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 13391a27b2aSJeff Layton static 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 4901f55a6ecSAl Viro struct nameidata { 4911f55a6ecSAl Viro struct path path; 4921f55a6ecSAl Viro struct qstr last; 4931f55a6ecSAl Viro struct path root; 4941f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 4951f55a6ecSAl Viro unsigned int flags; 4961f55a6ecSAl Viro unsigned seq, m_seq; 4971f55a6ecSAl Viro int last_type; 4981f55a6ecSAl Viro unsigned depth; 4995e53084dSAl Viro struct file *base; 5001f55a6ecSAl Viro char *saved_names[MAX_NESTED_LINKS + 1]; 5011f55a6ecSAl Viro }; 5021f55a6ecSAl Viro 50319660af7SAl Viro /* 50431e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 50519660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 50619660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 50719660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 50819660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 50919660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 51019660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 51119660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 51231e6b01fSNick Piggin */ 51331e6b01fSNick Piggin 51431e6b01fSNick Piggin /** 51519660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 51619660af7SAl Viro * @nd: nameidata pathwalk data 51719660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 51839191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 51931e6b01fSNick Piggin * 52019660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 52119660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 52219660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 52331e6b01fSNick Piggin */ 52419660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 52531e6b01fSNick Piggin { 52631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 52731e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 52831e6b01fSNick Piggin 52931e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 530e5c832d5SLinus Torvalds 531e5c832d5SLinus Torvalds /* 53248a066e7SAl Viro * After legitimizing the bastards, terminate_walk() 53348a066e7SAl Viro * will do the right thing for non-RCU mode, and all our 53448a066e7SAl Viro * subsequent exit cases should rcu_read_unlock() 53548a066e7SAl Viro * before returning. Do vfsmount first; if dentry 53648a066e7SAl Viro * can't be legitimized, just set nd->path.dentry to NULL 53748a066e7SAl Viro * and rely on dput(NULL) being a no-op. 538e5c832d5SLinus Torvalds */ 53948a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 540e5c832d5SLinus Torvalds return -ECHILD; 541e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 54215570086SLinus Torvalds 54348a066e7SAl Viro if (!lockref_get_not_dead(&parent->d_lockref)) { 54448a066e7SAl Viro nd->path.dentry = NULL; 545d870b4a1SAl Viro goto out; 54648a066e7SAl Viro } 54748a066e7SAl Viro 54815570086SLinus Torvalds /* 54915570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 55015570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 55115570086SLinus Torvalds * 55215570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 55315570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 55415570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 55515570086SLinus Torvalds * since we checked the sequence number of the parent after we got 55615570086SLinus Torvalds * the child sequence number. So we know the parent must still 55715570086SLinus Torvalds * be valid if the child sequence number is still valid. 55815570086SLinus Torvalds */ 55919660af7SAl Viro if (!dentry) { 560e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 561e5c832d5SLinus Torvalds goto out; 56219660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 56319660af7SAl Viro } else { 564e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 565e5c832d5SLinus Torvalds goto out; 566e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 567e5c832d5SLinus Torvalds goto drop_dentry; 56819660af7SAl Viro } 569e5c832d5SLinus Torvalds 570e5c832d5SLinus Torvalds /* 571e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 572e5c832d5SLinus Torvalds * still valid and get it if required. 573e5c832d5SLinus Torvalds */ 574e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 575e5c832d5SLinus Torvalds spin_lock(&fs->lock); 576e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 577e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 57831e6b01fSNick Piggin path_get(&nd->root); 57931e6b01fSNick Piggin spin_unlock(&fs->lock); 58031e6b01fSNick Piggin } 58131e6b01fSNick Piggin 5828b61e74fSAl Viro rcu_read_unlock(); 58331e6b01fSNick Piggin return 0; 58419660af7SAl Viro 585e5c832d5SLinus Torvalds unlock_and_drop_dentry: 58631e6b01fSNick Piggin spin_unlock(&fs->lock); 587e5c832d5SLinus Torvalds drop_dentry: 5888b61e74fSAl Viro rcu_read_unlock(); 589e5c832d5SLinus Torvalds dput(dentry); 590d0d27277SLinus Torvalds goto drop_root_mnt; 591e5c832d5SLinus Torvalds out: 5928b61e74fSAl Viro rcu_read_unlock(); 593d0d27277SLinus Torvalds drop_root_mnt: 594d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 595d0d27277SLinus Torvalds nd->root.mnt = NULL; 59631e6b01fSNick Piggin return -ECHILD; 59731e6b01fSNick Piggin } 59831e6b01fSNick Piggin 5994ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 60034286d66SNick Piggin { 6014ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 60234286d66SNick Piggin } 60334286d66SNick Piggin 6049f1fafeeSAl Viro /** 6059f1fafeeSAl Viro * complete_walk - successful completion of path walk 6069f1fafeeSAl Viro * @nd: pointer nameidata 60739159de2SJeff Layton * 6089f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 6099f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 6109f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 6119f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 6129f1fafeeSAl Viro * need to drop nd->path. 61339159de2SJeff Layton */ 6149f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 61539159de2SJeff Layton { 61616c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 61739159de2SJeff Layton int status; 61839159de2SJeff Layton 6199f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 6209f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 6219f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6229f1fafeeSAl Viro nd->root.mnt = NULL; 62315570086SLinus Torvalds 62448a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { 6258b61e74fSAl Viro rcu_read_unlock(); 62648a066e7SAl Viro return -ECHILD; 62748a066e7SAl Viro } 628e5c832d5SLinus Torvalds if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 6298b61e74fSAl Viro rcu_read_unlock(); 63048a066e7SAl Viro mntput(nd->path.mnt); 6319f1fafeeSAl Viro return -ECHILD; 6329f1fafeeSAl Viro } 633e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 6348b61e74fSAl Viro rcu_read_unlock(); 635e5c832d5SLinus Torvalds dput(dentry); 63648a066e7SAl Viro mntput(nd->path.mnt); 637e5c832d5SLinus Torvalds return -ECHILD; 638e5c832d5SLinus Torvalds } 6398b61e74fSAl Viro rcu_read_unlock(); 6409f1fafeeSAl Viro } 6419f1fafeeSAl Viro 64216c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 64339159de2SJeff Layton return 0; 64439159de2SJeff Layton 645ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 64616c2cd71SAl Viro return 0; 64716c2cd71SAl Viro 648ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 64939159de2SJeff Layton if (status > 0) 65039159de2SJeff Layton return 0; 65139159de2SJeff Layton 65216c2cd71SAl Viro if (!status) 65339159de2SJeff Layton status = -ESTALE; 65416c2cd71SAl Viro 6559f1fafeeSAl Viro path_put(&nd->path); 65639159de2SJeff Layton return status; 65739159de2SJeff Layton } 65839159de2SJeff Layton 6592a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6602a737871SAl Viro { 661f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6622a737871SAl Viro } 6632a737871SAl Viro 6646de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6656de88d72SAl Viro 6667bd88377SAl Viro static __always_inline unsigned set_root_rcu(struct nameidata *nd) 66731e6b01fSNick Piggin { 66831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 6697bd88377SAl Viro unsigned seq, res; 670c28cc364SNick Piggin 671c28cc364SNick Piggin do { 672c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 67331e6b01fSNick Piggin nd->root = fs->root; 6747bd88377SAl Viro res = __read_seqcount_begin(&nd->root.dentry->d_seq); 675c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 6767bd88377SAl Viro return res; 67731e6b01fSNick Piggin } 67831e6b01fSNick Piggin 6791d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 680051d3812SIan Kent { 681051d3812SIan Kent dput(path->dentry); 6824ac91378SJan Blunck if (path->mnt != nd->path.mnt) 683051d3812SIan Kent mntput(path->mnt); 684051d3812SIan Kent } 685051d3812SIan Kent 6867b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6877b9337aaSNick Piggin struct nameidata *nd) 688051d3812SIan Kent { 68931e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6904ac91378SJan Blunck dput(nd->path.dentry); 69131e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6924ac91378SJan Blunck mntput(nd->path.mnt); 6939a229683SHuang Shijie } 69431e6b01fSNick Piggin nd->path.mnt = path->mnt; 6954ac91378SJan Blunck nd->path.dentry = path->dentry; 696051d3812SIan Kent } 697051d3812SIan Kent 698b5fb63c1SChristoph Hellwig /* 699b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 700b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 701b5fb63c1SChristoph Hellwig */ 702b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 703b5fb63c1SChristoph Hellwig { 704b5fb63c1SChristoph Hellwig path_put(&nd->path); 705b5fb63c1SChristoph Hellwig 706b5fb63c1SChristoph Hellwig nd->path = *path; 707b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 708b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 709b5fb63c1SChristoph Hellwig } 710b5fb63c1SChristoph Hellwig 7111f55a6ecSAl Viro void nd_set_link(struct nameidata *nd, char *path) 7121f55a6ecSAl Viro { 7131f55a6ecSAl Viro nd->saved_names[nd->depth] = path; 7141f55a6ecSAl Viro } 7151f55a6ecSAl Viro EXPORT_SYMBOL(nd_set_link); 7161f55a6ecSAl Viro 7171f55a6ecSAl Viro char *nd_get_link(struct nameidata *nd) 7181f55a6ecSAl Viro { 7191f55a6ecSAl Viro return nd->saved_names[nd->depth]; 7201f55a6ecSAl Viro } 7211f55a6ecSAl Viro EXPORT_SYMBOL(nd_get_link); 7221f55a6ecSAl Viro 723574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 724574197e0SAl Viro { 725574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 7266d7b5aaeSAl Viro if (inode->i_op->put_link) 727574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 728574197e0SAl Viro path_put(link); 729574197e0SAl Viro } 730574197e0SAl Viro 731561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 732561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 733800179c9SKees Cook 734800179c9SKees Cook /** 735800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 736800179c9SKees Cook * @link: The path of the symlink 73755852635SRandy Dunlap * @nd: nameidata pathwalk data 738800179c9SKees Cook * 739800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 740800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 741800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 742800179c9SKees Cook * processes from failing races against path names that may change out 743800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 744800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 745800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 746800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 747800179c9SKees Cook * 748800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 749800179c9SKees Cook */ 750800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 751800179c9SKees Cook { 752800179c9SKees Cook const struct inode *inode; 753800179c9SKees Cook const struct inode *parent; 754800179c9SKees Cook 755800179c9SKees Cook if (!sysctl_protected_symlinks) 756800179c9SKees Cook return 0; 757800179c9SKees Cook 758800179c9SKees Cook /* Allowed if owner and follower match. */ 759800179c9SKees Cook inode = link->dentry->d_inode; 76081abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 761800179c9SKees Cook return 0; 762800179c9SKees Cook 763800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 764800179c9SKees Cook parent = nd->path.dentry->d_inode; 765800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 766800179c9SKees Cook return 0; 767800179c9SKees Cook 768800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 76981abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 770800179c9SKees Cook return 0; 771800179c9SKees Cook 772ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 773800179c9SKees Cook path_put_conditional(link, nd); 774800179c9SKees Cook path_put(&nd->path); 775800179c9SKees Cook return -EACCES; 776800179c9SKees Cook } 777800179c9SKees Cook 778800179c9SKees Cook /** 779800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 780800179c9SKees Cook * @inode: the source inode to hardlink from 781800179c9SKees Cook * 782800179c9SKees Cook * Return false if at least one of the following conditions: 783800179c9SKees Cook * - inode is not a regular file 784800179c9SKees Cook * - inode is setuid 785800179c9SKees Cook * - inode is setgid and group-exec 786800179c9SKees Cook * - access failure for read and write 787800179c9SKees Cook * 788800179c9SKees Cook * Otherwise returns true. 789800179c9SKees Cook */ 790800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 791800179c9SKees Cook { 792800179c9SKees Cook umode_t mode = inode->i_mode; 793800179c9SKees Cook 794800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 795800179c9SKees Cook if (!S_ISREG(mode)) 796800179c9SKees Cook return false; 797800179c9SKees Cook 798800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 799800179c9SKees Cook if (mode & S_ISUID) 800800179c9SKees Cook return false; 801800179c9SKees Cook 802800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 803800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 804800179c9SKees Cook return false; 805800179c9SKees Cook 806800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 807800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 808800179c9SKees Cook return false; 809800179c9SKees Cook 810800179c9SKees Cook return true; 811800179c9SKees Cook } 812800179c9SKees Cook 813800179c9SKees Cook /** 814800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 815800179c9SKees Cook * @link: the source to hardlink from 816800179c9SKees Cook * 817800179c9SKees Cook * Block hardlink when all of: 818800179c9SKees Cook * - sysctl_protected_hardlinks enabled 819800179c9SKees Cook * - fsuid does not match inode 820800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 821800179c9SKees Cook * - not CAP_FOWNER 822800179c9SKees Cook * 823800179c9SKees Cook * Returns 0 if successful, -ve on error. 824800179c9SKees Cook */ 825800179c9SKees Cook static int may_linkat(struct path *link) 826800179c9SKees Cook { 827800179c9SKees Cook const struct cred *cred; 828800179c9SKees Cook struct inode *inode; 829800179c9SKees Cook 830800179c9SKees Cook if (!sysctl_protected_hardlinks) 831800179c9SKees Cook return 0; 832800179c9SKees Cook 833800179c9SKees Cook cred = current_cred(); 834800179c9SKees Cook inode = link->dentry->d_inode; 835800179c9SKees Cook 836800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 837800179c9SKees Cook * otherwise, it must be a safe source. 838800179c9SKees Cook */ 83981abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 840800179c9SKees Cook capable(CAP_FOWNER)) 841800179c9SKees Cook return 0; 842800179c9SKees Cook 843a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 844800179c9SKees Cook return -EPERM; 845800179c9SKees Cook } 846800179c9SKees Cook 847def4af30SAl Viro static __always_inline int 848574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 8491da177e4SLinus Torvalds { 8507b9337aaSNick Piggin struct dentry *dentry = link->dentry; 8516d7b5aaeSAl Viro int error; 8526d7b5aaeSAl Viro char *s; 8531da177e4SLinus Torvalds 854844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 855844a3917SAl Viro 8560e794589SAl Viro if (link->mnt == nd->path.mnt) 8570e794589SAl Viro mntget(link->mnt); 8580e794589SAl Viro 8596d7b5aaeSAl Viro error = -ELOOP; 8606d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 8616d7b5aaeSAl Viro goto out_put_nd_path; 8626d7b5aaeSAl Viro 863574197e0SAl Viro cond_resched(); 864574197e0SAl Viro current->total_link_count++; 865574197e0SAl Viro 86668ac1234SAl Viro touch_atime(link); 8671da177e4SLinus Torvalds nd_set_link(nd, NULL); 868cd4e91d3SAl Viro 86936f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 8706d7b5aaeSAl Viro if (error) 8716d7b5aaeSAl Viro goto out_put_nd_path; 87236f3b4f6SAl Viro 87386acdca1SAl Viro nd->last_type = LAST_BIND; 874def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 875def4af30SAl Viro error = PTR_ERR(*p); 8766d7b5aaeSAl Viro if (IS_ERR(*p)) 877408ef013SChristoph Hellwig goto out_put_nd_path; 8786d7b5aaeSAl Viro 879cc314eefSLinus Torvalds error = 0; 8806d7b5aaeSAl Viro s = nd_get_link(nd); 8816d7b5aaeSAl Viro if (s) { 882443ed254SAl Viro if (unlikely(IS_ERR(s))) { 883443ed254SAl Viro path_put(&nd->path); 884443ed254SAl Viro put_link(nd, link, *p); 885443ed254SAl Viro return PTR_ERR(s); 886443ed254SAl Viro } 887443ed254SAl Viro if (*s == '/') { 8887bd88377SAl Viro if (!nd->root.mnt) 889443ed254SAl Viro set_root(nd); 890443ed254SAl Viro path_put(&nd->path); 891443ed254SAl Viro nd->path = nd->root; 892443ed254SAl Viro path_get(&nd->root); 893443ed254SAl Viro nd->flags |= LOOKUP_JUMPED; 894443ed254SAl Viro } 895443ed254SAl Viro nd->inode = nd->path.dentry->d_inode; 896443ed254SAl Viro error = link_path_walk(s, nd); 8976d7b5aaeSAl Viro if (unlikely(error)) 8986d7b5aaeSAl Viro put_link(nd, link, *p); 899b5fb63c1SChristoph Hellwig } 9006d7b5aaeSAl Viro 9016d7b5aaeSAl Viro return error; 9026d7b5aaeSAl Viro 9036d7b5aaeSAl Viro out_put_nd_path: 90498f6ef64SArnd Bergmann *p = NULL; 9056d7b5aaeSAl Viro path_put(&nd->path); 9066d7b5aaeSAl Viro path_put(link); 9071da177e4SLinus Torvalds return error; 9081da177e4SLinus Torvalds } 9091da177e4SLinus Torvalds 91031e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 91131e6b01fSNick Piggin { 9120714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9130714a533SAl Viro struct mount *parent; 91431e6b01fSNick Piggin struct dentry *mountpoint; 91531e6b01fSNick Piggin 9160714a533SAl Viro parent = mnt->mnt_parent; 9170714a533SAl Viro if (&parent->mnt == path->mnt) 91831e6b01fSNick Piggin return 0; 919a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 92031e6b01fSNick Piggin path->dentry = mountpoint; 9210714a533SAl Viro path->mnt = &parent->mnt; 92231e6b01fSNick Piggin return 1; 92331e6b01fSNick Piggin } 92431e6b01fSNick Piggin 925f015f126SDavid Howells /* 926f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 927f015f126SDavid Howells * 928f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 929f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 930f015f126SDavid Howells * Up is towards /. 931f015f126SDavid Howells * 932f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 933f015f126SDavid Howells * root. 934f015f126SDavid Howells */ 935bab77ebfSAl Viro int follow_up(struct path *path) 9361da177e4SLinus Torvalds { 9370714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9380714a533SAl Viro struct mount *parent; 9391da177e4SLinus Torvalds struct dentry *mountpoint; 94099b7db7bSNick Piggin 94148a066e7SAl Viro read_seqlock_excl(&mount_lock); 9420714a533SAl Viro parent = mnt->mnt_parent; 9433c0a6163SAl Viro if (parent == mnt) { 94448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 9451da177e4SLinus Torvalds return 0; 9461da177e4SLinus Torvalds } 9470714a533SAl Viro mntget(&parent->mnt); 948a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 94948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 950bab77ebfSAl Viro dput(path->dentry); 951bab77ebfSAl Viro path->dentry = mountpoint; 952bab77ebfSAl Viro mntput(path->mnt); 9530714a533SAl Viro path->mnt = &parent->mnt; 9541da177e4SLinus Torvalds return 1; 9551da177e4SLinus Torvalds } 9564d359507SAl Viro EXPORT_SYMBOL(follow_up); 9571da177e4SLinus Torvalds 958b5c84bf6SNick Piggin /* 9599875cf80SDavid Howells * Perform an automount 9609875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9619875cf80SDavid Howells * were called with. 9621da177e4SLinus Torvalds */ 9639875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 9649875cf80SDavid Howells bool *need_mntput) 96531e6b01fSNick Piggin { 9669875cf80SDavid Howells struct vfsmount *mnt; 967ea5b778aSDavid Howells int err; 9689875cf80SDavid Howells 9699875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9709875cf80SDavid Howells return -EREMOTE; 9719875cf80SDavid Howells 9720ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9730ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9740ec26fd0SMiklos Szeredi * the name. 9750ec26fd0SMiklos Szeredi * 9760ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9775a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9780ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9790ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9800ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9810ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9825a30d8a2SDavid Howells */ 9835a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 984d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 9855a30d8a2SDavid Howells path->dentry->d_inode) 9869875cf80SDavid Howells return -EISDIR; 9870ec26fd0SMiklos Szeredi 9889875cf80SDavid Howells current->total_link_count++; 9899875cf80SDavid Howells if (current->total_link_count >= 40) 9909875cf80SDavid Howells return -ELOOP; 9919875cf80SDavid Howells 9929875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9939875cf80SDavid Howells if (IS_ERR(mnt)) { 9949875cf80SDavid Howells /* 9959875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 9969875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 9979875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 9989875cf80SDavid Howells * 9999875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 10009875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 10019875cf80SDavid Howells * the path is inaccessible and we should say so. 10029875cf80SDavid Howells */ 100349084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 10049875cf80SDavid Howells return -EREMOTE; 10059875cf80SDavid Howells return PTR_ERR(mnt); 100631e6b01fSNick Piggin } 1007ea5b778aSDavid Howells 10089875cf80SDavid Howells if (!mnt) /* mount collision */ 10099875cf80SDavid Howells return 0; 10109875cf80SDavid Howells 10118aef1884SAl Viro if (!*need_mntput) { 10128aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 10138aef1884SAl Viro mntget(path->mnt); 10148aef1884SAl Viro *need_mntput = true; 10158aef1884SAl Viro } 101619a167afSAl Viro err = finish_automount(mnt, path); 1017ea5b778aSDavid Howells 1018ea5b778aSDavid Howells switch (err) { 1019ea5b778aSDavid Howells case -EBUSY: 1020ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 102119a167afSAl Viro return 0; 1022ea5b778aSDavid Howells case 0: 10238aef1884SAl Viro path_put(path); 10249875cf80SDavid Howells path->mnt = mnt; 10259875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 10269875cf80SDavid Howells return 0; 102719a167afSAl Viro default: 102819a167afSAl Viro return err; 10299875cf80SDavid Howells } 103019a167afSAl Viro 1031ea5b778aSDavid Howells } 10329875cf80SDavid Howells 10339875cf80SDavid Howells /* 10349875cf80SDavid Howells * Handle a dentry that is managed in some way. 1035cc53ce53SDavid Howells * - Flagged for transit management (autofs) 10369875cf80SDavid Howells * - Flagged as mountpoint 10379875cf80SDavid Howells * - Flagged as automount point 10389875cf80SDavid Howells * 10399875cf80SDavid Howells * This may only be called in refwalk mode. 10409875cf80SDavid Howells * 10419875cf80SDavid Howells * Serialization is taken care of in namespace.c 10429875cf80SDavid Howells */ 10439875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 10449875cf80SDavid Howells { 10458aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10469875cf80SDavid Howells unsigned managed; 10479875cf80SDavid Howells bool need_mntput = false; 10488aef1884SAl Viro int ret = 0; 10499875cf80SDavid Howells 10509875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10519875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10529875cf80SDavid Howells * the components of that value change under us */ 10539875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10549875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10559875cf80SDavid Howells unlikely(managed != 0)) { 1056cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1057cc53ce53SDavid Howells * being held. */ 1058cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1059cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1060cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10611aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1062cc53ce53SDavid Howells if (ret < 0) 10638aef1884SAl Viro break; 1064cc53ce53SDavid Howells } 1065cc53ce53SDavid Howells 10669875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10679875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10689875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10699875cf80SDavid Howells if (mounted) { 10709875cf80SDavid Howells dput(path->dentry); 10719875cf80SDavid Howells if (need_mntput) 1072463ffb2eSAl Viro mntput(path->mnt); 1073463ffb2eSAl Viro path->mnt = mounted; 1074463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10759875cf80SDavid Howells need_mntput = true; 10769875cf80SDavid Howells continue; 1077463ffb2eSAl Viro } 1078463ffb2eSAl Viro 10799875cf80SDavid Howells /* Something is mounted on this dentry in another 10809875cf80SDavid Howells * namespace and/or whatever was mounted there in this 108148a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 108248a066e7SAl Viro * get it */ 10831da177e4SLinus Torvalds } 10849875cf80SDavid Howells 10859875cf80SDavid Howells /* Handle an automount point */ 10869875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 10879875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 10889875cf80SDavid Howells if (ret < 0) 10898aef1884SAl Viro break; 10909875cf80SDavid Howells continue; 10919875cf80SDavid Howells } 10929875cf80SDavid Howells 10939875cf80SDavid Howells /* We didn't change the current path point */ 10949875cf80SDavid Howells break; 10959875cf80SDavid Howells } 10968aef1884SAl Viro 10978aef1884SAl Viro if (need_mntput && path->mnt == mnt) 10988aef1884SAl Viro mntput(path->mnt); 10998aef1884SAl Viro if (ret == -EISDIR) 11008aef1884SAl Viro ret = 0; 1101a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 11021da177e4SLinus Torvalds } 11031da177e4SLinus Torvalds 1104cc53ce53SDavid Howells int follow_down_one(struct path *path) 11051da177e4SLinus Torvalds { 11061da177e4SLinus Torvalds struct vfsmount *mounted; 11071da177e4SLinus Torvalds 11081c755af4SAl Viro mounted = lookup_mnt(path); 11091da177e4SLinus Torvalds if (mounted) { 11109393bd07SAl Viro dput(path->dentry); 11119393bd07SAl Viro mntput(path->mnt); 11129393bd07SAl Viro path->mnt = mounted; 11139393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 11141da177e4SLinus Torvalds return 1; 11151da177e4SLinus Torvalds } 11161da177e4SLinus Torvalds return 0; 11171da177e4SLinus Torvalds } 11184d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 11191da177e4SLinus Torvalds 1120b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 112162a7375eSIan Kent { 1122b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1123b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 112462a7375eSIan Kent } 112562a7375eSIan Kent 11269875cf80SDavid Howells /* 1127287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1128287548e4SAl Viro * we meet a managed dentry that would need blocking. 11299875cf80SDavid Howells */ 11309875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1131287548e4SAl Viro struct inode **inode) 11329875cf80SDavid Howells { 113362a7375eSIan Kent for (;;) { 1134c7105365SAl Viro struct mount *mounted; 113562a7375eSIan Kent /* 113662a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 113762a7375eSIan Kent * that wants to block transit. 113862a7375eSIan Kent */ 1139b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1140b8faf035SNeilBrown case -ECHILD: 1141b8faf035SNeilBrown default: 1142ab90911fSDavid Howells return false; 1143b8faf035SNeilBrown case -EISDIR: 1144b8faf035SNeilBrown return true; 1145b8faf035SNeilBrown case 0: 1146b8faf035SNeilBrown break; 1147b8faf035SNeilBrown } 114862a7375eSIan Kent 114962a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1150b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 115162a7375eSIan Kent 1152474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 11539875cf80SDavid Howells if (!mounted) 11549875cf80SDavid Howells break; 1155c7105365SAl Viro path->mnt = &mounted->mnt; 1156c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1157a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 11589875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 115959430262SLinus Torvalds /* 116059430262SLinus Torvalds * Update the inode too. We don't need to re-check the 116159430262SLinus Torvalds * dentry sequence number here after this d_inode read, 116259430262SLinus Torvalds * because a mount-point is always pinned. 116359430262SLinus Torvalds */ 116459430262SLinus Torvalds *inode = path->dentry->d_inode; 11659875cf80SDavid Howells } 1166f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1167b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1168287548e4SAl Viro } 1169287548e4SAl Viro 117031e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 117131e6b01fSNick Piggin { 11724023bfc9SAl Viro struct inode *inode = nd->inode; 11737bd88377SAl Viro if (!nd->root.mnt) 117431e6b01fSNick Piggin set_root_rcu(nd); 117531e6b01fSNick Piggin 117631e6b01fSNick Piggin while (1) { 117731e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 117831e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 117931e6b01fSNick Piggin break; 118031e6b01fSNick Piggin } 118131e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 118231e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 118331e6b01fSNick Piggin struct dentry *parent = old->d_parent; 118431e6b01fSNick Piggin unsigned seq; 118531e6b01fSNick Piggin 11864023bfc9SAl Viro inode = parent->d_inode; 118731e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 118831e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1189ef7562d5SAl Viro goto failed; 119031e6b01fSNick Piggin nd->path.dentry = parent; 119131e6b01fSNick Piggin nd->seq = seq; 119231e6b01fSNick Piggin break; 119331e6b01fSNick Piggin } 119431e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 119531e6b01fSNick Piggin break; 11964023bfc9SAl Viro inode = nd->path.dentry->d_inode; 119731e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 119831e6b01fSNick Piggin } 1199b37199e6SAl Viro while (d_mountpoint(nd->path.dentry)) { 1200b37199e6SAl Viro struct mount *mounted; 1201b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1202b37199e6SAl Viro if (!mounted) 1203b37199e6SAl Viro break; 1204b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1205b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 12064023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1207b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1208f5be3e29SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1209b37199e6SAl Viro goto failed; 1210b37199e6SAl Viro } 12114023bfc9SAl Viro nd->inode = inode; 121231e6b01fSNick Piggin return 0; 1213ef7562d5SAl Viro 1214ef7562d5SAl Viro failed: 1215ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 12165b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1217ef7562d5SAl Viro nd->root.mnt = NULL; 12188b61e74fSAl Viro rcu_read_unlock(); 1219ef7562d5SAl Viro return -ECHILD; 122031e6b01fSNick Piggin } 122131e6b01fSNick Piggin 12229875cf80SDavid Howells /* 1223cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1224cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1225cc53ce53SDavid Howells * caller is permitted to proceed or not. 1226cc53ce53SDavid Howells */ 12277cc90cc3SAl Viro int follow_down(struct path *path) 1228cc53ce53SDavid Howells { 1229cc53ce53SDavid Howells unsigned managed; 1230cc53ce53SDavid Howells int ret; 1231cc53ce53SDavid Howells 1232cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1233cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1234cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1235cc53ce53SDavid Howells * being held. 1236cc53ce53SDavid Howells * 1237cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1238cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1239cc53ce53SDavid Howells * other than its daemon the right to mount on its 1240cc53ce53SDavid Howells * superstructure. 1241cc53ce53SDavid Howells * 1242cc53ce53SDavid Howells * The filesystem may sleep at this point. 1243cc53ce53SDavid Howells */ 1244cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1245cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1246cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1247ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 12481aed3e42SAl Viro path->dentry, false); 1249cc53ce53SDavid Howells if (ret < 0) 1250cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1251cc53ce53SDavid Howells } 1252cc53ce53SDavid Howells 1253cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1254cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1255cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1256cc53ce53SDavid Howells if (!mounted) 1257cc53ce53SDavid Howells break; 1258cc53ce53SDavid Howells dput(path->dentry); 1259cc53ce53SDavid Howells mntput(path->mnt); 1260cc53ce53SDavid Howells path->mnt = mounted; 1261cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1262cc53ce53SDavid Howells continue; 1263cc53ce53SDavid Howells } 1264cc53ce53SDavid Howells 1265cc53ce53SDavid Howells /* Don't handle automount points here */ 1266cc53ce53SDavid Howells break; 1267cc53ce53SDavid Howells } 1268cc53ce53SDavid Howells return 0; 1269cc53ce53SDavid Howells } 12704d359507SAl Viro EXPORT_SYMBOL(follow_down); 1271cc53ce53SDavid Howells 1272cc53ce53SDavid Howells /* 12739875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12749875cf80SDavid Howells */ 12759875cf80SDavid Howells static void follow_mount(struct path *path) 12769875cf80SDavid Howells { 12779875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12789875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12799875cf80SDavid Howells if (!mounted) 12809875cf80SDavid Howells break; 12819875cf80SDavid Howells dput(path->dentry); 12829875cf80SDavid Howells mntput(path->mnt); 12839875cf80SDavid Howells path->mnt = mounted; 12849875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12859875cf80SDavid Howells } 12869875cf80SDavid Howells } 12879875cf80SDavid Howells 128831e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 12891da177e4SLinus Torvalds { 12907bd88377SAl Viro if (!nd->root.mnt) 12912a737871SAl Viro set_root(nd); 1292e518ddb7SAndreas Mohr 12931da177e4SLinus Torvalds while(1) { 12944ac91378SJan Blunck struct dentry *old = nd->path.dentry; 12951da177e4SLinus Torvalds 12962a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 12972a737871SAl Viro nd->path.mnt == nd->root.mnt) { 12981da177e4SLinus Torvalds break; 12991da177e4SLinus Torvalds } 13004ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 13013088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 13023088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 13031da177e4SLinus Torvalds dput(old); 13041da177e4SLinus Torvalds break; 13051da177e4SLinus Torvalds } 13063088dd70SAl Viro if (!follow_up(&nd->path)) 13071da177e4SLinus Torvalds break; 13081da177e4SLinus Torvalds } 130979ed0226SAl Viro follow_mount(&nd->path); 131031e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds /* 1314bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1315bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1316bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1317bad61189SMiklos Szeredi * 1318bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1319baa03890SNick Piggin */ 1320bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1321201f956eSAl Viro unsigned int flags, bool *need_lookup) 1322baa03890SNick Piggin { 1323baa03890SNick Piggin struct dentry *dentry; 1324bad61189SMiklos Szeredi int error; 1325baa03890SNick Piggin 1326bad61189SMiklos Szeredi *need_lookup = false; 1327bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1328bad61189SMiklos Szeredi if (dentry) { 132939e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1330201f956eSAl Viro error = d_revalidate(dentry, flags); 1331bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1332bad61189SMiklos Szeredi if (error < 0) { 1333bad61189SMiklos Szeredi dput(dentry); 1334bad61189SMiklos Szeredi return ERR_PTR(error); 13355542aa2fSEric W. Biederman } else { 13365542aa2fSEric W. Biederman d_invalidate(dentry); 1337bad61189SMiklos Szeredi dput(dentry); 1338bad61189SMiklos Szeredi dentry = NULL; 1339bad61189SMiklos Szeredi } 1340bad61189SMiklos Szeredi } 1341bad61189SMiklos Szeredi } 1342bad61189SMiklos Szeredi } 1343baa03890SNick Piggin 1344bad61189SMiklos Szeredi if (!dentry) { 1345bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1346baa03890SNick Piggin if (unlikely(!dentry)) 1347baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1348baa03890SNick Piggin 1349bad61189SMiklos Szeredi *need_lookup = true; 1350baa03890SNick Piggin } 1351baa03890SNick Piggin return dentry; 1352baa03890SNick Piggin } 1353baa03890SNick Piggin 1354baa03890SNick Piggin /* 135513a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 135613a2c3beSJ. Bruce Fields * unhashed. 1357bad61189SMiklos Szeredi * 1358bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 135944396f4bSJosef Bacik */ 1360bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 136172bd866aSAl Viro unsigned int flags) 136244396f4bSJosef Bacik { 136344396f4bSJosef Bacik struct dentry *old; 136444396f4bSJosef Bacik 136544396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1366bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1367e188dc02SMiklos Szeredi dput(dentry); 136844396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1369e188dc02SMiklos Szeredi } 137044396f4bSJosef Bacik 137172bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 137244396f4bSJosef Bacik if (unlikely(old)) { 137344396f4bSJosef Bacik dput(dentry); 137444396f4bSJosef Bacik dentry = old; 137544396f4bSJosef Bacik } 137644396f4bSJosef Bacik return dentry; 137744396f4bSJosef Bacik } 137844396f4bSJosef Bacik 1379a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 138072bd866aSAl Viro struct dentry *base, unsigned int flags) 1381a3255546SAl Viro { 1382bad61189SMiklos Szeredi bool need_lookup; 1383a3255546SAl Viro struct dentry *dentry; 1384a3255546SAl Viro 138572bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1386bad61189SMiklos Szeredi if (!need_lookup) 1387a3255546SAl Viro return dentry; 1388bad61189SMiklos Szeredi 138972bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1390a3255546SAl Viro } 1391a3255546SAl Viro 139244396f4bSJosef Bacik /* 13931da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 13941da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 13951da177e4SLinus Torvalds * It _is_ time-critical. 13961da177e4SLinus Torvalds */ 1397e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 139831e6b01fSNick Piggin struct path *path, struct inode **inode) 13991da177e4SLinus Torvalds { 14004ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 140131e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 14025a18fff2SAl Viro int need_reval = 1; 14035a18fff2SAl Viro int status = 1; 14049875cf80SDavid Howells int err; 14059875cf80SDavid Howells 14063cac260aSAl Viro /* 1407b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1408b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1409b04f784eSNick Piggin * do the non-racy lookup, below. 1410b04f784eSNick Piggin */ 141131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 141231e6b01fSNick Piggin unsigned seq; 1413da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 14145a18fff2SAl Viro if (!dentry) 14155a18fff2SAl Viro goto unlazy; 14165a18fff2SAl Viro 141712f8ad4bSLinus Torvalds /* 141812f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 141912f8ad4bSLinus Torvalds * the dentry name information from lookup. 142012f8ad4bSLinus Torvalds */ 142112f8ad4bSLinus Torvalds *inode = dentry->d_inode; 142212f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 142312f8ad4bSLinus Torvalds return -ECHILD; 142412f8ad4bSLinus Torvalds 142512f8ad4bSLinus Torvalds /* 142612f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 142712f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 142812f8ad4bSLinus Torvalds * 142912f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 143012f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 143112f8ad4bSLinus Torvalds */ 143231e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 143331e6b01fSNick Piggin return -ECHILD; 143431e6b01fSNick Piggin nd->seq = seq; 14355a18fff2SAl Viro 143624643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 14374ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14385a18fff2SAl Viro if (unlikely(status <= 0)) { 14395a18fff2SAl Viro if (status != -ECHILD) 14405a18fff2SAl Viro need_reval = 0; 14415a18fff2SAl Viro goto unlazy; 14425a18fff2SAl Viro } 144324643087SAl Viro } 144431e6b01fSNick Piggin path->mnt = mnt; 144531e6b01fSNick Piggin path->dentry = dentry; 1446b8faf035SNeilBrown if (likely(__follow_mount_rcu(nd, path, inode))) 14479875cf80SDavid Howells return 0; 14485a18fff2SAl Viro unlazy: 144919660af7SAl Viro if (unlazy_walk(nd, dentry)) 14505a18fff2SAl Viro return -ECHILD; 14515a18fff2SAl Viro } else { 1452e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 145324643087SAl Viro } 14545a18fff2SAl Viro 145581e6f520SAl Viro if (unlikely(!dentry)) 145681e6f520SAl Viro goto need_lookup; 14575a18fff2SAl Viro 14585a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14594ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14605a18fff2SAl Viro if (unlikely(status <= 0)) { 14615a18fff2SAl Viro if (status < 0) { 14625a18fff2SAl Viro dput(dentry); 14635a18fff2SAl Viro return status; 14645a18fff2SAl Viro } 14655542aa2fSEric W. Biederman d_invalidate(dentry); 14665a18fff2SAl Viro dput(dentry); 146781e6f520SAl Viro goto need_lookup; 14685a18fff2SAl Viro } 1469697f514dSMiklos Szeredi 14701da177e4SLinus Torvalds path->mnt = mnt; 14711da177e4SLinus Torvalds path->dentry = dentry; 14729875cf80SDavid Howells err = follow_managed(path, nd->flags); 147389312214SIan Kent if (unlikely(err < 0)) { 147489312214SIan Kent path_put_conditional(path, nd); 14759875cf80SDavid Howells return err; 147689312214SIan Kent } 1477a3fbbde7SAl Viro if (err) 1478a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 147931e6b01fSNick Piggin *inode = path->dentry->d_inode; 14801da177e4SLinus Torvalds return 0; 148181e6f520SAl Viro 148281e6f520SAl Viro need_lookup: 1483697f514dSMiklos Szeredi return 1; 1484697f514dSMiklos Szeredi } 1485697f514dSMiklos Szeredi 1486697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1487cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1488697f514dSMiklos Szeredi { 1489697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1490697f514dSMiklos Szeredi int err; 1491697f514dSMiklos Szeredi 1492697f514dSMiklos Szeredi parent = nd->path.dentry; 149381e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 149481e6f520SAl Viro 149581e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1496cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 149781e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 149881e6f520SAl Viro if (IS_ERR(dentry)) 149981e6f520SAl Viro return PTR_ERR(dentry); 1500697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1501697f514dSMiklos Szeredi path->dentry = dentry; 1502697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1503697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1504697f514dSMiklos Szeredi path_put_conditional(path, nd); 1505697f514dSMiklos Szeredi return err; 1506697f514dSMiklos Szeredi } 1507697f514dSMiklos Szeredi if (err) 1508697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1509697f514dSMiklos Szeredi return 0; 15101da177e4SLinus Torvalds } 15111da177e4SLinus Torvalds 151252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 151352094c8aSAl Viro { 151452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 15154ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 151652094c8aSAl Viro if (err != -ECHILD) 151752094c8aSAl Viro return err; 151819660af7SAl Viro if (unlazy_walk(nd, NULL)) 151952094c8aSAl Viro return -ECHILD; 152052094c8aSAl Viro } 15214ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 152252094c8aSAl Viro } 152352094c8aSAl Viro 15249856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 15259856fa1bSAl Viro { 15269856fa1bSAl Viro if (type == LAST_DOTDOT) { 15279856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 15289856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 15299856fa1bSAl Viro return -ECHILD; 15309856fa1bSAl Viro } else 15319856fa1bSAl Viro follow_dotdot(nd); 15329856fa1bSAl Viro } 15339856fa1bSAl Viro return 0; 15349856fa1bSAl Viro } 15359856fa1bSAl Viro 1536951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1537951361f9SAl Viro { 1538951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1539951361f9SAl Viro path_put(&nd->path); 1540951361f9SAl Viro } else { 1541951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 15425b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1543951361f9SAl Viro nd->root.mnt = NULL; 15448b61e74fSAl Viro rcu_read_unlock(); 1545951361f9SAl Viro } 1546951361f9SAl Viro } 1547951361f9SAl Viro 15483ddcd056SLinus Torvalds /* 15493ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15503ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15513ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15523ddcd056SLinus Torvalds * for the common case. 15533ddcd056SLinus Torvalds */ 1554b18825a7SDavid Howells static inline int should_follow_link(struct dentry *dentry, int follow) 15553ddcd056SLinus Torvalds { 1556b18825a7SDavid Howells return unlikely(d_is_symlink(dentry)) ? follow : 0; 15573ddcd056SLinus Torvalds } 15583ddcd056SLinus Torvalds 1559ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 156021b9b073SAl Viro int follow) 1561ce57dfc1SAl Viro { 1562ce57dfc1SAl Viro struct inode *inode; 1563ce57dfc1SAl Viro int err; 1564ce57dfc1SAl Viro /* 1565ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1566ce57dfc1SAl Viro * to be able to know about the current root directory and 1567ce57dfc1SAl Viro * parent relationships. 1568ce57dfc1SAl Viro */ 156921b9b073SAl Viro if (unlikely(nd->last_type != LAST_NORM)) 157021b9b073SAl Viro return handle_dots(nd, nd->last_type); 1571e97cdc87SAl Viro err = lookup_fast(nd, path, &inode); 1572ce57dfc1SAl Viro if (unlikely(err)) { 1573697f514dSMiklos Szeredi if (err < 0) 1574697f514dSMiklos Szeredi goto out_err; 1575697f514dSMiklos Szeredi 1576cc2a5271SAl Viro err = lookup_slow(nd, path); 1577697f514dSMiklos Szeredi if (err < 0) 1578697f514dSMiklos Szeredi goto out_err; 1579697f514dSMiklos Szeredi 1580697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1581ce57dfc1SAl Viro } 1582697f514dSMiklos Szeredi err = -ENOENT; 158322213318SAl Viro if (!inode || d_is_negative(path->dentry)) 1584697f514dSMiklos Szeredi goto out_path_put; 1585697f514dSMiklos Szeredi 1586b18825a7SDavid Howells if (should_follow_link(path->dentry, follow)) { 158719660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 158819660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 1589697f514dSMiklos Szeredi err = -ECHILD; 1590697f514dSMiklos Szeredi goto out_err; 159119660af7SAl Viro } 159219660af7SAl Viro } 1593ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1594ce57dfc1SAl Viro return 1; 1595ce57dfc1SAl Viro } 1596ce57dfc1SAl Viro path_to_nameidata(path, nd); 1597ce57dfc1SAl Viro nd->inode = inode; 1598ce57dfc1SAl Viro return 0; 1599697f514dSMiklos Szeredi 1600697f514dSMiklos Szeredi out_path_put: 1601697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1602697f514dSMiklos Szeredi out_err: 1603697f514dSMiklos Szeredi terminate_walk(nd); 1604697f514dSMiklos Szeredi return err; 1605ce57dfc1SAl Viro } 1606ce57dfc1SAl Viro 16071da177e4SLinus Torvalds /* 1608b356379aSAl Viro * This limits recursive symlink follows to 8, while 1609b356379aSAl Viro * limiting consecutive symlinks to 40. 1610b356379aSAl Viro * 1611b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1612b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1613b356379aSAl Viro */ 1614b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1615b356379aSAl Viro { 1616b356379aSAl Viro int res; 1617b356379aSAl Viro 1618b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1619b356379aSAl Viro path_put_conditional(path, nd); 1620b356379aSAl Viro path_put(&nd->path); 1621b356379aSAl Viro return -ELOOP; 1622b356379aSAl Viro } 16231a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1624b356379aSAl Viro 1625b356379aSAl Viro nd->depth++; 1626b356379aSAl Viro current->link_count++; 1627b356379aSAl Viro 1628b356379aSAl Viro do { 1629b356379aSAl Viro struct path link = *path; 1630b356379aSAl Viro void *cookie; 1631574197e0SAl Viro 1632574197e0SAl Viro res = follow_link(&link, nd, &cookie); 16336d7b5aaeSAl Viro if (res) 16346d7b5aaeSAl Viro break; 163521b9b073SAl Viro res = walk_component(nd, path, LOOKUP_FOLLOW); 1636574197e0SAl Viro put_link(nd, &link, cookie); 1637b356379aSAl Viro } while (res > 0); 1638b356379aSAl Viro 1639b356379aSAl Viro current->link_count--; 1640b356379aSAl Viro nd->depth--; 1641b356379aSAl Viro return res; 1642b356379aSAl Viro } 1643b356379aSAl Viro 1644b356379aSAl Viro /* 1645bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1646bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1647bfcfaa77SLinus Torvalds * 1648bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1649bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1650bfcfaa77SLinus Torvalds * fast. 1651bfcfaa77SLinus Torvalds * 1652bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1653bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1654bfcfaa77SLinus Torvalds * crossing operation. 1655bfcfaa77SLinus Torvalds * 1656bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1657bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1658bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1659bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1660bfcfaa77SLinus Torvalds */ 1661bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1662bfcfaa77SLinus Torvalds 1663f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1664bfcfaa77SLinus Torvalds 1665f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1666bfcfaa77SLinus Torvalds 1667bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1668bfcfaa77SLinus Torvalds { 166999d263d4SLinus Torvalds return hash_64(hash, 32); 1670bfcfaa77SLinus Torvalds } 1671bfcfaa77SLinus Torvalds 1672bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1673bfcfaa77SLinus Torvalds 1674bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1675bfcfaa77SLinus Torvalds 1676bfcfaa77SLinus Torvalds #endif 1677bfcfaa77SLinus Torvalds 1678bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1679bfcfaa77SLinus Torvalds { 1680bfcfaa77SLinus Torvalds unsigned long a, mask; 1681bfcfaa77SLinus Torvalds unsigned long hash = 0; 1682bfcfaa77SLinus Torvalds 1683bfcfaa77SLinus Torvalds for (;;) { 1684e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1685bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1686bfcfaa77SLinus Torvalds break; 1687bfcfaa77SLinus Torvalds hash += a; 1688f132c5beSAl Viro hash *= 9; 1689bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1690bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1691bfcfaa77SLinus Torvalds if (!len) 1692bfcfaa77SLinus Torvalds goto done; 1693bfcfaa77SLinus Torvalds } 1694a5c21dceSWill Deacon mask = bytemask_from_count(len); 1695bfcfaa77SLinus Torvalds hash += mask & a; 1696bfcfaa77SLinus Torvalds done: 1697bfcfaa77SLinus Torvalds return fold_hash(hash); 1698bfcfaa77SLinus Torvalds } 1699bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1700bfcfaa77SLinus Torvalds 1701bfcfaa77SLinus Torvalds /* 1702bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1703d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1704bfcfaa77SLinus Torvalds */ 1705d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1706bfcfaa77SLinus Torvalds { 170736126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 170836126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1709bfcfaa77SLinus Torvalds 1710bfcfaa77SLinus Torvalds hash = a = 0; 1711bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1712bfcfaa77SLinus Torvalds do { 1713bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1714bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1715e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 171636126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 171736126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1718bfcfaa77SLinus Torvalds 171936126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 172036126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 172136126f8fSLinus Torvalds 172236126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 172336126f8fSLinus Torvalds 172436126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 17259226b5b4SLinus Torvalds len += find_zero(mask); 1726d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1727bfcfaa77SLinus Torvalds } 1728bfcfaa77SLinus Torvalds 1729bfcfaa77SLinus Torvalds #else 1730bfcfaa77SLinus Torvalds 17310145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 17320145acc2SLinus Torvalds { 17330145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 17340145acc2SLinus Torvalds while (len--) 17350145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 17360145acc2SLinus Torvalds return end_name_hash(hash); 17370145acc2SLinus Torvalds } 1738ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 17390145acc2SLinus Torvalds 17403ddcd056SLinus Torvalds /* 1741200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1742200e9ef7SLinus Torvalds * one character. 1743200e9ef7SLinus Torvalds */ 1744d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1745200e9ef7SLinus Torvalds { 1746200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1747200e9ef7SLinus Torvalds unsigned long len = 0, c; 1748200e9ef7SLinus Torvalds 1749200e9ef7SLinus Torvalds c = (unsigned char)*name; 1750200e9ef7SLinus Torvalds do { 1751200e9ef7SLinus Torvalds len++; 1752200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1753200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1754200e9ef7SLinus Torvalds } while (c && c != '/'); 1755d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1756200e9ef7SLinus Torvalds } 1757200e9ef7SLinus Torvalds 1758bfcfaa77SLinus Torvalds #endif 1759bfcfaa77SLinus Torvalds 1760200e9ef7SLinus Torvalds /* 17611da177e4SLinus Torvalds * Name resolution. 1762ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1763ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17641da177e4SLinus Torvalds * 1765ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1766ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17671da177e4SLinus Torvalds */ 17686de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17691da177e4SLinus Torvalds { 17701da177e4SLinus Torvalds struct path next; 17711da177e4SLinus Torvalds int err; 17721da177e4SLinus Torvalds 17731da177e4SLinus Torvalds while (*name=='/') 17741da177e4SLinus Torvalds name++; 17751da177e4SLinus Torvalds if (!*name) 1776086e183aSAl Viro return 0; 17771da177e4SLinus Torvalds 17781da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17791da177e4SLinus Torvalds for(;;) { 1780d6bb3e90SLinus Torvalds u64 hash_len; 1781fe479a58SAl Viro int type; 17821da177e4SLinus Torvalds 178352094c8aSAl Viro err = may_lookup(nd); 17841da177e4SLinus Torvalds if (err) 17851da177e4SLinus Torvalds break; 17861da177e4SLinus Torvalds 1787d6bb3e90SLinus Torvalds hash_len = hash_name(name); 17881da177e4SLinus Torvalds 1789fe479a58SAl Viro type = LAST_NORM; 1790d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1791fe479a58SAl Viro case 2: 1792200e9ef7SLinus Torvalds if (name[1] == '.') { 1793fe479a58SAl Viro type = LAST_DOTDOT; 179416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 179516c2cd71SAl Viro } 1796fe479a58SAl Viro break; 1797fe479a58SAl Viro case 1: 1798fe479a58SAl Viro type = LAST_DOT; 1799fe479a58SAl Viro } 18005a202bcdSAl Viro if (likely(type == LAST_NORM)) { 18015a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 180216c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 18035a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1804a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1805da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 18065a202bcdSAl Viro if (err < 0) 18075a202bcdSAl Viro break; 1808d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1809d6bb3e90SLinus Torvalds name = this.name; 18105a202bcdSAl Viro } 18115a202bcdSAl Viro } 1812fe479a58SAl Viro 1813d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1814d6bb3e90SLinus Torvalds nd->last.name = name; 18155f4a6a69SAl Viro nd->last_type = type; 18165f4a6a69SAl Viro 1817d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1818d6bb3e90SLinus Torvalds if (!*name) 18195f4a6a69SAl Viro return 0; 1820200e9ef7SLinus Torvalds /* 1821200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1822200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1823200e9ef7SLinus Torvalds */ 1824200e9ef7SLinus Torvalds do { 1825d6bb3e90SLinus Torvalds name++; 1826d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 1827d6bb3e90SLinus Torvalds if (!*name) 18285f4a6a69SAl Viro return 0; 18295f4a6a69SAl Viro 183021b9b073SAl Viro err = walk_component(nd, &next, LOOKUP_FOLLOW); 1831ce57dfc1SAl Viro if (err < 0) 1832ce57dfc1SAl Viro return err; 1833fe479a58SAl Viro 1834ce57dfc1SAl Viro if (err) { 1835b356379aSAl Viro err = nested_symlink(&next, nd); 18361da177e4SLinus Torvalds if (err) 1837a7472babSAl Viro return err; 183831e6b01fSNick Piggin } 183944b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 18403ddcd056SLinus Torvalds err = -ENOTDIR; 18413ddcd056SLinus Torvalds break; 18425f4a6a69SAl Viro } 1843ce57dfc1SAl Viro } 1844951361f9SAl Viro terminate_walk(nd); 18451da177e4SLinus Torvalds return err; 18461da177e4SLinus Torvalds } 18471da177e4SLinus Torvalds 184870e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 18495e53084dSAl Viro struct nameidata *nd) 185031e6b01fSNick Piggin { 185131e6b01fSNick Piggin int retval = 0; 185231e6b01fSNick Piggin 185331e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1854980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 185531e6b01fSNick Piggin nd->depth = 0; 18565e53084dSAl Viro nd->base = NULL; 18575b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1858b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1859b18825a7SDavid Howells struct inode *inode = root->d_inode; 186073d049a4SAl Viro if (*name) { 186144b1d530SMiklos Szeredi if (!d_can_lookup(root)) 18625b6ca027SAl Viro return -ENOTDIR; 18635b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18645b6ca027SAl Viro if (retval) 18655b6ca027SAl Viro return retval; 186673d049a4SAl Viro } 18675b6ca027SAl Viro nd->path = nd->root; 18685b6ca027SAl Viro nd->inode = inode; 18695b6ca027SAl Viro if (flags & LOOKUP_RCU) { 18708b61e74fSAl Viro rcu_read_lock(); 18715b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 187248a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 18735b6ca027SAl Viro } else { 18745b6ca027SAl Viro path_get(&nd->path); 18755b6ca027SAl Viro } 18765b6ca027SAl Viro return 0; 18775b6ca027SAl Viro } 18785b6ca027SAl Viro 187931e6b01fSNick Piggin nd->root.mnt = NULL; 188031e6b01fSNick Piggin 188148a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 188231e6b01fSNick Piggin if (*name=='/') { 1883e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18848b61e74fSAl Viro rcu_read_lock(); 18857bd88377SAl Viro nd->seq = set_root_rcu(nd); 1886e41f7d4eSAl Viro } else { 1887e41f7d4eSAl Viro set_root(nd); 1888e41f7d4eSAl Viro path_get(&nd->root); 1889e41f7d4eSAl Viro } 189031e6b01fSNick Piggin nd->path = nd->root; 189131e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1892e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 189331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1894c28cc364SNick Piggin unsigned seq; 189531e6b01fSNick Piggin 18968b61e74fSAl Viro rcu_read_lock(); 189731e6b01fSNick Piggin 1898c28cc364SNick Piggin do { 1899c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 190031e6b01fSNick Piggin nd->path = fs->pwd; 1901c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1902c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1903e41f7d4eSAl Viro } else { 1904e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1905e41f7d4eSAl Viro } 190631e6b01fSNick Piggin } else { 1907582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 19082903ff01SAl Viro struct fd f = fdget_raw(dfd); 190931e6b01fSNick Piggin struct dentry *dentry; 191031e6b01fSNick Piggin 19112903ff01SAl Viro if (!f.file) 19122903ff01SAl Viro return -EBADF; 191331e6b01fSNick Piggin 19142903ff01SAl Viro dentry = f.file->f_path.dentry; 191531e6b01fSNick Piggin 1916f52e0c11SAl Viro if (*name) { 191744b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 19182903ff01SAl Viro fdput(f); 19192903ff01SAl Viro return -ENOTDIR; 1920f52e0c11SAl Viro } 19212903ff01SAl Viro } 19222903ff01SAl Viro 19232903ff01SAl Viro nd->path = f.file->f_path; 1924e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19259c225f26SLinus Torvalds if (f.flags & FDPUT_FPUT) 19265e53084dSAl Viro nd->base = f.file; 1927c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 19288b61e74fSAl Viro rcu_read_lock(); 19295590ff0dSUlrich Drepper } else { 19302903ff01SAl Viro path_get(&nd->path); 19312903ff01SAl Viro fdput(f); 19321da177e4SLinus Torvalds } 1933e41f7d4eSAl Viro } 1934e41f7d4eSAl Viro 193531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 19364023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 19379b4a9b14SAl Viro return 0; 19384023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 19394023bfc9SAl Viro return 0; 19404023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 19414023bfc9SAl Viro nd->root.mnt = NULL; 19424023bfc9SAl Viro rcu_read_unlock(); 19434023bfc9SAl Viro return -ECHILD; 19449b4a9b14SAl Viro } 19459b4a9b14SAl Viro 1946893b7775SAl Viro static void path_cleanup(struct nameidata *nd) 1947893b7775SAl Viro { 1948893b7775SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 1949893b7775SAl Viro path_put(&nd->root); 1950893b7775SAl Viro nd->root.mnt = NULL; 1951893b7775SAl Viro } 1952893b7775SAl Viro if (unlikely(nd->base)) 1953893b7775SAl Viro fput(nd->base); 1954893b7775SAl Viro } 1955893b7775SAl Viro 1956bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1957bd92d7feSAl Viro { 1958bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1959bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1960bd92d7feSAl Viro 1961bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 196221b9b073SAl Viro return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); 1963bd92d7feSAl Viro } 1964bd92d7feSAl Viro 19659b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1966ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 19679b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 19689b4a9b14SAl Viro { 1969bd92d7feSAl Viro struct path path; 1970bd92d7feSAl Viro int err; 197131e6b01fSNick Piggin 197231e6b01fSNick Piggin /* 197331e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 197431e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 197531e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 197631e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 197731e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 197831e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 197931e6b01fSNick Piggin * analogue, foo_rcu(). 198031e6b01fSNick Piggin * 198131e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 198231e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 198331e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 198431e6b01fSNick Piggin * be able to complete). 198531e6b01fSNick Piggin */ 1986980f3ea2SAl Viro err = path_init(dfd, name, flags, nd); 1987ee0827cdSAl Viro 1988bd92d7feSAl Viro if (unlikely(err)) 1989115cbfdcSAl Viro goto out; 1990ee0827cdSAl Viro 1991ee0827cdSAl Viro current->total_link_count = 0; 1992bd92d7feSAl Viro err = link_path_walk(name, nd); 1993bd92d7feSAl Viro 1994bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1995bd92d7feSAl Viro err = lookup_last(nd, &path); 1996bd92d7feSAl Viro while (err > 0) { 1997bd92d7feSAl Viro void *cookie; 1998bd92d7feSAl Viro struct path link = path; 1999800179c9SKees Cook err = may_follow_link(&link, nd); 2000800179c9SKees Cook if (unlikely(err)) 2001800179c9SKees Cook break; 2002bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 2003574197e0SAl Viro err = follow_link(&link, nd, &cookie); 20046d7b5aaeSAl Viro if (err) 20056d7b5aaeSAl Viro break; 2006bd92d7feSAl Viro err = lookup_last(nd, &path); 2007574197e0SAl Viro put_link(nd, &link, cookie); 2008bd92d7feSAl Viro } 2009bd92d7feSAl Viro } 2010ee0827cdSAl Viro 20119f1fafeeSAl Viro if (!err) 20129f1fafeeSAl Viro err = complete_walk(nd); 2013bd92d7feSAl Viro 2014bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 201544b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 2016bd92d7feSAl Viro path_put(&nd->path); 2017bd23a539SAl Viro err = -ENOTDIR; 2018bd92d7feSAl Viro } 2019bd92d7feSAl Viro } 202016c2cd71SAl Viro 2021115cbfdcSAl Viro out: 2022893b7775SAl Viro path_cleanup(nd); 2023bd92d7feSAl Viro return err; 202431e6b01fSNick Piggin } 202531e6b01fSNick Piggin 2026873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 2027873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 2028873f1eedSJeff Layton { 2029873f1eedSJeff Layton int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd); 2030873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2031873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, flags, nd); 2032873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2033873f1eedSJeff Layton retval = path_lookupat(dfd, name->name, 2034873f1eedSJeff Layton flags | LOOKUP_REVAL, nd); 2035873f1eedSJeff Layton 2036873f1eedSJeff Layton if (likely(!retval)) 2037adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2038873f1eedSJeff Layton return retval; 2039873f1eedSJeff Layton } 2040873f1eedSJeff Layton 2041ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 2042ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 2043ee0827cdSAl Viro { 2044873f1eedSJeff Layton struct filename filename = { .name = name }; 2045ee0827cdSAl Viro 2046873f1eedSJeff Layton return filename_lookup(dfd, &filename, flags, nd); 20471da177e4SLinus Torvalds } 20481da177e4SLinus Torvalds 204979714f72SAl Viro /* does lookup, returns the object with parent locked */ 205079714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 20515590ff0dSUlrich Drepper { 205279714f72SAl Viro struct nameidata nd; 205379714f72SAl Viro struct dentry *d; 205479714f72SAl Viro int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd); 205579714f72SAl Viro if (err) 205679714f72SAl Viro return ERR_PTR(err); 205779714f72SAl Viro if (nd.last_type != LAST_NORM) { 205879714f72SAl Viro path_put(&nd.path); 205979714f72SAl Viro return ERR_PTR(-EINVAL); 206079714f72SAl Viro } 206179714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 20621e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 206379714f72SAl Viro if (IS_ERR(d)) { 206479714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 206579714f72SAl Viro path_put(&nd.path); 206679714f72SAl Viro return d; 206779714f72SAl Viro } 206879714f72SAl Viro *path = nd.path; 206979714f72SAl Viro return d; 20705590ff0dSUlrich Drepper } 20715590ff0dSUlrich Drepper 2072d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2073d1811465SAl Viro { 2074d1811465SAl Viro struct nameidata nd; 2075d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 2076d1811465SAl Viro if (!res) 2077d1811465SAl Viro *path = nd.path; 2078d1811465SAl Viro return res; 2079d1811465SAl Viro } 20804d359507SAl Viro EXPORT_SYMBOL(kern_path); 2081d1811465SAl Viro 208216f18200SJosef 'Jeff' Sipek /** 208316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 208416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 208516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 208616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 208716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2088e0a01249SAl Viro * @path: pointer to struct path to fill 208916f18200SJosef 'Jeff' Sipek */ 209016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 209116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2092e0a01249SAl Viro struct path *path) 209316f18200SJosef 'Jeff' Sipek { 2094e0a01249SAl Viro struct nameidata nd; 2095e0a01249SAl Viro int err; 2096e0a01249SAl Viro nd.root.dentry = dentry; 2097e0a01249SAl Viro nd.root.mnt = mnt; 2098e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 20995b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 2100e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 2101e0a01249SAl Viro if (!err) 2102e0a01249SAl Viro *path = nd.path; 2103e0a01249SAl Viro return err; 210416f18200SJosef 'Jeff' Sipek } 21054d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 210616f18200SJosef 'Jeff' Sipek 2107057f6c01SJames Morris /* 2108057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 2109057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 2110057f6c01SJames Morris * SMP-safe. 2111057f6c01SJames Morris */ 2112a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 21131da177e4SLinus Torvalds { 211472bd866aSAl Viro return __lookup_hash(&nd->last, nd->path.dentry, nd->flags); 21151da177e4SLinus Torvalds } 21161da177e4SLinus Torvalds 2117eead1911SChristoph Hellwig /** 2118a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2119eead1911SChristoph Hellwig * @name: pathname component to lookup 2120eead1911SChristoph Hellwig * @base: base directory to lookup from 2121eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2122eead1911SChristoph Hellwig * 2123a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 2124a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 2125eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 2126eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 2127eead1911SChristoph Hellwig */ 2128057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2129057f6c01SJames Morris { 2130057f6c01SJames Morris struct qstr this; 21316a96ba54SAl Viro unsigned int c; 2132cda309deSMiklos Szeredi int err; 2133057f6c01SJames Morris 21342f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 21352f9092e1SDavid Woodhouse 21366a96ba54SAl Viro this.name = name; 21376a96ba54SAl Viro this.len = len; 21380145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 21396a96ba54SAl Viro if (!len) 21406a96ba54SAl Viro return ERR_PTR(-EACCES); 21416a96ba54SAl Viro 214221d8a15aSAl Viro if (unlikely(name[0] == '.')) { 214321d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 214421d8a15aSAl Viro return ERR_PTR(-EACCES); 214521d8a15aSAl Viro } 214621d8a15aSAl Viro 21476a96ba54SAl Viro while (len--) { 21486a96ba54SAl Viro c = *(const unsigned char *)name++; 21496a96ba54SAl Viro if (c == '/' || c == '\0') 21506a96ba54SAl Viro return ERR_PTR(-EACCES); 21516a96ba54SAl Viro } 21525a202bcdSAl Viro /* 21535a202bcdSAl Viro * See if the low-level filesystem might want 21545a202bcdSAl Viro * to use its own hash.. 21555a202bcdSAl Viro */ 21565a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2157da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 21585a202bcdSAl Viro if (err < 0) 21595a202bcdSAl Viro return ERR_PTR(err); 21605a202bcdSAl Viro } 2161eead1911SChristoph Hellwig 2162cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2163cda309deSMiklos Szeredi if (err) 2164cda309deSMiklos Szeredi return ERR_PTR(err); 2165cda309deSMiklos Szeredi 216672bd866aSAl Viro return __lookup_hash(&this, base, 0); 2167057f6c01SJames Morris } 21684d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2169057f6c01SJames Morris 21701fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 21711fa1e7f6SAndy Whitcroft struct path *path, int *empty) 21721da177e4SLinus Torvalds { 21732d8f3038SAl Viro struct nameidata nd; 217491a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 21751da177e4SLinus Torvalds int err = PTR_ERR(tmp); 21761da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 21772d8f3038SAl Viro 21782d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21792d8f3038SAl Viro 2180873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 21811da177e4SLinus Torvalds putname(tmp); 21822d8f3038SAl Viro if (!err) 21832d8f3038SAl Viro *path = nd.path; 21841da177e4SLinus Torvalds } 21851da177e4SLinus Torvalds return err; 21861da177e4SLinus Torvalds } 21871da177e4SLinus Torvalds 21881fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 21891fa1e7f6SAndy Whitcroft struct path *path) 21901fa1e7f6SAndy Whitcroft { 2191f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 21921fa1e7f6SAndy Whitcroft } 21934d359507SAl Viro EXPORT_SYMBOL(user_path_at); 21941fa1e7f6SAndy Whitcroft 2195873f1eedSJeff Layton /* 2196873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2197873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2198873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2199873f1eedSJeff Layton * path-walking is complete. 2200873f1eedSJeff Layton */ 220191a27b2aSJeff Layton static struct filename * 22029e790bd6SJeff Layton user_path_parent(int dfd, const char __user *path, struct nameidata *nd, 22039e790bd6SJeff Layton unsigned int flags) 22042ad94ae6SAl Viro { 220591a27b2aSJeff Layton struct filename *s = getname(path); 22062ad94ae6SAl Viro int error; 22072ad94ae6SAl Viro 22089e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 22099e790bd6SJeff Layton flags &= LOOKUP_REVAL; 22109e790bd6SJeff Layton 22112ad94ae6SAl Viro if (IS_ERR(s)) 221291a27b2aSJeff Layton return s; 22132ad94ae6SAl Viro 22149e790bd6SJeff Layton error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd); 221591a27b2aSJeff Layton if (error) { 22162ad94ae6SAl Viro putname(s); 221791a27b2aSJeff Layton return ERR_PTR(error); 221891a27b2aSJeff Layton } 22192ad94ae6SAl Viro 222091a27b2aSJeff Layton return s; 22212ad94ae6SAl Viro } 22222ad94ae6SAl Viro 22238033426eSJeff Layton /** 2224197df04cSAl Viro * mountpoint_last - look up last component for umount 22258033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 22268033426eSJeff Layton * @path: pointer to container for result 22278033426eSJeff Layton * 22288033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 22298033426eSJeff Layton * need to resolve the path without doing any revalidation. 22308033426eSJeff Layton * 22318033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 22328033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 22338033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 22348033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 22358033426eSJeff Layton * bogus and it doesn't exist. 22368033426eSJeff Layton * 22378033426eSJeff Layton * Returns: 22388033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 22398033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 22408033426eSJeff Layton * put in this case. 22418033426eSJeff Layton * 22428033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 22438033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 22448033426eSJeff Layton * The nd->path reference will also be put. 22458033426eSJeff Layton * 22468033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 22478033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 22488033426eSJeff Layton * to the link, and nd->path will *not* be put. 22498033426eSJeff Layton */ 22508033426eSJeff Layton static int 2251197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 22528033426eSJeff Layton { 22538033426eSJeff Layton int error = 0; 22548033426eSJeff Layton struct dentry *dentry; 22558033426eSJeff Layton struct dentry *dir = nd->path.dentry; 22568033426eSJeff Layton 225735759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 225835759521SAl Viro if (nd->flags & LOOKUP_RCU) { 225935759521SAl Viro if (unlazy_walk(nd, NULL)) { 22608033426eSJeff Layton error = -ECHILD; 226135759521SAl Viro goto out; 226235759521SAl Viro } 22638033426eSJeff Layton } 22648033426eSJeff Layton 22658033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 22668033426eSJeff Layton 22678033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 22688033426eSJeff Layton error = handle_dots(nd, nd->last_type); 226935759521SAl Viro if (error) 227035759521SAl Viro goto out; 22718033426eSJeff Layton dentry = dget(nd->path.dentry); 227235759521SAl Viro goto done; 22738033426eSJeff Layton } 22748033426eSJeff Layton 22758033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 22768033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 22778033426eSJeff Layton if (!dentry) { 22788033426eSJeff Layton /* 22798033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 22808033426eSJeff Layton * so that means that this dentry is probably a symlink or the 22818033426eSJeff Layton * path doesn't actually point to a mounted dentry. 22828033426eSJeff Layton */ 22838033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 22848033426eSJeff Layton if (!dentry) { 22858033426eSJeff Layton error = -ENOMEM; 2286bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 228735759521SAl Viro goto out; 22888033426eSJeff Layton } 228935759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 229035759521SAl Viro error = PTR_ERR(dentry); 2291bcceeebaSDave Jones if (IS_ERR(dentry)) { 2292bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 229335759521SAl Viro goto out; 22948033426eSJeff Layton } 2295bcceeebaSDave Jones } 22968033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 22978033426eSJeff Layton 229835759521SAl Viro done: 229922213318SAl Viro if (!dentry->d_inode || d_is_negative(dentry)) { 23008033426eSJeff Layton error = -ENOENT; 23018033426eSJeff Layton dput(dentry); 230235759521SAl Viro goto out; 230335759521SAl Viro } 23048033426eSJeff Layton path->dentry = dentry; 2305295dc39dSVasily Averin path->mnt = nd->path.mnt; 2306b18825a7SDavid Howells if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) 23078033426eSJeff Layton return 1; 2308295dc39dSVasily Averin mntget(path->mnt); 23098033426eSJeff Layton follow_mount(path); 231035759521SAl Viro error = 0; 231135759521SAl Viro out: 23128033426eSJeff Layton terminate_walk(nd); 23138033426eSJeff Layton return error; 23148033426eSJeff Layton } 23158033426eSJeff Layton 23168033426eSJeff Layton /** 2317197df04cSAl Viro * path_mountpoint - look up a path to be umounted 23188033426eSJeff Layton * @dfd: directory file descriptor to start walk from 23198033426eSJeff Layton * @name: full pathname to walk 2320606d6fe3SRandy Dunlap * @path: pointer to container for result 23218033426eSJeff Layton * @flags: lookup flags 23228033426eSJeff Layton * 23238033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2324606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 23258033426eSJeff Layton */ 23268033426eSJeff Layton static int 2327197df04cSAl Viro path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags) 23288033426eSJeff Layton { 23298033426eSJeff Layton struct nameidata nd; 23308033426eSJeff Layton int err; 23318033426eSJeff Layton 2332980f3ea2SAl Viro err = path_init(dfd, name, flags, &nd); 23338033426eSJeff Layton if (unlikely(err)) 2334115cbfdcSAl Viro goto out; 23358033426eSJeff Layton 23368033426eSJeff Layton current->total_link_count = 0; 23378033426eSJeff Layton err = link_path_walk(name, &nd); 23388033426eSJeff Layton if (err) 23398033426eSJeff Layton goto out; 23408033426eSJeff Layton 2341197df04cSAl Viro err = mountpoint_last(&nd, path); 23428033426eSJeff Layton while (err > 0) { 23438033426eSJeff Layton void *cookie; 23448033426eSJeff Layton struct path link = *path; 23458033426eSJeff Layton err = may_follow_link(&link, &nd); 23468033426eSJeff Layton if (unlikely(err)) 23478033426eSJeff Layton break; 23488033426eSJeff Layton nd.flags |= LOOKUP_PARENT; 23498033426eSJeff Layton err = follow_link(&link, &nd, &cookie); 23508033426eSJeff Layton if (err) 23518033426eSJeff Layton break; 2352197df04cSAl Viro err = mountpoint_last(&nd, path); 23538033426eSJeff Layton put_link(&nd, &link, cookie); 23548033426eSJeff Layton } 23558033426eSJeff Layton out: 2356893b7775SAl Viro path_cleanup(&nd); 23578033426eSJeff Layton return err; 23588033426eSJeff Layton } 23598033426eSJeff Layton 23602d864651SAl Viro static int 23612d864651SAl Viro filename_mountpoint(int dfd, struct filename *s, struct path *path, 23622d864651SAl Viro unsigned int flags) 23632d864651SAl Viro { 23642d864651SAl Viro int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU); 23652d864651SAl Viro if (unlikely(error == -ECHILD)) 23662d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags); 23672d864651SAl Viro if (unlikely(error == -ESTALE)) 23682d864651SAl Viro error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL); 23692d864651SAl Viro if (likely(!error)) 23702d864651SAl Viro audit_inode(s, path->dentry, 0); 23712d864651SAl Viro return error; 23722d864651SAl Viro } 23732d864651SAl Viro 23748033426eSJeff Layton /** 2375197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 23768033426eSJeff Layton * @dfd: directory file descriptor 23778033426eSJeff Layton * @name: pathname from userland 23788033426eSJeff Layton * @flags: lookup flags 23798033426eSJeff Layton * @path: pointer to container to hold result 23808033426eSJeff Layton * 23818033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 23828033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 23838033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 23848033426eSJeff Layton * and avoid revalidating the last component. 23858033426eSJeff Layton * 23868033426eSJeff Layton * Returns 0 and populates "path" on success. 23878033426eSJeff Layton */ 23888033426eSJeff Layton int 2389197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 23908033426eSJeff Layton struct path *path) 23918033426eSJeff Layton { 23928033426eSJeff Layton struct filename *s = getname(name); 23938033426eSJeff Layton int error; 23948033426eSJeff Layton if (IS_ERR(s)) 23958033426eSJeff Layton return PTR_ERR(s); 23962d864651SAl Viro error = filename_mountpoint(dfd, s, path, flags); 23978033426eSJeff Layton putname(s); 23988033426eSJeff Layton return error; 23998033426eSJeff Layton } 24008033426eSJeff Layton 24012d864651SAl Viro int 24022d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 24032d864651SAl Viro unsigned int flags) 24042d864651SAl Viro { 24052d864651SAl Viro struct filename s = {.name = name}; 24062d864651SAl Viro return filename_mountpoint(dfd, &s, path, flags); 24072d864651SAl Viro } 24082d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 24092d864651SAl Viro 2410cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 24111da177e4SLinus Torvalds { 24128e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2413da9592edSDavid Howells 24148e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 24151da177e4SLinus Torvalds return 0; 24168e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 24171da177e4SLinus Torvalds return 0; 241823adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 24191da177e4SLinus Torvalds } 2420cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 24211da177e4SLinus Torvalds 24221da177e4SLinus Torvalds /* 24231da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 24241da177e4SLinus Torvalds * whether the type of victim is right. 24251da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 24261da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 24271da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 24281da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 24291da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 24301da177e4SLinus Torvalds * a. be owner of dir, or 24311da177e4SLinus Torvalds * b. be owner of victim, or 24321da177e4SLinus Torvalds * c. have CAP_FOWNER capability 24331da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 24341da177e4SLinus Torvalds * links pointing to it. 24351da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 24361da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 24371da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 24381da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 24391da177e4SLinus Torvalds * nfs_async_unlink(). 24401da177e4SLinus Torvalds */ 2441b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 24421da177e4SLinus Torvalds { 2443b18825a7SDavid Howells struct inode *inode = victim->d_inode; 24441da177e4SLinus Torvalds int error; 24451da177e4SLinus Torvalds 2446b18825a7SDavid Howells if (d_is_negative(victim)) 24471da177e4SLinus Torvalds return -ENOENT; 2448b18825a7SDavid Howells BUG_ON(!inode); 24491da177e4SLinus Torvalds 24501da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24514fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 24521da177e4SLinus Torvalds 2453f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 24541da177e4SLinus Torvalds if (error) 24551da177e4SLinus Torvalds return error; 24561da177e4SLinus Torvalds if (IS_APPEND(dir)) 24571da177e4SLinus Torvalds return -EPERM; 2458b18825a7SDavid Howells 2459b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2460b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 24611da177e4SLinus Torvalds return -EPERM; 24621da177e4SLinus Torvalds if (isdir) { 246344b1d530SMiklos Szeredi if (!d_is_dir(victim)) 24641da177e4SLinus Torvalds return -ENOTDIR; 24651da177e4SLinus Torvalds if (IS_ROOT(victim)) 24661da177e4SLinus Torvalds return -EBUSY; 246744b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 24681da177e4SLinus Torvalds return -EISDIR; 24691da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24701da177e4SLinus Torvalds return -ENOENT; 24711da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 24721da177e4SLinus Torvalds return -EBUSY; 24731da177e4SLinus Torvalds return 0; 24741da177e4SLinus Torvalds } 24751da177e4SLinus Torvalds 24761da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 24771da177e4SLinus Torvalds * dir. 24781da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 24791da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 24801da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 24811da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 24821da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 24831da177e4SLinus Torvalds */ 2484a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 24851da177e4SLinus Torvalds { 248614e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 24871da177e4SLinus Torvalds if (child->d_inode) 24881da177e4SLinus Torvalds return -EEXIST; 24891da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24901da177e4SLinus Torvalds return -ENOENT; 2491f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds /* 24951da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 24961da177e4SLinus Torvalds */ 24971da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 24981da177e4SLinus Torvalds { 24991da177e4SLinus Torvalds struct dentry *p; 25001da177e4SLinus Torvalds 25011da177e4SLinus Torvalds if (p1 == p2) { 2502f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 25031da177e4SLinus Torvalds return NULL; 25041da177e4SLinus Torvalds } 25051da177e4SLinus Torvalds 2506a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25071da177e4SLinus Torvalds 2508e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2509e2761a11SOGAWA Hirofumi if (p) { 2510f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2511f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 25121da177e4SLinus Torvalds return p; 25131da177e4SLinus Torvalds } 25141da177e4SLinus Torvalds 2515e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2516e2761a11SOGAWA Hirofumi if (p) { 2517f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2518f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 25191da177e4SLinus Torvalds return p; 25201da177e4SLinus Torvalds } 25211da177e4SLinus Torvalds 2522f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2523d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 25241da177e4SLinus Torvalds return NULL; 25251da177e4SLinus Torvalds } 25264d359507SAl Viro EXPORT_SYMBOL(lock_rename); 25271da177e4SLinus Torvalds 25281da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 25291da177e4SLinus Torvalds { 25301b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 25311da177e4SLinus Torvalds if (p1 != p2) { 25321b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2533a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25341da177e4SLinus Torvalds } 25351da177e4SLinus Torvalds } 25364d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 25371da177e4SLinus Torvalds 25384acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2539312b63fbSAl Viro bool want_excl) 25401da177e4SLinus Torvalds { 2541a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25421da177e4SLinus Torvalds if (error) 25431da177e4SLinus Torvalds return error; 25441da177e4SLinus Torvalds 2545acfa4380SAl Viro if (!dir->i_op->create) 25461da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 25471da177e4SLinus Torvalds mode &= S_IALLUGO; 25481da177e4SLinus Torvalds mode |= S_IFREG; 25491da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 25501da177e4SLinus Torvalds if (error) 25511da177e4SLinus Torvalds return error; 2552312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2553a74574aaSStephen Smalley if (!error) 2554f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25551da177e4SLinus Torvalds return error; 25561da177e4SLinus Torvalds } 25574d359507SAl Viro EXPORT_SYMBOL(vfs_create); 25581da177e4SLinus Torvalds 255973d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 25601da177e4SLinus Torvalds { 25613fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 25621da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 25631da177e4SLinus Torvalds int error; 25641da177e4SLinus Torvalds 2565bcda7652SAl Viro /* O_PATH? */ 2566bcda7652SAl Viro if (!acc_mode) 2567bcda7652SAl Viro return 0; 2568bcda7652SAl Viro 25691da177e4SLinus Torvalds if (!inode) 25701da177e4SLinus Torvalds return -ENOENT; 25711da177e4SLinus Torvalds 2572c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2573c8fe8f30SChristoph Hellwig case S_IFLNK: 25741da177e4SLinus Torvalds return -ELOOP; 2575c8fe8f30SChristoph Hellwig case S_IFDIR: 2576c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 25771da177e4SLinus Torvalds return -EISDIR; 2578c8fe8f30SChristoph Hellwig break; 2579c8fe8f30SChristoph Hellwig case S_IFBLK: 2580c8fe8f30SChristoph Hellwig case S_IFCHR: 25813fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 25821da177e4SLinus Torvalds return -EACCES; 2583c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2584c8fe8f30SChristoph Hellwig case S_IFIFO: 2585c8fe8f30SChristoph Hellwig case S_IFSOCK: 25861da177e4SLinus Torvalds flag &= ~O_TRUNC; 2587c8fe8f30SChristoph Hellwig break; 25884a3fd211SDave Hansen } 2589b41572e9SDave Hansen 25903fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2591b41572e9SDave Hansen if (error) 2592b41572e9SDave Hansen return error; 25936146f0d5SMimi Zohar 25941da177e4SLinus Torvalds /* 25951da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 25961da177e4SLinus Torvalds */ 25971da177e4SLinus Torvalds if (IS_APPEND(inode)) { 25988737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 25997715b521SAl Viro return -EPERM; 26001da177e4SLinus Torvalds if (flag & O_TRUNC) 26017715b521SAl Viro return -EPERM; 26021da177e4SLinus Torvalds } 26031da177e4SLinus Torvalds 26041da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 26052e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 26067715b521SAl Viro return -EPERM; 26071da177e4SLinus Torvalds 2608f3c7691eSJ. Bruce Fields return 0; 26097715b521SAl Viro } 26107715b521SAl Viro 2611e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 26127715b521SAl Viro { 2613e1181ee6SJeff Layton struct path *path = &filp->f_path; 26147715b521SAl Viro struct inode *inode = path->dentry->d_inode; 26157715b521SAl Viro int error = get_write_access(inode); 26161da177e4SLinus Torvalds if (error) 26177715b521SAl Viro return error; 26181da177e4SLinus Torvalds /* 26191da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 26201da177e4SLinus Torvalds */ 2621d7a06983SJeff Layton error = locks_verify_locked(filp); 2622be6d3e56SKentaro Takeda if (!error) 2623ea0d3ab2STetsuo Handa error = security_path_truncate(path); 26241da177e4SLinus Torvalds if (!error) { 26257715b521SAl Viro error = do_truncate(path->dentry, 0, 2626d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2627e1181ee6SJeff Layton filp); 26281da177e4SLinus Torvalds } 26291da177e4SLinus Torvalds put_write_access(inode); 2630acd0c935SMimi Zohar return error; 26311da177e4SLinus Torvalds } 26321da177e4SLinus Torvalds 2633d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2634d57999e1SDave Hansen { 26358a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 26368a5e929dSAl Viro flag--; 2637d57999e1SDave Hansen return flag; 2638d57999e1SDave Hansen } 2639d57999e1SDave Hansen 2640d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2641d18e9008SMiklos Szeredi { 2642d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2643d18e9008SMiklos Szeredi if (error) 2644d18e9008SMiklos Szeredi return error; 2645d18e9008SMiklos Szeredi 2646d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2647d18e9008SMiklos Szeredi if (error) 2648d18e9008SMiklos Szeredi return error; 2649d18e9008SMiklos Szeredi 2650d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2651d18e9008SMiklos Szeredi } 2652d18e9008SMiklos Szeredi 26531acf0af9SDavid Howells /* 26541acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 26551acf0af9SDavid Howells * dentry. 26561acf0af9SDavid Howells * 26571acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 26581acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 26591acf0af9SDavid Howells * 26601acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 26611acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 26621acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 26631acf0af9SDavid Howells * 26641acf0af9SDavid Howells * Returns an error code otherwise. 26651acf0af9SDavid Howells */ 26662675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 266730d90494SAl Viro struct path *path, struct file *file, 2668015c3bbcSMiklos Szeredi const struct open_flags *op, 266964894cf8SAl Viro bool got_write, bool need_lookup, 267047237687SAl Viro int *opened) 2671d18e9008SMiklos Szeredi { 2672d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2673d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2674d18e9008SMiklos Szeredi umode_t mode; 2675d18e9008SMiklos Szeredi int error; 2676d18e9008SMiklos Szeredi int acc_mode; 2677d18e9008SMiklos Szeredi int create_error = 0; 2678d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2679116cc022SMiklos Szeredi bool excl; 2680d18e9008SMiklos Szeredi 2681d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2682d18e9008SMiklos Szeredi 2683d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2684d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 26852675a4ebSAl Viro error = -ENOENT; 2686d18e9008SMiklos Szeredi goto out; 2687d18e9008SMiklos Szeredi } 2688d18e9008SMiklos Szeredi 268962b259d8SMiklos Szeredi mode = op->mode; 2690d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2691d18e9008SMiklos Szeredi mode &= ~current_umask(); 2692d18e9008SMiklos Szeredi 2693116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2694116cc022SMiklos Szeredi if (excl) 2695d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2696d18e9008SMiklos Szeredi 2697d18e9008SMiklos Szeredi /* 2698d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2699d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2700d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2701d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2702d18e9008SMiklos Szeredi * 2703d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2704d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2705d18e9008SMiklos Szeredi */ 270664894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 270764894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 270864894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2709d18e9008SMiklos Szeredi /* 2710d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2711d18e9008SMiklos Szeredi * back to lookup + open 2712d18e9008SMiklos Szeredi */ 2713d18e9008SMiklos Szeredi goto no_open; 2714d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2715d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 271664894cf8SAl Viro create_error = -EROFS; 2717d18e9008SMiklos Szeredi goto no_open; 2718d18e9008SMiklos Szeredi } else { 2719d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 272064894cf8SAl Viro create_error = -EROFS; 2721d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2722d18e9008SMiklos Szeredi } 2723d18e9008SMiklos Szeredi } 2724d18e9008SMiklos Szeredi 2725d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 272638227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2727d18e9008SMiklos Szeredi if (error) { 2728d18e9008SMiklos Szeredi create_error = error; 2729d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2730d18e9008SMiklos Szeredi goto no_open; 2731d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2732d18e9008SMiklos Szeredi } 2733d18e9008SMiklos Szeredi } 2734d18e9008SMiklos Szeredi 2735d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2736d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2737d18e9008SMiklos Szeredi 273830d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 273930d90494SAl Viro file->f_path.mnt = nd->path.mnt; 274030d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 274147237687SAl Viro opened); 2742d9585277SAl Viro if (error < 0) { 2743d9585277SAl Viro if (create_error && error == -ENOENT) 2744d9585277SAl Viro error = create_error; 2745d18e9008SMiklos Szeredi goto out; 2746d18e9008SMiklos Szeredi } 2747d18e9008SMiklos Szeredi 2748d9585277SAl Viro if (error) { /* returned 1, that is */ 274930d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 27502675a4ebSAl Viro error = -EIO; 2751d18e9008SMiklos Szeredi goto out; 2752d18e9008SMiklos Szeredi } 275330d90494SAl Viro if (file->f_path.dentry) { 2754d18e9008SMiklos Szeredi dput(dentry); 275530d90494SAl Viro dentry = file->f_path.dentry; 2756d18e9008SMiklos Szeredi } 275703da633aSAl Viro if (*opened & FILE_CREATED) 275803da633aSAl Viro fsnotify_create(dir, dentry); 275903da633aSAl Viro if (!dentry->d_inode) { 276003da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 276103da633aSAl Viro if (create_error) { 276262b2ce96SSage Weil error = create_error; 276362b2ce96SSage Weil goto out; 276462b2ce96SSage Weil } 276503da633aSAl Viro } else { 276603da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 276703da633aSAl Viro error = -EEXIST; 276803da633aSAl Viro goto out; 276903da633aSAl Viro } 277003da633aSAl Viro } 2771d18e9008SMiklos Szeredi goto looked_up; 2772d18e9008SMiklos Szeredi } 2773d18e9008SMiklos Szeredi 2774d18e9008SMiklos Szeredi /* 2775d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2776d18e9008SMiklos Szeredi * here. 2777d18e9008SMiklos Szeredi */ 277803da633aSAl Viro acc_mode = op->acc_mode; 277903da633aSAl Viro if (*opened & FILE_CREATED) { 278003da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 278103da633aSAl Viro fsnotify_create(dir, dentry); 278203da633aSAl Viro acc_mode = MAY_OPEN; 278303da633aSAl Viro } 27842675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 27852675a4ebSAl Viro if (error) 27862675a4ebSAl Viro fput(file); 2787d18e9008SMiklos Szeredi 2788d18e9008SMiklos Szeredi out: 2789d18e9008SMiklos Szeredi dput(dentry); 27902675a4ebSAl Viro return error; 2791d18e9008SMiklos Szeredi 2792d18e9008SMiklos Szeredi no_open: 2793d18e9008SMiklos Szeredi if (need_lookup) { 279472bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2795d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 27962675a4ebSAl Viro return PTR_ERR(dentry); 2797d18e9008SMiklos Szeredi 2798d18e9008SMiklos Szeredi if (create_error) { 2799d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2800d18e9008SMiklos Szeredi 28012675a4ebSAl Viro error = create_error; 2802d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2803d18e9008SMiklos Szeredi if (!dentry->d_inode) 2804d18e9008SMiklos Szeredi goto out; 2805d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2806d18e9008SMiklos Szeredi goto out; 2807d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2808d18e9008SMiklos Szeredi S_ISREG(dentry->d_inode->i_mode)) { 2809d18e9008SMiklos Szeredi goto out; 2810d18e9008SMiklos Szeredi } 2811d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2812d18e9008SMiklos Szeredi } 2813d18e9008SMiklos Szeredi } 2814d18e9008SMiklos Szeredi looked_up: 2815d18e9008SMiklos Szeredi path->dentry = dentry; 2816d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 28172675a4ebSAl Viro return 1; 2818d18e9008SMiklos Szeredi } 2819d18e9008SMiklos Szeredi 282031e6b01fSNick Piggin /* 28211acf0af9SDavid Howells * Look up and maybe create and open the last component. 2822d58ffd35SMiklos Szeredi * 2823d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2824d58ffd35SMiklos Szeredi * 28251acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 28261acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 28271acf0af9SDavid Howells * 28281acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 28291acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 28301acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 28311acf0af9SDavid Howells * specified then a negative dentry may be returned. 28321acf0af9SDavid Howells * 28331acf0af9SDavid Howells * An error code is returned otherwise. 28341acf0af9SDavid Howells * 28351acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 28361acf0af9SDavid Howells * cleared otherwise prior to returning. 2837d58ffd35SMiklos Szeredi */ 28382675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 283930d90494SAl Viro struct file *file, 2840d58ffd35SMiklos Szeredi const struct open_flags *op, 284164894cf8SAl Viro bool got_write, int *opened) 2842d58ffd35SMiklos Szeredi { 2843d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 284454ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2845d58ffd35SMiklos Szeredi struct dentry *dentry; 2846d58ffd35SMiklos Szeredi int error; 284754ef4872SMiklos Szeredi bool need_lookup; 2848d58ffd35SMiklos Szeredi 284947237687SAl Viro *opened &= ~FILE_CREATED; 2850201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2851d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 28522675a4ebSAl Viro return PTR_ERR(dentry); 2853d58ffd35SMiklos Szeredi 2854d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2855d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2856d18e9008SMiklos Szeredi goto out_no_open; 2857d18e9008SMiklos Szeredi 2858d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 285964894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 286047237687SAl Viro need_lookup, opened); 2861d18e9008SMiklos Szeredi } 2862d18e9008SMiklos Szeredi 286354ef4872SMiklos Szeredi if (need_lookup) { 286454ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 286554ef4872SMiklos Szeredi 286672bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 286754ef4872SMiklos Szeredi if (IS_ERR(dentry)) 28682675a4ebSAl Viro return PTR_ERR(dentry); 286954ef4872SMiklos Szeredi } 287054ef4872SMiklos Szeredi 2871d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2872d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2873d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2874d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2875d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2876d58ffd35SMiklos Szeredi /* 2877d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2878d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2879d58ffd35SMiklos Szeredi * the time when the file is created and when 2880d58ffd35SMiklos Szeredi * a permanent write count is taken through 2881015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2882d58ffd35SMiklos Szeredi */ 288364894cf8SAl Viro if (!got_write) { 288464894cf8SAl Viro error = -EROFS; 2885d58ffd35SMiklos Szeredi goto out_dput; 288664894cf8SAl Viro } 288747237687SAl Viro *opened |= FILE_CREATED; 2888d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2889d58ffd35SMiklos Szeredi if (error) 2890d58ffd35SMiklos Szeredi goto out_dput; 2891312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2892312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2893d58ffd35SMiklos Szeredi if (error) 2894d58ffd35SMiklos Szeredi goto out_dput; 2895d58ffd35SMiklos Szeredi } 2896d18e9008SMiklos Szeredi out_no_open: 2897d58ffd35SMiklos Szeredi path->dentry = dentry; 2898d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 28992675a4ebSAl Viro return 1; 2900d58ffd35SMiklos Szeredi 2901d58ffd35SMiklos Szeredi out_dput: 2902d58ffd35SMiklos Szeredi dput(dentry); 29032675a4ebSAl Viro return error; 2904d58ffd35SMiklos Szeredi } 2905d58ffd35SMiklos Szeredi 2906d58ffd35SMiklos Szeredi /* 2907fe2d35ffSAl Viro * Handle the last step of open() 290831e6b01fSNick Piggin */ 29092675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 291030d90494SAl Viro struct file *file, const struct open_flags *op, 2911669abf4eSJeff Layton int *opened, struct filename *name) 2912fb1cc555SAl Viro { 2913a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2914ca344a89SAl Viro int open_flag = op->open_flag; 291577d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 291664894cf8SAl Viro bool got_write = false; 2917bcda7652SAl Viro int acc_mode = op->acc_mode; 2918a1eb3315SMiklos Szeredi struct inode *inode; 291977d660a8SMiklos Szeredi bool symlink_ok = false; 292016b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 292116b1c1cdSMiklos Szeredi bool retried = false; 292216c2cd71SAl Viro int error; 2923fb1cc555SAl Viro 2924c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2925c3e380b0SAl Viro nd->flags |= op->intent; 2926c3e380b0SAl Viro 2927bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2928fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2929fe2d35ffSAl Viro if (error) 29302675a4ebSAl Viro return error; 2931e83db167SMiklos Szeredi goto finish_open; 29321f36f774SAl Viro } 2933a2c36b45SAl Viro 2934ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2935fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2936fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2937bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 293877d660a8SMiklos Szeredi symlink_ok = true; 2939fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2940e97cdc87SAl Viro error = lookup_fast(nd, path, &inode); 294171574865SMiklos Szeredi if (likely(!error)) 294271574865SMiklos Szeredi goto finish_lookup; 294371574865SMiklos Szeredi 2944ce57dfc1SAl Viro if (error < 0) 29452675a4ebSAl Viro goto out; 2946a1eb3315SMiklos Szeredi 294737d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2948b6183df7SMiklos Szeredi } else { 2949fe2d35ffSAl Viro /* create side of things */ 2950a3fbbde7SAl Viro /* 2951b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2952b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2953b6183df7SMiklos Szeredi * about to look up 2954a3fbbde7SAl Viro */ 29559f1fafeeSAl Viro error = complete_walk(nd); 29569f1fafeeSAl Viro if (error) 29572675a4ebSAl Viro return error; 2958fe2d35ffSAl Viro 295933e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 296016c2cd71SAl Viro error = -EISDIR; 29611f36f774SAl Viro /* trailing slashes? */ 296231e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 29632675a4ebSAl Viro goto out; 2964b6183df7SMiklos Szeredi } 29651f36f774SAl Viro 296616b1c1cdSMiklos Szeredi retry_lookup: 296764894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 296864894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 296964894cf8SAl Viro if (!error) 297064894cf8SAl Viro got_write = true; 297164894cf8SAl Viro /* 297264894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 297364894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 297464894cf8SAl Viro * dropping this one anyway. 297564894cf8SAl Viro */ 297664894cf8SAl Viro } 2977a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 297864894cf8SAl Viro error = lookup_open(nd, path, file, op, got_write, opened); 2979fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2980fb1cc555SAl Viro 29812675a4ebSAl Viro if (error <= 0) { 29822675a4ebSAl Viro if (error) 2983d58ffd35SMiklos Szeredi goto out; 29846c0d46c4SAl Viro 298547237687SAl Viro if ((*opened & FILE_CREATED) || 2986496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 298777d660a8SMiklos Szeredi will_truncate = false; 2988d18e9008SMiklos Szeredi 2989adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 2990d18e9008SMiklos Szeredi goto opened; 2991d18e9008SMiklos Szeredi } 2992d18e9008SMiklos Szeredi 299347237687SAl Viro if (*opened & FILE_CREATED) { 29949b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2995ca344a89SAl Viro open_flag &= ~O_TRUNC; 299677d660a8SMiklos Szeredi will_truncate = false; 2997bcda7652SAl Viro acc_mode = MAY_OPEN; 2998d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 2999e83db167SMiklos Szeredi goto finish_open_created; 3000fb1cc555SAl Viro } 3001fb1cc555SAl Viro 3002fb1cc555SAl Viro /* 30033134f37eSJeff Layton * create/update audit record if it already exists. 3004fb1cc555SAl Viro */ 3005b18825a7SDavid Howells if (d_is_positive(path->dentry)) 3006adb5c247SJeff Layton audit_inode(name, path->dentry, 0); 3007fb1cc555SAl Viro 3008d18e9008SMiklos Szeredi /* 3009d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3010d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3011d18e9008SMiklos Szeredi * necessary...) 3012d18e9008SMiklos Szeredi */ 301364894cf8SAl Viro if (got_write) { 3014d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 301564894cf8SAl Viro got_write = false; 3016d18e9008SMiklos Szeredi } 3017d18e9008SMiklos Szeredi 3018fb1cc555SAl Viro error = -EEXIST; 3019f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 3020fb1cc555SAl Viro goto exit_dput; 3021fb1cc555SAl Viro 30229875cf80SDavid Howells error = follow_managed(path, nd->flags); 30239875cf80SDavid Howells if (error < 0) 3024fb1cc555SAl Viro goto exit_dput; 3025fb1cc555SAl Viro 3026a3fbbde7SAl Viro if (error) 3027a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 3028a3fbbde7SAl Viro 3029decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 3030decf3400SMiklos Szeredi inode = path->dentry->d_inode; 30315f5daac1SMiklos Szeredi finish_lookup: 30325f5daac1SMiklos Szeredi /* we _can_ be in RCU mode here */ 3033fb1cc555SAl Viro error = -ENOENT; 303422213318SAl Viro if (!inode || d_is_negative(path->dentry)) { 303554c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 30362675a4ebSAl Viro goto out; 303754c33e7fSMiklos Szeredi } 30389e67f361SAl Viro 3039b18825a7SDavid Howells if (should_follow_link(path->dentry, !symlink_ok)) { 3040d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 3041d45ea867SMiklos Szeredi if (unlikely(unlazy_walk(nd, path->dentry))) { 3042d45ea867SMiklos Szeredi error = -ECHILD; 30432675a4ebSAl Viro goto out; 3044d45ea867SMiklos Szeredi } 3045d45ea867SMiklos Szeredi } 3046d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 30472675a4ebSAl Viro return 1; 3048d45ea867SMiklos Szeredi } 3049fb1cc555SAl Viro 305016b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 3051fb1cc555SAl Viro path_to_nameidata(path, nd); 305216b1c1cdSMiklos Szeredi } else { 305316b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 305416b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 305516b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 305616b1c1cdSMiklos Szeredi 305716b1c1cdSMiklos Szeredi } 3058decf3400SMiklos Szeredi nd->inode = inode; 3059a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3060bc77daa7SAl Viro finish_open: 3061a3fbbde7SAl Viro error = complete_walk(nd); 306216b1c1cdSMiklos Szeredi if (error) { 306316b1c1cdSMiklos Szeredi path_put(&save_parent); 30642675a4ebSAl Viro return error; 306516b1c1cdSMiklos Szeredi } 3066bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3067fb1cc555SAl Viro error = -EISDIR; 306844b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 30692675a4ebSAl Viro goto out; 3070af2f5542SMiklos Szeredi error = -ENOTDIR; 307144b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 30722675a4ebSAl Viro goto out; 30736c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 307477d660a8SMiklos Szeredi will_truncate = false; 30756c0d46c4SAl Viro 30760f9d1a10SAl Viro if (will_truncate) { 30770f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 30780f9d1a10SAl Viro if (error) 30792675a4ebSAl Viro goto out; 308064894cf8SAl Viro got_write = true; 30810f9d1a10SAl Viro } 3082e83db167SMiklos Szeredi finish_open_created: 3083bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3084ca344a89SAl Viro if (error) 30852675a4ebSAl Viro goto out; 30864aa7c634SMiklos Szeredi 30874aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 30884aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 30894aa7c634SMiklos Szeredi if (!error) { 30904aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 30914aa7c634SMiklos Szeredi } else { 309230d90494SAl Viro if (error == -EOPENSTALE) 3093f60dc3dbSMiklos Szeredi goto stale_open; 3094015c3bbcSMiklos Szeredi goto out; 3095f60dc3dbSMiklos Szeredi } 3096a8277b9bSMiklos Szeredi opened: 30972675a4ebSAl Viro error = open_check_o_direct(file); 3098015c3bbcSMiklos Szeredi if (error) 3099015c3bbcSMiklos Szeredi goto exit_fput; 31003034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3101aa4caadbSMiklos Szeredi if (error) 3102aa4caadbSMiklos Szeredi goto exit_fput; 3103aa4caadbSMiklos Szeredi 31040f9d1a10SAl Viro if (will_truncate) { 31052675a4ebSAl Viro error = handle_truncate(file); 3106aa4caadbSMiklos Szeredi if (error) 3107aa4caadbSMiklos Szeredi goto exit_fput; 31080f9d1a10SAl Viro } 3109ca344a89SAl Viro out: 311064894cf8SAl Viro if (got_write) 31110f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 311216b1c1cdSMiklos Szeredi path_put(&save_parent); 3113e276ae67SMiklos Szeredi terminate_walk(nd); 31142675a4ebSAl Viro return error; 3115fb1cc555SAl Viro 3116fb1cc555SAl Viro exit_dput: 3117fb1cc555SAl Viro path_put_conditional(path, nd); 3118ca344a89SAl Viro goto out; 3119015c3bbcSMiklos Szeredi exit_fput: 31202675a4ebSAl Viro fput(file); 31212675a4ebSAl Viro goto out; 3122015c3bbcSMiklos Szeredi 3123f60dc3dbSMiklos Szeredi stale_open: 3124f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3125f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3126f60dc3dbSMiklos Szeredi goto out; 3127f60dc3dbSMiklos Szeredi 3128f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3129f60dc3dbSMiklos Szeredi path_put(&nd->path); 3130f60dc3dbSMiklos Szeredi nd->path = save_parent; 3131f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3132f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3133f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 313464894cf8SAl Viro if (got_write) { 3135f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 313664894cf8SAl Viro got_write = false; 3137f60dc3dbSMiklos Szeredi } 3138f60dc3dbSMiklos Szeredi retried = true; 3139f60dc3dbSMiklos Szeredi goto retry_lookup; 3140fb1cc555SAl Viro } 3141fb1cc555SAl Viro 314260545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 314360545d0dSAl Viro struct nameidata *nd, int flags, 314460545d0dSAl Viro const struct open_flags *op, 314560545d0dSAl Viro struct file *file, int *opened) 314660545d0dSAl Viro { 314760545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 314860545d0dSAl Viro struct dentry *dentry, *child; 314960545d0dSAl Viro struct inode *dir; 315060545d0dSAl Viro int error = path_lookupat(dfd, pathname->name, 315160545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 315260545d0dSAl Viro if (unlikely(error)) 315360545d0dSAl Viro return error; 315460545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 315560545d0dSAl Viro if (unlikely(error)) 315660545d0dSAl Viro goto out; 315760545d0dSAl Viro /* we want directory to be writable */ 315860545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 315960545d0dSAl Viro if (error) 316060545d0dSAl Viro goto out2; 316160545d0dSAl Viro dentry = nd->path.dentry; 316260545d0dSAl Viro dir = dentry->d_inode; 316360545d0dSAl Viro if (!dir->i_op->tmpfile) { 316460545d0dSAl Viro error = -EOPNOTSUPP; 316560545d0dSAl Viro goto out2; 316660545d0dSAl Viro } 316760545d0dSAl Viro child = d_alloc(dentry, &name); 316860545d0dSAl Viro if (unlikely(!child)) { 316960545d0dSAl Viro error = -ENOMEM; 317060545d0dSAl Viro goto out2; 317160545d0dSAl Viro } 317260545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 317360545d0dSAl Viro nd->flags |= op->intent; 317460545d0dSAl Viro dput(nd->path.dentry); 317560545d0dSAl Viro nd->path.dentry = child; 317660545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 317760545d0dSAl Viro if (error) 317860545d0dSAl Viro goto out2; 317960545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 318069a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 318169a91c23SEric Rannaud error = may_open(&nd->path, MAY_OPEN, op->open_flag); 318260545d0dSAl Viro if (error) 318360545d0dSAl Viro goto out2; 318460545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 318560545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 318660545d0dSAl Viro if (error) 318760545d0dSAl Viro goto out2; 318860545d0dSAl Viro error = open_check_o_direct(file); 3189f4e0c30cSAl Viro if (error) { 319060545d0dSAl Viro fput(file); 3191f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3192f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3193f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3194f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3195f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3196f4e0c30cSAl Viro } 319760545d0dSAl Viro out2: 319860545d0dSAl Viro mnt_drop_write(nd->path.mnt); 319960545d0dSAl Viro out: 320060545d0dSAl Viro path_put(&nd->path); 320160545d0dSAl Viro return error; 320260545d0dSAl Viro } 320360545d0dSAl Viro 3204669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 320573d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 32061da177e4SLinus Torvalds { 320730d90494SAl Viro struct file *file; 32089850c056SAl Viro struct path path; 320947237687SAl Viro int opened = 0; 321013aab428SAl Viro int error; 321131e6b01fSNick Piggin 321230d90494SAl Viro file = get_empty_filp(); 32131afc99beSAl Viro if (IS_ERR(file)) 32141afc99beSAl Viro return file; 321531e6b01fSNick Piggin 321630d90494SAl Viro file->f_flags = op->open_flag; 321731e6b01fSNick Piggin 3218bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 321960545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 322060545d0dSAl Viro goto out; 322160545d0dSAl Viro } 322260545d0dSAl Viro 3223980f3ea2SAl Viro error = path_init(dfd, pathname->name, flags, nd); 322431e6b01fSNick Piggin if (unlikely(error)) 32252675a4ebSAl Viro goto out; 322631e6b01fSNick Piggin 3227fe2d35ffSAl Viro current->total_link_count = 0; 3228669abf4eSJeff Layton error = link_path_walk(pathname->name, nd); 322931e6b01fSNick Piggin if (unlikely(error)) 32302675a4ebSAl Viro goto out; 32311da177e4SLinus Torvalds 32322675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 32332675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 32347b9337aaSNick Piggin struct path link = path; 3235def4af30SAl Viro void *cookie; 3236574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 323773d049a4SAl Viro path_put_conditional(&path, nd); 323873d049a4SAl Viro path_put(&nd->path); 32392675a4ebSAl Viro error = -ELOOP; 324040b39136SAl Viro break; 324140b39136SAl Viro } 3242800179c9SKees Cook error = may_follow_link(&link, nd); 3243800179c9SKees Cook if (unlikely(error)) 3244800179c9SKees Cook break; 324573d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 324673d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3247574197e0SAl Viro error = follow_link(&link, nd, &cookie); 3248c3e380b0SAl Viro if (unlikely(error)) 32492675a4ebSAl Viro break; 32502675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 3251574197e0SAl Viro put_link(nd, &link, cookie); 3252806b681cSAl Viro } 325310fa8e62SAl Viro out: 3254893b7775SAl Viro path_cleanup(nd); 32552675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32562675a4ebSAl Viro BUG_ON(!error); 325730d90494SAl Viro put_filp(file); 3258015c3bbcSMiklos Szeredi } 32592675a4ebSAl Viro if (unlikely(error)) { 32602675a4ebSAl Viro if (error == -EOPENSTALE) { 32612675a4ebSAl Viro if (flags & LOOKUP_RCU) 32622675a4ebSAl Viro error = -ECHILD; 32632675a4ebSAl Viro else 32642675a4ebSAl Viro error = -ESTALE; 32652675a4ebSAl Viro } 32662675a4ebSAl Viro file = ERR_PTR(error); 32672675a4ebSAl Viro } 32682675a4ebSAl Viro return file; 3269de459215SKirill Korotaev } 32701da177e4SLinus Torvalds 3271669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3272f9652e10SAl Viro const struct open_flags *op) 327313aab428SAl Viro { 327473d049a4SAl Viro struct nameidata nd; 3275f9652e10SAl Viro int flags = op->lookup_flags; 327613aab428SAl Viro struct file *filp; 327713aab428SAl Viro 327873d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 327913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 328073d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 328113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 328273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 328313aab428SAl Viro return filp; 328413aab428SAl Viro } 328513aab428SAl Viro 328673d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3287f9652e10SAl Viro const char *name, const struct open_flags *op) 328873d049a4SAl Viro { 328973d049a4SAl Viro struct nameidata nd; 329073d049a4SAl Viro struct file *file; 3291669abf4eSJeff Layton struct filename filename = { .name = name }; 3292f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 329373d049a4SAl Viro 329473d049a4SAl Viro nd.root.mnt = mnt; 329573d049a4SAl Viro nd.root.dentry = dentry; 329673d049a4SAl Viro 3297b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 329873d049a4SAl Viro return ERR_PTR(-ELOOP); 329973d049a4SAl Viro 3300669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU); 330173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3302669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags); 330373d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3304669abf4eSJeff Layton file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL); 330573d049a4SAl Viro return file; 330673d049a4SAl Viro } 330773d049a4SAl Viro 33081ac12b4bSJeff Layton struct dentry *kern_path_create(int dfd, const char *pathname, 33091ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33101da177e4SLinus Torvalds { 3311c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3312ed75e95dSAl Viro struct nameidata nd; 3313c30dabfeSJan Kara int err2; 33141ac12b4bSJeff Layton int error; 33151ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 33161ac12b4bSJeff Layton 33171ac12b4bSJeff Layton /* 33181ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 33191ac12b4bSJeff Layton * other flags passed in are ignored! 33201ac12b4bSJeff Layton */ 33211ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 33221ac12b4bSJeff Layton 33231ac12b4bSJeff Layton error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd); 3324ed75e95dSAl Viro if (error) 3325ed75e95dSAl Viro return ERR_PTR(error); 33261da177e4SLinus Torvalds 3327c663e5d8SChristoph Hellwig /* 3328c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3329c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3330c663e5d8SChristoph Hellwig */ 3331ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3332ed75e95dSAl Viro goto out; 3333ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3334ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3335c663e5d8SChristoph Hellwig 3336c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3337c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3338c663e5d8SChristoph Hellwig /* 3339c663e5d8SChristoph Hellwig * Do the final lookup. 3340c663e5d8SChristoph Hellwig */ 3341ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3342ed75e95dSAl Viro dentry = lookup_hash(&nd); 33431da177e4SLinus Torvalds if (IS_ERR(dentry)) 3344a8104a9fSAl Viro goto unlock; 3345c663e5d8SChristoph Hellwig 3346a8104a9fSAl Viro error = -EEXIST; 3347b18825a7SDavid Howells if (d_is_positive(dentry)) 3348a8104a9fSAl Viro goto fail; 3349b18825a7SDavid Howells 3350c663e5d8SChristoph Hellwig /* 3351c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3352c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3353c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3354c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3355c663e5d8SChristoph Hellwig */ 3356ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3357a8104a9fSAl Viro error = -ENOENT; 3358ed75e95dSAl Viro goto fail; 3359e9baf6e5SAl Viro } 3360c30dabfeSJan Kara if (unlikely(err2)) { 3361c30dabfeSJan Kara error = err2; 3362a8104a9fSAl Viro goto fail; 3363c30dabfeSJan Kara } 3364ed75e95dSAl Viro *path = nd.path; 3365e9baf6e5SAl Viro return dentry; 33661da177e4SLinus Torvalds fail: 3367a8104a9fSAl Viro dput(dentry); 3368a8104a9fSAl Viro dentry = ERR_PTR(error); 3369a8104a9fSAl Viro unlock: 3370dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3371c30dabfeSJan Kara if (!err2) 3372c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3373ed75e95dSAl Viro out: 3374dae6ad8fSAl Viro path_put(&nd.path); 3375ed75e95dSAl Viro return dentry; 3376dae6ad8fSAl Viro } 3377dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3378dae6ad8fSAl Viro 3379921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3380921a1650SAl Viro { 3381921a1650SAl Viro dput(dentry); 3382921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3383a8104a9fSAl Viro mnt_drop_write(path->mnt); 3384921a1650SAl Viro path_put(path); 3385921a1650SAl Viro } 3386921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3387921a1650SAl Viro 33881ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 33891ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3390dae6ad8fSAl Viro { 339191a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3392dae6ad8fSAl Viro struct dentry *res; 3393dae6ad8fSAl Viro if (IS_ERR(tmp)) 3394dae6ad8fSAl Viro return ERR_CAST(tmp); 33951ac12b4bSJeff Layton res = kern_path_create(dfd, tmp->name, path, lookup_flags); 3396dae6ad8fSAl Viro putname(tmp); 3397dae6ad8fSAl Viro return res; 3398dae6ad8fSAl Viro } 3399dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3400dae6ad8fSAl Viro 34011a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34021da177e4SLinus Torvalds { 3403a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34041da177e4SLinus Torvalds 34051da177e4SLinus Torvalds if (error) 34061da177e4SLinus Torvalds return error; 34071da177e4SLinus Torvalds 3408975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34091da177e4SLinus Torvalds return -EPERM; 34101da177e4SLinus Torvalds 3411acfa4380SAl Viro if (!dir->i_op->mknod) 34121da177e4SLinus Torvalds return -EPERM; 34131da177e4SLinus Torvalds 341408ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 341508ce5f16SSerge E. Hallyn if (error) 341608ce5f16SSerge E. Hallyn return error; 341708ce5f16SSerge E. Hallyn 34181da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 34191da177e4SLinus Torvalds if (error) 34201da177e4SLinus Torvalds return error; 34211da177e4SLinus Torvalds 34221da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3423a74574aaSStephen Smalley if (!error) 3424f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34251da177e4SLinus Torvalds return error; 34261da177e4SLinus Torvalds } 34274d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 34281da177e4SLinus Torvalds 3429f69aac00SAl Viro static int may_mknod(umode_t mode) 3430463c3197SDave Hansen { 3431463c3197SDave Hansen switch (mode & S_IFMT) { 3432463c3197SDave Hansen case S_IFREG: 3433463c3197SDave Hansen case S_IFCHR: 3434463c3197SDave Hansen case S_IFBLK: 3435463c3197SDave Hansen case S_IFIFO: 3436463c3197SDave Hansen case S_IFSOCK: 3437463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3438463c3197SDave Hansen return 0; 3439463c3197SDave Hansen case S_IFDIR: 3440463c3197SDave Hansen return -EPERM; 3441463c3197SDave Hansen default: 3442463c3197SDave Hansen return -EINVAL; 3443463c3197SDave Hansen } 3444463c3197SDave Hansen } 3445463c3197SDave Hansen 34468208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 34472e4d0924SHeiko Carstens unsigned, dev) 34481da177e4SLinus Torvalds { 34491da177e4SLinus Torvalds struct dentry *dentry; 3450dae6ad8fSAl Viro struct path path; 3451dae6ad8fSAl Viro int error; 3452972567f1SJeff Layton unsigned int lookup_flags = 0; 34531da177e4SLinus Torvalds 34548e4bfca1SAl Viro error = may_mknod(mode); 34558e4bfca1SAl Viro if (error) 34568e4bfca1SAl Viro return error; 3457972567f1SJeff Layton retry: 3458972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3459dae6ad8fSAl Viro if (IS_ERR(dentry)) 3460dae6ad8fSAl Viro return PTR_ERR(dentry); 34612ad94ae6SAl Viro 3462dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3463ce3b0f8dSAl Viro mode &= ~current_umask(); 3464dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3465be6d3e56SKentaro Takeda if (error) 3466a8104a9fSAl Viro goto out; 34671da177e4SLinus Torvalds switch (mode & S_IFMT) { 34681da177e4SLinus Torvalds case 0: case S_IFREG: 3469312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 34701da177e4SLinus Torvalds break; 34711da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3472dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 34731da177e4SLinus Torvalds new_decode_dev(dev)); 34741da177e4SLinus Torvalds break; 34751da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3476dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 34771da177e4SLinus Torvalds break; 34781da177e4SLinus Torvalds } 3479a8104a9fSAl Viro out: 3480921a1650SAl Viro done_path_create(&path, dentry); 3481972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3482972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3483972567f1SJeff Layton goto retry; 3484972567f1SJeff Layton } 34851da177e4SLinus Torvalds return error; 34861da177e4SLinus Torvalds } 34871da177e4SLinus Torvalds 34888208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 34895590ff0dSUlrich Drepper { 34905590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 34915590ff0dSUlrich Drepper } 34925590ff0dSUlrich Drepper 349318bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 34941da177e4SLinus Torvalds { 3495a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34968de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 34971da177e4SLinus Torvalds 34981da177e4SLinus Torvalds if (error) 34991da177e4SLinus Torvalds return error; 35001da177e4SLinus Torvalds 3501acfa4380SAl Viro if (!dir->i_op->mkdir) 35021da177e4SLinus Torvalds return -EPERM; 35031da177e4SLinus Torvalds 35041da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35051da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35061da177e4SLinus Torvalds if (error) 35071da177e4SLinus Torvalds return error; 35081da177e4SLinus Torvalds 35098de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 35108de52778SAl Viro return -EMLINK; 35118de52778SAl Viro 35121da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3513a74574aaSStephen Smalley if (!error) 3514f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 35151da177e4SLinus Torvalds return error; 35161da177e4SLinus Torvalds } 35174d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 35181da177e4SLinus Torvalds 3519a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 35201da177e4SLinus Torvalds { 35216902d925SDave Hansen struct dentry *dentry; 3522dae6ad8fSAl Viro struct path path; 3523dae6ad8fSAl Viro int error; 3524b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 35251da177e4SLinus Torvalds 3526b76d8b82SJeff Layton retry: 3527b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 35286902d925SDave Hansen if (IS_ERR(dentry)) 3529dae6ad8fSAl Viro return PTR_ERR(dentry); 35306902d925SDave Hansen 3531dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3532ce3b0f8dSAl Viro mode &= ~current_umask(); 3533dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3534a8104a9fSAl Viro if (!error) 3535dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3536921a1650SAl Viro done_path_create(&path, dentry); 3537b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3538b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3539b76d8b82SJeff Layton goto retry; 3540b76d8b82SJeff Layton } 35411da177e4SLinus Torvalds return error; 35421da177e4SLinus Torvalds } 35431da177e4SLinus Torvalds 3544a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 35455590ff0dSUlrich Drepper { 35465590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 35475590ff0dSUlrich Drepper } 35485590ff0dSUlrich Drepper 35491da177e4SLinus Torvalds /* 3550a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3551c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3552a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3553a71905f0SSage Weil * then we drop the dentry now. 35541da177e4SLinus Torvalds * 35551da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 35561da177e4SLinus Torvalds * do a 35571da177e4SLinus Torvalds * 35581da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 35591da177e4SLinus Torvalds * return -EBUSY; 35601da177e4SLinus Torvalds * 35611da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 35621da177e4SLinus Torvalds * that is still in use by something else.. 35631da177e4SLinus Torvalds */ 35641da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 35651da177e4SLinus Torvalds { 35661da177e4SLinus Torvalds shrink_dcache_parent(dentry); 35671da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 356898474236SWaiman Long if (dentry->d_lockref.count == 1) 35691da177e4SLinus Torvalds __d_drop(dentry); 35701da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 35711da177e4SLinus Torvalds } 35724d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 35731da177e4SLinus Torvalds 35741da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 35751da177e4SLinus Torvalds { 35761da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 35771da177e4SLinus Torvalds 35781da177e4SLinus Torvalds if (error) 35791da177e4SLinus Torvalds return error; 35801da177e4SLinus Torvalds 3581acfa4380SAl Viro if (!dir->i_op->rmdir) 35821da177e4SLinus Torvalds return -EPERM; 35831da177e4SLinus Torvalds 35841d2ef590SAl Viro dget(dentry); 35851b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3586912dbc15SSage Weil 35871da177e4SLinus Torvalds error = -EBUSY; 35887af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3589912dbc15SSage Weil goto out; 3590912dbc15SSage Weil 35911da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3592912dbc15SSage Weil if (error) 3593912dbc15SSage Weil goto out; 3594912dbc15SSage Weil 35953cebde24SSage Weil shrink_dcache_parent(dentry); 35961da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3597912dbc15SSage Weil if (error) 3598912dbc15SSage Weil goto out; 3599912dbc15SSage Weil 36001da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3601d83c49f3SAl Viro dont_mount(dentry); 36028ed936b5SEric W. Biederman detach_mounts(dentry); 36031da177e4SLinus Torvalds 3604912dbc15SSage Weil out: 3605912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36061d2ef590SAl Viro dput(dentry); 3607912dbc15SSage Weil if (!error) 3608912dbc15SSage Weil d_delete(dentry); 36091da177e4SLinus Torvalds return error; 36101da177e4SLinus Torvalds } 36114d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 36121da177e4SLinus Torvalds 36135590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 36141da177e4SLinus Torvalds { 36151da177e4SLinus Torvalds int error = 0; 361691a27b2aSJeff Layton struct filename *name; 36171da177e4SLinus Torvalds struct dentry *dentry; 36181da177e4SLinus Torvalds struct nameidata nd; 3619c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3620c6ee9206SJeff Layton retry: 3621c6ee9206SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 362291a27b2aSJeff Layton if (IS_ERR(name)) 362391a27b2aSJeff Layton return PTR_ERR(name); 36241da177e4SLinus Torvalds 36251da177e4SLinus Torvalds switch(nd.last_type) { 36261da177e4SLinus Torvalds case LAST_DOTDOT: 36271da177e4SLinus Torvalds error = -ENOTEMPTY; 36281da177e4SLinus Torvalds goto exit1; 36291da177e4SLinus Torvalds case LAST_DOT: 36301da177e4SLinus Torvalds error = -EINVAL; 36311da177e4SLinus Torvalds goto exit1; 36321da177e4SLinus Torvalds case LAST_ROOT: 36331da177e4SLinus Torvalds error = -EBUSY; 36341da177e4SLinus Torvalds goto exit1; 36351da177e4SLinus Torvalds } 36360612d9fbSOGAWA Hirofumi 36370612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3638c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3639c30dabfeSJan Kara if (error) 3640c30dabfeSJan Kara goto exit1; 36410612d9fbSOGAWA Hirofumi 36424ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 364349705b77SChristoph Hellwig dentry = lookup_hash(&nd); 36441da177e4SLinus Torvalds error = PTR_ERR(dentry); 36456902d925SDave Hansen if (IS_ERR(dentry)) 36466902d925SDave Hansen goto exit2; 3647e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3648e6bc45d6STheodore Ts'o error = -ENOENT; 3649e6bc45d6STheodore Ts'o goto exit3; 3650e6bc45d6STheodore Ts'o } 3651be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 3652be6d3e56SKentaro Takeda if (error) 3653c30dabfeSJan Kara goto exit3; 36544ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 36550622753bSDave Hansen exit3: 36561da177e4SLinus Torvalds dput(dentry); 36576902d925SDave Hansen exit2: 36584ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3659c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 36601da177e4SLinus Torvalds exit1: 36611d957f9bSJan Blunck path_put(&nd.path); 36621da177e4SLinus Torvalds putname(name); 3663c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3664c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3665c6ee9206SJeff Layton goto retry; 3666c6ee9206SJeff Layton } 36671da177e4SLinus Torvalds return error; 36681da177e4SLinus Torvalds } 36691da177e4SLinus Torvalds 36703cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 36715590ff0dSUlrich Drepper { 36725590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 36735590ff0dSUlrich Drepper } 36745590ff0dSUlrich Drepper 3675b21996e3SJ. Bruce Fields /** 3676b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3677b21996e3SJ. Bruce Fields * @dir: parent directory 3678b21996e3SJ. Bruce Fields * @dentry: victim 3679b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3680b21996e3SJ. Bruce Fields * 3681b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3682b21996e3SJ. Bruce Fields * 3683b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3684b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3685b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3686b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3687b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3688b21996e3SJ. Bruce Fields * 3689b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3690b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3691b21996e3SJ. Bruce Fields * to be NFS exported. 3692b21996e3SJ. Bruce Fields */ 3693b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 36941da177e4SLinus Torvalds { 36959accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 36961da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 36971da177e4SLinus Torvalds 36981da177e4SLinus Torvalds if (error) 36991da177e4SLinus Torvalds return error; 37001da177e4SLinus Torvalds 3701acfa4380SAl Viro if (!dir->i_op->unlink) 37021da177e4SLinus Torvalds return -EPERM; 37031da177e4SLinus Torvalds 37049accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37058ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37061da177e4SLinus Torvalds error = -EBUSY; 37071da177e4SLinus Torvalds else { 37081da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3709bec1052eSAl Viro if (!error) { 37105a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 37115a14696cSJ. Bruce Fields if (error) 3712b21996e3SJ. Bruce Fields goto out; 37131da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 37148ed936b5SEric W. Biederman if (!error) { 3715d83c49f3SAl Viro dont_mount(dentry); 37168ed936b5SEric W. Biederman detach_mounts(dentry); 37178ed936b5SEric W. Biederman } 3718bec1052eSAl Viro } 37191da177e4SLinus Torvalds } 3720b21996e3SJ. Bruce Fields out: 37219accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 37221da177e4SLinus Torvalds 37231da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 37241da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 37259accbb97SJ. Bruce Fields fsnotify_link_count(target); 37261da177e4SLinus Torvalds d_delete(dentry); 37271da177e4SLinus Torvalds } 37280eeca283SRobert Love 37291da177e4SLinus Torvalds return error; 37301da177e4SLinus Torvalds } 37314d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 37321da177e4SLinus Torvalds 37331da177e4SLinus Torvalds /* 37341da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 37351b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 37361da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 37371da177e4SLinus Torvalds * while waiting on the I/O. 37381da177e4SLinus Torvalds */ 37395590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 37401da177e4SLinus Torvalds { 37412ad94ae6SAl Viro int error; 374291a27b2aSJeff Layton struct filename *name; 37431da177e4SLinus Torvalds struct dentry *dentry; 37441da177e4SLinus Torvalds struct nameidata nd; 37451da177e4SLinus Torvalds struct inode *inode = NULL; 3746b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 37475d18f813SJeff Layton unsigned int lookup_flags = 0; 37485d18f813SJeff Layton retry: 37495d18f813SJeff Layton name = user_path_parent(dfd, pathname, &nd, lookup_flags); 375091a27b2aSJeff Layton if (IS_ERR(name)) 375191a27b2aSJeff Layton return PTR_ERR(name); 37522ad94ae6SAl Viro 37531da177e4SLinus Torvalds error = -EISDIR; 37541da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 37551da177e4SLinus Torvalds goto exit1; 37560612d9fbSOGAWA Hirofumi 37570612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 3758c30dabfeSJan Kara error = mnt_want_write(nd.path.mnt); 3759c30dabfeSJan Kara if (error) 3760c30dabfeSJan Kara goto exit1; 3761b21996e3SJ. Bruce Fields retry_deleg: 37624ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 376349705b77SChristoph Hellwig dentry = lookup_hash(&nd); 37641da177e4SLinus Torvalds error = PTR_ERR(dentry); 37651da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 37661da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 376750338b88STörök Edwin if (nd.last.name[nd.last.len]) 376850338b88STörök Edwin goto slashes; 37691da177e4SLinus Torvalds inode = dentry->d_inode; 3770b18825a7SDavid Howells if (d_is_negative(dentry)) 3771e6bc45d6STheodore Ts'o goto slashes; 37727de9c6eeSAl Viro ihold(inode); 3773be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 3774be6d3e56SKentaro Takeda if (error) 3775c30dabfeSJan Kara goto exit2; 3776b21996e3SJ. Bruce Fields error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode); 37771da177e4SLinus Torvalds exit2: 37781da177e4SLinus Torvalds dput(dentry); 37791da177e4SLinus Torvalds } 37804ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 37811da177e4SLinus Torvalds if (inode) 37821da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3783b21996e3SJ. Bruce Fields inode = NULL; 3784b21996e3SJ. Bruce Fields if (delegated_inode) { 37855a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3786b21996e3SJ. Bruce Fields if (!error) 3787b21996e3SJ. Bruce Fields goto retry_deleg; 3788b21996e3SJ. Bruce Fields } 3789c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 37901da177e4SLinus Torvalds exit1: 37911d957f9bSJan Blunck path_put(&nd.path); 37921da177e4SLinus Torvalds putname(name); 37935d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 37945d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 37955d18f813SJeff Layton inode = NULL; 37965d18f813SJeff Layton goto retry; 37975d18f813SJeff Layton } 37981da177e4SLinus Torvalds return error; 37991da177e4SLinus Torvalds 38001da177e4SLinus Torvalds slashes: 3801b18825a7SDavid Howells if (d_is_negative(dentry)) 3802b18825a7SDavid Howells error = -ENOENT; 380344b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3804b18825a7SDavid Howells error = -EISDIR; 3805b18825a7SDavid Howells else 3806b18825a7SDavid Howells error = -ENOTDIR; 38071da177e4SLinus Torvalds goto exit2; 38081da177e4SLinus Torvalds } 38091da177e4SLinus Torvalds 38102e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 38115590ff0dSUlrich Drepper { 38125590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 38135590ff0dSUlrich Drepper return -EINVAL; 38145590ff0dSUlrich Drepper 38155590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 38165590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 38175590ff0dSUlrich Drepper 38185590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 38195590ff0dSUlrich Drepper } 38205590ff0dSUlrich Drepper 38213480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 38225590ff0dSUlrich Drepper { 38235590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 38245590ff0dSUlrich Drepper } 38255590ff0dSUlrich Drepper 3826db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 38271da177e4SLinus Torvalds { 3828a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38291da177e4SLinus Torvalds 38301da177e4SLinus Torvalds if (error) 38311da177e4SLinus Torvalds return error; 38321da177e4SLinus Torvalds 3833acfa4380SAl Viro if (!dir->i_op->symlink) 38341da177e4SLinus Torvalds return -EPERM; 38351da177e4SLinus Torvalds 38361da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 38371da177e4SLinus Torvalds if (error) 38381da177e4SLinus Torvalds return error; 38391da177e4SLinus Torvalds 38401da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3841a74574aaSStephen Smalley if (!error) 3842f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38431da177e4SLinus Torvalds return error; 38441da177e4SLinus Torvalds } 38454d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 38461da177e4SLinus Torvalds 38472e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 38482e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 38491da177e4SLinus Torvalds { 38502ad94ae6SAl Viro int error; 385191a27b2aSJeff Layton struct filename *from; 38526902d925SDave Hansen struct dentry *dentry; 3853dae6ad8fSAl Viro struct path path; 3854f46d3567SJeff Layton unsigned int lookup_flags = 0; 38551da177e4SLinus Torvalds 38561da177e4SLinus Torvalds from = getname(oldname); 38571da177e4SLinus Torvalds if (IS_ERR(from)) 38581da177e4SLinus Torvalds return PTR_ERR(from); 3859f46d3567SJeff Layton retry: 3860f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 38611da177e4SLinus Torvalds error = PTR_ERR(dentry); 38626902d925SDave Hansen if (IS_ERR(dentry)) 3863dae6ad8fSAl Viro goto out_putname; 38646902d925SDave Hansen 386591a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3866a8104a9fSAl Viro if (!error) 386791a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3868921a1650SAl Viro done_path_create(&path, dentry); 3869f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3870f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3871f46d3567SJeff Layton goto retry; 3872f46d3567SJeff Layton } 38736902d925SDave Hansen out_putname: 38741da177e4SLinus Torvalds putname(from); 38751da177e4SLinus Torvalds return error; 38761da177e4SLinus Torvalds } 38771da177e4SLinus Torvalds 38783480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 38795590ff0dSUlrich Drepper { 38805590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 38815590ff0dSUlrich Drepper } 38825590ff0dSUlrich Drepper 3883146a8595SJ. Bruce Fields /** 3884146a8595SJ. Bruce Fields * vfs_link - create a new link 3885146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3886146a8595SJ. Bruce Fields * @dir: new parent 3887146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3888146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3889146a8595SJ. Bruce Fields * 3890146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3891146a8595SJ. Bruce Fields * 3892146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3893146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3894146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3895146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3896146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3897146a8595SJ. Bruce Fields * 3898146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3899146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3900146a8595SJ. Bruce Fields * to be NFS exported. 3901146a8595SJ. Bruce Fields */ 3902146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39031da177e4SLinus Torvalds { 39041da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39058de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39061da177e4SLinus Torvalds int error; 39071da177e4SLinus Torvalds 39081da177e4SLinus Torvalds if (!inode) 39091da177e4SLinus Torvalds return -ENOENT; 39101da177e4SLinus Torvalds 3911a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 39121da177e4SLinus Torvalds if (error) 39131da177e4SLinus Torvalds return error; 39141da177e4SLinus Torvalds 39151da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 39161da177e4SLinus Torvalds return -EXDEV; 39171da177e4SLinus Torvalds 39181da177e4SLinus Torvalds /* 39191da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 39201da177e4SLinus Torvalds */ 39211da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 39221da177e4SLinus Torvalds return -EPERM; 3923acfa4380SAl Viro if (!dir->i_op->link) 39241da177e4SLinus Torvalds return -EPERM; 39257e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 39261da177e4SLinus Torvalds return -EPERM; 39271da177e4SLinus Torvalds 39281da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 39291da177e4SLinus Torvalds if (error) 39301da177e4SLinus Torvalds return error; 39311da177e4SLinus Torvalds 39327e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3933aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3934f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3935aae8a97dSAneesh Kumar K.V error = -ENOENT; 39368de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 39378de52778SAl Viro error = -EMLINK; 3938146a8595SJ. Bruce Fields else { 3939146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 3940146a8595SJ. Bruce Fields if (!error) 39411da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 3942146a8595SJ. Bruce Fields } 3943f4e0c30cSAl Viro 3944f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 3945f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3946f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 3947f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3948f4e0c30cSAl Viro } 39497e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3950e31e14ecSStephen Smalley if (!error) 39517e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 39521da177e4SLinus Torvalds return error; 39531da177e4SLinus Torvalds } 39544d359507SAl Viro EXPORT_SYMBOL(vfs_link); 39551da177e4SLinus Torvalds 39561da177e4SLinus Torvalds /* 39571da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 39581da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 39591da177e4SLinus Torvalds * newname. --KAB 39601da177e4SLinus Torvalds * 39611da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 39621da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 39631da177e4SLinus Torvalds * and other special files. --ADM 39641da177e4SLinus Torvalds */ 39652e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 39662e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 39671da177e4SLinus Torvalds { 39681da177e4SLinus Torvalds struct dentry *new_dentry; 3969dae6ad8fSAl Viro struct path old_path, new_path; 3970146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 397111a7b371SAneesh Kumar K.V int how = 0; 39721da177e4SLinus Torvalds int error; 39731da177e4SLinus Torvalds 397411a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 3975c04030e1SUlrich Drepper return -EINVAL; 397611a7b371SAneesh Kumar K.V /* 3977f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 3978f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 3979f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 398011a7b371SAneesh Kumar K.V */ 3981f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 3982f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 3983f0cc6ffbSLinus Torvalds return -ENOENT; 398411a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 3985f0cc6ffbSLinus Torvalds } 3986c04030e1SUlrich Drepper 398711a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 398811a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 3989442e31caSJeff Layton retry: 399011a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 39911da177e4SLinus Torvalds if (error) 39922ad94ae6SAl Viro return error; 39932ad94ae6SAl Viro 3994442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 3995442e31caSJeff Layton (how & LOOKUP_REVAL)); 39961da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 39976902d925SDave Hansen if (IS_ERR(new_dentry)) 3998dae6ad8fSAl Viro goto out; 3999dae6ad8fSAl Viro 4000dae6ad8fSAl Viro error = -EXDEV; 4001dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4002dae6ad8fSAl Viro goto out_dput; 4003800179c9SKees Cook error = may_linkat(&old_path); 4004800179c9SKees Cook if (unlikely(error)) 4005800179c9SKees Cook goto out_dput; 4006dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4007be6d3e56SKentaro Takeda if (error) 4008a8104a9fSAl Viro goto out_dput; 4009146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 401075c3f29dSDave Hansen out_dput: 4011921a1650SAl Viro done_path_create(&new_path, new_dentry); 4012146a8595SJ. Bruce Fields if (delegated_inode) { 4013146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4014d22e6338SOleg Drokin if (!error) { 4015d22e6338SOleg Drokin path_put(&old_path); 4016146a8595SJ. Bruce Fields goto retry; 4017146a8595SJ. Bruce Fields } 4018d22e6338SOleg Drokin } 4019442e31caSJeff Layton if (retry_estale(error, how)) { 4020d22e6338SOleg Drokin path_put(&old_path); 4021442e31caSJeff Layton how |= LOOKUP_REVAL; 4022442e31caSJeff Layton goto retry; 4023442e31caSJeff Layton } 40241da177e4SLinus Torvalds out: 40252d8f3038SAl Viro path_put(&old_path); 40261da177e4SLinus Torvalds 40271da177e4SLinus Torvalds return error; 40281da177e4SLinus Torvalds } 40291da177e4SLinus Torvalds 40303480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 40315590ff0dSUlrich Drepper { 4032c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 40335590ff0dSUlrich Drepper } 40345590ff0dSUlrich Drepper 4035bc27027aSMiklos Szeredi /** 4036bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4037bc27027aSMiklos Szeredi * @old_dir: parent of source 4038bc27027aSMiklos Szeredi * @old_dentry: source 4039bc27027aSMiklos Szeredi * @new_dir: parent of destination 4040bc27027aSMiklos Szeredi * @new_dentry: destination 4041bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4042520c8b16SMiklos Szeredi * @flags: rename flags 4043bc27027aSMiklos Szeredi * 4044bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4045bc27027aSMiklos Szeredi * 4046bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4047bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4048bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4049bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4050bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4051bc27027aSMiklos Szeredi * so. 4052bc27027aSMiklos Szeredi * 4053bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4054bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4055bc27027aSMiklos Szeredi * to be NFS exported. 4056bc27027aSMiklos Szeredi * 40571da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 40581da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 40591da177e4SLinus Torvalds * Problems: 4060d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 40611da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 40621da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4063a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 40641da177e4SLinus Torvalds * story. 40656cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 40666cedba89SJ. Bruce Fields * and source (if it is not a directory). 40671b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 40681da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 40691da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4070a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 40711da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 40721da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 40731da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4074a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 40751da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 40761da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 40771da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4078e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 40791b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 40801da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4081c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 40821da177e4SLinus Torvalds * locking]. 40831da177e4SLinus Torvalds */ 40841da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 40858e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4086520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 40871da177e4SLinus Torvalds { 40881da177e4SLinus Torvalds int error; 4089bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 409059b0df21SEric Paris const unsigned char *old_name; 4091bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4092bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4093da1ce067SMiklos Szeredi bool new_is_dir = false; 4094da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 40951da177e4SLinus Torvalds 4096bc27027aSMiklos Szeredi if (source == target) 40971da177e4SLinus Torvalds return 0; 40981da177e4SLinus Torvalds 40991da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41001da177e4SLinus Torvalds if (error) 41011da177e4SLinus Torvalds return error; 41021da177e4SLinus Torvalds 4103da1ce067SMiklos Szeredi if (!target) { 4104a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4105da1ce067SMiklos Szeredi } else { 4106da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4107da1ce067SMiklos Szeredi 4108da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 41091da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4110da1ce067SMiklos Szeredi else 4111da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4112da1ce067SMiklos Szeredi } 41131da177e4SLinus Torvalds if (error) 41141da177e4SLinus Torvalds return error; 41151da177e4SLinus Torvalds 41167177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 41171da177e4SLinus Torvalds return -EPERM; 41181da177e4SLinus Torvalds 4119520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4120520c8b16SMiklos Szeredi return -EINVAL; 4121520c8b16SMiklos Szeredi 4122bc27027aSMiklos Szeredi /* 4123bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4124bc27027aSMiklos Szeredi * we'll need to flip '..'. 4125bc27027aSMiklos Szeredi */ 4126da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4127da1ce067SMiklos Szeredi if (is_dir) { 4128bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4129bc27027aSMiklos Szeredi if (error) 4130bc27027aSMiklos Szeredi return error; 4131bc27027aSMiklos Szeredi } 4132da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4133da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4134da1ce067SMiklos Szeredi if (error) 4135da1ce067SMiklos Szeredi return error; 4136da1ce067SMiklos Szeredi } 4137da1ce067SMiklos Szeredi } 41380eeca283SRobert Love 41390b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 41400b3974ebSMiklos Szeredi flags); 4141bc27027aSMiklos Szeredi if (error) 4142bc27027aSMiklos Szeredi return error; 4143bc27027aSMiklos Szeredi 4144bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4145bc27027aSMiklos Szeredi dget(new_dentry); 4146da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4147bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4148bc27027aSMiklos Szeredi else if (target) 4149bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4150bc27027aSMiklos Szeredi 4151bc27027aSMiklos Szeredi error = -EBUSY; 41527af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4153bc27027aSMiklos Szeredi goto out; 4154bc27027aSMiklos Szeredi 4155da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4156bc27027aSMiklos Szeredi error = -EMLINK; 4157da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4158bc27027aSMiklos Szeredi goto out; 4159da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4160da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4161da1ce067SMiklos Szeredi goto out; 4162da1ce067SMiklos Szeredi } 4163da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4164bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4165da1ce067SMiklos Szeredi if (!is_dir) { 4166bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4167bc27027aSMiklos Szeredi if (error) 4168bc27027aSMiklos Szeredi goto out; 4169da1ce067SMiklos Szeredi } 4170da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4171bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4172bc27027aSMiklos Szeredi if (error) 4173bc27027aSMiklos Szeredi goto out; 4174bc27027aSMiklos Szeredi } 41757177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4176520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4177520c8b16SMiklos Szeredi new_dir, new_dentry); 4178520c8b16SMiklos Szeredi } else { 41797177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4180520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4181520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4182520c8b16SMiklos Szeredi } 4183bc27027aSMiklos Szeredi if (error) 4184bc27027aSMiklos Szeredi goto out; 4185bc27027aSMiklos Szeredi 4186da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 41871da177e4SLinus Torvalds if (is_dir) 4188bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4189bc27027aSMiklos Szeredi dont_mount(new_dentry); 41908ed936b5SEric W. Biederman detach_mounts(new_dentry); 4191bc27027aSMiklos Szeredi } 4192da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4193da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4194bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4195da1ce067SMiklos Szeredi else 4196da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4197da1ce067SMiklos Szeredi } 4198bc27027aSMiklos Szeredi out: 4199da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4200bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4201bc27027aSMiklos Szeredi else if (target) 4202bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4203bc27027aSMiklos Szeredi dput(new_dentry); 4204da1ce067SMiklos Szeredi if (!error) { 4205123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4206da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4207da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4208da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4209da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4210da1ce067SMiklos Szeredi } 4211da1ce067SMiklos Szeredi } 42120eeca283SRobert Love fsnotify_oldname_free(old_name); 42130eeca283SRobert Love 42141da177e4SLinus Torvalds return error; 42151da177e4SLinus Torvalds } 42164d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 42171da177e4SLinus Torvalds 4218520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4219520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 42201da177e4SLinus Torvalds { 42211da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 42221da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 42231da177e4SLinus Torvalds struct dentry *trap; 42241da177e4SLinus Torvalds struct nameidata oldnd, newnd; 42258e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 422691a27b2aSJeff Layton struct filename *from; 422791a27b2aSJeff Layton struct filename *to; 4228c6a94284SJeff Layton unsigned int lookup_flags = 0; 4229c6a94284SJeff Layton bool should_retry = false; 42302ad94ae6SAl Viro int error; 4231520c8b16SMiklos Szeredi 42320d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4233da1ce067SMiklos Szeredi return -EINVAL; 4234da1ce067SMiklos Szeredi 42350d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 42360d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4237520c8b16SMiklos Szeredi return -EINVAL; 4238520c8b16SMiklos Szeredi 42390d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 42400d7a8555SMiklos Szeredi return -EPERM; 42410d7a8555SMiklos Szeredi 4242c6a94284SJeff Layton retry: 4243c6a94284SJeff Layton from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags); 424491a27b2aSJeff Layton if (IS_ERR(from)) { 424591a27b2aSJeff Layton error = PTR_ERR(from); 42461da177e4SLinus Torvalds goto exit; 424791a27b2aSJeff Layton } 42481da177e4SLinus Torvalds 4249c6a94284SJeff Layton to = user_path_parent(newdfd, newname, &newnd, lookup_flags); 425091a27b2aSJeff Layton if (IS_ERR(to)) { 425191a27b2aSJeff Layton error = PTR_ERR(to); 42521da177e4SLinus Torvalds goto exit1; 425391a27b2aSJeff Layton } 42541da177e4SLinus Torvalds 42551da177e4SLinus Torvalds error = -EXDEV; 42564ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 42571da177e4SLinus Torvalds goto exit2; 42581da177e4SLinus Torvalds 42594ac91378SJan Blunck old_dir = oldnd.path.dentry; 42601da177e4SLinus Torvalds error = -EBUSY; 42611da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 42621da177e4SLinus Torvalds goto exit2; 42631da177e4SLinus Torvalds 42644ac91378SJan Blunck new_dir = newnd.path.dentry; 42650a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 42660a7c3937SMiklos Szeredi error = -EEXIST; 42671da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 42681da177e4SLinus Torvalds goto exit2; 42691da177e4SLinus Torvalds 4270c30dabfeSJan Kara error = mnt_want_write(oldnd.path.mnt); 4271c30dabfeSJan Kara if (error) 4272c30dabfeSJan Kara goto exit2; 4273c30dabfeSJan Kara 42740612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 42750612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 4276da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42774e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 42780612d9fbSOGAWA Hirofumi 42798e6d782cSJ. Bruce Fields retry_deleg: 42801da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 42811da177e4SLinus Torvalds 428249705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 42831da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 42841da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 42851da177e4SLinus Torvalds goto exit3; 42861da177e4SLinus Torvalds /* source must exist */ 42871da177e4SLinus Torvalds error = -ENOENT; 4288b18825a7SDavid Howells if (d_is_negative(old_dentry)) 42891da177e4SLinus Torvalds goto exit4; 429049705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 42911da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42921da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 42931da177e4SLinus Torvalds goto exit4; 42940a7c3937SMiklos Szeredi error = -EEXIST; 42950a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 42960a7c3937SMiklos Szeredi goto exit5; 4297da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4298da1ce067SMiklos Szeredi error = -ENOENT; 4299da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4300da1ce067SMiklos Szeredi goto exit5; 4301da1ce067SMiklos Szeredi 4302da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4303da1ce067SMiklos Szeredi error = -ENOTDIR; 4304da1ce067SMiklos Szeredi if (newnd.last.name[newnd.last.len]) 4305da1ce067SMiklos Szeredi goto exit5; 4306da1ce067SMiklos Szeredi } 4307da1ce067SMiklos Szeredi } 43080a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 43090a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 43100a7c3937SMiklos Szeredi error = -ENOTDIR; 43110a7c3937SMiklos Szeredi if (oldnd.last.name[oldnd.last.len]) 43120a7c3937SMiklos Szeredi goto exit5; 4313da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len]) 43140a7c3937SMiklos Szeredi goto exit5; 43150a7c3937SMiklos Szeredi } 43160a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 43170a7c3937SMiklos Szeredi error = -EINVAL; 43180a7c3937SMiklos Szeredi if (old_dentry == trap) 43190a7c3937SMiklos Szeredi goto exit5; 43201da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4321da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43221da177e4SLinus Torvalds error = -ENOTEMPTY; 43231da177e4SLinus Torvalds if (new_dentry == trap) 43241da177e4SLinus Torvalds goto exit5; 43251da177e4SLinus Torvalds 4326be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 43270b3974ebSMiklos Szeredi &newnd.path, new_dentry, flags); 4328be6d3e56SKentaro Takeda if (error) 4329c30dabfeSJan Kara goto exit5; 43301da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 43318e6d782cSJ. Bruce Fields new_dir->d_inode, new_dentry, 4332520c8b16SMiklos Szeredi &delegated_inode, flags); 43331da177e4SLinus Torvalds exit5: 43341da177e4SLinus Torvalds dput(new_dentry); 43351da177e4SLinus Torvalds exit4: 43361da177e4SLinus Torvalds dput(old_dentry); 43371da177e4SLinus Torvalds exit3: 43381da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 43398e6d782cSJ. Bruce Fields if (delegated_inode) { 43408e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 43418e6d782cSJ. Bruce Fields if (!error) 43428e6d782cSJ. Bruce Fields goto retry_deleg; 43438e6d782cSJ. Bruce Fields } 4344c30dabfeSJan Kara mnt_drop_write(oldnd.path.mnt); 43451da177e4SLinus Torvalds exit2: 4346c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4347c6a94284SJeff Layton should_retry = true; 43481d957f9bSJan Blunck path_put(&newnd.path); 43492ad94ae6SAl Viro putname(to); 43501da177e4SLinus Torvalds exit1: 43511d957f9bSJan Blunck path_put(&oldnd.path); 43521da177e4SLinus Torvalds putname(from); 4353c6a94284SJeff Layton if (should_retry) { 4354c6a94284SJeff Layton should_retry = false; 4355c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4356c6a94284SJeff Layton goto retry; 4357c6a94284SJeff Layton } 43582ad94ae6SAl Viro exit: 43591da177e4SLinus Torvalds return error; 43601da177e4SLinus Torvalds } 43611da177e4SLinus Torvalds 4362520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4363520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4364520c8b16SMiklos Szeredi { 4365520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4366520c8b16SMiklos Szeredi } 4367520c8b16SMiklos Szeredi 4368a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 43695590ff0dSUlrich Drepper { 4370520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43715590ff0dSUlrich Drepper } 43725590ff0dSUlrich Drepper 4373787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4374787fb6bcSMiklos Szeredi { 4375787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4376787fb6bcSMiklos Szeredi if (error) 4377787fb6bcSMiklos Szeredi return error; 4378787fb6bcSMiklos Szeredi 4379787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4380787fb6bcSMiklos Szeredi return -EPERM; 4381787fb6bcSMiklos Szeredi 4382787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4383787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4384787fb6bcSMiklos Szeredi } 4385787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4386787fb6bcSMiklos Szeredi 43875d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 43881da177e4SLinus Torvalds { 43895d826c84SAl Viro int len = PTR_ERR(link); 43901da177e4SLinus Torvalds if (IS_ERR(link)) 43911da177e4SLinus Torvalds goto out; 43921da177e4SLinus Torvalds 43931da177e4SLinus Torvalds len = strlen(link); 43941da177e4SLinus Torvalds if (len > (unsigned) buflen) 43951da177e4SLinus Torvalds len = buflen; 43961da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 43971da177e4SLinus Torvalds len = -EFAULT; 43981da177e4SLinus Torvalds out: 43991da177e4SLinus Torvalds return len; 44001da177e4SLinus Torvalds } 44015d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44021da177e4SLinus Torvalds 44031da177e4SLinus Torvalds /* 44041da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44051da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44061da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 44071da177e4SLinus Torvalds */ 44081da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44091da177e4SLinus Torvalds { 44101da177e4SLinus Torvalds struct nameidata nd; 4411cc314eefSLinus Torvalds void *cookie; 4412694a1764SMarcin Slusarz int res; 4413cc314eefSLinus Torvalds 44141da177e4SLinus Torvalds nd.depth = 0; 4415cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 4416694a1764SMarcin Slusarz if (IS_ERR(cookie)) 4417694a1764SMarcin Slusarz return PTR_ERR(cookie); 4418694a1764SMarcin Slusarz 44195d826c84SAl Viro res = readlink_copy(buffer, buflen, nd_get_link(&nd)); 44201da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 4421cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 4422694a1764SMarcin Slusarz return res; 44231da177e4SLinus Torvalds } 44244d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 44251da177e4SLinus Torvalds 44261da177e4SLinus Torvalds /* get the link contents into pagecache */ 44271da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 44281da177e4SLinus Torvalds { 4429ebd09abbSDuane Griffin char *kaddr; 44301da177e4SLinus Torvalds struct page *page; 44311da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4432090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 44331da177e4SLinus Torvalds if (IS_ERR(page)) 44346fe6900eSNick Piggin return (char*)page; 44351da177e4SLinus Torvalds *ppage = page; 4436ebd09abbSDuane Griffin kaddr = kmap(page); 4437ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4438ebd09abbSDuane Griffin return kaddr; 44391da177e4SLinus Torvalds } 44401da177e4SLinus Torvalds 44411da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44421da177e4SLinus Torvalds { 44431da177e4SLinus Torvalds struct page *page = NULL; 44445d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 44451da177e4SLinus Torvalds if (page) { 44461da177e4SLinus Torvalds kunmap(page); 44471da177e4SLinus Torvalds page_cache_release(page); 44481da177e4SLinus Torvalds } 44491da177e4SLinus Torvalds return res; 44501da177e4SLinus Torvalds } 44514d359507SAl Viro EXPORT_SYMBOL(page_readlink); 44521da177e4SLinus Torvalds 4453cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 44541da177e4SLinus Torvalds { 4455cc314eefSLinus Torvalds struct page *page = NULL; 44561da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 4457cc314eefSLinus Torvalds return page; 44581da177e4SLinus Torvalds } 44594d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 44601da177e4SLinus Torvalds 4461cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 44621da177e4SLinus Torvalds { 4463cc314eefSLinus Torvalds struct page *page = cookie; 4464cc314eefSLinus Torvalds 4465cc314eefSLinus Torvalds if (page) { 44661da177e4SLinus Torvalds kunmap(page); 44671da177e4SLinus Torvalds page_cache_release(page); 44681da177e4SLinus Torvalds } 44691da177e4SLinus Torvalds } 44704d359507SAl Viro EXPORT_SYMBOL(page_put_link); 44711da177e4SLinus Torvalds 447254566b2cSNick Piggin /* 447354566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 447454566b2cSNick Piggin */ 447554566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 44761da177e4SLinus Torvalds { 44771da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 44780adb25d2SKirill Korotaev struct page *page; 4479afddba49SNick Piggin void *fsdata; 4480beb497abSDmitriy Monakhov int err; 44811da177e4SLinus Torvalds char *kaddr; 448254566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 448354566b2cSNick Piggin if (nofs) 448454566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 44851da177e4SLinus Torvalds 44867e53cac4SNeilBrown retry: 4487afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 448854566b2cSNick Piggin flags, &page, &fsdata); 44891da177e4SLinus Torvalds if (err) 4490afddba49SNick Piggin goto fail; 4491afddba49SNick Piggin 4492e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 44931da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4494e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4495afddba49SNick Piggin 4496afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4497afddba49SNick Piggin page, fsdata); 44981da177e4SLinus Torvalds if (err < 0) 44991da177e4SLinus Torvalds goto fail; 4500afddba49SNick Piggin if (err < len-1) 4501afddba49SNick Piggin goto retry; 4502afddba49SNick Piggin 45031da177e4SLinus Torvalds mark_inode_dirty(inode); 45041da177e4SLinus Torvalds return 0; 45051da177e4SLinus Torvalds fail: 45061da177e4SLinus Torvalds return err; 45071da177e4SLinus Torvalds } 45084d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 45091da177e4SLinus Torvalds 45100adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 45110adb25d2SKirill Korotaev { 45120adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 451354566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 45140adb25d2SKirill Korotaev } 45154d359507SAl Viro EXPORT_SYMBOL(page_symlink); 45160adb25d2SKirill Korotaev 451792e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 45181da177e4SLinus Torvalds .readlink = generic_readlink, 45191da177e4SLinus Torvalds .follow_link = page_follow_link_light, 45201da177e4SLinus Torvalds .put_link = page_put_link, 45211da177e4SLinus Torvalds }; 45221da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4523