source: src/linux/universal/linux-4.9/fs/namei.c @ 31662

Last change on this file since 31662 was 31662, checked in by brainslayer, 4 months ago

use new squashfs in all kernels

File size: 119.9 KB
Line 
1/*
2 *  linux/fs/namei.c
3 *
4 *  Copyright (C) 1991, 1992  Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/export.h>
19#include <linux/kernel.h>
20#include <linux/slab.h>
21#include <linux/fs.h>
22#include <linux/namei.h>
23#include <linux/pagemap.h>
24#include <linux/fsnotify.h>
25#include <linux/personality.h>
26#include <linux/security.h>
27#include <linux/ima.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
31#include <linux/capability.h>
32#include <linux/file.h>
33#include <linux/fcntl.h>
34#include <linux/device_cgroup.h>
35#include <linux/fs_struct.h>
36#include <linux/posix_acl.h>
37#include <linux/hash.h>
38#include <linux/bitops.h>
39#include <linux/init_task.h>
40#include <asm/uaccess.h>
41
42#include "internal.h"
43#include "mount.h"
44
45/* [Feb-1997 T. Schoebel-Theuer]
46 * Fundamental changes in the pathname lookup mechanisms (namei)
47 * were necessary because of omirr.  The reason is that omirr needs
48 * to know the _real_ pathname, not the user-supplied one, in case
49 * of symlinks (and also when transname replacements occur).
50 *
51 * The new code replaces the old recursive symlink resolution with
52 * an iterative one (in case of non-nested symlink chains).  It does
53 * this with calls to <fs>_follow_link().
54 * As a side effect, dir_namei(), _namei() and follow_link() are now
55 * replaced with a single function lookup_dentry() that can handle all
56 * the special cases of the former code.
57 *
58 * With the new dcache, the pathname is stored at each inode, at least as
59 * long as the refcount of the inode is positive.  As a side effect, the
60 * size of the dcache depends on the inode cache and thus is dynamic.
61 *
62 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63 * resolution to correspond with current state of the code.
64 *
65 * Note that the symlink resolution is not *completely* iterative.
66 * There is still a significant amount of tail- and mid- recursion in
67 * the algorithm.  Also, note that <fs>_readlink() is not used in
68 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69 * may return different results than <fs>_follow_link().  Many virtual
70 * filesystems (including /proc) exhibit this behavior.
71 */
72
73/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75 * and the name already exists in form of a symlink, try to create the new
76 * name indicated by the symlink. The old code always complained that the
77 * name already exists, due to not following the symlink even if its target
78 * is nonexistent.  The new semantics affects also mknod() and link() when
79 * the name is a symlink pointing to a non-existent name.
80 *
81 * I don't know which semantics is the right one, since I have no access
82 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84 * "old" one. Personally, I think the new semantics is much more logical.
85 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86 * file does succeed in both HP-UX and SunOs, but not in Solaris
87 * and in the old Linux semantics.
88 */
89
90/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91 * semantics.  See the comments in "open_namei" and "do_link" below.
92 *
93 * [10-Sep-98 Alan Modra] Another symlink change.
94 */
95
96/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97 *      inside the path - always follow.
98 *      in the last component in creation/removal/renaming - never follow.
99 *      if LOOKUP_FOLLOW passed - follow.
100 *      if the pathname has trailing slashes - follow.
101 *      otherwise - don't follow.
102 * (applied in that order).
103 *
104 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106 * During the 2.4 we need to fix the userland stuff depending on it -
107 * hopefully we will be able to get rid of that wart in 2.5. So far only
108 * XEmacs seems to be relying on it...
109 */
110/*
111 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
112 * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
113 * any extra contention...
114 */
115
116/* In order to reduce some races, while at the same time doing additional
117 * checking and hopefully speeding things up, we copy filenames to the
118 * kernel data space before using them..
119 *
120 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121 * PATH_MAX includes the nul terminator --RR.
122 */
123
124#define EMBEDDED_NAME_MAX       (PATH_MAX - offsetof(struct filename, iname))
125
126struct filename *
127getname_flags(const char __user *filename, int flags, int *empty)
128{
129        struct filename *result;
130        char *kname;
131        int len;
132
133        result = audit_reusename(filename);
134        if (result)
135                return result;
136
137        result = __getname();
138        if (unlikely(!result))
139                return ERR_PTR(-ENOMEM);
140
141        /*
142         * First, try to embed the struct filename inside the names_cache
143         * allocation
144         */
145        kname = (char *)result->iname;
146        result->name = kname;
147
148        len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
149        if (unlikely(len < 0)) {
150                __putname(result);
151                return ERR_PTR(len);
152        }
153
154        /*
155         * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156         * separate struct filename so we can dedicate the entire
157         * names_cache allocation for the pathname, and re-do the copy from
158         * userland.
159         */
160        if (unlikely(len == EMBEDDED_NAME_MAX)) {
161                const size_t size = offsetof(struct filename, iname[1]);
162                kname = (char *)result;
163
164                /*
165                 * size is chosen that way we to guarantee that
166                 * result->iname[0] is within the same object and that
167                 * kname can't be equal to result->iname, no matter what.
168                 */
169                result = kzalloc(size, GFP_KERNEL);
170                if (unlikely(!result)) {
171                        __putname(kname);
172                        return ERR_PTR(-ENOMEM);
173                }
174                result->name = kname;
175                len = strncpy_from_user(kname, filename, PATH_MAX);
176                if (unlikely(len < 0)) {
177                        __putname(kname);
178                        kfree(result);
179                        return ERR_PTR(len);
180                }
181                if (unlikely(len == PATH_MAX)) {
182                        __putname(kname);
183                        kfree(result);
184                        return ERR_PTR(-ENAMETOOLONG);
185                }
186        }
187
188        result->refcnt = 1;
189        /* The empty path is special. */
190        if (unlikely(!len)) {
191                if (empty)
192                        *empty = 1;
193                if (!(flags & LOOKUP_EMPTY)) {
194                        putname(result);
195                        return ERR_PTR(-ENOENT);
196                }
197        }
198
199        result->uptr = filename;
200        result->aname = NULL;
201        audit_getname(result);
202        return result;
203}
204
205struct filename *
206getname(const char __user * filename)
207{
208        return getname_flags(filename, 0, NULL);
209}
210
211struct filename *
212getname_kernel(const char * filename)
213{
214        struct filename *result;
215        int len = strlen(filename) + 1;
216
217        result = __getname();
218        if (unlikely(!result))
219                return ERR_PTR(-ENOMEM);
220
221        if (len <= EMBEDDED_NAME_MAX) {
222                result->name = (char *)result->iname;
223        } else if (len <= PATH_MAX) {
224                struct filename *tmp;
225
226                tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227                if (unlikely(!tmp)) {
228                        __putname(result);
229                        return ERR_PTR(-ENOMEM);
230                }
231                tmp->name = (char *)result;
232                result = tmp;
233        } else {
234                __putname(result);
235                return ERR_PTR(-ENAMETOOLONG);
236        }
237        memcpy((char *)result->name, filename, len);
238        result->uptr = NULL;
239        result->aname = NULL;
240        result->refcnt = 1;
241        audit_getname(result);
242
243        return result;
244}
245
246void putname(struct filename *name)
247{
248        BUG_ON(name->refcnt <= 0);
249
250        if (--name->refcnt > 0)
251                return;
252
253        if (name->name != name->iname) {
254                __putname(name->name);
255                kfree(name);
256        } else
257                __putname(name);
258}
259
260static int check_acl(struct inode *inode, int mask)
261{
262#ifdef CONFIG_FS_POSIX_ACL
263        struct posix_acl *acl;
264
265        if (mask & MAY_NOT_BLOCK) {
266                acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267                if (!acl)
268                        return -EAGAIN;
269                /* no ->get_acl() calls in RCU mode... */
270                if (is_uncached_acl(acl))
271                        return -ECHILD;
272                return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273        }
274
275        acl = get_acl(inode, ACL_TYPE_ACCESS);
276        if (IS_ERR(acl))
277                return PTR_ERR(acl);
278        if (acl) {
279                int error = posix_acl_permission(inode, acl, mask);
280                posix_acl_release(acl);
281                return error;
282        }
283#endif
284
285        return -EAGAIN;
286}
287
288/*
289 * This does the basic permission checking
290 */
291static int acl_permission_check(struct inode *inode, int mask)
292{
293        unsigned int mode = inode->i_mode;
294
295        if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296                mode >>= 6;
297        else {
298                if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
299                        int error = check_acl(inode, mask);
300                        if (error != -EAGAIN)
301                                return error;
302                }
303
304                if (in_group_p(inode->i_gid))
305                        mode >>= 3;
306        }
307
308        /*
309         * If the DACs are ok we don't need any capability check.
310         */
311        if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
312                return 0;
313        return -EACCES;
314}
315
316/**
317 * generic_permission -  check for access rights on a Posix-like filesystem
318 * @inode:      inode to check access rights for
319 * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
320 *
321 * Used to check for read/write/execute permissions on a file.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
324 * are used for other things.
325 *
326 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327 * request cannot be satisfied (eg. requires blocking or too much complexity).
328 * It would then be called again in ref-walk mode.
329 */
330int generic_permission(struct inode *inode, int mask)
331{
332        int ret;
333
334        /*
335         * Do the basic permission checks.
336         */
337        ret = acl_permission_check(inode, mask);
338        if (ret != -EACCES)
339                return ret;
340
341        if (S_ISDIR(inode->i_mode)) {
342                /* DACs are overridable for directories */
343                if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
344                        return 0;
345                if (!(mask & MAY_WRITE))
346                        if (capable_wrt_inode_uidgid(inode,
347                                                     CAP_DAC_READ_SEARCH))
348                                return 0;
349                return -EACCES;
350        }
351        /*
352         * Read/write DACs are always overridable.
353         * Executable DACs are overridable when there is
354         * at least one exec bit set.
355         */
356        if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
357                if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
358                        return 0;
359
360        /*
361         * Searching includes executable on directories, else just read.
362         */
363        mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
364        if (mask == MAY_READ)
365                if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
366                        return 0;
367
368        return -EACCES;
369}
370EXPORT_SYMBOL(generic_permission);
371
372/*
373 * We _really_ want to just do "generic_permission()" without
374 * even looking at the inode->i_op values. So we keep a cache
375 * flag in inode->i_opflags, that says "this has not special
376 * permission function, use the fast case".
377 */
378static inline int do_inode_permission(struct inode *inode, int mask)
379{
380        if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381                if (likely(inode->i_op->permission))
382                        return inode->i_op->permission(inode, mask);
383
384                /* This gets set once for the inode lifetime */
385                spin_lock(&inode->i_lock);
386                inode->i_opflags |= IOP_FASTPERM;
387                spin_unlock(&inode->i_lock);
388        }
389        return generic_permission(inode, mask);
390}
391
392/**
393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
396 *
397 * Check for read/write/execute permissions on an inode.
398 *
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
400 *
401 * This does not check for a read-only file system.  You probably want
402 * inode_permission().
403 */
404int __inode_permission(struct inode *inode, int mask)
405{
406        int retval;
407
408        if (unlikely(mask & MAY_WRITE)) {
409                /*
410                 * Nobody gets write access to an immutable file.
411                 */
412                if (IS_IMMUTABLE(inode))
413                        return -EPERM;
414
415                /*
416                 * Updating mtime will likely cause i_uid and i_gid to be
417                 * written back improperly if their true value is unknown
418                 * to the vfs.
419                 */
420                if (HAS_UNMAPPED_ID(inode))
421                        return -EACCES;
422        }
423
424        retval = do_inode_permission(inode, mask);
425        if (retval)
426                return retval;
427
428        retval = devcgroup_inode_permission(inode, mask);
429        if (retval)
430                return retval;
431
432        return security_inode_permission(inode, mask);
433}
434EXPORT_SYMBOL(__inode_permission);
435
436/**
437 * sb_permission - Check superblock-level permissions
438 * @sb: Superblock of inode to check permission on
439 * @inode: Inode to check permission on
440 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
441 *
442 * Separate out file-system wide checks from inode-specific permission checks.
443 */
444static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
445{
446        if (unlikely(mask & MAY_WRITE)) {
447                umode_t mode = inode->i_mode;
448
449                /* Nobody gets write access to a read-only fs. */
450                if ((sb->s_flags & MS_RDONLY) &&
451                    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
452                        return -EROFS;
453        }
454        return 0;
455}
456
457/**
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461 *
462 * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
465 *
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
467 */
468int inode_permission(struct inode *inode, int mask)
469{
470        int retval;
471
472        retval = sb_permission(inode->i_sb, inode, mask);
473        if (retval)
474                return retval;
475        return __inode_permission(inode, mask);
476}
477EXPORT_SYMBOL(inode_permission);
478
479/**
480 * path_get - get a reference to a path
481 * @path: path to get the reference to
482 *
483 * Given a path increment the reference count to the dentry and the vfsmount.
484 */
485void path_get(const struct path *path)
486{
487        mntget(path->mnt);
488        dget(path->dentry);
489}
490EXPORT_SYMBOL(path_get);
491
492/**
493 * path_put - put a reference to a path
494 * @path: path to put the reference to
495 *
496 * Given a path decrement the reference count to the dentry and the vfsmount.
497 */
498void path_put(const struct path *path)
499{
500        dput(path->dentry);
501        mntput(path->mnt);
502}
503EXPORT_SYMBOL(path_put);
504
505#define EMBEDDED_LEVELS 2
506struct nameidata {
507        struct path     path;
508        struct qstr     last;
509        struct path     root;
510        struct inode    *inode; /* path.dentry.d_inode */
511        unsigned int    flags;
512        unsigned        seq, m_seq;
513        int             last_type;
514        unsigned        depth;
515        int             total_link_count;
516        struct saved {
517                struct path link;
518                struct delayed_call done;
519                const char *name;
520                unsigned seq;
521        } *stack, internal[EMBEDDED_LEVELS];
522        struct filename *name;
523        struct nameidata *saved;
524        struct inode    *link_inode;
525        unsigned        root_seq;
526        int             dfd;
527};
528
529static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
530{
531        struct nameidata *old = current->nameidata;
532        p->stack = p->internal;
533        p->dfd = dfd;
534        p->name = name;
535        p->total_link_count = old ? old->total_link_count : 0;
536        p->saved = old;
537        current->nameidata = p;
538}
539
540static void restore_nameidata(void)
541{
542        struct nameidata *now = current->nameidata, *old = now->saved;
543
544        current->nameidata = old;
545        if (old)
546                old->total_link_count = now->total_link_count;
547        if (now->stack != now->internal)
548                kfree(now->stack);
549}
550
551static int __nd_alloc_stack(struct nameidata *nd)
552{
553        struct saved *p;
554
555        if (nd->flags & LOOKUP_RCU) {
556                p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
557                                  GFP_ATOMIC);
558                if (unlikely(!p))
559                        return -ECHILD;
560        } else {
561                p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
562                                  GFP_KERNEL);
563                if (unlikely(!p))
564                        return -ENOMEM;
565        }
566        memcpy(p, nd->internal, sizeof(nd->internal));
567        nd->stack = p;
568        return 0;
569}
570
571/**
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
574 *
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
577 */
578static bool path_connected(const struct path *path)
579{
580        struct vfsmount *mnt = path->mnt;
581
582        /* Only bind mounts can have disconnected paths */
583        if (mnt->mnt_root == mnt->mnt_sb->s_root)
584                return true;
585
586        return is_subdir(path->dentry, mnt->mnt_root);
587}
588
589static inline int nd_alloc_stack(struct nameidata *nd)
590{
591        if (likely(nd->depth != EMBEDDED_LEVELS))
592                return 0;
593        if (likely(nd->stack != nd->internal))
594                return 0;
595        return __nd_alloc_stack(nd);
596}
597
598static void drop_links(struct nameidata *nd)
599{
600        int i = nd->depth;
601        while (i--) {
602                struct saved *last = nd->stack + i;
603                do_delayed_call(&last->done);
604                clear_delayed_call(&last->done);
605        }
606}
607
608static void terminate_walk(struct nameidata *nd)
609{
610        drop_links(nd);
611        if (!(nd->flags & LOOKUP_RCU)) {
612                int i;
613                path_put(&nd->path);
614                for (i = 0; i < nd->depth; i++)
615                        path_put(&nd->stack[i].link);
616                if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
617                        path_put(&nd->root);
618                        nd->root.mnt = NULL;
619                }
620        } else {
621                nd->flags &= ~LOOKUP_RCU;
622                if (!(nd->flags & LOOKUP_ROOT))
623                        nd->root.mnt = NULL;
624                rcu_read_unlock();
625        }
626        nd->depth = 0;
627}
628
629/* path_put is needed afterwards regardless of success or failure */
630static bool legitimize_path(struct nameidata *nd,
631                            struct path *path, unsigned seq)
632{
633        int res = __legitimize_mnt(path->mnt, nd->m_seq);
634        if (unlikely(res)) {
635                if (res > 0)
636                        path->mnt = NULL;
637                path->dentry = NULL;
638                return false;
639        }
640        if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
641                path->dentry = NULL;
642                return false;
643        }
644        return !read_seqcount_retry(&path->dentry->d_seq, seq);
645}
646
647static bool legitimize_links(struct nameidata *nd)
648{
649        int i;
650        for (i = 0; i < nd->depth; i++) {
651                struct saved *last = nd->stack + i;
652                if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653                        drop_links(nd);
654                        nd->depth = i + 1;
655                        return false;
656                }
657        }
658        return true;
659}
660
661/*
662 * Path walking has 2 modes, rcu-walk and ref-walk (see
663 * Documentation/filesystems/path-lookup.txt).  In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
666 * mode.  Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
670 */
671
672/**
673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
675 * @dentry: child of nd->path.dentry or NULL
676 * @seq: seq number to check dentry against
677 * Returns: 0 on success, -ECHILD on failure
678 *
679 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
680 * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
681 * @nd or NULL.  Must be called from rcu-walk context.
682 * Nothing should touch nameidata between unlazy_walk() failure and
683 * terminate_walk().
684 */
685static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
686{
687        struct dentry *parent = nd->path.dentry;
688
689        BUG_ON(!(nd->flags & LOOKUP_RCU));
690
691        nd->flags &= ~LOOKUP_RCU;
692        if (unlikely(!legitimize_links(nd)))
693                goto out2;
694        if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
695                goto out2;
696        if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
697                goto out1;
698
699        /*
700         * For a negative lookup, the lookup sequence point is the parents
701         * sequence point, and it only needs to revalidate the parent dentry.
702         *
703         * For a positive lookup, we need to move both the parent and the
704         * dentry from the RCU domain to be properly refcounted. And the
705         * sequence number in the dentry validates *both* dentry counters,
706         * since we checked the sequence number of the parent after we got
707         * the child sequence number. So we know the parent must still
708         * be valid if the child sequence number is still valid.
709         */
710        if (!dentry) {
711                if (read_seqcount_retry(&parent->d_seq, nd->seq))
712                        goto out;
713                BUG_ON(nd->inode != parent->d_inode);
714        } else {
715                if (!lockref_get_not_dead(&dentry->d_lockref))
716                        goto out;
717                if (read_seqcount_retry(&dentry->d_seq, seq))
718                        goto drop_dentry;
719        }
720
721        /*
722         * Sequence counts matched. Now make sure that the root is
723         * still valid and get it if required.
724         */
725        if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
726                if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
727                        rcu_read_unlock();
728                        dput(dentry);
729                        return -ECHILD;
730                }
731        }
732
733        rcu_read_unlock();
734        return 0;
735
736drop_dentry:
737        rcu_read_unlock();
738        dput(dentry);
739        goto drop_root_mnt;
740out2:
741        nd->path.mnt = NULL;
742out1:
743        nd->path.dentry = NULL;
744out:
745        rcu_read_unlock();
746drop_root_mnt:
747        if (!(nd->flags & LOOKUP_ROOT))
748                nd->root.mnt = NULL;
749        return -ECHILD;
750}
751
752static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
753{
754        if (unlikely(!legitimize_path(nd, link, seq))) {
755                drop_links(nd);
756                nd->depth = 0;
757                nd->flags &= ~LOOKUP_RCU;
758                nd->path.mnt = NULL;
759                nd->path.dentry = NULL;
760                if (!(nd->flags & LOOKUP_ROOT))
761                        nd->root.mnt = NULL;
762                rcu_read_unlock();
763        } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
764                return 0;
765        }
766        path_put(link);
767        return -ECHILD;
768}
769
770static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
771{
772        return dentry->d_op->d_revalidate(dentry, flags);
773}
774
775/**
776 * complete_walk - successful completion of path walk
777 * @nd:  pointer nameidata
778 *
779 * If we had been in RCU mode, drop out of it and legitimize nd->path.
780 * Revalidate the final result, unless we'd already done that during
781 * the path walk or the filesystem doesn't ask for it.  Return 0 on
782 * success, -error on failure.  In case of failure caller does not
783 * need to drop nd->path.
784 */
785static int complete_walk(struct nameidata *nd)
786{
787        struct dentry *dentry = nd->path.dentry;
788        int status;
789
790        if (nd->flags & LOOKUP_RCU) {
791                if (!(nd->flags & LOOKUP_ROOT))
792                        nd->root.mnt = NULL;
793                if (unlikely(unlazy_walk(nd, NULL, 0)))
794                        return -ECHILD;
795        }
796
797        if (likely(!(nd->flags & LOOKUP_JUMPED)))
798                return 0;
799
800        if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
801                return 0;
802
803        status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
804        if (status > 0)
805                return 0;
806
807        if (!status)
808                status = -ESTALE;
809
810        return status;
811}
812
813static void set_root(struct nameidata *nd)
814{
815        struct fs_struct *fs = current->fs;
816
817        if (nd->flags & LOOKUP_RCU) {
818                unsigned seq;
819
820                do {
821                        seq = read_seqcount_begin(&fs->seq);
822                        nd->root = fs->root;
823                        nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
824                } while (read_seqcount_retry(&fs->seq, seq));
825        } else {
826                get_fs_root(fs, &nd->root);
827        }
828}
829
830static void path_put_conditional(struct path *path, struct nameidata *nd)
831{
832        dput(path->dentry);
833        if (path->mnt != nd->path.mnt)
834                mntput(path->mnt);
835}
836
837static inline void path_to_nameidata(const struct path *path,
838                                        struct nameidata *nd)
839{
840        if (!(nd->flags & LOOKUP_RCU)) {
841                dput(nd->path.dentry);
842                if (nd->path.mnt != path->mnt)
843                        mntput(nd->path.mnt);
844        }
845        nd->path.mnt = path->mnt;
846        nd->path.dentry = path->dentry;
847}
848
849static int nd_jump_root(struct nameidata *nd)
850{
851        if (nd->flags & LOOKUP_RCU) {
852                struct dentry *d;
853                nd->path = nd->root;
854                d = nd->path.dentry;
855                nd->inode = d->d_inode;
856                nd->seq = nd->root_seq;
857                if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
858                        return -ECHILD;
859        } else {
860                path_put(&nd->path);
861                nd->path = nd->root;
862                path_get(&nd->path);
863                nd->inode = nd->path.dentry->d_inode;
864        }
865        nd->flags |= LOOKUP_JUMPED;
866        return 0;
867}
868
869/*
870 * Helper to directly jump to a known parsed path from ->get_link,
871 * caller must have taken a reference to path beforehand.
872 */
873void nd_jump_link(struct path *path)
874{
875        struct nameidata *nd = current->nameidata;
876        path_put(&nd->path);
877
878        nd->path = *path;
879        nd->inode = nd->path.dentry->d_inode;
880        nd->flags |= LOOKUP_JUMPED;
881}
882
883static inline void put_link(struct nameidata *nd)
884{
885        struct saved *last = nd->stack + --nd->depth;
886        do_delayed_call(&last->done);
887        if (!(nd->flags & LOOKUP_RCU))
888                path_put(&last->link);
889}
890
891int sysctl_protected_symlinks __read_mostly = 0;
892int sysctl_protected_hardlinks __read_mostly = 0;
893
894/**
895 * may_follow_link - Check symlink following for unsafe situations
896 * @nd: nameidata pathwalk data
897 *
898 * In the case of the sysctl_protected_symlinks sysctl being enabled,
899 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
900 * in a sticky world-writable directory. This is to protect privileged
901 * processes from failing races against path names that may change out
902 * from under them by way of other users creating malicious symlinks.
903 * It will permit symlinks to be followed only when outside a sticky
904 * world-writable directory, or when the uid of the symlink and follower
905 * match, or when the directory owner matches the symlink's owner.
906 *
907 * Returns 0 if following the symlink is allowed, -ve on error.
908 */
909static inline int may_follow_link(struct nameidata *nd)
910{
911        const struct inode *inode;
912        const struct inode *parent;
913        kuid_t puid;
914
915        if (!sysctl_protected_symlinks)
916                return 0;
917
918        /* Allowed if owner and follower match. */
919        inode = nd->link_inode;
920        if (uid_eq(current_cred()->fsuid, inode->i_uid))
921                return 0;
922
923        /* Allowed if parent directory not sticky and world-writable. */
924        parent = nd->inode;
925        if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
926                return 0;
927
928        /* Allowed if parent directory and link owner match. */
929        puid = parent->i_uid;
930        if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
931                return 0;
932
933        if (nd->flags & LOOKUP_RCU)
934                return -ECHILD;
935
936        audit_log_link_denied("follow_link", &nd->stack[0].link);
937        return -EACCES;
938}
939
940/**
941 * safe_hardlink_source - Check for safe hardlink conditions
942 * @inode: the source inode to hardlink from
943 *
944 * Return false if at least one of the following conditions:
945 *    - inode is not a regular file
946 *    - inode is setuid
947 *    - inode is setgid and group-exec
948 *    - access failure for read and write
949 *
950 * Otherwise returns true.
951 */
952static bool safe_hardlink_source(struct inode *inode)
953{
954        umode_t mode = inode->i_mode;
955
956        /* Special files should not get pinned to the filesystem. */
957        if (!S_ISREG(mode))
958                return false;
959
960        /* Setuid files should not get pinned to the filesystem. */
961        if (mode & S_ISUID)
962                return false;
963
964        /* Executable setgid files should not get pinned to the filesystem. */
965        if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
966                return false;
967
968        /* Hardlinking to unreadable or unwritable sources is dangerous. */
969        if (inode_permission(inode, MAY_READ | MAY_WRITE))
970                return false;
971
972        return true;
973}
974
975/**
976 * may_linkat - Check permissions for creating a hardlink
977 * @link: the source to hardlink from
978 *
979 * Block hardlink when all of:
980 *  - sysctl_protected_hardlinks enabled
981 *  - fsuid does not match inode
982 *  - hardlink source is unsafe (see safe_hardlink_source() above)
983 *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
984 *
985 * Returns 0 if successful, -ve on error.
986 */
987static int may_linkat(struct path *link)
988{
989        struct inode *inode;
990
991        if (!sysctl_protected_hardlinks)
992                return 0;
993
994        inode = link->dentry->d_inode;
995
996        /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997         * otherwise, it must be a safe source.
998         */
999        if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
1000                return 0;
1001
1002        audit_log_link_denied("linkat", link);
1003        return -EPERM;
1004}
1005
1006static __always_inline
1007const char *get_link(struct nameidata *nd)
1008{
1009        struct saved *last = nd->stack + nd->depth - 1;
1010        struct dentry *dentry = last->link.dentry;
1011        struct inode *inode = nd->link_inode;
1012        int error;
1013        const char *res;
1014
1015        if (!(nd->flags & LOOKUP_RCU)) {
1016                touch_atime(&last->link);
1017                cond_resched();
1018        } else if (atime_needs_update_rcu(&last->link, inode)) {
1019                if (unlikely(unlazy_walk(nd, NULL, 0)))
1020                        return ERR_PTR(-ECHILD);
1021                touch_atime(&last->link);
1022        }
1023
1024        error = security_inode_follow_link(dentry, inode,
1025                                           nd->flags & LOOKUP_RCU);
1026        if (unlikely(error))
1027                return ERR_PTR(error);
1028
1029        nd->last_type = LAST_BIND;
1030        res = inode->i_link;
1031        if (!res) {
1032                const char * (*get)(struct dentry *, struct inode *,
1033                                struct delayed_call *);
1034                get = inode->i_op->get_link;
1035                if (nd->flags & LOOKUP_RCU) {
1036                        res = get(NULL, inode, &last->done);
1037                        if (res == ERR_PTR(-ECHILD)) {
1038                                if (unlikely(unlazy_walk(nd, NULL, 0)))
1039                                        return ERR_PTR(-ECHILD);
1040                                res = get(dentry, inode, &last->done);
1041                        }
1042                } else {
1043                        res = get(dentry, inode, &last->done);
1044                }
1045                if (IS_ERR_OR_NULL(res))
1046                        return res;
1047        }
1048        if (*res == '/') {
1049                if (!nd->root.mnt)
1050                        set_root(nd);
1051                if (unlikely(nd_jump_root(nd)))
1052                        return ERR_PTR(-ECHILD);
1053                while (unlikely(*++res == '/'))
1054                        ;
1055        }
1056        if (!*res)
1057                res = NULL;
1058        return res;
1059}
1060
1061/*
1062 * follow_up - Find the mountpoint of path's vfsmount
1063 *
1064 * Given a path, find the mountpoint of its source file system.
1065 * Replace @path with the path of the mountpoint in the parent mount.
1066 * Up is towards /.
1067 *
1068 * Return 1 if we went up a level and 0 if we were already at the
1069 * root.
1070 */
1071int follow_up(struct path *path)
1072{
1073        struct mount *mnt = real_mount(path->mnt);
1074        struct mount *parent;
1075        struct dentry *mountpoint;
1076
1077        read_seqlock_excl(&mount_lock);
1078        parent = mnt->mnt_parent;
1079        if (parent == mnt) {
1080                read_sequnlock_excl(&mount_lock);
1081                return 0;
1082        }
1083        mntget(&parent->mnt);
1084        mountpoint = dget(mnt->mnt_mountpoint);
1085        read_sequnlock_excl(&mount_lock);
1086        dput(path->dentry);
1087        path->dentry = mountpoint;
1088        mntput(path->mnt);
1089        path->mnt = &parent->mnt;
1090        return 1;
1091}
1092EXPORT_SYMBOL(follow_up);
1093
1094/*
1095 * Perform an automount
1096 * - return -EISDIR to tell follow_managed() to stop and return the path we
1097 *   were called with.
1098 */
1099static int follow_automount(struct path *path, struct nameidata *nd,
1100                            bool *need_mntput)
1101{
1102        struct vfsmount *mnt;
1103        int err;
1104
1105        if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1106                return -EREMOTE;
1107
1108        /* We don't want to mount if someone's just doing a stat -
1109         * unless they're stat'ing a directory and appended a '/' to
1110         * the name.
1111         *
1112         * We do, however, want to mount if someone wants to open or
1113         * create a file of any type under the mountpoint, wants to
1114         * traverse through the mountpoint or wants to open the
1115         * mounted directory.  Also, autofs may mark negative dentries
1116         * as being automount points.  These will need the attentions
1117         * of the daemon to instantiate them before they can be used.
1118         */
1119        if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1120                           LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1121            path->dentry->d_inode)
1122                return -EISDIR;
1123
1124        if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1125                return -EACCES;
1126
1127        nd->total_link_count++;
1128        if (nd->total_link_count >= 40)
1129                return -ELOOP;
1130
1131        mnt = path->dentry->d_op->d_automount(path);
1132        if (IS_ERR(mnt)) {
1133                /*
1134                 * The filesystem is allowed to return -EISDIR here to indicate
1135                 * it doesn't want to automount.  For instance, autofs would do
1136                 * this so that its userspace daemon can mount on this dentry.
1137                 *
1138                 * However, we can only permit this if it's a terminal point in
1139                 * the path being looked up; if it wasn't then the remainder of
1140                 * the path is inaccessible and we should say so.
1141                 */
1142                if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1143                        return -EREMOTE;
1144                return PTR_ERR(mnt);
1145        }
1146
1147        if (!mnt) /* mount collision */
1148                return 0;
1149
1150        if (!*need_mntput) {
1151                /* lock_mount() may release path->mnt on error */
1152                mntget(path->mnt);
1153                *need_mntput = true;
1154        }
1155        err = finish_automount(mnt, path);
1156
1157        switch (err) {
1158        case -EBUSY:
1159                /* Someone else made a mount here whilst we were busy */
1160                return 0;
1161        case 0:
1162                path_put(path);
1163                path->mnt = mnt;
1164                path->dentry = dget(mnt->mnt_root);
1165                return 0;
1166        default:
1167                return err;
1168        }
1169
1170}
1171
1172/*
1173 * Handle a dentry that is managed in some way.
1174 * - Flagged for transit management (autofs)
1175 * - Flagged as mountpoint
1176 * - Flagged as automount point
1177 *
1178 * This may only be called in refwalk mode.
1179 *
1180 * Serialization is taken care of in namespace.c
1181 */
1182static int follow_managed(struct path *path, struct nameidata *nd)
1183{
1184        struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1185        unsigned managed;
1186        bool need_mntput = false;
1187        int ret = 0;
1188
1189        /* Given that we're not holding a lock here, we retain the value in a
1190         * local variable for each dentry as we look at it so that we don't see
1191         * the components of that value change under us */
1192        while (managed = ACCESS_ONCE(path->dentry->d_flags),
1193               managed &= DCACHE_MANAGED_DENTRY,
1194               unlikely(managed != 0)) {
1195                /* Allow the filesystem to manage the transit without i_mutex
1196                 * being held. */
1197                if (managed & DCACHE_MANAGE_TRANSIT) {
1198                        BUG_ON(!path->dentry->d_op);
1199                        BUG_ON(!path->dentry->d_op->d_manage);
1200                        ret = path->dentry->d_op->d_manage(path->dentry, false);
1201                        if (ret < 0)
1202                                break;
1203                }
1204
1205                /* Transit to a mounted filesystem. */
1206                if (managed & DCACHE_MOUNTED) {
1207                        struct vfsmount *mounted = lookup_mnt(path);
1208                        if (mounted) {
1209                                dput(path->dentry);
1210                                if (need_mntput)
1211                                        mntput(path->mnt);
1212                                path->mnt = mounted;
1213                                path->dentry = dget(mounted->mnt_root);
1214                                need_mntput = true;
1215                                continue;
1216                        }
1217
1218                        /* Something is mounted on this dentry in another
1219                         * namespace and/or whatever was mounted there in this
1220                         * namespace got unmounted before lookup_mnt() could
1221                         * get it */
1222                }
1223
1224                /* Handle an automount point */
1225                if (managed & DCACHE_NEED_AUTOMOUNT) {
1226                        ret = follow_automount(path, nd, &need_mntput);
1227                        if (ret < 0)
1228                                break;
1229                        continue;
1230                }
1231
1232                /* We didn't change the current path point */
1233                break;
1234        }
1235
1236        if (need_mntput && path->mnt == mnt)
1237                mntput(path->mnt);
1238        if (ret == -EISDIR || !ret)
1239                ret = 1;
1240        if (need_mntput)
1241                nd->flags |= LOOKUP_JUMPED;
1242        if (unlikely(ret < 0))
1243                path_put_conditional(path, nd);
1244        return ret;
1245}
1246
1247int follow_down_one(struct path *path)
1248{
1249        struct vfsmount *mounted;
1250
1251        mounted = lookup_mnt(path);
1252        if (mounted) {
1253                dput(path->dentry);
1254                mntput(path->mnt);
1255                path->mnt = mounted;
1256                path->dentry = dget(mounted->mnt_root);
1257                return 1;
1258        }
1259        return 0;
1260}
1261EXPORT_SYMBOL(follow_down_one);
1262
1263static inline int managed_dentry_rcu(struct dentry *dentry)
1264{
1265        return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1266                dentry->d_op->d_manage(dentry, true) : 0;
1267}
1268
1269/*
1270 * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1271 * we meet a managed dentry that would need blocking.
1272 */
1273static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1274                               struct inode **inode, unsigned *seqp)
1275{
1276        for (;;) {
1277                struct mount *mounted;
1278                /*
1279                 * Don't forget we might have a non-mountpoint managed dentry
1280                 * that wants to block transit.
1281                 */
1282                switch (managed_dentry_rcu(path->dentry)) {
1283                case -ECHILD:
1284                default:
1285                        return false;
1286                case -EISDIR:
1287                        return true;
1288                case 0:
1289                        break;
1290                }
1291
1292                if (!d_mountpoint(path->dentry))
1293                        return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1294
1295                mounted = __lookup_mnt(path->mnt, path->dentry);
1296                if (!mounted)
1297                        break;
1298                path->mnt = &mounted->mnt;
1299                path->dentry = mounted->mnt.mnt_root;
1300                nd->flags |= LOOKUP_JUMPED;
1301                *seqp = read_seqcount_begin(&path->dentry->d_seq);
1302                /*
1303                 * Update the inode too. We don't need to re-check the
1304                 * dentry sequence number here after this d_inode read,
1305                 * because a mount-point is always pinned.
1306                 */
1307                *inode = path->dentry->d_inode;
1308        }
1309        return !read_seqretry(&mount_lock, nd->m_seq) &&
1310                !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1311}
1312
1313static int follow_dotdot_rcu(struct nameidata *nd)
1314{
1315        struct inode *inode = nd->inode;
1316
1317        while (1) {
1318                if (path_equal(&nd->path, &nd->root))
1319                        break;
1320                if (nd->path.dentry != nd->path.mnt->mnt_root) {
1321                        struct dentry *old = nd->path.dentry;
1322                        struct dentry *parent = old->d_parent;
1323                        unsigned seq;
1324
1325                        inode = parent->d_inode;
1326                        seq = read_seqcount_begin(&parent->d_seq);
1327                        if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1328                                return -ECHILD;
1329                        nd->path.dentry = parent;
1330                        nd->seq = seq;
1331                        if (unlikely(!path_connected(&nd->path)))
1332                                return -ENOENT;
1333                        break;
1334                } else {
1335                        struct mount *mnt = real_mount(nd->path.mnt);
1336                        struct mount *mparent = mnt->mnt_parent;
1337                        struct dentry *mountpoint = mnt->mnt_mountpoint;
1338                        struct inode *inode2 = mountpoint->d_inode;
1339                        unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1340                        if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1341                                return -ECHILD;
1342                        if (&mparent->mnt == nd->path.mnt)
1343                                break;
1344                        /* we know that mountpoint was pinned */
1345                        nd->path.dentry = mountpoint;
1346                        nd->path.mnt = &mparent->mnt;
1347                        inode = inode2;
1348                        nd->seq = seq;
1349                }
1350        }
1351        while (unlikely(d_mountpoint(nd->path.dentry))) {
1352                struct mount *mounted;
1353                mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1354                if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1355                        return -ECHILD;
1356                if (!mounted)
1357                        break;
1358                nd->path.mnt = &mounted->mnt;
1359                nd->path.dentry = mounted->mnt.mnt_root;
1360                inode = nd->path.dentry->d_inode;
1361                nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1362        }
1363        nd->inode = inode;
1364        return 0;
1365}
1366
1367/*
1368 * Follow down to the covering mount currently visible to userspace.  At each
1369 * point, the filesystem owning that dentry may be queried as to whether the
1370 * caller is permitted to proceed or not.
1371 */
1372int follow_down(struct path *path)
1373{
1374        unsigned managed;
1375        int ret;
1376
1377        while (managed = ACCESS_ONCE(path->dentry->d_flags),
1378               unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1379                /* Allow the filesystem to manage the transit without i_mutex
1380                 * being held.
1381                 *
1382                 * We indicate to the filesystem if someone is trying to mount
1383                 * something here.  This gives autofs the chance to deny anyone
1384                 * other than its daemon the right to mount on its
1385                 * superstructure.
1386                 *
1387                 * The filesystem may sleep at this point.
1388                 */
1389                if (managed & DCACHE_MANAGE_TRANSIT) {
1390                        BUG_ON(!path->dentry->d_op);
1391                        BUG_ON(!path->dentry->d_op->d_manage);
1392                        ret = path->dentry->d_op->d_manage(
1393                                path->dentry, false);
1394                        if (ret < 0)
1395                                return ret == -EISDIR ? 0 : ret;
1396                }
1397
1398                /* Transit to a mounted filesystem. */
1399                if (managed & DCACHE_MOUNTED) {
1400                        struct vfsmount *mounted = lookup_mnt(path);
1401                        if (!mounted)
1402                                break;
1403                        dput(path->dentry);
1404                        mntput(path->mnt);
1405                        path->mnt = mounted;
1406                        path->dentry = dget(mounted->mnt_root);
1407                        continue;
1408                }
1409
1410                /* Don't handle automount points here */
1411                break;
1412        }
1413        return 0;
1414}
1415EXPORT_SYMBOL(follow_down);
1416
1417/*
1418 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1419 */
1420static void follow_mount(struct path *path)
1421{
1422        while (d_mountpoint(path->dentry)) {
1423                struct vfsmount *mounted = lookup_mnt(path);
1424                if (!mounted)
1425                        break;
1426                dput(path->dentry);
1427                mntput(path->mnt);
1428                path->mnt = mounted;
1429                path->dentry = dget(mounted->mnt_root);
1430        }
1431}
1432
1433static int path_parent_directory(struct path *path)
1434{
1435        struct dentry *old = path->dentry;
1436        /* rare case of legitimate dget_parent()... */
1437        path->dentry = dget_parent(path->dentry);
1438        dput(old);
1439        if (unlikely(!path_connected(path)))
1440                return -ENOENT;
1441        return 0;
1442}
1443
1444static int follow_dotdot(struct nameidata *nd)
1445{
1446        while(1) {
1447                if (nd->path.dentry == nd->root.dentry &&
1448                    nd->path.mnt == nd->root.mnt) {
1449                        break;
1450                }
1451                if (nd->path.dentry != nd->path.mnt->mnt_root) {
1452                        int ret = path_parent_directory(&nd->path);
1453                        if (ret)
1454                                return ret;
1455                        break;
1456                }
1457                if (!follow_up(&nd->path))
1458                        break;
1459        }
1460        follow_mount(&nd->path);
1461        nd->inode = nd->path.dentry->d_inode;
1462        return 0;
1463}
1464
1465/*
1466 * This looks up the name in dcache and possibly revalidates the found dentry.
1467 * NULL is returned if the dentry does not exist in the cache.
1468 */
1469static struct dentry *lookup_dcache(const struct qstr *name,
1470                                    struct dentry *dir,
1471                                    unsigned int flags)
1472{
1473        struct dentry *dentry;
1474        int error;
1475
1476        dentry = d_lookup(dir, name);
1477        if (dentry) {
1478                if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1479                        error = d_revalidate(dentry, flags);
1480                        if (unlikely(error <= 0)) {
1481                                if (!error)
1482                                        d_invalidate(dentry);
1483                                dput(dentry);
1484                                return ERR_PTR(error);
1485                        }
1486                }
1487        }
1488        return dentry;
1489}
1490
1491/*
1492 * Call i_op->lookup on the dentry.  The dentry must be negative and
1493 * unhashed.
1494 *
1495 * dir->d_inode->i_mutex must be held
1496 */
1497static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1498                                  unsigned int flags)
1499{
1500        struct dentry *old;
1501
1502        /* Don't create child dentry for a dead directory. */
1503        if (unlikely(IS_DEADDIR(dir))) {
1504                dput(dentry);
1505                return ERR_PTR(-ENOENT);
1506        }
1507
1508        old = dir->i_op->lookup(dir, dentry, flags);
1509        if (unlikely(old)) {
1510                dput(dentry);
1511                dentry = old;
1512        }
1513        return dentry;
1514}
1515
1516static struct dentry *__lookup_hash(const struct qstr *name,
1517                struct dentry *base, unsigned int flags)
1518{
1519        struct dentry *dentry = lookup_dcache(name, base, flags);
1520
1521        if (dentry)
1522                return dentry;
1523
1524        dentry = d_alloc(base, name);
1525        if (unlikely(!dentry))
1526                return ERR_PTR(-ENOMEM);
1527
1528        return lookup_real(base->d_inode, dentry, flags);
1529}
1530
1531static int lookup_fast(struct nameidata *nd,
1532                       struct path *path, struct inode **inode,
1533                       unsigned *seqp)
1534{
1535        struct vfsmount *mnt = nd->path.mnt;
1536        struct dentry *dentry, *parent = nd->path.dentry;
1537        int status = 1;
1538        int err;
1539
1540        /*
1541         * Rename seqlock is not required here because in the off chance
1542         * of a false negative due to a concurrent rename, the caller is
1543         * going to fall back to non-racy lookup.
1544         */
1545        if (nd->flags & LOOKUP_RCU) {
1546                unsigned seq;
1547                bool negative;
1548                dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1549                if (unlikely(!dentry)) {
1550                        if (unlazy_walk(nd, NULL, 0))
1551                                return -ECHILD;
1552                        return 0;
1553                }
1554
1555                /*
1556                 * This sequence count validates that the inode matches
1557                 * the dentry name information from lookup.
1558                 */
1559                *inode = d_backing_inode(dentry);
1560                negative = d_is_negative(dentry);
1561                if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1562                        return -ECHILD;
1563
1564                /*
1565                 * This sequence count validates that the parent had no
1566                 * changes while we did the lookup of the dentry above.
1567                 *
1568                 * The memory barrier in read_seqcount_begin of child is
1569                 *  enough, we can use __read_seqcount_retry here.
1570                 */
1571                if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1572                        return -ECHILD;
1573
1574                *seqp = seq;
1575                if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1576                        status = d_revalidate(dentry, nd->flags);
1577                if (unlikely(status <= 0)) {
1578                        if (unlazy_walk(nd, dentry, seq))
1579                                return -ECHILD;
1580                        if (status == -ECHILD)
1581                                status = d_revalidate(dentry, nd->flags);
1582                } else {
1583                        /*
1584                         * Note: do negative dentry check after revalidation in
1585                         * case that drops it.
1586                         */
1587                        if (unlikely(negative))
1588                                return -ENOENT;
1589                        path->mnt = mnt;
1590                        path->dentry = dentry;
1591                        if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1592                                return 1;
1593                        if (unlazy_walk(nd, dentry, seq))
1594                                return -ECHILD;
1595                }
1596        } else {
1597                dentry = __d_lookup(parent, &nd->last);
1598                if (unlikely(!dentry))
1599                        return 0;
1600                if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1601                        status = d_revalidate(dentry, nd->flags);
1602        }
1603        if (unlikely(status <= 0)) {
1604                if (!status)
1605                        d_invalidate(dentry);
1606                dput(dentry);
1607                return status;
1608        }
1609        if (unlikely(d_is_negative(dentry))) {
1610                dput(dentry);
1611                return -ENOENT;
1612        }
1613
1614        path->mnt = mnt;
1615        path->dentry = dentry;
1616        err = follow_managed(path, nd);
1617        if (likely(err > 0))
1618                *inode = d_backing_inode(path->dentry);
1619        return err;
1620}
1621
1622/* Fast lookup failed, do it the slow way */
1623static struct dentry *lookup_slow(const struct qstr *name,
1624                                  struct dentry *dir,
1625                                  unsigned int flags)
1626{
1627        struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1628        struct inode *inode = dir->d_inode;
1629        DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1630
1631        inode_lock_shared(inode);
1632        /* Don't go there if it's already dead */
1633        if (unlikely(IS_DEADDIR(inode)))
1634                goto out;
1635again:
1636        dentry = d_alloc_parallel(dir, name, &wq);
1637        if (IS_ERR(dentry))
1638                goto out;
1639        if (unlikely(!d_in_lookup(dentry))) {
1640                if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1641                    !(flags & LOOKUP_NO_REVAL)) {
1642                        int error = d_revalidate(dentry, flags);
1643                        if (unlikely(error <= 0)) {
1644                                if (!error) {
1645                                        d_invalidate(dentry);
1646                                        dput(dentry);
1647                                        goto again;
1648                                }
1649                                dput(dentry);
1650                                dentry = ERR_PTR(error);
1651                        }
1652                }
1653        } else {
1654                old = inode->i_op->lookup(inode, dentry, flags);
1655                d_lookup_done(dentry);
1656                if (unlikely(old)) {
1657                        dput(dentry);
1658                        dentry = old;
1659                }
1660        }
1661out:
1662        inode_unlock_shared(inode);
1663        return dentry;
1664}
1665
1666static inline int may_lookup(struct nameidata *nd)
1667{
1668        if (nd->flags & LOOKUP_RCU) {
1669                int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1670                if (err != -ECHILD)
1671                        return err;
1672                if (unlazy_walk(nd, NULL, 0))
1673                        return -ECHILD;
1674        }
1675        return inode_permission(nd->inode, MAY_EXEC);
1676}
1677
1678static inline int handle_dots(struct nameidata *nd, int type)
1679{
1680        if (type == LAST_DOTDOT) {
1681                if (!nd->root.mnt)
1682                        set_root(nd);
1683                if (nd->flags & LOOKUP_RCU) {
1684                        return follow_dotdot_rcu(nd);
1685                } else
1686                        return follow_dotdot(nd);
1687        }
1688        return 0;
1689}
1690
1691static int pick_link(struct nameidata *nd, struct path *link,
1692                     struct inode *inode, unsigned seq)
1693{
1694        int error;
1695        struct saved *last;
1696        if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1697                path_to_nameidata(link, nd);
1698                return -ELOOP;
1699        }
1700        if (!(nd->flags & LOOKUP_RCU)) {
1701                if (link->mnt == nd->path.mnt)
1702                        mntget(link->mnt);
1703        }
1704        error = nd_alloc_stack(nd);
1705        if (unlikely(error)) {
1706                if (error == -ECHILD) {
1707                        if (unlikely(unlazy_link(nd, link, seq)))
1708                                return -ECHILD;
1709                        error = nd_alloc_stack(nd);
1710                }
1711                if (error) {
1712                        path_put(link);
1713                        return error;
1714                }
1715        }
1716
1717        last = nd->stack + nd->depth++;
1718        last->link = *link;
1719        clear_delayed_call(&last->done);
1720        nd->link_inode = inode;
1721        last->seq = seq;
1722        return 1;
1723}
1724
1725/*
1726 * Do we need to follow links? We _really_ want to be able
1727 * to do this check without having to look at inode->i_op,
1728 * so we keep a cache of "no, this doesn't need follow_link"
1729 * for the common case.
1730 */
1731static inline int should_follow_link(struct nameidata *nd, struct path *link,
1732                                     int follow,
1733                                     struct inode *inode, unsigned seq)
1734{
1735        if (likely(!d_is_symlink(link->dentry)))
1736                return 0;
1737        if (!follow)
1738                return 0;
1739        /* make sure that d_is_symlink above matches inode */
1740        if (nd->flags & LOOKUP_RCU) {
1741                if (read_seqcount_retry(&link->dentry->d_seq, seq))
1742                        return -ECHILD;
1743        }
1744        return pick_link(nd, link, inode, seq);
1745}
1746
1747enum {WALK_GET = 1, WALK_PUT = 2};
1748
1749static int walk_component(struct nameidata *nd, int flags)
1750{
1751        struct path path;
1752        struct inode *inode;
1753        unsigned seq;
1754        int err;
1755        /*
1756         * "." and ".." are special - ".." especially so because it has
1757         * to be able to know about the current root directory and
1758         * parent relationships.
1759         */
1760        if (unlikely(nd->last_type != LAST_NORM)) {
1761                err = handle_dots(nd, nd->last_type);
1762                if (flags & WALK_PUT)
1763                        put_link(nd);
1764                return err;
1765        }
1766        err = lookup_fast(nd, &path, &inode, &seq);
1767        if (unlikely(err <= 0)) {
1768                if (err < 0)
1769                        return err;
1770                path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1771                                          nd->flags);
1772                if (IS_ERR(path.dentry))
1773                        return PTR_ERR(path.dentry);
1774
1775                path.mnt = nd->path.mnt;
1776                err = follow_managed(&path, nd);
1777                if (unlikely(err < 0))
1778                        return err;
1779
1780                if (unlikely(d_is_negative(path.dentry))) {
1781                        path_to_nameidata(&path, nd);
1782                        return -ENOENT;
1783                }
1784
1785                seq = 0;        /* we are already out of RCU mode */
1786                inode = d_backing_inode(path.dentry);
1787        }
1788
1789        if (flags & WALK_PUT)
1790                put_link(nd);
1791        err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1792        if (unlikely(err))
1793                return err;
1794        path_to_nameidata(&path, nd);
1795        nd->inode = inode;
1796        nd->seq = seq;
1797        return 0;
1798}
1799
1800/*
1801 * We can do the critical dentry name comparison and hashing
1802 * operations one word at a time, but we are limited to:
1803 *
1804 * - Architectures with fast unaligned word accesses. We could
1805 *   do a "get_unaligned()" if this helps and is sufficiently
1806 *   fast.
1807 *
1808 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1809 *   do not trap on the (extremely unlikely) case of a page
1810 *   crossing operation.
1811 *
1812 * - Furthermore, we need an efficient 64-bit compile for the
1813 *   64-bit case in order to generate the "number of bytes in
1814 *   the final mask". Again, that could be replaced with a
1815 *   efficient population count instruction or similar.
1816 */
1817#ifdef CONFIG_DCACHE_WORD_ACCESS
1818
1819#include <asm/word-at-a-time.h>
1820
1821#ifdef HASH_MIX
1822
1823/* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1824
1825#elif defined(CONFIG_64BIT)
1826/*
1827 * Register pressure in the mixing function is an issue, particularly
1828 * on 32-bit x86, but almost any function requires one state value and
1829 * one temporary.  Instead, use a function designed for two state values
1830 * and no temporaries.
1831 *
1832 * This function cannot create a collision in only two iterations, so
1833 * we have two iterations to achieve avalanche.  In those two iterations,
1834 * we have six layers of mixing, which is enough to spread one bit's
1835 * influence out to 2^6 = 64 state bits.
1836 *
1837 * Rotate constants are scored by considering either 64 one-bit input
1838 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1839 * probability of that delta causing a change to each of the 128 output
1840 * bits, using a sample of random initial states.
1841 *
1842 * The Shannon entropy of the computed probabilities is then summed
1843 * to produce a score.  Ideally, any input change has a 50% chance of
1844 * toggling any given output bit.
1845 *
1846 * Mixing scores (in bits) for (12,45):
1847 * Input delta: 1-bit      2-bit
1848 * 1 round:     713.3    42542.6
1849 * 2 rounds:   2753.7   140389.8
1850 * 3 rounds:   5954.1   233458.2
1851 * 4 rounds:   7862.6   256672.2
1852 * Perfect:    8192     258048
1853 *            (64*128) (64*63/2 * 128)
1854 */
1855#define HASH_MIX(x, y, a)       \
1856        (       x ^= (a),       \
1857        y ^= x, x = rol64(x,12),\
1858        x += y, y = rol64(y,45),\
1859        y *= 9                  )
1860
1861/*
1862 * Fold two longs into one 32-bit hash value.  This must be fast, but
1863 * latency isn't quite as critical, as there is a fair bit of additional
1864 * work done before the hash value is used.
1865 */
1866static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1867{
1868        y ^= x * GOLDEN_RATIO_64;
1869        y *= GOLDEN_RATIO_64;
1870        return y >> 32;
1871}
1872
1873#else   /* 32-bit case */
1874
1875/*
1876 * Mixing scores (in bits) for (7,20):
1877 * Input delta: 1-bit      2-bit
1878 * 1 round:     330.3     9201.6
1879 * 2 rounds:   1246.4    25475.4
1880 * 3 rounds:   1907.1    31295.1
1881 * 4 rounds:   2042.3    31718.6
1882 * Perfect:    2048      31744
1883 *            (32*64)   (32*31/2 * 64)
1884 */
1885#define HASH_MIX(x, y, a)       \
1886        (       x ^= (a),       \
1887        y ^= x, x = rol32(x, 7),\
1888        x += y, y = rol32(y,20),\
1889        y *= 9                  )
1890
1891static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1892{
1893        /* Use arch-optimized multiply if one exists */
1894        return __hash_32(y ^ __hash_32(x));
1895}
1896
1897#endif
1898
1899/*
1900 * Return the hash of a string of known length.  This is carfully
1901 * designed to match hash_name(), which is the more critical function.
1902 * In particular, we must end by hashing a final word containing 0..7
1903 * payload bytes, to match the way that hash_name() iterates until it
1904 * finds the delimiter after the name.
1905 */
1906unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1907{
1908        unsigned long a, x = 0, y = (unsigned long)salt;
1909
1910        for (;;) {
1911                if (!len)
1912                        goto done;
1913                a = load_unaligned_zeropad(name);
1914                if (len < sizeof(unsigned long))
1915                        break;
1916                HASH_MIX(x, y, a);
1917                name += sizeof(unsigned long);
1918                len -= sizeof(unsigned long);
1919        }
1920        x ^= a & bytemask_from_count(len);
1921done:
1922        return fold_hash(x, y);
1923}
1924EXPORT_SYMBOL(full_name_hash);
1925
1926/* Return the "hash_len" (hash and length) of a null-terminated string */
1927u64 hashlen_string(const void *salt, const char *name)
1928{
1929        unsigned long a = 0, x = 0, y = (unsigned long)salt;
1930        unsigned long adata, mask, len;
1931        const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1932
1933        len = 0;
1934        goto inside;
1935
1936        do {
1937                HASH_MIX(x, y, a);
1938                len += sizeof(unsigned long);
1939inside:
1940                a = load_unaligned_zeropad(name+len);
1941        } while (!has_zero(a, &adata, &constants));
1942
1943        adata = prep_zero_mask(a, adata, &constants);
1944        mask = create_zero_mask(adata);
1945        x ^= a & zero_bytemask(mask);
1946
1947        return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1948}
1949EXPORT_SYMBOL(hashlen_string);
1950
1951/*
1952 * Calculate the length and hash of the path component, and
1953 * return the "hash_len" as the result.
1954 */
1955static inline u64 hash_name(const void *salt, const char *name)
1956{
1957        unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1958        unsigned long adata, bdata, mask, len;
1959        const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1960
1961        len = 0;
1962        goto inside;
1963
1964        do {
1965                HASH_MIX(x, y, a);
1966                len += sizeof(unsigned long);
1967inside:
1968                a = load_unaligned_zeropad(name+len);
1969                b = a ^ REPEAT_BYTE('/');
1970        } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1971
1972        adata = prep_zero_mask(a, adata, &constants);
1973        bdata = prep_zero_mask(b, bdata, &constants);
1974        mask = create_zero_mask(adata | bdata);
1975        x ^= a & zero_bytemask(mask);
1976
1977        return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1978}
1979
1980#else   /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1981
1982/* Return the hash of a string of known length */
1983unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1984{
1985        unsigned long hash = init_name_hash(salt);
1986        while (len--)
1987                hash = partial_name_hash((unsigned char)*name++, hash);
1988        return end_name_hash(hash);
1989}
1990EXPORT_SYMBOL(full_name_hash);
1991
1992/* Return the "hash_len" (hash and length) of a null-terminated string */
1993u64 hashlen_string(const void *salt, const char *name)
1994{
1995        unsigned long hash = init_name_hash(salt);
1996        unsigned long len = 0, c;
1997
1998        c = (unsigned char)*name;
1999        while (c) {
2000                len++;
2001                hash = partial_name_hash(c, hash);
2002                c = (unsigned char)name[len];
2003        }
2004        return hashlen_create(end_name_hash(hash), len);
2005}
2006EXPORT_SYMBOL(hashlen_string);
2007
2008/*
2009 * We know there's a real path component here of at least
2010 * one character.
2011 */
2012static inline u64 hash_name(const void *salt, const char *name)
2013{
2014        unsigned long hash = init_name_hash(salt);
2015        unsigned long len = 0, c;
2016
2017        c = (unsigned char)*name;
2018        do {
2019                len++;
2020                hash = partial_name_hash(c, hash);
2021                c = (unsigned char)name[len];
2022        } while (c && c != '/');
2023        return hashlen_create(end_name_hash(hash), len);
2024}
2025
2026#endif
2027
2028/*
2029 * Name resolution.
2030 * This is the basic name resolution function, turning a pathname into
2031 * the final dentry. We expect 'base' to be positive and a directory.
2032 *
2033 * Returns 0 and nd will have valid dentry and mnt on success.
2034 * Returns error and drops reference to input namei data on failure.
2035 */
2036static int link_path_walk(const char *name, struct nameidata *nd)
2037{
2038        int err;
2039
2040        while (*name=='/')
2041                name++;
2042        if (!*name)
2043                return 0;
2044
2045        /* At this point we know we have a real path component. */
2046        for(;;) {
2047                u64 hash_len;
2048                int type;
2049
2050                err = may_lookup(nd);
2051                if (err)
2052                        return err;
2053
2054                hash_len = hash_name(nd->path.dentry, name);
2055
2056                type = LAST_NORM;
2057                if (name[0] == '.') switch (hashlen_len(hash_len)) {
2058                        case 2:
2059                                if (name[1] == '.') {
2060                                        type = LAST_DOTDOT;
2061                                        nd->flags |= LOOKUP_JUMPED;
2062                                }
2063                                break;
2064                        case 1:
2065                                type = LAST_DOT;
2066                }
2067                if (likely(type == LAST_NORM)) {
2068                        struct dentry *parent = nd->path.dentry;
2069                        nd->flags &= ~LOOKUP_JUMPED;
2070                        if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2071                                struct qstr this = { { .hash_len = hash_len }, .name = name };
2072                                err = parent->d_op->d_hash(parent, &this);
2073                                if (err < 0)
2074                                        return err;
2075                                hash_len = this.hash_len;
2076                                name = this.name;
2077                        }
2078                }
2079
2080                nd->last.hash_len = hash_len;
2081                nd->last.name = name;
2082                nd->last_type = type;
2083
2084                name += hashlen_len(hash_len);
2085                if (!*name)
2086                        goto OK;
2087                /*
2088                 * If it wasn't NUL, we know it was '/'. Skip that
2089                 * slash, and continue until no more slashes.
2090                 */
2091                do {
2092                        name++;
2093                } while (unlikely(*name == '/'));
2094                if (unlikely(!*name)) {
2095OK:
2096                        /* pathname body, done */
2097                        if (!nd->depth)
2098                                return 0;
2099                        name = nd->stack[nd->depth - 1].name;
2100                        /* trailing symlink, done */
2101                        if (!name)
2102                                return 0;
2103                        /* last component of nested symlink */
2104                        err = walk_component(nd, WALK_GET | WALK_PUT);
2105                } else {
2106                        err = walk_component(nd, WALK_GET);
2107                }
2108                if (err < 0)
2109                        return err;
2110
2111                if (err) {
2112                        const char *s = get_link(nd);
2113
2114                        if (IS_ERR(s))
2115                                return PTR_ERR(s);
2116                        err = 0;
2117                        if (unlikely(!s)) {
2118                                /* jumped */
2119                                put_link(nd);
2120                        } else {
2121                                nd->stack[nd->depth - 1].name = name;
2122                                name = s;
2123                                continue;
2124                        }
2125                }
2126                if (unlikely(!d_can_lookup(nd->path.dentry))) {
2127                        if (nd->flags & LOOKUP_RCU) {
2128                                if (unlazy_walk(nd, NULL, 0))
2129                                        return -ECHILD;
2130                        }
2131                        return -ENOTDIR;
2132                }
2133        }
2134}
2135
2136static const char *path_init(struct nameidata *nd, unsigned flags)
2137{
2138        int retval = 0;
2139        const char *s = nd->name->name;
2140
2141        nd->last_type = LAST_ROOT; /* if there are only slashes... */
2142        nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2143        nd->depth = 0;
2144        if (flags & LOOKUP_ROOT) {
2145                struct dentry *root = nd->root.dentry;
2146                struct inode *inode = root->d_inode;
2147                if (*s) {
2148                        if (!d_can_lookup(root))
2149                                return ERR_PTR(-ENOTDIR);
2150                        retval = inode_permission(inode, MAY_EXEC);
2151                        if (retval)
2152                                return ERR_PTR(retval);
2153                }
2154                nd->path = nd->root;
2155                nd->inode = inode;
2156                if (flags & LOOKUP_RCU) {
2157                        rcu_read_lock();
2158                        nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2159                        nd->root_seq = nd->seq;
2160                        nd->m_seq = read_seqbegin(&mount_lock);
2161                } else {
2162                        path_get(&nd->path);
2163                }
2164                return s;
2165        }
2166
2167        nd->root.mnt = NULL;
2168        nd->path.mnt = NULL;
2169        nd->path.dentry = NULL;
2170
2171        nd->m_seq = read_seqbegin(&mount_lock);
2172        if (*s == '/') {
2173                if (flags & LOOKUP_RCU)
2174                        rcu_read_lock();
2175                set_root(nd);
2176                if (likely(!nd_jump_root(nd)))
2177                        return s;
2178                nd->root.mnt = NULL;
2179                rcu_read_unlock();
2180                return ERR_PTR(-ECHILD);
2181        } else if (nd->dfd == AT_FDCWD) {
2182                if (flags & LOOKUP_RCU) {
2183                        struct fs_struct *fs = current->fs;
2184                        unsigned seq;
2185
2186                        rcu_read_lock();
2187
2188                        do {
2189                                seq = read_seqcount_begin(&fs->seq);
2190                                nd->path = fs->pwd;
2191                                nd->inode = nd->path.dentry->d_inode;
2192                                nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2193                        } while (read_seqcount_retry(&fs->seq, seq));
2194                } else {
2195                        get_fs_pwd(current->fs, &nd->path);
2196                        nd->inode = nd->path.dentry->d_inode;
2197                }
2198                return s;
2199        } else {
2200                /* Caller must check execute permissions on the starting path component */
2201                struct fd f = fdget_raw(nd->dfd);
2202                struct dentry *dentry;
2203
2204                if (!f.file)
2205                        return ERR_PTR(-EBADF);
2206
2207                dentry = f.file->f_path.dentry;
2208
2209                if (*s) {
2210                        if (!d_can_lookup(dentry)) {
2211                                fdput(f);
2212                                return ERR_PTR(-ENOTDIR);
2213                        }
2214                }
2215
2216                nd->path = f.file->f_path;
2217                if (flags & LOOKUP_RCU) {
2218                        rcu_read_lock();
2219                        nd->inode = nd->path.dentry->d_inode;
2220                        nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2221                } else {
2222                        path_get(&nd->path);
2223                        nd->inode = nd->path.dentry->d_inode;
2224                }
2225                fdput(f);
2226                return s;
2227        }
2228}
2229
2230static const char *trailing_symlink(struct nameidata *nd)
2231{
2232        const char *s;
2233        int error = may_follow_link(nd);
2234        if (unlikely(error))
2235                return ERR_PTR(error);
2236        nd->flags |= LOOKUP_PARENT;
2237        nd->stack[0].name = NULL;
2238        s = get_link(nd);
2239        return s ? s : "";
2240}
2241
2242static inline int lookup_last(struct nameidata *nd)
2243{
2244        if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2245                nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2246
2247        nd->flags &= ~LOOKUP_PARENT;
2248        return walk_component(nd,
2249                        nd->flags & LOOKUP_FOLLOW
2250                                ? nd->depth
2251                                        ? WALK_PUT | WALK_GET
2252                                        : WALK_GET
2253                                : 0);
2254}
2255
2256/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2257static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2258{
2259        const char *s = path_init(nd, flags);
2260        int err;
2261
2262        if (IS_ERR(s))
2263                return PTR_ERR(s);
2264        while (!(err = link_path_walk(s, nd))
2265                && ((err = lookup_last(nd)) > 0)) {
2266                s = trailing_symlink(nd);
2267                if (IS_ERR(s)) {
2268                        err = PTR_ERR(s);
2269                        break;
2270                }
2271        }
2272        if (!err)
2273                err = complete_walk(nd);
2274
2275        if (!err && nd->flags & LOOKUP_DIRECTORY)
2276                if (!d_can_lookup(nd->path.dentry))
2277                        err = -ENOTDIR;
2278        if (!err) {
2279                *path = nd->path;
2280                nd->path.mnt = NULL;
2281                nd->path.dentry = NULL;
2282        }
2283        terminate_walk(nd);
2284        return err;
2285}
2286
2287static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2288                           struct path *path, struct path *root)
2289{
2290        int retval;
2291        struct nameidata nd;
2292        if (IS_ERR(name))
2293                return PTR_ERR(name);
2294        if (unlikely(root)) {
2295                nd.root = *root;
2296                flags |= LOOKUP_ROOT;
2297        }
2298        set_nameidata(&nd, dfd, name);
2299        retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2300        if (unlikely(retval == -ECHILD))
2301                retval = path_lookupat(&nd, flags, path);
2302        if (unlikely(retval == -ESTALE))
2303                retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2304
2305        if (likely(!retval))
2306                audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2307        restore_nameidata();
2308        putname(name);
2309        return retval;
2310}
2311
2312/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2313static int path_parentat(struct nameidata *nd, unsigned flags,
2314                                struct path *parent)
2315{
2316        const char *s = path_init(nd, flags);
2317        int err;
2318        if (IS_ERR(s))
2319                return PTR_ERR(s);
2320        err = link_path_walk(s, nd);
2321        if (!err)
2322                err = complete_walk(nd);
2323        if (!err) {
2324                *parent = nd->path;
2325                nd->path.mnt = NULL;
2326                nd->path.dentry = NULL;
2327        }
2328        terminate_walk(nd);
2329        return err;
2330}
2331
2332static struct filename *filename_parentat(int dfd, struct filename *name,
2333                                unsigned int flags, struct path *parent,
2334                                struct qstr *last, int *type)
2335{
2336        int retval;
2337        struct nameidata nd;
2338
2339        if (IS_ERR(name))
2340                return name;
2341        set_nameidata(&nd, dfd, name);
2342        retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2343        if (unlikely(retval == -ECHILD))
2344                retval = path_parentat(&nd, flags, parent);
2345        if (unlikely(retval == -ESTALE))
2346                retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2347        if (likely(!retval)) {
2348                *last = nd.last;
2349                *type = nd.last_type;
2350                audit_inode(name, parent->dentry, LOOKUP_PARENT);
2351        } else {
2352                putname(name);
2353                name = ERR_PTR(retval);
2354        }
2355        restore_nameidata();
2356        return name;
2357}
2358
2359/* does lookup, returns the object with parent locked */
2360struct dentry *kern_path_locked(const char *name, struct path *path)
2361{
2362        struct filename *filename;
2363        struct dentry *d;
2364        struct qstr last;
2365        int type;
2366
2367        filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2368                                    &last, &type);
2369        if (IS_ERR(filename))
2370                return ERR_CAST(filename);
2371        if (unlikely(type != LAST_NORM)) {
2372                path_put(path);
2373                putname(filename);
2374                return ERR_PTR(-EINVAL);
2375        }
2376        inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2377        d = __lookup_hash(&last, path->dentry, 0);
2378        if (IS_ERR(d)) {
2379                inode_unlock(path->dentry->d_inode);
2380                path_put(path);
2381        }
2382        putname(filename);
2383        return d;
2384}
2385
2386int kern_path(const char *name, unsigned int flags, struct path *path)
2387{
2388        return filename_lookup(AT_FDCWD, getname_kernel(name),
2389                               flags, path, NULL);
2390}
2391EXPORT_SYMBOL(kern_path);
2392
2393/**
2394 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2395 * @dentry:  pointer to dentry of the base directory
2396 * @mnt: pointer to vfs mount of the base directory
2397 * @name: pointer to file name
2398 * @flags: lookup flags
2399 * @path: pointer to struct path to fill
2400 */
2401int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2402                    const char *name, unsigned int flags,
2403                    struct path *path)
2404{
2405        struct path root = {.mnt = mnt, .dentry = dentry};
2406        /* the first argument of filename_lookup() is ignored with root */
2407        return filename_lookup(AT_FDCWD, getname_kernel(name),
2408                               flags , path, &root);
2409}
2410EXPORT_SYMBOL(vfs_path_lookup);
2411
2412/**
2413 * lookup_one_len - filesystem helper to lookup single pathname component
2414 * @name:       pathname component to lookup
2415 * @base:       base directory to lookup from
2416 * @len:        maximum length @len should be interpreted to
2417 *
2418 * Note that this routine is purely a helper for filesystem usage and should
2419 * not be called by generic code.
2420 *
2421 * The caller must hold base->i_mutex.
2422 */
2423struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2424{
2425        struct qstr this;
2426        unsigned int c;
2427        int err;
2428
2429        WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2430
2431        this.name = name;
2432        this.len = len;
2433        this.hash = full_name_hash(base, name, len);
2434        if (!len)
2435                return ERR_PTR(-EACCES);
2436
2437        if (unlikely(name[0] == '.')) {
2438                if (len < 2 || (len == 2 && name[1] == '.'))
2439                        return ERR_PTR(-EACCES);
2440        }
2441
2442        while (len--) {
2443                c = *(const unsigned char *)name++;
2444                if (c == '/' || c == '\0')
2445                        return ERR_PTR(-EACCES);
2446        }
2447        /*
2448         * See if the low-level filesystem might want
2449         * to use its own hash..
2450         */
2451        if (base->d_flags & DCACHE_OP_HASH) {
2452                int err = base->d_op->d_hash(base, &this);
2453                if (err < 0)
2454                        return ERR_PTR(err);
2455        }
2456
2457        err = inode_permission(base->d_inode, MAY_EXEC);
2458        if (err)
2459                return ERR_PTR(err);
2460
2461        return __lookup_hash(&this, base, 0);
2462}
2463EXPORT_SYMBOL(lookup_one_len);
2464
2465/**
2466 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2467 * @name:       pathname component to lookup
2468 * @base:       base directory to lookup from
2469 * @len:        maximum length @len should be interpreted to
2470 *
2471 * Note that this routine is purely a helper for filesystem usage and should
2472 * not be called by generic code.
2473 *
2474 * Unlike lookup_one_len, it should be called without the parent
2475 * i_mutex held, and will take the i_mutex itself if necessary.
2476 */
2477struct dentry *lookup_one_len_unlocked(const char *name,
2478                                       struct dentry *base, int len)
2479{
2480        struct qstr this;
2481        unsigned int c;
2482        int err;
2483        struct dentry *ret;
2484
2485        this.name = name;
2486        this.len = len;
2487        this.hash = full_name_hash(base, name, len);
2488        if (!len)
2489                return ERR_PTR(-EACCES);
2490
2491        if (unlikely(name[0] == '.')) {
2492                if (len < 2 || (len == 2 && name[1] == '.'))
2493                        return ERR_PTR(-EACCES);
2494        }
2495
2496        while (len--) {
2497                c = *(const unsigned char *)name++;
2498                if (c == '/' || c == '\0')
2499                        return ERR_PTR(-EACCES);
2500        }
2501        /*
2502         * See if the low-level filesystem might want
2503         * to use its own hash..
2504         */
2505        if (base->d_flags & DCACHE_OP_HASH) {
2506                int err = base->d_op->d_hash(base, &this);
2507                if (err < 0)
2508                        return ERR_PTR(err);
2509        }
2510
2511        err = inode_permission(base->d_inode, MAY_EXEC);
2512        if (err)
2513                return ERR_PTR(err);
2514
2515        ret = lookup_dcache(&this, base, 0);
2516        if (!ret)
2517                ret = lookup_slow(&this, base, 0);
2518        return ret;
2519}
2520EXPORT_SYMBOL(lookup_one_len_unlocked);
2521
2522#ifdef CONFIG_UNIX98_PTYS
2523int path_pts(struct path *path)
2524{
2525        /* Find something mounted on "pts" in the same directory as
2526         * the input path.
2527         */
2528        struct dentry *child, *parent;
2529        struct qstr this;
2530        int ret;
2531
2532        ret = path_parent_directory(path);
2533        if (ret)
2534                return ret;
2535
2536        parent = path->dentry;
2537        this.name = "pts";
2538        this.len = 3;
2539        child = d_hash_and_lookup(parent, &this);
2540        if (!child)
2541                return -ENOENT;
2542
2543        path->dentry = child;
2544        dput(parent);
2545        follow_mount(path);
2546        return 0;
2547}
2548#endif
2549
2550int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2551                 struct path *path, int *empty)
2552{
2553        return filename_lookup(dfd, getname_flags(name, flags, empty),
2554                               flags, path, NULL);
2555}
2556EXPORT_SYMBOL(user_path_at_empty);
2557
2558/*
2559 * NB: most callers don't do anything directly with the reference to the
2560 *     to struct filename, but the nd->last pointer points into the name string
2561 *     allocated by getname. So we must hold the reference to it until all
2562 *     path-walking is complete.
2563 */
2564static inline struct filename *
2565user_path_parent(int dfd, const char __user *path,
2566                 struct path *parent,
2567                 struct qstr *last,
2568                 int *type,
2569                 unsigned int flags)
2570{
2571        /* only LOOKUP_REVAL is allowed in extra flags */
2572        return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2573                                 parent, last, type);
2574}
2575
2576/**
2577 * mountpoint_last - look up last component for umount
2578 * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2579 * @path: pointer to container for result
2580 *
2581 * This is a special lookup_last function just for umount. In this case, we
2582 * need to resolve the path without doing any revalidation.
2583 *
2584 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2585 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2586 * in almost all cases, this lookup will be served out of the dcache. The only
2587 * cases where it won't are if nd->last refers to a symlink or the path is
2588 * bogus and it doesn't exist.
2589 *
2590 * Returns:
2591 * -error: if there was an error during lookup. This includes -ENOENT if the
2592 *         lookup found a negative dentry. The nd->path reference will also be
2593 *         put in this case.
2594 *
2595 * 0:      if we successfully resolved nd->path and found it to not to be a
2596 *         symlink that needs to be followed. "path" will also be populated.
2597 *         The nd->path reference will also be put.
2598 *
2599 * 1:      if we successfully resolved nd->last and found it to be a symlink
2600 *         that needs to be followed. "path" will be populated with the path
2601 *         to the link, and nd->path will *not* be put.
2602 */
2603static int
2604mountpoint_last(struct nameidata *nd, struct path *path)
2605{
2606        int error = 0;
2607        struct dentry *dentry;
2608        struct dentry *dir = nd->path.dentry;
2609
2610        /* If we're in rcuwalk, drop out of it to handle last component */
2611        if (nd->flags & LOOKUP_RCU) {
2612                if (unlazy_walk(nd, NULL, 0))
2613                        return -ECHILD;
2614        }
2615
2616        nd->flags &= ~LOOKUP_PARENT;
2617
2618        if (unlikely(nd->last_type != LAST_NORM)) {
2619                error = handle_dots(nd, nd->last_type);
2620                if (error)
2621                        return error;
2622                dentry = dget(nd->path.dentry);
2623        } else {
2624                dentry = d_lookup(dir, &nd->last);
2625                if (!dentry) {
2626                        /*
2627                         * No cached dentry. Mounted dentries are pinned in the
2628                         * cache, so that means that this dentry is probably
2629                         * a symlink or the path doesn't actually point
2630                         * to a mounted dentry.
2631                         */
2632                        dentry = lookup_slow(&nd->last, dir,
2633                                             nd->flags | LOOKUP_NO_REVAL);
2634                        if (IS_ERR(dentry))
2635                                return PTR_ERR(dentry);
2636                }
2637        }
2638        if (d_is_negative(dentry)) {
2639                dput(dentry);
2640                return -ENOENT;
2641        }
2642        if (nd->depth)
2643                put_link(nd);
2644        path->dentry = dentry;
2645        path->mnt = nd->path.mnt;
2646        error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2647                                   d_backing_inode(dentry), 0);
2648        if (unlikely(error))
2649                return error;
2650        mntget(path->mnt);
2651        follow_mount(path);
2652        return 0;
2653}
2654
2655/**
2656 * path_mountpoint - look up a path to be umounted
2657 * @nd:         lookup context
2658 * @flags:      lookup flags
2659 * @path:       pointer to container for result
2660 *
2661 * Look up the given name, but don't attempt to revalidate the last component.
2662 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2663 */
2664static int
2665path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2666{
2667        const char *s = path_init(nd, flags);
2668        int err;
2669        if (IS_ERR(s))
2670                return PTR_ERR(s);
2671        while (!(err = link_path_walk(s, nd)) &&
2672                (err = mountpoint_last(nd, path)) > 0) {
2673                s = trailing_symlink(nd);
2674                if (IS_ERR(s)) {
2675                        err = PTR_ERR(s);
2676                        break;
2677                }
2678        }
2679        terminate_walk(nd);
2680        return err;
2681}
2682
2683static int
2684filename_mountpoint(int dfd, struct filename *name, struct path *path,
2685                        unsigned int flags)
2686{
2687        struct nameidata nd;
2688        int error;
2689        if (IS_ERR(name))
2690                return PTR_ERR(name);
2691        set_nameidata(&nd, dfd, name);
2692        error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2693        if (unlikely(error == -ECHILD))
2694                error = path_mountpoint(&nd, flags, path);
2695        if (unlikely(error == -ESTALE))
2696                error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2697        if (likely(!error))
2698                audit_inode(name, path->dentry, 0);
2699        restore_nameidata();
2700        putname(name);
2701        return error;
2702}
2703
2704/**
2705 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2706 * @dfd:        directory file descriptor
2707 * @name:       pathname from userland
2708 * @flags:      lookup flags
2709 * @path:       pointer to container to hold result
2710 *
2711 * A umount is a special case for path walking. We're not actually interested
2712 * in the inode in this situation, and ESTALE errors can be a problem. We
2713 * simply want track down the dentry and vfsmount attached at the mountpoint
2714 * and avoid revalidating the last component.
2715 *
2716 * Returns 0 and populates "path" on success.
2717 */
2718int
2719user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2720                        struct path *path)
2721{
2722        return filename_mountpoint(dfd, getname(name), path, flags);
2723}
2724
2725int
2726kern_path_mountpoint(int dfd, const char *name, struct path *path,
2727                        unsigned int flags)
2728{
2729        return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2730}
2731EXPORT_SYMBOL(kern_path_mountpoint);
2732
2733int __check_sticky(struct inode *dir, struct inode *inode)
2734{
2735        kuid_t fsuid = current_fsuid();
2736
2737        if (uid_eq(inode->i_uid, fsuid))
2738                return 0;
2739        if (uid_eq(dir->i_uid, fsuid))
2740                return 0;
2741        return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2742}
2743EXPORT_SYMBOL(__check_sticky);
2744
2745/*
2746 *      Check whether we can remove a link victim from directory dir, check
2747 *  whether the type of victim is right.
2748 *  1. We can't do it if dir is read-only (done in permission())
2749 *  2. We should have write and exec permissions on dir
2750 *  3. We can't remove anything from append-only dir
2751 *  4. We can't do anything with immutable dir (done in permission())
2752 *  5. If the sticky bit on dir is set we should either
2753 *      a. be owner of dir, or
2754 *      b. be owner of victim, or
2755 *      c. have CAP_FOWNER capability
2756 *  6. If the victim is append-only or immutable we can't do antyhing with
2757 *     links pointing to it.
2758 *  7. If the victim has an unknown uid or gid we can't change the inode.
2759 *  8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2760 *  9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2761 * 10. We can't remove a root or mountpoint.
2762 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2763 *     nfs_async_unlink().
2764 */
2765static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2766{
2767        struct inode *inode = d_backing_inode(victim);
2768        int error;
2769
2770        if (d_is_negative(victim))
2771                return -ENOENT;
2772        BUG_ON(!inode);
2773
2774        BUG_ON(victim->d_parent->d_inode != dir);
2775        audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2776
2777        error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2778        if (error)
2779                return error;
2780        if (IS_APPEND(dir))
2781                return -EPERM;
2782
2783        if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2784            IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2785                return -EPERM;
2786        if (isdir) {
2787                if (!d_is_dir(victim))
2788                        return -ENOTDIR;
2789                if (IS_ROOT(victim))
2790                        return -EBUSY;
2791        } else if (d_is_dir(victim))
2792                return -EISDIR;
2793        if (IS_DEADDIR(dir))
2794                return -ENOENT;
2795        if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2796                return -EBUSY;
2797        return 0;
2798}
2799
2800/*      Check whether we can create an object with dentry child in directory
2801 *  dir.
2802 *  1. We can't do it if child already exists (open has special treatment for
2803 *     this case, but since we are inlined it's OK)
2804 *  2. We can't do it if dir is read-only (done in permission())
2805 *  3. We can't do it if the fs can't represent the fsuid or fsgid.
2806 *  4. We should have write and exec permissions on dir
2807 *  5. We can't do it if dir is immutable (done in permission())
2808 */
2809static inline int may_create(struct inode *dir, struct dentry *child)
2810{
2811        struct user_namespace *s_user_ns;
2812        audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2813        if (child->d_inode)
2814                return -EEXIST;
2815        if (IS_DEADDIR(dir))
2816                return -ENOENT;
2817        s_user_ns = dir->i_sb->s_user_ns;
2818        if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2819            !kgid_has_mapping(s_user_ns, current_fsgid()))
2820                return -EOVERFLOW;
2821        return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2822}
2823
2824/*
2825 * p1 and p2 should be directories on the same fs.
2826 */
2827struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2828{
2829        struct dentry *p;
2830
2831        if (p1 == p2) {
2832                inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2833                return NULL;
2834        }
2835
2836        mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2837
2838        p = d_ancestor(p2, p1);
2839        if (p) {
2840                inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2841                inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2842                return p;
2843        }
2844
2845        p = d_ancestor(p1, p2);
2846        if (p) {
2847                inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2848                inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2849                return p;
2850        }
2851
2852        inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2853        inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2854        return NULL;
2855}
2856EXPORT_SYMBOL(lock_rename);
2857
2858void unlock_rename(struct dentry *p1, struct dentry *p2)
2859{
2860        inode_unlock(p1->d_inode);
2861        if (p1 != p2) {
2862                inode_unlock(p2->d_inode);
2863                mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2864        }
2865}
2866EXPORT_SYMBOL(unlock_rename);
2867
2868int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2869                bool want_excl)
2870{
2871        int error = may_create(dir, dentry);
2872        if (error)
2873                return error;
2874
2875        if (!dir->i_op->create)
2876                return -EACCES; /* shouldn't it be ENOSYS? */
2877        mode &= S_IALLUGO;
2878        mode |= S_IFREG;
2879        error = security_inode_create(dir, dentry, mode);
2880        if (error)
2881                return error;
2882        error = dir->i_op->create(dir, dentry, mode, want_excl);
2883        if (!error)
2884                fsnotify_create(dir, dentry);
2885        return error;
2886}
2887EXPORT_SYMBOL(vfs_create);
2888
2889bool may_open_dev(const struct path *path)
2890{
2891        return !(path->mnt->mnt_flags & MNT_NODEV) &&
2892                !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2893}
2894
2895static int may_open(struct path *path, int acc_mode, int flag)
2896{
2897        struct dentry *dentry = path->dentry;
2898        struct inode *inode = dentry->d_inode;
2899        int error;
2900
2901        if (!inode)
2902                return -ENOENT;
2903
2904        switch (inode->i_mode & S_IFMT) {
2905        case S_IFLNK:
2906                return -ELOOP;
2907        case S_IFDIR:
2908                if (acc_mode & MAY_WRITE)
2909                        return -EISDIR;
2910                break;
2911        case S_IFBLK:
2912        case S_IFCHR:
2913                if (!may_open_dev(path))
2914                        return -EACCES;
2915                /*FALLTHRU*/
2916        case S_IFIFO:
2917        case S_IFSOCK:
2918                flag &= ~O_TRUNC;
2919                break;
2920        }
2921
2922        error = inode_permission(inode, MAY_OPEN | acc_mode);
2923        if (error)
2924                return error;
2925
2926        /*
2927         * An append-only file must be opened in append mode for writing.
2928         */
2929        if (IS_APPEND(inode)) {
2930                if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2931                        return -EPERM;
2932                if (flag & O_TRUNC)
2933                        return -EPERM;
2934        }
2935
2936        /* O_NOATIME can only be set by the owner or superuser */
2937        if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2938                return -EPERM;
2939
2940        return 0;
2941}
2942
2943static int handle_truncate(struct file *filp)
2944{
2945        struct path *path = &filp->f_path;
2946        struct inode *inode = path->dentry->d_inode;
2947        int error = get_write_access(inode);
2948        if (error)
2949                return error;
2950        /*
2951         * Refuse to truncate files with mandatory locks held on them.
2952         */
2953        error = locks_verify_locked(filp);
2954        if (!error)
2955                error = security_path_truncate(path);
2956        if (!error) {
2957                error = do_truncate(path->dentry, 0,
2958                                    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2959                                    filp);
2960        }
2961        put_write_access(inode);
2962        return error;
2963}
2964
2965static inline int open_to_namei_flags(int flag)
2966{
2967        if ((flag & O_ACCMODE) == 3)
2968                flag--;
2969        return flag;
2970}
2971
2972static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2973{
2974        int error = security_path_mknod(dir, dentry, mode, 0);
2975        if (error)
2976                return error;
2977
2978        error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2979        if (error)
2980                return error;
2981
2982        return security_inode_create(dir->dentry->d_inode, dentry, mode);
2983}
2984
2985/*
2986 * Attempt to atomically look up, create and open a file from a negative
2987 * dentry.
2988 *
2989 * Returns 0 if successful.  The file will have been created and attached to
2990 * @file by the filesystem calling finish_open().
2991 *
2992 * Returns 1 if the file was looked up only or didn't need creating.  The
2993 * caller will need to perform the open themselves.  @path will have been
2994 * updated to point to the new dentry.  This may be negative.
2995 *
2996 * Returns an error code otherwise.
2997 */
2998static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2999                        struct path *path, struct file *file,
3000                        const struct open_flags *op,
3001                        int open_flag, umode_t mode,
3002                        int *opened)
3003{
3004        struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3005        struct inode *dir =  nd->path.dentry->d_inode;
3006        int error;
3007
3008        if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3009                open_flag &= ~O_TRUNC;
3010
3011        if (nd->flags & LOOKUP_DIRECTORY)
3012                open_flag |= O_DIRECTORY;
3013
3014        file->f_path.dentry = DENTRY_NOT_SET;
3015        file->f_path.mnt = nd->path.mnt;
3016        error = dir->i_op->atomic_open(dir, dentry, file,
3017                                       open_to_namei_flags(open_flag),
3018                                       mode, opened);
3019        d_lookup_done(dentry);
3020        if (!error) {
3021                /*
3022                 * We didn't have the inode before the open, so check open
3023                 * permission here.
3024                 */
3025                int acc_mode = op->acc_mode;
3026                if (*opened & FILE_CREATED) {
3027                        WARN_ON(!(open_flag & O_CREAT));
3028                        fsnotify_create(dir, dentry);
3029                        acc_mode = 0;
3030                }
3031                error = may_open(&file->f_path, acc_mode, open_flag);
3032                if (WARN_ON(error > 0))
3033                        error = -EINVAL;
3034        } else if (error > 0) {
3035                if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3036                        error = -EIO;
3037                } else {
3038                        if (file->f_path.dentry) {
3039                                dput(dentry);
3040                                dentry = file->f_path.dentry;
3041                        }
3042                        if (*opened & FILE_CREATED)
3043                                fsnotify_create(dir, dentry);
3044                        if (unlikely(d_is_negative(dentry))) {
3045                                error = -ENOENT;
3046                        } else {
3047                                path->dentry = dentry;
3048                                path->mnt = nd->path.mnt;
3049                                return 1;
3050                        }
3051                }
3052        }
3053        dput(dentry);
3054        return error;
3055}
3056
3057/*
3058 * Look up and maybe create and open the last component.
3059 *
3060 * Must be called with i_mutex held on parent.
3061 *
3062 * Returns 0 if the file was successfully atomically created (if necessary) and
3063 * opened.  In this case the file will be returned attached to @file.
3064 *
3065 * Returns 1 if the file was not completely opened at this time, though lookups
3066 * and creations will have been performed and the dentry returned in @path will
3067 * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
3068 * specified then a negative dentry may be returned.
3069 *
3070 * An error code is returned otherwise.
3071 *
3072 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3073 * cleared otherwise prior to returning.
3074 */
3075static int lookup_open(struct nameidata *nd, struct path *path,
3076                        struct file *file,
3077                        const struct open_flags *op,
3078                        bool got_write, int *opened)
3079{
3080        struct dentry *dir = nd->path.dentry;
3081        struct inode *dir_inode = dir->d_inode;
3082        int open_flag = op->open_flag;
3083        struct dentry *dentry;
3084        int error, create_error = 0;
3085        umode_t mode = op->mode;
3086        DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3087
3088        if (unlikely(IS_DEADDIR(dir_inode)))
3089                return -ENOENT;
3090
3091        *opened &= ~FILE_CREATED;
3092        dentry = d_lookup(dir, &nd->last);
3093        for (;;) {
3094                if (!dentry) {
3095                        dentry = d_alloc_parallel(dir, &nd->last, &wq);
3096                        if (IS_ERR(dentry))
3097                                return PTR_ERR(dentry);
3098                }
3099                if (d_in_lookup(dentry))
3100                        break;
3101
3102                if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3103                        break;
3104
3105                error = d_revalidate(dentry, nd->flags);
3106                if (likely(error > 0))
3107                        break;
3108                if (error)
3109                        goto out_dput;
3110                d_invalidate(dentry);
3111                dput(dentry);
3112                dentry = NULL;
3113        }
3114        if (dentry->d_inode) {
3115                /* Cached positive dentry: will open in f_op->open */
3116                goto out_no_open;
3117        }
3118
3119        /*
3120         * Checking write permission is tricky, bacuse we don't know if we are
3121         * going to actually need it: O_CREAT opens should work as long as the
3122         * file exists.  But checking existence breaks atomicity.  The trick is
3123         * to check access and if not granted clear O_CREAT from the flags.
3124         *
3125         * Another problem is returing the "right" error value (e.g. for an
3126         * O_EXCL open we want to return EEXIST not EROFS).
3127         */
3128        if (open_flag & O_CREAT) {
3129                if (!IS_POSIXACL(dir->d_inode))
3130                        mode &= ~current_umask();
3131                if (unlikely(!got_write)) {
3132                        create_error = -EROFS;
3133                        open_flag &= ~O_CREAT;
3134                        if (open_flag & (O_EXCL | O_TRUNC))
3135                                goto no_open;
3136                        /* No side effects, safe to clear O_CREAT */
3137                } else {
3138                        create_error = may_o_create(&nd->path, dentry, mode);
3139                        if (create_error) {
3140                                open_flag &= ~O_CREAT;
3141                                if (open_flag & O_EXCL)
3142                                        goto no_open;
3143                        }
3144                }
3145        } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3146                   unlikely(!got_write)) {
3147                /*
3148                 * No O_CREATE -> atomicity not a requirement -> fall
3149                 * back to lookup + open
3150                 */
3151                goto no_open;
3152        }
3153
3154        if (dir_inode->i_op->atomic_open) {
3155                error = atomic_open(nd, dentry, path, file, op, open_flag,
3156                                    mode, opened);
3157                if (unlikely(error == -ENOENT) && create_error)
3158                        error = create_error;
3159                return error;
3160        }
3161
3162no_open:
3163        if (d_in_lookup(dentry)) {
3164                struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3165                                                             nd->flags);
3166                d_lookup_done(dentry);
3167                if (unlikely(res)) {
3168                        if (IS_ERR(res)) {
3169                                error = PTR_ERR(res);
3170                                goto out_dput;
3171                        }
3172                        dput(dentry);
3173                        dentry = res;
3174                }
3175        }
3176
3177        /* Negative dentry, just create the file */
3178        if (!dentry->d_inode && (open_flag & O_CREAT)) {
3179                *opened |= FILE_CREATED;
3180                audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3181                if (!dir_inode->i_op->create) {
3182                        error = -EACCES;
3183                        goto out_dput;
3184                }
3185                error = dir_inode->i_op->create(dir_inode, dentry, mode,
3186                                                open_flag & O_EXCL);
3187                if (error)
3188                        goto out_dput;
3189                fsnotify_create(dir_inode, dentry);
3190        }
3191        if (unlikely(create_error) && !dentry->d_inode) {
3192                error = create_error;
3193                goto out_dput;
3194        }
3195out_no_open:
3196        path->dentry = dentry;
3197        path->mnt = nd->path.mnt;
3198        return 1;
3199
3200out_dput:
3201        dput(dentry);
3202        return error;
3203}
3204
3205/*
3206 * Handle the last step of open()
3207 */
3208static int do_last(struct nameidata *nd,
3209                   struct file *file, const struct open_flags *op,
3210                   int *opened)
3211{
3212        struct dentry *dir = nd->path.dentry;
3213        int open_flag = op->open_flag;
3214        bool will_truncate = (open_flag & O_TRUNC) != 0;
3215        bool got_write = false;
3216        int acc_mode = op->acc_mode;
3217        unsigned seq;
3218        struct inode *inode;
3219        struct path path;
3220        int error;
3221
3222        nd->flags &= ~LOOKUP_PARENT;
3223        nd->flags |= op->intent;
3224
3225        if (nd->last_type != LAST_NORM) {
3226                error = handle_dots(nd, nd->last_type);
3227                if (unlikely(error))
3228                        return error;
3229                goto finish_open;
3230        }
3231
3232        if (!(open_flag & O_CREAT)) {
3233                if (nd->last.name[nd->last.len])
3234                        nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3235                /* we _can_ be in RCU mode here */
3236                error = lookup_fast(nd, &path, &inode, &seq);
3237                if (likely(error > 0))
3238                        goto finish_lookup;
3239
3240                if (error < 0)
3241                        return error;
3242
3243                BUG_ON(nd->inode != dir->d_inode);
3244                BUG_ON(nd->flags & LOOKUP_RCU);
3245        } else {
3246                /* create side of things */
3247                /*
3248                 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3249                 * has been cleared when we got to the last component we are
3250                 * about to look up
3251                 */
3252                error = complete_walk(nd);
3253                if (error)
3254                        return error;
3255
3256                audit_inode(nd->name, dir, LOOKUP_PARENT);
3257                /* trailing slashes? */
3258                if (unlikely(nd->last.name[nd->last.len]))
3259                        return -EISDIR;
3260        }
3261
3262        if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3263                error = mnt_want_write(nd->path.mnt);
3264                if (!error)
3265                        got_write = true;
3266                /*
3267                 * do _not_ fail yet - we might not need that or fail with
3268                 * a different error; let lookup_open() decide; we'll be
3269                 * dropping this one anyway.
3270                 */
3271        }
3272        if (open_flag & O_CREAT)
3273                inode_lock(dir->d_inode);
3274        else
3275                inode_lock_shared(dir->d_inode);
3276        error = lookup_open(nd, &path, file, op, got_write, opened);
3277        if (open_flag & O_CREAT)
3278                inode_unlock(dir->d_inode);
3279        else
3280                inode_unlock_shared(dir->d_inode);
3281
3282        if (error <= 0) {
3283                if (error)
3284                        goto out;
3285
3286                if ((*opened & FILE_CREATED) ||
3287                    !S_ISREG(file_inode(file)->i_mode))
3288                        will_truncate = false;
3289
3290                audit_inode(nd->name, file->f_path.dentry, 0);
3291                goto opened;
3292        }
3293
3294        if (*opened & FILE_CREATED) {
3295                /* Don't check for write permission, don't truncate */
3296                open_flag &= ~O_TRUNC;
3297                will_truncate = false;
3298                acc_mode = 0;
3299                path_to_nameidata(&path, nd);
3300                goto finish_open_created;
3301        }
3302
3303        /*
3304         * If atomic_open() acquired write access it is dropped now due to
3305         * possible mount and symlink following (this might be optimized away if
3306         * necessary...)
3307         */
3308        if (got_write) {
3309                mnt_drop_write(nd->path.mnt);
3310                got_write = false;
3311        }
3312
3313        error = follow_managed(&path, nd);
3314        if (unlikely(error < 0))
3315                return error;
3316
3317        if (unlikely(d_is_negative(path.dentry))) {
3318                path_to_nameidata(&path, nd);
3319                return -ENOENT;
3320        }
3321
3322        /*
3323         * create/update audit record if it already exists.
3324         */
3325        audit_inode(nd->name, path.dentry, 0);
3326
3327        if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3328                path_to_nameidata(&path, nd);
3329                return -EEXIST;
3330        }
3331
3332        seq = 0;        /* out of RCU mode, so the value doesn't matter */
3333        inode = d_backing_inode(path.dentry);
3334finish_lookup:
3335        if (nd->depth)
3336                put_link(nd);
3337        error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3338                                   inode, seq);
3339        if (unlikely(error))
3340                return error;
3341
3342        path_to_nameidata(&path, nd);
3343        nd->inode = inode;
3344        nd->seq = seq;
3345        /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3346finish_open:
3347        error = complete_walk(nd);
3348        if (error)
3349                return error;
3350        audit_inode(nd->name, nd->path.dentry, 0);
3351        error = -EISDIR;
3352        if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3353                goto out;
3354        error = -ENOTDIR;
3355        if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3356                goto out;
3357        if (!d_is_reg(nd->path.dentry))
3358                will_truncate = false;
3359
3360        if (will_truncate) {
3361                error = mnt_want_write(nd->path.mnt);
3362                if (error)
3363                        goto out;
3364                got_write = true;
3365        }
3366finish_open_created:
3367        error = may_open(&nd->path, acc_mode, open_flag);
3368        if (error)
3369                goto out;
3370        BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3371        error = vfs_open(&nd->path, file, current_cred());
3372        if (error)
3373                goto out;
3374        *opened |= FILE_OPENED;
3375opened:
3376        error = open_check_o_direct(file);
3377        if (!error)
3378                error = ima_file_check(file, op->acc_mode, *opened);
3379        if (!error && will_truncate)
3380                error = handle_truncate(file);
3381out:
3382        if (unlikely(error) && (*opened & FILE_OPENED))
3383                fput(file);
3384        if (unlikely(error > 0)) {
3385                WARN_ON(1);
3386                error = -EINVAL;
3387        }
3388        if (got_write)
3389                mnt_drop_write(nd->path.mnt);
3390        return error;
3391}
3392
3393static int do_tmpfile(struct nameidata *nd, unsigned flags,
3394                const struct open_flags *op,
3395                struct file *file, int *opened)
3396{
3397        static const struct qstr name = QSTR_INIT("/", 1);
3398        struct dentry *child;
3399        struct inode *dir;
3400        struct path path;
3401        int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3402        if (unlikely(error))
3403                return error;
3404        error = mnt_want_write(path.mnt);
3405        if (unlikely(error))
3406                goto out;
3407        dir = path.dentry->d_inode;
3408        /* we want directory to be writable */
3409        error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3410        if (error)
3411                goto out2;
3412        if (!dir->i_op->tmpfile) {
3413                error = -EOPNOTSUPP;
3414                goto out2;
3415        }
3416        child = d_alloc(path.dentry, &name);
3417        if (unlikely(!child)) {
3418                error = -ENOMEM;
3419                goto out2;
3420        }
3421        dput(path.dentry);
3422        path.dentry = child;
3423        error = dir->i_op->tmpfile(dir, child, op->mode);
3424        if (error)
3425                goto out2;
3426        audit_inode(nd->name, child, 0);
3427        /* Don't check for other permissions, the inode was just created */
3428        error = may_open(&path, 0, op->open_flag);
3429        if (error)
3430                goto out2;
3431        file->f_path.mnt = path.mnt;
3432        error = finish_open(file, child, NULL, opened);
3433        if (error)
3434                goto out2;
3435        error = open_check_o_direct(file);
3436        if (error) {
3437                fput(file);
3438        } else if (!(op->open_flag & O_EXCL)) {
3439                struct inode *inode = file_inode(file);
3440                spin_lock(&inode->i_lock);
3441                inode->i_state |= I_LINKABLE;
3442                spin_unlock(&inode->i_lock);
3443        }
3444out2:
3445        mnt_drop_write(path.mnt);
3446out:
3447        path_put(&path);
3448        return error;
3449}
3450
3451static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3452{
3453        struct path path;
3454        int error = path_lookupat(nd, flags, &path);
3455        if (!error) {
3456                audit_inode(nd->name, path.dentry, 0);
3457                error = vfs_open(&path, file, current_cred());
3458                path_put(&path);
3459        }
3460        return error;
3461}
3462
3463static struct file *path_openat(struct nameidata *nd,
3464                        const struct open_flags *op, unsigned flags)
3465{
3466        const char *s;
3467        struct file *file;
3468        int opened = 0;
3469        int error;
3470
3471        file = get_empty_filp();
3472        if (IS_ERR(file))
3473                return file;
3474
3475        file->f_flags = op->open_flag;
3476
3477        if (unlikely(file->f_flags & __O_TMPFILE)) {
3478                error = do_tmpfile(nd, flags, op, file, &opened);
3479                goto out2;
3480        }
3481
3482        if (unlikely(file->f_flags & O_PATH)) {
3483                error = do_o_path(nd, flags, file);
3484                if (!error)
3485                        opened |= FILE_OPENED;
3486                goto out2;
3487        }
3488
3489        s = path_init(nd, flags);
3490        if (IS_ERR(s)) {
3491                put_filp(file);
3492                return ERR_CAST(s);
3493        }
3494        while (!(error = link_path_walk(s, nd)) &&
3495                (error = do_last(nd, file, op, &opened)) > 0) {
3496                nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3497                s = trailing_symlink(nd);
3498                if (IS_ERR(s)) {
3499                        error = PTR_ERR(s);
3500                        break;
3501                }
3502        }
3503        terminate_walk(nd);
3504out2:
3505        if (!(opened & FILE_OPENED)) {
3506                BUG_ON(!error);
3507                put_filp(file);
3508        }
3509        if (unlikely(error)) {
3510                if (error == -EOPENSTALE) {
3511                        if (flags & LOOKUP_RCU)
3512                                error = -ECHILD;
3513                        else
3514                                error = -ESTALE;
3515                }
3516                file = ERR_PTR(error);
3517        }
3518        return file;
3519}
3520
3521struct file *do_filp_open(int dfd, struct filename *pathname,
3522                const struct open_flags *op)
3523{
3524        struct nameidata nd;
3525        int flags = op->lookup_flags;
3526        struct file *filp;
3527
3528        set_nameidata(&nd, dfd, pathname);
3529        filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3530        if (unlikely(filp == ERR_PTR(-ECHILD)))
3531                filp = path_openat(&nd, op, flags);
3532        if (unlikely(filp == ERR_PTR(-ESTALE)))
3533                filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3534        restore_nameidata();
3535        return filp;
3536}
3537
3538struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3539                const char *name, const struct open_flags *op)
3540{
3541        struct nameidata nd;
3542        struct file *file;
3543        struct filename *filename;
3544        int flags = op->lookup_flags | LOOKUP_ROOT;
3545
3546        nd.root.mnt = mnt;
3547        nd.root.dentry = dentry;
3548
3549        if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3550                return ERR_PTR(-ELOOP);
3551
3552        filename = getname_kernel(name);
3553        if (IS_ERR(filename))
3554                return ERR_CAST(filename);
3555
3556        set_nameidata(&nd, -1, filename);
3557        file = path_openat(&nd, op, flags | LOOKUP_RCU);
3558        if (unlikely(file == ERR_PTR(-ECHILD)))
3559                file = path_openat(&nd, op, flags);
3560        if (unlikely(file == ERR_PTR(-ESTALE)))
3561                file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3562        restore_nameidata();
3563        putname(filename);
3564        return file;
3565}
3566
3567static struct dentry *filename_create(int dfd, struct filename *name,
3568                                struct path *path, unsigned int lookup_flags)
3569{
3570        struct dentry *dentry = ERR_PTR(-EEXIST);
3571        struct qstr last;
3572        int type;
3573        int err2;
3574        int error;
3575        bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3576
3577        /*
3578         * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3579         * other flags passed in are ignored!
3580         */
3581        lookup_flags &= LOOKUP_REVAL;
3582
3583        name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3584        if (IS_ERR(name))
3585                return ERR_CAST(name);
3586
3587        /*
3588         * Yucky last component or no last component at all?
3589         * (foo/., foo/.., /////)
3590         */
3591        if (unlikely(type != LAST_NORM))
3592                goto out;
3593
3594        /* don't fail immediately if it's r/o, at least try to report other errors */
3595        err2 = mnt_want_write(path->mnt);
3596        /*
3597         * Do the final lookup.
3598         */
3599        lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3600        inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3601        dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3602        if (IS_ERR(dentry))
3603                goto unlock;
3604
3605        error = -EEXIST;
3606        if (d_is_positive(dentry))
3607                goto fail;
3608
3609        /*
3610         * Special case - lookup gave negative, but... we had foo/bar/
3611         * From the vfs_mknod() POV we just have a negative dentry -
3612         * all is fine. Let's be bastards - you had / on the end, you've
3613         * been asking for (non-existent) directory. -ENOENT for you.
3614         */
3615        if (unlikely(!is_dir && last.name[last.len])) {
3616                error = -ENOENT;
3617                goto fail;
3618        }
3619        if (unlikely(err2)) {
3620                error = err2;
3621                goto fail;
3622        }
3623        putname(name);
3624        return dentry;
3625fail:
3626        dput(dentry);
3627        dentry = ERR_PTR(error);
3628unlock:
3629        inode_unlock(path->dentry->d_inode);
3630        if (!err2)
3631                mnt_drop_write(path->mnt);
3632out:
3633        path_put(path);
3634        putname(name);
3635        return dentry;
3636}
3637
3638struct dentry *kern_path_create(int dfd, const char *pathname,
3639                                struct path *path, unsigned int lookup_flags)
3640{
3641        return filename_create(dfd, getname_kernel(pathname),
3642                                path, lookup_flags);
3643}
3644EXPORT_SYMBOL(kern_path_create);
3645
3646void done_path_create(struct path *path, struct dentry *dentry)
3647{
3648        dput(dentry);
3649        inode_unlock(path->dentry->d_inode);
3650        mnt_drop_write(path->mnt);
3651        path_put(path);
3652}
3653EXPORT_SYMBOL(done_path_create);
3654
3655inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3656                                struct path *path, unsigned int lookup_flags)
3657{
3658        return filename_create(dfd, getname(pathname), path, lookup_flags);
3659}
3660EXPORT_SYMBOL(user_path_create);
3661
3662int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3663{
3664        int error = may_create(dir, dentry);
3665
3666        if (error)
3667                return error;
3668
3669        if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3670                return -EPERM;
3671
3672        if (!dir->i_op->mknod)
3673                return -EPERM;
3674
3675        error = devcgroup_inode_mknod(mode, dev);
3676        if (error)
3677                return error;
3678
3679        error = security_inode_mknod(dir, dentry, mode, dev);
3680        if (error)
3681                return error;
3682
3683        error = dir->i_op->mknod(dir, dentry, mode, dev);
3684        if (!error)
3685                fsnotify_create(dir, dentry);
3686        return error;
3687}
3688EXPORT_SYMBOL(vfs_mknod);
3689
3690static int may_mknod(umode_t mode)
3691{
3692        switch (mode & S_IFMT) {
3693        case S_IFREG:
3694        case S_IFCHR:
3695        case S_IFBLK:
3696        case S_IFIFO:
3697        case S_IFSOCK:
3698        case 0: /* zero mode translates to S_IFREG */
3699                return 0;
3700        case S_IFDIR:
3701                return -EPERM;
3702        default:
3703                return -EINVAL;
3704        }
3705}
3706
3707SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3708                unsigned, dev)
3709{
3710        struct dentry *dentry;
3711        struct path path;
3712        int error;
3713        unsigned int lookup_flags = 0;
3714
3715        error = may_mknod(mode);
3716        if (error)
3717                return error;
3718retry:
3719        dentry = user_path_create(dfd, filename, &path, lookup_flags);
3720        if (IS_ERR(dentry))
3721                return PTR_ERR(dentry);
3722
3723        if (!IS_POSIXACL(path.dentry->d_inode))
3724                mode &= ~current_umask();
3725        error = security_path_mknod(&path, dentry, mode, dev);
3726        if (error)
3727                goto out;
3728        switch (mode & S_IFMT) {
3729                case 0: case S_IFREG:
3730                        error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3731                        if (!error)
3732                                ima_post_path_mknod(dentry);
3733                        break;
3734                case S_IFCHR: case S_IFBLK:
3735                        error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3736                                        new_decode_dev(dev));
3737                        break;
3738                case S_IFIFO: case S_IFSOCK:
3739                        error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3740                        break;
3741        }
3742out:
3743        done_path_create(&path, dentry);
3744        if (retry_estale(error, lookup_flags)) {
3745                lookup_flags |= LOOKUP_REVAL;
3746                goto retry;
3747        }
3748        return error;
3749}
3750
3751SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3752{
3753        return sys_mknodat(AT_FDCWD, filename, mode, dev);
3754}
3755
3756int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3757{
3758        int error = may_create(dir, dentry);
3759        unsigned max_links = dir->i_sb->s_max_links;
3760
3761        if (error)
3762                return error;
3763
3764        if (!dir->i_op->mkdir)
3765                return -EPERM;
3766
3767        mode &= (S_IRWXUGO|S_ISVTX);
3768        error = security_inode_mkdir(dir, dentry, mode);
3769        if (error)
3770                return error;
3771
3772        if (max_links && dir->i_nlink >= max_links)
3773                return -EMLINK;
3774
3775        error = dir->i_op->mkdir(dir, dentry, mode);
3776        if (!error)
3777                fsnotify_mkdir(dir, dentry);
3778        return error;
3779}
3780EXPORT_SYMBOL(vfs_mkdir);
3781
3782SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3783{
3784        struct dentry *dentry;
3785        struct path path;
3786        int error;
3787        unsigned int lookup_flags = LOOKUP_DIRECTORY;
3788
3789retry:
3790        dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3791        if (IS_ERR(dentry))
3792                return PTR_ERR(dentry);
3793
3794        if (!IS_POSIXACL(path.dentry->d_inode))
3795                mode &= ~current_umask();
3796        error = security_path_mkdir(&path, dentry, mode);
3797        if (!error)
3798                error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3799        done_path_create(&path, dentry);
3800        if (retry_estale(error, lookup_flags)) {
3801                lookup_flags |= LOOKUP_REVAL;
3802                goto retry;
3803        }
3804        return error;
3805}
3806
3807SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3808{
3809        return sys_mkdirat(AT_FDCWD, pathname, mode);
3810}
3811
3812int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3813{
3814        int error = may_delete(dir, dentry, 1);
3815
3816        if (error)
3817                return error;
3818
3819        if (!dir->i_op->rmdir)
3820                return -EPERM;
3821
3822        dget(dentry);
3823        inode_lock(dentry->d_inode);
3824
3825        error = -EBUSY;
3826        if (is_local_mountpoint(dentry))
3827                goto out;
3828
3829        error = security_inode_rmdir(dir, dentry);
3830        if (error)
3831                goto out;
3832
3833        shrink_dcache_parent(dentry);
3834        error = dir->i_op->rmdir(dir, dentry);
3835        if (error)
3836                goto out;
3837
3838        dentry->d_inode->i_flags |= S_DEAD;
3839        dont_mount(dentry);
3840        detach_mounts(dentry);
3841
3842out:
3843        inode_unlock(dentry->d_inode);
3844        dput(dentry);
3845        if (!error)
3846                d_delete(dentry);
3847        return error;
3848}
3849EXPORT_SYMBOL(vfs_rmdir);
3850
3851static long do_rmdir(int dfd, const char __user *pathname)
3852{
3853        int error = 0;
3854        struct filename *name;
3855        struct dentry *dentry;
3856        struct path path;
3857        struct qstr last;
3858        int type;
3859        unsigned int lookup_flags = 0;
3860retry:
3861        name = user_path_parent(dfd, pathname,
3862                                &path, &last, &type, lookup_flags);
3863        if (IS_ERR(name))
3864                return PTR_ERR(name);
3865
3866        switch (type) {
3867        case LAST_DOTDOT:
3868                error = -ENOTEMPTY;
3869                goto exit1;
3870        case LAST_DOT:
3871                error = -EINVAL;
3872                goto exit1;
3873        case LAST_ROOT:
3874                error = -EBUSY;
3875                goto exit1;
3876        }
3877
3878        error = mnt_want_write(path.mnt);
3879        if (error)
3880                goto exit1;
3881
3882        inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3883        dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3884        error = PTR_ERR(dentry);
3885        if (IS_ERR(dentry))
3886                goto exit2;
3887        if (!dentry->d_inode) {
3888                error = -ENOENT;
3889                goto exit3;
3890        }
3891        error = security_path_rmdir(&path, dentry);
3892        if (error)
3893                goto exit3;
3894        error = vfs_rmdir(path.dentry->d_inode, dentry);
3895exit3:
3896        dput(dentry);
3897exit2:
3898        inode_unlock(path.dentry->d_inode);
3899        mnt_drop_write(path.mnt);
3900exit1:
3901        path_put(&path);
3902        putname(name);
3903        if (retry_estale(error, lookup_flags)) {
3904                lookup_flags |= LOOKUP_REVAL;
3905                goto retry;
3906        }
3907        return error;
3908}
3909
3910SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3911{
3912        return do_rmdir(AT_FDCWD, pathname);
3913}
3914
3915/**
3916 * vfs_unlink - unlink a filesystem object
3917 * @dir:        parent directory
3918 * @dentry:     victim
3919 * @delegated_inode: returns victim inode, if the inode is delegated.
3920 *
3921 * The caller must hold dir->i_mutex.
3922 *
3923 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3924 * return a reference to the inode in delegated_inode.  The caller
3925 * should then break the delegation on that inode and retry.  Because
3926 * breaking a delegation may take a long time, the caller should drop
3927 * dir->i_mutex before doing so.
3928 *
3929 * Alternatively, a caller may pass NULL for delegated_inode.  This may
3930 * be appropriate for callers that expect the underlying filesystem not
3931 * to be NFS exported.
3932 */
3933int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3934{
3935        struct inode *target = dentry->d_inode;
3936        int error = may_delete(dir, dentry, 0);
3937
3938        if (error)
3939                return error;
3940
3941        if (!dir->i_op->unlink)
3942                return -EPERM;
3943
3944        inode_lock(target);
3945        if (is_local_mountpoint(dentry))
3946                error = -EBUSY;
3947        else {
3948                error = security_inode_unlink(dir, dentry);
3949                if (!error) {
3950                        error = try_break_deleg(target, delegated_inode);
3951                        if (error)
3952                                goto out;
3953                        error = dir->i_op->unlink(dir, dentry);
3954                        if (!error) {
3955                                dont_mount(dentry);
3956                                detach_mounts(dentry);
3957                        }
3958                }
3959        }
3960out:
3961        inode_unlock(target);
3962
3963        /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3964        if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3965                fsnotify_link_count(target);
3966                d_delete(dentry);
3967        }
3968
3969        return error;
3970}
3971EXPORT_SYMBOL(vfs_unlink);
3972
3973/*
3974 * Make sure that the actual truncation of the file will occur outside its
3975 * directory's i_mutex.  Truncate can take a long time if there is a lot of
3976 * writeout happening, and we don't want to prevent access to the directory
3977 * while waiting on the I/O.
3978 */
3979static long do_unlinkat(int dfd, const char __user *pathname)
3980{
3981        int error;
3982        struct filename *name;
3983        struct dentry *dentry;
3984        struct path path;
3985        struct qstr last;
3986        int type;
3987        struct inode *inode = NULL;
3988        struct inode *delegated_inode = NULL;
3989        unsigned int lookup_flags = 0;
3990retry:
3991        name = user_path_parent(dfd, pathname,
3992                                &path, &last, &type, lookup_flags);
3993        if (IS_ERR(name))
3994                return PTR_ERR(name);
3995
3996        error = -EISDIR;
3997        if (type != LAST_NORM)
3998                goto exit1;
3999
4000        error = mnt_want_write(path.mnt);
4001        if (error)
4002                goto exit1;
4003retry_deleg:
4004        inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4005        dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4006        error = PTR_ERR(dentry);
4007        if (!IS_ERR(dentry)) {
4008                /* Why not before? Because we want correct error value */
4009                if (last.name[last.len])
4010                        goto slashes;
4011                inode = dentry->d_inode;
4012                if (d_is_negative(dentry))
4013                        goto slashes;
4014                ihold(inode);
4015                error = security_path_unlink(&path, dentry);
4016                if (error)
4017                        goto exit2;
4018                error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4019exit2:
4020                dput(dentry);
4021        }
4022        inode_unlock(path.dentry->d_inode);
4023        if (inode)
4024                iput(inode);    /* truncate the inode here */
4025        inode = NULL;
4026        if (delegated_inode) {
4027                error = break_deleg_wait(&delegated_inode);
4028                if (!error)
4029                        goto retry_deleg;
4030        }
4031        mnt_drop_write(path.mnt);
4032exit1:
4033        path_put(&path);
4034        putname(name);
4035        if (retry_estale(error, lookup_flags)) {
4036                lookup_flags |= LOOKUP_REVAL;
4037                inode = NULL;
4038                goto retry;
4039        }
4040        return error;
4041
4042slashes:
4043        if (d_is_negative(dentry))
4044                error = -ENOENT;
4045        else if (d_is_dir(dentry))
4046                error = -EISDIR;
4047        else
4048                error = -ENOTDIR;
4049        goto exit2;
4050}
4051
4052SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4053{
4054        if ((flag & ~AT_REMOVEDIR) != 0)
4055                return -EINVAL;
4056
4057        if (flag & AT_REMOVEDIR)
4058                return do_rmdir(dfd, pathname);
4059
4060        return do_unlinkat(dfd, pathname);
4061}
4062
4063SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4064{
4065        return do_unlinkat(AT_FDCWD, pathname);
4066}
4067
4068int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4069{
4070        int error = may_create(dir, dentry);
4071
4072        if (error)
4073                return error;
4074
4075        if (!dir->i_op->symlink)
4076                return -EPERM;
4077
4078        error = security_inode_symlink(dir, dentry, oldname);
4079        if (error)
4080                return error;
4081
4082        error = dir->i_op->symlink(dir, dentry, oldname);
4083        if (!error)
4084                fsnotify_create(dir, dentry);
4085        return error;
4086}
4087EXPORT_SYMBOL(vfs_symlink);
4088
4089SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4090                int, newdfd, const char __user *, newname)
4091{
4092        int error;
4093        struct filename *from;
4094        struct dentry *dentry;
4095        struct path path;
4096        unsigned int lookup_flags = 0;
4097
4098        from = getname(oldname);
4099        if (IS_ERR(from))
4100                return PTR_ERR(from);
4101retry:
4102        dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4103        error = PTR_ERR(dentry);
4104        if (IS_ERR(dentry))
4105                goto out_putname;
4106
4107        error = security_path_symlink(&path, dentry, from->name);
4108        if (!error)
4109                error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4110        done_path_create(&path, dentry);
4111        if (retry_estale(error, lookup_flags)) {
4112                lookup_flags |= LOOKUP_REVAL;
4113                goto retry;
4114        }
4115out_putname:
4116        putname(from);
4117        return error;
4118}
4119
4120SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4121{
4122        return sys_symlinkat(oldname, AT_FDCWD, newname);
4123}
4124
4125/**
4126 * vfs_link - create a new link
4127 * @old_dentry: object to be linked
4128 * @dir:        new parent
4129 * @new_dentry: where to create the new link
4130 * @delegated_inode: returns inode needing a delegation break
4131 *
4132 * The caller must hold dir->i_mutex
4133 *
4134 * If vfs_link discovers a delegation on the to-be-linked file in need
4135 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4136 * inode in delegated_inode.  The caller should then break the delegation
4137 * and retry.  Because breaking a delegation may take a long time, the
4138 * caller should drop the i_mutex before doing so.
4139 *
4140 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4141 * be appropriate for callers that expect the underlying filesystem not
4142 * to be NFS exported.
4143 */
4144int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4145{
4146        struct inode *inode = old_dentry->d_inode;
4147        unsigned max_links = dir->i_sb->s_max_links;
4148        int error;
4149
4150        if (!inode)
4151                return -ENOENT;
4152
4153        error = may_create(dir, new_dentry);
4154        if (error)
4155                return error;
4156
4157        if (dir->i_sb != inode->i_sb)
4158                return -EXDEV;
4159
4160        /*
4161         * A link to an append-only or immutable file cannot be created.
4162         */
4163        if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4164                return -EPERM;
4165        /*
4166         * Updating the link count will likely cause i_uid and i_gid to
4167         * be writen back improperly if their true value is unknown to
4168         * the vfs.
4169         */
4170        if (HAS_UNMAPPED_ID(inode))
4171                return -EPERM;
4172        if (!dir->i_op->link)
4173                return -EPERM;
4174        if (S_ISDIR(inode->i_mode))
4175                return -EPERM;
4176
4177        error = security_inode_link(old_dentry, dir, new_dentry);
4178        if (error)
4179                return error;
4180
4181        inode_lock(inode);
4182        /* Make sure we don't allow creating hardlink to an unlinked file */
4183        if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4184                error =  -ENOENT;
4185        else if (max_links && inode->i_nlink >= max_links)
4186                error = -EMLINK;
4187        else {
4188                error = try_break_deleg(inode, delegated_inode);
4189                if (!error)
4190                        error = dir->i_op->link(old_dentry, dir, new_dentry);
4191        }
4192
4193        if (!error && (inode->i_state & I_LINKABLE)) {
4194                spin_lock(&inode->i_lock);
4195                inode->i_state &= ~I_LINKABLE;
4196                spin_unlock(&inode->i_lock);
4197        }
4198        inode_unlock(inode);
4199        if (!error)
4200                fsnotify_link(dir, inode, new_dentry);
4201        return error;
4202}
4203EXPORT_SYMBOL(vfs_link);
4204
4205/*
4206 * Hardlinks are often used in delicate situations.  We avoid
4207 * security-related surprises by not following symlinks on the
4208 * newname.  --KAB
4209 *
4210 * We don't follow them on the oldname either to be compatible
4211 * with linux 2.0, and to avoid hard-linking to directories
4212 * and other special files.  --ADM
4213 */
4214SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4215                int, newdfd, const char __user *, newname, int, flags)
4216{
4217        struct dentry *new_dentry;
4218        struct path old_path, new_path;
4219        struct inode *delegated_inode = NULL;
4220        int how = 0;
4221        int error;
4222
4223        if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4224                return -EINVAL;
4225        /*
4226         * To use null names we require CAP_DAC_READ_SEARCH
4227         * This ensures that not everyone will be able to create
4228         * handlink using the passed filedescriptor.
4229         */
4230        if (flags & AT_EMPTY_PATH) {
4231                if (!capable(CAP_DAC_READ_SEARCH))
4232                        return -ENOENT;
4233                how = LOOKUP_EMPTY;
4234        }
4235
4236        if (flags & AT_SYMLINK_FOLLOW)
4237                how |= LOOKUP_FOLLOW;
4238retry:
4239        error = user_path_at(olddfd, oldname, how, &old_path);
4240        if (error)
4241                return error;
4242
4243        new_dentry = user_path_create(newdfd, newname, &new_path,
4244                                        (how & LOOKUP_REVAL));
4245        error = PTR_ERR(new_dentry);
4246        if (IS_ERR(new_dentry))
4247                goto out;
4248
4249        error = -EXDEV;
4250        if (old_path.mnt != new_path.mnt)
4251                goto out_dput;
4252        error = may_linkat(&old_path);
4253        if (unlikely(error))
4254                goto out_dput;
4255        error = security_path_link(old_path.dentry, &new_path, new_dentry);
4256        if (error)
4257                goto out_dput;
4258        error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4259out_dput:
4260        done_path_create(&new_path, new_dentry);
4261        if (delegated_inode) {
4262                error = break_deleg_wait(&delegated_inode);
4263                if (!error) {
4264                        path_put(&old_path);
4265                        goto retry;
4266                }
4267        }
4268        if (retry_estale(error, how)) {
4269                path_put(&old_path);
4270                how |= LOOKUP_REVAL;
4271                goto retry;
4272        }
4273out:
4274        path_put(&old_path);
4275
4276        return error;
4277}
4278
4279SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4280{
4281        return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4282}
4283
4284/**
4285 * vfs_rename - rename a filesystem object
4286 * @old_dir:    parent of source
4287 * @old_dentry: source
4288 * @new_dir:    parent of destination
4289 * @new_dentry: destination
4290 * @delegated_inode: returns an inode needing a delegation break
4291 * @flags:      rename flags
4292 *
4293 * The caller must hold multiple mutexes--see lock_rename()).
4294 *
4295 * If vfs_rename discovers a delegation in need of breaking at either
4296 * the source or destination, it will return -EWOULDBLOCK and return a
4297 * reference to the inode in delegated_inode.  The caller should then
4298 * break the delegation and retry.  Because breaking a delegation may
4299 * take a long time, the caller should drop all locks before doing
4300 * so.
4301 *
4302 * Alternatively, a caller may pass NULL for delegated_inode.  This may
4303 * be appropriate for callers that expect the underlying filesystem not
4304 * to be NFS exported.
4305 *
4306 * The worst of all namespace operations - renaming directory. "Perverted"
4307 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4308 * Problems:
4309 *      a) we can get into loop creation.
4310 *      b) race potential - two innocent renames can create a loop together.
4311 *         That's where 4.4 screws up. Current fix: serialization on
4312 *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4313 *         story.
4314 *      c) we have to lock _four_ objects - parents and victim (if it exists),
4315 *         and source (if it is not a directory).
4316 *         And that - after we got ->i_mutex on parents (until then we don't know
4317 *         whether the target exists).  Solution: try to be smart with locking
4318 *         order for inodes.  We rely on the fact that tree topology may change
4319 *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4320 *         move will be locked.  Thus we can rank directories by the tree
4321 *         (ancestors first) and rank all non-directories after them.
4322 *         That works since everybody except rename does "lock parent, lookup,
4323 *         lock child" and rename is under ->s_vfs_rename_mutex.
4324 *         HOWEVER, it relies on the assumption that any object with ->lookup()
4325 *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4326 *         we'd better make sure that there's no link(2) for them.
4327 *      d) conversion from fhandle to dentry may come in the wrong moment - when
4328 *         we are removing the target. Solution: we will have to grab ->i_mutex
4329 *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4330 *         ->i_mutex on parents, which works but leads to some truly excessive
4331 *         locking].
4332 */
4333int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4334               struct inode *new_dir, struct dentry *new_dentry,
4335               struct inode **delegated_inode, unsigned int flags)
4336{
4337        int error;
4338        bool is_dir = d_is_dir(old_dentry);
4339        const unsigned char *old_name;
4340        struct inode *source = old_dentry->d_inode;
4341        struct inode *target = new_dentry->d_inode;
4342        bool new_is_dir = false;
4343        unsigned max_links = new_dir->i_sb->s_max_links;
4344
4345        /*
4346         * Check source == target.
4347         * On overlayfs need to look at underlying inodes.
4348         */
4349        if (d_real_inode(old_dentry) == d_real_inode(new_dentry))
4350                return 0;
4351
4352        error = may_delete(old_dir, old_dentry, is_dir);
4353        if (error)
4354                return error;
4355
4356        if (!target) {
4357                error = may_create(new_dir, new_dentry);
4358        } else {
4359                new_is_dir = d_is_dir(new_dentry);
4360
4361                if (!(flags & RENAME_EXCHANGE))
4362                        error = may_delete(new_dir, new_dentry, is_dir);
4363                else
4364                        error = may_delete(new_dir, new_dentry, new_is_dir);
4365        }
4366        if (error)
4367                return error;
4368
4369        if (!old_dir->i_op->rename)
4370                return -EPERM;
4371
4372        /*
4373         * If we are going to change the parent - check write permissions,
4374         * we'll need to flip '..'.
4375         */
4376        if (new_dir != old_dir) {
4377                if (is_dir) {
4378                        error = inode_permission(source, MAY_WRITE);
4379                        if (error)
4380                                return error;
4381                }
4382                if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4383                        error = inode_permission(target, MAY_WRITE);
4384                        if (error)
4385                                return error;
4386                }
4387        }
4388
4389        error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4390                                      flags);
4391        if (error)
4392                return error;
4393
4394        old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4395        dget(new_dentry);
4396        if (!is_dir || (flags & RENAME_EXCHANGE))
4397                lock_two_nondirectories(source, target);
4398        else if (target)
4399                inode_lock(target);
4400
4401        error = -EBUSY;
4402        if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4403                goto out;
4404
4405        if (max_links && new_dir != old_dir) {
4406                error = -EMLINK;
4407                if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4408                        goto out;
4409                if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4410                    old_dir->i_nlink >= max_links)
4411                        goto out;
4412        }
4413        if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4414                shrink_dcache_parent(new_dentry);
4415        if (!is_dir) {
4416                error = try_break_deleg(source, delegated_inode);
4417                if (error)
4418                        goto out;
4419        }
4420        if (target && !new_is_dir) {
4421                error = try_break_deleg(target, delegated_inode);
4422                if (error)
4423                        goto out;
4424        }
4425        error = old_dir->i_op->rename(old_dir, old_dentry,
4426                                       new_dir, new_dentry, flags);
4427        if (error)
4428                goto out;
4429
4430        if (!(flags & RENAME_EXCHANGE) && target) {
4431                if (is_dir)
4432                        target->i_flags |= S_DEAD;
4433                dont_mount(new_dentry);
4434                detach_mounts(new_dentry);
4435        }
4436        if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4437                if (!(flags & RENAME_EXCHANGE))
4438                        d_move(old_dentry, new_dentry);
4439                else
4440                        d_exchange(old_dentry, new_dentry);
4441        }
4442out:
4443        if (!is_dir || (flags & RENAME_EXCHANGE))
4444                unlock_two_nondirectories(source, target);
4445        else if (target)
4446                inode_unlock(target);
4447        dput(new_dentry);
4448        if (!error) {
4449                fsnotify_move(old_dir, new_dir, old_name, is_dir,
4450                              !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4451                if (flags & RENAME_EXCHANGE) {
4452                        fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4453                                      new_is_dir, NULL, new_dentry);
4454                }
4455        }
4456        fsnotify_oldname_free(old_name);
4457
4458        return error;
4459}
4460EXPORT_SYMBOL(vfs_rename);
4461
4462SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4463                int, newdfd, const char __user *, newname, unsigned int, flags)
4464{
4465        struct dentry *old_dentry, *new_dentry;
4466        struct dentry *trap;
4467        struct path old_path, new_path;
4468        struct qstr old_last, new_last;
4469        int old_type, new_type;
4470        struct inode *delegated_inode = NULL;
4471        struct filename *from;
4472        struct filename *to;
4473        unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4474        bool should_retry = false;
4475        int error;
4476
4477        if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4478                return -EINVAL;
4479
4480        if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4481            (flags & RENAME_EXCHANGE))
4482                return -EINVAL;
4483
4484        if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4485                return -EPERM;
4486
4487        if (flags & RENAME_EXCHANGE)
4488                target_flags = 0;
4489
4490retry:
4491        from = user_path_parent(olddfd, oldname,
4492                                &old_path, &old_last, &old_type, lookup_flags);
4493        if (IS_ERR(from)) {
4494                error = PTR_ERR(from);
4495                goto exit;
4496        }
4497
4498        to = user_path_parent(newdfd, newname,
4499                                &new_path, &new_last, &new_type, lookup_flags);
4500        if (IS_ERR(to)) {
4501                error = PTR_ERR(to);
4502                goto exit1;
4503        }
4504
4505        error = -EXDEV;
4506        if (old_path.mnt != new_path.mnt)
4507                goto exit2;
4508
4509        error = -EBUSY;
4510        if (old_type != LAST_NORM)
4511                goto exit2;
4512
4513        if (flags & RENAME_NOREPLACE)
4514                error = -EEXIST;
4515        if (new_type != LAST_NORM)
4516                goto exit2;
4517
4518        error = mnt_want_write(old_path.mnt);
4519        if (error)
4520                goto exit2;
4521
4522retry_deleg:
4523        trap = lock_rename(new_path.dentry, old_path.dentry);
4524
4525        old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4526        error = PTR_ERR(old_dentry);
4527        if (IS_ERR(old_dentry))
4528                goto exit3;
4529        /* source must exist */
4530        error = -ENOENT;
4531        if (d_is_negative(old_dentry))
4532                goto exit4;
4533        new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4534        error = PTR_ERR(new_dentry);
4535        if (IS_ERR(new_dentry))
4536                goto exit4;
4537        error = -EEXIST;
4538        if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4539                goto exit5;
4540        if (flags & RENAME_EXCHANGE) {
4541                error = -ENOENT;
4542                if (d_is_negative(new_dentry))
4543                        goto exit5;
4544
4545                if (!d_is_dir(new_dentry)) {
4546                        error = -ENOTDIR;
4547                        if (new_last.name[new_last.len])
4548                                goto exit5;
4549                }
4550        }
4551        /* unless the source is a directory trailing slashes give -ENOTDIR */
4552        if (!d_is_dir(old_dentry)) {
4553                error = -ENOTDIR;
4554                if (old_last.name[old_last.len])
4555                        goto exit5;
4556                if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4557                        goto exit5;
4558        }
4559        /* source should not be ancestor of target */
4560        error = -EINVAL;
4561        if (old_dentry == trap)
4562                goto exit5;
4563        /* target should not be an ancestor of source */
4564        if (!(flags & RENAME_EXCHANGE))
4565                error = -ENOTEMPTY;
4566        if (new_dentry == trap)
4567                goto exit5;
4568
4569        error = security_path_rename(&old_path, old_dentry,
4570                                     &new_path, new_dentry, flags);
4571        if (error)
4572                goto exit5;
4573        error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4574                           new_path.dentry->d_inode, new_dentry,
4575                           &delegated_inode, flags);
4576exit5:
4577        dput(new_dentry);
4578exit4:
4579        dput(old_dentry);
4580exit3:
4581        unlock_rename(new_path.dentry, old_path.dentry);
4582        if (delegated_inode) {
4583                error = break_deleg_wait(&delegated_inode);
4584                if (!error)
4585                        goto retry_deleg;
4586        }
4587        mnt_drop_write(old_path.mnt);
4588exit2:
4589        if (retry_estale(error, lookup_flags))
4590                should_retry = true;
4591        path_put(&new_path);
4592        putname(to);
4593exit1:
4594        path_put(&old_path);
4595        putname(from);
4596        if (should_retry) {
4597                should_retry = false;
4598                lookup_flags |= LOOKUP_REVAL;
4599                goto retry;
4600        }
4601exit:
4602        return error;
4603}
4604
4605SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4606                int, newdfd, const char __user *, newname)
4607{
4608        return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4609}
4610
4611SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4612{
4613        return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4614}
4615
4616int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4617{
4618        int error = may_create(dir, dentry);
4619        if (error)
4620                return error;
4621
4622        if (!dir->i_op->mknod)
4623                return -EPERM;
4624
4625        return dir->i_op->mknod(dir, dentry,
4626                                S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4627}
4628EXPORT_SYMBOL(vfs_whiteout);
4629
4630int readlink_copy(char __user *buffer, int buflen, const char *link)
4631{
4632        int len = PTR_ERR(link);
4633        if (IS_ERR(link))
4634                goto out;
4635
4636        len = strlen(link);
4637        if (len > (unsigned) buflen)
4638                len = buflen;
4639        if (copy_to_user(buffer, link, len))
4640                len = -EFAULT;
4641out:
4642        return len;
4643}
4644
4645/*
4646 * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4647 * have ->get_link() not calling nd_jump_link().  Using (or not using) it
4648 * for any given inode is up to filesystem.
4649 */
4650int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4651{
4652        DEFINE_DELAYED_CALL(done);
4653        struct inode *inode = d_inode(dentry);
4654        const char *link = inode->i_link;
4655        int res;
4656
4657        if (!link) {
4658                link = inode->i_op->get_link(dentry, inode, &done);
4659                if (IS_ERR(link))
4660                        return PTR_ERR(link);
4661        }
4662        res = readlink_copy(buffer, buflen, link);
4663        do_delayed_call(&done);
4664        return res;
4665}
4666EXPORT_SYMBOL(generic_readlink);
4667
4668/**
4669 * vfs_get_link - get symlink body
4670 * @dentry: dentry on which to get symbolic link
4671 * @done: caller needs to free returned data with this
4672 *
4673 * Calls security hook and i_op->get_link() on the supplied inode.
4674 *
4675 * It does not touch atime.  That's up to the caller if necessary.
4676 *
4677 * Does not work on "special" symlinks like /proc/$$/fd/N
4678 */
4679const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4680{
4681        const char *res = ERR_PTR(-EINVAL);
4682        struct inode *inode = d_inode(dentry);
4683
4684        if (d_is_symlink(dentry)) {
4685                res = ERR_PTR(security_inode_readlink(dentry));
4686                if (!res)
4687                        res = inode->i_op->get_link(dentry, inode, done);
4688        }
4689        return res;
4690}
4691EXPORT_SYMBOL(vfs_get_link);
4692
4693/* get the link contents into pagecache */
4694const char *page_get_link(struct dentry *dentry, struct inode *inode,
4695                          struct delayed_call *callback)
4696{
4697        char *kaddr;
4698        struct page *page;
4699        struct address_space *mapping = inode->i_mapping;
4700
4701        if (!dentry) {
4702                page = find_get_page(mapping, 0);
4703                if (!page)
4704                        return ERR_PTR(-ECHILD);
4705                if (!PageUptodate(page)) {
4706                        put_page(page);
4707                        return ERR_PTR(-ECHILD);
4708                }
4709        } else {
4710                page = read_mapping_page(mapping, 0, NULL);
4711                if (IS_ERR(page))
4712                        return (char*)page;
4713        }
4714        set_delayed_call(callback, page_put_link, page);
4715        BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4716        kaddr = page_address(page);
4717        nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4718        return kaddr;
4719}
4720
4721EXPORT_SYMBOL(page_get_link);
4722
4723void page_put_link(void *arg)
4724{
4725        put_page(arg);
4726}
4727EXPORT_SYMBOL(page_put_link);
4728
4729int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4730{
4731        DEFINE_DELAYED_CALL(done);
4732        int res = readlink_copy(buffer, buflen,
4733                                page_get_link(dentry, d_inode(dentry),
4734                                              &done));
4735        do_delayed_call(&done);
4736        return res;
4737}
4738EXPORT_SYMBOL(page_readlink);
4739
4740/*
4741 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4742 */
4743int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4744{
4745        struct address_space *mapping = inode->i_mapping;
4746        struct page *page;
4747        void *fsdata;
4748        int err;
4749        unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4750        if (nofs)
4751                flags |= AOP_FLAG_NOFS;
4752
4753retry:
4754        err = pagecache_write_begin(NULL, mapping, 0, len-1,
4755                                flags, &page, &fsdata);
4756        if (err)
4757                goto fail;
4758
4759        memcpy(page_address(page), symname, len-1);
4760
4761        err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4762                                                        page, fsdata);
4763        if (err < 0)
4764                goto fail;
4765        if (err < len-1)
4766                goto retry;
4767
4768        mark_inode_dirty(inode);
4769        return 0;
4770fail:
4771        return err;
4772}
4773EXPORT_SYMBOL(__page_symlink);
4774
4775int page_symlink(struct inode *inode, const char *symname, int len)
4776{
4777        return __page_symlink(inode, symname, len,
4778                        !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4779}
4780EXPORT_SYMBOL(page_symlink);
4781
4782const struct inode_operations page_symlink_inode_operations = {
4783        .readlink       = generic_readlink,
4784        .get_link       = page_get_link,
4785};
4786EXPORT_SYMBOL(page_symlink_inode_operations);
Note: See TracBrowser for help on using the repository browser.