;
917 
918         /* Allowed if owner and follower match. */
919         inode = nd->link_inode;
920         if (uid_eq(current_cred()->fsuid, inode->i_uid))
921                 return 0;
922 
923         /* Allowed if parent directory not sticky and world-writable. */
924         parent = nd->inode;
925         if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
926                 return 0;
927 
928         /* Allowed if parent directory and link owner match. */
929         puid = parent->i_uid;
930         if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
931                 return 0;
932 
933         if (nd->flags & LOOKUP_RCU)
934                 return -ECHILD;
935 
936         audit_log_link_denied("follow_link", &nd->stack[0].link);
937         return -EACCES;
938 }
939 
940 /**
941  * safe_hardlink_source - Check for safe hardlink conditions
942  * @inode: the source inode to hardlink from
943  *
944  * Return false if at least one of the following conditions:
945  *    - inode is not a regular file
946  *    - inode is setuid
947  *    - inode is setgid and group-exec
948  *    - access failure for read and write
949  *
950  * Otherwise returns true.
951  */
952 static bool safe_hardlink_source(struct inode *inode)
953 {
954         umode_t mode = inode->i_mode;
955 
956         /* Special files should not get pinned to the filesystem. */
957         if (!S_ISREG(mode))
958                 return false;
959 
960         /* Setuid files should not get pinned to the filesystem. */
961         if (mode & S_ISUID)
962                 return false;
963 
964         /* Executable setgid files should not get pinned to the filesystem. */
965         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
966                 return false;
967 
968         /* Hardlinking to unreadable or unwritable sources is dangerous. */
969         if (inode_permission(inode, MAY_READ | MAY_WRITE))
970                 return false;
971 
972         return true;
973 }
974 
975 /**
976  * may_linkat - Check permissions for creating a hardlink
977  * @link: the source to hardlink from
978  *
979  * Block hardlink when all of:
980  *  - sysctl_protected_hardlinks enabled
981  *  - fsuid does not match inode
982  *  - hardlink source is unsafe (see safe_hardlink_source() above)
983  *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
984  *
985  * Returns 0 if successful, -ve on error.
986  */
987 static int may_linkat(struct path *link)
988 {
989         struct inode *inode;
990 
991         if (!sysctl_protected_hardlinks)
992                 return 0;
993 
994         inode = link->dentry->d_inode;
995 
996         /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997          * otherwise, it must be a safe source.
998          */
999         if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
1000                 return 0;
1001 
1002         audit_log_link_denied("linkat", link);
1003         return -EPERM;
1004 }
1005 
1006 static __always_inline
1007 const char *get_link(struct nameidata *nd)
1008 {
1009         struct saved *last = nd->stack + nd->depth - 1;
1010         struct dentry *dentry = last->link.dentry;
1011         struct inode *inode = nd->link_inode;
1012         int error;
1013         const char *res;
1014 
1015         if (!(nd->flags & LOOKUP_RCU)) {
1016                 touch_atime(&last->link);
1017                 cond_resched();
1018         } else if (atime_needs_update(&last->link, inode)) {
1019                 if (unlikely(unlazy_walk(nd, NULL, 0)))
1020                         return ERR_PTR(-ECHILD);
1021                 touch_atime(&last->link);
1022         }
1023 
1024         error = security_inode_follow_link(dentry, inode,
1025                                            nd->flags & LOOKUP_RCU);
1026         if (unlikely(error))
1027                 return ERR_PTR(error);
1028 
1029         nd->last_type = LAST_BIND;
1030         res = inode->i_link;
1031         if (!res) {
1032                 const char * (*get)(struct dentry *, struct inode *,
1033                                 struct delayed_call *);
1034                 get = inode->i_op->get_link;
1035                 if (nd->flags & LOOKUP_RCU) {
1036                         res = get(NULL, inode, &last->done);
1037                         if (res == ERR_PTR(-ECHILD)) {
1038                                 if (unlikely(unlazy_walk(nd, NULL, 0)))
1039                                         return ERR_PTR(-ECHILD);
1040                                 res = get(dentry, inode, &last->done);
1041                         }
1042                 } else {
1043                         res = get(dentry, inode, &last->done);
1044                 }
1045                 if (IS_ERR_OR_NULL(res))
1046                         return res;
1047         }
1048         if (*res == '/') {
1049                 if (!nd->root.mnt)
1050                         set_root(nd);
1051                 if (unlikely(nd_jump_root(nd)))
1052                         return ERR_PTR(-ECHILD);
1053                 while (unlikely(*++res == '/'))
1054                         ;
1055         }
1056         if (!*res)
1057                 res = NULL;
1058         return res;
1059 }
1060 
1061 /*
1062  * follow_up - Find the mountpoint of path's vfsmount
1063  *
1064  * Given a path, find the mountpoint of its source file system.
1065  * Replace @path with the path of the mountpoint in the parent mount.
1066  * Up is towards /.
1067  *
1068  * Return 1 if we went up a level and 0 if we were already at the
1069  * root.
1070  */
1071 int follow_up(struct path *path)
1072 {
1073         struct mount *mnt = real_mount(path->mnt);
1074         struct mount *parent;
1075         struct dentry *mountpoint;
1076 
1077         read_seqlock_excl(&mount_lock);
1078         parent = mnt->mnt_parent;
1079         if (parent == mnt) {
1080                 read_sequnlock_excl(&mount_lock);
1081                 return 0;
1082         }
1083         mntget(&parent->mnt);
1084         mountpoint = dget(mnt->mnt_mountpoint);
1085         read_sequnlock_excl(&mount_lock);
1086         dput(path->dentry);
1087         path->dentry = mountpoint;
1088         mntput(path->mnt);
1089         path->mnt = &parent->mnt;
1090         return 1;
1091 }
1092 EXPORT_SYMBOL(follow_up);
1093 
1094 /*
1095  * Perform an automount
1096  * - return -EISDIR to tell follow_managed() to stop and return the path we
1097  *   were called with.
1098  */
1099 static int follow_automount(struct path *path, struct nameidata *nd,
1100                             bool *need_mntput)
1101 {
1102         struct vfsmount *mnt;
1103         const struct cred *old_cred;
1104         int err;
1105 
1106         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1107                 return -EREMOTE;
1108 
1109         /* We don't want to mount if someone's just doing a stat -
1110          * unless they're stat'ing a directory and appended a '/' to
1111          * the name.
1112          *
1113          * We do, however, want to mount if someone wants to open or
1114          * create a file of any type under the mountpoint, wants to
1115          * traverse through the mountpoint or wants to open the
1116          * mounted directory.  Also, autofs may mark negative dentries
1117          * as being automount points.  These will need the attentions
1118          * of the daemon to instantiate them before they can be used.
1119          */
1120         if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1121                            LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1122             path->dentry->d_inode)
1123                 return -EISDIR;
1124 
1125         if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1126                 return -EACCES;
1127 
1128         nd->total_link_count++;
1129         if (nd->total_link_count >= 40)
1130                 return -ELOOP;
1131 
1132         old_cred = override_creds(&init_cred);
1133         mnt = path->dentry->d_op->d_automount(path);
1134         revert_creds(old_cred);
1135         if (IS_ERR(mnt)) {
1136                 /*
1137                  * The filesystem is allowed to return -EISDIR here to indicate
1138                  * it doesn't want to automount.  For instance, autofs would do
1139                  * this so that its userspace daemon can mount on this dentry.
1140                  *
1141                  * However, we can only permit this if it's a terminal point in
1142                  * the path being looked up; if it wasn't then the remainder of
1143                  * the path is inaccessible and we should say so.
1144                  */
1145                 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1146                         return -EREMOTE;
1147                 return PTR_ERR(mnt);
1148         }
1149 
1150         if (!mnt) /* mount collision */
1151                 return 0;
1152 
1153         if (!*need_mntput) {
1154                 /* lock_mount() may release path->mnt on error */
1155                 mntget(path->mnt);
1156                 *need_mntput = true;
1157         }
1158         err = finish_automount(mnt, path);
1159 
1160         switch (err) {
1161         case -EBUSY:
1162                 /* Someone else made a mount here whilst we were busy */
1163                 return 0;
1164         case 0:
1165                 path_put(path);
1166                 path->mnt = mnt;
1167                 path->dentry = dget(mnt->mnt_root);
1168                 return 0;
1169         default:
1170                 return err;
1171         }
1172 
1173 }
1174 
1175 /*
1176  * Handle a dentry that is managed in some way.
1177  * - Flagged for transit management (autofs)
1178  * - Flagged as mountpoint
1179  * - Flagged as automount point
1180  *
1181  * This may only be called in refwalk mode.
1182  *
1183  * Serialization is taken care of in namespace.c
1184  */
1185 static int follow_managed(struct path *path, struct nameidata *nd)
1186 {
1187         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1188         unsigned managed;
1189         bool need_mntput = false;
1190         int ret = 0;
1191 
1192         /* Given that we're not holding a lock here, we retain the value in a
1193          * local variable for each dentry as we look at it so that we don't see
1194          * the components of that value change under us */
1195         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1196                managed &= DCACHE_MANAGED_DENTRY,
1197                unlikely(managed != 0)) {
1198                 /* Allow the filesystem to manage the transit without i_mutex
1199                  * being held. */
1200                 if (managed & DCACHE_MANAGE_TRANSIT) {
1201                         BUG_ON(!path->dentry->d_op);
1202                         BUG_ON(!path->dentry->d_op->d_manage);
1203                         ret = path->dentry->d_op->d_manage(path->dentry, false);
1204                         if (ret < 0)
1205                                 break;
1206                 }
1207 
1208                 /* Transit to a mounted filesystem. */
1209                 if (managed & DCACHE_MOUNTED) {
1210                         struct vfsmount *mounted = lookup_mnt(path);
1211                         if (mounted) {
1212                                 dput(path->dentry);
1213                                 if (need_mntput)
1214                                         mntput(path->mnt);
1215                                 path->mnt = mounted;
1216                                 path->dentry = dget(mounted->mnt_root);
1217                                 need_mntput = true;
1218                                 continue;
1219                         }
1220 
1221                         /* Something is mounted on this dentry in another
1222                          * namespace and/or whatever was mounted there in this
1223                          * namespace got unmounted before lookup_mnt() could
1224                          * get it */
1225                 }
1226 
1227                 /* Handle an automount point */
1228                 if (managed & DCACHE_NEED_AUTOMOUNT) {
1229                         ret = follow_automount(path, nd, &need_mntput);
1230                         if (ret < 0)
1231                                 break;
1232                         continue;
1233                 }
1234 
1235                 /* We didn't change the current path point */
1236                 break;
1237         }
1238 
1239         if (need_mntput && path->mnt == mnt)
1240                 mntput(path->mnt);
1241         if (ret == -EISDIR || !ret)
1242                 ret = 1;
1243         if (need_mntput)
1244                 nd->flags |= LOOKUP_JUMPED;
1245         if (unlikely(ret < 0))
1246                 path_put_conditional(path, nd);
1247         return ret;
1248 }
1249 
1250 int follow_down_one(struct path *path)
1251 {
1252         struct vfsmount *mounted;
1253 
1254         mounted = lookup_mnt(path);
1255         if (mounted) {
1256                 dput(path->dentry);
1257                 mntput(path->mnt);
1258                 path->mnt = mounted;
1259                 path->dentry = dget(mounted->mnt_root);
1260                 return 1;
1261         }
1262         return 0;
1263 }
1264 EXPORT_SYMBOL(follow_down_one);
1265 
1266 static inline int managed_dentry_rcu(struct dentry *dentry)
1267 {
1268         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1269                 dentry->d_op->d_manage(dentry, true) : 0;
1270 }
1271 
1272 /*
1273  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1274  * we meet a managed dentry that would need blocking.
1275  */
1276 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1277                                struct inode **inode, unsigned *seqp)
1278 {
1279         for (;;) {
1280                 struct mount *mounted;
1281                 /*
1282                  * Don't forget we might have a non-mountpoint managed dentry
1283                  * that wants to block transit.
1284                  */
1285                 switch (managed_dentry_rcu(path->dentry)) {
1286                 case -ECHILD:
1287                 default:
1288                         return false;
1289                 case -EISDIR:
1290                         return true;
1291                 case 0:
1292                         break;
1293                 }
1294 
1295                 if (!d_mountpoint(path->dentry))
1296                         return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1297 
1298                 mounted = __lookup_mnt(path->mnt, path->dentry);
1299                 if (!mounted)
1300                         break;
1301                 path->mnt = &mounted->mnt;
1302                 path->dentry = mounted->mnt.mnt_root;
1303                 nd->flags |= LOOKUP_JUMPED;
1304                 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1305                 /*
1306                  * Update the inode too. We don't need to re-check the
1307                  * dentry sequence number here after this d_inode read,
1308                  * because a mount-point is always pinned.
1309                  */
1310                 *inode = path->dentry->d_inode;
1311         }
1312         return !read_seqretry(&mount_lock, nd->m_seq) &&
1313                 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1314 }
1315 
1316 static int follow_dotdot_rcu(struct nameidata *nd)
1317 {
1318         struct inode *inode = nd->inode;
1319 
1320         while (1) {
1321                 if (path_equal(&nd->path, &nd->root))
1322                         break;
1323                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1324                         struct dentry *old = nd->path.dentry;
1325                         struct dentry *parent = old->d_parent;
1326                         unsigned seq;
1327 
1328                         inode = parent->d_inode;
1329                         seq = read_seqcount_begin(&parent->d_seq);
1330                         if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1331                                 return -ECHILD;
1332                         nd->path.dentry = parent;
1333                         nd->seq = seq;
1334                         if (unlikely(!path_connected(&nd->path)))
1335                                 return -ENOENT;
1336                         break;
1337                 } else {
1338                         struct mount *mnt = real_mount(nd->path.mnt);
1339                         struct mount *mparent = mnt->mnt_parent;
1340                         struct dentry *mountpoint = mnt->mnt_mountpoint;
1341                         struct inode *inode2 = mountpoint->d_inode;
1342                         unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1343                         if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1344                                 return -ECHILD;
1345                         if (&mparent->mnt == nd->path.mnt)
1346                                 break;
1347                         /* we know that mountpoint was pinned */
1348                         nd->path.dentry = mountpoint;
1349                         nd->path.mnt = &mparent->mnt;
1350                         inode = inode2;
1351                         nd->seq = seq;
1352                 }
1353         }
1354         while (unlikely(d_mountpoint(nd->path.dentry))) {
1355                 struct mount *mounted;
1356                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1357                 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1358                         return -ECHILD;
1359                 if (!mounted)
1360                         break;
1361                 nd->path.mnt = &mounted->mnt;
1362                 nd->path.dentry = mounted->mnt.mnt_root;
1363                 inode = nd->path.dentry->d_inode;
1364                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1365         }
1366         nd->inode = inode;
1367         return 0;
1368 }
1369 
1370 /*
1371  * Follow down to the covering mount currently visible to userspace.  At each
1372  * point, the filesystem owning that dentry may be queried as to whether the
1373  * caller is permitted to proceed or not.
1374  */
1375 int follow_down(struct path *path)
1376 {
1377         unsigned managed;
1378         int ret;
1379 
1380         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1381                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1382                 /* Allow the filesystem to manage the transit without i_mutex
1383                  * being held.
1384                  *
1385                  * We indicate to the filesystem if someone is trying to mount
1386                  * something here.  This gives autofs the chance to deny anyone
1387                  * other than its daemon the right to mount on its
1388                  * superstructure.
1389                  *
1390                  * The filesystem may sleep at this point.
1391                  */
1392                 if (managed & DCACHE_MANAGE_TRANSIT) {
1393                         BUG_ON(!path->dentry->d_op);
1394                         BUG_ON(!path->dentry->d_op->d_manage);
1395                         ret = path->dentry->d_op->d_manage(
1396                                 path->dentry, false);
1397                         if (ret < 0)
1398                                 return ret == -EISDIR ? 0 : ret;
1399                 }
1400 
1401                 /* Transit to a mounted filesystem. */
1402                 if (managed & DCACHE_MOUNTED) {
1403                         struct vfsmount *mounted = lookup_mnt(path);
1404                         if (!mounted)
1405                                 break;
1406                         dput(path->dentry);
1407                         mntput(path->mnt);
1408                         path->mnt = mounted;
1409                         path->dentry = dget(mounted->mnt_root);
1410                         continue;
1411                 }
1412 
1413                 /* Don't handle automount points here */
1414                 break;
1415         }
1416         return 0;
1417 }
1418 EXPORT_SYMBOL(follow_down);
1419 
1420 /*
1421  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1422  */
1423 static void follow_mount(struct path *path)
1424 {
1425         while (d_mountpoint(path->dentry)) {
1426                 struct vfsmount *mounted = lookup_mnt(path);
1427                 if (!mounted)
1428                         break;
1429                 dput(path->dentry);
1430                 mntput(path->mnt);
1431                 path->mnt = mounted;
1432                 path->dentry = dget(mounted->mnt_root);
1433         }
1434 }
1435 
1436 static int path_parent_directory(struct path *path)
1437 {
1438         struct dentry *old = path->dentry;
1439         /* rare case of legitimate dget_parent()... */
1440         path->dentry = dget_parent(path->dentry);
1441         dput(old);
1442         if (unlikely(!path_connected(path)))
1443                 return -ENOENT;
1444         return 0;
1445 }
1446 
1447 static int follow_dotdot(struct nameidata *nd)
1448 {
1449         while(1) {
1450                 if (nd->path.dentry == nd->root.dentry &&
1451                     nd->path.mnt == nd->root.mnt) {
1452                         break;
1453                 }
1454                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1455                         int ret = path_parent_directory(&nd->path);
1456                         if (ret)
1457                                 return ret;
1458                         break;
1459                 }
1460                 if (!follow_up(&nd->path))
1461                         break;
1462         }
1463         follow_mount(&nd->path);
1464         nd->inode = nd->path.dentry->d_inode;
1465         return 0;
1466 }
1467 
1468 /*
1469  * This looks up the name in dcache and possibly revalidates the found dentry.
1470  * NULL is returned if the dentry does not exist in the cache.
1471  */
1472 static struct dentry *lookup_dcache(const struct qstr *name,
1473                                     struct dentry *dir,
1474                                     unsigned int flags)
1475 {
1476         struct dentry *dentry;
1477         int error;
1478 
1479         dentry = d_lookup(dir, name);
1480         if (dentry) {
1481                 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1482                         error = d_revalidate(dentry, flags);
1483                         if (unlikely(error <= 0)) {
1484                                 if (!error)
1485                                         d_invalidate(dentry);
1486                                 dput(dentry);
1487                                 return ERR_PTR(error);
1488                         }
1489                 }
1490         }
1491         return dentry;
1492 }
1493 
1494 /*
1495  * Call i_op->lookup on the dentry.  The dentry must be negative and
1496  * unhashed.
1497  *
1498  * dir->d_inode->i_mutex must be held
1499  */
1500 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1501                                   unsigned int flags)
1502 {
1503         struct dentry *old;
1504 
1505         /* Don't create child dentry for a dead directory. */
1506         if (unlikely(IS_DEADDIR(dir))) {
1507                 dput(dentry);
1508                 return ERR_PTR(-ENOENT);
1509         }
1510 
1511         old = dir->i_op->lookup(dir, dentry, flags);
1512         if (unlikely(old)) {
1513                 dput(dentry);
1514                 dentry = old;
1515         }
1516         return dentry;
1517 }
1518 
1519 static struct dentry *__lookup_hash(const struct qstr *name,
1520                 struct dentry *base, unsigned int flags)
1521 {
1522         struct dentry *dentry = lookup_dcache(name, base, flags);
1523 
1524         if (dentry)
1525                 return dentry;
1526 
1527         dentry = d_alloc(base, name);
1528         if (unlikely(!dentry))
1529                 return ERR_PTR(-ENOMEM);
1530 
1531         return lookup_real(base->d_inode, dentry, flags);
1532 }
1533 
1534 static int lookup_fast(struct nameidata *nd,
1535                        struct path *path, struct inode **inode,
1536                        unsigned *seqp)
1537 {
1538         struct vfsmount *mnt = nd->path.mnt;
1539         struct dentry *dentry, *parent = nd->path.dentry;
1540         int status = 1;
1541         int err;
1542 
1543         /*
1544          * Rename seqlock is not required here because in the off chance
1545          * of a false negative due to a concurrent rename, the caller is
1546          * going to fall back to non-racy lookup.
1547          */
1548         if (nd->flags & LOOKUP_RCU) {
1549                 unsigned seq;
1550                 bool negative;
1551                 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1552                 if (unlikely(!dentry)) {
1553                         if (unlazy_walk(nd, NULL, 0))
1554                                 return -ECHILD;
1555                         return 0;
1556                 }
1557 
1558                 /*
1559                  * This sequence count validates that the inode matches
1560                  * the dentry name information from lookup.
1561                  */
1562                 *inode = d_backing_inode(dentry);
1563                 negative = d_is_negative(dentry);
1564                 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1565                         return -ECHILD;
1566 
1567                 /*
1568                  * This sequence count validates that the parent had no
1569                  * changes while we did the lookup of the dentry above.
1570                  *
1571                  * The memory barrier in read_seqcount_begin of child is
1572                  *  enough, we can use __read_seqcount_retry here.
1573                  */
1574                 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1575                         return -ECHILD;
1576 
1577                 *seqp = seq;
1578                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1579                         status = d_revalidate(dentry, nd->flags);
1580                 if (unlikely(status <= 0)) {
1581                         if (unlazy_walk(nd, dentry, seq))
1582                                 return -ECHILD;
1583                         if (status == -ECHILD)
1584                                 status = d_revalidate(dentry, nd->flags);
1585                 } else {
1586                         /*
1587                          * Note: do negative dentry check after revalidation in
1588                          * case that drops it.
1589                          */
1590                         if (unlikely(negative))
1591                                 return -ENOENT;
1592                         path->mnt = mnt;
1593                         path->dentry = dentry;
1594                         if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1595                                 return 1;
1596                         if (unlazy_walk(nd, dentry, seq))
1597                                 return -ECHILD;
1598                 }
1599         } else {
1600                 dentry = __d_lookup(parent, &nd->last);
1601                 if (unlikely(!dentry))
1602                         return 0;
1603                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1604                         status = d_revalidate(dentry, nd->flags);
1605         }
1606         if (unlikely(status <= 0)) {
1607                 if (!status)
1608                         d_invalidate(dentry);
1609                 dput(dentry);
1610                 return status;
1611         }
1612         if (unlikely(d_is_negative(dentry))) {
1613                 dput(dentry);
1614                 return -ENOENT;
1615         }
1616 
1617         path->mnt = mnt;
1618         path->dentry = dentry;
1619         err = follow_managed(path, nd);
1620         if (likely(err > 0))
1621                 *inode = d_backing_inode(path->dentry);
1622         return err;
1623 }
1624 
1625 /* Fast lookup failed, do it the slow way */
1626 static struct dentry *lookup_slow(const struct qstr *name,
1627                                   struct dentry *dir,
1628                                   unsigned int flags)
1629 {
1630         struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1631         struct inode *inode = dir->d_inode;
1632         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1633 
1634         inode_lock_shared(inode);
1635         /* Don't go there if it's already dead */
1636         if (unlikely(IS_DEADDIR(inode)))
1637                 goto out;
1638 again:
1639         dentry = d_alloc_parallel(dir, name, &wq);
1640         if (IS_ERR(dentry))
1641                 goto out;
1642         if (unlikely(!d_in_lookup(dentry))) {
1643                 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1644                     !(flags & LOOKUP_NO_REVAL)) {
1645                         int error = d_revalidate(dentry, flags);
1646                         if (unlikely(error <= 0)) {
1647                                 if (!error) {
1648                                         d_invalidate(dentry);
1649                                         dput(dentry);
1650                                         goto again;
1651                                 }
1652                                 dput(dentry);
1653                                 dentry = ERR_PTR(error);
1654                         }
1655                 }
1656         } else {
1657                 old = inode->i_op->lookup(inode, dentry, flags);
1658                 d_lookup_done(dentry);
1659                 if (unlikely(old)) {
1660                         dput(dentry);
1661                         dentry = old;
1662                 }
1663         }
1664 out:
1665         inode_unlock_shared(inode);
1666         return dentry;
1667 }
1668 
1669 static inline int may_lookup(struct nameidata *nd)
1670 {
1671         if (nd->flags & LOOKUP_RCU) {
1672                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1673                 if (err != -ECHILD)
1674                         return err;
1675                 if (unlazy_walk(nd, NULL, 0))
1676                         return -ECHILD;
1677         }
1678         return inode_permission(nd->inode, MAY_EXEC);
1679 }
1680 
1681 static inline int handle_dots(struct nameidata *nd, int type)
1682 {
1683         if (type == LAST_DOTDOT) {
1684                 if (!nd->root.mnt)
1685                         set_root(nd);
1686                 if (nd->flags & LOOKUP_RCU) {
1687                         return follow_dotdot_rcu(nd);
1688                 } else
1689                         return follow_dotdot(nd);
1690         }
1691         return 0;
1692 }
1693 
1694 static int pick_link(struct nameidata *nd, struct path *link,
1695                      struct inode *inode, unsigned seq)
1696 {
1697         int error;
1698         struct saved *last;
1699         if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1700                 path_to_nameidata(link, nd);
1701                 return -ELOOP;
1702         }
1703         if (!(nd->flags & LOOKUP_RCU)) {
1704                 if (link->mnt == nd->path.mnt)
1705                         mntget(link->mnt);
1706         }
1707         error = nd_alloc_stack(nd);
1708         if (unlikely(error)) {
1709                 if (error == -ECHILD) {
1710                         if (unlikely(unlazy_link(nd, link, seq)))
1711                                 return -ECHILD;
1712                         error = nd_alloc_stack(nd);
1713                 }
1714                 if (error) {
1715                         path_put(link);
1716                         return error;
1717                 }
1718         }
1719 
1720         last = nd->stack + nd->depth++;
1721         last->link = *link;
1722         clear_delayed_call(&last->done);
1723         nd->link_inode = inode;
1724         last->seq = seq;
1725         return 1;
1726 }
1727 
1728 /*
1729  * Do we need to follow links? We _really_ want to be able
1730  * to do this check without having to look at inode->i_op,
1731  * so we keep a cache of "no, this doesn't need follow_link"
1732  * for the common case.
1733  */
1734 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1735                                      int follow,
1736                                      struct inode *inode, unsigned seq)
1737 {
1738         if (likely(!d_is_symlink(link->dentry)))
1739                 return 0;
1740         if (!follow)
1741                 return 0;
1742         /* make sure that d_is_symlink above matches inode */
1743         if (nd->flags & LOOKUP_RCU) {
1744                 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1745                         return -ECHILD;
1746         }
1747         return pick_link(nd, link, inode, seq);
1748 }
1749 
1750 enum {WALK_GET = 1, WALK_PUT = 2};
1751 
1752 static int walk_component(struct nameidata *nd, int flags)
1753 {
1754         struct path path;
1755         struct inode *inode;
1756         unsigned seq;
1757         int err;
1758         /*
1759          * "." and ".." are special - ".." especially so because it has
1760          * to be able to know about the current root directory and
1761          * parent relationships.
1762          */
1763         if (unlikely(nd->last_type != LAST_NORM)) {
1764                 err = handle_dots(nd, nd->last_type);
1765                 if (flags & WALK_PUT)
1766                         put_link(nd);
1767                 return err;
1768         }
1769         err = lookup_fast(nd, &path, &inode, &seq);
1770         if (unlikely(err <= 0)) {
1771                 if (err < 0)
1772                         return err;
1773                 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1774                                           nd->flags);
1775                 if (IS_ERR(path.dentry))
1776                         return PTR_ERR(path.dentry);
1777 
1778                 path.mnt = nd->path.mnt;
1779                 err = follow_managed(&path, nd);
1780                 if (unlikely(err < 0))
1781                         return err;
1782 
1783                 if (unlikely(d_is_negative(path.dentry))) {
1784                         path_to_nameidata(&path, nd);
1785                         return -ENOENT;
1786                 }
1787 
1788                 seq = 0;        /* we are already out of RCU mode */
1789                 inode = d_backing_inode(path.dentry);
1790         }
1791 
1792         if (flags & WALK_PUT)
1793                 put_link(nd);
1794         err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1795         if (unlikely(err))
1796                 return err;
1797         path_to_nameidata(&path, nd);
1798         nd->inode = inode;
1799         nd->seq = seq;
1800         return 0;
1801 }
1802 
1803 /*
1804  * We can do the critical dentry name comparison and hashing
1805  * operations one word at a time, but we are limited to:
1806  *
1807  * - Architectures with fast unaligned word accesses. We could
1808  *   do a "get_unaligned()" if this helps and is sufficiently
1809  *   fast.
1810  *
1811  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1812  *   do not trap on the (extremely unlikely) case of a page
1813  *   crossing operation.
1814  *
1815  * - Furthermore, we need an efficient 64-bit compile for the
1816  *   64-bit case in order to generate the "number of bytes in
1817  *   the final mask". Again, that could be replaced with a
1818  *   efficient population count instruction or similar.
1819  */
1820 #ifdef CONFIG_DCACHE_WORD_ACCESS
1821 
1822 #include <asm/word-at-a-time.h>
1823 
1824 #ifdef HASH_MIX
1825 
1826 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1827 
1828 #elif defined(CONFIG_64BIT)
1829 /*
1830  * Register pressure in the mixing function is an issue, particularly
1831  * on 32-bit x86, but almost any function requires one state value and
1832  * one temporary.  Instead, use a function designed for two state values
1833  * and no temporaries.
1834  *
1835  * This function cannot create a collision in only two iterations, so
1836  * we have two iterations to achieve avalanche.  In those two iterations,
1837  * we have six layers of mixing, which is enough to spread one bit's
1838  * influence out to 2^6 = 64 state bits.
1839  *
1840  * Rotate constants are scored by considering either 64 one-bit input
1841  * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1842  * probability of that delta causing a change to each of the 128 output
1843  * bits, using a sample of random initial states.
1844  *
1845  * The Shannon entropy of the computed probabilities is then summed
1846  * to produce a score.  Ideally, any input change has a 50% chance of
1847  * toggling any given output bit.
1848  *
1849  * Mixing scores (in bits) for (12,45):
1850  * Input delta: 1-bit      2-bit
1851  * 1 round:     713.3    42542.6
1852  * 2 rounds:   2753.7   140389.8
1853  * 3 rounds:   5954.1   233458.2
1854  * 4 rounds:   7862.6   256672.2
1855  * Perfect:    8192     258048
1856  *            (64*128) (64*63/2 * 128)
1857  */
1858 #define HASH_MIX(x, y, a)       \
1859         (       x ^= (a),       \
1860         y ^= x, x = rol64(x,12),\
1861         x += y, y = rol64(y,45),\
1862         y *= 9                  )
1863 
1864 /*
1865  * Fold two longs into one 32-bit hash value.  This must be fast, but
1866  * latency isn't quite as critical, as there is a fair bit of additional
1867  * work done before the hash value is used.
1868  */
1869 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1870 {
1871         y ^= x * GOLDEN_RATIO_64;
1872         y *= GOLDEN_RATIO_64;
1873         return y >> 32;
1874 }
1875 
1876 #else   /* 32-bit case */
1877 
1878 /*
1879  * Mixing scores (in bits) for (7,20):
1880  * Input delta: 1-bit      2-bit
1881  * 1 round:     330.3     9201.6
1882  * 2 rounds:   1246.4    25475.4
1883  * 3 rounds:   1907.1    31295.1
1884  * 4 rounds:   2042.3    31718.6
1885  * Perfect:    2048      31744
1886  *            (32*64)   (32*31/2 * 64)
1887  */
1888 #define HASH_MIX(x, y, a)       \
1889         (       x ^= (a),       \
1890         y ^= x, x = rol32(x, 7),\
1891         x += y, y = rol32(y,20),\
1892         y *= 9                  )
1893 
1894 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1895 {
1896         /* Use arch-optimized multiply if one exists */
1897         return __hash_32(y ^ __hash_32(x));
1898 }
1899 
1900 #endif
1901 
1902 /*
1903  * Return the hash of a string of known length.  This is carfully
1904  * designed to match hash_name(), which is the more critical function.
1905  * In particular, we must end by hashing a final word containing 0..7
1906  * payload bytes, to match the way that hash_name() iterates until it
1907  * finds the delimiter after the name.
1908  */
1909 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1910 {
1911         unsigned long a, x = 0, y = (unsigned long)salt;
1912 
1913         for (;;) {
1914                 if (!len)
1915                         goto done;
1916                 a = load_unaligned_zeropad(name);
1917                 if (len < sizeof(unsigned long))
1918                         break;
1919                 HASH_MIX(x, y, a);
1920                 name += sizeof(unsigned long);
1921                 len -= sizeof(unsigned long);
1922         }
1923         x ^= a & bytemask_from_count(len);
1924 done:
1925         return fold_hash(x, y);
1926 }
1927 EXPORT_SYMBOL(full_name_hash);
1928 
1929 /* Return the "hash_len" (hash and length) of a null-terminated string */
1930 u64 hashlen_string(const void *salt, const char *name)
1931 {
1932         unsigned long a = 0, x = 0, y = (unsigned long)salt;
1933         unsigned long adata, mask, len;
1934         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1935 
1936         len = 0;
1937         goto inside;
1938 
1939         do {
1940                 HASH_MIX(x, y, a);
1941                 len += sizeof(unsigned long);
1942 inside:
1943                 a = load_unaligned_zeropad(name+len);
1944         } while (!has_zero(a, &adata, &constants));
1945 
1946         adata = prep_zero_mask(a, adata, &constants);
1947         mask = create_zero_mask(adata);
1948         x ^= a & zero_bytemask(mask);
1949 
1950         return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1951 }
1952 EXPORT_SYMBOL(hashlen_string);
1953 
1954 /*
1955  * Calculate the length and hash of the path component, and
1956  * return the "hash_len" as the result.
1957  */
1958 static inline u64 hash_name(const void *salt, const char *name)
1959 {
1960         unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1961         unsigned long adata, bdata, mask, len;
1962         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1963 
1964         len = 0;
1965         goto inside;
1966 
1967         do {
1968                 HASH_MIX(x, y, a);
1969                 len += sizeof(unsigned long);
1970 inside:
1971                 a = load_unaligned_zeropad(name+len);
1972                 b = a ^ REPEAT_BYTE('/');
1973         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1974 
1975         adata = prep_zero_mask(a, adata, &constants);
1976         bdata = prep_zero_mask(b, bdata, &constants);
1977         mask = create_zero_mask(adata | bdata);
1978         x ^= a & zero_bytemask(mask);
1979 
1980         return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1981 }
1982 
1983 #else   /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1984 
1985 /* Return the hash of a string of known length */
1986 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1987 {
1988         unsigned long hash = init_name_hash(salt);
1989         while (len--)
1990                 hash = partial_name_hash((unsigned char)*name++, hash);
1991         return end_name_hash(hash);
1992 }
1993 EXPORT_SYMBOL(full_name_hash);
1994 
1995 /* Return the "hash_len" (hash and length) of a null-terminated string */
1996 u64 hashlen_string(const void *salt, const char *name)
1997 {
1998         unsigned long hash = init_name_hash(salt);
1999         unsigned long len = 0, c;
2000 
2001         c = (unsigned char)*name;
2002         while (c) {
2003                 len++;
2004                 hash = partial_name_hash(c, hash);
2005                 c = (unsigned char)name[len];
2006         }
2007         return hashlen_create(end_name_hash(hash), len);
2008 }
2009 EXPORT_SYMBOL(hashlen_string);
2010 
2011 /*
2012  * We know there's a real path component here of at least
2013  * one character.
2014  */
2015 static inline u64 hash_name(const void *salt, const char *name)
2016 {
2017         unsigned long hash = init_name_hash(salt);
2018         unsigned long len = 0, c;
2019 
2020         c = (unsigned char)*name;
2021         do {
2022                 len++;
2023                 hash = partial_name_hash(c, hash);
2024                 c = (unsigned char)name[len];
2025         } while (c && c != '/');
2026         return hashlen_create(end_name_hash(hash), len);
2027 }
2028 
2029 #endif
2030 
2031 /*
2032  * Name resolution.
2033  * This is the basic name resolution function, turning a pathname into
2034  * the final dentry. We expect 'base' to be positive and a directory.
2035  *
2036  * Returns 0 and nd will have valid dentry and mnt on success.
2037  * Returns error and drops reference to input namei data on failure.
2038  */
2039 static int link_path_walk(const char *name, struct nameidata *nd)
2040 {
2041         int err;
2042 
2043         while (*name=='/')
2044                 name++;
2045         if (!*name)
2046                 return 0;
2047 
2048         /* At this point we know we have a real path component. */
2049         for(;;) {
2050                 u64 hash_len;
2051                 int type;
2052 
2053                 err = may_lookup(nd);
2054                 if (err)
2055                         return err;
2056 
2057                 hash_len = hash_name(nd->path.dentry, name);
2058 
2059                 type = LAST_NORM;
2060                 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2061                         case 2:
2062                                 if (name[1] == '.') {
2063                                         type = LAST_DOTDOT;
2064                                         nd->flags |= LOOKUP_JUMPED;
2065                                 }
2066                                 break;
2067                         case 1:
2068                                 type = LAST_DOT;
2069                 }
2070                 if (likely(type == LAST_NORM)) {
2071                         struct dentry *parent = nd->path.dentry;
2072                         nd->flags &= ~LOOKUP_JUMPED;
2073                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2074                                 struct qstr this = { { .hash_len = hash_len }, .name = name };
2075                                 err = parent->d_op->d_hash(parent, &this);
2076                                 if (err < 0)
2077                                         return err;
2078                                 hash_len = this.hash_len;
2079                                 name = this.name;
2080                         }
2081                 }
2082 
2083                 nd->last.hash_len = hash_len;
2084                 nd->last.name = name;
2085                 nd->last_type = type;
2086 
2087                 name += hashlen_len(hash_len);
2088                 if (!*name)
2089                         goto OK;
2090                 /*
2091                  * If it wasn't NUL, we know it was '/'. Skip that
2092                  * slash, and continue until no more slashes.
2093                  */
2094                 do {
2095                         name++;
2096                 } while (unlikely(*name == '/'));
2097                 if (unlikely(!*name)) {
2098 OK:
2099                         /* pathname body, done */
2100                         if (!nd->depth)
2101                                 return 0;
2102                         name = nd->stack[nd->depth - 1].name;
2103                         /* trailing symlink, done */
2104                         if (!name)
2105                                 return 0;
2106                         /* last component of nested symlink */
2107                         err = walk_component(nd, WALK_GET | WALK_PUT);
2108                 } else {
2109                         err = walk_component(nd, WALK_GET);
2110                 }
2111                 if (err < 0)
2112                         return err;
2113 
2114                 if (err) {
2115                         const char *s = get_link(nd);
2116 
2117                         if (IS_ERR(s))
2118                                 return PTR_ERR(s);
2119                         err = 0;
2120                         if (unlikely(!s)) {
2121                                 /* jumped */
2122                                 put_link(nd);
2123                         } else {
2124                                 nd->stack[nd->depth - 1].name = name;
2125                                 name = s;
2126                                 continue;
2127                         }
2128                 }
2129                 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2130                         if (nd->flags & LOOKUP_RCU) {
2131                                 if (unlazy_walk(nd, NULL, 0))
2132                                         return -ECHILD;
2133                         }
2134                         return -ENOTDIR;
2135                 }
2136         }
2137 }
2138 
2139 static const char *path_init(struct nameidata *nd, unsigned flags)
2140 {
2141         int retval = 0;
2142         const char *s = nd->name->name;
2143 
2144         nd->last_type = LAST_ROOT; /* if there are only slashes... */
2145         nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2146         nd->depth = 0;
2147         if (flags & LOOKUP_ROOT) {
2148                 struct dentry *root = nd->root.dentry;
2149                 struct inode *inode = root->d_inode;
2150                 if (*s) {
2151                         if (!d_can_lookup(root))
2152                                 return ERR_PTR(-ENOTDIR);
2153                         retval = inode_permission(inode, MAY_EXEC);
2154                         if (retval)
2155                                 return ERR_PTR(retval);
2156                 }
2157                 nd->path = nd->root;
2158                 nd->inode = inode;
2159                 if (flags & LOOKUP_RCU) {
2160                         rcu_read_lock();
2161                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2162                         nd->root_seq = nd->seq;
2163                         nd->m_seq = read_seqbegin(&mount_lock);
2164                 } else {
2165                         path_get(&nd->path);
2166                 }
2167                 return s;
2168         }
2169 
2170         nd->root.mnt = NULL;
2171         nd->path.mnt = NULL;
2172         nd->path.dentry = NULL;
2173 
2174         nd->m_seq = read_seqbegin(&mount_lock);
2175         if (*s == '/') {
2176                 if (flags & LOOKUP_RCU)
2177                         rcu_read_lock();
2178                 set_root(nd);
2179                 if (likely(!nd_jump_root(nd)))
2180                         return s;
2181                 nd->root.mnt = NULL;
2182                 rcu_read_unlock();
2183                 return ERR_PTR(-ECHILD);
2184         } else if (nd->dfd == AT_FDCWD) {
2185                 if (flags & LOOKUP_RCU) {
2186                         struct fs_struct *fs = current->fs;
2187                         unsigned seq;
2188 
2189                         rcu_read_lock();
2190 
2191                         do {
2192                                 seq = read_seqcount_begin(&fs->seq);
2193                                 nd->path = fs->pwd;
2194                                 nd->inode = nd->path.dentry->d_inode;
2195                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2196                         } while (read_seqcount_retry(&fs->seq, seq));
2197                 } else {
2198                         get_fs_pwd(current->fs, &nd->path);
2199                         nd->inode = nd->path.dentry->d_inode;
2200                 }
2201                 return s;
2202         } else {
2203                 /* Caller must check execute permissions on the starting path component */
2204                 struct fd f = fdget_raw(nd->dfd);
2205                 struct dentry *dentry;
2206 
2207                 if (!f.file)
2208                         return ERR_PTR(-EBADF);
2209 
2210                 dentry = f.file->f_path.dentry;
2211 
2212                 if (*s) {
2213                         if (!d_can_lookup(dentry)) {
2214                                 fdput(f);
2215                                 return ERR_PTR(-ENOTDIR);
2216                         }
2217                 }
2218 
2219                 nd->path = f.file->f_path;
2220                 if (flags & LOOKUP_RCU) {
2221                         rcu_read_lock();
2222                         nd->inode = nd->path.dentry->d_inode;
2223                         nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2224                 } else {
2225                         path_get(&nd->path);
2226                         nd->inode = nd->path.dentry->d_inode;
2227                 }
2228                 fdput(f);
2229                 return s;
2230         }
2231 }
2232 
2233 static const char *trailing_symlink(struct nameidata *nd)
2234 {
2235         const char *s;
2236         int error = may_follow_link(nd);
2237         if (unlikely(error))
2238                 return ERR_PTR(error);
2239         nd->flags |= LOOKUP_PARENT;
2240         nd->stack[0].name = NULL;
2241         s = get_link(nd);
2242         return s ? s : "";
2243 }
2244 
2245 static inline int lookup_last(struct nameidata *nd)
2246 {
2247         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2248                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2249 
2250         nd->flags &= ~LOOKUP_PARENT;
2251         return walk_component(nd,
2252                         nd->flags & LOOKUP_FOLLOW
2253                                 ? nd->depth
2254                                         ? WALK_PUT | WALK_GET
2255                                         : WALK_GET
2256                                 : 0);
2257 }
2258 
2259 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2260 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2261 {
2262         const char *s = path_init(nd, flags);
2263         int err;
2264 
2265         if (IS_ERR(s))
2266                 return PTR_ERR(s);
2267         while (!(err = link_path_walk(s, nd))
2268                 && ((err = lookup_last(nd)) > 0)) {
2269                 s = trailing_symlink(nd);
2270                 if (IS_ERR(s)) {
2271                         err = PTR_ERR(s);
2272                         break;
2273                 }
2274         }
2275         if (!err)
2276                 err = complete_walk(nd);
2277 
2278         if (!err && nd->flags & LOOKUP_DIRECTORY)
2279                 if (!d_can_lookup(nd->path.dentry))
2280                         err = -ENOTDIR;
2281         if (!err) {
2282                 *path = nd->path;
2283                 nd->path.mnt = NULL;
2284                 nd->path.dentry = NULL;
2285         }
2286         terminate_walk(nd);
2287         return err;
2288 }
2289 
2290 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2291                            struct path *path, struct path *root)
2292 {
2293         int retval;
2294         struct nameidata nd;
2295         if (IS_ERR(name))
2296                 return PTR_ERR(name);
2297         if (unlikely(root)) {
2298                 nd.root = *root;
2299                 flags |= LOOKUP_ROOT;
2300         }
2301         set_nameidata(&nd, dfd, name);
2302         retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2303         if (unlikely(retval == -ECHILD))
2304                 retval = path_lookupat(&nd, flags, path);
2305         if (unlikely(retval == -ESTALE))
2306                 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2307 
2308         if (likely(!retval))
2309                 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2310         restore_nameidata();
2311         putname(name);
2312         return retval;
2313 }
2314 
2315 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2316 static int path_parentat(struct nameidata *nd, unsigned flags,
2317                                 struct path *parent)
2318 {
2319         const char *s = path_init(nd, flags);
2320         int err;
2321         if (IS_ERR(s))
2322                 return PTR_ERR(s);
2323         err = link_path_walk(s, nd);
2324         if (!err)
2325                 err = complete_walk(nd);
2326         if (!err) {
2327                 *parent = nd->path;
2328                 nd->path.mnt = NULL;
2329                 nd->path.dentry = NULL;
2330         }
2331         terminate_walk(nd);
2332         return err;
2333 }
2334 
2335 static struct filename *filename_parentat(int dfd, struct filename *name,
2336                                 unsigned int flags, struct path *parent,
2337                                 struct qstr *last, int *type)
2338 {
2339         int retval;
2340         struct nameidata nd;
2341 
2342         if (IS_ERR(name))
2343                 return name;
2344         set_nameidata(&nd, dfd, name);
2345         retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2346         if (unlikely(retval == -ECHILD))
2347                 retval = path_parentat(&nd, flags, parent);
2348         if (unlikely(retval == -ESTALE))
2349                 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2350         if (likely(!retval)) {
2351                 *last = nd.last;
2352                 *type = nd.last_type;
2353                 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2354         } else {
2355                 putname(name);
2356                 name = ERR_PTR(retval);
2357         }
2358         restore_nameidata();
2359         return name;
2360 }
2361 
2362 /* does lookup, returns the object with parent locked */
2363 struct dentry *kern_path_locked(const char *name, struct path *path)
2364 {
2365         struct filename *filename;
2366         struct dentry *d;
2367         struct qstr last;
2368         int type;
2369 
2370         filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2371                                     &last, &type);
2372         if (IS_ERR(filename))
2373                 return ERR_CAST(filename);
2374         if (unlikely(type != LAST_NORM)) {
2375                 path_put(path);
2376                 putname(filename);
2377                 return ERR_PTR(-EINVAL);
2378         }
2379         inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2380         d = __lookup_hash(&last, path->dentry, 0);
2381         if (IS_ERR(d)) {
2382                 inode_unlock(path->dentry->d_inode);
2383                 path_put(path);
2384         }
2385         putname(filename);
2386         return d;
2387 }
2388 
2389 int kern_path(const char *name, unsigned int flags, struct path *path)
2390 {
2391         return filename_lookup(AT_FDCWD, getname_kernel(name),
2392                                flags, path, NULL);
2393 }
2394 EXPORT_SYMBOL(kern_path);
2395 
2396 /**
2397  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2398  * @dentry:  pointer to dentry of the base directory
2399  * @mnt: pointer to vfs mount of the base directory
2400  * @name: pointer to file name
2401  * @flags: lookup flags
2402  * @path: pointer to struct path to fill
2403  */
2404 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2405                     const char *name, unsigned int flags,
2406                     struct path *path)
2407 {
2408         struct path root = {.mnt = mnt, .dentry = dentry};
2409         /* the first argument of filename_lookup() is ignored with root */
2410         return filename_lookup(AT_FDCWD, getname_kernel(name),
2411                                flags , path, &root);
2412 }
2413 EXPORT_SYMBOL(vfs_path_lookup);
2414 
2415 /**
2416  * lookup_one_len - filesystem helper to lookup single pathname component
2417  * @name:       pathname component to lookup
2418  * @base:       base directory to lookup from
2419  * @len:        maximum length @len should be interpreted to
2420  *
2421  * Note that this routine is purely a helper for filesystem usage and should
2422  * not be called by generic code.
2423  *
2424  * The caller must hold base->i_mutex.
2425  */
2426 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2427 {
2428         struct qstr this;
2429         unsigned int c;
2430         int err;
2431 
2432         WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2433 
2434         this.name = name;
2435         this.len = len;
2436         this.hash = full_name_hash(base, name, len);
2437         if (!len)
2438                 return ERR_PTR(-EACCES);
2439 
2440         if (unlikely(name[0] == '.')) {
2441                 if (len < 2 || (len == 2 && name[1] == '.'))
2442                         return ERR_PTR(-EACCES);
2443         }
2444 
2445         while (len--) {
2446                 c = *(const unsigned char *)name++;
2447                 if (c == '/' || c == '\0')
2448                         return ERR_PTR(-EACCES);
2449         }
2450         /*
2451          * See if the low-level filesystem might want
2452          * to use its own hash..
2453          */
2454         if (base->d_flags & DCACHE_OP_HASH) {
2455                 int err = base->d_op->d_hash(base, &this);
2456                 if (err < 0)
2457                         return ERR_PTR(err);
2458         }
2459 
2460         err = inode_permission(base->d_inode, MAY_EXEC);
2461         if (err)
2462                 return ERR_PTR(err);
2463 
2464         return __lookup_hash(&this, base, 0);
2465 }
2466 EXPORT_SYMBOL(lookup_one_len);
2467 
2468 /**
2469  * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2470  * @name:       pathname component to lookup
2471  * @base:       base directory to lookup from
2472  * @len:        maximum length @len should be interpreted to
2473  *
2474  * Note that this routine is purely a helper for filesystem usage and should
2475  * not be called by generic code.
2476  *
2477  * Unlike lookup_one_len, it should be called without the parent
2478  * i_mutex held, and will take the i_mutex itself if necessary.
2479  */
2480 struct dentry *lookup_one_len_unlocked(const char *name,
2481                                        struct dentry *base, int len)
2482 {
2483         struct qstr this;
2484         unsigned int c;
2485         int err;
2486         struct dentry *ret;
2487 
2488         this.name = name;
2489         this.len = len;
2490         this.hash = full_name_hash(base, name, len);
2491         if (!len)
2492                 return ERR_PTR(-EACCES);
2493 
2494         if (unlikely(name[0] == '.')) {
2495                 if (len < 2 || (len == 2 && name[1] == '.'))
2496                         return ERR_PTR(-EACCES);
2497         }
2498 
2499         while (len--) {
2500                 c = *(const unsigned char *)name++;
2501                 if (c == '/' || c == '\0')
2502                         return ERR_PTR(-EACCES);
2503         }
2504         /*
2505          * See if the low-level filesystem might want
2506          * to use its own hash..
2507          */
2508         if (base->d_flags & DCACHE_OP_HASH) {
2509                 int err = base->d_op->d_hash(base, &this);
2510                 if (err < 0)
2511                         return ERR_PTR(err);
2512         }
2513 
2514         err = inode_permission(base->d_inode, MAY_EXEC);
2515         if (err)
2516                 return ERR_PTR(err);
2517 
2518         ret = lookup_dcache(&this, base, 0);
2519         if (!ret)
2520                 ret = lookup_slow(&this, base, 0);
2521         return ret;
2522 }
2523 EXPORT_SYMBOL(lookup_one_len_unlocked);
2524 
2525 #ifdef CONFIG_UNIX98_PTYS
2526 int path_pts(struct path *path)
2527 {
2528         /* Find something mounted on "pts" in the same directory as
2529          * the input path.
2530          */
2531         struct dentry *child, *parent;
2532         struct qstr this;
2533         int ret;
2534 
2535         ret = path_parent_directory(path);
2536         if (ret)
2537                 return ret;
2538 
2539         parent = path->dentry;
2540         this.name = "pts";
2541         this.len = 3;
2542         child = d_hash_and_lookup(parent, &this);
2543         if (!child)
2544                 return -ENOENT;
2545 
2546         path->dentry = child;
2547         dput(parent);
2548         follow_mount(path);
2549         return 0;
2550 }
2551 #endif
2552 
2553 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2554                  struct path *path, int *empty)
2555 {
2556         return filename_lookup(dfd, getname_flags(name, flags, empty),
2557                                flags, path, NULL);
2558 }
2559 EXPORT_SYMBOL(user_path_at_empty);
2560 
2561 /*
2562  * NB: most callers don't do anything directly with the reference to the
2563  *     to struct filename, but the nd->last pointer points into the name string
2564  *     allocated by getname. So we must hold the reference to it until all
2565  *     path-walking is complete.
2566  */
2567 static inline struct filename *
2568 user_path_parent(int dfd, const char __user *path,
2569                  struct path *parent,
2570                  struct qstr *last,
2571                  int *type,
2572                  unsigned int flags)
2573 {
2574         /* only LOOKUP_REVAL is allowed in extra flags */
2575         return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2576                                  parent, last, type);
2577 }
2578 
2579 /**
2580  * mountpoint_last - look up last component for umount
2581  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2582  * @path: pointer to container for result
2583  *
2584  * This is a special lookup_last function just for umount. In this case, we
2585  * need to resolve the path without doing any revalidation.
2586  *
2587  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2588  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2589  * in almost all cases, this lookup will be served out of the dcache. The only
2590  * cases where it won't are if nd->last refers to a symlink or the path is
2591  * bogus and it doesn't exist.
2592  *
2593  * Returns:
2594  * -error: if there was an error during lookup. This includes -ENOENT if the
2595  *         lookup found a negative dentry. The nd->path reference will also be
2596  *         put in this case.
2597  *
2598  * 0:      if we successfully resolved nd->path and found it to not to be a
2599  *         symlink that needs to be followed. "path" will also be populated.
2600  *         The nd->path reference will also be put.
2601  *
2602  * 1:      if we successfully resolved nd->last and found it to be a symlink
2603  *         that needs to be followed. "path" will be populated with the path
2604  *         to the link, and nd->path will *not* be put.
2605  */
2606 static int
2607 mountpoint_last(struct nameidata *nd, struct path *path)
2608 {
2609         int error = 0;
2610         struct dentry *dentry;
2611         struct dentry *dir = nd->path.dentry;
2612 
2613         /* If we're in rcuwalk, drop out of it to handle last component */
2614         if (nd->flags & LOOKUP_RCU) {
2615                 if (unlazy_walk(nd, NULL, 0))
2616                         return -ECHILD;
2617         }
2618 
2619         nd->flags &= ~LOOKUP_PARENT;
2620 
2621         if (unlikely(nd->last_type != LAST_NORM)) {
2622                 error = handle_dots(nd, nd->last_type);
2623                 if (error)
2624                         return error;
2625                 dentry = dget(nd->path.dentry);
2626         } else {
2627                 dentry = d_lookup(dir, &nd->last);
2628                 if (!dentry) {
2629                         /*
2630                          * No cached dentry. Mounted dentries are pinned in the
2631                          * cache, so that means that this dentry is probably
2632                          * a symlink or the path doesn't actually point
2633                          * to a mounted dentry.
2634                          */
2635                         dentry = lookup_slow(&nd->last, dir,
2636                                              nd->flags | LOOKUP_NO_REVAL);
2637                         if (IS_ERR(dentry))
2638                                 return PTR_ERR(dentry);
2639                 }
2640         }
2641         if (d_is_negative(dentry)) {
2642                 dput(dentry);
2643                 return -ENOENT;
2644         }
2645         if (nd->depth)
2646                 put_link(nd);
2647         path->dentry = dentry;
2648         path->mnt = nd->path.mnt;
2649         error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2650                                    d_backing_inode(dentry), 0);
2651         if (unlikely(error))
2652                 return error;
2653         mntget(path->mnt);
2654         follow_mount(path);
2655         return 0;
2656 }
2657 
2658 /**
2659  * path_mountpoint - look up a path to be umounted
2660  * @nd:         lookup context
2661  * @flags:      lookup flags
2662  * @path:       pointer to container for result
2663  *
2664  * Look up the given name, but don't attempt to revalidate the last component.
2665  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2666  */
2667 static int
2668 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2669 {
2670         const char *s = path_init(nd, flags);
2671         int err;
2672         if (IS_ERR(s))
2673                 return PTR_ERR(s);
2674         while (!(err = link_path_walk(s, nd)) &&
2675                 (err = mountpoint_last(nd, path)) > 0) {
2676                 s = trailing_symlink(nd);
2677                 if (IS_ERR(s)) {
2678                         err = PTR_ERR(s);
2679                         break;
2680                 }
2681         }
2682         terminate_walk(nd);
2683         return err;
2684 }
2685 
2686 static int
2687 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2688                         unsigned int flags)
2689 {
2690         struct nameidata nd;
2691         int error;
2692         if (IS_ERR(name))
2693                 return PTR_ERR(name);
2694         set_nameidata(&nd, dfd, name);
2695         error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2696         if (unlikely(error == -ECHILD))
2697                 error = path_mountpoint(&nd, flags, path);
2698         if (unlikely(error == -ESTALE))
2699                 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2700         if (likely(!error))
2701                 audit_inode(name, path->dentry, 0);
2702         restore_nameidata();
2703         putname(name);
2704         return error;
2705 }
2706 
2707 /**
2708  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2709  * @dfd:        directory file descriptor
2710  * @name:       pathname from userland
2711  * @flags:      lookup flags
2712  * @path:       pointer to container to hold result
2713  *
2714  * A umount is a special case for path walking. We're not actually interested
2715  * in the inode in this situation, and ESTALE errors can be a problem. We
2716  * simply want track down the dentry and vfsmount attached at the mountpoint
2717  * and avoid revalidating the last component.
2718  *
2719  * Returns 0 and populates "path" on success.
2720  */
2721 int
2722 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2723                         struct path *path)
2724 {
2725         return filename_mountpoint(dfd, getname(name), path, flags);
2726 }
2727 
2728 int
2729 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2730                         unsigned int flags)
2731 {
2732         return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2733 }
2734 EXPORT_SYMBOL(kern_path_mountpoint);
2735 
2736 int __check_sticky(struct inode *dir, struct inode *inode)
2737 {
2738         kuid_t fsuid = current_fsuid();
2739 
2740         if (uid_eq(inode->i_uid, fsuid))
2741                 return 0;
2742         if (uid_eq(dir->i_uid, fsuid))
2743                 return 0;
2744         return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2745 }
2746 EXPORT_SYMBOL(__check_sticky);
2747 
2748 /*
2749  *      Check whether we can remove a link victim from directory dir, check
2750  *  whether the type of victim is right.
2751  *  1. We can't do it if dir is read-only (done in permission())
2752  *  2. We should have write and exec permissions on dir
2753  *  3. We can't remove anything from append-only dir
2754  *  4. We can't do anything with immutable dir (done in permission())
2755  *  5. If the sticky bit on dir is set we should either
2756  *      a. be owner of dir, or
2757  *      b. be owner of victim, or
2758  *      c. have CAP_FOWNER capability
2759  *  6. If the victim is append-only or immutable we can't do antyhing with
2760  *     links pointing to it.
2761  *  7. If the victim has an unknown uid or gid we can't change the inode.
2762  *  8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2763  *  9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2764  * 10. We can't remove a root or mountpoint.
2765  * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2766  *     nfs_async_unlink().
2767  */
2768 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2769 {
2770         struct inode *inode = d_backing_inode(victim);
2771         int error;
2772 
2773         if (d_is_negative(victim))
2774                 return -ENOENT;
2775         BUG_ON(!inode);
2776 
2777         BUG_ON(victim->d_parent->d_inode != dir);
2778         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2779 
2780         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2781         if (error)
2782                 return error;
2783         if (IS_APPEND(dir))
2784                 return -EPERM;
2785 
2786         if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2787             IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2788                 return -EPERM;
2789         if (isdir) {
2790                 if (!d_is_dir(victim))
2791                         return -ENOTDIR;
2792                 if (IS_ROOT(victim))
2793                         return -EBUSY;
2794         } else if (d_is_dir(victim))
2795                 return -EISDIR;
2796         if (IS_DEADDIR(dir))
2797                 return -ENOENT;
2798         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2799                 return -EBUSY;
2800         return 0;
2801 }
2802 
2803 /*      Check whether we can create an object with dentry child in directory
2804  *  dir.
2805  *  1. We can't do it if child already exists (open has special treatment for
2806  *     this case, but since we are inlined it's OK)
2807  *  2. We can't do it if dir is read-only (done in permission())
2808  *  3. We can't do it if the fs can't represent the fsuid or fsgid.
2809  *  4. We should have write and exec permissions on dir
2810  *  5. We can't do it if dir is immutable (done in permission())
2811  */
2812 static inline int may_create(struct inode *dir, struct dentry *child)
2813 {
2814         struct user_namespace *s_user_ns;
2815         audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2816         if (child->d_inode)
2817                 return -EEXIST;
2818         if (IS_DEADDIR(dir))
2819                 return -ENOENT;
2820         s_user_ns = dir->i_sb->s_user_ns;
2821         if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2822             !kgid_has_mapping(s_user_ns, current_fsgid()))
2823                 return -EOVERFLOW;
2824         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2825 }
2826 
2827 /*
2828  * p1 and p2 should be directories on the same fs.
2829  */
2830 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2831 {
2832         struct dentry *p;
2833 
2834         if (p1 == p2) {
2835                 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2836                 return NULL;
2837         }
2838 
2839         mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2840 
2841         p = d_ancestor(p2, p1);
2842         if (p) {
2843                 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2844                 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2845                 return p;
2846         }
2847 
2848         p = d_ancestor(p1, p2);
2849         if (p) {
2850                 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2851                 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2852                 return p;
2853         }
2854 
2855         inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2856         inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2857         return NULL;
2858 }
2859 EXPORT_SYMBOL(lock_rename);
2860 
2861 void unlock_rename(struct dentry *p1, struct dentry *p2)
2862 {
2863         inode_unlock(p1->d_inode);
2864         if (p1 != p2) {
2865                 inode_unlock(p2->d_inode);
2866                 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2867         }
2868 }
2869 EXPORT_SYMBOL(unlock_rename);
2870 
2871 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2872                 bool want_excl)
2873 {
2874         int error = may_create(dir, dentry);
2875         if (error)
2876                 return error;
2877 
2878         if (!dir->i_op->create)
2879                 return -EACCES; /* shouldn't it be ENOSYS? */
2880         mode &= S_IALLUGO;
2881         mode |= S_IFREG;
2882         error = security_inode_create(dir, dentry, mode);
2883         if (error)
2884                 return error;
2885         error = dir->i_op->create(dir, dentry, mode, want_excl);
2886         if (!error)
2887                 fsnotify_create(dir, dentry);
2888         return error;
2889 }
2890 EXPORT_SYMBOL(vfs_create);
2891 
2892 bool may_open_dev(const struct path *path)
2893 {
2894         return !(path->mnt->mnt_flags & MNT_NODEV) &&
2895                 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2896 }
2897 
2898 static int may_open(struct path *path, int acc_mode, int flag)
2899 {
2900         struct dentry *dentry = path->dentry;
2901         struct inode *inode = dentry->d_inode;
2902         int error;
2903 
2904         if (!inode)
2905                 return -ENOENT;
2906 
2907         switch (inode->i_mode & S_IFMT) {
2908         case S_IFLNK:
2909                 return -ELOOP;
2910         case S_IFDIR:
2911                 if (acc_mode & MAY_WRITE)
2912                         return -EISDIR;
2913                 break;
2914         case S_IFBLK:
2915         case S_IFCHR:
2916                 if (!may_open_dev(path))
2917                         return -EACCES;
2918                 /*FALLTHRU*/
2919         case S_IFIFO:
2920         case S_IFSOCK:
2921                 flag &= ~O_TRUNC;
2922                 break;
2923         }
2924 
2925         error = inode_permission(inode, MAY_OPEN | acc_mode);
2926         if (error)
2927                 return error;
2928 
2929         /*
2930          * An append-only file must be opened in append mode for writing.
2931          */
2932         if (IS_APPEND(inode)) {
2933                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2934                         return -EPERM;
2935                 if (flag & O_TRUNC)
2936                         return -EPERM;
2937         }
2938 
2939         /* O_NOATIME can only be set by the owner or superuser */
2940         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2941                 return -EPERM;
2942 
2943         return 0;
2944 }
2945 
2946 static int handle_truncate(struct file *filp)
2947 {
2948         struct path *path = &filp->f_path;
2949         struct inode *inode = path->dentry->d_inode;
2950         int error = get_write_access(inode);
2951         if (error)
2952                 return error;
2953         /*
2954          * Refuse to truncate files with mandatory locks held on them.
2955          */
2956         error = locks_verify_locked(filp);
2957         if (!error)
2958                 error = security_path_truncate(path);
2959         if (!error) {
2960                 error = do_truncate(path->dentry, 0,
2961                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2962                                     filp);
2963         }
2964         put_write_access(inode);
2965         return error;
2966 }
2967 
2968 static inline int open_to_namei_flags(int flag)
2969 {
2970         if ((flag & O_ACCMODE) == 3)
2971                 flag--;
2972         return flag;
2973 }
2974 
2975 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2976 {
2977         int error = security_path_mknod(dir, dentry, mode, 0);
2978         if (error)
2979                 return error;
2980 
2981         error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2982         if (error)
2983                 return error;
2984 
2985         return security_inode_create(dir->dentry->d_inode, dentry, mode);
2986 }
2987 
2988 /*
2989  * Attempt to atomically look up, create and open a file from a negative
2990  * dentry.
2991  *
2992  * Returns 0 if successful.  The file will have been created and attached to
2993  * @file by the filesystem calling finish_open().
2994  *
2995  * Returns 1 if the file was looked up only or didn't need creating.  The
2996  * caller will need to perform the open themselves.  @path will have been
2997  * updated to point to the new dentry.  This may be negative.
2998  *
2999  * Returns an error code otherwise.
3000  */
3001 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3002                         struct path *path, struct file *file,
3003                         const struct open_flags *op,
3004                         int open_flag, umode_t mode,
3005                         int *opened)
3006 {
3007         struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3008         struct inode *dir =  nd->path.dentry->d_inode;
3009         int error;
3010 
3011         if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3012                 open_flag &= ~O_TRUNC;
3013 
3014         if (nd->flags & LOOKUP_DIRECTORY)
3015                 open_flag |= O_DIRECTORY;
3016 
3017         file->f_path.dentry = DENTRY_NOT_SET;
3018         file->f_path.mnt = nd->path.mnt;
3019         error = dir->i_op->atomic_open(dir, dentry, file,
3020                                        open_to_namei_flags(open_flag),
3021                                        mode, opened);
3022         d_lookup_done(dentry);
3023         if (!error) {
3024                 /*
3025                  * We didn't have the inode before the open, so check open
3026                  * permission here.
3027                  */
3028                 int acc_mode = op->acc_mode;
3029                 if (*opened & FILE_CREATED) {
3030                         WARN_ON(!(open_flag & O_CREAT));
3031                         fsnotify_create(dir, dentry);
3032                         acc_mode = 0;
3033                 }
3034                 error = may_open(&file->f_path, acc_mode, open_flag);
3035                 if (WARN_ON(error > 0))
3036                         error = -EINVAL;
3037         } else if (error > 0) {
3038                 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3039                         error = -EIO;
3040                 } else {
3041                         if (file->f_path.dentry) {
3042                                 dput(dentry);
3043                                 dentry = file->f_path.dentry;
3044                         }
3045                         if (*opened & FILE_CREATED)
3046                                 fsnotify_create(dir, dentry);
3047                         if (unlikely(d_is_negative(dentry))) {
3048                                 error = -ENOENT;
3049                         } else {
3050                                 path->dentry = dentry;
3051                                 path->mnt = nd->path.mnt;
3052                                 return 1;
3053                         }
3054                 }
3055         }
3056         dput(dentry);
3057         return error;
3058 }
3059 
3060 /*
3061  * Look up and maybe create and open the last component.
3062  *
3063  * Must be called with i_mutex held on parent.
3064  *
3065  * Returns 0 if the file was successfully atomically created (if necessary) and
3066  * opened.  In this case the file will be returned attached to @file.
3067  *
3068  * Returns 1 if the file was not completely opened at this time, though lookups
3069  * and creations will have been performed and the dentry returned in @path will
3070  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
3071  * specified then a negative dentry may be returned.
3072  *
3073  * An error code is returned otherwise.
3074  *
3075  * FILE_CREATE will be set in @*opened if the dentry was created and will be
3076  * cleared otherwise prior to returning.
3077  */
3078 static int lookup_open(struct nameidata *nd, struct path *path,
3079                         struct file *file,
3080                         const struct open_flags *op,
3081                         bool got_write, int *opened)
3082 {
3083         struct dentry *dir = nd->path.dentry;
3084         struct inode *dir_inode = dir->d_inode;
3085         int open_flag = op->open_flag;
3086         struct dentry *dentry;
3087         int error, create_error = 0;
3088         umode_t mode = op->mode;
3089         DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3090 
3091         if (unlikely(IS_DEADDIR(dir_inode)))
3092                 return -ENOENT;
3093 
3094         *opened &= ~FILE_CREATED;
3095         dentry = d_lookup(dir, &nd->last);
3096         for (;;) {
3097                 if (!dentry) {
3098                         dentry = d_alloc_parallel(dir, &nd->last, &wq);
3099                         if (IS_ERR(dentry))
3100                                 return PTR_ERR(dentry);
3101                 }
3102                 if (d_in_lookup(dentry))
3103                         break;
3104 
3105                 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3106                         break;
3107 
3108                 error = d_revalidate(dentry, nd->flags);
3109                 if (likely(error > 0))
3110                         break;
3111                 if (error)
3112                         goto out_dput;
3113                 d_invalidate(dentry);
3114                 dput(dentry);
3115                 dentry = NULL;
3116         }
3117         if (dentry->d_inode) {
3118                 /* Cached positive dentry: will open in f_op->open */
3119                 goto out_no_open;
3120         }
3121 
3122         /*
3123          * Checking write permission is tricky, bacuse we don't know if we are
3124          * going to actually need it: O_CREAT opens should work as long as the
3125          * file exists.  But checking existence breaks atomicity.  The trick is
3126          * to check access and if not granted clear O_CREAT from the flags.
3127          *
3128          * Another problem is returing the "right" error value (e.g. for an
3129          * O_EXCL open we want to return EEXIST not EROFS).
3130          */
3131         if (open_flag & O_CREAT) {
3132                 if (!IS_POSIXACL(dir->d_inode))
3133                         mode &= ~current_umask();
3134                 if (unlikely(!got_write)) {
3135                         create_error = -EROFS;
3136                         open_flag &= ~O_CREAT;
3137                         if (open_flag & (O_EXCL | O_TRUNC))
3138                                 goto no_open;
3139                         /* No side effects, safe to clear O_CREAT */
3140                 } else {
3141                         create_error = may_o_create(&nd->path, dentry, mode);
3142                         if (create_error) {
3143                                 open_flag &= ~O_CREAT;
3144                                 if (open_flag & O_EXCL)
3145                                         goto no_open;
3146                         }
3147                 }
3148         } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3149                    unlikely(!got_write)) {
3150                 /*
3151                  * No O_CREATE -> atomicity not a requirement -> fall
3152                  * back to lookup + open
3153                  */
3154                 goto no_open;
3155         }
3156 
3157         if (dir_inode->i_op->atomic_open) {
3158                 error = atomic_open(nd, dentry, path, file, op, open_flag,
3159                                     mode, opened);
3160                 if (unlikely(error == -ENOENT) && create_error)
3161                         error = create_error;
3162                 return error;
3163         }
3164 
3165 no_open:
3166         if (d_in_lookup(dentry)) {
3167                 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3168                                                              nd->flags);
3169                 d_lookup_done(dentry);
3170                 if (unlikely(res)) {
3171                         if (IS_ERR(res)) {
3172                                 error = PTR_ERR(res);
3173                                 goto out_dput;
3174                         }
3175                         dput(dentry);
3176                         dentry = res;
3177                 }
3178         }
3179 
3180         /* Negative dentry, just create the file */
3181         if (!dentry->d_inode && (open_flag & O_CREAT)) {
3182                 *opened |= FILE_CREATED;
3183                 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3184                 if (!dir_inode->i_op->create) {
3185                         error = -EACCES;
3186                         goto out_dput;
3187                 }
3188                 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3189                                                 open_flag & O_EXCL);
3190                 if (error)
3191                         goto out_dput;
3192                 fsnotify_create(dir_inode, dentry);
3193         }
3194         if (unlikely(create_error) && !dentry->d_inode) {
3195                 error = create_error;
3196                 goto out_dput;
3197         }
3198 out_no_open:
3199         path->dentry = dentry;
3200         path->mnt = nd->path.mnt;
3201         return 1;
3202 
3203 out_dput:
3204         dput(dentry);
3205         return error;
3206 }
3207 
3208 /*
3209  * Handle the last step of open()
3210  */
3211 static int do_last(struct nameidata *nd,
3212                    struct file *file, const struct open_flags *op,
3213                    int *opened)
3214 {
3215         struct dentry *dir = nd->path.dentry;
3216         int open_flag = op->open_flag;
3217         bool will_truncate = (open_flag & O_TRUNC) != 0;
3218         bool got_write = false;
3219         int acc_mode = op->acc_mode;
3220         unsigned seq;
3221         struct inode *inode;
3222         struct path path;
3223         int error;
3224 
3225         nd->flags &= ~LOOKUP_PARENT;
3226         nd->flags |= op->intent;
3227 
3228         if (nd->last_type != LAST_NORM) {
3229                 error = handle_dots(nd, nd->last_type);
3230                 if (unlikely(error))
3231                         return error;
3232                 goto finish_open;
3233         }
3234 
3235         if (!(open_flag & O_CREAT)) {
3236                 if (nd->last.name[nd->last.len])
3237                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3238                 /* we _can_ be in RCU mode here */
3239                 error = lookup_fast(nd, &path, &inode, &seq);
3240                 if (likely(error > 0))
3241                         goto finish_lookup;
3242 
3243                 if (error < 0)
3244                         return error;
3245 
3246                 BUG_ON(nd->inode != dir->d_inode);
3247                 BUG_ON(nd->flags & LOOKUP_RCU);
3248         } else {
3249                 /* create side of things */
3250                 /*
3251                  * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3252                  * has been cleared when we got to the last component we are
3253                  * about to look up
3254                  */
3255                 error = complete_walk(nd);
3256                 if (error)
3257                         return error;
3258 
3259                 audit_inode(nd->name, dir, LOOKUP_PARENT);
3260                 /* trailing slashes? */
3261                 if (unlikely(nd->last.name[nd->last.len]))
3262                         return -EISDIR;
3263         }
3264 
3265         if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3266                 error = mnt_want_write(nd->path.mnt);
3267                 if (!error)
3268                         got_write = true;
3269                 /*
3270                  * do _not_ fail yet - we might not need that or fail with
3271                  * a different error; let lookup_open() decide; we'll be
3272                  * dropping this one anyway.
3273                  */
3274         }
3275         if (open_flag & O_CREAT)
3276                 inode_lock(dir->d_inode);
3277         else
3278                 inode_lock_shared(dir->d_inode);
3279         error = lookup_open(nd, &path, file, op, got_write, opened);
3280         if (open_flag & O_CREAT)
3281                 inode_unlock(dir->d_inode);
3282         else
3283                 inode_unlock_shared(dir->d_inode);
3284 
3285         if (error <= 0) {
3286                 if (error)
3287                         goto out;
3288 
3289                 if ((*opened & FILE_CREATED) ||
3290                     !S_ISREG(file_inode(file)->i_mode))
3291                         will_truncate = false;
3292 
3293                 audit_inode(nd->name, file->f_path.dentry, 0);
3294                 goto opened;
3295         }
3296 
3297         if (*opened & FILE_CREATED) {
3298                 /* Don't check for write permission, don't truncate */
3299                 open_flag &= ~O_TRUNC;
3300                 will_truncate = false;
3301                 acc_mode = 0;
3302                 path_to_nameidata(&path, nd);
3303                 goto finish_open_created;
3304         }
3305 
3306         /*
3307          * If atomic_open() acquired write access it is dropped now due to
3308          * possible mount and symlink following (this might be optimized away if
3309          * necessary...)
3310          */
3311         if (got_write) {
3312                 mnt_drop_write(nd->path.mnt);
3313                 got_write = false;
3314         }
3315 
3316         error = follow_managed(&path, nd);
3317         if (unlikely(error < 0))
3318                 return error;
3319 
3320         if (unlikely(d_is_negative(path.dentry))) {
3321                 path_to_nameidata(&path, nd);
3322                 return -ENOENT;
3323         }
3324 
3325         /*
3326          * create/update audit record if it already exists.
3327          */
3328         audit_inode(nd->name, path.dentry, 0);
3329 
3330         if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3331                 path_to_nameidata(&path, nd);
3332                 return -EEXIST;
3333         }
3334 
3335         seq = 0;        /* out of RCU mode, so the value doesn't matter */
3336         inode = d_backing_inode(path.dentry);
3337 finish_lookup:
3338         if (nd->depth)
3339                 put_link(nd);
3340         error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3341                                    inode, seq);
3342         if (unlikely(error))
3343                 return error;
3344 
3345         path_to_nameidata(&path, nd);
3346         nd->inode = inode;
3347         nd->seq = seq;
3348         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3349 finish_open:
3350         error = complete_walk(nd);
3351         if (error)
3352                 return error;
3353         audit_inode(nd->name, nd->path.dentry, 0);
3354         error = -EISDIR;
3355         if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3356                 goto out;
3357         error = -ENOTDIR;
3358         if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3359                 goto out;
3360         if (!d_is_reg(nd->path.dentry))
3361                 will_truncate = false;
3362 
3363         if (will_truncate) {
3364                 error = mnt_want_write(nd->path.mnt);
3365                 if (error)
3366                         goto out;
3367                 got_write = true;
3368         }
3369 finish_open_created:
3370         error = may_open(&nd->path, acc_mode, open_flag);
3371         if (error)
3372                 goto out;
3373         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3374         error = vfs_open(&nd->path, file, current_cred());
3375         if (error)
3376                 goto out;
3377         *opened |= FILE_OPENED;
3378 opened:
3379         error = open_check_o_direct(file);
3380         if (!error)
3381                 error = ima_file_check(file, op->acc_mode, *opened);
3382         if (!error && will_truncate)
3383                 error = handle_truncate(file);
3384 out:
3385         if (unlikely(error) && (*opened & FILE_OPENED))
3386                 fput(file);
3387         if (unlikely(error > 0)) {
3388                 WARN_ON(1);
3389                 error = -EINVAL;
3390         }
3391         if (got_write)
3392                 mnt_drop_write(nd->path.mnt);
3393         return error;
3394 }
3395 
3396 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3397                 const struct open_flags *op,
3398                 struct file *file, int *opened)
3399 {
3400         static const struct qstr name = QSTR_INIT("/", 1);
3401         struct dentry *child;
3402         struct inode *dir;
3403         struct path path;
3404         int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3405         if (unlikely(error))
3406                 return error;
3407         error = mnt_want_write(path.mnt);
3408         if (unlikely(error))
3409                 goto out;
3410         dir = path.dentry->d_inode;
3411         /* we want directory to be writable */
3412         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3413         if (error)
3414                 goto out2;
3415         if (!dir->i_op->tmpfile) {
3416                 error = -EOPNOTSUPP;
3417                 goto out2;
3418         }
3419         child = d_alloc(path.dentry, &name);
3420         if (unlikely(!child)) {
3421                 error = -ENOMEM;
3422                 goto out2;
3423         }
3424         dput(path.dentry);
3425         path.dentry = child;
3426         error = dir->i_op->tmpfile(dir, child, op->mode);
3427         if (error)
3428                 goto out2;
3429         audit_inode(nd->name, child, 0);
3430         /* Don't check for other permissions, the inode was just created */
3431         error = may_open(&path, 0, op->open_flag);
3432         if (error)
3433                 goto out2;
3434         file->f_path.mnt = path.mnt;
3435         error = finish_open(file, child, NULL, opened);
3436         if (error)
3437                 goto out2;
3438         error = open_check_o_direct(file);
3439         if (error) {
3440                 fput(file);
3441         } else if (!(op->open_flag & O_EXCL)) {
3442                 struct inode *inode = file_inode(file);
3443                 spin_lock(&inode->i_lock);
3444                 inode->i_state |= I_LINKABLE;
3445                 spin_unlock(&inode->i_lock);
3446         }
3447 out2:
3448         mnt_drop_write(path.mnt);
3449 out:
3450         path_put(&path);
3451         return error;
3452 }
3453 
3454 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3455 {
3456         struct path path;
3457         int error = path_lookupat(nd, flags, &path);
3458         if (!error) {
3459                 audit_inode(nd->name, path.dentry, 0);
3460                 error = vfs_open(&path, file, current_cred());
3461                 path_put(&path);
3462         }
3463         return error;
3464 }
3465 
3466 static struct file *path_openat(struct nameidata *nd,
3467                         const struct open_flags *op, unsigned flags)
3468 {
3469         const char *s;
3470         struct file *file;
3471         int opened = 0;
3472         int error;
3473 
3474         file = get_empty_filp();
3475         if (IS_ERR(file))
3476                 return file;
3477 
3478         file->f_flags = op->open_flag;
3479 
3480         if (unlikely(file->f_flags & __O_TMPFILE)) {
3481                 error = do_tmpfile(nd, flags, op, file, &opened);
3482                 goto out2;
3483         }
3484 
3485         if (unlikely(file->f_flags & O_PATH)) {
3486                 error = do_o_path(nd, flags, file);
3487                 if (!error)
3488                         opened |= FILE_OPENED;
3489                 goto out2;
3490         }
3491 
3492         s = path_init(nd, flags);
3493         if (IS_ERR(s)) {
3494                 put_filp(file);
3495                 return ERR_CAST(s);
3496         }
3497         while (!(error = link_path_walk(s, nd)) &&
3498                 (error = do_last(nd, file, op, &opened)) > 0) {
3499                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3500                 s = trailing_symlink(nd);
3501                 if (IS_ERR(s)) {
3502                         error = PTR_ERR(s);
3503                         break;
3504                 }
3505         }
3506         terminate_walk(nd);
3507 out2:
3508         if (!(opened & FILE_OPENED)) {
3509                 BUG_ON(!error);
3510                 put_filp(file);
3511         }
3512         if (unlikely(error)) {
3513                 if (error == -EOPENSTALE) {
3514                         if (flags & LOOKUP_RCU)
3515                                 error = -ECHILD;
3516                         else
3517                                 error = -ESTALE;
3518                 }
3519                 file = ERR_PTR(error);
3520         }
3521         return file;
3522 }
3523 
3524 struct file *do_filp_open(int dfd, struct filename *pathname,
3525                 const struct open_flags *op)
3526 {
3527         struct nameidata nd;
3528         int flags = op->lookup_flags;
3529         struct file *filp;
3530 
3531         set_nameidata(&nd, dfd, pathname);
3532         filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3533         if (unlikely(filp == ERR_PTR(-ECHILD)))
3534                 filp = path_openat(&nd, op, flags);
3535         if (unlikely(filp == ERR_PTR(-ESTALE)))
3536                 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3537         restore_nameidata();
3538         return filp;
3539 }
3540 
3541 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3542                 const char *name, const struct open_flags *op)
3543 {
3544         struct nameidata nd;
3545         struct file *file;
3546         struct filename *filename;
3547         int flags = op->lookup_flags | LOOKUP_ROOT;
3548 
3549         nd.root.mnt = mnt;
3550         nd.root.dentry = dentry;
3551 
3552         if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3553                 return ERR_PTR(-ELOOP);
3554 
3555         filename = getname_kernel(name);
3556         if (IS_ERR(filename))
3557                 return ERR_CAST(filename);
3558 
3559         set_nameidata(&nd, -1, filename);
3560         file = path_openat(&nd, op, flags | LOOKUP_RCU);
3561         if (unlikely(file == ERR_PTR(-ECHILD)))
3562                 file = path_openat(&nd, op, flags);
3563         if (unlikely(file == ERR_PTR(-ESTALE)))
3564                 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3565         restore_nameidata();
3566         putname(filename);
3567         return file;
3568 }
3569 
3570 static struct dentry *filename_create(int dfd, struct filename *name,
3571                                 struct path *path, unsigned int lookup_flags)
3572 {
3573         struct dentry *dentry = ERR_PTR(-EEXIST);
3574         struct qstr last;
3575         int type;
3576         int err2;
3577         int error;
3578         bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3579 
3580         /*
3581          * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3582          * other flags passed in are ignored!
3583          */
3584         lookup_flags &= LOOKUP_REVAL;
3585 
3586         name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3587         if (IS_ERR(name))
3588                 return ERR_CAST(name);
3589 
3590         /*
3591          * Yucky last component or no last component at all?
3592          * (foo/., foo/.., /////)
3593          */
3594         if (unlikely(type != LAST_NORM))
3595                 goto out;
3596 
3597         /* don't fail immediately if it's r/o, at least try to report other errors */
3598         err2 = mnt_want_write(path->mnt);
3599         /*
3600          * Do the final lookup.
3601          */
3602         lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3603         inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3604         dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3605         if (IS_ERR(dentry))
3606                 goto unlock;
3607 
3608         error = -EEXIST;
3609         if (d_is_positive(dentry))
3610                 goto fail;
3611 
3612         /*
3613          * Special case - lookup gave negative, but... we had foo/bar/
3614          * From the vfs_mknod() POV we just have a negative dentry -
3615          * all is fine. Let's be bastards - you had / on the end, you've
3616          * been asking for (non-existent) directory. -ENOENT for you.
3617          */
3618         if (unlikely(!is_dir && last.name[last.len])) {
3619                 error = -ENOENT;
3620                 goto fail;
3621         }
3622         if (unlikely(err2)) {
3623                 error = err2;
3624                 goto fail;
3625         }
3626         putname(name);
3627         return dentry;
3628 fail:
3629         dput(dentry);
3630         dentry = ERR_PTR(error);
3631 unlock:
3632         inode_unlock(path->dentry->d_inode);
3633         if (!err2)
3634                 mnt_drop_write(path->mnt);
3635 out:
3636         path_put(path);
3637         putname(name);
3638         return dentry;
3639 }
3640 
3641 struct dentry *kern_path_create(int dfd, const char *pathname,
3642                                 struct path *path, unsigned int lookup_flags)
3643 {
3644         return filename_create(dfd, getname_kernel(pathname),
3645                                 path, lookup_flags);
3646 }
3647 EXPORT_SYMBOL(kern_path_create);
3648 
3649 void done_path_create(struct path *path, struct dentry *dentry)
3650 {
3651         dput(dentry);
3652         inode_unlock(path->dentry->d_inode);
3653         mnt_drop_write(path->mnt);
3654         path_put(path);
3655 }
3656 EXPORT_SYMBOL(done_path_create);
3657 
3658 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3659                                 struct path *path, unsigned int lookup_flags)
3660 {
3661         return filename_create(dfd, getname(pathname), path, lookup_flags);
3662 }
3663 EXPORT_SYMBOL(user_path_create);
3664 
3665 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3666 {
3667         int error = may_create(dir, dentry);
3668 
3669         if (error)
3670                 return error;
3671 
3672         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3673                 return -EPERM;
3674 
3675         if (!dir->i_op->mknod)
3676                 return -EPERM;
3677 
3678         error = devcgroup_inode_mknod(mode, dev);
3679         if (error)
3680                 return error;
3681 
3682         error = security_inode_mknod(dir, dentry, mode, dev);
3683         if (error)
3684                 return error;
3685 
3686         error = dir->i_op->mknod(dir, dentry, mode, dev);
3687         if (!error)
3688                 fsnotify_create(dir, dentry);
3689         return error;
3690 }
3691 EXPORT_SYMBOL(vfs_mknod);
3692 
3693 static int may_mknod(umode_t mode)
3694 {
3695         switch (mode & S_IFMT) {
3696         case S_IFREG:
3697         case S_IFCHR:
3698         case S_IFBLK:
3699         case S_IFIFO:
3700         case S_IFSOCK:
3701         case 0: /* zero mode translates to S_IFREG */
3702                 return 0;
3703         case S_IFDIR:
3704                 return -EPERM;
3705         default:
3706                 return -EINVAL;
3707         }
3708 }
3709 
3710 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3711                 unsigned, dev)
3712 {
3713         struct dentry *dentry;
3714         struct path path;
3715         int error;
3716         unsigned int lookup_flags = 0;
3717 
3718         error = may_mknod(mode);
3719         if (error)
3720                 return error;
3721 retry:
3722         dentry = user_path_create(dfd, filename, &path, lookup_flags);
3723         if (IS_ERR(dentry))
3724                 return PTR_ERR(dentry);
3725 
3726         if (!IS_POSIXACL(path.dentry->d_inode))
3727                 mode &= ~current_umask();
3728         error = security_path_mknod(&path, dentry, mode, dev);
3729         if (error)
3730                 goto out;
3731         switch (mode & S_IFMT) {
3732                 case 0: case S_IFREG:
3733                         error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3734                         if (!error)
3735                                 ima_post_path_mknod(dentry);
3736                         break;
3737                 case S_IFCHR: case S_IFBLK:
3738                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3739                                         new_decode_dev(dev));
3740                         break;
3741                 case S_IFIFO: case S_IFSOCK:
3742                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3743                         break;
3744         }
3745 out:
3746         done_path_create(&path, dentry);
3747         if (retry_estale(error, lookup_flags)) {
3748                 lookup_flags |= LOOKUP_REVAL;
3749                 goto retry;
3750         }
3751         return error;
3752 }
3753 
3754 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3755 {
3756         return sys_mknodat(AT_FDCWD, filename, mode, dev);
3757 }
3758 
3759 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3760 {
3761         int error = may_create(dir, dentry);
3762         unsigned max_links = dir->i_sb->s_max_links;
3763 
3764         if (error)
3765                 return error;
3766 
3767         if (!dir->i_op->mkdir)
3768                 return -EPERM;
3769 
3770         mode &= (S_IRWXUGO|S_ISVTX);
3771         error = security_inode_mkdir(dir, dentry, mode);
3772         if (error)
3773                 return error;
3774 
3775         if (max_links && dir->i_nlink >= max_links)
3776                 return -EMLINK;
3777 
3778         error = dir->i_op->mkdir(dir, dentry, mode);
3779         if (!error)
3780                 fsnotify_mkdir(dir, dentry);
3781         return error;
3782 }
3783 EXPORT_SYMBOL(vfs_mkdir);
3784 
3785 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3786 {
3787         struct dentry *dentry;
3788         struct path path;
3789         int error;
3790         unsigned int lookup_flags = LOOKUP_DIRECTORY;
3791 
3792 retry:
3793         dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3794         if (IS_ERR(dentry))
3795                 return PTR_ERR(dentry);
3796 
3797         if (!IS_POSIXACL(path.dentry->d_inode))
3798                 mode &= ~current_umask();
3799         error = security_path_mkdir(&path, dentry, mode);
3800         if (!error)
3801                 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3802         done_path_create(&path, dentry);
3803         if (retry_estale(error, lookup_flags)) {
3804                 lookup_flags |= LOOKUP_REVAL;
3805                 goto retry;
3806         }
3807         return error;
3808 }
3809 
3810 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3811 {
3812         return sys_mkdirat(AT_FDCWD, pathname, mode);
3813 }
3814 
3815 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3816 {
3817         int error = may_delete(dir, dentry, 1);
3818 
3819         if (error)
3820                 return error;
3821 
3822         if (!dir->i_op->rmdir)
3823                 return -EPERM;
3824 
3825         dget(dentry);
3826         inode_lock(dentry->d_inode);
3827 
3828         error = -EBUSY;
3829         if (is_local_mountpoint(dentry))
3830                 goto out;
3831 
3832         error = security_inode_rmdir(dir, dentry);
3833         if (error)
3834                 goto out;
3835 
3836         shrink_dcache_parent(dentry);
3837         error = dir->i_op->rmdir(dir, dentry);
3838         if (error)
3839                 goto out;
3840 
3841         dentry->d_inode->i_flags |= S_DEAD;
3842         dont_mount(dentry);
3843         detach_mounts(dentry);
3844 
3845 out:
3846         inode_unlock(dentry->d_inode);
3847         dput(dentry);
3848         if (!error)
3849                 d_delete(dentry);
3850         return error;
3851 }
3852 EXPORT_SYMBOL(vfs_rmdir);
3853 
3854 static long do_rmdir(int dfd, const char __user *pathname)
3855 {
3856         int error = 0;
3857         struct filename *name;
3858         struct dentry *dentry;
3859         struct path path;
3860         struct qstr last;
3861         int type;
3862         unsigned int lookup_flags = 0;
3863 retry:
3864         name = user_path_parent(dfd, pathname,
3865                                 &path, &last, &type, lookup_flags);
3866         if (IS_ERR(name))
3867                 return PTR_ERR(name);
3868 
3869         switch (type) {
3870         case LAST_DOTDOT:
3871                 error = -ENOTEMPTY;
3872                 goto exit1;
3873         case LAST_DOT:
3874                 error = -EINVAL;
3875                 goto exit1;
3876         case LAST_ROOT:
3877                 error = -EBUSY;
3878                 goto exit1;
3879         }
3880 
3881         error = mnt_want_write(path.mnt);
3882         if (error)
3883                 goto exit1;
3884 
3885         inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3886         dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3887         error = PTR_ERR(dentry);
3888         if (IS_ERR(dentry))
3889                 goto exit2;
3890         if (!dentry->d_inode) {
3891                 error = -ENOENT;
3892                 goto exit3;
3893         }
3894         error = security_path_rmdir(&path, dentry);
3895         if (error)
3896                 goto exit3;
3897         error = vfs_rmdir(path.dentry->d_inode, dentry);
3898 exit3:
3899         dput(dentry);
3900 exit2:
3901         inode_unlock(path.dentry->d_inode);
3902         mnt_drop_write(path.mnt);
3903 exit1:
3904         path_put(&path);
3905         putname(name);
3906         if (retry_estale(error, lookup_flags)) {
3907                 lookup_flags |= LOOKUP_REVAL;
3908                 goto retry;
3909         }
3910         return error;
3911 }
3912 
3913 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3914 {
3915         return do_rmdir(AT_FDCWD, pathname);
3916 }
3917 
3918 /**
3919  * vfs_unlink - unlink a filesystem object
3920  * @dir:        parent directory
3921  * @dentry:     victim
3922  * @delegated_inode: returns victim inode, if the inode is delegated.
3923  *
3924  * The caller must hold dir->i_mutex.
3925  *
3926  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3927  * return a reference to the inode in delegated_inode.  The caller
3928  * should then break the delegation on that inode and retry.  Because
3929  * breaking a delegation may take a long time, the caller should drop
3930  * dir->i_mutex before doing so.
3931  *
3932  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3933  * be appropriate for callers that expect the underlying filesystem not
3934  * to be NFS exported.
3935  */
3936 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3937 {
3938         struct inode *target = dentry->d_inode;
3939         int error = may_delete(dir, dentry, 0);
3940 
3941         if (error)
3942                 return error;
3943 
3944         if (!dir->i_op->unlink)
3945                 return -EPERM;
3946 
3947         inode_lock(target);
3948         if (is_local_mountpoint(dentry))
3949                 error = -EBUSY;
3950         else {
3951                 error = security_inode_unlink(dir, dentry);
3952                 if (!error) {
3953                         error = try_break_deleg(target, delegated_inode);
3954                         if (error)
3955                                 goto out;
3956                         error = dir->i_op->unlink(dir, dentry);
3957                         if (!error) {
3958                                 dont_mount(dentry);
3959                                 detach_mounts(dentry);
3960                         }
3961                 }
3962         }
3963 out:
3964         inode_unlock(target);
3965 
3966         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3967         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3968                 fsnotify_link_count(target);
3969                 d_delete(dentry);
3970         }
3971 
3972         return error;
3973 }
3974 EXPORT_SYMBOL(vfs_unlink);
3975 
3976 /*
3977  * Make sure that the actual truncation of the file will occur outside its
3978  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3979  * writeout happening, and we don't want to prevent access to the directory
3980  * while waiting on the I/O.
3981  */
3982 static long do_unlinkat(int dfd, const char __user *pathname)
3983 {
3984         int error;
3985         struct filename *name;
3986         struct dentry *dentry;
3987         struct path path;
3988         struct qstr last;
3989         int type;
3990         struct inode *inode = NULL;
3991         struct inode *delegated_inode = NULL;
3992         unsigned int lookup_flags = 0;
3993 retry:
3994         name = user_path_parent(dfd, pathname,
3995                                 &path, &last, &type, lookup_flags);
3996         if (IS_ERR(name))
3997                 return PTR_ERR(name);
3998 
3999         error = -EISDIR;
4000         if (type != LAST_NORM)
4001                 goto exit1;
4002 
4003         error = mnt_want_write(path.mnt);
4004         if (error)
4005                 goto exit1;
4006 retry_deleg:
4007         inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4008         dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4009         error = PTR_ERR(dentry);
4010         if (!IS_ERR(dentry)) {
4011                 /* Why not before? Because we want correct error value */
4012                 if (last.name[last.len])
4013                         goto slashes;
4014                 inode = dentry->d_inode;
4015                 if (d_is_negative(dentry))
4016                         goto slashes;
4017                 ihold(inode);
4018                 error = security_path_unlink(&path, dentry);
4019                 if (error)
4020                         goto exit2;
4021                 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4022 exit2:
4023                 dput(dentry);
4024         }
4025         inode_unlock(path.dentry->d_inode);
4026         if (inode)
4027                 iput(inode);    /* truncate the inode here */
4028         inode = NULL;
4029         if (delegated_inode) {
4030                 error = break_deleg_wait(&delegated_inode);
4031                 if (!error)
4032                         goto retry_deleg;
4033         }
4034         mnt_drop_write(path.mnt);
4035 exit1:
4036         path_put(&path);
4037         putname(name);
4038         if (retry_estale(error, lookup_flags)) {
4039                 lookup_flags |= LOOKUP_REVAL;
4040                 inode = NULL;
4041                 goto retry;
4042         }
4043         return error;
4044 
4045 slashes:
4046         if (d_is_negative(dentry))
4047                 error = -ENOENT;
4048         else if (d_is_dir(dentry))
4049                 error = -EISDIR;
4050         else
4051                 error = -ENOTDIR;
4052         goto exit2;
4053 }
4054 
4055 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4056 {
4057         if ((flag & ~AT_REMOVEDIR) != 0)
4058                 return -EINVAL;
4059 
4060         if (flag & AT_REMOVEDIR)
4061                 return do_rmdir(dfd, pathname);
4062 
4063         return do_unlinkat(dfd, pathname);
4064 }
4065 
4066 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4067 {
4068         return do_unlinkat(AT_FDCWD, pathname);
4069 }
4070 
4071 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4072 {
4073         int error = may_create(dir, dentry);
4074 
4075         if (error)
4076                 return error;
4077 
4078         if (!dir->i_op->symlink)
4079                 return -EPERM;
4080 
4081         error = security_inode_symlink(dir, dentry, oldname);
4082         if (error)
4083                 return error;
4084 
4085         error = dir->i_op->symlink(dir, dentry, oldname);
4086         if (!error)
4087                 fsnotify_create(dir, dentry);
4088         return error;
4089 }
4090 EXPORT_SYMBOL(vfs_symlink);
4091 
4092 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4093                 int, newdfd, const char __user *, newname)
4094 {
4095         int error;
4096         struct filename *from;
4097         struct dentry *dentry;
4098         struct path path;
4099         unsigned int lookup_flags = 0;
4100 
4101         from = getname(oldname);
4102         if (IS_ERR(from))
4103                 return PTR_ERR(from);
4104 retry:
4105         dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4106         error = PTR_ERR(dentry);
4107         if (IS_ERR(dentry))
4108                 goto out_putname;
4109 
4110         error = security_path_symlink(&path, dentry, from->name);
4111         if (!error)
4112                 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4113         done_path_create(&path, dentry);
4114         if (retry_estale(error, lookup_flags)) {
4115                 lookup_flags |= LOOKUP_REVAL;
4116                 goto retry;
4117         }
4118 out_putname:
4119         putname(from);
4120         return error;
4121 }
4122 
4123 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4124 {
4125         return sys_symlinkat(oldname, AT_FDCWD, newname);
4126 }
4127 
4128 /**
4129  * vfs_link - create a new link
4130  * @old_dentry: object to be linked
4131  * @dir:        new parent
4132  * @new_dentry: where to create the new link
4133  * @delegated_inode: returns inode needing a delegation break
4134  *
4135  * The caller must hold dir->i_mutex
4136  *
4137  * If vfs_link discovers a delegation on the to-be-linked file in need
4138  * of breaking, it will return -EWOULDBLOCK and return a reference to the
4139  * inode in delegated_inode.  The caller should then break the delegation
4140  * and retry.  Because breaking a delegation may take a long time, the
4141  * caller should drop the i_mutex before doing so.
4142  *
4143  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4144  * be appropriate for callers that expect the underlying filesystem not
4145  * to be NFS exported.
4146  */
4147 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4148 {
4149         struct inode *inode = old_dentry->d_inode;
4150         unsigned max_links = dir->i_sb->s_max_links;
4151         int error;
4152 
4153         if (!inode)
4154                 return -ENOENT;
4155 
4156         error = may_create(dir, new_dentry);
4157         if (error)
4158                 return error;
4159 
4160         if (dir->i_sb != inode->i_sb)
4161                 return -EXDEV;
4162 
4163         /*
4164          * A link to an append-only or immutable file cannot be created.
4165          */
4166         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4167                 return -EPERM;
4168         /*
4169          * Updating the link count will likely cause i_uid and i_gid to
4170          * be writen back improperly if their true value is unknown to
4171          * the vfs.
4172          */
4173         if (HAS_UNMAPPED_ID(inode))
4174                 return -EPERM;
4175         if (!dir->i_op->link)
4176                 return -EPERM;
4177         if (S_ISDIR(inode->i_mode))
4178                 return -EPERM;
4179 
4180         error = security_inode_link(old_dentry, dir, new_dentry);
4181         if (error)
4182                 return error;
4183 
4184         inode_lock(inode);
4185         /* Make sure we don't allow creating hardlink to an unlinked file */
4186         if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4187                 error =  -ENOENT;
4188         else if (max_links && inode->i_nlink >= max_links)
4189                 error = -EMLINK;
4190         else {
4191                 error = try_break_deleg(inode, delegated_inode);
4192                 if (!error)
4193                         error = dir->i_op->link(old_dentry, dir, new_dentry);
4194         }
4195 
4196         if (!error && (inode->i_state & I_LINKABLE)) {
4197                 spin_lock(&inode->i_lock);
4198                 inode->i_state &= ~I_LINKABLE;
4199                 spin_unlock(&inode->i_lock);
4200         }
4201         inode_unlock(inode);
4202         if (!error)
4203                 fsnotify_link(dir, inode, new_dentry);
4204         return error;
4205 }
4206 EXPORT_SYMBOL(vfs_link);
4207 
4208 /*
4209  * Hardlinks are often used in delicate situations.  We avoid
4210  * security-related surprises by not following symlinks on the
4211  * newname.  --KAB
4212  *
4213  * We don't follow them on the oldname either to be compatible
4214  * with linux 2.0, and to avoid hard-linking to directories
4215  * and other special files.  --ADM
4216  */
4217 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4218                 int, newdfd, const char __user *, newname, int, flags)
4219 {
4220         struct dentry *new_dentry;
4221         struct path old_path, new_path;
4222         struct inode *delegated_inode = NULL;
4223         int how = 0;
4224         int error;
4225 
4226         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4227                 return -EINVAL;
4228         /*
4229          * To use null names we require CAP_DAC_READ_SEARCH
4230          * This ensures that not everyone will be able to create
4231          * handlink using the passed filedescriptor.
4232          */
4233         if (flags & AT_EMPTY_PATH) {
4234                 if (!capable(CAP_DAC_READ_SEARCH))
4235                         return -ENOENT;
4236                 how = LOOKUP_EMPTY;
4237         }
4238 
4239         if (flags & AT_SYMLINK_FOLLOW)
4240                 how |= LOOKUP_FOLLOW;
4241 retry:
4242         error = user_path_at(olddfd, oldname, how, &old_path);
4243         if (error)
4244                 return error;
4245 
4246         new_dentry = user_path_create(newdfd, newname, &new_path,
4247                                         (how & LOOKUP_REVAL));
4248         error = PTR_ERR(new_dentry);
4249         if (IS_ERR(new_dentry))
4250                 goto out;
4251 
4252         error = -EXDEV;
4253         if (old_path.mnt != new_path.mnt)
4254                 goto out_dput;
4255         error = may_linkat(&old_path);
4256         if (unlikely(error))
4257                 goto out_dput;
4258         error = security_path_link(old_path.dentry, &new_path, new_dentry);
4259         if (error)
4260                 goto out_dput;
4261         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4262 out_dput:
4263         done_path_create(&new_path, new_dentry);
4264         if (delegated_inode) {
4265                 error = break_deleg_wait(&delegated_inode);
4266                 if (!error) {
4267                         path_put(&old_path);
4268                         goto retry;
4269                 }
4270         }
4271         if (retry_estale(error, how)) {
4272                 path_put(&old_path);
4273                 how |= LOOKUP_REVAL;
4274                 goto retry;
4275         }
4276 out:
4277         path_put(&old_path);
4278 
4279         return error;
4280 }
4281 
4282 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4283 {
4284         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4285 }
4286 
4287 /**
4288  * vfs_rename - rename a filesystem object
4289  * @old_dir:    parent of source
4290  * @old_dentry: source
4291  * @new_dir:    parent of destination
4292  * @new_dentry: destination
4293  * @delegated_inode: returns an inode needing a delegation break
4294  * @flags:      rename flags
4295  *
4296  * The caller must hold multiple mutexes--see lock_rename()).
4297  *
4298  * If vfs_rename discovers a delegation in need of breaking at either
4299  * the source or destination, it will return -EWOULDBLOCK and return a
4300  * reference to the inode in delegated_inode.  The caller should then
4301  * break the delegation and retry.  Because breaking a delegation may
4302  * take a long time, the caller should drop all locks before doing
4303  * so.
4304  *
4305  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4306  * be appropriate for callers that expect the underlying filesystem not
4307  * to be NFS exported.
4308  *
4309  * The worst of all namespace operations - renaming directory. "Perverted"
4310  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4311  * Problems:
4312  *      a) we can get into loop creation.
4313  *      b) race potential - two innocent renames can create a loop together.
4314  *         That's where 4.4 screws up. Current fix: serialization on
4315  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4316  *         story.
4317  *      c) we have to lock _four_ objects - parents and victim (if it exists),
4318  *         and source (if it is not a directory).
4319  *         And that - after we got ->i_mutex on parents (until then we don't know
4320  *         whether the target exists).  Solution: try to be smart with locking
4321  *         order for inodes.  We rely on the fact that tree topology may change
4322  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4323  *         move will be locked.  Thus we can rank directories by the tree
4324  *         (ancestors first) and rank all non-directories after them.
4325  *         That works since everybody except rename does "lock parent, lookup,
4326  *         lock child" and rename is under ->s_vfs_rename_mutex.
4327  *         HOWEVER, it relies on the assumption that any object with ->lookup()
4328  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4329  *         we'd better make sure that there's no link(2) for them.
4330  *      d) conversion from fhandle to dentry may come in the wrong moment - when
4331  *         we are removing the target. Solution: we will have to grab ->i_mutex
4332  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4333  *         ->i_mutex on parents, which works but leads to some truly excessive
4334  *         locking].
4335  */
4336 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4337                struct inode *new_dir, struct dentry *new_dentry,
4338                struct inode **delegated_inode, unsigned int flags)
4339 {
4340         int error;
4341         bool is_dir = d_is_dir(old_dentry);
4342         const unsigned char *old_name;
4343         struct inode *source = old_dentry->d_inode;
4344         struct inode *target = new_dentry->d_inode;
4345         bool new_is_dir = false;
4346         unsigned max_links = new_dir->i_sb->s_max_links;
4347 
4348         /*
4349          * Check source == target.
4350          * On overlayfs need to look at underlying inodes.
4351          */
4352         if (d_real_inode(old_dentry) == d_real_inode(new_dentry))
4353                 return 0;
4354 
4355         error = may_delete(old_dir, old_dentry, is_dir);
4356         if (error)
4357                 return error;
4358 
4359         if (!target) {
4360                 error = may_create(new_dir, new_dentry);
4361         } else {
4362                 new_is_dir = d_is_dir(new_dentry);
4363 
4364                 if (!(flags & RENAME_EXCHANGE))
4365                         error = may_delete(new_dir, new_dentry, is_dir);
4366                 else
4367                         error = may_delete(new_dir, new_dentry, new_is_dir);
4368         }
4369         if (error)
4370                 return error;
4371 
4372         if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4373                 return -EPERM;
4374 
4375         if (flags && !old_dir->i_op->rename2)
4376                 return -EINVAL;
4377 
4378         /*
4379          * If we are going to change the parent - check write permissions,
4380          * we'll need to flip '..'.
4381          */
4382         if (new_dir != old_dir) {
4383                 if (is_dir) {
4384                         error = inode_permission(source, MAY_WRITE);
4385                         if (error)
4386                                 return error;
4387                 }
4388                 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4389                         error = inode_permission(target, MAY_WRITE);
4390                         if (error)
4391                                 return error;
4392                 }
4393         }
4394 
4395         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4396                                       flags);
4397         if (error)
4398                 return error;
4399 
4400         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4401         dget(new_dentry);
4402         if (!is_dir || (flags & RENAME_EXCHANGE))
4403                 lock_two_nondirectories(source, target);
4404         else if (target)
4405                 inode_lock(target);
4406 
4407         error = -EBUSY;
4408         if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4409                 goto out;
4410 
4411         if (max_links && new_dir != old_dir) {
4412                 error = -EMLINK;
4413                 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4414                         goto out;
4415                 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4416                     old_dir->i_nlink >= max_links)
4417                         goto out;
4418         }
4419         if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4420                 shrink_dcache_parent(new_dentry);
4421         if (!is_dir) {
4422                 error = try_break_deleg(source, delegated_inode);
4423                 if (error)
4424                         goto out;
4425         }
4426         if (target && !new_is_dir) {
4427                 error = try_break_deleg(target, delegated_inode);
4428                 if (error)
4429                         goto out;
4430         }
4431         if (!old_dir->i_op->rename2) {
4432                 error = old_dir->i_op->rename(old_dir, old_dentry,
4433                                               new_dir, new_dentry);
4434         } else {
4435                 WARN_ON(old_dir->i_op->rename != NULL);
4436                 error = old_dir->i_op->rename2(old_dir, old_dentry,
4437                                                new_dir, new_dentry, flags);
4438         }
4439         if (error)
4440                 goto out;
4441 
4442         if (!(flags & RENAME_EXCHANGE) && target) {
4443                 if (is_dir)
4444                         target->i_flags |= S_DEAD;
4445                 dont_mount(new_dentry);
4446                 detach_mounts(new_dentry);
4447         }
4448         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4449                 if (!(flags & RENAME_EXCHANGE))
4450                         d_move(old_dentry, new_dentry);
4451                 else
4452                         d_exchange(old_dentry, new_dentry);
4453         }
4454 out:
4455         if (!is_dir || (flags & RENAME_EXCHANGE))
4456                 unlock_two_nondirectories(source, target);
4457         else if (target)
4458                 inode_unlock(target);
4459         dput(new_dentry);
4460         if (!error) {
4461                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4462                               !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4463                 if (flags & RENAME_EXCHANGE) {
4464                         fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4465                                       new_is_dir, NULL, new_dentry);
4466                 }
4467         }
4468         fsnotify_oldname_free(old_name);
4469 
4470         return error;
4471 }
4472 EXPORT_SYMBOL(vfs_rename);
4473 
4474 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4475                 int, newdfd, const char __user *, newname, unsigned int, flags)
4476 {
4477         struct dentry *old_dentry, *new_dentry;
4478         struct dentry *trap;
4479         struct path old_path, new_path;
4480         struct qstr old_last, new_last;
4481         int old_type, new_type;
4482         struct inode *delegated_inode = NULL;
4483         struct filename *from;
4484         struct filename *to;
4485         unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4486         bool should_retry = false;
4487         int error;
4488 
4489         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4490                 return -EINVAL;
4491 
4492         if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4493             (flags & RENAME_EXCHANGE))
4494                 return -EINVAL;
4495 
4496         if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4497                 return -EPERM;
4498 
4499         if (flags & RENAME_EXCHANGE)
4500                 target_flags = 0;
4501 
4502 retry:
4503         from = user_path_parent(olddfd, oldname,
4504                                 &old_path, &old_last, &old_type, lookup_flags);
4505         if (IS_ERR(from)) {
4506                 error = PTR_ERR(from);
4507                 goto exit;
4508         }
4509 
4510         to = user_path_parent(newdfd, newname,
4511                                 &new_path, &new_last, &new_type, lookup_flags);
4512         if (IS_ERR(to)) {
4513                 error = PTR_ERR(to);
4514                 goto exit1;
4515         }
4516 
4517         error = -EXDEV;
4518         if (old_path.mnt != new_path.mnt)
4519                 goto exit2;
4520 
4521         error = -EBUSY;
4522         if (old_type != LAST_NORM)
4523                 goto exit2;
4524 
4525         if (flags & RENAME_NOREPLACE)
4526                 error = -EEXIST;
4527         if (new_type != LAST_NORM)
4528                 goto exit2;
4529 
4530         error = mnt_want_write(old_path.mnt);
4531         if (error)
4532                 goto exit2;
4533 
4534 retry_deleg:
4535         trap = lock_rename(new_path.dentry, old_path.dentry);
4536 
4537         old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4538         error = PTR_ERR(old_dentry);
4539         if (IS_ERR(old_dentry))
4540                 goto exit3;
4541         /* source must exist */
4542         error = -ENOENT;
4543         if (d_is_negative(old_dentry))
4544                 goto exit4;
4545         new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4546         error = PTR_ERR(new_dentry);
4547         if (IS_ERR(new_dentry))
4548                 goto exit4;
4549         error = -EEXIST;
4550         if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4551                 goto exit5;
4552         if (flags & RENAME_EXCHANGE) {
4553                 error = -ENOENT;
4554                 if (d_is_negative(new_dentry))
4555                         goto exit5;
4556 
4557                 if (!d_is_dir(new_dentry)) {
4558                         error = -ENOTDIR;
4559                         if (new_last.name[new_last.len])
4560                                 goto exit5;
4561                 }
4562         }
4563         /* unless the source is a directory trailing slashes give -ENOTDIR */
4564         if (!d_is_dir(old_dentry)) {
4565                 error = -ENOTDIR;
4566                 if (old_last.name[old_last.len])
4567                         goto exit5;
4568                 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4569                         goto exit5;
4570         }
4571         /* source should not be ancestor of target */
4572         error = -EINVAL;
4573         if (old_dentry == trap)
4574                 goto exit5;
4575         /* target should not be an ancestor of source */
4576         if (!(flags & RENAME_EXCHANGE))
4577                 error = -ENOTEMPTY;
4578         if (new_dentry == trap)
4579                 goto exit5;
4580 
4581         error = security_path_rename(&old_path, old_dentry,
4582                                      &new_path, new_dentry, flags);
4583         if (error)
4584                 goto exit5;
4585         error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4586                            new_path.dentry->d_inode, new_dentry,
4587                            &delegated_inode, flags);
4588 exit5:
4589         dput(new_dentry);
4590 exit4:
4591         dput(old_dentry);
4592 exit3:
4593         unlock_rename(new_path.dentry, old_path.dentry);
4594         if (delegated_inode) {
4595                 error = break_deleg_wait(&delegated_inode);
4596                 if (!error)
4597                         goto retry_deleg;
4598         }
4599         mnt_drop_write(old_path.mnt);
4600 exit2:
4601         if (retry_estale(error, lookup_flags))
4602                 should_retry = true;
4603         path_put(&new_path);
4604         putname(to);
4605 exit1:
4606         path_put(&old_path);
4607         putname(from);
4608         if (should_retry) {
4609                 should_retry = false;
4610                 lookup_flags |= LOOKUP_REVAL;
4611                 goto retry;
4612         }
4613 exit:
4614         return error;
4615 }
4616 
4617 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4618                 int, newdfd, const char __user *, newname)
4619 {
4620         return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4621 }
4622 
4623 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4624 {
4625         return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4626 }
4627 
4628 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4629 {
4630         int error = may_create(dir, dentry);
4631         if (error)
4632                 return error;
4633 
4634         if (!dir->i_op->mknod)
4635                 return -EPERM;
4636 
4637         return dir->i_op->mknod(dir, dentry,
4638                                 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4639 }
4640 EXPORT_SYMBOL(vfs_whiteout);
4641 
4642 int readlink_copy(char __user *buffer, int buflen, const char *link)
4643 {
4644         int len = PTR_ERR(link);
4645         if (IS_ERR(link))
4646                 goto out;
4647 
4648         len = strlen(link);
4649         if (len > (unsigned) buflen)
4650                 len = buflen;
4651         if (copy_to_user(buffer, link, len))
4652                 len = -EFAULT;
4653 out:
4654         return len;
4655 }
4656 
4657 /*
4658  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4659  * have ->get_link() not calling nd_jump_link().  Using (or not using) it
4660  * for any given inode is up to filesystem.
4661  */
4662 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4663 {
4664         DEFINE_DELAYED_CALL(done);
4665         struct inode *inode = d_inode(dentry);
4666         const char *link = inode->i_link;
4667         int res;
4668 
4669         if (!link) {
4670                 link = inode->i_op->get_link(dentry, inode, &done);
4671                 if (IS_ERR(link))
4672                         return PTR_ERR(link);
4673         }
4674         res = readlink_copy(buffer, buflen, link);
4675         do_delayed_call(&done);
4676         return res;
4677 }
4678 EXPORT_SYMBOL(generic_readlink);
4679 
4680 /* get the link contents into pagecache */
4681 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4682                           struct delayed_call *callback)
4683 {
4684         char *kaddr;
4685         struct page *page;
4686         struct address_space *mapping = inode->i_mapping;
4687 
4688         if (!dentry) {
4689                 page = find_get_page(mapping, 0);
4690                 if (!page)
4691                         return ERR_PTR(-ECHILD);
4692                 if (!PageUptodate(page)) {
4693                         put_page(page);
4694                         return ERR_PTR(-ECHILD);
4695                 }
4696         } else {
4697                 page = read_mapping_page(mapping, 0, NULL);
4698                 if (IS_ERR(page))
4699                         return (char*)page;
4700         }
4701         set_delayed_call(callback, page_put_link, page);
4702         BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4703         kaddr = page_address(page);
4704         nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4705         return kaddr;
4706 }
4707 
4708 EXPORT_SYMBOL(page_get_link);
4709 
4710 void page_put_link(void *arg)
4711 {
4712         put_page(arg);
4713 }
4714 EXPORT_SYMBOL(page_put_link);
4715 
4716 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4717 {
4718         DEFINE_DELAYED_CALL(done);
4719         int res = readlink_copy(buffer, buflen,
4720                                 page_get_link(dentry, d_inode(dentry),
4721                                               &done));
4722         do_delayed_call(&done);
4723         return res;
4724 }
4725 EXPORT_SYMBOL(page_readlink);
4726 
4727 /*
4728  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4729  */
4730 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4731 {
4732         struct address_space *mapping = inode->i_mapping;
4733         struct page *page;
4734         void *fsdata;
4735         int err;
4736         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4737         if (nofs)
4738                 flags |= AOP_FLAG_NOFS;
4739 
4740 retry:
4741         err = pagecache_write_begin(NULL, mapping, 0, len-1,
4742                                 flags, &page, &fsdata);
4743         if (err)
4744                 goto fail;
4745 
4746         memcpy(page_address(page), symname, len-1);
4747 
4748         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4749                                                         page, fsdata);
4750         if (err < 0)
4751                 goto fail;
4752         if (err < len-1)
4753                 goto retry;
4754 
4755         mark_inode_dirty(inode);
4756         return 0;
4757 fail:
4758         return err;
4759 }
4760 EXPORT_SYMBOL(__page_symlink);
4761 
4762 int page_symlink(struct inode *inode, const char *symname, int len)
4763 {
4764         return __page_symlink(inode, symname, len,
4765                         !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4766 }
4767 EXPORT_SYMBOL(page_symlink);
4768 
4769 const struct inode_operations page_symlink_inode_operations = {
4770         .readlink       = generic_readlink,
4771         .get_link       = page_get_link,
4772 };
4773 EXPORT_SYMBOL(page_symlink_inode_operations);
4774 

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