~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/namei.c

Version: ~ [ linux-5.16-rc3 ] ~ [ linux-5.15.5 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.82 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.162 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.218 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.256 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.291 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.293 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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                 struct delayed_call done;
509                 const char *name;
510                 unsigned seq;
511         } *stack, internal[EMBEDDED_LEVELS];
512         struct filename *name;
513         struct nameidata *saved;
514         struct inode    *link_inode;
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 }
540 
541 static int __nd_alloc_stack(struct nameidata *nd)
542 {
543         struct saved *p;
544 
545         if (nd->flags & LOOKUP_RCU) {
546                 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
547                                   GFP_ATOMIC);
548                 if (unlikely(!p))
549                         return -ECHILD;
550         } else {
551                 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
552                                   GFP_KERNEL);
553                 if (unlikely(!p))
554                         return -ENOMEM;
555         }
556         memcpy(p, nd->internal, sizeof(nd->internal));
557         nd->stack = p;
558         return 0;
559 }
560 
561 /**
562  * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
563  * @path: nameidate to verify
564  *
565  * Rename can sometimes move a file or directory outside of a bind
566  * mount, path_connected allows those cases to be detected.
567  */
568 static bool path_connected(const struct path *path)
569 {
570         struct vfsmount *mnt = path->mnt;
571 
572         /* Only bind mounts can have disconnected paths */
573         if (mnt->mnt_root == mnt->mnt_sb->s_root)
574                 return true;
575 
576         return is_subdir(path->dentry, mnt->mnt_root);
577 }
578 
579 static inline int nd_alloc_stack(struct nameidata *nd)
580 {
581         if (likely(nd->depth != EMBEDDED_LEVELS))
582                 return 0;
583         if (likely(nd->stack != nd->internal))
584                 return 0;
585         return __nd_alloc_stack(nd);
586 }
587 
588 static void drop_links(struct nameidata *nd)
589 {
590         int i = nd->depth;
591         while (i--) {
592                 struct saved *last = nd->stack + i;
593                 do_delayed_call(&last->done);
594                 clear_delayed_call(&last->done);
595         }
596 }
597 
598 static void terminate_walk(struct nameidata *nd)
599 {
600         drop_links(nd);
601         if (!(nd->flags & LOOKUP_RCU)) {
602                 int i;
603                 path_put(&nd->path);
604                 for (i = 0; i < nd->depth; i++)
605                         path_put(&nd->stack[i].link);
606                 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
607                         path_put(&nd->root);
608                         nd->root.mnt = NULL;
609                 }
610         } else {
611                 nd->flags &= ~LOOKUP_RCU;
612                 if (!(nd->flags & LOOKUP_ROOT))
613                         nd->root.mnt = NULL;
614                 rcu_read_unlock();
615         }
616         nd->depth = 0;
617 }
618 
619 /* path_put is needed afterwards regardless of success or failure */
620 static bool legitimize_path(struct nameidata *nd,
621                             struct path *path, unsigned seq)
622 {
623         int res = __legitimize_mnt(path->mnt, nd->m_seq);
624         if (unlikely(res)) {
625                 if (res > 0)
626                         path->mnt = NULL;
627                 path->dentry = NULL;
628                 return false;
629         }
630         if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
631                 path->dentry = NULL;
632                 return false;
633         }
634         return !read_seqcount_retry(&path->dentry->d_seq, seq);
635 }
636 
637 static bool legitimize_links(struct nameidata *nd)
638 {
639         int i;
640         for (i = 0; i < nd->depth; i++) {
641                 struct saved *last = nd->stack + i;
642                 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
643                         drop_links(nd);
644                         nd->depth = i + 1;
645                         return false;
646                 }
647         }
648         return true;
649 }
650 
651 /*
652  * Path walking has 2 modes, rcu-walk and ref-walk (see
653  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
654  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
655  * normal reference counts on dentries and vfsmounts to transition to ref-walk
656  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
657  * got stuck, so ref-walk may continue from there. If this is not successful
658  * (eg. a seqcount has changed), then failure is returned and it's up to caller
659  * to restart the path walk from the beginning in ref-walk mode.
660  */
661 
662 /**
663  * unlazy_walk - try to switch to ref-walk mode.
664  * @nd: nameidata pathwalk data
665  * @dentry: child of nd->path.dentry or NULL
666  * @seq: seq number to check dentry against
667  * Returns: 0 on success, -ECHILD on failure
668  *
669  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
670  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
671  * @nd or NULL.  Must be called from rcu-walk context.
672  * Nothing should touch nameidata between unlazy_walk() failure and
673  * terminate_walk().
674  */
675 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
676 {
677         struct dentry *parent = nd->path.dentry;
678 
679         BUG_ON(!(nd->flags & LOOKUP_RCU));
680 
681         nd->flags &= ~LOOKUP_RCU;
682         if (unlikely(!legitimize_links(nd)))
683                 goto out2;
684         if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
685                 goto out2;
686         if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
687                 goto out1;
688 
689         /*
690          * For a negative lookup, the lookup sequence point is the parents
691          * sequence point, and it only needs to revalidate the parent dentry.
692          *
693          * For a positive lookup, we need to move both the parent and the
694          * dentry from the RCU domain to be properly refcounted. And the
695          * sequence number in the dentry validates *both* dentry counters,
696          * since we checked the sequence number of the parent after we got
697          * the child sequence number. So we know the parent must still
698          * be valid if the child sequence number is still valid.
699          */
700         if (!dentry) {
701                 if (read_seqcount_retry(&parent->d_seq, nd->seq))
702                         goto out;
703                 BUG_ON(nd->inode != parent->d_inode);
704         } else {
705                 if (!lockref_get_not_dead(&dentry->d_lockref))
706                         goto out;
707                 if (read_seqcount_retry(&dentry->d_seq, seq))
708                         goto drop_dentry;
709         }
710 
711         /*
712          * Sequence counts matched. Now make sure that the root is
713          * still valid and get it if required.
714          */
715         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
716                 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
717                         rcu_read_unlock();
718                         dput(dentry);
719                         return -ECHILD;
720                 }
721         }
722 
723         rcu_read_unlock();
724         return 0;
725 
726 drop_dentry:
727         rcu_read_unlock();
728         dput(dentry);
729         goto drop_root_mnt;
730 out2:
731         nd->path.mnt = NULL;
732 out1:
733         nd->path.dentry = NULL;
734 out:
735         rcu_read_unlock();
736 drop_root_mnt:
737         if (!(nd->flags & LOOKUP_ROOT))
738                 nd->root.mnt = NULL;
739         return -ECHILD;
740 }
741 
742 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
743 {
744         if (unlikely(!legitimize_path(nd, link, seq))) {
745                 drop_links(nd);
746                 nd->depth = 0;
747                 nd->flags &= ~LOOKUP_RCU;
748                 nd->path.mnt = NULL;
749                 nd->path.dentry = NULL;
750                 if (!(nd->flags & LOOKUP_ROOT))
751                         nd->root.mnt = NULL;
752                 rcu_read_unlock();
753         } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
754                 return 0;
755         }
756         path_put(link);
757         return -ECHILD;
758 }
759 
760 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
761 {
762         return dentry->d_op->d_revalidate(dentry, flags);
763 }
764 
765 /**
766  * complete_walk - successful completion of path walk
767  * @nd:  pointer nameidata
768  *
769  * If we had been in RCU mode, drop out of it and legitimize nd->path.
770  * Revalidate the final result, unless we'd already done that during
771  * the path walk or the filesystem doesn't ask for it.  Return 0 on
772  * success, -error on failure.  In case of failure caller does not
773  * need to drop nd->path.
774  */
775 static int complete_walk(struct nameidata *nd)
776 {
777         struct dentry *dentry = nd->path.dentry;
778         int status;
779 
780         if (nd->flags & LOOKUP_RCU) {
781                 if (!(nd->flags & LOOKUP_ROOT))
782                         nd->root.mnt = NULL;
783                 if (unlikely(unlazy_walk(nd, NULL, 0)))
784                         return -ECHILD;
785         }
786 
787         if (likely(!(nd->flags & LOOKUP_JUMPED)))
788                 return 0;
789 
790         if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
791                 return 0;
792 
793         status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
794         if (status > 0)
795                 return 0;
796 
797         if (!status)
798                 status = -ESTALE;
799 
800         return status;
801 }
802 
803 static void set_root(struct nameidata *nd)
804 {
805         struct fs_struct *fs = current->fs;
806 
807         if (nd->flags & LOOKUP_RCU) {
808                 unsigned seq;
809 
810                 do {
811                         seq = read_seqcount_begin(&fs->seq);
812                         nd->root = fs->root;
813                         nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
814                 } while (read_seqcount_retry(&fs->seq, seq));
815         } else {
816                 get_fs_root(fs, &nd->root);
817         }
818 }
819 
820 static void path_put_conditional(struct path *path, struct nameidata *nd)
821 {
822         dput(path->dentry);
823         if (path->mnt != nd->path.mnt)
824                 mntput(path->mnt);
825 }
826 
827 static inline void path_to_nameidata(const struct path *path,
828                                         struct nameidata *nd)
829 {
830         if (!(nd->flags & LOOKUP_RCU)) {
831                 dput(nd->path.dentry);
832                 if (nd->path.mnt != path->mnt)
833                         mntput(nd->path.mnt);
834         }
835         nd->path.mnt = path->mnt;
836         nd->path.dentry = path->dentry;
837 }
838 
839 static int nd_jump_root(struct nameidata *nd)
840 {
841         if (nd->flags & LOOKUP_RCU) {
842                 struct dentry *d;
843                 nd->path = nd->root;
844                 d = nd->path.dentry;
845                 nd->inode = d->d_inode;
846                 nd->seq = nd->root_seq;
847                 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
848                         return -ECHILD;
849         } else {
850                 path_put(&nd->path);
851                 nd->path = nd->root;
852                 path_get(&nd->path);
853                 nd->inode = nd->path.dentry->d_inode;
854         }
855         nd->flags |= LOOKUP_JUMPED;
856         return 0;
857 }
858 
859 /*
860  * Helper to directly jump to a known parsed path from ->get_link,
861  * caller must have taken a reference to path beforehand.
862  */
863 void nd_jump_link(struct path *path)
864 {
865         struct nameidata *nd = current->nameidata;
866         path_put(&nd->path);
867 
868         nd->path = *path;
869         nd->inode = nd->path.dentry->d_inode;
870         nd->flags |= LOOKUP_JUMPED;
871 }
872 
873 static inline void put_link(struct nameidata *nd)
874 {
875         struct saved *last = nd->stack + --nd->depth;
876         do_delayed_call(&last->done);
877         if (!(nd->flags & LOOKUP_RCU))
878                 path_put(&last->link);
879 }
880 
881 int sysctl_protected_symlinks __read_mostly = 0;
882 int sysctl_protected_hardlinks __read_mostly = 0;
883 
884 /**
885  * may_follow_link - Check symlink following for unsafe situations
886  * @nd: nameidata pathwalk data
887  *
888  * In the case of the sysctl_protected_symlinks sysctl being enabled,
889  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
890  * in a sticky world-writable directory. This is to protect privileged
891  * processes from failing races against path names that may change out
892  * from under them by way of other users creating malicious symlinks.
893  * It will permit symlinks to be followed only when outside a sticky
894  * world-writable directory, or when the uid of the symlink and follower
895  * match, or when the directory owner matches the symlink's owner.
896  *
897  * Returns 0 if following the symlink is allowed, -ve on error.
898  */
899 static inline int may_follow_link(struct nameidata *nd)
900 {
901         const struct inode *inode;
902         const struct inode *parent;
903 
904         if (!sysctl_protected_symlinks)
905                 return 0;
906 
907         /* Allowed if owner and follower match. */
908         inode = nd->link_inode;
909         if (uid_eq(current_cred()->fsuid, inode->i_uid))
910                 return 0;
911 
912         /* Allowed if parent directory not sticky and world-writable. */
913         parent = nd->inode;
914         if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
915                 return 0;
916 
917         /* Allowed if parent directory and link owner match. */
918         if (uid_eq(parent->i_uid, inode->i_uid))
919                 return 0;
920 
921         if (nd->flags & LOOKUP_RCU)
922                 return -ECHILD;
923 
924         audit_log_link_denied("follow_link", &nd->stack[0].link);
925         return -EACCES;
926 }
927 
928 /**
929  * safe_hardlink_source - Check for safe hardlink conditions
930  * @inode: the source inode to hardlink from
931  *
932  * Return false if at least one of the following conditions:
933  *    - inode is not a regular file
934  *    - inode is setuid
935  *    - inode is setgid and group-exec
936  *    - access failure for read and write
937  *
938  * Otherwise returns true.
939  */
940 static bool safe_hardlink_source(struct inode *inode)
941 {
942         umode_t mode = inode->i_mode;
943 
944         /* Special files should not get pinned to the filesystem. */
945         if (!S_ISREG(mode))
946                 return false;
947 
948         /* Setuid files should not get pinned to the filesystem. */
949         if (mode & S_ISUID)
950                 return false;
951 
952         /* Executable setgid files should not get pinned to the filesystem. */
953         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
954                 return false;
955 
956         /* Hardlinking to unreadable or unwritable sources is dangerous. */
957         if (inode_permission(inode, MAY_READ | MAY_WRITE))
958                 return false;
959 
960         return true;
961 }
962 
963 /**
964  * may_linkat - Check permissions for creating a hardlink
965  * @link: the source to hardlink from
966  *
967  * Block hardlink when all of:
968  *  - sysctl_protected_hardlinks enabled
969  *  - fsuid does not match inode
970  *  - hardlink source is unsafe (see safe_hardlink_source() above)
971  *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
972  *
973  * Returns 0 if successful, -ve on error.
974  */
975 static int may_linkat(struct path *link)
976 {
977         struct inode *inode;
978 
979         if (!sysctl_protected_hardlinks)
980                 return 0;
981 
982         inode = link->dentry->d_inode;
983 
984         /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
985          * otherwise, it must be a safe source.
986          */
987         if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
988                 return 0;
989 
990         audit_log_link_denied("linkat", link);
991         return -EPERM;
992 }
993 
994 static __always_inline
995 const char *get_link(struct nameidata *nd)
996 {
997         struct saved *last = nd->stack + nd->depth - 1;
998         struct dentry *dentry = last->link.dentry;
999         struct inode *inode = nd->link_inode;
1000         int error;
1001         const char *res;
1002 
1003         if (!(nd->flags & LOOKUP_RCU)) {
1004                 touch_atime(&last->link);
1005                 cond_resched();
1006         } else if (atime_needs_update(&last->link, inode)) {
1007                 if (unlikely(unlazy_walk(nd, NULL, 0)))
1008                         return ERR_PTR(-ECHILD);
1009                 touch_atime(&last->link);
1010         }
1011 
1012         error = security_inode_follow_link(dentry, inode,
1013                                            nd->flags & LOOKUP_RCU);
1014         if (unlikely(error))
1015                 return ERR_PTR(error);
1016 
1017         nd->last_type = LAST_BIND;
1018         res = inode->i_link;
1019         if (!res) {
1020                 const char * (*get)(struct dentry *, struct inode *,
1021                                 struct delayed_call *);
1022                 get = inode->i_op->get_link;
1023                 if (nd->flags & LOOKUP_RCU) {
1024                         res = get(NULL, inode, &last->done);
1025                         if (res == ERR_PTR(-ECHILD)) {
1026                                 if (unlikely(unlazy_walk(nd, NULL, 0)))
1027                                         return ERR_PTR(-ECHILD);
1028                                 res = get(dentry, inode, &last->done);
1029                         }
1030                 } else {
1031                         res = get(dentry, inode, &last->done);
1032                 }
1033                 if (IS_ERR_OR_NULL(res))
1034                         return res;
1035         }
1036         if (*res == '/') {
1037                 if (!nd->root.mnt)
1038                         set_root(nd);
1039                 if (unlikely(nd_jump_root(nd)))
1040                         return ERR_PTR(-ECHILD);
1041                 while (unlikely(*++res == '/'))
1042                         ;
1043         }
1044         if (!*res)
1045                 res = NULL;
1046         return res;
1047 }
1048 
1049 /*
1050  * follow_up - Find the mountpoint of path's vfsmount
1051  *
1052  * Given a path, find the mountpoint of its source file system.
1053  * Replace @path with the path of the mountpoint in the parent mount.
1054  * Up is towards /.
1055  *
1056  * Return 1 if we went up a level and 0 if we were already at the
1057  * root.
1058  */
1059 int follow_up(struct path *path)
1060 {
1061         struct mount *mnt = real_mount(path->mnt);
1062         struct mount *parent;
1063         struct dentry *mountpoint;
1064 
1065         read_seqlock_excl(&mount_lock);
1066         parent = mnt->mnt_parent;
1067         if (parent == mnt) {
1068                 read_sequnlock_excl(&mount_lock);
1069                 return 0;
1070         }
1071         mntget(&parent->mnt);
1072         mountpoint = dget(mnt->mnt_mountpoint);
1073         read_sequnlock_excl(&mount_lock);
1074         dput(path->dentry);
1075         path->dentry = mountpoint;
1076         mntput(path->mnt);
1077         path->mnt = &parent->mnt;
1078         return 1;
1079 }
1080 EXPORT_SYMBOL(follow_up);
1081 
1082 /*
1083  * Perform an automount
1084  * - return -EISDIR to tell follow_managed() to stop and return the path we
1085  *   were called with.
1086  */
1087 static int follow_automount(struct path *path, struct nameidata *nd,
1088                             bool *need_mntput)
1089 {
1090         struct vfsmount *mnt;
1091         int err;
1092 
1093         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1094                 return -EREMOTE;
1095 
1096         /* We don't want to mount if someone's just doing a stat -
1097          * unless they're stat'ing a directory and appended a '/' to
1098          * the name.
1099          *
1100          * We do, however, want to mount if someone wants to open or
1101          * create a file of any type under the mountpoint, wants to
1102          * traverse through the mountpoint or wants to open the
1103          * mounted directory.  Also, autofs may mark negative dentries
1104          * as being automount points.  These will need the attentions
1105          * of the daemon to instantiate them before they can be used.
1106          */
1107         if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1108                            LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1109             path->dentry->d_inode)
1110                 return -EISDIR;
1111 
1112         nd->total_link_count++;
1113         if (nd->total_link_count >= 40)
1114                 return -ELOOP;
1115 
1116         mnt = path->dentry->d_op->d_automount(path);
1117         if (IS_ERR(mnt)) {
1118                 /*
1119                  * The filesystem is allowed to return -EISDIR here to indicate
1120                  * it doesn't want to automount.  For instance, autofs would do
1121                  * this so that its userspace daemon can mount on this dentry.
1122                  *
1123                  * However, we can only permit this if it's a terminal point in
1124                  * the path being looked up; if it wasn't then the remainder of
1125                  * the path is inaccessible and we should say so.
1126                  */
1127                 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1128                         return -EREMOTE;
1129                 return PTR_ERR(mnt);
1130         }
1131 
1132         if (!mnt) /* mount collision */
1133                 return 0;
1134 
1135         if (!*need_mntput) {
1136                 /* lock_mount() may release path->mnt on error */
1137                 mntget(path->mnt);
1138                 *need_mntput = true;
1139         }
1140         err = finish_automount(mnt, path);
1141 
1142         switch (err) {
1143         case -EBUSY:
1144                 /* Someone else made a mount here whilst we were busy */
1145                 return 0;
1146         case 0:
1147                 path_put(path);
1148                 path->mnt = mnt;
1149                 path->dentry = dget(mnt->mnt_root);
1150                 return 0;
1151         default:
1152                 return err;
1153         }
1154 
1155 }
1156 
1157 /*
1158  * Handle a dentry that is managed in some way.
1159  * - Flagged for transit management (autofs)
1160  * - Flagged as mountpoint
1161  * - Flagged as automount point
1162  *
1163  * This may only be called in refwalk mode.
1164  *
1165  * Serialization is taken care of in namespace.c
1166  */
1167 static int follow_managed(struct path *path, struct nameidata *nd)
1168 {
1169         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1170         unsigned managed;
1171         bool need_mntput = false;
1172         int ret = 0;
1173 
1174         /* Given that we're not holding a lock here, we retain the value in a
1175          * local variable for each dentry as we look at it so that we don't see
1176          * the components of that value change under us */
1177         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1178                managed &= DCACHE_MANAGED_DENTRY,
1179                unlikely(managed != 0)) {
1180                 /* Allow the filesystem to manage the transit without i_mutex
1181                  * being held. */
1182                 if (managed & DCACHE_MANAGE_TRANSIT) {
1183                         BUG_ON(!path->dentry->d_op);
1184                         BUG_ON(!path->dentry->d_op->d_manage);
1185                         ret = path->dentry->d_op->d_manage(path->dentry, false);
1186                         if (ret < 0)
1187                                 break;
1188                 }
1189 
1190                 /* Transit to a mounted filesystem. */
1191                 if (managed & DCACHE_MOUNTED) {
1192                         struct vfsmount *mounted = lookup_mnt(path);
1193                         if (mounted) {
1194                                 dput(path->dentry);
1195                                 if (need_mntput)
1196                                         mntput(path->mnt);
1197                                 path->mnt = mounted;
1198                                 path->dentry = dget(mounted->mnt_root);
1199                                 need_mntput = true;
1200                                 continue;
1201                         }
1202 
1203                         /* Something is mounted on this dentry in another
1204                          * namespace and/or whatever was mounted there in this
1205                          * namespace got unmounted before lookup_mnt() could
1206                          * get it */
1207                 }
1208 
1209                 /* Handle an automount point */
1210                 if (managed & DCACHE_NEED_AUTOMOUNT) {
1211                         ret = follow_automount(path, nd, &need_mntput);
1212                         if (ret < 0)
1213                                 break;
1214                         continue;
1215                 }
1216 
1217                 /* We didn't change the current path point */
1218                 break;
1219         }
1220 
1221         if (need_mntput && path->mnt == mnt)
1222                 mntput(path->mnt);
1223         if (ret == -EISDIR)
1224                 ret = 0;
1225         if (need_mntput)
1226                 nd->flags |= LOOKUP_JUMPED;
1227         if (unlikely(ret < 0))
1228                 path_put_conditional(path, nd);
1229         return ret;
1230 }
1231 
1232 int follow_down_one(struct path *path)
1233 {
1234         struct vfsmount *mounted;
1235 
1236         mounted = lookup_mnt(path);
1237         if (mounted) {
1238                 dput(path->dentry);
1239                 mntput(path->mnt);
1240                 path->mnt = mounted;
1241                 path->dentry = dget(mounted->mnt_root);
1242                 return 1;
1243         }
1244         return 0;
1245 }
1246 EXPORT_SYMBOL(follow_down_one);
1247 
1248 static inline int managed_dentry_rcu(struct dentry *dentry)
1249 {
1250         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1251                 dentry->d_op->d_manage(dentry, true) : 0;
1252 }
1253 
1254 /*
1255  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1256  * we meet a managed dentry that would need blocking.
1257  */
1258 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1259                                struct inode **inode, unsigned *seqp)
1260 {
1261         for (;;) {
1262                 struct mount *mounted;
1263                 /*
1264                  * Don't forget we might have a non-mountpoint managed dentry
1265                  * that wants to block transit.
1266                  */
1267                 switch (managed_dentry_rcu(path->dentry)) {
1268                 case -ECHILD:
1269                 default:
1270                         return false;
1271                 case -EISDIR:
1272                         return true;
1273                 case 0:
1274                         break;
1275                 }
1276 
1277                 if (!d_mountpoint(path->dentry))
1278                         return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1279 
1280                 mounted = __lookup_mnt(path->mnt, path->dentry);
1281                 if (!mounted)
1282                         break;
1283                 path->mnt = &mounted->mnt;
1284                 path->dentry = mounted->mnt.mnt_root;
1285                 nd->flags |= LOOKUP_JUMPED;
1286                 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1287                 /*
1288                  * Update the inode too. We don't need to re-check the
1289                  * dentry sequence number here after this d_inode read,
1290                  * because a mount-point is always pinned.
1291                  */
1292                 *inode = path->dentry->d_inode;
1293         }
1294         return !read_seqretry(&mount_lock, nd->m_seq) &&
1295                 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1296 }
1297 
1298 static int follow_dotdot_rcu(struct nameidata *nd)
1299 {
1300         struct inode *inode = nd->inode;
1301 
1302         while (1) {
1303                 if (path_equal(&nd->path, &nd->root))
1304                         break;
1305                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1306                         struct dentry *old = nd->path.dentry;
1307                         struct dentry *parent = old->d_parent;
1308                         unsigned seq;
1309 
1310                         inode = parent->d_inode;
1311                         seq = read_seqcount_begin(&parent->d_seq);
1312                         if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1313                                 return -ECHILD;
1314                         nd->path.dentry = parent;
1315                         nd->seq = seq;
1316                         if (unlikely(!path_connected(&nd->path)))
1317                                 return -ENOENT;
1318                         break;
1319                 } else {
1320                         struct mount *mnt = real_mount(nd->path.mnt);
1321                         struct mount *mparent = mnt->mnt_parent;
1322                         struct dentry *mountpoint = mnt->mnt_mountpoint;
1323                         struct inode *inode2 = mountpoint->d_inode;
1324                         unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1325                         if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1326                                 return -ECHILD;
1327                         if (&mparent->mnt == nd->path.mnt)
1328                                 break;
1329                         /* we know that mountpoint was pinned */
1330                         nd->path.dentry = mountpoint;
1331                         nd->path.mnt = &mparent->mnt;
1332                         inode = inode2;
1333                         nd->seq = seq;
1334                 }
1335         }
1336         while (unlikely(d_mountpoint(nd->path.dentry))) {
1337                 struct mount *mounted;
1338                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1339                 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1340                         return -ECHILD;
1341                 if (!mounted)
1342                         break;
1343                 nd->path.mnt = &mounted->mnt;
1344                 nd->path.dentry = mounted->mnt.mnt_root;
1345                 inode = nd->path.dentry->d_inode;
1346                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1347         }
1348         nd->inode = inode;
1349         return 0;
1350 }
1351 
1352 /*
1353  * Follow down to the covering mount currently visible to userspace.  At each
1354  * point, the filesystem owning that dentry may be queried as to whether the
1355  * caller is permitted to proceed or not.
1356  */
1357 int follow_down(struct path *path)
1358 {
1359         unsigned managed;
1360         int ret;
1361 
1362         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1363                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1364                 /* Allow the filesystem to manage the transit without i_mutex
1365                  * being held.
1366                  *
1367                  * We indicate to the filesystem if someone is trying to mount
1368                  * something here.  This gives autofs the chance to deny anyone
1369                  * other than its daemon the right to mount on its
1370                  * superstructure.
1371                  *
1372                  * The filesystem may sleep at this point.
1373                  */
1374                 if (managed & DCACHE_MANAGE_TRANSIT) {
1375                         BUG_ON(!path->dentry->d_op);
1376                         BUG_ON(!path->dentry->d_op->d_manage);
1377                         ret = path->dentry->d_op->d_manage(
1378                                 path->dentry, false);
1379                         if (ret < 0)
1380                                 return ret == -EISDIR ? 0 : ret;
1381                 }
1382 
1383                 /* Transit to a mounted filesystem. */
1384                 if (managed & DCACHE_MOUNTED) {
1385                         struct vfsmount *mounted = lookup_mnt(path);
1386                         if (!mounted)
1387                                 break;
1388                         dput(path->dentry);
1389                         mntput(path->mnt);
1390                         path->mnt = mounted;
1391                         path->dentry = dget(mounted->mnt_root);
1392                         continue;
1393                 }
1394 
1395                 /* Don't handle automount points here */
1396                 break;
1397         }
1398         return 0;
1399 }
1400 EXPORT_SYMBOL(follow_down);
1401 
1402 /*
1403  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1404  */
1405 static void follow_mount(struct path *path)
1406 {
1407         while (d_mountpoint(path->dentry)) {
1408                 struct vfsmount *mounted = lookup_mnt(path);
1409                 if (!mounted)
1410                         break;
1411                 dput(path->dentry);
1412                 mntput(path->mnt);
1413                 path->mnt = mounted;
1414                 path->dentry = dget(mounted->mnt_root);
1415         }
1416 }
1417 
1418 static int follow_dotdot(struct nameidata *nd)
1419 {
1420         while(1) {
1421                 struct dentry *old = nd->path.dentry;
1422 
1423                 if (nd->path.dentry == nd->root.dentry &&
1424                     nd->path.mnt == nd->root.mnt) {
1425                         break;
1426                 }
1427                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1428                         /* rare case of legitimate dget_parent()... */
1429                         nd->path.dentry = dget_parent(nd->path.dentry);
1430                         dput(old);
1431                         if (unlikely(!path_connected(&nd->path)))
1432                                 return -ENOENT;
1433                         break;
1434                 }
1435                 if (!follow_up(&nd->path))
1436                         break;
1437         }
1438         follow_mount(&nd->path);
1439         nd->inode = nd->path.dentry->d_inode;
1440         return 0;
1441 }
1442 
1443 /*
1444  * This looks up the name in dcache, possibly revalidates the old dentry and
1445  * allocates a new one if not found or not valid.  In the need_lookup argument
1446  * returns whether i_op->lookup is necessary.
1447  *
1448  * dir->d_inode->i_mutex must be held
1449  */
1450 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1451                                     unsigned int flags, bool *need_lookup)
1452 {
1453         struct dentry *dentry;
1454         int error;
1455 
1456         *need_lookup = false;
1457         dentry = d_lookup(dir, name);
1458         if (dentry) {
1459                 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1460                         error = d_revalidate(dentry, flags);
1461                         if (unlikely(error <= 0)) {
1462                                 if (error < 0) {
1463                                         dput(dentry);
1464                                         return ERR_PTR(error);
1465                                 } else {
1466                                         d_invalidate(dentry);
1467                                         dput(dentry);
1468                                         dentry = NULL;
1469                                 }
1470                         }
1471                 }
1472         }
1473 
1474         if (!dentry) {
1475                 dentry = d_alloc(dir, name);
1476                 if (unlikely(!dentry))
1477                         return ERR_PTR(-ENOMEM);
1478 
1479                 *need_lookup = true;
1480         }
1481         return dentry;
1482 }
1483 
1484 /*
1485  * Call i_op->lookup on the dentry.  The dentry must be negative and
1486  * unhashed.
1487  *
1488  * dir->d_inode->i_mutex must be held
1489  */
1490 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1491                                   unsigned int flags)
1492 {
1493         struct dentry *old;
1494 
1495         /* Don't create child dentry for a dead directory. */
1496         if (unlikely(IS_DEADDIR(dir))) {
1497                 dput(dentry);
1498                 return ERR_PTR(-ENOENT);
1499         }
1500 
1501         old = dir->i_op->lookup(dir, dentry, flags);
1502         if (unlikely(old)) {
1503                 dput(dentry);
1504                 dentry = old;
1505         }
1506         return dentry;
1507 }
1508 
1509 static struct dentry *__lookup_hash(struct qstr *name,
1510                 struct dentry *base, unsigned int flags)
1511 {
1512         bool need_lookup;
1513         struct dentry *dentry;
1514 
1515         dentry = lookup_dcache(name, base, flags, &need_lookup);
1516         if (!need_lookup)
1517                 return dentry;
1518 
1519         return lookup_real(base->d_inode, dentry, flags);
1520 }
1521 
1522 /*
1523  *  It's more convoluted than I'd like it to be, but... it's still fairly
1524  *  small and for now I'd prefer to have fast path as straight as possible.
1525  *  It _is_ time-critical.
1526  */
1527 static int lookup_fast(struct nameidata *nd,
1528                        struct path *path, struct inode **inode,
1529                        unsigned *seqp)
1530 {
1531         struct vfsmount *mnt = nd->path.mnt;
1532         struct dentry *dentry, *parent = nd->path.dentry;
1533         int need_reval = 1;
1534         int status = 1;
1535         int err;
1536 
1537         /*
1538          * Rename seqlock is not required here because in the off chance
1539          * of a false negative due to a concurrent rename, we're going to
1540          * do the non-racy lookup, below.
1541          */
1542         if (nd->flags & LOOKUP_RCU) {
1543                 unsigned seq;
1544                 bool negative;
1545                 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1546                 if (!dentry)
1547                         goto unlazy;
1548 
1549                 /*
1550                  * This sequence count validates that the inode matches
1551                  * the dentry name information from lookup.
1552                  */
1553                 *inode = d_backing_inode(dentry);
1554                 negative = d_is_negative(dentry);
1555                 if (read_seqcount_retry(&dentry->d_seq, seq))
1556                         return -ECHILD;
1557 
1558                 /*
1559                  * This sequence count validates that the parent had no
1560                  * changes while we did the lookup of the dentry above.
1561                  *
1562                  * The memory barrier in read_seqcount_begin of child is
1563                  *  enough, we can use __read_seqcount_retry here.
1564                  */
1565                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1566                         return -ECHILD;
1567 
1568                 *seqp = seq;
1569                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1570                         status = d_revalidate(dentry, nd->flags);
1571                         if (unlikely(status <= 0)) {
1572                                 if (status != -ECHILD)
1573                                         need_reval = 0;
1574                                 goto unlazy;
1575                         }
1576                 }
1577                 /*
1578                  * Note: do negative dentry check after revalidation in
1579                  * case that drops it.
1580                  */
1581                 if (negative)
1582                         return -ENOENT;
1583                 path->mnt = mnt;
1584                 path->dentry = dentry;
1585                 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1586                         return 0;
1587 unlazy:
1588                 if (unlazy_walk(nd, dentry, seq))
1589                         return -ECHILD;
1590         } else {
1591                 dentry = __d_lookup(parent, &nd->last);
1592         }
1593 
1594         if (unlikely(!dentry))
1595                 goto need_lookup;
1596 
1597         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1598                 status = d_revalidate(dentry, nd->flags);
1599         if (unlikely(status <= 0)) {
1600                 if (status < 0) {
1601                         dput(dentry);
1602                         return status;
1603                 }
1604                 d_invalidate(dentry);
1605                 dput(dentry);
1606                 goto need_lookup;
1607         }
1608 
1609         if (unlikely(d_is_negative(dentry))) {
1610                 dput(dentry);
1611                 return -ENOENT;
1612         }
1613         path->mnt = mnt;
1614         path->dentry = dentry;
1615         err = follow_managed(path, nd);
1616         if (likely(!err))
1617                 *inode = d_backing_inode(path->dentry);
1618         return err;
1619 
1620 need_lookup:
1621         return 1;
1622 }
1623 
1624 /* Fast lookup failed, do it the slow way */
1625 static int lookup_slow(struct nameidata *nd, struct path *path)
1626 {
1627         struct dentry *dentry, *parent;
1628 
1629         parent = nd->path.dentry;
1630         BUG_ON(nd->inode != parent->d_inode);
1631 
1632         inode_lock(parent->d_inode);
1633         dentry = __lookup_hash(&nd->last, parent, nd->flags);
1634         inode_unlock(parent->d_inode);
1635         if (IS_ERR(dentry))
1636                 return PTR_ERR(dentry);
1637         path->mnt = nd->path.mnt;
1638         path->dentry = dentry;
1639         return follow_managed(path, nd);
1640 }
1641 
1642 static inline int may_lookup(struct nameidata *nd)
1643 {
1644         if (nd->flags & LOOKUP_RCU) {
1645                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1646                 if (err != -ECHILD)
1647                         return err;
1648                 if (unlazy_walk(nd, NULL, 0))
1649                         return -ECHILD;
1650         }
1651         return inode_permission(nd->inode, MAY_EXEC);
1652 }
1653 
1654 static inline int handle_dots(struct nameidata *nd, int type)
1655 {
1656         if (type == LAST_DOTDOT) {
1657                 if (!nd->root.mnt)
1658                         set_root(nd);
1659                 if (nd->flags & LOOKUP_RCU) {
1660                         return follow_dotdot_rcu(nd);
1661                 } else
1662                         return follow_dotdot(nd);
1663         }
1664         return 0;
1665 }
1666 
1667 static int pick_link(struct nameidata *nd, struct path *link,
1668                      struct inode *inode, unsigned seq)
1669 {
1670         int error;
1671         struct saved *last;
1672         if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1673                 path_to_nameidata(link, nd);
1674                 return -ELOOP;
1675         }
1676         if (!(nd->flags & LOOKUP_RCU)) {
1677                 if (link->mnt == nd->path.mnt)
1678                         mntget(link->mnt);
1679         }
1680         error = nd_alloc_stack(nd);
1681         if (unlikely(error)) {
1682                 if (error == -ECHILD) {
1683                         if (unlikely(unlazy_link(nd, link, seq)))
1684                                 return -ECHILD;
1685                         error = nd_alloc_stack(nd);
1686                 }
1687                 if (error) {
1688                         path_put(link);
1689                         return error;
1690                 }
1691         }
1692 
1693         last = nd->stack + nd->depth++;
1694         last->link = *link;
1695         clear_delayed_call(&last->done);
1696         nd->link_inode = inode;
1697         last->seq = seq;
1698         return 1;
1699 }
1700 
1701 /*
1702  * Do we need to follow links? We _really_ want to be able
1703  * to do this check without having to look at inode->i_op,
1704  * so we keep a cache of "no, this doesn't need follow_link"
1705  * for the common case.
1706  */
1707 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1708                                      int follow,
1709                                      struct inode *inode, unsigned seq)
1710 {
1711         if (likely(!d_is_symlink(link->dentry)))
1712                 return 0;
1713         if (!follow)
1714                 return 0;
1715         /* make sure that d_is_symlink above matches inode */
1716         if (nd->flags & LOOKUP_RCU) {
1717                 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1718                         return -ECHILD;
1719         }
1720         return pick_link(nd, link, inode, seq);
1721 }
1722 
1723 enum {WALK_GET = 1, WALK_PUT = 2};
1724 
1725 static int walk_component(struct nameidata *nd, int flags)
1726 {
1727         struct path path;
1728         struct inode *inode;
1729         unsigned seq;
1730         int err;
1731         /*
1732          * "." and ".." are special - ".." especially so because it has
1733          * to be able to know about the current root directory and
1734          * parent relationships.
1735          */
1736         if (unlikely(nd->last_type != LAST_NORM)) {
1737                 err = handle_dots(nd, nd->last_type);
1738                 if (flags & WALK_PUT)
1739                         put_link(nd);
1740                 return err;
1741         }
1742         err = lookup_fast(nd, &path, &inode, &seq);
1743         if (unlikely(err)) {
1744                 if (err < 0)
1745                         return err;
1746 
1747                 err = lookup_slow(nd, &path);
1748                 if (err < 0)
1749                         return err;
1750 
1751                 seq = 0;        /* we are already out of RCU mode */
1752                 err = -ENOENT;
1753                 if (d_is_negative(path.dentry))
1754                         goto out_path_put;
1755                 inode = d_backing_inode(path.dentry);
1756         }
1757 
1758         if (flags & WALK_PUT)
1759                 put_link(nd);
1760         err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1761         if (unlikely(err))
1762                 return err;
1763         path_to_nameidata(&path, nd);
1764         nd->inode = inode;
1765         nd->seq = seq;
1766         return 0;
1767 
1768 out_path_put:
1769         path_to_nameidata(&path, nd);
1770         return err;
1771 }
1772 
1773 /*
1774  * We can do the critical dentry name comparison and hashing
1775  * operations one word at a time, but we are limited to:
1776  *
1777  * - Architectures with fast unaligned word accesses. We could
1778  *   do a "get_unaligned()" if this helps and is sufficiently
1779  *   fast.
1780  *
1781  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1782  *   do not trap on the (extremely unlikely) case of a page
1783  *   crossing operation.
1784  *
1785  * - Furthermore, we need an efficient 64-bit compile for the
1786  *   64-bit case in order to generate the "number of bytes in
1787  *   the final mask". Again, that could be replaced with a
1788  *   efficient population count instruction or similar.
1789  */
1790 #ifdef CONFIG_DCACHE_WORD_ACCESS
1791 
1792 #include <asm/word-at-a-time.h>
1793 
1794 #ifdef CONFIG_64BIT
1795 
1796 static inline unsigned int fold_hash(unsigned long hash)
1797 {
1798         return hash_64(hash, 32);
1799 }
1800 
1801 #else   /* 32-bit case */
1802 
1803 #define fold_hash(x) (x)
1804 
1805 #endif
1806 
1807 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1808 {
1809         unsigned long a, mask;
1810         unsigned long hash = 0;
1811 
1812         for (;;) {
1813                 a = load_unaligned_zeropad(name);
1814                 if (len < sizeof(unsigned long))
1815                         break;
1816                 hash += a;
1817                 hash *= 9;
1818                 name += sizeof(unsigned long);
1819                 len -= sizeof(unsigned long);
1820                 if (!len)
1821                         goto done;
1822         }
1823         mask = bytemask_from_count(len);
1824         hash += mask & a;
1825 done:
1826         return fold_hash(hash);
1827 }
1828 EXPORT_SYMBOL(full_name_hash);
1829 
1830 /*
1831  * Calculate the length and hash of the path component, and
1832  * return the "hash_len" as the result.
1833  */
1834 static inline u64 hash_name(const char *name)
1835 {
1836         unsigned long a, b, adata, bdata, mask, hash, len;
1837         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1838 
1839         hash = a = 0;
1840         len = -sizeof(unsigned long);
1841         do {
1842                 hash = (hash + a) * 9;
1843                 len += sizeof(unsigned long);
1844                 a = load_unaligned_zeropad(name+len);
1845                 b = a ^ REPEAT_BYTE('/');
1846         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1847 
1848         adata = prep_zero_mask(a, adata, &constants);
1849         bdata = prep_zero_mask(b, bdata, &constants);
1850 
1851         mask = create_zero_mask(adata | bdata);
1852 
1853         hash += a & zero_bytemask(mask);
1854         len += find_zero(mask);
1855         return hashlen_create(fold_hash(hash), len);
1856 }
1857 
1858 #else
1859 
1860 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1861 {
1862         unsigned long hash = init_name_hash();
1863         while (len--)
1864                 hash = partial_name_hash(*name++, hash);
1865         return end_name_hash(hash);
1866 }
1867 EXPORT_SYMBOL(full_name_hash);
1868 
1869 /*
1870  * We know there's a real path component here of at least
1871  * one character.
1872  */
1873 static inline u64 hash_name(const char *name)
1874 {
1875         unsigned long hash = init_name_hash();
1876         unsigned long len = 0, c;
1877 
1878         c = (unsigned char)*name;
1879         do {
1880                 len++;
1881                 hash = partial_name_hash(c, hash);
1882                 c = (unsigned char)name[len];
1883         } while (c && c != '/');
1884         return hashlen_create(end_name_hash(hash), len);
1885 }
1886 
1887 #endif
1888 
1889 /*
1890  * Name resolution.
1891  * This is the basic name resolution function, turning a pathname into
1892  * the final dentry. We expect 'base' to be positive and a directory.
1893  *
1894  * Returns 0 and nd will have valid dentry and mnt on success.
1895  * Returns error and drops reference to input namei data on failure.
1896  */
1897 static int link_path_walk(const char *name, struct nameidata *nd)
1898 {
1899         int err;
1900 
1901         while (*name=='/')
1902                 name++;
1903         if (!*name)
1904                 return 0;
1905 
1906         /* At this point we know we have a real path component. */
1907         for(;;) {
1908                 u64 hash_len;
1909                 int type;
1910 
1911                 err = may_lookup(nd);
1912                 if (err)
1913                         return err;
1914 
1915                 hash_len = hash_name(name);
1916 
1917                 type = LAST_NORM;
1918                 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1919                         case 2:
1920                                 if (name[1] == '.') {
1921                                         type = LAST_DOTDOT;
1922                                         nd->flags |= LOOKUP_JUMPED;
1923                                 }
1924                                 break;
1925                         case 1:
1926                                 type = LAST_DOT;
1927                 }
1928                 if (likely(type == LAST_NORM)) {
1929                         struct dentry *parent = nd->path.dentry;
1930                         nd->flags &= ~LOOKUP_JUMPED;
1931                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1932                                 struct qstr this = { { .hash_len = hash_len }, .name = name };
1933                                 err = parent->d_op->d_hash(parent, &this);
1934                                 if (err < 0)
1935                                         return err;
1936                                 hash_len = this.hash_len;
1937                                 name = this.name;
1938                         }
1939                 }
1940 
1941                 nd->last.hash_len = hash_len;
1942                 nd->last.name = name;
1943                 nd->last_type = type;
1944 
1945                 name += hashlen_len(hash_len);
1946                 if (!*name)
1947                         goto OK;
1948                 /*
1949                  * If it wasn't NUL, we know it was '/'. Skip that
1950                  * slash, and continue until no more slashes.
1951                  */
1952                 do {
1953                         name++;
1954                 } while (unlikely(*name == '/'));
1955                 if (unlikely(!*name)) {
1956 OK:
1957                         /* pathname body, done */
1958                         if (!nd->depth)
1959                                 return 0;
1960                         name = nd->stack[nd->depth - 1].name;
1961                         /* trailing symlink, done */
1962                         if (!name)
1963                                 return 0;
1964                         /* last component of nested symlink */
1965                         err = walk_component(nd, WALK_GET | WALK_PUT);
1966                 } else {
1967                         err = walk_component(nd, WALK_GET);
1968                 }
1969                 if (err < 0)
1970                         return err;
1971 
1972                 if (err) {
1973                         const char *s = get_link(nd);
1974 
1975                         if (IS_ERR(s))
1976                                 return PTR_ERR(s);
1977                         err = 0;
1978                         if (unlikely(!s)) {
1979                                 /* jumped */
1980                                 put_link(nd);
1981                         } else {
1982                                 nd->stack[nd->depth - 1].name = name;
1983                                 name = s;
1984                                 continue;
1985                         }
1986                 }
1987                 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1988                         if (nd->flags & LOOKUP_RCU) {
1989                                 if (unlazy_walk(nd, NULL, 0))
1990                                         return -ECHILD;
1991                         }
1992                         return -ENOTDIR;
1993                 }
1994         }
1995 }
1996 
1997 static const char *path_init(struct nameidata *nd, unsigned flags)
1998 {
1999         int retval = 0;
2000         const char *s = nd->name->name;
2001 
2002         nd->last_type = LAST_ROOT; /* if there are only slashes... */
2003         nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2004         nd->depth = 0;
2005         if (flags & LOOKUP_ROOT) {
2006                 struct dentry *root = nd->root.dentry;
2007                 struct inode *inode = root->d_inode;
2008                 if (*s) {
2009                         if (!d_can_lookup(root))
2010                                 return ERR_PTR(-ENOTDIR);
2011                         retval = inode_permission(inode, MAY_EXEC);
2012                         if (retval)
2013                                 return ERR_PTR(retval);
2014                 }
2015                 nd->path = nd->root;
2016                 nd->inode = inode;
2017                 if (flags & LOOKUP_RCU) {
2018                         rcu_read_lock();
2019                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2020                         nd->root_seq = nd->seq;
2021                         nd->m_seq = read_seqbegin(&mount_lock);
2022                 } else {
2023                         path_get(&nd->path);
2024                 }
2025                 return s;
2026         }
2027 
2028         nd->root.mnt = NULL;
2029         nd->path.mnt = NULL;
2030         nd->path.dentry = NULL;
2031 
2032         nd->m_seq = read_seqbegin(&mount_lock);
2033         if (*s == '/') {
2034                 if (flags & LOOKUP_RCU)
2035                         rcu_read_lock();
2036                 set_root(nd);
2037                 if (likely(!nd_jump_root(nd)))
2038                         return s;
2039                 nd->root.mnt = NULL;
2040                 rcu_read_unlock();
2041                 return ERR_PTR(-ECHILD);
2042         } else if (nd->dfd == AT_FDCWD) {
2043                 if (flags & LOOKUP_RCU) {
2044                         struct fs_struct *fs = current->fs;
2045                         unsigned seq;
2046 
2047                         rcu_read_lock();
2048 
2049                         do {
2050                                 seq = read_seqcount_begin(&fs->seq);
2051                                 nd->path = fs->pwd;
2052                                 nd->inode = nd->path.dentry->d_inode;
2053                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2054                         } while (read_seqcount_retry(&fs->seq, seq));
2055                 } else {
2056                         get_fs_pwd(current->fs, &nd->path);
2057                         nd->inode = nd->path.dentry->d_inode;
2058                 }
2059                 return s;
2060         } else {
2061                 /* Caller must check execute permissions on the starting path component */
2062                 struct fd f = fdget_raw(nd->dfd);
2063                 struct dentry *dentry;
2064 
2065                 if (!f.file)
2066                         return ERR_PTR(-EBADF);
2067 
2068                 dentry = f.file->f_path.dentry;
2069 
2070                 if (*s) {
2071                         if (!d_can_lookup(dentry)) {
2072                                 fdput(f);
2073                                 return ERR_PTR(-ENOTDIR);
2074                         }
2075                 }
2076 
2077                 nd->path = f.file->f_path;
2078                 if (flags & LOOKUP_RCU) {
2079                         rcu_read_lock();
2080                         nd->inode = nd->path.dentry->d_inode;
2081                         nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2082                 } else {
2083                         path_get(&nd->path);
2084                         nd->inode = nd->path.dentry->d_inode;
2085                 }
2086                 fdput(f);
2087                 return s;
2088         }
2089 }
2090 
2091 static const char *trailing_symlink(struct nameidata *nd)
2092 {
2093         const char *s;
2094         int error = may_follow_link(nd);
2095         if (unlikely(error))
2096                 return ERR_PTR(error);
2097         nd->flags |= LOOKUP_PARENT;
2098         nd->stack[0].name = NULL;
2099         s = get_link(nd);
2100         return s ? s : "";
2101 }
2102 
2103 static inline int lookup_last(struct nameidata *nd)
2104 {
2105         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2106                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2107 
2108         nd->flags &= ~LOOKUP_PARENT;
2109         return walk_component(nd,
2110                         nd->flags & LOOKUP_FOLLOW
2111                                 ? nd->depth
2112                                         ? WALK_PUT | WALK_GET
2113                                         : WALK_GET
2114                                 : 0);
2115 }
2116 
2117 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2118 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2119 {
2120         const char *s = path_init(nd, flags);
2121         int err;
2122 
2123         if (IS_ERR(s))
2124                 return PTR_ERR(s);
2125         while (!(err = link_path_walk(s, nd))
2126                 && ((err = lookup_last(nd)) > 0)) {
2127                 s = trailing_symlink(nd);
2128                 if (IS_ERR(s)) {
2129                         err = PTR_ERR(s);
2130                         break;
2131                 }
2132         }
2133         if (!err)
2134                 err = complete_walk(nd);
2135 
2136         if (!err && nd->flags & LOOKUP_DIRECTORY)
2137                 if (!d_can_lookup(nd->path.dentry))
2138                         err = -ENOTDIR;
2139         if (!err) {
2140                 *path = nd->path;
2141                 nd->path.mnt = NULL;
2142                 nd->path.dentry = NULL;
2143         }
2144         terminate_walk(nd);
2145         return err;
2146 }
2147 
2148 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2149                            struct path *path, struct path *root)
2150 {
2151         int retval;
2152         struct nameidata nd;
2153         if (IS_ERR(name))
2154                 return PTR_ERR(name);
2155         if (unlikely(root)) {
2156                 nd.root = *root;
2157                 flags |= LOOKUP_ROOT;
2158         }
2159         set_nameidata(&nd, dfd, name);
2160         retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2161         if (unlikely(retval == -ECHILD))
2162                 retval = path_lookupat(&nd, flags, path);
2163         if (unlikely(retval == -ESTALE))
2164                 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2165 
2166         if (likely(!retval))
2167                 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2168         restore_nameidata();
2169         putname(name);
2170         return retval;
2171 }
2172 
2173 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2174 static int path_parentat(struct nameidata *nd, unsigned flags,
2175                                 struct path *parent)
2176 {
2177         const char *s = path_init(nd, flags);
2178         int err;
2179         if (IS_ERR(s))
2180                 return PTR_ERR(s);
2181         err = link_path_walk(s, nd);
2182         if (!err)
2183                 err = complete_walk(nd);
2184         if (!err) {
2185                 *parent = nd->path;
2186                 nd->path.mnt = NULL;
2187                 nd->path.dentry = NULL;
2188         }
2189         terminate_walk(nd);
2190         return err;
2191 }
2192 
2193 static struct filename *filename_parentat(int dfd, struct filename *name,
2194                                 unsigned int flags, struct path *parent,
2195                                 struct qstr *last, int *type)
2196 {
2197         int retval;
2198         struct nameidata nd;
2199 
2200         if (IS_ERR(name))
2201                 return name;
2202         set_nameidata(&nd, dfd, name);
2203         retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2204         if (unlikely(retval == -ECHILD))
2205                 retval = path_parentat(&nd, flags, parent);
2206         if (unlikely(retval == -ESTALE))
2207                 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2208         if (likely(!retval)) {
2209                 *last = nd.last;
2210                 *type = nd.last_type;
2211                 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2212         } else {
2213                 putname(name);
2214                 name = ERR_PTR(retval);
2215         }
2216         restore_nameidata();
2217         return name;
2218 }
2219 
2220 /* does lookup, returns the object with parent locked */
2221 struct dentry *kern_path_locked(const char *name, struct path *path)
2222 {
2223         struct filename *filename;
2224         struct dentry *d;
2225         struct qstr last;
2226         int type;
2227 
2228         filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2229                                     &last, &type);
2230         if (IS_ERR(filename))
2231                 return ERR_CAST(filename);
2232         if (unlikely(type != LAST_NORM)) {
2233                 path_put(path);
2234                 putname(filename);
2235                 return ERR_PTR(-EINVAL);
2236         }
2237         inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2238         d = __lookup_hash(&last, path->dentry, 0);
2239         if (IS_ERR(d)) {
2240                 inode_unlock(path->dentry->d_inode);
2241                 path_put(path);
2242         }
2243         putname(filename);
2244         return d;
2245 }
2246 
2247 int kern_path(const char *name, unsigned int flags, struct path *path)
2248 {
2249         return filename_lookup(AT_FDCWD, getname_kernel(name),
2250                                flags, path, NULL);
2251 }
2252 EXPORT_SYMBOL(kern_path);
2253 
2254 /**
2255  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2256  * @dentry:  pointer to dentry of the base directory
2257  * @mnt: pointer to vfs mount of the base directory
2258  * @name: pointer to file name
2259  * @flags: lookup flags
2260  * @path: pointer to struct path to fill
2261  */
2262 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2263                     const char *name, unsigned int flags,
2264                     struct path *path)
2265 {
2266         struct path root = {.mnt = mnt, .dentry = dentry};
2267         /* the first argument of filename_lookup() is ignored with root */
2268         return filename_lookup(AT_FDCWD, getname_kernel(name),
2269                                flags , path, &root);
2270 }
2271 EXPORT_SYMBOL(vfs_path_lookup);
2272 
2273 /**
2274  * lookup_one_len - filesystem helper to lookup single pathname component
2275  * @name:       pathname component to lookup
2276  * @base:       base directory to lookup from
2277  * @len:        maximum length @len should be interpreted to
2278  *
2279  * Note that this routine is purely a helper for filesystem usage and should
2280  * not be called by generic code.
2281  *
2282  * The caller must hold base->i_mutex.
2283  */
2284 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2285 {
2286         struct qstr this;
2287         unsigned int c;
2288         int err;
2289 
2290         WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2291 
2292         this.name = name;
2293         this.len = len;
2294         this.hash = full_name_hash(name, len);
2295         if (!len)
2296                 return ERR_PTR(-EACCES);
2297 
2298         if (unlikely(name[0] == '.')) {
2299                 if (len < 2 || (len == 2 && name[1] == '.'))
2300                         return ERR_PTR(-EACCES);
2301         }
2302 
2303         while (len--) {
2304                 c = *(const unsigned char *)name++;
2305                 if (c == '/' || c == '\0')
2306                         return ERR_PTR(-EACCES);
2307         }
2308         /*
2309          * See if the low-level filesystem might want
2310          * to use its own hash..
2311          */
2312         if (base->d_flags & DCACHE_OP_HASH) {
2313                 int err = base->d_op->d_hash(base, &this);
2314                 if (err < 0)
2315                         return ERR_PTR(err);
2316         }
2317 
2318         err = inode_permission(base->d_inode, MAY_EXEC);
2319         if (err)
2320                 return ERR_PTR(err);
2321 
2322         return __lookup_hash(&this, base, 0);
2323 }
2324 EXPORT_SYMBOL(lookup_one_len);
2325 
2326 /**
2327  * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2328  * @name:       pathname component to lookup
2329  * @base:       base directory to lookup from
2330  * @len:        maximum length @len should be interpreted to
2331  *
2332  * Note that this routine is purely a helper for filesystem usage and should
2333  * not be called by generic code.
2334  *
2335  * Unlike lookup_one_len, it should be called without the parent
2336  * i_mutex held, and will take the i_mutex itself if necessary.
2337  */
2338 struct dentry *lookup_one_len_unlocked(const char *name,
2339                                        struct dentry *base, int len)
2340 {
2341         struct qstr this;
2342         unsigned int c;
2343         int err;
2344         struct dentry *ret;
2345 
2346         this.name = name;
2347         this.len = len;
2348         this.hash = full_name_hash(name, len);
2349         if (!len)
2350                 return ERR_PTR(-EACCES);
2351 
2352         if (unlikely(name[0] == '.')) {
2353                 if (len < 2 || (len == 2 && name[1] == '.'))
2354                         return ERR_PTR(-EACCES);
2355         }
2356 
2357         while (len--) {
2358                 c = *(const unsigned char *)name++;
2359                 if (c == '/' || c == '\0')
2360                         return ERR_PTR(-EACCES);
2361         }
2362         /*
2363          * See if the low-level filesystem might want
2364          * to use its own hash..
2365          */
2366         if (base->d_flags & DCACHE_OP_HASH) {
2367                 int err = base->d_op->d_hash(base, &this);
2368                 if (err < 0)
2369                         return ERR_PTR(err);
2370         }
2371 
2372         err = inode_permission(base->d_inode, MAY_EXEC);
2373         if (err)
2374                 return ERR_PTR(err);
2375 
2376         /*
2377          * __d_lookup() is used to try to get a quick answer and avoid the
2378          * mutex.  A false-negative does no harm.
2379          */
2380         ret = __d_lookup(base, &this);
2381         if (ret && unlikely(ret->d_flags & DCACHE_OP_REVALIDATE)) {
2382                 dput(ret);
2383                 ret = NULL;
2384         }
2385         if (ret)
2386                 return ret;
2387 
2388         inode_lock(base->d_inode);
2389         ret =  __lookup_hash(&this, base, 0);
2390         inode_unlock(base->d_inode);
2391         return ret;
2392 }
2393 EXPORT_SYMBOL(lookup_one_len_unlocked);
2394 
2395 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2396                  struct path *path, int *empty)
2397 {
2398         return filename_lookup(dfd, getname_flags(name, flags, empty),
2399                                flags, path, NULL);
2400 }
2401 EXPORT_SYMBOL(user_path_at_empty);
2402 
2403 /*
2404  * NB: most callers don't do anything directly with the reference to the
2405  *     to struct filename, but the nd->last pointer points into the name string
2406  *     allocated by getname. So we must hold the reference to it until all
2407  *     path-walking is complete.
2408  */
2409 static inline struct filename *
2410 user_path_parent(int dfd, const char __user *path,
2411                  struct path *parent,
2412                  struct qstr *last,
2413                  int *type,
2414                  unsigned int flags)
2415 {
2416         /* only LOOKUP_REVAL is allowed in extra flags */
2417         return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2418                                  parent, last, type);
2419 }
2420 
2421 /**
2422  * mountpoint_last - look up last component for umount
2423  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2424  * @path: pointer to container for result
2425  *
2426  * This is a special lookup_last function just for umount. In this case, we
2427  * need to resolve the path without doing any revalidation.
2428  *
2429  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2430  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2431  * in almost all cases, this lookup will be served out of the dcache. The only
2432  * cases where it won't are if nd->last refers to a symlink or the path is
2433  * bogus and it doesn't exist.
2434  *
2435  * Returns:
2436  * -error: if there was an error during lookup. This includes -ENOENT if the
2437  *         lookup found a negative dentry. The nd->path reference will also be
2438  *         put in this case.
2439  *
2440  * 0:      if we successfully resolved nd->path and found it to not to be a
2441  *         symlink that needs to be followed. "path" will also be populated.
2442  *         The nd->path reference will also be put.
2443  *
2444  * 1:      if we successfully resolved nd->last and found it to be a symlink
2445  *         that needs to be followed. "path" will be populated with the path
2446  *         to the link, and nd->path will *not* be put.
2447  */
2448 static int
2449 mountpoint_last(struct nameidata *nd, struct path *path)
2450 {
2451         int error = 0;
2452         struct dentry *dentry;
2453         struct dentry *dir = nd->path.dentry;
2454 
2455         /* If we're in rcuwalk, drop out of it to handle last component */
2456         if (nd->flags & LOOKUP_RCU) {
2457                 if (unlazy_walk(nd, NULL, 0))
2458                         return -ECHILD;
2459         }
2460 
2461         nd->flags &= ~LOOKUP_PARENT;
2462 
2463         if (unlikely(nd->last_type != LAST_NORM)) {
2464                 error = handle_dots(nd, nd->last_type);
2465                 if (error)
2466                         return error;
2467                 dentry = dget(nd->path.dentry);
2468                 goto done;
2469         }
2470 
2471         inode_lock(dir->d_inode);
2472         dentry = d_lookup(dir, &nd->last);
2473         if (!dentry) {
2474                 /*
2475                  * No cached dentry. Mounted dentries are pinned in the cache,
2476                  * so that means that this dentry is probably a symlink or the
2477                  * path doesn't actually point to a mounted dentry.
2478                  */
2479                 dentry = d_alloc(dir, &nd->last);
2480                 if (!dentry) {
2481                         inode_unlock(dir->d_inode);
2482                         return -ENOMEM;
2483                 }
2484                 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2485                 if (IS_ERR(dentry)) {
2486                         inode_unlock(dir->d_inode);
2487                         return PTR_ERR(dentry);
2488                 }
2489         }
2490         inode_unlock(dir->d_inode);
2491 
2492 done:
2493         if (d_is_negative(dentry)) {
2494                 dput(dentry);
2495                 return -ENOENT;
2496         }
2497         if (nd->depth)
2498                 put_link(nd);
2499         path->dentry = dentry;
2500         path->mnt = nd->path.mnt;
2501         error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2502                                    d_backing_inode(dentry), 0);
2503         if (unlikely(error))
2504                 return error;
2505         mntget(path->mnt);
2506         follow_mount(path);
2507         return 0;
2508 }
2509 
2510 /**
2511  * path_mountpoint - look up a path to be umounted
2512  * @nd:         lookup context
2513  * @flags:      lookup flags
2514  * @path:       pointer to container for result
2515  *
2516  * Look up the given name, but don't attempt to revalidate the last component.
2517  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2518  */
2519 static int
2520 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2521 {
2522         const char *s = path_init(nd, flags);
2523         int err;
2524         if (IS_ERR(s))
2525                 return PTR_ERR(s);
2526         while (!(err = link_path_walk(s, nd)) &&
2527                 (err = mountpoint_last(nd, path)) > 0) {
2528                 s = trailing_symlink(nd);
2529                 if (IS_ERR(s)) {
2530                         err = PTR_ERR(s);
2531                         break;
2532                 }
2533         }
2534         terminate_walk(nd);
2535         return err;
2536 }
2537 
2538 static int
2539 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2540                         unsigned int flags)
2541 {
2542         struct nameidata nd;
2543         int error;
2544         if (IS_ERR(name))
2545                 return PTR_ERR(name);
2546         set_nameidata(&nd, dfd, name);
2547         error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2548         if (unlikely(error == -ECHILD))
2549                 error = path_mountpoint(&nd, flags, path);
2550         if (unlikely(error == -ESTALE))
2551                 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2552         if (likely(!error))
2553                 audit_inode(name, path->dentry, 0);
2554         restore_nameidata();
2555         putname(name);
2556         return error;
2557 }
2558 
2559 /**
2560  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2561  * @dfd:        directory file descriptor
2562  * @name:       pathname from userland
2563  * @flags:      lookup flags
2564  * @path:       pointer to container to hold result
2565  *
2566  * A umount is a special case for path walking. We're not actually interested
2567  * in the inode in this situation, and ESTALE errors can be a problem. We
2568  * simply want track down the dentry and vfsmount attached at the mountpoint
2569  * and avoid revalidating the last component.
2570  *
2571  * Returns 0 and populates "path" on success.
2572  */
2573 int
2574 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2575                         struct path *path)
2576 {
2577         return filename_mountpoint(dfd, getname(name), path, flags);
2578 }
2579 
2580 int
2581 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2582                         unsigned int flags)
2583 {
2584         return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2585 }
2586 EXPORT_SYMBOL(kern_path_mountpoint);
2587 
2588 int __check_sticky(struct inode *dir, struct inode *inode)
2589 {
2590         kuid_t fsuid = current_fsuid();
2591 
2592         if (uid_eq(inode->i_uid, fsuid))
2593                 return 0;
2594         if (uid_eq(dir->i_uid, fsuid))
2595                 return 0;
2596         return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2597 }
2598 EXPORT_SYMBOL(__check_sticky);
2599 
2600 /*
2601  *      Check whether we can remove a link victim from directory dir, check
2602  *  whether the type of victim is right.
2603  *  1. We can't do it if dir is read-only (done in permission())
2604  *  2. We should have write and exec permissions on dir
2605  *  3. We can't remove anything from append-only dir
2606  *  4. We can't do anything with immutable dir (done in permission())
2607  *  5. If the sticky bit on dir is set we should either
2608  *      a. be owner of dir, or
2609  *      b. be owner of victim, or
2610  *      c. have CAP_FOWNER capability
2611  *  6. If the victim is append-only or immutable we can't do antyhing with
2612  *     links pointing to it.
2613  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2614  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2615  *  9. We can't remove a root or mountpoint.
2616  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2617  *     nfs_async_unlink().
2618  */
2619 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2620 {
2621         struct inode *inode = d_backing_inode(victim);
2622         int error;
2623 
2624         if (d_is_negative(victim))
2625                 return -ENOENT;
2626         BUG_ON(!inode);
2627 
2628         BUG_ON(victim->d_parent->d_inode != dir);
2629         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2630 
2631         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2632         if (error)
2633                 return error;
2634         if (IS_APPEND(dir))
2635                 return -EPERM;
2636 
2637         if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2638             IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2639                 return -EPERM;
2640         if (isdir) {
2641                 if (!d_is_dir(victim))
2642                         return -ENOTDIR;
2643                 if (IS_ROOT(victim))
2644                         return -EBUSY;
2645         } else if (d_is_dir(victim))
2646                 return -EISDIR;
2647         if (IS_DEADDIR(dir))
2648                 return -ENOENT;
2649         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2650                 return -EBUSY;
2651         return 0;
2652 }
2653 
2654 /*      Check whether we can create an object with dentry child in directory
2655  *  dir.
2656  *  1. We can't do it if child already exists (open has special treatment for
2657  *     this case, but since we are inlined it's OK)
2658  *  2. We can't do it if dir is read-only (done in permission())
2659  *  3. We should have write and exec permissions on dir
2660  *  4. We can't do it if dir is immutable (done in permission())
2661  */
2662 static inline int may_create(struct inode *dir, struct dentry *child)
2663 {
2664         audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2665         if (child->d_inode)
2666                 return -EEXIST;
2667         if (IS_DEADDIR(dir))
2668                 return -ENOENT;
2669         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2670 }
2671 
2672 /*
2673  * p1 and p2 should be directories on the same fs.
2674  */
2675 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2676 {
2677         struct dentry *p;
2678 
2679         if (p1 == p2) {
2680                 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2681                 return NULL;
2682         }
2683 
2684         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2685 
2686         p = d_ancestor(p2, p1);
2687         if (p) {
2688                 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2689                 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2690                 return p;
2691         }
2692 
2693         p = d_ancestor(p1, p2);
2694         if (p) {
2695                 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2696                 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2697                 return p;
2698         }
2699 
2700         inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2701         inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2702         return NULL;
2703 }
2704 EXPORT_SYMBOL(lock_rename);
2705 
2706 void unlock_rename(struct dentry *p1, struct dentry *p2)
2707 {
2708         inode_unlock(p1->d_inode);
2709         if (p1 != p2) {
2710                 inode_unlock(p2->d_inode);
2711                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2712         }
2713 }
2714 EXPORT_SYMBOL(unlock_rename);
2715 
2716 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2717                 bool want_excl)
2718 {
2719         int error = may_create(dir, dentry);
2720         if (error)
2721                 return error;
2722 
2723         if (!dir->i_op->create)
2724                 return -EACCES; /* shouldn't it be ENOSYS? */
2725         mode &= S_IALLUGO;
2726         mode |= S_IFREG;
2727         error = security_inode_create(dir, dentry, mode);
2728         if (error)
2729                 return error;
2730         error = dir->i_op->create(dir, dentry, mode, want_excl);
2731         if (!error)
2732                 fsnotify_create(dir, dentry);
2733         return error;
2734 }
2735 EXPORT_SYMBOL(vfs_create);
2736 
2737 static int may_open(struct path *path, int acc_mode, int flag)
2738 {
2739         struct dentry *dentry = path->dentry;
2740         struct inode *inode = dentry->d_inode;
2741         int error;
2742 
2743         if (!inode)
2744                 return -ENOENT;
2745 
2746         switch (inode->i_mode & S_IFMT) {
2747         case S_IFLNK:
2748                 return -ELOOP;
2749         case S_IFDIR:
2750                 if (acc_mode & MAY_WRITE)
2751                         return -EISDIR;
2752                 break;
2753         case S_IFBLK:
2754         case S_IFCHR:
2755                 if (path->mnt->mnt_flags & MNT_NODEV)
2756                         return -EACCES;
2757                 /*FALLTHRU*/
2758         case S_IFIFO:
2759         case S_IFSOCK:
2760                 flag &= ~O_TRUNC;
2761                 break;
2762         }
2763 
2764         error = inode_permission(inode, MAY_OPEN | acc_mode);
2765         if (error)
2766                 return error;
2767 
2768         /*
2769          * An append-only file must be opened in append mode for writing.
2770          */
2771         if (IS_APPEND(inode)) {
2772                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2773                         return -EPERM;
2774                 if (flag & O_TRUNC)
2775                         return -EPERM;
2776         }
2777 
2778         /* O_NOATIME can only be set by the owner or superuser */
2779         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2780                 return -EPERM;
2781 
2782         return 0;
2783 }
2784 
2785 static int handle_truncate(struct file *filp)
2786 {
2787         struct path *path = &filp->f_path;
2788         struct inode *inode = path->dentry->d_inode;
2789         int error = get_write_access(inode);
2790         if (error)
2791                 return error;
2792         /*
2793          * Refuse to truncate files with mandatory locks held on them.
2794          */
2795         error = locks_verify_locked(filp);
2796         if (!error)
2797                 error = security_path_truncate(path);
2798         if (!error) {
2799                 error = do_truncate(path->dentry, 0,
2800                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2801                                     filp);
2802         }
2803         put_write_access(inode);
2804         return error;
2805 }
2806 
2807 static inline int open_to_namei_flags(int flag)
2808 {
2809         if ((flag & O_ACCMODE) == 3)
2810                 flag--;
2811         return flag;
2812 }
2813 
2814 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2815 {
2816         int error = security_path_mknod(dir, dentry, mode, 0);
2817         if (error)
2818                 return error;
2819 
2820         error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2821         if (error)
2822                 return error;
2823 
2824         return security_inode_create(dir->dentry->d_inode, dentry, mode);
2825 }
2826 
2827 /*
2828  * Attempt to atomically look up, create and open a file from a negative
2829  * dentry.
2830  *
2831  * Returns 0 if successful.  The file will have been created and attached to
2832  * @file by the filesystem calling finish_open().
2833  *
2834  * Returns 1 if the file was looked up only or didn't need creating.  The
2835  * caller will need to perform the open themselves.  @path will have been
2836  * updated to point to the new dentry.  This may be negative.
2837  *
2838  * Returns an error code otherwise.
2839  */
2840 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2841                         struct path *path, struct file *file,
2842                         const struct open_flags *op,
2843                         bool got_write, bool need_lookup,
2844                         int *opened)
2845 {
2846         struct inode *dir =  nd->path.dentry->d_inode;
2847         unsigned open_flag = open_to_namei_flags(op->open_flag);
2848         umode_t mode;
2849         int error;
2850         int acc_mode;
2851         int create_error = 0;
2852         struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2853         bool excl;
2854 
2855         BUG_ON(dentry->d_inode);
2856 
2857         /* Don't create child dentry for a dead directory. */
2858         if (unlikely(IS_DEADDIR(dir))) {
2859                 error = -ENOENT;
2860                 goto out;
2861         }
2862 
2863         mode = op->mode;
2864         if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2865                 mode &= ~current_umask();
2866 
2867         excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2868         if (excl)
2869                 open_flag &= ~O_TRUNC;
2870 
2871         /*
2872          * Checking write permission is tricky, bacuse we don't know if we are
2873          * going to actually need it: O_CREAT opens should work as long as the
2874          * file exists.  But checking existence breaks atomicity.  The trick is
2875          * to check access and if not granted clear O_CREAT from the flags.
2876          *
2877          * Another problem is returing the "right" error value (e.g. for an
2878          * O_EXCL open we want to return EEXIST not EROFS).
2879          */
2880         if (((open_flag & (O_CREAT | O_TRUNC)) ||
2881             (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2882                 if (!(open_flag & O_CREAT)) {
2883                         /*
2884                          * No O_CREATE -> atomicity not a requirement -> fall
2885                          * back to lookup + open
2886                          */
2887                         goto no_open;
2888                 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2889                         /* Fall back and fail with the right error */
2890                         create_error = -EROFS;
2891                         goto no_open;
2892                 } else {
2893                         /* No side effects, safe to clear O_CREAT */
2894                         create_error = -EROFS;
2895                         open_flag &= ~O_CREAT;
2896                 }
2897         }
2898 
2899         if (open_flag & O_CREAT) {
2900                 error = may_o_create(&nd->path, dentry, mode);
2901                 if (error) {
2902                         create_error = error;
2903                         if (open_flag & O_EXCL)
2904                                 goto no_open;
2905                         open_flag &= ~O_CREAT;
2906                 }
2907         }
2908 
2909         if (nd->flags & LOOKUP_DIRECTORY)
2910                 open_flag |= O_DIRECTORY;
2911 
2912         file->f_path.dentry = DENTRY_NOT_SET;
2913         file->f_path.mnt = nd->path.mnt;
2914         error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2915                                       opened);
2916         if (error < 0) {
2917                 if (create_error && error == -ENOENT)
2918                         error = create_error;
2919                 goto out;
2920         }
2921 
2922         if (error) {    /* returned 1, that is */
2923                 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2924                         error = -EIO;
2925                         goto out;
2926                 }
2927                 if (file->f_path.dentry) {
2928                         dput(dentry);
2929                         dentry = file->f_path.dentry;
2930                 }
2931                 if (*opened & FILE_CREATED)
2932                         fsnotify_create(dir, dentry);
2933                 if (!dentry->d_inode) {
2934                         WARN_ON(*opened & FILE_CREATED);
2935                         if (create_error) {
2936                                 error = create_error;
2937                                 goto out;
2938                         }
2939                 } else {
2940                         if (excl && !(*opened & FILE_CREATED)) {
2941                                 error = -EEXIST;
2942                                 goto out;
2943                         }
2944                 }
2945                 goto looked_up;
2946         }
2947 
2948         /*
2949          * We didn't have the inode before the open, so check open permission
2950          * here.
2951          */
2952         acc_mode = op->acc_mode;
2953         if (*opened & FILE_CREATED) {
2954                 WARN_ON(!(open_flag & O_CREAT));
2955                 fsnotify_create(dir, dentry);
2956                 acc_mode = 0;
2957         }
2958         error = may_open(&file->f_path, acc_mode, open_flag);
2959         if (error)
2960                 fput(file);
2961 
2962 out:
2963         dput(dentry);
2964         return error;
2965 
2966 no_open:
2967         if (need_lookup) {
2968                 dentry = lookup_real(dir, dentry, nd->flags);
2969                 if (IS_ERR(dentry))
2970                         return PTR_ERR(dentry);
2971         }
2972         if (create_error && !dentry->d_inode) {
2973                 error = create_error;
2974                 goto out;
2975         }
2976 looked_up:
2977         path->dentry = dentry;
2978         path->mnt = nd->path.mnt;
2979         return 1;
2980 }
2981 
2982 /*
2983  * Look up and maybe create and open the last component.
2984  *
2985  * Must be called with i_mutex held on parent.
2986  *
2987  * Returns 0 if the file was successfully atomically created (if necessary) and
2988  * opened.  In this case the file will be returned attached to @file.
2989  *
2990  * Returns 1 if the file was not completely opened at this time, though lookups
2991  * and creations will have been performed and the dentry returned in @path will
2992  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2993  * specified then a negative dentry may be returned.
2994  *
2995  * An error code is returned otherwise.
2996  *
2997  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2998  * cleared otherwise prior to returning.
2999  */
3000 static int lookup_open(struct nameidata *nd, struct path *path,
3001                         struct file *file,
3002                         const struct open_flags *op,
3003                         bool got_write, int *opened)
3004 {
3005         struct dentry *dir = nd->path.dentry;
3006         struct inode *dir_inode = dir->d_inode;
3007         struct dentry *dentry;
3008         int error;
3009         bool need_lookup;
3010 
3011         *opened &= ~FILE_CREATED;
3012         dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
3013         if (IS_ERR(dentry))
3014                 return PTR_ERR(dentry);
3015 
3016         /* Cached positive dentry: will open in f_op->open */
3017         if (!need_lookup && dentry->d_inode)
3018                 goto out_no_open;
3019 
3020         if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
3021                 return atomic_open(nd, dentry, path, file, op, got_write,
3022                                    need_lookup, opened);
3023         }
3024 
3025         if (need_lookup) {
3026                 BUG_ON(dentry->d_inode);
3027 
3028                 dentry = lookup_real(dir_inode, dentry, nd->flags);
3029                 if (IS_ERR(dentry))
3030                         return PTR_ERR(dentry);
3031         }
3032 
3033         /* Negative dentry, just create the file */
3034         if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3035                 umode_t mode = op->mode;
3036                 if (!IS_POSIXACL(dir->d_inode))
3037                         mode &= ~current_umask();
3038                 /*
3039                  * This write is needed to ensure that a
3040                  * rw->ro transition does not occur between
3041                  * the time when the file is created and when
3042                  * a permanent write count is taken through
3043                  * the 'struct file' in finish_open().
3044                  */
3045                 if (!got_write) {
3046                         error = -EROFS;
3047                         goto out_dput;
3048                 }
3049                 *opened |= FILE_CREATED;
3050                 error = security_path_mknod(&nd->path, dentry, mode, 0);
3051                 if (error)
3052                         goto out_dput;
3053                 error = vfs_create(dir->d_inode, dentry, mode,
3054                                    nd->flags & LOOKUP_EXCL);
3055                 if (error)
3056                         goto out_dput;
3057         }
3058 out_no_open:
3059         path->dentry = dentry;
3060         path->mnt = nd->path.mnt;
3061         return 1;
3062 
3063 out_dput:
3064         dput(dentry);
3065         return error;
3066 }
3067 
3068 /*
3069  * Handle the last step of open()
3070  */
3071 static int do_last(struct nameidata *nd,
3072                    struct file *file, const struct open_flags *op,
3073                    int *opened)
3074 {
3075         struct dentry *dir = nd->path.dentry;
3076         int open_flag = op->open_flag;
3077         bool will_truncate = (open_flag & O_TRUNC) != 0;
3078         bool got_write = false;
3079         int acc_mode = op->acc_mode;
3080         unsigned seq;
3081         struct inode *inode;
3082         struct path save_parent = { .dentry = NULL, .mnt = NULL };
3083         struct path path;
3084         bool retried = false;
3085         int error;
3086 
3087         nd->flags &= ~LOOKUP_PARENT;
3088         nd->flags |= op->intent;
3089 
3090         if (nd->last_type != LAST_NORM) {
3091                 error = handle_dots(nd, nd->last_type);
3092                 if (unlikely(error))
3093                         return error;
3094                 goto finish_open;
3095         }
3096 
3097         if (!(open_flag & O_CREAT)) {
3098                 if (nd->last.name[nd->last.len])
3099                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3100                 /* we _can_ be in RCU mode here */
3101                 error = lookup_fast(nd, &path, &inode, &seq);
3102                 if (likely(!error))
3103                         goto finish_lookup;
3104 
3105                 if (error < 0)
3106                         return error;
3107 
3108                 BUG_ON(nd->inode != dir->d_inode);
3109         } else {
3110                 /* create side of things */
3111                 /*
3112                  * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3113                  * has been cleared when we got to the last component we are
3114                  * about to look up
3115                  */
3116                 error = complete_walk(nd);
3117                 if (error)
3118                         return error;
3119 
3120                 audit_inode(nd->name, dir, LOOKUP_PARENT);
3121                 /* trailing slashes? */
3122                 if (unlikely(nd->last.name[nd->last.len]))
3123                         return -EISDIR;
3124         }
3125 
3126 retry_lookup:
3127         if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3128                 error = mnt_want_write(nd->path.mnt);
3129                 if (!error)
3130                         got_write = true;
3131                 /*
3132                  * do _not_ fail yet - we might not need that or fail with
3133                  * a different error; let lookup_open() decide; we'll be
3134                  * dropping this one anyway.
3135                  */
3136         }
3137         inode_lock(dir->d_inode);
3138         error = lookup_open(nd, &path, file, op, got_write, opened);
3139         inode_unlock(dir->d_inode);
3140 
3141         if (error <= 0) {
3142                 if (error)
3143                         goto out;
3144 
3145                 if ((*opened & FILE_CREATED) ||
3146                     !S_ISREG(file_inode(file)->i_mode))
3147                         will_truncate = false;
3148 
3149                 audit_inode(nd->name, file->f_path.dentry, 0);
3150                 goto opened;
3151         }
3152 
3153         if (*opened & FILE_CREATED) {
3154                 /* Don't check for write permission, don't truncate */
3155                 open_flag &= ~O_TRUNC;
3156                 will_truncate = false;
3157                 acc_mode = 0;
3158                 path_to_nameidata(&path, nd);
3159                 goto finish_open_created;
3160         }
3161 
3162         /*
3163          * create/update audit record if it already exists.
3164          */
3165         if (d_is_positive(path.dentry))
3166                 audit_inode(nd->name, path.dentry, 0);
3167 
3168         /*
3169          * If atomic_open() acquired write access it is dropped now due to
3170          * possible mount and symlink following (this might be optimized away if
3171          * necessary...)
3172          */
3173         if (got_write) {
3174                 mnt_drop_write(nd->path.mnt);
3175                 got_write = false;
3176         }
3177 
3178         if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3179                 path_to_nameidata(&path, nd);
3180                 return -EEXIST;
3181         }
3182 
3183         error = follow_managed(&path, nd);
3184         if (unlikely(error < 0))
3185                 return error;
3186 
3187         BUG_ON(nd->flags & LOOKUP_RCU);
3188         seq = 0;        /* out of RCU mode, so the value doesn't matter */
3189         if (unlikely(d_is_negative(path.dentry))) {
3190                 path_to_nameidata(&path, nd);
3191                 return -ENOENT;
3192         }
3193         inode = d_backing_inode(path.dentry);
3194 finish_lookup:
3195         if (nd->depth)
3196                 put_link(nd);
3197         error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3198                                    inode, seq);
3199         if (unlikely(error))
3200                 return error;
3201 
3202         if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3203                 path_to_nameidata(&path, nd);
3204         } else {
3205                 save_parent.dentry = nd->path.dentry;
3206                 save_parent.mnt = mntget(path.mnt);
3207                 nd->path.dentry = path.dentry;
3208 
3209         }
3210         nd->inode = inode;
3211         nd->seq = seq;
3212         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3213 finish_open:
3214         error = complete_walk(nd);
3215         if (error) {
3216                 path_put(&save_parent);
3217                 return error;
3218         }
3219         audit_inode(nd->name, nd->path.dentry, 0);
3220         if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) {
3221                 error = -ELOOP;
3222                 goto out;
3223         }
3224         error = -EISDIR;
3225         if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3226                 goto out;
3227         error = -ENOTDIR;
3228         if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3229                 goto out;
3230         if (!d_is_reg(nd->path.dentry))
3231                 will_truncate = false;
3232 
3233         if (will_truncate) {
3234                 error = mnt_want_write(nd->path.mnt);
3235                 if (error)
3236                         goto out;
3237                 got_write = true;
3238         }
3239 finish_open_created:
3240         if (likely(!(open_flag & O_PATH))) {
3241                 error = may_open(&nd->path, acc_mode, open_flag);
3242                 if (error)
3243                         goto out;
3244         }
3245         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3246         error = vfs_open(&nd->path, file, current_cred());
3247         if (!error) {
3248                 *opened |= FILE_OPENED;
3249         } else {
3250                 if (error == -EOPENSTALE)
3251                         goto stale_open;
3252                 goto out;
3253         }
3254 opened:
3255         error = open_check_o_direct(file);
3256         if (error)
3257                 goto exit_fput;
3258         error = ima_file_check(file, op->acc_mode, *opened);
3259         if (error)
3260                 goto exit_fput;
3261 
3262         if (will_truncate) {
3263                 error = handle_truncate(file);
3264                 if (error)
3265                         goto exit_fput;
3266         }
3267 out:
3268         if (unlikely(error > 0)) {
3269                 WARN_ON(1);
3270                 error = -EINVAL;
3271         }
3272         if (got_write)
3273                 mnt_drop_write(nd->path.mnt);
3274         path_put(&save_parent);
3275         return error;
3276 
3277 exit_fput:
3278         fput(file);
3279         goto out;
3280 
3281 stale_open:
3282         /* If no saved parent or already retried then can't retry */
3283         if (!save_parent.dentry || retried)
3284                 goto out;
3285 
3286         BUG_ON(save_parent.dentry != dir);
3287         path_put(&nd->path);
3288         nd->path = save_parent;
3289         nd->inode = dir->d_inode;
3290         save_parent.mnt = NULL;
3291         save_parent.dentry = NULL;
3292         if (got_write) {
3293                 mnt_drop_write(nd->path.mnt);
3294                 got_write = false;
3295         }
3296         retried = true;
3297         goto retry_lookup;
3298 }
3299 
3300 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3301                 const struct open_flags *op,
3302                 struct file *file, int *opened)
3303 {
3304         static const struct qstr name = QSTR_INIT("/", 1);
3305         struct dentry *child;
3306         struct inode *dir;
3307         struct path path;
3308         int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3309         if (unlikely(error))
3310                 return error;
3311         error = mnt_want_write(path.mnt);
3312         if (unlikely(error))
3313                 goto out;
3314         dir = path.dentry->d_inode;
3315         /* we want directory to be writable */
3316         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3317         if (error)
3318                 goto out2;
3319         if (!dir->i_op->tmpfile) {
3320                 error = -EOPNOTSUPP;
3321                 goto out2;
3322         }
3323         child = d_alloc(path.dentry, &name);
3324         if (unlikely(!child)) {
3325                 error = -ENOMEM;
3326                 goto out2;
3327         }
3328         dput(path.dentry);
3329         path.dentry = child;
3330         error = dir->i_op->tmpfile(dir, child, op->mode);
3331         if (error)
3332                 goto out2;
3333         audit_inode(nd->name, child, 0);
3334         /* Don't check for other permissions, the inode was just created */
3335         error = may_open(&path, 0, op->open_flag);
3336         if (error)
3337                 goto out2;
3338         file->f_path.mnt = path.mnt;
3339         error = finish_open(file, child, NULL, opened);
3340         if (error)
3341                 goto out2;
3342         error = open_check_o_direct(file);
3343         if (error) {
3344                 fput(file);
3345         } else if (!(op->open_flag & O_EXCL)) {
3346                 struct inode *inode = file_inode(file);
3347                 spin_lock(&inode->i_lock);
3348                 inode->i_state |= I_LINKABLE;
3349                 spin_unlock(&inode->i_lock);
3350         }
3351 out2:
3352         mnt_drop_write(path.mnt);
3353 out:
3354         path_put(&path);
3355         return error;
3356 }
3357 
3358 static struct file *path_openat(struct nameidata *nd,
3359                         const struct open_flags *op, unsigned flags)
3360 {
3361         const char *s;
3362         struct file *file;
3363         int opened = 0;
3364         int error;
3365 
3366         file = get_empty_filp();
3367         if (IS_ERR(file))
3368                 return file;
3369 
3370         file->f_flags = op->open_flag;
3371 
3372         if (unlikely(file->f_flags & __O_TMPFILE)) {
3373                 error = do_tmpfile(nd, flags, op, file, &opened);
3374                 goto out2;
3375         }
3376 
3377         s = path_init(nd, flags);
3378         if (IS_ERR(s)) {
3379                 put_filp(file);
3380                 return ERR_CAST(s);
3381         }
3382         while (!(error = link_path_walk(s, nd)) &&
3383                 (error = do_last(nd, file, op, &opened)) > 0) {
3384                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3385                 s = trailing_symlink(nd);
3386                 if (IS_ERR(s)) {
3387                         error = PTR_ERR(s);
3388                         break;
3389                 }
3390         }
3391         terminate_walk(nd);
3392 out2:
3393         if (!(opened & FILE_OPENED)) {
3394                 BUG_ON(!error);
3395                 put_filp(file);
3396         }
3397         if (unlikely(error)) {
3398                 if (error == -EOPENSTALE) {
3399                         if (flags & LOOKUP_RCU)
3400                                 error = -ECHILD;
3401                         else
3402                                 error = -ESTALE;
3403                 }
3404                 file = ERR_PTR(error);
3405         }
3406         return file;
3407 }
3408 
3409 struct file *do_filp_open(int dfd, struct filename *pathname,
3410                 const struct open_flags *op)
3411 {
3412         struct nameidata nd;
3413         int flags = op->lookup_flags;
3414         struct file *filp;
3415 
3416         set_nameidata(&nd, dfd, pathname);
3417         filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3418         if (unlikely(filp == ERR_PTR(-ECHILD)))
3419                 filp = path_openat(&nd, op, flags);
3420         if (unlikely(filp == ERR_PTR(-ESTALE)))
3421                 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3422         restore_nameidata();
3423         return filp;
3424 }
3425 
3426 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3427                 const char *name, const struct open_flags *op)
3428 {
3429         struct nameidata nd;
3430         struct file *file;
3431         struct filename *filename;
3432         int flags = op->lookup_flags | LOOKUP_ROOT;
3433 
3434         nd.root.mnt = mnt;
3435         nd.root.dentry = dentry;
3436 
3437         if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3438                 return ERR_PTR(-ELOOP);
3439 
3440         filename = getname_kernel(name);
3441         if (IS_ERR(filename))
3442                 return ERR_CAST(filename);
3443 
3444         set_nameidata(&nd, -1, filename);
3445         file = path_openat(&nd, op, flags | LOOKUP_RCU);
3446         if (unlikely(file == ERR_PTR(-ECHILD)))
3447                 file = path_openat(&nd, op, flags);
3448         if (unlikely(file == ERR_PTR(-ESTALE)))
3449                 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3450         restore_nameidata();
3451         putname(filename);
3452         return file;
3453 }
3454 
3455 static struct dentry *filename_create(int dfd, struct filename *name,
3456                                 struct path *path, unsigned int lookup_flags)
3457 {
3458         struct dentry *dentry = ERR_PTR(-EEXIST);
3459         struct qstr last;
3460         int type;
3461         int err2;
3462         int error;
3463         bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3464 
3465         /*
3466          * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3467          * other flags passed in are ignored!
3468          */
3469         lookup_flags &= LOOKUP_REVAL;
3470 
3471         name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3472         if (IS_ERR(name))
3473                 return ERR_CAST(name);
3474 
3475         /*
3476          * Yucky last component or no last component at all?
3477          * (foo/., foo/.., /////)
3478          */
3479         if (unlikely(type != LAST_NORM))
3480                 goto out;
3481 
3482         /* don't fail immediately if it's r/o, at least try to report other errors */
3483         err2 = mnt_want_write(path->mnt);
3484         /*
3485          * Do the final lookup.
3486          */
3487         lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3488         inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3489         dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3490         if (IS_ERR(dentry))
3491                 goto unlock;
3492 
3493         error = -EEXIST;
3494         if (d_is_positive(dentry))
3495                 goto fail;
3496 
3497         /*
3498          * Special case - lookup gave negative, but... we had foo/bar/
3499          * From the vfs_mknod() POV we just have a negative dentry -
3500          * all is fine. Let's be bastards - you had / on the end, you've
3501          * been asking for (non-existent) directory. -ENOENT for you.
3502          */
3503         if (unlikely(!is_dir && last.name[last.len])) {
3504                 error = -ENOENT;
3505                 goto fail;
3506         }
3507         if (unlikely(err2)) {
3508                 error = err2;
3509                 goto fail;
3510         }
3511         putname(name);
3512         return dentry;
3513 fail:
3514         dput(dentry);
3515         dentry = ERR_PTR(error);
3516 unlock:
3517         inode_unlock(path->dentry->d_inode);
3518         if (!err2)
3519                 mnt_drop_write(path->mnt);
3520 out:
3521         path_put(path);
3522         putname(name);
3523         return dentry;
3524 }
3525 
3526 struct dentry *kern_path_create(int dfd, const char *pathname,
3527                                 struct path *path, unsigned int lookup_flags)
3528 {
3529         return filename_create(dfd, getname_kernel(pathname),
3530                                 path, lookup_flags);
3531 }
3532 EXPORT_SYMBOL(kern_path_create);
3533 
3534 void done_path_create(struct path *path, struct dentry *dentry)
3535 {
3536         dput(dentry);
3537         inode_unlock(path->dentry->d_inode);
3538         mnt_drop_write(path->mnt);
3539         path_put(path);
3540 }
3541 EXPORT_SYMBOL(done_path_create);
3542 
3543 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3544                                 struct path *path, unsigned int lookup_flags)
3545 {
3546         return filename_create(dfd, getname(pathname), path, lookup_flags);
3547 }
3548 EXPORT_SYMBOL(user_path_create);
3549 
3550 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3551 {
3552         int error = may_create(dir, dentry);
3553 
3554         if (error)
3555                 return error;
3556 
3557         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3558                 return -EPERM;
3559 
3560         if (!dir->i_op->mknod)
3561                 return -EPERM;
3562 
3563         error = devcgroup_inode_mknod(mode, dev);
3564         if (error)
3565                 return error;
3566 
3567         error = security_inode_mknod(dir, dentry, mode, dev);
3568         if (error)
3569                 return error;
3570 
3571         error = dir->i_op->mknod(dir, dentry, mode, dev);
3572         if (!error)
3573                 fsnotify_create(dir, dentry);
3574         return error;
3575 }
3576 EXPORT_SYMBOL(vfs_mknod);
3577 
3578 static int may_mknod(umode_t mode)
3579 {
3580         switch (mode & S_IFMT) {
3581         case S_IFREG:
3582         case S_IFCHR:
3583         case S_IFBLK:
3584         case S_IFIFO:
3585         case S_IFSOCK:
3586         case 0: /* zero mode translates to S_IFREG */
3587                 return 0;
3588         case S_IFDIR:
3589                 return -EPERM;
3590         default:
3591                 return -EINVAL;
3592         }
3593 }
3594 
3595 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3596                 unsigned, dev)
3597 {
3598         struct dentry *dentry;
3599         struct path path;
3600         int error;
3601         unsigned int lookup_flags = 0;
3602 
3603         error = may_mknod(mode);
3604         if (error)
3605                 return error;
3606 retry:
3607         dentry = user_path_create(dfd, filename, &path, lookup_flags);
3608         if (IS_ERR(dentry))
3609                 return PTR_ERR(dentry);
3610 
3611         if (!IS_POSIXACL(path.dentry->d_inode))
3612                 mode &= ~current_umask();
3613         error = security_path_mknod(&path, dentry, mode, dev);
3614         if (error)
3615                 goto out;
3616         switch (mode & S_IFMT) {
3617                 case 0: case S_IFREG:
3618                         error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3619                         break;
3620                 case S_IFCHR: case S_IFBLK:
3621                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3622                                         new_decode_dev(dev));
3623                         break;
3624                 case S_IFIFO: case S_IFSOCK:
3625                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3626                         break;
3627         }
3628 out:
3629         done_path_create(&path, dentry);
3630         if (retry_estale(error, lookup_flags)) {
3631                 lookup_flags |= LOOKUP_REVAL;
3632                 goto retry;
3633         }
3634         return error;
3635 }
3636 
3637 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3638 {
3639         return sys_mknodat(AT_FDCWD, filename, mode, dev);
3640 }
3641 
3642 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3643 {
3644         int error = may_create(dir, dentry);
3645         unsigned max_links = dir->i_sb->s_max_links;
3646 
3647         if (error)
3648                 return error;
3649 
3650         if (!dir->i_op->mkdir)
3651                 return -EPERM;
3652 
3653         mode &= (S_IRWXUGO|S_ISVTX);
3654         error = security_inode_mkdir(dir, dentry, mode);
3655         if (error)
3656                 return error;
3657 
3658         if (max_links && dir->i_nlink >= max_links)
3659                 return -EMLINK;
3660 
3661         error = dir->i_op->mkdir(dir, dentry, mode);
3662         if (!error)
3663                 fsnotify_mkdir(dir, dentry);
3664         return error;
3665 }
3666 EXPORT_SYMBOL(vfs_mkdir);
3667 
3668 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3669 {
3670         struct dentry *dentry;
3671         struct path path;
3672         int error;
3673         unsigned int lookup_flags = LOOKUP_DIRECTORY;
3674 
3675 retry:
3676         dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3677         if (IS_ERR(dentry))
3678                 return PTR_ERR(dentry);
3679 
3680         if (!IS_POSIXACL(path.dentry->d_inode))
3681                 mode &= ~current_umask();
3682         error = security_path_mkdir(&path, dentry, mode);
3683         if (!error)
3684                 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3685         done_path_create(&path, dentry);
3686         if (retry_estale(error, lookup_flags)) {
3687                 lookup_flags |= LOOKUP_REVAL;
3688                 goto retry;
3689         }
3690         return error;
3691 }
3692 
3693 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3694 {
3695         return sys_mkdirat(AT_FDCWD, pathname, mode);
3696 }
3697 
3698 /*
3699  * The dentry_unhash() helper will try to drop the dentry early: we
3700  * should have a usage count of 1 if we're the only user of this
3701  * dentry, and if that is true (possibly after pruning the dcache),
3702  * then we drop the dentry now.
3703  *
3704  * A low-level filesystem can, if it choses, legally
3705  * do a
3706  *
3707  *      if (!d_unhashed(dentry))
3708  *              return -EBUSY;
3709  *
3710  * if it cannot handle the case of removing a directory
3711  * that is still in use by something else..
3712  */
3713 void dentry_unhash(struct dentry *dentry)
3714 {
3715         shrink_dcache_parent(dentry);
3716         spin_lock(&dentry->d_lock);
3717         if (dentry->d_lockref.count == 1)
3718                 __d_drop(dentry);
3719         spin_unlock(&dentry->d_lock);
3720 }
3721 EXPORT_SYMBOL(dentry_unhash);
3722 
3723 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3724 {
3725         int error = may_delete(dir, dentry, 1);
3726 
3727         if (error)
3728                 return error;
3729 
3730         if (!dir->i_op->rmdir)
3731                 return -EPERM;
3732 
3733         dget(dentry);
3734         inode_lock(dentry->d_inode);
3735 
3736         error = -EBUSY;
3737         if (is_local_mountpoint(dentry))
3738                 goto out;
3739 
3740         error = security_inode_rmdir(dir, dentry);
3741         if (error)
3742                 goto out;
3743 
3744         shrink_dcache_parent(dentry);
3745         error = dir->i_op->rmdir(dir, dentry);
3746         if (error)
3747                 goto out;
3748 
3749         dentry->d_inode->i_flags |= S_DEAD;
3750         dont_mount(dentry);
3751         detach_mounts(dentry);
3752 
3753 out:
3754         inode_unlock(dentry->d_inode);
3755         dput(dentry);
3756         if (!error)
3757                 d_delete(dentry);
3758         return error;
3759 }
3760 EXPORT_SYMBOL(vfs_rmdir);
3761 
3762 static long do_rmdir(int dfd, const char __user *pathname)
3763 {
3764         int error = 0;
3765         struct filename *name;
3766         struct dentry *dentry;
3767         struct path path;
3768         struct qstr last;
3769         int type;
3770         unsigned int lookup_flags = 0;
3771 retry:
3772         name = user_path_parent(dfd, pathname,
3773                                 &path, &last, &type, lookup_flags);
3774         if (IS_ERR(name))
3775                 return PTR_ERR(name);
3776 
3777         switch (type) {
3778         case LAST_DOTDOT:
3779                 error = -ENOTEMPTY;
3780                 goto exit1;
3781         case LAST_DOT:
3782                 error = -EINVAL;
3783                 goto exit1;
3784         case LAST_ROOT:
3785                 error = -EBUSY;
3786                 goto exit1;
3787         }
3788 
3789         error = mnt_want_write(path.mnt);
3790         if (error)
3791                 goto exit1;
3792 
3793         inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3794         dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3795         error = PTR_ERR(dentry);
3796         if (IS_ERR(dentry))
3797                 goto exit2;
3798         if (!dentry->d_inode) {
3799                 error = -ENOENT;
3800                 goto exit3;
3801         }
3802         error = security_path_rmdir(&path, dentry);
3803         if (error)
3804                 goto exit3;
3805         error = vfs_rmdir(path.dentry->d_inode, dentry);
3806 exit3:
3807         dput(dentry);
3808 exit2:
3809         inode_unlock(path.dentry->d_inode);
3810         mnt_drop_write(path.mnt);
3811 exit1:
3812         path_put(&path);
3813         putname(name);
3814         if (retry_estale(error, lookup_flags)) {
3815                 lookup_flags |= LOOKUP_REVAL;
3816                 goto retry;
3817         }
3818         return error;
3819 }
3820 
3821 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3822 {
3823         return do_rmdir(AT_FDCWD, pathname);
3824 }
3825 
3826 /**
3827  * vfs_unlink - unlink a filesystem object
3828  * @dir:        parent directory
3829  * @dentry:     victim
3830  * @delegated_inode: returns victim inode, if the inode is delegated.
3831  *
3832  * The caller must hold dir->i_mutex.
3833  *
3834  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3835  * return a reference to the inode in delegated_inode.  The caller
3836  * should then break the delegation on that inode and retry.  Because
3837  * breaking a delegation may take a long time, the caller should drop
3838  * dir->i_mutex before doing so.
3839  *
3840  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3841  * be appropriate for callers that expect the underlying filesystem not
3842  * to be NFS exported.
3843  */
3844 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3845 {
3846         struct inode *target = dentry->d_inode;
3847         int error = may_delete(dir, dentry, 0);
3848 
3849         if (error)
3850                 return error;
3851 
3852         if (!dir->i_op->unlink)
3853                 return -EPERM;
3854 
3855         inode_lock(target);
3856         if (is_local_mountpoint(dentry))
3857                 error = -EBUSY;
3858         else {
3859                 error = security_inode_unlink(dir, dentry);
3860                 if (!error) {
3861                         error = try_break_deleg(target, delegated_inode);
3862                         if (error)
3863                                 goto out;
3864                         error = dir->i_op->unlink(dir, dentry);
3865                         if (!error) {
3866                                 dont_mount(dentry);
3867                                 detach_mounts(dentry);
3868                         }
3869                 }
3870         }
3871 out:
3872         inode_unlock(target);
3873 
3874         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3875         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3876                 fsnotify_link_count(target);
3877                 d_delete(dentry);
3878         }
3879 
3880         return error;
3881 }
3882 EXPORT_SYMBOL(vfs_unlink);
3883 
3884 /*
3885  * Make sure that the actual truncation of the file will occur outside its
3886  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3887  * writeout happening, and we don't want to prevent access to the directory
3888  * while waiting on the I/O.
3889  */
3890 static long do_unlinkat(int dfd, const char __user *pathname)
3891 {
3892         int error;
3893         struct filename *name;
3894         struct dentry *dentry;
3895         struct path path;
3896         struct qstr last;
3897         int type;
3898         struct inode *inode = NULL;
3899         struct inode *delegated_inode = NULL;
3900         unsigned int lookup_flags = 0;
3901 retry:
3902         name = user_path_parent(dfd, pathname,
3903                                 &path, &last, &type, lookup_flags);
3904         if (IS_ERR(name))
3905                 return PTR_ERR(name);
3906 
3907         error = -EISDIR;
3908         if (type != LAST_NORM)
3909                 goto exit1;
3910 
3911         error = mnt_want_write(path.mnt);
3912         if (error)
3913                 goto exit1;
3914 retry_deleg:
3915         inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3916         dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3917         error = PTR_ERR(dentry);
3918         if (!IS_ERR(dentry)) {
3919                 /* Why not before? Because we want correct error value */
3920                 if (last.name[last.len])
3921                         goto slashes;
3922                 inode = dentry->d_inode;
3923                 if (d_is_negative(dentry))
3924                         goto slashes;
3925                 ihold(inode);
3926                 error = security_path_unlink(&path, dentry);
3927                 if (error)
3928                         goto exit2;
3929                 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3930 exit2:
3931                 dput(dentry);
3932         }
3933         inode_unlock(path.dentry->d_inode);
3934         if (inode)
3935                 iput(inode);    /* truncate the inode here */
3936         inode = NULL;
3937         if (delegated_inode) {
3938                 error = break_deleg_wait(&delegated_inode);
3939                 if (!error)
3940                         goto retry_deleg;
3941         }
3942         mnt_drop_write(path.mnt);
3943 exit1:
3944         path_put(&path);
3945         putname(name);
3946         if (retry_estale(error, lookup_flags)) {
3947                 lookup_flags |= LOOKUP_REVAL;
3948                 inode = NULL;
3949                 goto retry;
3950         }
3951         return error;
3952 
3953 slashes:
3954         if (d_is_negative(dentry))
3955                 error = -ENOENT;
3956         else if (d_is_dir(dentry))
3957                 error = -EISDIR;
3958         else
3959                 error = -ENOTDIR;
3960         goto exit2;
3961 }
3962 
3963 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3964 {
3965         if ((flag & ~AT_REMOVEDIR) != 0)
3966                 return -EINVAL;
3967 
3968         if (flag & AT_REMOVEDIR)
3969                 return do_rmdir(dfd, pathname);
3970 
3971         return do_unlinkat(dfd, pathname);
3972 }
3973 
3974 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3975 {
3976         return do_unlinkat(AT_FDCWD, pathname);
3977 }
3978 
3979 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3980 {
3981         int error = may_create(dir, dentry);
3982 
3983         if (error)
3984                 return error;
3985 
3986         if (!dir->i_op->symlink)
3987                 return -EPERM;
3988 
3989         error = security_inode_symlink(dir, dentry, oldname);
3990         if (error)
3991                 return error;
3992 
3993         error = dir->i_op->symlink(dir, dentry, oldname);
3994         if (!error)
3995                 fsnotify_create(dir, dentry);
3996         return error;
3997 }
3998 EXPORT_SYMBOL(vfs_symlink);
3999 
4000 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4001                 int, newdfd, const char __user *, newname)
4002 {
4003         int error;
4004         struct filename *from;
4005         struct dentry *dentry;
4006         struct path path;
4007         unsigned int lookup_flags = 0;
4008 
4009         from = getname(oldname);
4010         if (IS_ERR(from))
4011                 return PTR_ERR(from);
4012 retry:
4013         dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4014         error = PTR_ERR(dentry);
4015         if (IS_ERR(dentry))
4016                 goto out_putname;
4017 
4018         error = security_path_symlink(&path, dentry, from->name);
4019         if (!error)
4020                 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4021         done_path_create(&path, dentry);
4022         if (retry_estale(error, lookup_flags)) {
4023                 lookup_flags |= LOOKUP_REVAL;
4024                 goto retry;
4025         }
4026 out_putname:
4027         putname(from);
4028         return error;
4029 }
4030 
4031 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4032 {
4033         return sys_symlinkat(oldname, AT_FDCWD, newname);
4034 }
4035 
4036 /**
4037  * vfs_link - create a new link
4038  * @old_dentry: object to be linked
4039  * @dir:        new parent
4040  * @new_dentry: where to create the new link
4041  * @delegated_inode: returns inode needing a delegation break
4042  *
4043  * The caller must hold dir->i_mutex
4044  *
4045  * If vfs_link discovers a delegation on the to-be-linked file in need
4046  * of breaking, it will return -EWOULDBLOCK and return a reference to the
4047  * inode in delegated_inode.  The caller should then break the delegation
4048  * and retry.  Because breaking a delegation may take a long time, the
4049  * caller should drop the i_mutex before doing so.
4050  *
4051  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4052  * be appropriate for callers that expect the underlying filesystem not
4053  * to be NFS exported.
4054  */
4055 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4056 {
4057         struct inode *inode = old_dentry->d_inode;
4058         unsigned max_links = dir->i_sb->s_max_links;
4059         int error;
4060 
4061         if (!inode)
4062                 return -ENOENT;
4063 
4064         error = may_create(dir, new_dentry);
4065         if (error)
4066                 return error;
4067 
4068         if (dir->i_sb != inode->i_sb)
4069                 return -EXDEV;
4070 
4071         /*
4072          * A link to an append-only or immutable file cannot be created.
4073          */
4074         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4075                 return -EPERM;
4076         if (!dir->i_op->link)
4077                 return -EPERM;
4078         if (S_ISDIR(inode->i_mode))
4079                 return -EPERM;
4080 
4081         error = security_inode_link(old_dentry, dir, new_dentry);
4082         if (error)
4083                 return error;
4084 
4085         inode_lock(inode);
4086         /* Make sure we don't allow creating hardlink to an unlinked file */
4087         if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4088                 error =  -ENOENT;
4089         else if (max_links && inode->i_nlink >= max_links)
4090                 error = -EMLINK;
4091         else {
4092                 error = try_break_deleg(inode, delegated_inode);
4093                 if (!error)
4094                         error = dir->i_op->link(old_dentry, dir, new_dentry);
4095         }
4096 
4097         if (!error && (inode->i_state & I_LINKABLE)) {
4098                 spin_lock(&inode->i_lock);
4099                 inode->i_state &= ~I_LINKABLE;
4100                 spin_unlock(&inode->i_lock);
4101         }
4102         inode_unlock(inode);
4103         if (!error)
4104                 fsnotify_link(dir, inode, new_dentry);
4105         return error;
4106 }
4107 EXPORT_SYMBOL(vfs_link);
4108 
4109 /*
4110  * Hardlinks are often used in delicate situations.  We avoid
4111  * security-related surprises by not following symlinks on the
4112  * newname.  --KAB
4113  *
4114  * We don't follow them on the oldname either to be compatible
4115  * with linux 2.0, and to avoid hard-linking to directories
4116  * and other special files.  --ADM
4117  */
4118 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4119                 int, newdfd, const char __user *, newname, int, flags)
4120 {
4121         struct dentry *new_dentry;
4122         struct path old_path, new_path;
4123         struct inode *delegated_inode = NULL;
4124         int how = 0;
4125         int error;
4126 
4127         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4128                 return -EINVAL;
4129         /*
4130          * To use null names we require CAP_DAC_READ_SEARCH
4131          * This ensures that not everyone will be able to create
4132          * handlink using the passed filedescriptor.
4133          */
4134         if (flags & AT_EMPTY_PATH) {
4135                 if (!capable(CAP_DAC_READ_SEARCH))
4136                         return -ENOENT;
4137                 how = LOOKUP_EMPTY;
4138         }
4139 
4140         if (flags & AT_SYMLINK_FOLLOW)
4141                 how |= LOOKUP_FOLLOW;
4142 retry:
4143         error = user_path_at(olddfd, oldname, how, &old_path);
4144         if (error)
4145                 return error;
4146 
4147         new_dentry = user_path_create(newdfd, newname, &new_path,
4148                                         (how & LOOKUP_REVAL));
4149         error = PTR_ERR(new_dentry);
4150         if (IS_ERR(new_dentry))
4151                 goto out;
4152 
4153         error = -EXDEV;
4154         if (old_path.mnt != new_path.mnt)
4155                 goto out_dput;
4156         error = may_linkat(&old_path);
4157         if (unlikely(error))
4158                 goto out_dput;
4159         error = security_path_link(old_path.dentry, &new_path, new_dentry);
4160         if (error)
4161                 goto out_dput;
4162         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4163 out_dput:
4164         done_path_create(&new_path, new_dentry);
4165         if (delegated_inode) {
4166                 error = break_deleg_wait(&delegated_inode);
4167                 if (!error) {
4168                         path_put(&old_path);
4169                         goto retry;
4170                 }
4171         }
4172         if (retry_estale(error, how)) {
4173                 path_put(&old_path);
4174                 how |= LOOKUP_REVAL;
4175                 goto retry;
4176         }
4177 out:
4178         path_put(&old_path);
4179 
4180         return error;
4181 }
4182 
4183 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4184 {
4185         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4186 }
4187 
4188 /**
4189  * vfs_rename - rename a filesystem object
4190  * @old_dir:    parent of source
4191  * @old_dentry: source
4192  * @new_dir:    parent of destination
4193  * @new_dentry: destination
4194  * @delegated_inode: returns an inode needing a delegation break
4195  * @flags:      rename flags
4196  *
4197  * The caller must hold multiple mutexes--see lock_rename()).
4198  *
4199  * If vfs_rename discovers a delegation in need of breaking at either
4200  * the source or destination, it will return -EWOULDBLOCK and return a
4201  * reference to the inode in delegated_inode.  The caller should then
4202  * break the delegation and retry.  Because breaking a delegation may
4203  * take a long time, the caller should drop all locks before doing
4204  * so.
4205  *
4206  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4207  * be appropriate for callers that expect the underlying filesystem not
4208  * to be NFS exported.
4209  *
4210  * The worst of all namespace operations - renaming directory. "Perverted"
4211  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4212  * Problems:
4213  *      a) we can get into loop creation.
4214  *      b) race potential - two innocent renames can create a loop together.
4215  *         That's where 4.4 screws up. Current fix: serialization on
4216  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4217  *         story.
4218  *      c) we have to lock _four_ objects - parents and victim (if it exists),
4219  *         and source (if it is not a directory).
4220  *         And that - after we got ->i_mutex on parents (until then we don't know
4221  *         whether the target exists).  Solution: try to be smart with locking
4222  *         order for inodes.  We rely on the fact that tree topology may change
4223  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4224  *         move will be locked.  Thus we can rank directories by the tree
4225  *         (ancestors first) and rank all non-directories after them.
4226  *         That works since everybody except rename does "lock parent, lookup,
4227  *         lock child" and rename is under ->s_vfs_rename_mutex.
4228  *         HOWEVER, it relies on the assumption that any object with ->lookup()
4229  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4230  *         we'd better make sure that there's no link(2) for them.
4231  *      d) conversion from fhandle to dentry may come in the wrong moment - when
4232  *         we are removing the target. Solution: we will have to grab ->i_mutex
4233  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4234  *         ->i_mutex on parents, which works but leads to some truly excessive
4235  *         locking].
4236  */
4237 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4238                struct inode *new_dir, struct dentry *new_dentry,
4239                struct inode **delegated_inode, unsigned int flags)
4240 {
4241         int error;
4242         bool is_dir = d_is_dir(old_dentry);
4243         const unsigned char *old_name;
4244         struct inode *source = old_dentry->d_inode;
4245         struct inode *target = new_dentry->d_inode;
4246         bool new_is_dir = false;
4247         unsigned max_links = new_dir->i_sb->s_max_links;
4248 
4249         /*
4250          * Check source == target.
4251          * On overlayfs need to look at underlying inodes.
4252          */
4253         if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0))
4254                 return 0;
4255 
4256         error = may_delete(old_dir, old_dentry, is_dir);
4257         if (error)
4258                 return error;
4259 
4260         if (!target) {
4261                 error = may_create(new_dir, new_dentry);
4262         } else {
4263                 new_is_dir = d_is_dir(new_dentry);
4264 
4265                 if (!(flags & RENAME_EXCHANGE))
4266                         error = may_delete(new_dir, new_dentry, is_dir);
4267                 else
4268                         error = may_delete(new_dir, new_dentry, new_is_dir);
4269         }
4270         if (error)
4271                 return error;
4272 
4273         if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4274                 return -EPERM;
4275 
4276         if (flags && !old_dir->i_op->rename2)
4277                 return -EINVAL;
4278 
4279         /*
4280          * If we are going to change the parent - check write permissions,
4281          * we'll need to flip '..'.
4282          */
4283         if (new_dir != old_dir) {
4284                 if (is_dir) {
4285                         error = inode_permission(source, MAY_WRITE);
4286                         if (error)
4287                                 return error;
4288                 }
4289                 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4290                         error = inode_permission(target, MAY_WRITE);
4291                         if (error)
4292                                 return error;
4293                 }
4294         }
4295 
4296         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4297                                       flags);
4298         if (error)
4299                 return error;
4300 
4301         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4302         dget(new_dentry);
4303         if (!is_dir || (flags & RENAME_EXCHANGE))
4304                 lock_two_nondirectories(source, target);
4305         else if (target)
4306                 inode_lock(target);
4307 
4308         error = -EBUSY;
4309         if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4310                 goto out;
4311 
4312         if (max_links && new_dir != old_dir) {
4313                 error = -EMLINK;
4314                 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4315                         goto out;
4316                 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4317                     old_dir->i_nlink >= max_links)
4318                         goto out;
4319         }
4320         if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4321                 shrink_dcache_parent(new_dentry);
4322         if (!is_dir) {
4323                 error = try_break_deleg(source, delegated_inode);
4324                 if (error)
4325                         goto out;
4326         }
4327         if (target && !new_is_dir) {
4328                 error = try_break_deleg(target, delegated_inode);
4329                 if (error)
4330                         goto out;
4331         }
4332         if (!old_dir->i_op->rename2) {
4333                 error = old_dir->i_op->rename(old_dir, old_dentry,
4334                                               new_dir, new_dentry);
4335         } else {
4336                 WARN_ON(old_dir->i_op->rename != NULL);
4337                 error = old_dir->i_op->rename2(old_dir, old_dentry,
4338                                                new_dir, new_dentry, flags);
4339         }
4340         if (error)
4341                 goto out;
4342 
4343         if (!(flags & RENAME_EXCHANGE) && target) {
4344                 if (is_dir)
4345                         target->i_flags |= S_DEAD;
4346                 dont_mount(new_dentry);
4347                 detach_mounts(new_dentry);
4348         }
4349         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4350                 if (!(flags & RENAME_EXCHANGE))
4351                         d_move(old_dentry, new_dentry);
4352                 else
4353                         d_exchange(old_dentry, new_dentry);
4354         }
4355 out:
4356         if (!is_dir || (flags & RENAME_EXCHANGE))
4357                 unlock_two_nondirectories(source, target);
4358         else if (target)
4359                 inode_unlock(target);
4360         dput(new_dentry);
4361         if (!error) {
4362                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4363                               !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4364                 if (flags & RENAME_EXCHANGE) {
4365                         fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4366                                       new_is_dir, NULL, new_dentry);
4367                 }
4368         }
4369         fsnotify_oldname_free(old_name);
4370 
4371         return error;
4372 }
4373 EXPORT_SYMBOL(vfs_rename);
4374 
4375 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4376                 int, newdfd, const char __user *, newname, unsigned int, flags)
4377 {
4378         struct dentry *old_dentry, *new_dentry;
4379         struct dentry *trap;
4380         struct path old_path, new_path;
4381         struct qstr old_last, new_last;
4382         int old_type, new_type;
4383         struct inode *delegated_inode = NULL;
4384         struct filename *from;
4385         struct filename *to;
4386         unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4387         bool should_retry = false;
4388         int error;
4389 
4390         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4391                 return -EINVAL;
4392 
4393         if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4394             (flags & RENAME_EXCHANGE))
4395                 return -EINVAL;
4396 
4397         if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4398                 return -EPERM;
4399 
4400         if (flags & RENAME_EXCHANGE)
4401                 target_flags = 0;
4402 
4403 retry:
4404         from = user_path_parent(olddfd, oldname,
4405                                 &old_path, &old_last, &old_type, lookup_flags);
4406         if (IS_ERR(from)) {
4407                 error = PTR_ERR(from);
4408                 goto exit;
4409         }
4410 
4411         to = user_path_parent(newdfd, newname,
4412                                 &new_path, &new_last, &new_type, lookup_flags);
4413         if (IS_ERR(to)) {
4414                 error = PTR_ERR(to);
4415                 goto exit1;
4416         }
4417 
4418         error = -EXDEV;
4419         if (old_path.mnt != new_path.mnt)
4420                 goto exit2;
4421 
4422         error = -EBUSY;
4423         if (old_type != LAST_NORM)
4424                 goto exit2;
4425 
4426         if (flags & RENAME_NOREPLACE)
4427                 error = -EEXIST;
4428         if (new_type != LAST_NORM)
4429                 goto exit2;
4430 
4431         error = mnt_want_write(old_path.mnt);
4432         if (error)
4433                 goto exit2;
4434 
4435 retry_deleg:
4436         trap = lock_rename(new_path.dentry, old_path.dentry);
4437 
4438         old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4439         error = PTR_ERR(old_dentry);
4440         if (IS_ERR(old_dentry))
4441                 goto exit3;
4442         /* source must exist */
4443         error = -ENOENT;
4444         if (d_is_negative(old_dentry))
4445                 goto exit4;
4446         new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4447         error = PTR_ERR(new_dentry);
4448         if (IS_ERR(new_dentry))
4449                 goto exit4;
4450         error = -EEXIST;
4451         if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4452                 goto exit5;
4453         if (flags & RENAME_EXCHANGE) {
4454                 error = -ENOENT;
4455                 if (d_is_negative(new_dentry))
4456                         goto exit5;
4457 
4458                 if (!d_is_dir(new_dentry)) {
4459                         error = -ENOTDIR;
4460                         if (new_last.name[new_last.len])
4461                                 goto exit5;
4462                 }
4463         }
4464         /* unless the source is a directory trailing slashes give -ENOTDIR */
4465         if (!d_is_dir(old_dentry)) {
4466                 error = -ENOTDIR;
4467                 if (old_last.name[old_last.len])
4468                         goto exit5;
4469                 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4470                         goto exit5;
4471         }
4472         /* source should not be ancestor of target */
4473         error = -EINVAL;
4474         if (old_dentry == trap)
4475                 goto exit5;
4476         /* target should not be an ancestor of source */
4477         if (!(flags & RENAME_EXCHANGE))
4478                 error = -ENOTEMPTY;
4479         if (new_dentry == trap)
4480                 goto exit5;
4481 
4482         error = security_path_rename(&old_path, old_dentry,
4483                                      &new_path, new_dentry, flags);
4484         if (error)
4485                 goto exit5;
4486         error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4487                            new_path.dentry->d_inode, new_dentry,
4488                            &delegated_inode, flags);
4489 exit5:
4490         dput(new_dentry);
4491 exit4:
4492         dput(old_dentry);
4493 exit3:
4494         unlock_rename(new_path.dentry, old_path.dentry);
4495         if (delegated_inode) {
4496                 error = break_deleg_wait(&delegated_inode);
4497                 if (!error)
4498                         goto retry_deleg;
4499         }
4500         mnt_drop_write(old_path.mnt);
4501 exit2:
4502         if (retry_estale(error, lookup_flags))
4503                 should_retry = true;
4504         path_put(&new_path);
4505         putname(to);
4506 exit1:
4507         path_put(&old_path);
4508         putname(from);
4509         if (should_retry) {
4510                 should_retry = false;
4511                 lookup_flags |= LOOKUP_REVAL;
4512                 goto retry;
4513         }
4514 exit:
4515         return error;
4516 }
4517 
4518 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4519                 int, newdfd, const char __user *, newname)
4520 {
4521         return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4522 }
4523 
4524 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4525 {
4526         return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4527 }
4528 
4529 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4530 {
4531         int error = may_create(dir, dentry);
4532         if (error)
4533                 return error;
4534 
4535         if (!dir->i_op->mknod)
4536                 return -EPERM;
4537 
4538         return dir->i_op->mknod(dir, dentry,
4539                                 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4540 }
4541 EXPORT_SYMBOL(vfs_whiteout);
4542 
4543 int readlink_copy(char __user *buffer, int buflen, const char *link)
4544 {
4545         int len = PTR_ERR(link);
4546         if (IS_ERR(link))
4547                 goto out;
4548 
4549         len = strlen(link);
4550         if (len > (unsigned) buflen)
4551                 len = buflen;
4552         if (copy_to_user(buffer, link, len))
4553                 len = -EFAULT;
4554 out:
4555         return len;
4556 }
4557 EXPORT_SYMBOL(readlink_copy);
4558 
4559 /*
4560  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4561  * have ->get_link() not calling nd_jump_link().  Using (or not using) it
4562  * for any given inode is up to filesystem.
4563  */
4564 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4565 {
4566         DEFINE_DELAYED_CALL(done);
4567         struct inode *inode = d_inode(dentry);
4568         const char *link = inode->i_link;
4569         int res;
4570 
4571         if (!link) {
4572                 link = inode->i_op->get_link(dentry, inode, &done);
4573                 if (IS_ERR(link))
4574                         return PTR_ERR(link);
4575         }
4576         res = readlink_copy(buffer, buflen, link);
4577         do_delayed_call(&done);
4578         return res;
4579 }
4580 EXPORT_SYMBOL(generic_readlink);
4581 
4582 /* get the link contents into pagecache */
4583 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4584                           struct delayed_call *callback)
4585 {
4586         char *kaddr;
4587         struct page *page;
4588         struct address_space *mapping = inode->i_mapping;
4589 
4590         if (!dentry) {
4591                 page = find_get_page(mapping, 0);
4592                 if (!page)
4593                         return ERR_PTR(-ECHILD);
4594                 if (!PageUptodate(page)) {
4595                         put_page(page);
4596                         return ERR_PTR(-ECHILD);
4597                 }
4598         } else {
4599                 page = read_mapping_page(mapping, 0, NULL);
4600                 if (IS_ERR(page))
4601                         return (char*)page;
4602         }
4603         set_delayed_call(callback, page_put_link, page);
4604         BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4605         kaddr = page_address(page);
4606         nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4607         return kaddr;
4608 }
4609 
4610 EXPORT_SYMBOL(page_get_link);
4611 
4612 void page_put_link(void *arg)
4613 {
4614         put_page(arg);
4615 }
4616 EXPORT_SYMBOL(page_put_link);
4617 
4618 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4619 {
4620         DEFINE_DELAYED_CALL(done);
4621         int res = readlink_copy(buffer, buflen,
4622                                 page_get_link(dentry, d_inode(dentry),
4623                                               &done));
4624         do_delayed_call(&done);
4625         return res;
4626 }
4627 EXPORT_SYMBOL(page_readlink);
4628 
4629 /*
4630  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4631  */
4632 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4633 {
4634         struct address_space *mapping = inode->i_mapping;
4635         struct page *page;
4636         void *fsdata;
4637         int err;
4638         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4639         if (nofs)
4640                 flags |= AOP_FLAG_NOFS;
4641 
4642 retry:
4643         err = pagecache_write_begin(NULL, mapping, 0, len-1,
4644                                 flags, &page, &fsdata);
4645         if (err)
4646                 goto fail;
4647 
4648         memcpy(page_address(page), symname, len-1);
4649 
4650         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4651                                                         page, fsdata);
4652         if (err < 0)
4653                 goto fail;
4654         if (err < len-1)
4655                 goto retry;
4656 
4657         mark_inode_dirty(inode);
4658         return 0;
4659 fail:
4660         return err;
4661 }
4662 EXPORT_SYMBOL(__page_symlink);
4663 
4664 int page_symlink(struct inode *inode, const char *symname, int len)
4665 {
4666         return __page_symlink(inode, symname, len,
4667                         !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4668 }
4669 EXPORT_SYMBOL(page_symlink);
4670 
4671 const struct inode_operations page_symlink_inode_operations = {
4672         .readlink       = generic_readlink,
4673         .get_link       = page_get_link,
4674 };
4675 EXPORT_SYMBOL(page_symlink_inode_operations);
4676 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp