11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namei.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Some corrections by tytso. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 121da177e4SLinus Torvalds * lookup logic. 131da177e4SLinus Torvalds */ 141da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/init.h> 18630d9c47SPaul Gortmaker #include <linux/export.h> 1944696908SDavid S. Miller #include <linux/kernel.h> 201da177e4SLinus Torvalds #include <linux/slab.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/pagemap.h> 240eeca283SRobert Love #include <linux/fsnotify.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 276146f0d5SMimi Zohar #include <linux/ima.h> 281da177e4SLinus Torvalds #include <linux/syscalls.h> 291da177e4SLinus Torvalds #include <linux/mount.h> 301da177e4SLinus Torvalds #include <linux/audit.h> 3116f7e0feSRandy Dunlap #include <linux/capability.h> 32834f2a4aSTrond Myklebust #include <linux/file.h> 335590ff0dSUlrich Drepper #include <linux/fcntl.h> 3408ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 355ad4e53bSAl Viro #include <linux/fs_struct.h> 36e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3799d263d4SLinus Torvalds #include <linux/hash.h> 381da177e4SLinus Torvalds #include <asm/uaccess.h> 391da177e4SLinus Torvalds 40e81e3f4dSEric Paris #include "internal.h" 41c7105365SAl Viro #include "mount.h" 42e81e3f4dSEric Paris 431da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 441da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 451da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 461da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 471da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 481da177e4SLinus Torvalds * 491da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 501da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 511da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 521da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 531da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 541da177e4SLinus Torvalds * the special cases of the former code. 551da177e4SLinus Torvalds * 561da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 571da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 581da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 611da177e4SLinus Torvalds * resolution to correspond with current state of the code. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 641da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 651da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 661da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 671da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 681da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 691da177e4SLinus Torvalds */ 701da177e4SLinus Torvalds 711da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 721da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 731da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 741da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 751da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 761da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7725985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 781da177e4SLinus Torvalds * 791da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 801da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 811da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 821da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 831da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 841da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 851da177e4SLinus Torvalds * and in the old Linux semantics. 861da177e4SLinus Torvalds */ 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 891da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 901da177e4SLinus Torvalds * 911da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 921da177e4SLinus Torvalds */ 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 951da177e4SLinus Torvalds * inside the path - always follow. 961da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 971da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 981da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 991da177e4SLinus Torvalds * otherwise - don't follow. 1001da177e4SLinus Torvalds * (applied in that order). 1011da177e4SLinus Torvalds * 1021da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1031da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1041da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1051da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1061da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1071da177e4SLinus Torvalds */ 1081da177e4SLinus Torvalds /* 1091da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 110a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1111da177e4SLinus Torvalds * any extra contention... 1121da177e4SLinus Torvalds */ 1131da177e4SLinus Torvalds 1141da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1151da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1161da177e4SLinus Torvalds * kernel data space before using them.. 1171da177e4SLinus Torvalds * 1181da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1191da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1201da177e4SLinus Torvalds */ 1217950e385SJeff Layton 1227950e385SJeff Layton #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) 12391a27b2aSJeff Layton 12451f39a1fSDavid Drysdale struct filename * 12591a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12691a27b2aSJeff Layton { 12791a27b2aSJeff Layton struct filename *result, *err; 1283f9f0aa6SLinus Torvalds int len; 1297950e385SJeff Layton long max; 1307950e385SJeff Layton char *kname; 1311da177e4SLinus Torvalds 1327ac86265SJeff Layton result = audit_reusename(filename); 1337ac86265SJeff Layton if (result) 1347ac86265SJeff Layton return result; 1357ac86265SJeff Layton 1367950e385SJeff Layton result = __getname(); 1373f9f0aa6SLinus Torvalds if (unlikely(!result)) 1384043cde8SEric Paris return ERR_PTR(-ENOMEM); 13955422d0bSPaul Moore result->refcnt = 1; 1401da177e4SLinus Torvalds 1417950e385SJeff Layton /* 1427950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1437950e385SJeff Layton * allocation 1447950e385SJeff Layton */ 1457950e385SJeff Layton kname = (char *)result + sizeof(*result); 14691a27b2aSJeff Layton result->name = kname; 1477950e385SJeff Layton result->separate = false; 1487950e385SJeff Layton max = EMBEDDED_NAME_MAX; 1497950e385SJeff Layton 1507950e385SJeff Layton recopy: 1517950e385SJeff Layton len = strncpy_from_user(kname, filename, max); 15291a27b2aSJeff Layton if (unlikely(len < 0)) { 1533f9f0aa6SLinus Torvalds err = ERR_PTR(len); 1543f9f0aa6SLinus Torvalds goto error; 15591a27b2aSJeff Layton } 1563f9f0aa6SLinus Torvalds 1577950e385SJeff Layton /* 1587950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1597950e385SJeff Layton * separate struct filename so we can dedicate the entire 1607950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1617950e385SJeff Layton * userland. 1627950e385SJeff Layton */ 1637950e385SJeff Layton if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { 1647950e385SJeff Layton kname = (char *)result; 1657950e385SJeff Layton 1667950e385SJeff Layton result = kzalloc(sizeof(*result), GFP_KERNEL); 1677950e385SJeff Layton if (!result) { 1687950e385SJeff Layton err = ERR_PTR(-ENOMEM); 1697950e385SJeff Layton result = (struct filename *)kname; 1707950e385SJeff Layton goto error; 1717950e385SJeff Layton } 1727950e385SJeff Layton result->name = kname; 1737950e385SJeff Layton result->separate = true; 17455422d0bSPaul Moore result->refcnt = 1; 1757950e385SJeff Layton max = PATH_MAX; 1767950e385SJeff Layton goto recopy; 1777950e385SJeff Layton } 1787950e385SJeff Layton 1793f9f0aa6SLinus Torvalds /* The empty path is special. */ 1803f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1813f9f0aa6SLinus Torvalds if (empty) 1821fa1e7f6SAndy Whitcroft *empty = 1; 1833f9f0aa6SLinus Torvalds err = ERR_PTR(-ENOENT); 1843f9f0aa6SLinus Torvalds if (!(flags & LOOKUP_EMPTY)) 1853f9f0aa6SLinus Torvalds goto error; 1861da177e4SLinus Torvalds } 1873f9f0aa6SLinus Torvalds 1883f9f0aa6SLinus Torvalds err = ERR_PTR(-ENAMETOOLONG); 1897950e385SJeff Layton if (unlikely(len >= PATH_MAX)) 1907950e385SJeff Layton goto error; 1917950e385SJeff Layton 1927950e385SJeff Layton result->uptr = filename; 193c4ad8f98SLinus Torvalds result->aname = NULL; 1941da177e4SLinus Torvalds audit_getname(result); 1951da177e4SLinus Torvalds return result; 1961da177e4SLinus Torvalds 1973f9f0aa6SLinus Torvalds error: 19855422d0bSPaul Moore putname(result); 1993f9f0aa6SLinus Torvalds return err; 2003f9f0aa6SLinus Torvalds } 2013f9f0aa6SLinus Torvalds 20291a27b2aSJeff Layton struct filename * 20391a27b2aSJeff Layton getname(const char __user * filename) 204f52e0c11SAl Viro { 205f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 206f52e0c11SAl Viro } 207f52e0c11SAl Viro 208c4ad8f98SLinus Torvalds struct filename * 209c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 210c4ad8f98SLinus Torvalds { 211c4ad8f98SLinus Torvalds struct filename *result; 21208518549SPaul Moore int len = strlen(filename) + 1; 213c4ad8f98SLinus Torvalds 214c4ad8f98SLinus Torvalds result = __getname(); 215c4ad8f98SLinus Torvalds if (unlikely(!result)) 216c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 217c4ad8f98SLinus Torvalds 21808518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 21908518549SPaul Moore result->name = (char *)(result) + sizeof(*result); 22008518549SPaul Moore result->separate = false; 22108518549SPaul Moore } else if (len <= PATH_MAX) { 22208518549SPaul Moore struct filename *tmp; 22308518549SPaul Moore 22408518549SPaul Moore tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 22508518549SPaul Moore if (unlikely(!tmp)) { 22608518549SPaul Moore __putname(result); 22708518549SPaul Moore return ERR_PTR(-ENOMEM); 22808518549SPaul Moore } 22908518549SPaul Moore tmp->name = (char *)result; 23008518549SPaul Moore tmp->separate = true; 23108518549SPaul Moore result = tmp; 23208518549SPaul Moore } else { 23308518549SPaul Moore __putname(result); 23408518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23508518549SPaul Moore } 23608518549SPaul Moore memcpy((char *)result->name, filename, len); 237c4ad8f98SLinus Torvalds result->uptr = NULL; 238c4ad8f98SLinus Torvalds result->aname = NULL; 23955422d0bSPaul Moore result->refcnt = 1; 240fd3522fdSPaul Moore audit_getname(result); 241c4ad8f98SLinus Torvalds 242c4ad8f98SLinus Torvalds return result; 243c4ad8f98SLinus Torvalds } 244c4ad8f98SLinus Torvalds 24591a27b2aSJeff Layton void putname(struct filename *name) 2461da177e4SLinus Torvalds { 24755422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 24855422d0bSPaul Moore 24955422d0bSPaul Moore if (--name->refcnt > 0) 25055422d0bSPaul Moore return; 25155422d0bSPaul Moore 25255422d0bSPaul Moore if (name->separate) { 25355422d0bSPaul Moore __putname(name->name); 25455422d0bSPaul Moore kfree(name); 25555422d0bSPaul Moore } else 25655422d0bSPaul Moore __putname(name); 2571da177e4SLinus Torvalds } 2581da177e4SLinus Torvalds 259e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 260e77819e5SLinus Torvalds { 26184635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 262e77819e5SLinus Torvalds struct posix_acl *acl; 263e77819e5SLinus Torvalds 264e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2653567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2663567866bSAl Viro if (!acl) 267e77819e5SLinus Torvalds return -EAGAIN; 2683567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 2693567866bSAl Viro if (acl == ACL_NOT_CACHED) 270e77819e5SLinus Torvalds return -ECHILD; 271206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 272e77819e5SLinus Torvalds } 273e77819e5SLinus Torvalds 2742982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2754e34e719SChristoph Hellwig if (IS_ERR(acl)) 2764e34e719SChristoph Hellwig return PTR_ERR(acl); 277e77819e5SLinus Torvalds if (acl) { 278e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 279e77819e5SLinus Torvalds posix_acl_release(acl); 280e77819e5SLinus Torvalds return error; 281e77819e5SLinus Torvalds } 28284635d68SLinus Torvalds #endif 283e77819e5SLinus Torvalds 284e77819e5SLinus Torvalds return -EAGAIN; 285e77819e5SLinus Torvalds } 286e77819e5SLinus Torvalds 2875909ccaaSLinus Torvalds /* 288948409c7SAndreas Gruenbacher * This does the basic permission checking 2895909ccaaSLinus Torvalds */ 2907e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2915909ccaaSLinus Torvalds { 29226cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2935909ccaaSLinus Torvalds 2948e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2955909ccaaSLinus Torvalds mode >>= 6; 2965909ccaaSLinus Torvalds else { 297e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2987e40145eSAl Viro int error = check_acl(inode, mask); 2995909ccaaSLinus Torvalds if (error != -EAGAIN) 3005909ccaaSLinus Torvalds return error; 3015909ccaaSLinus Torvalds } 3025909ccaaSLinus Torvalds 3035909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3045909ccaaSLinus Torvalds mode >>= 3; 3055909ccaaSLinus Torvalds } 3065909ccaaSLinus Torvalds 3075909ccaaSLinus Torvalds /* 3085909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3095909ccaaSLinus Torvalds */ 3109c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3115909ccaaSLinus Torvalds return 0; 3125909ccaaSLinus Torvalds return -EACCES; 3135909ccaaSLinus Torvalds } 3141da177e4SLinus Torvalds 3151da177e4SLinus Torvalds /** 3161da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3171da177e4SLinus Torvalds * @inode: inode to check access rights for 3188fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3191da177e4SLinus Torvalds * 3201da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3211da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3221da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 323b74c79e9SNick Piggin * are used for other things. 324b74c79e9SNick Piggin * 325b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 326b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 327b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3281da177e4SLinus Torvalds */ 3292830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3301da177e4SLinus Torvalds { 3315909ccaaSLinus Torvalds int ret; 3321da177e4SLinus Torvalds 3331da177e4SLinus Torvalds /* 334948409c7SAndreas Gruenbacher * Do the basic permission checks. 3351da177e4SLinus Torvalds */ 3367e40145eSAl Viro ret = acl_permission_check(inode, mask); 3375909ccaaSLinus Torvalds if (ret != -EACCES) 3385909ccaaSLinus Torvalds return ret; 3391da177e4SLinus Torvalds 340d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 341d594e7ecSAl Viro /* DACs are overridable for directories */ 34223adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 343d594e7ecSAl Viro return 0; 344d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34523adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34623adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 347d594e7ecSAl Viro return 0; 348d594e7ecSAl Viro return -EACCES; 349d594e7ecSAl Viro } 3501da177e4SLinus Torvalds /* 3511da177e4SLinus Torvalds * Read/write DACs are always overridable. 352d594e7ecSAl Viro * Executable DACs are overridable when there is 353d594e7ecSAl Viro * at least one exec bit set. 3541da177e4SLinus Torvalds */ 355d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 35623adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3571da177e4SLinus Torvalds return 0; 3581da177e4SLinus Torvalds 3591da177e4SLinus Torvalds /* 3601da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3611da177e4SLinus Torvalds */ 3627ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 363d594e7ecSAl Viro if (mask == MAY_READ) 36423adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3651da177e4SLinus Torvalds return 0; 3661da177e4SLinus Torvalds 3671da177e4SLinus Torvalds return -EACCES; 3681da177e4SLinus Torvalds } 3694d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3701da177e4SLinus Torvalds 3713ddcd056SLinus Torvalds /* 3723ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3733ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3743ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3753ddcd056SLinus Torvalds * permission function, use the fast case". 3763ddcd056SLinus Torvalds */ 3773ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3783ddcd056SLinus Torvalds { 3793ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3803ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3813ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3823ddcd056SLinus Torvalds 3833ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3843ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3853ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3863ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3873ddcd056SLinus Torvalds } 3883ddcd056SLinus Torvalds return generic_permission(inode, mask); 3893ddcd056SLinus Torvalds } 3903ddcd056SLinus Torvalds 391cb23beb5SChristoph Hellwig /** 3920bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3930bdaea90SDavid Howells * @inode: Inode to check permission on 3940bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 395cb23beb5SChristoph Hellwig * 3960bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 397948409c7SAndreas Gruenbacher * 398948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3990bdaea90SDavid Howells * 4000bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 4010bdaea90SDavid Howells * inode_permission(). 402cb23beb5SChristoph Hellwig */ 4030bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 4041da177e4SLinus Torvalds { 405e6305c43SAl Viro int retval; 4061da177e4SLinus Torvalds 4073ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 4081da177e4SLinus Torvalds /* 4091da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4101da177e4SLinus Torvalds */ 4111da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 4121da177e4SLinus Torvalds return -EACCES; 4131da177e4SLinus Torvalds } 4141da177e4SLinus Torvalds 4153ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4161da177e4SLinus Torvalds if (retval) 4171da177e4SLinus Torvalds return retval; 4181da177e4SLinus Torvalds 41908ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 42008ce5f16SSerge E. Hallyn if (retval) 42108ce5f16SSerge E. Hallyn return retval; 42208ce5f16SSerge E. Hallyn 423d09ca739SEric Paris return security_inode_permission(inode, mask); 4241da177e4SLinus Torvalds } 425bd5d0856SMiklos Szeredi EXPORT_SYMBOL(__inode_permission); 4261da177e4SLinus Torvalds 427f4d6ff89SAl Viro /** 4280bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4290bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 43055852635SRandy Dunlap * @inode: Inode to check permission on 4310bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4320bdaea90SDavid Howells * 4330bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4340bdaea90SDavid Howells */ 4350bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4360bdaea90SDavid Howells { 4370bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4380bdaea90SDavid Howells umode_t mode = inode->i_mode; 4390bdaea90SDavid Howells 4400bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4410bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4420bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4430bdaea90SDavid Howells return -EROFS; 4440bdaea90SDavid Howells } 4450bdaea90SDavid Howells return 0; 4460bdaea90SDavid Howells } 4470bdaea90SDavid Howells 4480bdaea90SDavid Howells /** 4490bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4500bdaea90SDavid Howells * @inode: Inode to check permission on 4510bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4520bdaea90SDavid Howells * 4530bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4540bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4550bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4560bdaea90SDavid Howells * 4570bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4580bdaea90SDavid Howells */ 4590bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4600bdaea90SDavid Howells { 4610bdaea90SDavid Howells int retval; 4620bdaea90SDavid Howells 4630bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4640bdaea90SDavid Howells if (retval) 4650bdaea90SDavid Howells return retval; 4660bdaea90SDavid Howells return __inode_permission(inode, mask); 4670bdaea90SDavid Howells } 4684d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4690bdaea90SDavid Howells 4700bdaea90SDavid Howells /** 4715dd784d0SJan Blunck * path_get - get a reference to a path 4725dd784d0SJan Blunck * @path: path to get the reference to 4735dd784d0SJan Blunck * 4745dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4755dd784d0SJan Blunck */ 476dcf787f3SAl Viro void path_get(const struct path *path) 4775dd784d0SJan Blunck { 4785dd784d0SJan Blunck mntget(path->mnt); 4795dd784d0SJan Blunck dget(path->dentry); 4805dd784d0SJan Blunck } 4815dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4825dd784d0SJan Blunck 4835dd784d0SJan Blunck /** 4841d957f9bSJan Blunck * path_put - put a reference to a path 4851d957f9bSJan Blunck * @path: path to put the reference to 4861d957f9bSJan Blunck * 4871d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4881d957f9bSJan Blunck */ 489dcf787f3SAl Viro void path_put(const struct path *path) 4901da177e4SLinus Torvalds { 4911d957f9bSJan Blunck dput(path->dentry); 4921d957f9bSJan Blunck mntput(path->mnt); 4931da177e4SLinus Torvalds } 4941d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4951da177e4SLinus Torvalds 4961f55a6ecSAl Viro struct nameidata { 4971f55a6ecSAl Viro struct path path; 4981f55a6ecSAl Viro struct qstr last; 4991f55a6ecSAl Viro struct path root; 5001f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5011f55a6ecSAl Viro unsigned int flags; 5021f55a6ecSAl Viro unsigned seq, m_seq; 5031f55a6ecSAl Viro int last_type; 5041f55a6ecSAl Viro unsigned depth; 5055e53084dSAl Viro struct file *base; 5061f55a6ecSAl Viro char *saved_names[MAX_NESTED_LINKS + 1]; 5071f55a6ecSAl Viro }; 5081f55a6ecSAl Viro 50919660af7SAl Viro /* 51031e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 51119660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 51219660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 51319660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 51419660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 51519660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 51619660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 51719660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 51831e6b01fSNick Piggin */ 51931e6b01fSNick Piggin 52031e6b01fSNick Piggin /** 52119660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 52219660af7SAl Viro * @nd: nameidata pathwalk data 52319660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 52439191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 52531e6b01fSNick Piggin * 52619660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 52719660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 52819660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 52931e6b01fSNick Piggin */ 53019660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 53131e6b01fSNick Piggin { 53231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 53331e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 53431e6b01fSNick Piggin 53531e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 536e5c832d5SLinus Torvalds 537e5c832d5SLinus Torvalds /* 53848a066e7SAl Viro * After legitimizing the bastards, terminate_walk() 53948a066e7SAl Viro * will do the right thing for non-RCU mode, and all our 54048a066e7SAl Viro * subsequent exit cases should rcu_read_unlock() 54148a066e7SAl Viro * before returning. Do vfsmount first; if dentry 54248a066e7SAl Viro * can't be legitimized, just set nd->path.dentry to NULL 54348a066e7SAl Viro * and rely on dput(NULL) being a no-op. 544e5c832d5SLinus Torvalds */ 54548a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 546e5c832d5SLinus Torvalds return -ECHILD; 547e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 54815570086SLinus Torvalds 54948a066e7SAl Viro if (!lockref_get_not_dead(&parent->d_lockref)) { 55048a066e7SAl Viro nd->path.dentry = NULL; 551d870b4a1SAl Viro goto out; 55248a066e7SAl Viro } 55348a066e7SAl Viro 55415570086SLinus Torvalds /* 55515570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 55615570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 55715570086SLinus Torvalds * 55815570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 55915570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 56015570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 56115570086SLinus Torvalds * since we checked the sequence number of the parent after we got 56215570086SLinus Torvalds * the child sequence number. So we know the parent must still 56315570086SLinus Torvalds * be valid if the child sequence number is still valid. 56415570086SLinus Torvalds */ 56519660af7SAl Viro if (!dentry) { 566e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 567e5c832d5SLinus Torvalds goto out; 56819660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 56919660af7SAl Viro } else { 570e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 571e5c832d5SLinus Torvalds goto out; 572e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 573e5c832d5SLinus Torvalds goto drop_dentry; 57419660af7SAl Viro } 575e5c832d5SLinus Torvalds 576e5c832d5SLinus Torvalds /* 577e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 578e5c832d5SLinus Torvalds * still valid and get it if required. 579e5c832d5SLinus Torvalds */ 580e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 581e5c832d5SLinus Torvalds spin_lock(&fs->lock); 582e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 583e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 58431e6b01fSNick Piggin path_get(&nd->root); 58531e6b01fSNick Piggin spin_unlock(&fs->lock); 58631e6b01fSNick Piggin } 58731e6b01fSNick Piggin 5888b61e74fSAl Viro rcu_read_unlock(); 58931e6b01fSNick Piggin return 0; 59019660af7SAl Viro 591e5c832d5SLinus Torvalds unlock_and_drop_dentry: 59231e6b01fSNick Piggin spin_unlock(&fs->lock); 593e5c832d5SLinus Torvalds drop_dentry: 5948b61e74fSAl Viro rcu_read_unlock(); 595e5c832d5SLinus Torvalds dput(dentry); 596d0d27277SLinus Torvalds goto drop_root_mnt; 597e5c832d5SLinus Torvalds out: 5988b61e74fSAl Viro rcu_read_unlock(); 599d0d27277SLinus Torvalds drop_root_mnt: 600d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 601d0d27277SLinus Torvalds nd->root.mnt = NULL; 60231e6b01fSNick Piggin return -ECHILD; 60331e6b01fSNick Piggin } 60431e6b01fSNick Piggin 6054ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 60634286d66SNick Piggin { 6074ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 60834286d66SNick Piggin } 60934286d66SNick Piggin 6109f1fafeeSAl Viro /** 6119f1fafeeSAl Viro * complete_walk - successful completion of path walk 6129f1fafeeSAl Viro * @nd: pointer nameidata 61339159de2SJeff Layton * 6149f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 6159f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 6169f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 6179f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 6189f1fafeeSAl Viro * need to drop nd->path. 61939159de2SJeff Layton */ 6209f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 62139159de2SJeff Layton { 62216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 62339159de2SJeff Layton int status; 62439159de2SJeff Layton 6259f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 6269f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 6279f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6289f1fafeeSAl Viro nd->root.mnt = NULL; 62915570086SLinus Torvalds 63048a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { 6318b61e74fSAl Viro rcu_read_unlock(); 63248a066e7SAl Viro return -ECHILD; 63348a066e7SAl Viro } 634e5c832d5SLinus Torvalds if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 6358b61e74fSAl Viro rcu_read_unlock(); 63648a066e7SAl Viro mntput(nd->path.mnt); 6379f1fafeeSAl Viro return -ECHILD; 6389f1fafeeSAl Viro } 639e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 6408b61e74fSAl Viro rcu_read_unlock(); 641e5c832d5SLinus Torvalds dput(dentry); 64248a066e7SAl Viro mntput(nd->path.mnt); 643e5c832d5SLinus Torvalds return -ECHILD; 644e5c832d5SLinus Torvalds } 6458b61e74fSAl Viro rcu_read_unlock(); 6469f1fafeeSAl Viro } 6479f1fafeeSAl Viro 64816c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 64939159de2SJeff Layton return 0; 65039159de2SJeff Layton 651ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 65216c2cd71SAl Viro return 0; 65316c2cd71SAl Viro 654ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 65539159de2SJeff Layton if (status > 0) 65639159de2SJeff Layton return 0; 65739159de2SJeff Layton 65816c2cd71SAl Viro if (!status) 65939159de2SJeff Layton status = -ESTALE; 66016c2cd71SAl Viro 6619f1fafeeSAl Viro path_put(&nd->path); 66239159de2SJeff Layton return status; 66339159de2SJeff Layton } 66439159de2SJeff Layton 6652a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6662a737871SAl Viro { 667f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6682a737871SAl Viro } 6692a737871SAl Viro 6706de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6716de88d72SAl Viro 6727bd88377SAl Viro static __always_inline unsigned set_root_rcu(struct nameidata *nd) 67331e6b01fSNick Piggin { 67431e6b01fSNick Piggin struct fs_struct *fs = current->fs; 6757bd88377SAl Viro unsigned seq, res; 676c28cc364SNick Piggin 677c28cc364SNick Piggin do { 678c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 67931e6b01fSNick Piggin nd->root = fs->root; 6807bd88377SAl Viro res = __read_seqcount_begin(&nd->root.dentry->d_seq); 681c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 6827bd88377SAl Viro return res; 68331e6b01fSNick Piggin } 68431e6b01fSNick Piggin 6851d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 686051d3812SIan Kent { 687051d3812SIan Kent dput(path->dentry); 6884ac91378SJan Blunck if (path->mnt != nd->path.mnt) 689051d3812SIan Kent mntput(path->mnt); 690051d3812SIan Kent } 691051d3812SIan Kent 6927b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6937b9337aaSNick Piggin struct nameidata *nd) 694051d3812SIan Kent { 69531e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6964ac91378SJan Blunck dput(nd->path.dentry); 69731e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6984ac91378SJan Blunck mntput(nd->path.mnt); 6999a229683SHuang Shijie } 70031e6b01fSNick Piggin nd->path.mnt = path->mnt; 7014ac91378SJan Blunck nd->path.dentry = path->dentry; 702051d3812SIan Kent } 703051d3812SIan Kent 704b5fb63c1SChristoph Hellwig /* 705b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 706b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 707b5fb63c1SChristoph Hellwig */ 708b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 709b5fb63c1SChristoph Hellwig { 710b5fb63c1SChristoph Hellwig path_put(&nd->path); 711b5fb63c1SChristoph Hellwig 712b5fb63c1SChristoph Hellwig nd->path = *path; 713b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 714b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 715b5fb63c1SChristoph Hellwig } 716b5fb63c1SChristoph Hellwig 7171f55a6ecSAl Viro void nd_set_link(struct nameidata *nd, char *path) 7181f55a6ecSAl Viro { 7191f55a6ecSAl Viro nd->saved_names[nd->depth] = path; 7201f55a6ecSAl Viro } 7211f55a6ecSAl Viro EXPORT_SYMBOL(nd_set_link); 7221f55a6ecSAl Viro 7231f55a6ecSAl Viro char *nd_get_link(struct nameidata *nd) 7241f55a6ecSAl Viro { 7251f55a6ecSAl Viro return nd->saved_names[nd->depth]; 7261f55a6ecSAl Viro } 7271f55a6ecSAl Viro EXPORT_SYMBOL(nd_get_link); 7281f55a6ecSAl Viro 729574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 730574197e0SAl Viro { 731574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 7326d7b5aaeSAl Viro if (inode->i_op->put_link) 733574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 734574197e0SAl Viro path_put(link); 735574197e0SAl Viro } 736574197e0SAl Viro 737561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 738561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 739800179c9SKees Cook 740800179c9SKees Cook /** 741800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 742800179c9SKees Cook * @link: The path of the symlink 74355852635SRandy Dunlap * @nd: nameidata pathwalk data 744800179c9SKees Cook * 745800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 746800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 747800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 748800179c9SKees Cook * processes from failing races against path names that may change out 749800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 750800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 751800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 752800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 753800179c9SKees Cook * 754800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 755800179c9SKees Cook */ 756800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 757800179c9SKees Cook { 758800179c9SKees Cook const struct inode *inode; 759800179c9SKees Cook const struct inode *parent; 760800179c9SKees Cook 761800179c9SKees Cook if (!sysctl_protected_symlinks) 762800179c9SKees Cook return 0; 763800179c9SKees Cook 764800179c9SKees Cook /* Allowed if owner and follower match. */ 765800179c9SKees Cook inode = link->dentry->d_inode; 76681abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 767800179c9SKees Cook return 0; 768800179c9SKees Cook 769800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 770800179c9SKees Cook parent = nd->path.dentry->d_inode; 771800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 772800179c9SKees Cook return 0; 773800179c9SKees Cook 774800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 77581abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 776800179c9SKees Cook return 0; 777800179c9SKees Cook 778ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 779800179c9SKees Cook path_put_conditional(link, nd); 780800179c9SKees Cook path_put(&nd->path); 781800179c9SKees Cook return -EACCES; 782800179c9SKees Cook } 783800179c9SKees Cook 784800179c9SKees Cook /** 785800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 786800179c9SKees Cook * @inode: the source inode to hardlink from 787800179c9SKees Cook * 788800179c9SKees Cook * Return false if at least one of the following conditions: 789800179c9SKees Cook * - inode is not a regular file 790800179c9SKees Cook * - inode is setuid 791800179c9SKees Cook * - inode is setgid and group-exec 792800179c9SKees Cook * - access failure for read and write 793800179c9SKees Cook * 794800179c9SKees Cook * Otherwise returns true. 795800179c9SKees Cook */ 796800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 797800179c9SKees Cook { 798800179c9SKees Cook umode_t mode = inode->i_mode; 799800179c9SKees Cook 800800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 801800179c9SKees Cook if (!S_ISREG(mode)) 802800179c9SKees Cook return false; 803800179c9SKees Cook 804800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 805800179c9SKees Cook if (mode & S_ISUID) 806800179c9SKees Cook return false; 807800179c9SKees Cook 808800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 809800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 810800179c9SKees Cook return false; 811800179c9SKees Cook 812800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 813800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 814800179c9SKees Cook return false; 815800179c9SKees Cook 816800179c9SKees Cook return true; 817800179c9SKees Cook } 818800179c9SKees Cook 819800179c9SKees Cook /** 820800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 821800179c9SKees Cook * @link: the source to hardlink from 822800179c9SKees Cook * 823800179c9SKees Cook * Block hardlink when all of: 824800179c9SKees Cook * - sysctl_protected_hardlinks enabled 825800179c9SKees Cook * - fsuid does not match inode 826800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 827800179c9SKees Cook * - not CAP_FOWNER 828800179c9SKees Cook * 829800179c9SKees Cook * Returns 0 if successful, -ve on error. 830800179c9SKees Cook */ 831800179c9SKees Cook static int may_linkat(struct path *link) 832800179c9SKees Cook { 833800179c9SKees Cook const struct cred *cred; 834800179c9SKees Cook struct inode *inode; 835800179c9SKees Cook 836800179c9SKees Cook if (!sysctl_protected_hardlinks) 837800179c9SKees Cook return 0; 838800179c9SKees Cook 839800179c9SKees Cook cred = current_cred(); 840800179c9SKees Cook inode = link->dentry->d_inode; 841800179c9SKees Cook 842800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 843800179c9SKees Cook * otherwise, it must be a safe source. 844800179c9SKees Cook */ 84581abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 846800179c9SKees Cook capable(CAP_FOWNER)) 847800179c9SKees Cook return 0; 848800179c9SKees Cook 849a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 850800179c9SKees Cook return -EPERM; 851800179c9SKees Cook } 852800179c9SKees Cook 853def4af30SAl Viro static __always_inline int 854574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 8551da177e4SLinus Torvalds { 8567b9337aaSNick Piggin struct dentry *dentry = link->dentry; 8576d7b5aaeSAl Viro int error; 8586d7b5aaeSAl Viro char *s; 8591da177e4SLinus Torvalds 860844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 861844a3917SAl Viro 8620e794589SAl Viro if (link->mnt == nd->path.mnt) 8630e794589SAl Viro mntget(link->mnt); 8640e794589SAl Viro 8656d7b5aaeSAl Viro error = -ELOOP; 8666d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 8676d7b5aaeSAl Viro goto out_put_nd_path; 8686d7b5aaeSAl Viro 869574197e0SAl Viro cond_resched(); 870574197e0SAl Viro current->total_link_count++; 871574197e0SAl Viro 87268ac1234SAl Viro touch_atime(link); 8731da177e4SLinus Torvalds nd_set_link(nd, NULL); 874cd4e91d3SAl Viro 87536f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 8766d7b5aaeSAl Viro if (error) 8776d7b5aaeSAl Viro goto out_put_nd_path; 87836f3b4f6SAl Viro 87986acdca1SAl Viro nd->last_type = LAST_BIND; 880def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 881def4af30SAl Viro error = PTR_ERR(*p); 8826d7b5aaeSAl Viro if (IS_ERR(*p)) 883408ef013SChristoph Hellwig goto out_put_nd_path; 8846d7b5aaeSAl Viro 885cc314eefSLinus Torvalds error = 0; 8866d7b5aaeSAl Viro s = nd_get_link(nd); 8876d7b5aaeSAl Viro if (s) { 888443ed254SAl Viro if (unlikely(IS_ERR(s))) { 889443ed254SAl Viro path_put(&nd->path); 890443ed254SAl Viro put_link(nd, link, *p); 891443ed254SAl Viro return PTR_ERR(s); 892443ed254SAl Viro } 893443ed254SAl Viro if (*s == '/') { 8947bd88377SAl Viro if (!nd->root.mnt) 895443ed254SAl Viro set_root(nd); 896443ed254SAl Viro path_put(&nd->path); 897443ed254SAl Viro nd->path = nd->root; 898443ed254SAl Viro path_get(&nd->root); 899443ed254SAl Viro nd->flags |= LOOKUP_JUMPED; 900443ed254SAl Viro } 901443ed254SAl Viro nd->inode = nd->path.dentry->d_inode; 902443ed254SAl Viro error = link_path_walk(s, nd); 9036d7b5aaeSAl Viro if (unlikely(error)) 9046d7b5aaeSAl Viro put_link(nd, link, *p); 905b5fb63c1SChristoph Hellwig } 9066d7b5aaeSAl Viro 9076d7b5aaeSAl Viro return error; 9086d7b5aaeSAl Viro 9096d7b5aaeSAl Viro out_put_nd_path: 91098f6ef64SArnd Bergmann *p = NULL; 9116d7b5aaeSAl Viro path_put(&nd->path); 9126d7b5aaeSAl Viro path_put(link); 9131da177e4SLinus Torvalds return error; 9141da177e4SLinus Torvalds } 9151da177e4SLinus Torvalds 91631e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 91731e6b01fSNick Piggin { 9180714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9190714a533SAl Viro struct mount *parent; 92031e6b01fSNick Piggin struct dentry *mountpoint; 92131e6b01fSNick Piggin 9220714a533SAl Viro parent = mnt->mnt_parent; 9230714a533SAl Viro if (&parent->mnt == path->mnt) 92431e6b01fSNick Piggin return 0; 925a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 92631e6b01fSNick Piggin path->dentry = mountpoint; 9270714a533SAl Viro path->mnt = &parent->mnt; 92831e6b01fSNick Piggin return 1; 92931e6b01fSNick Piggin } 93031e6b01fSNick Piggin 931f015f126SDavid Howells /* 932f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 933f015f126SDavid Howells * 934f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 935f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 936f015f126SDavid Howells * Up is towards /. 937f015f126SDavid Howells * 938f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 939f015f126SDavid Howells * root. 940f015f126SDavid Howells */ 941bab77ebfSAl Viro int follow_up(struct path *path) 9421da177e4SLinus Torvalds { 9430714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9440714a533SAl Viro struct mount *parent; 9451da177e4SLinus Torvalds struct dentry *mountpoint; 94699b7db7bSNick Piggin 94748a066e7SAl Viro read_seqlock_excl(&mount_lock); 9480714a533SAl Viro parent = mnt->mnt_parent; 9493c0a6163SAl Viro if (parent == mnt) { 95048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 9511da177e4SLinus Torvalds return 0; 9521da177e4SLinus Torvalds } 9530714a533SAl Viro mntget(&parent->mnt); 954a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 95548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 956bab77ebfSAl Viro dput(path->dentry); 957bab77ebfSAl Viro path->dentry = mountpoint; 958bab77ebfSAl Viro mntput(path->mnt); 9590714a533SAl Viro path->mnt = &parent->mnt; 9601da177e4SLinus Torvalds return 1; 9611da177e4SLinus Torvalds } 9624d359507SAl Viro EXPORT_SYMBOL(follow_up); 9631da177e4SLinus Torvalds 964b5c84bf6SNick Piggin /* 9659875cf80SDavid Howells * Perform an automount 9669875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9679875cf80SDavid Howells * were called with. 9681da177e4SLinus Torvalds */ 9699875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 9709875cf80SDavid Howells bool *need_mntput) 97131e6b01fSNick Piggin { 9729875cf80SDavid Howells struct vfsmount *mnt; 973ea5b778aSDavid Howells int err; 9749875cf80SDavid Howells 9759875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9769875cf80SDavid Howells return -EREMOTE; 9779875cf80SDavid Howells 9780ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9790ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9800ec26fd0SMiklos Szeredi * the name. 9810ec26fd0SMiklos Szeredi * 9820ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9835a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9840ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9850ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9860ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9870ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9885a30d8a2SDavid Howells */ 9895a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 990d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 9915a30d8a2SDavid Howells path->dentry->d_inode) 9929875cf80SDavid Howells return -EISDIR; 9930ec26fd0SMiklos Szeredi 9949875cf80SDavid Howells current->total_link_count++; 9959875cf80SDavid Howells if (current->total_link_count >= 40) 9969875cf80SDavid Howells return -ELOOP; 9979875cf80SDavid Howells 9989875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9999875cf80SDavid Howells if (IS_ERR(mnt)) { 10009875cf80SDavid Howells /* 10019875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 10029875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 10039875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 10049875cf80SDavid Howells * 10059875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 10069875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 10079875cf80SDavid Howells * the path is inaccessible and we should say so. 10089875cf80SDavid Howells */ 100949084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 10109875cf80SDavid Howells return -EREMOTE; 10119875cf80SDavid Howells return PTR_ERR(mnt); 101231e6b01fSNick Piggin } 1013ea5b778aSDavid Howells 10149875cf80SDavid Howells if (!mnt) /* mount collision */ 10159875cf80SDavid Howells return 0; 10169875cf80SDavid Howells 10178aef1884SAl Viro if (!*need_mntput) { 10188aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 10198aef1884SAl Viro mntget(path->mnt); 10208aef1884SAl Viro *need_mntput = true; 10218aef1884SAl Viro } 102219a167afSAl Viro err = finish_automount(mnt, path); 1023ea5b778aSDavid Howells 1024ea5b778aSDavid Howells switch (err) { 1025ea5b778aSDavid Howells case -EBUSY: 1026ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 102719a167afSAl Viro return 0; 1028ea5b778aSDavid Howells case 0: 10298aef1884SAl Viro path_put(path); 10309875cf80SDavid Howells path->mnt = mnt; 10319875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 10329875cf80SDavid Howells return 0; 103319a167afSAl Viro default: 103419a167afSAl Viro return err; 10359875cf80SDavid Howells } 103619a167afSAl Viro 1037ea5b778aSDavid Howells } 10389875cf80SDavid Howells 10399875cf80SDavid Howells /* 10409875cf80SDavid Howells * Handle a dentry that is managed in some way. 1041cc53ce53SDavid Howells * - Flagged for transit management (autofs) 10429875cf80SDavid Howells * - Flagged as mountpoint 10439875cf80SDavid Howells * - Flagged as automount point 10449875cf80SDavid Howells * 10459875cf80SDavid Howells * This may only be called in refwalk mode. 10469875cf80SDavid Howells * 10479875cf80SDavid Howells * Serialization is taken care of in namespace.c 10489875cf80SDavid Howells */ 10499875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 10509875cf80SDavid Howells { 10518aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10529875cf80SDavid Howells unsigned managed; 10539875cf80SDavid Howells bool need_mntput = false; 10548aef1884SAl Viro int ret = 0; 10559875cf80SDavid Howells 10569875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10579875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10589875cf80SDavid Howells * the components of that value change under us */ 10599875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10609875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10619875cf80SDavid Howells unlikely(managed != 0)) { 1062cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1063cc53ce53SDavid Howells * being held. */ 1064cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1065cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1066cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10671aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1068cc53ce53SDavid Howells if (ret < 0) 10698aef1884SAl Viro break; 1070cc53ce53SDavid Howells } 1071cc53ce53SDavid Howells 10729875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10739875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10749875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10759875cf80SDavid Howells if (mounted) { 10769875cf80SDavid Howells dput(path->dentry); 10779875cf80SDavid Howells if (need_mntput) 1078463ffb2eSAl Viro mntput(path->mnt); 1079463ffb2eSAl Viro path->mnt = mounted; 1080463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10819875cf80SDavid Howells need_mntput = true; 10829875cf80SDavid Howells continue; 1083463ffb2eSAl Viro } 1084463ffb2eSAl Viro 10859875cf80SDavid Howells /* Something is mounted on this dentry in another 10869875cf80SDavid Howells * namespace and/or whatever was mounted there in this 108748a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 108848a066e7SAl Viro * get it */ 10891da177e4SLinus Torvalds } 10909875cf80SDavid Howells 10919875cf80SDavid Howells /* Handle an automount point */ 10929875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 10939875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 10949875cf80SDavid Howells if (ret < 0) 10958aef1884SAl Viro break; 10969875cf80SDavid Howells continue; 10979875cf80SDavid Howells } 10989875cf80SDavid Howells 10999875cf80SDavid Howells /* We didn't change the current path point */ 11009875cf80SDavid Howells break; 11019875cf80SDavid Howells } 11028aef1884SAl Viro 11038aef1884SAl Viro if (need_mntput && path->mnt == mnt) 11048aef1884SAl Viro mntput(path->mnt); 11058aef1884SAl Viro if (ret == -EISDIR) 11068aef1884SAl Viro ret = 0; 1107a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 11081da177e4SLinus Torvalds } 11091da177e4SLinus Torvalds 1110cc53ce53SDavid Howells int follow_down_one(struct path *path) 11111da177e4SLinus Torvalds { 11121da177e4SLinus Torvalds struct vfsmount *mounted; 11131da177e4SLinus Torvalds 11141c755af4SAl Viro mounted = lookup_mnt(path); 11151da177e4SLinus Torvalds if (mounted) { 11169393bd07SAl Viro dput(path->dentry); 11179393bd07SAl Viro mntput(path->mnt); 11189393bd07SAl Viro path->mnt = mounted; 11199393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 11201da177e4SLinus Torvalds return 1; 11211da177e4SLinus Torvalds } 11221da177e4SLinus Torvalds return 0; 11231da177e4SLinus Torvalds } 11244d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 11251da177e4SLinus Torvalds 1126b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 112762a7375eSIan Kent { 1128b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1129b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 113062a7375eSIan Kent } 113162a7375eSIan Kent 11329875cf80SDavid Howells /* 1133287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1134287548e4SAl Viro * we meet a managed dentry that would need blocking. 11359875cf80SDavid Howells */ 11369875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1137287548e4SAl Viro struct inode **inode) 11389875cf80SDavid Howells { 113962a7375eSIan Kent for (;;) { 1140c7105365SAl Viro struct mount *mounted; 114162a7375eSIan Kent /* 114262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 114362a7375eSIan Kent * that wants to block transit. 114462a7375eSIan Kent */ 1145b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1146b8faf035SNeilBrown case -ECHILD: 1147b8faf035SNeilBrown default: 1148ab90911fSDavid Howells return false; 1149b8faf035SNeilBrown case -EISDIR: 1150b8faf035SNeilBrown return true; 1151b8faf035SNeilBrown case 0: 1152b8faf035SNeilBrown break; 1153b8faf035SNeilBrown } 115462a7375eSIan Kent 115562a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1156b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 115762a7375eSIan Kent 1158474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 11599875cf80SDavid Howells if (!mounted) 11609875cf80SDavid Howells break; 1161c7105365SAl Viro path->mnt = &mounted->mnt; 1162c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1163a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 11649875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 116559430262SLinus Torvalds /* 116659430262SLinus Torvalds * Update the inode too. We don't need to re-check the 116759430262SLinus Torvalds * dentry sequence number here after this d_inode read, 116859430262SLinus Torvalds * because a mount-point is always pinned. 116959430262SLinus Torvalds */ 117059430262SLinus Torvalds *inode = path->dentry->d_inode; 11719875cf80SDavid Howells } 1172f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1173b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1174287548e4SAl Viro } 1175287548e4SAl Viro 117631e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 117731e6b01fSNick Piggin { 11784023bfc9SAl Viro struct inode *inode = nd->inode; 11797bd88377SAl Viro if (!nd->root.mnt) 118031e6b01fSNick Piggin set_root_rcu(nd); 118131e6b01fSNick Piggin 118231e6b01fSNick Piggin while (1) { 118331e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 118431e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 118531e6b01fSNick Piggin break; 118631e6b01fSNick Piggin } 118731e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 118831e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 118931e6b01fSNick Piggin struct dentry *parent = old->d_parent; 119031e6b01fSNick Piggin unsigned seq; 119131e6b01fSNick Piggin 11924023bfc9SAl Viro inode = parent->d_inode; 119331e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 119431e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1195ef7562d5SAl Viro goto failed; 119631e6b01fSNick Piggin nd->path.dentry = parent; 119731e6b01fSNick Piggin nd->seq = seq; 119831e6b01fSNick Piggin break; 119931e6b01fSNick Piggin } 120031e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 120131e6b01fSNick Piggin break; 12024023bfc9SAl Viro inode = nd->path.dentry->d_inode; 120331e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 120431e6b01fSNick Piggin } 1205b37199e6SAl Viro while (d_mountpoint(nd->path.dentry)) { 1206b37199e6SAl Viro struct mount *mounted; 1207b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1208b37199e6SAl Viro if (!mounted) 1209b37199e6SAl Viro break; 1210b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1211b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 12124023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1213b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1214f5be3e29SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1215b37199e6SAl Viro goto failed; 1216b37199e6SAl Viro } 12174023bfc9SAl Viro nd->inode = inode; 121831e6b01fSNick Piggin return 0; 1219ef7562d5SAl Viro 1220ef7562d5SAl Viro failed: 1221ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 12225b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1223ef7562d5SAl Viro nd->root.mnt = NULL; 12248b61e74fSAl Viro rcu_read_unlock(); 1225ef7562d5SAl Viro return -ECHILD; 122631e6b01fSNick Piggin } 122731e6b01fSNick Piggin 12289875cf80SDavid Howells /* 1229cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1230cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1231cc53ce53SDavid Howells * caller is permitted to proceed or not. 1232cc53ce53SDavid Howells */ 12337cc90cc3SAl Viro int follow_down(struct path *path) 1234cc53ce53SDavid Howells { 1235cc53ce53SDavid Howells unsigned managed; 1236cc53ce53SDavid Howells int ret; 1237cc53ce53SDavid Howells 1238cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1239cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1240cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1241cc53ce53SDavid Howells * being held. 1242cc53ce53SDavid Howells * 1243cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1244cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1245cc53ce53SDavid Howells * other than its daemon the right to mount on its 1246cc53ce53SDavid Howells * superstructure. 1247cc53ce53SDavid Howells * 1248cc53ce53SDavid Howells * The filesystem may sleep at this point. 1249cc53ce53SDavid Howells */ 1250cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1251cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1252cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1253ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 12541aed3e42SAl Viro path->dentry, false); 1255cc53ce53SDavid Howells if (ret < 0) 1256cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1257cc53ce53SDavid Howells } 1258cc53ce53SDavid Howells 1259cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1260cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1261cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1262cc53ce53SDavid Howells if (!mounted) 1263cc53ce53SDavid Howells break; 1264cc53ce53SDavid Howells dput(path->dentry); 1265cc53ce53SDavid Howells mntput(path->mnt); 1266cc53ce53SDavid Howells path->mnt = mounted; 1267cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1268cc53ce53SDavid Howells continue; 1269cc53ce53SDavid Howells } 1270cc53ce53SDavid Howells 1271cc53ce53SDavid Howells /* Don't handle automount points here */ 1272cc53ce53SDavid Howells break; 1273cc53ce53SDavid Howells } 1274cc53ce53SDavid Howells return 0; 1275cc53ce53SDavid Howells } 12764d359507SAl Viro EXPORT_SYMBOL(follow_down); 1277cc53ce53SDavid Howells 1278cc53ce53SDavid Howells /* 12799875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12809875cf80SDavid Howells */ 12819875cf80SDavid Howells static void follow_mount(struct path *path) 12829875cf80SDavid Howells { 12839875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12849875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12859875cf80SDavid Howells if (!mounted) 12869875cf80SDavid Howells break; 12879875cf80SDavid Howells dput(path->dentry); 12889875cf80SDavid Howells mntput(path->mnt); 12899875cf80SDavid Howells path->mnt = mounted; 12909875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12919875cf80SDavid Howells } 12929875cf80SDavid Howells } 12939875cf80SDavid Howells 129431e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 12951da177e4SLinus Torvalds { 12967bd88377SAl Viro if (!nd->root.mnt) 12972a737871SAl Viro set_root(nd); 1298e518ddb7SAndreas Mohr 12991da177e4SLinus Torvalds while(1) { 13004ac91378SJan Blunck struct dentry *old = nd->path.dentry; 13011da177e4SLinus Torvalds 13022a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 13032a737871SAl Viro nd->path.mnt == nd->root.mnt) { 13041da177e4SLinus Torvalds break; 13051da177e4SLinus Torvalds } 13064ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 13073088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 13083088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 13091da177e4SLinus Torvalds dput(old); 13101da177e4SLinus Torvalds break; 13111da177e4SLinus Torvalds } 13123088dd70SAl Viro if (!follow_up(&nd->path)) 13131da177e4SLinus Torvalds break; 13141da177e4SLinus Torvalds } 131579ed0226SAl Viro follow_mount(&nd->path); 131631e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 13171da177e4SLinus Torvalds } 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds /* 1320bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1321bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1322bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1323bad61189SMiklos Szeredi * 1324bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1325baa03890SNick Piggin */ 1326bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1327201f956eSAl Viro unsigned int flags, bool *need_lookup) 1328baa03890SNick Piggin { 1329baa03890SNick Piggin struct dentry *dentry; 1330bad61189SMiklos Szeredi int error; 1331baa03890SNick Piggin 1332bad61189SMiklos Szeredi *need_lookup = false; 1333bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1334bad61189SMiklos Szeredi if (dentry) { 133539e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1336201f956eSAl Viro error = d_revalidate(dentry, flags); 1337bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1338bad61189SMiklos Szeredi if (error < 0) { 1339bad61189SMiklos Szeredi dput(dentry); 1340bad61189SMiklos Szeredi return ERR_PTR(error); 13415542aa2fSEric W. Biederman } else { 13425542aa2fSEric W. Biederman d_invalidate(dentry); 1343bad61189SMiklos Szeredi dput(dentry); 1344bad61189SMiklos Szeredi dentry = NULL; 1345bad61189SMiklos Szeredi } 1346bad61189SMiklos Szeredi } 1347bad61189SMiklos Szeredi } 1348bad61189SMiklos Szeredi } 1349baa03890SNick Piggin 1350bad61189SMiklos Szeredi if (!dentry) { 1351bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1352baa03890SNick Piggin if (unlikely(!dentry)) 1353baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1354baa03890SNick Piggin 1355bad61189SMiklos Szeredi *need_lookup = true; 1356baa03890SNick Piggin } 1357baa03890SNick Piggin return dentry; 1358baa03890SNick Piggin } 1359baa03890SNick Piggin 1360baa03890SNick Piggin /* 136113a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 136213a2c3beSJ. Bruce Fields * unhashed. 1363bad61189SMiklos Szeredi * 1364bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 136544396f4bSJosef Bacik */ 1366bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 136772bd866aSAl Viro unsigned int flags) 136844396f4bSJosef Bacik { 136944396f4bSJosef Bacik struct dentry *old; 137044396f4bSJosef Bacik 137144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1372bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1373e188dc02SMiklos Szeredi dput(dentry); 137444396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1375e188dc02SMiklos Szeredi } 137644396f4bSJosef Bacik 137772bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 137844396f4bSJosef Bacik if (unlikely(old)) { 137944396f4bSJosef Bacik dput(dentry); 138044396f4bSJosef Bacik dentry = old; 138144396f4bSJosef Bacik } 138244396f4bSJosef Bacik return dentry; 138344396f4bSJosef Bacik } 138444396f4bSJosef Bacik 1385a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 138672bd866aSAl Viro struct dentry *base, unsigned int flags) 1387a3255546SAl Viro { 1388bad61189SMiklos Szeredi bool need_lookup; 1389a3255546SAl Viro struct dentry *dentry; 1390a3255546SAl Viro 139172bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1392bad61189SMiklos Szeredi if (!need_lookup) 1393a3255546SAl Viro return dentry; 1394bad61189SMiklos Szeredi 139572bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1396a3255546SAl Viro } 1397a3255546SAl Viro 139844396f4bSJosef Bacik /* 13991da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 14001da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 14011da177e4SLinus Torvalds * It _is_ time-critical. 14021da177e4SLinus Torvalds */ 1403e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 140431e6b01fSNick Piggin struct path *path, struct inode **inode) 14051da177e4SLinus Torvalds { 14064ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 140731e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 14085a18fff2SAl Viro int need_reval = 1; 14095a18fff2SAl Viro int status = 1; 14109875cf80SDavid Howells int err; 14119875cf80SDavid Howells 14123cac260aSAl Viro /* 1413b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1414b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1415b04f784eSNick Piggin * do the non-racy lookup, below. 1416b04f784eSNick Piggin */ 141731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 141831e6b01fSNick Piggin unsigned seq; 1419da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 14205a18fff2SAl Viro if (!dentry) 14215a18fff2SAl Viro goto unlazy; 14225a18fff2SAl Viro 142312f8ad4bSLinus Torvalds /* 142412f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 142512f8ad4bSLinus Torvalds * the dentry name information from lookup. 142612f8ad4bSLinus Torvalds */ 142712f8ad4bSLinus Torvalds *inode = dentry->d_inode; 142812f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 142912f8ad4bSLinus Torvalds return -ECHILD; 143012f8ad4bSLinus Torvalds 143112f8ad4bSLinus Torvalds /* 143212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 143312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 143412f8ad4bSLinus Torvalds * 143512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 143612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 143712f8ad4bSLinus Torvalds */ 143831e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 143931e6b01fSNick Piggin return -ECHILD; 144031e6b01fSNick Piggin nd->seq = seq; 14415a18fff2SAl Viro 144224643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 14434ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14445a18fff2SAl Viro if (unlikely(status <= 0)) { 14455a18fff2SAl Viro if (status != -ECHILD) 14465a18fff2SAl Viro need_reval = 0; 14475a18fff2SAl Viro goto unlazy; 14485a18fff2SAl Viro } 144924643087SAl Viro } 145031e6b01fSNick Piggin path->mnt = mnt; 145131e6b01fSNick Piggin path->dentry = dentry; 1452b8faf035SNeilBrown if (likely(__follow_mount_rcu(nd, path, inode))) 14539875cf80SDavid Howells return 0; 14545a18fff2SAl Viro unlazy: 145519660af7SAl Viro if (unlazy_walk(nd, dentry)) 14565a18fff2SAl Viro return -ECHILD; 14575a18fff2SAl Viro } else { 1458e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 145924643087SAl Viro } 14605a18fff2SAl Viro 146181e6f520SAl Viro if (unlikely(!dentry)) 146281e6f520SAl Viro goto need_lookup; 14635a18fff2SAl Viro 14645a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14654ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14665a18fff2SAl Viro if (unlikely(status <= 0)) { 14675a18fff2SAl Viro if (status < 0) { 14685a18fff2SAl Viro dput(dentry); 14695a18fff2SAl Viro return status; 14705a18fff2SAl Viro } 14715542aa2fSEric W. Biederman d_invalidate(dentry); 14725a18fff2SAl Viro dput(dentry); 147381e6f520SAl Viro goto need_lookup; 14745a18fff2SAl Viro } 1475697f514dSMiklos Szeredi 14761da177e4SLinus Torvalds path->mnt = mnt; 14771da177e4SLinus Torvalds path->dentry = dentry; 14789875cf80SDavid Howells err = follow_managed(path, nd->flags); 147989312214SIan Kent if (unlikely(err < 0)) { 148089312214SIan Kent path_put_conditional(path, nd); 14819875cf80SDavid Howells return err; 148289312214SIan Kent } 1483a3fbbde7SAl Viro if (err) 1484a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 148531e6b01fSNick Piggin *inode = path->dentry->d_inode; 14861da177e4SLinus Torvalds return 0; 148781e6f520SAl Viro 148881e6f520SAl Viro need_lookup: 1489697f514dSMiklos Szeredi return 1; 1490697f514dSMiklos Szeredi } 1491697f514dSMiklos Szeredi 1492697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1493cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1494697f514dSMiklos Szeredi { 1495697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1496697f514dSMiklos Szeredi int err; 1497697f514dSMiklos Szeredi 1498697f514dSMiklos Szeredi parent = nd->path.dentry; 149981e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 150081e6f520SAl Viro 150181e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1502cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 150381e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 150481e6f520SAl Viro if (IS_ERR(dentry)) 150581e6f520SAl Viro return PTR_ERR(dentry); 1506697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1507697f514dSMiklos Szeredi path->dentry = dentry; 1508697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1509697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1510697f514dSMiklos Szeredi path_put_conditional(path, nd); 1511697f514dSMiklos Szeredi return err; 1512697f514dSMiklos Szeredi } 1513697f514dSMiklos Szeredi if (err) 1514697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1515697f514dSMiklos Szeredi return 0; 15161da177e4SLinus Torvalds } 15171da177e4SLinus Torvalds 151852094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 151952094c8aSAl Viro { 152052094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 15214ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 152252094c8aSAl Viro if (err != -ECHILD) 152352094c8aSAl Viro return err; 152419660af7SAl Viro if (unlazy_walk(nd, NULL)) 152552094c8aSAl Viro return -ECHILD; 152652094c8aSAl Viro } 15274ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 152852094c8aSAl Viro } 152952094c8aSAl Viro 15309856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 15319856fa1bSAl Viro { 15329856fa1bSAl Viro if (type == LAST_DOTDOT) { 15339856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 15349856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 15359856fa1bSAl Viro return -ECHILD; 15369856fa1bSAl Viro } else 15379856fa1bSAl Viro follow_dotdot(nd); 15389856fa1bSAl Viro } 15399856fa1bSAl Viro return 0; 15409856fa1bSAl Viro } 15419856fa1bSAl Viro 1542951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1543951361f9SAl Viro { 1544951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1545951361f9SAl Viro path_put(&nd->path); 1546951361f9SAl Viro } else { 1547951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 15485b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1549951361f9SAl Viro nd->root.mnt = NULL; 15508b61e74fSAl Viro rcu_read_unlock(); 1551951361f9SAl Viro } 1552951361f9SAl Viro } 1553951361f9SAl Viro 15543ddcd056SLinus Torvalds /* 15553ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15563ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15573ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15583ddcd056SLinus Torvalds * for the common case. 15593ddcd056SLinus Torvalds */ 1560b18825a7SDavid Howells static inline int should_follow_link(struct dentry *dentry, int follow) 15613ddcd056SLinus Torvalds { 1562b18825a7SDavid Howells return unlikely(d_is_symlink(dentry)) ? follow : 0; 15633ddcd056SLinus Torvalds } 15643ddcd056SLinus Torvalds 1565ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 156621b9b073SAl Viro int follow) 1567ce57dfc1SAl Viro { 1568ce57dfc1SAl Viro struct inode *inode; 1569ce57dfc1SAl Viro int err; 1570ce57dfc1SAl Viro /* 1571ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1572ce57dfc1SAl Viro * to be able to know about the current root directory and 1573ce57dfc1SAl Viro * parent relationships. 1574ce57dfc1SAl Viro */ 157521b9b073SAl Viro if (unlikely(nd->last_type != LAST_NORM)) 157621b9b073SAl Viro return handle_dots(nd, nd->last_type); 1577e97cdc87SAl Viro err = lookup_fast(nd, path, &inode); 1578ce57dfc1SAl Viro if (unlikely(err)) { 1579697f514dSMiklos Szeredi if (err < 0) 1580697f514dSMiklos Szeredi goto out_err; 1581697f514dSMiklos Szeredi 1582cc2a5271SAl Viro err = lookup_slow(nd, path); 1583697f514dSMiklos Szeredi if (err < 0) 1584697f514dSMiklos Szeredi goto out_err; 1585697f514dSMiklos Szeredi 1586697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1587ce57dfc1SAl Viro } 1588697f514dSMiklos Szeredi err = -ENOENT; 158922213318SAl Viro if (!inode || d_is_negative(path->dentry)) 1590697f514dSMiklos Szeredi goto out_path_put; 1591697f514dSMiklos Szeredi 1592b18825a7SDavid Howells if (should_follow_link(path->dentry, follow)) { 159319660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 159419660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1595697f514dSMiklos Szeredi err = -ECHILD; 1596697f514dSMiklos Szeredi goto out_err; 159719660af7SAl Viro } 159819660af7SAl Viro } 1599ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1600ce57dfc1SAl Viro return 1; 1601ce57dfc1SAl Viro } 1602ce57dfc1SAl Viro path_to_nameidata(path, nd); 1603ce57dfc1SAl Viro nd->inode = inode; 1604ce57dfc1SAl Viro return 0; 1605697f514dSMiklos Szeredi 1606697f514dSMiklos Szeredi out_path_put: 1607697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1608697f514dSMiklos Szeredi out_err: 1609697f514dSMiklos Szeredi terminate_walk(nd); 1610697f514dSMiklos Szeredi return err; 1611ce57dfc1SAl Viro } 1612ce57dfc1SAl Viro 16131da177e4SLinus Torvalds /* 1614b356379aSAl Viro * This limits recursive symlink follows to 8, while 1615b356379aSAl Viro * limiting consecutive symlinks to 40. 1616b356379aSAl Viro * 1617b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1618b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1619b356379aSAl Viro */ 1620b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1621b356379aSAl Viro { 1622b356379aSAl Viro int res; 1623b356379aSAl Viro 1624b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1625b356379aSAl Viro path_put_conditional(path, nd); 1626b356379aSAl Viro path_put(&nd->path); 1627b356379aSAl Viro return -ELOOP; 1628b356379aSAl Viro } 16291a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1630b356379aSAl Viro 1631b356379aSAl Viro nd->depth++; 1632b356379aSAl Viro current->link_count++; 1633b356379aSAl Viro 1634b356379aSAl Viro do { 1635b356379aSAl Viro struct path link = *path; 1636b356379aSAl Viro void *cookie; 1637574197e0SAl Viro 1638574197e0SAl Viro res = follow_link(&link, nd, &cookie); 16396d7b5aaeSAl Viro if (res) 16406d7b5aaeSAl Viro break; 164121b9b073SAl Viro res = walk_component(nd, path, LOOKUP_FOLLOW); 1642574197e0SAl Viro put_link(nd, &link, cookie); 1643b356379aSAl Viro } while (res > 0); 1644b356379aSAl Viro 1645b356379aSAl Viro current->link_count--; 1646b356379aSAl Viro nd->depth--; 1647b356379aSAl Viro return res; 1648b356379aSAl Viro } 1649b356379aSAl Viro 1650b356379aSAl Viro /* 1651bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1652bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1653bfcfaa77SLinus Torvalds * 1654bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1655bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1656bfcfaa77SLinus Torvalds * fast. 1657bfcfaa77SLinus Torvalds * 1658bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1659bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1660bfcfaa77SLinus Torvalds * crossing operation. 1661bfcfaa77SLinus Torvalds * 1662bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1663bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1664bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1665bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1666bfcfaa77SLinus Torvalds */ 1667bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1668bfcfaa77SLinus Torvalds 1669f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1670bfcfaa77SLinus Torvalds 1671f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1672bfcfaa77SLinus Torvalds 1673bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1674bfcfaa77SLinus Torvalds { 167599d263d4SLinus Torvalds return hash_64(hash, 32); 1676bfcfaa77SLinus Torvalds } 1677bfcfaa77SLinus Torvalds 1678bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1679bfcfaa77SLinus Torvalds 1680bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1681bfcfaa77SLinus Torvalds 1682bfcfaa77SLinus Torvalds #endif 1683bfcfaa77SLinus Torvalds 1684bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1685bfcfaa77SLinus Torvalds { 1686bfcfaa77SLinus Torvalds unsigned long a, mask; 1687bfcfaa77SLinus Torvalds unsigned long hash = 0; 1688bfcfaa77SLinus Torvalds 1689bfcfaa77SLinus Torvalds for (;;) { 1690e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1691bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1692bfcfaa77SLinus Torvalds break; 1693bfcfaa77SLinus Torvalds hash += a; 1694f132c5beSAl Viro hash *= 9; 1695bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1696bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1697bfcfaa77SLinus Torvalds if (!len) 1698bfcfaa77SLinus Torvalds goto done; 1699bfcfaa77SLinus Torvalds } 1700a5c21dceSWill Deacon mask = bytemask_from_count(len); 1701bfcfaa77SLinus Torvalds hash += mask & a; 1702bfcfaa77SLinus Torvalds done: 1703bfcfaa77SLinus Torvalds return fold_hash(hash); 1704bfcfaa77SLinus Torvalds } 1705bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1706bfcfaa77SLinus Torvalds 1707bfcfaa77SLinus Torvalds /* 1708bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1709d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1710bfcfaa77SLinus Torvalds */ 1711d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1712bfcfaa77SLinus Torvalds { 171336126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 171436126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1715bfcfaa77SLinus Torvalds 1716bfcfaa77SLinus Torvalds hash = a = 0; 1717bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1718bfcfaa77SLinus Torvalds do { 1719bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1720bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1721e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 172236126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 172336126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1724bfcfaa77SLinus Torvalds 172536126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 172636126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 172736126f8fSLinus Torvalds 172836126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 172936126f8fSLinus Torvalds 173036126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 17319226b5b4SLinus Torvalds len += find_zero(mask); 1732d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1733bfcfaa77SLinus Torvalds } 1734bfcfaa77SLinus Torvalds 1735bfcfaa77SLinus Torvalds #else 1736bfcfaa77SLinus Torvalds 17370145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 17380145acc2SLinus Torvalds { 17390145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 17400145acc2SLinus Torvalds while (len--) 17410145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 17420145acc2SLinus Torvalds return end_name_hash(hash); 17430145acc2SLinus Torvalds } 1744ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 17450145acc2SLinus Torvalds 17463ddcd056SLinus Torvalds /* 1747200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1748200e9ef7SLinus Torvalds * one character. 1749200e9ef7SLinus Torvalds */ 1750d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1751200e9ef7SLinus Torvalds { 1752200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1753200e9ef7SLinus Torvalds unsigned long len = 0, c; 1754200e9ef7SLinus Torvalds 1755200e9ef7SLinus Torvalds c = (unsigned char)*name; 1756200e9ef7SLinus Torvalds do { 1757200e9ef7SLinus Torvalds len++; 1758200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1759200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1760200e9ef7SLinus Torvalds } while (c && c != '/'); 1761d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1762200e9ef7SLinus Torvalds } 1763200e9ef7SLinus Torvalds 1764bfcfaa77SLinus Torvalds #endif 1765bfcfaa77SLinus Torvalds 1766200e9ef7SLinus Torvalds /* 17671da177e4SLinus Torvalds * Name resolution. 1768ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1769ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17701da177e4SLinus Torvalds * 1771ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1772ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17731da177e4SLinus Torvalds */ 17746de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17751da177e4SLinus Torvalds { 17761da177e4SLinus Torvalds struct path next; 17771da177e4SLinus Torvalds int err; 17781da177e4SLinus Torvalds 17791da177e4SLinus Torvalds while (*name=='/') 17801da177e4SLinus Torvalds name++; 17811da177e4SLinus Torvalds if (!*name) 1782086e183aSAl Viro return 0; 17831da177e4SLinus Torvalds 17841da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17851da177e4SLinus Torvalds for(;;) { 1786d6bb3e90SLinus Torvalds u64 hash_len; 1787fe479a58SAl Viro int type; 17881da177e4SLinus Torvalds 178952094c8aSAl Viro err = may_lookup(nd); 17901da177e4SLinus Torvalds if (err) 17911da177e4SLinus Torvalds break; 17921da177e4SLinus Torvalds 1793d6bb3e90SLinus Torvalds hash_len = hash_name(name); 17941da177e4SLinus Torvalds 1795fe479a58SAl Viro type = LAST_NORM; 1796d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1797fe479a58SAl Viro case 2: 1798200e9ef7SLinus Torvalds if (name[1] == '.') { 1799fe479a58SAl Viro type = LAST_DOTDOT; 180016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 180116c2cd71SAl Viro } 1802fe479a58SAl Viro break; 1803fe479a58SAl Viro case 1: 1804fe479a58SAl Viro type = LAST_DOT; 1805fe479a58SAl Viro } 18065a202bcdSAl Viro if (likely(type == LAST_NORM)) { 18075a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 180816c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 18095a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1810a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1811da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 18125a202bcdSAl Viro if (err < 0) 18135a202bcdSAl Viro break; 1814d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1815d6bb3e90SLinus Torvalds name = this.name; 18165a202bcdSAl Viro } 18175a202bcdSAl Viro } 1818fe479a58SAl Viro 1819d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1820d6bb3e90SLinus Torvalds nd->last.name = name; 18215f4a6a69SAl Viro nd->last_type = type; 18225f4a6a69SAl Viro 1823d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1824d6bb3e90SLinus Torvalds if (!*name) 18255f4a6a69SAl Viro return 0; 1826200e9ef7SLinus Torvalds /* 1827200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1828200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1829200e9ef7SLinus Torvalds */ 1830200e9ef7SLinus Torvalds do { 1831d6bb3e90SLinus Torvalds name++; 1832d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 1833d6bb3e90SLinus Torvalds if (!*name) 18345f4a6a69SAl Viro return 0; 18355f4a6a69SAl Viro 183621b9b073SAl Viro err = walk_component(nd, &next, LOOKUP_FOLLOW); 1837ce57dfc1SAl Viro if (err < 0) 1838ce57dfc1SAl Viro return err; 1839fe479a58SAl Viro 1840ce57dfc1SAl Viro if (err) { 1841b356379aSAl Viro err = nested_symlink(&next, nd); 18421da177e4SLinus Torvalds if (err) 1843a7472babSAl Viro return err; 184431e6b01fSNick Piggin } 184544b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 18463ddcd056SLinus Torvalds err = -ENOTDIR; 18473ddcd056SLinus Torvalds break; 18485f4a6a69SAl Viro } 1849ce57dfc1SAl Viro } 1850951361f9SAl Viro terminate_walk(nd); 18511da177e4SLinus Torvalds return err; 18521da177e4SLinus Torvalds } 18531da177e4SLinus Torvalds 185470e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 18555e53084dSAl Viro struct nameidata *nd) 185631e6b01fSNick Piggin { 185731e6b01fSNick Piggin int retval = 0; 185831e6b01fSNick Piggin 185931e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1860980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 186131e6b01fSNick Piggin nd->depth = 0; 18625e53084dSAl Viro nd->base = NULL; 18635b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1864b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1865b18825a7SDavid Howells struct inode *inode = root->d_inode; 186673d049a4SAl Viro if (*name) { 186744b1d530SMiklos Szeredi if (!d_can_lookup(root)) 18685b6ca027SAl Viro return -ENOTDIR; 18695b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18705b6ca027SAl Viro if (retval) 18715b6ca027SAl Viro return retval; 187273d049a4SAl Viro } 18735b6ca027SAl Viro nd->path = nd->root; 18745b6ca027SAl Viro nd->inode = inode; 18755b6ca027SAl Viro if (flags & LOOKUP_RCU) { 18768b61e74fSAl Viro rcu_read_lock(); 18775b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 187848a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 18795b6ca027SAl Viro } else { 18805b6ca027SAl Viro path_get(&nd->path); 18815b6ca027SAl Viro } 1882d465887fSAl Viro goto done; 18835b6ca027SAl Viro } 18845b6ca027SAl Viro 188531e6b01fSNick Piggin nd->root.mnt = NULL; 188631e6b01fSNick Piggin 188748a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 188831e6b01fSNick Piggin if (*name=='/') { 1889e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18908b61e74fSAl Viro rcu_read_lock(); 18917bd88377SAl Viro nd->seq = set_root_rcu(nd); 1892e41f7d4eSAl Viro } else { 1893e41f7d4eSAl Viro set_root(nd); 1894e41f7d4eSAl Viro path_get(&nd->root); 1895e41f7d4eSAl Viro } 189631e6b01fSNick Piggin nd->path = nd->root; 189731e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1898e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 189931e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1900c28cc364SNick Piggin unsigned seq; 190131e6b01fSNick Piggin 19028b61e74fSAl Viro rcu_read_lock(); 190331e6b01fSNick Piggin 1904c28cc364SNick Piggin do { 1905c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 190631e6b01fSNick Piggin nd->path = fs->pwd; 1907c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1908c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1909e41f7d4eSAl Viro } else { 1910e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1911e41f7d4eSAl Viro } 191231e6b01fSNick Piggin } else { 1913582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 19142903ff01SAl Viro struct fd f = fdget_raw(dfd); 191531e6b01fSNick Piggin struct dentry *dentry; 191631e6b01fSNick Piggin 19172903ff01SAl Viro if (!f.file) 19182903ff01SAl Viro return -EBADF; 191931e6b01fSNick Piggin 19202903ff01SAl Viro dentry = f.file->f_path.dentry; 192131e6b01fSNick Piggin 1922f52e0c11SAl Viro if (*name) { 192344b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 19242903ff01SAl Viro fdput(f); 19252903ff01SAl Viro return -ENOTDIR; 1926f52e0c11SAl Viro } 19272903ff01SAl Viro } 19282903ff01SAl Viro 19292903ff01SAl Viro nd->path = f.file->f_path; 1930e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19319c225f26SLinus Torvalds if (f.flags & FDPUT_FPUT) 19325e53084dSAl Viro nd->base = f.file; 1933c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 19348b61e74fSAl Viro rcu_read_lock(); 19355590ff0dSUlrich Drepper } else { 19362903ff01SAl Viro path_get(&nd->path); 19372903ff01SAl Viro fdput(f); 19381da177e4SLinus Torvalds } 1939e41f7d4eSAl Viro } 1940e41f7d4eSAl Viro 194131e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 19424023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 1943d465887fSAl Viro goto done; 19444023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 1945d465887fSAl Viro goto done; 19464023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 19474023bfc9SAl Viro nd->root.mnt = NULL; 19484023bfc9SAl Viro rcu_read_unlock(); 19494023bfc9SAl Viro return -ECHILD; 1950d465887fSAl Viro done: 1951d465887fSAl Viro current->total_link_count = 0; 1952d465887fSAl Viro return link_path_walk(name, nd); 19539b4a9b14SAl Viro } 19549b4a9b14SAl Viro 1955893b7775SAl Viro static void path_cleanup(struct nameidata *nd) 1956893b7775SAl Viro { 1957893b7775SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 1958893b7775SAl Viro path_put(&nd->root); 1959893b7775SAl Viro nd->root.mnt = NULL; 1960893b7775SAl Viro } 1961893b7775SAl Viro if (unlikely(nd->base)) 1962893b7775SAl Viro fput(nd->base); 19639b4a9b14SAl Viro } 19649b4a9b14SAl Viro 1965bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1966bd92d7feSAl Viro { 1967bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1968bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1969bd92d7feSAl Viro 1970bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 197121b9b073SAl Viro return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); 1972bd92d7feSAl Viro } 1973bd92d7feSAl Viro 19749b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1975ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 19769b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 19779b4a9b14SAl Viro { 1978bd92d7feSAl Viro struct path path; 1979bd92d7feSAl Viro int err; 198031e6b01fSNick Piggin 198131e6b01fSNick Piggin /* 198231e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 198331e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 198431e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 198531e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 198631e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 198731e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 198831e6b01fSNick Piggin * analogue, foo_rcu(). 198931e6b01fSNick Piggin * 199031e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 199131e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 199231e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 199331e6b01fSNick Piggin * be able to complete). 199431e6b01fSNick Piggin */ 1995980f3ea2SAl Viro err = path_init(dfd, name, flags, nd); 1996bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1997bd92d7feSAl Viro err = lookup_last(nd, &path); 1998bd92d7feSAl Viro while (err > 0) { 1999bd92d7feSAl Viro void *cookie; 2000bd92d7feSAl Viro struct path link = path; 2001800179c9SKees Cook err = may_follow_link(&link, nd); 2002800179c9SKees Cook if (unlikely(err)) 2003800179c9SKees Cook break; 2004bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 2005574197e0SAl Viro err = follow_link(&link, nd, &cookie); 20066d7b5aaeSAl Viro if (err) 20076d7b5aaeSAl Viro break; 2008bd92d7feSAl Viro err = lookup_last(nd, &path); 2009574197e0SAl Viro put_link(nd, &link, cookie); 2010bd92d7feSAl Viro } 2011bd92d7feSAl Viro } 2012ee0827cdSAl Viro 20139f1fafeeSAl Viro if (!err) 20149f1fafeeSAl Viro err = complete_walk(nd); 2015bd92d7feSAl Viro 2016bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 201744b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 2018bd92d7feSAl Viro path_put(&nd->path); 2019bd23a539SAl Viro err = -ENOTDIR; 2020bd92d7feSAl Viro } 2021bd92d7feSAl Viro } 202216c2cd71SAl Viro 2023893b7775SAl Viro path_cleanup(nd); 2024bd92d7feSAl Viro return err; 202531e6b01fSNick Piggin } 202631e6b01fSNick Piggin 2027873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 2028873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 2029873f1eedSJeff Layton { 2030873f1eedSJeff Layton int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); 2031873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2032873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, flags, nd); 2033873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2034873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, 2035873f1eedSJeff Layton flags | LOOKUP_REVAL, nd); 2036873f1eedSJeff Layton 2037873f1eedSJeff Layton if (likely(!retval)) 2038adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2039873f1eedSJeff Layton return retval; 2040873f1eedSJeff Layton } 2041873f1eedSJeff Layton 2042ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 2043ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 2044ee0827cdSAl Viro { 204551689104SPaul Moore struct filename *filename = getname_kernel(name); 204651689104SPaul Moore int retval = PTR_ERR(filename); 2047ee0827cdSAl Viro 204851689104SPaul Moore if (!IS_ERR(filename)) { 204951689104SPaul Moore retval = filename_lookup(dfd, filename, flags, nd); 205051689104SPaul Moore putname(filename); 205151689104SPaul Moore } 205251689104SPaul Moore return retval; 20531da177e4SLinus Torvalds } 20541da177e4SLinus Torvalds 205579714f72SAl Viro /* does lookup, returns the object with parent locked */ 205679714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 20575590ff0dSUlrich Drepper { 205851689104SPaul Moore struct filename *filename = getname_kernel(name); 205979714f72SAl Viro struct nameidata nd; 206079714f72SAl Viro struct dentry *d; 206151689104SPaul Moore int err; 206251689104SPaul Moore 206351689104SPaul Moore if (IS_ERR(filename)) 206451689104SPaul Moore return ERR_CAST(filename); 206551689104SPaul Moore 206651689104SPaul Moore err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd); 206751689104SPaul Moore if (err) { 206851689104SPaul Moore d = ERR_PTR(err); 206951689104SPaul Moore goto out; 207051689104SPaul Moore } 207179714f72SAl Viro if (nd.last_type != LAST_NORM) { 207279714f72SAl Viro path_put(&nd.path); 207351689104SPaul Moore d = ERR_PTR(-EINVAL); 207451689104SPaul Moore goto out; 207579714f72SAl Viro } 207679714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 20771e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 207879714f72SAl Viro if (IS_ERR(d)) { 207979714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 208079714f72SAl Viro path_put(&nd.path); 208151689104SPaul Moore goto out; 208279714f72SAl Viro } 208379714f72SAl Viro *path = nd.path; 208451689104SPaul Moore out: 208551689104SPaul Moore putname(filename); 208679714f72SAl Viro return d; 20875590ff0dSUlrich Drepper } 20885590ff0dSUlrich Drepper 2089d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2090d1811465SAl Viro { 2091d1811465SAl Viro struct nameidata nd; 2092d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 2093d1811465SAl Viro if (!res) 2094d1811465SAl Viro *path = nd.path; 2095d1811465SAl Viro return res; 2096d1811465SAl Viro } 20974d359507SAl Viro EXPORT_SYMBOL(kern_path); 2098d1811465SAl Viro 209916f18200SJosef 'Jeff' Sipek /** 210016f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 210116f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 210216f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 210316f18200SJosef 'Jeff' Sipek * @name: pointer to file name 210416f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2105e0a01249SAl Viro * @path: pointer to struct path to fill 210616f18200SJosef 'Jeff' Sipek */ 210716f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 210816f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2109e0a01249SAl Viro struct path *path) 211016f18200SJosef 'Jeff' Sipek { 2111e0a01249SAl Viro struct nameidata nd; 2112e0a01249SAl Viro int err; 2113e0a01249SAl Viro nd.root.dentry = dentry; 2114e0a01249SAl Viro nd.root.mnt = mnt; 2115e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21165b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 2117e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 2118e0a01249SAl Viro if (!err) 2119e0a01249SAl Viro *path = nd.path; 2120e0a01249SAl Viro return err; 212116f18200SJosef 'Jeff' Sipek } 21224d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 212316f18200SJosef 'Jeff' Sipek 2124057f6c01SJames Morris /* 2125057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 2126057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 2127057f6c01SJames Morris * SMP-safe. 2128057f6c01SJames Morris */ 2129a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 21301da177e4SLinus Torvalds { 213172bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 21321da177e4SLinus Torvalds } 21331da177e4SLinus Torvalds 2134eead1911SChristoph Hellwig /** 2135a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2136eead1911SChristoph Hellwig * @name: pathname component to lookup 2137eead1911SChristoph Hellwig * @base: base directory to lookup from 2138eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2139eead1911SChristoph Hellwig * 2140a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 2141a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 2142eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 2143eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 2144eead1911SChristoph Hellwig */ 2145057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2146057f6c01SJames Morris { 2147057f6c01SJames Morris struct qstr this; 21486a96ba54SAl Viro unsigned int c; 2149cda309deSMiklos Szeredi int err; 2150057f6c01SJames Morris 21512f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 21522f9092e1SDavid Woodhouse 21536a96ba54SAl Viro this.name = name; 21546a96ba54SAl Viro this.len = len; 21550145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 21566a96ba54SAl Viro if (!len) 21576a96ba54SAl Viro return ERR_PTR(-EACCES); 21586a96ba54SAl Viro 215921d8a15aSAl Viro if (unlikely(name[0] == '.')) { 216021d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 216121d8a15aSAl Viro return ERR_PTR(-EACCES); 216221d8a15aSAl Viro } 216321d8a15aSAl Viro 21646a96ba54SAl Viro while (len--) { 21656a96ba54SAl Viro c = *(const unsigned char *)name++; 21666a96ba54SAl Viro if (c == '/' || c == '\0') 21676a96ba54SAl Viro return ERR_PTR(-EACCES); 21686a96ba54SAl Viro } 21695a202bcdSAl Viro /* 21705a202bcdSAl Viro * See if the low-level filesystem might want 21715a202bcdSAl Viro * to use its own hash.. 21725a202bcdSAl Viro */ 21735a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2174da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 21755a202bcdSAl Viro if (err < 0) 21765a202bcdSAl Viro return ERR_PTR(err); 21775a202bcdSAl Viro } 2178eead1911SChristoph Hellwig 2179cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2180cda309deSMiklos Szeredi if (err) 2181cda309deSMiklos Szeredi return ERR_PTR(err); 2182cda309deSMiklos Szeredi 218372bd866aSAl Viro return __lookup_hash(&this, base, 0); 2184057f6c01SJames Morris } 21854d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2186057f6c01SJames Morris 21871fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 21881fa1e7f6SAndy Whitcroft struct path *path, int *empty) 21891da177e4SLinus Torvalds { 21902d8f3038SAl Viro struct nameidata nd; 219191a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 21921da177e4SLinus Torvalds int err = PTR_ERR(tmp); 21931da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 21942d8f3038SAl Viro 21952d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21962d8f3038SAl Viro 2197873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 21981da177e4SLinus Torvalds putname(tmp); 21992d8f3038SAl Viro if (!err) 22002d8f3038SAl Viro *path = nd.path; 22011da177e4SLinus Torvalds } 22021da177e4SLinus Torvalds return err; 22031da177e4SLinus Torvalds } 22041da177e4SLinus Torvalds 22051fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 22061fa1e7f6SAndy Whitcroft struct path *path) 22071fa1e7f6SAndy Whitcroft { 2208f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 22091fa1e7f6SAndy Whitcroft } 22104d359507SAl Viro EXPORT_SYMBOL(user_path_at); 22111fa1e7f6SAndy Whitcroft 2212873f1eedSJeff Layton /* 2213873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2214873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2215873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2216873f1eedSJeff Layton * path-walking is complete. 2217873f1eedSJeff Layton */ 221891a27b2aSJeff Layton static struct filename * 22199e790bd6SJeff Layton user_path_parent(int dfd, const char __user *path, struct nameidata *nd, 22209e790bd6SJeff Layton unsigned int flags) 22212ad94ae6SAl Viro { 222291a27b2aSJeff Layton struct filename *s = getname(path); 22232ad94ae6SAl Viro int error; 22242ad94ae6SAl Viro 22259e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 22269e790bd6SJeff Layton flags &= LOOKUP_REVAL; 22279e790bd6SJeff Layton 22282ad94ae6SAl Viro if (IS_ERR(s)) 222991a27b2aSJeff Layton return s; 22302ad94ae6SAl Viro 22319e790bd6SJeff Layton error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); 223291a27b2aSJeff Layton if (error) { 22332ad94ae6SAl Viro putname(s); 223491a27b2aSJeff Layton return ERR_PTR(error); 223591a27b2aSJeff Layton } 22362ad94ae6SAl Viro 223791a27b2aSJeff Layton return s; 22382ad94ae6SAl Viro } 22392ad94ae6SAl Viro 22408033426eSJeff Layton /** 2241197df04cSAl Viro * mountpoint_last - look up last component for umount 22428033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 22438033426eSJeff Layton * @path: pointer to container for result 22448033426eSJeff Layton * 22458033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 22468033426eSJeff Layton * need to resolve the path without doing any revalidation. 22478033426eSJeff Layton * 22488033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 22498033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 22508033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 22518033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 22528033426eSJeff Layton * bogus and it doesn't exist. 22538033426eSJeff Layton * 22548033426eSJeff Layton * Returns: 22558033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 22568033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 22578033426eSJeff Layton * put in this case. 22588033426eSJeff Layton * 22598033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 22608033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 22618033426eSJeff Layton * The nd->path reference will also be put. 22628033426eSJeff Layton * 22638033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 22648033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 22658033426eSJeff Layton * to the link, and nd->path will *not* be put. 22668033426eSJeff Layton */ 22678033426eSJeff Layton static int 2268197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 22698033426eSJeff Layton { 22708033426eSJeff Layton int error = 0; 22718033426eSJeff Layton struct dentry *dentry; 22728033426eSJeff Layton struct dentry *dir = nd->path.dentry; 22738033426eSJeff Layton 227435759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 227535759521SAl Viro if (nd->flags & LOOKUP_RCU) { 227635759521SAl Viro if (unlazy_walk(nd, NULL)) { 22778033426eSJeff Layton error = -ECHILD; 227835759521SAl Viro goto out; 227935759521SAl Viro } 22808033426eSJeff Layton } 22818033426eSJeff Layton 22828033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 22838033426eSJeff Layton 22848033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 22858033426eSJeff Layton error = handle_dots(nd, nd->last_type); 228635759521SAl Viro if (error) 228735759521SAl Viro goto out; 22888033426eSJeff Layton dentry = dget(nd->path.dentry); 228935759521SAl Viro goto done; 22908033426eSJeff Layton } 22918033426eSJeff Layton 22928033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 22938033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 22948033426eSJeff Layton if (!dentry) { 22958033426eSJeff Layton /* 22968033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 22978033426eSJeff Layton * so that means that this dentry is probably a symlink or the 22988033426eSJeff Layton * path doesn't actually point to a mounted dentry. 22998033426eSJeff Layton */ 23008033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 23018033426eSJeff Layton if (!dentry) { 23028033426eSJeff Layton error = -ENOMEM; 2303bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 230435759521SAl Viro goto out; 23058033426eSJeff Layton } 230635759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 230735759521SAl Viro error = PTR_ERR(dentry); 2308bcceeebaSDave Jones if (IS_ERR(dentry)) { 2309bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 231035759521SAl Viro goto out; 23118033426eSJeff Layton } 2312bcceeebaSDave Jones } 23138033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 23148033426eSJeff Layton 231535759521SAl Viro done: 231622213318SAl Viro if (!dentry->d_inode || d_is_negative(dentry)) { 23178033426eSJeff Layton error = -ENOENT; 23188033426eSJeff Layton dput(dentry); 231935759521SAl Viro goto out; 232035759521SAl Viro } 23218033426eSJeff Layton path->dentry = dentry; 2322295dc39dSVasily Averin path->mnt = nd->path.mnt; 2323b18825a7SDavid Howells if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) 23248033426eSJeff Layton return 1; 2325295dc39dSVasily Averin mntget(path->mnt); 23268033426eSJeff Layton follow_mount(path); 232735759521SAl Viro error = 0; 232835759521SAl Viro out: 23298033426eSJeff Layton terminate_walk(nd); 23308033426eSJeff Layton return error; 23318033426eSJeff Layton } 23328033426eSJeff Layton 23338033426eSJeff Layton /** 2334197df04cSAl Viro * path_mountpoint - look up a path to be umounted 23358033426eSJeff Layton * @dfd: directory file descriptor to start walk from 23368033426eSJeff Layton * @name: full pathname to walk 2337606d6fe3SRandy Dunlap * @path: pointer to container for result 23388033426eSJeff Layton * @flags: lookup flags 23398033426eSJeff Layton * 23408033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2341606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 23428033426eSJeff Layton */ 23438033426eSJeff Layton static int 2344197df04cSAl Viro path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) 23458033426eSJeff Layton { 23468033426eSJeff Layton struct nameidata nd; 23478033426eSJeff Layton int err; 23488033426eSJeff Layton 2349980f3ea2SAl Viro err = path_init(dfd, name, flags, &nd); 23508033426eSJeff Layton if (unlikely(err)) 2351115cbfdcSAl Viro goto out; 23528033426eSJeff Layton 2353197df04cSAl Viro err = mountpoint_last(&nd, path); 23548033426eSJeff Layton while (err > 0) { 23558033426eSJeff Layton void *cookie; 23568033426eSJeff Layton struct path link = *path; 23578033426eSJeff Layton err = may_follow_link(&link, &nd); 23588033426eSJeff Layton if (unlikely(err)) 23598033426eSJeff Layton break; 23608033426eSJeff Layton nd.flags |= LOOKUP_PARENT; 23618033426eSJeff Layton err = follow_link(&link, &nd, &cookie); 23628033426eSJeff Layton if (err) 23638033426eSJeff Layton break; 2364197df04cSAl Viro err = mountpoint_last(&nd, path); 23658033426eSJeff Layton put_link(&nd, &link, cookie); 23668033426eSJeff Layton } 23678033426eSJeff Layton out: 2368893b7775SAl Viro path_cleanup(&nd); 23698033426eSJeff Layton return err; 23708033426eSJeff Layton } 23718033426eSJeff Layton 23722d864651SAl Viro static int 23732d864651SAl Viro filename_mountpoint(int dfd, struct filename *s, struct path *path, 23742d864651SAl Viro unsigned int flags) 23752d864651SAl Viro { 2376cbaab2dbSAl Viro int error; 2377cbaab2dbSAl Viro if (IS_ERR(s)) 2378cbaab2dbSAl Viro return PTR_ERR(s); 2379cbaab2dbSAl Viro error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); 23802d864651SAl Viro if (unlikely(error == -ECHILD)) 23812d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags); 23822d864651SAl Viro if (unlikely(error == -ESTALE)) 23832d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); 23842d864651SAl Viro if (likely(!error)) 23852d864651SAl Viro audit_inode(s, path->dentry, 0); 2386cbaab2dbSAl Viro putname(s); 23872d864651SAl Viro return error; 23882d864651SAl Viro } 23892d864651SAl Viro 23908033426eSJeff Layton /** 2391197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 23928033426eSJeff Layton * @dfd: directory file descriptor 23938033426eSJeff Layton * @name: pathname from userland 23948033426eSJeff Layton * @flags: lookup flags 23958033426eSJeff Layton * @path: pointer to container to hold result 23968033426eSJeff Layton * 23978033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 23988033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 23998033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 24008033426eSJeff Layton * and avoid revalidating the last component. 24018033426eSJeff Layton * 24028033426eSJeff Layton * Returns 0 and populates "path" on success. 24038033426eSJeff Layton */ 24048033426eSJeff Layton int 2405197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 24068033426eSJeff Layton struct path *path) 24078033426eSJeff Layton { 2408cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 24098033426eSJeff Layton } 24108033426eSJeff Layton 24112d864651SAl Viro int 24122d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 24132d864651SAl Viro unsigned int flags) 24142d864651SAl Viro { 2415cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 24162d864651SAl Viro } 24172d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 24182d864651SAl Viro 2419cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 24201da177e4SLinus Torvalds { 24218e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2422da9592edSDavid Howells 24238e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 24241da177e4SLinus Torvalds return 0; 24258e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 24261da177e4SLinus Torvalds return 0; 242723adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 24281da177e4SLinus Torvalds } 2429cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 24301da177e4SLinus Torvalds 24311da177e4SLinus Torvalds /* 24321da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 24331da177e4SLinus Torvalds * whether the type of victim is right. 24341da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 24351da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 24361da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 24371da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 24381da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 24391da177e4SLinus Torvalds * a. be owner of dir, or 24401da177e4SLinus Torvalds * b. be owner of victim, or 24411da177e4SLinus Torvalds * c. have CAP_FOWNER capability 24421da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 24431da177e4SLinus Torvalds * links pointing to it. 24441da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 24451da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 24461da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 24471da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 24481da177e4SLinus Torvalds * nfs_async_unlink(). 24491da177e4SLinus Torvalds */ 2450b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 24511da177e4SLinus Torvalds { 2452b18825a7SDavid Howells struct inode *inode = victim->d_inode; 24531da177e4SLinus Torvalds int error; 24541da177e4SLinus Torvalds 2455b18825a7SDavid Howells if (d_is_negative(victim)) 24561da177e4SLinus Torvalds return -ENOENT; 2457b18825a7SDavid Howells BUG_ON(!inode); 24581da177e4SLinus Torvalds 24591da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24604fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 24611da177e4SLinus Torvalds 2462f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 24631da177e4SLinus Torvalds if (error) 24641da177e4SLinus Torvalds return error; 24651da177e4SLinus Torvalds if (IS_APPEND(dir)) 24661da177e4SLinus Torvalds return -EPERM; 2467b18825a7SDavid Howells 2468b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2469b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 24701da177e4SLinus Torvalds return -EPERM; 24711da177e4SLinus Torvalds if (isdir) { 247244b1d530SMiklos Szeredi if (!d_is_dir(victim)) 24731da177e4SLinus Torvalds return -ENOTDIR; 24741da177e4SLinus Torvalds if (IS_ROOT(victim)) 24751da177e4SLinus Torvalds return -EBUSY; 247644b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 24771da177e4SLinus Torvalds return -EISDIR; 24781da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24791da177e4SLinus Torvalds return -ENOENT; 24801da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 24811da177e4SLinus Torvalds return -EBUSY; 24821da177e4SLinus Torvalds return 0; 24831da177e4SLinus Torvalds } 24841da177e4SLinus Torvalds 24851da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 24861da177e4SLinus Torvalds * dir. 24871da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 24881da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 24891da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 24901da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 24911da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 24921da177e4SLinus Torvalds */ 2493a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 24941da177e4SLinus Torvalds { 249514e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 24961da177e4SLinus Torvalds if (child->d_inode) 24971da177e4SLinus Torvalds return -EEXIST; 24981da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24991da177e4SLinus Torvalds return -ENOENT; 2500f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 25011da177e4SLinus Torvalds } 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds /* 25041da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 25051da177e4SLinus Torvalds */ 25061da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 25071da177e4SLinus Torvalds { 25081da177e4SLinus Torvalds struct dentry *p; 25091da177e4SLinus Torvalds 25101da177e4SLinus Torvalds if (p1 == p2) { 2511f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 25121da177e4SLinus Torvalds return NULL; 25131da177e4SLinus Torvalds } 25141da177e4SLinus Torvalds 2515a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25161da177e4SLinus Torvalds 2517e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2518e2761a11SOGAWA Hirofumi if (p) { 2519f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2520f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 25211da177e4SLinus Torvalds return p; 25221da177e4SLinus Torvalds } 25231da177e4SLinus Torvalds 2524e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2525e2761a11SOGAWA Hirofumi if (p) { 2526f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2527f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 25281da177e4SLinus Torvalds return p; 25291da177e4SLinus Torvalds } 25301da177e4SLinus Torvalds 2531f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2532d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 25331da177e4SLinus Torvalds return NULL; 25341da177e4SLinus Torvalds } 25354d359507SAl Viro EXPORT_SYMBOL(lock_rename); 25361da177e4SLinus Torvalds 25371da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 25381da177e4SLinus Torvalds { 25391b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 25401da177e4SLinus Torvalds if (p1 != p2) { 25411b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2542a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25431da177e4SLinus Torvalds } 25441da177e4SLinus Torvalds } 25454d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 25461da177e4SLinus Torvalds 25474acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2548312b63fbSAl Viro bool want_excl) 25491da177e4SLinus Torvalds { 2550a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25511da177e4SLinus Torvalds if (error) 25521da177e4SLinus Torvalds return error; 25531da177e4SLinus Torvalds 2554acfa4380SAl Viro if (!dir->i_op->create) 25551da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 25561da177e4SLinus Torvalds mode &= S_IALLUGO; 25571da177e4SLinus Torvalds mode |= S_IFREG; 25581da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 25591da177e4SLinus Torvalds if (error) 25601da177e4SLinus Torvalds return error; 2561312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2562a74574aaSStephen Smalley if (!error) 2563f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25641da177e4SLinus Torvalds return error; 25651da177e4SLinus Torvalds } 25664d359507SAl Viro EXPORT_SYMBOL(vfs_create); 25671da177e4SLinus Torvalds 256873d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 25691da177e4SLinus Torvalds { 25703fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 25711da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 25721da177e4SLinus Torvalds int error; 25731da177e4SLinus Torvalds 2574bcda7652SAl Viro /* O_PATH? */ 2575bcda7652SAl Viro if (!acc_mode) 2576bcda7652SAl Viro return 0; 2577bcda7652SAl Viro 25781da177e4SLinus Torvalds if (!inode) 25791da177e4SLinus Torvalds return -ENOENT; 25801da177e4SLinus Torvalds 2581c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2582c8fe8f30SChristoph Hellwig case S_IFLNK: 25831da177e4SLinus Torvalds return -ELOOP; 2584c8fe8f30SChristoph Hellwig case S_IFDIR: 2585c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 25861da177e4SLinus Torvalds return -EISDIR; 2587c8fe8f30SChristoph Hellwig break; 2588c8fe8f30SChristoph Hellwig case S_IFBLK: 2589c8fe8f30SChristoph Hellwig case S_IFCHR: 25903fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 25911da177e4SLinus Torvalds return -EACCES; 2592c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2593c8fe8f30SChristoph Hellwig case S_IFIFO: 2594c8fe8f30SChristoph Hellwig case S_IFSOCK: 25951da177e4SLinus Torvalds flag &= ~O_TRUNC; 2596c8fe8f30SChristoph Hellwig break; 25974a3fd211SDave Hansen } 2598b41572e9SDave Hansen 25993fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2600b41572e9SDave Hansen if (error) 2601b41572e9SDave Hansen return error; 26026146f0d5SMimi Zohar 26031da177e4SLinus Torvalds /* 26041da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 26051da177e4SLinus Torvalds */ 26061da177e4SLinus Torvalds if (IS_APPEND(inode)) { 26078737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 26087715b521SAl Viro return -EPERM; 26091da177e4SLinus Torvalds if (flag & O_TRUNC) 26107715b521SAl Viro return -EPERM; 26111da177e4SLinus Torvalds } 26121da177e4SLinus Torvalds 26131da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 26142e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 26157715b521SAl Viro return -EPERM; 26161da177e4SLinus Torvalds 2617f3c7691eSJ. Bruce Fields return 0; 26187715b521SAl Viro } 26197715b521SAl Viro 2620e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 26217715b521SAl Viro { 2622e1181ee6SJeff Layton struct path *path = &filp->f_path; 26237715b521SAl Viro struct inode *inode = path->dentry->d_inode; 26247715b521SAl Viro int error = get_write_access(inode); 26251da177e4SLinus Torvalds if (error) 26267715b521SAl Viro return error; 26271da177e4SLinus Torvalds /* 26281da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 26291da177e4SLinus Torvalds */ 2630d7a06983SJeff Layton error = locks_verify_locked(filp); 2631be6d3e56SKentaro Takeda if (!error) 2632ea0d3ab2STetsuo Handa error = security_path_truncate(path); 26331da177e4SLinus Torvalds if (!error) { 26347715b521SAl Viro error = do_truncate(path->dentry, 0, 2635d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2636e1181ee6SJeff Layton filp); 26371da177e4SLinus Torvalds } 26381da177e4SLinus Torvalds put_write_access(inode); 2639acd0c935SMimi Zohar return error; 26401da177e4SLinus Torvalds } 26411da177e4SLinus Torvalds 2642d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2643d57999e1SDave Hansen { 26448a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 26458a5e929dSAl Viro flag--; 2646d57999e1SDave Hansen return flag; 2647d57999e1SDave Hansen } 2648d57999e1SDave Hansen 2649d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2650d18e9008SMiklos Szeredi { 2651d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2652d18e9008SMiklos Szeredi if (error) 2653d18e9008SMiklos Szeredi return error; 2654d18e9008SMiklos Szeredi 2655d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2656d18e9008SMiklos Szeredi if (error) 2657d18e9008SMiklos Szeredi return error; 2658d18e9008SMiklos Szeredi 2659d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2660d18e9008SMiklos Szeredi } 2661d18e9008SMiklos Szeredi 26621acf0af9SDavid Howells /* 26631acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 26641acf0af9SDavid Howells * dentry. 26651acf0af9SDavid Howells * 26661acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 26671acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 26681acf0af9SDavid Howells * 26691acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 26701acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 26711acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 26721acf0af9SDavid Howells * 26731acf0af9SDavid Howells * Returns an error code otherwise. 26741acf0af9SDavid Howells */ 26752675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 267630d90494SAl Viro struct path *path, struct file *file, 2677015c3bbcSMiklos Szeredi const struct open_flags *op, 267864894cf8SAl Viro bool got_write, bool need_lookup, 267947237687SAl Viro int *opened) 2680d18e9008SMiklos Szeredi { 2681d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2682d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2683d18e9008SMiklos Szeredi umode_t mode; 2684d18e9008SMiklos Szeredi int error; 2685d18e9008SMiklos Szeredi int acc_mode; 2686d18e9008SMiklos Szeredi int create_error = 0; 2687d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2688116cc022SMiklos Szeredi bool excl; 2689d18e9008SMiklos Szeredi 2690d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2691d18e9008SMiklos Szeredi 2692d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2693d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 26942675a4ebSAl Viro error = -ENOENT; 2695d18e9008SMiklos Szeredi goto out; 2696d18e9008SMiklos Szeredi } 2697d18e9008SMiklos Szeredi 269862b259d8SMiklos Szeredi mode = op->mode; 2699d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2700d18e9008SMiklos Szeredi mode &= ~current_umask(); 2701d18e9008SMiklos Szeredi 2702116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2703116cc022SMiklos Szeredi if (excl) 2704d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2705d18e9008SMiklos Szeredi 2706d18e9008SMiklos Szeredi /* 2707d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2708d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2709d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2710d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2711d18e9008SMiklos Szeredi * 2712d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2713d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2714d18e9008SMiklos Szeredi */ 271564894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 271664894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 271764894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2718d18e9008SMiklos Szeredi /* 2719d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2720d18e9008SMiklos Szeredi * back to lookup + open 2721d18e9008SMiklos Szeredi */ 2722d18e9008SMiklos Szeredi goto no_open; 2723d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2724d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 272564894cf8SAl Viro create_error = -EROFS; 2726d18e9008SMiklos Szeredi goto no_open; 2727d18e9008SMiklos Szeredi } else { 2728d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 272964894cf8SAl Viro create_error = -EROFS; 2730d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2731d18e9008SMiklos Szeredi } 2732d18e9008SMiklos Szeredi } 2733d18e9008SMiklos Szeredi 2734d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 273538227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2736d18e9008SMiklos Szeredi if (error) { 2737d18e9008SMiklos Szeredi create_error = error; 2738d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2739d18e9008SMiklos Szeredi goto no_open; 2740d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2741d18e9008SMiklos Szeredi } 2742d18e9008SMiklos Szeredi } 2743d18e9008SMiklos Szeredi 2744d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2745d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2746d18e9008SMiklos Szeredi 274730d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 274830d90494SAl Viro file->f_path.mnt = nd->path.mnt; 274930d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 275047237687SAl Viro opened); 2751d9585277SAl Viro if (error < 0) { 2752d9585277SAl Viro if (create_error && error == -ENOENT) 2753d9585277SAl Viro error = create_error; 2754d18e9008SMiklos Szeredi goto out; 2755d18e9008SMiklos Szeredi } 2756d18e9008SMiklos Szeredi 2757d9585277SAl Viro if (error) { /* returned 1, that is */ 275830d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 27592675a4ebSAl Viro error = -EIO; 2760d18e9008SMiklos Szeredi goto out; 2761d18e9008SMiklos Szeredi } 276230d90494SAl Viro if (file->f_path.dentry) { 2763d18e9008SMiklos Szeredi dput(dentry); 276430d90494SAl Viro dentry = file->f_path.dentry; 2765d18e9008SMiklos Szeredi } 276603da633aSAl Viro if (*opened & FILE_CREATED) 276703da633aSAl Viro fsnotify_create(dir, dentry); 276803da633aSAl Viro if (!dentry->d_inode) { 276903da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 277003da633aSAl Viro if (create_error) { 277162b2ce96SSage Weil error = create_error; 277262b2ce96SSage Weil goto out; 277362b2ce96SSage Weil } 277403da633aSAl Viro } else { 277503da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 277603da633aSAl Viro error = -EEXIST; 277703da633aSAl Viro goto out; 277803da633aSAl Viro } 277903da633aSAl Viro } 2780d18e9008SMiklos Szeredi goto looked_up; 2781d18e9008SMiklos Szeredi } 2782d18e9008SMiklos Szeredi 2783d18e9008SMiklos Szeredi /* 2784d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2785d18e9008SMiklos Szeredi * here. 2786d18e9008SMiklos Szeredi */ 278703da633aSAl Viro acc_mode = op->acc_mode; 278803da633aSAl Viro if (*opened & FILE_CREATED) { 278903da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 279003da633aSAl Viro fsnotify_create(dir, dentry); 279103da633aSAl Viro acc_mode = MAY_OPEN; 279203da633aSAl Viro } 27932675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 27942675a4ebSAl Viro if (error) 27952675a4ebSAl Viro fput(file); 2796d18e9008SMiklos Szeredi 2797d18e9008SMiklos Szeredi out: 2798d18e9008SMiklos Szeredi dput(dentry); 27992675a4ebSAl Viro return error; 2800d18e9008SMiklos Szeredi 2801d18e9008SMiklos Szeredi no_open: 2802d18e9008SMiklos Szeredi if (need_lookup) { 280372bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2804d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 28052675a4ebSAl Viro return PTR_ERR(dentry); 2806d18e9008SMiklos Szeredi 2807d18e9008SMiklos Szeredi if (create_error) { 2808d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2809d18e9008SMiklos Szeredi 28102675a4ebSAl Viro error = create_error; 2811d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2812d18e9008SMiklos Szeredi if (!dentry->d_inode) 2813d18e9008SMiklos Szeredi goto out; 2814d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2815d18e9008SMiklos Szeredi goto out; 2816d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2817d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2818d18e9008SMiklos Szeredi goto out; 2819d18e9008SMiklos Szeredi } 2820d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2821d18e9008SMiklos Szeredi } 2822d18e9008SMiklos Szeredi } 2823d18e9008SMiklos Szeredi looked_up: 2824d18e9008SMiklos Szeredi path->dentry = dentry; 2825d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 28262675a4ebSAl Viro return 1; 2827d18e9008SMiklos Szeredi } 2828d18e9008SMiklos Szeredi 282931e6b01fSNick Piggin /* 28301acf0af9SDavid Howells * Look up and maybe create and open the last component. 2831d58ffd35SMiklos Szeredi * 2832d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2833d58ffd35SMiklos Szeredi * 28341acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 28351acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 28361acf0af9SDavid Howells * 28371acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 28381acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 28391acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 28401acf0af9SDavid Howells * specified then a negative dentry may be returned. 28411acf0af9SDavid Howells * 28421acf0af9SDavid Howells * An error code is returned otherwise. 28431acf0af9SDavid Howells * 28441acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 28451acf0af9SDavid Howells * cleared otherwise prior to returning. 2846d58ffd35SMiklos Szeredi */ 28472675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 284830d90494SAl Viro struct file *file, 2849d58ffd35SMiklos Szeredi const struct open_flags *op, 285064894cf8SAl Viro bool got_write, int *opened) 2851d58ffd35SMiklos Szeredi { 2852d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 285354ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2854d58ffd35SMiklos Szeredi struct dentry *dentry; 2855d58ffd35SMiklos Szeredi int error; 285654ef4872SMiklos Szeredi bool need_lookup; 2857d58ffd35SMiklos Szeredi 285847237687SAl Viro *opened &= ~FILE_CREATED; 2859201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2860d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 28612675a4ebSAl Viro return PTR_ERR(dentry); 2862d58ffd35SMiklos Szeredi 2863d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2864d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2865d18e9008SMiklos Szeredi goto out_no_open; 2866d18e9008SMiklos Szeredi 2867d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 286864894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 286947237687SAl Viro need_lookup, opened); 2870d18e9008SMiklos Szeredi } 2871d18e9008SMiklos Szeredi 287254ef4872SMiklos Szeredi if (need_lookup) { 287354ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 287454ef4872SMiklos Szeredi 287572bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 287654ef4872SMiklos Szeredi if (IS_ERR(dentry)) 28772675a4ebSAl Viro return PTR_ERR(dentry); 287854ef4872SMiklos Szeredi } 287954ef4872SMiklos Szeredi 2880d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2881d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2882d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2883d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2884d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2885d58ffd35SMiklos Szeredi /* 2886d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2887d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2888d58ffd35SMiklos Szeredi * the time when the file is created and when 2889d58ffd35SMiklos Szeredi * a permanent write count is taken through 2890015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2891d58ffd35SMiklos Szeredi */ 289264894cf8SAl Viro if (!got_write) { 289364894cf8SAl Viro error = -EROFS; 2894d58ffd35SMiklos Szeredi goto out_dput; 289564894cf8SAl Viro } 289647237687SAl Viro *opened |= FILE_CREATED; 2897d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2898d58ffd35SMiklos Szeredi if (error) 2899d58ffd35SMiklos Szeredi goto out_dput; 2900312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2901312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2902d58ffd35SMiklos Szeredi if (error) 2903d58ffd35SMiklos Szeredi goto out_dput; 2904d58ffd35SMiklos Szeredi } 2905d18e9008SMiklos Szeredi out_no_open: 2906d58ffd35SMiklos Szeredi path->dentry = dentry; 2907d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 29082675a4ebSAl Viro return 1; 2909d58ffd35SMiklos Szeredi 2910d58ffd35SMiklos Szeredi out_dput: 2911d58ffd35SMiklos Szeredi dput(dentry); 29122675a4ebSAl Viro return error; 2913d58ffd35SMiklos Szeredi } 2914d58ffd35SMiklos Szeredi 2915d58ffd35SMiklos Szeredi /* 2916fe2d35ffSAl Viro * Handle the last step of open() 291731e6b01fSNick Piggin */ 29182675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 291930d90494SAl Viro struct file *file, const struct open_flags *op, 2920669abf4eSJeff Layton int *opened, struct filename *name) 2921fb1cc555SAl Viro { 2922a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2923ca344a89SAl Viro int open_flag = op->open_flag; 292477d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 292564894cf8SAl Viro bool got_write = false; 2926bcda7652SAl Viro int acc_mode = op->acc_mode; 2927a1eb3315SMiklos Szeredi struct inode *inode; 292877d660a8SMiklos Szeredi bool symlink_ok = false; 292916b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 293016b1c1cdSMiklos Szeredi bool retried = false; 293116c2cd71SAl Viro int error; 2932fb1cc555SAl Viro 2933c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2934c3e380b0SAl Viro nd->flags |= op->intent; 2935c3e380b0SAl Viro 2936bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2937fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2938fe2d35ffSAl Viro if (error) 29392675a4ebSAl Viro return error; 2940e83db167SMiklos Szeredi goto finish_open; 29411f36f774SAl Viro } 2942a2c36b45SAl Viro 2943ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2944fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2945fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2946bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 294777d660a8SMiklos Szeredi symlink_ok = true; 2948fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2949e97cdc87SAl Viro error = lookup_fast(nd, path, &inode); 295071574865SMiklos Szeredi if (likely(!error)) 295171574865SMiklos Szeredi goto finish_lookup; 295271574865SMiklos Szeredi 2953ce57dfc1SAl Viro if (error < 0) 29542675a4ebSAl Viro goto out; 2955a1eb3315SMiklos Szeredi 295637d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2957b6183df7SMiklos Szeredi } else { 2958fe2d35ffSAl Viro /* create side of things */ 2959a3fbbde7SAl Viro /* 2960b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2961b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2962b6183df7SMiklos Szeredi * about to look up 2963a3fbbde7SAl Viro */ 29649f1fafeeSAl Viro error = complete_walk(nd); 29659f1fafeeSAl Viro if (error) 29662675a4ebSAl Viro return error; 2967fe2d35ffSAl Viro 296833e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 296916c2cd71SAl Viro error = -EISDIR; 29701f36f774SAl Viro /* trailing slashes? */ 297131e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 29722675a4ebSAl Viro goto out; 2973b6183df7SMiklos Szeredi } 29741f36f774SAl Viro 297516b1c1cdSMiklos Szeredi retry_lookup: 297664894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 297764894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 297864894cf8SAl Viro if (!error) 297964894cf8SAl Viro got_write = true; 298064894cf8SAl Viro /* 298164894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 298264894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 298364894cf8SAl Viro * dropping this one anyway. 298464894cf8SAl Viro */ 298564894cf8SAl Viro } 2986a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 298764894cf8SAl Viro error = lookup_open(nd, path, file, op, got_write, opened); 2988fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2989fb1cc555SAl Viro 29902675a4ebSAl Viro if (error <= 0) { 29912675a4ebSAl Viro if (error) 2992d58ffd35SMiklos Szeredi goto out; 29936c0d46c4SAl Viro 299447237687SAl Viro if ((*opened & FILE_CREATED) || 2995496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 299677d660a8SMiklos Szeredi will_truncate = false; 2997d18e9008SMiklos Szeredi 2998adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 2999d18e9008SMiklos Szeredi goto opened; 3000d18e9008SMiklos Szeredi } 3001d18e9008SMiklos Szeredi 300247237687SAl Viro if (*opened & FILE_CREATED) { 30039b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3004ca344a89SAl Viro open_flag &= ~O_TRUNC; 300577d660a8SMiklos Szeredi will_truncate = false; 3006bcda7652SAl Viro acc_mode = MAY_OPEN; 3007d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 3008e83db167SMiklos Szeredi goto finish_open_created; 3009fb1cc555SAl Viro } 3010fb1cc555SAl Viro 3011fb1cc555SAl Viro /* 30123134f37eSJeff Layton * create/update audit record if it already exists. 3013fb1cc555SAl Viro */ 3014b18825a7SDavid Howells if (d_is_positive(path->dentry)) 3015adb5c247SJeff Layton audit_inode(name, path->dentry, 0); 3016fb1cc555SAl Viro 3017d18e9008SMiklos Szeredi /* 3018d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3019d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3020d18e9008SMiklos Szeredi * necessary...) 3021d18e9008SMiklos Szeredi */ 302264894cf8SAl Viro if (got_write) { 3023d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 302464894cf8SAl Viro got_write = false; 3025d18e9008SMiklos Szeredi } 3026d18e9008SMiklos Szeredi 3027fb1cc555SAl Viro error = -EEXIST; 3028f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 3029fb1cc555SAl Viro goto exit_dput; 3030fb1cc555SAl Viro 30319875cf80SDavid Howells error = follow_managed(path, nd->flags); 30329875cf80SDavid Howells if (error < 0) 3033fb1cc555SAl Viro goto exit_dput; 3034fb1cc555SAl Viro 3035a3fbbde7SAl Viro if (error) 3036a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 3037a3fbbde7SAl Viro 3038decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 3039decf3400SMiklos Szeredi inode = path->dentry->d_inode; 30405f5daac1SMiklos Szeredi finish_lookup: 30415f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 3042fb1cc555SAl Viro error = -ENOENT; 304322213318SAl Viro if (!inode || d_is_negative(path->dentry)) { 304454c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 30452675a4ebSAl Viro goto out; 304654c33e7fSMiklos Szeredi } 30479e67f361SAl Viro 3048b18825a7SDavid Howells if (should_follow_link(path->dentry, !symlink_ok)) { 3049d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 3050d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 3051d45ea867SMiklos Szeredi error = -ECHILD; 30522675a4ebSAl Viro goto out; 3053d45ea867SMiklos Szeredi } 3054d45ea867SMiklos Szeredi } 3055d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 30562675a4ebSAl Viro return 1; 3057d45ea867SMiklos Szeredi } 3058fb1cc555SAl Viro 305916b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 3060fb1cc555SAl Viro path_to_nameidata(path, nd); 306116b1c1cdSMiklos Szeredi } else { 306216b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 306316b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 306416b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 306516b1c1cdSMiklos Szeredi 306616b1c1cdSMiklos Szeredi } 3067decf3400SMiklos Szeredi nd->inode = inode; 3068a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3069bc77daa7SAl Viro finish_open: 3070a3fbbde7SAl Viro error = complete_walk(nd); 307116b1c1cdSMiklos Szeredi if (error) { 307216b1c1cdSMiklos Szeredi path_put(&save_parent); 30732675a4ebSAl Viro return error; 307416b1c1cdSMiklos Szeredi } 3075bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3076fb1cc555SAl Viro error = -EISDIR; 307744b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 30782675a4ebSAl Viro goto out; 3079af2f5542SMiklos Szeredi error = -ENOTDIR; 308044b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 30812675a4ebSAl Viro goto out; 30826c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 308377d660a8SMiklos Szeredi will_truncate = false; 30846c0d46c4SAl Viro 30850f9d1a10SAl Viro if (will_truncate) { 30860f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 30870f9d1a10SAl Viro if (error) 30882675a4ebSAl Viro goto out; 308964894cf8SAl Viro got_write = true; 30900f9d1a10SAl Viro } 3091e83db167SMiklos Szeredi finish_open_created: 3092bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3093ca344a89SAl Viro if (error) 30942675a4ebSAl Viro goto out; 30954aa7c634SMiklos Szeredi 30964aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 30974aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 30984aa7c634SMiklos Szeredi if (!error) { 30994aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31004aa7c634SMiklos Szeredi } else { 310130d90494SAl Viro if (error == -EOPENSTALE) 3102f60dc3dbSMiklos Szeredi goto stale_open; 3103015c3bbcSMiklos Szeredi goto out; 3104f60dc3dbSMiklos Szeredi } 3105a8277b9bSMiklos Szeredi opened: 31062675a4ebSAl Viro error = open_check_o_direct(file); 3107015c3bbcSMiklos Szeredi if (error) 3108015c3bbcSMiklos Szeredi goto exit_fput; 31093034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3110aa4caadbSMiklos Szeredi if (error) 3111aa4caadbSMiklos Szeredi goto exit_fput; 3112aa4caadbSMiklos Szeredi 31130f9d1a10SAl Viro if (will_truncate) { 31142675a4ebSAl Viro error = handle_truncate(file); 3115aa4caadbSMiklos Szeredi if (error) 3116aa4caadbSMiklos Szeredi goto exit_fput; 31170f9d1a10SAl Viro } 3118ca344a89SAl Viro out: 311964894cf8SAl Viro if (got_write) 31200f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 312116b1c1cdSMiklos Szeredi path_put(&save_parent); 3122e276ae67SMiklos Szeredi terminate_walk(nd); 31232675a4ebSAl Viro return error; 3124fb1cc555SAl Viro 3125fb1cc555SAl Viro exit_dput: 3126fb1cc555SAl Viro path_put_conditional(path, nd); 3127ca344a89SAl Viro goto out; 3128015c3bbcSMiklos Szeredi exit_fput: 31292675a4ebSAl Viro fput(file); 31302675a4ebSAl Viro goto out; 3131015c3bbcSMiklos Szeredi 3132f60dc3dbSMiklos Szeredi stale_open: 3133f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3134f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3135f60dc3dbSMiklos Szeredi goto out; 3136f60dc3dbSMiklos Szeredi 3137f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3138f60dc3dbSMiklos Szeredi path_put(&nd->path); 3139f60dc3dbSMiklos Szeredi nd->path = save_parent; 3140f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3141f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3142f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 314364894cf8SAl Viro if (got_write) { 3144f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 314564894cf8SAl Viro got_write = false; 3146f60dc3dbSMiklos Szeredi } 3147f60dc3dbSMiklos Szeredi retried = true; 3148f60dc3dbSMiklos Szeredi goto retry_lookup; 3149fb1cc555SAl Viro } 3150fb1cc555SAl Viro 315160545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 315260545d0dSAl Viro struct nameidata *nd, int flags, 315360545d0dSAl Viro const struct open_flags *op, 315460545d0dSAl Viro struct file *file, int *opened) 315560545d0dSAl Viro { 315660545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 315760545d0dSAl Viro struct dentry *dentry, *child; 315860545d0dSAl Viro struct inode *dir; 315960545d0dSAl Viro int error = path_lookupat(dfd, pathname->name, 316060545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 316160545d0dSAl Viro if (unlikely(error)) 316260545d0dSAl Viro return error; 316360545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 316460545d0dSAl Viro if (unlikely(error)) 316560545d0dSAl Viro goto out; 316660545d0dSAl Viro /* we want directory to be writable */ 316760545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 316860545d0dSAl Viro if (error) 316960545d0dSAl Viro goto out2; 317060545d0dSAl Viro dentry = nd->path.dentry; 317160545d0dSAl Viro dir = dentry->d_inode; 317260545d0dSAl Viro if (!dir->i_op->tmpfile) { 317360545d0dSAl Viro error = -EOPNOTSUPP; 317460545d0dSAl Viro goto out2; 317560545d0dSAl Viro } 317660545d0dSAl Viro child = d_alloc(dentry, &name); 317760545d0dSAl Viro if (unlikely(!child)) { 317860545d0dSAl Viro error = -ENOMEM; 317960545d0dSAl Viro goto out2; 318060545d0dSAl Viro } 318160545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 318260545d0dSAl Viro nd->flags |= op->intent; 318360545d0dSAl Viro dput(nd->path.dentry); 318460545d0dSAl Viro nd->path.dentry = child; 318560545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 318660545d0dSAl Viro if (error) 318760545d0dSAl Viro goto out2; 318860545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 318969a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 319069a91c23SEric Rannaud error = may_open(&nd->path, MAY_OPEN, op->open_flag); 319160545d0dSAl Viro if (error) 319260545d0dSAl Viro goto out2; 319360545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 319460545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 319560545d0dSAl Viro if (error) 319660545d0dSAl Viro goto out2; 319760545d0dSAl Viro error = open_check_o_direct(file); 3198f4e0c30cSAl Viro if (error) { 319960545d0dSAl Viro fput(file); 3200f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3201f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3202f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3203f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3204f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3205f4e0c30cSAl Viro } 320660545d0dSAl Viro out2: 320760545d0dSAl Viro mnt_drop_write(nd->path.mnt); 320860545d0dSAl Viro out: 320960545d0dSAl Viro path_put(&nd->path); 321060545d0dSAl Viro return error; 321160545d0dSAl Viro } 321260545d0dSAl Viro 3213669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 321473d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 32151da177e4SLinus Torvalds { 321630d90494SAl Viro struct file *file; 32179850c056SAl Viro struct path path; 321847237687SAl Viro int opened = 0; 321913aab428SAl Viro int error; 322031e6b01fSNick Piggin 322130d90494SAl Viro file = get_empty_filp(); 32221afc99beSAl Viro if (IS_ERR(file)) 32231afc99beSAl Viro return file; 322431e6b01fSNick Piggin 322530d90494SAl Viro file->f_flags = op->open_flag; 322631e6b01fSNick Piggin 3227bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 322860545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 322960545d0dSAl Viro goto out; 323060545d0dSAl Viro } 323160545d0dSAl Viro 3232980f3ea2SAl Viro error = path_init(dfd, pathname->name, flags, nd); 323331e6b01fSNick Piggin if (unlikely(error)) 32342675a4ebSAl Viro goto out; 32351da177e4SLinus Torvalds 32362675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 32372675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 32387b9337aaSNick Piggin struct path link = path; 3239def4af30SAl Viro void *cookie; 3240574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 324173d049a4SAl Viro path_put_conditional(&path, nd); 324273d049a4SAl Viro path_put(&nd->path); 32432675a4ebSAl Viro error = -ELOOP; 324440b39136SAl Viro break; 324540b39136SAl Viro } 3246800179c9SKees Cook error = may_follow_link(&link, nd); 3247800179c9SKees Cook if (unlikely(error)) 3248800179c9SKees Cook break; 324973d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 325073d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3251574197e0SAl Viro error = follow_link(&link, nd, &cookie); 3252c3e380b0SAl Viro if (unlikely(error)) 32532675a4ebSAl Viro break; 32542675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 3255574197e0SAl Viro put_link(nd, &link, cookie); 3256806b681cSAl Viro } 325710fa8e62SAl Viro out: 3258893b7775SAl Viro path_cleanup(nd); 32592675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32602675a4ebSAl Viro BUG_ON(!error); 326130d90494SAl Viro put_filp(file); 3262015c3bbcSMiklos Szeredi } 32632675a4ebSAl Viro if (unlikely(error)) { 32642675a4ebSAl Viro if (error == -EOPENSTALE) { 32652675a4ebSAl Viro if (flags & LOOKUP_RCU) 32662675a4ebSAl Viro error = -ECHILD; 32672675a4ebSAl Viro else 32682675a4ebSAl Viro error = -ESTALE; 32692675a4ebSAl Viro } 32702675a4ebSAl Viro file = ERR_PTR(error); 32712675a4ebSAl Viro } 32722675a4ebSAl Viro return file; 3273de459215SKirill Korotaev } 32741da177e4SLinus Torvalds 3275669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3276f9652e10SAl Viro const struct open_flags *op) 327713aab428SAl Viro { 327873d049a4SAl Viro struct nameidata nd; 3279f9652e10SAl Viro int flags = op->lookup_flags; 328013aab428SAl Viro struct file *filp; 328113aab428SAl Viro 328273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 328313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 328473d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 328513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 328673d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 328713aab428SAl Viro return filp; 328813aab428SAl Viro } 328913aab428SAl Viro 329073d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3291f9652e10SAl Viro const char *name, const struct open_flags *op) 329273d049a4SAl Viro { 329373d049a4SAl Viro struct nameidata nd; 329473d049a4SAl Viro struct file *file; 329551689104SPaul Moore struct filename *filename; 3296f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 329773d049a4SAl Viro 329873d049a4SAl Viro nd.root.mnt = mnt; 329973d049a4SAl Viro nd.root.dentry = dentry; 330073d049a4SAl Viro 3301b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 330273d049a4SAl Viro return ERR_PTR(-ELOOP); 330373d049a4SAl Viro 330451689104SPaul Moore filename = getname_kernel(name); 330551689104SPaul Moore if (unlikely(IS_ERR(filename))) 330651689104SPaul Moore return ERR_CAST(filename); 330751689104SPaul Moore 330851689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU); 330973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 331051689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags); 331173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 331251689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL); 331351689104SPaul Moore putname(filename); 331473d049a4SAl Viro return file; 331573d049a4SAl Viro } 331673d049a4SAl Viro 3317fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 33181ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33191da177e4SLinus Torvalds { 3320c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3321ed75e95dSAl Viro struct nameidata nd; 3322c30dabfeSJan Kara int err2; 33231ac12b4bSJeff Layton int error; 33241ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 33251ac12b4bSJeff Layton 33261ac12b4bSJeff Layton /* 33271ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 33281ac12b4bSJeff Layton * other flags passed in are ignored! 33291ac12b4bSJeff Layton */ 33301ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 33311ac12b4bSJeff Layton 3332fa14a0b8SAl Viro error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd); 3333ed75e95dSAl Viro if (error) 3334ed75e95dSAl Viro return ERR_PTR(error); 33351da177e4SLinus Torvalds 3336c663e5d8SChristoph Hellwig /* 3337c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3338c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3339c663e5d8SChristoph Hellwig */ 3340ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3341ed75e95dSAl Viro goto out; 3342ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3343ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3344c663e5d8SChristoph Hellwig 3345c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3346c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3347c663e5d8SChristoph Hellwig /* 3348c663e5d8SChristoph Hellwig * Do the final lookup. 3349c663e5d8SChristoph Hellwig */ 3350ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3351ed75e95dSAl Viro dentry = lookup_hash(&nd); 33521da177e4SLinus Torvalds if (IS_ERR(dentry)) 3353a8104a9fSAl Viro goto unlock; 3354c663e5d8SChristoph Hellwig 3355a8104a9fSAl Viro error = -EEXIST; 3356b18825a7SDavid Howells if (d_is_positive(dentry)) 3357a8104a9fSAl Viro goto fail; 3358b18825a7SDavid Howells 3359c663e5d8SChristoph Hellwig /* 3360c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3361c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3362c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3363c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3364c663e5d8SChristoph Hellwig */ 3365ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3366a8104a9fSAl Viro error = -ENOENT; 3367ed75e95dSAl Viro goto fail; 3368e9baf6e5SAl Viro } 3369c30dabfeSJan Kara if (unlikely(err2)) { 3370c30dabfeSJan Kara error = err2; 3371a8104a9fSAl Viro goto fail; 3372c30dabfeSJan Kara } 3373ed75e95dSAl Viro *path = nd.path; 3374e9baf6e5SAl Viro return dentry; 33751da177e4SLinus Torvalds fail: 3376a8104a9fSAl Viro dput(dentry); 3377a8104a9fSAl Viro dentry = ERR_PTR(error); 3378a8104a9fSAl Viro unlock: 3379dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3380c30dabfeSJan Kara if (!err2) 3381c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3382ed75e95dSAl Viro out: 3383dae6ad8fSAl Viro path_put(&nd.path); 3384ed75e95dSAl Viro return dentry; 3385dae6ad8fSAl Viro } 3386fa14a0b8SAl Viro 3387fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3388fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3389fa14a0b8SAl Viro { 339051689104SPaul Moore struct filename *filename = getname_kernel(pathname); 339151689104SPaul Moore struct dentry *res; 339251689104SPaul Moore 339351689104SPaul Moore if (IS_ERR(filename)) 339451689104SPaul Moore return ERR_CAST(filename); 339551689104SPaul Moore res = filename_create(dfd, filename, path, lookup_flags); 339651689104SPaul Moore putname(filename); 339751689104SPaul Moore return res; 3398fa14a0b8SAl Viro } 3399dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3400dae6ad8fSAl Viro 3401921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3402921a1650SAl Viro { 3403921a1650SAl Viro dput(dentry); 3404921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3405a8104a9fSAl Viro mnt_drop_write(path->mnt); 3406921a1650SAl Viro path_put(path); 3407921a1650SAl Viro } 3408921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3409921a1650SAl Viro 34101ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 34111ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3412dae6ad8fSAl Viro { 341391a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3414dae6ad8fSAl Viro struct dentry *res; 3415dae6ad8fSAl Viro if (IS_ERR(tmp)) 3416dae6ad8fSAl Viro return ERR_CAST(tmp); 3417fa14a0b8SAl Viro res = filename_create(dfd, tmp, path, lookup_flags); 3418dae6ad8fSAl Viro putname(tmp); 3419dae6ad8fSAl Viro return res; 3420dae6ad8fSAl Viro } 3421dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3422dae6ad8fSAl Viro 34231a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34241da177e4SLinus Torvalds { 3425a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34261da177e4SLinus Torvalds 34271da177e4SLinus Torvalds if (error) 34281da177e4SLinus Torvalds return error; 34291da177e4SLinus Torvalds 3430975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34311da177e4SLinus Torvalds return -EPERM; 34321da177e4SLinus Torvalds 3433acfa4380SAl Viro if (!dir->i_op->mknod) 34341da177e4SLinus Torvalds return -EPERM; 34351da177e4SLinus Torvalds 343608ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 343708ce5f16SSerge E. Hallyn if (error) 343808ce5f16SSerge E. Hallyn return error; 343908ce5f16SSerge E. Hallyn 34401da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 34411da177e4SLinus Torvalds if (error) 34421da177e4SLinus Torvalds return error; 34431da177e4SLinus Torvalds 34441da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3445a74574aaSStephen Smalley if (!error) 3446f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34471da177e4SLinus Torvalds return error; 34481da177e4SLinus Torvalds } 34494d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 34501da177e4SLinus Torvalds 3451f69aac00SAl Viro static int may_mknod(umode_t mode) 3452463c3197SDave Hansen { 3453463c3197SDave Hansen switch (mode & S_IFMT) { 3454463c3197SDave Hansen case S_IFREG: 3455463c3197SDave Hansen case S_IFCHR: 3456463c3197SDave Hansen case S_IFBLK: 3457463c3197SDave Hansen case S_IFIFO: 3458463c3197SDave Hansen case S_IFSOCK: 3459463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3460463c3197SDave Hansen return 0; 3461463c3197SDave Hansen case S_IFDIR: 3462463c3197SDave Hansen return -EPERM; 3463463c3197SDave Hansen default: 3464463c3197SDave Hansen return -EINVAL; 3465463c3197SDave Hansen } 3466463c3197SDave Hansen } 3467463c3197SDave Hansen 34688208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 34692e4d0924SHeiko Carstens unsigned, dev) 34701da177e4SLinus Torvalds { 34711da177e4SLinus Torvalds struct dentry *dentry; 3472dae6ad8fSAl Viro struct path path; 3473dae6ad8fSAl Viro int error; 3474972567f1SJeff Layton unsigned int lookup_flags = 0; 34751da177e4SLinus Torvalds 34768e4bfca1SAl Viro error = may_mknod(mode); 34778e4bfca1SAl Viro if (error) 34788e4bfca1SAl Viro return error; 3479972567f1SJeff Layton retry: 3480972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3481dae6ad8fSAl Viro if (IS_ERR(dentry)) 3482dae6ad8fSAl Viro return PTR_ERR(dentry); 34832ad94ae6SAl Viro 3484dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3485ce3b0f8dSAl Viro mode &= ~current_umask(); 3486dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3487be6d3e56SKentaro Takeda if (error) 3488a8104a9fSAl Viro goto out; 34891da177e4SLinus Torvalds switch (mode & S_IFMT) { 34901da177e4SLinus Torvalds case 0: case S_IFREG: 3491312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 34921da177e4SLinus Torvalds break; 34931da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3494dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 34951da177e4SLinus Torvalds new_decode_dev(dev)); 34961da177e4SLinus Torvalds break; 34971da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3498dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 34991da177e4SLinus Torvalds break; 35001da177e4SLinus Torvalds } 3501a8104a9fSAl Viro out: 3502921a1650SAl Viro done_path_create(&path, dentry); 3503972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3504972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3505972567f1SJeff Layton goto retry; 3506972567f1SJeff Layton } 35071da177e4SLinus Torvalds return error; 35081da177e4SLinus Torvalds } 35091da177e4SLinus Torvalds 35108208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35115590ff0dSUlrich Drepper { 35125590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35135590ff0dSUlrich Drepper } 35145590ff0dSUlrich Drepper 351518bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35161da177e4SLinus Torvalds { 3517a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35188de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35191da177e4SLinus Torvalds 35201da177e4SLinus Torvalds if (error) 35211da177e4SLinus Torvalds return error; 35221da177e4SLinus Torvalds 3523acfa4380SAl Viro if (!dir->i_op->mkdir) 35241da177e4SLinus Torvalds return -EPERM; 35251da177e4SLinus Torvalds 35261da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35271da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35281da177e4SLinus Torvalds if (error) 35291da177e4SLinus Torvalds return error; 35301da177e4SLinus Torvalds 35318de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 35328de52778SAl Viro return -EMLINK; 35338de52778SAl Viro 35341da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3535a74574aaSStephen Smalley if (!error) 3536f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 35371da177e4SLinus Torvalds return error; 35381da177e4SLinus Torvalds } 35394d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 35401da177e4SLinus Torvalds 3541a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 35421da177e4SLinus Torvalds { 35436902d925SDave Hansen struct dentry *dentry; 3544dae6ad8fSAl Viro struct path path; 3545dae6ad8fSAl Viro int error; 3546b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 35471da177e4SLinus Torvalds 3548b76d8b82SJeff Layton retry: 3549b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 35506902d925SDave Hansen if (IS_ERR(dentry)) 3551dae6ad8fSAl Viro return PTR_ERR(dentry); 35526902d925SDave Hansen 3553dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3554ce3b0f8dSAl Viro mode &= ~current_umask(); 3555dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3556a8104a9fSAl Viro if (!error) 3557dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3558921a1650SAl Viro done_path_create(&path, dentry); 3559b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3560b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3561b76d8b82SJeff Layton goto retry; 3562b76d8b82SJeff Layton } 35631da177e4SLinus Torvalds return error; 35641da177e4SLinus Torvalds } 35651da177e4SLinus Torvalds 3566a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 35675590ff0dSUlrich Drepper { 35685590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 35695590ff0dSUlrich Drepper } 35705590ff0dSUlrich Drepper 35711da177e4SLinus Torvalds /* 3572a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3573c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3574a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3575a71905f0SSage Weil * then we drop the dentry now. 35761da177e4SLinus Torvalds * 35771da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 35781da177e4SLinus Torvalds * do a 35791da177e4SLinus Torvalds * 35801da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 35811da177e4SLinus Torvalds * return -EBUSY; 35821da177e4SLinus Torvalds * 35831da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 35841da177e4SLinus Torvalds * that is still in use by something else.. 35851da177e4SLinus Torvalds */ 35861da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 35871da177e4SLinus Torvalds { 35881da177e4SLinus Torvalds shrink_dcache_parent(dentry); 35891da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 359098474236SWaiman Long if (dentry->d_lockref.count == 1) 35911da177e4SLinus Torvalds __d_drop(dentry); 35921da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 35931da177e4SLinus Torvalds } 35944d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 35951da177e4SLinus Torvalds 35961da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 35971da177e4SLinus Torvalds { 35981da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 35991da177e4SLinus Torvalds 36001da177e4SLinus Torvalds if (error) 36011da177e4SLinus Torvalds return error; 36021da177e4SLinus Torvalds 3603acfa4380SAl Viro if (!dir->i_op->rmdir) 36041da177e4SLinus Torvalds return -EPERM; 36051da177e4SLinus Torvalds 36061d2ef590SAl Viro dget(dentry); 36071b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3608912dbc15SSage Weil 36091da177e4SLinus Torvalds error = -EBUSY; 36107af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3611912dbc15SSage Weil goto out; 3612912dbc15SSage Weil 36131da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3614912dbc15SSage Weil if (error) 3615912dbc15SSage Weil goto out; 3616912dbc15SSage Weil 36173cebde24SSage Weil shrink_dcache_parent(dentry); 36181da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3619912dbc15SSage Weil if (error) 3620912dbc15SSage Weil goto out; 3621912dbc15SSage Weil 36221da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3623d83c49f3SAl Viro dont_mount(dentry); 36248ed936b5SEric W. Biederman detach_mounts(dentry); 36251da177e4SLinus Torvalds 3626912dbc15SSage Weil out: 3627912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36281d2ef590SAl Viro dput(dentry); 3629912dbc15SSage Weil if (!error) 3630912dbc15SSage Weil d_delete(dentry); 36311da177e4SLinus Torvalds return error; 36321da177e4SLinus Torvalds } 36334d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 36341da177e4SLinus Torvalds 36355590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 36361da177e4SLinus Torvalds { 36371da177e4SLinus Torvalds int error = 0; 363891a27b2aSJeff Layton struct filename *name; 36391da177e4SLinus Torvalds struct dentry *dentry; 36401da177e4SLinus Torvalds struct nameidata nd; 3641c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3642c6ee9206SJeff Layton retry: 3643c6ee9206SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 364491a27b2aSJeff Layton if (IS_ERR(name)) 364591a27b2aSJeff Layton return PTR_ERR(name); 36461da177e4SLinus Torvalds 36471da177e4SLinus Torvalds switch(nd.last_type) { 36481da177e4SLinus Torvalds case LAST_DOTDOT: 36491da177e4SLinus Torvalds error = -ENOTEMPTY; 36501da177e4SLinus Torvalds goto exit1; 36511da177e4SLinus Torvalds case LAST_DOT: 36521da177e4SLinus Torvalds error = -EINVAL; 36531da177e4SLinus Torvalds goto exit1; 36541da177e4SLinus Torvalds case LAST_ROOT: 36551da177e4SLinus Torvalds error = -EBUSY; 36561da177e4SLinus Torvalds goto exit1; 36571da177e4SLinus Torvalds } 36580612d9fbSOGAWA Hirofumi 36590612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3660c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3661c30dabfeSJan Kara if (error) 3662c30dabfeSJan Kara goto exit1; 36630612d9fbSOGAWA Hirofumi 36644ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 366549705b77SChristoph Hellwig dentry = lookup_hash(&nd); 36661da177e4SLinus Torvalds error = PTR_ERR(dentry); 36676902d925SDave Hansen if (IS_ERR(dentry)) 36686902d925SDave Hansen goto exit2; 3669e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3670e6bc45d6STheodore Ts'o error = -ENOENT; 3671e6bc45d6STheodore Ts'o goto exit3; 3672e6bc45d6STheodore Ts'o } 3673be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3674be6d3e56SKentaro Takeda if (error) 3675c30dabfeSJan Kara goto exit3; 36764ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 36770622753bSDave Hansen exit3: 36781da177e4SLinus Torvalds dput(dentry); 36796902d925SDave Hansen exit2: 36804ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3681c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 36821da177e4SLinus Torvalds exit1: 36831d957f9bSJan Blunck path_put(&nd.path); 36841da177e4SLinus Torvalds putname(name); 3685c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3686c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3687c6ee9206SJeff Layton goto retry; 3688c6ee9206SJeff Layton } 36891da177e4SLinus Torvalds return error; 36901da177e4SLinus Torvalds } 36911da177e4SLinus Torvalds 36923cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 36935590ff0dSUlrich Drepper { 36945590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 36955590ff0dSUlrich Drepper } 36965590ff0dSUlrich Drepper 3697b21996e3SJ. Bruce Fields /** 3698b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3699b21996e3SJ. Bruce Fields * @dir: parent directory 3700b21996e3SJ. Bruce Fields * @dentry: victim 3701b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3702b21996e3SJ. Bruce Fields * 3703b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3704b21996e3SJ. Bruce Fields * 3705b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3706b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3707b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3708b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3709b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3710b21996e3SJ. Bruce Fields * 3711b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3712b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3713b21996e3SJ. Bruce Fields * to be NFS exported. 3714b21996e3SJ. Bruce Fields */ 3715b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37161da177e4SLinus Torvalds { 37179accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37181da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37191da177e4SLinus Torvalds 37201da177e4SLinus Torvalds if (error) 37211da177e4SLinus Torvalds return error; 37221da177e4SLinus Torvalds 3723acfa4380SAl Viro if (!dir->i_op->unlink) 37241da177e4SLinus Torvalds return -EPERM; 37251da177e4SLinus Torvalds 37269accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37278ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37281da177e4SLinus Torvalds error = -EBUSY; 37291da177e4SLinus Torvalds else { 37301da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3731bec1052eSAl Viro if (!error) { 37325a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 37335a14696cSJ. Bruce Fields if (error) 3734b21996e3SJ. Bruce Fields goto out; 37351da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 37368ed936b5SEric W. Biederman if (!error) { 3737d83c49f3SAl Viro dont_mount(dentry); 37388ed936b5SEric W. Biederman detach_mounts(dentry); 37398ed936b5SEric W. Biederman } 3740bec1052eSAl Viro } 37411da177e4SLinus Torvalds } 3742b21996e3SJ. Bruce Fields out: 37439accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 37441da177e4SLinus Torvalds 37451da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 37461da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 37479accbb97SJ. Bruce Fields fsnotify_link_count(target); 37481da177e4SLinus Torvalds d_delete(dentry); 37491da177e4SLinus Torvalds } 37500eeca283SRobert Love 37511da177e4SLinus Torvalds return error; 37521da177e4SLinus Torvalds } 37534d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 37541da177e4SLinus Torvalds 37551da177e4SLinus Torvalds /* 37561da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 37571b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 37581da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 37591da177e4SLinus Torvalds * while waiting on the I/O. 37601da177e4SLinus Torvalds */ 37615590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 37621da177e4SLinus Torvalds { 37632ad94ae6SAl Viro int error; 376491a27b2aSJeff Layton struct filename *name; 37651da177e4SLinus Torvalds struct dentry *dentry; 37661da177e4SLinus Torvalds struct nameidata nd; 37671da177e4SLinus Torvalds struct inode *inode = NULL; 3768b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 37695d18f813SJeff Layton unsigned int lookup_flags = 0; 37705d18f813SJeff Layton retry: 37715d18f813SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 377291a27b2aSJeff Layton if (IS_ERR(name)) 377391a27b2aSJeff Layton return PTR_ERR(name); 37742ad94ae6SAl Viro 37751da177e4SLinus Torvalds error = -EISDIR; 37761da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 37771da177e4SLinus Torvalds goto exit1; 37780612d9fbSOGAWA Hirofumi 37790612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3780c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3781c30dabfeSJan Kara if (error) 3782c30dabfeSJan Kara goto exit1; 3783b21996e3SJ. Bruce Fields retry_deleg: 37844ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 378549705b77SChristoph Hellwig dentry = lookup_hash(&nd); 37861da177e4SLinus Torvalds error = PTR_ERR(dentry); 37871da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 37881da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 378950338b88STörök Edwin if (nd.last.name[nd.last.len]) 379050338b88STörök Edwin goto slashes; 37911da177e4SLinus Torvalds inode = dentry->d_inode; 3792b18825a7SDavid Howells if (d_is_negative(dentry)) 3793e6bc45d6STheodore Ts'o goto slashes; 37947de9c6eeSAl Viro ihold(inode); 3795be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3796be6d3e56SKentaro Takeda if (error) 3797c30dabfeSJan Kara goto exit2; 3798b21996e3SJ. Bruce Fields error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); 37991da177e4SLinus Torvalds exit2: 38001da177e4SLinus Torvalds dput(dentry); 38011da177e4SLinus Torvalds } 38024ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 38031da177e4SLinus Torvalds if (inode) 38041da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3805b21996e3SJ. Bruce Fields inode = NULL; 3806b21996e3SJ. Bruce Fields if (delegated_inode) { 38075a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3808b21996e3SJ. Bruce Fields if (!error) 3809b21996e3SJ. Bruce Fields goto retry_deleg; 3810b21996e3SJ. Bruce Fields } 3811c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 38121da177e4SLinus Torvalds exit1: 38131d957f9bSJan Blunck path_put(&nd.path); 38141da177e4SLinus Torvalds putname(name); 38155d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38165d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38175d18f813SJeff Layton inode = NULL; 38185d18f813SJeff Layton goto retry; 38195d18f813SJeff Layton } 38201da177e4SLinus Torvalds return error; 38211da177e4SLinus Torvalds 38221da177e4SLinus Torvalds slashes: 3823b18825a7SDavid Howells if (d_is_negative(dentry)) 3824b18825a7SDavid Howells error = -ENOENT; 382544b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3826b18825a7SDavid Howells error = -EISDIR; 3827b18825a7SDavid Howells else 3828b18825a7SDavid Howells error = -ENOTDIR; 38291da177e4SLinus Torvalds goto exit2; 38301da177e4SLinus Torvalds } 38311da177e4SLinus Torvalds 38322e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 38335590ff0dSUlrich Drepper { 38345590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 38355590ff0dSUlrich Drepper return -EINVAL; 38365590ff0dSUlrich Drepper 38375590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 38385590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 38395590ff0dSUlrich Drepper 38405590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 38415590ff0dSUlrich Drepper } 38425590ff0dSUlrich Drepper 38433480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 38445590ff0dSUlrich Drepper { 38455590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 38465590ff0dSUlrich Drepper } 38475590ff0dSUlrich Drepper 3848db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 38491da177e4SLinus Torvalds { 3850a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38511da177e4SLinus Torvalds 38521da177e4SLinus Torvalds if (error) 38531da177e4SLinus Torvalds return error; 38541da177e4SLinus Torvalds 3855acfa4380SAl Viro if (!dir->i_op->symlink) 38561da177e4SLinus Torvalds return -EPERM; 38571da177e4SLinus Torvalds 38581da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 38591da177e4SLinus Torvalds if (error) 38601da177e4SLinus Torvalds return error; 38611da177e4SLinus Torvalds 38621da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3863a74574aaSStephen Smalley if (!error) 3864f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38651da177e4SLinus Torvalds return error; 38661da177e4SLinus Torvalds } 38674d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 38681da177e4SLinus Torvalds 38692e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 38702e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 38711da177e4SLinus Torvalds { 38722ad94ae6SAl Viro int error; 387391a27b2aSJeff Layton struct filename *from; 38746902d925SDave Hansen struct dentry *dentry; 3875dae6ad8fSAl Viro struct path path; 3876f46d3567SJeff Layton unsigned int lookup_flags = 0; 38771da177e4SLinus Torvalds 38781da177e4SLinus Torvalds from = getname(oldname); 38791da177e4SLinus Torvalds if (IS_ERR(from)) 38801da177e4SLinus Torvalds return PTR_ERR(from); 3881f46d3567SJeff Layton retry: 3882f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 38831da177e4SLinus Torvalds error = PTR_ERR(dentry); 38846902d925SDave Hansen if (IS_ERR(dentry)) 3885dae6ad8fSAl Viro goto out_putname; 38866902d925SDave Hansen 388791a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3888a8104a9fSAl Viro if (!error) 388991a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3890921a1650SAl Viro done_path_create(&path, dentry); 3891f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3892f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3893f46d3567SJeff Layton goto retry; 3894f46d3567SJeff Layton } 38956902d925SDave Hansen out_putname: 38961da177e4SLinus Torvalds putname(from); 38971da177e4SLinus Torvalds return error; 38981da177e4SLinus Torvalds } 38991da177e4SLinus Torvalds 39003480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39015590ff0dSUlrich Drepper { 39025590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39035590ff0dSUlrich Drepper } 39045590ff0dSUlrich Drepper 3905146a8595SJ. Bruce Fields /** 3906146a8595SJ. Bruce Fields * vfs_link - create a new link 3907146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3908146a8595SJ. Bruce Fields * @dir: new parent 3909146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3910146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3911146a8595SJ. Bruce Fields * 3912146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3913146a8595SJ. Bruce Fields * 3914146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3915146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3916146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3917146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3918146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3919146a8595SJ. Bruce Fields * 3920146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3921146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3922146a8595SJ. Bruce Fields * to be NFS exported. 3923146a8595SJ. Bruce Fields */ 3924146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39251da177e4SLinus Torvalds { 39261da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39278de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39281da177e4SLinus Torvalds int error; 39291da177e4SLinus Torvalds 39301da177e4SLinus Torvalds if (!inode) 39311da177e4SLinus Torvalds return -ENOENT; 39321da177e4SLinus Torvalds 3933a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 39341da177e4SLinus Torvalds if (error) 39351da177e4SLinus Torvalds return error; 39361da177e4SLinus Torvalds 39371da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 39381da177e4SLinus Torvalds return -EXDEV; 39391da177e4SLinus Torvalds 39401da177e4SLinus Torvalds /* 39411da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 39421da177e4SLinus Torvalds */ 39431da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 39441da177e4SLinus Torvalds return -EPERM; 3945acfa4380SAl Viro if (!dir->i_op->link) 39461da177e4SLinus Torvalds return -EPERM; 39477e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 39481da177e4SLinus Torvalds return -EPERM; 39491da177e4SLinus Torvalds 39501da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 39511da177e4SLinus Torvalds if (error) 39521da177e4SLinus Torvalds return error; 39531da177e4SLinus Torvalds 39547e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3955aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3956f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3957aae8a97dSAneesh Kumar K.V error = -ENOENT; 39588de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 39598de52778SAl Viro error = -EMLINK; 3960146a8595SJ. Bruce Fields else { 3961146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 3962146a8595SJ. Bruce Fields if (!error) 39631da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 3964146a8595SJ. Bruce Fields } 3965f4e0c30cSAl Viro 3966f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 3967f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3968f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 3969f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3970f4e0c30cSAl Viro } 39717e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3972e31e14ecSStephen Smalley if (!error) 39737e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 39741da177e4SLinus Torvalds return error; 39751da177e4SLinus Torvalds } 39764d359507SAl Viro EXPORT_SYMBOL(vfs_link); 39771da177e4SLinus Torvalds 39781da177e4SLinus Torvalds /* 39791da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 39801da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 39811da177e4SLinus Torvalds * newname. --KAB 39821da177e4SLinus Torvalds * 39831da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 39841da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 39851da177e4SLinus Torvalds * and other special files. --ADM 39861da177e4SLinus Torvalds */ 39872e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 39882e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 39891da177e4SLinus Torvalds { 39901da177e4SLinus Torvalds struct dentry *new_dentry; 3991dae6ad8fSAl Viro struct path old_path, new_path; 3992146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 399311a7b371SAneesh Kumar K.V int how = 0; 39941da177e4SLinus Torvalds int error; 39951da177e4SLinus Torvalds 399611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3997c04030e1SUlrich Drepper return -EINVAL; 399811a7b371SAneesh Kumar K.V /* 3999f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4000f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4001f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 400211a7b371SAneesh Kumar K.V */ 4003f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4004f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4005f0cc6ffbSLinus Torvalds return -ENOENT; 400611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4007f0cc6ffbSLinus Torvalds } 4008c04030e1SUlrich Drepper 400911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 401011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4011442e31caSJeff Layton retry: 401211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40131da177e4SLinus Torvalds if (error) 40142ad94ae6SAl Viro return error; 40152ad94ae6SAl Viro 4016442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4017442e31caSJeff Layton (how & LOOKUP_REVAL)); 40181da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40196902d925SDave Hansen if (IS_ERR(new_dentry)) 4020dae6ad8fSAl Viro goto out; 4021dae6ad8fSAl Viro 4022dae6ad8fSAl Viro error = -EXDEV; 4023dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4024dae6ad8fSAl Viro goto out_dput; 4025800179c9SKees Cook error = may_linkat(&old_path); 4026800179c9SKees Cook if (unlikely(error)) 4027800179c9SKees Cook goto out_dput; 4028dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4029be6d3e56SKentaro Takeda if (error) 4030a8104a9fSAl Viro goto out_dput; 4031146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 403275c3f29dSDave Hansen out_dput: 4033921a1650SAl Viro done_path_create(&new_path, new_dentry); 4034146a8595SJ. Bruce Fields if (delegated_inode) { 4035146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4036d22e6338SOleg Drokin if (!error) { 4037d22e6338SOleg Drokin path_put(&old_path); 4038146a8595SJ. Bruce Fields goto retry; 4039146a8595SJ. Bruce Fields } 4040d22e6338SOleg Drokin } 4041442e31caSJeff Layton if (retry_estale(error, how)) { 4042d22e6338SOleg Drokin path_put(&old_path); 4043442e31caSJeff Layton how |= LOOKUP_REVAL; 4044442e31caSJeff Layton goto retry; 4045442e31caSJeff Layton } 40461da177e4SLinus Torvalds out: 40472d8f3038SAl Viro path_put(&old_path); 40481da177e4SLinus Torvalds 40491da177e4SLinus Torvalds return error; 40501da177e4SLinus Torvalds } 40511da177e4SLinus Torvalds 40523480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 40535590ff0dSUlrich Drepper { 4054c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 40555590ff0dSUlrich Drepper } 40565590ff0dSUlrich Drepper 4057bc27027aSMiklos Szeredi /** 4058bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4059bc27027aSMiklos Szeredi * @old_dir: parent of source 4060bc27027aSMiklos Szeredi * @old_dentry: source 4061bc27027aSMiklos Szeredi * @new_dir: parent of destination 4062bc27027aSMiklos Szeredi * @new_dentry: destination 4063bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4064520c8b16SMiklos Szeredi * @flags: rename flags 4065bc27027aSMiklos Szeredi * 4066bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4067bc27027aSMiklos Szeredi * 4068bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4069bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4070bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4071bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4072bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4073bc27027aSMiklos Szeredi * so. 4074bc27027aSMiklos Szeredi * 4075bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4076bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4077bc27027aSMiklos Szeredi * to be NFS exported. 4078bc27027aSMiklos Szeredi * 40791da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 40801da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 40811da177e4SLinus Torvalds * Problems: 4082d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 40831da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 40841da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4085a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 40861da177e4SLinus Torvalds * story. 40876cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 40886cedba89SJ. Bruce Fields * and source (if it is not a directory). 40891b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 40901da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 40911da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4092a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 40931da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 40941da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 40951da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4096a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 40971da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 40981da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 40991da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4100e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41011b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41021da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4103c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41041da177e4SLinus Torvalds * locking]. 41051da177e4SLinus Torvalds */ 41061da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41078e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4108520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41091da177e4SLinus Torvalds { 41101da177e4SLinus Torvalds int error; 4111bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 411259b0df21SEric Paris const unsigned char *old_name; 4113bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4114bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4115da1ce067SMiklos Szeredi bool new_is_dir = false; 4116da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41171da177e4SLinus Torvalds 4118bc27027aSMiklos Szeredi if (source == target) 41191da177e4SLinus Torvalds return 0; 41201da177e4SLinus Torvalds 41211da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41221da177e4SLinus Torvalds if (error) 41231da177e4SLinus Torvalds return error; 41241da177e4SLinus Torvalds 4125da1ce067SMiklos Szeredi if (!target) { 4126a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4127da1ce067SMiklos Szeredi } else { 4128da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4129da1ce067SMiklos Szeredi 4130da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 41311da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4132da1ce067SMiklos Szeredi else 4133da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4134da1ce067SMiklos Szeredi } 41351da177e4SLinus Torvalds if (error) 41361da177e4SLinus Torvalds return error; 41371da177e4SLinus Torvalds 41387177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 41391da177e4SLinus Torvalds return -EPERM; 41401da177e4SLinus Torvalds 4141520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4142520c8b16SMiklos Szeredi return -EINVAL; 4143520c8b16SMiklos Szeredi 4144bc27027aSMiklos Szeredi /* 4145bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4146bc27027aSMiklos Szeredi * we'll need to flip '..'. 4147bc27027aSMiklos Szeredi */ 4148da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4149da1ce067SMiklos Szeredi if (is_dir) { 4150bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4151bc27027aSMiklos Szeredi if (error) 4152bc27027aSMiklos Szeredi return error; 4153bc27027aSMiklos Szeredi } 4154da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4155da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4156da1ce067SMiklos Szeredi if (error) 4157da1ce067SMiklos Szeredi return error; 4158da1ce067SMiklos Szeredi } 4159da1ce067SMiklos Szeredi } 41600eeca283SRobert Love 41610b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 41620b3974ebSMiklos Szeredi flags); 4163bc27027aSMiklos Szeredi if (error) 4164bc27027aSMiklos Szeredi return error; 4165bc27027aSMiklos Szeredi 4166bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4167bc27027aSMiklos Szeredi dget(new_dentry); 4168da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4169bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4170bc27027aSMiklos Szeredi else if (target) 4171bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4172bc27027aSMiklos Szeredi 4173bc27027aSMiklos Szeredi error = -EBUSY; 41747af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4175bc27027aSMiklos Szeredi goto out; 4176bc27027aSMiklos Szeredi 4177da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4178bc27027aSMiklos Szeredi error = -EMLINK; 4179da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4180bc27027aSMiklos Szeredi goto out; 4181da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4182da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4183da1ce067SMiklos Szeredi goto out; 4184da1ce067SMiklos Szeredi } 4185da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4186bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4187da1ce067SMiklos Szeredi if (!is_dir) { 4188bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4189bc27027aSMiklos Szeredi if (error) 4190bc27027aSMiklos Szeredi goto out; 4191da1ce067SMiklos Szeredi } 4192da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4193bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4194bc27027aSMiklos Szeredi if (error) 4195bc27027aSMiklos Szeredi goto out; 4196bc27027aSMiklos Szeredi } 41977177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4198520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4199520c8b16SMiklos Szeredi new_dir, new_dentry); 4200520c8b16SMiklos Szeredi } else { 42017177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4202520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4203520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4204520c8b16SMiklos Szeredi } 4205bc27027aSMiklos Szeredi if (error) 4206bc27027aSMiklos Szeredi goto out; 4207bc27027aSMiklos Szeredi 4208da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42091da177e4SLinus Torvalds if (is_dir) 4210bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4211bc27027aSMiklos Szeredi dont_mount(new_dentry); 42128ed936b5SEric W. Biederman detach_mounts(new_dentry); 4213bc27027aSMiklos Szeredi } 4214da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4215da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4216bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4217da1ce067SMiklos Szeredi else 4218da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4219da1ce067SMiklos Szeredi } 4220bc27027aSMiklos Szeredi out: 4221da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4222bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4223bc27027aSMiklos Szeredi else if (target) 4224bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4225bc27027aSMiklos Szeredi dput(new_dentry); 4226da1ce067SMiklos Szeredi if (!error) { 4227123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4228da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4229da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4230da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4231da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4232da1ce067SMiklos Szeredi } 4233da1ce067SMiklos Szeredi } 42340eeca283SRobert Love fsnotify_oldname_free(old_name); 42350eeca283SRobert Love 42361da177e4SLinus Torvalds return error; 42371da177e4SLinus Torvalds } 42384d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 42391da177e4SLinus Torvalds 4240520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4241520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 42421da177e4SLinus Torvalds { 42431da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 42441da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 42451da177e4SLinus Torvalds struct dentry *trap; 42461da177e4SLinus Torvalds struct nameidata oldnd, newnd; 42478e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 424891a27b2aSJeff Layton struct filename *from; 424991a27b2aSJeff Layton struct filename *to; 4250c6a94284SJeff Layton unsigned int lookup_flags = 0; 4251c6a94284SJeff Layton bool should_retry = false; 42522ad94ae6SAl Viro int error; 4253520c8b16SMiklos Szeredi 42540d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4255da1ce067SMiklos Szeredi return -EINVAL; 4256da1ce067SMiklos Szeredi 42570d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 42580d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4259520c8b16SMiklos Szeredi return -EINVAL; 4260520c8b16SMiklos Szeredi 42610d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 42620d7a8555SMiklos Szeredi return -EPERM; 42630d7a8555SMiklos Szeredi 4264c6a94284SJeff Layton retry: 4265c6a94284SJeff Layton from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); 426691a27b2aSJeff Layton if (IS_ERR(from)) { 426791a27b2aSJeff Layton error = PTR_ERR(from); 42681da177e4SLinus Torvalds goto exit; 426991a27b2aSJeff Layton } 42701da177e4SLinus Torvalds 4271c6a94284SJeff Layton to = user_path_parent(newdfd, newname, &newnd, lookup_flags); 427291a27b2aSJeff Layton if (IS_ERR(to)) { 427391a27b2aSJeff Layton error = PTR_ERR(to); 42741da177e4SLinus Torvalds goto exit1; 427591a27b2aSJeff Layton } 42761da177e4SLinus Torvalds 42771da177e4SLinus Torvalds error = -EXDEV; 42784ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 42791da177e4SLinus Torvalds goto exit2; 42801da177e4SLinus Torvalds 42814ac91378SJan Blunck old_dir = oldnd.path.dentry; 42821da177e4SLinus Torvalds error = -EBUSY; 42831da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 42841da177e4SLinus Torvalds goto exit2; 42851da177e4SLinus Torvalds 42864ac91378SJan Blunck new_dir = newnd.path.dentry; 42870a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 42880a7c3937SMiklos Szeredi error = -EEXIST; 42891da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 42901da177e4SLinus Torvalds goto exit2; 42911da177e4SLinus Torvalds 4292c30dabfeSJan Kara error = mnt_want_write(oldnd.path.mnt); 4293c30dabfeSJan Kara if (error) 4294c30dabfeSJan Kara goto exit2; 4295c30dabfeSJan Kara 42960612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 42970612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 4298da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42994e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 43000612d9fbSOGAWA Hirofumi 43018e6d782cSJ. Bruce Fields retry_deleg: 43021da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 43031da177e4SLinus Torvalds 430449705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 43051da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43061da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43071da177e4SLinus Torvalds goto exit3; 43081da177e4SLinus Torvalds /* source must exist */ 43091da177e4SLinus Torvalds error = -ENOENT; 4310b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43111da177e4SLinus Torvalds goto exit4; 431249705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 43131da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43141da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43151da177e4SLinus Torvalds goto exit4; 43160a7c3937SMiklos Szeredi error = -EEXIST; 43170a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43180a7c3937SMiklos Szeredi goto exit5; 4319da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4320da1ce067SMiklos Szeredi error = -ENOENT; 4321da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4322da1ce067SMiklos Szeredi goto exit5; 4323da1ce067SMiklos Szeredi 4324da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4325da1ce067SMiklos Szeredi error = -ENOTDIR; 4326da1ce067SMiklos Szeredi if (newnd.last.name[newnd.last.len]) 4327da1ce067SMiklos Szeredi goto exit5; 4328da1ce067SMiklos Szeredi } 4329da1ce067SMiklos Szeredi } 43300a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 43310a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 43320a7c3937SMiklos Szeredi error = -ENOTDIR; 43330a7c3937SMiklos Szeredi if (oldnd.last.name[oldnd.last.len]) 43340a7c3937SMiklos Szeredi goto exit5; 4335da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len]) 43360a7c3937SMiklos Szeredi goto exit5; 43370a7c3937SMiklos Szeredi } 43380a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 43390a7c3937SMiklos Szeredi error = -EINVAL; 43400a7c3937SMiklos Szeredi if (old_dentry == trap) 43410a7c3937SMiklos Szeredi goto exit5; 43421da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4343da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43441da177e4SLinus Torvalds error = -ENOTEMPTY; 43451da177e4SLinus Torvalds if (new_dentry == trap) 43461da177e4SLinus Torvalds goto exit5; 43471da177e4SLinus Torvalds 4348be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 43490b3974ebSMiklos Szeredi &newnd.path, new_dentry, flags); 4350be6d3e56SKentaro Takeda if (error) 4351c30dabfeSJan Kara goto exit5; 43521da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 43538e6d782cSJ. Bruce Fields new_dir->d_inode, new_dentry, 4354520c8b16SMiklos Szeredi &delegated_inode, flags); 43551da177e4SLinus Torvalds exit5: 43561da177e4SLinus Torvalds dput(new_dentry); 43571da177e4SLinus Torvalds exit4: 43581da177e4SLinus Torvalds dput(old_dentry); 43591da177e4SLinus Torvalds exit3: 43601da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 43618e6d782cSJ. Bruce Fields if (delegated_inode) { 43628e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 43638e6d782cSJ. Bruce Fields if (!error) 43648e6d782cSJ. Bruce Fields goto retry_deleg; 43658e6d782cSJ. Bruce Fields } 4366c30dabfeSJan Kara mnt_drop_write(oldnd.path.mnt); 43671da177e4SLinus Torvalds exit2: 4368c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4369c6a94284SJeff Layton should_retry = true; 43701d957f9bSJan Blunck path_put(&newnd.path); 43712ad94ae6SAl Viro putname(to); 43721da177e4SLinus Torvalds exit1: 43731d957f9bSJan Blunck path_put(&oldnd.path); 43741da177e4SLinus Torvalds putname(from); 4375c6a94284SJeff Layton if (should_retry) { 4376c6a94284SJeff Layton should_retry = false; 4377c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4378c6a94284SJeff Layton goto retry; 4379c6a94284SJeff Layton } 43802ad94ae6SAl Viro exit: 43811da177e4SLinus Torvalds return error; 43821da177e4SLinus Torvalds } 43831da177e4SLinus Torvalds 4384520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4385520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4386520c8b16SMiklos Szeredi { 4387520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4388520c8b16SMiklos Szeredi } 4389520c8b16SMiklos Szeredi 4390a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 43915590ff0dSUlrich Drepper { 4392520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43935590ff0dSUlrich Drepper } 43945590ff0dSUlrich Drepper 4395787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4396787fb6bcSMiklos Szeredi { 4397787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4398787fb6bcSMiklos Szeredi if (error) 4399787fb6bcSMiklos Szeredi return error; 4400787fb6bcSMiklos Szeredi 4401787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4402787fb6bcSMiklos Szeredi return -EPERM; 4403787fb6bcSMiklos Szeredi 4404787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4405787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4406787fb6bcSMiklos Szeredi } 4407787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4408787fb6bcSMiklos Szeredi 44095d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44101da177e4SLinus Torvalds { 44115d826c84SAl Viro int len = PTR_ERR(link); 44121da177e4SLinus Torvalds if (IS_ERR(link)) 44131da177e4SLinus Torvalds goto out; 44141da177e4SLinus Torvalds 44151da177e4SLinus Torvalds len = strlen(link); 44161da177e4SLinus Torvalds if (len > (unsigned) buflen) 44171da177e4SLinus Torvalds len = buflen; 44181da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44191da177e4SLinus Torvalds len = -EFAULT; 44201da177e4SLinus Torvalds out: 44211da177e4SLinus Torvalds return len; 44221da177e4SLinus Torvalds } 44235d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44241da177e4SLinus Torvalds 44251da177e4SLinus Torvalds /* 44261da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44271da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44281da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 44291da177e4SLinus Torvalds */ 44301da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44311da177e4SLinus Torvalds { 44321da177e4SLinus Torvalds struct nameidata nd; 4433cc314eefSLinus Torvalds void *cookie; 4434694a1764SMarcin Slusarz int res; 4435cc314eefSLinus Torvalds 44361da177e4SLinus Torvalds nd.depth = 0; 4437cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 4438694a1764SMarcin Slusarz if (IS_ERR(cookie)) 4439694a1764SMarcin Slusarz return PTR_ERR(cookie); 4440694a1764SMarcin Slusarz 44415d826c84SAl Viro res = readlink_copy(buffer, buflen, nd_get_link(&nd)); 44421da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 4443cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 4444694a1764SMarcin Slusarz return res; 44451da177e4SLinus Torvalds } 44464d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 44471da177e4SLinus Torvalds 44481da177e4SLinus Torvalds /* get the link contents into pagecache */ 44491da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 44501da177e4SLinus Torvalds { 4451ebd09abbSDuane Griffin char *kaddr; 44521da177e4SLinus Torvalds struct page *page; 44531da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4454090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 44551da177e4SLinus Torvalds if (IS_ERR(page)) 44566fe6900eSNick Piggin return (char*)page; 44571da177e4SLinus Torvalds *ppage = page; 4458ebd09abbSDuane Griffin kaddr = kmap(page); 4459ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4460ebd09abbSDuane Griffin return kaddr; 44611da177e4SLinus Torvalds } 44621da177e4SLinus Torvalds 44631da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44641da177e4SLinus Torvalds { 44651da177e4SLinus Torvalds struct page *page = NULL; 44665d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 44671da177e4SLinus Torvalds if (page) { 44681da177e4SLinus Torvalds kunmap(page); 44691da177e4SLinus Torvalds page_cache_release(page); 44701da177e4SLinus Torvalds } 44711da177e4SLinus Torvalds return res; 44721da177e4SLinus Torvalds } 44734d359507SAl Viro EXPORT_SYMBOL(page_readlink); 44741da177e4SLinus Torvalds 4475cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 44761da177e4SLinus Torvalds { 4477cc314eefSLinus Torvalds struct page *page = NULL; 44781da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 4479cc314eefSLinus Torvalds return page; 44801da177e4SLinus Torvalds } 44814d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 44821da177e4SLinus Torvalds 4483cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 44841da177e4SLinus Torvalds { 4485cc314eefSLinus Torvalds struct page *page = cookie; 4486cc314eefSLinus Torvalds 4487cc314eefSLinus Torvalds if (page) { 44881da177e4SLinus Torvalds kunmap(page); 44891da177e4SLinus Torvalds page_cache_release(page); 44901da177e4SLinus Torvalds } 44911da177e4SLinus Torvalds } 44924d359507SAl Viro EXPORT_SYMBOL(page_put_link); 44931da177e4SLinus Torvalds 449454566b2cSNick Piggin /* 449554566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 449654566b2cSNick Piggin */ 449754566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 44981da177e4SLinus Torvalds { 44991da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45000adb25d2SKirill Korotaev struct page *page; 4501afddba49SNick Piggin void *fsdata; 4502beb497abSDmitriy Monakhov int err; 45031da177e4SLinus Torvalds char *kaddr; 450454566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 450554566b2cSNick Piggin if (nofs) 450654566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45071da177e4SLinus Torvalds 45087e53cac4SNeilBrown retry: 4509afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 451054566b2cSNick Piggin flags, &page, &fsdata); 45111da177e4SLinus Torvalds if (err) 4512afddba49SNick Piggin goto fail; 4513afddba49SNick Piggin 4514e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 45151da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4516e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4517afddba49SNick Piggin 4518afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4519afddba49SNick Piggin page, fsdata); 45201da177e4SLinus Torvalds if (err < 0) 45211da177e4SLinus Torvalds goto fail; 4522afddba49SNick Piggin if (err < len-1) 4523afddba49SNick Piggin goto retry; 4524afddba49SNick Piggin 45251da177e4SLinus Torvalds mark_inode_dirty(inode); 45261da177e4SLinus Torvalds return 0; 45271da177e4SLinus Torvalds fail: 45281da177e4SLinus Torvalds return err; 45291da177e4SLinus Torvalds } 45304d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 45311da177e4SLinus Torvalds 45320adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 45330adb25d2SKirill Korotaev { 45340adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 453554566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 45360adb25d2SKirill Korotaev } 45374d359507SAl Viro EXPORT_SYMBOL(page_symlink); 45380adb25d2SKirill Korotaev 453992e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 45401da177e4SLinus Torvalds .readlink = generic_readlink, 45411da177e4SLinus Torvalds .follow_link = page_follow_link_light, 45421da177e4SLinus Torvalds .put_link = page_put_link, 45431da177e4SLinus Torvalds }; 45441da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4545