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 */ 12191a27b2aSJeff Layton void final_putname(struct filename *name) 1221da177e4SLinus Torvalds { 1237950e385SJeff Layton if (name->separate) { 12491a27b2aSJeff Layton __putname(name->name); 12591a27b2aSJeff Layton kfree(name); 1267950e385SJeff Layton } else { 1277950e385SJeff Layton __putname(name); 12891a27b2aSJeff Layton } 1297950e385SJeff Layton } 1307950e385SJeff Layton 1317950e385SJeff Layton #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename)) 13291a27b2aSJeff Layton 13351f39a1fSDavid Drysdale struct filename * 13491a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13591a27b2aSJeff Layton { 13691a27b2aSJeff Layton struct filename *result, *err; 1373f9f0aa6SLinus Torvalds int len; 1387950e385SJeff Layton long max; 1397950e385SJeff Layton char *kname; 1401da177e4SLinus Torvalds 1417ac86265SJeff Layton result = audit_reusename(filename); 1427ac86265SJeff Layton if (result) 1437ac86265SJeff Layton return result; 1447ac86265SJeff Layton 1457950e385SJeff Layton result = __getname(); 1463f9f0aa6SLinus Torvalds if (unlikely(!result)) 1474043cde8SEric Paris return ERR_PTR(-ENOMEM); 1481da177e4SLinus Torvalds 1497950e385SJeff Layton /* 1507950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1517950e385SJeff Layton * allocation 1527950e385SJeff Layton */ 1537950e385SJeff Layton kname = (char *)result + sizeof(*result); 15491a27b2aSJeff Layton result->name = kname; 1557950e385SJeff Layton result->separate = false; 1567950e385SJeff Layton max = EMBEDDED_NAME_MAX; 1577950e385SJeff Layton 1587950e385SJeff Layton recopy: 1597950e385SJeff Layton len = strncpy_from_user(kname, filename, max); 16091a27b2aSJeff Layton if (unlikely(len < 0)) { 1613f9f0aa6SLinus Torvalds err = ERR_PTR(len); 1623f9f0aa6SLinus Torvalds goto error; 16391a27b2aSJeff Layton } 1643f9f0aa6SLinus Torvalds 1657950e385SJeff Layton /* 1667950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1677950e385SJeff Layton * separate struct filename so we can dedicate the entire 1687950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1697950e385SJeff Layton * userland. 1707950e385SJeff Layton */ 1717950e385SJeff Layton if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) { 1727950e385SJeff Layton kname = (char *)result; 1737950e385SJeff Layton 1747950e385SJeff Layton result = kzalloc(sizeof(*result), GFP_KERNEL); 1757950e385SJeff Layton if (!result) { 1767950e385SJeff Layton err = ERR_PTR(-ENOMEM); 1777950e385SJeff Layton result = (struct filename *)kname; 1787950e385SJeff Layton goto error; 1797950e385SJeff Layton } 1807950e385SJeff Layton result->name = kname; 1817950e385SJeff Layton result->separate = true; 1827950e385SJeff Layton max = PATH_MAX; 1837950e385SJeff Layton goto recopy; 1847950e385SJeff Layton } 1857950e385SJeff Layton 1863f9f0aa6SLinus Torvalds /* The empty path is special. */ 1873f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1883f9f0aa6SLinus Torvalds if (empty) 1891fa1e7f6SAndy Whitcroft *empty = 1; 1903f9f0aa6SLinus Torvalds err = ERR_PTR(-ENOENT); 1913f9f0aa6SLinus Torvalds if (!(flags & LOOKUP_EMPTY)) 1923f9f0aa6SLinus Torvalds goto error; 1931da177e4SLinus Torvalds } 1943f9f0aa6SLinus Torvalds 1953f9f0aa6SLinus Torvalds err = ERR_PTR(-ENAMETOOLONG); 1967950e385SJeff Layton if (unlikely(len >= PATH_MAX)) 1977950e385SJeff Layton goto error; 1987950e385SJeff Layton 1997950e385SJeff Layton result->uptr = filename; 200c4ad8f98SLinus Torvalds result->aname = NULL; 2011da177e4SLinus Torvalds audit_getname(result); 2021da177e4SLinus Torvalds return result; 2031da177e4SLinus Torvalds 2043f9f0aa6SLinus Torvalds error: 2057950e385SJeff Layton final_putname(result); 2063f9f0aa6SLinus Torvalds return err; 2073f9f0aa6SLinus Torvalds } 2083f9f0aa6SLinus Torvalds 20991a27b2aSJeff Layton struct filename * 21091a27b2aSJeff Layton getname(const char __user * filename) 211f52e0c11SAl Viro { 212f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 213f52e0c11SAl Viro } 214f52e0c11SAl Viro 215c4ad8f98SLinus Torvalds /* 216c4ad8f98SLinus Torvalds * The "getname_kernel()" interface doesn't do pathnames longer 217c4ad8f98SLinus Torvalds * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user. 218c4ad8f98SLinus Torvalds */ 219c4ad8f98SLinus Torvalds struct filename * 220c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 221c4ad8f98SLinus Torvalds { 222c4ad8f98SLinus Torvalds struct filename *result; 223c4ad8f98SLinus Torvalds char *kname; 224c4ad8f98SLinus Torvalds int len; 225c4ad8f98SLinus Torvalds 226c4ad8f98SLinus Torvalds len = strlen(filename); 227c4ad8f98SLinus Torvalds if (len >= EMBEDDED_NAME_MAX) 228c4ad8f98SLinus Torvalds return ERR_PTR(-ENAMETOOLONG); 229c4ad8f98SLinus Torvalds 230c4ad8f98SLinus Torvalds result = __getname(); 231c4ad8f98SLinus Torvalds if (unlikely(!result)) 232c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 233c4ad8f98SLinus Torvalds 234c4ad8f98SLinus Torvalds kname = (char *)result + sizeof(*result); 235c4ad8f98SLinus Torvalds result->name = kname; 236c4ad8f98SLinus Torvalds result->uptr = NULL; 237c4ad8f98SLinus Torvalds result->aname = NULL; 238c4ad8f98SLinus Torvalds result->separate = false; 239c4ad8f98SLinus Torvalds 240c4ad8f98SLinus Torvalds strlcpy(kname, filename, EMBEDDED_NAME_MAX); 241c4ad8f98SLinus Torvalds return result; 242c4ad8f98SLinus Torvalds } 243c4ad8f98SLinus Torvalds 2441da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 24591a27b2aSJeff Layton void putname(struct filename *name) 2461da177e4SLinus Torvalds { 2475ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 24891a27b2aSJeff Layton return audit_putname(name); 24991a27b2aSJeff Layton final_putname(name); 2501da177e4SLinus Torvalds } 2511da177e4SLinus Torvalds #endif 2521da177e4SLinus Torvalds 253e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 254e77819e5SLinus Torvalds { 25584635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 256e77819e5SLinus Torvalds struct posix_acl *acl; 257e77819e5SLinus Torvalds 258e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2593567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2603567866bSAl Viro if (!acl) 261e77819e5SLinus Torvalds return -EAGAIN; 2623567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 2633567866bSAl Viro if (acl == ACL_NOT_CACHED) 264e77819e5SLinus Torvalds return -ECHILD; 265206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 266e77819e5SLinus Torvalds } 267e77819e5SLinus Torvalds 2682982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2694e34e719SChristoph Hellwig if (IS_ERR(acl)) 2704e34e719SChristoph Hellwig return PTR_ERR(acl); 271e77819e5SLinus Torvalds if (acl) { 272e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 273e77819e5SLinus Torvalds posix_acl_release(acl); 274e77819e5SLinus Torvalds return error; 275e77819e5SLinus Torvalds } 27684635d68SLinus Torvalds #endif 277e77819e5SLinus Torvalds 278e77819e5SLinus Torvalds return -EAGAIN; 279e77819e5SLinus Torvalds } 280e77819e5SLinus Torvalds 2815909ccaaSLinus Torvalds /* 282948409c7SAndreas Gruenbacher * This does the basic permission checking 2835909ccaaSLinus Torvalds */ 2847e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2855909ccaaSLinus Torvalds { 28626cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2875909ccaaSLinus Torvalds 2888e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2895909ccaaSLinus Torvalds mode >>= 6; 2905909ccaaSLinus Torvalds else { 291e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2927e40145eSAl Viro int error = check_acl(inode, mask); 2935909ccaaSLinus Torvalds if (error != -EAGAIN) 2945909ccaaSLinus Torvalds return error; 2955909ccaaSLinus Torvalds } 2965909ccaaSLinus Torvalds 2975909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 2985909ccaaSLinus Torvalds mode >>= 3; 2995909ccaaSLinus Torvalds } 3005909ccaaSLinus Torvalds 3015909ccaaSLinus Torvalds /* 3025909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3035909ccaaSLinus Torvalds */ 3049c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3055909ccaaSLinus Torvalds return 0; 3065909ccaaSLinus Torvalds return -EACCES; 3075909ccaaSLinus Torvalds } 3081da177e4SLinus Torvalds 3091da177e4SLinus Torvalds /** 3101da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3111da177e4SLinus Torvalds * @inode: inode to check access rights for 3128fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3131da177e4SLinus Torvalds * 3141da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3151da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3161da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 317b74c79e9SNick Piggin * are used for other things. 318b74c79e9SNick Piggin * 319b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 320b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 321b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3221da177e4SLinus Torvalds */ 3232830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3241da177e4SLinus Torvalds { 3255909ccaaSLinus Torvalds int ret; 3261da177e4SLinus Torvalds 3271da177e4SLinus Torvalds /* 328948409c7SAndreas Gruenbacher * Do the basic permission checks. 3291da177e4SLinus Torvalds */ 3307e40145eSAl Viro ret = acl_permission_check(inode, mask); 3315909ccaaSLinus Torvalds if (ret != -EACCES) 3325909ccaaSLinus Torvalds return ret; 3331da177e4SLinus Torvalds 334d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 335d594e7ecSAl Viro /* DACs are overridable for directories */ 33623adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 337d594e7ecSAl Viro return 0; 338d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 33923adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34023adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 341d594e7ecSAl Viro return 0; 342d594e7ecSAl Viro return -EACCES; 343d594e7ecSAl Viro } 3441da177e4SLinus Torvalds /* 3451da177e4SLinus Torvalds * Read/write DACs are always overridable. 346d594e7ecSAl Viro * Executable DACs are overridable when there is 347d594e7ecSAl Viro * at least one exec bit set. 3481da177e4SLinus Torvalds */ 349d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 35023adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3511da177e4SLinus Torvalds return 0; 3521da177e4SLinus Torvalds 3531da177e4SLinus Torvalds /* 3541da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3551da177e4SLinus Torvalds */ 3567ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 357d594e7ecSAl Viro if (mask == MAY_READ) 35823adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3591da177e4SLinus Torvalds return 0; 3601da177e4SLinus Torvalds 3611da177e4SLinus Torvalds return -EACCES; 3621da177e4SLinus Torvalds } 3634d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3641da177e4SLinus Torvalds 3653ddcd056SLinus Torvalds /* 3663ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3673ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3683ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3693ddcd056SLinus Torvalds * permission function, use the fast case". 3703ddcd056SLinus Torvalds */ 3713ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3723ddcd056SLinus Torvalds { 3733ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3743ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3753ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3763ddcd056SLinus Torvalds 3773ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3783ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3793ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3803ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3813ddcd056SLinus Torvalds } 3823ddcd056SLinus Torvalds return generic_permission(inode, mask); 3833ddcd056SLinus Torvalds } 3843ddcd056SLinus Torvalds 385cb23beb5SChristoph Hellwig /** 3860bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3870bdaea90SDavid Howells * @inode: Inode to check permission on 3880bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 389cb23beb5SChristoph Hellwig * 3900bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 391948409c7SAndreas Gruenbacher * 392948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3930bdaea90SDavid Howells * 3940bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 3950bdaea90SDavid Howells * inode_permission(). 396cb23beb5SChristoph Hellwig */ 3970bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 3981da177e4SLinus Torvalds { 399e6305c43SAl Viro int retval; 4001da177e4SLinus Torvalds 4013ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 4021da177e4SLinus Torvalds /* 4031da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4041da177e4SLinus Torvalds */ 4051da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 4061da177e4SLinus Torvalds return -EACCES; 4071da177e4SLinus Torvalds } 4081da177e4SLinus Torvalds 4093ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4101da177e4SLinus Torvalds if (retval) 4111da177e4SLinus Torvalds return retval; 4121da177e4SLinus Torvalds 41308ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 41408ce5f16SSerge E. Hallyn if (retval) 41508ce5f16SSerge E. Hallyn return retval; 41608ce5f16SSerge E. Hallyn 417d09ca739SEric Paris return security_inode_permission(inode, mask); 4181da177e4SLinus Torvalds } 419bd5d0856SMiklos Szeredi EXPORT_SYMBOL(__inode_permission); 4201da177e4SLinus Torvalds 421f4d6ff89SAl Viro /** 4220bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4230bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 42455852635SRandy Dunlap * @inode: Inode to check permission on 4250bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4260bdaea90SDavid Howells * 4270bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4280bdaea90SDavid Howells */ 4290bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4300bdaea90SDavid Howells { 4310bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4320bdaea90SDavid Howells umode_t mode = inode->i_mode; 4330bdaea90SDavid Howells 4340bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4350bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4360bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4370bdaea90SDavid Howells return -EROFS; 4380bdaea90SDavid Howells } 4390bdaea90SDavid Howells return 0; 4400bdaea90SDavid Howells } 4410bdaea90SDavid Howells 4420bdaea90SDavid Howells /** 4430bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4440bdaea90SDavid Howells * @inode: Inode to check permission on 4450bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4460bdaea90SDavid Howells * 4470bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4480bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4490bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4500bdaea90SDavid Howells * 4510bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4520bdaea90SDavid Howells */ 4530bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4540bdaea90SDavid Howells { 4550bdaea90SDavid Howells int retval; 4560bdaea90SDavid Howells 4570bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4580bdaea90SDavid Howells if (retval) 4590bdaea90SDavid Howells return retval; 4600bdaea90SDavid Howells return __inode_permission(inode, mask); 4610bdaea90SDavid Howells } 4624d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4630bdaea90SDavid Howells 4640bdaea90SDavid Howells /** 4655dd784d0SJan Blunck * path_get - get a reference to a path 4665dd784d0SJan Blunck * @path: path to get the reference to 4675dd784d0SJan Blunck * 4685dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4695dd784d0SJan Blunck */ 470dcf787f3SAl Viro void path_get(const struct path *path) 4715dd784d0SJan Blunck { 4725dd784d0SJan Blunck mntget(path->mnt); 4735dd784d0SJan Blunck dget(path->dentry); 4745dd784d0SJan Blunck } 4755dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4765dd784d0SJan Blunck 4775dd784d0SJan Blunck /** 4781d957f9bSJan Blunck * path_put - put a reference to a path 4791d957f9bSJan Blunck * @path: path to put the reference to 4801d957f9bSJan Blunck * 4811d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4821d957f9bSJan Blunck */ 483dcf787f3SAl Viro void path_put(const struct path *path) 4841da177e4SLinus Torvalds { 4851d957f9bSJan Blunck dput(path->dentry); 4861d957f9bSJan Blunck mntput(path->mnt); 4871da177e4SLinus Torvalds } 4881d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4891da177e4SLinus Torvalds 49019660af7SAl Viro /* 49131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 49219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 49319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 49419660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 49519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 49619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 49719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 49819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 49931e6b01fSNick Piggin */ 50031e6b01fSNick Piggin 50131e6b01fSNick Piggin /** 50219660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 50319660af7SAl Viro * @nd: nameidata pathwalk data 50419660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 50539191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 50631e6b01fSNick Piggin * 50719660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 50819660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 50919660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 51031e6b01fSNick Piggin */ 51119660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 51231e6b01fSNick Piggin { 51331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 51431e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 51531e6b01fSNick Piggin 51631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 517e5c832d5SLinus Torvalds 518e5c832d5SLinus Torvalds /* 51948a066e7SAl Viro * After legitimizing the bastards, terminate_walk() 52048a066e7SAl Viro * will do the right thing for non-RCU mode, and all our 52148a066e7SAl Viro * subsequent exit cases should rcu_read_unlock() 52248a066e7SAl Viro * before returning. Do vfsmount first; if dentry 52348a066e7SAl Viro * can't be legitimized, just set nd->path.dentry to NULL 52448a066e7SAl Viro * and rely on dput(NULL) being a no-op. 525e5c832d5SLinus Torvalds */ 52648a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 527e5c832d5SLinus Torvalds return -ECHILD; 528e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 52915570086SLinus Torvalds 53048a066e7SAl Viro if (!lockref_get_not_dead(&parent->d_lockref)) { 53148a066e7SAl Viro nd->path.dentry = NULL; 532d870b4a1SAl Viro goto out; 53348a066e7SAl Viro } 53448a066e7SAl Viro 53515570086SLinus Torvalds /* 53615570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 53715570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 53815570086SLinus Torvalds * 53915570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 54015570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 54115570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 54215570086SLinus Torvalds * since we checked the sequence number of the parent after we got 54315570086SLinus Torvalds * the child sequence number. So we know the parent must still 54415570086SLinus Torvalds * be valid if the child sequence number is still valid. 54515570086SLinus Torvalds */ 54619660af7SAl Viro if (!dentry) { 547e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 548e5c832d5SLinus Torvalds goto out; 54919660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 55019660af7SAl Viro } else { 551e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 552e5c832d5SLinus Torvalds goto out; 553e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 554e5c832d5SLinus Torvalds goto drop_dentry; 55519660af7SAl Viro } 556e5c832d5SLinus Torvalds 557e5c832d5SLinus Torvalds /* 558e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 559e5c832d5SLinus Torvalds * still valid and get it if required. 560e5c832d5SLinus Torvalds */ 561e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 562e5c832d5SLinus Torvalds spin_lock(&fs->lock); 563e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 564e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 56531e6b01fSNick Piggin path_get(&nd->root); 56631e6b01fSNick Piggin spin_unlock(&fs->lock); 56731e6b01fSNick Piggin } 56831e6b01fSNick Piggin 5698b61e74fSAl Viro rcu_read_unlock(); 57031e6b01fSNick Piggin return 0; 57119660af7SAl Viro 572e5c832d5SLinus Torvalds unlock_and_drop_dentry: 57331e6b01fSNick Piggin spin_unlock(&fs->lock); 574e5c832d5SLinus Torvalds drop_dentry: 5758b61e74fSAl Viro rcu_read_unlock(); 576e5c832d5SLinus Torvalds dput(dentry); 577d0d27277SLinus Torvalds goto drop_root_mnt; 578e5c832d5SLinus Torvalds out: 5798b61e74fSAl Viro rcu_read_unlock(); 580d0d27277SLinus Torvalds drop_root_mnt: 581d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 582d0d27277SLinus Torvalds nd->root.mnt = NULL; 58331e6b01fSNick Piggin return -ECHILD; 58431e6b01fSNick Piggin } 58531e6b01fSNick Piggin 5864ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 58734286d66SNick Piggin { 5884ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 58934286d66SNick Piggin } 59034286d66SNick Piggin 5919f1fafeeSAl Viro /** 5929f1fafeeSAl Viro * complete_walk - successful completion of path walk 5939f1fafeeSAl Viro * @nd: pointer nameidata 59439159de2SJeff Layton * 5959f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 5969f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 5979f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 5989f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 5999f1fafeeSAl Viro * need to drop nd->path. 60039159de2SJeff Layton */ 6019f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 60239159de2SJeff Layton { 60316c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 60439159de2SJeff Layton int status; 60539159de2SJeff Layton 6069f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 6079f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 6089f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6099f1fafeeSAl Viro nd->root.mnt = NULL; 61015570086SLinus Torvalds 61148a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { 6128b61e74fSAl Viro rcu_read_unlock(); 61348a066e7SAl Viro return -ECHILD; 61448a066e7SAl Viro } 615e5c832d5SLinus Torvalds if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 6168b61e74fSAl Viro rcu_read_unlock(); 61748a066e7SAl Viro mntput(nd->path.mnt); 6189f1fafeeSAl Viro return -ECHILD; 6199f1fafeeSAl Viro } 620e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 6218b61e74fSAl Viro rcu_read_unlock(); 622e5c832d5SLinus Torvalds dput(dentry); 62348a066e7SAl Viro mntput(nd->path.mnt); 624e5c832d5SLinus Torvalds return -ECHILD; 625e5c832d5SLinus Torvalds } 6268b61e74fSAl Viro rcu_read_unlock(); 6279f1fafeeSAl Viro } 6289f1fafeeSAl Viro 62916c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 63039159de2SJeff Layton return 0; 63139159de2SJeff Layton 632ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 63316c2cd71SAl Viro return 0; 63416c2cd71SAl Viro 635ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 63639159de2SJeff Layton if (status > 0) 63739159de2SJeff Layton return 0; 63839159de2SJeff Layton 63916c2cd71SAl Viro if (!status) 64039159de2SJeff Layton status = -ESTALE; 64116c2cd71SAl Viro 6429f1fafeeSAl Viro path_put(&nd->path); 64339159de2SJeff Layton return status; 64439159de2SJeff Layton } 64539159de2SJeff Layton 6462a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6472a737871SAl Viro { 648f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6492a737871SAl Viro } 6502a737871SAl Viro 6516de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6526de88d72SAl Viro 6537bd88377SAl Viro static __always_inline unsigned set_root_rcu(struct nameidata *nd) 65431e6b01fSNick Piggin { 65531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 6567bd88377SAl Viro unsigned seq, res; 657c28cc364SNick Piggin 658c28cc364SNick Piggin do { 659c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 66031e6b01fSNick Piggin nd->root = fs->root; 6617bd88377SAl Viro res = __read_seqcount_begin(&nd->root.dentry->d_seq); 662c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 6637bd88377SAl Viro return res; 66431e6b01fSNick Piggin } 66531e6b01fSNick Piggin 6661d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 667051d3812SIan Kent { 668051d3812SIan Kent dput(path->dentry); 6694ac91378SJan Blunck if (path->mnt != nd->path.mnt) 670051d3812SIan Kent mntput(path->mnt); 671051d3812SIan Kent } 672051d3812SIan Kent 6737b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6747b9337aaSNick Piggin struct nameidata *nd) 675051d3812SIan Kent { 67631e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6774ac91378SJan Blunck dput(nd->path.dentry); 67831e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6794ac91378SJan Blunck mntput(nd->path.mnt); 6809a229683SHuang Shijie } 68131e6b01fSNick Piggin nd->path.mnt = path->mnt; 6824ac91378SJan Blunck nd->path.dentry = path->dentry; 683051d3812SIan Kent } 684051d3812SIan Kent 685b5fb63c1SChristoph Hellwig /* 686b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 687b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 688b5fb63c1SChristoph Hellwig */ 689b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 690b5fb63c1SChristoph Hellwig { 691b5fb63c1SChristoph Hellwig path_put(&nd->path); 692b5fb63c1SChristoph Hellwig 693b5fb63c1SChristoph Hellwig nd->path = *path; 694b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 695b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 696b5fb63c1SChristoph Hellwig } 697b5fb63c1SChristoph Hellwig 698574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 699574197e0SAl Viro { 700574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 7016d7b5aaeSAl Viro if (inode->i_op->put_link) 702574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 703574197e0SAl Viro path_put(link); 704574197e0SAl Viro } 705574197e0SAl Viro 706561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 707561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 708800179c9SKees Cook 709800179c9SKees Cook /** 710800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 711800179c9SKees Cook * @link: The path of the symlink 71255852635SRandy Dunlap * @nd: nameidata pathwalk data 713800179c9SKees Cook * 714800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 715800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 716800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 717800179c9SKees Cook * processes from failing races against path names that may change out 718800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 719800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 720800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 721800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 722800179c9SKees Cook * 723800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 724800179c9SKees Cook */ 725800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 726800179c9SKees Cook { 727800179c9SKees Cook const struct inode *inode; 728800179c9SKees Cook const struct inode *parent; 729800179c9SKees Cook 730800179c9SKees Cook if (!sysctl_protected_symlinks) 731800179c9SKees Cook return 0; 732800179c9SKees Cook 733800179c9SKees Cook /* Allowed if owner and follower match. */ 734800179c9SKees Cook inode = link->dentry->d_inode; 73581abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 736800179c9SKees Cook return 0; 737800179c9SKees Cook 738800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 739800179c9SKees Cook parent = nd->path.dentry->d_inode; 740800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 741800179c9SKees Cook return 0; 742800179c9SKees Cook 743800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 74481abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 745800179c9SKees Cook return 0; 746800179c9SKees Cook 747ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 748800179c9SKees Cook path_put_conditional(link, nd); 749800179c9SKees Cook path_put(&nd->path); 750800179c9SKees Cook return -EACCES; 751800179c9SKees Cook } 752800179c9SKees Cook 753800179c9SKees Cook /** 754800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 755800179c9SKees Cook * @inode: the source inode to hardlink from 756800179c9SKees Cook * 757800179c9SKees Cook * Return false if at least one of the following conditions: 758800179c9SKees Cook * - inode is not a regular file 759800179c9SKees Cook * - inode is setuid 760800179c9SKees Cook * - inode is setgid and group-exec 761800179c9SKees Cook * - access failure for read and write 762800179c9SKees Cook * 763800179c9SKees Cook * Otherwise returns true. 764800179c9SKees Cook */ 765800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 766800179c9SKees Cook { 767800179c9SKees Cook umode_t mode = inode->i_mode; 768800179c9SKees Cook 769800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 770800179c9SKees Cook if (!S_ISREG(mode)) 771800179c9SKees Cook return false; 772800179c9SKees Cook 773800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 774800179c9SKees Cook if (mode & S_ISUID) 775800179c9SKees Cook return false; 776800179c9SKees Cook 777800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 778800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 779800179c9SKees Cook return false; 780800179c9SKees Cook 781800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 782800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 783800179c9SKees Cook return false; 784800179c9SKees Cook 785800179c9SKees Cook return true; 786800179c9SKees Cook } 787800179c9SKees Cook 788800179c9SKees Cook /** 789800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 790800179c9SKees Cook * @link: the source to hardlink from 791800179c9SKees Cook * 792800179c9SKees Cook * Block hardlink when all of: 793800179c9SKees Cook * - sysctl_protected_hardlinks enabled 794800179c9SKees Cook * - fsuid does not match inode 795800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 796800179c9SKees Cook * - not CAP_FOWNER 797800179c9SKees Cook * 798800179c9SKees Cook * Returns 0 if successful, -ve on error. 799800179c9SKees Cook */ 800800179c9SKees Cook static int may_linkat(struct path *link) 801800179c9SKees Cook { 802800179c9SKees Cook const struct cred *cred; 803800179c9SKees Cook struct inode *inode; 804800179c9SKees Cook 805800179c9SKees Cook if (!sysctl_protected_hardlinks) 806800179c9SKees Cook return 0; 807800179c9SKees Cook 808800179c9SKees Cook cred = current_cred(); 809800179c9SKees Cook inode = link->dentry->d_inode; 810800179c9SKees Cook 811800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 812800179c9SKees Cook * otherwise, it must be a safe source. 813800179c9SKees Cook */ 81481abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 815800179c9SKees Cook capable(CAP_FOWNER)) 816800179c9SKees Cook return 0; 817800179c9SKees Cook 818a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 819800179c9SKees Cook return -EPERM; 820800179c9SKees Cook } 821800179c9SKees Cook 822def4af30SAl Viro static __always_inline int 823574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 8241da177e4SLinus Torvalds { 8257b9337aaSNick Piggin struct dentry *dentry = link->dentry; 8266d7b5aaeSAl Viro int error; 8276d7b5aaeSAl Viro char *s; 8281da177e4SLinus Torvalds 829844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 830844a3917SAl Viro 8310e794589SAl Viro if (link->mnt == nd->path.mnt) 8320e794589SAl Viro mntget(link->mnt); 8330e794589SAl Viro 8346d7b5aaeSAl Viro error = -ELOOP; 8356d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 8366d7b5aaeSAl Viro goto out_put_nd_path; 8376d7b5aaeSAl Viro 838574197e0SAl Viro cond_resched(); 839574197e0SAl Viro current->total_link_count++; 840574197e0SAl Viro 84168ac1234SAl Viro touch_atime(link); 8421da177e4SLinus Torvalds nd_set_link(nd, NULL); 843cd4e91d3SAl Viro 84436f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 8456d7b5aaeSAl Viro if (error) 8466d7b5aaeSAl Viro goto out_put_nd_path; 84736f3b4f6SAl Viro 84886acdca1SAl Viro nd->last_type = LAST_BIND; 849def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 850def4af30SAl Viro error = PTR_ERR(*p); 8516d7b5aaeSAl Viro if (IS_ERR(*p)) 852408ef013SChristoph Hellwig goto out_put_nd_path; 8536d7b5aaeSAl Viro 854cc314eefSLinus Torvalds error = 0; 8556d7b5aaeSAl Viro s = nd_get_link(nd); 8566d7b5aaeSAl Viro if (s) { 857443ed254SAl Viro if (unlikely(IS_ERR(s))) { 858443ed254SAl Viro path_put(&nd->path); 859443ed254SAl Viro put_link(nd, link, *p); 860443ed254SAl Viro return PTR_ERR(s); 861443ed254SAl Viro } 862443ed254SAl Viro if (*s == '/') { 8637bd88377SAl Viro if (!nd->root.mnt) 864443ed254SAl Viro set_root(nd); 865443ed254SAl Viro path_put(&nd->path); 866443ed254SAl Viro nd->path = nd->root; 867443ed254SAl Viro path_get(&nd->root); 868443ed254SAl Viro nd->flags |= LOOKUP_JUMPED; 869443ed254SAl Viro } 870443ed254SAl Viro nd->inode = nd->path.dentry->d_inode; 871443ed254SAl Viro error = link_path_walk(s, nd); 8726d7b5aaeSAl Viro if (unlikely(error)) 8736d7b5aaeSAl Viro put_link(nd, link, *p); 874b5fb63c1SChristoph Hellwig } 8756d7b5aaeSAl Viro 8766d7b5aaeSAl Viro return error; 8776d7b5aaeSAl Viro 8786d7b5aaeSAl Viro out_put_nd_path: 87998f6ef64SArnd Bergmann *p = NULL; 8806d7b5aaeSAl Viro path_put(&nd->path); 8816d7b5aaeSAl Viro path_put(link); 8821da177e4SLinus Torvalds return error; 8831da177e4SLinus Torvalds } 8841da177e4SLinus Torvalds 88531e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 88631e6b01fSNick Piggin { 8870714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 8880714a533SAl Viro struct mount *parent; 88931e6b01fSNick Piggin struct dentry *mountpoint; 89031e6b01fSNick Piggin 8910714a533SAl Viro parent = mnt->mnt_parent; 8920714a533SAl Viro if (&parent->mnt == path->mnt) 89331e6b01fSNick Piggin return 0; 894a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 89531e6b01fSNick Piggin path->dentry = mountpoint; 8960714a533SAl Viro path->mnt = &parent->mnt; 89731e6b01fSNick Piggin return 1; 89831e6b01fSNick Piggin } 89931e6b01fSNick Piggin 900f015f126SDavid Howells /* 901f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 902f015f126SDavid Howells * 903f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 904f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 905f015f126SDavid Howells * Up is towards /. 906f015f126SDavid Howells * 907f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 908f015f126SDavid Howells * root. 909f015f126SDavid Howells */ 910bab77ebfSAl Viro int follow_up(struct path *path) 9111da177e4SLinus Torvalds { 9120714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9130714a533SAl Viro struct mount *parent; 9141da177e4SLinus Torvalds struct dentry *mountpoint; 91599b7db7bSNick Piggin 91648a066e7SAl Viro read_seqlock_excl(&mount_lock); 9170714a533SAl Viro parent = mnt->mnt_parent; 9183c0a6163SAl Viro if (parent == mnt) { 91948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 9201da177e4SLinus Torvalds return 0; 9211da177e4SLinus Torvalds } 9220714a533SAl Viro mntget(&parent->mnt); 923a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 92448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 925bab77ebfSAl Viro dput(path->dentry); 926bab77ebfSAl Viro path->dentry = mountpoint; 927bab77ebfSAl Viro mntput(path->mnt); 9280714a533SAl Viro path->mnt = &parent->mnt; 9291da177e4SLinus Torvalds return 1; 9301da177e4SLinus Torvalds } 9314d359507SAl Viro EXPORT_SYMBOL(follow_up); 9321da177e4SLinus Torvalds 933b5c84bf6SNick Piggin /* 9349875cf80SDavid Howells * Perform an automount 9359875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9369875cf80SDavid Howells * were called with. 9371da177e4SLinus Torvalds */ 9389875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 9399875cf80SDavid Howells bool *need_mntput) 94031e6b01fSNick Piggin { 9419875cf80SDavid Howells struct vfsmount *mnt; 942ea5b778aSDavid Howells int err; 9439875cf80SDavid Howells 9449875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9459875cf80SDavid Howells return -EREMOTE; 9469875cf80SDavid Howells 9470ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9480ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9490ec26fd0SMiklos Szeredi * the name. 9500ec26fd0SMiklos Szeredi * 9510ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9525a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9530ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9540ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9550ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9560ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9575a30d8a2SDavid Howells */ 9585a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 959d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 9605a30d8a2SDavid Howells path->dentry->d_inode) 9619875cf80SDavid Howells return -EISDIR; 9620ec26fd0SMiklos Szeredi 9639875cf80SDavid Howells current->total_link_count++; 9649875cf80SDavid Howells if (current->total_link_count >= 40) 9659875cf80SDavid Howells return -ELOOP; 9669875cf80SDavid Howells 9679875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9689875cf80SDavid Howells if (IS_ERR(mnt)) { 9699875cf80SDavid Howells /* 9709875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 9719875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 9729875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 9739875cf80SDavid Howells * 9749875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 9759875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 9769875cf80SDavid Howells * the path is inaccessible and we should say so. 9779875cf80SDavid Howells */ 97849084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 9799875cf80SDavid Howells return -EREMOTE; 9809875cf80SDavid Howells return PTR_ERR(mnt); 98131e6b01fSNick Piggin } 982ea5b778aSDavid Howells 9839875cf80SDavid Howells if (!mnt) /* mount collision */ 9849875cf80SDavid Howells return 0; 9859875cf80SDavid Howells 9868aef1884SAl Viro if (!*need_mntput) { 9878aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 9888aef1884SAl Viro mntget(path->mnt); 9898aef1884SAl Viro *need_mntput = true; 9908aef1884SAl Viro } 99119a167afSAl Viro err = finish_automount(mnt, path); 992ea5b778aSDavid Howells 993ea5b778aSDavid Howells switch (err) { 994ea5b778aSDavid Howells case -EBUSY: 995ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 99619a167afSAl Viro return 0; 997ea5b778aSDavid Howells case 0: 9988aef1884SAl Viro path_put(path); 9999875cf80SDavid Howells path->mnt = mnt; 10009875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 10019875cf80SDavid Howells return 0; 100219a167afSAl Viro default: 100319a167afSAl Viro return err; 10049875cf80SDavid Howells } 100519a167afSAl Viro 1006ea5b778aSDavid Howells } 10079875cf80SDavid Howells 10089875cf80SDavid Howells /* 10099875cf80SDavid Howells * Handle a dentry that is managed in some way. 1010cc53ce53SDavid Howells * - Flagged for transit management (autofs) 10119875cf80SDavid Howells * - Flagged as mountpoint 10129875cf80SDavid Howells * - Flagged as automount point 10139875cf80SDavid Howells * 10149875cf80SDavid Howells * This may only be called in refwalk mode. 10159875cf80SDavid Howells * 10169875cf80SDavid Howells * Serialization is taken care of in namespace.c 10179875cf80SDavid Howells */ 10189875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 10199875cf80SDavid Howells { 10208aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10219875cf80SDavid Howells unsigned managed; 10229875cf80SDavid Howells bool need_mntput = false; 10238aef1884SAl Viro int ret = 0; 10249875cf80SDavid Howells 10259875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10269875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10279875cf80SDavid Howells * the components of that value change under us */ 10289875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10299875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10309875cf80SDavid Howells unlikely(managed != 0)) { 1031cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1032cc53ce53SDavid Howells * being held. */ 1033cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1034cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1035cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10361aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1037cc53ce53SDavid Howells if (ret < 0) 10388aef1884SAl Viro break; 1039cc53ce53SDavid Howells } 1040cc53ce53SDavid Howells 10419875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10429875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10439875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10449875cf80SDavid Howells if (mounted) { 10459875cf80SDavid Howells dput(path->dentry); 10469875cf80SDavid Howells if (need_mntput) 1047463ffb2eSAl Viro mntput(path->mnt); 1048463ffb2eSAl Viro path->mnt = mounted; 1049463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10509875cf80SDavid Howells need_mntput = true; 10519875cf80SDavid Howells continue; 1052463ffb2eSAl Viro } 1053463ffb2eSAl Viro 10549875cf80SDavid Howells /* Something is mounted on this dentry in another 10559875cf80SDavid Howells * namespace and/or whatever was mounted there in this 105648a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 105748a066e7SAl Viro * get it */ 10581da177e4SLinus Torvalds } 10599875cf80SDavid Howells 10609875cf80SDavid Howells /* Handle an automount point */ 10619875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 10629875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 10639875cf80SDavid Howells if (ret < 0) 10648aef1884SAl Viro break; 10659875cf80SDavid Howells continue; 10669875cf80SDavid Howells } 10679875cf80SDavid Howells 10689875cf80SDavid Howells /* We didn't change the current path point */ 10699875cf80SDavid Howells break; 10709875cf80SDavid Howells } 10718aef1884SAl Viro 10728aef1884SAl Viro if (need_mntput && path->mnt == mnt) 10738aef1884SAl Viro mntput(path->mnt); 10748aef1884SAl Viro if (ret == -EISDIR) 10758aef1884SAl Viro ret = 0; 1076a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 1079cc53ce53SDavid Howells int follow_down_one(struct path *path) 10801da177e4SLinus Torvalds { 10811da177e4SLinus Torvalds struct vfsmount *mounted; 10821da177e4SLinus Torvalds 10831c755af4SAl Viro mounted = lookup_mnt(path); 10841da177e4SLinus Torvalds if (mounted) { 10859393bd07SAl Viro dput(path->dentry); 10869393bd07SAl Viro mntput(path->mnt); 10879393bd07SAl Viro path->mnt = mounted; 10889393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 10891da177e4SLinus Torvalds return 1; 10901da177e4SLinus Torvalds } 10911da177e4SLinus Torvalds return 0; 10921da177e4SLinus Torvalds } 10934d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 10941da177e4SLinus Torvalds 1095b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 109662a7375eSIan Kent { 1097b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1098b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 109962a7375eSIan Kent } 110062a7375eSIan Kent 11019875cf80SDavid Howells /* 1102287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1103287548e4SAl Viro * we meet a managed dentry that would need blocking. 11049875cf80SDavid Howells */ 11059875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1106287548e4SAl Viro struct inode **inode) 11079875cf80SDavid Howells { 110862a7375eSIan Kent for (;;) { 1109c7105365SAl Viro struct mount *mounted; 111062a7375eSIan Kent /* 111162a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 111262a7375eSIan Kent * that wants to block transit. 111362a7375eSIan Kent */ 1114b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1115b8faf035SNeilBrown case -ECHILD: 1116b8faf035SNeilBrown default: 1117ab90911fSDavid Howells return false; 1118b8faf035SNeilBrown case -EISDIR: 1119b8faf035SNeilBrown return true; 1120b8faf035SNeilBrown case 0: 1121b8faf035SNeilBrown break; 1122b8faf035SNeilBrown } 112362a7375eSIan Kent 112462a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1125b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 112662a7375eSIan Kent 1127474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 11289875cf80SDavid Howells if (!mounted) 11299875cf80SDavid Howells break; 1130c7105365SAl Viro path->mnt = &mounted->mnt; 1131c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1132a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 11339875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 113459430262SLinus Torvalds /* 113559430262SLinus Torvalds * Update the inode too. We don't need to re-check the 113659430262SLinus Torvalds * dentry sequence number here after this d_inode read, 113759430262SLinus Torvalds * because a mount-point is always pinned. 113859430262SLinus Torvalds */ 113959430262SLinus Torvalds *inode = path->dentry->d_inode; 11409875cf80SDavid Howells } 1141f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1142b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1143287548e4SAl Viro } 1144287548e4SAl Viro 114531e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 114631e6b01fSNick Piggin { 11474023bfc9SAl Viro struct inode *inode = nd->inode; 11487bd88377SAl Viro if (!nd->root.mnt) 114931e6b01fSNick Piggin set_root_rcu(nd); 115031e6b01fSNick Piggin 115131e6b01fSNick Piggin while (1) { 115231e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 115331e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 115431e6b01fSNick Piggin break; 115531e6b01fSNick Piggin } 115631e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 115731e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 115831e6b01fSNick Piggin struct dentry *parent = old->d_parent; 115931e6b01fSNick Piggin unsigned seq; 116031e6b01fSNick Piggin 11614023bfc9SAl Viro inode = parent->d_inode; 116231e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 116331e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1164ef7562d5SAl Viro goto failed; 116531e6b01fSNick Piggin nd->path.dentry = parent; 116631e6b01fSNick Piggin nd->seq = seq; 116731e6b01fSNick Piggin break; 116831e6b01fSNick Piggin } 116931e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 117031e6b01fSNick Piggin break; 11714023bfc9SAl Viro inode = nd->path.dentry->d_inode; 117231e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 117331e6b01fSNick Piggin } 1174b37199e6SAl Viro while (d_mountpoint(nd->path.dentry)) { 1175b37199e6SAl Viro struct mount *mounted; 1176b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1177b37199e6SAl Viro if (!mounted) 1178b37199e6SAl Viro break; 1179b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1180b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 11814023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1182b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1183f5be3e29SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1184b37199e6SAl Viro goto failed; 1185b37199e6SAl Viro } 11864023bfc9SAl Viro nd->inode = inode; 118731e6b01fSNick Piggin return 0; 1188ef7562d5SAl Viro 1189ef7562d5SAl Viro failed: 1190ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 11915b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1192ef7562d5SAl Viro nd->root.mnt = NULL; 11938b61e74fSAl Viro rcu_read_unlock(); 1194ef7562d5SAl Viro return -ECHILD; 119531e6b01fSNick Piggin } 119631e6b01fSNick Piggin 11979875cf80SDavid Howells /* 1198cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1199cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1200cc53ce53SDavid Howells * caller is permitted to proceed or not. 1201cc53ce53SDavid Howells */ 12027cc90cc3SAl Viro int follow_down(struct path *path) 1203cc53ce53SDavid Howells { 1204cc53ce53SDavid Howells unsigned managed; 1205cc53ce53SDavid Howells int ret; 1206cc53ce53SDavid Howells 1207cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1208cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1209cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1210cc53ce53SDavid Howells * being held. 1211cc53ce53SDavid Howells * 1212cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1213cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1214cc53ce53SDavid Howells * other than its daemon the right to mount on its 1215cc53ce53SDavid Howells * superstructure. 1216cc53ce53SDavid Howells * 1217cc53ce53SDavid Howells * The filesystem may sleep at this point. 1218cc53ce53SDavid Howells */ 1219cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1220cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1221cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1222ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 12231aed3e42SAl Viro path->dentry, false); 1224cc53ce53SDavid Howells if (ret < 0) 1225cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1226cc53ce53SDavid Howells } 1227cc53ce53SDavid Howells 1228cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1229cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1230cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1231cc53ce53SDavid Howells if (!mounted) 1232cc53ce53SDavid Howells break; 1233cc53ce53SDavid Howells dput(path->dentry); 1234cc53ce53SDavid Howells mntput(path->mnt); 1235cc53ce53SDavid Howells path->mnt = mounted; 1236cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1237cc53ce53SDavid Howells continue; 1238cc53ce53SDavid Howells } 1239cc53ce53SDavid Howells 1240cc53ce53SDavid Howells /* Don't handle automount points here */ 1241cc53ce53SDavid Howells break; 1242cc53ce53SDavid Howells } 1243cc53ce53SDavid Howells return 0; 1244cc53ce53SDavid Howells } 12454d359507SAl Viro EXPORT_SYMBOL(follow_down); 1246cc53ce53SDavid Howells 1247cc53ce53SDavid Howells /* 12489875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12499875cf80SDavid Howells */ 12509875cf80SDavid Howells static void follow_mount(struct path *path) 12519875cf80SDavid Howells { 12529875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12539875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12549875cf80SDavid Howells if (!mounted) 12559875cf80SDavid Howells break; 12569875cf80SDavid Howells dput(path->dentry); 12579875cf80SDavid Howells mntput(path->mnt); 12589875cf80SDavid Howells path->mnt = mounted; 12599875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12609875cf80SDavid Howells } 12619875cf80SDavid Howells } 12629875cf80SDavid Howells 126331e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 12641da177e4SLinus Torvalds { 12657bd88377SAl Viro if (!nd->root.mnt) 12662a737871SAl Viro set_root(nd); 1267e518ddb7SAndreas Mohr 12681da177e4SLinus Torvalds while(1) { 12694ac91378SJan Blunck struct dentry *old = nd->path.dentry; 12701da177e4SLinus Torvalds 12712a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 12722a737871SAl Viro nd->path.mnt == nd->root.mnt) { 12731da177e4SLinus Torvalds break; 12741da177e4SLinus Torvalds } 12754ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 12763088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 12773088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 12781da177e4SLinus Torvalds dput(old); 12791da177e4SLinus Torvalds break; 12801da177e4SLinus Torvalds } 12813088dd70SAl Viro if (!follow_up(&nd->path)) 12821da177e4SLinus Torvalds break; 12831da177e4SLinus Torvalds } 128479ed0226SAl Viro follow_mount(&nd->path); 128531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 12861da177e4SLinus Torvalds } 12871da177e4SLinus Torvalds 12881da177e4SLinus Torvalds /* 1289bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1290bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1291bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1292bad61189SMiklos Szeredi * 1293bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1294baa03890SNick Piggin */ 1295bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1296201f956eSAl Viro unsigned int flags, bool *need_lookup) 1297baa03890SNick Piggin { 1298baa03890SNick Piggin struct dentry *dentry; 1299bad61189SMiklos Szeredi int error; 1300baa03890SNick Piggin 1301bad61189SMiklos Szeredi *need_lookup = false; 1302bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1303bad61189SMiklos Szeredi if (dentry) { 130439e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1305201f956eSAl Viro error = d_revalidate(dentry, flags); 1306bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1307bad61189SMiklos Szeredi if (error < 0) { 1308bad61189SMiklos Szeredi dput(dentry); 1309bad61189SMiklos Szeredi return ERR_PTR(error); 13105542aa2fSEric W. Biederman } else { 13115542aa2fSEric W. Biederman d_invalidate(dentry); 1312bad61189SMiklos Szeredi dput(dentry); 1313bad61189SMiklos Szeredi dentry = NULL; 1314bad61189SMiklos Szeredi } 1315bad61189SMiklos Szeredi } 1316bad61189SMiklos Szeredi } 1317bad61189SMiklos Szeredi } 1318baa03890SNick Piggin 1319bad61189SMiklos Szeredi if (!dentry) { 1320bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1321baa03890SNick Piggin if (unlikely(!dentry)) 1322baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1323baa03890SNick Piggin 1324bad61189SMiklos Szeredi *need_lookup = true; 1325baa03890SNick Piggin } 1326baa03890SNick Piggin return dentry; 1327baa03890SNick Piggin } 1328baa03890SNick Piggin 1329baa03890SNick Piggin /* 133013a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 133113a2c3beSJ. Bruce Fields * unhashed. 1332bad61189SMiklos Szeredi * 1333bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 133444396f4bSJosef Bacik */ 1335bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 133672bd866aSAl Viro unsigned int flags) 133744396f4bSJosef Bacik { 133844396f4bSJosef Bacik struct dentry *old; 133944396f4bSJosef Bacik 134044396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1341bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1342e188dc02SMiklos Szeredi dput(dentry); 134344396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1344e188dc02SMiklos Szeredi } 134544396f4bSJosef Bacik 134672bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 134744396f4bSJosef Bacik if (unlikely(old)) { 134844396f4bSJosef Bacik dput(dentry); 134944396f4bSJosef Bacik dentry = old; 135044396f4bSJosef Bacik } 135144396f4bSJosef Bacik return dentry; 135244396f4bSJosef Bacik } 135344396f4bSJosef Bacik 1354a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 135572bd866aSAl Viro struct dentry *base, unsigned int flags) 1356a3255546SAl Viro { 1357bad61189SMiklos Szeredi bool need_lookup; 1358a3255546SAl Viro struct dentry *dentry; 1359a3255546SAl Viro 136072bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1361bad61189SMiklos Szeredi if (!need_lookup) 1362a3255546SAl Viro return dentry; 1363bad61189SMiklos Szeredi 136472bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1365a3255546SAl Viro } 1366a3255546SAl Viro 136744396f4bSJosef Bacik /* 13681da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 13691da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 13701da177e4SLinus Torvalds * It _is_ time-critical. 13711da177e4SLinus Torvalds */ 1372e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 137331e6b01fSNick Piggin struct path *path, struct inode **inode) 13741da177e4SLinus Torvalds { 13754ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 137631e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 13775a18fff2SAl Viro int need_reval = 1; 13785a18fff2SAl Viro int status = 1; 13799875cf80SDavid Howells int err; 13809875cf80SDavid Howells 13813cac260aSAl Viro /* 1382b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1383b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1384b04f784eSNick Piggin * do the non-racy lookup, below. 1385b04f784eSNick Piggin */ 138631e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 138731e6b01fSNick Piggin unsigned seq; 1388da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 13895a18fff2SAl Viro if (!dentry) 13905a18fff2SAl Viro goto unlazy; 13915a18fff2SAl Viro 139212f8ad4bSLinus Torvalds /* 139312f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 139412f8ad4bSLinus Torvalds * the dentry name information from lookup. 139512f8ad4bSLinus Torvalds */ 139612f8ad4bSLinus Torvalds *inode = dentry->d_inode; 139712f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 139812f8ad4bSLinus Torvalds return -ECHILD; 139912f8ad4bSLinus Torvalds 140012f8ad4bSLinus Torvalds /* 140112f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 140212f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 140312f8ad4bSLinus Torvalds * 140412f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 140512f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 140612f8ad4bSLinus Torvalds */ 140731e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 140831e6b01fSNick Piggin return -ECHILD; 140931e6b01fSNick Piggin nd->seq = seq; 14105a18fff2SAl Viro 141124643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 14124ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14135a18fff2SAl Viro if (unlikely(status <= 0)) { 14145a18fff2SAl Viro if (status != -ECHILD) 14155a18fff2SAl Viro need_reval = 0; 14165a18fff2SAl Viro goto unlazy; 14175a18fff2SAl Viro } 141824643087SAl Viro } 141931e6b01fSNick Piggin path->mnt = mnt; 142031e6b01fSNick Piggin path->dentry = dentry; 1421b8faf035SNeilBrown if (likely(__follow_mount_rcu(nd, path, inode))) 14229875cf80SDavid Howells return 0; 14235a18fff2SAl Viro unlazy: 142419660af7SAl Viro if (unlazy_walk(nd, dentry)) 14255a18fff2SAl Viro return -ECHILD; 14265a18fff2SAl Viro } else { 1427e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 142824643087SAl Viro } 14295a18fff2SAl Viro 143081e6f520SAl Viro if (unlikely(!dentry)) 143181e6f520SAl Viro goto need_lookup; 14325a18fff2SAl Viro 14335a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14344ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14355a18fff2SAl Viro if (unlikely(status <= 0)) { 14365a18fff2SAl Viro if (status < 0) { 14375a18fff2SAl Viro dput(dentry); 14385a18fff2SAl Viro return status; 14395a18fff2SAl Viro } 14405542aa2fSEric W. Biederman d_invalidate(dentry); 14415a18fff2SAl Viro dput(dentry); 144281e6f520SAl Viro goto need_lookup; 14435a18fff2SAl Viro } 1444697f514dSMiklos Szeredi 14451da177e4SLinus Torvalds path->mnt = mnt; 14461da177e4SLinus Torvalds path->dentry = dentry; 14479875cf80SDavid Howells err = follow_managed(path, nd->flags); 144889312214SIan Kent if (unlikely(err < 0)) { 144989312214SIan Kent path_put_conditional(path, nd); 14509875cf80SDavid Howells return err; 145189312214SIan Kent } 1452a3fbbde7SAl Viro if (err) 1453a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 145431e6b01fSNick Piggin *inode = path->dentry->d_inode; 14551da177e4SLinus Torvalds return 0; 145681e6f520SAl Viro 145781e6f520SAl Viro need_lookup: 1458697f514dSMiklos Szeredi return 1; 1459697f514dSMiklos Szeredi } 1460697f514dSMiklos Szeredi 1461697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1462cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1463697f514dSMiklos Szeredi { 1464697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1465697f514dSMiklos Szeredi int err; 1466697f514dSMiklos Szeredi 1467697f514dSMiklos Szeredi parent = nd->path.dentry; 146881e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 146981e6f520SAl Viro 147081e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1471cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 147281e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 147381e6f520SAl Viro if (IS_ERR(dentry)) 147481e6f520SAl Viro return PTR_ERR(dentry); 1475697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1476697f514dSMiklos Szeredi path->dentry = dentry; 1477697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1478697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1479697f514dSMiklos Szeredi path_put_conditional(path, nd); 1480697f514dSMiklos Szeredi return err; 1481697f514dSMiklos Szeredi } 1482697f514dSMiklos Szeredi if (err) 1483697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1484697f514dSMiklos Szeredi return 0; 14851da177e4SLinus Torvalds } 14861da177e4SLinus Torvalds 148752094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 148852094c8aSAl Viro { 148952094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 14904ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 149152094c8aSAl Viro if (err != -ECHILD) 149252094c8aSAl Viro return err; 149319660af7SAl Viro if (unlazy_walk(nd, NULL)) 149452094c8aSAl Viro return -ECHILD; 149552094c8aSAl Viro } 14964ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 149752094c8aSAl Viro } 149852094c8aSAl Viro 14999856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 15009856fa1bSAl Viro { 15019856fa1bSAl Viro if (type == LAST_DOTDOT) { 15029856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 15039856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 15049856fa1bSAl Viro return -ECHILD; 15059856fa1bSAl Viro } else 15069856fa1bSAl Viro follow_dotdot(nd); 15079856fa1bSAl Viro } 15089856fa1bSAl Viro return 0; 15099856fa1bSAl Viro } 15109856fa1bSAl Viro 1511951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1512951361f9SAl Viro { 1513951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1514951361f9SAl Viro path_put(&nd->path); 1515951361f9SAl Viro } else { 1516951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 15175b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1518951361f9SAl Viro nd->root.mnt = NULL; 15198b61e74fSAl Viro rcu_read_unlock(); 1520951361f9SAl Viro } 1521951361f9SAl Viro } 1522951361f9SAl Viro 15233ddcd056SLinus Torvalds /* 15243ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15253ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15263ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15273ddcd056SLinus Torvalds * for the common case. 15283ddcd056SLinus Torvalds */ 1529b18825a7SDavid Howells static inline int should_follow_link(struct dentry *dentry, int follow) 15303ddcd056SLinus Torvalds { 1531b18825a7SDavid Howells return unlikely(d_is_symlink(dentry)) ? follow : 0; 15323ddcd056SLinus Torvalds } 15333ddcd056SLinus Torvalds 1534ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 153521b9b073SAl Viro int follow) 1536ce57dfc1SAl Viro { 1537ce57dfc1SAl Viro struct inode *inode; 1538ce57dfc1SAl Viro int err; 1539ce57dfc1SAl Viro /* 1540ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1541ce57dfc1SAl Viro * to be able to know about the current root directory and 1542ce57dfc1SAl Viro * parent relationships. 1543ce57dfc1SAl Viro */ 154421b9b073SAl Viro if (unlikely(nd->last_type != LAST_NORM)) 154521b9b073SAl Viro return handle_dots(nd, nd->last_type); 1546e97cdc87SAl Viro err = lookup_fast(nd, path, &inode); 1547ce57dfc1SAl Viro if (unlikely(err)) { 1548697f514dSMiklos Szeredi if (err < 0) 1549697f514dSMiklos Szeredi goto out_err; 1550697f514dSMiklos Szeredi 1551cc2a5271SAl Viro err = lookup_slow(nd, path); 1552697f514dSMiklos Szeredi if (err < 0) 1553697f514dSMiklos Szeredi goto out_err; 1554697f514dSMiklos Szeredi 1555697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1556ce57dfc1SAl Viro } 1557697f514dSMiklos Szeredi err = -ENOENT; 155822213318SAl Viro if (!inode || d_is_negative(path->dentry)) 1559697f514dSMiklos Szeredi goto out_path_put; 1560697f514dSMiklos Szeredi 1561b18825a7SDavid Howells if (should_follow_link(path->dentry, follow)) { 156219660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 156319660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1564697f514dSMiklos Szeredi err = -ECHILD; 1565697f514dSMiklos Szeredi goto out_err; 156619660af7SAl Viro } 156719660af7SAl Viro } 1568ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1569ce57dfc1SAl Viro return 1; 1570ce57dfc1SAl Viro } 1571ce57dfc1SAl Viro path_to_nameidata(path, nd); 1572ce57dfc1SAl Viro nd->inode = inode; 1573ce57dfc1SAl Viro return 0; 1574697f514dSMiklos Szeredi 1575697f514dSMiklos Szeredi out_path_put: 1576697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1577697f514dSMiklos Szeredi out_err: 1578697f514dSMiklos Szeredi terminate_walk(nd); 1579697f514dSMiklos Szeredi return err; 1580ce57dfc1SAl Viro } 1581ce57dfc1SAl Viro 15821da177e4SLinus Torvalds /* 1583b356379aSAl Viro * This limits recursive symlink follows to 8, while 1584b356379aSAl Viro * limiting consecutive symlinks to 40. 1585b356379aSAl Viro * 1586b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1587b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1588b356379aSAl Viro */ 1589b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1590b356379aSAl Viro { 1591b356379aSAl Viro int res; 1592b356379aSAl Viro 1593b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1594b356379aSAl Viro path_put_conditional(path, nd); 1595b356379aSAl Viro path_put(&nd->path); 1596b356379aSAl Viro return -ELOOP; 1597b356379aSAl Viro } 15981a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1599b356379aSAl Viro 1600b356379aSAl Viro nd->depth++; 1601b356379aSAl Viro current->link_count++; 1602b356379aSAl Viro 1603b356379aSAl Viro do { 1604b356379aSAl Viro struct path link = *path; 1605b356379aSAl Viro void *cookie; 1606574197e0SAl Viro 1607574197e0SAl Viro res = follow_link(&link, nd, &cookie); 16086d7b5aaeSAl Viro if (res) 16096d7b5aaeSAl Viro break; 161021b9b073SAl Viro res = walk_component(nd, path, LOOKUP_FOLLOW); 1611574197e0SAl Viro put_link(nd, &link, cookie); 1612b356379aSAl Viro } while (res > 0); 1613b356379aSAl Viro 1614b356379aSAl Viro current->link_count--; 1615b356379aSAl Viro nd->depth--; 1616b356379aSAl Viro return res; 1617b356379aSAl Viro } 1618b356379aSAl Viro 1619b356379aSAl Viro /* 1620bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1621bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1622bfcfaa77SLinus Torvalds * 1623bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1624bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1625bfcfaa77SLinus Torvalds * fast. 1626bfcfaa77SLinus Torvalds * 1627bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1628bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1629bfcfaa77SLinus Torvalds * crossing operation. 1630bfcfaa77SLinus Torvalds * 1631bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1632bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1633bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1634bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1635bfcfaa77SLinus Torvalds */ 1636bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1637bfcfaa77SLinus Torvalds 1638f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1639bfcfaa77SLinus Torvalds 1640f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1641bfcfaa77SLinus Torvalds 1642bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1643bfcfaa77SLinus Torvalds { 164499d263d4SLinus Torvalds return hash_64(hash, 32); 1645bfcfaa77SLinus Torvalds } 1646bfcfaa77SLinus Torvalds 1647bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1648bfcfaa77SLinus Torvalds 1649bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1650bfcfaa77SLinus Torvalds 1651bfcfaa77SLinus Torvalds #endif 1652bfcfaa77SLinus Torvalds 1653bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1654bfcfaa77SLinus Torvalds { 1655bfcfaa77SLinus Torvalds unsigned long a, mask; 1656bfcfaa77SLinus Torvalds unsigned long hash = 0; 1657bfcfaa77SLinus Torvalds 1658bfcfaa77SLinus Torvalds for (;;) { 1659e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1660bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1661bfcfaa77SLinus Torvalds break; 1662bfcfaa77SLinus Torvalds hash += a; 1663f132c5beSAl Viro hash *= 9; 1664bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1665bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1666bfcfaa77SLinus Torvalds if (!len) 1667bfcfaa77SLinus Torvalds goto done; 1668bfcfaa77SLinus Torvalds } 1669a5c21dceSWill Deacon mask = bytemask_from_count(len); 1670bfcfaa77SLinus Torvalds hash += mask & a; 1671bfcfaa77SLinus Torvalds done: 1672bfcfaa77SLinus Torvalds return fold_hash(hash); 1673bfcfaa77SLinus Torvalds } 1674bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1675bfcfaa77SLinus Torvalds 1676bfcfaa77SLinus Torvalds /* 1677bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1678d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1679bfcfaa77SLinus Torvalds */ 1680d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1681bfcfaa77SLinus Torvalds { 168236126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 168336126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1684bfcfaa77SLinus Torvalds 1685bfcfaa77SLinus Torvalds hash = a = 0; 1686bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1687bfcfaa77SLinus Torvalds do { 1688bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1689bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1690e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 169136126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 169236126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1693bfcfaa77SLinus Torvalds 169436126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 169536126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 169636126f8fSLinus Torvalds 169736126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 169836126f8fSLinus Torvalds 169936126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 17009226b5b4SLinus Torvalds len += find_zero(mask); 1701d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1702bfcfaa77SLinus Torvalds } 1703bfcfaa77SLinus Torvalds 1704bfcfaa77SLinus Torvalds #else 1705bfcfaa77SLinus Torvalds 17060145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 17070145acc2SLinus Torvalds { 17080145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 17090145acc2SLinus Torvalds while (len--) 17100145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 17110145acc2SLinus Torvalds return end_name_hash(hash); 17120145acc2SLinus Torvalds } 1713ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 17140145acc2SLinus Torvalds 17153ddcd056SLinus Torvalds /* 1716200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1717200e9ef7SLinus Torvalds * one character. 1718200e9ef7SLinus Torvalds */ 1719d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1720200e9ef7SLinus Torvalds { 1721200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1722200e9ef7SLinus Torvalds unsigned long len = 0, c; 1723200e9ef7SLinus Torvalds 1724200e9ef7SLinus Torvalds c = (unsigned char)*name; 1725200e9ef7SLinus Torvalds do { 1726200e9ef7SLinus Torvalds len++; 1727200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1728200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1729200e9ef7SLinus Torvalds } while (c && c != '/'); 1730d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1731200e9ef7SLinus Torvalds } 1732200e9ef7SLinus Torvalds 1733bfcfaa77SLinus Torvalds #endif 1734bfcfaa77SLinus Torvalds 1735200e9ef7SLinus Torvalds /* 17361da177e4SLinus Torvalds * Name resolution. 1737ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1738ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17391da177e4SLinus Torvalds * 1740ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1741ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17421da177e4SLinus Torvalds */ 17436de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17441da177e4SLinus Torvalds { 17451da177e4SLinus Torvalds struct path next; 17461da177e4SLinus Torvalds int err; 17471da177e4SLinus Torvalds 17481da177e4SLinus Torvalds while (*name=='/') 17491da177e4SLinus Torvalds name++; 17501da177e4SLinus Torvalds if (!*name) 1751086e183aSAl Viro return 0; 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17541da177e4SLinus Torvalds for(;;) { 1755d6bb3e90SLinus Torvalds u64 hash_len; 1756fe479a58SAl Viro int type; 17571da177e4SLinus Torvalds 175852094c8aSAl Viro err = may_lookup(nd); 17591da177e4SLinus Torvalds if (err) 17601da177e4SLinus Torvalds break; 17611da177e4SLinus Torvalds 1762d6bb3e90SLinus Torvalds hash_len = hash_name(name); 17631da177e4SLinus Torvalds 1764fe479a58SAl Viro type = LAST_NORM; 1765d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1766fe479a58SAl Viro case 2: 1767200e9ef7SLinus Torvalds if (name[1] == '.') { 1768fe479a58SAl Viro type = LAST_DOTDOT; 176916c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 177016c2cd71SAl Viro } 1771fe479a58SAl Viro break; 1772fe479a58SAl Viro case 1: 1773fe479a58SAl Viro type = LAST_DOT; 1774fe479a58SAl Viro } 17755a202bcdSAl Viro if (likely(type == LAST_NORM)) { 17765a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 177716c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 17785a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1779a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1780da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 17815a202bcdSAl Viro if (err < 0) 17825a202bcdSAl Viro break; 1783d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1784d6bb3e90SLinus Torvalds name = this.name; 17855a202bcdSAl Viro } 17865a202bcdSAl Viro } 1787fe479a58SAl Viro 1788d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1789d6bb3e90SLinus Torvalds nd->last.name = name; 17905f4a6a69SAl Viro nd->last_type = type; 17915f4a6a69SAl Viro 1792d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1793d6bb3e90SLinus Torvalds if (!*name) 17945f4a6a69SAl Viro return 0; 1795200e9ef7SLinus Torvalds /* 1796200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1797200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1798200e9ef7SLinus Torvalds */ 1799200e9ef7SLinus Torvalds do { 1800d6bb3e90SLinus Torvalds name++; 1801d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 1802d6bb3e90SLinus Torvalds if (!*name) 18035f4a6a69SAl Viro return 0; 18045f4a6a69SAl Viro 180521b9b073SAl Viro err = walk_component(nd, &next, LOOKUP_FOLLOW); 1806ce57dfc1SAl Viro if (err < 0) 1807ce57dfc1SAl Viro return err; 1808fe479a58SAl Viro 1809ce57dfc1SAl Viro if (err) { 1810b356379aSAl Viro err = nested_symlink(&next, nd); 18111da177e4SLinus Torvalds if (err) 1812a7472babSAl Viro return err; 181331e6b01fSNick Piggin } 181444b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 18153ddcd056SLinus Torvalds err = -ENOTDIR; 18163ddcd056SLinus Torvalds break; 18175f4a6a69SAl Viro } 1818ce57dfc1SAl Viro } 1819951361f9SAl Viro terminate_walk(nd); 18201da177e4SLinus Torvalds return err; 18211da177e4SLinus Torvalds } 18221da177e4SLinus Torvalds 182370e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 182470e9b357SAl Viro struct nameidata *nd, struct file **fp) 182531e6b01fSNick Piggin { 182631e6b01fSNick Piggin int retval = 0; 182731e6b01fSNick Piggin 182831e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 182916c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 183031e6b01fSNick Piggin nd->depth = 0; 18315b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1832b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1833b18825a7SDavid Howells struct inode *inode = root->d_inode; 183473d049a4SAl Viro if (*name) { 183544b1d530SMiklos Szeredi if (!d_can_lookup(root)) 18365b6ca027SAl Viro return -ENOTDIR; 18375b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18385b6ca027SAl Viro if (retval) 18395b6ca027SAl Viro return retval; 184073d049a4SAl Viro } 18415b6ca027SAl Viro nd->path = nd->root; 18425b6ca027SAl Viro nd->inode = inode; 18435b6ca027SAl Viro if (flags & LOOKUP_RCU) { 18448b61e74fSAl Viro rcu_read_lock(); 18455b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 184648a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 18475b6ca027SAl Viro } else { 18485b6ca027SAl Viro path_get(&nd->path); 18495b6ca027SAl Viro } 18505b6ca027SAl Viro return 0; 18515b6ca027SAl Viro } 18525b6ca027SAl Viro 185331e6b01fSNick Piggin nd->root.mnt = NULL; 185431e6b01fSNick Piggin 185548a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 185631e6b01fSNick Piggin if (*name=='/') { 1857e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18588b61e74fSAl Viro rcu_read_lock(); 18597bd88377SAl Viro nd->seq = set_root_rcu(nd); 1860e41f7d4eSAl Viro } else { 1861e41f7d4eSAl Viro set_root(nd); 1862e41f7d4eSAl Viro path_get(&nd->root); 1863e41f7d4eSAl Viro } 186431e6b01fSNick Piggin nd->path = nd->root; 186531e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1866e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 186731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1868c28cc364SNick Piggin unsigned seq; 186931e6b01fSNick Piggin 18708b61e74fSAl Viro rcu_read_lock(); 187131e6b01fSNick Piggin 1872c28cc364SNick Piggin do { 1873c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 187431e6b01fSNick Piggin nd->path = fs->pwd; 1875c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1876c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1877e41f7d4eSAl Viro } else { 1878e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1879e41f7d4eSAl Viro } 188031e6b01fSNick Piggin } else { 1881582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 18822903ff01SAl Viro struct fd f = fdget_raw(dfd); 188331e6b01fSNick Piggin struct dentry *dentry; 188431e6b01fSNick Piggin 18852903ff01SAl Viro if (!f.file) 18862903ff01SAl Viro return -EBADF; 188731e6b01fSNick Piggin 18882903ff01SAl Viro dentry = f.file->f_path.dentry; 188931e6b01fSNick Piggin 1890f52e0c11SAl Viro if (*name) { 189144b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 18922903ff01SAl Viro fdput(f); 18932903ff01SAl Viro return -ENOTDIR; 1894f52e0c11SAl Viro } 18952903ff01SAl Viro } 18962903ff01SAl Viro 18972903ff01SAl Viro nd->path = f.file->f_path; 1898e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18999c225f26SLinus Torvalds if (f.flags & FDPUT_FPUT) 19002903ff01SAl Viro *fp = f.file; 1901c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 19028b61e74fSAl Viro rcu_read_lock(); 19035590ff0dSUlrich Drepper } else { 19042903ff01SAl Viro path_get(&nd->path); 19052903ff01SAl Viro fdput(f); 19061da177e4SLinus Torvalds } 1907e41f7d4eSAl Viro } 1908e41f7d4eSAl Viro 190931e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 19104023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 19119b4a9b14SAl Viro return 0; 19124023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 19134023bfc9SAl Viro return 0; 19144023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 19154023bfc9SAl Viro nd->root.mnt = NULL; 19164023bfc9SAl Viro rcu_read_unlock(); 19174023bfc9SAl Viro return -ECHILD; 19189b4a9b14SAl Viro } 19199b4a9b14SAl Viro 1920bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1921bd92d7feSAl Viro { 1922bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1923bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1924bd92d7feSAl Viro 1925bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 192621b9b073SAl Viro return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); 1927bd92d7feSAl Viro } 1928bd92d7feSAl Viro 19299b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1930ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 19319b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 19329b4a9b14SAl Viro { 193370e9b357SAl Viro struct file *base = NULL; 1934bd92d7feSAl Viro struct path path; 1935bd92d7feSAl Viro int err; 193631e6b01fSNick Piggin 193731e6b01fSNick Piggin /* 193831e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 193931e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 194031e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 194131e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 194231e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 194331e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 194431e6b01fSNick Piggin * analogue, foo_rcu(). 194531e6b01fSNick Piggin * 194631e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 194731e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 194831e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 194931e6b01fSNick Piggin * be able to complete). 195031e6b01fSNick Piggin */ 1951bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1952ee0827cdSAl Viro 1953bd92d7feSAl Viro if (unlikely(err)) 1954115cbfdcSAl Viro goto out; 1955ee0827cdSAl Viro 1956ee0827cdSAl Viro current->total_link_count = 0; 1957bd92d7feSAl Viro err = link_path_walk(name, nd); 1958bd92d7feSAl Viro 1959bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1960bd92d7feSAl Viro err = lookup_last(nd, &path); 1961bd92d7feSAl Viro while (err > 0) { 1962bd92d7feSAl Viro void *cookie; 1963bd92d7feSAl Viro struct path link = path; 1964800179c9SKees Cook err = may_follow_link(&link, nd); 1965800179c9SKees Cook if (unlikely(err)) 1966800179c9SKees Cook break; 1967bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1968574197e0SAl Viro err = follow_link(&link, nd, &cookie); 19696d7b5aaeSAl Viro if (err) 19706d7b5aaeSAl Viro break; 1971bd92d7feSAl Viro err = lookup_last(nd, &path); 1972574197e0SAl Viro put_link(nd, &link, cookie); 1973bd92d7feSAl Viro } 1974bd92d7feSAl Viro } 1975ee0827cdSAl Viro 19769f1fafeeSAl Viro if (!err) 19779f1fafeeSAl Viro err = complete_walk(nd); 1978bd92d7feSAl Viro 1979bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 198044b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 1981bd92d7feSAl Viro path_put(&nd->path); 1982bd23a539SAl Viro err = -ENOTDIR; 1983bd92d7feSAl Viro } 1984bd92d7feSAl Viro } 198516c2cd71SAl Viro 1986115cbfdcSAl Viro out: 198770e9b357SAl Viro if (base) 198870e9b357SAl Viro fput(base); 1989ee0827cdSAl Viro 19905b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 199131e6b01fSNick Piggin path_put(&nd->root); 199231e6b01fSNick Piggin nd->root.mnt = NULL; 199331e6b01fSNick Piggin } 1994bd92d7feSAl Viro return err; 199531e6b01fSNick Piggin } 199631e6b01fSNick Piggin 1997873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 1998873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 1999873f1eedSJeff Layton { 2000873f1eedSJeff Layton int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); 2001873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2002873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, flags, nd); 2003873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2004873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, 2005873f1eedSJeff Layton flags | LOOKUP_REVAL, nd); 2006873f1eedSJeff Layton 2007873f1eedSJeff Layton if (likely(!retval)) 2008adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2009873f1eedSJeff Layton return retval; 2010873f1eedSJeff Layton } 2011873f1eedSJeff Layton 2012ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 2013ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 2014ee0827cdSAl Viro { 2015873f1eedSJeff Layton struct filename filename = { .name = name }; 2016ee0827cdSAl Viro 2017873f1eedSJeff Layton return filename_lookup(dfd, &filename, flags, nd); 20181da177e4SLinus Torvalds } 20191da177e4SLinus Torvalds 202079714f72SAl Viro /* does lookup, returns the object with parent locked */ 202179714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 20225590ff0dSUlrich Drepper { 202379714f72SAl Viro struct nameidata nd; 202479714f72SAl Viro struct dentry *d; 202579714f72SAl Viro int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 202679714f72SAl Viro if (err) 202779714f72SAl Viro return ERR_PTR(err); 202879714f72SAl Viro if (nd.last_type != LAST_NORM) { 202979714f72SAl Viro path_put(&nd.path); 203079714f72SAl Viro return ERR_PTR(-EINVAL); 203179714f72SAl Viro } 203279714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 20331e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 203479714f72SAl Viro if (IS_ERR(d)) { 203579714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 203679714f72SAl Viro path_put(&nd.path); 203779714f72SAl Viro return d; 203879714f72SAl Viro } 203979714f72SAl Viro *path = nd.path; 204079714f72SAl Viro return d; 20415590ff0dSUlrich Drepper } 20425590ff0dSUlrich Drepper 2043d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2044d1811465SAl Viro { 2045d1811465SAl Viro struct nameidata nd; 2046d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 2047d1811465SAl Viro if (!res) 2048d1811465SAl Viro *path = nd.path; 2049d1811465SAl Viro return res; 2050d1811465SAl Viro } 20514d359507SAl Viro EXPORT_SYMBOL(kern_path); 2052d1811465SAl Viro 205316f18200SJosef 'Jeff' Sipek /** 205416f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 205516f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 205616f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 205716f18200SJosef 'Jeff' Sipek * @name: pointer to file name 205816f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2059e0a01249SAl Viro * @path: pointer to struct path to fill 206016f18200SJosef 'Jeff' Sipek */ 206116f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 206216f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2063e0a01249SAl Viro struct path *path) 206416f18200SJosef 'Jeff' Sipek { 2065e0a01249SAl Viro struct nameidata nd; 2066e0a01249SAl Viro int err; 2067e0a01249SAl Viro nd.root.dentry = dentry; 2068e0a01249SAl Viro nd.root.mnt = mnt; 2069e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 20705b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 2071e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 2072e0a01249SAl Viro if (!err) 2073e0a01249SAl Viro *path = nd.path; 2074e0a01249SAl Viro return err; 207516f18200SJosef 'Jeff' Sipek } 20764d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 207716f18200SJosef 'Jeff' Sipek 2078057f6c01SJames Morris /* 2079057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 2080057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 2081057f6c01SJames Morris * SMP-safe. 2082057f6c01SJames Morris */ 2083a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 20841da177e4SLinus Torvalds { 208572bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 20861da177e4SLinus Torvalds } 20871da177e4SLinus Torvalds 2088eead1911SChristoph Hellwig /** 2089a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2090eead1911SChristoph Hellwig * @name: pathname component to lookup 2091eead1911SChristoph Hellwig * @base: base directory to lookup from 2092eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2093eead1911SChristoph Hellwig * 2094a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 2095a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 2096eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 2097eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 2098eead1911SChristoph Hellwig */ 2099057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2100057f6c01SJames Morris { 2101057f6c01SJames Morris struct qstr this; 21026a96ba54SAl Viro unsigned int c; 2103cda309deSMiklos Szeredi int err; 2104057f6c01SJames Morris 21052f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 21062f9092e1SDavid Woodhouse 21076a96ba54SAl Viro this.name = name; 21086a96ba54SAl Viro this.len = len; 21090145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 21106a96ba54SAl Viro if (!len) 21116a96ba54SAl Viro return ERR_PTR(-EACCES); 21126a96ba54SAl Viro 211321d8a15aSAl Viro if (unlikely(name[0] == '.')) { 211421d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 211521d8a15aSAl Viro return ERR_PTR(-EACCES); 211621d8a15aSAl Viro } 211721d8a15aSAl Viro 21186a96ba54SAl Viro while (len--) { 21196a96ba54SAl Viro c = *(const unsigned char *)name++; 21206a96ba54SAl Viro if (c == '/' || c == '\0') 21216a96ba54SAl Viro return ERR_PTR(-EACCES); 21226a96ba54SAl Viro } 21235a202bcdSAl Viro /* 21245a202bcdSAl Viro * See if the low-level filesystem might want 21255a202bcdSAl Viro * to use its own hash.. 21265a202bcdSAl Viro */ 21275a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2128da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 21295a202bcdSAl Viro if (err < 0) 21305a202bcdSAl Viro return ERR_PTR(err); 21315a202bcdSAl Viro } 2132eead1911SChristoph Hellwig 2133cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2134cda309deSMiklos Szeredi if (err) 2135cda309deSMiklos Szeredi return ERR_PTR(err); 2136cda309deSMiklos Szeredi 213772bd866aSAl Viro return __lookup_hash(&this, base, 0); 2138057f6c01SJames Morris } 21394d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2140057f6c01SJames Morris 21411fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 21421fa1e7f6SAndy Whitcroft struct path *path, int *empty) 21431da177e4SLinus Torvalds { 21442d8f3038SAl Viro struct nameidata nd; 214591a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 21461da177e4SLinus Torvalds int err = PTR_ERR(tmp); 21471da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 21482d8f3038SAl Viro 21492d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21502d8f3038SAl Viro 2151873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 21521da177e4SLinus Torvalds putname(tmp); 21532d8f3038SAl Viro if (!err) 21542d8f3038SAl Viro *path = nd.path; 21551da177e4SLinus Torvalds } 21561da177e4SLinus Torvalds return err; 21571da177e4SLinus Torvalds } 21581da177e4SLinus Torvalds 21591fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 21601fa1e7f6SAndy Whitcroft struct path *path) 21611fa1e7f6SAndy Whitcroft { 2162f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 21631fa1e7f6SAndy Whitcroft } 21644d359507SAl Viro EXPORT_SYMBOL(user_path_at); 21651fa1e7f6SAndy Whitcroft 2166873f1eedSJeff Layton /* 2167873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2168873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2169873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2170873f1eedSJeff Layton * path-walking is complete. 2171873f1eedSJeff Layton */ 217291a27b2aSJeff Layton static struct filename * 21739e790bd6SJeff Layton user_path_parent(int dfd, const char __user *path, struct nameidata *nd, 21749e790bd6SJeff Layton unsigned int flags) 21752ad94ae6SAl Viro { 217691a27b2aSJeff Layton struct filename *s = getname(path); 21772ad94ae6SAl Viro int error; 21782ad94ae6SAl Viro 21799e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 21809e790bd6SJeff Layton flags &= LOOKUP_REVAL; 21819e790bd6SJeff Layton 21822ad94ae6SAl Viro if (IS_ERR(s)) 218391a27b2aSJeff Layton return s; 21842ad94ae6SAl Viro 21859e790bd6SJeff Layton error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); 218691a27b2aSJeff Layton if (error) { 21872ad94ae6SAl Viro putname(s); 218891a27b2aSJeff Layton return ERR_PTR(error); 218991a27b2aSJeff Layton } 21902ad94ae6SAl Viro 219191a27b2aSJeff Layton return s; 21922ad94ae6SAl Viro } 21932ad94ae6SAl Viro 21948033426eSJeff Layton /** 2195197df04cSAl Viro * mountpoint_last - look up last component for umount 21968033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 21978033426eSJeff Layton * @path: pointer to container for result 21988033426eSJeff Layton * 21998033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 22008033426eSJeff Layton * need to resolve the path without doing any revalidation. 22018033426eSJeff Layton * 22028033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 22038033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 22048033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 22058033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 22068033426eSJeff Layton * bogus and it doesn't exist. 22078033426eSJeff Layton * 22088033426eSJeff Layton * Returns: 22098033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 22108033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 22118033426eSJeff Layton * put in this case. 22128033426eSJeff Layton * 22138033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 22148033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 22158033426eSJeff Layton * The nd->path reference will also be put. 22168033426eSJeff Layton * 22178033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 22188033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 22198033426eSJeff Layton * to the link, and nd->path will *not* be put. 22208033426eSJeff Layton */ 22218033426eSJeff Layton static int 2222197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 22238033426eSJeff Layton { 22248033426eSJeff Layton int error = 0; 22258033426eSJeff Layton struct dentry *dentry; 22268033426eSJeff Layton struct dentry *dir = nd->path.dentry; 22278033426eSJeff Layton 222835759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 222935759521SAl Viro if (nd->flags & LOOKUP_RCU) { 223035759521SAl Viro if (unlazy_walk(nd, NULL)) { 22318033426eSJeff Layton error = -ECHILD; 223235759521SAl Viro goto out; 223335759521SAl Viro } 22348033426eSJeff Layton } 22358033426eSJeff Layton 22368033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 22378033426eSJeff Layton 22388033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 22398033426eSJeff Layton error = handle_dots(nd, nd->last_type); 224035759521SAl Viro if (error) 224135759521SAl Viro goto out; 22428033426eSJeff Layton dentry = dget(nd->path.dentry); 224335759521SAl Viro goto done; 22448033426eSJeff Layton } 22458033426eSJeff Layton 22468033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 22478033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 22488033426eSJeff Layton if (!dentry) { 22498033426eSJeff Layton /* 22508033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 22518033426eSJeff Layton * so that means that this dentry is probably a symlink or the 22528033426eSJeff Layton * path doesn't actually point to a mounted dentry. 22538033426eSJeff Layton */ 22548033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 22558033426eSJeff Layton if (!dentry) { 22568033426eSJeff Layton error = -ENOMEM; 2257bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 225835759521SAl Viro goto out; 22598033426eSJeff Layton } 226035759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 226135759521SAl Viro error = PTR_ERR(dentry); 2262bcceeebaSDave Jones if (IS_ERR(dentry)) { 2263bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 226435759521SAl Viro goto out; 22658033426eSJeff Layton } 2266bcceeebaSDave Jones } 22678033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 22688033426eSJeff Layton 226935759521SAl Viro done: 227022213318SAl Viro if (!dentry->d_inode || d_is_negative(dentry)) { 22718033426eSJeff Layton error = -ENOENT; 22728033426eSJeff Layton dput(dentry); 227335759521SAl Viro goto out; 227435759521SAl Viro } 22758033426eSJeff Layton path->dentry = dentry; 2276295dc39dSVasily Averin path->mnt = nd->path.mnt; 2277b18825a7SDavid Howells if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) 22788033426eSJeff Layton return 1; 2279295dc39dSVasily Averin mntget(path->mnt); 22808033426eSJeff Layton follow_mount(path); 228135759521SAl Viro error = 0; 228235759521SAl Viro out: 22838033426eSJeff Layton terminate_walk(nd); 22848033426eSJeff Layton return error; 22858033426eSJeff Layton } 22868033426eSJeff Layton 22878033426eSJeff Layton /** 2288197df04cSAl Viro * path_mountpoint - look up a path to be umounted 22898033426eSJeff Layton * @dfd: directory file descriptor to start walk from 22908033426eSJeff Layton * @name: full pathname to walk 2291606d6fe3SRandy Dunlap * @path: pointer to container for result 22928033426eSJeff Layton * @flags: lookup flags 22938033426eSJeff Layton * 22948033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2295606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 22968033426eSJeff Layton */ 22978033426eSJeff Layton static int 2298197df04cSAl Viro path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) 22998033426eSJeff Layton { 23008033426eSJeff Layton struct file *base = NULL; 23018033426eSJeff Layton struct nameidata nd; 23028033426eSJeff Layton int err; 23038033426eSJeff Layton 23048033426eSJeff Layton err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base); 23058033426eSJeff Layton if (unlikely(err)) 2306115cbfdcSAl Viro goto out; 23078033426eSJeff Layton 23088033426eSJeff Layton current->total_link_count = 0; 23098033426eSJeff Layton err = link_path_walk(name, &nd); 23108033426eSJeff Layton if (err) 23118033426eSJeff Layton goto out; 23128033426eSJeff Layton 2313197df04cSAl Viro err = mountpoint_last(&nd, path); 23148033426eSJeff Layton while (err > 0) { 23158033426eSJeff Layton void *cookie; 23168033426eSJeff Layton struct path link = *path; 23178033426eSJeff Layton err = may_follow_link(&link, &nd); 23188033426eSJeff Layton if (unlikely(err)) 23198033426eSJeff Layton break; 23208033426eSJeff Layton nd.flags |= LOOKUP_PARENT; 23218033426eSJeff Layton err = follow_link(&link, &nd, &cookie); 23228033426eSJeff Layton if (err) 23238033426eSJeff Layton break; 2324197df04cSAl Viro err = mountpoint_last(&nd, path); 23258033426eSJeff Layton put_link(&nd, &link, cookie); 23268033426eSJeff Layton } 23278033426eSJeff Layton out: 23288033426eSJeff Layton if (base) 23298033426eSJeff Layton fput(base); 23308033426eSJeff Layton 23318033426eSJeff Layton if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) 23328033426eSJeff Layton path_put(&nd.root); 23338033426eSJeff Layton 23348033426eSJeff Layton return err; 23358033426eSJeff Layton } 23368033426eSJeff Layton 23372d864651SAl Viro static int 23382d864651SAl Viro filename_mountpoint(int dfd, struct filename *s, struct path *path, 23392d864651SAl Viro unsigned int flags) 23402d864651SAl Viro { 23412d864651SAl Viro int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); 23422d864651SAl Viro if (unlikely(error == -ECHILD)) 23432d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags); 23442d864651SAl Viro if (unlikely(error == -ESTALE)) 23452d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); 23462d864651SAl Viro if (likely(!error)) 23472d864651SAl Viro audit_inode(s, path->dentry, 0); 23482d864651SAl Viro return error; 23492d864651SAl Viro } 23502d864651SAl Viro 23518033426eSJeff Layton /** 2352197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 23538033426eSJeff Layton * @dfd: directory file descriptor 23548033426eSJeff Layton * @name: pathname from userland 23558033426eSJeff Layton * @flags: lookup flags 23568033426eSJeff Layton * @path: pointer to container to hold result 23578033426eSJeff Layton * 23588033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 23598033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 23608033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 23618033426eSJeff Layton * and avoid revalidating the last component. 23628033426eSJeff Layton * 23638033426eSJeff Layton * Returns 0 and populates "path" on success. 23648033426eSJeff Layton */ 23658033426eSJeff Layton int 2366197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 23678033426eSJeff Layton struct path *path) 23688033426eSJeff Layton { 23698033426eSJeff Layton struct filename *s = getname(name); 23708033426eSJeff Layton int error; 23718033426eSJeff Layton if (IS_ERR(s)) 23728033426eSJeff Layton return PTR_ERR(s); 23732d864651SAl Viro error = filename_mountpoint(dfd, s, path, flags); 23748033426eSJeff Layton putname(s); 23758033426eSJeff Layton return error; 23768033426eSJeff Layton } 23778033426eSJeff Layton 23782d864651SAl Viro int 23792d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 23802d864651SAl Viro unsigned int flags) 23812d864651SAl Viro { 23822d864651SAl Viro struct filename s = {.name = name}; 23832d864651SAl Viro return filename_mountpoint(dfd, &s, path, flags); 23842d864651SAl Viro } 23852d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 23862d864651SAl Viro 2387cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 23881da177e4SLinus Torvalds { 23898e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2390da9592edSDavid Howells 23918e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 23921da177e4SLinus Torvalds return 0; 23938e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 23941da177e4SLinus Torvalds return 0; 239523adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 23961da177e4SLinus Torvalds } 2397cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 23981da177e4SLinus Torvalds 23991da177e4SLinus Torvalds /* 24001da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 24011da177e4SLinus Torvalds * whether the type of victim is right. 24021da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 24031da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 24041da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 24051da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 24061da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 24071da177e4SLinus Torvalds * a. be owner of dir, or 24081da177e4SLinus Torvalds * b. be owner of victim, or 24091da177e4SLinus Torvalds * c. have CAP_FOWNER capability 24101da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 24111da177e4SLinus Torvalds * links pointing to it. 24121da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 24131da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 24141da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 24151da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 24161da177e4SLinus Torvalds * nfs_async_unlink(). 24171da177e4SLinus Torvalds */ 2418b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 24191da177e4SLinus Torvalds { 2420b18825a7SDavid Howells struct inode *inode = victim->d_inode; 24211da177e4SLinus Torvalds int error; 24221da177e4SLinus Torvalds 2423b18825a7SDavid Howells if (d_is_negative(victim)) 24241da177e4SLinus Torvalds return -ENOENT; 2425b18825a7SDavid Howells BUG_ON(!inode); 24261da177e4SLinus Torvalds 24271da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24284fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 24291da177e4SLinus Torvalds 2430f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 24311da177e4SLinus Torvalds if (error) 24321da177e4SLinus Torvalds return error; 24331da177e4SLinus Torvalds if (IS_APPEND(dir)) 24341da177e4SLinus Torvalds return -EPERM; 2435b18825a7SDavid Howells 2436b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2437b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 24381da177e4SLinus Torvalds return -EPERM; 24391da177e4SLinus Torvalds if (isdir) { 244044b1d530SMiklos Szeredi if (!d_is_dir(victim)) 24411da177e4SLinus Torvalds return -ENOTDIR; 24421da177e4SLinus Torvalds if (IS_ROOT(victim)) 24431da177e4SLinus Torvalds return -EBUSY; 244444b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 24451da177e4SLinus Torvalds return -EISDIR; 24461da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24471da177e4SLinus Torvalds return -ENOENT; 24481da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 24491da177e4SLinus Torvalds return -EBUSY; 24501da177e4SLinus Torvalds return 0; 24511da177e4SLinus Torvalds } 24521da177e4SLinus Torvalds 24531da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 24541da177e4SLinus Torvalds * dir. 24551da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 24561da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 24571da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 24581da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 24591da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 24601da177e4SLinus Torvalds */ 2461a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 24621da177e4SLinus Torvalds { 246314e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 24641da177e4SLinus Torvalds if (child->d_inode) 24651da177e4SLinus Torvalds return -EEXIST; 24661da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24671da177e4SLinus Torvalds return -ENOENT; 2468f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 24691da177e4SLinus Torvalds } 24701da177e4SLinus Torvalds 24711da177e4SLinus Torvalds /* 24721da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 24731da177e4SLinus Torvalds */ 24741da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 24751da177e4SLinus Torvalds { 24761da177e4SLinus Torvalds struct dentry *p; 24771da177e4SLinus Torvalds 24781da177e4SLinus Torvalds if (p1 == p2) { 2479f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 24801da177e4SLinus Torvalds return NULL; 24811da177e4SLinus Torvalds } 24821da177e4SLinus Torvalds 2483a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 24841da177e4SLinus Torvalds 2485e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2486e2761a11SOGAWA Hirofumi if (p) { 2487f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2488f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 24891da177e4SLinus Torvalds return p; 24901da177e4SLinus Torvalds } 24911da177e4SLinus Torvalds 2492e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2493e2761a11SOGAWA Hirofumi if (p) { 2494f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2495f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 24961da177e4SLinus Torvalds return p; 24971da177e4SLinus Torvalds } 24981da177e4SLinus Torvalds 2499f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2500d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 25011da177e4SLinus Torvalds return NULL; 25021da177e4SLinus Torvalds } 25034d359507SAl Viro EXPORT_SYMBOL(lock_rename); 25041da177e4SLinus Torvalds 25051da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 25061da177e4SLinus Torvalds { 25071b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 25081da177e4SLinus Torvalds if (p1 != p2) { 25091b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2510a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25111da177e4SLinus Torvalds } 25121da177e4SLinus Torvalds } 25134d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 25141da177e4SLinus Torvalds 25154acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2516312b63fbSAl Viro bool want_excl) 25171da177e4SLinus Torvalds { 2518a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25191da177e4SLinus Torvalds if (error) 25201da177e4SLinus Torvalds return error; 25211da177e4SLinus Torvalds 2522acfa4380SAl Viro if (!dir->i_op->create) 25231da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 25241da177e4SLinus Torvalds mode &= S_IALLUGO; 25251da177e4SLinus Torvalds mode |= S_IFREG; 25261da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 25271da177e4SLinus Torvalds if (error) 25281da177e4SLinus Torvalds return error; 2529312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2530a74574aaSStephen Smalley if (!error) 2531f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25321da177e4SLinus Torvalds return error; 25331da177e4SLinus Torvalds } 25344d359507SAl Viro EXPORT_SYMBOL(vfs_create); 25351da177e4SLinus Torvalds 253673d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 25371da177e4SLinus Torvalds { 25383fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 25391da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 25401da177e4SLinus Torvalds int error; 25411da177e4SLinus Torvalds 2542bcda7652SAl Viro /* O_PATH? */ 2543bcda7652SAl Viro if (!acc_mode) 2544bcda7652SAl Viro return 0; 2545bcda7652SAl Viro 25461da177e4SLinus Torvalds if (!inode) 25471da177e4SLinus Torvalds return -ENOENT; 25481da177e4SLinus Torvalds 2549c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2550c8fe8f30SChristoph Hellwig case S_IFLNK: 25511da177e4SLinus Torvalds return -ELOOP; 2552c8fe8f30SChristoph Hellwig case S_IFDIR: 2553c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 25541da177e4SLinus Torvalds return -EISDIR; 2555c8fe8f30SChristoph Hellwig break; 2556c8fe8f30SChristoph Hellwig case S_IFBLK: 2557c8fe8f30SChristoph Hellwig case S_IFCHR: 25583fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 25591da177e4SLinus Torvalds return -EACCES; 2560c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2561c8fe8f30SChristoph Hellwig case S_IFIFO: 2562c8fe8f30SChristoph Hellwig case S_IFSOCK: 25631da177e4SLinus Torvalds flag &= ~O_TRUNC; 2564c8fe8f30SChristoph Hellwig break; 25654a3fd211SDave Hansen } 2566b41572e9SDave Hansen 25673fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2568b41572e9SDave Hansen if (error) 2569b41572e9SDave Hansen return error; 25706146f0d5SMimi Zohar 25711da177e4SLinus Torvalds /* 25721da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 25731da177e4SLinus Torvalds */ 25741da177e4SLinus Torvalds if (IS_APPEND(inode)) { 25758737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 25767715b521SAl Viro return -EPERM; 25771da177e4SLinus Torvalds if (flag & O_TRUNC) 25787715b521SAl Viro return -EPERM; 25791da177e4SLinus Torvalds } 25801da177e4SLinus Torvalds 25811da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 25822e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 25837715b521SAl Viro return -EPERM; 25841da177e4SLinus Torvalds 2585f3c7691eSJ. Bruce Fields return 0; 25867715b521SAl Viro } 25877715b521SAl Viro 2588e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 25897715b521SAl Viro { 2590e1181ee6SJeff Layton struct path *path = &filp->f_path; 25917715b521SAl Viro struct inode *inode = path->dentry->d_inode; 25927715b521SAl Viro int error = get_write_access(inode); 25931da177e4SLinus Torvalds if (error) 25947715b521SAl Viro return error; 25951da177e4SLinus Torvalds /* 25961da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 25971da177e4SLinus Torvalds */ 2598d7a06983SJeff Layton error = locks_verify_locked(filp); 2599be6d3e56SKentaro Takeda if (!error) 2600ea0d3ab2STetsuo Handa error = security_path_truncate(path); 26011da177e4SLinus Torvalds if (!error) { 26027715b521SAl Viro error = do_truncate(path->dentry, 0, 2603d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2604e1181ee6SJeff Layton filp); 26051da177e4SLinus Torvalds } 26061da177e4SLinus Torvalds put_write_access(inode); 2607acd0c935SMimi Zohar return error; 26081da177e4SLinus Torvalds } 26091da177e4SLinus Torvalds 2610d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2611d57999e1SDave Hansen { 26128a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 26138a5e929dSAl Viro flag--; 2614d57999e1SDave Hansen return flag; 2615d57999e1SDave Hansen } 2616d57999e1SDave Hansen 2617d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2618d18e9008SMiklos Szeredi { 2619d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2620d18e9008SMiklos Szeredi if (error) 2621d18e9008SMiklos Szeredi return error; 2622d18e9008SMiklos Szeredi 2623d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2624d18e9008SMiklos Szeredi if (error) 2625d18e9008SMiklos Szeredi return error; 2626d18e9008SMiklos Szeredi 2627d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2628d18e9008SMiklos Szeredi } 2629d18e9008SMiklos Szeredi 26301acf0af9SDavid Howells /* 26311acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 26321acf0af9SDavid Howells * dentry. 26331acf0af9SDavid Howells * 26341acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 26351acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 26361acf0af9SDavid Howells * 26371acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 26381acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 26391acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 26401acf0af9SDavid Howells * 26411acf0af9SDavid Howells * Returns an error code otherwise. 26421acf0af9SDavid Howells */ 26432675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 264430d90494SAl Viro struct path *path, struct file *file, 2645015c3bbcSMiklos Szeredi const struct open_flags *op, 264664894cf8SAl Viro bool got_write, bool need_lookup, 264747237687SAl Viro int *opened) 2648d18e9008SMiklos Szeredi { 2649d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2650d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2651d18e9008SMiklos Szeredi umode_t mode; 2652d18e9008SMiklos Szeredi int error; 2653d18e9008SMiklos Szeredi int acc_mode; 2654d18e9008SMiklos Szeredi int create_error = 0; 2655d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2656116cc022SMiklos Szeredi bool excl; 2657d18e9008SMiklos Szeredi 2658d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2659d18e9008SMiklos Szeredi 2660d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2661d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 26622675a4ebSAl Viro error = -ENOENT; 2663d18e9008SMiklos Szeredi goto out; 2664d18e9008SMiklos Szeredi } 2665d18e9008SMiklos Szeredi 266662b259d8SMiklos Szeredi mode = op->mode; 2667d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2668d18e9008SMiklos Szeredi mode &= ~current_umask(); 2669d18e9008SMiklos Szeredi 2670116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2671116cc022SMiklos Szeredi if (excl) 2672d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2673d18e9008SMiklos Szeredi 2674d18e9008SMiklos Szeredi /* 2675d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2676d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2677d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2678d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2679d18e9008SMiklos Szeredi * 2680d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2681d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2682d18e9008SMiklos Szeredi */ 268364894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 268464894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 268564894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2686d18e9008SMiklos Szeredi /* 2687d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2688d18e9008SMiklos Szeredi * back to lookup + open 2689d18e9008SMiklos Szeredi */ 2690d18e9008SMiklos Szeredi goto no_open; 2691d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2692d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 269364894cf8SAl Viro create_error = -EROFS; 2694d18e9008SMiklos Szeredi goto no_open; 2695d18e9008SMiklos Szeredi } else { 2696d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 269764894cf8SAl Viro create_error = -EROFS; 2698d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2699d18e9008SMiklos Szeredi } 2700d18e9008SMiklos Szeredi } 2701d18e9008SMiklos Szeredi 2702d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 270338227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2704d18e9008SMiklos Szeredi if (error) { 2705d18e9008SMiklos Szeredi create_error = error; 2706d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2707d18e9008SMiklos Szeredi goto no_open; 2708d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2709d18e9008SMiklos Szeredi } 2710d18e9008SMiklos Szeredi } 2711d18e9008SMiklos Szeredi 2712d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2713d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2714d18e9008SMiklos Szeredi 271530d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 271630d90494SAl Viro file->f_path.mnt = nd->path.mnt; 271730d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 271847237687SAl Viro opened); 2719d9585277SAl Viro if (error < 0) { 2720d9585277SAl Viro if (create_error && error == -ENOENT) 2721d9585277SAl Viro error = create_error; 2722d18e9008SMiklos Szeredi goto out; 2723d18e9008SMiklos Szeredi } 2724d18e9008SMiklos Szeredi 2725d9585277SAl Viro if (error) { /* returned 1, that is */ 272630d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 27272675a4ebSAl Viro error = -EIO; 2728d18e9008SMiklos Szeredi goto out; 2729d18e9008SMiklos Szeredi } 273030d90494SAl Viro if (file->f_path.dentry) { 2731d18e9008SMiklos Szeredi dput(dentry); 273230d90494SAl Viro dentry = file->f_path.dentry; 2733d18e9008SMiklos Szeredi } 273403da633aSAl Viro if (*opened & FILE_CREATED) 273503da633aSAl Viro fsnotify_create(dir, dentry); 273603da633aSAl Viro if (!dentry->d_inode) { 273703da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 273803da633aSAl Viro if (create_error) { 273962b2ce96SSage Weil error = create_error; 274062b2ce96SSage Weil goto out; 274162b2ce96SSage Weil } 274203da633aSAl Viro } else { 274303da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 274403da633aSAl Viro error = -EEXIST; 274503da633aSAl Viro goto out; 274603da633aSAl Viro } 274703da633aSAl Viro } 2748d18e9008SMiklos Szeredi goto looked_up; 2749d18e9008SMiklos Szeredi } 2750d18e9008SMiklos Szeredi 2751d18e9008SMiklos Szeredi /* 2752d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2753d18e9008SMiklos Szeredi * here. 2754d18e9008SMiklos Szeredi */ 275503da633aSAl Viro acc_mode = op->acc_mode; 275603da633aSAl Viro if (*opened & FILE_CREATED) { 275703da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 275803da633aSAl Viro fsnotify_create(dir, dentry); 275903da633aSAl Viro acc_mode = MAY_OPEN; 276003da633aSAl Viro } 27612675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 27622675a4ebSAl Viro if (error) 27632675a4ebSAl Viro fput(file); 2764d18e9008SMiklos Szeredi 2765d18e9008SMiklos Szeredi out: 2766d18e9008SMiklos Szeredi dput(dentry); 27672675a4ebSAl Viro return error; 2768d18e9008SMiklos Szeredi 2769d18e9008SMiklos Szeredi no_open: 2770d18e9008SMiklos Szeredi if (need_lookup) { 277172bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2772d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 27732675a4ebSAl Viro return PTR_ERR(dentry); 2774d18e9008SMiklos Szeredi 2775d18e9008SMiklos Szeredi if (create_error) { 2776d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2777d18e9008SMiklos Szeredi 27782675a4ebSAl Viro error = create_error; 2779d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2780d18e9008SMiklos Szeredi if (!dentry->d_inode) 2781d18e9008SMiklos Szeredi goto out; 2782d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2783d18e9008SMiklos Szeredi goto out; 2784d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2785d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2786d18e9008SMiklos Szeredi goto out; 2787d18e9008SMiklos Szeredi } 2788d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2789d18e9008SMiklos Szeredi } 2790d18e9008SMiklos Szeredi } 2791d18e9008SMiklos Szeredi looked_up: 2792d18e9008SMiklos Szeredi path->dentry = dentry; 2793d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 27942675a4ebSAl Viro return 1; 2795d18e9008SMiklos Szeredi } 2796d18e9008SMiklos Szeredi 279731e6b01fSNick Piggin /* 27981acf0af9SDavid Howells * Look up and maybe create and open the last component. 2799d58ffd35SMiklos Szeredi * 2800d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2801d58ffd35SMiklos Szeredi * 28021acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 28031acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 28041acf0af9SDavid Howells * 28051acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 28061acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 28071acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 28081acf0af9SDavid Howells * specified then a negative dentry may be returned. 28091acf0af9SDavid Howells * 28101acf0af9SDavid Howells * An error code is returned otherwise. 28111acf0af9SDavid Howells * 28121acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 28131acf0af9SDavid Howells * cleared otherwise prior to returning. 2814d58ffd35SMiklos Szeredi */ 28152675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 281630d90494SAl Viro struct file *file, 2817d58ffd35SMiklos Szeredi const struct open_flags *op, 281864894cf8SAl Viro bool got_write, int *opened) 2819d58ffd35SMiklos Szeredi { 2820d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 282154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2822d58ffd35SMiklos Szeredi struct dentry *dentry; 2823d58ffd35SMiklos Szeredi int error; 282454ef4872SMiklos Szeredi bool need_lookup; 2825d58ffd35SMiklos Szeredi 282647237687SAl Viro *opened &= ~FILE_CREATED; 2827201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2828d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 28292675a4ebSAl Viro return PTR_ERR(dentry); 2830d58ffd35SMiklos Szeredi 2831d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2832d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2833d18e9008SMiklos Szeredi goto out_no_open; 2834d18e9008SMiklos Szeredi 2835d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 283664894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 283747237687SAl Viro need_lookup, opened); 2838d18e9008SMiklos Szeredi } 2839d18e9008SMiklos Szeredi 284054ef4872SMiklos Szeredi if (need_lookup) { 284154ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 284254ef4872SMiklos Szeredi 284372bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 284454ef4872SMiklos Szeredi if (IS_ERR(dentry)) 28452675a4ebSAl Viro return PTR_ERR(dentry); 284654ef4872SMiklos Szeredi } 284754ef4872SMiklos Szeredi 2848d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2849d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2850d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2851d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2852d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2853d58ffd35SMiklos Szeredi /* 2854d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2855d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2856d58ffd35SMiklos Szeredi * the time when the file is created and when 2857d58ffd35SMiklos Szeredi * a permanent write count is taken through 2858015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2859d58ffd35SMiklos Szeredi */ 286064894cf8SAl Viro if (!got_write) { 286164894cf8SAl Viro error = -EROFS; 2862d58ffd35SMiklos Szeredi goto out_dput; 286364894cf8SAl Viro } 286447237687SAl Viro *opened |= FILE_CREATED; 2865d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2866d58ffd35SMiklos Szeredi if (error) 2867d58ffd35SMiklos Szeredi goto out_dput; 2868312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2869312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2870d58ffd35SMiklos Szeredi if (error) 2871d58ffd35SMiklos Szeredi goto out_dput; 2872d58ffd35SMiklos Szeredi } 2873d18e9008SMiklos Szeredi out_no_open: 2874d58ffd35SMiklos Szeredi path->dentry = dentry; 2875d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 28762675a4ebSAl Viro return 1; 2877d58ffd35SMiklos Szeredi 2878d58ffd35SMiklos Szeredi out_dput: 2879d58ffd35SMiklos Szeredi dput(dentry); 28802675a4ebSAl Viro return error; 2881d58ffd35SMiklos Szeredi } 2882d58ffd35SMiklos Szeredi 2883d58ffd35SMiklos Szeredi /* 2884fe2d35ffSAl Viro * Handle the last step of open() 288531e6b01fSNick Piggin */ 28862675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 288730d90494SAl Viro struct file *file, const struct open_flags *op, 2888669abf4eSJeff Layton int *opened, struct filename *name) 2889fb1cc555SAl Viro { 2890a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2891ca344a89SAl Viro int open_flag = op->open_flag; 289277d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 289364894cf8SAl Viro bool got_write = false; 2894bcda7652SAl Viro int acc_mode = op->acc_mode; 2895a1eb3315SMiklos Szeredi struct inode *inode; 289677d660a8SMiklos Szeredi bool symlink_ok = false; 289716b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 289816b1c1cdSMiklos Szeredi bool retried = false; 289916c2cd71SAl Viro int error; 2900fb1cc555SAl Viro 2901c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2902c3e380b0SAl Viro nd->flags |= op->intent; 2903c3e380b0SAl Viro 2904bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2905fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2906fe2d35ffSAl Viro if (error) 29072675a4ebSAl Viro return error; 2908e83db167SMiklos Szeredi goto finish_open; 29091f36f774SAl Viro } 2910a2c36b45SAl Viro 2911ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2912fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2913fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2914bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 291577d660a8SMiklos Szeredi symlink_ok = true; 2916fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2917e97cdc87SAl Viro error = lookup_fast(nd, path, &inode); 291871574865SMiklos Szeredi if (likely(!error)) 291971574865SMiklos Szeredi goto finish_lookup; 292071574865SMiklos Szeredi 2921ce57dfc1SAl Viro if (error < 0) 29222675a4ebSAl Viro goto out; 2923a1eb3315SMiklos Szeredi 292437d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2925b6183df7SMiklos Szeredi } else { 2926fe2d35ffSAl Viro /* create side of things */ 2927a3fbbde7SAl Viro /* 2928b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2929b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2930b6183df7SMiklos Szeredi * about to look up 2931a3fbbde7SAl Viro */ 29329f1fafeeSAl Viro error = complete_walk(nd); 29339f1fafeeSAl Viro if (error) 29342675a4ebSAl Viro return error; 2935fe2d35ffSAl Viro 293633e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 293716c2cd71SAl Viro error = -EISDIR; 29381f36f774SAl Viro /* trailing slashes? */ 293931e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 29402675a4ebSAl Viro goto out; 2941b6183df7SMiklos Szeredi } 29421f36f774SAl Viro 294316b1c1cdSMiklos Szeredi retry_lookup: 294464894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 294564894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 294664894cf8SAl Viro if (!error) 294764894cf8SAl Viro got_write = true; 294864894cf8SAl Viro /* 294964894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 295064894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 295164894cf8SAl Viro * dropping this one anyway. 295264894cf8SAl Viro */ 295364894cf8SAl Viro } 2954a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 295564894cf8SAl Viro error = lookup_open(nd, path, file, op, got_write, opened); 2956fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2957fb1cc555SAl Viro 29582675a4ebSAl Viro if (error <= 0) { 29592675a4ebSAl Viro if (error) 2960d58ffd35SMiklos Szeredi goto out; 29616c0d46c4SAl Viro 296247237687SAl Viro if ((*opened & FILE_CREATED) || 2963496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 296477d660a8SMiklos Szeredi will_truncate = false; 2965d18e9008SMiklos Szeredi 2966adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 2967d18e9008SMiklos Szeredi goto opened; 2968d18e9008SMiklos Szeredi } 2969d18e9008SMiklos Szeredi 297047237687SAl Viro if (*opened & FILE_CREATED) { 29719b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2972ca344a89SAl Viro open_flag &= ~O_TRUNC; 297377d660a8SMiklos Szeredi will_truncate = false; 2974bcda7652SAl Viro acc_mode = MAY_OPEN; 2975d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 2976e83db167SMiklos Szeredi goto finish_open_created; 2977fb1cc555SAl Viro } 2978fb1cc555SAl Viro 2979fb1cc555SAl Viro /* 29803134f37eSJeff Layton * create/update audit record if it already exists. 2981fb1cc555SAl Viro */ 2982b18825a7SDavid Howells if (d_is_positive(path->dentry)) 2983adb5c247SJeff Layton audit_inode(name, path->dentry, 0); 2984fb1cc555SAl Viro 2985d18e9008SMiklos Szeredi /* 2986d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 2987d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 2988d18e9008SMiklos Szeredi * necessary...) 2989d18e9008SMiklos Szeredi */ 299064894cf8SAl Viro if (got_write) { 2991d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 299264894cf8SAl Viro got_write = false; 2993d18e9008SMiklos Szeredi } 2994d18e9008SMiklos Szeredi 2995fb1cc555SAl Viro error = -EEXIST; 2996f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 2997fb1cc555SAl Viro goto exit_dput; 2998fb1cc555SAl Viro 29999875cf80SDavid Howells error = follow_managed(path, nd->flags); 30009875cf80SDavid Howells if (error < 0) 3001fb1cc555SAl Viro goto exit_dput; 3002fb1cc555SAl Viro 3003a3fbbde7SAl Viro if (error) 3004a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 3005a3fbbde7SAl Viro 3006decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 3007decf3400SMiklos Szeredi inode = path->dentry->d_inode; 30085f5daac1SMiklos Szeredi finish_lookup: 30095f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 3010fb1cc555SAl Viro error = -ENOENT; 301122213318SAl Viro if (!inode || d_is_negative(path->dentry)) { 301254c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 30132675a4ebSAl Viro goto out; 301454c33e7fSMiklos Szeredi } 30159e67f361SAl Viro 3016b18825a7SDavid Howells if (should_follow_link(path->dentry, !symlink_ok)) { 3017d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 3018d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 3019d45ea867SMiklos Szeredi error = -ECHILD; 30202675a4ebSAl Viro goto out; 3021d45ea867SMiklos Szeredi } 3022d45ea867SMiklos Szeredi } 3023d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 30242675a4ebSAl Viro return 1; 3025d45ea867SMiklos Szeredi } 3026fb1cc555SAl Viro 302716b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 3028fb1cc555SAl Viro path_to_nameidata(path, nd); 302916b1c1cdSMiklos Szeredi } else { 303016b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 303116b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 303216b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 303316b1c1cdSMiklos Szeredi 303416b1c1cdSMiklos Szeredi } 3035decf3400SMiklos Szeredi nd->inode = inode; 3036a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3037bc77daa7SAl Viro finish_open: 3038a3fbbde7SAl Viro error = complete_walk(nd); 303916b1c1cdSMiklos Szeredi if (error) { 304016b1c1cdSMiklos Szeredi path_put(&save_parent); 30412675a4ebSAl Viro return error; 304216b1c1cdSMiklos Szeredi } 3043bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3044fb1cc555SAl Viro error = -EISDIR; 304544b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 30462675a4ebSAl Viro goto out; 3047af2f5542SMiklos Szeredi error = -ENOTDIR; 304844b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 30492675a4ebSAl Viro goto out; 30506c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 305177d660a8SMiklos Szeredi will_truncate = false; 30526c0d46c4SAl Viro 30530f9d1a10SAl Viro if (will_truncate) { 30540f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 30550f9d1a10SAl Viro if (error) 30562675a4ebSAl Viro goto out; 305764894cf8SAl Viro got_write = true; 30580f9d1a10SAl Viro } 3059e83db167SMiklos Szeredi finish_open_created: 3060bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3061ca344a89SAl Viro if (error) 30622675a4ebSAl Viro goto out; 30634aa7c634SMiklos Szeredi 30644aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 30654aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 30664aa7c634SMiklos Szeredi if (!error) { 30674aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 30684aa7c634SMiklos Szeredi } else { 306930d90494SAl Viro if (error == -EOPENSTALE) 3070f60dc3dbSMiklos Szeredi goto stale_open; 3071015c3bbcSMiklos Szeredi goto out; 3072f60dc3dbSMiklos Szeredi } 3073a8277b9bSMiklos Szeredi opened: 30742675a4ebSAl Viro error = open_check_o_direct(file); 3075015c3bbcSMiklos Szeredi if (error) 3076015c3bbcSMiklos Szeredi goto exit_fput; 30773034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3078aa4caadbSMiklos Szeredi if (error) 3079aa4caadbSMiklos Szeredi goto exit_fput; 3080aa4caadbSMiklos Szeredi 30810f9d1a10SAl Viro if (will_truncate) { 30822675a4ebSAl Viro error = handle_truncate(file); 3083aa4caadbSMiklos Szeredi if (error) 3084aa4caadbSMiklos Szeredi goto exit_fput; 30850f9d1a10SAl Viro } 3086ca344a89SAl Viro out: 308764894cf8SAl Viro if (got_write) 30880f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 308916b1c1cdSMiklos Szeredi path_put(&save_parent); 3090e276ae67SMiklos Szeredi terminate_walk(nd); 30912675a4ebSAl Viro return error; 3092fb1cc555SAl Viro 3093fb1cc555SAl Viro exit_dput: 3094fb1cc555SAl Viro path_put_conditional(path, nd); 3095ca344a89SAl Viro goto out; 3096015c3bbcSMiklos Szeredi exit_fput: 30972675a4ebSAl Viro fput(file); 30982675a4ebSAl Viro goto out; 3099015c3bbcSMiklos Szeredi 3100f60dc3dbSMiklos Szeredi stale_open: 3101f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3102f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3103f60dc3dbSMiklos Szeredi goto out; 3104f60dc3dbSMiklos Szeredi 3105f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3106f60dc3dbSMiklos Szeredi path_put(&nd->path); 3107f60dc3dbSMiklos Szeredi nd->path = save_parent; 3108f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3109f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3110f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 311164894cf8SAl Viro if (got_write) { 3112f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 311364894cf8SAl Viro got_write = false; 3114f60dc3dbSMiklos Szeredi } 3115f60dc3dbSMiklos Szeredi retried = true; 3116f60dc3dbSMiklos Szeredi goto retry_lookup; 3117fb1cc555SAl Viro } 3118fb1cc555SAl Viro 311960545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 312060545d0dSAl Viro struct nameidata *nd, int flags, 312160545d0dSAl Viro const struct open_flags *op, 312260545d0dSAl Viro struct file *file, int *opened) 312360545d0dSAl Viro { 312460545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 312560545d0dSAl Viro struct dentry *dentry, *child; 312660545d0dSAl Viro struct inode *dir; 312760545d0dSAl Viro int error = path_lookupat(dfd, pathname->name, 312860545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 312960545d0dSAl Viro if (unlikely(error)) 313060545d0dSAl Viro return error; 313160545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 313260545d0dSAl Viro if (unlikely(error)) 313360545d0dSAl Viro goto out; 313460545d0dSAl Viro /* we want directory to be writable */ 313560545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 313660545d0dSAl Viro if (error) 313760545d0dSAl Viro goto out2; 313860545d0dSAl Viro dentry = nd->path.dentry; 313960545d0dSAl Viro dir = dentry->d_inode; 314060545d0dSAl Viro if (!dir->i_op->tmpfile) { 314160545d0dSAl Viro error = -EOPNOTSUPP; 314260545d0dSAl Viro goto out2; 314360545d0dSAl Viro } 314460545d0dSAl Viro child = d_alloc(dentry, &name); 314560545d0dSAl Viro if (unlikely(!child)) { 314660545d0dSAl Viro error = -ENOMEM; 314760545d0dSAl Viro goto out2; 314860545d0dSAl Viro } 314960545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 315060545d0dSAl Viro nd->flags |= op->intent; 315160545d0dSAl Viro dput(nd->path.dentry); 315260545d0dSAl Viro nd->path.dentry = child; 315360545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 315460545d0dSAl Viro if (error) 315560545d0dSAl Viro goto out2; 315660545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 315769a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 315869a91c23SEric Rannaud error = may_open(&nd->path, MAY_OPEN, op->open_flag); 315960545d0dSAl Viro if (error) 316060545d0dSAl Viro goto out2; 316160545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 316260545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 316360545d0dSAl Viro if (error) 316460545d0dSAl Viro goto out2; 316560545d0dSAl Viro error = open_check_o_direct(file); 3166f4e0c30cSAl Viro if (error) { 316760545d0dSAl Viro fput(file); 3168f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3169f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3170f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3171f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3172f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3173f4e0c30cSAl Viro } 317460545d0dSAl Viro out2: 317560545d0dSAl Viro mnt_drop_write(nd->path.mnt); 317660545d0dSAl Viro out: 317760545d0dSAl Viro path_put(&nd->path); 317860545d0dSAl Viro return error; 317960545d0dSAl Viro } 318060545d0dSAl Viro 3181669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 318273d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 31831da177e4SLinus Torvalds { 3184fe2d35ffSAl Viro struct file *base = NULL; 318530d90494SAl Viro struct file *file; 31869850c056SAl Viro struct path path; 318747237687SAl Viro int opened = 0; 318813aab428SAl Viro int error; 318931e6b01fSNick Piggin 319030d90494SAl Viro file = get_empty_filp(); 31911afc99beSAl Viro if (IS_ERR(file)) 31921afc99beSAl Viro return file; 319331e6b01fSNick Piggin 319430d90494SAl Viro file->f_flags = op->open_flag; 319531e6b01fSNick Piggin 3196bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 319760545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 319860545d0dSAl Viro goto out; 319960545d0dSAl Viro } 320060545d0dSAl Viro 3201669abf4eSJeff Layton error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base); 320231e6b01fSNick Piggin if (unlikely(error)) 32032675a4ebSAl Viro goto out; 320431e6b01fSNick Piggin 3205fe2d35ffSAl Viro current->total_link_count = 0; 3206669abf4eSJeff Layton error = link_path_walk(pathname->name, nd); 320731e6b01fSNick Piggin if (unlikely(error)) 32082675a4ebSAl Viro goto out; 32091da177e4SLinus Torvalds 32102675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 32112675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 32127b9337aaSNick Piggin struct path link = path; 3213def4af30SAl Viro void *cookie; 3214574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 321573d049a4SAl Viro path_put_conditional(&path, nd); 321673d049a4SAl Viro path_put(&nd->path); 32172675a4ebSAl Viro error = -ELOOP; 321840b39136SAl Viro break; 321940b39136SAl Viro } 3220800179c9SKees Cook error = may_follow_link(&link, nd); 3221800179c9SKees Cook if (unlikely(error)) 3222800179c9SKees Cook break; 322373d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 322473d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3225574197e0SAl Viro error = follow_link(&link, nd, &cookie); 3226c3e380b0SAl Viro if (unlikely(error)) 32272675a4ebSAl Viro break; 32282675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 3229574197e0SAl Viro put_link(nd, &link, cookie); 3230806b681cSAl Viro } 323110fa8e62SAl Viro out: 323273d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 323373d049a4SAl Viro path_put(&nd->root); 3234fe2d35ffSAl Viro if (base) 3235fe2d35ffSAl Viro fput(base); 32362675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32372675a4ebSAl Viro BUG_ON(!error); 323830d90494SAl Viro put_filp(file); 3239015c3bbcSMiklos Szeredi } 32402675a4ebSAl Viro if (unlikely(error)) { 32412675a4ebSAl Viro if (error == -EOPENSTALE) { 32422675a4ebSAl Viro if (flags & LOOKUP_RCU) 32432675a4ebSAl Viro error = -ECHILD; 32442675a4ebSAl Viro else 32452675a4ebSAl Viro error = -ESTALE; 32462675a4ebSAl Viro } 32472675a4ebSAl Viro file = ERR_PTR(error); 32482675a4ebSAl Viro } 32492675a4ebSAl Viro return file; 3250de459215SKirill Korotaev } 32511da177e4SLinus Torvalds 3252669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3253f9652e10SAl Viro const struct open_flags *op) 325413aab428SAl Viro { 325573d049a4SAl Viro struct nameidata nd; 3256f9652e10SAl Viro int flags = op->lookup_flags; 325713aab428SAl Viro struct file *filp; 325813aab428SAl Viro 325973d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 326013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 326173d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 326213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 326373d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 326413aab428SAl Viro return filp; 326513aab428SAl Viro } 326613aab428SAl Viro 326773d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3268f9652e10SAl Viro const char *name, const struct open_flags *op) 326973d049a4SAl Viro { 327073d049a4SAl Viro struct nameidata nd; 327173d049a4SAl Viro struct file *file; 3272669abf4eSJeff Layton struct filename filename = { .name = name }; 3273f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 327473d049a4SAl Viro 327573d049a4SAl Viro nd.root.mnt = mnt; 327673d049a4SAl Viro nd.root.dentry = dentry; 327773d049a4SAl Viro 3278b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 327973d049a4SAl Viro return ERR_PTR(-ELOOP); 328073d049a4SAl Viro 3281669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); 328273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3283669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags); 328473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3285669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); 328673d049a4SAl Viro return file; 328773d049a4SAl Viro } 328873d049a4SAl Viro 32891ac12b4bSJeff Layton struct dentry *kern_path_create(int dfd, const char *pathname, 32901ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 32911da177e4SLinus Torvalds { 3292c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3293ed75e95dSAl Viro struct nameidata nd; 3294c30dabfeSJan Kara int err2; 32951ac12b4bSJeff Layton int error; 32961ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 32971ac12b4bSJeff Layton 32981ac12b4bSJeff Layton /* 32991ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 33001ac12b4bSJeff Layton * other flags passed in are ignored! 33011ac12b4bSJeff Layton */ 33021ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 33031ac12b4bSJeff Layton 33041ac12b4bSJeff Layton error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); 3305ed75e95dSAl Viro if (error) 3306ed75e95dSAl Viro return ERR_PTR(error); 33071da177e4SLinus Torvalds 3308c663e5d8SChristoph Hellwig /* 3309c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3310c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3311c663e5d8SChristoph Hellwig */ 3312ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3313ed75e95dSAl Viro goto out; 3314ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3315ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3316c663e5d8SChristoph Hellwig 3317c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3318c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3319c663e5d8SChristoph Hellwig /* 3320c663e5d8SChristoph Hellwig * Do the final lookup. 3321c663e5d8SChristoph Hellwig */ 3322ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3323ed75e95dSAl Viro dentry = lookup_hash(&nd); 33241da177e4SLinus Torvalds if (IS_ERR(dentry)) 3325a8104a9fSAl Viro goto unlock; 3326c663e5d8SChristoph Hellwig 3327a8104a9fSAl Viro error = -EEXIST; 3328b18825a7SDavid Howells if (d_is_positive(dentry)) 3329a8104a9fSAl Viro goto fail; 3330b18825a7SDavid Howells 3331c663e5d8SChristoph Hellwig /* 3332c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3333c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3334c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3335c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3336c663e5d8SChristoph Hellwig */ 3337ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3338a8104a9fSAl Viro error = -ENOENT; 3339ed75e95dSAl Viro goto fail; 3340e9baf6e5SAl Viro } 3341c30dabfeSJan Kara if (unlikely(err2)) { 3342c30dabfeSJan Kara error = err2; 3343a8104a9fSAl Viro goto fail; 3344c30dabfeSJan Kara } 3345ed75e95dSAl Viro *path = nd.path; 3346e9baf6e5SAl Viro return dentry; 33471da177e4SLinus Torvalds fail: 3348a8104a9fSAl Viro dput(dentry); 3349a8104a9fSAl Viro dentry = ERR_PTR(error); 3350a8104a9fSAl Viro unlock: 3351dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3352c30dabfeSJan Kara if (!err2) 3353c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3354ed75e95dSAl Viro out: 3355dae6ad8fSAl Viro path_put(&nd.path); 3356ed75e95dSAl Viro return dentry; 3357dae6ad8fSAl Viro } 3358dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3359dae6ad8fSAl Viro 3360921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3361921a1650SAl Viro { 3362921a1650SAl Viro dput(dentry); 3363921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3364a8104a9fSAl Viro mnt_drop_write(path->mnt); 3365921a1650SAl Viro path_put(path); 3366921a1650SAl Viro } 3367921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3368921a1650SAl Viro 33691ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 33701ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3371dae6ad8fSAl Viro { 337291a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3373dae6ad8fSAl Viro struct dentry *res; 3374dae6ad8fSAl Viro if (IS_ERR(tmp)) 3375dae6ad8fSAl Viro return ERR_CAST(tmp); 33761ac12b4bSJeff Layton res = kern_path_create(dfd, tmp->name, path, lookup_flags); 3377dae6ad8fSAl Viro putname(tmp); 3378dae6ad8fSAl Viro return res; 3379dae6ad8fSAl Viro } 3380dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3381dae6ad8fSAl Viro 33821a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 33831da177e4SLinus Torvalds { 3384a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 33851da177e4SLinus Torvalds 33861da177e4SLinus Torvalds if (error) 33871da177e4SLinus Torvalds return error; 33881da177e4SLinus Torvalds 3389975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 33901da177e4SLinus Torvalds return -EPERM; 33911da177e4SLinus Torvalds 3392acfa4380SAl Viro if (!dir->i_op->mknod) 33931da177e4SLinus Torvalds return -EPERM; 33941da177e4SLinus Torvalds 339508ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 339608ce5f16SSerge E. Hallyn if (error) 339708ce5f16SSerge E. Hallyn return error; 339808ce5f16SSerge E. Hallyn 33991da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 34001da177e4SLinus Torvalds if (error) 34011da177e4SLinus Torvalds return error; 34021da177e4SLinus Torvalds 34031da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3404a74574aaSStephen Smalley if (!error) 3405f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34061da177e4SLinus Torvalds return error; 34071da177e4SLinus Torvalds } 34084d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 34091da177e4SLinus Torvalds 3410f69aac00SAl Viro static int may_mknod(umode_t mode) 3411463c3197SDave Hansen { 3412463c3197SDave Hansen switch (mode & S_IFMT) { 3413463c3197SDave Hansen case S_IFREG: 3414463c3197SDave Hansen case S_IFCHR: 3415463c3197SDave Hansen case S_IFBLK: 3416463c3197SDave Hansen case S_IFIFO: 3417463c3197SDave Hansen case S_IFSOCK: 3418463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3419463c3197SDave Hansen return 0; 3420463c3197SDave Hansen case S_IFDIR: 3421463c3197SDave Hansen return -EPERM; 3422463c3197SDave Hansen default: 3423463c3197SDave Hansen return -EINVAL; 3424463c3197SDave Hansen } 3425463c3197SDave Hansen } 3426463c3197SDave Hansen 34278208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 34282e4d0924SHeiko Carstens unsigned, dev) 34291da177e4SLinus Torvalds { 34301da177e4SLinus Torvalds struct dentry *dentry; 3431dae6ad8fSAl Viro struct path path; 3432dae6ad8fSAl Viro int error; 3433972567f1SJeff Layton unsigned int lookup_flags = 0; 34341da177e4SLinus Torvalds 34358e4bfca1SAl Viro error = may_mknod(mode); 34368e4bfca1SAl Viro if (error) 34378e4bfca1SAl Viro return error; 3438972567f1SJeff Layton retry: 3439972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3440dae6ad8fSAl Viro if (IS_ERR(dentry)) 3441dae6ad8fSAl Viro return PTR_ERR(dentry); 34422ad94ae6SAl Viro 3443dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3444ce3b0f8dSAl Viro mode &= ~current_umask(); 3445dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3446be6d3e56SKentaro Takeda if (error) 3447a8104a9fSAl Viro goto out; 34481da177e4SLinus Torvalds switch (mode & S_IFMT) { 34491da177e4SLinus Torvalds case 0: case S_IFREG: 3450312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 34511da177e4SLinus Torvalds break; 34521da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3453dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 34541da177e4SLinus Torvalds new_decode_dev(dev)); 34551da177e4SLinus Torvalds break; 34561da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3457dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 34581da177e4SLinus Torvalds break; 34591da177e4SLinus Torvalds } 3460a8104a9fSAl Viro out: 3461921a1650SAl Viro done_path_create(&path, dentry); 3462972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3463972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3464972567f1SJeff Layton goto retry; 3465972567f1SJeff Layton } 34661da177e4SLinus Torvalds return error; 34671da177e4SLinus Torvalds } 34681da177e4SLinus Torvalds 34698208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 34705590ff0dSUlrich Drepper { 34715590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 34725590ff0dSUlrich Drepper } 34735590ff0dSUlrich Drepper 347418bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 34751da177e4SLinus Torvalds { 3476a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34778de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 34781da177e4SLinus Torvalds 34791da177e4SLinus Torvalds if (error) 34801da177e4SLinus Torvalds return error; 34811da177e4SLinus Torvalds 3482acfa4380SAl Viro if (!dir->i_op->mkdir) 34831da177e4SLinus Torvalds return -EPERM; 34841da177e4SLinus Torvalds 34851da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 34861da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 34871da177e4SLinus Torvalds if (error) 34881da177e4SLinus Torvalds return error; 34891da177e4SLinus Torvalds 34908de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 34918de52778SAl Viro return -EMLINK; 34928de52778SAl Viro 34931da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3494a74574aaSStephen Smalley if (!error) 3495f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 34961da177e4SLinus Torvalds return error; 34971da177e4SLinus Torvalds } 34984d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 34991da177e4SLinus Torvalds 3500a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 35011da177e4SLinus Torvalds { 35026902d925SDave Hansen struct dentry *dentry; 3503dae6ad8fSAl Viro struct path path; 3504dae6ad8fSAl Viro int error; 3505b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 35061da177e4SLinus Torvalds 3507b76d8b82SJeff Layton retry: 3508b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 35096902d925SDave Hansen if (IS_ERR(dentry)) 3510dae6ad8fSAl Viro return PTR_ERR(dentry); 35116902d925SDave Hansen 3512dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3513ce3b0f8dSAl Viro mode &= ~current_umask(); 3514dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3515a8104a9fSAl Viro if (!error) 3516dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3517921a1650SAl Viro done_path_create(&path, dentry); 3518b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3519b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3520b76d8b82SJeff Layton goto retry; 3521b76d8b82SJeff Layton } 35221da177e4SLinus Torvalds return error; 35231da177e4SLinus Torvalds } 35241da177e4SLinus Torvalds 3525a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 35265590ff0dSUlrich Drepper { 35275590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 35285590ff0dSUlrich Drepper } 35295590ff0dSUlrich Drepper 35301da177e4SLinus Torvalds /* 3531a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3532c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3533a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3534a71905f0SSage Weil * then we drop the dentry now. 35351da177e4SLinus Torvalds * 35361da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 35371da177e4SLinus Torvalds * do a 35381da177e4SLinus Torvalds * 35391da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 35401da177e4SLinus Torvalds * return -EBUSY; 35411da177e4SLinus Torvalds * 35421da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 35431da177e4SLinus Torvalds * that is still in use by something else.. 35441da177e4SLinus Torvalds */ 35451da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 35461da177e4SLinus Torvalds { 35471da177e4SLinus Torvalds shrink_dcache_parent(dentry); 35481da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 354998474236SWaiman Long if (dentry->d_lockref.count == 1) 35501da177e4SLinus Torvalds __d_drop(dentry); 35511da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 35521da177e4SLinus Torvalds } 35534d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 35541da177e4SLinus Torvalds 35551da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 35561da177e4SLinus Torvalds { 35571da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 35581da177e4SLinus Torvalds 35591da177e4SLinus Torvalds if (error) 35601da177e4SLinus Torvalds return error; 35611da177e4SLinus Torvalds 3562acfa4380SAl Viro if (!dir->i_op->rmdir) 35631da177e4SLinus Torvalds return -EPERM; 35641da177e4SLinus Torvalds 35651d2ef590SAl Viro dget(dentry); 35661b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3567912dbc15SSage Weil 35681da177e4SLinus Torvalds error = -EBUSY; 35697af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3570912dbc15SSage Weil goto out; 3571912dbc15SSage Weil 35721da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3573912dbc15SSage Weil if (error) 3574912dbc15SSage Weil goto out; 3575912dbc15SSage Weil 35763cebde24SSage Weil shrink_dcache_parent(dentry); 35771da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3578912dbc15SSage Weil if (error) 3579912dbc15SSage Weil goto out; 3580912dbc15SSage Weil 35811da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3582d83c49f3SAl Viro dont_mount(dentry); 35838ed936b5SEric W. Biederman detach_mounts(dentry); 35841da177e4SLinus Torvalds 3585912dbc15SSage Weil out: 3586912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 35871d2ef590SAl Viro dput(dentry); 3588912dbc15SSage Weil if (!error) 3589912dbc15SSage Weil d_delete(dentry); 35901da177e4SLinus Torvalds return error; 35911da177e4SLinus Torvalds } 35924d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 35931da177e4SLinus Torvalds 35945590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 35951da177e4SLinus Torvalds { 35961da177e4SLinus Torvalds int error = 0; 359791a27b2aSJeff Layton struct filename *name; 35981da177e4SLinus Torvalds struct dentry *dentry; 35991da177e4SLinus Torvalds struct nameidata nd; 3600c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3601c6ee9206SJeff Layton retry: 3602c6ee9206SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 360391a27b2aSJeff Layton if (IS_ERR(name)) 360491a27b2aSJeff Layton return PTR_ERR(name); 36051da177e4SLinus Torvalds 36061da177e4SLinus Torvalds switch(nd.last_type) { 36071da177e4SLinus Torvalds case LAST_DOTDOT: 36081da177e4SLinus Torvalds error = -ENOTEMPTY; 36091da177e4SLinus Torvalds goto exit1; 36101da177e4SLinus Torvalds case LAST_DOT: 36111da177e4SLinus Torvalds error = -EINVAL; 36121da177e4SLinus Torvalds goto exit1; 36131da177e4SLinus Torvalds case LAST_ROOT: 36141da177e4SLinus Torvalds error = -EBUSY; 36151da177e4SLinus Torvalds goto exit1; 36161da177e4SLinus Torvalds } 36170612d9fbSOGAWA Hirofumi 36180612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3619c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3620c30dabfeSJan Kara if (error) 3621c30dabfeSJan Kara goto exit1; 36220612d9fbSOGAWA Hirofumi 36234ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 362449705b77SChristoph Hellwig dentry = lookup_hash(&nd); 36251da177e4SLinus Torvalds error = PTR_ERR(dentry); 36266902d925SDave Hansen if (IS_ERR(dentry)) 36276902d925SDave Hansen goto exit2; 3628e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3629e6bc45d6STheodore Ts'o error = -ENOENT; 3630e6bc45d6STheodore Ts'o goto exit3; 3631e6bc45d6STheodore Ts'o } 3632be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3633be6d3e56SKentaro Takeda if (error) 3634c30dabfeSJan Kara goto exit3; 36354ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 36360622753bSDave Hansen exit3: 36371da177e4SLinus Torvalds dput(dentry); 36386902d925SDave Hansen exit2: 36394ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3640c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 36411da177e4SLinus Torvalds exit1: 36421d957f9bSJan Blunck path_put(&nd.path); 36431da177e4SLinus Torvalds putname(name); 3644c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3645c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3646c6ee9206SJeff Layton goto retry; 3647c6ee9206SJeff Layton } 36481da177e4SLinus Torvalds return error; 36491da177e4SLinus Torvalds } 36501da177e4SLinus Torvalds 36513cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 36525590ff0dSUlrich Drepper { 36535590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 36545590ff0dSUlrich Drepper } 36555590ff0dSUlrich Drepper 3656b21996e3SJ. Bruce Fields /** 3657b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3658b21996e3SJ. Bruce Fields * @dir: parent directory 3659b21996e3SJ. Bruce Fields * @dentry: victim 3660b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3661b21996e3SJ. Bruce Fields * 3662b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3663b21996e3SJ. Bruce Fields * 3664b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3665b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3666b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3667b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3668b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3669b21996e3SJ. Bruce Fields * 3670b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3671b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3672b21996e3SJ. Bruce Fields * to be NFS exported. 3673b21996e3SJ. Bruce Fields */ 3674b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 36751da177e4SLinus Torvalds { 36769accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 36771da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 36781da177e4SLinus Torvalds 36791da177e4SLinus Torvalds if (error) 36801da177e4SLinus Torvalds return error; 36811da177e4SLinus Torvalds 3682acfa4380SAl Viro if (!dir->i_op->unlink) 36831da177e4SLinus Torvalds return -EPERM; 36841da177e4SLinus Torvalds 36859accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 36868ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 36871da177e4SLinus Torvalds error = -EBUSY; 36881da177e4SLinus Torvalds else { 36891da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3690bec1052eSAl Viro if (!error) { 36915a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 36925a14696cSJ. Bruce Fields if (error) 3693b21996e3SJ. Bruce Fields goto out; 36941da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 36958ed936b5SEric W. Biederman if (!error) { 3696d83c49f3SAl Viro dont_mount(dentry); 36978ed936b5SEric W. Biederman detach_mounts(dentry); 36988ed936b5SEric W. Biederman } 3699bec1052eSAl Viro } 37001da177e4SLinus Torvalds } 3701b21996e3SJ. Bruce Fields out: 37029accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 37031da177e4SLinus Torvalds 37041da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 37051da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 37069accbb97SJ. Bruce Fields fsnotify_link_count(target); 37071da177e4SLinus Torvalds d_delete(dentry); 37081da177e4SLinus Torvalds } 37090eeca283SRobert Love 37101da177e4SLinus Torvalds return error; 37111da177e4SLinus Torvalds } 37124d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 37131da177e4SLinus Torvalds 37141da177e4SLinus Torvalds /* 37151da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 37161b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 37171da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 37181da177e4SLinus Torvalds * while waiting on the I/O. 37191da177e4SLinus Torvalds */ 37205590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 37211da177e4SLinus Torvalds { 37222ad94ae6SAl Viro int error; 372391a27b2aSJeff Layton struct filename *name; 37241da177e4SLinus Torvalds struct dentry *dentry; 37251da177e4SLinus Torvalds struct nameidata nd; 37261da177e4SLinus Torvalds struct inode *inode = NULL; 3727b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 37285d18f813SJeff Layton unsigned int lookup_flags = 0; 37295d18f813SJeff Layton retry: 37305d18f813SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 373191a27b2aSJeff Layton if (IS_ERR(name)) 373291a27b2aSJeff Layton return PTR_ERR(name); 37332ad94ae6SAl Viro 37341da177e4SLinus Torvalds error = -EISDIR; 37351da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 37361da177e4SLinus Torvalds goto exit1; 37370612d9fbSOGAWA Hirofumi 37380612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3739c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3740c30dabfeSJan Kara if (error) 3741c30dabfeSJan Kara goto exit1; 3742b21996e3SJ. Bruce Fields retry_deleg: 37434ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 374449705b77SChristoph Hellwig dentry = lookup_hash(&nd); 37451da177e4SLinus Torvalds error = PTR_ERR(dentry); 37461da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 37471da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 374850338b88STörök Edwin if (nd.last.name[nd.last.len]) 374950338b88STörök Edwin goto slashes; 37501da177e4SLinus Torvalds inode = dentry->d_inode; 3751b18825a7SDavid Howells if (d_is_negative(dentry)) 3752e6bc45d6STheodore Ts'o goto slashes; 37537de9c6eeSAl Viro ihold(inode); 3754be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3755be6d3e56SKentaro Takeda if (error) 3756c30dabfeSJan Kara goto exit2; 3757b21996e3SJ. Bruce Fields error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); 37581da177e4SLinus Torvalds exit2: 37591da177e4SLinus Torvalds dput(dentry); 37601da177e4SLinus Torvalds } 37614ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 37621da177e4SLinus Torvalds if (inode) 37631da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3764b21996e3SJ. Bruce Fields inode = NULL; 3765b21996e3SJ. Bruce Fields if (delegated_inode) { 37665a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3767b21996e3SJ. Bruce Fields if (!error) 3768b21996e3SJ. Bruce Fields goto retry_deleg; 3769b21996e3SJ. Bruce Fields } 3770c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 37711da177e4SLinus Torvalds exit1: 37721d957f9bSJan Blunck path_put(&nd.path); 37731da177e4SLinus Torvalds putname(name); 37745d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 37755d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 37765d18f813SJeff Layton inode = NULL; 37775d18f813SJeff Layton goto retry; 37785d18f813SJeff Layton } 37791da177e4SLinus Torvalds return error; 37801da177e4SLinus Torvalds 37811da177e4SLinus Torvalds slashes: 3782b18825a7SDavid Howells if (d_is_negative(dentry)) 3783b18825a7SDavid Howells error = -ENOENT; 378444b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3785b18825a7SDavid Howells error = -EISDIR; 3786b18825a7SDavid Howells else 3787b18825a7SDavid Howells error = -ENOTDIR; 37881da177e4SLinus Torvalds goto exit2; 37891da177e4SLinus Torvalds } 37901da177e4SLinus Torvalds 37912e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 37925590ff0dSUlrich Drepper { 37935590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 37945590ff0dSUlrich Drepper return -EINVAL; 37955590ff0dSUlrich Drepper 37965590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 37975590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 37985590ff0dSUlrich Drepper 37995590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 38005590ff0dSUlrich Drepper } 38015590ff0dSUlrich Drepper 38023480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 38035590ff0dSUlrich Drepper { 38045590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 38055590ff0dSUlrich Drepper } 38065590ff0dSUlrich Drepper 3807db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 38081da177e4SLinus Torvalds { 3809a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38101da177e4SLinus Torvalds 38111da177e4SLinus Torvalds if (error) 38121da177e4SLinus Torvalds return error; 38131da177e4SLinus Torvalds 3814acfa4380SAl Viro if (!dir->i_op->symlink) 38151da177e4SLinus Torvalds return -EPERM; 38161da177e4SLinus Torvalds 38171da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 38181da177e4SLinus Torvalds if (error) 38191da177e4SLinus Torvalds return error; 38201da177e4SLinus Torvalds 38211da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3822a74574aaSStephen Smalley if (!error) 3823f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38241da177e4SLinus Torvalds return error; 38251da177e4SLinus Torvalds } 38264d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 38271da177e4SLinus Torvalds 38282e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 38292e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 38301da177e4SLinus Torvalds { 38312ad94ae6SAl Viro int error; 383291a27b2aSJeff Layton struct filename *from; 38336902d925SDave Hansen struct dentry *dentry; 3834dae6ad8fSAl Viro struct path path; 3835f46d3567SJeff Layton unsigned int lookup_flags = 0; 38361da177e4SLinus Torvalds 38371da177e4SLinus Torvalds from = getname(oldname); 38381da177e4SLinus Torvalds if (IS_ERR(from)) 38391da177e4SLinus Torvalds return PTR_ERR(from); 3840f46d3567SJeff Layton retry: 3841f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 38421da177e4SLinus Torvalds error = PTR_ERR(dentry); 38436902d925SDave Hansen if (IS_ERR(dentry)) 3844dae6ad8fSAl Viro goto out_putname; 38456902d925SDave Hansen 384691a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3847a8104a9fSAl Viro if (!error) 384891a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3849921a1650SAl Viro done_path_create(&path, dentry); 3850f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3851f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3852f46d3567SJeff Layton goto retry; 3853f46d3567SJeff Layton } 38546902d925SDave Hansen out_putname: 38551da177e4SLinus Torvalds putname(from); 38561da177e4SLinus Torvalds return error; 38571da177e4SLinus Torvalds } 38581da177e4SLinus Torvalds 38593480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 38605590ff0dSUlrich Drepper { 38615590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 38625590ff0dSUlrich Drepper } 38635590ff0dSUlrich Drepper 3864146a8595SJ. Bruce Fields /** 3865146a8595SJ. Bruce Fields * vfs_link - create a new link 3866146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3867146a8595SJ. Bruce Fields * @dir: new parent 3868146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3869146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3870146a8595SJ. Bruce Fields * 3871146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3872146a8595SJ. Bruce Fields * 3873146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3874146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3875146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3876146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3877146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3878146a8595SJ. Bruce Fields * 3879146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3880146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3881146a8595SJ. Bruce Fields * to be NFS exported. 3882146a8595SJ. Bruce Fields */ 3883146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 38841da177e4SLinus Torvalds { 38851da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 38868de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38871da177e4SLinus Torvalds int error; 38881da177e4SLinus Torvalds 38891da177e4SLinus Torvalds if (!inode) 38901da177e4SLinus Torvalds return -ENOENT; 38911da177e4SLinus Torvalds 3892a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 38931da177e4SLinus Torvalds if (error) 38941da177e4SLinus Torvalds return error; 38951da177e4SLinus Torvalds 38961da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 38971da177e4SLinus Torvalds return -EXDEV; 38981da177e4SLinus Torvalds 38991da177e4SLinus Torvalds /* 39001da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 39011da177e4SLinus Torvalds */ 39021da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 39031da177e4SLinus Torvalds return -EPERM; 3904acfa4380SAl Viro if (!dir->i_op->link) 39051da177e4SLinus Torvalds return -EPERM; 39067e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 39071da177e4SLinus Torvalds return -EPERM; 39081da177e4SLinus Torvalds 39091da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 39101da177e4SLinus Torvalds if (error) 39111da177e4SLinus Torvalds return error; 39121da177e4SLinus Torvalds 39137e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3914aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3915f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3916aae8a97dSAneesh Kumar K.V error = -ENOENT; 39178de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 39188de52778SAl Viro error = -EMLINK; 3919146a8595SJ. Bruce Fields else { 3920146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 3921146a8595SJ. Bruce Fields if (!error) 39221da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 3923146a8595SJ. Bruce Fields } 3924f4e0c30cSAl Viro 3925f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 3926f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3927f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 3928f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3929f4e0c30cSAl Viro } 39307e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3931e31e14ecSStephen Smalley if (!error) 39327e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 39331da177e4SLinus Torvalds return error; 39341da177e4SLinus Torvalds } 39354d359507SAl Viro EXPORT_SYMBOL(vfs_link); 39361da177e4SLinus Torvalds 39371da177e4SLinus Torvalds /* 39381da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 39391da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 39401da177e4SLinus Torvalds * newname. --KAB 39411da177e4SLinus Torvalds * 39421da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 39431da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 39441da177e4SLinus Torvalds * and other special files. --ADM 39451da177e4SLinus Torvalds */ 39462e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 39472e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 39481da177e4SLinus Torvalds { 39491da177e4SLinus Torvalds struct dentry *new_dentry; 3950dae6ad8fSAl Viro struct path old_path, new_path; 3951146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 395211a7b371SAneesh Kumar K.V int how = 0; 39531da177e4SLinus Torvalds int error; 39541da177e4SLinus Torvalds 395511a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3956c04030e1SUlrich Drepper return -EINVAL; 395711a7b371SAneesh Kumar K.V /* 3958f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 3959f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 3960f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 396111a7b371SAneesh Kumar K.V */ 3962f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 3963f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 3964f0cc6ffbSLinus Torvalds return -ENOENT; 396511a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 3966f0cc6ffbSLinus Torvalds } 3967c04030e1SUlrich Drepper 396811a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 396911a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 3970442e31caSJeff Layton retry: 397111a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 39721da177e4SLinus Torvalds if (error) 39732ad94ae6SAl Viro return error; 39742ad94ae6SAl Viro 3975442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 3976442e31caSJeff Layton (how & LOOKUP_REVAL)); 39771da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 39786902d925SDave Hansen if (IS_ERR(new_dentry)) 3979dae6ad8fSAl Viro goto out; 3980dae6ad8fSAl Viro 3981dae6ad8fSAl Viro error = -EXDEV; 3982dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 3983dae6ad8fSAl Viro goto out_dput; 3984800179c9SKees Cook error = may_linkat(&old_path); 3985800179c9SKees Cook if (unlikely(error)) 3986800179c9SKees Cook goto out_dput; 3987dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 3988be6d3e56SKentaro Takeda if (error) 3989a8104a9fSAl Viro goto out_dput; 3990146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 399175c3f29dSDave Hansen out_dput: 3992921a1650SAl Viro done_path_create(&new_path, new_dentry); 3993146a8595SJ. Bruce Fields if (delegated_inode) { 3994146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3995d22e6338SOleg Drokin if (!error) { 3996d22e6338SOleg Drokin path_put(&old_path); 3997146a8595SJ. Bruce Fields goto retry; 3998146a8595SJ. Bruce Fields } 3999d22e6338SOleg Drokin } 4000442e31caSJeff Layton if (retry_estale(error, how)) { 4001d22e6338SOleg Drokin path_put(&old_path); 4002442e31caSJeff Layton how |= LOOKUP_REVAL; 4003442e31caSJeff Layton goto retry; 4004442e31caSJeff Layton } 40051da177e4SLinus Torvalds out: 40062d8f3038SAl Viro path_put(&old_path); 40071da177e4SLinus Torvalds 40081da177e4SLinus Torvalds return error; 40091da177e4SLinus Torvalds } 40101da177e4SLinus Torvalds 40113480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 40125590ff0dSUlrich Drepper { 4013c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 40145590ff0dSUlrich Drepper } 40155590ff0dSUlrich Drepper 4016bc27027aSMiklos Szeredi /** 4017bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4018bc27027aSMiklos Szeredi * @old_dir: parent of source 4019bc27027aSMiklos Szeredi * @old_dentry: source 4020bc27027aSMiklos Szeredi * @new_dir: parent of destination 4021bc27027aSMiklos Szeredi * @new_dentry: destination 4022bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4023520c8b16SMiklos Szeredi * @flags: rename flags 4024bc27027aSMiklos Szeredi * 4025bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4026bc27027aSMiklos Szeredi * 4027bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4028bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4029bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4030bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4031bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4032bc27027aSMiklos Szeredi * so. 4033bc27027aSMiklos Szeredi * 4034bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4035bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4036bc27027aSMiklos Szeredi * to be NFS exported. 4037bc27027aSMiklos Szeredi * 40381da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 40391da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 40401da177e4SLinus Torvalds * Problems: 4041d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 40421da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 40431da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4044a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 40451da177e4SLinus Torvalds * story. 40466cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 40476cedba89SJ. Bruce Fields * and source (if it is not a directory). 40481b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 40491da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 40501da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4051a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 40521da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 40531da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 40541da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4055a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 40561da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 40571da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 40581da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4059e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 40601b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 40611da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4062c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 40631da177e4SLinus Torvalds * locking]. 40641da177e4SLinus Torvalds */ 40651da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 40668e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4067520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 40681da177e4SLinus Torvalds { 40691da177e4SLinus Torvalds int error; 4070bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 407159b0df21SEric Paris const unsigned char *old_name; 4072bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4073bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4074da1ce067SMiklos Szeredi bool new_is_dir = false; 4075da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 40761da177e4SLinus Torvalds 4077bc27027aSMiklos Szeredi if (source == target) 40781da177e4SLinus Torvalds return 0; 40791da177e4SLinus Torvalds 40801da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 40811da177e4SLinus Torvalds if (error) 40821da177e4SLinus Torvalds return error; 40831da177e4SLinus Torvalds 4084da1ce067SMiklos Szeredi if (!target) { 4085a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4086da1ce067SMiklos Szeredi } else { 4087da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4088da1ce067SMiklos Szeredi 4089da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 40901da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4091da1ce067SMiklos Szeredi else 4092da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4093da1ce067SMiklos Szeredi } 40941da177e4SLinus Torvalds if (error) 40951da177e4SLinus Torvalds return error; 40961da177e4SLinus Torvalds 40977177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 40981da177e4SLinus Torvalds return -EPERM; 40991da177e4SLinus Torvalds 4100520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4101520c8b16SMiklos Szeredi return -EINVAL; 4102520c8b16SMiklos Szeredi 4103bc27027aSMiklos Szeredi /* 4104bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4105bc27027aSMiklos Szeredi * we'll need to flip '..'. 4106bc27027aSMiklos Szeredi */ 4107da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4108da1ce067SMiklos Szeredi if (is_dir) { 4109bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4110bc27027aSMiklos Szeredi if (error) 4111bc27027aSMiklos Szeredi return error; 4112bc27027aSMiklos Szeredi } 4113da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4114da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4115da1ce067SMiklos Szeredi if (error) 4116da1ce067SMiklos Szeredi return error; 4117da1ce067SMiklos Szeredi } 4118da1ce067SMiklos Szeredi } 41190eeca283SRobert Love 41200b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 41210b3974ebSMiklos Szeredi flags); 4122bc27027aSMiklos Szeredi if (error) 4123bc27027aSMiklos Szeredi return error; 4124bc27027aSMiklos Szeredi 4125bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4126bc27027aSMiklos Szeredi dget(new_dentry); 4127da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4128bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4129bc27027aSMiklos Szeredi else if (target) 4130bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4131bc27027aSMiklos Szeredi 4132bc27027aSMiklos Szeredi error = -EBUSY; 41337af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4134bc27027aSMiklos Szeredi goto out; 4135bc27027aSMiklos Szeredi 4136da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4137bc27027aSMiklos Szeredi error = -EMLINK; 4138da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4139bc27027aSMiklos Szeredi goto out; 4140da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4141da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4142da1ce067SMiklos Szeredi goto out; 4143da1ce067SMiklos Szeredi } 4144da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4145bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4146da1ce067SMiklos Szeredi if (!is_dir) { 4147bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4148bc27027aSMiklos Szeredi if (error) 4149bc27027aSMiklos Szeredi goto out; 4150da1ce067SMiklos Szeredi } 4151da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4152bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4153bc27027aSMiklos Szeredi if (error) 4154bc27027aSMiklos Szeredi goto out; 4155bc27027aSMiklos Szeredi } 41567177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4157520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4158520c8b16SMiklos Szeredi new_dir, new_dentry); 4159520c8b16SMiklos Szeredi } else { 41607177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4161520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4162520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4163520c8b16SMiklos Szeredi } 4164bc27027aSMiklos Szeredi if (error) 4165bc27027aSMiklos Szeredi goto out; 4166bc27027aSMiklos Szeredi 4167da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 41681da177e4SLinus Torvalds if (is_dir) 4169bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4170bc27027aSMiklos Szeredi dont_mount(new_dentry); 41718ed936b5SEric W. Biederman detach_mounts(new_dentry); 4172bc27027aSMiklos Szeredi } 4173da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4174da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4175bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4176da1ce067SMiklos Szeredi else 4177da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4178da1ce067SMiklos Szeredi } 4179bc27027aSMiklos Szeredi out: 4180da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4181bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4182bc27027aSMiklos Szeredi else if (target) 4183bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4184bc27027aSMiklos Szeredi dput(new_dentry); 4185da1ce067SMiklos Szeredi if (!error) { 4186123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4187da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4188da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4189da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4190da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4191da1ce067SMiklos Szeredi } 4192da1ce067SMiklos Szeredi } 41930eeca283SRobert Love fsnotify_oldname_free(old_name); 41940eeca283SRobert Love 41951da177e4SLinus Torvalds return error; 41961da177e4SLinus Torvalds } 41974d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 41981da177e4SLinus Torvalds 4199520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4200520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 42011da177e4SLinus Torvalds { 42021da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 42031da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 42041da177e4SLinus Torvalds struct dentry *trap; 42051da177e4SLinus Torvalds struct nameidata oldnd, newnd; 42068e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 420791a27b2aSJeff Layton struct filename *from; 420891a27b2aSJeff Layton struct filename *to; 4209c6a94284SJeff Layton unsigned int lookup_flags = 0; 4210c6a94284SJeff Layton bool should_retry = false; 42112ad94ae6SAl Viro int error; 4212520c8b16SMiklos Szeredi 42130d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4214da1ce067SMiklos Szeredi return -EINVAL; 4215da1ce067SMiklos Szeredi 42160d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 42170d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4218520c8b16SMiklos Szeredi return -EINVAL; 4219520c8b16SMiklos Szeredi 42200d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 42210d7a8555SMiklos Szeredi return -EPERM; 42220d7a8555SMiklos Szeredi 4223c6a94284SJeff Layton retry: 4224c6a94284SJeff Layton from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); 422591a27b2aSJeff Layton if (IS_ERR(from)) { 422691a27b2aSJeff Layton error = PTR_ERR(from); 42271da177e4SLinus Torvalds goto exit; 422891a27b2aSJeff Layton } 42291da177e4SLinus Torvalds 4230c6a94284SJeff Layton to = user_path_parent(newdfd, newname, &newnd, lookup_flags); 423191a27b2aSJeff Layton if (IS_ERR(to)) { 423291a27b2aSJeff Layton error = PTR_ERR(to); 42331da177e4SLinus Torvalds goto exit1; 423491a27b2aSJeff Layton } 42351da177e4SLinus Torvalds 42361da177e4SLinus Torvalds error = -EXDEV; 42374ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 42381da177e4SLinus Torvalds goto exit2; 42391da177e4SLinus Torvalds 42404ac91378SJan Blunck old_dir = oldnd.path.dentry; 42411da177e4SLinus Torvalds error = -EBUSY; 42421da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 42431da177e4SLinus Torvalds goto exit2; 42441da177e4SLinus Torvalds 42454ac91378SJan Blunck new_dir = newnd.path.dentry; 42460a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 42470a7c3937SMiklos Szeredi error = -EEXIST; 42481da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 42491da177e4SLinus Torvalds goto exit2; 42501da177e4SLinus Torvalds 4251c30dabfeSJan Kara error = mnt_want_write(oldnd.path.mnt); 4252c30dabfeSJan Kara if (error) 4253c30dabfeSJan Kara goto exit2; 4254c30dabfeSJan Kara 42550612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 42560612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 4257da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42584e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 42590612d9fbSOGAWA Hirofumi 42608e6d782cSJ. Bruce Fields retry_deleg: 42611da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 42621da177e4SLinus Torvalds 426349705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 42641da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 42651da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 42661da177e4SLinus Torvalds goto exit3; 42671da177e4SLinus Torvalds /* source must exist */ 42681da177e4SLinus Torvalds error = -ENOENT; 4269b18825a7SDavid Howells if (d_is_negative(old_dentry)) 42701da177e4SLinus Torvalds goto exit4; 427149705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 42721da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42731da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 42741da177e4SLinus Torvalds goto exit4; 42750a7c3937SMiklos Szeredi error = -EEXIST; 42760a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 42770a7c3937SMiklos Szeredi goto exit5; 4278da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4279da1ce067SMiklos Szeredi error = -ENOENT; 4280da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4281da1ce067SMiklos Szeredi goto exit5; 4282da1ce067SMiklos Szeredi 4283da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4284da1ce067SMiklos Szeredi error = -ENOTDIR; 4285da1ce067SMiklos Szeredi if (newnd.last.name[newnd.last.len]) 4286da1ce067SMiklos Szeredi goto exit5; 4287da1ce067SMiklos Szeredi } 4288da1ce067SMiklos Szeredi } 42890a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 42900a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 42910a7c3937SMiklos Szeredi error = -ENOTDIR; 42920a7c3937SMiklos Szeredi if (oldnd.last.name[oldnd.last.len]) 42930a7c3937SMiklos Szeredi goto exit5; 4294da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len]) 42950a7c3937SMiklos Szeredi goto exit5; 42960a7c3937SMiklos Szeredi } 42970a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 42980a7c3937SMiklos Szeredi error = -EINVAL; 42990a7c3937SMiklos Szeredi if (old_dentry == trap) 43000a7c3937SMiklos Szeredi goto exit5; 43011da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4302da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43031da177e4SLinus Torvalds error = -ENOTEMPTY; 43041da177e4SLinus Torvalds if (new_dentry == trap) 43051da177e4SLinus Torvalds goto exit5; 43061da177e4SLinus Torvalds 4307be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 43080b3974ebSMiklos Szeredi &newnd.path, new_dentry, flags); 4309be6d3e56SKentaro Takeda if (error) 4310c30dabfeSJan Kara goto exit5; 43111da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 43128e6d782cSJ. Bruce Fields new_dir->d_inode, new_dentry, 4313520c8b16SMiklos Szeredi &delegated_inode, flags); 43141da177e4SLinus Torvalds exit5: 43151da177e4SLinus Torvalds dput(new_dentry); 43161da177e4SLinus Torvalds exit4: 43171da177e4SLinus Torvalds dput(old_dentry); 43181da177e4SLinus Torvalds exit3: 43191da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 43208e6d782cSJ. Bruce Fields if (delegated_inode) { 43218e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 43228e6d782cSJ. Bruce Fields if (!error) 43238e6d782cSJ. Bruce Fields goto retry_deleg; 43248e6d782cSJ. Bruce Fields } 4325c30dabfeSJan Kara mnt_drop_write(oldnd.path.mnt); 43261da177e4SLinus Torvalds exit2: 4327c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4328c6a94284SJeff Layton should_retry = true; 43291d957f9bSJan Blunck path_put(&newnd.path); 43302ad94ae6SAl Viro putname(to); 43311da177e4SLinus Torvalds exit1: 43321d957f9bSJan Blunck path_put(&oldnd.path); 43331da177e4SLinus Torvalds putname(from); 4334c6a94284SJeff Layton if (should_retry) { 4335c6a94284SJeff Layton should_retry = false; 4336c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4337c6a94284SJeff Layton goto retry; 4338c6a94284SJeff Layton } 43392ad94ae6SAl Viro exit: 43401da177e4SLinus Torvalds return error; 43411da177e4SLinus Torvalds } 43421da177e4SLinus Torvalds 4343520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4344520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4345520c8b16SMiklos Szeredi { 4346520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4347520c8b16SMiklos Szeredi } 4348520c8b16SMiklos Szeredi 4349a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 43505590ff0dSUlrich Drepper { 4351520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43525590ff0dSUlrich Drepper } 43535590ff0dSUlrich Drepper 4354787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4355787fb6bcSMiklos Szeredi { 4356787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4357787fb6bcSMiklos Szeredi if (error) 4358787fb6bcSMiklos Szeredi return error; 4359787fb6bcSMiklos Szeredi 4360787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4361787fb6bcSMiklos Szeredi return -EPERM; 4362787fb6bcSMiklos Szeredi 4363787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4364787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4365787fb6bcSMiklos Szeredi } 4366787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4367787fb6bcSMiklos Szeredi 43685d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 43691da177e4SLinus Torvalds { 43705d826c84SAl Viro int len = PTR_ERR(link); 43711da177e4SLinus Torvalds if (IS_ERR(link)) 43721da177e4SLinus Torvalds goto out; 43731da177e4SLinus Torvalds 43741da177e4SLinus Torvalds len = strlen(link); 43751da177e4SLinus Torvalds if (len > (unsigned) buflen) 43761da177e4SLinus Torvalds len = buflen; 43771da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 43781da177e4SLinus Torvalds len = -EFAULT; 43791da177e4SLinus Torvalds out: 43801da177e4SLinus Torvalds return len; 43811da177e4SLinus Torvalds } 43825d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 43831da177e4SLinus Torvalds 43841da177e4SLinus Torvalds /* 43851da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 43861da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 43871da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 43881da177e4SLinus Torvalds */ 43891da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 43901da177e4SLinus Torvalds { 43911da177e4SLinus Torvalds struct nameidata nd; 4392cc314eefSLinus Torvalds void *cookie; 4393694a1764SMarcin Slusarz int res; 4394cc314eefSLinus Torvalds 43951da177e4SLinus Torvalds nd.depth = 0; 4396cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 4397694a1764SMarcin Slusarz if (IS_ERR(cookie)) 4398694a1764SMarcin Slusarz return PTR_ERR(cookie); 4399694a1764SMarcin Slusarz 44005d826c84SAl Viro res = readlink_copy(buffer, buflen, nd_get_link(&nd)); 44011da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 4402cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 4403694a1764SMarcin Slusarz return res; 44041da177e4SLinus Torvalds } 44054d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 44061da177e4SLinus Torvalds 44071da177e4SLinus Torvalds /* get the link contents into pagecache */ 44081da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 44091da177e4SLinus Torvalds { 4410ebd09abbSDuane Griffin char *kaddr; 44111da177e4SLinus Torvalds struct page *page; 44121da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4413090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 44141da177e4SLinus Torvalds if (IS_ERR(page)) 44156fe6900eSNick Piggin return (char*)page; 44161da177e4SLinus Torvalds *ppage = page; 4417ebd09abbSDuane Griffin kaddr = kmap(page); 4418ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4419ebd09abbSDuane Griffin return kaddr; 44201da177e4SLinus Torvalds } 44211da177e4SLinus Torvalds 44221da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44231da177e4SLinus Torvalds { 44241da177e4SLinus Torvalds struct page *page = NULL; 44255d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 44261da177e4SLinus Torvalds if (page) { 44271da177e4SLinus Torvalds kunmap(page); 44281da177e4SLinus Torvalds page_cache_release(page); 44291da177e4SLinus Torvalds } 44301da177e4SLinus Torvalds return res; 44311da177e4SLinus Torvalds } 44324d359507SAl Viro EXPORT_SYMBOL(page_readlink); 44331da177e4SLinus Torvalds 4434cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 44351da177e4SLinus Torvalds { 4436cc314eefSLinus Torvalds struct page *page = NULL; 44371da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 4438cc314eefSLinus Torvalds return page; 44391da177e4SLinus Torvalds } 44404d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 44411da177e4SLinus Torvalds 4442cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 44431da177e4SLinus Torvalds { 4444cc314eefSLinus Torvalds struct page *page = cookie; 4445cc314eefSLinus Torvalds 4446cc314eefSLinus Torvalds if (page) { 44471da177e4SLinus Torvalds kunmap(page); 44481da177e4SLinus Torvalds page_cache_release(page); 44491da177e4SLinus Torvalds } 44501da177e4SLinus Torvalds } 44514d359507SAl Viro EXPORT_SYMBOL(page_put_link); 44521da177e4SLinus Torvalds 445354566b2cSNick Piggin /* 445454566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 445554566b2cSNick Piggin */ 445654566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 44571da177e4SLinus Torvalds { 44581da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 44590adb25d2SKirill Korotaev struct page *page; 4460afddba49SNick Piggin void *fsdata; 4461beb497abSDmitriy Monakhov int err; 44621da177e4SLinus Torvalds char *kaddr; 446354566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 446454566b2cSNick Piggin if (nofs) 446554566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 44661da177e4SLinus Torvalds 44677e53cac4SNeilBrown retry: 4468afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 446954566b2cSNick Piggin flags, &page, &fsdata); 44701da177e4SLinus Torvalds if (err) 4471afddba49SNick Piggin goto fail; 4472afddba49SNick Piggin 4473e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 44741da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4475e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4476afddba49SNick Piggin 4477afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4478afddba49SNick Piggin page, fsdata); 44791da177e4SLinus Torvalds if (err < 0) 44801da177e4SLinus Torvalds goto fail; 4481afddba49SNick Piggin if (err < len-1) 4482afddba49SNick Piggin goto retry; 4483afddba49SNick Piggin 44841da177e4SLinus Torvalds mark_inode_dirty(inode); 44851da177e4SLinus Torvalds return 0; 44861da177e4SLinus Torvalds fail: 44871da177e4SLinus Torvalds return err; 44881da177e4SLinus Torvalds } 44894d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 44901da177e4SLinus Torvalds 44910adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 44920adb25d2SKirill Korotaev { 44930adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 449454566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 44950adb25d2SKirill Korotaev } 44964d359507SAl Viro EXPORT_SYMBOL(page_symlink); 44970adb25d2SKirill Korotaev 449892e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 44991da177e4SLinus Torvalds .readlink = generic_readlink, 45001da177e4SLinus Torvalds .follow_link = page_follow_link_light, 45011da177e4SLinus Torvalds .put_link = page_put_link, 45021da177e4SLinus Torvalds }; 45031da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4504