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 1391da177e4SLinus Torvalds char * getname(const char __user * filename) 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) { 1501da177e4SLinus Torvalds __putname(tmp); 1511da177e4SLinus Torvalds result = ERR_PTR(retval); 1521da177e4SLinus Torvalds } 1531da177e4SLinus Torvalds } 1541da177e4SLinus Torvalds audit_getname(result); 1551da177e4SLinus Torvalds return result; 1561da177e4SLinus Torvalds } 1571da177e4SLinus Torvalds 1581da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 1591da177e4SLinus Torvalds void putname(const char *name) 1601da177e4SLinus Torvalds { 1615ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 1621da177e4SLinus Torvalds audit_putname(name); 1631da177e4SLinus Torvalds else 1641da177e4SLinus Torvalds __putname(name); 1651da177e4SLinus Torvalds } 1661da177e4SLinus Torvalds EXPORT_SYMBOL(putname); 1671da177e4SLinus Torvalds #endif 1681da177e4SLinus Torvalds 1695909ccaaSLinus Torvalds /* 1705909ccaaSLinus Torvalds * This does basic POSIX ACL permission checking 1715909ccaaSLinus Torvalds */ 172b74c79e9SNick Piggin static int acl_permission_check(struct inode *inode, int mask, unsigned int flags, 173b74c79e9SNick Piggin int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) 1745909ccaaSLinus Torvalds { 1755909ccaaSLinus Torvalds umode_t mode = inode->i_mode; 1765909ccaaSLinus Torvalds 1775909ccaaSLinus Torvalds mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 1785909ccaaSLinus Torvalds 1795909ccaaSLinus Torvalds if (current_fsuid() == inode->i_uid) 1805909ccaaSLinus Torvalds mode >>= 6; 1815909ccaaSLinus Torvalds else { 1825909ccaaSLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { 183b74c79e9SNick Piggin int error = check_acl(inode, mask, flags); 1845909ccaaSLinus Torvalds if (error != -EAGAIN) 1855909ccaaSLinus Torvalds return error; 1865909ccaaSLinus Torvalds } 1875909ccaaSLinus Torvalds 1885909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 1895909ccaaSLinus Torvalds mode >>= 3; 1905909ccaaSLinus Torvalds } 1915909ccaaSLinus Torvalds 1925909ccaaSLinus Torvalds /* 1935909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 1945909ccaaSLinus Torvalds */ 1955909ccaaSLinus Torvalds if ((mask & ~mode) == 0) 1965909ccaaSLinus Torvalds return 0; 1975909ccaaSLinus Torvalds return -EACCES; 1985909ccaaSLinus Torvalds } 1991da177e4SLinus Torvalds 2001da177e4SLinus Torvalds /** 2011da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 2021da177e4SLinus Torvalds * @inode: inode to check access rights for 2031da177e4SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 2041da177e4SLinus Torvalds * @check_acl: optional callback to check for Posix ACLs 20539191628SRandy Dunlap * @flags: IPERM_FLAG_ flags. 2061da177e4SLinus Torvalds * 2071da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 2081da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 2091da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 210b74c79e9SNick Piggin * are used for other things. 211b74c79e9SNick Piggin * 212b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 213b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 214b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 2151da177e4SLinus Torvalds */ 216b74c79e9SNick Piggin int generic_permission(struct inode *inode, int mask, unsigned int flags, 217b74c79e9SNick Piggin int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) 2181da177e4SLinus Torvalds { 2195909ccaaSLinus Torvalds int ret; 2201da177e4SLinus Torvalds 2211da177e4SLinus Torvalds /* 2225909ccaaSLinus Torvalds * Do the basic POSIX ACL permission checks. 2231da177e4SLinus Torvalds */ 224b74c79e9SNick Piggin ret = acl_permission_check(inode, mask, flags, check_acl); 2255909ccaaSLinus Torvalds if (ret != -EACCES) 2265909ccaaSLinus Torvalds return ret; 2271da177e4SLinus Torvalds 2281da177e4SLinus Torvalds /* 2291da177e4SLinus Torvalds * Read/write DACs are always overridable. 2301da177e4SLinus Torvalds * Executable DACs are overridable if at least one exec bit is set. 2311da177e4SLinus Torvalds */ 232f696a365SMiklos Szeredi if (!(mask & MAY_EXEC) || execute_ok(inode)) 2331da177e4SLinus Torvalds if (capable(CAP_DAC_OVERRIDE)) 2341da177e4SLinus Torvalds return 0; 2351da177e4SLinus Torvalds 2361da177e4SLinus Torvalds /* 2371da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 2381da177e4SLinus Torvalds */ 2397ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 2401da177e4SLinus Torvalds if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) 2411da177e4SLinus Torvalds if (capable(CAP_DAC_READ_SEARCH)) 2421da177e4SLinus Torvalds return 0; 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds return -EACCES; 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 247cb23beb5SChristoph Hellwig /** 248cb23beb5SChristoph Hellwig * inode_permission - check for access rights to a given inode 249cb23beb5SChristoph Hellwig * @inode: inode to check permission on 250cb23beb5SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 251cb23beb5SChristoph Hellwig * 252cb23beb5SChristoph Hellwig * Used to check for read/write/execute permissions on an inode. 253cb23beb5SChristoph Hellwig * We use "fsuid" for this, letting us set arbitrary permissions 254cb23beb5SChristoph Hellwig * for filesystem access without changing the "normal" uids which 255cb23beb5SChristoph Hellwig * are used for other things. 256cb23beb5SChristoph Hellwig */ 257f419a2e3SAl Viro int inode_permission(struct inode *inode, int mask) 2581da177e4SLinus Torvalds { 259e6305c43SAl Viro int retval; 2601da177e4SLinus Torvalds 2611da177e4SLinus Torvalds if (mask & MAY_WRITE) { 26222590e41SMiklos Szeredi umode_t mode = inode->i_mode; 2631da177e4SLinus Torvalds 2641da177e4SLinus Torvalds /* 2651da177e4SLinus Torvalds * Nobody gets write access to a read-only fs. 2661da177e4SLinus Torvalds */ 2671da177e4SLinus Torvalds if (IS_RDONLY(inode) && 2681da177e4SLinus Torvalds (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 2691da177e4SLinus Torvalds return -EROFS; 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds /* 2721da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 2731da177e4SLinus Torvalds */ 2741da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 2751da177e4SLinus Torvalds return -EACCES; 2761da177e4SLinus Torvalds } 2771da177e4SLinus Torvalds 278acfa4380SAl Viro if (inode->i_op->permission) 279b74c79e9SNick Piggin retval = inode->i_op->permission(inode, mask, 0); 280f696a365SMiklos Szeredi else 281b74c79e9SNick Piggin retval = generic_permission(inode, mask, 0, 282b74c79e9SNick Piggin inode->i_op->check_acl); 283f696a365SMiklos Szeredi 2841da177e4SLinus Torvalds if (retval) 2851da177e4SLinus Torvalds return retval; 2861da177e4SLinus Torvalds 28708ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 28808ce5f16SSerge E. Hallyn if (retval) 28908ce5f16SSerge E. Hallyn return retval; 29008ce5f16SSerge E. Hallyn 291d09ca739SEric Paris return security_inode_permission(inode, mask); 2921da177e4SLinus Torvalds } 2931da177e4SLinus Torvalds 294e4543eddSChristoph Hellwig /** 2958c744fb8SChristoph Hellwig * file_permission - check for additional access rights to a given file 2968c744fb8SChristoph Hellwig * @file: file to check access rights for 2978c744fb8SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 2988c744fb8SChristoph Hellwig * 2998c744fb8SChristoph Hellwig * Used to check for read/write/execute permissions on an already opened 3008c744fb8SChristoph Hellwig * file. 3018c744fb8SChristoph Hellwig * 3028c744fb8SChristoph Hellwig * Note: 3038c744fb8SChristoph Hellwig * Do not use this function in new code. All access checks should 304cb23beb5SChristoph Hellwig * be done using inode_permission(). 3058c744fb8SChristoph Hellwig */ 3068c744fb8SChristoph Hellwig int file_permission(struct file *file, int mask) 3078c744fb8SChristoph Hellwig { 308f419a2e3SAl Viro return inode_permission(file->f_path.dentry->d_inode, mask); 3098c744fb8SChristoph Hellwig } 3108c744fb8SChristoph Hellwig 3111da177e4SLinus Torvalds /* 3121da177e4SLinus Torvalds * get_write_access() gets write permission for a file. 3131da177e4SLinus Torvalds * put_write_access() releases this write permission. 3141da177e4SLinus Torvalds * This is used for regular files. 3151da177e4SLinus Torvalds * We cannot support write (and maybe mmap read-write shared) accesses and 3161da177e4SLinus Torvalds * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode 3171da177e4SLinus Torvalds * can have the following values: 3181da177e4SLinus Torvalds * 0: no writers, no VM_DENYWRITE mappings 3191da177e4SLinus Torvalds * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist 3201da177e4SLinus Torvalds * > 0: (i_writecount) users are writing to the file. 3211da177e4SLinus Torvalds * 3221da177e4SLinus Torvalds * Normally we operate on that counter with atomic_{inc,dec} and it's safe 3231da177e4SLinus Torvalds * except for the cases where we don't hold i_writecount yet. Then we need to 3241da177e4SLinus Torvalds * use {get,deny}_write_access() - these functions check the sign and refuse 3251da177e4SLinus Torvalds * to do the change if sign is wrong. Exclusion between them is provided by 3261da177e4SLinus Torvalds * the inode->i_lock spinlock. 3271da177e4SLinus Torvalds */ 3281da177e4SLinus Torvalds 3291da177e4SLinus Torvalds int get_write_access(struct inode * inode) 3301da177e4SLinus Torvalds { 3311da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3321da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) < 0) { 3331da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3341da177e4SLinus Torvalds return -ETXTBSY; 3351da177e4SLinus Torvalds } 3361da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 3371da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3381da177e4SLinus Torvalds 3391da177e4SLinus Torvalds return 0; 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds int deny_write_access(struct file * file) 3431da177e4SLinus Torvalds { 3440f7fc9e4SJosef "Jeff" Sipek struct inode *inode = file->f_path.dentry->d_inode; 3451da177e4SLinus Torvalds 3461da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3471da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) > 0) { 3481da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3491da177e4SLinus Torvalds return -ETXTBSY; 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds atomic_dec(&inode->i_writecount); 3521da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3531da177e4SLinus Torvalds 3541da177e4SLinus Torvalds return 0; 3551da177e4SLinus Torvalds } 3561da177e4SLinus Torvalds 3571d957f9bSJan Blunck /** 3585dd784d0SJan Blunck * path_get - get a reference to a path 3595dd784d0SJan Blunck * @path: path to get the reference to 3605dd784d0SJan Blunck * 3615dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3625dd784d0SJan Blunck */ 3635dd784d0SJan Blunck void path_get(struct path *path) 3645dd784d0SJan Blunck { 3655dd784d0SJan Blunck mntget(path->mnt); 3665dd784d0SJan Blunck dget(path->dentry); 3675dd784d0SJan Blunck } 3685dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 3695dd784d0SJan Blunck 3705dd784d0SJan Blunck /** 3711d957f9bSJan Blunck * path_put - put a reference to a path 3721d957f9bSJan Blunck * @path: path to put the reference to 3731d957f9bSJan Blunck * 3741d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 3751d957f9bSJan Blunck */ 3761d957f9bSJan Blunck void path_put(struct path *path) 3771da177e4SLinus Torvalds { 3781d957f9bSJan Blunck dput(path->dentry); 3791d957f9bSJan Blunck mntput(path->mnt); 3801da177e4SLinus Torvalds } 3811d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 3821da177e4SLinus Torvalds 383834f2a4aSTrond Myklebust /** 38431e6b01fSNick Piggin * nameidata_drop_rcu - drop this nameidata out of rcu-walk 38531e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 38639191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 38731e6b01fSNick Piggin * 38831e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 38931e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt 39031e6b01fSNick Piggin * to drop out of rcu-walk mode and take normal reference counts on dentries 39131e6b01fSNick Piggin * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take 39231e6b01fSNick Piggin * refcounts at the last known good point before rcu-walk got stuck, so 39331e6b01fSNick Piggin * ref-walk may continue from there. If this is not successful (eg. a seqcount 39431e6b01fSNick Piggin * has changed), then failure is returned and path walk restarts from the 39531e6b01fSNick Piggin * beginning in ref-walk mode. 39631e6b01fSNick Piggin * 39731e6b01fSNick Piggin * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into 39831e6b01fSNick Piggin * ref-walk. Must be called from rcu-walk context. 39931e6b01fSNick Piggin */ 40031e6b01fSNick Piggin static int nameidata_drop_rcu(struct nameidata *nd) 40131e6b01fSNick Piggin { 40231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 40331e6b01fSNick Piggin struct dentry *dentry = nd->path.dentry; 40431e6b01fSNick Piggin 40531e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 40631e6b01fSNick Piggin if (nd->root.mnt) { 40731e6b01fSNick Piggin spin_lock(&fs->lock); 40831e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 40931e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 41031e6b01fSNick Piggin goto err_root; 41131e6b01fSNick Piggin } 41231e6b01fSNick Piggin spin_lock(&dentry->d_lock); 41331e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 41431e6b01fSNick Piggin goto err; 41531e6b01fSNick Piggin BUG_ON(nd->inode != dentry->d_inode); 41631e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 41731e6b01fSNick Piggin if (nd->root.mnt) { 41831e6b01fSNick Piggin path_get(&nd->root); 41931e6b01fSNick Piggin spin_unlock(&fs->lock); 42031e6b01fSNick Piggin } 42131e6b01fSNick Piggin mntget(nd->path.mnt); 42231e6b01fSNick Piggin 42331e6b01fSNick Piggin rcu_read_unlock(); 42431e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 42531e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 42631e6b01fSNick Piggin return 0; 42731e6b01fSNick Piggin err: 42831e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 42931e6b01fSNick Piggin err_root: 43031e6b01fSNick Piggin if (nd->root.mnt) 43131e6b01fSNick Piggin spin_unlock(&fs->lock); 43231e6b01fSNick Piggin return -ECHILD; 43331e6b01fSNick Piggin } 43431e6b01fSNick Piggin 43531e6b01fSNick Piggin /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ 43631e6b01fSNick Piggin static inline int nameidata_drop_rcu_maybe(struct nameidata *nd) 43731e6b01fSNick Piggin { 43831e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) 43931e6b01fSNick Piggin return nameidata_drop_rcu(nd); 44031e6b01fSNick Piggin return 0; 44131e6b01fSNick Piggin } 44231e6b01fSNick Piggin 44331e6b01fSNick Piggin /** 44431e6b01fSNick Piggin * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk 44531e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 44631e6b01fSNick Piggin * @dentry: dentry to drop 44739191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 44831e6b01fSNick Piggin * 44931e6b01fSNick Piggin * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root, 45031e6b01fSNick Piggin * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on 45131e6b01fSNick Piggin * @nd. Must be called from rcu-walk context. 45231e6b01fSNick Piggin */ 45331e6b01fSNick Piggin static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry) 45431e6b01fSNick Piggin { 45531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 45631e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 45731e6b01fSNick Piggin 45831e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 45931e6b01fSNick Piggin if (nd->root.mnt) { 46031e6b01fSNick Piggin spin_lock(&fs->lock); 46131e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 46231e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 46331e6b01fSNick Piggin goto err_root; 46431e6b01fSNick Piggin } 46531e6b01fSNick Piggin spin_lock(&parent->d_lock); 46631e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 46731e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 46831e6b01fSNick Piggin goto err; 46931e6b01fSNick Piggin /* 47031e6b01fSNick Piggin * If the sequence check on the child dentry passed, then the child has 47131e6b01fSNick Piggin * not been removed from its parent. This means the parent dentry must 47231e6b01fSNick Piggin * be valid and able to take a reference at this point. 47331e6b01fSNick Piggin */ 47431e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 47531e6b01fSNick Piggin BUG_ON(!parent->d_count); 47631e6b01fSNick Piggin parent->d_count++; 47731e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 47831e6b01fSNick Piggin spin_unlock(&parent->d_lock); 47931e6b01fSNick Piggin if (nd->root.mnt) { 48031e6b01fSNick Piggin path_get(&nd->root); 48131e6b01fSNick Piggin spin_unlock(&fs->lock); 48231e6b01fSNick Piggin } 48331e6b01fSNick Piggin mntget(nd->path.mnt); 48431e6b01fSNick Piggin 48531e6b01fSNick Piggin rcu_read_unlock(); 48631e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 48731e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 48831e6b01fSNick Piggin return 0; 48931e6b01fSNick Piggin err: 49031e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 49131e6b01fSNick Piggin spin_unlock(&parent->d_lock); 49231e6b01fSNick Piggin err_root: 49331e6b01fSNick Piggin if (nd->root.mnt) 49431e6b01fSNick Piggin spin_unlock(&fs->lock); 49531e6b01fSNick Piggin return -ECHILD; 49631e6b01fSNick Piggin } 49731e6b01fSNick Piggin 49831e6b01fSNick Piggin /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ 49931e6b01fSNick Piggin static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry) 50031e6b01fSNick Piggin { 50131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) 50231e6b01fSNick Piggin return nameidata_dentry_drop_rcu(nd, dentry); 50331e6b01fSNick Piggin return 0; 50431e6b01fSNick Piggin } 50531e6b01fSNick Piggin 50631e6b01fSNick Piggin /** 50731e6b01fSNick Piggin * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk 50831e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 50939191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 51031e6b01fSNick Piggin * 51131e6b01fSNick Piggin * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk. 51231e6b01fSNick Piggin * nd->path should be the final element of the lookup, so nd->root is discarded. 51331e6b01fSNick Piggin * Must be called from rcu-walk context. 51431e6b01fSNick Piggin */ 51531e6b01fSNick Piggin static int nameidata_drop_rcu_last(struct nameidata *nd) 51631e6b01fSNick Piggin { 51731e6b01fSNick Piggin struct dentry *dentry = nd->path.dentry; 51831e6b01fSNick Piggin 51931e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 52031e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 52131e6b01fSNick Piggin nd->root.mnt = NULL; 52231e6b01fSNick Piggin spin_lock(&dentry->d_lock); 52331e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 52431e6b01fSNick Piggin goto err_unlock; 52531e6b01fSNick Piggin BUG_ON(nd->inode != dentry->d_inode); 52631e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 52731e6b01fSNick Piggin 52831e6b01fSNick Piggin mntget(nd->path.mnt); 52931e6b01fSNick Piggin 53031e6b01fSNick Piggin rcu_read_unlock(); 53131e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 53231e6b01fSNick Piggin 53331e6b01fSNick Piggin return 0; 53431e6b01fSNick Piggin 53531e6b01fSNick Piggin err_unlock: 53631e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 53731e6b01fSNick Piggin rcu_read_unlock(); 53831e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 53931e6b01fSNick Piggin return -ECHILD; 54031e6b01fSNick Piggin } 54131e6b01fSNick Piggin 54231e6b01fSNick Piggin /** 543834f2a4aSTrond Myklebust * release_open_intent - free up open intent resources 544834f2a4aSTrond Myklebust * @nd: pointer to nameidata 545834f2a4aSTrond Myklebust */ 546834f2a4aSTrond Myklebust void release_open_intent(struct nameidata *nd) 547834f2a4aSTrond Myklebust { 5482dab5974SLinus Torvalds struct file *file = nd->intent.open.file; 5492dab5974SLinus Torvalds 5502dab5974SLinus Torvalds if (file && !IS_ERR(file)) { 5512dab5974SLinus Torvalds if (file->f_path.dentry == NULL) 5522dab5974SLinus Torvalds put_filp(file); 553834f2a4aSTrond Myklebust else 5542dab5974SLinus Torvalds fput(file); 5552dab5974SLinus Torvalds } 556834f2a4aSTrond Myklebust } 557834f2a4aSTrond Myklebust 558f60aef7eSAl Viro static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) 55934286d66SNick Piggin { 560f60aef7eSAl Viro return dentry->d_op->d_revalidate(dentry, nd); 56134286d66SNick Piggin } 56234286d66SNick Piggin 563f5e1c1c1SAl Viro static struct dentry * 564bcdc5e01SIan Kent do_revalidate(struct dentry *dentry, struct nameidata *nd) 565bcdc5e01SIan Kent { 566f5e1c1c1SAl Viro int status = d_revalidate(dentry, nd); 567bcdc5e01SIan Kent if (unlikely(status <= 0)) { 568bcdc5e01SIan Kent /* 569bcdc5e01SIan Kent * The dentry failed validation. 570bcdc5e01SIan Kent * If d_revalidate returned 0 attempt to invalidate 571bcdc5e01SIan Kent * the dentry otherwise d_revalidate is asking us 572bcdc5e01SIan Kent * to return a fail status. 573bcdc5e01SIan Kent */ 57434286d66SNick Piggin if (status < 0) { 57534286d66SNick Piggin dput(dentry); 57634286d66SNick Piggin dentry = ERR_PTR(status); 577f5e1c1c1SAl Viro } else if (!d_invalidate(dentry)) { 578bcdc5e01SIan Kent dput(dentry); 579bcdc5e01SIan Kent dentry = NULL; 580bcdc5e01SIan Kent } 581bcdc5e01SIan Kent } 582f5e1c1c1SAl Viro return dentry; 583f5e1c1c1SAl Viro } 584f5e1c1c1SAl Viro 585f5e1c1c1SAl Viro static inline struct dentry * 586f5e1c1c1SAl Viro do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd) 587f5e1c1c1SAl Viro { 588f60aef7eSAl Viro int status = d_revalidate(dentry, nd); 589f5e1c1c1SAl Viro if (likely(status > 0)) 590f5e1c1c1SAl Viro return dentry; 591f5e1c1c1SAl Viro if (status == -ECHILD) { 592f5e1c1c1SAl Viro if (nameidata_dentry_drop_rcu(nd, dentry)) 593f5e1c1c1SAl Viro return ERR_PTR(-ECHILD); 594f5e1c1c1SAl Viro return do_revalidate(dentry, nd); 595f5e1c1c1SAl Viro } 596f5e1c1c1SAl Viro if (status < 0) 597f5e1c1c1SAl Viro return ERR_PTR(status); 598f5e1c1c1SAl Viro /* Don't d_invalidate in rcu-walk mode */ 599f5e1c1c1SAl Viro if (nameidata_dentry_drop_rcu(nd, dentry)) 600f5e1c1c1SAl Viro return ERR_PTR(-ECHILD); 601f5e1c1c1SAl Viro if (!d_invalidate(dentry)) { 602f5e1c1c1SAl Viro dput(dentry); 603f5e1c1c1SAl Viro dentry = NULL; 604bcdc5e01SIan Kent } 605bcdc5e01SIan Kent return dentry; 606bcdc5e01SIan Kent } 607bcdc5e01SIan Kent 6081da177e4SLinus Torvalds /* 60916c2cd71SAl Viro * handle_reval_path - force revalidation of a dentry 61039159de2SJeff Layton * 61139159de2SJeff Layton * In some situations the path walking code will trust dentries without 61239159de2SJeff Layton * revalidating them. This causes problems for filesystems that depend on 61339159de2SJeff Layton * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set 61439159de2SJeff Layton * (which indicates that it's possible for the dentry to go stale), force 61539159de2SJeff Layton * a d_revalidate call before proceeding. 61639159de2SJeff Layton * 61739159de2SJeff Layton * Returns 0 if the revalidation was successful. If the revalidation fails, 61839159de2SJeff Layton * either return the error returned by d_revalidate or -ESTALE if the 61939159de2SJeff Layton * revalidation it just returned 0. If d_revalidate returns 0, we attempt to 62039159de2SJeff Layton * invalidate the dentry. It's up to the caller to handle putting references 62139159de2SJeff Layton * to the path if necessary. 62239159de2SJeff Layton */ 62316c2cd71SAl Viro static inline int handle_reval_path(struct nameidata *nd) 62439159de2SJeff Layton { 62516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 62639159de2SJeff Layton int status; 62739159de2SJeff Layton 62816c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 62939159de2SJeff Layton return 0; 63039159de2SJeff Layton 63116c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 63216c2cd71SAl Viro return 0; 63316c2cd71SAl Viro 63416c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 63516c2cd71SAl Viro return 0; 63616c2cd71SAl Viro 63716c2cd71SAl Viro /* Note: we do not d_invalidate() */ 63834286d66SNick Piggin status = d_revalidate(dentry, nd); 63939159de2SJeff Layton if (status > 0) 64039159de2SJeff Layton return 0; 64139159de2SJeff Layton 64216c2cd71SAl Viro if (!status) 64339159de2SJeff Layton status = -ESTALE; 64416c2cd71SAl Viro 64539159de2SJeff Layton return status; 64639159de2SJeff Layton } 64739159de2SJeff Layton 64839159de2SJeff Layton /* 649b75b5086SAl Viro * Short-cut version of permission(), for calling on directories 650b75b5086SAl Viro * during pathname resolution. Combines parts of permission() 651b75b5086SAl Viro * and generic_permission(), and tests ONLY for MAY_EXEC permission. 6521da177e4SLinus Torvalds * 6531da177e4SLinus Torvalds * If appropriate, check DAC only. If not appropriate, or 654b75b5086SAl Viro * short-cut DAC fails, then call ->permission() to do more 6551da177e4SLinus Torvalds * complete permission check. 6561da177e4SLinus Torvalds */ 657b74c79e9SNick Piggin static inline int exec_permission(struct inode *inode, unsigned int flags) 6581da177e4SLinus Torvalds { 6595909ccaaSLinus Torvalds int ret; 6601da177e4SLinus Torvalds 661cb9179eaSLinus Torvalds if (inode->i_op->permission) { 662b74c79e9SNick Piggin ret = inode->i_op->permission(inode, MAY_EXEC, flags); 663b74c79e9SNick Piggin } else { 664b74c79e9SNick Piggin ret = acl_permission_check(inode, MAY_EXEC, flags, 665b74c79e9SNick Piggin inode->i_op->check_acl); 666cb9179eaSLinus Torvalds } 667b74c79e9SNick Piggin if (likely(!ret)) 6681da177e4SLinus Torvalds goto ok; 669b74c79e9SNick Piggin if (ret == -ECHILD) 67031e6b01fSNick Piggin return ret; 6711da177e4SLinus Torvalds 672f1ac9f6bSLinus Torvalds if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) 6731da177e4SLinus Torvalds goto ok; 6741da177e4SLinus Torvalds 6755909ccaaSLinus Torvalds return ret; 6761da177e4SLinus Torvalds ok: 677b74c79e9SNick Piggin return security_inode_exec_permission(inode, flags); 6781da177e4SLinus Torvalds } 6791da177e4SLinus Torvalds 6802a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6812a737871SAl Viro { 682f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 683f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6842a737871SAl Viro } 6852a737871SAl Viro 6866de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6876de88d72SAl Viro 68831e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 68931e6b01fSNick Piggin { 69031e6b01fSNick Piggin if (!nd->root.mnt) { 69131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 692c28cc364SNick Piggin unsigned seq; 693c28cc364SNick Piggin 694c28cc364SNick Piggin do { 695c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 69631e6b01fSNick Piggin nd->root = fs->root; 697c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 69831e6b01fSNick Piggin } 69931e6b01fSNick Piggin } 70031e6b01fSNick Piggin 701f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 7021da177e4SLinus Torvalds { 70331e6b01fSNick Piggin int ret; 70431e6b01fSNick Piggin 7051da177e4SLinus Torvalds if (IS_ERR(link)) 7061da177e4SLinus Torvalds goto fail; 7071da177e4SLinus Torvalds 7081da177e4SLinus Torvalds if (*link == '/') { 7092a737871SAl Viro set_root(nd); 7101d957f9bSJan Blunck path_put(&nd->path); 7112a737871SAl Viro nd->path = nd->root; 7122a737871SAl Viro path_get(&nd->root); 71316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 7141da177e4SLinus Torvalds } 71531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 716b4091d5fSChristoph Hellwig 71731e6b01fSNick Piggin ret = link_path_walk(link, nd); 71831e6b01fSNick Piggin return ret; 7191da177e4SLinus Torvalds fail: 7201d957f9bSJan Blunck path_put(&nd->path); 7211da177e4SLinus Torvalds return PTR_ERR(link); 7221da177e4SLinus Torvalds } 7231da177e4SLinus Torvalds 7241d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 725051d3812SIan Kent { 726051d3812SIan Kent dput(path->dentry); 7274ac91378SJan Blunck if (path->mnt != nd->path.mnt) 728051d3812SIan Kent mntput(path->mnt); 729051d3812SIan Kent } 730051d3812SIan Kent 7317b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 7327b9337aaSNick Piggin struct nameidata *nd) 733051d3812SIan Kent { 73431e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 7354ac91378SJan Blunck dput(nd->path.dentry); 73631e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 7374ac91378SJan Blunck mntput(nd->path.mnt); 7389a229683SHuang Shijie } 73931e6b01fSNick Piggin nd->path.mnt = path->mnt; 7404ac91378SJan Blunck nd->path.dentry = path->dentry; 741051d3812SIan Kent } 742051d3812SIan Kent 743def4af30SAl Viro static __always_inline int 7447b9337aaSNick Piggin __do_follow_link(const struct path *link, struct nameidata *nd, void **p) 7451da177e4SLinus Torvalds { 7461da177e4SLinus Torvalds int error; 7477b9337aaSNick Piggin struct dentry *dentry = link->dentry; 7481da177e4SLinus Torvalds 749844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 750844a3917SAl Viro 7517b9337aaSNick Piggin touch_atime(link->mnt, dentry); 7521da177e4SLinus Torvalds nd_set_link(nd, NULL); 753cd4e91d3SAl Viro 75487556ef1SDavid Howells if (link->mnt == nd->path.mnt) 7557b9337aaSNick Piggin mntget(link->mnt); 75631e6b01fSNick Piggin 75736f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 75836f3b4f6SAl Viro if (error) { 75936f3b4f6SAl Viro *p = ERR_PTR(error); /* no ->put_link(), please */ 76036f3b4f6SAl Viro path_put(&nd->path); 76136f3b4f6SAl Viro return error; 76236f3b4f6SAl Viro } 76336f3b4f6SAl Viro 76486acdca1SAl Viro nd->last_type = LAST_BIND; 765def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 766def4af30SAl Viro error = PTR_ERR(*p); 767def4af30SAl Viro if (!IS_ERR(*p)) { 7681da177e4SLinus Torvalds char *s = nd_get_link(nd); 769cc314eefSLinus Torvalds error = 0; 7701da177e4SLinus Torvalds if (s) 7711da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 77216c2cd71SAl Viro else if (nd->last_type == LAST_BIND) 77316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds return error; 7761da177e4SLinus Torvalds } 7771da177e4SLinus Torvalds 7781da177e4SLinus Torvalds /* 7791da177e4SLinus Torvalds * This limits recursive symlink follows to 8, while 7801da177e4SLinus Torvalds * limiting consecutive symlinks to 40. 7811da177e4SLinus Torvalds * 7821da177e4SLinus Torvalds * Without that kind of total limit, nasty chains of consecutive 7831da177e4SLinus Torvalds * symlinks can cause almost arbitrarily long lookups. 7841da177e4SLinus Torvalds */ 7853abb17e8SLinus Torvalds static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd) 7861da177e4SLinus Torvalds { 787def4af30SAl Viro void *cookie; 7881da177e4SLinus Torvalds int err = -ELOOP; 789844a3917SAl Viro 790844a3917SAl Viro /* We drop rcu-walk here */ 791844a3917SAl Viro if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) 792844a3917SAl Viro return -ECHILD; 7933abb17e8SLinus Torvalds BUG_ON(inode != path->dentry->d_inode); 794844a3917SAl Viro 7951da177e4SLinus Torvalds if (current->link_count >= MAX_NESTED_LINKS) 7961da177e4SLinus Torvalds goto loop; 7971da177e4SLinus Torvalds if (current->total_link_count >= 40) 7981da177e4SLinus Torvalds goto loop; 7991da177e4SLinus Torvalds BUG_ON(nd->depth >= MAX_NESTED_LINKS); 8001da177e4SLinus Torvalds cond_resched(); 8011da177e4SLinus Torvalds current->link_count++; 8021da177e4SLinus Torvalds current->total_link_count++; 8031da177e4SLinus Torvalds nd->depth++; 804def4af30SAl Viro err = __do_follow_link(path, nd, &cookie); 805def4af30SAl Viro if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) 806def4af30SAl Viro path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); 807258fa999SAl Viro path_put(path); 8081da177e4SLinus Torvalds current->link_count--; 8091da177e4SLinus Torvalds nd->depth--; 8101da177e4SLinus Torvalds return err; 8111da177e4SLinus Torvalds loop: 8121d957f9bSJan Blunck path_put_conditional(path, nd); 8131d957f9bSJan Blunck path_put(&nd->path); 8141da177e4SLinus Torvalds return err; 8151da177e4SLinus Torvalds } 8161da177e4SLinus Torvalds 81731e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 81831e6b01fSNick Piggin { 81931e6b01fSNick Piggin struct vfsmount *parent; 82031e6b01fSNick Piggin struct dentry *mountpoint; 82131e6b01fSNick Piggin 82231e6b01fSNick Piggin parent = path->mnt->mnt_parent; 82331e6b01fSNick Piggin if (parent == path->mnt) 82431e6b01fSNick Piggin return 0; 82531e6b01fSNick Piggin mountpoint = path->mnt->mnt_mountpoint; 82631e6b01fSNick Piggin path->dentry = mountpoint; 82731e6b01fSNick Piggin path->mnt = parent; 82831e6b01fSNick Piggin return 1; 82931e6b01fSNick Piggin } 83031e6b01fSNick Piggin 831bab77ebfSAl Viro int follow_up(struct path *path) 8321da177e4SLinus Torvalds { 8331da177e4SLinus Torvalds struct vfsmount *parent; 8341da177e4SLinus Torvalds struct dentry *mountpoint; 83599b7db7bSNick Piggin 83699b7db7bSNick Piggin br_read_lock(vfsmount_lock); 837bab77ebfSAl Viro parent = path->mnt->mnt_parent; 838bab77ebfSAl Viro if (parent == path->mnt) { 83999b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 8401da177e4SLinus Torvalds return 0; 8411da177e4SLinus Torvalds } 8421da177e4SLinus Torvalds mntget(parent); 843bab77ebfSAl Viro mountpoint = dget(path->mnt->mnt_mountpoint); 84499b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 845bab77ebfSAl Viro dput(path->dentry); 846bab77ebfSAl Viro path->dentry = mountpoint; 847bab77ebfSAl Viro mntput(path->mnt); 848bab77ebfSAl Viro path->mnt = parent; 8491da177e4SLinus Torvalds return 1; 8501da177e4SLinus Torvalds } 8511da177e4SLinus Torvalds 852b5c84bf6SNick Piggin /* 8539875cf80SDavid Howells * Perform an automount 8549875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 8559875cf80SDavid Howells * were called with. 8561da177e4SLinus Torvalds */ 8579875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 8589875cf80SDavid Howells bool *need_mntput) 85931e6b01fSNick Piggin { 8609875cf80SDavid Howells struct vfsmount *mnt; 861ea5b778aSDavid Howells int err; 8629875cf80SDavid Howells 8639875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 8649875cf80SDavid Howells return -EREMOTE; 8659875cf80SDavid Howells 8666f45b656SDavid Howells /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT 8676f45b656SDavid Howells * and this is the terminal part of the path. 8686f45b656SDavid Howells */ 8696f45b656SDavid Howells if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE)) 8706f45b656SDavid Howells return -EISDIR; /* we actually want to stop here */ 8716f45b656SDavid Howells 8729875cf80SDavid Howells /* We want to mount if someone is trying to open/create a file of any 8739875cf80SDavid Howells * type under the mountpoint, wants to traverse through the mountpoint 8749875cf80SDavid Howells * or wants to open the mounted directory. 8759875cf80SDavid Howells * 8769875cf80SDavid Howells * We don't want to mount if someone's just doing a stat and they've 8779875cf80SDavid Howells * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and 8789875cf80SDavid Howells * appended a '/' to the name. 8799875cf80SDavid Howells */ 8809875cf80SDavid Howells if (!(flags & LOOKUP_FOLLOW) && 8819875cf80SDavid Howells !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | 8829875cf80SDavid Howells LOOKUP_OPEN | LOOKUP_CREATE))) 8839875cf80SDavid Howells return -EISDIR; 8849875cf80SDavid Howells 8859875cf80SDavid Howells current->total_link_count++; 8869875cf80SDavid Howells if (current->total_link_count >= 40) 8879875cf80SDavid Howells return -ELOOP; 8889875cf80SDavid Howells 8899875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 8909875cf80SDavid Howells if (IS_ERR(mnt)) { 8919875cf80SDavid Howells /* 8929875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 8939875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 8949875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 8959875cf80SDavid Howells * 8969875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 8979875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 8989875cf80SDavid Howells * the path is inaccessible and we should say so. 8999875cf80SDavid Howells */ 9009875cf80SDavid Howells if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) 9019875cf80SDavid Howells return -EREMOTE; 9029875cf80SDavid Howells return PTR_ERR(mnt); 90331e6b01fSNick Piggin } 904ea5b778aSDavid Howells 9059875cf80SDavid Howells if (!mnt) /* mount collision */ 9069875cf80SDavid Howells return 0; 9079875cf80SDavid Howells 90819a167afSAl Viro err = finish_automount(mnt, path); 909ea5b778aSDavid Howells 910ea5b778aSDavid Howells switch (err) { 911ea5b778aSDavid Howells case -EBUSY: 912ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 91319a167afSAl Viro return 0; 914ea5b778aSDavid Howells case 0: 915463ffb2eSAl Viro dput(path->dentry); 9169875cf80SDavid Howells if (*need_mntput) 9179875cf80SDavid Howells mntput(path->mnt); 9189875cf80SDavid Howells path->mnt = mnt; 9199875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 9209875cf80SDavid Howells *need_mntput = true; 9219875cf80SDavid Howells return 0; 92219a167afSAl Viro default: 92319a167afSAl Viro return err; 9249875cf80SDavid Howells } 92519a167afSAl Viro 926ea5b778aSDavid Howells } 9279875cf80SDavid Howells 9289875cf80SDavid Howells /* 9299875cf80SDavid Howells * Handle a dentry that is managed in some way. 930cc53ce53SDavid Howells * - Flagged for transit management (autofs) 9319875cf80SDavid Howells * - Flagged as mountpoint 9329875cf80SDavid Howells * - Flagged as automount point 9339875cf80SDavid Howells * 9349875cf80SDavid Howells * This may only be called in refwalk mode. 9359875cf80SDavid Howells * 9369875cf80SDavid Howells * Serialization is taken care of in namespace.c 9379875cf80SDavid Howells */ 9389875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 9399875cf80SDavid Howells { 9409875cf80SDavid Howells unsigned managed; 9419875cf80SDavid Howells bool need_mntput = false; 9429875cf80SDavid Howells int ret; 9439875cf80SDavid Howells 9449875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 9459875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 9469875cf80SDavid Howells * the components of that value change under us */ 9479875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 9489875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 9499875cf80SDavid Howells unlikely(managed != 0)) { 950cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 951cc53ce53SDavid Howells * being held. */ 952cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 953cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 954cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 955ab90911fSDavid Howells ret = path->dentry->d_op->d_manage(path->dentry, 956ab90911fSDavid Howells false, false); 957cc53ce53SDavid Howells if (ret < 0) 958cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 959cc53ce53SDavid Howells } 960cc53ce53SDavid Howells 9619875cf80SDavid Howells /* Transit to a mounted filesystem. */ 9629875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 9639875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 9649875cf80SDavid Howells if (mounted) { 9659875cf80SDavid Howells dput(path->dentry); 9669875cf80SDavid Howells if (need_mntput) 967463ffb2eSAl Viro mntput(path->mnt); 968463ffb2eSAl Viro path->mnt = mounted; 969463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 9709875cf80SDavid Howells need_mntput = true; 9719875cf80SDavid Howells continue; 972463ffb2eSAl Viro } 973463ffb2eSAl Viro 9749875cf80SDavid Howells /* Something is mounted on this dentry in another 9759875cf80SDavid Howells * namespace and/or whatever was mounted there in this 9769875cf80SDavid Howells * namespace got unmounted before we managed to get the 9779875cf80SDavid Howells * vfsmount_lock */ 9781da177e4SLinus Torvalds } 9799875cf80SDavid Howells 9809875cf80SDavid Howells /* Handle an automount point */ 9819875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 9829875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 9839875cf80SDavid Howells if (ret < 0) 9849875cf80SDavid Howells return ret == -EISDIR ? 0 : ret; 9859875cf80SDavid Howells continue; 9869875cf80SDavid Howells } 9879875cf80SDavid Howells 9889875cf80SDavid Howells /* We didn't change the current path point */ 9899875cf80SDavid Howells break; 9909875cf80SDavid Howells } 9919875cf80SDavid Howells return 0; 9921da177e4SLinus Torvalds } 9931da177e4SLinus Torvalds 994cc53ce53SDavid Howells int follow_down_one(struct path *path) 9951da177e4SLinus Torvalds { 9961da177e4SLinus Torvalds struct vfsmount *mounted; 9971da177e4SLinus Torvalds 9981c755af4SAl Viro mounted = lookup_mnt(path); 9991da177e4SLinus Torvalds if (mounted) { 10009393bd07SAl Viro dput(path->dentry); 10019393bd07SAl Viro mntput(path->mnt); 10029393bd07SAl Viro path->mnt = mounted; 10039393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 10041da177e4SLinus Torvalds return 1; 10051da177e4SLinus Torvalds } 10061da177e4SLinus Torvalds return 0; 10071da177e4SLinus Torvalds } 10081da177e4SLinus Torvalds 10099875cf80SDavid Howells /* 10109875cf80SDavid Howells * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we 1011cc53ce53SDavid Howells * meet a managed dentry and we're not walking to "..". True is returned to 10129875cf80SDavid Howells * continue, false to abort. 10139875cf80SDavid Howells */ 10149875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 10159875cf80SDavid Howells struct inode **inode, bool reverse_transit) 10169875cf80SDavid Howells { 10179875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 10189875cf80SDavid Howells struct vfsmount *mounted; 1019ab90911fSDavid Howells if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) && 1020ab90911fSDavid Howells !reverse_transit && 1021ab90911fSDavid Howells path->dentry->d_op->d_manage(path->dentry, false, true) < 0) 1022ab90911fSDavid Howells return false; 10239875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 10249875cf80SDavid Howells if (!mounted) 10259875cf80SDavid Howells break; 10269875cf80SDavid Howells path->mnt = mounted; 10279875cf80SDavid Howells path->dentry = mounted->mnt_root; 10289875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 10299875cf80SDavid Howells *inode = path->dentry->d_inode; 10309875cf80SDavid Howells } 10319875cf80SDavid Howells 10329875cf80SDavid Howells if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 10339875cf80SDavid Howells return reverse_transit; 10349875cf80SDavid Howells return true; 10359875cf80SDavid Howells } 10369875cf80SDavid Howells 103731e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 103831e6b01fSNick Piggin { 103931e6b01fSNick Piggin struct inode *inode = nd->inode; 104031e6b01fSNick Piggin 104131e6b01fSNick Piggin set_root_rcu(nd); 104231e6b01fSNick Piggin 104331e6b01fSNick Piggin while (1) { 104431e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 104531e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 104631e6b01fSNick Piggin break; 104731e6b01fSNick Piggin } 104831e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 104931e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 105031e6b01fSNick Piggin struct dentry *parent = old->d_parent; 105131e6b01fSNick Piggin unsigned seq; 105231e6b01fSNick Piggin 105331e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 105431e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 105531e6b01fSNick Piggin return -ECHILD; 105631e6b01fSNick Piggin inode = parent->d_inode; 105731e6b01fSNick Piggin nd->path.dentry = parent; 105831e6b01fSNick Piggin nd->seq = seq; 105931e6b01fSNick Piggin break; 106031e6b01fSNick Piggin } 106131e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 106231e6b01fSNick Piggin break; 106331e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 106431e6b01fSNick Piggin inode = nd->path.dentry->d_inode; 106531e6b01fSNick Piggin } 10669875cf80SDavid Howells __follow_mount_rcu(nd, &nd->path, &inode, true); 106731e6b01fSNick Piggin nd->inode = inode; 106831e6b01fSNick Piggin 106931e6b01fSNick Piggin return 0; 107031e6b01fSNick Piggin } 107131e6b01fSNick Piggin 10729875cf80SDavid Howells /* 1073cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1074cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1075cc53ce53SDavid Howells * caller is permitted to proceed or not. 1076cc53ce53SDavid Howells * 1077cc53ce53SDavid Howells * Care must be taken as namespace_sem may be held (indicated by mounting_here 1078cc53ce53SDavid Howells * being true). 1079cc53ce53SDavid Howells */ 1080cc53ce53SDavid Howells int follow_down(struct path *path, bool mounting_here) 1081cc53ce53SDavid Howells { 1082cc53ce53SDavid Howells unsigned managed; 1083cc53ce53SDavid Howells int ret; 1084cc53ce53SDavid Howells 1085cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1086cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1087cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1088cc53ce53SDavid Howells * being held. 1089cc53ce53SDavid Howells * 1090cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1091cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1092cc53ce53SDavid Howells * other than its daemon the right to mount on its 1093cc53ce53SDavid Howells * superstructure. 1094cc53ce53SDavid Howells * 1095cc53ce53SDavid Howells * The filesystem may sleep at this point. 1096cc53ce53SDavid Howells */ 1097cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1098cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1099cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1100ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 1101ab90911fSDavid Howells path->dentry, mounting_here, false); 1102cc53ce53SDavid Howells if (ret < 0) 1103cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1104cc53ce53SDavid Howells } 1105cc53ce53SDavid Howells 1106cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1107cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1108cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1109cc53ce53SDavid Howells if (!mounted) 1110cc53ce53SDavid Howells break; 1111cc53ce53SDavid Howells dput(path->dentry); 1112cc53ce53SDavid Howells mntput(path->mnt); 1113cc53ce53SDavid Howells path->mnt = mounted; 1114cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1115cc53ce53SDavid Howells continue; 1116cc53ce53SDavid Howells } 1117cc53ce53SDavid Howells 1118cc53ce53SDavid Howells /* Don't handle automount points here */ 1119cc53ce53SDavid Howells break; 1120cc53ce53SDavid Howells } 1121cc53ce53SDavid Howells return 0; 1122cc53ce53SDavid Howells } 1123cc53ce53SDavid Howells 1124cc53ce53SDavid Howells /* 11259875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 11269875cf80SDavid Howells */ 11279875cf80SDavid Howells static void follow_mount(struct path *path) 11289875cf80SDavid Howells { 11299875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 11309875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11319875cf80SDavid Howells if (!mounted) 11329875cf80SDavid Howells break; 11339875cf80SDavid Howells dput(path->dentry); 11349875cf80SDavid Howells mntput(path->mnt); 11359875cf80SDavid Howells path->mnt = mounted; 11369875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 11379875cf80SDavid Howells } 11389875cf80SDavid Howells } 11399875cf80SDavid Howells 114031e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 11411da177e4SLinus Torvalds { 11422a737871SAl Viro set_root(nd); 1143e518ddb7SAndreas Mohr 11441da177e4SLinus Torvalds while(1) { 11454ac91378SJan Blunck struct dentry *old = nd->path.dentry; 11461da177e4SLinus Torvalds 11472a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 11482a737871SAl Viro nd->path.mnt == nd->root.mnt) { 11491da177e4SLinus Torvalds break; 11501da177e4SLinus Torvalds } 11514ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 11523088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 11533088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 11541da177e4SLinus Torvalds dput(old); 11551da177e4SLinus Torvalds break; 11561da177e4SLinus Torvalds } 11573088dd70SAl Viro if (!follow_up(&nd->path)) 11581da177e4SLinus Torvalds break; 11591da177e4SLinus Torvalds } 116079ed0226SAl Viro follow_mount(&nd->path); 116131e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 11621da177e4SLinus Torvalds } 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds /* 1165baa03890SNick Piggin * Allocate a dentry with name and parent, and perform a parent 1166baa03890SNick Piggin * directory ->lookup on it. Returns the new dentry, or ERR_PTR 1167baa03890SNick Piggin * on error. parent->d_inode->i_mutex must be held. d_lookup must 1168baa03890SNick Piggin * have verified that no child exists while under i_mutex. 1169baa03890SNick Piggin */ 1170baa03890SNick Piggin static struct dentry *d_alloc_and_lookup(struct dentry *parent, 1171baa03890SNick Piggin struct qstr *name, struct nameidata *nd) 1172baa03890SNick Piggin { 1173baa03890SNick Piggin struct inode *inode = parent->d_inode; 1174baa03890SNick Piggin struct dentry *dentry; 1175baa03890SNick Piggin struct dentry *old; 1176baa03890SNick Piggin 1177baa03890SNick Piggin /* Don't create child dentry for a dead directory. */ 1178baa03890SNick Piggin if (unlikely(IS_DEADDIR(inode))) 1179baa03890SNick Piggin return ERR_PTR(-ENOENT); 1180baa03890SNick Piggin 1181baa03890SNick Piggin dentry = d_alloc(parent, name); 1182baa03890SNick Piggin if (unlikely(!dentry)) 1183baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1184baa03890SNick Piggin 1185baa03890SNick Piggin old = inode->i_op->lookup(inode, dentry, nd); 1186baa03890SNick Piggin if (unlikely(old)) { 1187baa03890SNick Piggin dput(dentry); 1188baa03890SNick Piggin dentry = old; 1189baa03890SNick Piggin } 1190baa03890SNick Piggin return dentry; 1191baa03890SNick Piggin } 1192baa03890SNick Piggin 1193baa03890SNick Piggin /* 11941da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 11951da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 11961da177e4SLinus Torvalds * It _is_ time-critical. 11971da177e4SLinus Torvalds */ 11981da177e4SLinus Torvalds static int do_lookup(struct nameidata *nd, struct qstr *name, 119931e6b01fSNick Piggin struct path *path, struct inode **inode) 12001da177e4SLinus Torvalds { 12014ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 120231e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 12036e6b1bd1SAl Viro struct inode *dir; 12049875cf80SDavid Howells int err; 12059875cf80SDavid Howells 12063cac260aSAl Viro /* 12073cac260aSAl Viro * See if the low-level filesystem might want 12083cac260aSAl Viro * to use its own hash.. 12093cac260aSAl Viro */ 1210fb045adbSNick Piggin if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 12119875cf80SDavid Howells err = parent->d_op->d_hash(parent, nd->inode, name); 12123cac260aSAl Viro if (err < 0) 12133cac260aSAl Viro return err; 12143cac260aSAl Viro } 12151da177e4SLinus Torvalds 1216b04f784eSNick Piggin /* 1217b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1218b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1219b04f784eSNick Piggin * do the non-racy lookup, below. 1220b04f784eSNick Piggin */ 122131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 122231e6b01fSNick Piggin unsigned seq; 122331e6b01fSNick Piggin 122431e6b01fSNick Piggin *inode = nd->inode; 122531e6b01fSNick Piggin dentry = __d_lookup_rcu(parent, name, &seq, inode); 122631e6b01fSNick Piggin if (!dentry) { 122731e6b01fSNick Piggin if (nameidata_drop_rcu(nd)) 122831e6b01fSNick Piggin return -ECHILD; 122931e6b01fSNick Piggin goto need_lookup; 123031e6b01fSNick Piggin } 123131e6b01fSNick Piggin /* Memory barrier in read_seqcount_begin of child is enough */ 123231e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 123331e6b01fSNick Piggin return -ECHILD; 123431e6b01fSNick Piggin 123531e6b01fSNick Piggin nd->seq = seq; 123624643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 1237f5e1c1c1SAl Viro dentry = do_revalidate_rcu(dentry, nd); 123824643087SAl Viro if (!dentry) 123924643087SAl Viro goto need_lookup; 124024643087SAl Viro if (IS_ERR(dentry)) 124124643087SAl Viro goto fail; 124224643087SAl Viro if (!(nd->flags & LOOKUP_RCU)) 124324643087SAl Viro goto done; 124424643087SAl Viro } 124531e6b01fSNick Piggin path->mnt = mnt; 124631e6b01fSNick Piggin path->dentry = dentry; 12479875cf80SDavid Howells if (likely(__follow_mount_rcu(nd, path, inode, false))) 12489875cf80SDavid Howells return 0; 12499875cf80SDavid Howells if (nameidata_drop_rcu(nd)) 12509875cf80SDavid Howells return -ECHILD; 12519875cf80SDavid Howells /* fallthru */ 12529875cf80SDavid Howells } 125331e6b01fSNick Piggin dentry = __d_lookup(parent, name); 12541da177e4SLinus Torvalds if (!dentry) 12551da177e4SLinus Torvalds goto need_lookup; 12562e2e88eaSNick Piggin found: 125724643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 125824643087SAl Viro dentry = do_revalidate(dentry, nd); 125924643087SAl Viro if (!dentry) 126024643087SAl Viro goto need_lookup; 126124643087SAl Viro if (IS_ERR(dentry)) 126224643087SAl Viro goto fail; 126324643087SAl Viro } 12641da177e4SLinus Torvalds done: 12651da177e4SLinus Torvalds path->mnt = mnt; 12661da177e4SLinus Torvalds path->dentry = dentry; 12679875cf80SDavid Howells err = follow_managed(path, nd->flags); 126889312214SIan Kent if (unlikely(err < 0)) { 126989312214SIan Kent path_put_conditional(path, nd); 12709875cf80SDavid Howells return err; 127189312214SIan Kent } 127231e6b01fSNick Piggin *inode = path->dentry->d_inode; 12731da177e4SLinus Torvalds return 0; 12741da177e4SLinus Torvalds 12751da177e4SLinus Torvalds need_lookup: 12766e6b1bd1SAl Viro dir = parent->d_inode; 127731e6b01fSNick Piggin BUG_ON(nd->inode != dir); 12786e6b1bd1SAl Viro 12796e6b1bd1SAl Viro mutex_lock(&dir->i_mutex); 12806e6b1bd1SAl Viro /* 12816e6b1bd1SAl Viro * First re-do the cached lookup just in case it was created 1282b04f784eSNick Piggin * while we waited for the directory semaphore, or the first 1283b04f784eSNick Piggin * lookup failed due to an unrelated rename. 12846e6b1bd1SAl Viro * 1285b04f784eSNick Piggin * This could use version numbering or similar to avoid unnecessary 1286b04f784eSNick Piggin * cache lookups, but then we'd have to do the first lookup in the 1287b04f784eSNick Piggin * non-racy way. However in the common case here, everything should 1288b04f784eSNick Piggin * be hot in cache, so would it be a big win? 12896e6b1bd1SAl Viro */ 12906e6b1bd1SAl Viro dentry = d_lookup(parent, name); 1291baa03890SNick Piggin if (likely(!dentry)) { 1292baa03890SNick Piggin dentry = d_alloc_and_lookup(parent, name, nd); 12936e6b1bd1SAl Viro mutex_unlock(&dir->i_mutex); 12946e6b1bd1SAl Viro if (IS_ERR(dentry)) 12956e6b1bd1SAl Viro goto fail; 12966e6b1bd1SAl Viro goto done; 12976e6b1bd1SAl Viro } 12986e6b1bd1SAl Viro /* 12996e6b1bd1SAl Viro * Uhhuh! Nasty case: the cache was re-populated while 13006e6b1bd1SAl Viro * we waited on the semaphore. Need to revalidate. 13016e6b1bd1SAl Viro */ 13026e6b1bd1SAl Viro mutex_unlock(&dir->i_mutex); 13032e2e88eaSNick Piggin goto found; 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds fail: 13061da177e4SLinus Torvalds return PTR_ERR(dentry); 13071da177e4SLinus Torvalds } 13081da177e4SLinus Torvalds 130952094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 131052094c8aSAl Viro { 131152094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 131252094c8aSAl Viro int err = exec_permission(nd->inode, IPERM_FLAG_RCU); 131352094c8aSAl Viro if (err != -ECHILD) 131452094c8aSAl Viro return err; 131552094c8aSAl Viro if (nameidata_drop_rcu(nd)) 131652094c8aSAl Viro return -ECHILD; 131752094c8aSAl Viro } 131852094c8aSAl Viro return exec_permission(nd->inode, 0); 131952094c8aSAl Viro } 132052094c8aSAl Viro 13211da177e4SLinus Torvalds /* 13221da177e4SLinus Torvalds * Name resolution. 1323ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1324ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 13251da177e4SLinus Torvalds * 1326ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1327ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 13281da177e4SLinus Torvalds */ 13296de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 13301da177e4SLinus Torvalds { 13311da177e4SLinus Torvalds struct path next; 13321da177e4SLinus Torvalds int err; 13331da177e4SLinus Torvalds unsigned int lookup_flags = nd->flags; 13341da177e4SLinus Torvalds 13351da177e4SLinus Torvalds while (*name=='/') 13361da177e4SLinus Torvalds name++; 13371da177e4SLinus Torvalds if (!*name) 1338086e183aSAl Viro return 0; 13391da177e4SLinus Torvalds 13401da177e4SLinus Torvalds if (nd->depth) 1341f55eab82STrond Myklebust lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); 13421da177e4SLinus Torvalds 13431da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 13441da177e4SLinus Torvalds for(;;) { 134531e6b01fSNick Piggin struct inode *inode; 13461da177e4SLinus Torvalds unsigned long hash; 13471da177e4SLinus Torvalds struct qstr this; 13481da177e4SLinus Torvalds unsigned int c; 1349fe479a58SAl Viro int type; 13501da177e4SLinus Torvalds 1351cdce5d6bSTrond Myklebust nd->flags |= LOOKUP_CONTINUE; 135252094c8aSAl Viro 135352094c8aSAl Viro err = may_lookup(nd); 13541da177e4SLinus Torvalds if (err) 13551da177e4SLinus Torvalds break; 13561da177e4SLinus Torvalds 13571da177e4SLinus Torvalds this.name = name; 13581da177e4SLinus Torvalds c = *(const unsigned char *)name; 13591da177e4SLinus Torvalds 13601da177e4SLinus Torvalds hash = init_name_hash(); 13611da177e4SLinus Torvalds do { 13621da177e4SLinus Torvalds name++; 13631da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 13641da177e4SLinus Torvalds c = *(const unsigned char *)name; 13651da177e4SLinus Torvalds } while (c && (c != '/')); 13661da177e4SLinus Torvalds this.len = name - (const char *) this.name; 13671da177e4SLinus Torvalds this.hash = end_name_hash(hash); 13681da177e4SLinus Torvalds 1369fe479a58SAl Viro type = LAST_NORM; 1370fe479a58SAl Viro if (this.name[0] == '.') switch (this.len) { 1371fe479a58SAl Viro case 2: 137216c2cd71SAl Viro if (this.name[1] == '.') { 1373fe479a58SAl Viro type = LAST_DOTDOT; 137416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 137516c2cd71SAl Viro } 1376fe479a58SAl Viro break; 1377fe479a58SAl Viro case 1: 1378fe479a58SAl Viro type = LAST_DOT; 1379fe479a58SAl Viro } 138016c2cd71SAl Viro if (likely(type == LAST_NORM)) 138116c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 1382fe479a58SAl Viro 13831da177e4SLinus Torvalds /* remove trailing slashes? */ 13841da177e4SLinus Torvalds if (!c) 13851da177e4SLinus Torvalds goto last_component; 13861da177e4SLinus Torvalds while (*++name == '/'); 13871da177e4SLinus Torvalds if (!*name) 13881da177e4SLinus Torvalds goto last_with_slashes; 13891da177e4SLinus Torvalds 13901da177e4SLinus Torvalds /* 13911da177e4SLinus Torvalds * "." and ".." are special - ".." especially so because it has 13921da177e4SLinus Torvalds * to be able to know about the current root directory and 13931da177e4SLinus Torvalds * parent relationships. 13941da177e4SLinus Torvalds */ 1395fe479a58SAl Viro if (unlikely(type != LAST_NORM)) { 1396fe479a58SAl Viro if (type == LAST_DOTDOT) { 139731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 139831e6b01fSNick Piggin if (follow_dotdot_rcu(nd)) 139931e6b01fSNick Piggin return -ECHILD; 140031e6b01fSNick Piggin } else 140158c465ebSAl Viro follow_dotdot(nd); 1402fe479a58SAl Viro } 14031da177e4SLinus Torvalds continue; 14041da177e4SLinus Torvalds } 1405fe479a58SAl Viro 14061da177e4SLinus Torvalds /* This does the actual lookups.. */ 140731e6b01fSNick Piggin err = do_lookup(nd, &this, &next, &inode); 14081da177e4SLinus Torvalds if (err) 14091da177e4SLinus Torvalds break; 14101da177e4SLinus Torvalds err = -ENOENT; 14111da177e4SLinus Torvalds if (!inode) 14121da177e4SLinus Torvalds goto out_dput; 14131da177e4SLinus Torvalds 14141da177e4SLinus Torvalds if (inode->i_op->follow_link) { 14153abb17e8SLinus Torvalds err = do_follow_link(inode, &next, nd); 14161da177e4SLinus Torvalds if (err) 14171da177e4SLinus Torvalds goto return_err; 141831e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 14191da177e4SLinus Torvalds err = -ENOENT; 142031e6b01fSNick Piggin if (!nd->inode) 14211da177e4SLinus Torvalds break; 142231e6b01fSNick Piggin } else { 142309dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 142431e6b01fSNick Piggin nd->inode = inode; 142531e6b01fSNick Piggin } 14261da177e4SLinus Torvalds err = -ENOTDIR; 142731e6b01fSNick Piggin if (!nd->inode->i_op->lookup) 14281da177e4SLinus Torvalds break; 14291da177e4SLinus Torvalds continue; 14301da177e4SLinus Torvalds /* here ends the main loop */ 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds last_with_slashes: 14331da177e4SLinus Torvalds lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 14341da177e4SLinus Torvalds last_component: 1435f55eab82STrond Myklebust /* Clear LOOKUP_CONTINUE iff it was previously unset */ 1436f55eab82STrond Myklebust nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; 14371da177e4SLinus Torvalds if (lookup_flags & LOOKUP_PARENT) 14381da177e4SLinus Torvalds goto lookup_parent; 1439fe479a58SAl Viro if (unlikely(type != LAST_NORM)) { 1440fe479a58SAl Viro if (type == LAST_DOTDOT) { 144131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 144231e6b01fSNick Piggin if (follow_dotdot_rcu(nd)) 144331e6b01fSNick Piggin return -ECHILD; 144431e6b01fSNick Piggin } else 144558c465ebSAl Viro follow_dotdot(nd); 1446fe479a58SAl Viro } 1447086e183aSAl Viro return 0; 14481da177e4SLinus Torvalds } 144931e6b01fSNick Piggin err = do_lookup(nd, &this, &next, &inode); 14501da177e4SLinus Torvalds if (err) 14511da177e4SLinus Torvalds break; 1452db372915SDavid Howells if (inode && unlikely(inode->i_op->follow_link) && 1453db372915SDavid Howells (lookup_flags & LOOKUP_FOLLOW)) { 14543abb17e8SLinus Torvalds err = do_follow_link(inode, &next, nd); 14551da177e4SLinus Torvalds if (err) 14561da177e4SLinus Torvalds goto return_err; 145731e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 145831e6b01fSNick Piggin } else { 145909dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 146031e6b01fSNick Piggin nd->inode = inode; 146131e6b01fSNick Piggin } 14621da177e4SLinus Torvalds err = -ENOENT; 146331e6b01fSNick Piggin if (!nd->inode) 14641da177e4SLinus Torvalds break; 14651da177e4SLinus Torvalds if (lookup_flags & LOOKUP_DIRECTORY) { 14661da177e4SLinus Torvalds err = -ENOTDIR; 146731e6b01fSNick Piggin if (!nd->inode->i_op->lookup) 14681da177e4SLinus Torvalds break; 14691da177e4SLinus Torvalds } 1470086e183aSAl Viro return 0; 14711da177e4SLinus Torvalds lookup_parent: 14721da177e4SLinus Torvalds nd->last = this; 1473fe479a58SAl Viro nd->last_type = type; 14741da177e4SLinus Torvalds return 0; 14751da177e4SLinus Torvalds out_dput: 147631e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) 14771d957f9bSJan Blunck path_put_conditional(&next, nd); 14781da177e4SLinus Torvalds break; 14791da177e4SLinus Torvalds } 148031e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) 14811d957f9bSJan Blunck path_put(&nd->path); 14821da177e4SLinus Torvalds return_err: 14831da177e4SLinus Torvalds return err; 14841da177e4SLinus Torvalds } 14851da177e4SLinus Torvalds 1486e41f7d4eSAl Viro static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd) 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; 149531e6b01fSNick Piggin nd->root.mnt = NULL; 149631e6b01fSNick Piggin nd->file = NULL; 149731e6b01fSNick Piggin 149831e6b01fSNick Piggin if (*name=='/') { 1499e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 150031e6b01fSNick Piggin br_read_lock(vfsmount_lock); 150131e6b01fSNick Piggin rcu_read_lock(); 1502e41f7d4eSAl Viro set_root_rcu(nd); 1503e41f7d4eSAl Viro } else { 1504e41f7d4eSAl Viro set_root(nd); 1505e41f7d4eSAl Viro path_get(&nd->root); 1506e41f7d4eSAl Viro } 150731e6b01fSNick Piggin nd->path = nd->root; 150831e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1509e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 151031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1511c28cc364SNick Piggin unsigned seq; 151231e6b01fSNick Piggin 151331e6b01fSNick Piggin br_read_lock(vfsmount_lock); 151431e6b01fSNick Piggin rcu_read_lock(); 151531e6b01fSNick Piggin 1516c28cc364SNick Piggin do { 1517c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 151831e6b01fSNick Piggin nd->path = fs->pwd; 1519c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1520c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1521e41f7d4eSAl Viro } else { 1522e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1523e41f7d4eSAl Viro } 152431e6b01fSNick Piggin } else { 152531e6b01fSNick Piggin struct dentry *dentry; 152631e6b01fSNick Piggin 152731e6b01fSNick Piggin file = fget_light(dfd, &fput_needed); 152831e6b01fSNick Piggin retval = -EBADF; 152931e6b01fSNick Piggin if (!file) 153031e6b01fSNick Piggin goto out_fail; 153131e6b01fSNick Piggin 153231e6b01fSNick Piggin dentry = file->f_path.dentry; 153331e6b01fSNick Piggin 153431e6b01fSNick Piggin retval = -ENOTDIR; 153531e6b01fSNick Piggin if (!S_ISDIR(dentry->d_inode->i_mode)) 153631e6b01fSNick Piggin goto fput_fail; 153731e6b01fSNick Piggin 153831e6b01fSNick Piggin retval = file_permission(file, MAY_EXEC); 153931e6b01fSNick Piggin if (retval) 154031e6b01fSNick Piggin goto fput_fail; 154131e6b01fSNick Piggin 154231e6b01fSNick Piggin nd->path = file->f_path; 1543e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 154431e6b01fSNick Piggin if (fput_needed) 154531e6b01fSNick Piggin nd->file = file; 1546c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 154731e6b01fSNick Piggin br_read_lock(vfsmount_lock); 154831e6b01fSNick Piggin rcu_read_lock(); 15495590ff0dSUlrich Drepper } else { 15505dd784d0SJan Blunck path_get(&file->f_path); 15515590ff0dSUlrich Drepper fput_light(file, fput_needed); 15521da177e4SLinus Torvalds } 1553e41f7d4eSAl Viro } 1554e41f7d4eSAl Viro 155531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 15569b4a9b14SAl Viro return 0; 15572dfdd266SJosef 'Jeff' Sipek 15589b4a9b14SAl Viro fput_fail: 15599b4a9b14SAl Viro fput_light(file, fput_needed); 15609b4a9b14SAl Viro out_fail: 15619b4a9b14SAl Viro return retval; 15629b4a9b14SAl Viro } 15639b4a9b14SAl Viro 15649b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1565ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 15669b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 15679b4a9b14SAl Viro { 156831e6b01fSNick Piggin int retval; 156931e6b01fSNick Piggin 157031e6b01fSNick Piggin /* 157131e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 157231e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 157331e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 157431e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 157531e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 157631e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 157731e6b01fSNick Piggin * analogue, foo_rcu(). 157831e6b01fSNick Piggin * 157931e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 158031e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 158131e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 158231e6b01fSNick Piggin * be able to complete). 158331e6b01fSNick Piggin */ 1584ee0827cdSAl Viro retval = path_init(dfd, name, flags, nd); 1585ee0827cdSAl Viro 158631e6b01fSNick Piggin if (unlikely(retval)) 158731e6b01fSNick Piggin return retval; 1588ee0827cdSAl Viro 1589ee0827cdSAl Viro current->total_link_count = 0; 1590ee0827cdSAl Viro retval = link_path_walk(name, nd); 1591ee0827cdSAl Viro 1592ee0827cdSAl Viro if (nd->flags & LOOKUP_RCU) { 1593ee0827cdSAl Viro /* RCU dangling. Cancel it. */ 1594086e183aSAl Viro if (!retval) { 1595086e183aSAl Viro if (nameidata_drop_rcu_last(nd)) 1596086e183aSAl Viro retval = -ECHILD; 1597086e183aSAl Viro } else { 1598ee0827cdSAl Viro nd->flags &= ~LOOKUP_RCU; 15992a737871SAl Viro nd->root.mnt = NULL; 1600ee0827cdSAl Viro rcu_read_unlock(); 1601ee0827cdSAl Viro br_read_unlock(vfsmount_lock); 16022a737871SAl Viro } 1603086e183aSAl Viro } 160431e6b01fSNick Piggin 160516c2cd71SAl Viro if (!retval) 160616c2cd71SAl Viro retval = handle_reval_path(nd); 160716c2cd71SAl Viro 1608ee0827cdSAl Viro if (nd->file) { 1609ee0827cdSAl Viro fput(nd->file); 1610ee0827cdSAl Viro nd->file = NULL; 1611ee0827cdSAl Viro } 1612ee0827cdSAl Viro 161331e6b01fSNick Piggin if (nd->root.mnt) { 161431e6b01fSNick Piggin path_put(&nd->root); 161531e6b01fSNick Piggin nd->root.mnt = NULL; 161631e6b01fSNick Piggin } 1617ee0827cdSAl Viro return retval; 161831e6b01fSNick Piggin } 161931e6b01fSNick Piggin 1620ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1621ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1622ee0827cdSAl Viro { 1623ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1624ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1625ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1626ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1627ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1628ee0827cdSAl Viro 162931e6b01fSNick Piggin if (likely(!retval)) { 163031e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 163131e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 163231e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 163331e6b01fSNick Piggin } 163431e6b01fSNick Piggin } 1635170aa3d0SUlrich Drepper return retval; 16361da177e4SLinus Torvalds } 16371da177e4SLinus Torvalds 1638c9c6cac0SAl Viro int kern_path_parent(const char *name, struct nameidata *nd) 16395590ff0dSUlrich Drepper { 1640c9c6cac0SAl Viro return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); 16415590ff0dSUlrich Drepper } 16425590ff0dSUlrich Drepper 1643d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1644d1811465SAl Viro { 1645d1811465SAl Viro struct nameidata nd; 1646d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1647d1811465SAl Viro if (!res) 1648d1811465SAl Viro *path = nd.path; 1649d1811465SAl Viro return res; 1650d1811465SAl Viro } 1651d1811465SAl Viro 165216f18200SJosef 'Jeff' Sipek /** 165316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 165416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 165516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 165616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 165716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 165816f18200SJosef 'Jeff' Sipek * @nd: pointer to nameidata 165916f18200SJosef 'Jeff' Sipek */ 166016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 166116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 166216f18200SJosef 'Jeff' Sipek struct nameidata *nd) 166316f18200SJosef 'Jeff' Sipek { 1664ee0827cdSAl Viro int result; 166516f18200SJosef 'Jeff' Sipek 166616f18200SJosef 'Jeff' Sipek /* same as do_path_lookup */ 166716f18200SJosef 'Jeff' Sipek nd->last_type = LAST_ROOT; 166816c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 166916f18200SJosef 'Jeff' Sipek nd->depth = 0; 167016f18200SJosef 'Jeff' Sipek 1671c8e7f449SJan Blunck nd->path.dentry = dentry; 1672c8e7f449SJan Blunck nd->path.mnt = mnt; 1673c8e7f449SJan Blunck path_get(&nd->path); 16745b857119SAl Viro nd->root = nd->path; 16755b857119SAl Viro path_get(&nd->root); 167631e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 167716f18200SJosef 'Jeff' Sipek 1678ee0827cdSAl Viro current->total_link_count = 0; 1679ee0827cdSAl Viro 1680ee0827cdSAl Viro result = link_path_walk(name, nd); 168116c2cd71SAl Viro if (!result) 168216c2cd71SAl Viro result = handle_reval_path(nd); 1683ee0827cdSAl Viro if (result == -ESTALE) { 1684ee0827cdSAl Viro /* nd->path had been dropped */ 1685ee0827cdSAl Viro current->total_link_count = 0; 1686ee0827cdSAl Viro nd->path.dentry = dentry; 1687ee0827cdSAl Viro nd->path.mnt = mnt; 1688ee0827cdSAl Viro nd->inode = dentry->d_inode; 1689ee0827cdSAl Viro path_get(&nd->path); 169016c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_REVAL; 169116c2cd71SAl Viro 1692ee0827cdSAl Viro result = link_path_walk(name, nd); 169316c2cd71SAl Viro if (!result) 169416c2cd71SAl Viro result = handle_reval_path(nd); 1695ee0827cdSAl Viro } 1696ee0827cdSAl Viro if (unlikely(!result && !audit_dummy_context() && nd->path.dentry && 169731e6b01fSNick Piggin nd->inode)) 16984ac91378SJan Blunck audit_inode(name, nd->path.dentry); 169916f18200SJosef 'Jeff' Sipek 17002a737871SAl Viro path_put(&nd->root); 17012a737871SAl Viro nd->root.mnt = NULL; 170216f18200SJosef 'Jeff' Sipek 1703ee0827cdSAl Viro return result; 170416f18200SJosef 'Jeff' Sipek } 170516f18200SJosef 'Jeff' Sipek 1706eead1911SChristoph Hellwig static struct dentry *__lookup_hash(struct qstr *name, 1707eead1911SChristoph Hellwig struct dentry *base, struct nameidata *nd) 17081da177e4SLinus Torvalds { 170981fca444SChristoph Hellwig struct inode *inode = base->d_inode; 17101da177e4SLinus Torvalds struct dentry *dentry; 17111da177e4SLinus Torvalds int err; 17121da177e4SLinus Torvalds 1713b74c79e9SNick Piggin err = exec_permission(inode, 0); 171481fca444SChristoph Hellwig if (err) 171581fca444SChristoph Hellwig return ERR_PTR(err); 17161da177e4SLinus Torvalds 17171da177e4SLinus Torvalds /* 17181da177e4SLinus Torvalds * See if the low-level filesystem might want 17191da177e4SLinus Torvalds * to use its own hash.. 17201da177e4SLinus Torvalds */ 1721fb045adbSNick Piggin if (base->d_flags & DCACHE_OP_HASH) { 1722b1e6a015SNick Piggin err = base->d_op->d_hash(base, inode, name); 17231da177e4SLinus Torvalds dentry = ERR_PTR(err); 17241da177e4SLinus Torvalds if (err < 0) 17251da177e4SLinus Torvalds goto out; 17261da177e4SLinus Torvalds } 17271da177e4SLinus Torvalds 1728b04f784eSNick Piggin /* 1729b04f784eSNick Piggin * Don't bother with __d_lookup: callers are for creat as 1730b04f784eSNick Piggin * well as unlink, so a lot of the time it would cost 1731b04f784eSNick Piggin * a double lookup. 17326e6b1bd1SAl Viro */ 17336e6b1bd1SAl Viro dentry = d_lookup(base, name); 17346e6b1bd1SAl Viro 1735fb045adbSNick Piggin if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) 17366e6b1bd1SAl Viro dentry = do_revalidate(dentry, nd); 17376e6b1bd1SAl Viro 17381da177e4SLinus Torvalds if (!dentry) 1739baa03890SNick Piggin dentry = d_alloc_and_lookup(base, name, nd); 17401da177e4SLinus Torvalds out: 17411da177e4SLinus Torvalds return dentry; 17421da177e4SLinus Torvalds } 17431da177e4SLinus Torvalds 1744057f6c01SJames Morris /* 1745057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1746057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1747057f6c01SJames Morris * SMP-safe. 1748057f6c01SJames Morris */ 1749a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 17501da177e4SLinus Torvalds { 17514ac91378SJan Blunck return __lookup_hash(&nd->last, nd->path.dentry, nd); 17521da177e4SLinus Torvalds } 17531da177e4SLinus Torvalds 1754eead1911SChristoph Hellwig static int __lookup_one_len(const char *name, struct qstr *this, 1755eead1911SChristoph Hellwig struct dentry *base, int len) 17561da177e4SLinus Torvalds { 17571da177e4SLinus Torvalds unsigned long hash; 17581da177e4SLinus Torvalds unsigned int c; 17591da177e4SLinus Torvalds 1760057f6c01SJames Morris this->name = name; 1761057f6c01SJames Morris this->len = len; 17621da177e4SLinus Torvalds if (!len) 1763057f6c01SJames Morris return -EACCES; 17641da177e4SLinus Torvalds 17651da177e4SLinus Torvalds hash = init_name_hash(); 17661da177e4SLinus Torvalds while (len--) { 17671da177e4SLinus Torvalds c = *(const unsigned char *)name++; 17681da177e4SLinus Torvalds if (c == '/' || c == '\0') 1769057f6c01SJames Morris return -EACCES; 17701da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 17711da177e4SLinus Torvalds } 1772057f6c01SJames Morris this->hash = end_name_hash(hash); 1773057f6c01SJames Morris return 0; 1774057f6c01SJames Morris } 17751da177e4SLinus Torvalds 1776eead1911SChristoph Hellwig /** 1777a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1778eead1911SChristoph Hellwig * @name: pathname component to lookup 1779eead1911SChristoph Hellwig * @base: base directory to lookup from 1780eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1781eead1911SChristoph Hellwig * 1782a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1783a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1784eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1785eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1786eead1911SChristoph Hellwig */ 1787057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1788057f6c01SJames Morris { 1789057f6c01SJames Morris int err; 1790057f6c01SJames Morris struct qstr this; 1791057f6c01SJames Morris 17922f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 17932f9092e1SDavid Woodhouse 1794057f6c01SJames Morris err = __lookup_one_len(name, &this, base, len); 1795057f6c01SJames Morris if (err) 1796057f6c01SJames Morris return ERR_PTR(err); 1797eead1911SChristoph Hellwig 179849705b77SChristoph Hellwig return __lookup_hash(&this, base, NULL); 1799057f6c01SJames Morris } 1800057f6c01SJames Morris 18012d8f3038SAl Viro int user_path_at(int dfd, const char __user *name, unsigned flags, 18022d8f3038SAl Viro struct path *path) 18031da177e4SLinus Torvalds { 18042d8f3038SAl Viro struct nameidata nd; 18051da177e4SLinus Torvalds char *tmp = getname(name); 18061da177e4SLinus Torvalds int err = PTR_ERR(tmp); 18071da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 18082d8f3038SAl Viro 18092d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 18102d8f3038SAl Viro 18112d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 18121da177e4SLinus Torvalds putname(tmp); 18132d8f3038SAl Viro if (!err) 18142d8f3038SAl Viro *path = nd.path; 18151da177e4SLinus Torvalds } 18161da177e4SLinus Torvalds return err; 18171da177e4SLinus Torvalds } 18181da177e4SLinus Torvalds 18192ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 18202ad94ae6SAl Viro struct nameidata *nd, char **name) 18212ad94ae6SAl Viro { 18222ad94ae6SAl Viro char *s = getname(path); 18232ad94ae6SAl Viro int error; 18242ad94ae6SAl Viro 18252ad94ae6SAl Viro if (IS_ERR(s)) 18262ad94ae6SAl Viro return PTR_ERR(s); 18272ad94ae6SAl Viro 18282ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 18292ad94ae6SAl Viro if (error) 18302ad94ae6SAl Viro putname(s); 18312ad94ae6SAl Viro else 18322ad94ae6SAl Viro *name = s; 18332ad94ae6SAl Viro 18342ad94ae6SAl Viro return error; 18352ad94ae6SAl Viro } 18362ad94ae6SAl Viro 18371da177e4SLinus Torvalds /* 18381da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 18391da177e4SLinus Torvalds * minimal. 18401da177e4SLinus Torvalds */ 18411da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 18421da177e4SLinus Torvalds { 1843da9592edSDavid Howells uid_t fsuid = current_fsuid(); 1844da9592edSDavid Howells 18451da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 18461da177e4SLinus Torvalds return 0; 1847da9592edSDavid Howells if (inode->i_uid == fsuid) 18481da177e4SLinus Torvalds return 0; 1849da9592edSDavid Howells if (dir->i_uid == fsuid) 18501da177e4SLinus Torvalds return 0; 18511da177e4SLinus Torvalds return !capable(CAP_FOWNER); 18521da177e4SLinus Torvalds } 18531da177e4SLinus Torvalds 18541da177e4SLinus Torvalds /* 18551da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 18561da177e4SLinus Torvalds * whether the type of victim is right. 18571da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 18581da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 18591da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 18601da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 18611da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 18621da177e4SLinus Torvalds * a. be owner of dir, or 18631da177e4SLinus Torvalds * b. be owner of victim, or 18641da177e4SLinus Torvalds * c. have CAP_FOWNER capability 18651da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 18661da177e4SLinus Torvalds * links pointing to it. 18671da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 18681da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 18691da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 18701da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 18711da177e4SLinus Torvalds * nfs_async_unlink(). 18721da177e4SLinus Torvalds */ 1873858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 18741da177e4SLinus Torvalds { 18751da177e4SLinus Torvalds int error; 18761da177e4SLinus Torvalds 18771da177e4SLinus Torvalds if (!victim->d_inode) 18781da177e4SLinus Torvalds return -ENOENT; 18791da177e4SLinus Torvalds 18801da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 1881cccc6bbaSAl Viro audit_inode_child(victim, dir); 18821da177e4SLinus Torvalds 1883f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 18841da177e4SLinus Torvalds if (error) 18851da177e4SLinus Torvalds return error; 18861da177e4SLinus Torvalds if (IS_APPEND(dir)) 18871da177e4SLinus Torvalds return -EPERM; 18881da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 1889f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 18901da177e4SLinus Torvalds return -EPERM; 18911da177e4SLinus Torvalds if (isdir) { 18921da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 18931da177e4SLinus Torvalds return -ENOTDIR; 18941da177e4SLinus Torvalds if (IS_ROOT(victim)) 18951da177e4SLinus Torvalds return -EBUSY; 18961da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 18971da177e4SLinus Torvalds return -EISDIR; 18981da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 18991da177e4SLinus Torvalds return -ENOENT; 19001da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 19011da177e4SLinus Torvalds return -EBUSY; 19021da177e4SLinus Torvalds return 0; 19031da177e4SLinus Torvalds } 19041da177e4SLinus Torvalds 19051da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 19061da177e4SLinus Torvalds * dir. 19071da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 19081da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 19091da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 19101da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 19111da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 19121da177e4SLinus Torvalds */ 1913a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 19141da177e4SLinus Torvalds { 19151da177e4SLinus Torvalds if (child->d_inode) 19161da177e4SLinus Torvalds return -EEXIST; 19171da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 19181da177e4SLinus Torvalds return -ENOENT; 1919f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 19201da177e4SLinus Torvalds } 19211da177e4SLinus Torvalds 19221da177e4SLinus Torvalds /* 19231da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 19241da177e4SLinus Torvalds */ 19251da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 19261da177e4SLinus Torvalds { 19271da177e4SLinus Torvalds struct dentry *p; 19281da177e4SLinus Torvalds 19291da177e4SLinus Torvalds if (p1 == p2) { 1930f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 19311da177e4SLinus Torvalds return NULL; 19321da177e4SLinus Torvalds } 19331da177e4SLinus Torvalds 1934a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19351da177e4SLinus Torvalds 1936e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 1937e2761a11SOGAWA Hirofumi if (p) { 1938f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1939f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 19401da177e4SLinus Torvalds return p; 19411da177e4SLinus Torvalds } 19421da177e4SLinus Torvalds 1943e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 1944e2761a11SOGAWA Hirofumi if (p) { 1945f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1946f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19471da177e4SLinus Torvalds return p; 19481da177e4SLinus Torvalds } 19491da177e4SLinus Torvalds 1950f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1951f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19521da177e4SLinus Torvalds return NULL; 19531da177e4SLinus Torvalds } 19541da177e4SLinus Torvalds 19551da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 19561da177e4SLinus Torvalds { 19571b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 19581da177e4SLinus Torvalds if (p1 != p2) { 19591b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 1960a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19611da177e4SLinus Torvalds } 19621da177e4SLinus Torvalds } 19631da177e4SLinus Torvalds 19641da177e4SLinus Torvalds int vfs_create(struct inode *dir, struct dentry *dentry, int mode, 19651da177e4SLinus Torvalds struct nameidata *nd) 19661da177e4SLinus Torvalds { 1967a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 19681da177e4SLinus Torvalds 19691da177e4SLinus Torvalds if (error) 19701da177e4SLinus Torvalds return error; 19711da177e4SLinus Torvalds 1972acfa4380SAl Viro if (!dir->i_op->create) 19731da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 19741da177e4SLinus Torvalds mode &= S_IALLUGO; 19751da177e4SLinus Torvalds mode |= S_IFREG; 19761da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 19771da177e4SLinus Torvalds if (error) 19781da177e4SLinus Torvalds return error; 19791da177e4SLinus Torvalds error = dir->i_op->create(dir, dentry, mode, nd); 1980a74574aaSStephen Smalley if (!error) 1981f38aa942SAmy Griffis fsnotify_create(dir, dentry); 19821da177e4SLinus Torvalds return error; 19831da177e4SLinus Torvalds } 19841da177e4SLinus Torvalds 19853fb64190SChristoph Hellwig int may_open(struct path *path, int acc_mode, int flag) 19861da177e4SLinus Torvalds { 19873fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 19881da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 19891da177e4SLinus Torvalds int error; 19901da177e4SLinus Torvalds 19911da177e4SLinus Torvalds if (!inode) 19921da177e4SLinus Torvalds return -ENOENT; 19931da177e4SLinus Torvalds 1994c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 1995c8fe8f30SChristoph Hellwig case S_IFLNK: 19961da177e4SLinus Torvalds return -ELOOP; 1997c8fe8f30SChristoph Hellwig case S_IFDIR: 1998c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 19991da177e4SLinus Torvalds return -EISDIR; 2000c8fe8f30SChristoph Hellwig break; 2001c8fe8f30SChristoph Hellwig case S_IFBLK: 2002c8fe8f30SChristoph Hellwig case S_IFCHR: 20033fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 20041da177e4SLinus Torvalds return -EACCES; 2005c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2006c8fe8f30SChristoph Hellwig case S_IFIFO: 2007c8fe8f30SChristoph Hellwig case S_IFSOCK: 20081da177e4SLinus Torvalds flag &= ~O_TRUNC; 2009c8fe8f30SChristoph Hellwig break; 20104a3fd211SDave Hansen } 2011b41572e9SDave Hansen 20123fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2013b41572e9SDave Hansen if (error) 2014b41572e9SDave Hansen return error; 20156146f0d5SMimi Zohar 20161da177e4SLinus Torvalds /* 20171da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 20181da177e4SLinus Torvalds */ 20191da177e4SLinus Torvalds if (IS_APPEND(inode)) { 20208737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 20217715b521SAl Viro return -EPERM; 20221da177e4SLinus Torvalds if (flag & O_TRUNC) 20237715b521SAl Viro return -EPERM; 20241da177e4SLinus Torvalds } 20251da177e4SLinus Torvalds 20261da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 20277715b521SAl Viro if (flag & O_NOATIME && !is_owner_or_cap(inode)) 20287715b521SAl Viro return -EPERM; 20291da177e4SLinus Torvalds 20301da177e4SLinus Torvalds /* 20311da177e4SLinus Torvalds * Ensure there are no outstanding leases on the file. 20321da177e4SLinus Torvalds */ 2033b65a9cfcSAl Viro return break_lease(inode, flag); 20347715b521SAl Viro } 20357715b521SAl Viro 2036e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 20377715b521SAl Viro { 2038e1181ee6SJeff Layton struct path *path = &filp->f_path; 20397715b521SAl Viro struct inode *inode = path->dentry->d_inode; 20407715b521SAl Viro int error = get_write_access(inode); 20411da177e4SLinus Torvalds if (error) 20427715b521SAl Viro return error; 20431da177e4SLinus Torvalds /* 20441da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 20451da177e4SLinus Torvalds */ 20461da177e4SLinus Torvalds error = locks_verify_locked(inode); 2047be6d3e56SKentaro Takeda if (!error) 2048ea0d3ab2STetsuo Handa error = security_path_truncate(path); 20491da177e4SLinus Torvalds if (!error) { 20507715b521SAl Viro error = do_truncate(path->dentry, 0, 2051d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2052e1181ee6SJeff Layton filp); 20531da177e4SLinus Torvalds } 20541da177e4SLinus Torvalds put_write_access(inode); 2055acd0c935SMimi Zohar return error; 20561da177e4SLinus Torvalds } 20571da177e4SLinus Torvalds 2058d57999e1SDave Hansen /* 2059d57999e1SDave Hansen * Be careful about ever adding any more callers of this 2060d57999e1SDave Hansen * function. Its flags must be in the namei format, not 2061d57999e1SDave Hansen * what get passed to sys_open(). 2062d57999e1SDave Hansen */ 2063d57999e1SDave Hansen static int __open_namei_create(struct nameidata *nd, struct path *path, 20648737c930SAl Viro int open_flag, int mode) 2065aab520e2SDave Hansen { 2066aab520e2SDave Hansen int error; 20674ac91378SJan Blunck struct dentry *dir = nd->path.dentry; 2068aab520e2SDave Hansen 2069aab520e2SDave Hansen if (!IS_POSIXACL(dir->d_inode)) 2070ce3b0f8dSAl Viro mode &= ~current_umask(); 2071be6d3e56SKentaro Takeda error = security_path_mknod(&nd->path, path->dentry, mode, 0); 2072be6d3e56SKentaro Takeda if (error) 2073be6d3e56SKentaro Takeda goto out_unlock; 2074aab520e2SDave Hansen error = vfs_create(dir->d_inode, path->dentry, mode, nd); 2075be6d3e56SKentaro Takeda out_unlock: 2076aab520e2SDave Hansen mutex_unlock(&dir->d_inode->i_mutex); 20774ac91378SJan Blunck dput(nd->path.dentry); 20784ac91378SJan Blunck nd->path.dentry = path->dentry; 207931e6b01fSNick Piggin 2080aab520e2SDave Hansen if (error) 2081aab520e2SDave Hansen return error; 2082aab520e2SDave Hansen /* Don't check for write permission, don't truncate */ 20838737c930SAl Viro return may_open(&nd->path, 0, open_flag & ~O_TRUNC); 2084aab520e2SDave Hansen } 2085aab520e2SDave Hansen 20861da177e4SLinus Torvalds /* 2087d57999e1SDave Hansen * Note that while the flag value (low two bits) for sys_open means: 2088d57999e1SDave Hansen * 00 - read-only 2089d57999e1SDave Hansen * 01 - write-only 2090d57999e1SDave Hansen * 10 - read-write 2091d57999e1SDave Hansen * 11 - special 2092d57999e1SDave Hansen * it is changed into 2093d57999e1SDave Hansen * 00 - no permissions needed 2094d57999e1SDave Hansen * 01 - read-permission 2095d57999e1SDave Hansen * 10 - write-permission 2096d57999e1SDave Hansen * 11 - read-write 2097d57999e1SDave Hansen * for the internal routines (ie open_namei()/follow_link() etc) 2098d57999e1SDave Hansen * This is more logical, and also allows the 00 "no perm needed" 2099d57999e1SDave Hansen * to be used for symlinks (where the permissions are checked 2100d57999e1SDave Hansen * later). 2101d57999e1SDave Hansen * 2102d57999e1SDave Hansen */ 2103d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2104d57999e1SDave Hansen { 2105d57999e1SDave Hansen if ((flag+1) & O_ACCMODE) 2106d57999e1SDave Hansen flag++; 2107d57999e1SDave Hansen return flag; 2108d57999e1SDave Hansen } 2109d57999e1SDave Hansen 21107715b521SAl Viro static int open_will_truncate(int flag, struct inode *inode) 21114a3fd211SDave Hansen { 2112d57999e1SDave Hansen /* 21134a3fd211SDave Hansen * We'll never write to the fs underlying 21144a3fd211SDave Hansen * a device file. 21154a3fd211SDave Hansen */ 21164a3fd211SDave Hansen if (special_file(inode->i_mode)) 21174a3fd211SDave Hansen return 0; 21184a3fd211SDave Hansen return (flag & O_TRUNC); 21194a3fd211SDave Hansen } 21204a3fd211SDave Hansen 2121648fa861SAl Viro static struct file *finish_open(struct nameidata *nd, 21229a66179eSAl Viro int open_flag, int acc_mode) 2123648fa861SAl Viro { 2124648fa861SAl Viro struct file *filp; 2125648fa861SAl Viro int will_truncate; 2126648fa861SAl Viro int error; 2127648fa861SAl Viro 21289a66179eSAl Viro will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode); 2129648fa861SAl Viro if (will_truncate) { 2130648fa861SAl Viro error = mnt_want_write(nd->path.mnt); 2131648fa861SAl Viro if (error) 2132648fa861SAl Viro goto exit; 2133648fa861SAl Viro } 2134648fa861SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 2135648fa861SAl Viro if (error) { 2136648fa861SAl Viro if (will_truncate) 2137648fa861SAl Viro mnt_drop_write(nd->path.mnt); 2138648fa861SAl Viro goto exit; 2139648fa861SAl Viro } 2140648fa861SAl Viro filp = nameidata_to_filp(nd); 2141648fa861SAl Viro if (!IS_ERR(filp)) { 2142648fa861SAl Viro error = ima_file_check(filp, acc_mode); 2143648fa861SAl Viro if (error) { 2144648fa861SAl Viro fput(filp); 2145648fa861SAl Viro filp = ERR_PTR(error); 2146648fa861SAl Viro } 2147648fa861SAl Viro } 2148648fa861SAl Viro if (!IS_ERR(filp)) { 2149648fa861SAl Viro if (will_truncate) { 2150e1181ee6SJeff Layton error = handle_truncate(filp); 2151648fa861SAl Viro if (error) { 2152648fa861SAl Viro fput(filp); 2153648fa861SAl Viro filp = ERR_PTR(error); 2154648fa861SAl Viro } 2155648fa861SAl Viro } 2156648fa861SAl Viro } 2157648fa861SAl Viro /* 2158648fa861SAl Viro * It is now safe to drop the mnt write 2159648fa861SAl Viro * because the filp has had a write taken 2160648fa861SAl Viro * on its behalf. 2161648fa861SAl Viro */ 2162648fa861SAl Viro if (will_truncate) 2163648fa861SAl Viro mnt_drop_write(nd->path.mnt); 2164d893f1bcSAl Viro path_put(&nd->path); 2165648fa861SAl Viro return filp; 2166648fa861SAl Viro 2167648fa861SAl Viro exit: 2168648fa861SAl Viro path_put(&nd->path); 2169648fa861SAl Viro return ERR_PTR(error); 2170648fa861SAl Viro } 2171648fa861SAl Viro 217231e6b01fSNick Piggin /* 217331e6b01fSNick Piggin * Handle O_CREAT case for do_filp_open 217431e6b01fSNick Piggin */ 2175fb1cc555SAl Viro static struct file *do_last(struct nameidata *nd, struct path *path, 2176c3e380b0SAl Viro const struct open_flags *op, const char *pathname) 2177fb1cc555SAl Viro { 2178a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2179fb1cc555SAl Viro struct file *filp; 218016c2cd71SAl Viro int error; 2181fb1cc555SAl Viro 2182c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2183c3e380b0SAl Viro nd->flags |= op->intent; 2184c3e380b0SAl Viro 21851f36f774SAl Viro switch (nd->last_type) { 21861f36f774SAl Viro case LAST_DOTDOT: 21871f36f774SAl Viro follow_dotdot(nd); 21881f36f774SAl Viro dir = nd->path.dentry; 2189176306f5SNeil Brown case LAST_DOT: 21901f36f774SAl Viro /* fallthrough */ 21911f36f774SAl Viro case LAST_ROOT: 219216c2cd71SAl Viro error = handle_reval_path(nd); 219316c2cd71SAl Viro if (error) 219416c2cd71SAl Viro goto exit; 219516c2cd71SAl Viro error = -EISDIR; 21961f36f774SAl Viro goto exit; 21971f36f774SAl Viro case LAST_BIND: 219816c2cd71SAl Viro error = handle_reval_path(nd); 219916c2cd71SAl Viro if (error) 220016c2cd71SAl Viro goto exit; 22011f36f774SAl Viro audit_inode(pathname, dir); 22021f36f774SAl Viro goto ok; 22031f36f774SAl Viro } 2204a2c36b45SAl Viro 220516c2cd71SAl Viro error = -EISDIR; 22061f36f774SAl Viro /* trailing slashes? */ 220731e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 22081f36f774SAl Viro goto exit; 22091f36f774SAl Viro 2210a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 2211a1e28038SAl Viro 2212a1e28038SAl Viro path->dentry = lookup_hash(nd); 2213a1e28038SAl Viro path->mnt = nd->path.mnt; 2214a1e28038SAl Viro 2215fb1cc555SAl Viro error = PTR_ERR(path->dentry); 2216fb1cc555SAl Viro if (IS_ERR(path->dentry)) { 2217fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2218fb1cc555SAl Viro goto exit; 2219fb1cc555SAl Viro } 2220fb1cc555SAl Viro 2221fb1cc555SAl Viro if (IS_ERR(nd->intent.open.file)) { 2222fb1cc555SAl Viro error = PTR_ERR(nd->intent.open.file); 2223fb1cc555SAl Viro goto exit_mutex_unlock; 2224fb1cc555SAl Viro } 2225fb1cc555SAl Viro 2226fb1cc555SAl Viro /* Negative dentry, just create the file */ 2227fb1cc555SAl Viro if (!path->dentry->d_inode) { 2228fb1cc555SAl Viro /* 2229fb1cc555SAl Viro * This write is needed to ensure that a 2230fb1cc555SAl Viro * ro->rw transition does not occur between 2231fb1cc555SAl Viro * the time when the file is created and when 2232fb1cc555SAl Viro * a permanent write count is taken through 2233fb1cc555SAl Viro * the 'struct file' in nameidata_to_filp(). 2234fb1cc555SAl Viro */ 2235fb1cc555SAl Viro error = mnt_want_write(nd->path.mnt); 2236fb1cc555SAl Viro if (error) 2237fb1cc555SAl Viro goto exit_mutex_unlock; 2238c3e380b0SAl Viro error = __open_namei_create(nd, path, op->open_flag, op->mode); 2239fb1cc555SAl Viro if (error) { 2240fb1cc555SAl Viro mnt_drop_write(nd->path.mnt); 2241fb1cc555SAl Viro goto exit; 2242fb1cc555SAl Viro } 2243fb1cc555SAl Viro filp = nameidata_to_filp(nd); 2244fb1cc555SAl Viro mnt_drop_write(nd->path.mnt); 2245d893f1bcSAl Viro path_put(&nd->path); 2246fb1cc555SAl Viro if (!IS_ERR(filp)) { 2247c3e380b0SAl Viro error = ima_file_check(filp, op->acc_mode); 2248fb1cc555SAl Viro if (error) { 2249fb1cc555SAl Viro fput(filp); 2250fb1cc555SAl Viro filp = ERR_PTR(error); 2251fb1cc555SAl Viro } 2252fb1cc555SAl Viro } 2253fb1cc555SAl Viro return filp; 2254fb1cc555SAl Viro } 2255fb1cc555SAl Viro 2256fb1cc555SAl Viro /* 2257fb1cc555SAl Viro * It already exists. 2258fb1cc555SAl Viro */ 2259fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2260fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2261fb1cc555SAl Viro 2262fb1cc555SAl Viro error = -EEXIST; 2263c3e380b0SAl Viro if (op->open_flag & O_EXCL) 2264fb1cc555SAl Viro goto exit_dput; 2265fb1cc555SAl Viro 22669875cf80SDavid Howells error = follow_managed(path, nd->flags); 22679875cf80SDavid Howells if (error < 0) 2268fb1cc555SAl Viro goto exit_dput; 2269fb1cc555SAl Viro 2270fb1cc555SAl Viro error = -ENOENT; 2271fb1cc555SAl Viro if (!path->dentry->d_inode) 2272fb1cc555SAl Viro goto exit_dput; 22739e67f361SAl Viro 22749e67f361SAl Viro if (path->dentry->d_inode->i_op->follow_link) 2275fb1cc555SAl Viro return NULL; 2276fb1cc555SAl Viro 2277fb1cc555SAl Viro path_to_nameidata(path, nd); 227831e6b01fSNick Piggin nd->inode = path->dentry->d_inode; 2279fb1cc555SAl Viro error = -EISDIR; 228031e6b01fSNick Piggin if (S_ISDIR(nd->inode->i_mode)) 2281fb1cc555SAl Viro goto exit; 228267ee3ad2SAl Viro ok: 2283c3e380b0SAl Viro filp = finish_open(nd, op->open_flag, op->acc_mode); 2284fb1cc555SAl Viro return filp; 2285fb1cc555SAl Viro 2286fb1cc555SAl Viro exit_mutex_unlock: 2287fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2288fb1cc555SAl Viro exit_dput: 2289fb1cc555SAl Viro path_put_conditional(path, nd); 2290fb1cc555SAl Viro exit: 2291fb1cc555SAl Viro path_put(&nd->path); 2292fb1cc555SAl Viro return ERR_PTR(error); 2293fb1cc555SAl Viro } 2294fb1cc555SAl Viro 229513aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 229647c805dcSAl Viro const struct open_flags *op, int flags) 22971da177e4SLinus Torvalds { 22984a3fd211SDave Hansen struct file *filp; 2299a70e65dfSChristoph Hellwig struct nameidata nd; 23009850c056SAl Viro struct path path; 23011da177e4SLinus Torvalds int count = 0; 230213aab428SAl Viro int error; 230331e6b01fSNick Piggin 230431e6b01fSNick Piggin filp = get_empty_filp(); 230531e6b01fSNick Piggin if (!filp) 230631e6b01fSNick Piggin return ERR_PTR(-ENFILE); 230731e6b01fSNick Piggin 230847c805dcSAl Viro filp->f_flags = op->open_flag; 230931e6b01fSNick Piggin nd.intent.open.file = filp; 231047c805dcSAl Viro nd.intent.open.flags = open_to_namei_flags(op->open_flag); 231147c805dcSAl Viro nd.intent.open.create_mode = op->mode; 231231e6b01fSNick Piggin 231347c805dcSAl Viro if (op->open_flag & O_CREAT) 231431e6b01fSNick Piggin goto creat; 231531e6b01fSNick Piggin 231631e6b01fSNick Piggin /* !O_CREAT, simple open */ 231713aab428SAl Viro error = path_lookupat(dfd, pathname, flags | op->intent, &nd); 231831e6b01fSNick Piggin if (unlikely(error)) 231913aab428SAl Viro goto out_filp; 232031e6b01fSNick Piggin error = -ELOOP; 232131e6b01fSNick Piggin if (!(nd.flags & LOOKUP_FOLLOW)) { 232231e6b01fSNick Piggin if (nd.inode->i_op->follow_link) 232313aab428SAl Viro goto out_path; 232431e6b01fSNick Piggin } 232531e6b01fSNick Piggin error = -ENOTDIR; 232631e6b01fSNick Piggin if (nd.flags & LOOKUP_DIRECTORY) { 232731e6b01fSNick Piggin if (!nd.inode->i_op->lookup) 232813aab428SAl Viro goto out_path; 232931e6b01fSNick Piggin } 233031e6b01fSNick Piggin audit_inode(pathname, nd.path.dentry); 233147c805dcSAl Viro filp = finish_open(&nd, op->open_flag, op->acc_mode); 23322dab5974SLinus Torvalds release_open_intent(&nd); 233331e6b01fSNick Piggin return filp; 233431e6b01fSNick Piggin 233531e6b01fSNick Piggin creat: 233631e6b01fSNick Piggin /* OK, have to create the file. Find the parent. */ 2337c3e380b0SAl Viro error = path_lookupat(dfd, pathname, LOOKUP_PARENT | flags, &nd); 233831e6b01fSNick Piggin if (unlikely(error)) 233931e6b01fSNick Piggin goto out_filp; 234031e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) 23419b4a9b14SAl Viro audit_inode(pathname, nd.path.dentry); 23421da177e4SLinus Torvalds 23431da177e4SLinus Torvalds /* 2344a2c36b45SAl Viro * We have the parent and last component. 23451da177e4SLinus Torvalds */ 234647c805dcSAl Viro filp = do_last(&nd, &path, op, pathname); 2347806b681cSAl Viro while (unlikely(!filp)) { /* trailing symlink */ 23487b9337aaSNick Piggin struct path link = path; 23497b9337aaSNick Piggin struct inode *linki = link.dentry->d_inode; 2350def4af30SAl Viro void *cookie; 2351806b681cSAl Viro error = -ELOOP; 2352db372915SDavid Howells if (!(nd.flags & LOOKUP_FOLLOW)) 23531f36f774SAl Viro goto exit_dput; 23541f36f774SAl Viro if (count++ == 32) 2355806b681cSAl Viro goto exit_dput; 2356806b681cSAl Viro /* 2357806b681cSAl Viro * This is subtle. Instead of calling do_follow_link() we do 2358806b681cSAl Viro * the thing by hands. The reason is that this way we have zero 2359806b681cSAl Viro * link_count and path_walk() (called from ->follow_link) 2360806b681cSAl Viro * honoring LOOKUP_PARENT. After that we have the parent and 2361806b681cSAl Viro * last component, i.e. we are in the same situation as after 2362806b681cSAl Viro * the first path_walk(). Well, almost - if the last component 2363806b681cSAl Viro * is normal we get its copy stored in nd->last.name and we will 2364806b681cSAl Viro * have to putname() it when we are done. Procfs-like symlinks 2365806b681cSAl Viro * just set LAST_BIND. 2366806b681cSAl Viro */ 2367806b681cSAl Viro nd.flags |= LOOKUP_PARENT; 2368c3e380b0SAl Viro nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 23697b9337aaSNick Piggin error = __do_follow_link(&link, &nd, &cookie); 2370c3e380b0SAl Viro if (unlikely(error)) 2371f1afe9efSAl Viro filp = ERR_PTR(error); 2372c3e380b0SAl Viro else 237347c805dcSAl Viro filp = do_last(&nd, &path, op, pathname); 2374f1afe9efSAl Viro if (!IS_ERR(cookie) && linki->i_op->put_link) 23757b9337aaSNick Piggin linki->i_op->put_link(link.dentry, &nd, cookie); 23767b9337aaSNick Piggin path_put(&link); 2377806b681cSAl Viro } 237810fa8e62SAl Viro out: 23792a737871SAl Viro if (nd.root.mnt) 23802a737871SAl Viro path_put(&nd.root); 23812dab5974SLinus Torvalds release_open_intent(&nd); 238210fa8e62SAl Viro return filp; 23831da177e4SLinus Torvalds 2384806b681cSAl Viro exit_dput: 2385806b681cSAl Viro path_put_conditional(&path, &nd); 238631e6b01fSNick Piggin out_path: 238731e6b01fSNick Piggin path_put(&nd.path); 238831e6b01fSNick Piggin out_filp: 238910fa8e62SAl Viro filp = ERR_PTR(error); 239010fa8e62SAl Viro goto out; 2391de459215SKirill Korotaev } 23921da177e4SLinus Torvalds 239313aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 239413aab428SAl Viro const struct open_flags *op, int flags) 239513aab428SAl Viro { 239613aab428SAl Viro struct file *filp; 239713aab428SAl Viro 239813aab428SAl Viro filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU); 239913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 240013aab428SAl Viro filp = path_openat(dfd, pathname, op, flags); 240113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 240213aab428SAl Viro filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL); 240313aab428SAl Viro return filp; 240413aab428SAl Viro } 240513aab428SAl Viro 24061da177e4SLinus Torvalds /** 24071da177e4SLinus Torvalds * lookup_create - lookup a dentry, creating it if it doesn't exist 24081da177e4SLinus Torvalds * @nd: nameidata info 24091da177e4SLinus Torvalds * @is_dir: directory flag 24101da177e4SLinus Torvalds * 24111da177e4SLinus Torvalds * Simple function to lookup and return a dentry and create it 24121da177e4SLinus Torvalds * if it doesn't exist. Is SMP-safe. 2413c663e5d8SChristoph Hellwig * 24144ac91378SJan Blunck * Returns with nd->path.dentry->d_inode->i_mutex locked. 24151da177e4SLinus Torvalds */ 24161da177e4SLinus Torvalds struct dentry *lookup_create(struct nameidata *nd, int is_dir) 24171da177e4SLinus Torvalds { 2418c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 24191da177e4SLinus Torvalds 24204ac91378SJan Blunck mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2421c663e5d8SChristoph Hellwig /* 2422c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2423c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2424c663e5d8SChristoph Hellwig */ 24251da177e4SLinus Torvalds if (nd->last_type != LAST_NORM) 24261da177e4SLinus Torvalds goto fail; 24271da177e4SLinus Torvalds nd->flags &= ~LOOKUP_PARENT; 24283516586aSAl Viro nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2429a634904aSASANO Masahiro nd->intent.open.flags = O_EXCL; 2430c663e5d8SChristoph Hellwig 2431c663e5d8SChristoph Hellwig /* 2432c663e5d8SChristoph Hellwig * Do the final lookup. 2433c663e5d8SChristoph Hellwig */ 243449705b77SChristoph Hellwig dentry = lookup_hash(nd); 24351da177e4SLinus Torvalds if (IS_ERR(dentry)) 24361da177e4SLinus Torvalds goto fail; 2437c663e5d8SChristoph Hellwig 2438e9baf6e5SAl Viro if (dentry->d_inode) 2439e9baf6e5SAl Viro goto eexist; 2440c663e5d8SChristoph Hellwig /* 2441c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 2442c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 2443c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 2444c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 2445c663e5d8SChristoph Hellwig */ 2446e9baf6e5SAl Viro if (unlikely(!is_dir && nd->last.name[nd->last.len])) { 24471da177e4SLinus Torvalds dput(dentry); 24481da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 2449e9baf6e5SAl Viro } 2450e9baf6e5SAl Viro return dentry; 2451e9baf6e5SAl Viro eexist: 2452e9baf6e5SAl Viro dput(dentry); 2453e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 24541da177e4SLinus Torvalds fail: 24551da177e4SLinus Torvalds return dentry; 24561da177e4SLinus Torvalds } 2457f81a0bffSChristoph Hellwig EXPORT_SYMBOL_GPL(lookup_create); 24581da177e4SLinus Torvalds 24591da177e4SLinus Torvalds int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 24601da177e4SLinus Torvalds { 2461a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 24621da177e4SLinus Torvalds 24631da177e4SLinus Torvalds if (error) 24641da177e4SLinus Torvalds return error; 24651da177e4SLinus Torvalds 24661da177e4SLinus Torvalds if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 24671da177e4SLinus Torvalds return -EPERM; 24681da177e4SLinus Torvalds 2469acfa4380SAl Viro if (!dir->i_op->mknod) 24701da177e4SLinus Torvalds return -EPERM; 24711da177e4SLinus Torvalds 247208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 247308ce5f16SSerge E. Hallyn if (error) 247408ce5f16SSerge E. Hallyn return error; 247508ce5f16SSerge E. Hallyn 24761da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 24771da177e4SLinus Torvalds if (error) 24781da177e4SLinus Torvalds return error; 24791da177e4SLinus Torvalds 24801da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 2481a74574aaSStephen Smalley if (!error) 2482f38aa942SAmy Griffis fsnotify_create(dir, dentry); 24831da177e4SLinus Torvalds return error; 24841da177e4SLinus Torvalds } 24851da177e4SLinus Torvalds 2486463c3197SDave Hansen static int may_mknod(mode_t mode) 2487463c3197SDave Hansen { 2488463c3197SDave Hansen switch (mode & S_IFMT) { 2489463c3197SDave Hansen case S_IFREG: 2490463c3197SDave Hansen case S_IFCHR: 2491463c3197SDave Hansen case S_IFBLK: 2492463c3197SDave Hansen case S_IFIFO: 2493463c3197SDave Hansen case S_IFSOCK: 2494463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 2495463c3197SDave Hansen return 0; 2496463c3197SDave Hansen case S_IFDIR: 2497463c3197SDave Hansen return -EPERM; 2498463c3197SDave Hansen default: 2499463c3197SDave Hansen return -EINVAL; 2500463c3197SDave Hansen } 2501463c3197SDave Hansen } 2502463c3197SDave Hansen 25032e4d0924SHeiko Carstens SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, 25042e4d0924SHeiko Carstens unsigned, dev) 25051da177e4SLinus Torvalds { 25062ad94ae6SAl Viro int error; 25071da177e4SLinus Torvalds char *tmp; 25081da177e4SLinus Torvalds struct dentry *dentry; 25091da177e4SLinus Torvalds struct nameidata nd; 25101da177e4SLinus Torvalds 25111da177e4SLinus Torvalds if (S_ISDIR(mode)) 25121da177e4SLinus Torvalds return -EPERM; 25131da177e4SLinus Torvalds 25142ad94ae6SAl Viro error = user_path_parent(dfd, filename, &nd, &tmp); 25151da177e4SLinus Torvalds if (error) 25162ad94ae6SAl Viro return error; 25172ad94ae6SAl Viro 25181da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 2519463c3197SDave Hansen if (IS_ERR(dentry)) { 25201da177e4SLinus Torvalds error = PTR_ERR(dentry); 2521463c3197SDave Hansen goto out_unlock; 2522463c3197SDave Hansen } 25234ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2524ce3b0f8dSAl Viro mode &= ~current_umask(); 2525463c3197SDave Hansen error = may_mknod(mode); 2526463c3197SDave Hansen if (error) 2527463c3197SDave Hansen goto out_dput; 2528463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2529463c3197SDave Hansen if (error) 2530463c3197SDave Hansen goto out_dput; 2531be6d3e56SKentaro Takeda error = security_path_mknod(&nd.path, dentry, mode, dev); 2532be6d3e56SKentaro Takeda if (error) 2533be6d3e56SKentaro Takeda goto out_drop_write; 25341da177e4SLinus Torvalds switch (mode & S_IFMT) { 25351da177e4SLinus Torvalds case 0: case S_IFREG: 25364ac91378SJan Blunck error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); 25371da177e4SLinus Torvalds break; 25381da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 25394ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, 25401da177e4SLinus Torvalds new_decode_dev(dev)); 25411da177e4SLinus Torvalds break; 25421da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 25434ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); 25441da177e4SLinus Torvalds break; 25451da177e4SLinus Torvalds } 2546be6d3e56SKentaro Takeda out_drop_write: 2547463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2548463c3197SDave Hansen out_dput: 25491da177e4SLinus Torvalds dput(dentry); 2550463c3197SDave Hansen out_unlock: 25514ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 25521d957f9bSJan Blunck path_put(&nd.path); 25531da177e4SLinus Torvalds putname(tmp); 25541da177e4SLinus Torvalds 25551da177e4SLinus Torvalds return error; 25561da177e4SLinus Torvalds } 25571da177e4SLinus Torvalds 25583480b257SHeiko Carstens SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) 25595590ff0dSUlrich Drepper { 25605590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 25615590ff0dSUlrich Drepper } 25625590ff0dSUlrich Drepper 25631da177e4SLinus Torvalds int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 25641da177e4SLinus Torvalds { 2565a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25661da177e4SLinus Torvalds 25671da177e4SLinus Torvalds if (error) 25681da177e4SLinus Torvalds return error; 25691da177e4SLinus Torvalds 2570acfa4380SAl Viro if (!dir->i_op->mkdir) 25711da177e4SLinus Torvalds return -EPERM; 25721da177e4SLinus Torvalds 25731da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 25741da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 25751da177e4SLinus Torvalds if (error) 25761da177e4SLinus Torvalds return error; 25771da177e4SLinus Torvalds 25781da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 2579a74574aaSStephen Smalley if (!error) 2580f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 25811da177e4SLinus Torvalds return error; 25821da177e4SLinus Torvalds } 25831da177e4SLinus Torvalds 25842e4d0924SHeiko Carstens SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) 25851da177e4SLinus Torvalds { 25861da177e4SLinus Torvalds int error = 0; 25871da177e4SLinus Torvalds char * tmp; 25886902d925SDave Hansen struct dentry *dentry; 25896902d925SDave Hansen struct nameidata nd; 25901da177e4SLinus Torvalds 25912ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &tmp); 25922ad94ae6SAl Viro if (error) 25936902d925SDave Hansen goto out_err; 25941da177e4SLinus Torvalds 25951da177e4SLinus Torvalds dentry = lookup_create(&nd, 1); 25961da177e4SLinus Torvalds error = PTR_ERR(dentry); 25976902d925SDave Hansen if (IS_ERR(dentry)) 25986902d925SDave Hansen goto out_unlock; 25996902d925SDave Hansen 26004ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2601ce3b0f8dSAl Viro mode &= ~current_umask(); 2602463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2603463c3197SDave Hansen if (error) 2604463c3197SDave Hansen goto out_dput; 2605be6d3e56SKentaro Takeda error = security_path_mkdir(&nd.path, dentry, mode); 2606be6d3e56SKentaro Takeda if (error) 2607be6d3e56SKentaro Takeda goto out_drop_write; 26084ac91378SJan Blunck error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); 2609be6d3e56SKentaro Takeda out_drop_write: 2610463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2611463c3197SDave Hansen out_dput: 26121da177e4SLinus Torvalds dput(dentry); 26136902d925SDave Hansen out_unlock: 26144ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 26151d957f9bSJan Blunck path_put(&nd.path); 26161da177e4SLinus Torvalds putname(tmp); 26176902d925SDave Hansen out_err: 26181da177e4SLinus Torvalds return error; 26191da177e4SLinus Torvalds } 26201da177e4SLinus Torvalds 26213cdad428SHeiko Carstens SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) 26225590ff0dSUlrich Drepper { 26235590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 26245590ff0dSUlrich Drepper } 26255590ff0dSUlrich Drepper 26261da177e4SLinus Torvalds /* 26271da177e4SLinus Torvalds * We try to drop the dentry early: we should have 26281da177e4SLinus Torvalds * a usage count of 2 if we're the only user of this 26291da177e4SLinus Torvalds * dentry, and if that is true (possibly after pruning 26301da177e4SLinus Torvalds * the dcache), then we drop the dentry now. 26311da177e4SLinus Torvalds * 26321da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 26331da177e4SLinus Torvalds * do a 26341da177e4SLinus Torvalds * 26351da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 26361da177e4SLinus Torvalds * return -EBUSY; 26371da177e4SLinus Torvalds * 26381da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 26391da177e4SLinus Torvalds * that is still in use by something else.. 26401da177e4SLinus Torvalds */ 26411da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 26421da177e4SLinus Torvalds { 26431da177e4SLinus Torvalds dget(dentry); 26441da177e4SLinus Torvalds shrink_dcache_parent(dentry); 26451da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 2646b7ab39f6SNick Piggin if (dentry->d_count == 2) 26471da177e4SLinus Torvalds __d_drop(dentry); 26481da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 26491da177e4SLinus Torvalds } 26501da177e4SLinus Torvalds 26511da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 26521da177e4SLinus Torvalds { 26531da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 26541da177e4SLinus Torvalds 26551da177e4SLinus Torvalds if (error) 26561da177e4SLinus Torvalds return error; 26571da177e4SLinus Torvalds 2658acfa4380SAl Viro if (!dir->i_op->rmdir) 26591da177e4SLinus Torvalds return -EPERM; 26601da177e4SLinus Torvalds 26611b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 26621da177e4SLinus Torvalds dentry_unhash(dentry); 26631da177e4SLinus Torvalds if (d_mountpoint(dentry)) 26641da177e4SLinus Torvalds error = -EBUSY; 26651da177e4SLinus Torvalds else { 26661da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 26671da177e4SLinus Torvalds if (!error) { 26681da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 2669d83c49f3SAl Viro if (!error) { 26701da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 2671d83c49f3SAl Viro dont_mount(dentry); 2672d83c49f3SAl Viro } 26731da177e4SLinus Torvalds } 26741da177e4SLinus Torvalds } 26751b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 26761da177e4SLinus Torvalds if (!error) { 26771da177e4SLinus Torvalds d_delete(dentry); 26781da177e4SLinus Torvalds } 26791da177e4SLinus Torvalds dput(dentry); 26801da177e4SLinus Torvalds 26811da177e4SLinus Torvalds return error; 26821da177e4SLinus Torvalds } 26831da177e4SLinus Torvalds 26845590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 26851da177e4SLinus Torvalds { 26861da177e4SLinus Torvalds int error = 0; 26871da177e4SLinus Torvalds char * name; 26881da177e4SLinus Torvalds struct dentry *dentry; 26891da177e4SLinus Torvalds struct nameidata nd; 26901da177e4SLinus Torvalds 26912ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 26921da177e4SLinus Torvalds if (error) 26932ad94ae6SAl Viro return error; 26941da177e4SLinus Torvalds 26951da177e4SLinus Torvalds switch(nd.last_type) { 26961da177e4SLinus Torvalds case LAST_DOTDOT: 26971da177e4SLinus Torvalds error = -ENOTEMPTY; 26981da177e4SLinus Torvalds goto exit1; 26991da177e4SLinus Torvalds case LAST_DOT: 27001da177e4SLinus Torvalds error = -EINVAL; 27011da177e4SLinus Torvalds goto exit1; 27021da177e4SLinus Torvalds case LAST_ROOT: 27031da177e4SLinus Torvalds error = -EBUSY; 27041da177e4SLinus Torvalds goto exit1; 27051da177e4SLinus Torvalds } 27060612d9fbSOGAWA Hirofumi 27070612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 27080612d9fbSOGAWA Hirofumi 27094ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 271049705b77SChristoph Hellwig dentry = lookup_hash(&nd); 27111da177e4SLinus Torvalds error = PTR_ERR(dentry); 27126902d925SDave Hansen if (IS_ERR(dentry)) 27136902d925SDave Hansen goto exit2; 27140622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 27150622753bSDave Hansen if (error) 27160622753bSDave Hansen goto exit3; 2717be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 2718be6d3e56SKentaro Takeda if (error) 2719be6d3e56SKentaro Takeda goto exit4; 27204ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 2721be6d3e56SKentaro Takeda exit4: 27220622753bSDave Hansen mnt_drop_write(nd.path.mnt); 27230622753bSDave Hansen exit3: 27241da177e4SLinus Torvalds dput(dentry); 27256902d925SDave Hansen exit2: 27264ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 27271da177e4SLinus Torvalds exit1: 27281d957f9bSJan Blunck path_put(&nd.path); 27291da177e4SLinus Torvalds putname(name); 27301da177e4SLinus Torvalds return error; 27311da177e4SLinus Torvalds } 27321da177e4SLinus Torvalds 27333cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 27345590ff0dSUlrich Drepper { 27355590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 27365590ff0dSUlrich Drepper } 27375590ff0dSUlrich Drepper 27381da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 27391da177e4SLinus Torvalds { 27401da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 27411da177e4SLinus Torvalds 27421da177e4SLinus Torvalds if (error) 27431da177e4SLinus Torvalds return error; 27441da177e4SLinus Torvalds 2745acfa4380SAl Viro if (!dir->i_op->unlink) 27461da177e4SLinus Torvalds return -EPERM; 27471da177e4SLinus Torvalds 27481b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 27491da177e4SLinus Torvalds if (d_mountpoint(dentry)) 27501da177e4SLinus Torvalds error = -EBUSY; 27511da177e4SLinus Torvalds else { 27521da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 2753bec1052eSAl Viro if (!error) { 27541da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 2755bec1052eSAl Viro if (!error) 2756d83c49f3SAl Viro dont_mount(dentry); 2757bec1052eSAl Viro } 27581da177e4SLinus Torvalds } 27591b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 27601da177e4SLinus Torvalds 27611da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 27621da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 2763ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 27641da177e4SLinus Torvalds d_delete(dentry); 27651da177e4SLinus Torvalds } 27660eeca283SRobert Love 27671da177e4SLinus Torvalds return error; 27681da177e4SLinus Torvalds } 27691da177e4SLinus Torvalds 27701da177e4SLinus Torvalds /* 27711da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 27721b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 27731da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 27741da177e4SLinus Torvalds * while waiting on the I/O. 27751da177e4SLinus Torvalds */ 27765590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 27771da177e4SLinus Torvalds { 27782ad94ae6SAl Viro int error; 27791da177e4SLinus Torvalds char *name; 27801da177e4SLinus Torvalds struct dentry *dentry; 27811da177e4SLinus Torvalds struct nameidata nd; 27821da177e4SLinus Torvalds struct inode *inode = NULL; 27831da177e4SLinus Torvalds 27842ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 27851da177e4SLinus Torvalds if (error) 27862ad94ae6SAl Viro return error; 27872ad94ae6SAl Viro 27881da177e4SLinus Torvalds error = -EISDIR; 27891da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 27901da177e4SLinus Torvalds goto exit1; 27910612d9fbSOGAWA Hirofumi 27920612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 27930612d9fbSOGAWA Hirofumi 27944ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 279549705b77SChristoph Hellwig dentry = lookup_hash(&nd); 27961da177e4SLinus Torvalds error = PTR_ERR(dentry); 27971da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 27981da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 27991da177e4SLinus Torvalds if (nd.last.name[nd.last.len]) 28001da177e4SLinus Torvalds goto slashes; 28011da177e4SLinus Torvalds inode = dentry->d_inode; 28021da177e4SLinus Torvalds if (inode) 28037de9c6eeSAl Viro ihold(inode); 28040622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 28050622753bSDave Hansen if (error) 28060622753bSDave Hansen goto exit2; 2807be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 2808be6d3e56SKentaro Takeda if (error) 2809be6d3e56SKentaro Takeda goto exit3; 28104ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 2811be6d3e56SKentaro Takeda exit3: 28120622753bSDave Hansen mnt_drop_write(nd.path.mnt); 28131da177e4SLinus Torvalds exit2: 28141da177e4SLinus Torvalds dput(dentry); 28151da177e4SLinus Torvalds } 28164ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 28171da177e4SLinus Torvalds if (inode) 28181da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 28191da177e4SLinus Torvalds exit1: 28201d957f9bSJan Blunck path_put(&nd.path); 28211da177e4SLinus Torvalds putname(name); 28221da177e4SLinus Torvalds return error; 28231da177e4SLinus Torvalds 28241da177e4SLinus Torvalds slashes: 28251da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 28261da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 28271da177e4SLinus Torvalds goto exit2; 28281da177e4SLinus Torvalds } 28291da177e4SLinus Torvalds 28302e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 28315590ff0dSUlrich Drepper { 28325590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 28335590ff0dSUlrich Drepper return -EINVAL; 28345590ff0dSUlrich Drepper 28355590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 28365590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 28375590ff0dSUlrich Drepper 28385590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 28395590ff0dSUlrich Drepper } 28405590ff0dSUlrich Drepper 28413480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 28425590ff0dSUlrich Drepper { 28435590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 28445590ff0dSUlrich Drepper } 28455590ff0dSUlrich Drepper 2846db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 28471da177e4SLinus Torvalds { 2848a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28491da177e4SLinus Torvalds 28501da177e4SLinus Torvalds if (error) 28511da177e4SLinus Torvalds return error; 28521da177e4SLinus Torvalds 2853acfa4380SAl Viro if (!dir->i_op->symlink) 28541da177e4SLinus Torvalds return -EPERM; 28551da177e4SLinus Torvalds 28561da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 28571da177e4SLinus Torvalds if (error) 28581da177e4SLinus Torvalds return error; 28591da177e4SLinus Torvalds 28601da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 2861a74574aaSStephen Smalley if (!error) 2862f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28631da177e4SLinus Torvalds return error; 28641da177e4SLinus Torvalds } 28651da177e4SLinus Torvalds 28662e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 28672e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 28681da177e4SLinus Torvalds { 28692ad94ae6SAl Viro int error; 28701da177e4SLinus Torvalds char *from; 28711da177e4SLinus Torvalds char *to; 28726902d925SDave Hansen struct dentry *dentry; 28736902d925SDave Hansen struct nameidata nd; 28741da177e4SLinus Torvalds 28751da177e4SLinus Torvalds from = getname(oldname); 28761da177e4SLinus Torvalds if (IS_ERR(from)) 28771da177e4SLinus Torvalds return PTR_ERR(from); 28782ad94ae6SAl Viro 28792ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 28802ad94ae6SAl Viro if (error) 28816902d925SDave Hansen goto out_putname; 28821da177e4SLinus Torvalds 28831da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 28841da177e4SLinus Torvalds error = PTR_ERR(dentry); 28856902d925SDave Hansen if (IS_ERR(dentry)) 28866902d925SDave Hansen goto out_unlock; 28876902d925SDave Hansen 288875c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 288975c3f29dSDave Hansen if (error) 289075c3f29dSDave Hansen goto out_dput; 2891be6d3e56SKentaro Takeda error = security_path_symlink(&nd.path, dentry, from); 2892be6d3e56SKentaro Takeda if (error) 2893be6d3e56SKentaro Takeda goto out_drop_write; 2894db2e747bSMiklos Szeredi error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); 2895be6d3e56SKentaro Takeda out_drop_write: 289675c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 289775c3f29dSDave Hansen out_dput: 28981da177e4SLinus Torvalds dput(dentry); 28996902d925SDave Hansen out_unlock: 29004ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 29011d957f9bSJan Blunck path_put(&nd.path); 29021da177e4SLinus Torvalds putname(to); 29036902d925SDave Hansen out_putname: 29041da177e4SLinus Torvalds putname(from); 29051da177e4SLinus Torvalds return error; 29061da177e4SLinus Torvalds } 29071da177e4SLinus Torvalds 29083480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 29095590ff0dSUlrich Drepper { 29105590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 29115590ff0dSUlrich Drepper } 29125590ff0dSUlrich Drepper 29131da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 29141da177e4SLinus Torvalds { 29151da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 29161da177e4SLinus Torvalds int error; 29171da177e4SLinus Torvalds 29181da177e4SLinus Torvalds if (!inode) 29191da177e4SLinus Torvalds return -ENOENT; 29201da177e4SLinus Torvalds 2921a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 29221da177e4SLinus Torvalds if (error) 29231da177e4SLinus Torvalds return error; 29241da177e4SLinus Torvalds 29251da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 29261da177e4SLinus Torvalds return -EXDEV; 29271da177e4SLinus Torvalds 29281da177e4SLinus Torvalds /* 29291da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 29301da177e4SLinus Torvalds */ 29311da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 29321da177e4SLinus Torvalds return -EPERM; 2933acfa4380SAl Viro if (!dir->i_op->link) 29341da177e4SLinus Torvalds return -EPERM; 29357e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 29361da177e4SLinus Torvalds return -EPERM; 29371da177e4SLinus Torvalds 29381da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 29391da177e4SLinus Torvalds if (error) 29401da177e4SLinus Torvalds return error; 29411da177e4SLinus Torvalds 29427e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 29431da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 29447e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 2945e31e14ecSStephen Smalley if (!error) 29467e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 29471da177e4SLinus Torvalds return error; 29481da177e4SLinus Torvalds } 29491da177e4SLinus Torvalds 29501da177e4SLinus Torvalds /* 29511da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 29521da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 29531da177e4SLinus Torvalds * newname. --KAB 29541da177e4SLinus Torvalds * 29551da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 29561da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 29571da177e4SLinus Torvalds * and other special files. --ADM 29581da177e4SLinus Torvalds */ 29592e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 29602e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 29611da177e4SLinus Torvalds { 29621da177e4SLinus Torvalds struct dentry *new_dentry; 29632d8f3038SAl Viro struct nameidata nd; 29642d8f3038SAl Viro struct path old_path; 29651da177e4SLinus Torvalds int error; 29661da177e4SLinus Torvalds char *to; 29671da177e4SLinus Torvalds 296845c9b11aSUlrich Drepper if ((flags & ~AT_SYMLINK_FOLLOW) != 0) 2969c04030e1SUlrich Drepper return -EINVAL; 2970c04030e1SUlrich Drepper 29712d8f3038SAl Viro error = user_path_at(olddfd, oldname, 297245c9b11aSUlrich Drepper flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, 29732d8f3038SAl Viro &old_path); 29741da177e4SLinus Torvalds if (error) 29752ad94ae6SAl Viro return error; 29762ad94ae6SAl Viro 29772ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 29781da177e4SLinus Torvalds if (error) 29791da177e4SLinus Torvalds goto out; 29801da177e4SLinus Torvalds error = -EXDEV; 29812d8f3038SAl Viro if (old_path.mnt != nd.path.mnt) 29821da177e4SLinus Torvalds goto out_release; 29831da177e4SLinus Torvalds new_dentry = lookup_create(&nd, 0); 29841da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 29856902d925SDave Hansen if (IS_ERR(new_dentry)) 29866902d925SDave Hansen goto out_unlock; 298775c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 298875c3f29dSDave Hansen if (error) 298975c3f29dSDave Hansen goto out_dput; 2990be6d3e56SKentaro Takeda error = security_path_link(old_path.dentry, &nd.path, new_dentry); 2991be6d3e56SKentaro Takeda if (error) 2992be6d3e56SKentaro Takeda goto out_drop_write; 29932d8f3038SAl Viro error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); 2994be6d3e56SKentaro Takeda out_drop_write: 299575c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 299675c3f29dSDave Hansen out_dput: 29971da177e4SLinus Torvalds dput(new_dentry); 29986902d925SDave Hansen out_unlock: 29994ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 30001da177e4SLinus Torvalds out_release: 30011d957f9bSJan Blunck path_put(&nd.path); 30022ad94ae6SAl Viro putname(to); 30031da177e4SLinus Torvalds out: 30042d8f3038SAl Viro path_put(&old_path); 30051da177e4SLinus Torvalds 30061da177e4SLinus Torvalds return error; 30071da177e4SLinus Torvalds } 30081da177e4SLinus Torvalds 30093480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 30105590ff0dSUlrich Drepper { 3011c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 30125590ff0dSUlrich Drepper } 30135590ff0dSUlrich Drepper 30141da177e4SLinus Torvalds /* 30151da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 30161da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 30171da177e4SLinus Torvalds * Problems: 30181da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 30191da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 30201da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3021a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 30221da177e4SLinus Torvalds * story. 30231da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 30241b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 30251da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 30261da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3027a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 30281da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 30291da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 30301da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3031a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 30321da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 30331da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 30341da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 30351da177e4SLinus Torvalds * d) some filesystems don't support opened-but-unlinked directories, 30361da177e4SLinus Torvalds * either because of layout or because they are not ready to deal with 30371da177e4SLinus Torvalds * all cases correctly. The latter will be fixed (taking this sort of 30381da177e4SLinus Torvalds * stuff into VFS), but the former is not going away. Solution: the same 30391da177e4SLinus Torvalds * trick as in rmdir(). 30401da177e4SLinus Torvalds * e) conversion from fhandle to dentry may come in the wrong moment - when 30411b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 30421da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3043c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 30441da177e4SLinus Torvalds * locking]. 30451da177e4SLinus Torvalds */ 304675c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 30471da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30481da177e4SLinus Torvalds { 30491da177e4SLinus Torvalds int error = 0; 30501da177e4SLinus Torvalds struct inode *target; 30511da177e4SLinus Torvalds 30521da177e4SLinus Torvalds /* 30531da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 30541da177e4SLinus Torvalds * we'll need to flip '..'. 30551da177e4SLinus Torvalds */ 30561da177e4SLinus Torvalds if (new_dir != old_dir) { 3057f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 30581da177e4SLinus Torvalds if (error) 30591da177e4SLinus Torvalds return error; 30601da177e4SLinus Torvalds } 30611da177e4SLinus Torvalds 30621da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30631da177e4SLinus Torvalds if (error) 30641da177e4SLinus Torvalds return error; 30651da177e4SLinus Torvalds 30661da177e4SLinus Torvalds target = new_dentry->d_inode; 3067d83c49f3SAl Viro if (target) 30681b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 30691da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 30701da177e4SLinus Torvalds error = -EBUSY; 3071d83c49f3SAl Viro else { 3072d83c49f3SAl Viro if (target) 3073d83c49f3SAl Viro dentry_unhash(new_dentry); 30741da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 3075d83c49f3SAl Viro } 30761da177e4SLinus Torvalds if (target) { 3077d83c49f3SAl Viro if (!error) { 30781da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3079d83c49f3SAl Viro dont_mount(new_dentry); 3080d83c49f3SAl Viro } 30811b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 30821da177e4SLinus Torvalds if (d_unhashed(new_dentry)) 30831da177e4SLinus Torvalds d_rehash(new_dentry); 30841da177e4SLinus Torvalds dput(new_dentry); 30851da177e4SLinus Torvalds } 3086e31e14ecSStephen Smalley if (!error) 3087349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 30881da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 30891da177e4SLinus Torvalds return error; 30901da177e4SLinus Torvalds } 30911da177e4SLinus Torvalds 309275c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 30931da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30941da177e4SLinus Torvalds { 30951da177e4SLinus Torvalds struct inode *target; 30961da177e4SLinus Torvalds int error; 30971da177e4SLinus Torvalds 30981da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30991da177e4SLinus Torvalds if (error) 31001da177e4SLinus Torvalds return error; 31011da177e4SLinus Torvalds 31021da177e4SLinus Torvalds dget(new_dentry); 31031da177e4SLinus Torvalds target = new_dentry->d_inode; 31041da177e4SLinus Torvalds if (target) 31051b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 31061da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 31071da177e4SLinus Torvalds error = -EBUSY; 31081da177e4SLinus Torvalds else 31091da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 31101da177e4SLinus Torvalds if (!error) { 3111bec1052eSAl Viro if (target) 3112d83c49f3SAl Viro dont_mount(new_dentry); 3113349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 31141da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 31151da177e4SLinus Torvalds } 31161da177e4SLinus Torvalds if (target) 31171b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 31181da177e4SLinus Torvalds dput(new_dentry); 31191da177e4SLinus Torvalds return error; 31201da177e4SLinus Torvalds } 31211da177e4SLinus Torvalds 31221da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 31231da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 31241da177e4SLinus Torvalds { 31251da177e4SLinus Torvalds int error; 31261da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 312759b0df21SEric Paris const unsigned char *old_name; 31281da177e4SLinus Torvalds 31291da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 31301da177e4SLinus Torvalds return 0; 31311da177e4SLinus Torvalds 31321da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 31331da177e4SLinus Torvalds if (error) 31341da177e4SLinus Torvalds return error; 31351da177e4SLinus Torvalds 31361da177e4SLinus Torvalds if (!new_dentry->d_inode) 3137a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 31381da177e4SLinus Torvalds else 31391da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 31401da177e4SLinus Torvalds if (error) 31411da177e4SLinus Torvalds return error; 31421da177e4SLinus Torvalds 3143acfa4380SAl Viro if (!old_dir->i_op->rename) 31441da177e4SLinus Torvalds return -EPERM; 31451da177e4SLinus Torvalds 31460eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 31470eeca283SRobert Love 31481da177e4SLinus Torvalds if (is_dir) 31491da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 31501da177e4SLinus Torvalds else 31511da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3152123df294SAl Viro if (!error) 3153123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 31545a190ae6SAl Viro new_dentry->d_inode, old_dentry); 31550eeca283SRobert Love fsnotify_oldname_free(old_name); 31560eeca283SRobert Love 31571da177e4SLinus Torvalds return error; 31581da177e4SLinus Torvalds } 31591da177e4SLinus Torvalds 31602e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 31612e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 31621da177e4SLinus Torvalds { 31631da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 31641da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 31651da177e4SLinus Torvalds struct dentry *trap; 31661da177e4SLinus Torvalds struct nameidata oldnd, newnd; 31672ad94ae6SAl Viro char *from; 31682ad94ae6SAl Viro char *to; 31692ad94ae6SAl Viro int error; 31701da177e4SLinus Torvalds 31712ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 31721da177e4SLinus Torvalds if (error) 31731da177e4SLinus Torvalds goto exit; 31741da177e4SLinus Torvalds 31752ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 31761da177e4SLinus Torvalds if (error) 31771da177e4SLinus Torvalds goto exit1; 31781da177e4SLinus Torvalds 31791da177e4SLinus Torvalds error = -EXDEV; 31804ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 31811da177e4SLinus Torvalds goto exit2; 31821da177e4SLinus Torvalds 31834ac91378SJan Blunck old_dir = oldnd.path.dentry; 31841da177e4SLinus Torvalds error = -EBUSY; 31851da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 31861da177e4SLinus Torvalds goto exit2; 31871da177e4SLinus Torvalds 31884ac91378SJan Blunck new_dir = newnd.path.dentry; 31891da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 31901da177e4SLinus Torvalds goto exit2; 31911da177e4SLinus Torvalds 31920612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 31930612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 31944e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 31950612d9fbSOGAWA Hirofumi 31961da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 31971da177e4SLinus Torvalds 319849705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 31991da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 32001da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 32011da177e4SLinus Torvalds goto exit3; 32021da177e4SLinus Torvalds /* source must exist */ 32031da177e4SLinus Torvalds error = -ENOENT; 32041da177e4SLinus Torvalds if (!old_dentry->d_inode) 32051da177e4SLinus Torvalds goto exit4; 32061da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 32071da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 32081da177e4SLinus Torvalds error = -ENOTDIR; 32091da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 32101da177e4SLinus Torvalds goto exit4; 32111da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 32121da177e4SLinus Torvalds goto exit4; 32131da177e4SLinus Torvalds } 32141da177e4SLinus Torvalds /* source should not be ancestor of target */ 32151da177e4SLinus Torvalds error = -EINVAL; 32161da177e4SLinus Torvalds if (old_dentry == trap) 32171da177e4SLinus Torvalds goto exit4; 321849705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 32191da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 32201da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 32211da177e4SLinus Torvalds goto exit4; 32221da177e4SLinus Torvalds /* target should not be an ancestor of source */ 32231da177e4SLinus Torvalds error = -ENOTEMPTY; 32241da177e4SLinus Torvalds if (new_dentry == trap) 32251da177e4SLinus Torvalds goto exit5; 32261da177e4SLinus Torvalds 32279079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 32289079b1ebSDave Hansen if (error) 32299079b1ebSDave Hansen goto exit5; 3230be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3231be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3232be6d3e56SKentaro Takeda if (error) 3233be6d3e56SKentaro Takeda goto exit6; 32341da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 32351da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 3236be6d3e56SKentaro Takeda exit6: 32379079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 32381da177e4SLinus Torvalds exit5: 32391da177e4SLinus Torvalds dput(new_dentry); 32401da177e4SLinus Torvalds exit4: 32411da177e4SLinus Torvalds dput(old_dentry); 32421da177e4SLinus Torvalds exit3: 32431da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 32441da177e4SLinus Torvalds exit2: 32451d957f9bSJan Blunck path_put(&newnd.path); 32462ad94ae6SAl Viro putname(to); 32471da177e4SLinus Torvalds exit1: 32481d957f9bSJan Blunck path_put(&oldnd.path); 32491da177e4SLinus Torvalds putname(from); 32502ad94ae6SAl Viro exit: 32511da177e4SLinus Torvalds return error; 32521da177e4SLinus Torvalds } 32531da177e4SLinus Torvalds 3254a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 32555590ff0dSUlrich Drepper { 32565590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 32575590ff0dSUlrich Drepper } 32585590ff0dSUlrich Drepper 32591da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 32601da177e4SLinus Torvalds { 32611da177e4SLinus Torvalds int len; 32621da177e4SLinus Torvalds 32631da177e4SLinus Torvalds len = PTR_ERR(link); 32641da177e4SLinus Torvalds if (IS_ERR(link)) 32651da177e4SLinus Torvalds goto out; 32661da177e4SLinus Torvalds 32671da177e4SLinus Torvalds len = strlen(link); 32681da177e4SLinus Torvalds if (len > (unsigned) buflen) 32691da177e4SLinus Torvalds len = buflen; 32701da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 32711da177e4SLinus Torvalds len = -EFAULT; 32721da177e4SLinus Torvalds out: 32731da177e4SLinus Torvalds return len; 32741da177e4SLinus Torvalds } 32751da177e4SLinus Torvalds 32761da177e4SLinus Torvalds /* 32771da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 32781da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 32791da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 32801da177e4SLinus Torvalds */ 32811da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 32821da177e4SLinus Torvalds { 32831da177e4SLinus Torvalds struct nameidata nd; 3284cc314eefSLinus Torvalds void *cookie; 3285694a1764SMarcin Slusarz int res; 3286cc314eefSLinus Torvalds 32871da177e4SLinus Torvalds nd.depth = 0; 3288cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3289694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3290694a1764SMarcin Slusarz return PTR_ERR(cookie); 3291694a1764SMarcin Slusarz 3292694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 32931da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3294cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3295694a1764SMarcin Slusarz return res; 32961da177e4SLinus Torvalds } 32971da177e4SLinus Torvalds 32981da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 32991da177e4SLinus Torvalds { 33001da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 33011da177e4SLinus Torvalds } 33021da177e4SLinus Torvalds 33031da177e4SLinus Torvalds /* get the link contents into pagecache */ 33041da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 33051da177e4SLinus Torvalds { 3306ebd09abbSDuane Griffin char *kaddr; 33071da177e4SLinus Torvalds struct page *page; 33081da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3309090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 33101da177e4SLinus Torvalds if (IS_ERR(page)) 33116fe6900eSNick Piggin return (char*)page; 33121da177e4SLinus Torvalds *ppage = page; 3313ebd09abbSDuane Griffin kaddr = kmap(page); 3314ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3315ebd09abbSDuane Griffin return kaddr; 33161da177e4SLinus Torvalds } 33171da177e4SLinus Torvalds 33181da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 33191da177e4SLinus Torvalds { 33201da177e4SLinus Torvalds struct page *page = NULL; 33211da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 33221da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 33231da177e4SLinus Torvalds if (page) { 33241da177e4SLinus Torvalds kunmap(page); 33251da177e4SLinus Torvalds page_cache_release(page); 33261da177e4SLinus Torvalds } 33271da177e4SLinus Torvalds return res; 33281da177e4SLinus Torvalds } 33291da177e4SLinus Torvalds 3330cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 33311da177e4SLinus Torvalds { 3332cc314eefSLinus Torvalds struct page *page = NULL; 33331da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3334cc314eefSLinus Torvalds return page; 33351da177e4SLinus Torvalds } 33361da177e4SLinus Torvalds 3337cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 33381da177e4SLinus Torvalds { 3339cc314eefSLinus Torvalds struct page *page = cookie; 3340cc314eefSLinus Torvalds 3341cc314eefSLinus Torvalds if (page) { 33421da177e4SLinus Torvalds kunmap(page); 33431da177e4SLinus Torvalds page_cache_release(page); 33441da177e4SLinus Torvalds } 33451da177e4SLinus Torvalds } 33461da177e4SLinus Torvalds 334754566b2cSNick Piggin /* 334854566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 334954566b2cSNick Piggin */ 335054566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 33511da177e4SLinus Torvalds { 33521da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 33530adb25d2SKirill Korotaev struct page *page; 3354afddba49SNick Piggin void *fsdata; 3355beb497abSDmitriy Monakhov int err; 33561da177e4SLinus Torvalds char *kaddr; 335754566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 335854566b2cSNick Piggin if (nofs) 335954566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 33601da177e4SLinus Torvalds 33617e53cac4SNeilBrown retry: 3362afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 336354566b2cSNick Piggin flags, &page, &fsdata); 33641da177e4SLinus Torvalds if (err) 3365afddba49SNick Piggin goto fail; 3366afddba49SNick Piggin 33671da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 33681da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 33691da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 3370afddba49SNick Piggin 3371afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3372afddba49SNick Piggin page, fsdata); 33731da177e4SLinus Torvalds if (err < 0) 33741da177e4SLinus Torvalds goto fail; 3375afddba49SNick Piggin if (err < len-1) 3376afddba49SNick Piggin goto retry; 3377afddba49SNick Piggin 33781da177e4SLinus Torvalds mark_inode_dirty(inode); 33791da177e4SLinus Torvalds return 0; 33801da177e4SLinus Torvalds fail: 33811da177e4SLinus Torvalds return err; 33821da177e4SLinus Torvalds } 33831da177e4SLinus Torvalds 33840adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 33850adb25d2SKirill Korotaev { 33860adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 338754566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 33880adb25d2SKirill Korotaev } 33890adb25d2SKirill Korotaev 339092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 33911da177e4SLinus Torvalds .readlink = generic_readlink, 33921da177e4SLinus Torvalds .follow_link = page_follow_link_light, 33931da177e4SLinus Torvalds .put_link = page_put_link, 33941da177e4SLinus Torvalds }; 33951da177e4SLinus Torvalds 33962d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3397cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 33981da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 33991da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 34001da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 34011da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 34021da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 34031da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 34041da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 34051da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 34061da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 34070adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 34081da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 34091da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3410c9c6cac0SAl Viro EXPORT_SYMBOL(kern_path_parent); 3411d1811465SAl Viro EXPORT_SYMBOL(kern_path); 341216f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3413f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 34148c744fb8SChristoph Hellwig EXPORT_SYMBOL(file_permission); 34151da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 34161da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 34171da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 34181da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 34191da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 34201da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 34211da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 34221da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 34231da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 34241da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 34251da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 34261da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 34271da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 34281da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3429