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> 351da177e4SLinus Torvalds #include <asm/uaccess.h> 361da177e4SLinus Torvalds 37e81e3f4dSEric Paris #include "internal.h" 38e81e3f4dSEric Paris 391da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 401da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 411da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 421da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 431da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 441da177e4SLinus Torvalds * 451da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 461da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 471da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 481da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 491da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 501da177e4SLinus Torvalds * the special cases of the former code. 511da177e4SLinus Torvalds * 521da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 531da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 541da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 551da177e4SLinus Torvalds * 561da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 571da177e4SLinus Torvalds * resolution to correspond with current state of the code. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 601da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 611da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 621da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 631da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 641da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 651da177e4SLinus Torvalds */ 661da177e4SLinus Torvalds 671da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 681da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 691da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 701da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 711da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 721da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 731da177e4SLinus Torvalds * the name is a symlink pointing to a non-existant name. 741da177e4SLinus Torvalds * 751da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 761da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 771da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 781da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 791da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 801da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 811da177e4SLinus Torvalds * and in the old Linux semantics. 821da177e4SLinus Torvalds */ 831da177e4SLinus Torvalds 841da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 851da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 861da177e4SLinus Torvalds * 871da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 881da177e4SLinus Torvalds */ 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 911da177e4SLinus Torvalds * inside the path - always follow. 921da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 931da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 941da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 951da177e4SLinus Torvalds * otherwise - don't follow. 961da177e4SLinus Torvalds * (applied in that order). 971da177e4SLinus Torvalds * 981da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 991da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1001da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1011da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1021da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1031da177e4SLinus Torvalds */ 1041da177e4SLinus Torvalds /* 1051da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 106a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1071da177e4SLinus Torvalds * any extra contention... 1081da177e4SLinus Torvalds */ 1091da177e4SLinus Torvalds 1101da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1111da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1121da177e4SLinus Torvalds * kernel data space before using them.. 1131da177e4SLinus Torvalds * 1141da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1151da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1161da177e4SLinus Torvalds */ 117858119e1SArjan van de Ven static int do_getname(const char __user *filename, char *page) 1181da177e4SLinus Torvalds { 1191da177e4SLinus Torvalds int retval; 1201da177e4SLinus Torvalds unsigned long len = PATH_MAX; 1211da177e4SLinus Torvalds 1221da177e4SLinus Torvalds if (!segment_eq(get_fs(), KERNEL_DS)) { 1231da177e4SLinus Torvalds if ((unsigned long) filename >= TASK_SIZE) 1241da177e4SLinus Torvalds return -EFAULT; 1251da177e4SLinus Torvalds if (TASK_SIZE - (unsigned long) filename < PATH_MAX) 1261da177e4SLinus Torvalds len = TASK_SIZE - (unsigned long) filename; 1271da177e4SLinus Torvalds } 1281da177e4SLinus Torvalds 1291da177e4SLinus Torvalds retval = strncpy_from_user(page, filename, len); 1301da177e4SLinus Torvalds if (retval > 0) { 1311da177e4SLinus Torvalds if (retval < len) 1321da177e4SLinus Torvalds return 0; 1331da177e4SLinus Torvalds return -ENAMETOOLONG; 1341da177e4SLinus Torvalds } else if (!retval) 1351da177e4SLinus Torvalds retval = -ENOENT; 1361da177e4SLinus Torvalds return retval; 1371da177e4SLinus Torvalds } 1381da177e4SLinus Torvalds 139f52e0c11SAl Viro static char *getname_flags(const char __user * filename, int flags) 1401da177e4SLinus Torvalds { 1411da177e4SLinus Torvalds char *tmp, *result; 1421da177e4SLinus Torvalds 1431da177e4SLinus Torvalds result = ERR_PTR(-ENOMEM); 1441da177e4SLinus Torvalds tmp = __getname(); 1451da177e4SLinus Torvalds if (tmp) { 1461da177e4SLinus Torvalds int retval = do_getname(filename, tmp); 1471da177e4SLinus Torvalds 1481da177e4SLinus Torvalds result = tmp; 1491da177e4SLinus Torvalds if (retval < 0) { 150f52e0c11SAl Viro if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) { 1511da177e4SLinus Torvalds __putname(tmp); 1521da177e4SLinus Torvalds result = ERR_PTR(retval); 1531da177e4SLinus Torvalds } 1541da177e4SLinus Torvalds } 155f52e0c11SAl Viro } 1561da177e4SLinus Torvalds audit_getname(result); 1571da177e4SLinus Torvalds return result; 1581da177e4SLinus Torvalds } 1591da177e4SLinus Torvalds 160f52e0c11SAl Viro char *getname(const char __user * filename) 161f52e0c11SAl Viro { 162f52e0c11SAl Viro return getname_flags(filename, 0); 163f52e0c11SAl Viro } 164f52e0c11SAl Viro 1651da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 1661da177e4SLinus Torvalds void putname(const char *name) 1671da177e4SLinus Torvalds { 1685ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 1691da177e4SLinus Torvalds audit_putname(name); 1701da177e4SLinus Torvalds else 1711da177e4SLinus Torvalds __putname(name); 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds EXPORT_SYMBOL(putname); 1741da177e4SLinus Torvalds #endif 1751da177e4SLinus Torvalds 1765909ccaaSLinus Torvalds /* 1775909ccaaSLinus Torvalds * This does basic POSIX ACL permission checking 1785909ccaaSLinus Torvalds */ 179b74c79e9SNick Piggin static int acl_permission_check(struct inode *inode, int mask, unsigned int flags, 180b74c79e9SNick Piggin int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) 1815909ccaaSLinus Torvalds { 1825909ccaaSLinus Torvalds umode_t mode = inode->i_mode; 1835909ccaaSLinus Torvalds 1845909ccaaSLinus Torvalds mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 1855909ccaaSLinus Torvalds 1865909ccaaSLinus Torvalds if (current_fsuid() == inode->i_uid) 1875909ccaaSLinus Torvalds mode >>= 6; 1885909ccaaSLinus Torvalds else { 1895909ccaaSLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { 190b74c79e9SNick Piggin int error = check_acl(inode, mask, flags); 1915909ccaaSLinus Torvalds if (error != -EAGAIN) 1925909ccaaSLinus Torvalds return error; 1935909ccaaSLinus Torvalds } 1945909ccaaSLinus Torvalds 1955909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 1965909ccaaSLinus Torvalds mode >>= 3; 1975909ccaaSLinus Torvalds } 1985909ccaaSLinus Torvalds 1995909ccaaSLinus Torvalds /* 2005909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 2015909ccaaSLinus Torvalds */ 2025909ccaaSLinus Torvalds if ((mask & ~mode) == 0) 2035909ccaaSLinus Torvalds return 0; 2045909ccaaSLinus Torvalds return -EACCES; 2055909ccaaSLinus Torvalds } 2061da177e4SLinus Torvalds 2071da177e4SLinus Torvalds /** 2081da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 2091da177e4SLinus Torvalds * @inode: inode to check access rights for 2101da177e4SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 2111da177e4SLinus Torvalds * @check_acl: optional callback to check for Posix ACLs 21239191628SRandy Dunlap * @flags: IPERM_FLAG_ flags. 2131da177e4SLinus Torvalds * 2141da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 2151da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 2161da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 217b74c79e9SNick Piggin * are used for other things. 218b74c79e9SNick Piggin * 219b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 220b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 221b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 2221da177e4SLinus Torvalds */ 223b74c79e9SNick Piggin int generic_permission(struct inode *inode, int mask, unsigned int flags, 224b74c79e9SNick Piggin int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) 2251da177e4SLinus Torvalds { 2265909ccaaSLinus Torvalds int ret; 2271da177e4SLinus Torvalds 2281da177e4SLinus Torvalds /* 2295909ccaaSLinus Torvalds * Do the basic POSIX ACL permission checks. 2301da177e4SLinus Torvalds */ 231b74c79e9SNick Piggin ret = acl_permission_check(inode, mask, flags, check_acl); 2325909ccaaSLinus Torvalds if (ret != -EACCES) 2335909ccaaSLinus Torvalds return ret; 2341da177e4SLinus Torvalds 2351da177e4SLinus Torvalds /* 2361da177e4SLinus Torvalds * Read/write DACs are always overridable. 2371da177e4SLinus Torvalds * Executable DACs are overridable if at least one exec bit is set. 2381da177e4SLinus Torvalds */ 239f696a365SMiklos Szeredi if (!(mask & MAY_EXEC) || execute_ok(inode)) 2401da177e4SLinus Torvalds if (capable(CAP_DAC_OVERRIDE)) 2411da177e4SLinus Torvalds return 0; 2421da177e4SLinus Torvalds 2431da177e4SLinus Torvalds /* 2441da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 2451da177e4SLinus Torvalds */ 2467ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 2471da177e4SLinus Torvalds if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) 2481da177e4SLinus Torvalds if (capable(CAP_DAC_READ_SEARCH)) 2491da177e4SLinus Torvalds return 0; 2501da177e4SLinus Torvalds 2511da177e4SLinus Torvalds return -EACCES; 2521da177e4SLinus Torvalds } 2531da177e4SLinus Torvalds 254cb23beb5SChristoph Hellwig /** 255cb23beb5SChristoph Hellwig * inode_permission - check for access rights to a given inode 256cb23beb5SChristoph Hellwig * @inode: inode to check permission on 257cb23beb5SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 258cb23beb5SChristoph Hellwig * 259cb23beb5SChristoph Hellwig * Used to check for read/write/execute permissions on an inode. 260cb23beb5SChristoph Hellwig * We use "fsuid" for this, letting us set arbitrary permissions 261cb23beb5SChristoph Hellwig * for filesystem access without changing the "normal" uids which 262cb23beb5SChristoph Hellwig * are used for other things. 263cb23beb5SChristoph Hellwig */ 264f419a2e3SAl Viro int inode_permission(struct inode *inode, int mask) 2651da177e4SLinus Torvalds { 266e6305c43SAl Viro int retval; 2671da177e4SLinus Torvalds 2681da177e4SLinus Torvalds if (mask & MAY_WRITE) { 26922590e41SMiklos Szeredi umode_t mode = inode->i_mode; 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds /* 2721da177e4SLinus Torvalds * Nobody gets write access to a read-only fs. 2731da177e4SLinus Torvalds */ 2741da177e4SLinus Torvalds if (IS_RDONLY(inode) && 2751da177e4SLinus Torvalds (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 2761da177e4SLinus Torvalds return -EROFS; 2771da177e4SLinus Torvalds 2781da177e4SLinus Torvalds /* 2791da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 2801da177e4SLinus Torvalds */ 2811da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 2821da177e4SLinus Torvalds return -EACCES; 2831da177e4SLinus Torvalds } 2841da177e4SLinus Torvalds 285acfa4380SAl Viro if (inode->i_op->permission) 286b74c79e9SNick Piggin retval = inode->i_op->permission(inode, mask, 0); 287f696a365SMiklos Szeredi else 288b74c79e9SNick Piggin retval = generic_permission(inode, mask, 0, 289b74c79e9SNick Piggin inode->i_op->check_acl); 290f696a365SMiklos Szeredi 2911da177e4SLinus Torvalds if (retval) 2921da177e4SLinus Torvalds return retval; 2931da177e4SLinus Torvalds 29408ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 29508ce5f16SSerge E. Hallyn if (retval) 29608ce5f16SSerge E. Hallyn return retval; 29708ce5f16SSerge E. Hallyn 298d09ca739SEric Paris return security_inode_permission(inode, mask); 2991da177e4SLinus Torvalds } 3001da177e4SLinus Torvalds 301e4543eddSChristoph Hellwig /** 3028c744fb8SChristoph Hellwig * file_permission - check for additional access rights to a given file 3038c744fb8SChristoph Hellwig * @file: file to check access rights for 3048c744fb8SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3058c744fb8SChristoph Hellwig * 3068c744fb8SChristoph Hellwig * Used to check for read/write/execute permissions on an already opened 3078c744fb8SChristoph Hellwig * file. 3088c744fb8SChristoph Hellwig * 3098c744fb8SChristoph Hellwig * Note: 3108c744fb8SChristoph Hellwig * Do not use this function in new code. All access checks should 311cb23beb5SChristoph Hellwig * be done using inode_permission(). 3128c744fb8SChristoph Hellwig */ 3138c744fb8SChristoph Hellwig int file_permission(struct file *file, int mask) 3148c744fb8SChristoph Hellwig { 315f419a2e3SAl Viro return inode_permission(file->f_path.dentry->d_inode, mask); 3168c744fb8SChristoph Hellwig } 3178c744fb8SChristoph Hellwig 3181da177e4SLinus Torvalds /* 3191da177e4SLinus Torvalds * get_write_access() gets write permission for a file. 3201da177e4SLinus Torvalds * put_write_access() releases this write permission. 3211da177e4SLinus Torvalds * This is used for regular files. 3221da177e4SLinus Torvalds * We cannot support write (and maybe mmap read-write shared) accesses and 3231da177e4SLinus Torvalds * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode 3241da177e4SLinus Torvalds * can have the following values: 3251da177e4SLinus Torvalds * 0: no writers, no VM_DENYWRITE mappings 3261da177e4SLinus Torvalds * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist 3271da177e4SLinus Torvalds * > 0: (i_writecount) users are writing to the file. 3281da177e4SLinus Torvalds * 3291da177e4SLinus Torvalds * Normally we operate on that counter with atomic_{inc,dec} and it's safe 3301da177e4SLinus Torvalds * except for the cases where we don't hold i_writecount yet. Then we need to 3311da177e4SLinus Torvalds * use {get,deny}_write_access() - these functions check the sign and refuse 3321da177e4SLinus Torvalds * to do the change if sign is wrong. Exclusion between them is provided by 3331da177e4SLinus Torvalds * the inode->i_lock spinlock. 3341da177e4SLinus Torvalds */ 3351da177e4SLinus Torvalds 3361da177e4SLinus Torvalds int get_write_access(struct inode * inode) 3371da177e4SLinus Torvalds { 3381da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3391da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) < 0) { 3401da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3411da177e4SLinus Torvalds return -ETXTBSY; 3421da177e4SLinus Torvalds } 3431da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 3441da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3451da177e4SLinus Torvalds 3461da177e4SLinus Torvalds return 0; 3471da177e4SLinus Torvalds } 3481da177e4SLinus Torvalds 3491da177e4SLinus Torvalds int deny_write_access(struct file * file) 3501da177e4SLinus Torvalds { 3510f7fc9e4SJosef "Jeff" Sipek struct inode *inode = file->f_path.dentry->d_inode; 3521da177e4SLinus Torvalds 3531da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3541da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) > 0) { 3551da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3561da177e4SLinus Torvalds return -ETXTBSY; 3571da177e4SLinus Torvalds } 3581da177e4SLinus Torvalds atomic_dec(&inode->i_writecount); 3591da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3601da177e4SLinus Torvalds 3611da177e4SLinus Torvalds return 0; 3621da177e4SLinus Torvalds } 3631da177e4SLinus Torvalds 3641d957f9bSJan Blunck /** 3655dd784d0SJan Blunck * path_get - get a reference to a path 3665dd784d0SJan Blunck * @path: path to get the reference to 3675dd784d0SJan Blunck * 3685dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3695dd784d0SJan Blunck */ 3705dd784d0SJan Blunck void path_get(struct path *path) 3715dd784d0SJan Blunck { 3725dd784d0SJan Blunck mntget(path->mnt); 3735dd784d0SJan Blunck dget(path->dentry); 3745dd784d0SJan Blunck } 3755dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 3765dd784d0SJan Blunck 3775dd784d0SJan Blunck /** 3781d957f9bSJan Blunck * path_put - put a reference to a path 3791d957f9bSJan Blunck * @path: path to put the reference to 3801d957f9bSJan Blunck * 3811d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 3821d957f9bSJan Blunck */ 3831d957f9bSJan Blunck void path_put(struct path *path) 3841da177e4SLinus Torvalds { 3851d957f9bSJan Blunck dput(path->dentry); 3861d957f9bSJan Blunck mntput(path->mnt); 3871da177e4SLinus Torvalds } 3881d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 3891da177e4SLinus Torvalds 390834f2a4aSTrond Myklebust /** 39131e6b01fSNick Piggin * nameidata_drop_rcu - drop this nameidata out of rcu-walk 39231e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 39339191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 39431e6b01fSNick Piggin * 39531e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 39631e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt 39731e6b01fSNick Piggin * to drop out of rcu-walk mode and take normal reference counts on dentries 39831e6b01fSNick Piggin * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take 39931e6b01fSNick Piggin * refcounts at the last known good point before rcu-walk got stuck, so 40031e6b01fSNick Piggin * ref-walk may continue from there. If this is not successful (eg. a seqcount 40131e6b01fSNick Piggin * has changed), then failure is returned and path walk restarts from the 40231e6b01fSNick Piggin * beginning in ref-walk mode. 40331e6b01fSNick Piggin * 40431e6b01fSNick Piggin * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into 40531e6b01fSNick Piggin * ref-walk. Must be called from rcu-walk context. 40631e6b01fSNick Piggin */ 40731e6b01fSNick Piggin static int nameidata_drop_rcu(struct nameidata *nd) 40831e6b01fSNick Piggin { 40931e6b01fSNick Piggin struct fs_struct *fs = current->fs; 41031e6b01fSNick Piggin struct dentry *dentry = nd->path.dentry; 4115b6ca027SAl Viro int want_root = 0; 41231e6b01fSNick Piggin 41331e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4145b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4155b6ca027SAl Viro want_root = 1; 41631e6b01fSNick Piggin spin_lock(&fs->lock); 41731e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 41831e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 41931e6b01fSNick Piggin goto err_root; 42031e6b01fSNick Piggin } 42131e6b01fSNick Piggin spin_lock(&dentry->d_lock); 42231e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 42331e6b01fSNick Piggin goto err; 42431e6b01fSNick Piggin BUG_ON(nd->inode != dentry->d_inode); 42531e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 4265b6ca027SAl Viro if (want_root) { 42731e6b01fSNick Piggin path_get(&nd->root); 42831e6b01fSNick Piggin spin_unlock(&fs->lock); 42931e6b01fSNick Piggin } 43031e6b01fSNick Piggin mntget(nd->path.mnt); 43131e6b01fSNick Piggin 43231e6b01fSNick Piggin rcu_read_unlock(); 43331e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 43431e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 43531e6b01fSNick Piggin return 0; 43631e6b01fSNick Piggin err: 43731e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 43831e6b01fSNick Piggin err_root: 4395b6ca027SAl Viro if (want_root) 44031e6b01fSNick Piggin spin_unlock(&fs->lock); 44131e6b01fSNick Piggin return -ECHILD; 44231e6b01fSNick Piggin } 44331e6b01fSNick Piggin 44431e6b01fSNick Piggin /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ 44531e6b01fSNick Piggin static inline int nameidata_drop_rcu_maybe(struct nameidata *nd) 44631e6b01fSNick Piggin { 44731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) 44831e6b01fSNick Piggin return nameidata_drop_rcu(nd); 44931e6b01fSNick Piggin return 0; 45031e6b01fSNick Piggin } 45131e6b01fSNick Piggin 45231e6b01fSNick Piggin /** 45331e6b01fSNick Piggin * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk 45431e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 45531e6b01fSNick Piggin * @dentry: dentry to drop 45639191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 45731e6b01fSNick Piggin * 45831e6b01fSNick Piggin * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root, 45931e6b01fSNick Piggin * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on 46031e6b01fSNick Piggin * @nd. Must be called from rcu-walk context. 46131e6b01fSNick Piggin */ 46231e6b01fSNick Piggin static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry) 46331e6b01fSNick Piggin { 46431e6b01fSNick Piggin struct fs_struct *fs = current->fs; 46531e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 4665b6ca027SAl Viro int want_root = 0; 46731e6b01fSNick Piggin 46831e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4695b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4705b6ca027SAl Viro want_root = 1; 47131e6b01fSNick Piggin spin_lock(&fs->lock); 47231e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 47331e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 47431e6b01fSNick Piggin goto err_root; 47531e6b01fSNick Piggin } 47631e6b01fSNick Piggin spin_lock(&parent->d_lock); 47731e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 47831e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 47931e6b01fSNick Piggin goto err; 48031e6b01fSNick Piggin /* 48131e6b01fSNick Piggin * If the sequence check on the child dentry passed, then the child has 48231e6b01fSNick Piggin * not been removed from its parent. This means the parent dentry must 48331e6b01fSNick Piggin * be valid and able to take a reference at this point. 48431e6b01fSNick Piggin */ 48531e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 48631e6b01fSNick Piggin BUG_ON(!parent->d_count); 48731e6b01fSNick Piggin parent->d_count++; 48831e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 48931e6b01fSNick Piggin spin_unlock(&parent->d_lock); 4905b6ca027SAl Viro if (want_root) { 49131e6b01fSNick Piggin path_get(&nd->root); 49231e6b01fSNick Piggin spin_unlock(&fs->lock); 49331e6b01fSNick Piggin } 49431e6b01fSNick Piggin mntget(nd->path.mnt); 49531e6b01fSNick Piggin 49631e6b01fSNick Piggin rcu_read_unlock(); 49731e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 49831e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 49931e6b01fSNick Piggin return 0; 50031e6b01fSNick Piggin err: 50131e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 50231e6b01fSNick Piggin spin_unlock(&parent->d_lock); 50331e6b01fSNick Piggin err_root: 5045b6ca027SAl Viro if (want_root) 50531e6b01fSNick Piggin spin_unlock(&fs->lock); 50631e6b01fSNick Piggin return -ECHILD; 50731e6b01fSNick Piggin } 50831e6b01fSNick Piggin 50931e6b01fSNick Piggin /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ 51031e6b01fSNick Piggin static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry) 51131e6b01fSNick Piggin { 512a7472babSAl Viro if (nd->flags & LOOKUP_RCU) { 513a7472babSAl Viro if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) { 514a7472babSAl Viro nd->flags &= ~LOOKUP_RCU; 5155b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 516a7472babSAl Viro nd->root.mnt = NULL; 517a7472babSAl Viro rcu_read_unlock(); 518a7472babSAl Viro br_read_unlock(vfsmount_lock); 519a7472babSAl Viro return -ECHILD; 520a7472babSAl Viro } 521a7472babSAl Viro } 52231e6b01fSNick Piggin return 0; 52331e6b01fSNick Piggin } 52431e6b01fSNick Piggin 52531e6b01fSNick Piggin /** 52631e6b01fSNick Piggin * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk 52731e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 52839191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 52931e6b01fSNick Piggin * 53031e6b01fSNick Piggin * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk. 53131e6b01fSNick Piggin * nd->path should be the final element of the lookup, so nd->root is discarded. 53231e6b01fSNick Piggin * Must be called from rcu-walk context. 53331e6b01fSNick Piggin */ 53431e6b01fSNick Piggin static int nameidata_drop_rcu_last(struct nameidata *nd) 53531e6b01fSNick Piggin { 53631e6b01fSNick Piggin struct dentry *dentry = nd->path.dentry; 53731e6b01fSNick Piggin 53831e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 53931e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 5405b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 54131e6b01fSNick Piggin nd->root.mnt = NULL; 54231e6b01fSNick Piggin spin_lock(&dentry->d_lock); 54331e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 54431e6b01fSNick Piggin goto err_unlock; 54531e6b01fSNick Piggin BUG_ON(nd->inode != dentry->d_inode); 54631e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 54731e6b01fSNick Piggin 54831e6b01fSNick Piggin mntget(nd->path.mnt); 54931e6b01fSNick Piggin 55031e6b01fSNick Piggin rcu_read_unlock(); 55131e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 55231e6b01fSNick Piggin 55331e6b01fSNick Piggin return 0; 55431e6b01fSNick Piggin 55531e6b01fSNick Piggin err_unlock: 55631e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 55731e6b01fSNick Piggin rcu_read_unlock(); 55831e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 55931e6b01fSNick Piggin return -ECHILD; 56031e6b01fSNick Piggin } 56131e6b01fSNick Piggin 56231e6b01fSNick Piggin /** 563834f2a4aSTrond Myklebust * release_open_intent - free up open intent resources 564834f2a4aSTrond Myklebust * @nd: pointer to nameidata 565834f2a4aSTrond Myklebust */ 566834f2a4aSTrond Myklebust void release_open_intent(struct nameidata *nd) 567834f2a4aSTrond Myklebust { 5682dab5974SLinus Torvalds struct file *file = nd->intent.open.file; 5692dab5974SLinus Torvalds 5702dab5974SLinus Torvalds if (file && !IS_ERR(file)) { 5712dab5974SLinus Torvalds if (file->f_path.dentry == NULL) 5722dab5974SLinus Torvalds put_filp(file); 573834f2a4aSTrond Myklebust else 5742dab5974SLinus Torvalds fput(file); 5752dab5974SLinus Torvalds } 576834f2a4aSTrond Myklebust } 577834f2a4aSTrond Myklebust 578f60aef7eSAl Viro static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) 57934286d66SNick Piggin { 580f60aef7eSAl Viro return dentry->d_op->d_revalidate(dentry, nd); 58134286d66SNick Piggin } 58234286d66SNick Piggin 583f5e1c1c1SAl Viro static struct dentry * 584bcdc5e01SIan Kent do_revalidate(struct dentry *dentry, struct nameidata *nd) 585bcdc5e01SIan Kent { 586f5e1c1c1SAl Viro int status = d_revalidate(dentry, nd); 587bcdc5e01SIan Kent if (unlikely(status <= 0)) { 588bcdc5e01SIan Kent /* 589bcdc5e01SIan Kent * The dentry failed validation. 590bcdc5e01SIan Kent * If d_revalidate returned 0 attempt to invalidate 591bcdc5e01SIan Kent * the dentry otherwise d_revalidate is asking us 592bcdc5e01SIan Kent * to return a fail status. 593bcdc5e01SIan Kent */ 59434286d66SNick Piggin if (status < 0) { 59534286d66SNick Piggin dput(dentry); 59634286d66SNick Piggin dentry = ERR_PTR(status); 597f5e1c1c1SAl Viro } else if (!d_invalidate(dentry)) { 598bcdc5e01SIan Kent dput(dentry); 599bcdc5e01SIan Kent dentry = NULL; 600bcdc5e01SIan Kent } 601bcdc5e01SIan Kent } 602f5e1c1c1SAl Viro return dentry; 603f5e1c1c1SAl Viro } 604f5e1c1c1SAl Viro 6051da177e4SLinus Torvalds /* 60616c2cd71SAl Viro * handle_reval_path - force revalidation of a dentry 60739159de2SJeff Layton * 60839159de2SJeff Layton * In some situations the path walking code will trust dentries without 60939159de2SJeff Layton * revalidating them. This causes problems for filesystems that depend on 61039159de2SJeff Layton * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set 61139159de2SJeff Layton * (which indicates that it's possible for the dentry to go stale), force 61239159de2SJeff Layton * a d_revalidate call before proceeding. 61339159de2SJeff Layton * 61439159de2SJeff Layton * Returns 0 if the revalidation was successful. If the revalidation fails, 61539159de2SJeff Layton * either return the error returned by d_revalidate or -ESTALE if the 61639159de2SJeff Layton * revalidation it just returned 0. If d_revalidate returns 0, we attempt to 61739159de2SJeff Layton * invalidate the dentry. It's up to the caller to handle putting references 61839159de2SJeff Layton * to the path if necessary. 61939159de2SJeff Layton */ 62016c2cd71SAl Viro static inline int handle_reval_path(struct nameidata *nd) 62139159de2SJeff Layton { 62216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 62339159de2SJeff Layton int status; 62439159de2SJeff Layton 62516c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 62639159de2SJeff Layton return 0; 62739159de2SJeff Layton 62816c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 62916c2cd71SAl Viro return 0; 63016c2cd71SAl Viro 63116c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 63216c2cd71SAl Viro return 0; 63316c2cd71SAl Viro 63416c2cd71SAl Viro /* Note: we do not d_invalidate() */ 63534286d66SNick Piggin status = d_revalidate(dentry, nd); 63639159de2SJeff Layton if (status > 0) 63739159de2SJeff Layton return 0; 63839159de2SJeff Layton 63916c2cd71SAl Viro if (!status) 64039159de2SJeff Layton status = -ESTALE; 64116c2cd71SAl Viro 64239159de2SJeff Layton return status; 64339159de2SJeff Layton } 64439159de2SJeff Layton 64539159de2SJeff Layton /* 646b75b5086SAl Viro * Short-cut version of permission(), for calling on directories 647b75b5086SAl Viro * during pathname resolution. Combines parts of permission() 648b75b5086SAl Viro * and generic_permission(), and tests ONLY for MAY_EXEC permission. 6491da177e4SLinus Torvalds * 6501da177e4SLinus Torvalds * If appropriate, check DAC only. If not appropriate, or 651b75b5086SAl Viro * short-cut DAC fails, then call ->permission() to do more 6521da177e4SLinus Torvalds * complete permission check. 6531da177e4SLinus Torvalds */ 654b74c79e9SNick Piggin static inline int exec_permission(struct inode *inode, unsigned int flags) 6551da177e4SLinus Torvalds { 6565909ccaaSLinus Torvalds int ret; 6571da177e4SLinus Torvalds 658cb9179eaSLinus Torvalds if (inode->i_op->permission) { 659b74c79e9SNick Piggin ret = inode->i_op->permission(inode, MAY_EXEC, flags); 660b74c79e9SNick Piggin } else { 661b74c79e9SNick Piggin ret = acl_permission_check(inode, MAY_EXEC, flags, 662b74c79e9SNick Piggin inode->i_op->check_acl); 663cb9179eaSLinus Torvalds } 664b74c79e9SNick Piggin if (likely(!ret)) 6651da177e4SLinus Torvalds goto ok; 666b74c79e9SNick Piggin if (ret == -ECHILD) 66731e6b01fSNick Piggin return ret; 6681da177e4SLinus Torvalds 669f1ac9f6bSLinus Torvalds if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) 6701da177e4SLinus Torvalds goto ok; 6711da177e4SLinus Torvalds 6725909ccaaSLinus Torvalds return ret; 6731da177e4SLinus Torvalds ok: 674b74c79e9SNick Piggin return security_inode_exec_permission(inode, flags); 6751da177e4SLinus Torvalds } 6761da177e4SLinus Torvalds 6772a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6782a737871SAl Viro { 679f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 680f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6812a737871SAl Viro } 6822a737871SAl Viro 6836de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6846de88d72SAl Viro 68531e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 68631e6b01fSNick Piggin { 68731e6b01fSNick Piggin if (!nd->root.mnt) { 68831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 689c28cc364SNick Piggin unsigned seq; 690c28cc364SNick Piggin 691c28cc364SNick Piggin do { 692c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 69331e6b01fSNick Piggin nd->root = fs->root; 694c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 69531e6b01fSNick Piggin } 69631e6b01fSNick Piggin } 69731e6b01fSNick Piggin 698f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 6991da177e4SLinus Torvalds { 70031e6b01fSNick Piggin int ret; 70131e6b01fSNick Piggin 7021da177e4SLinus Torvalds if (IS_ERR(link)) 7031da177e4SLinus Torvalds goto fail; 7041da177e4SLinus Torvalds 7051da177e4SLinus Torvalds if (*link == '/') { 7062a737871SAl Viro set_root(nd); 7071d957f9bSJan Blunck path_put(&nd->path); 7082a737871SAl Viro nd->path = nd->root; 7092a737871SAl Viro path_get(&nd->root); 71016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 7111da177e4SLinus Torvalds } 71231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 713b4091d5fSChristoph Hellwig 71431e6b01fSNick Piggin ret = link_path_walk(link, nd); 71531e6b01fSNick Piggin return ret; 7161da177e4SLinus Torvalds fail: 7171d957f9bSJan Blunck path_put(&nd->path); 7181da177e4SLinus Torvalds return PTR_ERR(link); 7191da177e4SLinus Torvalds } 7201da177e4SLinus Torvalds 7211d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 722051d3812SIan Kent { 723051d3812SIan Kent dput(path->dentry); 7244ac91378SJan Blunck if (path->mnt != nd->path.mnt) 725051d3812SIan Kent mntput(path->mnt); 726051d3812SIan Kent } 727051d3812SIan Kent 7287b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 7297b9337aaSNick Piggin struct nameidata *nd) 730051d3812SIan Kent { 73131e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 7324ac91378SJan Blunck dput(nd->path.dentry); 73331e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 7344ac91378SJan Blunck mntput(nd->path.mnt); 7359a229683SHuang Shijie } 73631e6b01fSNick Piggin nd->path.mnt = path->mnt; 7374ac91378SJan Blunck nd->path.dentry = path->dentry; 738051d3812SIan Kent } 739051d3812SIan Kent 740def4af30SAl Viro static __always_inline int 7417b9337aaSNick Piggin __do_follow_link(const struct path *link, struct nameidata *nd, void **p) 7421da177e4SLinus Torvalds { 7431da177e4SLinus Torvalds int error; 7447b9337aaSNick Piggin struct dentry *dentry = link->dentry; 7451da177e4SLinus Torvalds 746844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 747844a3917SAl Viro 7487b9337aaSNick Piggin touch_atime(link->mnt, dentry); 7491da177e4SLinus Torvalds nd_set_link(nd, NULL); 750cd4e91d3SAl Viro 75187556ef1SDavid Howells if (link->mnt == nd->path.mnt) 7527b9337aaSNick Piggin mntget(link->mnt); 75331e6b01fSNick Piggin 75436f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 75536f3b4f6SAl Viro if (error) { 75636f3b4f6SAl Viro *p = ERR_PTR(error); /* no ->put_link(), please */ 75736f3b4f6SAl Viro path_put(&nd->path); 75836f3b4f6SAl Viro return error; 75936f3b4f6SAl Viro } 76036f3b4f6SAl Viro 76186acdca1SAl Viro nd->last_type = LAST_BIND; 762def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 763def4af30SAl Viro error = PTR_ERR(*p); 764def4af30SAl Viro if (!IS_ERR(*p)) { 7651da177e4SLinus Torvalds char *s = nd_get_link(nd); 766cc314eefSLinus Torvalds error = 0; 7671da177e4SLinus Torvalds if (s) 7681da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 76916c2cd71SAl Viro else if (nd->last_type == LAST_BIND) 77016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds return error; 7731da177e4SLinus Torvalds } 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds /* 7761da177e4SLinus Torvalds * This limits recursive symlink follows to 8, while 7771da177e4SLinus Torvalds * limiting consecutive symlinks to 40. 7781da177e4SLinus Torvalds * 7791da177e4SLinus Torvalds * Without that kind of total limit, nasty chains of consecutive 7801da177e4SLinus Torvalds * symlinks can cause almost arbitrarily long lookups. 7811da177e4SLinus Torvalds */ 7823abb17e8SLinus Torvalds static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd) 7831da177e4SLinus Torvalds { 784def4af30SAl Viro void *cookie; 7851da177e4SLinus Torvalds int err = -ELOOP; 786844a3917SAl Viro 787844a3917SAl Viro /* We drop rcu-walk here */ 788844a3917SAl Viro if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) 789844a3917SAl Viro return -ECHILD; 7903abb17e8SLinus Torvalds BUG_ON(inode != path->dentry->d_inode); 791844a3917SAl Viro 7921da177e4SLinus Torvalds if (current->link_count >= MAX_NESTED_LINKS) 7931da177e4SLinus Torvalds goto loop; 7941da177e4SLinus Torvalds if (current->total_link_count >= 40) 7951da177e4SLinus Torvalds goto loop; 7961da177e4SLinus Torvalds BUG_ON(nd->depth >= MAX_NESTED_LINKS); 7971da177e4SLinus Torvalds cond_resched(); 7981da177e4SLinus Torvalds current->link_count++; 7991da177e4SLinus Torvalds current->total_link_count++; 8001da177e4SLinus Torvalds nd->depth++; 801def4af30SAl Viro err = __do_follow_link(path, nd, &cookie); 802def4af30SAl Viro if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) 803def4af30SAl Viro path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); 804258fa999SAl Viro path_put(path); 8051da177e4SLinus Torvalds current->link_count--; 8061da177e4SLinus Torvalds nd->depth--; 8071da177e4SLinus Torvalds return err; 8081da177e4SLinus Torvalds loop: 8091d957f9bSJan Blunck path_put_conditional(path, nd); 8101d957f9bSJan Blunck path_put(&nd->path); 8111da177e4SLinus Torvalds return err; 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds 81431e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 81531e6b01fSNick Piggin { 81631e6b01fSNick Piggin struct vfsmount *parent; 81731e6b01fSNick Piggin struct dentry *mountpoint; 81831e6b01fSNick Piggin 81931e6b01fSNick Piggin parent = path->mnt->mnt_parent; 82031e6b01fSNick Piggin if (parent == path->mnt) 82131e6b01fSNick Piggin return 0; 82231e6b01fSNick Piggin mountpoint = path->mnt->mnt_mountpoint; 82331e6b01fSNick Piggin path->dentry = mountpoint; 82431e6b01fSNick Piggin path->mnt = parent; 82531e6b01fSNick Piggin return 1; 82631e6b01fSNick Piggin } 82731e6b01fSNick Piggin 828bab77ebfSAl Viro int follow_up(struct path *path) 8291da177e4SLinus Torvalds { 8301da177e4SLinus Torvalds struct vfsmount *parent; 8311da177e4SLinus Torvalds struct dentry *mountpoint; 83299b7db7bSNick Piggin 83399b7db7bSNick Piggin br_read_lock(vfsmount_lock); 834bab77ebfSAl Viro parent = path->mnt->mnt_parent; 835bab77ebfSAl Viro if (parent == path->mnt) { 83699b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 8371da177e4SLinus Torvalds return 0; 8381da177e4SLinus Torvalds } 8391da177e4SLinus Torvalds mntget(parent); 840bab77ebfSAl Viro mountpoint = dget(path->mnt->mnt_mountpoint); 84199b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 842bab77ebfSAl Viro dput(path->dentry); 843bab77ebfSAl Viro path->dentry = mountpoint; 844bab77ebfSAl Viro mntput(path->mnt); 845bab77ebfSAl Viro path->mnt = parent; 8461da177e4SLinus Torvalds return 1; 8471da177e4SLinus Torvalds } 8481da177e4SLinus Torvalds 849b5c84bf6SNick Piggin /* 8509875cf80SDavid Howells * Perform an automount 8519875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 8529875cf80SDavid Howells * were called with. 8531da177e4SLinus Torvalds */ 8549875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 8559875cf80SDavid Howells bool *need_mntput) 85631e6b01fSNick Piggin { 8579875cf80SDavid Howells struct vfsmount *mnt; 858ea5b778aSDavid Howells int err; 8599875cf80SDavid Howells 8609875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 8619875cf80SDavid Howells return -EREMOTE; 8629875cf80SDavid Howells 8636f45b656SDavid Howells /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT 8646f45b656SDavid Howells * and this is the terminal part of the path. 8656f45b656SDavid Howells */ 8666f45b656SDavid Howells if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE)) 8676f45b656SDavid Howells return -EISDIR; /* we actually want to stop here */ 8686f45b656SDavid Howells 8699875cf80SDavid Howells /* We want to mount if someone is trying to open/create a file of any 8709875cf80SDavid Howells * type under the mountpoint, wants to traverse through the mountpoint 8719875cf80SDavid Howells * or wants to open the mounted directory. 8729875cf80SDavid Howells * 8739875cf80SDavid Howells * We don't want to mount if someone's just doing a stat and they've 8749875cf80SDavid Howells * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and 8759875cf80SDavid Howells * appended a '/' to the name. 8769875cf80SDavid Howells */ 8779875cf80SDavid Howells if (!(flags & LOOKUP_FOLLOW) && 8789875cf80SDavid Howells !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | 8799875cf80SDavid Howells LOOKUP_OPEN | LOOKUP_CREATE))) 8809875cf80SDavid Howells return -EISDIR; 8819875cf80SDavid Howells 8829875cf80SDavid Howells current->total_link_count++; 8839875cf80SDavid Howells if (current->total_link_count >= 40) 8849875cf80SDavid Howells return -ELOOP; 8859875cf80SDavid Howells 8869875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 8879875cf80SDavid Howells if (IS_ERR(mnt)) { 8889875cf80SDavid Howells /* 8899875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 8909875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 8919875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 8929875cf80SDavid Howells * 8939875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 8949875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 8959875cf80SDavid Howells * the path is inaccessible and we should say so. 8969875cf80SDavid Howells */ 8979875cf80SDavid Howells if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) 8989875cf80SDavid Howells return -EREMOTE; 8999875cf80SDavid Howells return PTR_ERR(mnt); 90031e6b01fSNick Piggin } 901ea5b778aSDavid Howells 9029875cf80SDavid Howells if (!mnt) /* mount collision */ 9039875cf80SDavid Howells return 0; 9049875cf80SDavid Howells 90519a167afSAl Viro err = finish_automount(mnt, path); 906ea5b778aSDavid Howells 907ea5b778aSDavid Howells switch (err) { 908ea5b778aSDavid Howells case -EBUSY: 909ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 91019a167afSAl Viro return 0; 911ea5b778aSDavid Howells case 0: 912463ffb2eSAl Viro dput(path->dentry); 9139875cf80SDavid Howells if (*need_mntput) 9149875cf80SDavid Howells mntput(path->mnt); 9159875cf80SDavid Howells path->mnt = mnt; 9169875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 9179875cf80SDavid Howells *need_mntput = true; 9189875cf80SDavid Howells return 0; 91919a167afSAl Viro default: 92019a167afSAl Viro return err; 9219875cf80SDavid Howells } 92219a167afSAl Viro 923ea5b778aSDavid Howells } 9249875cf80SDavid Howells 9259875cf80SDavid Howells /* 9269875cf80SDavid Howells * Handle a dentry that is managed in some way. 927cc53ce53SDavid Howells * - Flagged for transit management (autofs) 9289875cf80SDavid Howells * - Flagged as mountpoint 9299875cf80SDavid Howells * - Flagged as automount point 9309875cf80SDavid Howells * 9319875cf80SDavid Howells * This may only be called in refwalk mode. 9329875cf80SDavid Howells * 9339875cf80SDavid Howells * Serialization is taken care of in namespace.c 9349875cf80SDavid Howells */ 9359875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 9369875cf80SDavid Howells { 9379875cf80SDavid Howells unsigned managed; 9389875cf80SDavid Howells bool need_mntput = false; 9399875cf80SDavid Howells int ret; 9409875cf80SDavid Howells 9419875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 9429875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 9439875cf80SDavid Howells * the components of that value change under us */ 9449875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 9459875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 9469875cf80SDavid Howells unlikely(managed != 0)) { 947cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 948cc53ce53SDavid Howells * being held. */ 949cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 950cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 951cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 952ab90911fSDavid Howells ret = path->dentry->d_op->d_manage(path->dentry, 953ab90911fSDavid Howells false, false); 954cc53ce53SDavid Howells if (ret < 0) 955cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 956cc53ce53SDavid Howells } 957cc53ce53SDavid Howells 9589875cf80SDavid Howells /* Transit to a mounted filesystem. */ 9599875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 9609875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 9619875cf80SDavid Howells if (mounted) { 9629875cf80SDavid Howells dput(path->dentry); 9639875cf80SDavid Howells if (need_mntput) 964463ffb2eSAl Viro mntput(path->mnt); 965463ffb2eSAl Viro path->mnt = mounted; 966463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 9679875cf80SDavid Howells need_mntput = true; 9689875cf80SDavid Howells continue; 969463ffb2eSAl Viro } 970463ffb2eSAl Viro 9719875cf80SDavid Howells /* Something is mounted on this dentry in another 9729875cf80SDavid Howells * namespace and/or whatever was mounted there in this 9739875cf80SDavid Howells * namespace got unmounted before we managed to get the 9749875cf80SDavid Howells * vfsmount_lock */ 9751da177e4SLinus Torvalds } 9769875cf80SDavid Howells 9779875cf80SDavid Howells /* Handle an automount point */ 9789875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 9799875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 9809875cf80SDavid Howells if (ret < 0) 9819875cf80SDavid Howells return ret == -EISDIR ? 0 : ret; 9829875cf80SDavid Howells continue; 9839875cf80SDavid Howells } 9849875cf80SDavid Howells 9859875cf80SDavid Howells /* We didn't change the current path point */ 9869875cf80SDavid Howells break; 9879875cf80SDavid Howells } 9889875cf80SDavid Howells return 0; 9891da177e4SLinus Torvalds } 9901da177e4SLinus Torvalds 991cc53ce53SDavid Howells int follow_down_one(struct path *path) 9921da177e4SLinus Torvalds { 9931da177e4SLinus Torvalds struct vfsmount *mounted; 9941da177e4SLinus Torvalds 9951c755af4SAl Viro mounted = lookup_mnt(path); 9961da177e4SLinus Torvalds if (mounted) { 9979393bd07SAl Viro dput(path->dentry); 9989393bd07SAl Viro mntput(path->mnt); 9999393bd07SAl Viro path->mnt = mounted; 10009393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 10011da177e4SLinus Torvalds return 1; 10021da177e4SLinus Torvalds } 10031da177e4SLinus Torvalds return 0; 10041da177e4SLinus Torvalds } 10051da177e4SLinus Torvalds 10069875cf80SDavid Howells /* 10079875cf80SDavid Howells * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we 1008cc53ce53SDavid Howells * meet a managed dentry and we're not walking to "..". True is returned to 10099875cf80SDavid Howells * continue, false to abort. 10109875cf80SDavid Howells */ 10119875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 10129875cf80SDavid Howells struct inode **inode, bool reverse_transit) 10139875cf80SDavid Howells { 10149875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 10159875cf80SDavid Howells struct vfsmount *mounted; 1016ab90911fSDavid Howells if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) && 1017ab90911fSDavid Howells !reverse_transit && 1018ab90911fSDavid Howells path->dentry->d_op->d_manage(path->dentry, false, true) < 0) 1019ab90911fSDavid Howells return false; 10209875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 10219875cf80SDavid Howells if (!mounted) 10229875cf80SDavid Howells break; 10239875cf80SDavid Howells path->mnt = mounted; 10249875cf80SDavid Howells path->dentry = mounted->mnt_root; 10259875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 10269875cf80SDavid Howells *inode = path->dentry->d_inode; 10279875cf80SDavid Howells } 10289875cf80SDavid Howells 10299875cf80SDavid Howells if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 10309875cf80SDavid Howells return reverse_transit; 10319875cf80SDavid Howells return true; 10329875cf80SDavid Howells } 10339875cf80SDavid Howells 103431e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 103531e6b01fSNick Piggin { 103631e6b01fSNick Piggin struct inode *inode = nd->inode; 103731e6b01fSNick Piggin 103831e6b01fSNick Piggin set_root_rcu(nd); 103931e6b01fSNick Piggin 104031e6b01fSNick Piggin while (1) { 104131e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 104231e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 104331e6b01fSNick Piggin break; 104431e6b01fSNick Piggin } 104531e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 104631e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 104731e6b01fSNick Piggin struct dentry *parent = old->d_parent; 104831e6b01fSNick Piggin unsigned seq; 104931e6b01fSNick Piggin 105031e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 105131e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1052ef7562d5SAl Viro goto failed; 105331e6b01fSNick Piggin inode = parent->d_inode; 105431e6b01fSNick Piggin nd->path.dentry = parent; 105531e6b01fSNick Piggin nd->seq = seq; 105631e6b01fSNick Piggin break; 105731e6b01fSNick Piggin } 105831e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 105931e6b01fSNick Piggin break; 106031e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 106131e6b01fSNick Piggin inode = nd->path.dentry->d_inode; 106231e6b01fSNick Piggin } 10639875cf80SDavid Howells __follow_mount_rcu(nd, &nd->path, &inode, true); 106431e6b01fSNick Piggin nd->inode = inode; 106531e6b01fSNick Piggin return 0; 1066ef7562d5SAl Viro 1067ef7562d5SAl Viro failed: 1068ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 10695b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1070ef7562d5SAl Viro nd->root.mnt = NULL; 1071ef7562d5SAl Viro rcu_read_unlock(); 1072ef7562d5SAl Viro br_read_unlock(vfsmount_lock); 1073ef7562d5SAl Viro return -ECHILD; 107431e6b01fSNick Piggin } 107531e6b01fSNick Piggin 10769875cf80SDavid Howells /* 1077cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1078cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1079cc53ce53SDavid Howells * caller is permitted to proceed or not. 1080cc53ce53SDavid Howells * 1081cc53ce53SDavid Howells * Care must be taken as namespace_sem may be held (indicated by mounting_here 1082cc53ce53SDavid Howells * being true). 1083cc53ce53SDavid Howells */ 1084cc53ce53SDavid Howells int follow_down(struct path *path, bool mounting_here) 1085cc53ce53SDavid Howells { 1086cc53ce53SDavid Howells unsigned managed; 1087cc53ce53SDavid Howells int ret; 1088cc53ce53SDavid Howells 1089cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1090cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1091cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1092cc53ce53SDavid Howells * being held. 1093cc53ce53SDavid Howells * 1094cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1095cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1096cc53ce53SDavid Howells * other than its daemon the right to mount on its 1097cc53ce53SDavid Howells * superstructure. 1098cc53ce53SDavid Howells * 1099cc53ce53SDavid Howells * The filesystem may sleep at this point. 1100cc53ce53SDavid Howells */ 1101cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1102cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1103cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1104ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 1105ab90911fSDavid Howells path->dentry, mounting_here, false); 1106cc53ce53SDavid Howells if (ret < 0) 1107cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1108cc53ce53SDavid Howells } 1109cc53ce53SDavid Howells 1110cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1111cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1112cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1113cc53ce53SDavid Howells if (!mounted) 1114cc53ce53SDavid Howells break; 1115cc53ce53SDavid Howells dput(path->dentry); 1116cc53ce53SDavid Howells mntput(path->mnt); 1117cc53ce53SDavid Howells path->mnt = mounted; 1118cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1119cc53ce53SDavid Howells continue; 1120cc53ce53SDavid Howells } 1121cc53ce53SDavid Howells 1122cc53ce53SDavid Howells /* Don't handle automount points here */ 1123cc53ce53SDavid Howells break; 1124cc53ce53SDavid Howells } 1125cc53ce53SDavid Howells return 0; 1126cc53ce53SDavid Howells } 1127cc53ce53SDavid Howells 1128cc53ce53SDavid Howells /* 11299875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 11309875cf80SDavid Howells */ 11319875cf80SDavid Howells static void follow_mount(struct path *path) 11329875cf80SDavid Howells { 11339875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 11349875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11359875cf80SDavid Howells if (!mounted) 11369875cf80SDavid Howells break; 11379875cf80SDavid Howells dput(path->dentry); 11389875cf80SDavid Howells mntput(path->mnt); 11399875cf80SDavid Howells path->mnt = mounted; 11409875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 11419875cf80SDavid Howells } 11429875cf80SDavid Howells } 11439875cf80SDavid Howells 114431e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 11451da177e4SLinus Torvalds { 11462a737871SAl Viro set_root(nd); 1147e518ddb7SAndreas Mohr 11481da177e4SLinus Torvalds while(1) { 11494ac91378SJan Blunck struct dentry *old = nd->path.dentry; 11501da177e4SLinus Torvalds 11512a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 11522a737871SAl Viro nd->path.mnt == nd->root.mnt) { 11531da177e4SLinus Torvalds break; 11541da177e4SLinus Torvalds } 11554ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 11563088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 11573088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 11581da177e4SLinus Torvalds dput(old); 11591da177e4SLinus Torvalds break; 11601da177e4SLinus Torvalds } 11613088dd70SAl Viro if (!follow_up(&nd->path)) 11621da177e4SLinus Torvalds break; 11631da177e4SLinus Torvalds } 116479ed0226SAl Viro follow_mount(&nd->path); 116531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds /* 1169baa03890SNick Piggin * Allocate a dentry with name and parent, and perform a parent 1170baa03890SNick Piggin * directory ->lookup on it. Returns the new dentry, or ERR_PTR 1171baa03890SNick Piggin * on error. parent->d_inode->i_mutex must be held. d_lookup must 1172baa03890SNick Piggin * have verified that no child exists while under i_mutex. 1173baa03890SNick Piggin */ 1174baa03890SNick Piggin static struct dentry *d_alloc_and_lookup(struct dentry *parent, 1175baa03890SNick Piggin struct qstr *name, struct nameidata *nd) 1176baa03890SNick Piggin { 1177baa03890SNick Piggin struct inode *inode = parent->d_inode; 1178baa03890SNick Piggin struct dentry *dentry; 1179baa03890SNick Piggin struct dentry *old; 1180baa03890SNick Piggin 1181baa03890SNick Piggin /* Don't create child dentry for a dead directory. */ 1182baa03890SNick Piggin if (unlikely(IS_DEADDIR(inode))) 1183baa03890SNick Piggin return ERR_PTR(-ENOENT); 1184baa03890SNick Piggin 1185baa03890SNick Piggin dentry = d_alloc(parent, name); 1186baa03890SNick Piggin if (unlikely(!dentry)) 1187baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1188baa03890SNick Piggin 1189baa03890SNick Piggin old = inode->i_op->lookup(inode, dentry, nd); 1190baa03890SNick Piggin if (unlikely(old)) { 1191baa03890SNick Piggin dput(dentry); 1192baa03890SNick Piggin dentry = old; 1193baa03890SNick Piggin } 1194baa03890SNick Piggin return dentry; 1195baa03890SNick Piggin } 1196baa03890SNick Piggin 1197baa03890SNick Piggin /* 11981da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 11991da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 12001da177e4SLinus Torvalds * It _is_ time-critical. 12011da177e4SLinus Torvalds */ 12021da177e4SLinus Torvalds static int do_lookup(struct nameidata *nd, struct qstr *name, 120331e6b01fSNick Piggin struct path *path, struct inode **inode) 12041da177e4SLinus Torvalds { 12054ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 120631e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 12075a18fff2SAl Viro int need_reval = 1; 12085a18fff2SAl Viro int status = 1; 12099875cf80SDavid Howells int err; 12109875cf80SDavid Howells 12113cac260aSAl Viro /* 1212b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1213b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1214b04f784eSNick Piggin * do the non-racy lookup, below. 1215b04f784eSNick Piggin */ 121631e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 121731e6b01fSNick Piggin unsigned seq; 121831e6b01fSNick Piggin *inode = nd->inode; 121931e6b01fSNick Piggin dentry = __d_lookup_rcu(parent, name, &seq, inode); 12205a18fff2SAl Viro if (!dentry) 12215a18fff2SAl Viro goto unlazy; 12225a18fff2SAl Viro 122331e6b01fSNick Piggin /* Memory barrier in read_seqcount_begin of child is enough */ 122431e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 122531e6b01fSNick Piggin return -ECHILD; 122631e6b01fSNick Piggin nd->seq = seq; 12275a18fff2SAl Viro 122824643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 12295a18fff2SAl Viro status = d_revalidate(dentry, nd); 12305a18fff2SAl Viro if (unlikely(status <= 0)) { 12315a18fff2SAl Viro if (status != -ECHILD) 12325a18fff2SAl Viro need_reval = 0; 12335a18fff2SAl Viro goto unlazy; 12345a18fff2SAl Viro } 123524643087SAl Viro } 123631e6b01fSNick Piggin path->mnt = mnt; 123731e6b01fSNick Piggin path->dentry = dentry; 12389875cf80SDavid Howells if (likely(__follow_mount_rcu(nd, path, inode, false))) 12399875cf80SDavid Howells return 0; 12405a18fff2SAl Viro unlazy: 12415a18fff2SAl Viro if (dentry) { 12425a18fff2SAl Viro if (nameidata_dentry_drop_rcu(nd, dentry)) 12435a18fff2SAl Viro return -ECHILD; 12445a18fff2SAl Viro } else { 12459875cf80SDavid Howells if (nameidata_drop_rcu(nd)) 12469875cf80SDavid Howells return -ECHILD; 12479875cf80SDavid Howells } 12485a18fff2SAl Viro } else { 124931e6b01fSNick Piggin dentry = __d_lookup(parent, name); 125024643087SAl Viro } 12515a18fff2SAl Viro 12525a18fff2SAl Viro retry: 12535a18fff2SAl Viro if (unlikely(!dentry)) { 12545a18fff2SAl Viro struct inode *dir = parent->d_inode; 12555a18fff2SAl Viro BUG_ON(nd->inode != dir); 12565a18fff2SAl Viro 12575a18fff2SAl Viro mutex_lock(&dir->i_mutex); 12585a18fff2SAl Viro dentry = d_lookup(parent, name); 12595a18fff2SAl Viro if (likely(!dentry)) { 12605a18fff2SAl Viro dentry = d_alloc_and_lookup(parent, name, nd); 12615a18fff2SAl Viro if (IS_ERR(dentry)) { 12625a18fff2SAl Viro mutex_unlock(&dir->i_mutex); 12635a18fff2SAl Viro return PTR_ERR(dentry); 12645a18fff2SAl Viro } 12655a18fff2SAl Viro /* known good */ 12665a18fff2SAl Viro need_reval = 0; 12675a18fff2SAl Viro status = 1; 12685a18fff2SAl Viro } 12695a18fff2SAl Viro mutex_unlock(&dir->i_mutex); 12705a18fff2SAl Viro } 12715a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 12725a18fff2SAl Viro status = d_revalidate(dentry, nd); 12735a18fff2SAl Viro if (unlikely(status <= 0)) { 12745a18fff2SAl Viro if (status < 0) { 12755a18fff2SAl Viro dput(dentry); 12765a18fff2SAl Viro return status; 12775a18fff2SAl Viro } 12785a18fff2SAl Viro if (!d_invalidate(dentry)) { 12795a18fff2SAl Viro dput(dentry); 12805a18fff2SAl Viro dentry = NULL; 12815a18fff2SAl Viro need_reval = 1; 12825a18fff2SAl Viro goto retry; 12835a18fff2SAl Viro } 12845a18fff2SAl Viro } 12855a18fff2SAl Viro 12861da177e4SLinus Torvalds path->mnt = mnt; 12871da177e4SLinus Torvalds path->dentry = dentry; 12889875cf80SDavid Howells err = follow_managed(path, nd->flags); 128989312214SIan Kent if (unlikely(err < 0)) { 129089312214SIan Kent path_put_conditional(path, nd); 12919875cf80SDavid Howells return err; 129289312214SIan Kent } 129331e6b01fSNick Piggin *inode = path->dentry->d_inode; 12941da177e4SLinus Torvalds return 0; 12951da177e4SLinus Torvalds } 12961da177e4SLinus Torvalds 129752094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 129852094c8aSAl Viro { 129952094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 130052094c8aSAl Viro int err = exec_permission(nd->inode, IPERM_FLAG_RCU); 130152094c8aSAl Viro if (err != -ECHILD) 130252094c8aSAl Viro return err; 130352094c8aSAl Viro if (nameidata_drop_rcu(nd)) 130452094c8aSAl Viro return -ECHILD; 130552094c8aSAl Viro } 130652094c8aSAl Viro return exec_permission(nd->inode, 0); 130752094c8aSAl Viro } 130852094c8aSAl Viro 13099856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 13109856fa1bSAl Viro { 13119856fa1bSAl Viro if (type == LAST_DOTDOT) { 13129856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 13139856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 13149856fa1bSAl Viro return -ECHILD; 13159856fa1bSAl Viro } else 13169856fa1bSAl Viro follow_dotdot(nd); 13179856fa1bSAl Viro } 13189856fa1bSAl Viro return 0; 13199856fa1bSAl Viro } 13209856fa1bSAl Viro 1321951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1322951361f9SAl Viro { 1323951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1324951361f9SAl Viro path_put(&nd->path); 1325951361f9SAl Viro } else { 1326951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 13275b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1328951361f9SAl Viro nd->root.mnt = NULL; 1329951361f9SAl Viro rcu_read_unlock(); 1330951361f9SAl Viro br_read_unlock(vfsmount_lock); 1331951361f9SAl Viro } 1332951361f9SAl Viro } 1333951361f9SAl Viro 13341da177e4SLinus Torvalds /* 13351da177e4SLinus Torvalds * Name resolution. 1336ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1337ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 13381da177e4SLinus Torvalds * 1339ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1340ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 13411da177e4SLinus Torvalds */ 13426de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 13431da177e4SLinus Torvalds { 13441da177e4SLinus Torvalds struct path next; 13451da177e4SLinus Torvalds int err; 13461da177e4SLinus Torvalds unsigned int lookup_flags = nd->flags; 13471da177e4SLinus Torvalds 13481da177e4SLinus Torvalds while (*name=='/') 13491da177e4SLinus Torvalds name++; 13501da177e4SLinus Torvalds if (!*name) 1351086e183aSAl Viro return 0; 13521da177e4SLinus Torvalds 13531da177e4SLinus Torvalds if (nd->depth) 1354f55eab82STrond Myklebust lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); 13551da177e4SLinus Torvalds 13561da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 13571da177e4SLinus Torvalds for(;;) { 135831e6b01fSNick Piggin struct inode *inode; 13591da177e4SLinus Torvalds unsigned long hash; 13601da177e4SLinus Torvalds struct qstr this; 13611da177e4SLinus Torvalds unsigned int c; 1362fe479a58SAl Viro int type; 13631da177e4SLinus Torvalds 1364cdce5d6bSTrond Myklebust nd->flags |= LOOKUP_CONTINUE; 136552094c8aSAl Viro 136652094c8aSAl Viro err = may_lookup(nd); 13671da177e4SLinus Torvalds if (err) 13681da177e4SLinus Torvalds break; 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds this.name = name; 13711da177e4SLinus Torvalds c = *(const unsigned char *)name; 13721da177e4SLinus Torvalds 13731da177e4SLinus Torvalds hash = init_name_hash(); 13741da177e4SLinus Torvalds do { 13751da177e4SLinus Torvalds name++; 13761da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 13771da177e4SLinus Torvalds c = *(const unsigned char *)name; 13781da177e4SLinus Torvalds } while (c && (c != '/')); 13791da177e4SLinus Torvalds this.len = name - (const char *) this.name; 13801da177e4SLinus Torvalds this.hash = end_name_hash(hash); 13811da177e4SLinus Torvalds 1382fe479a58SAl Viro type = LAST_NORM; 1383fe479a58SAl Viro if (this.name[0] == '.') switch (this.len) { 1384fe479a58SAl Viro case 2: 138516c2cd71SAl Viro if (this.name[1] == '.') { 1386fe479a58SAl Viro type = LAST_DOTDOT; 138716c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 138816c2cd71SAl Viro } 1389fe479a58SAl Viro break; 1390fe479a58SAl Viro case 1: 1391fe479a58SAl Viro type = LAST_DOT; 1392fe479a58SAl Viro } 13935a202bcdSAl Viro if (likely(type == LAST_NORM)) { 13945a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 139516c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 13965a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 13975a202bcdSAl Viro err = parent->d_op->d_hash(parent, nd->inode, 13985a202bcdSAl Viro &this); 13995a202bcdSAl Viro if (err < 0) 14005a202bcdSAl Viro break; 14015a202bcdSAl Viro } 14025a202bcdSAl Viro } 1403fe479a58SAl Viro 14041da177e4SLinus Torvalds /* remove trailing slashes? */ 14051da177e4SLinus Torvalds if (!c) 14061da177e4SLinus Torvalds goto last_component; 14071da177e4SLinus Torvalds while (*++name == '/'); 14081da177e4SLinus Torvalds if (!*name) 14091da177e4SLinus Torvalds goto last_with_slashes; 14101da177e4SLinus Torvalds 14111da177e4SLinus Torvalds /* 14121da177e4SLinus Torvalds * "." and ".." are special - ".." especially so because it has 14131da177e4SLinus Torvalds * to be able to know about the current root directory and 14141da177e4SLinus Torvalds * parent relationships. 14151da177e4SLinus Torvalds */ 1416fe479a58SAl Viro if (unlikely(type != LAST_NORM)) { 1417ef7562d5SAl Viro if (handle_dots(nd, type)) 1418ef7562d5SAl Viro return -ECHILD; 14191da177e4SLinus Torvalds continue; 14201da177e4SLinus Torvalds } 1421fe479a58SAl Viro 14221da177e4SLinus Torvalds /* This does the actual lookups.. */ 142331e6b01fSNick Piggin err = do_lookup(nd, &this, &next, &inode); 14241da177e4SLinus Torvalds if (err) 14251da177e4SLinus Torvalds break; 14261da177e4SLinus Torvalds 14277bc055d1SAl Viro if (inode && inode->i_op->follow_link) { 14283abb17e8SLinus Torvalds err = do_follow_link(inode, &next, nd); 14291da177e4SLinus Torvalds if (err) 1430a7472babSAl Viro return err; 143131e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 143231e6b01fSNick Piggin } else { 143309dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 143431e6b01fSNick Piggin nd->inode = inode; 143531e6b01fSNick Piggin } 14367bc055d1SAl Viro err = -ENOENT; 14377bc055d1SAl Viro if (!nd->inode) 14387bc055d1SAl Viro break; 14391da177e4SLinus Torvalds err = -ENOTDIR; 144031e6b01fSNick Piggin if (!nd->inode->i_op->lookup) 14411da177e4SLinus Torvalds break; 14421da177e4SLinus Torvalds continue; 14431da177e4SLinus Torvalds /* here ends the main loop */ 14441da177e4SLinus Torvalds 14451da177e4SLinus Torvalds last_with_slashes: 14461da177e4SLinus Torvalds lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 14471da177e4SLinus Torvalds last_component: 1448f55eab82STrond Myklebust /* Clear LOOKUP_CONTINUE iff it was previously unset */ 1449f55eab82STrond Myklebust nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; 14501da177e4SLinus Torvalds if (lookup_flags & LOOKUP_PARENT) 14511da177e4SLinus Torvalds goto lookup_parent; 1452ef7562d5SAl Viro if (unlikely(type != LAST_NORM)) 1453ef7562d5SAl Viro return handle_dots(nd, type); 145431e6b01fSNick Piggin err = do_lookup(nd, &this, &next, &inode); 14551da177e4SLinus Torvalds if (err) 14561da177e4SLinus Torvalds break; 1457db372915SDavid Howells if (inode && unlikely(inode->i_op->follow_link) && 1458db372915SDavid Howells (lookup_flags & LOOKUP_FOLLOW)) { 14593abb17e8SLinus Torvalds err = do_follow_link(inode, &next, nd); 14601da177e4SLinus Torvalds if (err) 1461a7472babSAl Viro return err; 146231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 146331e6b01fSNick Piggin } else { 146409dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 146531e6b01fSNick Piggin nd->inode = inode; 146631e6b01fSNick Piggin } 14671da177e4SLinus Torvalds err = -ENOENT; 146831e6b01fSNick Piggin if (!nd->inode) 14691da177e4SLinus Torvalds break; 14701da177e4SLinus Torvalds if (lookup_flags & LOOKUP_DIRECTORY) { 14711da177e4SLinus Torvalds err = -ENOTDIR; 147231e6b01fSNick Piggin if (!nd->inode->i_op->lookup) 14731da177e4SLinus Torvalds break; 14741da177e4SLinus Torvalds } 1475086e183aSAl Viro return 0; 14761da177e4SLinus Torvalds lookup_parent: 14771da177e4SLinus Torvalds nd->last = this; 1478fe479a58SAl Viro nd->last_type = type; 14791da177e4SLinus Torvalds return 0; 14801da177e4SLinus Torvalds } 1481951361f9SAl Viro terminate_walk(nd); 14821da177e4SLinus Torvalds return err; 14831da177e4SLinus Torvalds } 14841da177e4SLinus Torvalds 148570e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 148670e9b357SAl Viro struct nameidata *nd, struct file **fp) 148731e6b01fSNick Piggin { 148831e6b01fSNick Piggin int retval = 0; 148931e6b01fSNick Piggin int fput_needed; 149031e6b01fSNick Piggin struct file *file; 149131e6b01fSNick Piggin 149231e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 149316c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 149431e6b01fSNick Piggin nd->depth = 0; 14955b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 14965b6ca027SAl Viro struct inode *inode = nd->root.dentry->d_inode; 149773d049a4SAl Viro if (*name) { 14985b6ca027SAl Viro if (!inode->i_op->lookup) 14995b6ca027SAl Viro return -ENOTDIR; 15005b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 15015b6ca027SAl Viro if (retval) 15025b6ca027SAl Viro return retval; 150373d049a4SAl Viro } 15045b6ca027SAl Viro nd->path = nd->root; 15055b6ca027SAl Viro nd->inode = inode; 15065b6ca027SAl Viro if (flags & LOOKUP_RCU) { 15075b6ca027SAl Viro br_read_lock(vfsmount_lock); 15085b6ca027SAl Viro rcu_read_lock(); 15095b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 15105b6ca027SAl Viro } else { 15115b6ca027SAl Viro path_get(&nd->path); 15125b6ca027SAl Viro } 15135b6ca027SAl Viro return 0; 15145b6ca027SAl Viro } 15155b6ca027SAl Viro 151631e6b01fSNick Piggin nd->root.mnt = NULL; 151731e6b01fSNick Piggin 151831e6b01fSNick Piggin if (*name=='/') { 1519e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 152031e6b01fSNick Piggin br_read_lock(vfsmount_lock); 152131e6b01fSNick Piggin rcu_read_lock(); 1522e41f7d4eSAl Viro set_root_rcu(nd); 1523e41f7d4eSAl Viro } else { 1524e41f7d4eSAl Viro set_root(nd); 1525e41f7d4eSAl Viro path_get(&nd->root); 1526e41f7d4eSAl Viro } 152731e6b01fSNick Piggin nd->path = nd->root; 152831e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1529e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 153031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1531c28cc364SNick Piggin unsigned seq; 153231e6b01fSNick Piggin 153331e6b01fSNick Piggin br_read_lock(vfsmount_lock); 153431e6b01fSNick Piggin rcu_read_lock(); 153531e6b01fSNick Piggin 1536c28cc364SNick Piggin do { 1537c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 153831e6b01fSNick Piggin nd->path = fs->pwd; 1539c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1540c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1541e41f7d4eSAl Viro } else { 1542e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1543e41f7d4eSAl Viro } 154431e6b01fSNick Piggin } else { 154531e6b01fSNick Piggin struct dentry *dentry; 154631e6b01fSNick Piggin 154731e6b01fSNick Piggin file = fget_light(dfd, &fput_needed); 154831e6b01fSNick Piggin retval = -EBADF; 154931e6b01fSNick Piggin if (!file) 155031e6b01fSNick Piggin goto out_fail; 155131e6b01fSNick Piggin 155231e6b01fSNick Piggin dentry = file->f_path.dentry; 155331e6b01fSNick Piggin 1554f52e0c11SAl Viro if (*name) { 155531e6b01fSNick Piggin retval = -ENOTDIR; 155631e6b01fSNick Piggin if (!S_ISDIR(dentry->d_inode->i_mode)) 155731e6b01fSNick Piggin goto fput_fail; 155831e6b01fSNick Piggin 155931e6b01fSNick Piggin retval = file_permission(file, MAY_EXEC); 156031e6b01fSNick Piggin if (retval) 156131e6b01fSNick Piggin goto fput_fail; 1562f52e0c11SAl Viro } 156331e6b01fSNick Piggin 156431e6b01fSNick Piggin nd->path = file->f_path; 1565e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 156631e6b01fSNick Piggin if (fput_needed) 156770e9b357SAl Viro *fp = file; 1568c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 156931e6b01fSNick Piggin br_read_lock(vfsmount_lock); 157031e6b01fSNick Piggin rcu_read_lock(); 15715590ff0dSUlrich Drepper } else { 15725dd784d0SJan Blunck path_get(&file->f_path); 15735590ff0dSUlrich Drepper fput_light(file, fput_needed); 15741da177e4SLinus Torvalds } 1575e41f7d4eSAl Viro } 1576e41f7d4eSAl Viro 157731e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 15789b4a9b14SAl Viro return 0; 15792dfdd266SJosef 'Jeff' Sipek 15809b4a9b14SAl Viro fput_fail: 15819b4a9b14SAl Viro fput_light(file, fput_needed); 15829b4a9b14SAl Viro out_fail: 15839b4a9b14SAl Viro return retval; 15849b4a9b14SAl Viro } 15859b4a9b14SAl Viro 15869b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1587ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 15889b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 15899b4a9b14SAl Viro { 159070e9b357SAl Viro struct file *base = NULL; 159131e6b01fSNick Piggin int retval; 159231e6b01fSNick Piggin 159331e6b01fSNick Piggin /* 159431e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 159531e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 159631e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 159731e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 159831e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 159931e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 160031e6b01fSNick Piggin * analogue, foo_rcu(). 160131e6b01fSNick Piggin * 160231e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 160331e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 160431e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 160531e6b01fSNick Piggin * be able to complete). 160631e6b01fSNick Piggin */ 160770e9b357SAl Viro retval = path_init(dfd, name, flags, nd, &base); 1608ee0827cdSAl Viro 160931e6b01fSNick Piggin if (unlikely(retval)) 161031e6b01fSNick Piggin return retval; 1611ee0827cdSAl Viro 1612ee0827cdSAl Viro current->total_link_count = 0; 1613ee0827cdSAl Viro retval = link_path_walk(name, nd); 1614ee0827cdSAl Viro 1615ee0827cdSAl Viro if (nd->flags & LOOKUP_RCU) { 16164455ca62SAl Viro /* went all way through without dropping RCU */ 16174455ca62SAl Viro BUG_ON(retval); 1618086e183aSAl Viro if (nameidata_drop_rcu_last(nd)) 1619086e183aSAl Viro retval = -ECHILD; 1620086e183aSAl Viro } 162131e6b01fSNick Piggin 162216c2cd71SAl Viro if (!retval) 162316c2cd71SAl Viro retval = handle_reval_path(nd); 162416c2cd71SAl Viro 162570e9b357SAl Viro if (base) 162670e9b357SAl Viro fput(base); 1627ee0827cdSAl Viro 16285b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 162931e6b01fSNick Piggin path_put(&nd->root); 163031e6b01fSNick Piggin nd->root.mnt = NULL; 163131e6b01fSNick Piggin } 1632ee0827cdSAl Viro return retval; 163331e6b01fSNick Piggin } 163431e6b01fSNick Piggin 1635ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1636ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1637ee0827cdSAl Viro { 1638ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1639ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1640ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1641ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1642ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1643ee0827cdSAl Viro 164431e6b01fSNick Piggin if (likely(!retval)) { 164531e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 164631e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 164731e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 164831e6b01fSNick Piggin } 164931e6b01fSNick Piggin } 1650170aa3d0SUlrich Drepper return retval; 16511da177e4SLinus Torvalds } 16521da177e4SLinus Torvalds 1653c9c6cac0SAl Viro int kern_path_parent(const char *name, struct nameidata *nd) 16545590ff0dSUlrich Drepper { 1655c9c6cac0SAl Viro return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); 16565590ff0dSUlrich Drepper } 16575590ff0dSUlrich Drepper 1658d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1659d1811465SAl Viro { 1660d1811465SAl Viro struct nameidata nd; 1661d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1662d1811465SAl Viro if (!res) 1663d1811465SAl Viro *path = nd.path; 1664d1811465SAl Viro return res; 1665d1811465SAl Viro } 1666d1811465SAl Viro 166716f18200SJosef 'Jeff' Sipek /** 166816f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 166916f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 167016f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 167116f18200SJosef 'Jeff' Sipek * @name: pointer to file name 167216f18200SJosef 'Jeff' Sipek * @flags: lookup flags 167316f18200SJosef 'Jeff' Sipek * @nd: pointer to nameidata 167416f18200SJosef 'Jeff' Sipek */ 167516f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 167616f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 167716f18200SJosef 'Jeff' Sipek struct nameidata *nd) 167816f18200SJosef 'Jeff' Sipek { 16795b6ca027SAl Viro nd->root.dentry = dentry; 16805b6ca027SAl Viro nd->root.mnt = mnt; 16815b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 16825b6ca027SAl Viro return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd); 168316f18200SJosef 'Jeff' Sipek } 168416f18200SJosef 'Jeff' Sipek 1685eead1911SChristoph Hellwig static struct dentry *__lookup_hash(struct qstr *name, 1686eead1911SChristoph Hellwig struct dentry *base, struct nameidata *nd) 16871da177e4SLinus Torvalds { 168881fca444SChristoph Hellwig struct inode *inode = base->d_inode; 16891da177e4SLinus Torvalds struct dentry *dentry; 16901da177e4SLinus Torvalds int err; 16911da177e4SLinus Torvalds 1692b74c79e9SNick Piggin err = exec_permission(inode, 0); 169381fca444SChristoph Hellwig if (err) 169481fca444SChristoph Hellwig return ERR_PTR(err); 16951da177e4SLinus Torvalds 16961da177e4SLinus Torvalds /* 1697b04f784eSNick Piggin * Don't bother with __d_lookup: callers are for creat as 1698b04f784eSNick Piggin * well as unlink, so a lot of the time it would cost 1699b04f784eSNick Piggin * a double lookup. 17006e6b1bd1SAl Viro */ 17016e6b1bd1SAl Viro dentry = d_lookup(base, name); 17026e6b1bd1SAl Viro 1703fb045adbSNick Piggin if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) 17046e6b1bd1SAl Viro dentry = do_revalidate(dentry, nd); 17056e6b1bd1SAl Viro 17061da177e4SLinus Torvalds if (!dentry) 1707baa03890SNick Piggin dentry = d_alloc_and_lookup(base, name, nd); 17085a202bcdSAl Viro 17091da177e4SLinus Torvalds return dentry; 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds 1712057f6c01SJames Morris /* 1713057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1714057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1715057f6c01SJames Morris * SMP-safe. 1716057f6c01SJames Morris */ 1717a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 17181da177e4SLinus Torvalds { 17194ac91378SJan Blunck return __lookup_hash(&nd->last, nd->path.dentry, nd); 17201da177e4SLinus Torvalds } 17211da177e4SLinus Torvalds 1722eead1911SChristoph Hellwig /** 1723a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1724eead1911SChristoph Hellwig * @name: pathname component to lookup 1725eead1911SChristoph Hellwig * @base: base directory to lookup from 1726eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1727eead1911SChristoph Hellwig * 1728a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1729a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1730eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1731eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1732eead1911SChristoph Hellwig */ 1733057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1734057f6c01SJames Morris { 1735057f6c01SJames Morris struct qstr this; 17366a96ba54SAl Viro unsigned long hash; 17376a96ba54SAl Viro unsigned int c; 1738057f6c01SJames Morris 17392f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 17402f9092e1SDavid Woodhouse 17416a96ba54SAl Viro this.name = name; 17426a96ba54SAl Viro this.len = len; 17436a96ba54SAl Viro if (!len) 17446a96ba54SAl Viro return ERR_PTR(-EACCES); 17456a96ba54SAl Viro 17466a96ba54SAl Viro hash = init_name_hash(); 17476a96ba54SAl Viro while (len--) { 17486a96ba54SAl Viro c = *(const unsigned char *)name++; 17496a96ba54SAl Viro if (c == '/' || c == '\0') 17506a96ba54SAl Viro return ERR_PTR(-EACCES); 17516a96ba54SAl Viro hash = partial_name_hash(c, hash); 17526a96ba54SAl Viro } 17536a96ba54SAl Viro this.hash = end_name_hash(hash); 17545a202bcdSAl Viro /* 17555a202bcdSAl Viro * See if the low-level filesystem might want 17565a202bcdSAl Viro * to use its own hash.. 17575a202bcdSAl Viro */ 17585a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 17595a202bcdSAl Viro int err = base->d_op->d_hash(base, base->d_inode, &this); 17605a202bcdSAl Viro if (err < 0) 17615a202bcdSAl Viro return ERR_PTR(err); 17625a202bcdSAl Viro } 1763eead1911SChristoph Hellwig 176449705b77SChristoph Hellwig return __lookup_hash(&this, base, NULL); 1765057f6c01SJames Morris } 1766057f6c01SJames Morris 17672d8f3038SAl Viro int user_path_at(int dfd, const char __user *name, unsigned flags, 17682d8f3038SAl Viro struct path *path) 17691da177e4SLinus Torvalds { 17702d8f3038SAl Viro struct nameidata nd; 1771f52e0c11SAl Viro char *tmp = getname_flags(name, flags); 17721da177e4SLinus Torvalds int err = PTR_ERR(tmp); 17731da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 17742d8f3038SAl Viro 17752d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 17762d8f3038SAl Viro 17772d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 17781da177e4SLinus Torvalds putname(tmp); 17792d8f3038SAl Viro if (!err) 17802d8f3038SAl Viro *path = nd.path; 17811da177e4SLinus Torvalds } 17821da177e4SLinus Torvalds return err; 17831da177e4SLinus Torvalds } 17841da177e4SLinus Torvalds 17852ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 17862ad94ae6SAl Viro struct nameidata *nd, char **name) 17872ad94ae6SAl Viro { 17882ad94ae6SAl Viro char *s = getname(path); 17892ad94ae6SAl Viro int error; 17902ad94ae6SAl Viro 17912ad94ae6SAl Viro if (IS_ERR(s)) 17922ad94ae6SAl Viro return PTR_ERR(s); 17932ad94ae6SAl Viro 17942ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 17952ad94ae6SAl Viro if (error) 17962ad94ae6SAl Viro putname(s); 17972ad94ae6SAl Viro else 17982ad94ae6SAl Viro *name = s; 17992ad94ae6SAl Viro 18002ad94ae6SAl Viro return error; 18012ad94ae6SAl Viro } 18022ad94ae6SAl Viro 18031da177e4SLinus Torvalds /* 18041da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 18051da177e4SLinus Torvalds * minimal. 18061da177e4SLinus Torvalds */ 18071da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 18081da177e4SLinus Torvalds { 1809da9592edSDavid Howells uid_t fsuid = current_fsuid(); 1810da9592edSDavid Howells 18111da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 18121da177e4SLinus Torvalds return 0; 1813da9592edSDavid Howells if (inode->i_uid == fsuid) 18141da177e4SLinus Torvalds return 0; 1815da9592edSDavid Howells if (dir->i_uid == fsuid) 18161da177e4SLinus Torvalds return 0; 18171da177e4SLinus Torvalds return !capable(CAP_FOWNER); 18181da177e4SLinus Torvalds } 18191da177e4SLinus Torvalds 18201da177e4SLinus Torvalds /* 18211da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 18221da177e4SLinus Torvalds * whether the type of victim is right. 18231da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 18241da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 18251da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 18261da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 18271da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 18281da177e4SLinus Torvalds * a. be owner of dir, or 18291da177e4SLinus Torvalds * b. be owner of victim, or 18301da177e4SLinus Torvalds * c. have CAP_FOWNER capability 18311da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 18321da177e4SLinus Torvalds * links pointing to it. 18331da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 18341da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 18351da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 18361da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 18371da177e4SLinus Torvalds * nfs_async_unlink(). 18381da177e4SLinus Torvalds */ 1839858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 18401da177e4SLinus Torvalds { 18411da177e4SLinus Torvalds int error; 18421da177e4SLinus Torvalds 18431da177e4SLinus Torvalds if (!victim->d_inode) 18441da177e4SLinus Torvalds return -ENOENT; 18451da177e4SLinus Torvalds 18461da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 1847cccc6bbaSAl Viro audit_inode_child(victim, dir); 18481da177e4SLinus Torvalds 1849f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 18501da177e4SLinus Torvalds if (error) 18511da177e4SLinus Torvalds return error; 18521da177e4SLinus Torvalds if (IS_APPEND(dir)) 18531da177e4SLinus Torvalds return -EPERM; 18541da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 1855f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 18561da177e4SLinus Torvalds return -EPERM; 18571da177e4SLinus Torvalds if (isdir) { 18581da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 18591da177e4SLinus Torvalds return -ENOTDIR; 18601da177e4SLinus Torvalds if (IS_ROOT(victim)) 18611da177e4SLinus Torvalds return -EBUSY; 18621da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 18631da177e4SLinus Torvalds return -EISDIR; 18641da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 18651da177e4SLinus Torvalds return -ENOENT; 18661da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 18671da177e4SLinus Torvalds return -EBUSY; 18681da177e4SLinus Torvalds return 0; 18691da177e4SLinus Torvalds } 18701da177e4SLinus Torvalds 18711da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 18721da177e4SLinus Torvalds * dir. 18731da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 18741da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 18751da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 18761da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 18771da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 18781da177e4SLinus Torvalds */ 1879a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 18801da177e4SLinus Torvalds { 18811da177e4SLinus Torvalds if (child->d_inode) 18821da177e4SLinus Torvalds return -EEXIST; 18831da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 18841da177e4SLinus Torvalds return -ENOENT; 1885f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds 18881da177e4SLinus Torvalds /* 18891da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 18901da177e4SLinus Torvalds */ 18911da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 18921da177e4SLinus Torvalds { 18931da177e4SLinus Torvalds struct dentry *p; 18941da177e4SLinus Torvalds 18951da177e4SLinus Torvalds if (p1 == p2) { 1896f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 18971da177e4SLinus Torvalds return NULL; 18981da177e4SLinus Torvalds } 18991da177e4SLinus Torvalds 1900a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19011da177e4SLinus Torvalds 1902e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 1903e2761a11SOGAWA Hirofumi if (p) { 1904f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1905f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 19061da177e4SLinus Torvalds return p; 19071da177e4SLinus Torvalds } 19081da177e4SLinus Torvalds 1909e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 1910e2761a11SOGAWA Hirofumi if (p) { 1911f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1912f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19131da177e4SLinus Torvalds return p; 19141da177e4SLinus Torvalds } 19151da177e4SLinus Torvalds 1916f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1917f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19181da177e4SLinus Torvalds return NULL; 19191da177e4SLinus Torvalds } 19201da177e4SLinus Torvalds 19211da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 19221da177e4SLinus Torvalds { 19231b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 19241da177e4SLinus Torvalds if (p1 != p2) { 19251b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 1926a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19271da177e4SLinus Torvalds } 19281da177e4SLinus Torvalds } 19291da177e4SLinus Torvalds 19301da177e4SLinus Torvalds int vfs_create(struct inode *dir, struct dentry *dentry, int mode, 19311da177e4SLinus Torvalds struct nameidata *nd) 19321da177e4SLinus Torvalds { 1933a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 19341da177e4SLinus Torvalds 19351da177e4SLinus Torvalds if (error) 19361da177e4SLinus Torvalds return error; 19371da177e4SLinus Torvalds 1938acfa4380SAl Viro if (!dir->i_op->create) 19391da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 19401da177e4SLinus Torvalds mode &= S_IALLUGO; 19411da177e4SLinus Torvalds mode |= S_IFREG; 19421da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 19431da177e4SLinus Torvalds if (error) 19441da177e4SLinus Torvalds return error; 19451da177e4SLinus Torvalds error = dir->i_op->create(dir, dentry, mode, nd); 1946a74574aaSStephen Smalley if (!error) 1947f38aa942SAmy Griffis fsnotify_create(dir, dentry); 19481da177e4SLinus Torvalds return error; 19491da177e4SLinus Torvalds } 19501da177e4SLinus Torvalds 195173d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 19521da177e4SLinus Torvalds { 19533fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 19541da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 19551da177e4SLinus Torvalds int error; 19561da177e4SLinus Torvalds 19571da177e4SLinus Torvalds if (!inode) 19581da177e4SLinus Torvalds return -ENOENT; 19591da177e4SLinus Torvalds 1960c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 1961c8fe8f30SChristoph Hellwig case S_IFLNK: 19621da177e4SLinus Torvalds return -ELOOP; 1963c8fe8f30SChristoph Hellwig case S_IFDIR: 1964c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 19651da177e4SLinus Torvalds return -EISDIR; 1966c8fe8f30SChristoph Hellwig break; 1967c8fe8f30SChristoph Hellwig case S_IFBLK: 1968c8fe8f30SChristoph Hellwig case S_IFCHR: 19693fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 19701da177e4SLinus Torvalds return -EACCES; 1971c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 1972c8fe8f30SChristoph Hellwig case S_IFIFO: 1973c8fe8f30SChristoph Hellwig case S_IFSOCK: 19741da177e4SLinus Torvalds flag &= ~O_TRUNC; 1975c8fe8f30SChristoph Hellwig break; 19764a3fd211SDave Hansen } 1977b41572e9SDave Hansen 19783fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 1979b41572e9SDave Hansen if (error) 1980b41572e9SDave Hansen return error; 19816146f0d5SMimi Zohar 19821da177e4SLinus Torvalds /* 19831da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 19841da177e4SLinus Torvalds */ 19851da177e4SLinus Torvalds if (IS_APPEND(inode)) { 19868737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 19877715b521SAl Viro return -EPERM; 19881da177e4SLinus Torvalds if (flag & O_TRUNC) 19897715b521SAl Viro return -EPERM; 19901da177e4SLinus Torvalds } 19911da177e4SLinus Torvalds 19921da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 19937715b521SAl Viro if (flag & O_NOATIME && !is_owner_or_cap(inode)) 19947715b521SAl Viro return -EPERM; 19951da177e4SLinus Torvalds 19961da177e4SLinus Torvalds /* 19971da177e4SLinus Torvalds * Ensure there are no outstanding leases on the file. 19981da177e4SLinus Torvalds */ 1999b65a9cfcSAl Viro return break_lease(inode, flag); 20007715b521SAl Viro } 20017715b521SAl Viro 2002e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 20037715b521SAl Viro { 2004e1181ee6SJeff Layton struct path *path = &filp->f_path; 20057715b521SAl Viro struct inode *inode = path->dentry->d_inode; 20067715b521SAl Viro int error = get_write_access(inode); 20071da177e4SLinus Torvalds if (error) 20087715b521SAl Viro return error; 20091da177e4SLinus Torvalds /* 20101da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 20111da177e4SLinus Torvalds */ 20121da177e4SLinus Torvalds error = locks_verify_locked(inode); 2013be6d3e56SKentaro Takeda if (!error) 2014ea0d3ab2STetsuo Handa error = security_path_truncate(path); 20151da177e4SLinus Torvalds if (!error) { 20167715b521SAl Viro error = do_truncate(path->dentry, 0, 2017d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2018e1181ee6SJeff Layton filp); 20191da177e4SLinus Torvalds } 20201da177e4SLinus Torvalds put_write_access(inode); 2021acd0c935SMimi Zohar return error; 20221da177e4SLinus Torvalds } 20231da177e4SLinus Torvalds 2024d57999e1SDave Hansen /* 2025d57999e1SDave Hansen * Note that while the flag value (low two bits) for sys_open means: 2026d57999e1SDave Hansen * 00 - read-only 2027d57999e1SDave Hansen * 01 - write-only 2028d57999e1SDave Hansen * 10 - read-write 2029d57999e1SDave Hansen * 11 - special 2030d57999e1SDave Hansen * it is changed into 2031d57999e1SDave Hansen * 00 - no permissions needed 2032d57999e1SDave Hansen * 01 - read-permission 2033d57999e1SDave Hansen * 10 - write-permission 2034d57999e1SDave Hansen * 11 - read-write 2035d57999e1SDave Hansen * for the internal routines (ie open_namei()/follow_link() etc) 2036d57999e1SDave Hansen * This is more logical, and also allows the 00 "no perm needed" 2037d57999e1SDave Hansen * to be used for symlinks (where the permissions are checked 2038d57999e1SDave Hansen * later). 2039d57999e1SDave Hansen * 2040d57999e1SDave Hansen */ 2041d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2042d57999e1SDave Hansen { 2043d57999e1SDave Hansen if ((flag+1) & O_ACCMODE) 2044d57999e1SDave Hansen flag++; 2045d57999e1SDave Hansen return flag; 2046d57999e1SDave Hansen } 2047d57999e1SDave Hansen 204831e6b01fSNick Piggin /* 2049fe2d35ffSAl Viro * Handle the last step of open() 205031e6b01fSNick Piggin */ 2051fb1cc555SAl Viro static struct file *do_last(struct nameidata *nd, struct path *path, 2052c3e380b0SAl Viro const struct open_flags *op, const char *pathname) 2053fb1cc555SAl Viro { 2054a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 20556c0d46c4SAl Viro struct dentry *dentry; 2056ca344a89SAl Viro int open_flag = op->open_flag; 20576c0d46c4SAl Viro int will_truncate = open_flag & O_TRUNC; 2058ca344a89SAl Viro int want_write = 0; 2059ca344a89SAl Viro int skip_perm = 0; 2060fb1cc555SAl Viro struct file *filp; 2061fe2d35ffSAl Viro struct inode *inode; 206216c2cd71SAl Viro int error; 2063fb1cc555SAl Viro 2064c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2065c3e380b0SAl Viro nd->flags |= op->intent; 2066c3e380b0SAl Viro 20671f36f774SAl Viro switch (nd->last_type) { 20681f36f774SAl Viro case LAST_DOTDOT: 2069176306f5SNeil Brown case LAST_DOT: 2070fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2071fe2d35ffSAl Viro if (error) 2072fe2d35ffSAl Viro return ERR_PTR(error); 20731f36f774SAl Viro /* fallthrough */ 20741f36f774SAl Viro case LAST_ROOT: 2075fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2076fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2077fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2078fe2d35ffSAl Viro } 207916c2cd71SAl Viro error = handle_reval_path(nd); 208016c2cd71SAl Viro if (error) 208116c2cd71SAl Viro goto exit; 2082fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2083ca344a89SAl Viro if (open_flag & O_CREAT) { 208416c2cd71SAl Viro error = -EISDIR; 20851f36f774SAl Viro goto exit; 2086fe2d35ffSAl Viro } 2087fe2d35ffSAl Viro goto ok; 20881f36f774SAl Viro case LAST_BIND: 2089fe2d35ffSAl Viro /* can't be RCU mode here */ 209016c2cd71SAl Viro error = handle_reval_path(nd); 209116c2cd71SAl Viro if (error) 209216c2cd71SAl Viro goto exit; 20931f36f774SAl Viro audit_inode(pathname, dir); 20941f36f774SAl Viro goto ok; 20951f36f774SAl Viro } 2096a2c36b45SAl Viro 2097ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2098fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2099fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2100fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2101fe2d35ffSAl Viro error = do_lookup(nd, &nd->last, path, &inode); 2102fe2d35ffSAl Viro if (error) { 2103fe2d35ffSAl Viro terminate_walk(nd); 2104fe2d35ffSAl Viro return ERR_PTR(error); 2105fe2d35ffSAl Viro } 2106fe2d35ffSAl Viro if (!inode) { 2107fe2d35ffSAl Viro path_to_nameidata(path, nd); 2108fe2d35ffSAl Viro terminate_walk(nd); 2109fe2d35ffSAl Viro return ERR_PTR(-ENOENT); 2110fe2d35ffSAl Viro } 2111fe2d35ffSAl Viro if (unlikely(inode->i_op->follow_link)) { 2112fe2d35ffSAl Viro /* We drop rcu-walk here */ 2113fe2d35ffSAl Viro if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) 2114fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2115fe2d35ffSAl Viro return NULL; 2116fe2d35ffSAl Viro } 2117fe2d35ffSAl Viro path_to_nameidata(path, nd); 2118fe2d35ffSAl Viro nd->inode = inode; 2119fe2d35ffSAl Viro /* sayonara */ 2120fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2121fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2122fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2123fe2d35ffSAl Viro } 2124fe2d35ffSAl Viro 2125fe2d35ffSAl Viro error = -ENOTDIR; 2126fe2d35ffSAl Viro if (nd->flags & LOOKUP_DIRECTORY) { 2127fe2d35ffSAl Viro if (!inode->i_op->lookup) 2128fe2d35ffSAl Viro goto exit; 2129fe2d35ffSAl Viro } 2130fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2131fe2d35ffSAl Viro goto ok; 2132fe2d35ffSAl Viro } 2133fe2d35ffSAl Viro 2134fe2d35ffSAl Viro /* create side of things */ 2135fe2d35ffSAl Viro 2136fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2137fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2138fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2139fe2d35ffSAl Viro } 2140fe2d35ffSAl Viro 2141fe2d35ffSAl Viro audit_inode(pathname, dir); 214216c2cd71SAl Viro error = -EISDIR; 21431f36f774SAl Viro /* trailing slashes? */ 214431e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 21451f36f774SAl Viro goto exit; 21461f36f774SAl Viro 2147a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 2148a1e28038SAl Viro 21496c0d46c4SAl Viro dentry = lookup_hash(nd); 21506c0d46c4SAl Viro error = PTR_ERR(dentry); 21516c0d46c4SAl Viro if (IS_ERR(dentry)) { 2152fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2153fb1cc555SAl Viro goto exit; 2154fb1cc555SAl Viro } 2155fb1cc555SAl Viro 21566c0d46c4SAl Viro path->dentry = dentry; 21576c0d46c4SAl Viro path->mnt = nd->path.mnt; 21586c0d46c4SAl Viro 2159fb1cc555SAl Viro /* Negative dentry, just create the file */ 21606c0d46c4SAl Viro if (!dentry->d_inode) { 21616c0d46c4SAl Viro int mode = op->mode; 21626c0d46c4SAl Viro if (!IS_POSIXACL(dir->d_inode)) 21636c0d46c4SAl Viro mode &= ~current_umask(); 2164fb1cc555SAl Viro /* 2165fb1cc555SAl Viro * This write is needed to ensure that a 21666c0d46c4SAl Viro * rw->ro transition does not occur between 2167fb1cc555SAl Viro * the time when the file is created and when 2168fb1cc555SAl Viro * a permanent write count is taken through 2169fb1cc555SAl Viro * the 'struct file' in nameidata_to_filp(). 2170fb1cc555SAl Viro */ 2171fb1cc555SAl Viro error = mnt_want_write(nd->path.mnt); 2172fb1cc555SAl Viro if (error) 2173fb1cc555SAl Viro goto exit_mutex_unlock; 2174ca344a89SAl Viro want_write = 1; 21759b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2176ca344a89SAl Viro open_flag &= ~O_TRUNC; 21776c0d46c4SAl Viro will_truncate = 0; 2178ca344a89SAl Viro skip_perm = 1; 21796c0d46c4SAl Viro error = security_path_mknod(&nd->path, dentry, mode, 0); 21806c0d46c4SAl Viro if (error) 21816c0d46c4SAl Viro goto exit_mutex_unlock; 21826c0d46c4SAl Viro error = vfs_create(dir->d_inode, dentry, mode, nd); 21836c0d46c4SAl Viro if (error) 21846c0d46c4SAl Viro goto exit_mutex_unlock; 21856c0d46c4SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 21866c0d46c4SAl Viro dput(nd->path.dentry); 21876c0d46c4SAl Viro nd->path.dentry = dentry; 2188ca344a89SAl Viro goto common; 2189fb1cc555SAl Viro } 2190fb1cc555SAl Viro 2191fb1cc555SAl Viro /* 2192fb1cc555SAl Viro * It already exists. 2193fb1cc555SAl Viro */ 2194fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2195fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2196fb1cc555SAl Viro 2197fb1cc555SAl Viro error = -EEXIST; 2198ca344a89SAl Viro if (open_flag & O_EXCL) 2199fb1cc555SAl Viro goto exit_dput; 2200fb1cc555SAl Viro 22019875cf80SDavid Howells error = follow_managed(path, nd->flags); 22029875cf80SDavid Howells if (error < 0) 2203fb1cc555SAl Viro goto exit_dput; 2204fb1cc555SAl Viro 2205fb1cc555SAl Viro error = -ENOENT; 2206fb1cc555SAl Viro if (!path->dentry->d_inode) 2207fb1cc555SAl Viro goto exit_dput; 22089e67f361SAl Viro 22099e67f361SAl Viro if (path->dentry->d_inode->i_op->follow_link) 2210fb1cc555SAl Viro return NULL; 2211fb1cc555SAl Viro 2212fb1cc555SAl Viro path_to_nameidata(path, nd); 221331e6b01fSNick Piggin nd->inode = path->dentry->d_inode; 2214fb1cc555SAl Viro error = -EISDIR; 221531e6b01fSNick Piggin if (S_ISDIR(nd->inode->i_mode)) 2216fb1cc555SAl Viro goto exit; 221767ee3ad2SAl Viro ok: 22186c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 22196c0d46c4SAl Viro will_truncate = 0; 22206c0d46c4SAl Viro 22210f9d1a10SAl Viro if (will_truncate) { 22220f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 22230f9d1a10SAl Viro if (error) 22240f9d1a10SAl Viro goto exit; 2225ca344a89SAl Viro want_write = 1; 22260f9d1a10SAl Viro } 2227ca344a89SAl Viro common: 2228ca344a89SAl Viro error = may_open(&nd->path, skip_perm ? 0 : op->acc_mode, open_flag); 2229ca344a89SAl Viro if (error) 22300f9d1a10SAl Viro goto exit; 22310f9d1a10SAl Viro filp = nameidata_to_filp(nd); 22320f9d1a10SAl Viro if (!IS_ERR(filp)) { 22330f9d1a10SAl Viro error = ima_file_check(filp, op->acc_mode); 22340f9d1a10SAl Viro if (error) { 22350f9d1a10SAl Viro fput(filp); 22360f9d1a10SAl Viro filp = ERR_PTR(error); 22370f9d1a10SAl Viro } 22380f9d1a10SAl Viro } 22390f9d1a10SAl Viro if (!IS_ERR(filp)) { 22400f9d1a10SAl Viro if (will_truncate) { 22410f9d1a10SAl Viro error = handle_truncate(filp); 22420f9d1a10SAl Viro if (error) { 22430f9d1a10SAl Viro fput(filp); 22440f9d1a10SAl Viro filp = ERR_PTR(error); 22450f9d1a10SAl Viro } 22460f9d1a10SAl Viro } 22470f9d1a10SAl Viro } 2248ca344a89SAl Viro out: 2249ca344a89SAl Viro if (want_write) 22500f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 22510f9d1a10SAl Viro path_put(&nd->path); 2252fb1cc555SAl Viro return filp; 2253fb1cc555SAl Viro 2254fb1cc555SAl Viro exit_mutex_unlock: 2255fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2256fb1cc555SAl Viro exit_dput: 2257fb1cc555SAl Viro path_put_conditional(path, nd); 2258fb1cc555SAl Viro exit: 2259ca344a89SAl Viro filp = ERR_PTR(error); 2260ca344a89SAl Viro goto out; 2261fb1cc555SAl Viro } 2262fb1cc555SAl Viro 226313aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 226473d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 22651da177e4SLinus Torvalds { 2266fe2d35ffSAl Viro struct file *base = NULL; 22674a3fd211SDave Hansen struct file *filp; 22689850c056SAl Viro struct path path; 22691da177e4SLinus Torvalds int count = 0; 227013aab428SAl Viro int error; 227131e6b01fSNick Piggin 227231e6b01fSNick Piggin filp = get_empty_filp(); 227331e6b01fSNick Piggin if (!filp) 227431e6b01fSNick Piggin return ERR_PTR(-ENFILE); 227531e6b01fSNick Piggin 227647c805dcSAl Viro filp->f_flags = op->open_flag; 227773d049a4SAl Viro nd->intent.open.file = filp; 227873d049a4SAl Viro nd->intent.open.flags = open_to_namei_flags(op->open_flag); 227973d049a4SAl Viro nd->intent.open.create_mode = op->mode; 228031e6b01fSNick Piggin 228173d049a4SAl Viro error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); 228231e6b01fSNick Piggin if (unlikely(error)) 228313aab428SAl Viro goto out_filp; 228431e6b01fSNick Piggin 2285fe2d35ffSAl Viro current->total_link_count = 0; 228673d049a4SAl Viro error = link_path_walk(pathname, nd); 228731e6b01fSNick Piggin if (unlikely(error)) 228831e6b01fSNick Piggin goto out_filp; 22891da177e4SLinus Torvalds 229073d049a4SAl Viro filp = do_last(nd, &path, op, pathname); 2291806b681cSAl Viro while (unlikely(!filp)) { /* trailing symlink */ 22927b9337aaSNick Piggin struct path link = path; 22937b9337aaSNick Piggin struct inode *linki = link.dentry->d_inode; 2294def4af30SAl Viro void *cookie; 229573d049a4SAl Viro if (!(nd->flags & LOOKUP_FOLLOW) || count++ == 32) { 229673d049a4SAl Viro path_put_conditional(&path, nd); 229773d049a4SAl Viro path_put(&nd->path); 229840b39136SAl Viro filp = ERR_PTR(-ELOOP); 229940b39136SAl Viro break; 230040b39136SAl Viro } 2301806b681cSAl Viro /* 2302806b681cSAl Viro * This is subtle. Instead of calling do_follow_link() we do 2303806b681cSAl Viro * the thing by hands. The reason is that this way we have zero 2304806b681cSAl Viro * link_count and path_walk() (called from ->follow_link) 2305806b681cSAl Viro * honoring LOOKUP_PARENT. After that we have the parent and 2306806b681cSAl Viro * last component, i.e. we are in the same situation as after 2307806b681cSAl Viro * the first path_walk(). Well, almost - if the last component 2308806b681cSAl Viro * is normal we get its copy stored in nd->last.name and we will 2309806b681cSAl Viro * have to putname() it when we are done. Procfs-like symlinks 2310806b681cSAl Viro * just set LAST_BIND. 2311806b681cSAl Viro */ 231273d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 231373d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 231473d049a4SAl Viro error = __do_follow_link(&link, nd, &cookie); 2315c3e380b0SAl Viro if (unlikely(error)) 2316f1afe9efSAl Viro filp = ERR_PTR(error); 2317c3e380b0SAl Viro else 231873d049a4SAl Viro filp = do_last(nd, &path, op, pathname); 2319f1afe9efSAl Viro if (!IS_ERR(cookie) && linki->i_op->put_link) 232073d049a4SAl Viro linki->i_op->put_link(link.dentry, nd, cookie); 23217b9337aaSNick Piggin path_put(&link); 2322806b681cSAl Viro } 232310fa8e62SAl Viro out: 232473d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 232573d049a4SAl Viro path_put(&nd->root); 2326fe2d35ffSAl Viro if (base) 2327fe2d35ffSAl Viro fput(base); 232873d049a4SAl Viro release_open_intent(nd); 232910fa8e62SAl Viro return filp; 23301da177e4SLinus Torvalds 233131e6b01fSNick Piggin out_filp: 233210fa8e62SAl Viro filp = ERR_PTR(error); 233310fa8e62SAl Viro goto out; 2334de459215SKirill Korotaev } 23351da177e4SLinus Torvalds 233613aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 233713aab428SAl Viro const struct open_flags *op, int flags) 233813aab428SAl Viro { 233973d049a4SAl Viro struct nameidata nd; 234013aab428SAl Viro struct file *filp; 234113aab428SAl Viro 234273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 234313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 234473d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 234513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 234673d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 234713aab428SAl Viro return filp; 234813aab428SAl Viro } 234913aab428SAl Viro 235073d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 235173d049a4SAl Viro const char *name, const struct open_flags *op, int flags) 235273d049a4SAl Viro { 235373d049a4SAl Viro struct nameidata nd; 235473d049a4SAl Viro struct file *file; 235573d049a4SAl Viro 235673d049a4SAl Viro nd.root.mnt = mnt; 235773d049a4SAl Viro nd.root.dentry = dentry; 235873d049a4SAl Viro 235973d049a4SAl Viro flags |= LOOKUP_ROOT; 236073d049a4SAl Viro 236173d049a4SAl Viro if (dentry->d_inode->i_op->follow_link) 236273d049a4SAl Viro return ERR_PTR(-ELOOP); 236373d049a4SAl Viro 236473d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); 236573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 236673d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags); 236773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 236873d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); 236973d049a4SAl Viro return file; 237073d049a4SAl Viro } 237173d049a4SAl Viro 23721da177e4SLinus Torvalds /** 23731da177e4SLinus Torvalds * lookup_create - lookup a dentry, creating it if it doesn't exist 23741da177e4SLinus Torvalds * @nd: nameidata info 23751da177e4SLinus Torvalds * @is_dir: directory flag 23761da177e4SLinus Torvalds * 23771da177e4SLinus Torvalds * Simple function to lookup and return a dentry and create it 23781da177e4SLinus Torvalds * if it doesn't exist. Is SMP-safe. 2379c663e5d8SChristoph Hellwig * 23804ac91378SJan Blunck * Returns with nd->path.dentry->d_inode->i_mutex locked. 23811da177e4SLinus Torvalds */ 23821da177e4SLinus Torvalds struct dentry *lookup_create(struct nameidata *nd, int is_dir) 23831da177e4SLinus Torvalds { 2384c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 23851da177e4SLinus Torvalds 23864ac91378SJan Blunck mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2387c663e5d8SChristoph Hellwig /* 2388c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2389c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2390c663e5d8SChristoph Hellwig */ 23911da177e4SLinus Torvalds if (nd->last_type != LAST_NORM) 23921da177e4SLinus Torvalds goto fail; 23931da177e4SLinus Torvalds nd->flags &= ~LOOKUP_PARENT; 23943516586aSAl Viro nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2395a634904aSASANO Masahiro nd->intent.open.flags = O_EXCL; 2396c663e5d8SChristoph Hellwig 2397c663e5d8SChristoph Hellwig /* 2398c663e5d8SChristoph Hellwig * Do the final lookup. 2399c663e5d8SChristoph Hellwig */ 240049705b77SChristoph Hellwig dentry = lookup_hash(nd); 24011da177e4SLinus Torvalds if (IS_ERR(dentry)) 24021da177e4SLinus Torvalds goto fail; 2403c663e5d8SChristoph Hellwig 2404e9baf6e5SAl Viro if (dentry->d_inode) 2405e9baf6e5SAl Viro goto eexist; 2406c663e5d8SChristoph Hellwig /* 2407c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 2408c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 2409c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 2410c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 2411c663e5d8SChristoph Hellwig */ 2412e9baf6e5SAl Viro if (unlikely(!is_dir && nd->last.name[nd->last.len])) { 24131da177e4SLinus Torvalds dput(dentry); 24141da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 2415e9baf6e5SAl Viro } 2416e9baf6e5SAl Viro return dentry; 2417e9baf6e5SAl Viro eexist: 2418e9baf6e5SAl Viro dput(dentry); 2419e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 24201da177e4SLinus Torvalds fail: 24211da177e4SLinus Torvalds return dentry; 24221da177e4SLinus Torvalds } 2423f81a0bffSChristoph Hellwig EXPORT_SYMBOL_GPL(lookup_create); 24241da177e4SLinus Torvalds 24251da177e4SLinus Torvalds int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 24261da177e4SLinus Torvalds { 2427a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 24281da177e4SLinus Torvalds 24291da177e4SLinus Torvalds if (error) 24301da177e4SLinus Torvalds return error; 24311da177e4SLinus Torvalds 24321da177e4SLinus Torvalds if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 24331da177e4SLinus Torvalds return -EPERM; 24341da177e4SLinus Torvalds 2435acfa4380SAl Viro if (!dir->i_op->mknod) 24361da177e4SLinus Torvalds return -EPERM; 24371da177e4SLinus Torvalds 243808ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 243908ce5f16SSerge E. Hallyn if (error) 244008ce5f16SSerge E. Hallyn return error; 244108ce5f16SSerge E. Hallyn 24421da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 24431da177e4SLinus Torvalds if (error) 24441da177e4SLinus Torvalds return error; 24451da177e4SLinus Torvalds 24461da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 2447a74574aaSStephen Smalley if (!error) 2448f38aa942SAmy Griffis fsnotify_create(dir, dentry); 24491da177e4SLinus Torvalds return error; 24501da177e4SLinus Torvalds } 24511da177e4SLinus Torvalds 2452463c3197SDave Hansen static int may_mknod(mode_t mode) 2453463c3197SDave Hansen { 2454463c3197SDave Hansen switch (mode & S_IFMT) { 2455463c3197SDave Hansen case S_IFREG: 2456463c3197SDave Hansen case S_IFCHR: 2457463c3197SDave Hansen case S_IFBLK: 2458463c3197SDave Hansen case S_IFIFO: 2459463c3197SDave Hansen case S_IFSOCK: 2460463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 2461463c3197SDave Hansen return 0; 2462463c3197SDave Hansen case S_IFDIR: 2463463c3197SDave Hansen return -EPERM; 2464463c3197SDave Hansen default: 2465463c3197SDave Hansen return -EINVAL; 2466463c3197SDave Hansen } 2467463c3197SDave Hansen } 2468463c3197SDave Hansen 24692e4d0924SHeiko Carstens SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, 24702e4d0924SHeiko Carstens unsigned, dev) 24711da177e4SLinus Torvalds { 24722ad94ae6SAl Viro int error; 24731da177e4SLinus Torvalds char *tmp; 24741da177e4SLinus Torvalds struct dentry *dentry; 24751da177e4SLinus Torvalds struct nameidata nd; 24761da177e4SLinus Torvalds 24771da177e4SLinus Torvalds if (S_ISDIR(mode)) 24781da177e4SLinus Torvalds return -EPERM; 24791da177e4SLinus Torvalds 24802ad94ae6SAl Viro error = user_path_parent(dfd, filename, &nd, &tmp); 24811da177e4SLinus Torvalds if (error) 24822ad94ae6SAl Viro return error; 24832ad94ae6SAl Viro 24841da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 2485463c3197SDave Hansen if (IS_ERR(dentry)) { 24861da177e4SLinus Torvalds error = PTR_ERR(dentry); 2487463c3197SDave Hansen goto out_unlock; 2488463c3197SDave Hansen } 24894ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2490ce3b0f8dSAl Viro mode &= ~current_umask(); 2491463c3197SDave Hansen error = may_mknod(mode); 2492463c3197SDave Hansen if (error) 2493463c3197SDave Hansen goto out_dput; 2494463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2495463c3197SDave Hansen if (error) 2496463c3197SDave Hansen goto out_dput; 2497be6d3e56SKentaro Takeda error = security_path_mknod(&nd.path, dentry, mode, dev); 2498be6d3e56SKentaro Takeda if (error) 2499be6d3e56SKentaro Takeda goto out_drop_write; 25001da177e4SLinus Torvalds switch (mode & S_IFMT) { 25011da177e4SLinus Torvalds case 0: case S_IFREG: 25024ac91378SJan Blunck error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); 25031da177e4SLinus Torvalds break; 25041da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 25054ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, 25061da177e4SLinus Torvalds new_decode_dev(dev)); 25071da177e4SLinus Torvalds break; 25081da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 25094ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); 25101da177e4SLinus Torvalds break; 25111da177e4SLinus Torvalds } 2512be6d3e56SKentaro Takeda out_drop_write: 2513463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2514463c3197SDave Hansen out_dput: 25151da177e4SLinus Torvalds dput(dentry); 2516463c3197SDave Hansen out_unlock: 25174ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 25181d957f9bSJan Blunck path_put(&nd.path); 25191da177e4SLinus Torvalds putname(tmp); 25201da177e4SLinus Torvalds 25211da177e4SLinus Torvalds return error; 25221da177e4SLinus Torvalds } 25231da177e4SLinus Torvalds 25243480b257SHeiko Carstens SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) 25255590ff0dSUlrich Drepper { 25265590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 25275590ff0dSUlrich Drepper } 25285590ff0dSUlrich Drepper 25291da177e4SLinus Torvalds int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 25301da177e4SLinus Torvalds { 2531a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25321da177e4SLinus Torvalds 25331da177e4SLinus Torvalds if (error) 25341da177e4SLinus Torvalds return error; 25351da177e4SLinus Torvalds 2536acfa4380SAl Viro if (!dir->i_op->mkdir) 25371da177e4SLinus Torvalds return -EPERM; 25381da177e4SLinus Torvalds 25391da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 25401da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 25411da177e4SLinus Torvalds if (error) 25421da177e4SLinus Torvalds return error; 25431da177e4SLinus Torvalds 25441da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 2545a74574aaSStephen Smalley if (!error) 2546f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 25471da177e4SLinus Torvalds return error; 25481da177e4SLinus Torvalds } 25491da177e4SLinus Torvalds 25502e4d0924SHeiko Carstens SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) 25511da177e4SLinus Torvalds { 25521da177e4SLinus Torvalds int error = 0; 25531da177e4SLinus Torvalds char * tmp; 25546902d925SDave Hansen struct dentry *dentry; 25556902d925SDave Hansen struct nameidata nd; 25561da177e4SLinus Torvalds 25572ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &tmp); 25582ad94ae6SAl Viro if (error) 25596902d925SDave Hansen goto out_err; 25601da177e4SLinus Torvalds 25611da177e4SLinus Torvalds dentry = lookup_create(&nd, 1); 25621da177e4SLinus Torvalds error = PTR_ERR(dentry); 25636902d925SDave Hansen if (IS_ERR(dentry)) 25646902d925SDave Hansen goto out_unlock; 25656902d925SDave Hansen 25664ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2567ce3b0f8dSAl Viro mode &= ~current_umask(); 2568463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2569463c3197SDave Hansen if (error) 2570463c3197SDave Hansen goto out_dput; 2571be6d3e56SKentaro Takeda error = security_path_mkdir(&nd.path, dentry, mode); 2572be6d3e56SKentaro Takeda if (error) 2573be6d3e56SKentaro Takeda goto out_drop_write; 25744ac91378SJan Blunck error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); 2575be6d3e56SKentaro Takeda out_drop_write: 2576463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2577463c3197SDave Hansen out_dput: 25781da177e4SLinus Torvalds dput(dentry); 25796902d925SDave Hansen out_unlock: 25804ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 25811d957f9bSJan Blunck path_put(&nd.path); 25821da177e4SLinus Torvalds putname(tmp); 25836902d925SDave Hansen out_err: 25841da177e4SLinus Torvalds return error; 25851da177e4SLinus Torvalds } 25861da177e4SLinus Torvalds 25873cdad428SHeiko Carstens SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) 25885590ff0dSUlrich Drepper { 25895590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 25905590ff0dSUlrich Drepper } 25915590ff0dSUlrich Drepper 25921da177e4SLinus Torvalds /* 25931da177e4SLinus Torvalds * We try to drop the dentry early: we should have 25941da177e4SLinus Torvalds * a usage count of 2 if we're the only user of this 25951da177e4SLinus Torvalds * dentry, and if that is true (possibly after pruning 25961da177e4SLinus Torvalds * the dcache), then we drop the dentry now. 25971da177e4SLinus Torvalds * 25981da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 25991da177e4SLinus Torvalds * do a 26001da177e4SLinus Torvalds * 26011da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 26021da177e4SLinus Torvalds * return -EBUSY; 26031da177e4SLinus Torvalds * 26041da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 26051da177e4SLinus Torvalds * that is still in use by something else.. 26061da177e4SLinus Torvalds */ 26071da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 26081da177e4SLinus Torvalds { 26091da177e4SLinus Torvalds dget(dentry); 26101da177e4SLinus Torvalds shrink_dcache_parent(dentry); 26111da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 2612b7ab39f6SNick Piggin if (dentry->d_count == 2) 26131da177e4SLinus Torvalds __d_drop(dentry); 26141da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 26151da177e4SLinus Torvalds } 26161da177e4SLinus Torvalds 26171da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 26181da177e4SLinus Torvalds { 26191da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 26201da177e4SLinus Torvalds 26211da177e4SLinus Torvalds if (error) 26221da177e4SLinus Torvalds return error; 26231da177e4SLinus Torvalds 2624acfa4380SAl Viro if (!dir->i_op->rmdir) 26251da177e4SLinus Torvalds return -EPERM; 26261da177e4SLinus Torvalds 26271b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 26281da177e4SLinus Torvalds dentry_unhash(dentry); 26291da177e4SLinus Torvalds if (d_mountpoint(dentry)) 26301da177e4SLinus Torvalds error = -EBUSY; 26311da177e4SLinus Torvalds else { 26321da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 26331da177e4SLinus Torvalds if (!error) { 26341da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 2635d83c49f3SAl Viro if (!error) { 26361da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 2637d83c49f3SAl Viro dont_mount(dentry); 2638d83c49f3SAl Viro } 26391da177e4SLinus Torvalds } 26401da177e4SLinus Torvalds } 26411b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 26421da177e4SLinus Torvalds if (!error) { 26431da177e4SLinus Torvalds d_delete(dentry); 26441da177e4SLinus Torvalds } 26451da177e4SLinus Torvalds dput(dentry); 26461da177e4SLinus Torvalds 26471da177e4SLinus Torvalds return error; 26481da177e4SLinus Torvalds } 26491da177e4SLinus Torvalds 26505590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 26511da177e4SLinus Torvalds { 26521da177e4SLinus Torvalds int error = 0; 26531da177e4SLinus Torvalds char * name; 26541da177e4SLinus Torvalds struct dentry *dentry; 26551da177e4SLinus Torvalds struct nameidata nd; 26561da177e4SLinus Torvalds 26572ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 26581da177e4SLinus Torvalds if (error) 26592ad94ae6SAl Viro return error; 26601da177e4SLinus Torvalds 26611da177e4SLinus Torvalds switch(nd.last_type) { 26621da177e4SLinus Torvalds case LAST_DOTDOT: 26631da177e4SLinus Torvalds error = -ENOTEMPTY; 26641da177e4SLinus Torvalds goto exit1; 26651da177e4SLinus Torvalds case LAST_DOT: 26661da177e4SLinus Torvalds error = -EINVAL; 26671da177e4SLinus Torvalds goto exit1; 26681da177e4SLinus Torvalds case LAST_ROOT: 26691da177e4SLinus Torvalds error = -EBUSY; 26701da177e4SLinus Torvalds goto exit1; 26711da177e4SLinus Torvalds } 26720612d9fbSOGAWA Hirofumi 26730612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 26740612d9fbSOGAWA Hirofumi 26754ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 267649705b77SChristoph Hellwig dentry = lookup_hash(&nd); 26771da177e4SLinus Torvalds error = PTR_ERR(dentry); 26786902d925SDave Hansen if (IS_ERR(dentry)) 26796902d925SDave Hansen goto exit2; 26800622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 26810622753bSDave Hansen if (error) 26820622753bSDave Hansen goto exit3; 2683be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 2684be6d3e56SKentaro Takeda if (error) 2685be6d3e56SKentaro Takeda goto exit4; 26864ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 2687be6d3e56SKentaro Takeda exit4: 26880622753bSDave Hansen mnt_drop_write(nd.path.mnt); 26890622753bSDave Hansen exit3: 26901da177e4SLinus Torvalds dput(dentry); 26916902d925SDave Hansen exit2: 26924ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 26931da177e4SLinus Torvalds exit1: 26941d957f9bSJan Blunck path_put(&nd.path); 26951da177e4SLinus Torvalds putname(name); 26961da177e4SLinus Torvalds return error; 26971da177e4SLinus Torvalds } 26981da177e4SLinus Torvalds 26993cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 27005590ff0dSUlrich Drepper { 27015590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 27025590ff0dSUlrich Drepper } 27035590ff0dSUlrich Drepper 27041da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 27051da177e4SLinus Torvalds { 27061da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 27071da177e4SLinus Torvalds 27081da177e4SLinus Torvalds if (error) 27091da177e4SLinus Torvalds return error; 27101da177e4SLinus Torvalds 2711acfa4380SAl Viro if (!dir->i_op->unlink) 27121da177e4SLinus Torvalds return -EPERM; 27131da177e4SLinus Torvalds 27141b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 27151da177e4SLinus Torvalds if (d_mountpoint(dentry)) 27161da177e4SLinus Torvalds error = -EBUSY; 27171da177e4SLinus Torvalds else { 27181da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 2719bec1052eSAl Viro if (!error) { 27201da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 2721bec1052eSAl Viro if (!error) 2722d83c49f3SAl Viro dont_mount(dentry); 2723bec1052eSAl Viro } 27241da177e4SLinus Torvalds } 27251b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 27261da177e4SLinus Torvalds 27271da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 27281da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 2729ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 27301da177e4SLinus Torvalds d_delete(dentry); 27311da177e4SLinus Torvalds } 27320eeca283SRobert Love 27331da177e4SLinus Torvalds return error; 27341da177e4SLinus Torvalds } 27351da177e4SLinus Torvalds 27361da177e4SLinus Torvalds /* 27371da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 27381b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 27391da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 27401da177e4SLinus Torvalds * while waiting on the I/O. 27411da177e4SLinus Torvalds */ 27425590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 27431da177e4SLinus Torvalds { 27442ad94ae6SAl Viro int error; 27451da177e4SLinus Torvalds char *name; 27461da177e4SLinus Torvalds struct dentry *dentry; 27471da177e4SLinus Torvalds struct nameidata nd; 27481da177e4SLinus Torvalds struct inode *inode = NULL; 27491da177e4SLinus Torvalds 27502ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 27511da177e4SLinus Torvalds if (error) 27522ad94ae6SAl Viro return error; 27532ad94ae6SAl Viro 27541da177e4SLinus Torvalds error = -EISDIR; 27551da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 27561da177e4SLinus Torvalds goto exit1; 27570612d9fbSOGAWA Hirofumi 27580612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 27590612d9fbSOGAWA Hirofumi 27604ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 276149705b77SChristoph Hellwig dentry = lookup_hash(&nd); 27621da177e4SLinus Torvalds error = PTR_ERR(dentry); 27631da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 27641da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 27651da177e4SLinus Torvalds if (nd.last.name[nd.last.len]) 27661da177e4SLinus Torvalds goto slashes; 27671da177e4SLinus Torvalds inode = dentry->d_inode; 27681da177e4SLinus Torvalds if (inode) 27697de9c6eeSAl Viro ihold(inode); 27700622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 27710622753bSDave Hansen if (error) 27720622753bSDave Hansen goto exit2; 2773be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 2774be6d3e56SKentaro Takeda if (error) 2775be6d3e56SKentaro Takeda goto exit3; 27764ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 2777be6d3e56SKentaro Takeda exit3: 27780622753bSDave Hansen mnt_drop_write(nd.path.mnt); 27791da177e4SLinus Torvalds exit2: 27801da177e4SLinus Torvalds dput(dentry); 27811da177e4SLinus Torvalds } 27824ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 27831da177e4SLinus Torvalds if (inode) 27841da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 27851da177e4SLinus Torvalds exit1: 27861d957f9bSJan Blunck path_put(&nd.path); 27871da177e4SLinus Torvalds putname(name); 27881da177e4SLinus Torvalds return error; 27891da177e4SLinus Torvalds 27901da177e4SLinus Torvalds slashes: 27911da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 27921da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 27931da177e4SLinus Torvalds goto exit2; 27941da177e4SLinus Torvalds } 27951da177e4SLinus Torvalds 27962e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 27975590ff0dSUlrich Drepper { 27985590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 27995590ff0dSUlrich Drepper return -EINVAL; 28005590ff0dSUlrich Drepper 28015590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 28025590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 28035590ff0dSUlrich Drepper 28045590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 28055590ff0dSUlrich Drepper } 28065590ff0dSUlrich Drepper 28073480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 28085590ff0dSUlrich Drepper { 28095590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 28105590ff0dSUlrich Drepper } 28115590ff0dSUlrich Drepper 2812db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 28131da177e4SLinus Torvalds { 2814a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28151da177e4SLinus Torvalds 28161da177e4SLinus Torvalds if (error) 28171da177e4SLinus Torvalds return error; 28181da177e4SLinus Torvalds 2819acfa4380SAl Viro if (!dir->i_op->symlink) 28201da177e4SLinus Torvalds return -EPERM; 28211da177e4SLinus Torvalds 28221da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 28231da177e4SLinus Torvalds if (error) 28241da177e4SLinus Torvalds return error; 28251da177e4SLinus Torvalds 28261da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 2827a74574aaSStephen Smalley if (!error) 2828f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28291da177e4SLinus Torvalds return error; 28301da177e4SLinus Torvalds } 28311da177e4SLinus Torvalds 28322e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 28332e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 28341da177e4SLinus Torvalds { 28352ad94ae6SAl Viro int error; 28361da177e4SLinus Torvalds char *from; 28371da177e4SLinus Torvalds char *to; 28386902d925SDave Hansen struct dentry *dentry; 28396902d925SDave Hansen struct nameidata nd; 28401da177e4SLinus Torvalds 28411da177e4SLinus Torvalds from = getname(oldname); 28421da177e4SLinus Torvalds if (IS_ERR(from)) 28431da177e4SLinus Torvalds return PTR_ERR(from); 28442ad94ae6SAl Viro 28452ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 28462ad94ae6SAl Viro if (error) 28476902d925SDave Hansen goto out_putname; 28481da177e4SLinus Torvalds 28491da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 28501da177e4SLinus Torvalds error = PTR_ERR(dentry); 28516902d925SDave Hansen if (IS_ERR(dentry)) 28526902d925SDave Hansen goto out_unlock; 28536902d925SDave Hansen 285475c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 285575c3f29dSDave Hansen if (error) 285675c3f29dSDave Hansen goto out_dput; 2857be6d3e56SKentaro Takeda error = security_path_symlink(&nd.path, dentry, from); 2858be6d3e56SKentaro Takeda if (error) 2859be6d3e56SKentaro Takeda goto out_drop_write; 2860db2e747bSMiklos Szeredi error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); 2861be6d3e56SKentaro Takeda out_drop_write: 286275c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 286375c3f29dSDave Hansen out_dput: 28641da177e4SLinus Torvalds dput(dentry); 28656902d925SDave Hansen out_unlock: 28664ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 28671d957f9bSJan Blunck path_put(&nd.path); 28681da177e4SLinus Torvalds putname(to); 28696902d925SDave Hansen out_putname: 28701da177e4SLinus Torvalds putname(from); 28711da177e4SLinus Torvalds return error; 28721da177e4SLinus Torvalds } 28731da177e4SLinus Torvalds 28743480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 28755590ff0dSUlrich Drepper { 28765590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 28775590ff0dSUlrich Drepper } 28785590ff0dSUlrich Drepper 28791da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 28801da177e4SLinus Torvalds { 28811da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 28821da177e4SLinus Torvalds int error; 28831da177e4SLinus Torvalds 28841da177e4SLinus Torvalds if (!inode) 28851da177e4SLinus Torvalds return -ENOENT; 28861da177e4SLinus Torvalds 2887a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 28881da177e4SLinus Torvalds if (error) 28891da177e4SLinus Torvalds return error; 28901da177e4SLinus Torvalds 28911da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 28921da177e4SLinus Torvalds return -EXDEV; 28931da177e4SLinus Torvalds 28941da177e4SLinus Torvalds /* 28951da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 28961da177e4SLinus Torvalds */ 28971da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 28981da177e4SLinus Torvalds return -EPERM; 2899acfa4380SAl Viro if (!dir->i_op->link) 29001da177e4SLinus Torvalds return -EPERM; 29017e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 29021da177e4SLinus Torvalds return -EPERM; 29031da177e4SLinus Torvalds 29041da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 29051da177e4SLinus Torvalds if (error) 29061da177e4SLinus Torvalds return error; 29071da177e4SLinus Torvalds 29087e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 2909aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 2910aae8a97dSAneesh Kumar K.V if (inode->i_nlink == 0) 2911aae8a97dSAneesh Kumar K.V error = -ENOENT; 2912aae8a97dSAneesh Kumar K.V else 29131da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 29147e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 2915e31e14ecSStephen Smalley if (!error) 29167e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 29171da177e4SLinus Torvalds return error; 29181da177e4SLinus Torvalds } 29191da177e4SLinus Torvalds 29201da177e4SLinus Torvalds /* 29211da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 29221da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 29231da177e4SLinus Torvalds * newname. --KAB 29241da177e4SLinus Torvalds * 29251da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 29261da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 29271da177e4SLinus Torvalds * and other special files. --ADM 29281da177e4SLinus Torvalds */ 29292e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 29302e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 29311da177e4SLinus Torvalds { 29321da177e4SLinus Torvalds struct dentry *new_dentry; 29332d8f3038SAl Viro struct nameidata nd; 29342d8f3038SAl Viro struct path old_path; 29351da177e4SLinus Torvalds int error; 29361da177e4SLinus Torvalds char *to; 29371da177e4SLinus Torvalds 293845c9b11aSUlrich Drepper if ((flags & ~AT_SYMLINK_FOLLOW) != 0) 2939c04030e1SUlrich Drepper return -EINVAL; 2940c04030e1SUlrich Drepper 29412d8f3038SAl Viro error = user_path_at(olddfd, oldname, 294245c9b11aSUlrich Drepper flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, 29432d8f3038SAl Viro &old_path); 29441da177e4SLinus Torvalds if (error) 29452ad94ae6SAl Viro return error; 29462ad94ae6SAl Viro 29472ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 29481da177e4SLinus Torvalds if (error) 29491da177e4SLinus Torvalds goto out; 29501da177e4SLinus Torvalds error = -EXDEV; 29512d8f3038SAl Viro if (old_path.mnt != nd.path.mnt) 29521da177e4SLinus Torvalds goto out_release; 29531da177e4SLinus Torvalds new_dentry = lookup_create(&nd, 0); 29541da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 29556902d925SDave Hansen if (IS_ERR(new_dentry)) 29566902d925SDave Hansen goto out_unlock; 295775c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 295875c3f29dSDave Hansen if (error) 295975c3f29dSDave Hansen goto out_dput; 2960be6d3e56SKentaro Takeda error = security_path_link(old_path.dentry, &nd.path, new_dentry); 2961be6d3e56SKentaro Takeda if (error) 2962be6d3e56SKentaro Takeda goto out_drop_write; 29632d8f3038SAl Viro error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); 2964be6d3e56SKentaro Takeda out_drop_write: 296575c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 296675c3f29dSDave Hansen out_dput: 29671da177e4SLinus Torvalds dput(new_dentry); 29686902d925SDave Hansen out_unlock: 29694ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 29701da177e4SLinus Torvalds out_release: 29711d957f9bSJan Blunck path_put(&nd.path); 29722ad94ae6SAl Viro putname(to); 29731da177e4SLinus Torvalds out: 29742d8f3038SAl Viro path_put(&old_path); 29751da177e4SLinus Torvalds 29761da177e4SLinus Torvalds return error; 29771da177e4SLinus Torvalds } 29781da177e4SLinus Torvalds 29793480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 29805590ff0dSUlrich Drepper { 2981c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 29825590ff0dSUlrich Drepper } 29835590ff0dSUlrich Drepper 29841da177e4SLinus Torvalds /* 29851da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 29861da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 29871da177e4SLinus Torvalds * Problems: 29881da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 29891da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 29901da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 2991a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 29921da177e4SLinus Torvalds * story. 29931da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 29941b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 29951da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 29961da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 2997a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 29981da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 29991da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 30001da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3001a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 30021da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 30031da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 30041da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 30051da177e4SLinus Torvalds * d) some filesystems don't support opened-but-unlinked directories, 30061da177e4SLinus Torvalds * either because of layout or because they are not ready to deal with 30071da177e4SLinus Torvalds * all cases correctly. The latter will be fixed (taking this sort of 30081da177e4SLinus Torvalds * stuff into VFS), but the former is not going away. Solution: the same 30091da177e4SLinus Torvalds * trick as in rmdir(). 30101da177e4SLinus Torvalds * e) conversion from fhandle to dentry may come in the wrong moment - when 30111b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 30121da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3013c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 30141da177e4SLinus Torvalds * locking]. 30151da177e4SLinus Torvalds */ 301675c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 30171da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30181da177e4SLinus Torvalds { 30191da177e4SLinus Torvalds int error = 0; 30201da177e4SLinus Torvalds struct inode *target; 30211da177e4SLinus Torvalds 30221da177e4SLinus Torvalds /* 30231da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 30241da177e4SLinus Torvalds * we'll need to flip '..'. 30251da177e4SLinus Torvalds */ 30261da177e4SLinus Torvalds if (new_dir != old_dir) { 3027f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 30281da177e4SLinus Torvalds if (error) 30291da177e4SLinus Torvalds return error; 30301da177e4SLinus Torvalds } 30311da177e4SLinus Torvalds 30321da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30331da177e4SLinus Torvalds if (error) 30341da177e4SLinus Torvalds return error; 30351da177e4SLinus Torvalds 30361da177e4SLinus Torvalds target = new_dentry->d_inode; 3037d83c49f3SAl Viro if (target) 30381b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 30391da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 30401da177e4SLinus Torvalds error = -EBUSY; 3041d83c49f3SAl Viro else { 3042d83c49f3SAl Viro if (target) 3043d83c49f3SAl Viro dentry_unhash(new_dentry); 30441da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 3045d83c49f3SAl Viro } 30461da177e4SLinus Torvalds if (target) { 3047d83c49f3SAl Viro if (!error) { 30481da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3049d83c49f3SAl Viro dont_mount(new_dentry); 3050d83c49f3SAl Viro } 30511b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 30521da177e4SLinus Torvalds if (d_unhashed(new_dentry)) 30531da177e4SLinus Torvalds d_rehash(new_dentry); 30541da177e4SLinus Torvalds dput(new_dentry); 30551da177e4SLinus Torvalds } 3056e31e14ecSStephen Smalley if (!error) 3057349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 30581da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 30591da177e4SLinus Torvalds return error; 30601da177e4SLinus Torvalds } 30611da177e4SLinus Torvalds 306275c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 30631da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30641da177e4SLinus Torvalds { 30651da177e4SLinus Torvalds struct inode *target; 30661da177e4SLinus Torvalds int error; 30671da177e4SLinus Torvalds 30681da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30691da177e4SLinus Torvalds if (error) 30701da177e4SLinus Torvalds return error; 30711da177e4SLinus Torvalds 30721da177e4SLinus Torvalds dget(new_dentry); 30731da177e4SLinus Torvalds target = new_dentry->d_inode; 30741da177e4SLinus Torvalds if (target) 30751b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 30761da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 30771da177e4SLinus Torvalds error = -EBUSY; 30781da177e4SLinus Torvalds else 30791da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 30801da177e4SLinus Torvalds if (!error) { 3081bec1052eSAl Viro if (target) 3082d83c49f3SAl Viro dont_mount(new_dentry); 3083349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 30841da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 30851da177e4SLinus Torvalds } 30861da177e4SLinus Torvalds if (target) 30871b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 30881da177e4SLinus Torvalds dput(new_dentry); 30891da177e4SLinus Torvalds return error; 30901da177e4SLinus Torvalds } 30911da177e4SLinus Torvalds 30921da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 30931da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30941da177e4SLinus Torvalds { 30951da177e4SLinus Torvalds int error; 30961da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 309759b0df21SEric Paris const unsigned char *old_name; 30981da177e4SLinus Torvalds 30991da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 31001da177e4SLinus Torvalds return 0; 31011da177e4SLinus Torvalds 31021da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 31031da177e4SLinus Torvalds if (error) 31041da177e4SLinus Torvalds return error; 31051da177e4SLinus Torvalds 31061da177e4SLinus Torvalds if (!new_dentry->d_inode) 3107a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 31081da177e4SLinus Torvalds else 31091da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 31101da177e4SLinus Torvalds if (error) 31111da177e4SLinus Torvalds return error; 31121da177e4SLinus Torvalds 3113acfa4380SAl Viro if (!old_dir->i_op->rename) 31141da177e4SLinus Torvalds return -EPERM; 31151da177e4SLinus Torvalds 31160eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 31170eeca283SRobert Love 31181da177e4SLinus Torvalds if (is_dir) 31191da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 31201da177e4SLinus Torvalds else 31211da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3122123df294SAl Viro if (!error) 3123123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 31245a190ae6SAl Viro new_dentry->d_inode, old_dentry); 31250eeca283SRobert Love fsnotify_oldname_free(old_name); 31260eeca283SRobert Love 31271da177e4SLinus Torvalds return error; 31281da177e4SLinus Torvalds } 31291da177e4SLinus Torvalds 31302e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 31312e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 31321da177e4SLinus Torvalds { 31331da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 31341da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 31351da177e4SLinus Torvalds struct dentry *trap; 31361da177e4SLinus Torvalds struct nameidata oldnd, newnd; 31372ad94ae6SAl Viro char *from; 31382ad94ae6SAl Viro char *to; 31392ad94ae6SAl Viro int error; 31401da177e4SLinus Torvalds 31412ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 31421da177e4SLinus Torvalds if (error) 31431da177e4SLinus Torvalds goto exit; 31441da177e4SLinus Torvalds 31452ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 31461da177e4SLinus Torvalds if (error) 31471da177e4SLinus Torvalds goto exit1; 31481da177e4SLinus Torvalds 31491da177e4SLinus Torvalds error = -EXDEV; 31504ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 31511da177e4SLinus Torvalds goto exit2; 31521da177e4SLinus Torvalds 31534ac91378SJan Blunck old_dir = oldnd.path.dentry; 31541da177e4SLinus Torvalds error = -EBUSY; 31551da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 31561da177e4SLinus Torvalds goto exit2; 31571da177e4SLinus Torvalds 31584ac91378SJan Blunck new_dir = newnd.path.dentry; 31591da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 31601da177e4SLinus Torvalds goto exit2; 31611da177e4SLinus Torvalds 31620612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 31630612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 31644e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 31650612d9fbSOGAWA Hirofumi 31661da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 31671da177e4SLinus Torvalds 316849705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 31691da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 31701da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 31711da177e4SLinus Torvalds goto exit3; 31721da177e4SLinus Torvalds /* source must exist */ 31731da177e4SLinus Torvalds error = -ENOENT; 31741da177e4SLinus Torvalds if (!old_dentry->d_inode) 31751da177e4SLinus Torvalds goto exit4; 31761da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 31771da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 31781da177e4SLinus Torvalds error = -ENOTDIR; 31791da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 31801da177e4SLinus Torvalds goto exit4; 31811da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 31821da177e4SLinus Torvalds goto exit4; 31831da177e4SLinus Torvalds } 31841da177e4SLinus Torvalds /* source should not be ancestor of target */ 31851da177e4SLinus Torvalds error = -EINVAL; 31861da177e4SLinus Torvalds if (old_dentry == trap) 31871da177e4SLinus Torvalds goto exit4; 318849705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 31891da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 31901da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 31911da177e4SLinus Torvalds goto exit4; 31921da177e4SLinus Torvalds /* target should not be an ancestor of source */ 31931da177e4SLinus Torvalds error = -ENOTEMPTY; 31941da177e4SLinus Torvalds if (new_dentry == trap) 31951da177e4SLinus Torvalds goto exit5; 31961da177e4SLinus Torvalds 31979079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 31989079b1ebSDave Hansen if (error) 31999079b1ebSDave Hansen goto exit5; 3200be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3201be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3202be6d3e56SKentaro Takeda if (error) 3203be6d3e56SKentaro Takeda goto exit6; 32041da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 32051da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 3206be6d3e56SKentaro Takeda exit6: 32079079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 32081da177e4SLinus Torvalds exit5: 32091da177e4SLinus Torvalds dput(new_dentry); 32101da177e4SLinus Torvalds exit4: 32111da177e4SLinus Torvalds dput(old_dentry); 32121da177e4SLinus Torvalds exit3: 32131da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 32141da177e4SLinus Torvalds exit2: 32151d957f9bSJan Blunck path_put(&newnd.path); 32162ad94ae6SAl Viro putname(to); 32171da177e4SLinus Torvalds exit1: 32181d957f9bSJan Blunck path_put(&oldnd.path); 32191da177e4SLinus Torvalds putname(from); 32202ad94ae6SAl Viro exit: 32211da177e4SLinus Torvalds return error; 32221da177e4SLinus Torvalds } 32231da177e4SLinus Torvalds 3224a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 32255590ff0dSUlrich Drepper { 32265590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 32275590ff0dSUlrich Drepper } 32285590ff0dSUlrich Drepper 32291da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 32301da177e4SLinus Torvalds { 32311da177e4SLinus Torvalds int len; 32321da177e4SLinus Torvalds 32331da177e4SLinus Torvalds len = PTR_ERR(link); 32341da177e4SLinus Torvalds if (IS_ERR(link)) 32351da177e4SLinus Torvalds goto out; 32361da177e4SLinus Torvalds 32371da177e4SLinus Torvalds len = strlen(link); 32381da177e4SLinus Torvalds if (len > (unsigned) buflen) 32391da177e4SLinus Torvalds len = buflen; 32401da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 32411da177e4SLinus Torvalds len = -EFAULT; 32421da177e4SLinus Torvalds out: 32431da177e4SLinus Torvalds return len; 32441da177e4SLinus Torvalds } 32451da177e4SLinus Torvalds 32461da177e4SLinus Torvalds /* 32471da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 32481da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 32491da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 32501da177e4SLinus Torvalds */ 32511da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 32521da177e4SLinus Torvalds { 32531da177e4SLinus Torvalds struct nameidata nd; 3254cc314eefSLinus Torvalds void *cookie; 3255694a1764SMarcin Slusarz int res; 3256cc314eefSLinus Torvalds 32571da177e4SLinus Torvalds nd.depth = 0; 3258cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3259694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3260694a1764SMarcin Slusarz return PTR_ERR(cookie); 3261694a1764SMarcin Slusarz 3262694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 32631da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3264cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3265694a1764SMarcin Slusarz return res; 32661da177e4SLinus Torvalds } 32671da177e4SLinus Torvalds 32681da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 32691da177e4SLinus Torvalds { 32701da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 32711da177e4SLinus Torvalds } 32721da177e4SLinus Torvalds 32731da177e4SLinus Torvalds /* get the link contents into pagecache */ 32741da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 32751da177e4SLinus Torvalds { 3276ebd09abbSDuane Griffin char *kaddr; 32771da177e4SLinus Torvalds struct page *page; 32781da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3279090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 32801da177e4SLinus Torvalds if (IS_ERR(page)) 32816fe6900eSNick Piggin return (char*)page; 32821da177e4SLinus Torvalds *ppage = page; 3283ebd09abbSDuane Griffin kaddr = kmap(page); 3284ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3285ebd09abbSDuane Griffin return kaddr; 32861da177e4SLinus Torvalds } 32871da177e4SLinus Torvalds 32881da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 32891da177e4SLinus Torvalds { 32901da177e4SLinus Torvalds struct page *page = NULL; 32911da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 32921da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 32931da177e4SLinus Torvalds if (page) { 32941da177e4SLinus Torvalds kunmap(page); 32951da177e4SLinus Torvalds page_cache_release(page); 32961da177e4SLinus Torvalds } 32971da177e4SLinus Torvalds return res; 32981da177e4SLinus Torvalds } 32991da177e4SLinus Torvalds 3300cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 33011da177e4SLinus Torvalds { 3302cc314eefSLinus Torvalds struct page *page = NULL; 33031da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3304cc314eefSLinus Torvalds return page; 33051da177e4SLinus Torvalds } 33061da177e4SLinus Torvalds 3307cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 33081da177e4SLinus Torvalds { 3309cc314eefSLinus Torvalds struct page *page = cookie; 3310cc314eefSLinus Torvalds 3311cc314eefSLinus Torvalds if (page) { 33121da177e4SLinus Torvalds kunmap(page); 33131da177e4SLinus Torvalds page_cache_release(page); 33141da177e4SLinus Torvalds } 33151da177e4SLinus Torvalds } 33161da177e4SLinus Torvalds 331754566b2cSNick Piggin /* 331854566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 331954566b2cSNick Piggin */ 332054566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 33211da177e4SLinus Torvalds { 33221da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 33230adb25d2SKirill Korotaev struct page *page; 3324afddba49SNick Piggin void *fsdata; 3325beb497abSDmitriy Monakhov int err; 33261da177e4SLinus Torvalds char *kaddr; 332754566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 332854566b2cSNick Piggin if (nofs) 332954566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 33301da177e4SLinus Torvalds 33317e53cac4SNeilBrown retry: 3332afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 333354566b2cSNick Piggin flags, &page, &fsdata); 33341da177e4SLinus Torvalds if (err) 3335afddba49SNick Piggin goto fail; 3336afddba49SNick Piggin 33371da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 33381da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 33391da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 3340afddba49SNick Piggin 3341afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3342afddba49SNick Piggin page, fsdata); 33431da177e4SLinus Torvalds if (err < 0) 33441da177e4SLinus Torvalds goto fail; 3345afddba49SNick Piggin if (err < len-1) 3346afddba49SNick Piggin goto retry; 3347afddba49SNick Piggin 33481da177e4SLinus Torvalds mark_inode_dirty(inode); 33491da177e4SLinus Torvalds return 0; 33501da177e4SLinus Torvalds fail: 33511da177e4SLinus Torvalds return err; 33521da177e4SLinus Torvalds } 33531da177e4SLinus Torvalds 33540adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 33550adb25d2SKirill Korotaev { 33560adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 335754566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 33580adb25d2SKirill Korotaev } 33590adb25d2SKirill Korotaev 336092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 33611da177e4SLinus Torvalds .readlink = generic_readlink, 33621da177e4SLinus Torvalds .follow_link = page_follow_link_light, 33631da177e4SLinus Torvalds .put_link = page_put_link, 33641da177e4SLinus Torvalds }; 33651da177e4SLinus Torvalds 33662d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3367cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 33681da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 33691da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 33701da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 33711da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 33721da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 33731da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 33741da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 33751da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 33761da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 33770adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 33781da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 33791da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3380c9c6cac0SAl Viro EXPORT_SYMBOL(kern_path_parent); 3381d1811465SAl Viro EXPORT_SYMBOL(kern_path); 338216f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3383f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 33848c744fb8SChristoph Hellwig EXPORT_SYMBOL(file_permission); 33851da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 33861da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 33871da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 33881da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 33891da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 33901da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 33911da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 33921da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 33931da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 33941da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 33951da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 33961da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 33971da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 33981da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3399