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> 181da177e4SLinus Torvalds #include <linux/module.h> 191da177e4SLinus Torvalds #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/fs.h> 211da177e4SLinus Torvalds #include <linux/namei.h> 221da177e4SLinus Torvalds #include <linux/pagemap.h> 230eeca283SRobert Love #include <linux/fsnotify.h> 241da177e4SLinus Torvalds #include <linux/personality.h> 251da177e4SLinus Torvalds #include <linux/security.h> 266146f0d5SMimi Zohar #include <linux/ima.h> 271da177e4SLinus Torvalds #include <linux/syscalls.h> 281da177e4SLinus Torvalds #include <linux/mount.h> 291da177e4SLinus Torvalds #include <linux/audit.h> 3016f7e0feSRandy Dunlap #include <linux/capability.h> 31834f2a4aSTrond Myklebust #include <linux/file.h> 325590ff0dSUlrich Drepper #include <linux/fcntl.h> 3308ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 345ad4e53bSAl Viro #include <linux/fs_struct.h> 35e77819e5SLinus Torvalds #include <linux/posix_acl.h> 361da177e4SLinus Torvalds #include <asm/uaccess.h> 371da177e4SLinus Torvalds 38e81e3f4dSEric Paris #include "internal.h" 39c7105365SAl Viro #include "mount.h" 40e81e3f4dSEric Paris 411da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 421da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 431da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 441da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 451da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 461da177e4SLinus Torvalds * 471da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 481da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 491da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 501da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 511da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 521da177e4SLinus Torvalds * the special cases of the former code. 531da177e4SLinus Torvalds * 541da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 551da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 561da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 571da177e4SLinus Torvalds * 581da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 591da177e4SLinus Torvalds * resolution to correspond with current state of the code. 601da177e4SLinus Torvalds * 611da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 621da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 631da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 641da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 651da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 661da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 671da177e4SLinus Torvalds */ 681da177e4SLinus Torvalds 691da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 701da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 711da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 721da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 731da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 741da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7525985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 761da177e4SLinus Torvalds * 771da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 781da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 791da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 801da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 811da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 821da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 831da177e4SLinus Torvalds * and in the old Linux semantics. 841da177e4SLinus Torvalds */ 851da177e4SLinus Torvalds 861da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 871da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 881da177e4SLinus Torvalds * 891da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 901da177e4SLinus Torvalds */ 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 931da177e4SLinus Torvalds * inside the path - always follow. 941da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 951da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 961da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 971da177e4SLinus Torvalds * otherwise - don't follow. 981da177e4SLinus Torvalds * (applied in that order). 991da177e4SLinus Torvalds * 1001da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1011da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1021da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1031da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1041da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1051da177e4SLinus Torvalds */ 1061da177e4SLinus Torvalds /* 1071da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 108a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1091da177e4SLinus Torvalds * any extra contention... 1101da177e4SLinus Torvalds */ 1111da177e4SLinus Torvalds 1121da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1131da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1141da177e4SLinus Torvalds * kernel data space before using them.. 1151da177e4SLinus Torvalds * 1161da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1171da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1181da177e4SLinus Torvalds */ 119858119e1SArjan van de Ven static int do_getname(const char __user *filename, char *page) 1201da177e4SLinus Torvalds { 1211da177e4SLinus Torvalds int retval; 1221da177e4SLinus Torvalds unsigned long len = PATH_MAX; 1231da177e4SLinus Torvalds 1241da177e4SLinus Torvalds if (!segment_eq(get_fs(), KERNEL_DS)) { 1251da177e4SLinus Torvalds if ((unsigned long) filename >= TASK_SIZE) 1261da177e4SLinus Torvalds return -EFAULT; 1271da177e4SLinus Torvalds if (TASK_SIZE - (unsigned long) filename < PATH_MAX) 1281da177e4SLinus Torvalds len = TASK_SIZE - (unsigned long) filename; 1291da177e4SLinus Torvalds } 1301da177e4SLinus Torvalds 1311da177e4SLinus Torvalds retval = strncpy_from_user(page, filename, len); 1321da177e4SLinus Torvalds if (retval > 0) { 1331da177e4SLinus Torvalds if (retval < len) 1341da177e4SLinus Torvalds return 0; 1351da177e4SLinus Torvalds return -ENAMETOOLONG; 1361da177e4SLinus Torvalds } else if (!retval) 1371da177e4SLinus Torvalds retval = -ENOENT; 1381da177e4SLinus Torvalds return retval; 1391da177e4SLinus Torvalds } 1401da177e4SLinus Torvalds 1411fa1e7f6SAndy Whitcroft static char *getname_flags(const char __user *filename, int flags, int *empty) 1421da177e4SLinus Torvalds { 1434043cde8SEric Paris char *result = __getname(); 1444043cde8SEric Paris int retval; 1451da177e4SLinus Torvalds 1464043cde8SEric Paris if (!result) 1474043cde8SEric Paris return ERR_PTR(-ENOMEM); 1481da177e4SLinus Torvalds 1494043cde8SEric Paris retval = do_getname(filename, result); 1501da177e4SLinus Torvalds if (retval < 0) { 1511fa1e7f6SAndy Whitcroft if (retval == -ENOENT && empty) 1521fa1e7f6SAndy Whitcroft *empty = 1; 153f52e0c11SAl Viro if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) { 1544043cde8SEric Paris __putname(result); 1554043cde8SEric Paris return ERR_PTR(retval); 1561da177e4SLinus Torvalds } 157f52e0c11SAl Viro } 1581da177e4SLinus Torvalds audit_getname(result); 1591da177e4SLinus Torvalds return result; 1601da177e4SLinus Torvalds } 1611da177e4SLinus Torvalds 162f52e0c11SAl Viro char *getname(const char __user * filename) 163f52e0c11SAl Viro { 1641fa1e7f6SAndy Whitcroft return getname_flags(filename, 0, 0); 165f52e0c11SAl Viro } 166f52e0c11SAl Viro 1671da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 1681da177e4SLinus Torvalds void putname(const char *name) 1691da177e4SLinus Torvalds { 1705ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 1711da177e4SLinus Torvalds audit_putname(name); 1721da177e4SLinus Torvalds else 1731da177e4SLinus Torvalds __putname(name); 1741da177e4SLinus Torvalds } 1751da177e4SLinus Torvalds EXPORT_SYMBOL(putname); 1761da177e4SLinus Torvalds #endif 1771da177e4SLinus Torvalds 178e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 179e77819e5SLinus Torvalds { 18084635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 181e77819e5SLinus Torvalds struct posix_acl *acl; 182e77819e5SLinus Torvalds 183e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 1843567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 1853567866bSAl Viro if (!acl) 186e77819e5SLinus Torvalds return -EAGAIN; 1873567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 1883567866bSAl Viro if (acl == ACL_NOT_CACHED) 189e77819e5SLinus Torvalds return -ECHILD; 190206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 191e77819e5SLinus Torvalds } 192e77819e5SLinus Torvalds 193e77819e5SLinus Torvalds acl = get_cached_acl(inode, ACL_TYPE_ACCESS); 194e77819e5SLinus Torvalds 195e77819e5SLinus Torvalds /* 1964e34e719SChristoph Hellwig * A filesystem can force a ACL callback by just never filling the 1974e34e719SChristoph Hellwig * ACL cache. But normally you'd fill the cache either at inode 1984e34e719SChristoph Hellwig * instantiation time, or on the first ->get_acl call. 199e77819e5SLinus Torvalds * 2004e34e719SChristoph Hellwig * If the filesystem doesn't have a get_acl() function at all, we'll 2014e34e719SChristoph Hellwig * just create the negative cache entry. 202e77819e5SLinus Torvalds */ 203e77819e5SLinus Torvalds if (acl == ACL_NOT_CACHED) { 2044e34e719SChristoph Hellwig if (inode->i_op->get_acl) { 2054e34e719SChristoph Hellwig acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); 2064e34e719SChristoph Hellwig if (IS_ERR(acl)) 2074e34e719SChristoph Hellwig return PTR_ERR(acl); 2084e34e719SChristoph Hellwig } else { 209e77819e5SLinus Torvalds set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); 210e77819e5SLinus Torvalds return -EAGAIN; 211e77819e5SLinus Torvalds } 2124e34e719SChristoph Hellwig } 213e77819e5SLinus Torvalds 214e77819e5SLinus Torvalds if (acl) { 215e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 216e77819e5SLinus Torvalds posix_acl_release(acl); 217e77819e5SLinus Torvalds return error; 218e77819e5SLinus Torvalds } 21984635d68SLinus Torvalds #endif 220e77819e5SLinus Torvalds 221e77819e5SLinus Torvalds return -EAGAIN; 222e77819e5SLinus Torvalds } 223e77819e5SLinus Torvalds 2245909ccaaSLinus Torvalds /* 225948409c7SAndreas Gruenbacher * This does the basic permission checking 2265909ccaaSLinus Torvalds */ 2277e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2285909ccaaSLinus Torvalds { 22926cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2305909ccaaSLinus Torvalds 231e795b717SSerge E. Hallyn if (current_user_ns() != inode_userns(inode)) 232e795b717SSerge E. Hallyn goto other_perms; 233e795b717SSerge E. Hallyn 23414067ff5SLinus Torvalds if (likely(current_fsuid() == inode->i_uid)) 2355909ccaaSLinus Torvalds mode >>= 6; 2365909ccaaSLinus Torvalds else { 237e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2387e40145eSAl Viro int error = check_acl(inode, mask); 2395909ccaaSLinus Torvalds if (error != -EAGAIN) 2405909ccaaSLinus Torvalds return error; 2415909ccaaSLinus Torvalds } 2425909ccaaSLinus Torvalds 2435909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 2445909ccaaSLinus Torvalds mode >>= 3; 2455909ccaaSLinus Torvalds } 2465909ccaaSLinus Torvalds 247e795b717SSerge E. Hallyn other_perms: 2485909ccaaSLinus Torvalds /* 2495909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 2505909ccaaSLinus Torvalds */ 2519c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2525909ccaaSLinus Torvalds return 0; 2535909ccaaSLinus Torvalds return -EACCES; 2545909ccaaSLinus Torvalds } 2551da177e4SLinus Torvalds 2561da177e4SLinus Torvalds /** 2571da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 2581da177e4SLinus Torvalds * @inode: inode to check access rights for 2598fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 2601da177e4SLinus Torvalds * 2611da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 2621da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 2631da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 264b74c79e9SNick Piggin * are used for other things. 265b74c79e9SNick Piggin * 266b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 267b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 268b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 2691da177e4SLinus Torvalds */ 2702830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 2711da177e4SLinus Torvalds { 2725909ccaaSLinus Torvalds int ret; 2731da177e4SLinus Torvalds 2741da177e4SLinus Torvalds /* 275948409c7SAndreas Gruenbacher * Do the basic permission checks. 2761da177e4SLinus Torvalds */ 2777e40145eSAl Viro ret = acl_permission_check(inode, mask); 2785909ccaaSLinus Torvalds if (ret != -EACCES) 2795909ccaaSLinus Torvalds return ret; 2801da177e4SLinus Torvalds 281d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 282d594e7ecSAl Viro /* DACs are overridable for directories */ 283d594e7ecSAl Viro if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) 284d594e7ecSAl Viro return 0; 285d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 286d594e7ecSAl Viro if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) 287d594e7ecSAl Viro return 0; 288d594e7ecSAl Viro return -EACCES; 289d594e7ecSAl Viro } 2901da177e4SLinus Torvalds /* 2911da177e4SLinus Torvalds * Read/write DACs are always overridable. 292d594e7ecSAl Viro * Executable DACs are overridable when there is 293d594e7ecSAl Viro * at least one exec bit set. 2941da177e4SLinus Torvalds */ 295d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 296e795b717SSerge E. Hallyn if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) 2971da177e4SLinus Torvalds return 0; 2981da177e4SLinus Torvalds 2991da177e4SLinus Torvalds /* 3001da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3011da177e4SLinus Torvalds */ 3027ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 303d594e7ecSAl Viro if (mask == MAY_READ) 304e795b717SSerge E. Hallyn if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) 3051da177e4SLinus Torvalds return 0; 3061da177e4SLinus Torvalds 3071da177e4SLinus Torvalds return -EACCES; 3081da177e4SLinus Torvalds } 3091da177e4SLinus Torvalds 3103ddcd056SLinus Torvalds /* 3113ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3123ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3133ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3143ddcd056SLinus Torvalds * permission function, use the fast case". 3153ddcd056SLinus Torvalds */ 3163ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3173ddcd056SLinus Torvalds { 3183ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3193ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3203ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3213ddcd056SLinus Torvalds 3223ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3233ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3243ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3253ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3263ddcd056SLinus Torvalds } 3273ddcd056SLinus Torvalds return generic_permission(inode, mask); 3283ddcd056SLinus Torvalds } 3293ddcd056SLinus Torvalds 330cb23beb5SChristoph Hellwig /** 331cb23beb5SChristoph Hellwig * inode_permission - check for access rights to a given inode 332cb23beb5SChristoph Hellwig * @inode: inode to check permission on 3338fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 334cb23beb5SChristoph Hellwig * 335cb23beb5SChristoph Hellwig * Used to check for read/write/execute permissions on an inode. 336cb23beb5SChristoph Hellwig * We use "fsuid" for this, letting us set arbitrary permissions 337cb23beb5SChristoph Hellwig * for filesystem access without changing the "normal" uids which 338cb23beb5SChristoph Hellwig * are used for other things. 339948409c7SAndreas Gruenbacher * 340948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 341cb23beb5SChristoph Hellwig */ 342f419a2e3SAl Viro int inode_permission(struct inode *inode, int mask) 3431da177e4SLinus Torvalds { 344e6305c43SAl Viro int retval; 3451da177e4SLinus Torvalds 3463ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 34722590e41SMiklos Szeredi umode_t mode = inode->i_mode; 3481da177e4SLinus Torvalds 3491da177e4SLinus Torvalds /* 3501da177e4SLinus Torvalds * Nobody gets write access to a read-only fs. 3511da177e4SLinus Torvalds */ 3521da177e4SLinus Torvalds if (IS_RDONLY(inode) && 3531da177e4SLinus Torvalds (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 3541da177e4SLinus Torvalds return -EROFS; 3551da177e4SLinus Torvalds 3561da177e4SLinus Torvalds /* 3571da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 3581da177e4SLinus Torvalds */ 3591da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 3601da177e4SLinus Torvalds return -EACCES; 3611da177e4SLinus Torvalds } 3621da177e4SLinus Torvalds 3633ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 3641da177e4SLinus Torvalds if (retval) 3651da177e4SLinus Torvalds return retval; 3661da177e4SLinus Torvalds 36708ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 36808ce5f16SSerge E. Hallyn if (retval) 36908ce5f16SSerge E. Hallyn return retval; 37008ce5f16SSerge E. Hallyn 371d09ca739SEric Paris return security_inode_permission(inode, mask); 3721da177e4SLinus Torvalds } 3731da177e4SLinus Torvalds 374f4d6ff89SAl Viro /** 3755dd784d0SJan Blunck * path_get - get a reference to a path 3765dd784d0SJan Blunck * @path: path to get the reference to 3775dd784d0SJan Blunck * 3785dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3795dd784d0SJan Blunck */ 3805dd784d0SJan Blunck void path_get(struct path *path) 3815dd784d0SJan Blunck { 3825dd784d0SJan Blunck mntget(path->mnt); 3835dd784d0SJan Blunck dget(path->dentry); 3845dd784d0SJan Blunck } 3855dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 3865dd784d0SJan Blunck 3875dd784d0SJan Blunck /** 3881d957f9bSJan Blunck * path_put - put a reference to a path 3891d957f9bSJan Blunck * @path: path to put the reference to 3901d957f9bSJan Blunck * 3911d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 3921d957f9bSJan Blunck */ 3931d957f9bSJan Blunck void path_put(struct path *path) 3941da177e4SLinus Torvalds { 3951d957f9bSJan Blunck dput(path->dentry); 3961d957f9bSJan Blunck mntput(path->mnt); 3971da177e4SLinus Torvalds } 3981d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 3991da177e4SLinus Torvalds 40019660af7SAl Viro /* 40131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 40219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 40319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 40419660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 40519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 40619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 40719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 40819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 40931e6b01fSNick Piggin */ 41031e6b01fSNick Piggin 41131e6b01fSNick Piggin /** 41219660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 41319660af7SAl Viro * @nd: nameidata pathwalk data 41419660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 41539191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 41631e6b01fSNick Piggin * 41719660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 41819660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 41919660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 42031e6b01fSNick Piggin */ 42119660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 42231e6b01fSNick Piggin { 42331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 42431e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 4255b6ca027SAl Viro int want_root = 0; 42631e6b01fSNick Piggin 42731e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4285b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4295b6ca027SAl Viro want_root = 1; 43031e6b01fSNick Piggin spin_lock(&fs->lock); 43131e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 43231e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 43331e6b01fSNick Piggin goto err_root; 43431e6b01fSNick Piggin } 43531e6b01fSNick Piggin spin_lock(&parent->d_lock); 43619660af7SAl Viro if (!dentry) { 43719660af7SAl Viro if (!__d_rcu_to_refcount(parent, nd->seq)) 43819660af7SAl Viro goto err_parent; 43919660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 44019660af7SAl Viro } else { 44194c0d4ecSAl Viro if (dentry->d_parent != parent) 44294c0d4ecSAl Viro goto err_parent; 44331e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 44431e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 44519660af7SAl Viro goto err_child; 44631e6b01fSNick Piggin /* 44719660af7SAl Viro * If the sequence check on the child dentry passed, then 44819660af7SAl Viro * the child has not been removed from its parent. This 44919660af7SAl Viro * means the parent dentry must be valid and able to take 45019660af7SAl Viro * a reference at this point. 45131e6b01fSNick Piggin */ 45231e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 45331e6b01fSNick Piggin BUG_ON(!parent->d_count); 45431e6b01fSNick Piggin parent->d_count++; 45531e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 45619660af7SAl Viro } 45731e6b01fSNick Piggin spin_unlock(&parent->d_lock); 4585b6ca027SAl Viro if (want_root) { 45931e6b01fSNick Piggin path_get(&nd->root); 46031e6b01fSNick Piggin spin_unlock(&fs->lock); 46131e6b01fSNick Piggin } 46231e6b01fSNick Piggin mntget(nd->path.mnt); 46331e6b01fSNick Piggin 46431e6b01fSNick Piggin rcu_read_unlock(); 46531e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 46631e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 46731e6b01fSNick Piggin return 0; 46819660af7SAl Viro 46919660af7SAl Viro err_child: 47031e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 47119660af7SAl Viro err_parent: 47231e6b01fSNick Piggin spin_unlock(&parent->d_lock); 47331e6b01fSNick Piggin err_root: 4745b6ca027SAl Viro if (want_root) 47531e6b01fSNick Piggin spin_unlock(&fs->lock); 47631e6b01fSNick Piggin return -ECHILD; 47731e6b01fSNick Piggin } 47831e6b01fSNick Piggin 47931e6b01fSNick Piggin /** 480834f2a4aSTrond Myklebust * release_open_intent - free up open intent resources 481834f2a4aSTrond Myklebust * @nd: pointer to nameidata 482834f2a4aSTrond Myklebust */ 483834f2a4aSTrond Myklebust void release_open_intent(struct nameidata *nd) 484834f2a4aSTrond Myklebust { 4852dab5974SLinus Torvalds struct file *file = nd->intent.open.file; 4862dab5974SLinus Torvalds 4872dab5974SLinus Torvalds if (file && !IS_ERR(file)) { 4882dab5974SLinus Torvalds if (file->f_path.dentry == NULL) 4892dab5974SLinus Torvalds put_filp(file); 490834f2a4aSTrond Myklebust else 4912dab5974SLinus Torvalds fput(file); 4922dab5974SLinus Torvalds } 493834f2a4aSTrond Myklebust } 494834f2a4aSTrond Myklebust 495f60aef7eSAl Viro static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) 49634286d66SNick Piggin { 497f60aef7eSAl Viro return dentry->d_op->d_revalidate(dentry, nd); 49834286d66SNick Piggin } 49934286d66SNick Piggin 5009f1fafeeSAl Viro /** 5019f1fafeeSAl Viro * complete_walk - successful completion of path walk 5029f1fafeeSAl Viro * @nd: pointer nameidata 50339159de2SJeff Layton * 5049f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 5059f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 5069f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 5079f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 5089f1fafeeSAl Viro * need to drop nd->path. 50939159de2SJeff Layton */ 5109f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 51139159de2SJeff Layton { 51216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 51339159de2SJeff Layton int status; 51439159de2SJeff Layton 5159f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 5169f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 5179f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 5189f1fafeeSAl Viro nd->root.mnt = NULL; 5199f1fafeeSAl Viro spin_lock(&dentry->d_lock); 5209f1fafeeSAl Viro if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { 5219f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 5229f1fafeeSAl Viro rcu_read_unlock(); 5239f1fafeeSAl Viro br_read_unlock(vfsmount_lock); 5249f1fafeeSAl Viro return -ECHILD; 5259f1fafeeSAl Viro } 5269f1fafeeSAl Viro BUG_ON(nd->inode != dentry->d_inode); 5279f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 5289f1fafeeSAl Viro mntget(nd->path.mnt); 5299f1fafeeSAl Viro rcu_read_unlock(); 5309f1fafeeSAl Viro br_read_unlock(vfsmount_lock); 5319f1fafeeSAl Viro } 5329f1fafeeSAl Viro 53316c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 53439159de2SJeff Layton return 0; 53539159de2SJeff Layton 53616c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 53716c2cd71SAl Viro return 0; 53816c2cd71SAl Viro 53916c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 54016c2cd71SAl Viro return 0; 54116c2cd71SAl Viro 54216c2cd71SAl Viro /* Note: we do not d_invalidate() */ 54334286d66SNick Piggin status = d_revalidate(dentry, nd); 54439159de2SJeff Layton if (status > 0) 54539159de2SJeff Layton return 0; 54639159de2SJeff Layton 54716c2cd71SAl Viro if (!status) 54839159de2SJeff Layton status = -ESTALE; 54916c2cd71SAl Viro 5509f1fafeeSAl Viro path_put(&nd->path); 55139159de2SJeff Layton return status; 55239159de2SJeff Layton } 55339159de2SJeff Layton 5542a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 5552a737871SAl Viro { 556f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 557f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 5582a737871SAl Viro } 5592a737871SAl Viro 5606de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 5616de88d72SAl Viro 56231e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 56331e6b01fSNick Piggin { 56431e6b01fSNick Piggin if (!nd->root.mnt) { 56531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 566c28cc364SNick Piggin unsigned seq; 567c28cc364SNick Piggin 568c28cc364SNick Piggin do { 569c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 57031e6b01fSNick Piggin nd->root = fs->root; 571c1530019STim Chen nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 572c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 57331e6b01fSNick Piggin } 57431e6b01fSNick Piggin } 57531e6b01fSNick Piggin 576f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 5771da177e4SLinus Torvalds { 57831e6b01fSNick Piggin int ret; 57931e6b01fSNick Piggin 5801da177e4SLinus Torvalds if (IS_ERR(link)) 5811da177e4SLinus Torvalds goto fail; 5821da177e4SLinus Torvalds 5831da177e4SLinus Torvalds if (*link == '/') { 5842a737871SAl Viro set_root(nd); 5851d957f9bSJan Blunck path_put(&nd->path); 5862a737871SAl Viro nd->path = nd->root; 5872a737871SAl Viro path_get(&nd->root); 58816c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 5891da177e4SLinus Torvalds } 59031e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 591b4091d5fSChristoph Hellwig 59231e6b01fSNick Piggin ret = link_path_walk(link, nd); 59331e6b01fSNick Piggin return ret; 5941da177e4SLinus Torvalds fail: 5951d957f9bSJan Blunck path_put(&nd->path); 5961da177e4SLinus Torvalds return PTR_ERR(link); 5971da177e4SLinus Torvalds } 5981da177e4SLinus Torvalds 5991d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 600051d3812SIan Kent { 601051d3812SIan Kent dput(path->dentry); 6024ac91378SJan Blunck if (path->mnt != nd->path.mnt) 603051d3812SIan Kent mntput(path->mnt); 604051d3812SIan Kent } 605051d3812SIan Kent 6067b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6077b9337aaSNick Piggin struct nameidata *nd) 608051d3812SIan Kent { 60931e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6104ac91378SJan Blunck dput(nd->path.dentry); 61131e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6124ac91378SJan Blunck mntput(nd->path.mnt); 6139a229683SHuang Shijie } 61431e6b01fSNick Piggin nd->path.mnt = path->mnt; 6154ac91378SJan Blunck nd->path.dentry = path->dentry; 616051d3812SIan Kent } 617051d3812SIan Kent 618574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 619574197e0SAl Viro { 620574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 621574197e0SAl Viro if (!IS_ERR(cookie) && inode->i_op->put_link) 622574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 623574197e0SAl Viro path_put(link); 624574197e0SAl Viro } 625574197e0SAl Viro 626def4af30SAl Viro static __always_inline int 627574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 6281da177e4SLinus Torvalds { 6291da177e4SLinus Torvalds int error; 6307b9337aaSNick Piggin struct dentry *dentry = link->dentry; 6311da177e4SLinus Torvalds 632844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 633844a3917SAl Viro 6340e794589SAl Viro if (link->mnt == nd->path.mnt) 6350e794589SAl Viro mntget(link->mnt); 6360e794589SAl Viro 637574197e0SAl Viro if (unlikely(current->total_link_count >= 40)) { 638574197e0SAl Viro *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */ 639574197e0SAl Viro path_put(&nd->path); 640574197e0SAl Viro return -ELOOP; 641574197e0SAl Viro } 642574197e0SAl Viro cond_resched(); 643574197e0SAl Viro current->total_link_count++; 644574197e0SAl Viro 6457b9337aaSNick Piggin touch_atime(link->mnt, dentry); 6461da177e4SLinus Torvalds nd_set_link(nd, NULL); 647cd4e91d3SAl Viro 64836f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 64936f3b4f6SAl Viro if (error) { 65036f3b4f6SAl Viro *p = ERR_PTR(error); /* no ->put_link(), please */ 65136f3b4f6SAl Viro path_put(&nd->path); 65236f3b4f6SAl Viro return error; 65336f3b4f6SAl Viro } 65436f3b4f6SAl Viro 65586acdca1SAl Viro nd->last_type = LAST_BIND; 656def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 657def4af30SAl Viro error = PTR_ERR(*p); 658def4af30SAl Viro if (!IS_ERR(*p)) { 6591da177e4SLinus Torvalds char *s = nd_get_link(nd); 660cc314eefSLinus Torvalds error = 0; 6611da177e4SLinus Torvalds if (s) 6621da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 663bcda7652SAl Viro else if (nd->last_type == LAST_BIND) { 66416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 665b21041d0SAl Viro nd->inode = nd->path.dentry->d_inode; 666b21041d0SAl Viro if (nd->inode->i_op->follow_link) { 667bcda7652SAl Viro /* stepped on a _really_ weird one */ 668bcda7652SAl Viro path_put(&nd->path); 669bcda7652SAl Viro error = -ELOOP; 670bcda7652SAl Viro } 671bcda7652SAl Viro } 6721da177e4SLinus Torvalds } 6731da177e4SLinus Torvalds return error; 6741da177e4SLinus Torvalds } 6751da177e4SLinus Torvalds 67631e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 67731e6b01fSNick Piggin { 6780714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 6790714a533SAl Viro struct mount *parent; 68031e6b01fSNick Piggin struct dentry *mountpoint; 68131e6b01fSNick Piggin 6820714a533SAl Viro parent = mnt->mnt_parent; 6830714a533SAl Viro if (&parent->mnt == path->mnt) 68431e6b01fSNick Piggin return 0; 685a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 68631e6b01fSNick Piggin path->dentry = mountpoint; 6870714a533SAl Viro path->mnt = &parent->mnt; 68831e6b01fSNick Piggin return 1; 68931e6b01fSNick Piggin } 69031e6b01fSNick Piggin 691bab77ebfSAl Viro int follow_up(struct path *path) 6921da177e4SLinus Torvalds { 6930714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 6940714a533SAl Viro struct mount *parent; 6951da177e4SLinus Torvalds struct dentry *mountpoint; 69699b7db7bSNick Piggin 69799b7db7bSNick Piggin br_read_lock(vfsmount_lock); 6980714a533SAl Viro parent = mnt->mnt_parent; 6990714a533SAl Viro if (&parent->mnt == path->mnt) { 70099b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 7011da177e4SLinus Torvalds return 0; 7021da177e4SLinus Torvalds } 7030714a533SAl Viro mntget(&parent->mnt); 704a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 70599b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 706bab77ebfSAl Viro dput(path->dentry); 707bab77ebfSAl Viro path->dentry = mountpoint; 708bab77ebfSAl Viro mntput(path->mnt); 7090714a533SAl Viro path->mnt = &parent->mnt; 7101da177e4SLinus Torvalds return 1; 7111da177e4SLinus Torvalds } 7121da177e4SLinus Torvalds 713b5c84bf6SNick Piggin /* 7149875cf80SDavid Howells * Perform an automount 7159875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 7169875cf80SDavid Howells * were called with. 7171da177e4SLinus Torvalds */ 7189875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 7199875cf80SDavid Howells bool *need_mntput) 72031e6b01fSNick Piggin { 7219875cf80SDavid Howells struct vfsmount *mnt; 722ea5b778aSDavid Howells int err; 7239875cf80SDavid Howells 7249875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 7259875cf80SDavid Howells return -EREMOTE; 7269875cf80SDavid Howells 7270ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 7280ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 7290ec26fd0SMiklos Szeredi * the name. 7300ec26fd0SMiklos Szeredi * 7310ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 7325a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 7330ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 7340ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 7350ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 7360ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 7375a30d8a2SDavid Howells */ 7385a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 739d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 7405a30d8a2SDavid Howells path->dentry->d_inode) 7419875cf80SDavid Howells return -EISDIR; 7420ec26fd0SMiklos Szeredi 7439875cf80SDavid Howells current->total_link_count++; 7449875cf80SDavid Howells if (current->total_link_count >= 40) 7459875cf80SDavid Howells return -ELOOP; 7469875cf80SDavid Howells 7479875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 7489875cf80SDavid Howells if (IS_ERR(mnt)) { 7499875cf80SDavid Howells /* 7509875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 7519875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 7529875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 7539875cf80SDavid Howells * 7549875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 7559875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 7569875cf80SDavid Howells * the path is inaccessible and we should say so. 7579875cf80SDavid Howells */ 75849084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 7599875cf80SDavid Howells return -EREMOTE; 7609875cf80SDavid Howells return PTR_ERR(mnt); 76131e6b01fSNick Piggin } 762ea5b778aSDavid Howells 7639875cf80SDavid Howells if (!mnt) /* mount collision */ 7649875cf80SDavid Howells return 0; 7659875cf80SDavid Howells 7668aef1884SAl Viro if (!*need_mntput) { 7678aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 7688aef1884SAl Viro mntget(path->mnt); 7698aef1884SAl Viro *need_mntput = true; 7708aef1884SAl Viro } 77119a167afSAl Viro err = finish_automount(mnt, path); 772ea5b778aSDavid Howells 773ea5b778aSDavid Howells switch (err) { 774ea5b778aSDavid Howells case -EBUSY: 775ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 77619a167afSAl Viro return 0; 777ea5b778aSDavid Howells case 0: 7788aef1884SAl Viro path_put(path); 7799875cf80SDavid Howells path->mnt = mnt; 7809875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 7819875cf80SDavid Howells return 0; 78219a167afSAl Viro default: 78319a167afSAl Viro return err; 7849875cf80SDavid Howells } 78519a167afSAl Viro 786ea5b778aSDavid Howells } 7879875cf80SDavid Howells 7889875cf80SDavid Howells /* 7899875cf80SDavid Howells * Handle a dentry that is managed in some way. 790cc53ce53SDavid Howells * - Flagged for transit management (autofs) 7919875cf80SDavid Howells * - Flagged as mountpoint 7929875cf80SDavid Howells * - Flagged as automount point 7939875cf80SDavid Howells * 7949875cf80SDavid Howells * This may only be called in refwalk mode. 7959875cf80SDavid Howells * 7969875cf80SDavid Howells * Serialization is taken care of in namespace.c 7979875cf80SDavid Howells */ 7989875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 7999875cf80SDavid Howells { 8008aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 8019875cf80SDavid Howells unsigned managed; 8029875cf80SDavid Howells bool need_mntput = false; 8038aef1884SAl Viro int ret = 0; 8049875cf80SDavid Howells 8059875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 8069875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 8079875cf80SDavid Howells * the components of that value change under us */ 8089875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 8099875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 8109875cf80SDavid Howells unlikely(managed != 0)) { 811cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 812cc53ce53SDavid Howells * being held. */ 813cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 814cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 815cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 8161aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 817cc53ce53SDavid Howells if (ret < 0) 8188aef1884SAl Viro break; 819cc53ce53SDavid Howells } 820cc53ce53SDavid Howells 8219875cf80SDavid Howells /* Transit to a mounted filesystem. */ 8229875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 8239875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 8249875cf80SDavid Howells if (mounted) { 8259875cf80SDavid Howells dput(path->dentry); 8269875cf80SDavid Howells if (need_mntput) 827463ffb2eSAl Viro mntput(path->mnt); 828463ffb2eSAl Viro path->mnt = mounted; 829463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 8309875cf80SDavid Howells need_mntput = true; 8319875cf80SDavid Howells continue; 832463ffb2eSAl Viro } 833463ffb2eSAl Viro 8349875cf80SDavid Howells /* Something is mounted on this dentry in another 8359875cf80SDavid Howells * namespace and/or whatever was mounted there in this 8369875cf80SDavid Howells * namespace got unmounted before we managed to get the 8379875cf80SDavid Howells * vfsmount_lock */ 8381da177e4SLinus Torvalds } 8399875cf80SDavid Howells 8409875cf80SDavid Howells /* Handle an automount point */ 8419875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 8429875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 8439875cf80SDavid Howells if (ret < 0) 8448aef1884SAl Viro break; 8459875cf80SDavid Howells continue; 8469875cf80SDavid Howells } 8479875cf80SDavid Howells 8489875cf80SDavid Howells /* We didn't change the current path point */ 8499875cf80SDavid Howells break; 8509875cf80SDavid Howells } 8518aef1884SAl Viro 8528aef1884SAl Viro if (need_mntput && path->mnt == mnt) 8538aef1884SAl Viro mntput(path->mnt); 8548aef1884SAl Viro if (ret == -EISDIR) 8558aef1884SAl Viro ret = 0; 856a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds 859cc53ce53SDavid Howells int follow_down_one(struct path *path) 8601da177e4SLinus Torvalds { 8611da177e4SLinus Torvalds struct vfsmount *mounted; 8621da177e4SLinus Torvalds 8631c755af4SAl Viro mounted = lookup_mnt(path); 8641da177e4SLinus Torvalds if (mounted) { 8659393bd07SAl Viro dput(path->dentry); 8669393bd07SAl Viro mntput(path->mnt); 8679393bd07SAl Viro path->mnt = mounted; 8689393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 8691da177e4SLinus Torvalds return 1; 8701da177e4SLinus Torvalds } 8711da177e4SLinus Torvalds return 0; 8721da177e4SLinus Torvalds } 8731da177e4SLinus Torvalds 87462a7375eSIan Kent static inline bool managed_dentry_might_block(struct dentry *dentry) 87562a7375eSIan Kent { 87662a7375eSIan Kent return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 87762a7375eSIan Kent dentry->d_op->d_manage(dentry, true) < 0); 87862a7375eSIan Kent } 87962a7375eSIan Kent 8809875cf80SDavid Howells /* 881287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 882287548e4SAl Viro * we meet a managed dentry that would need blocking. 8839875cf80SDavid Howells */ 8849875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 885287548e4SAl Viro struct inode **inode) 8869875cf80SDavid Howells { 88762a7375eSIan Kent for (;;) { 888c7105365SAl Viro struct mount *mounted; 88962a7375eSIan Kent /* 89062a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 89162a7375eSIan Kent * that wants to block transit. 89262a7375eSIan Kent */ 893287548e4SAl Viro if (unlikely(managed_dentry_might_block(path->dentry))) 894ab90911fSDavid Howells return false; 89562a7375eSIan Kent 89662a7375eSIan Kent if (!d_mountpoint(path->dentry)) 89762a7375eSIan Kent break; 89862a7375eSIan Kent 8999875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 9009875cf80SDavid Howells if (!mounted) 9019875cf80SDavid Howells break; 902c7105365SAl Viro path->mnt = &mounted->mnt; 903c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 904a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 9059875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 90659430262SLinus Torvalds /* 90759430262SLinus Torvalds * Update the inode too. We don't need to re-check the 90859430262SLinus Torvalds * dentry sequence number here after this d_inode read, 90959430262SLinus Torvalds * because a mount-point is always pinned. 91059430262SLinus Torvalds */ 91159430262SLinus Torvalds *inode = path->dentry->d_inode; 9129875cf80SDavid Howells } 9139875cf80SDavid Howells return true; 9149875cf80SDavid Howells } 9159875cf80SDavid Howells 916dea39376SAl Viro static void follow_mount_rcu(struct nameidata *nd) 917287548e4SAl Viro { 918dea39376SAl Viro while (d_mountpoint(nd->path.dentry)) { 919c7105365SAl Viro struct mount *mounted; 920dea39376SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); 921287548e4SAl Viro if (!mounted) 922287548e4SAl Viro break; 923c7105365SAl Viro nd->path.mnt = &mounted->mnt; 924c7105365SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 925dea39376SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 926287548e4SAl Viro } 927287548e4SAl Viro } 928287548e4SAl Viro 92931e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 93031e6b01fSNick Piggin { 93131e6b01fSNick Piggin set_root_rcu(nd); 93231e6b01fSNick Piggin 93331e6b01fSNick Piggin while (1) { 93431e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 93531e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 93631e6b01fSNick Piggin break; 93731e6b01fSNick Piggin } 93831e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 93931e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 94031e6b01fSNick Piggin struct dentry *parent = old->d_parent; 94131e6b01fSNick Piggin unsigned seq; 94231e6b01fSNick Piggin 94331e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 94431e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 945ef7562d5SAl Viro goto failed; 94631e6b01fSNick Piggin nd->path.dentry = parent; 94731e6b01fSNick Piggin nd->seq = seq; 94831e6b01fSNick Piggin break; 94931e6b01fSNick Piggin } 95031e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 95131e6b01fSNick Piggin break; 95231e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 95331e6b01fSNick Piggin } 954dea39376SAl Viro follow_mount_rcu(nd); 955dea39376SAl Viro nd->inode = nd->path.dentry->d_inode; 95631e6b01fSNick Piggin return 0; 957ef7562d5SAl Viro 958ef7562d5SAl Viro failed: 959ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 9605b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 961ef7562d5SAl Viro nd->root.mnt = NULL; 962ef7562d5SAl Viro rcu_read_unlock(); 963ef7562d5SAl Viro br_read_unlock(vfsmount_lock); 964ef7562d5SAl Viro return -ECHILD; 96531e6b01fSNick Piggin } 96631e6b01fSNick Piggin 9679875cf80SDavid Howells /* 968cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 969cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 970cc53ce53SDavid Howells * caller is permitted to proceed or not. 971cc53ce53SDavid Howells */ 9727cc90cc3SAl Viro int follow_down(struct path *path) 973cc53ce53SDavid Howells { 974cc53ce53SDavid Howells unsigned managed; 975cc53ce53SDavid Howells int ret; 976cc53ce53SDavid Howells 977cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 978cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 979cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 980cc53ce53SDavid Howells * being held. 981cc53ce53SDavid Howells * 982cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 983cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 984cc53ce53SDavid Howells * other than its daemon the right to mount on its 985cc53ce53SDavid Howells * superstructure. 986cc53ce53SDavid Howells * 987cc53ce53SDavid Howells * The filesystem may sleep at this point. 988cc53ce53SDavid Howells */ 989cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 990cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 991cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 992ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 9931aed3e42SAl Viro path->dentry, false); 994cc53ce53SDavid Howells if (ret < 0) 995cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 996cc53ce53SDavid Howells } 997cc53ce53SDavid Howells 998cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 999cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1000cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1001cc53ce53SDavid Howells if (!mounted) 1002cc53ce53SDavid Howells break; 1003cc53ce53SDavid Howells dput(path->dentry); 1004cc53ce53SDavid Howells mntput(path->mnt); 1005cc53ce53SDavid Howells path->mnt = mounted; 1006cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1007cc53ce53SDavid Howells continue; 1008cc53ce53SDavid Howells } 1009cc53ce53SDavid Howells 1010cc53ce53SDavid Howells /* Don't handle automount points here */ 1011cc53ce53SDavid Howells break; 1012cc53ce53SDavid Howells } 1013cc53ce53SDavid Howells return 0; 1014cc53ce53SDavid Howells } 1015cc53ce53SDavid Howells 1016cc53ce53SDavid Howells /* 10179875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 10189875cf80SDavid Howells */ 10199875cf80SDavid Howells static void follow_mount(struct path *path) 10209875cf80SDavid Howells { 10219875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 10229875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10239875cf80SDavid Howells if (!mounted) 10249875cf80SDavid Howells break; 10259875cf80SDavid Howells dput(path->dentry); 10269875cf80SDavid Howells mntput(path->mnt); 10279875cf80SDavid Howells path->mnt = mounted; 10289875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 10299875cf80SDavid Howells } 10309875cf80SDavid Howells } 10319875cf80SDavid Howells 103231e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 10331da177e4SLinus Torvalds { 10342a737871SAl Viro set_root(nd); 1035e518ddb7SAndreas Mohr 10361da177e4SLinus Torvalds while(1) { 10374ac91378SJan Blunck struct dentry *old = nd->path.dentry; 10381da177e4SLinus Torvalds 10392a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 10402a737871SAl Viro nd->path.mnt == nd->root.mnt) { 10411da177e4SLinus Torvalds break; 10421da177e4SLinus Torvalds } 10434ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 10443088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 10453088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 10461da177e4SLinus Torvalds dput(old); 10471da177e4SLinus Torvalds break; 10481da177e4SLinus Torvalds } 10493088dd70SAl Viro if (!follow_up(&nd->path)) 10501da177e4SLinus Torvalds break; 10511da177e4SLinus Torvalds } 105279ed0226SAl Viro follow_mount(&nd->path); 105331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 10541da177e4SLinus Torvalds } 10551da177e4SLinus Torvalds 10561da177e4SLinus Torvalds /* 1057baa03890SNick Piggin * Allocate a dentry with name and parent, and perform a parent 1058baa03890SNick Piggin * directory ->lookup on it. Returns the new dentry, or ERR_PTR 1059baa03890SNick Piggin * on error. parent->d_inode->i_mutex must be held. d_lookup must 1060baa03890SNick Piggin * have verified that no child exists while under i_mutex. 1061baa03890SNick Piggin */ 1062baa03890SNick Piggin static struct dentry *d_alloc_and_lookup(struct dentry *parent, 1063baa03890SNick Piggin struct qstr *name, struct nameidata *nd) 1064baa03890SNick Piggin { 1065baa03890SNick Piggin struct inode *inode = parent->d_inode; 1066baa03890SNick Piggin struct dentry *dentry; 1067baa03890SNick Piggin struct dentry *old; 1068baa03890SNick Piggin 1069baa03890SNick Piggin /* Don't create child dentry for a dead directory. */ 1070baa03890SNick Piggin if (unlikely(IS_DEADDIR(inode))) 1071baa03890SNick Piggin return ERR_PTR(-ENOENT); 1072baa03890SNick Piggin 1073baa03890SNick Piggin dentry = d_alloc(parent, name); 1074baa03890SNick Piggin if (unlikely(!dentry)) 1075baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1076baa03890SNick Piggin 1077baa03890SNick Piggin old = inode->i_op->lookup(inode, dentry, nd); 1078baa03890SNick Piggin if (unlikely(old)) { 1079baa03890SNick Piggin dput(dentry); 1080baa03890SNick Piggin dentry = old; 1081baa03890SNick Piggin } 1082baa03890SNick Piggin return dentry; 1083baa03890SNick Piggin } 1084baa03890SNick Piggin 1085baa03890SNick Piggin /* 108644396f4bSJosef Bacik * We already have a dentry, but require a lookup to be performed on the parent 108744396f4bSJosef Bacik * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error. 108844396f4bSJosef Bacik * parent->d_inode->i_mutex must be held. d_lookup must have verified that no 108944396f4bSJosef Bacik * child exists while under i_mutex. 109044396f4bSJosef Bacik */ 109144396f4bSJosef Bacik static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry, 109244396f4bSJosef Bacik struct nameidata *nd) 109344396f4bSJosef Bacik { 109444396f4bSJosef Bacik struct inode *inode = parent->d_inode; 109544396f4bSJosef Bacik struct dentry *old; 109644396f4bSJosef Bacik 109744396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1098e188dc02SMiklos Szeredi if (unlikely(IS_DEADDIR(inode))) { 1099e188dc02SMiklos Szeredi dput(dentry); 110044396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1101e188dc02SMiklos Szeredi } 110244396f4bSJosef Bacik 110344396f4bSJosef Bacik old = inode->i_op->lookup(inode, dentry, nd); 110444396f4bSJosef Bacik if (unlikely(old)) { 110544396f4bSJosef Bacik dput(dentry); 110644396f4bSJosef Bacik dentry = old; 110744396f4bSJosef Bacik } 110844396f4bSJosef Bacik return dentry; 110944396f4bSJosef Bacik } 111044396f4bSJosef Bacik 111144396f4bSJosef Bacik /* 11121da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 11131da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 11141da177e4SLinus Torvalds * It _is_ time-critical. 11151da177e4SLinus Torvalds */ 11161da177e4SLinus Torvalds static int do_lookup(struct nameidata *nd, struct qstr *name, 111731e6b01fSNick Piggin struct path *path, struct inode **inode) 11181da177e4SLinus Torvalds { 11194ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 112031e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 11215a18fff2SAl Viro int need_reval = 1; 11225a18fff2SAl Viro int status = 1; 11239875cf80SDavid Howells int err; 11249875cf80SDavid Howells 11253cac260aSAl Viro /* 1126b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1127b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1128b04f784eSNick Piggin * do the non-racy lookup, below. 1129b04f784eSNick Piggin */ 113031e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 113131e6b01fSNick Piggin unsigned seq; 113231e6b01fSNick Piggin *inode = nd->inode; 113331e6b01fSNick Piggin dentry = __d_lookup_rcu(parent, name, &seq, inode); 11345a18fff2SAl Viro if (!dentry) 11355a18fff2SAl Viro goto unlazy; 11365a18fff2SAl Viro 113731e6b01fSNick Piggin /* Memory barrier in read_seqcount_begin of child is enough */ 113831e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 113931e6b01fSNick Piggin return -ECHILD; 114031e6b01fSNick Piggin nd->seq = seq; 11415a18fff2SAl Viro 114224643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 11435a18fff2SAl Viro status = d_revalidate(dentry, nd); 11445a18fff2SAl Viro if (unlikely(status <= 0)) { 11455a18fff2SAl Viro if (status != -ECHILD) 11465a18fff2SAl Viro need_reval = 0; 11475a18fff2SAl Viro goto unlazy; 11485a18fff2SAl Viro } 114924643087SAl Viro } 115044396f4bSJosef Bacik if (unlikely(d_need_lookup(dentry))) 115144396f4bSJosef Bacik goto unlazy; 115231e6b01fSNick Piggin path->mnt = mnt; 115331e6b01fSNick Piggin path->dentry = dentry; 1154d6e9bd25SAl Viro if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1155d6e9bd25SAl Viro goto unlazy; 1156d6e9bd25SAl Viro if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1157d6e9bd25SAl Viro goto unlazy; 11589875cf80SDavid Howells return 0; 11595a18fff2SAl Viro unlazy: 116019660af7SAl Viro if (unlazy_walk(nd, dentry)) 11615a18fff2SAl Viro return -ECHILD; 11625a18fff2SAl Viro } else { 116331e6b01fSNick Piggin dentry = __d_lookup(parent, name); 116424643087SAl Viro } 11655a18fff2SAl Viro 116644396f4bSJosef Bacik if (dentry && unlikely(d_need_lookup(dentry))) { 116744396f4bSJosef Bacik dput(dentry); 116844396f4bSJosef Bacik dentry = NULL; 116944396f4bSJosef Bacik } 11705a18fff2SAl Viro retry: 11715a18fff2SAl Viro if (unlikely(!dentry)) { 11725a18fff2SAl Viro struct inode *dir = parent->d_inode; 11735a18fff2SAl Viro BUG_ON(nd->inode != dir); 11745a18fff2SAl Viro 11755a18fff2SAl Viro mutex_lock(&dir->i_mutex); 11765a18fff2SAl Viro dentry = d_lookup(parent, name); 11775a18fff2SAl Viro if (likely(!dentry)) { 11785a18fff2SAl Viro dentry = d_alloc_and_lookup(parent, name, nd); 11795a18fff2SAl Viro if (IS_ERR(dentry)) { 11805a18fff2SAl Viro mutex_unlock(&dir->i_mutex); 11815a18fff2SAl Viro return PTR_ERR(dentry); 11825a18fff2SAl Viro } 11835a18fff2SAl Viro /* known good */ 11845a18fff2SAl Viro need_reval = 0; 11855a18fff2SAl Viro status = 1; 118644396f4bSJosef Bacik } else if (unlikely(d_need_lookup(dentry))) { 118744396f4bSJosef Bacik dentry = d_inode_lookup(parent, dentry, nd); 118844396f4bSJosef Bacik if (IS_ERR(dentry)) { 118944396f4bSJosef Bacik mutex_unlock(&dir->i_mutex); 119044396f4bSJosef Bacik return PTR_ERR(dentry); 119144396f4bSJosef Bacik } 119244396f4bSJosef Bacik /* known good */ 119344396f4bSJosef Bacik need_reval = 0; 119444396f4bSJosef Bacik status = 1; 11955a18fff2SAl Viro } 11965a18fff2SAl Viro mutex_unlock(&dir->i_mutex); 11975a18fff2SAl Viro } 11985a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 11995a18fff2SAl Viro status = d_revalidate(dentry, nd); 12005a18fff2SAl Viro if (unlikely(status <= 0)) { 12015a18fff2SAl Viro if (status < 0) { 12025a18fff2SAl Viro dput(dentry); 12035a18fff2SAl Viro return status; 12045a18fff2SAl Viro } 12055a18fff2SAl Viro if (!d_invalidate(dentry)) { 12065a18fff2SAl Viro dput(dentry); 12075a18fff2SAl Viro dentry = NULL; 12085a18fff2SAl Viro need_reval = 1; 12095a18fff2SAl Viro goto retry; 12105a18fff2SAl Viro } 12115a18fff2SAl Viro } 12125a18fff2SAl Viro 12131da177e4SLinus Torvalds path->mnt = mnt; 12141da177e4SLinus Torvalds path->dentry = dentry; 12159875cf80SDavid Howells err = follow_managed(path, nd->flags); 121689312214SIan Kent if (unlikely(err < 0)) { 121789312214SIan Kent path_put_conditional(path, nd); 12189875cf80SDavid Howells return err; 121989312214SIan Kent } 1220a3fbbde7SAl Viro if (err) 1221a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 122231e6b01fSNick Piggin *inode = path->dentry->d_inode; 12231da177e4SLinus Torvalds return 0; 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds 122652094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 122752094c8aSAl Viro { 122852094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 12294ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 123052094c8aSAl Viro if (err != -ECHILD) 123152094c8aSAl Viro return err; 123219660af7SAl Viro if (unlazy_walk(nd, NULL)) 123352094c8aSAl Viro return -ECHILD; 123452094c8aSAl Viro } 12354ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 123652094c8aSAl Viro } 123752094c8aSAl Viro 12389856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 12399856fa1bSAl Viro { 12409856fa1bSAl Viro if (type == LAST_DOTDOT) { 12419856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 12429856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 12439856fa1bSAl Viro return -ECHILD; 12449856fa1bSAl Viro } else 12459856fa1bSAl Viro follow_dotdot(nd); 12469856fa1bSAl Viro } 12479856fa1bSAl Viro return 0; 12489856fa1bSAl Viro } 12499856fa1bSAl Viro 1250951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1251951361f9SAl Viro { 1252951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1253951361f9SAl Viro path_put(&nd->path); 1254951361f9SAl Viro } else { 1255951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 12565b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1257951361f9SAl Viro nd->root.mnt = NULL; 1258951361f9SAl Viro rcu_read_unlock(); 1259951361f9SAl Viro br_read_unlock(vfsmount_lock); 1260951361f9SAl Viro } 1261951361f9SAl Viro } 1262951361f9SAl Viro 12633ddcd056SLinus Torvalds /* 12643ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 12653ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 12663ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 12673ddcd056SLinus Torvalds * for the common case. 12683ddcd056SLinus Torvalds */ 12697813b94aSLinus Torvalds static inline int should_follow_link(struct inode *inode, int follow) 12703ddcd056SLinus Torvalds { 12713ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { 12723ddcd056SLinus Torvalds if (likely(inode->i_op->follow_link)) 12733ddcd056SLinus Torvalds return follow; 12743ddcd056SLinus Torvalds 12753ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 12763ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 12773ddcd056SLinus Torvalds inode->i_opflags |= IOP_NOFOLLOW; 12783ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 12793ddcd056SLinus Torvalds } 12803ddcd056SLinus Torvalds return 0; 12813ddcd056SLinus Torvalds } 12823ddcd056SLinus Torvalds 1283ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 1284ce57dfc1SAl Viro struct qstr *name, int type, int follow) 1285ce57dfc1SAl Viro { 1286ce57dfc1SAl Viro struct inode *inode; 1287ce57dfc1SAl Viro int err; 1288ce57dfc1SAl Viro /* 1289ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1290ce57dfc1SAl Viro * to be able to know about the current root directory and 1291ce57dfc1SAl Viro * parent relationships. 1292ce57dfc1SAl Viro */ 1293ce57dfc1SAl Viro if (unlikely(type != LAST_NORM)) 1294ce57dfc1SAl Viro return handle_dots(nd, type); 1295ce57dfc1SAl Viro err = do_lookup(nd, name, path, &inode); 1296ce57dfc1SAl Viro if (unlikely(err)) { 1297ce57dfc1SAl Viro terminate_walk(nd); 1298ce57dfc1SAl Viro return err; 1299ce57dfc1SAl Viro } 1300ce57dfc1SAl Viro if (!inode) { 1301ce57dfc1SAl Viro path_to_nameidata(path, nd); 1302ce57dfc1SAl Viro terminate_walk(nd); 1303ce57dfc1SAl Viro return -ENOENT; 1304ce57dfc1SAl Viro } 13057813b94aSLinus Torvalds if (should_follow_link(inode, follow)) { 130619660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 130719660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 130819660af7SAl Viro terminate_walk(nd); 1309ce57dfc1SAl Viro return -ECHILD; 131019660af7SAl Viro } 131119660af7SAl Viro } 1312ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1313ce57dfc1SAl Viro return 1; 1314ce57dfc1SAl Viro } 1315ce57dfc1SAl Viro path_to_nameidata(path, nd); 1316ce57dfc1SAl Viro nd->inode = inode; 1317ce57dfc1SAl Viro return 0; 1318ce57dfc1SAl Viro } 1319ce57dfc1SAl Viro 13201da177e4SLinus Torvalds /* 1321b356379aSAl Viro * This limits recursive symlink follows to 8, while 1322b356379aSAl Viro * limiting consecutive symlinks to 40. 1323b356379aSAl Viro * 1324b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1325b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1326b356379aSAl Viro */ 1327b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1328b356379aSAl Viro { 1329b356379aSAl Viro int res; 1330b356379aSAl Viro 1331b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1332b356379aSAl Viro path_put_conditional(path, nd); 1333b356379aSAl Viro path_put(&nd->path); 1334b356379aSAl Viro return -ELOOP; 1335b356379aSAl Viro } 13361a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1337b356379aSAl Viro 1338b356379aSAl Viro nd->depth++; 1339b356379aSAl Viro current->link_count++; 1340b356379aSAl Viro 1341b356379aSAl Viro do { 1342b356379aSAl Viro struct path link = *path; 1343b356379aSAl Viro void *cookie; 1344574197e0SAl Viro 1345574197e0SAl Viro res = follow_link(&link, nd, &cookie); 1346b356379aSAl Viro if (!res) 1347b356379aSAl Viro res = walk_component(nd, path, &nd->last, 1348b356379aSAl Viro nd->last_type, LOOKUP_FOLLOW); 1349574197e0SAl Viro put_link(nd, &link, cookie); 1350b356379aSAl Viro } while (res > 0); 1351b356379aSAl Viro 1352b356379aSAl Viro current->link_count--; 1353b356379aSAl Viro nd->depth--; 1354b356379aSAl Viro return res; 1355b356379aSAl Viro } 1356b356379aSAl Viro 1357b356379aSAl Viro /* 13583ddcd056SLinus Torvalds * We really don't want to look at inode->i_op->lookup 13593ddcd056SLinus Torvalds * when we don't have to. So we keep a cache bit in 13603ddcd056SLinus Torvalds * the inode ->i_opflags field that says "yes, we can 13613ddcd056SLinus Torvalds * do lookup on this inode". 13623ddcd056SLinus Torvalds */ 13633ddcd056SLinus Torvalds static inline int can_lookup(struct inode *inode) 13643ddcd056SLinus Torvalds { 13653ddcd056SLinus Torvalds if (likely(inode->i_opflags & IOP_LOOKUP)) 13663ddcd056SLinus Torvalds return 1; 13673ddcd056SLinus Torvalds if (likely(!inode->i_op->lookup)) 13683ddcd056SLinus Torvalds return 0; 13693ddcd056SLinus Torvalds 13703ddcd056SLinus Torvalds /* We do this once for the lifetime of the inode */ 13713ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 13723ddcd056SLinus Torvalds inode->i_opflags |= IOP_LOOKUP; 13733ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 13743ddcd056SLinus Torvalds return 1; 13753ddcd056SLinus Torvalds } 13763ddcd056SLinus Torvalds 13770145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 13780145acc2SLinus Torvalds { 13790145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 13800145acc2SLinus Torvalds while (len--) 13810145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 13820145acc2SLinus Torvalds return end_name_hash(hash); 13830145acc2SLinus Torvalds } 1384ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 13850145acc2SLinus Torvalds 13863ddcd056SLinus Torvalds /* 1387200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1388200e9ef7SLinus Torvalds * one character. 1389200e9ef7SLinus Torvalds */ 1390200e9ef7SLinus Torvalds static inline unsigned long hash_name(const char *name, unsigned int *hashp) 1391200e9ef7SLinus Torvalds { 1392200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1393200e9ef7SLinus Torvalds unsigned long len = 0, c; 1394200e9ef7SLinus Torvalds 1395200e9ef7SLinus Torvalds c = (unsigned char)*name; 1396200e9ef7SLinus Torvalds do { 1397200e9ef7SLinus Torvalds len++; 1398200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1399200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1400200e9ef7SLinus Torvalds } while (c && c != '/'); 1401200e9ef7SLinus Torvalds *hashp = end_name_hash(hash); 1402200e9ef7SLinus Torvalds return len; 1403200e9ef7SLinus Torvalds } 1404200e9ef7SLinus Torvalds 1405200e9ef7SLinus Torvalds /* 14061da177e4SLinus Torvalds * Name resolution. 1407ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1408ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 14091da177e4SLinus Torvalds * 1410ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1411ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 14121da177e4SLinus Torvalds */ 14136de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 14141da177e4SLinus Torvalds { 14151da177e4SLinus Torvalds struct path next; 14161da177e4SLinus Torvalds int err; 14171da177e4SLinus Torvalds 14181da177e4SLinus Torvalds while (*name=='/') 14191da177e4SLinus Torvalds name++; 14201da177e4SLinus Torvalds if (!*name) 1421086e183aSAl Viro return 0; 14221da177e4SLinus Torvalds 14231da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 14241da177e4SLinus Torvalds for(;;) { 14251da177e4SLinus Torvalds struct qstr this; 1426200e9ef7SLinus Torvalds long len; 1427fe479a58SAl Viro int type; 14281da177e4SLinus Torvalds 142952094c8aSAl Viro err = may_lookup(nd); 14301da177e4SLinus Torvalds if (err) 14311da177e4SLinus Torvalds break; 14321da177e4SLinus Torvalds 1433200e9ef7SLinus Torvalds len = hash_name(name, &this.hash); 14341da177e4SLinus Torvalds this.name = name; 1435200e9ef7SLinus Torvalds this.len = len; 14361da177e4SLinus Torvalds 1437fe479a58SAl Viro type = LAST_NORM; 1438200e9ef7SLinus Torvalds if (name[0] == '.') switch (len) { 1439fe479a58SAl Viro case 2: 1440200e9ef7SLinus Torvalds if (name[1] == '.') { 1441fe479a58SAl Viro type = LAST_DOTDOT; 144216c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 144316c2cd71SAl Viro } 1444fe479a58SAl Viro break; 1445fe479a58SAl Viro case 1: 1446fe479a58SAl Viro type = LAST_DOT; 1447fe479a58SAl Viro } 14485a202bcdSAl Viro if (likely(type == LAST_NORM)) { 14495a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 145016c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 14515a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 14525a202bcdSAl Viro err = parent->d_op->d_hash(parent, nd->inode, 14535a202bcdSAl Viro &this); 14545a202bcdSAl Viro if (err < 0) 14555a202bcdSAl Viro break; 14565a202bcdSAl Viro } 14575a202bcdSAl Viro } 1458fe479a58SAl Viro 1459200e9ef7SLinus Torvalds if (!name[len]) 14601da177e4SLinus Torvalds goto last_component; 1461200e9ef7SLinus Torvalds /* 1462200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1463200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1464200e9ef7SLinus Torvalds */ 1465200e9ef7SLinus Torvalds do { 1466200e9ef7SLinus Torvalds len++; 1467200e9ef7SLinus Torvalds } while (unlikely(name[len] == '/')); 1468200e9ef7SLinus Torvalds if (!name[len]) 1469b356379aSAl Viro goto last_component; 1470200e9ef7SLinus Torvalds name += len; 14711da177e4SLinus Torvalds 1472ce57dfc1SAl Viro err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); 1473ce57dfc1SAl Viro if (err < 0) 1474ce57dfc1SAl Viro return err; 1475fe479a58SAl Viro 1476ce57dfc1SAl Viro if (err) { 1477b356379aSAl Viro err = nested_symlink(&next, nd); 14781da177e4SLinus Torvalds if (err) 1479a7472babSAl Viro return err; 148031e6b01fSNick Piggin } 14813ddcd056SLinus Torvalds if (can_lookup(nd->inode)) 14821da177e4SLinus Torvalds continue; 14833ddcd056SLinus Torvalds err = -ENOTDIR; 14843ddcd056SLinus Torvalds break; 14851da177e4SLinus Torvalds /* here ends the main loop */ 14861da177e4SLinus Torvalds 14871da177e4SLinus Torvalds last_component: 1488ce57dfc1SAl Viro nd->last = this; 1489ce57dfc1SAl Viro nd->last_type = type; 1490ce57dfc1SAl Viro return 0; 1491ce57dfc1SAl Viro } 1492951361f9SAl Viro terminate_walk(nd); 14931da177e4SLinus Torvalds return err; 14941da177e4SLinus Torvalds } 14951da177e4SLinus Torvalds 149670e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 149770e9b357SAl Viro struct nameidata *nd, struct file **fp) 149831e6b01fSNick Piggin { 149931e6b01fSNick Piggin int retval = 0; 150031e6b01fSNick Piggin int fput_needed; 150131e6b01fSNick Piggin struct file *file; 150231e6b01fSNick Piggin 150331e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 150416c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 150531e6b01fSNick Piggin nd->depth = 0; 15065b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 15075b6ca027SAl Viro struct inode *inode = nd->root.dentry->d_inode; 150873d049a4SAl Viro if (*name) { 15095b6ca027SAl Viro if (!inode->i_op->lookup) 15105b6ca027SAl Viro return -ENOTDIR; 15115b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 15125b6ca027SAl Viro if (retval) 15135b6ca027SAl Viro return retval; 151473d049a4SAl Viro } 15155b6ca027SAl Viro nd->path = nd->root; 15165b6ca027SAl Viro nd->inode = inode; 15175b6ca027SAl Viro if (flags & LOOKUP_RCU) { 15185b6ca027SAl Viro br_read_lock(vfsmount_lock); 15195b6ca027SAl Viro rcu_read_lock(); 15205b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 15215b6ca027SAl Viro } else { 15225b6ca027SAl Viro path_get(&nd->path); 15235b6ca027SAl Viro } 15245b6ca027SAl Viro return 0; 15255b6ca027SAl Viro } 15265b6ca027SAl Viro 152731e6b01fSNick Piggin nd->root.mnt = NULL; 152831e6b01fSNick Piggin 152931e6b01fSNick Piggin if (*name=='/') { 1530e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 153131e6b01fSNick Piggin br_read_lock(vfsmount_lock); 153231e6b01fSNick Piggin rcu_read_lock(); 1533e41f7d4eSAl Viro set_root_rcu(nd); 1534e41f7d4eSAl Viro } else { 1535e41f7d4eSAl Viro set_root(nd); 1536e41f7d4eSAl Viro path_get(&nd->root); 1537e41f7d4eSAl Viro } 153831e6b01fSNick Piggin nd->path = nd->root; 153931e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1540e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 154131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1542c28cc364SNick Piggin unsigned seq; 154331e6b01fSNick Piggin 154431e6b01fSNick Piggin br_read_lock(vfsmount_lock); 154531e6b01fSNick Piggin rcu_read_lock(); 154631e6b01fSNick Piggin 1547c28cc364SNick Piggin do { 1548c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 154931e6b01fSNick Piggin nd->path = fs->pwd; 1550c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1551c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1552e41f7d4eSAl Viro } else { 1553e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1554e41f7d4eSAl Viro } 155531e6b01fSNick Piggin } else { 155631e6b01fSNick Piggin struct dentry *dentry; 155731e6b01fSNick Piggin 15581abf0c71SAl Viro file = fget_raw_light(dfd, &fput_needed); 155931e6b01fSNick Piggin retval = -EBADF; 156031e6b01fSNick Piggin if (!file) 156131e6b01fSNick Piggin goto out_fail; 156231e6b01fSNick Piggin 156331e6b01fSNick Piggin dentry = file->f_path.dentry; 156431e6b01fSNick Piggin 1565f52e0c11SAl Viro if (*name) { 156631e6b01fSNick Piggin retval = -ENOTDIR; 156731e6b01fSNick Piggin if (!S_ISDIR(dentry->d_inode->i_mode)) 156831e6b01fSNick Piggin goto fput_fail; 156931e6b01fSNick Piggin 15704ad5abb3SAl Viro retval = inode_permission(dentry->d_inode, MAY_EXEC); 157131e6b01fSNick Piggin if (retval) 157231e6b01fSNick Piggin goto fput_fail; 1573f52e0c11SAl Viro } 157431e6b01fSNick Piggin 157531e6b01fSNick Piggin nd->path = file->f_path; 1576e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 157731e6b01fSNick Piggin if (fput_needed) 157870e9b357SAl Viro *fp = file; 1579c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 158031e6b01fSNick Piggin br_read_lock(vfsmount_lock); 158131e6b01fSNick Piggin rcu_read_lock(); 15825590ff0dSUlrich Drepper } else { 15835dd784d0SJan Blunck path_get(&file->f_path); 15845590ff0dSUlrich Drepper fput_light(file, fput_needed); 15851da177e4SLinus Torvalds } 1586e41f7d4eSAl Viro } 1587e41f7d4eSAl Viro 158831e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 15899b4a9b14SAl Viro return 0; 15902dfdd266SJosef 'Jeff' Sipek 15919b4a9b14SAl Viro fput_fail: 15929b4a9b14SAl Viro fput_light(file, fput_needed); 15939b4a9b14SAl Viro out_fail: 15949b4a9b14SAl Viro return retval; 15959b4a9b14SAl Viro } 15969b4a9b14SAl Viro 1597bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1598bd92d7feSAl Viro { 1599bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1600bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1601bd92d7feSAl Viro 1602bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 1603bd92d7feSAl Viro return walk_component(nd, path, &nd->last, nd->last_type, 1604bd92d7feSAl Viro nd->flags & LOOKUP_FOLLOW); 1605bd92d7feSAl Viro } 1606bd92d7feSAl Viro 16079b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1608ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 16099b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 16109b4a9b14SAl Viro { 161170e9b357SAl Viro struct file *base = NULL; 1612bd92d7feSAl Viro struct path path; 1613bd92d7feSAl Viro int err; 161431e6b01fSNick Piggin 161531e6b01fSNick Piggin /* 161631e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 161731e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 161831e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 161931e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 162031e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 162131e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 162231e6b01fSNick Piggin * analogue, foo_rcu(). 162331e6b01fSNick Piggin * 162431e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 162531e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 162631e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 162731e6b01fSNick Piggin * be able to complete). 162831e6b01fSNick Piggin */ 1629bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1630ee0827cdSAl Viro 1631bd92d7feSAl Viro if (unlikely(err)) 1632bd92d7feSAl Viro return err; 1633ee0827cdSAl Viro 1634ee0827cdSAl Viro current->total_link_count = 0; 1635bd92d7feSAl Viro err = link_path_walk(name, nd); 1636bd92d7feSAl Viro 1637bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1638bd92d7feSAl Viro err = lookup_last(nd, &path); 1639bd92d7feSAl Viro while (err > 0) { 1640bd92d7feSAl Viro void *cookie; 1641bd92d7feSAl Viro struct path link = path; 1642bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1643574197e0SAl Viro err = follow_link(&link, nd, &cookie); 1644bd92d7feSAl Viro if (!err) 1645bd92d7feSAl Viro err = lookup_last(nd, &path); 1646574197e0SAl Viro put_link(nd, &link, cookie); 1647bd92d7feSAl Viro } 1648bd92d7feSAl Viro } 1649ee0827cdSAl Viro 16509f1fafeeSAl Viro if (!err) 16519f1fafeeSAl Viro err = complete_walk(nd); 1652bd92d7feSAl Viro 1653bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 1654bd92d7feSAl Viro if (!nd->inode->i_op->lookup) { 1655bd92d7feSAl Viro path_put(&nd->path); 1656bd23a539SAl Viro err = -ENOTDIR; 1657bd92d7feSAl Viro } 1658bd92d7feSAl Viro } 165916c2cd71SAl Viro 166070e9b357SAl Viro if (base) 166170e9b357SAl Viro fput(base); 1662ee0827cdSAl Viro 16635b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 166431e6b01fSNick Piggin path_put(&nd->root); 166531e6b01fSNick Piggin nd->root.mnt = NULL; 166631e6b01fSNick Piggin } 1667bd92d7feSAl Viro return err; 166831e6b01fSNick Piggin } 166931e6b01fSNick Piggin 1670ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1671ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1672ee0827cdSAl Viro { 1673ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1674ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1675ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1676ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1677ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1678ee0827cdSAl Viro 167931e6b01fSNick Piggin if (likely(!retval)) { 168031e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 168131e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 168231e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 168331e6b01fSNick Piggin } 168431e6b01fSNick Piggin } 1685170aa3d0SUlrich Drepper return retval; 16861da177e4SLinus Torvalds } 16871da177e4SLinus Torvalds 1688c9c6cac0SAl Viro int kern_path_parent(const char *name, struct nameidata *nd) 16895590ff0dSUlrich Drepper { 1690c9c6cac0SAl Viro return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); 16915590ff0dSUlrich Drepper } 16925590ff0dSUlrich Drepper 1693d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1694d1811465SAl Viro { 1695d1811465SAl Viro struct nameidata nd; 1696d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1697d1811465SAl Viro if (!res) 1698d1811465SAl Viro *path = nd.path; 1699d1811465SAl Viro return res; 1700d1811465SAl Viro } 1701d1811465SAl Viro 170216f18200SJosef 'Jeff' Sipek /** 170316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 170416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 170516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 170616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 170716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 1708e0a01249SAl Viro * @path: pointer to struct path to fill 170916f18200SJosef 'Jeff' Sipek */ 171016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 171116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 1712e0a01249SAl Viro struct path *path) 171316f18200SJosef 'Jeff' Sipek { 1714e0a01249SAl Viro struct nameidata nd; 1715e0a01249SAl Viro int err; 1716e0a01249SAl Viro nd.root.dentry = dentry; 1717e0a01249SAl Viro nd.root.mnt = mnt; 1718e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 17195b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 1720e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 1721e0a01249SAl Viro if (!err) 1722e0a01249SAl Viro *path = nd.path; 1723e0a01249SAl Viro return err; 172416f18200SJosef 'Jeff' Sipek } 172516f18200SJosef 'Jeff' Sipek 1726eead1911SChristoph Hellwig static struct dentry *__lookup_hash(struct qstr *name, 1727eead1911SChristoph Hellwig struct dentry *base, struct nameidata *nd) 17281da177e4SLinus Torvalds { 172981fca444SChristoph Hellwig struct inode *inode = base->d_inode; 17301da177e4SLinus Torvalds struct dentry *dentry; 17311da177e4SLinus Torvalds int err; 17321da177e4SLinus Torvalds 17334ad5abb3SAl Viro err = inode_permission(inode, MAY_EXEC); 173481fca444SChristoph Hellwig if (err) 173581fca444SChristoph Hellwig return ERR_PTR(err); 17361da177e4SLinus Torvalds 17371da177e4SLinus Torvalds /* 1738b04f784eSNick Piggin * Don't bother with __d_lookup: callers are for creat as 1739b04f784eSNick Piggin * well as unlink, so a lot of the time it would cost 1740b04f784eSNick Piggin * a double lookup. 17416e6b1bd1SAl Viro */ 17426e6b1bd1SAl Viro dentry = d_lookup(base, name); 17436e6b1bd1SAl Viro 174444396f4bSJosef Bacik if (dentry && d_need_lookup(dentry)) { 174544396f4bSJosef Bacik /* 174644396f4bSJosef Bacik * __lookup_hash is called with the parent dir's i_mutex already 174744396f4bSJosef Bacik * held, so we are good to go here. 174844396f4bSJosef Bacik */ 174944396f4bSJosef Bacik dentry = d_inode_lookup(base, dentry, nd); 175044396f4bSJosef Bacik if (IS_ERR(dentry)) 175144396f4bSJosef Bacik return dentry; 175244396f4bSJosef Bacik } 175344396f4bSJosef Bacik 1754d2d9e9fbSAl Viro if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) { 1755d2d9e9fbSAl Viro int status = d_revalidate(dentry, nd); 1756d2d9e9fbSAl Viro if (unlikely(status <= 0)) { 1757d2d9e9fbSAl Viro /* 1758d2d9e9fbSAl Viro * The dentry failed validation. 1759d2d9e9fbSAl Viro * If d_revalidate returned 0 attempt to invalidate 1760d2d9e9fbSAl Viro * the dentry otherwise d_revalidate is asking us 1761d2d9e9fbSAl Viro * to return a fail status. 1762d2d9e9fbSAl Viro */ 1763d2d9e9fbSAl Viro if (status < 0) { 1764d2d9e9fbSAl Viro dput(dentry); 1765d2d9e9fbSAl Viro return ERR_PTR(status); 1766d2d9e9fbSAl Viro } else if (!d_invalidate(dentry)) { 1767d2d9e9fbSAl Viro dput(dentry); 1768d2d9e9fbSAl Viro dentry = NULL; 1769d2d9e9fbSAl Viro } 1770d2d9e9fbSAl Viro } 1771d2d9e9fbSAl Viro } 17726e6b1bd1SAl Viro 17731da177e4SLinus Torvalds if (!dentry) 1774baa03890SNick Piggin dentry = d_alloc_and_lookup(base, name, nd); 17755a202bcdSAl Viro 17761da177e4SLinus Torvalds return dentry; 17771da177e4SLinus Torvalds } 17781da177e4SLinus Torvalds 1779057f6c01SJames Morris /* 1780057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1781057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1782057f6c01SJames Morris * SMP-safe. 1783057f6c01SJames Morris */ 1784a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 17851da177e4SLinus Torvalds { 17864ac91378SJan Blunck return __lookup_hash(&nd->last, nd->path.dentry, nd); 17871da177e4SLinus Torvalds } 17881da177e4SLinus Torvalds 1789eead1911SChristoph Hellwig /** 1790a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1791eead1911SChristoph Hellwig * @name: pathname component to lookup 1792eead1911SChristoph Hellwig * @base: base directory to lookup from 1793eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1794eead1911SChristoph Hellwig * 1795a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1796a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1797eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1798eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1799eead1911SChristoph Hellwig */ 1800057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1801057f6c01SJames Morris { 1802057f6c01SJames Morris struct qstr this; 18036a96ba54SAl Viro unsigned int c; 1804057f6c01SJames Morris 18052f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 18062f9092e1SDavid Woodhouse 18076a96ba54SAl Viro this.name = name; 18086a96ba54SAl Viro this.len = len; 18090145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 18106a96ba54SAl Viro if (!len) 18116a96ba54SAl Viro return ERR_PTR(-EACCES); 18126a96ba54SAl Viro 18136a96ba54SAl Viro while (len--) { 18146a96ba54SAl Viro c = *(const unsigned char *)name++; 18156a96ba54SAl Viro if (c == '/' || c == '\0') 18166a96ba54SAl Viro return ERR_PTR(-EACCES); 18176a96ba54SAl Viro } 18185a202bcdSAl Viro /* 18195a202bcdSAl Viro * See if the low-level filesystem might want 18205a202bcdSAl Viro * to use its own hash.. 18215a202bcdSAl Viro */ 18225a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 18235a202bcdSAl Viro int err = base->d_op->d_hash(base, base->d_inode, &this); 18245a202bcdSAl Viro if (err < 0) 18255a202bcdSAl Viro return ERR_PTR(err); 18265a202bcdSAl Viro } 1827eead1911SChristoph Hellwig 182849705b77SChristoph Hellwig return __lookup_hash(&this, base, NULL); 1829057f6c01SJames Morris } 1830057f6c01SJames Morris 18311fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 18321fa1e7f6SAndy Whitcroft struct path *path, int *empty) 18331da177e4SLinus Torvalds { 18342d8f3038SAl Viro struct nameidata nd; 18351fa1e7f6SAndy Whitcroft char *tmp = getname_flags(name, flags, empty); 18361da177e4SLinus Torvalds int err = PTR_ERR(tmp); 18371da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 18382d8f3038SAl Viro 18392d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 18402d8f3038SAl Viro 18412d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 18421da177e4SLinus Torvalds putname(tmp); 18432d8f3038SAl Viro if (!err) 18442d8f3038SAl Viro *path = nd.path; 18451da177e4SLinus Torvalds } 18461da177e4SLinus Torvalds return err; 18471da177e4SLinus Torvalds } 18481da177e4SLinus Torvalds 18491fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 18501fa1e7f6SAndy Whitcroft struct path *path) 18511fa1e7f6SAndy Whitcroft { 18521fa1e7f6SAndy Whitcroft return user_path_at_empty(dfd, name, flags, path, 0); 18531fa1e7f6SAndy Whitcroft } 18541fa1e7f6SAndy Whitcroft 18552ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 18562ad94ae6SAl Viro struct nameidata *nd, char **name) 18572ad94ae6SAl Viro { 18582ad94ae6SAl Viro char *s = getname(path); 18592ad94ae6SAl Viro int error; 18602ad94ae6SAl Viro 18612ad94ae6SAl Viro if (IS_ERR(s)) 18622ad94ae6SAl Viro return PTR_ERR(s); 18632ad94ae6SAl Viro 18642ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 18652ad94ae6SAl Viro if (error) 18662ad94ae6SAl Viro putname(s); 18672ad94ae6SAl Viro else 18682ad94ae6SAl Viro *name = s; 18692ad94ae6SAl Viro 18702ad94ae6SAl Viro return error; 18712ad94ae6SAl Viro } 18722ad94ae6SAl Viro 18731da177e4SLinus Torvalds /* 18741da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 18751da177e4SLinus Torvalds * minimal. 18761da177e4SLinus Torvalds */ 18771da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 18781da177e4SLinus Torvalds { 1879da9592edSDavid Howells uid_t fsuid = current_fsuid(); 1880da9592edSDavid Howells 18811da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 18821da177e4SLinus Torvalds return 0; 1883e795b717SSerge E. Hallyn if (current_user_ns() != inode_userns(inode)) 1884e795b717SSerge E. Hallyn goto other_userns; 1885da9592edSDavid Howells if (inode->i_uid == fsuid) 18861da177e4SLinus Torvalds return 0; 1887da9592edSDavid Howells if (dir->i_uid == fsuid) 18881da177e4SLinus Torvalds return 0; 1889e795b717SSerge E. Hallyn 1890e795b717SSerge E. Hallyn other_userns: 1891e795b717SSerge E. Hallyn return !ns_capable(inode_userns(inode), CAP_FOWNER); 18921da177e4SLinus Torvalds } 18931da177e4SLinus Torvalds 18941da177e4SLinus Torvalds /* 18951da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 18961da177e4SLinus Torvalds * whether the type of victim is right. 18971da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 18981da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 18991da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 19001da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 19011da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 19021da177e4SLinus Torvalds * a. be owner of dir, or 19031da177e4SLinus Torvalds * b. be owner of victim, or 19041da177e4SLinus Torvalds * c. have CAP_FOWNER capability 19051da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 19061da177e4SLinus Torvalds * links pointing to it. 19071da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 19081da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 19091da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 19101da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 19111da177e4SLinus Torvalds * nfs_async_unlink(). 19121da177e4SLinus Torvalds */ 1913858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 19141da177e4SLinus Torvalds { 19151da177e4SLinus Torvalds int error; 19161da177e4SLinus Torvalds 19171da177e4SLinus Torvalds if (!victim->d_inode) 19181da177e4SLinus Torvalds return -ENOENT; 19191da177e4SLinus Torvalds 19201da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 1921cccc6bbaSAl Viro audit_inode_child(victim, dir); 19221da177e4SLinus Torvalds 1923f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 19241da177e4SLinus Torvalds if (error) 19251da177e4SLinus Torvalds return error; 19261da177e4SLinus Torvalds if (IS_APPEND(dir)) 19271da177e4SLinus Torvalds return -EPERM; 19281da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 1929f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 19301da177e4SLinus Torvalds return -EPERM; 19311da177e4SLinus Torvalds if (isdir) { 19321da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 19331da177e4SLinus Torvalds return -ENOTDIR; 19341da177e4SLinus Torvalds if (IS_ROOT(victim)) 19351da177e4SLinus Torvalds return -EBUSY; 19361da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 19371da177e4SLinus Torvalds return -EISDIR; 19381da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 19391da177e4SLinus Torvalds return -ENOENT; 19401da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 19411da177e4SLinus Torvalds return -EBUSY; 19421da177e4SLinus Torvalds return 0; 19431da177e4SLinus Torvalds } 19441da177e4SLinus Torvalds 19451da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 19461da177e4SLinus Torvalds * dir. 19471da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 19481da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 19491da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 19501da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 19511da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 19521da177e4SLinus Torvalds */ 1953a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 19541da177e4SLinus Torvalds { 19551da177e4SLinus Torvalds if (child->d_inode) 19561da177e4SLinus Torvalds return -EEXIST; 19571da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 19581da177e4SLinus Torvalds return -ENOENT; 1959f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 19601da177e4SLinus Torvalds } 19611da177e4SLinus Torvalds 19621da177e4SLinus Torvalds /* 19631da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 19641da177e4SLinus Torvalds */ 19651da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 19661da177e4SLinus Torvalds { 19671da177e4SLinus Torvalds struct dentry *p; 19681da177e4SLinus Torvalds 19691da177e4SLinus Torvalds if (p1 == p2) { 1970f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 19711da177e4SLinus Torvalds return NULL; 19721da177e4SLinus Torvalds } 19731da177e4SLinus Torvalds 1974a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19751da177e4SLinus Torvalds 1976e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 1977e2761a11SOGAWA Hirofumi if (p) { 1978f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1979f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 19801da177e4SLinus Torvalds return p; 19811da177e4SLinus Torvalds } 19821da177e4SLinus Torvalds 1983e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 1984e2761a11SOGAWA Hirofumi if (p) { 1985f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1986f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19871da177e4SLinus Torvalds return p; 19881da177e4SLinus Torvalds } 19891da177e4SLinus Torvalds 1990f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1991f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19921da177e4SLinus Torvalds return NULL; 19931da177e4SLinus Torvalds } 19941da177e4SLinus Torvalds 19951da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 19961da177e4SLinus Torvalds { 19971b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 19981da177e4SLinus Torvalds if (p1 != p2) { 19991b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2000a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 20011da177e4SLinus Torvalds } 20021da177e4SLinus Torvalds } 20031da177e4SLinus Torvalds 20044acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 20051da177e4SLinus Torvalds struct nameidata *nd) 20061da177e4SLinus Torvalds { 2007a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 20081da177e4SLinus Torvalds 20091da177e4SLinus Torvalds if (error) 20101da177e4SLinus Torvalds return error; 20111da177e4SLinus Torvalds 2012acfa4380SAl Viro if (!dir->i_op->create) 20131da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 20141da177e4SLinus Torvalds mode &= S_IALLUGO; 20151da177e4SLinus Torvalds mode |= S_IFREG; 20161da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 20171da177e4SLinus Torvalds if (error) 20181da177e4SLinus Torvalds return error; 20191da177e4SLinus Torvalds error = dir->i_op->create(dir, dentry, mode, nd); 2020a74574aaSStephen Smalley if (!error) 2021f38aa942SAmy Griffis fsnotify_create(dir, dentry); 20221da177e4SLinus Torvalds return error; 20231da177e4SLinus Torvalds } 20241da177e4SLinus Torvalds 202573d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 20261da177e4SLinus Torvalds { 20273fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 20281da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 20291da177e4SLinus Torvalds int error; 20301da177e4SLinus Torvalds 2031bcda7652SAl Viro /* O_PATH? */ 2032bcda7652SAl Viro if (!acc_mode) 2033bcda7652SAl Viro return 0; 2034bcda7652SAl Viro 20351da177e4SLinus Torvalds if (!inode) 20361da177e4SLinus Torvalds return -ENOENT; 20371da177e4SLinus Torvalds 2038c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2039c8fe8f30SChristoph Hellwig case S_IFLNK: 20401da177e4SLinus Torvalds return -ELOOP; 2041c8fe8f30SChristoph Hellwig case S_IFDIR: 2042c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 20431da177e4SLinus Torvalds return -EISDIR; 2044c8fe8f30SChristoph Hellwig break; 2045c8fe8f30SChristoph Hellwig case S_IFBLK: 2046c8fe8f30SChristoph Hellwig case S_IFCHR: 20473fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 20481da177e4SLinus Torvalds return -EACCES; 2049c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2050c8fe8f30SChristoph Hellwig case S_IFIFO: 2051c8fe8f30SChristoph Hellwig case S_IFSOCK: 20521da177e4SLinus Torvalds flag &= ~O_TRUNC; 2053c8fe8f30SChristoph Hellwig break; 20544a3fd211SDave Hansen } 2055b41572e9SDave Hansen 20563fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2057b41572e9SDave Hansen if (error) 2058b41572e9SDave Hansen return error; 20596146f0d5SMimi Zohar 20601da177e4SLinus Torvalds /* 20611da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 20621da177e4SLinus Torvalds */ 20631da177e4SLinus Torvalds if (IS_APPEND(inode)) { 20648737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 20657715b521SAl Viro return -EPERM; 20661da177e4SLinus Torvalds if (flag & O_TRUNC) 20677715b521SAl Viro return -EPERM; 20681da177e4SLinus Torvalds } 20691da177e4SLinus Torvalds 20701da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 20712e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 20727715b521SAl Viro return -EPERM; 20731da177e4SLinus Torvalds 2074f3c7691eSJ. Bruce Fields return 0; 20757715b521SAl Viro } 20767715b521SAl Viro 2077e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 20787715b521SAl Viro { 2079e1181ee6SJeff Layton struct path *path = &filp->f_path; 20807715b521SAl Viro struct inode *inode = path->dentry->d_inode; 20817715b521SAl Viro int error = get_write_access(inode); 20821da177e4SLinus Torvalds if (error) 20837715b521SAl Viro return error; 20841da177e4SLinus Torvalds /* 20851da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 20861da177e4SLinus Torvalds */ 20871da177e4SLinus Torvalds error = locks_verify_locked(inode); 2088be6d3e56SKentaro Takeda if (!error) 2089ea0d3ab2STetsuo Handa error = security_path_truncate(path); 20901da177e4SLinus Torvalds if (!error) { 20917715b521SAl Viro error = do_truncate(path->dentry, 0, 2092d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2093e1181ee6SJeff Layton filp); 20941da177e4SLinus Torvalds } 20951da177e4SLinus Torvalds put_write_access(inode); 2096acd0c935SMimi Zohar return error; 20971da177e4SLinus Torvalds } 20981da177e4SLinus Torvalds 2099d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2100d57999e1SDave Hansen { 21018a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 21028a5e929dSAl Viro flag--; 2103d57999e1SDave Hansen return flag; 2104d57999e1SDave Hansen } 2105d57999e1SDave Hansen 210631e6b01fSNick Piggin /* 2107fe2d35ffSAl Viro * Handle the last step of open() 210831e6b01fSNick Piggin */ 2109fb1cc555SAl Viro static struct file *do_last(struct nameidata *nd, struct path *path, 2110c3e380b0SAl Viro const struct open_flags *op, const char *pathname) 2111fb1cc555SAl Viro { 2112a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 21136c0d46c4SAl Viro struct dentry *dentry; 2114ca344a89SAl Viro int open_flag = op->open_flag; 21156c0d46c4SAl Viro int will_truncate = open_flag & O_TRUNC; 2116ca344a89SAl Viro int want_write = 0; 2117bcda7652SAl Viro int acc_mode = op->acc_mode; 2118fb1cc555SAl Viro struct file *filp; 211916c2cd71SAl Viro int error; 2120fb1cc555SAl Viro 2121c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2122c3e380b0SAl Viro nd->flags |= op->intent; 2123c3e380b0SAl Viro 21241f36f774SAl Viro switch (nd->last_type) { 21251f36f774SAl Viro case LAST_DOTDOT: 2126176306f5SNeil Brown case LAST_DOT: 2127fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2128fe2d35ffSAl Viro if (error) 2129fe2d35ffSAl Viro return ERR_PTR(error); 21301f36f774SAl Viro /* fallthrough */ 21311f36f774SAl Viro case LAST_ROOT: 21329f1fafeeSAl Viro error = complete_walk(nd); 213316c2cd71SAl Viro if (error) 21349f1fafeeSAl Viro return ERR_PTR(error); 2135fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2136ca344a89SAl Viro if (open_flag & O_CREAT) { 213716c2cd71SAl Viro error = -EISDIR; 21381f36f774SAl Viro goto exit; 2139fe2d35ffSAl Viro } 2140fe2d35ffSAl Viro goto ok; 21411f36f774SAl Viro case LAST_BIND: 21429f1fafeeSAl Viro error = complete_walk(nd); 214316c2cd71SAl Viro if (error) 21449f1fafeeSAl Viro return ERR_PTR(error); 21451f36f774SAl Viro audit_inode(pathname, dir); 21461f36f774SAl Viro goto ok; 21471f36f774SAl Viro } 2148a2c36b45SAl Viro 2149ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2150bcda7652SAl Viro int symlink_ok = 0; 2151fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2152fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2153bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 2154bcda7652SAl Viro symlink_ok = 1; 2155fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2156ce57dfc1SAl Viro error = walk_component(nd, path, &nd->last, LAST_NORM, 2157ce57dfc1SAl Viro !symlink_ok); 2158ce57dfc1SAl Viro if (error < 0) 2159fe2d35ffSAl Viro return ERR_PTR(error); 2160ce57dfc1SAl Viro if (error) /* symlink */ 2161fe2d35ffSAl Viro return NULL; 2162fe2d35ffSAl Viro /* sayonara */ 21639f1fafeeSAl Viro error = complete_walk(nd); 21649f1fafeeSAl Viro if (error) 2165fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2166fe2d35ffSAl Viro 2167fe2d35ffSAl Viro error = -ENOTDIR; 2168fe2d35ffSAl Viro if (nd->flags & LOOKUP_DIRECTORY) { 2169ce57dfc1SAl Viro if (!nd->inode->i_op->lookup) 2170fe2d35ffSAl Viro goto exit; 2171fe2d35ffSAl Viro } 2172fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2173fe2d35ffSAl Viro goto ok; 2174fe2d35ffSAl Viro } 2175fe2d35ffSAl Viro 2176fe2d35ffSAl Viro /* create side of things */ 2177a3fbbde7SAl Viro /* 2178a3fbbde7SAl Viro * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been 2179a3fbbde7SAl Viro * cleared when we got to the last component we are about to look up 2180a3fbbde7SAl Viro */ 21819f1fafeeSAl Viro error = complete_walk(nd); 21829f1fafeeSAl Viro if (error) 21839f1fafeeSAl Viro return ERR_PTR(error); 2184fe2d35ffSAl Viro 2185fe2d35ffSAl Viro audit_inode(pathname, dir); 218616c2cd71SAl Viro error = -EISDIR; 21871f36f774SAl Viro /* trailing slashes? */ 218831e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 21891f36f774SAl Viro goto exit; 21901f36f774SAl Viro 2191a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 2192a1e28038SAl Viro 21936c0d46c4SAl Viro dentry = lookup_hash(nd); 21946c0d46c4SAl Viro error = PTR_ERR(dentry); 21956c0d46c4SAl Viro if (IS_ERR(dentry)) { 2196fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2197fb1cc555SAl Viro goto exit; 2198fb1cc555SAl Viro } 2199fb1cc555SAl Viro 22006c0d46c4SAl Viro path->dentry = dentry; 22016c0d46c4SAl Viro path->mnt = nd->path.mnt; 22026c0d46c4SAl Viro 2203fb1cc555SAl Viro /* Negative dentry, just create the file */ 22046c0d46c4SAl Viro if (!dentry->d_inode) { 2205a218d0fdSAl Viro umode_t mode = op->mode; 22066c0d46c4SAl Viro if (!IS_POSIXACL(dir->d_inode)) 22076c0d46c4SAl Viro mode &= ~current_umask(); 2208fb1cc555SAl Viro /* 2209fb1cc555SAl Viro * This write is needed to ensure that a 22106c0d46c4SAl Viro * rw->ro transition does not occur between 2211fb1cc555SAl Viro * the time when the file is created and when 2212fb1cc555SAl Viro * a permanent write count is taken through 2213fb1cc555SAl Viro * the 'struct file' in nameidata_to_filp(). 2214fb1cc555SAl Viro */ 2215fb1cc555SAl Viro error = mnt_want_write(nd->path.mnt); 2216fb1cc555SAl Viro if (error) 2217fb1cc555SAl Viro goto exit_mutex_unlock; 2218ca344a89SAl Viro want_write = 1; 22199b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2220ca344a89SAl Viro open_flag &= ~O_TRUNC; 22216c0d46c4SAl Viro will_truncate = 0; 2222bcda7652SAl Viro acc_mode = MAY_OPEN; 22236c0d46c4SAl Viro error = security_path_mknod(&nd->path, dentry, mode, 0); 22246c0d46c4SAl Viro if (error) 22256c0d46c4SAl Viro goto exit_mutex_unlock; 22266c0d46c4SAl Viro error = vfs_create(dir->d_inode, dentry, mode, nd); 22276c0d46c4SAl Viro if (error) 22286c0d46c4SAl Viro goto exit_mutex_unlock; 22296c0d46c4SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 22306c0d46c4SAl Viro dput(nd->path.dentry); 22316c0d46c4SAl Viro nd->path.dentry = dentry; 2232ca344a89SAl Viro goto common; 2233fb1cc555SAl Viro } 2234fb1cc555SAl Viro 2235fb1cc555SAl Viro /* 2236fb1cc555SAl Viro * It already exists. 2237fb1cc555SAl Viro */ 2238fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2239fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2240fb1cc555SAl Viro 2241fb1cc555SAl Viro error = -EEXIST; 2242ca344a89SAl Viro if (open_flag & O_EXCL) 2243fb1cc555SAl Viro goto exit_dput; 2244fb1cc555SAl Viro 22459875cf80SDavid Howells error = follow_managed(path, nd->flags); 22469875cf80SDavid Howells if (error < 0) 2247fb1cc555SAl Viro goto exit_dput; 2248fb1cc555SAl Viro 2249a3fbbde7SAl Viro if (error) 2250a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 2251a3fbbde7SAl Viro 2252fb1cc555SAl Viro error = -ENOENT; 2253fb1cc555SAl Viro if (!path->dentry->d_inode) 2254fb1cc555SAl Viro goto exit_dput; 22559e67f361SAl Viro 22569e67f361SAl Viro if (path->dentry->d_inode->i_op->follow_link) 2257fb1cc555SAl Viro return NULL; 2258fb1cc555SAl Viro 2259fb1cc555SAl Viro path_to_nameidata(path, nd); 226031e6b01fSNick Piggin nd->inode = path->dentry->d_inode; 2261a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 2262a3fbbde7SAl Viro error = complete_walk(nd); 2263a3fbbde7SAl Viro if (error) 2264a3fbbde7SAl Viro goto exit; 2265fb1cc555SAl Viro error = -EISDIR; 226631e6b01fSNick Piggin if (S_ISDIR(nd->inode->i_mode)) 2267fb1cc555SAl Viro goto exit; 226867ee3ad2SAl Viro ok: 22696c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 22706c0d46c4SAl Viro will_truncate = 0; 22716c0d46c4SAl Viro 22720f9d1a10SAl Viro if (will_truncate) { 22730f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 22740f9d1a10SAl Viro if (error) 22750f9d1a10SAl Viro goto exit; 2276ca344a89SAl Viro want_write = 1; 22770f9d1a10SAl Viro } 2278ca344a89SAl Viro common: 2279bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 2280ca344a89SAl Viro if (error) 22810f9d1a10SAl Viro goto exit; 22820f9d1a10SAl Viro filp = nameidata_to_filp(nd); 22830f9d1a10SAl Viro if (!IS_ERR(filp)) { 22840f9d1a10SAl Viro error = ima_file_check(filp, op->acc_mode); 22850f9d1a10SAl Viro if (error) { 22860f9d1a10SAl Viro fput(filp); 22870f9d1a10SAl Viro filp = ERR_PTR(error); 22880f9d1a10SAl Viro } 22890f9d1a10SAl Viro } 22900f9d1a10SAl Viro if (!IS_ERR(filp)) { 22910f9d1a10SAl Viro if (will_truncate) { 22920f9d1a10SAl Viro error = handle_truncate(filp); 22930f9d1a10SAl Viro if (error) { 22940f9d1a10SAl Viro fput(filp); 22950f9d1a10SAl Viro filp = ERR_PTR(error); 22960f9d1a10SAl Viro } 22970f9d1a10SAl Viro } 22980f9d1a10SAl Viro } 2299ca344a89SAl Viro out: 2300ca344a89SAl Viro if (want_write) 23010f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 23020f9d1a10SAl Viro path_put(&nd->path); 2303fb1cc555SAl Viro return filp; 2304fb1cc555SAl Viro 2305fb1cc555SAl Viro exit_mutex_unlock: 2306fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2307fb1cc555SAl Viro exit_dput: 2308fb1cc555SAl Viro path_put_conditional(path, nd); 2309fb1cc555SAl Viro exit: 2310ca344a89SAl Viro filp = ERR_PTR(error); 2311ca344a89SAl Viro goto out; 2312fb1cc555SAl Viro } 2313fb1cc555SAl Viro 231413aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 231573d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 23161da177e4SLinus Torvalds { 2317fe2d35ffSAl Viro struct file *base = NULL; 23184a3fd211SDave Hansen struct file *filp; 23199850c056SAl Viro struct path path; 232013aab428SAl Viro int error; 232131e6b01fSNick Piggin 232231e6b01fSNick Piggin filp = get_empty_filp(); 232331e6b01fSNick Piggin if (!filp) 232431e6b01fSNick Piggin return ERR_PTR(-ENFILE); 232531e6b01fSNick Piggin 232647c805dcSAl Viro filp->f_flags = op->open_flag; 232773d049a4SAl Viro nd->intent.open.file = filp; 232873d049a4SAl Viro nd->intent.open.flags = open_to_namei_flags(op->open_flag); 232973d049a4SAl Viro nd->intent.open.create_mode = op->mode; 233031e6b01fSNick Piggin 233173d049a4SAl Viro error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); 233231e6b01fSNick Piggin if (unlikely(error)) 233313aab428SAl Viro goto out_filp; 233431e6b01fSNick Piggin 2335fe2d35ffSAl Viro current->total_link_count = 0; 233673d049a4SAl Viro error = link_path_walk(pathname, nd); 233731e6b01fSNick Piggin if (unlikely(error)) 233831e6b01fSNick Piggin goto out_filp; 23391da177e4SLinus Torvalds 234073d049a4SAl Viro filp = do_last(nd, &path, op, pathname); 2341806b681cSAl Viro while (unlikely(!filp)) { /* trailing symlink */ 23427b9337aaSNick Piggin struct path link = path; 2343def4af30SAl Viro void *cookie; 2344574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 234573d049a4SAl Viro path_put_conditional(&path, nd); 234673d049a4SAl Viro path_put(&nd->path); 234740b39136SAl Viro filp = ERR_PTR(-ELOOP); 234840b39136SAl Viro break; 234940b39136SAl Viro } 235073d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 235173d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 2352574197e0SAl Viro error = follow_link(&link, nd, &cookie); 2353c3e380b0SAl Viro if (unlikely(error)) 2354f1afe9efSAl Viro filp = ERR_PTR(error); 2355c3e380b0SAl Viro else 235673d049a4SAl Viro filp = do_last(nd, &path, op, pathname); 2357574197e0SAl Viro put_link(nd, &link, cookie); 2358806b681cSAl Viro } 235910fa8e62SAl Viro out: 236073d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 236173d049a4SAl Viro path_put(&nd->root); 2362fe2d35ffSAl Viro if (base) 2363fe2d35ffSAl Viro fput(base); 236473d049a4SAl Viro release_open_intent(nd); 236510fa8e62SAl Viro return filp; 23661da177e4SLinus Torvalds 236731e6b01fSNick Piggin out_filp: 236810fa8e62SAl Viro filp = ERR_PTR(error); 236910fa8e62SAl Viro goto out; 2370de459215SKirill Korotaev } 23711da177e4SLinus Torvalds 237213aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 237313aab428SAl Viro const struct open_flags *op, int flags) 237413aab428SAl Viro { 237573d049a4SAl Viro struct nameidata nd; 237613aab428SAl Viro struct file *filp; 237713aab428SAl Viro 237873d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 237913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 238073d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 238113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 238273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 238313aab428SAl Viro return filp; 238413aab428SAl Viro } 238513aab428SAl Viro 238673d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 238773d049a4SAl Viro const char *name, const struct open_flags *op, int flags) 238873d049a4SAl Viro { 238973d049a4SAl Viro struct nameidata nd; 239073d049a4SAl Viro struct file *file; 239173d049a4SAl Viro 239273d049a4SAl Viro nd.root.mnt = mnt; 239373d049a4SAl Viro nd.root.dentry = dentry; 239473d049a4SAl Viro 239573d049a4SAl Viro flags |= LOOKUP_ROOT; 239673d049a4SAl Viro 2397bcda7652SAl Viro if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) 239873d049a4SAl Viro return ERR_PTR(-ELOOP); 239973d049a4SAl Viro 240073d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); 240173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 240273d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags); 240373d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 240473d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); 240573d049a4SAl Viro return file; 240673d049a4SAl Viro } 240773d049a4SAl Viro 2408ed75e95dSAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) 24091da177e4SLinus Torvalds { 2410c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 2411ed75e95dSAl Viro struct nameidata nd; 2412ed75e95dSAl Viro int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); 2413ed75e95dSAl Viro if (error) 2414ed75e95dSAl Viro return ERR_PTR(error); 24151da177e4SLinus Torvalds 2416c663e5d8SChristoph Hellwig /* 2417c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2418c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2419c663e5d8SChristoph Hellwig */ 2420ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 2421ed75e95dSAl Viro goto out; 2422ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 2423ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2424ed75e95dSAl Viro nd.intent.open.flags = O_EXCL; 2425c663e5d8SChristoph Hellwig 2426c663e5d8SChristoph Hellwig /* 2427c663e5d8SChristoph Hellwig * Do the final lookup. 2428c663e5d8SChristoph Hellwig */ 2429ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2430ed75e95dSAl Viro dentry = lookup_hash(&nd); 24311da177e4SLinus Torvalds if (IS_ERR(dentry)) 24321da177e4SLinus Torvalds goto fail; 2433c663e5d8SChristoph Hellwig 2434e9baf6e5SAl Viro if (dentry->d_inode) 2435e9baf6e5SAl Viro goto eexist; 2436c663e5d8SChristoph Hellwig /* 2437c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 2438c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 2439c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 2440c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 2441c663e5d8SChristoph Hellwig */ 2442ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 24431da177e4SLinus Torvalds dput(dentry); 24441da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 2445ed75e95dSAl Viro goto fail; 2446e9baf6e5SAl Viro } 2447ed75e95dSAl Viro *path = nd.path; 2448e9baf6e5SAl Viro return dentry; 2449e9baf6e5SAl Viro eexist: 2450e9baf6e5SAl Viro dput(dentry); 2451e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 24521da177e4SLinus Torvalds fail: 2453dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 2454ed75e95dSAl Viro out: 2455dae6ad8fSAl Viro path_put(&nd.path); 2456ed75e95dSAl Viro return dentry; 2457dae6ad8fSAl Viro } 2458dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 2459dae6ad8fSAl Viro 2460dae6ad8fSAl Viro struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) 2461dae6ad8fSAl Viro { 2462dae6ad8fSAl Viro char *tmp = getname(pathname); 2463dae6ad8fSAl Viro struct dentry *res; 2464dae6ad8fSAl Viro if (IS_ERR(tmp)) 2465dae6ad8fSAl Viro return ERR_CAST(tmp); 2466dae6ad8fSAl Viro res = kern_path_create(dfd, tmp, path, is_dir); 2467dae6ad8fSAl Viro putname(tmp); 2468dae6ad8fSAl Viro return res; 2469dae6ad8fSAl Viro } 2470dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 2471dae6ad8fSAl Viro 24721a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 24731da177e4SLinus Torvalds { 2474a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 24751da177e4SLinus Torvalds 24761da177e4SLinus Torvalds if (error) 24771da177e4SLinus Torvalds return error; 24781da177e4SLinus Torvalds 2479e795b717SSerge E. Hallyn if ((S_ISCHR(mode) || S_ISBLK(mode)) && 2480e795b717SSerge E. Hallyn !ns_capable(inode_userns(dir), CAP_MKNOD)) 24811da177e4SLinus Torvalds return -EPERM; 24821da177e4SLinus Torvalds 2483acfa4380SAl Viro if (!dir->i_op->mknod) 24841da177e4SLinus Torvalds return -EPERM; 24851da177e4SLinus Torvalds 248608ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 248708ce5f16SSerge E. Hallyn if (error) 248808ce5f16SSerge E. Hallyn return error; 248908ce5f16SSerge E. Hallyn 24901da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 24911da177e4SLinus Torvalds if (error) 24921da177e4SLinus Torvalds return error; 24931da177e4SLinus Torvalds 24941da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 2495a74574aaSStephen Smalley if (!error) 2496f38aa942SAmy Griffis fsnotify_create(dir, dentry); 24971da177e4SLinus Torvalds return error; 24981da177e4SLinus Torvalds } 24991da177e4SLinus Torvalds 2500f69aac00SAl Viro static int may_mknod(umode_t mode) 2501463c3197SDave Hansen { 2502463c3197SDave Hansen switch (mode & S_IFMT) { 2503463c3197SDave Hansen case S_IFREG: 2504463c3197SDave Hansen case S_IFCHR: 2505463c3197SDave Hansen case S_IFBLK: 2506463c3197SDave Hansen case S_IFIFO: 2507463c3197SDave Hansen case S_IFSOCK: 2508463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 2509463c3197SDave Hansen return 0; 2510463c3197SDave Hansen case S_IFDIR: 2511463c3197SDave Hansen return -EPERM; 2512463c3197SDave Hansen default: 2513463c3197SDave Hansen return -EINVAL; 2514463c3197SDave Hansen } 2515463c3197SDave Hansen } 2516463c3197SDave Hansen 25178208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 25182e4d0924SHeiko Carstens unsigned, dev) 25191da177e4SLinus Torvalds { 25201da177e4SLinus Torvalds struct dentry *dentry; 2521dae6ad8fSAl Viro struct path path; 2522dae6ad8fSAl Viro int error; 25231da177e4SLinus Torvalds 25241da177e4SLinus Torvalds if (S_ISDIR(mode)) 25251da177e4SLinus Torvalds return -EPERM; 25261da177e4SLinus Torvalds 2527dae6ad8fSAl Viro dentry = user_path_create(dfd, filename, &path, 0); 2528dae6ad8fSAl Viro if (IS_ERR(dentry)) 2529dae6ad8fSAl Viro return PTR_ERR(dentry); 25302ad94ae6SAl Viro 2531dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 2532ce3b0f8dSAl Viro mode &= ~current_umask(); 2533463c3197SDave Hansen error = may_mknod(mode); 2534463c3197SDave Hansen if (error) 2535463c3197SDave Hansen goto out_dput; 2536dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 2537463c3197SDave Hansen if (error) 2538463c3197SDave Hansen goto out_dput; 2539dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 2540be6d3e56SKentaro Takeda if (error) 2541be6d3e56SKentaro Takeda goto out_drop_write; 25421da177e4SLinus Torvalds switch (mode & S_IFMT) { 25431da177e4SLinus Torvalds case 0: case S_IFREG: 2544dae6ad8fSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,NULL); 25451da177e4SLinus Torvalds break; 25461da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 2547dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 25481da177e4SLinus Torvalds new_decode_dev(dev)); 25491da177e4SLinus Torvalds break; 25501da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 2551dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 25521da177e4SLinus Torvalds break; 25531da177e4SLinus Torvalds } 2554be6d3e56SKentaro Takeda out_drop_write: 2555dae6ad8fSAl Viro mnt_drop_write(path.mnt); 2556463c3197SDave Hansen out_dput: 25571da177e4SLinus Torvalds dput(dentry); 2558dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 2559dae6ad8fSAl Viro path_put(&path); 25601da177e4SLinus Torvalds 25611da177e4SLinus Torvalds return error; 25621da177e4SLinus Torvalds } 25631da177e4SLinus Torvalds 25648208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 25655590ff0dSUlrich Drepper { 25665590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 25675590ff0dSUlrich Drepper } 25685590ff0dSUlrich Drepper 256918bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 25701da177e4SLinus Torvalds { 2571a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25721da177e4SLinus Torvalds 25731da177e4SLinus Torvalds if (error) 25741da177e4SLinus Torvalds return error; 25751da177e4SLinus Torvalds 2576acfa4380SAl Viro if (!dir->i_op->mkdir) 25771da177e4SLinus Torvalds return -EPERM; 25781da177e4SLinus Torvalds 25791da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 25801da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 25811da177e4SLinus Torvalds if (error) 25821da177e4SLinus Torvalds return error; 25831da177e4SLinus Torvalds 25841da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 2585a74574aaSStephen Smalley if (!error) 2586f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 25871da177e4SLinus Torvalds return error; 25881da177e4SLinus Torvalds } 25891da177e4SLinus Torvalds 2590a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 25911da177e4SLinus Torvalds { 25926902d925SDave Hansen struct dentry *dentry; 2593dae6ad8fSAl Viro struct path path; 2594dae6ad8fSAl Viro int error; 25951da177e4SLinus Torvalds 2596dae6ad8fSAl Viro dentry = user_path_create(dfd, pathname, &path, 1); 25976902d925SDave Hansen if (IS_ERR(dentry)) 2598dae6ad8fSAl Viro return PTR_ERR(dentry); 25996902d925SDave Hansen 2600dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 2601ce3b0f8dSAl Viro mode &= ~current_umask(); 2602dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 2603463c3197SDave Hansen if (error) 2604463c3197SDave Hansen goto out_dput; 2605dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 2606be6d3e56SKentaro Takeda if (error) 2607be6d3e56SKentaro Takeda goto out_drop_write; 2608dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 2609be6d3e56SKentaro Takeda out_drop_write: 2610dae6ad8fSAl Viro mnt_drop_write(path.mnt); 2611463c3197SDave Hansen out_dput: 26121da177e4SLinus Torvalds dput(dentry); 2613dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 2614dae6ad8fSAl Viro path_put(&path); 26151da177e4SLinus Torvalds return error; 26161da177e4SLinus Torvalds } 26171da177e4SLinus Torvalds 2618a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 26195590ff0dSUlrich Drepper { 26205590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 26215590ff0dSUlrich Drepper } 26225590ff0dSUlrich Drepper 26231da177e4SLinus Torvalds /* 2624a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 2625a71905f0SSage Weil * should have a usage count of 2 if we're the only user of this 2626a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 2627a71905f0SSage Weil * then we drop the dentry now. 26281da177e4SLinus Torvalds * 26291da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 26301da177e4SLinus Torvalds * do a 26311da177e4SLinus Torvalds * 26321da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 26331da177e4SLinus Torvalds * return -EBUSY; 26341da177e4SLinus Torvalds * 26351da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 26361da177e4SLinus Torvalds * that is still in use by something else.. 26371da177e4SLinus Torvalds */ 26381da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 26391da177e4SLinus Torvalds { 26401da177e4SLinus Torvalds shrink_dcache_parent(dentry); 26411da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 264264252c75SSage Weil if (dentry->d_count == 1) 26431da177e4SLinus Torvalds __d_drop(dentry); 26441da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 26451da177e4SLinus Torvalds } 26461da177e4SLinus Torvalds 26471da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 26481da177e4SLinus Torvalds { 26491da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 26501da177e4SLinus Torvalds 26511da177e4SLinus Torvalds if (error) 26521da177e4SLinus Torvalds return error; 26531da177e4SLinus Torvalds 2654acfa4380SAl Viro if (!dir->i_op->rmdir) 26551da177e4SLinus Torvalds return -EPERM; 26561da177e4SLinus Torvalds 26571d2ef590SAl Viro dget(dentry); 26581b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 2659912dbc15SSage Weil 26601da177e4SLinus Torvalds error = -EBUSY; 2661912dbc15SSage Weil if (d_mountpoint(dentry)) 2662912dbc15SSage Weil goto out; 2663912dbc15SSage Weil 26641da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 2665912dbc15SSage Weil if (error) 2666912dbc15SSage Weil goto out; 2667912dbc15SSage Weil 26683cebde24SSage Weil shrink_dcache_parent(dentry); 26691da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 2670912dbc15SSage Weil if (error) 2671912dbc15SSage Weil goto out; 2672912dbc15SSage Weil 26731da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 2674d83c49f3SAl Viro dont_mount(dentry); 26751da177e4SLinus Torvalds 2676912dbc15SSage Weil out: 2677912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 26781d2ef590SAl Viro dput(dentry); 2679912dbc15SSage Weil if (!error) 2680912dbc15SSage Weil d_delete(dentry); 26811da177e4SLinus Torvalds return error; 26821da177e4SLinus Torvalds } 26831da177e4SLinus Torvalds 26845590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 26851da177e4SLinus Torvalds { 26861da177e4SLinus Torvalds int error = 0; 26871da177e4SLinus Torvalds char * name; 26881da177e4SLinus Torvalds struct dentry *dentry; 26891da177e4SLinus Torvalds struct nameidata nd; 26901da177e4SLinus Torvalds 26912ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 26921da177e4SLinus Torvalds if (error) 26932ad94ae6SAl Viro return error; 26941da177e4SLinus Torvalds 26951da177e4SLinus Torvalds switch(nd.last_type) { 26961da177e4SLinus Torvalds case LAST_DOTDOT: 26971da177e4SLinus Torvalds error = -ENOTEMPTY; 26981da177e4SLinus Torvalds goto exit1; 26991da177e4SLinus Torvalds case LAST_DOT: 27001da177e4SLinus Torvalds error = -EINVAL; 27011da177e4SLinus Torvalds goto exit1; 27021da177e4SLinus Torvalds case LAST_ROOT: 27031da177e4SLinus Torvalds error = -EBUSY; 27041da177e4SLinus Torvalds goto exit1; 27051da177e4SLinus Torvalds } 27060612d9fbSOGAWA Hirofumi 27070612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 27080612d9fbSOGAWA Hirofumi 27094ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 271049705b77SChristoph Hellwig dentry = lookup_hash(&nd); 27111da177e4SLinus Torvalds error = PTR_ERR(dentry); 27126902d925SDave Hansen if (IS_ERR(dentry)) 27136902d925SDave Hansen goto exit2; 2714e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 2715e6bc45d6STheodore Ts'o error = -ENOENT; 2716e6bc45d6STheodore Ts'o goto exit3; 2717e6bc45d6STheodore Ts'o } 27180622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 27190622753bSDave Hansen if (error) 27200622753bSDave Hansen goto exit3; 2721be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 2722be6d3e56SKentaro Takeda if (error) 2723be6d3e56SKentaro Takeda goto exit4; 27244ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 2725be6d3e56SKentaro Takeda exit4: 27260622753bSDave Hansen mnt_drop_write(nd.path.mnt); 27270622753bSDave Hansen exit3: 27281da177e4SLinus Torvalds dput(dentry); 27296902d925SDave Hansen exit2: 27304ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 27311da177e4SLinus Torvalds exit1: 27321d957f9bSJan Blunck path_put(&nd.path); 27331da177e4SLinus Torvalds putname(name); 27341da177e4SLinus Torvalds return error; 27351da177e4SLinus Torvalds } 27361da177e4SLinus Torvalds 27373cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 27385590ff0dSUlrich Drepper { 27395590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 27405590ff0dSUlrich Drepper } 27415590ff0dSUlrich Drepper 27421da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 27431da177e4SLinus Torvalds { 27441da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 27451da177e4SLinus Torvalds 27461da177e4SLinus Torvalds if (error) 27471da177e4SLinus Torvalds return error; 27481da177e4SLinus Torvalds 2749acfa4380SAl Viro if (!dir->i_op->unlink) 27501da177e4SLinus Torvalds return -EPERM; 27511da177e4SLinus Torvalds 27521b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 27531da177e4SLinus Torvalds if (d_mountpoint(dentry)) 27541da177e4SLinus Torvalds error = -EBUSY; 27551da177e4SLinus Torvalds else { 27561da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 2757bec1052eSAl Viro if (!error) { 27581da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 2759bec1052eSAl Viro if (!error) 2760d83c49f3SAl Viro dont_mount(dentry); 2761bec1052eSAl Viro } 27621da177e4SLinus Torvalds } 27631b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 27661da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 2767ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 27681da177e4SLinus Torvalds d_delete(dentry); 27691da177e4SLinus Torvalds } 27700eeca283SRobert Love 27711da177e4SLinus Torvalds return error; 27721da177e4SLinus Torvalds } 27731da177e4SLinus Torvalds 27741da177e4SLinus Torvalds /* 27751da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 27761b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 27771da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 27781da177e4SLinus Torvalds * while waiting on the I/O. 27791da177e4SLinus Torvalds */ 27805590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 27811da177e4SLinus Torvalds { 27822ad94ae6SAl Viro int error; 27831da177e4SLinus Torvalds char *name; 27841da177e4SLinus Torvalds struct dentry *dentry; 27851da177e4SLinus Torvalds struct nameidata nd; 27861da177e4SLinus Torvalds struct inode *inode = NULL; 27871da177e4SLinus Torvalds 27882ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 27891da177e4SLinus Torvalds if (error) 27902ad94ae6SAl Viro return error; 27912ad94ae6SAl Viro 27921da177e4SLinus Torvalds error = -EISDIR; 27931da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 27941da177e4SLinus Torvalds goto exit1; 27950612d9fbSOGAWA Hirofumi 27960612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 27970612d9fbSOGAWA Hirofumi 27984ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 279949705b77SChristoph Hellwig dentry = lookup_hash(&nd); 28001da177e4SLinus Torvalds error = PTR_ERR(dentry); 28011da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 28021da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 280350338b88STörök Edwin if (nd.last.name[nd.last.len]) 280450338b88STörök Edwin goto slashes; 28051da177e4SLinus Torvalds inode = dentry->d_inode; 280650338b88STörök Edwin if (!inode) 2807e6bc45d6STheodore Ts'o goto slashes; 28087de9c6eeSAl Viro ihold(inode); 28090622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 28100622753bSDave Hansen if (error) 28110622753bSDave Hansen goto exit2; 2812be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 2813be6d3e56SKentaro Takeda if (error) 2814be6d3e56SKentaro Takeda goto exit3; 28154ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 2816be6d3e56SKentaro Takeda exit3: 28170622753bSDave Hansen mnt_drop_write(nd.path.mnt); 28181da177e4SLinus Torvalds exit2: 28191da177e4SLinus Torvalds dput(dentry); 28201da177e4SLinus Torvalds } 28214ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 28221da177e4SLinus Torvalds if (inode) 28231da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 28241da177e4SLinus Torvalds exit1: 28251d957f9bSJan Blunck path_put(&nd.path); 28261da177e4SLinus Torvalds putname(name); 28271da177e4SLinus Torvalds return error; 28281da177e4SLinus Torvalds 28291da177e4SLinus Torvalds slashes: 28301da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 28311da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 28321da177e4SLinus Torvalds goto exit2; 28331da177e4SLinus Torvalds } 28341da177e4SLinus Torvalds 28352e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 28365590ff0dSUlrich Drepper { 28375590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 28385590ff0dSUlrich Drepper return -EINVAL; 28395590ff0dSUlrich Drepper 28405590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 28415590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 28425590ff0dSUlrich Drepper 28435590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 28445590ff0dSUlrich Drepper } 28455590ff0dSUlrich Drepper 28463480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 28475590ff0dSUlrich Drepper { 28485590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 28495590ff0dSUlrich Drepper } 28505590ff0dSUlrich Drepper 2851db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 28521da177e4SLinus Torvalds { 2853a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28541da177e4SLinus Torvalds 28551da177e4SLinus Torvalds if (error) 28561da177e4SLinus Torvalds return error; 28571da177e4SLinus Torvalds 2858acfa4380SAl Viro if (!dir->i_op->symlink) 28591da177e4SLinus Torvalds return -EPERM; 28601da177e4SLinus Torvalds 28611da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 28621da177e4SLinus Torvalds if (error) 28631da177e4SLinus Torvalds return error; 28641da177e4SLinus Torvalds 28651da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 2866a74574aaSStephen Smalley if (!error) 2867f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28681da177e4SLinus Torvalds return error; 28691da177e4SLinus Torvalds } 28701da177e4SLinus Torvalds 28712e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 28722e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 28731da177e4SLinus Torvalds { 28742ad94ae6SAl Viro int error; 28751da177e4SLinus Torvalds char *from; 28766902d925SDave Hansen struct dentry *dentry; 2877dae6ad8fSAl Viro struct path path; 28781da177e4SLinus Torvalds 28791da177e4SLinus Torvalds from = getname(oldname); 28801da177e4SLinus Torvalds if (IS_ERR(from)) 28811da177e4SLinus Torvalds return PTR_ERR(from); 28822ad94ae6SAl Viro 2883dae6ad8fSAl Viro dentry = user_path_create(newdfd, newname, &path, 0); 28841da177e4SLinus Torvalds error = PTR_ERR(dentry); 28856902d925SDave Hansen if (IS_ERR(dentry)) 2886dae6ad8fSAl Viro goto out_putname; 28876902d925SDave Hansen 2888dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 288975c3f29dSDave Hansen if (error) 289075c3f29dSDave Hansen goto out_dput; 2891dae6ad8fSAl Viro error = security_path_symlink(&path, dentry, from); 2892be6d3e56SKentaro Takeda if (error) 2893be6d3e56SKentaro Takeda goto out_drop_write; 2894dae6ad8fSAl Viro error = vfs_symlink(path.dentry->d_inode, dentry, from); 2895be6d3e56SKentaro Takeda out_drop_write: 2896dae6ad8fSAl Viro mnt_drop_write(path.mnt); 289775c3f29dSDave Hansen out_dput: 28981da177e4SLinus Torvalds dput(dentry); 2899dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 2900dae6ad8fSAl Viro path_put(&path); 29016902d925SDave Hansen out_putname: 29021da177e4SLinus Torvalds putname(from); 29031da177e4SLinus Torvalds return error; 29041da177e4SLinus Torvalds } 29051da177e4SLinus Torvalds 29063480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 29075590ff0dSUlrich Drepper { 29085590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 29095590ff0dSUlrich Drepper } 29105590ff0dSUlrich Drepper 29111da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 29121da177e4SLinus Torvalds { 29131da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 29141da177e4SLinus Torvalds int error; 29151da177e4SLinus Torvalds 29161da177e4SLinus Torvalds if (!inode) 29171da177e4SLinus Torvalds return -ENOENT; 29181da177e4SLinus Torvalds 2919a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 29201da177e4SLinus Torvalds if (error) 29211da177e4SLinus Torvalds return error; 29221da177e4SLinus Torvalds 29231da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 29241da177e4SLinus Torvalds return -EXDEV; 29251da177e4SLinus Torvalds 29261da177e4SLinus Torvalds /* 29271da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 29281da177e4SLinus Torvalds */ 29291da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 29301da177e4SLinus Torvalds return -EPERM; 2931acfa4380SAl Viro if (!dir->i_op->link) 29321da177e4SLinus Torvalds return -EPERM; 29337e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 29341da177e4SLinus Torvalds return -EPERM; 29351da177e4SLinus Torvalds 29361da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 29371da177e4SLinus Torvalds if (error) 29381da177e4SLinus Torvalds return error; 29391da177e4SLinus Torvalds 29407e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 2941aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 2942aae8a97dSAneesh Kumar K.V if (inode->i_nlink == 0) 2943aae8a97dSAneesh Kumar K.V error = -ENOENT; 2944aae8a97dSAneesh Kumar K.V else 29451da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 29467e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 2947e31e14ecSStephen Smalley if (!error) 29487e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 29491da177e4SLinus Torvalds return error; 29501da177e4SLinus Torvalds } 29511da177e4SLinus Torvalds 29521da177e4SLinus Torvalds /* 29531da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 29541da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 29551da177e4SLinus Torvalds * newname. --KAB 29561da177e4SLinus Torvalds * 29571da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 29581da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 29591da177e4SLinus Torvalds * and other special files. --ADM 29601da177e4SLinus Torvalds */ 29612e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 29622e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 29631da177e4SLinus Torvalds { 29641da177e4SLinus Torvalds struct dentry *new_dentry; 2965dae6ad8fSAl Viro struct path old_path, new_path; 296611a7b371SAneesh Kumar K.V int how = 0; 29671da177e4SLinus Torvalds int error; 29681da177e4SLinus Torvalds 296911a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 2970c04030e1SUlrich Drepper return -EINVAL; 297111a7b371SAneesh Kumar K.V /* 297211a7b371SAneesh Kumar K.V * To use null names we require CAP_DAC_READ_SEARCH 297311a7b371SAneesh Kumar K.V * This ensures that not everyone will be able to create 297411a7b371SAneesh Kumar K.V * handlink using the passed filedescriptor. 297511a7b371SAneesh Kumar K.V */ 297611a7b371SAneesh Kumar K.V if (flags & AT_EMPTY_PATH) { 297711a7b371SAneesh Kumar K.V if (!capable(CAP_DAC_READ_SEARCH)) 297811a7b371SAneesh Kumar K.V return -ENOENT; 297911a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 298011a7b371SAneesh Kumar K.V } 2981c04030e1SUlrich Drepper 298211a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 298311a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 298411a7b371SAneesh Kumar K.V 298511a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 29861da177e4SLinus Torvalds if (error) 29872ad94ae6SAl Viro return error; 29882ad94ae6SAl Viro 2989dae6ad8fSAl Viro new_dentry = user_path_create(newdfd, newname, &new_path, 0); 29901da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 29916902d925SDave Hansen if (IS_ERR(new_dentry)) 2992dae6ad8fSAl Viro goto out; 2993dae6ad8fSAl Viro 2994dae6ad8fSAl Viro error = -EXDEV; 2995dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 2996dae6ad8fSAl Viro goto out_dput; 2997dae6ad8fSAl Viro error = mnt_want_write(new_path.mnt); 299875c3f29dSDave Hansen if (error) 299975c3f29dSDave Hansen goto out_dput; 3000dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 3001be6d3e56SKentaro Takeda if (error) 3002be6d3e56SKentaro Takeda goto out_drop_write; 3003dae6ad8fSAl Viro error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); 3004be6d3e56SKentaro Takeda out_drop_write: 3005dae6ad8fSAl Viro mnt_drop_write(new_path.mnt); 300675c3f29dSDave Hansen out_dput: 30071da177e4SLinus Torvalds dput(new_dentry); 3008dae6ad8fSAl Viro mutex_unlock(&new_path.dentry->d_inode->i_mutex); 3009dae6ad8fSAl Viro path_put(&new_path); 30101da177e4SLinus Torvalds out: 30112d8f3038SAl Viro path_put(&old_path); 30121da177e4SLinus Torvalds 30131da177e4SLinus Torvalds return error; 30141da177e4SLinus Torvalds } 30151da177e4SLinus Torvalds 30163480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 30175590ff0dSUlrich Drepper { 3018c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 30195590ff0dSUlrich Drepper } 30205590ff0dSUlrich Drepper 30211da177e4SLinus Torvalds /* 30221da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 30231da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 30241da177e4SLinus Torvalds * Problems: 30251da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 30261da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 30271da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3028a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 30291da177e4SLinus Torvalds * story. 30301da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 30311b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 30321da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 30331da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3034a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 30351da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 30361da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 30371da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3038a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 30391da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 30401da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 30411da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 3042e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 30431b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 30441da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3045c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 30461da177e4SLinus Torvalds * locking]. 30471da177e4SLinus Torvalds */ 304875c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 30491da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30501da177e4SLinus Torvalds { 30511da177e4SLinus Torvalds int error = 0; 30529055cba7SSage Weil struct inode *target = new_dentry->d_inode; 30531da177e4SLinus Torvalds 30541da177e4SLinus Torvalds /* 30551da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 30561da177e4SLinus Torvalds * we'll need to flip '..'. 30571da177e4SLinus Torvalds */ 30581da177e4SLinus Torvalds if (new_dir != old_dir) { 3059f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 30601da177e4SLinus Torvalds if (error) 30611da177e4SLinus Torvalds return error; 30621da177e4SLinus Torvalds } 30631da177e4SLinus Torvalds 30641da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30651da177e4SLinus Torvalds if (error) 30661da177e4SLinus Torvalds return error; 30671da177e4SLinus Torvalds 30681d2ef590SAl Viro dget(new_dentry); 3069d83c49f3SAl Viro if (target) 30701b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 30719055cba7SSage Weil 30721da177e4SLinus Torvalds error = -EBUSY; 30739055cba7SSage Weil if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 30749055cba7SSage Weil goto out; 30759055cba7SSage Weil 30763cebde24SSage Weil if (target) 30773cebde24SSage Weil shrink_dcache_parent(new_dentry); 30781da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 30799055cba7SSage Weil if (error) 30809055cba7SSage Weil goto out; 30819055cba7SSage Weil 30821da177e4SLinus Torvalds if (target) { 30831da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3084d83c49f3SAl Viro dont_mount(new_dentry); 3085d83c49f3SAl Viro } 30869055cba7SSage Weil out: 30879055cba7SSage Weil if (target) 30881b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 30891d2ef590SAl Viro dput(new_dentry); 3090e31e14ecSStephen Smalley if (!error) 3091349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 30921da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 30931da177e4SLinus Torvalds return error; 30941da177e4SLinus Torvalds } 30951da177e4SLinus Torvalds 309675c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 30971da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30981da177e4SLinus Torvalds { 309951892bbbSSage Weil struct inode *target = new_dentry->d_inode; 31001da177e4SLinus Torvalds int error; 31011da177e4SLinus Torvalds 31021da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 31031da177e4SLinus Torvalds if (error) 31041da177e4SLinus Torvalds return error; 31051da177e4SLinus Torvalds 31061da177e4SLinus Torvalds dget(new_dentry); 31071da177e4SLinus Torvalds if (target) 31081b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 310951892bbbSSage Weil 31101da177e4SLinus Torvalds error = -EBUSY; 311151892bbbSSage Weil if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 311251892bbbSSage Weil goto out; 311351892bbbSSage Weil 31141da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 311551892bbbSSage Weil if (error) 311651892bbbSSage Weil goto out; 311751892bbbSSage Weil 3118bec1052eSAl Viro if (target) 3119d83c49f3SAl Viro dont_mount(new_dentry); 3120349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 31211da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 312251892bbbSSage Weil out: 31231da177e4SLinus Torvalds if (target) 31241b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 31251da177e4SLinus Torvalds dput(new_dentry); 31261da177e4SLinus Torvalds return error; 31271da177e4SLinus Torvalds } 31281da177e4SLinus Torvalds 31291da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 31301da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 31311da177e4SLinus Torvalds { 31321da177e4SLinus Torvalds int error; 31331da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 313459b0df21SEric Paris const unsigned char *old_name; 31351da177e4SLinus Torvalds 31361da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 31371da177e4SLinus Torvalds return 0; 31381da177e4SLinus Torvalds 31391da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 31401da177e4SLinus Torvalds if (error) 31411da177e4SLinus Torvalds return error; 31421da177e4SLinus Torvalds 31431da177e4SLinus Torvalds if (!new_dentry->d_inode) 3144a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 31451da177e4SLinus Torvalds else 31461da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 31471da177e4SLinus Torvalds if (error) 31481da177e4SLinus Torvalds return error; 31491da177e4SLinus Torvalds 3150acfa4380SAl Viro if (!old_dir->i_op->rename) 31511da177e4SLinus Torvalds return -EPERM; 31521da177e4SLinus Torvalds 31530eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 31540eeca283SRobert Love 31551da177e4SLinus Torvalds if (is_dir) 31561da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 31571da177e4SLinus Torvalds else 31581da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3159123df294SAl Viro if (!error) 3160123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 31615a190ae6SAl Viro new_dentry->d_inode, old_dentry); 31620eeca283SRobert Love fsnotify_oldname_free(old_name); 31630eeca283SRobert Love 31641da177e4SLinus Torvalds return error; 31651da177e4SLinus Torvalds } 31661da177e4SLinus Torvalds 31672e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 31682e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 31691da177e4SLinus Torvalds { 31701da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 31711da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 31721da177e4SLinus Torvalds struct dentry *trap; 31731da177e4SLinus Torvalds struct nameidata oldnd, newnd; 31742ad94ae6SAl Viro char *from; 31752ad94ae6SAl Viro char *to; 31762ad94ae6SAl Viro int error; 31771da177e4SLinus Torvalds 31782ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 31791da177e4SLinus Torvalds if (error) 31801da177e4SLinus Torvalds goto exit; 31811da177e4SLinus Torvalds 31822ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 31831da177e4SLinus Torvalds if (error) 31841da177e4SLinus Torvalds goto exit1; 31851da177e4SLinus Torvalds 31861da177e4SLinus Torvalds error = -EXDEV; 31874ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 31881da177e4SLinus Torvalds goto exit2; 31891da177e4SLinus Torvalds 31904ac91378SJan Blunck old_dir = oldnd.path.dentry; 31911da177e4SLinus Torvalds error = -EBUSY; 31921da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 31931da177e4SLinus Torvalds goto exit2; 31941da177e4SLinus Torvalds 31954ac91378SJan Blunck new_dir = newnd.path.dentry; 31961da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 31971da177e4SLinus Torvalds goto exit2; 31981da177e4SLinus Torvalds 31990612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 32000612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 32014e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 32020612d9fbSOGAWA Hirofumi 32031da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 32041da177e4SLinus Torvalds 320549705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 32061da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 32071da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 32081da177e4SLinus Torvalds goto exit3; 32091da177e4SLinus Torvalds /* source must exist */ 32101da177e4SLinus Torvalds error = -ENOENT; 32111da177e4SLinus Torvalds if (!old_dentry->d_inode) 32121da177e4SLinus Torvalds goto exit4; 32131da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 32141da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 32151da177e4SLinus Torvalds error = -ENOTDIR; 32161da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 32171da177e4SLinus Torvalds goto exit4; 32181da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 32191da177e4SLinus Torvalds goto exit4; 32201da177e4SLinus Torvalds } 32211da177e4SLinus Torvalds /* source should not be ancestor of target */ 32221da177e4SLinus Torvalds error = -EINVAL; 32231da177e4SLinus Torvalds if (old_dentry == trap) 32241da177e4SLinus Torvalds goto exit4; 322549705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 32261da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 32271da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 32281da177e4SLinus Torvalds goto exit4; 32291da177e4SLinus Torvalds /* target should not be an ancestor of source */ 32301da177e4SLinus Torvalds error = -ENOTEMPTY; 32311da177e4SLinus Torvalds if (new_dentry == trap) 32321da177e4SLinus Torvalds goto exit5; 32331da177e4SLinus Torvalds 32349079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 32359079b1ebSDave Hansen if (error) 32369079b1ebSDave Hansen goto exit5; 3237be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3238be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3239be6d3e56SKentaro Takeda if (error) 3240be6d3e56SKentaro Takeda goto exit6; 32411da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 32421da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 3243be6d3e56SKentaro Takeda exit6: 32449079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 32451da177e4SLinus Torvalds exit5: 32461da177e4SLinus Torvalds dput(new_dentry); 32471da177e4SLinus Torvalds exit4: 32481da177e4SLinus Torvalds dput(old_dentry); 32491da177e4SLinus Torvalds exit3: 32501da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 32511da177e4SLinus Torvalds exit2: 32521d957f9bSJan Blunck path_put(&newnd.path); 32532ad94ae6SAl Viro putname(to); 32541da177e4SLinus Torvalds exit1: 32551d957f9bSJan Blunck path_put(&oldnd.path); 32561da177e4SLinus Torvalds putname(from); 32572ad94ae6SAl Viro exit: 32581da177e4SLinus Torvalds return error; 32591da177e4SLinus Torvalds } 32601da177e4SLinus Torvalds 3261a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 32625590ff0dSUlrich Drepper { 32635590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 32645590ff0dSUlrich Drepper } 32655590ff0dSUlrich Drepper 32661da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 32671da177e4SLinus Torvalds { 32681da177e4SLinus Torvalds int len; 32691da177e4SLinus Torvalds 32701da177e4SLinus Torvalds len = PTR_ERR(link); 32711da177e4SLinus Torvalds if (IS_ERR(link)) 32721da177e4SLinus Torvalds goto out; 32731da177e4SLinus Torvalds 32741da177e4SLinus Torvalds len = strlen(link); 32751da177e4SLinus Torvalds if (len > (unsigned) buflen) 32761da177e4SLinus Torvalds len = buflen; 32771da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 32781da177e4SLinus Torvalds len = -EFAULT; 32791da177e4SLinus Torvalds out: 32801da177e4SLinus Torvalds return len; 32811da177e4SLinus Torvalds } 32821da177e4SLinus Torvalds 32831da177e4SLinus Torvalds /* 32841da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 32851da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 32861da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 32871da177e4SLinus Torvalds */ 32881da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 32891da177e4SLinus Torvalds { 32901da177e4SLinus Torvalds struct nameidata nd; 3291cc314eefSLinus Torvalds void *cookie; 3292694a1764SMarcin Slusarz int res; 3293cc314eefSLinus Torvalds 32941da177e4SLinus Torvalds nd.depth = 0; 3295cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3296694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3297694a1764SMarcin Slusarz return PTR_ERR(cookie); 3298694a1764SMarcin Slusarz 3299694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 33001da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3301cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3302694a1764SMarcin Slusarz return res; 33031da177e4SLinus Torvalds } 33041da177e4SLinus Torvalds 33051da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 33061da177e4SLinus Torvalds { 33071da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 33081da177e4SLinus Torvalds } 33091da177e4SLinus Torvalds 33101da177e4SLinus Torvalds /* get the link contents into pagecache */ 33111da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 33121da177e4SLinus Torvalds { 3313ebd09abbSDuane Griffin char *kaddr; 33141da177e4SLinus Torvalds struct page *page; 33151da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3316090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 33171da177e4SLinus Torvalds if (IS_ERR(page)) 33186fe6900eSNick Piggin return (char*)page; 33191da177e4SLinus Torvalds *ppage = page; 3320ebd09abbSDuane Griffin kaddr = kmap(page); 3321ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3322ebd09abbSDuane Griffin return kaddr; 33231da177e4SLinus Torvalds } 33241da177e4SLinus Torvalds 33251da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 33261da177e4SLinus Torvalds { 33271da177e4SLinus Torvalds struct page *page = NULL; 33281da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 33291da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 33301da177e4SLinus Torvalds if (page) { 33311da177e4SLinus Torvalds kunmap(page); 33321da177e4SLinus Torvalds page_cache_release(page); 33331da177e4SLinus Torvalds } 33341da177e4SLinus Torvalds return res; 33351da177e4SLinus Torvalds } 33361da177e4SLinus Torvalds 3337cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 33381da177e4SLinus Torvalds { 3339cc314eefSLinus Torvalds struct page *page = NULL; 33401da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3341cc314eefSLinus Torvalds return page; 33421da177e4SLinus Torvalds } 33431da177e4SLinus Torvalds 3344cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 33451da177e4SLinus Torvalds { 3346cc314eefSLinus Torvalds struct page *page = cookie; 3347cc314eefSLinus Torvalds 3348cc314eefSLinus Torvalds if (page) { 33491da177e4SLinus Torvalds kunmap(page); 33501da177e4SLinus Torvalds page_cache_release(page); 33511da177e4SLinus Torvalds } 33521da177e4SLinus Torvalds } 33531da177e4SLinus Torvalds 335454566b2cSNick Piggin /* 335554566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 335654566b2cSNick Piggin */ 335754566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 33581da177e4SLinus Torvalds { 33591da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 33600adb25d2SKirill Korotaev struct page *page; 3361afddba49SNick Piggin void *fsdata; 3362beb497abSDmitriy Monakhov int err; 33631da177e4SLinus Torvalds char *kaddr; 336454566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 336554566b2cSNick Piggin if (nofs) 336654566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 33671da177e4SLinus Torvalds 33687e53cac4SNeilBrown retry: 3369afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 337054566b2cSNick Piggin flags, &page, &fsdata); 33711da177e4SLinus Torvalds if (err) 3372afddba49SNick Piggin goto fail; 3373afddba49SNick Piggin 33741da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 33751da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 33761da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 3377afddba49SNick Piggin 3378afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3379afddba49SNick Piggin page, fsdata); 33801da177e4SLinus Torvalds if (err < 0) 33811da177e4SLinus Torvalds goto fail; 3382afddba49SNick Piggin if (err < len-1) 3383afddba49SNick Piggin goto retry; 3384afddba49SNick Piggin 33851da177e4SLinus Torvalds mark_inode_dirty(inode); 33861da177e4SLinus Torvalds return 0; 33871da177e4SLinus Torvalds fail: 33881da177e4SLinus Torvalds return err; 33891da177e4SLinus Torvalds } 33901da177e4SLinus Torvalds 33910adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 33920adb25d2SKirill Korotaev { 33930adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 339454566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 33950adb25d2SKirill Korotaev } 33960adb25d2SKirill Korotaev 339792e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 33981da177e4SLinus Torvalds .readlink = generic_readlink, 33991da177e4SLinus Torvalds .follow_link = page_follow_link_light, 34001da177e4SLinus Torvalds .put_link = page_put_link, 34011da177e4SLinus Torvalds }; 34021da177e4SLinus Torvalds 34032d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3404cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 34051da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 34061da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 34071da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 34081da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 34091da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 34101da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 34111da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 34121da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 34131da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 34140adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 34151da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 34161da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3417d1811465SAl Viro EXPORT_SYMBOL(kern_path); 341816f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3419f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 34201da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 34211da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 34221da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 34231da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 34241da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 34251da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 34261da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 34271da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 34281da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 34291da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 34301da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 34311da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 34321da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 34331da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3434