Version:  2.0.40 2.2.26 2.4.37 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2

Linux/fs/namei.c

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

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