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; 4045b6ca027SAl Viro int want_root = 0; 40531e6b01fSNick Piggin 40631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4075b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4085b6ca027SAl Viro want_root = 1; 40931e6b01fSNick Piggin spin_lock(&fs->lock); 41031e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 41131e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 41231e6b01fSNick Piggin goto err_root; 41331e6b01fSNick Piggin } 41431e6b01fSNick Piggin spin_lock(&dentry->d_lock); 41531e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 41631e6b01fSNick Piggin goto err; 41731e6b01fSNick Piggin BUG_ON(nd->inode != dentry->d_inode); 41831e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 4195b6ca027SAl Viro if (want_root) { 42031e6b01fSNick Piggin path_get(&nd->root); 42131e6b01fSNick Piggin spin_unlock(&fs->lock); 42231e6b01fSNick Piggin } 42331e6b01fSNick Piggin mntget(nd->path.mnt); 42431e6b01fSNick Piggin 42531e6b01fSNick Piggin rcu_read_unlock(); 42631e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 42731e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 42831e6b01fSNick Piggin return 0; 42931e6b01fSNick Piggin err: 43031e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 43131e6b01fSNick Piggin err_root: 4325b6ca027SAl Viro if (want_root) 43331e6b01fSNick Piggin spin_unlock(&fs->lock); 43431e6b01fSNick Piggin return -ECHILD; 43531e6b01fSNick Piggin } 43631e6b01fSNick Piggin 43731e6b01fSNick Piggin /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ 43831e6b01fSNick Piggin static inline int nameidata_drop_rcu_maybe(struct nameidata *nd) 43931e6b01fSNick Piggin { 44031e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) 44131e6b01fSNick Piggin return nameidata_drop_rcu(nd); 44231e6b01fSNick Piggin return 0; 44331e6b01fSNick Piggin } 44431e6b01fSNick Piggin 44531e6b01fSNick Piggin /** 44631e6b01fSNick Piggin * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk 44731e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 44831e6b01fSNick Piggin * @dentry: dentry to drop 44939191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 45031e6b01fSNick Piggin * 45131e6b01fSNick Piggin * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root, 45231e6b01fSNick Piggin * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on 45331e6b01fSNick Piggin * @nd. Must be called from rcu-walk context. 45431e6b01fSNick Piggin */ 45531e6b01fSNick Piggin static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry) 45631e6b01fSNick Piggin { 45731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 45831e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 4595b6ca027SAl Viro int want_root = 0; 46031e6b01fSNick Piggin 46131e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4625b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4635b6ca027SAl Viro want_root = 1; 46431e6b01fSNick Piggin spin_lock(&fs->lock); 46531e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 46631e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 46731e6b01fSNick Piggin goto err_root; 46831e6b01fSNick Piggin } 46931e6b01fSNick Piggin spin_lock(&parent->d_lock); 47031e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 47131e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 47231e6b01fSNick Piggin goto err; 47331e6b01fSNick Piggin /* 47431e6b01fSNick Piggin * If the sequence check on the child dentry passed, then the child has 47531e6b01fSNick Piggin * not been removed from its parent. This means the parent dentry must 47631e6b01fSNick Piggin * be valid and able to take a reference at this point. 47731e6b01fSNick Piggin */ 47831e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 47931e6b01fSNick Piggin BUG_ON(!parent->d_count); 48031e6b01fSNick Piggin parent->d_count++; 48131e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 48231e6b01fSNick Piggin spin_unlock(&parent->d_lock); 4835b6ca027SAl Viro if (want_root) { 48431e6b01fSNick Piggin path_get(&nd->root); 48531e6b01fSNick Piggin spin_unlock(&fs->lock); 48631e6b01fSNick Piggin } 48731e6b01fSNick Piggin mntget(nd->path.mnt); 48831e6b01fSNick Piggin 48931e6b01fSNick Piggin rcu_read_unlock(); 49031e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 49131e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 49231e6b01fSNick Piggin return 0; 49331e6b01fSNick Piggin err: 49431e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 49531e6b01fSNick Piggin spin_unlock(&parent->d_lock); 49631e6b01fSNick Piggin err_root: 4975b6ca027SAl Viro if (want_root) 49831e6b01fSNick Piggin spin_unlock(&fs->lock); 49931e6b01fSNick Piggin return -ECHILD; 50031e6b01fSNick Piggin } 50131e6b01fSNick Piggin 50231e6b01fSNick Piggin /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ 50331e6b01fSNick Piggin static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry) 50431e6b01fSNick Piggin { 505a7472babSAl Viro if (nd->flags & LOOKUP_RCU) { 506a7472babSAl Viro if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) { 507a7472babSAl Viro nd->flags &= ~LOOKUP_RCU; 5085b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 509a7472babSAl Viro nd->root.mnt = NULL; 510a7472babSAl Viro rcu_read_unlock(); 511a7472babSAl Viro br_read_unlock(vfsmount_lock); 512a7472babSAl Viro return -ECHILD; 513a7472babSAl Viro } 514a7472babSAl Viro } 51531e6b01fSNick Piggin return 0; 51631e6b01fSNick Piggin } 51731e6b01fSNick Piggin 51831e6b01fSNick Piggin /** 51931e6b01fSNick Piggin * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk 52031e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 52139191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 52231e6b01fSNick Piggin * 52331e6b01fSNick Piggin * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk. 52431e6b01fSNick Piggin * nd->path should be the final element of the lookup, so nd->root is discarded. 52531e6b01fSNick Piggin * Must be called from rcu-walk context. 52631e6b01fSNick Piggin */ 52731e6b01fSNick Piggin static int nameidata_drop_rcu_last(struct nameidata *nd) 52831e6b01fSNick Piggin { 52931e6b01fSNick Piggin struct dentry *dentry = nd->path.dentry; 53031e6b01fSNick Piggin 53131e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 53231e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 5335b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 53431e6b01fSNick Piggin nd->root.mnt = NULL; 53531e6b01fSNick Piggin spin_lock(&dentry->d_lock); 53631e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 53731e6b01fSNick Piggin goto err_unlock; 53831e6b01fSNick Piggin BUG_ON(nd->inode != dentry->d_inode); 53931e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 54031e6b01fSNick Piggin 54131e6b01fSNick Piggin mntget(nd->path.mnt); 54231e6b01fSNick Piggin 54331e6b01fSNick Piggin rcu_read_unlock(); 54431e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 54531e6b01fSNick Piggin 54631e6b01fSNick Piggin return 0; 54731e6b01fSNick Piggin 54831e6b01fSNick Piggin err_unlock: 54931e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 55031e6b01fSNick Piggin rcu_read_unlock(); 55131e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 55231e6b01fSNick Piggin return -ECHILD; 55331e6b01fSNick Piggin } 55431e6b01fSNick Piggin 55531e6b01fSNick Piggin /** 556834f2a4aSTrond Myklebust * release_open_intent - free up open intent resources 557834f2a4aSTrond Myklebust * @nd: pointer to nameidata 558834f2a4aSTrond Myklebust */ 559834f2a4aSTrond Myklebust void release_open_intent(struct nameidata *nd) 560834f2a4aSTrond Myklebust { 5612dab5974SLinus Torvalds struct file *file = nd->intent.open.file; 5622dab5974SLinus Torvalds 5632dab5974SLinus Torvalds if (file && !IS_ERR(file)) { 5642dab5974SLinus Torvalds if (file->f_path.dentry == NULL) 5652dab5974SLinus Torvalds put_filp(file); 566834f2a4aSTrond Myklebust else 5672dab5974SLinus Torvalds fput(file); 5682dab5974SLinus Torvalds } 569834f2a4aSTrond Myklebust } 570834f2a4aSTrond Myklebust 571f60aef7eSAl Viro static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) 57234286d66SNick Piggin { 573f60aef7eSAl Viro return dentry->d_op->d_revalidate(dentry, nd); 57434286d66SNick Piggin } 57534286d66SNick Piggin 576f5e1c1c1SAl Viro static struct dentry * 577bcdc5e01SIan Kent do_revalidate(struct dentry *dentry, struct nameidata *nd) 578bcdc5e01SIan Kent { 579f5e1c1c1SAl Viro int status = d_revalidate(dentry, nd); 580bcdc5e01SIan Kent if (unlikely(status <= 0)) { 581bcdc5e01SIan Kent /* 582bcdc5e01SIan Kent * The dentry failed validation. 583bcdc5e01SIan Kent * If d_revalidate returned 0 attempt to invalidate 584bcdc5e01SIan Kent * the dentry otherwise d_revalidate is asking us 585bcdc5e01SIan Kent * to return a fail status. 586bcdc5e01SIan Kent */ 58734286d66SNick Piggin if (status < 0) { 58834286d66SNick Piggin dput(dentry); 58934286d66SNick Piggin dentry = ERR_PTR(status); 590f5e1c1c1SAl Viro } else if (!d_invalidate(dentry)) { 591bcdc5e01SIan Kent dput(dentry); 592bcdc5e01SIan Kent dentry = NULL; 593bcdc5e01SIan Kent } 594bcdc5e01SIan Kent } 595f5e1c1c1SAl Viro return dentry; 596f5e1c1c1SAl Viro } 597f5e1c1c1SAl Viro 5981da177e4SLinus Torvalds /* 59916c2cd71SAl Viro * handle_reval_path - force revalidation of a dentry 60039159de2SJeff Layton * 60139159de2SJeff Layton * In some situations the path walking code will trust dentries without 60239159de2SJeff Layton * revalidating them. This causes problems for filesystems that depend on 60339159de2SJeff Layton * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set 60439159de2SJeff Layton * (which indicates that it's possible for the dentry to go stale), force 60539159de2SJeff Layton * a d_revalidate call before proceeding. 60639159de2SJeff Layton * 60739159de2SJeff Layton * Returns 0 if the revalidation was successful. If the revalidation fails, 60839159de2SJeff Layton * either return the error returned by d_revalidate or -ESTALE if the 60939159de2SJeff Layton * revalidation it just returned 0. If d_revalidate returns 0, we attempt to 61039159de2SJeff Layton * invalidate the dentry. It's up to the caller to handle putting references 61139159de2SJeff Layton * to the path if necessary. 61239159de2SJeff Layton */ 61316c2cd71SAl Viro static inline int handle_reval_path(struct nameidata *nd) 61439159de2SJeff Layton { 61516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 61639159de2SJeff Layton int status; 61739159de2SJeff Layton 61816c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 61939159de2SJeff Layton return 0; 62039159de2SJeff Layton 62116c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 62216c2cd71SAl Viro return 0; 62316c2cd71SAl Viro 62416c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 62516c2cd71SAl Viro return 0; 62616c2cd71SAl Viro 62716c2cd71SAl Viro /* Note: we do not d_invalidate() */ 62834286d66SNick Piggin status = d_revalidate(dentry, nd); 62939159de2SJeff Layton if (status > 0) 63039159de2SJeff Layton return 0; 63139159de2SJeff Layton 63216c2cd71SAl Viro if (!status) 63339159de2SJeff Layton status = -ESTALE; 63416c2cd71SAl Viro 63539159de2SJeff Layton return status; 63639159de2SJeff Layton } 63739159de2SJeff Layton 63839159de2SJeff Layton /* 639b75b5086SAl Viro * Short-cut version of permission(), for calling on directories 640b75b5086SAl Viro * during pathname resolution. Combines parts of permission() 641b75b5086SAl Viro * and generic_permission(), and tests ONLY for MAY_EXEC permission. 6421da177e4SLinus Torvalds * 6431da177e4SLinus Torvalds * If appropriate, check DAC only. If not appropriate, or 644b75b5086SAl Viro * short-cut DAC fails, then call ->permission() to do more 6451da177e4SLinus Torvalds * complete permission check. 6461da177e4SLinus Torvalds */ 647b74c79e9SNick Piggin static inline int exec_permission(struct inode *inode, unsigned int flags) 6481da177e4SLinus Torvalds { 6495909ccaaSLinus Torvalds int ret; 6501da177e4SLinus Torvalds 651cb9179eaSLinus Torvalds if (inode->i_op->permission) { 652b74c79e9SNick Piggin ret = inode->i_op->permission(inode, MAY_EXEC, flags); 653b74c79e9SNick Piggin } else { 654b74c79e9SNick Piggin ret = acl_permission_check(inode, MAY_EXEC, flags, 655b74c79e9SNick Piggin inode->i_op->check_acl); 656cb9179eaSLinus Torvalds } 657b74c79e9SNick Piggin if (likely(!ret)) 6581da177e4SLinus Torvalds goto ok; 659b74c79e9SNick Piggin if (ret == -ECHILD) 66031e6b01fSNick Piggin return ret; 6611da177e4SLinus Torvalds 662f1ac9f6bSLinus Torvalds if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) 6631da177e4SLinus Torvalds goto ok; 6641da177e4SLinus Torvalds 6655909ccaaSLinus Torvalds return ret; 6661da177e4SLinus Torvalds ok: 667b74c79e9SNick Piggin return security_inode_exec_permission(inode, flags); 6681da177e4SLinus Torvalds } 6691da177e4SLinus Torvalds 6702a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6712a737871SAl Viro { 672f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 673f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6742a737871SAl Viro } 6752a737871SAl Viro 6766de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6776de88d72SAl Viro 67831e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 67931e6b01fSNick Piggin { 68031e6b01fSNick Piggin if (!nd->root.mnt) { 68131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 682c28cc364SNick Piggin unsigned seq; 683c28cc364SNick Piggin 684c28cc364SNick Piggin do { 685c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 68631e6b01fSNick Piggin nd->root = fs->root; 687c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 68831e6b01fSNick Piggin } 68931e6b01fSNick Piggin } 69031e6b01fSNick Piggin 691f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 6921da177e4SLinus Torvalds { 69331e6b01fSNick Piggin int ret; 69431e6b01fSNick Piggin 6951da177e4SLinus Torvalds if (IS_ERR(link)) 6961da177e4SLinus Torvalds goto fail; 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds if (*link == '/') { 6992a737871SAl Viro set_root(nd); 7001d957f9bSJan Blunck path_put(&nd->path); 7012a737871SAl Viro nd->path = nd->root; 7022a737871SAl Viro path_get(&nd->root); 70316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 7041da177e4SLinus Torvalds } 70531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 706b4091d5fSChristoph Hellwig 70731e6b01fSNick Piggin ret = link_path_walk(link, nd); 70831e6b01fSNick Piggin return ret; 7091da177e4SLinus Torvalds fail: 7101d957f9bSJan Blunck path_put(&nd->path); 7111da177e4SLinus Torvalds return PTR_ERR(link); 7121da177e4SLinus Torvalds } 7131da177e4SLinus Torvalds 7141d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 715051d3812SIan Kent { 716051d3812SIan Kent dput(path->dentry); 7174ac91378SJan Blunck if (path->mnt != nd->path.mnt) 718051d3812SIan Kent mntput(path->mnt); 719051d3812SIan Kent } 720051d3812SIan Kent 7217b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 7227b9337aaSNick Piggin struct nameidata *nd) 723051d3812SIan Kent { 72431e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 7254ac91378SJan Blunck dput(nd->path.dentry); 72631e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 7274ac91378SJan Blunck mntput(nd->path.mnt); 7289a229683SHuang Shijie } 72931e6b01fSNick Piggin nd->path.mnt = path->mnt; 7304ac91378SJan Blunck nd->path.dentry = path->dentry; 731051d3812SIan Kent } 732051d3812SIan Kent 733def4af30SAl Viro static __always_inline int 7347b9337aaSNick Piggin __do_follow_link(const struct path *link, struct nameidata *nd, void **p) 7351da177e4SLinus Torvalds { 7361da177e4SLinus Torvalds int error; 7377b9337aaSNick Piggin struct dentry *dentry = link->dentry; 7381da177e4SLinus Torvalds 739844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 740844a3917SAl Viro 7417b9337aaSNick Piggin touch_atime(link->mnt, dentry); 7421da177e4SLinus Torvalds nd_set_link(nd, NULL); 743cd4e91d3SAl Viro 74487556ef1SDavid Howells if (link->mnt == nd->path.mnt) 7457b9337aaSNick Piggin mntget(link->mnt); 74631e6b01fSNick Piggin 74736f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 74836f3b4f6SAl Viro if (error) { 74936f3b4f6SAl Viro *p = ERR_PTR(error); /* no ->put_link(), please */ 75036f3b4f6SAl Viro path_put(&nd->path); 75136f3b4f6SAl Viro return error; 75236f3b4f6SAl Viro } 75336f3b4f6SAl Viro 75486acdca1SAl Viro nd->last_type = LAST_BIND; 755def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 756def4af30SAl Viro error = PTR_ERR(*p); 757def4af30SAl Viro if (!IS_ERR(*p)) { 7581da177e4SLinus Torvalds char *s = nd_get_link(nd); 759cc314eefSLinus Torvalds error = 0; 7601da177e4SLinus Torvalds if (s) 7611da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 76216c2cd71SAl Viro else if (nd->last_type == LAST_BIND) 76316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 7641da177e4SLinus Torvalds } 7651da177e4SLinus Torvalds return error; 7661da177e4SLinus Torvalds } 7671da177e4SLinus Torvalds 7681da177e4SLinus Torvalds /* 7691da177e4SLinus Torvalds * This limits recursive symlink follows to 8, while 7701da177e4SLinus Torvalds * limiting consecutive symlinks to 40. 7711da177e4SLinus Torvalds * 7721da177e4SLinus Torvalds * Without that kind of total limit, nasty chains of consecutive 7731da177e4SLinus Torvalds * symlinks can cause almost arbitrarily long lookups. 7741da177e4SLinus Torvalds */ 7753abb17e8SLinus Torvalds static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd) 7761da177e4SLinus Torvalds { 777def4af30SAl Viro void *cookie; 7781da177e4SLinus Torvalds int err = -ELOOP; 779844a3917SAl Viro 780844a3917SAl Viro /* We drop rcu-walk here */ 781844a3917SAl Viro if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) 782844a3917SAl Viro return -ECHILD; 7833abb17e8SLinus Torvalds BUG_ON(inode != path->dentry->d_inode); 784844a3917SAl Viro 7851da177e4SLinus Torvalds if (current->link_count >= MAX_NESTED_LINKS) 7861da177e4SLinus Torvalds goto loop; 7871da177e4SLinus Torvalds if (current->total_link_count >= 40) 7881da177e4SLinus Torvalds goto loop; 7891da177e4SLinus Torvalds BUG_ON(nd->depth >= MAX_NESTED_LINKS); 7901da177e4SLinus Torvalds cond_resched(); 7911da177e4SLinus Torvalds current->link_count++; 7921da177e4SLinus Torvalds current->total_link_count++; 7931da177e4SLinus Torvalds nd->depth++; 794def4af30SAl Viro err = __do_follow_link(path, nd, &cookie); 795def4af30SAl Viro if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) 796def4af30SAl Viro path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); 797258fa999SAl Viro path_put(path); 7981da177e4SLinus Torvalds current->link_count--; 7991da177e4SLinus Torvalds nd->depth--; 8001da177e4SLinus Torvalds return err; 8011da177e4SLinus Torvalds loop: 8021d957f9bSJan Blunck path_put_conditional(path, nd); 8031d957f9bSJan Blunck path_put(&nd->path); 8041da177e4SLinus Torvalds return err; 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 80731e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 80831e6b01fSNick Piggin { 80931e6b01fSNick Piggin struct vfsmount *parent; 81031e6b01fSNick Piggin struct dentry *mountpoint; 81131e6b01fSNick Piggin 81231e6b01fSNick Piggin parent = path->mnt->mnt_parent; 81331e6b01fSNick Piggin if (parent == path->mnt) 81431e6b01fSNick Piggin return 0; 81531e6b01fSNick Piggin mountpoint = path->mnt->mnt_mountpoint; 81631e6b01fSNick Piggin path->dentry = mountpoint; 81731e6b01fSNick Piggin path->mnt = parent; 81831e6b01fSNick Piggin return 1; 81931e6b01fSNick Piggin } 82031e6b01fSNick Piggin 821bab77ebfSAl Viro int follow_up(struct path *path) 8221da177e4SLinus Torvalds { 8231da177e4SLinus Torvalds struct vfsmount *parent; 8241da177e4SLinus Torvalds struct dentry *mountpoint; 82599b7db7bSNick Piggin 82699b7db7bSNick Piggin br_read_lock(vfsmount_lock); 827bab77ebfSAl Viro parent = path->mnt->mnt_parent; 828bab77ebfSAl Viro if (parent == path->mnt) { 82999b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 8301da177e4SLinus Torvalds return 0; 8311da177e4SLinus Torvalds } 8321da177e4SLinus Torvalds mntget(parent); 833bab77ebfSAl Viro mountpoint = dget(path->mnt->mnt_mountpoint); 83499b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 835bab77ebfSAl Viro dput(path->dentry); 836bab77ebfSAl Viro path->dentry = mountpoint; 837bab77ebfSAl Viro mntput(path->mnt); 838bab77ebfSAl Viro path->mnt = parent; 8391da177e4SLinus Torvalds return 1; 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds 842b5c84bf6SNick Piggin /* 8439875cf80SDavid Howells * Perform an automount 8449875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 8459875cf80SDavid Howells * were called with. 8461da177e4SLinus Torvalds */ 8479875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 8489875cf80SDavid Howells bool *need_mntput) 84931e6b01fSNick Piggin { 8509875cf80SDavid Howells struct vfsmount *mnt; 851ea5b778aSDavid Howells int err; 8529875cf80SDavid Howells 8539875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 8549875cf80SDavid Howells return -EREMOTE; 8559875cf80SDavid Howells 8566f45b656SDavid Howells /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT 8576f45b656SDavid Howells * and this is the terminal part of the path. 8586f45b656SDavid Howells */ 8596f45b656SDavid Howells if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE)) 8606f45b656SDavid Howells return -EISDIR; /* we actually want to stop here */ 8616f45b656SDavid Howells 8629875cf80SDavid Howells /* We want to mount if someone is trying to open/create a file of any 8639875cf80SDavid Howells * type under the mountpoint, wants to traverse through the mountpoint 8649875cf80SDavid Howells * or wants to open the mounted directory. 8659875cf80SDavid Howells * 8669875cf80SDavid Howells * We don't want to mount if someone's just doing a stat and they've 8679875cf80SDavid Howells * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and 8689875cf80SDavid Howells * appended a '/' to the name. 8699875cf80SDavid Howells */ 8709875cf80SDavid Howells if (!(flags & LOOKUP_FOLLOW) && 8719875cf80SDavid Howells !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | 8729875cf80SDavid Howells LOOKUP_OPEN | LOOKUP_CREATE))) 8739875cf80SDavid Howells return -EISDIR; 8749875cf80SDavid Howells 8759875cf80SDavid Howells current->total_link_count++; 8769875cf80SDavid Howells if (current->total_link_count >= 40) 8779875cf80SDavid Howells return -ELOOP; 8789875cf80SDavid Howells 8799875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 8809875cf80SDavid Howells if (IS_ERR(mnt)) { 8819875cf80SDavid Howells /* 8829875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 8839875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 8849875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 8859875cf80SDavid Howells * 8869875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 8879875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 8889875cf80SDavid Howells * the path is inaccessible and we should say so. 8899875cf80SDavid Howells */ 8909875cf80SDavid Howells if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) 8919875cf80SDavid Howells return -EREMOTE; 8929875cf80SDavid Howells return PTR_ERR(mnt); 89331e6b01fSNick Piggin } 894ea5b778aSDavid Howells 8959875cf80SDavid Howells if (!mnt) /* mount collision */ 8969875cf80SDavid Howells return 0; 8979875cf80SDavid Howells 89819a167afSAl Viro err = finish_automount(mnt, path); 899ea5b778aSDavid Howells 900ea5b778aSDavid Howells switch (err) { 901ea5b778aSDavid Howells case -EBUSY: 902ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 90319a167afSAl Viro return 0; 904ea5b778aSDavid Howells case 0: 905463ffb2eSAl Viro dput(path->dentry); 9069875cf80SDavid Howells if (*need_mntput) 9079875cf80SDavid Howells mntput(path->mnt); 9089875cf80SDavid Howells path->mnt = mnt; 9099875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 9109875cf80SDavid Howells *need_mntput = true; 9119875cf80SDavid Howells return 0; 91219a167afSAl Viro default: 91319a167afSAl Viro return err; 9149875cf80SDavid Howells } 91519a167afSAl Viro 916ea5b778aSDavid Howells } 9179875cf80SDavid Howells 9189875cf80SDavid Howells /* 9199875cf80SDavid Howells * Handle a dentry that is managed in some way. 920cc53ce53SDavid Howells * - Flagged for transit management (autofs) 9219875cf80SDavid Howells * - Flagged as mountpoint 9229875cf80SDavid Howells * - Flagged as automount point 9239875cf80SDavid Howells * 9249875cf80SDavid Howells * This may only be called in refwalk mode. 9259875cf80SDavid Howells * 9269875cf80SDavid Howells * Serialization is taken care of in namespace.c 9279875cf80SDavid Howells */ 9289875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 9299875cf80SDavid Howells { 9309875cf80SDavid Howells unsigned managed; 9319875cf80SDavid Howells bool need_mntput = false; 9329875cf80SDavid Howells int ret; 9339875cf80SDavid Howells 9349875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 9359875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 9369875cf80SDavid Howells * the components of that value change under us */ 9379875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 9389875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 9399875cf80SDavid Howells unlikely(managed != 0)) { 940cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 941cc53ce53SDavid Howells * being held. */ 942cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 943cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 944cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 945ab90911fSDavid Howells ret = path->dentry->d_op->d_manage(path->dentry, 946ab90911fSDavid Howells false, false); 947cc53ce53SDavid Howells if (ret < 0) 948cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 949cc53ce53SDavid Howells } 950cc53ce53SDavid Howells 9519875cf80SDavid Howells /* Transit to a mounted filesystem. */ 9529875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 9539875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 9549875cf80SDavid Howells if (mounted) { 9559875cf80SDavid Howells dput(path->dentry); 9569875cf80SDavid Howells if (need_mntput) 957463ffb2eSAl Viro mntput(path->mnt); 958463ffb2eSAl Viro path->mnt = mounted; 959463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 9609875cf80SDavid Howells need_mntput = true; 9619875cf80SDavid Howells continue; 962463ffb2eSAl Viro } 963463ffb2eSAl Viro 9649875cf80SDavid Howells /* Something is mounted on this dentry in another 9659875cf80SDavid Howells * namespace and/or whatever was mounted there in this 9669875cf80SDavid Howells * namespace got unmounted before we managed to get the 9679875cf80SDavid Howells * vfsmount_lock */ 9681da177e4SLinus Torvalds } 9699875cf80SDavid Howells 9709875cf80SDavid Howells /* Handle an automount point */ 9719875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 9729875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 9739875cf80SDavid Howells if (ret < 0) 9749875cf80SDavid Howells return ret == -EISDIR ? 0 : ret; 9759875cf80SDavid Howells continue; 9769875cf80SDavid Howells } 9779875cf80SDavid Howells 9789875cf80SDavid Howells /* We didn't change the current path point */ 9799875cf80SDavid Howells break; 9809875cf80SDavid Howells } 9819875cf80SDavid Howells return 0; 9821da177e4SLinus Torvalds } 9831da177e4SLinus Torvalds 984cc53ce53SDavid Howells int follow_down_one(struct path *path) 9851da177e4SLinus Torvalds { 9861da177e4SLinus Torvalds struct vfsmount *mounted; 9871da177e4SLinus Torvalds 9881c755af4SAl Viro mounted = lookup_mnt(path); 9891da177e4SLinus Torvalds if (mounted) { 9909393bd07SAl Viro dput(path->dentry); 9919393bd07SAl Viro mntput(path->mnt); 9929393bd07SAl Viro path->mnt = mounted; 9939393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 9941da177e4SLinus Torvalds return 1; 9951da177e4SLinus Torvalds } 9961da177e4SLinus Torvalds return 0; 9971da177e4SLinus Torvalds } 9981da177e4SLinus Torvalds 9999875cf80SDavid Howells /* 10009875cf80SDavid Howells * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we 1001cc53ce53SDavid Howells * meet a managed dentry and we're not walking to "..". True is returned to 10029875cf80SDavid Howells * continue, false to abort. 10039875cf80SDavid Howells */ 10049875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 10059875cf80SDavid Howells struct inode **inode, bool reverse_transit) 10069875cf80SDavid Howells { 10079875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 10089875cf80SDavid Howells struct vfsmount *mounted; 1009ab90911fSDavid Howells if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) && 1010ab90911fSDavid Howells !reverse_transit && 1011ab90911fSDavid Howells path->dentry->d_op->d_manage(path->dentry, false, true) < 0) 1012ab90911fSDavid Howells return false; 10139875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 10149875cf80SDavid Howells if (!mounted) 10159875cf80SDavid Howells break; 10169875cf80SDavid Howells path->mnt = mounted; 10179875cf80SDavid Howells path->dentry = mounted->mnt_root; 10189875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 10199875cf80SDavid Howells *inode = path->dentry->d_inode; 10209875cf80SDavid Howells } 10219875cf80SDavid Howells 10229875cf80SDavid Howells if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 10239875cf80SDavid Howells return reverse_transit; 10249875cf80SDavid Howells return true; 10259875cf80SDavid Howells } 10269875cf80SDavid Howells 102731e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 102831e6b01fSNick Piggin { 102931e6b01fSNick Piggin struct inode *inode = nd->inode; 103031e6b01fSNick Piggin 103131e6b01fSNick Piggin set_root_rcu(nd); 103231e6b01fSNick Piggin 103331e6b01fSNick Piggin while (1) { 103431e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 103531e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 103631e6b01fSNick Piggin break; 103731e6b01fSNick Piggin } 103831e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 103931e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 104031e6b01fSNick Piggin struct dentry *parent = old->d_parent; 104131e6b01fSNick Piggin unsigned seq; 104231e6b01fSNick Piggin 104331e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 104431e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1045ef7562d5SAl Viro goto failed; 104631e6b01fSNick Piggin inode = parent->d_inode; 104731e6b01fSNick Piggin nd->path.dentry = parent; 104831e6b01fSNick Piggin nd->seq = seq; 104931e6b01fSNick Piggin break; 105031e6b01fSNick Piggin } 105131e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 105231e6b01fSNick Piggin break; 105331e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 105431e6b01fSNick Piggin inode = nd->path.dentry->d_inode; 105531e6b01fSNick Piggin } 10569875cf80SDavid Howells __follow_mount_rcu(nd, &nd->path, &inode, true); 105731e6b01fSNick Piggin nd->inode = inode; 105831e6b01fSNick Piggin return 0; 1059ef7562d5SAl Viro 1060ef7562d5SAl Viro failed: 1061ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 10625b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1063ef7562d5SAl Viro nd->root.mnt = NULL; 1064ef7562d5SAl Viro rcu_read_unlock(); 1065ef7562d5SAl Viro br_read_unlock(vfsmount_lock); 1066ef7562d5SAl Viro return -ECHILD; 106731e6b01fSNick Piggin } 106831e6b01fSNick Piggin 10699875cf80SDavid Howells /* 1070cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1071cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1072cc53ce53SDavid Howells * caller is permitted to proceed or not. 1073cc53ce53SDavid Howells * 1074cc53ce53SDavid Howells * Care must be taken as namespace_sem may be held (indicated by mounting_here 1075cc53ce53SDavid Howells * being true). 1076cc53ce53SDavid Howells */ 1077cc53ce53SDavid Howells int follow_down(struct path *path, bool mounting_here) 1078cc53ce53SDavid Howells { 1079cc53ce53SDavid Howells unsigned managed; 1080cc53ce53SDavid Howells int ret; 1081cc53ce53SDavid Howells 1082cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1083cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1084cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1085cc53ce53SDavid Howells * being held. 1086cc53ce53SDavid Howells * 1087cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1088cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1089cc53ce53SDavid Howells * other than its daemon the right to mount on its 1090cc53ce53SDavid Howells * superstructure. 1091cc53ce53SDavid Howells * 1092cc53ce53SDavid Howells * The filesystem may sleep at this point. 1093cc53ce53SDavid Howells */ 1094cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1095cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1096cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1097ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 1098ab90911fSDavid Howells path->dentry, mounting_here, false); 1099cc53ce53SDavid Howells if (ret < 0) 1100cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1101cc53ce53SDavid Howells } 1102cc53ce53SDavid Howells 1103cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1104cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1105cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1106cc53ce53SDavid Howells if (!mounted) 1107cc53ce53SDavid Howells break; 1108cc53ce53SDavid Howells dput(path->dentry); 1109cc53ce53SDavid Howells mntput(path->mnt); 1110cc53ce53SDavid Howells path->mnt = mounted; 1111cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1112cc53ce53SDavid Howells continue; 1113cc53ce53SDavid Howells } 1114cc53ce53SDavid Howells 1115cc53ce53SDavid Howells /* Don't handle automount points here */ 1116cc53ce53SDavid Howells break; 1117cc53ce53SDavid Howells } 1118cc53ce53SDavid Howells return 0; 1119cc53ce53SDavid Howells } 1120cc53ce53SDavid Howells 1121cc53ce53SDavid Howells /* 11229875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 11239875cf80SDavid Howells */ 11249875cf80SDavid Howells static void follow_mount(struct path *path) 11259875cf80SDavid Howells { 11269875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 11279875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11289875cf80SDavid Howells if (!mounted) 11299875cf80SDavid Howells break; 11309875cf80SDavid Howells dput(path->dentry); 11319875cf80SDavid Howells mntput(path->mnt); 11329875cf80SDavid Howells path->mnt = mounted; 11339875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 11349875cf80SDavid Howells } 11359875cf80SDavid Howells } 11369875cf80SDavid Howells 113731e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 11381da177e4SLinus Torvalds { 11392a737871SAl Viro set_root(nd); 1140e518ddb7SAndreas Mohr 11411da177e4SLinus Torvalds while(1) { 11424ac91378SJan Blunck struct dentry *old = nd->path.dentry; 11431da177e4SLinus Torvalds 11442a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 11452a737871SAl Viro nd->path.mnt == nd->root.mnt) { 11461da177e4SLinus Torvalds break; 11471da177e4SLinus Torvalds } 11484ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 11493088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 11503088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 11511da177e4SLinus Torvalds dput(old); 11521da177e4SLinus Torvalds break; 11531da177e4SLinus Torvalds } 11543088dd70SAl Viro if (!follow_up(&nd->path)) 11551da177e4SLinus Torvalds break; 11561da177e4SLinus Torvalds } 115779ed0226SAl Viro follow_mount(&nd->path); 115831e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 11591da177e4SLinus Torvalds } 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds /* 1162baa03890SNick Piggin * Allocate a dentry with name and parent, and perform a parent 1163baa03890SNick Piggin * directory ->lookup on it. Returns the new dentry, or ERR_PTR 1164baa03890SNick Piggin * on error. parent->d_inode->i_mutex must be held. d_lookup must 1165baa03890SNick Piggin * have verified that no child exists while under i_mutex. 1166baa03890SNick Piggin */ 1167baa03890SNick Piggin static struct dentry *d_alloc_and_lookup(struct dentry *parent, 1168baa03890SNick Piggin struct qstr *name, struct nameidata *nd) 1169baa03890SNick Piggin { 1170baa03890SNick Piggin struct inode *inode = parent->d_inode; 1171baa03890SNick Piggin struct dentry *dentry; 1172baa03890SNick Piggin struct dentry *old; 1173baa03890SNick Piggin 1174baa03890SNick Piggin /* Don't create child dentry for a dead directory. */ 1175baa03890SNick Piggin if (unlikely(IS_DEADDIR(inode))) 1176baa03890SNick Piggin return ERR_PTR(-ENOENT); 1177baa03890SNick Piggin 1178baa03890SNick Piggin dentry = d_alloc(parent, name); 1179baa03890SNick Piggin if (unlikely(!dentry)) 1180baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1181baa03890SNick Piggin 1182baa03890SNick Piggin old = inode->i_op->lookup(inode, dentry, nd); 1183baa03890SNick Piggin if (unlikely(old)) { 1184baa03890SNick Piggin dput(dentry); 1185baa03890SNick Piggin dentry = old; 1186baa03890SNick Piggin } 1187baa03890SNick Piggin return dentry; 1188baa03890SNick Piggin } 1189baa03890SNick Piggin 1190baa03890SNick Piggin /* 11911da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 11921da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 11931da177e4SLinus Torvalds * It _is_ time-critical. 11941da177e4SLinus Torvalds */ 11951da177e4SLinus Torvalds static int do_lookup(struct nameidata *nd, struct qstr *name, 119631e6b01fSNick Piggin struct path *path, struct inode **inode) 11971da177e4SLinus Torvalds { 11984ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 119931e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 12005a18fff2SAl Viro int need_reval = 1; 12015a18fff2SAl Viro int status = 1; 12029875cf80SDavid Howells int err; 12039875cf80SDavid Howells 12043cac260aSAl Viro /* 1205b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1206b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1207b04f784eSNick Piggin * do the non-racy lookup, below. 1208b04f784eSNick Piggin */ 120931e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 121031e6b01fSNick Piggin unsigned seq; 121131e6b01fSNick Piggin *inode = nd->inode; 121231e6b01fSNick Piggin dentry = __d_lookup_rcu(parent, name, &seq, inode); 12135a18fff2SAl Viro if (!dentry) 12145a18fff2SAl Viro goto unlazy; 12155a18fff2SAl Viro 121631e6b01fSNick Piggin /* Memory barrier in read_seqcount_begin of child is enough */ 121731e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 121831e6b01fSNick Piggin return -ECHILD; 121931e6b01fSNick Piggin nd->seq = seq; 12205a18fff2SAl Viro 122124643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 12225a18fff2SAl Viro status = d_revalidate(dentry, nd); 12235a18fff2SAl Viro if (unlikely(status <= 0)) { 12245a18fff2SAl Viro if (status != -ECHILD) 12255a18fff2SAl Viro need_reval = 0; 12265a18fff2SAl Viro goto unlazy; 12275a18fff2SAl Viro } 122824643087SAl Viro } 122931e6b01fSNick Piggin path->mnt = mnt; 123031e6b01fSNick Piggin path->dentry = dentry; 12319875cf80SDavid Howells if (likely(__follow_mount_rcu(nd, path, inode, false))) 12329875cf80SDavid Howells return 0; 12335a18fff2SAl Viro unlazy: 12345a18fff2SAl Viro if (dentry) { 12355a18fff2SAl Viro if (nameidata_dentry_drop_rcu(nd, dentry)) 12365a18fff2SAl Viro return -ECHILD; 12375a18fff2SAl Viro } else { 12389875cf80SDavid Howells if (nameidata_drop_rcu(nd)) 12399875cf80SDavid Howells return -ECHILD; 12409875cf80SDavid Howells } 12415a18fff2SAl Viro } else { 124231e6b01fSNick Piggin dentry = __d_lookup(parent, name); 124324643087SAl Viro } 12445a18fff2SAl Viro 12455a18fff2SAl Viro retry: 12465a18fff2SAl Viro if (unlikely(!dentry)) { 12475a18fff2SAl Viro struct inode *dir = parent->d_inode; 12485a18fff2SAl Viro BUG_ON(nd->inode != dir); 12495a18fff2SAl Viro 12505a18fff2SAl Viro mutex_lock(&dir->i_mutex); 12515a18fff2SAl Viro dentry = d_lookup(parent, name); 12525a18fff2SAl Viro if (likely(!dentry)) { 12535a18fff2SAl Viro dentry = d_alloc_and_lookup(parent, name, nd); 12545a18fff2SAl Viro if (IS_ERR(dentry)) { 12555a18fff2SAl Viro mutex_unlock(&dir->i_mutex); 12565a18fff2SAl Viro return PTR_ERR(dentry); 12575a18fff2SAl Viro } 12585a18fff2SAl Viro /* known good */ 12595a18fff2SAl Viro need_reval = 0; 12605a18fff2SAl Viro status = 1; 12615a18fff2SAl Viro } 12625a18fff2SAl Viro mutex_unlock(&dir->i_mutex); 12635a18fff2SAl Viro } 12645a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 12655a18fff2SAl Viro status = d_revalidate(dentry, nd); 12665a18fff2SAl Viro if (unlikely(status <= 0)) { 12675a18fff2SAl Viro if (status < 0) { 12685a18fff2SAl Viro dput(dentry); 12695a18fff2SAl Viro return status; 12705a18fff2SAl Viro } 12715a18fff2SAl Viro if (!d_invalidate(dentry)) { 12725a18fff2SAl Viro dput(dentry); 12735a18fff2SAl Viro dentry = NULL; 12745a18fff2SAl Viro need_reval = 1; 12755a18fff2SAl Viro goto retry; 12765a18fff2SAl Viro } 12775a18fff2SAl Viro } 12785a18fff2SAl Viro 12791da177e4SLinus Torvalds path->mnt = mnt; 12801da177e4SLinus Torvalds path->dentry = dentry; 12819875cf80SDavid Howells err = follow_managed(path, nd->flags); 128289312214SIan Kent if (unlikely(err < 0)) { 128389312214SIan Kent path_put_conditional(path, nd); 12849875cf80SDavid Howells return err; 128589312214SIan Kent } 128631e6b01fSNick Piggin *inode = path->dentry->d_inode; 12871da177e4SLinus Torvalds return 0; 12881da177e4SLinus Torvalds } 12891da177e4SLinus Torvalds 129052094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 129152094c8aSAl Viro { 129252094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 129352094c8aSAl Viro int err = exec_permission(nd->inode, IPERM_FLAG_RCU); 129452094c8aSAl Viro if (err != -ECHILD) 129552094c8aSAl Viro return err; 129652094c8aSAl Viro if (nameidata_drop_rcu(nd)) 129752094c8aSAl Viro return -ECHILD; 129852094c8aSAl Viro } 129952094c8aSAl Viro return exec_permission(nd->inode, 0); 130052094c8aSAl Viro } 130152094c8aSAl Viro 13029856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 13039856fa1bSAl Viro { 13049856fa1bSAl Viro if (type == LAST_DOTDOT) { 13059856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 13069856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 13079856fa1bSAl Viro return -ECHILD; 13089856fa1bSAl Viro } else 13099856fa1bSAl Viro follow_dotdot(nd); 13109856fa1bSAl Viro } 13119856fa1bSAl Viro return 0; 13129856fa1bSAl Viro } 13139856fa1bSAl Viro 1314951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1315951361f9SAl Viro { 1316951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1317951361f9SAl Viro path_put(&nd->path); 1318951361f9SAl Viro } else { 1319951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 13205b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1321951361f9SAl Viro nd->root.mnt = NULL; 1322951361f9SAl Viro rcu_read_unlock(); 1323951361f9SAl Viro br_read_unlock(vfsmount_lock); 1324951361f9SAl Viro } 1325951361f9SAl Viro } 1326951361f9SAl Viro 13271da177e4SLinus Torvalds /* 13281da177e4SLinus Torvalds * Name resolution. 1329ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1330ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 13311da177e4SLinus Torvalds * 1332ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1333ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 13341da177e4SLinus Torvalds */ 13356de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 13361da177e4SLinus Torvalds { 13371da177e4SLinus Torvalds struct path next; 13381da177e4SLinus Torvalds int err; 13391da177e4SLinus Torvalds unsigned int lookup_flags = nd->flags; 13401da177e4SLinus Torvalds 13411da177e4SLinus Torvalds while (*name=='/') 13421da177e4SLinus Torvalds name++; 13431da177e4SLinus Torvalds if (!*name) 1344086e183aSAl Viro return 0; 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds if (nd->depth) 1347f55eab82STrond Myklebust lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 13501da177e4SLinus Torvalds for(;;) { 135131e6b01fSNick Piggin struct inode *inode; 13521da177e4SLinus Torvalds unsigned long hash; 13531da177e4SLinus Torvalds struct qstr this; 13541da177e4SLinus Torvalds unsigned int c; 1355fe479a58SAl Viro int type; 13561da177e4SLinus Torvalds 1357cdce5d6bSTrond Myklebust nd->flags |= LOOKUP_CONTINUE; 135852094c8aSAl Viro 135952094c8aSAl Viro err = may_lookup(nd); 13601da177e4SLinus Torvalds if (err) 13611da177e4SLinus Torvalds break; 13621da177e4SLinus Torvalds 13631da177e4SLinus Torvalds this.name = name; 13641da177e4SLinus Torvalds c = *(const unsigned char *)name; 13651da177e4SLinus Torvalds 13661da177e4SLinus Torvalds hash = init_name_hash(); 13671da177e4SLinus Torvalds do { 13681da177e4SLinus Torvalds name++; 13691da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 13701da177e4SLinus Torvalds c = *(const unsigned char *)name; 13711da177e4SLinus Torvalds } while (c && (c != '/')); 13721da177e4SLinus Torvalds this.len = name - (const char *) this.name; 13731da177e4SLinus Torvalds this.hash = end_name_hash(hash); 13741da177e4SLinus Torvalds 1375fe479a58SAl Viro type = LAST_NORM; 1376fe479a58SAl Viro if (this.name[0] == '.') switch (this.len) { 1377fe479a58SAl Viro case 2: 137816c2cd71SAl Viro if (this.name[1] == '.') { 1379fe479a58SAl Viro type = LAST_DOTDOT; 138016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 138116c2cd71SAl Viro } 1382fe479a58SAl Viro break; 1383fe479a58SAl Viro case 1: 1384fe479a58SAl Viro type = LAST_DOT; 1385fe479a58SAl Viro } 13865a202bcdSAl Viro if (likely(type == LAST_NORM)) { 13875a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 138816c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 13895a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 13905a202bcdSAl Viro err = parent->d_op->d_hash(parent, nd->inode, 13915a202bcdSAl Viro &this); 13925a202bcdSAl Viro if (err < 0) 13935a202bcdSAl Viro break; 13945a202bcdSAl Viro } 13955a202bcdSAl Viro } 1396fe479a58SAl Viro 13971da177e4SLinus Torvalds /* remove trailing slashes? */ 13981da177e4SLinus Torvalds if (!c) 13991da177e4SLinus Torvalds goto last_component; 14001da177e4SLinus Torvalds while (*++name == '/'); 14011da177e4SLinus Torvalds if (!*name) 14021da177e4SLinus Torvalds goto last_with_slashes; 14031da177e4SLinus Torvalds 14041da177e4SLinus Torvalds /* 14051da177e4SLinus Torvalds * "." and ".." are special - ".." especially so because it has 14061da177e4SLinus Torvalds * to be able to know about the current root directory and 14071da177e4SLinus Torvalds * parent relationships. 14081da177e4SLinus Torvalds */ 1409fe479a58SAl Viro if (unlikely(type != LAST_NORM)) { 1410ef7562d5SAl Viro if (handle_dots(nd, type)) 1411ef7562d5SAl Viro return -ECHILD; 14121da177e4SLinus Torvalds continue; 14131da177e4SLinus Torvalds } 1414fe479a58SAl Viro 14151da177e4SLinus Torvalds /* This does the actual lookups.. */ 141631e6b01fSNick Piggin err = do_lookup(nd, &this, &next, &inode); 14171da177e4SLinus Torvalds if (err) 14181da177e4SLinus Torvalds break; 14191da177e4SLinus Torvalds 14207bc055d1SAl Viro if (inode && inode->i_op->follow_link) { 14213abb17e8SLinus Torvalds err = do_follow_link(inode, &next, nd); 14221da177e4SLinus Torvalds if (err) 1423a7472babSAl Viro return err; 142431e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 142531e6b01fSNick Piggin } else { 142609dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 142731e6b01fSNick Piggin nd->inode = inode; 142831e6b01fSNick Piggin } 14297bc055d1SAl Viro err = -ENOENT; 14307bc055d1SAl Viro if (!nd->inode) 14317bc055d1SAl Viro break; 14321da177e4SLinus Torvalds err = -ENOTDIR; 143331e6b01fSNick Piggin if (!nd->inode->i_op->lookup) 14341da177e4SLinus Torvalds break; 14351da177e4SLinus Torvalds continue; 14361da177e4SLinus Torvalds /* here ends the main loop */ 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds last_with_slashes: 14391da177e4SLinus Torvalds lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 14401da177e4SLinus Torvalds last_component: 1441f55eab82STrond Myklebust /* Clear LOOKUP_CONTINUE iff it was previously unset */ 1442f55eab82STrond Myklebust nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; 14431da177e4SLinus Torvalds if (lookup_flags & LOOKUP_PARENT) 14441da177e4SLinus Torvalds goto lookup_parent; 1445ef7562d5SAl Viro if (unlikely(type != LAST_NORM)) 1446ef7562d5SAl Viro return handle_dots(nd, type); 144731e6b01fSNick Piggin err = do_lookup(nd, &this, &next, &inode); 14481da177e4SLinus Torvalds if (err) 14491da177e4SLinus Torvalds break; 1450db372915SDavid Howells if (inode && unlikely(inode->i_op->follow_link) && 1451db372915SDavid Howells (lookup_flags & LOOKUP_FOLLOW)) { 14523abb17e8SLinus Torvalds err = do_follow_link(inode, &next, nd); 14531da177e4SLinus Torvalds if (err) 1454a7472babSAl Viro return err; 145531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 145631e6b01fSNick Piggin } else { 145709dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 145831e6b01fSNick Piggin nd->inode = inode; 145931e6b01fSNick Piggin } 14601da177e4SLinus Torvalds err = -ENOENT; 146131e6b01fSNick Piggin if (!nd->inode) 14621da177e4SLinus Torvalds break; 14631da177e4SLinus Torvalds if (lookup_flags & LOOKUP_DIRECTORY) { 14641da177e4SLinus Torvalds err = -ENOTDIR; 146531e6b01fSNick Piggin if (!nd->inode->i_op->lookup) 14661da177e4SLinus Torvalds break; 14671da177e4SLinus Torvalds } 1468086e183aSAl Viro return 0; 14691da177e4SLinus Torvalds lookup_parent: 14701da177e4SLinus Torvalds nd->last = this; 1471fe479a58SAl Viro nd->last_type = type; 14721da177e4SLinus Torvalds return 0; 14731da177e4SLinus Torvalds } 1474951361f9SAl Viro terminate_walk(nd); 14751da177e4SLinus Torvalds return err; 14761da177e4SLinus Torvalds } 14771da177e4SLinus Torvalds 147870e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 147970e9b357SAl Viro struct nameidata *nd, struct file **fp) 148031e6b01fSNick Piggin { 148131e6b01fSNick Piggin int retval = 0; 148231e6b01fSNick Piggin int fput_needed; 148331e6b01fSNick Piggin struct file *file; 148431e6b01fSNick Piggin 148531e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 148616c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 148731e6b01fSNick Piggin nd->depth = 0; 14885b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 14895b6ca027SAl Viro struct inode *inode = nd->root.dentry->d_inode; 14905b6ca027SAl Viro if (!inode->i_op->lookup) 14915b6ca027SAl Viro return -ENOTDIR; 14925b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 14935b6ca027SAl Viro if (retval) 14945b6ca027SAl Viro return retval; 14955b6ca027SAl Viro nd->path = nd->root; 14965b6ca027SAl Viro nd->inode = inode; 14975b6ca027SAl Viro if (flags & LOOKUP_RCU) { 14985b6ca027SAl Viro br_read_lock(vfsmount_lock); 14995b6ca027SAl Viro rcu_read_lock(); 15005b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 15015b6ca027SAl Viro } else { 15025b6ca027SAl Viro path_get(&nd->path); 15035b6ca027SAl Viro } 15045b6ca027SAl Viro return 0; 15055b6ca027SAl Viro } 15065b6ca027SAl Viro 150731e6b01fSNick Piggin nd->root.mnt = NULL; 150831e6b01fSNick Piggin 150931e6b01fSNick Piggin if (*name=='/') { 1510e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 151131e6b01fSNick Piggin br_read_lock(vfsmount_lock); 151231e6b01fSNick Piggin rcu_read_lock(); 1513e41f7d4eSAl Viro set_root_rcu(nd); 1514e41f7d4eSAl Viro } else { 1515e41f7d4eSAl Viro set_root(nd); 1516e41f7d4eSAl Viro path_get(&nd->root); 1517e41f7d4eSAl Viro } 151831e6b01fSNick Piggin nd->path = nd->root; 151931e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1520e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 152131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1522c28cc364SNick Piggin unsigned seq; 152331e6b01fSNick Piggin 152431e6b01fSNick Piggin br_read_lock(vfsmount_lock); 152531e6b01fSNick Piggin rcu_read_lock(); 152631e6b01fSNick Piggin 1527c28cc364SNick Piggin do { 1528c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 152931e6b01fSNick Piggin nd->path = fs->pwd; 1530c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1531c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1532e41f7d4eSAl Viro } else { 1533e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1534e41f7d4eSAl Viro } 153531e6b01fSNick Piggin } else { 153631e6b01fSNick Piggin struct dentry *dentry; 153731e6b01fSNick Piggin 153831e6b01fSNick Piggin file = fget_light(dfd, &fput_needed); 153931e6b01fSNick Piggin retval = -EBADF; 154031e6b01fSNick Piggin if (!file) 154131e6b01fSNick Piggin goto out_fail; 154231e6b01fSNick Piggin 154331e6b01fSNick Piggin dentry = file->f_path.dentry; 154431e6b01fSNick Piggin 154531e6b01fSNick Piggin retval = -ENOTDIR; 154631e6b01fSNick Piggin if (!S_ISDIR(dentry->d_inode->i_mode)) 154731e6b01fSNick Piggin goto fput_fail; 154831e6b01fSNick Piggin 154931e6b01fSNick Piggin retval = file_permission(file, MAY_EXEC); 155031e6b01fSNick Piggin if (retval) 155131e6b01fSNick Piggin goto fput_fail; 155231e6b01fSNick Piggin 155331e6b01fSNick Piggin nd->path = file->f_path; 1554e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 155531e6b01fSNick Piggin if (fput_needed) 155670e9b357SAl Viro *fp = file; 1557c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 155831e6b01fSNick Piggin br_read_lock(vfsmount_lock); 155931e6b01fSNick Piggin rcu_read_lock(); 15605590ff0dSUlrich Drepper } else { 15615dd784d0SJan Blunck path_get(&file->f_path); 15625590ff0dSUlrich Drepper fput_light(file, fput_needed); 15631da177e4SLinus Torvalds } 1564e41f7d4eSAl Viro } 1565e41f7d4eSAl Viro 156631e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 15679b4a9b14SAl Viro return 0; 15682dfdd266SJosef 'Jeff' Sipek 15699b4a9b14SAl Viro fput_fail: 15709b4a9b14SAl Viro fput_light(file, fput_needed); 15719b4a9b14SAl Viro out_fail: 15729b4a9b14SAl Viro return retval; 15739b4a9b14SAl Viro } 15749b4a9b14SAl Viro 15759b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1576ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 15779b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 15789b4a9b14SAl Viro { 157970e9b357SAl Viro struct file *base = NULL; 158031e6b01fSNick Piggin int retval; 158131e6b01fSNick Piggin 158231e6b01fSNick Piggin /* 158331e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 158431e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 158531e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 158631e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 158731e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 158831e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 158931e6b01fSNick Piggin * analogue, foo_rcu(). 159031e6b01fSNick Piggin * 159131e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 159231e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 159331e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 159431e6b01fSNick Piggin * be able to complete). 159531e6b01fSNick Piggin */ 159670e9b357SAl Viro retval = path_init(dfd, name, flags, nd, &base); 1597ee0827cdSAl Viro 159831e6b01fSNick Piggin if (unlikely(retval)) 159931e6b01fSNick Piggin return retval; 1600ee0827cdSAl Viro 1601ee0827cdSAl Viro current->total_link_count = 0; 1602ee0827cdSAl Viro retval = link_path_walk(name, nd); 1603ee0827cdSAl Viro 1604ee0827cdSAl Viro if (nd->flags & LOOKUP_RCU) { 16054455ca62SAl Viro /* went all way through without dropping RCU */ 16064455ca62SAl Viro BUG_ON(retval); 1607086e183aSAl Viro if (nameidata_drop_rcu_last(nd)) 1608086e183aSAl Viro retval = -ECHILD; 1609086e183aSAl Viro } 161031e6b01fSNick Piggin 161116c2cd71SAl Viro if (!retval) 161216c2cd71SAl Viro retval = handle_reval_path(nd); 161316c2cd71SAl Viro 161470e9b357SAl Viro if (base) 161570e9b357SAl Viro fput(base); 1616ee0827cdSAl Viro 16175b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 161831e6b01fSNick Piggin path_put(&nd->root); 161931e6b01fSNick Piggin nd->root.mnt = NULL; 162031e6b01fSNick Piggin } 1621ee0827cdSAl Viro return retval; 162231e6b01fSNick Piggin } 162331e6b01fSNick Piggin 1624ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1625ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1626ee0827cdSAl Viro { 1627ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1628ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1629ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1630ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1631ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1632ee0827cdSAl Viro 163331e6b01fSNick Piggin if (likely(!retval)) { 163431e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 163531e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 163631e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 163731e6b01fSNick Piggin } 163831e6b01fSNick Piggin } 1639170aa3d0SUlrich Drepper return retval; 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 1642c9c6cac0SAl Viro int kern_path_parent(const char *name, struct nameidata *nd) 16435590ff0dSUlrich Drepper { 1644c9c6cac0SAl Viro return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); 16455590ff0dSUlrich Drepper } 16465590ff0dSUlrich Drepper 1647d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1648d1811465SAl Viro { 1649d1811465SAl Viro struct nameidata nd; 1650d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1651d1811465SAl Viro if (!res) 1652d1811465SAl Viro *path = nd.path; 1653d1811465SAl Viro return res; 1654d1811465SAl Viro } 1655d1811465SAl Viro 165616f18200SJosef 'Jeff' Sipek /** 165716f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 165816f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 165916f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 166016f18200SJosef 'Jeff' Sipek * @name: pointer to file name 166116f18200SJosef 'Jeff' Sipek * @flags: lookup flags 166216f18200SJosef 'Jeff' Sipek * @nd: pointer to nameidata 166316f18200SJosef 'Jeff' Sipek */ 166416f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 166516f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 166616f18200SJosef 'Jeff' Sipek struct nameidata *nd) 166716f18200SJosef 'Jeff' Sipek { 16685b6ca027SAl Viro nd->root.dentry = dentry; 16695b6ca027SAl Viro nd->root.mnt = mnt; 16705b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 16715b6ca027SAl Viro return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd); 167216f18200SJosef 'Jeff' Sipek } 167316f18200SJosef 'Jeff' Sipek 1674eead1911SChristoph Hellwig static struct dentry *__lookup_hash(struct qstr *name, 1675eead1911SChristoph Hellwig struct dentry *base, struct nameidata *nd) 16761da177e4SLinus Torvalds { 167781fca444SChristoph Hellwig struct inode *inode = base->d_inode; 16781da177e4SLinus Torvalds struct dentry *dentry; 16791da177e4SLinus Torvalds int err; 16801da177e4SLinus Torvalds 1681b74c79e9SNick Piggin err = exec_permission(inode, 0); 168281fca444SChristoph Hellwig if (err) 168381fca444SChristoph Hellwig return ERR_PTR(err); 16841da177e4SLinus Torvalds 16851da177e4SLinus Torvalds /* 1686b04f784eSNick Piggin * Don't bother with __d_lookup: callers are for creat as 1687b04f784eSNick Piggin * well as unlink, so a lot of the time it would cost 1688b04f784eSNick Piggin * a double lookup. 16896e6b1bd1SAl Viro */ 16906e6b1bd1SAl Viro dentry = d_lookup(base, name); 16916e6b1bd1SAl Viro 1692fb045adbSNick Piggin if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) 16936e6b1bd1SAl Viro dentry = do_revalidate(dentry, nd); 16946e6b1bd1SAl Viro 16951da177e4SLinus Torvalds if (!dentry) 1696baa03890SNick Piggin dentry = d_alloc_and_lookup(base, name, nd); 16975a202bcdSAl Viro 16981da177e4SLinus Torvalds return dentry; 16991da177e4SLinus Torvalds } 17001da177e4SLinus Torvalds 1701057f6c01SJames Morris /* 1702057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1703057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1704057f6c01SJames Morris * SMP-safe. 1705057f6c01SJames Morris */ 1706a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 17071da177e4SLinus Torvalds { 17084ac91378SJan Blunck return __lookup_hash(&nd->last, nd->path.dentry, nd); 17091da177e4SLinus Torvalds } 17101da177e4SLinus Torvalds 1711eead1911SChristoph Hellwig /** 1712a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1713eead1911SChristoph Hellwig * @name: pathname component to lookup 1714eead1911SChristoph Hellwig * @base: base directory to lookup from 1715eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1716eead1911SChristoph Hellwig * 1717a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1718a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1719eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1720eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1721eead1911SChristoph Hellwig */ 1722057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1723057f6c01SJames Morris { 1724057f6c01SJames Morris struct qstr this; 17256a96ba54SAl Viro unsigned long hash; 17266a96ba54SAl Viro unsigned int c; 1727057f6c01SJames Morris 17282f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 17292f9092e1SDavid Woodhouse 17306a96ba54SAl Viro this.name = name; 17316a96ba54SAl Viro this.len = len; 17326a96ba54SAl Viro if (!len) 17336a96ba54SAl Viro return ERR_PTR(-EACCES); 17346a96ba54SAl Viro 17356a96ba54SAl Viro hash = init_name_hash(); 17366a96ba54SAl Viro while (len--) { 17376a96ba54SAl Viro c = *(const unsigned char *)name++; 17386a96ba54SAl Viro if (c == '/' || c == '\0') 17396a96ba54SAl Viro return ERR_PTR(-EACCES); 17406a96ba54SAl Viro hash = partial_name_hash(c, hash); 17416a96ba54SAl Viro } 17426a96ba54SAl Viro this.hash = end_name_hash(hash); 17435a202bcdSAl Viro /* 17445a202bcdSAl Viro * See if the low-level filesystem might want 17455a202bcdSAl Viro * to use its own hash.. 17465a202bcdSAl Viro */ 17475a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 17485a202bcdSAl Viro int err = base->d_op->d_hash(base, base->d_inode, &this); 17495a202bcdSAl Viro if (err < 0) 17505a202bcdSAl Viro return ERR_PTR(err); 17515a202bcdSAl Viro } 1752eead1911SChristoph Hellwig 175349705b77SChristoph Hellwig return __lookup_hash(&this, base, NULL); 1754057f6c01SJames Morris } 1755057f6c01SJames Morris 17562d8f3038SAl Viro int user_path_at(int dfd, const char __user *name, unsigned flags, 17572d8f3038SAl Viro struct path *path) 17581da177e4SLinus Torvalds { 17592d8f3038SAl Viro struct nameidata nd; 17601da177e4SLinus Torvalds char *tmp = getname(name); 17611da177e4SLinus Torvalds int err = PTR_ERR(tmp); 17621da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 17632d8f3038SAl Viro 17642d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 17652d8f3038SAl Viro 17662d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 17671da177e4SLinus Torvalds putname(tmp); 17682d8f3038SAl Viro if (!err) 17692d8f3038SAl Viro *path = nd.path; 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds return err; 17721da177e4SLinus Torvalds } 17731da177e4SLinus Torvalds 17742ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 17752ad94ae6SAl Viro struct nameidata *nd, char **name) 17762ad94ae6SAl Viro { 17772ad94ae6SAl Viro char *s = getname(path); 17782ad94ae6SAl Viro int error; 17792ad94ae6SAl Viro 17802ad94ae6SAl Viro if (IS_ERR(s)) 17812ad94ae6SAl Viro return PTR_ERR(s); 17822ad94ae6SAl Viro 17832ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 17842ad94ae6SAl Viro if (error) 17852ad94ae6SAl Viro putname(s); 17862ad94ae6SAl Viro else 17872ad94ae6SAl Viro *name = s; 17882ad94ae6SAl Viro 17892ad94ae6SAl Viro return error; 17902ad94ae6SAl Viro } 17912ad94ae6SAl Viro 17921da177e4SLinus Torvalds /* 17931da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 17941da177e4SLinus Torvalds * minimal. 17951da177e4SLinus Torvalds */ 17961da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 17971da177e4SLinus Torvalds { 1798da9592edSDavid Howells uid_t fsuid = current_fsuid(); 1799da9592edSDavid Howells 18001da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 18011da177e4SLinus Torvalds return 0; 1802da9592edSDavid Howells if (inode->i_uid == fsuid) 18031da177e4SLinus Torvalds return 0; 1804da9592edSDavid Howells if (dir->i_uid == fsuid) 18051da177e4SLinus Torvalds return 0; 18061da177e4SLinus Torvalds return !capable(CAP_FOWNER); 18071da177e4SLinus Torvalds } 18081da177e4SLinus Torvalds 18091da177e4SLinus Torvalds /* 18101da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 18111da177e4SLinus Torvalds * whether the type of victim is right. 18121da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 18131da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 18141da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 18151da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 18161da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 18171da177e4SLinus Torvalds * a. be owner of dir, or 18181da177e4SLinus Torvalds * b. be owner of victim, or 18191da177e4SLinus Torvalds * c. have CAP_FOWNER capability 18201da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 18211da177e4SLinus Torvalds * links pointing to it. 18221da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 18231da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 18241da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 18251da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 18261da177e4SLinus Torvalds * nfs_async_unlink(). 18271da177e4SLinus Torvalds */ 1828858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 18291da177e4SLinus Torvalds { 18301da177e4SLinus Torvalds int error; 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds if (!victim->d_inode) 18331da177e4SLinus Torvalds return -ENOENT; 18341da177e4SLinus Torvalds 18351da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 1836cccc6bbaSAl Viro audit_inode_child(victim, dir); 18371da177e4SLinus Torvalds 1838f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 18391da177e4SLinus Torvalds if (error) 18401da177e4SLinus Torvalds return error; 18411da177e4SLinus Torvalds if (IS_APPEND(dir)) 18421da177e4SLinus Torvalds return -EPERM; 18431da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 1844f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 18451da177e4SLinus Torvalds return -EPERM; 18461da177e4SLinus Torvalds if (isdir) { 18471da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 18481da177e4SLinus Torvalds return -ENOTDIR; 18491da177e4SLinus Torvalds if (IS_ROOT(victim)) 18501da177e4SLinus Torvalds return -EBUSY; 18511da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 18521da177e4SLinus Torvalds return -EISDIR; 18531da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 18541da177e4SLinus Torvalds return -ENOENT; 18551da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 18561da177e4SLinus Torvalds return -EBUSY; 18571da177e4SLinus Torvalds return 0; 18581da177e4SLinus Torvalds } 18591da177e4SLinus Torvalds 18601da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 18611da177e4SLinus Torvalds * dir. 18621da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 18631da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 18641da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 18651da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 18661da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 18671da177e4SLinus Torvalds */ 1868a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 18691da177e4SLinus Torvalds { 18701da177e4SLinus Torvalds if (child->d_inode) 18711da177e4SLinus Torvalds return -EEXIST; 18721da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 18731da177e4SLinus Torvalds return -ENOENT; 1874f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 18751da177e4SLinus Torvalds } 18761da177e4SLinus Torvalds 18771da177e4SLinus Torvalds /* 18781da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 18791da177e4SLinus Torvalds */ 18801da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 18811da177e4SLinus Torvalds { 18821da177e4SLinus Torvalds struct dentry *p; 18831da177e4SLinus Torvalds 18841da177e4SLinus Torvalds if (p1 == p2) { 1885f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 18861da177e4SLinus Torvalds return NULL; 18871da177e4SLinus Torvalds } 18881da177e4SLinus Torvalds 1889a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 18901da177e4SLinus Torvalds 1891e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 1892e2761a11SOGAWA Hirofumi if (p) { 1893f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1894f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 18951da177e4SLinus Torvalds return p; 18961da177e4SLinus Torvalds } 18971da177e4SLinus Torvalds 1898e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 1899e2761a11SOGAWA Hirofumi if (p) { 1900f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1901f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19021da177e4SLinus Torvalds return p; 19031da177e4SLinus Torvalds } 19041da177e4SLinus Torvalds 1905f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1906f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19071da177e4SLinus Torvalds return NULL; 19081da177e4SLinus Torvalds } 19091da177e4SLinus Torvalds 19101da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 19111da177e4SLinus Torvalds { 19121b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 19131da177e4SLinus Torvalds if (p1 != p2) { 19141b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 1915a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19161da177e4SLinus Torvalds } 19171da177e4SLinus Torvalds } 19181da177e4SLinus Torvalds 19191da177e4SLinus Torvalds int vfs_create(struct inode *dir, struct dentry *dentry, int mode, 19201da177e4SLinus Torvalds struct nameidata *nd) 19211da177e4SLinus Torvalds { 1922a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds if (error) 19251da177e4SLinus Torvalds return error; 19261da177e4SLinus Torvalds 1927acfa4380SAl Viro if (!dir->i_op->create) 19281da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 19291da177e4SLinus Torvalds mode &= S_IALLUGO; 19301da177e4SLinus Torvalds mode |= S_IFREG; 19311da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 19321da177e4SLinus Torvalds if (error) 19331da177e4SLinus Torvalds return error; 19341da177e4SLinus Torvalds error = dir->i_op->create(dir, dentry, mode, nd); 1935a74574aaSStephen Smalley if (!error) 1936f38aa942SAmy Griffis fsnotify_create(dir, dentry); 19371da177e4SLinus Torvalds return error; 19381da177e4SLinus Torvalds } 19391da177e4SLinus Torvalds 19403fb64190SChristoph Hellwig int may_open(struct path *path, int acc_mode, int flag) 19411da177e4SLinus Torvalds { 19423fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 19431da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 19441da177e4SLinus Torvalds int error; 19451da177e4SLinus Torvalds 19461da177e4SLinus Torvalds if (!inode) 19471da177e4SLinus Torvalds return -ENOENT; 19481da177e4SLinus Torvalds 1949c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 1950c8fe8f30SChristoph Hellwig case S_IFLNK: 19511da177e4SLinus Torvalds return -ELOOP; 1952c8fe8f30SChristoph Hellwig case S_IFDIR: 1953c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 19541da177e4SLinus Torvalds return -EISDIR; 1955c8fe8f30SChristoph Hellwig break; 1956c8fe8f30SChristoph Hellwig case S_IFBLK: 1957c8fe8f30SChristoph Hellwig case S_IFCHR: 19583fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 19591da177e4SLinus Torvalds return -EACCES; 1960c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 1961c8fe8f30SChristoph Hellwig case S_IFIFO: 1962c8fe8f30SChristoph Hellwig case S_IFSOCK: 19631da177e4SLinus Torvalds flag &= ~O_TRUNC; 1964c8fe8f30SChristoph Hellwig break; 19654a3fd211SDave Hansen } 1966b41572e9SDave Hansen 19673fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 1968b41572e9SDave Hansen if (error) 1969b41572e9SDave Hansen return error; 19706146f0d5SMimi Zohar 19711da177e4SLinus Torvalds /* 19721da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 19731da177e4SLinus Torvalds */ 19741da177e4SLinus Torvalds if (IS_APPEND(inode)) { 19758737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 19767715b521SAl Viro return -EPERM; 19771da177e4SLinus Torvalds if (flag & O_TRUNC) 19787715b521SAl Viro return -EPERM; 19791da177e4SLinus Torvalds } 19801da177e4SLinus Torvalds 19811da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 19827715b521SAl Viro if (flag & O_NOATIME && !is_owner_or_cap(inode)) 19837715b521SAl Viro return -EPERM; 19841da177e4SLinus Torvalds 19851da177e4SLinus Torvalds /* 19861da177e4SLinus Torvalds * Ensure there are no outstanding leases on the file. 19871da177e4SLinus Torvalds */ 1988b65a9cfcSAl Viro return break_lease(inode, flag); 19897715b521SAl Viro } 19907715b521SAl Viro 1991e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 19927715b521SAl Viro { 1993e1181ee6SJeff Layton struct path *path = &filp->f_path; 19947715b521SAl Viro struct inode *inode = path->dentry->d_inode; 19957715b521SAl Viro int error = get_write_access(inode); 19961da177e4SLinus Torvalds if (error) 19977715b521SAl Viro return error; 19981da177e4SLinus Torvalds /* 19991da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 20001da177e4SLinus Torvalds */ 20011da177e4SLinus Torvalds error = locks_verify_locked(inode); 2002be6d3e56SKentaro Takeda if (!error) 2003ea0d3ab2STetsuo Handa error = security_path_truncate(path); 20041da177e4SLinus Torvalds if (!error) { 20057715b521SAl Viro error = do_truncate(path->dentry, 0, 2006d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2007e1181ee6SJeff Layton filp); 20081da177e4SLinus Torvalds } 20091da177e4SLinus Torvalds put_write_access(inode); 2010acd0c935SMimi Zohar return error; 20111da177e4SLinus Torvalds } 20121da177e4SLinus Torvalds 2013d57999e1SDave Hansen /* 2014d57999e1SDave Hansen * Note that while the flag value (low two bits) for sys_open means: 2015d57999e1SDave Hansen * 00 - read-only 2016d57999e1SDave Hansen * 01 - write-only 2017d57999e1SDave Hansen * 10 - read-write 2018d57999e1SDave Hansen * 11 - special 2019d57999e1SDave Hansen * it is changed into 2020d57999e1SDave Hansen * 00 - no permissions needed 2021d57999e1SDave Hansen * 01 - read-permission 2022d57999e1SDave Hansen * 10 - write-permission 2023d57999e1SDave Hansen * 11 - read-write 2024d57999e1SDave Hansen * for the internal routines (ie open_namei()/follow_link() etc) 2025d57999e1SDave Hansen * This is more logical, and also allows the 00 "no perm needed" 2026d57999e1SDave Hansen * to be used for symlinks (where the permissions are checked 2027d57999e1SDave Hansen * later). 2028d57999e1SDave Hansen * 2029d57999e1SDave Hansen */ 2030d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2031d57999e1SDave Hansen { 2032d57999e1SDave Hansen if ((flag+1) & O_ACCMODE) 2033d57999e1SDave Hansen flag++; 2034d57999e1SDave Hansen return flag; 2035d57999e1SDave Hansen } 2036d57999e1SDave Hansen 203731e6b01fSNick Piggin /* 2038fe2d35ffSAl Viro * Handle the last step of open() 203931e6b01fSNick Piggin */ 2040fb1cc555SAl Viro static struct file *do_last(struct nameidata *nd, struct path *path, 2041c3e380b0SAl Viro const struct open_flags *op, const char *pathname) 2042fb1cc555SAl Viro { 2043a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 20446c0d46c4SAl Viro struct dentry *dentry; 2045ca344a89SAl Viro int open_flag = op->open_flag; 20466c0d46c4SAl Viro int will_truncate = open_flag & O_TRUNC; 2047ca344a89SAl Viro int want_write = 0; 2048ca344a89SAl Viro int skip_perm = 0; 2049fb1cc555SAl Viro struct file *filp; 2050fe2d35ffSAl Viro struct inode *inode; 205116c2cd71SAl Viro int error; 2052fb1cc555SAl Viro 2053c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2054c3e380b0SAl Viro nd->flags |= op->intent; 2055c3e380b0SAl Viro 20561f36f774SAl Viro switch (nd->last_type) { 20571f36f774SAl Viro case LAST_DOTDOT: 2058176306f5SNeil Brown case LAST_DOT: 2059fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2060fe2d35ffSAl Viro if (error) 2061fe2d35ffSAl Viro return ERR_PTR(error); 20621f36f774SAl Viro /* fallthrough */ 20631f36f774SAl Viro case LAST_ROOT: 2064fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2065fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2066fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2067fe2d35ffSAl Viro } 206816c2cd71SAl Viro error = handle_reval_path(nd); 206916c2cd71SAl Viro if (error) 207016c2cd71SAl Viro goto exit; 2071fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2072ca344a89SAl Viro if (open_flag & O_CREAT) { 207316c2cd71SAl Viro error = -EISDIR; 20741f36f774SAl Viro goto exit; 2075fe2d35ffSAl Viro } 2076fe2d35ffSAl Viro goto ok; 20771f36f774SAl Viro case LAST_BIND: 2078fe2d35ffSAl Viro /* can't be RCU mode here */ 207916c2cd71SAl Viro error = handle_reval_path(nd); 208016c2cd71SAl Viro if (error) 208116c2cd71SAl Viro goto exit; 20821f36f774SAl Viro audit_inode(pathname, dir); 20831f36f774SAl Viro goto ok; 20841f36f774SAl Viro } 2085a2c36b45SAl Viro 2086ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2087fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2088fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2089fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2090fe2d35ffSAl Viro error = do_lookup(nd, &nd->last, path, &inode); 2091fe2d35ffSAl Viro if (error) { 2092fe2d35ffSAl Viro terminate_walk(nd); 2093fe2d35ffSAl Viro return ERR_PTR(error); 2094fe2d35ffSAl Viro } 2095fe2d35ffSAl Viro if (!inode) { 2096fe2d35ffSAl Viro path_to_nameidata(path, nd); 2097fe2d35ffSAl Viro terminate_walk(nd); 2098fe2d35ffSAl Viro return ERR_PTR(-ENOENT); 2099fe2d35ffSAl Viro } 2100fe2d35ffSAl Viro if (unlikely(inode->i_op->follow_link)) { 2101fe2d35ffSAl Viro /* We drop rcu-walk here */ 2102fe2d35ffSAl Viro if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) 2103fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2104fe2d35ffSAl Viro return NULL; 2105fe2d35ffSAl Viro } 2106fe2d35ffSAl Viro path_to_nameidata(path, nd); 2107fe2d35ffSAl Viro nd->inode = inode; 2108fe2d35ffSAl Viro /* sayonara */ 2109fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2110fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2111fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2112fe2d35ffSAl Viro } 2113fe2d35ffSAl Viro 2114fe2d35ffSAl Viro error = -ENOTDIR; 2115fe2d35ffSAl Viro if (nd->flags & LOOKUP_DIRECTORY) { 2116fe2d35ffSAl Viro if (!inode->i_op->lookup) 2117fe2d35ffSAl Viro goto exit; 2118fe2d35ffSAl Viro } 2119fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2120fe2d35ffSAl Viro goto ok; 2121fe2d35ffSAl Viro } 2122fe2d35ffSAl Viro 2123fe2d35ffSAl Viro /* create side of things */ 2124fe2d35ffSAl Viro 2125fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2126fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2127fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2128fe2d35ffSAl Viro } 2129fe2d35ffSAl Viro 2130fe2d35ffSAl Viro audit_inode(pathname, dir); 213116c2cd71SAl Viro error = -EISDIR; 21321f36f774SAl Viro /* trailing slashes? */ 213331e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 21341f36f774SAl Viro goto exit; 21351f36f774SAl Viro 2136a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 2137a1e28038SAl Viro 21386c0d46c4SAl Viro dentry = lookup_hash(nd); 21396c0d46c4SAl Viro error = PTR_ERR(dentry); 21406c0d46c4SAl Viro if (IS_ERR(dentry)) { 2141fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2142fb1cc555SAl Viro goto exit; 2143fb1cc555SAl Viro } 2144fb1cc555SAl Viro 21456c0d46c4SAl Viro path->dentry = dentry; 21466c0d46c4SAl Viro path->mnt = nd->path.mnt; 21476c0d46c4SAl Viro 2148fb1cc555SAl Viro /* Negative dentry, just create the file */ 21496c0d46c4SAl Viro if (!dentry->d_inode) { 21506c0d46c4SAl Viro int mode = op->mode; 21516c0d46c4SAl Viro if (!IS_POSIXACL(dir->d_inode)) 21526c0d46c4SAl Viro mode &= ~current_umask(); 2153fb1cc555SAl Viro /* 2154fb1cc555SAl Viro * This write is needed to ensure that a 21556c0d46c4SAl Viro * rw->ro transition does not occur between 2156fb1cc555SAl Viro * the time when the file is created and when 2157fb1cc555SAl Viro * a permanent write count is taken through 2158fb1cc555SAl Viro * the 'struct file' in nameidata_to_filp(). 2159fb1cc555SAl Viro */ 2160fb1cc555SAl Viro error = mnt_want_write(nd->path.mnt); 2161fb1cc555SAl Viro if (error) 2162fb1cc555SAl Viro goto exit_mutex_unlock; 2163ca344a89SAl Viro want_write = 1; 21649b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2165ca344a89SAl Viro open_flag &= ~O_TRUNC; 21666c0d46c4SAl Viro will_truncate = 0; 2167ca344a89SAl Viro skip_perm = 1; 21686c0d46c4SAl Viro error = security_path_mknod(&nd->path, dentry, mode, 0); 21696c0d46c4SAl Viro if (error) 21706c0d46c4SAl Viro goto exit_mutex_unlock; 21716c0d46c4SAl Viro error = vfs_create(dir->d_inode, dentry, mode, nd); 21726c0d46c4SAl Viro if (error) 21736c0d46c4SAl Viro goto exit_mutex_unlock; 21746c0d46c4SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 21756c0d46c4SAl Viro dput(nd->path.dentry); 21766c0d46c4SAl Viro nd->path.dentry = dentry; 2177ca344a89SAl Viro goto common; 2178fb1cc555SAl Viro } 2179fb1cc555SAl Viro 2180fb1cc555SAl Viro /* 2181fb1cc555SAl Viro * It already exists. 2182fb1cc555SAl Viro */ 2183fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2184fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2185fb1cc555SAl Viro 2186fb1cc555SAl Viro error = -EEXIST; 2187ca344a89SAl Viro if (open_flag & O_EXCL) 2188fb1cc555SAl Viro goto exit_dput; 2189fb1cc555SAl Viro 21909875cf80SDavid Howells error = follow_managed(path, nd->flags); 21919875cf80SDavid Howells if (error < 0) 2192fb1cc555SAl Viro goto exit_dput; 2193fb1cc555SAl Viro 2194fb1cc555SAl Viro error = -ENOENT; 2195fb1cc555SAl Viro if (!path->dentry->d_inode) 2196fb1cc555SAl Viro goto exit_dput; 21979e67f361SAl Viro 21989e67f361SAl Viro if (path->dentry->d_inode->i_op->follow_link) 2199fb1cc555SAl Viro return NULL; 2200fb1cc555SAl Viro 2201fb1cc555SAl Viro path_to_nameidata(path, nd); 220231e6b01fSNick Piggin nd->inode = path->dentry->d_inode; 2203fb1cc555SAl Viro error = -EISDIR; 220431e6b01fSNick Piggin if (S_ISDIR(nd->inode->i_mode)) 2205fb1cc555SAl Viro goto exit; 220667ee3ad2SAl Viro ok: 22076c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 22086c0d46c4SAl Viro will_truncate = 0; 22096c0d46c4SAl Viro 22100f9d1a10SAl Viro if (will_truncate) { 22110f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 22120f9d1a10SAl Viro if (error) 22130f9d1a10SAl Viro goto exit; 2214ca344a89SAl Viro want_write = 1; 22150f9d1a10SAl Viro } 2216ca344a89SAl Viro common: 2217ca344a89SAl Viro error = may_open(&nd->path, skip_perm ? 0 : op->acc_mode, open_flag); 2218ca344a89SAl Viro if (error) 22190f9d1a10SAl Viro goto exit; 22200f9d1a10SAl Viro filp = nameidata_to_filp(nd); 22210f9d1a10SAl Viro if (!IS_ERR(filp)) { 22220f9d1a10SAl Viro error = ima_file_check(filp, op->acc_mode); 22230f9d1a10SAl Viro if (error) { 22240f9d1a10SAl Viro fput(filp); 22250f9d1a10SAl Viro filp = ERR_PTR(error); 22260f9d1a10SAl Viro } 22270f9d1a10SAl Viro } 22280f9d1a10SAl Viro if (!IS_ERR(filp)) { 22290f9d1a10SAl Viro if (will_truncate) { 22300f9d1a10SAl Viro error = handle_truncate(filp); 22310f9d1a10SAl Viro if (error) { 22320f9d1a10SAl Viro fput(filp); 22330f9d1a10SAl Viro filp = ERR_PTR(error); 22340f9d1a10SAl Viro } 22350f9d1a10SAl Viro } 22360f9d1a10SAl Viro } 2237ca344a89SAl Viro out: 2238ca344a89SAl Viro if (want_write) 22390f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 22400f9d1a10SAl Viro path_put(&nd->path); 2241fb1cc555SAl Viro return filp; 2242fb1cc555SAl Viro 2243fb1cc555SAl Viro exit_mutex_unlock: 2244fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2245fb1cc555SAl Viro exit_dput: 2246fb1cc555SAl Viro path_put_conditional(path, nd); 2247fb1cc555SAl Viro exit: 2248ca344a89SAl Viro filp = ERR_PTR(error); 2249ca344a89SAl Viro goto out; 2250fb1cc555SAl Viro } 2251fb1cc555SAl Viro 225213aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 225347c805dcSAl Viro const struct open_flags *op, int flags) 22541da177e4SLinus Torvalds { 2255fe2d35ffSAl Viro struct file *base = NULL; 22564a3fd211SDave Hansen struct file *filp; 2257a70e65dfSChristoph Hellwig struct nameidata nd; 22589850c056SAl Viro struct path path; 22591da177e4SLinus Torvalds int count = 0; 226013aab428SAl Viro int error; 226131e6b01fSNick Piggin 226231e6b01fSNick Piggin filp = get_empty_filp(); 226331e6b01fSNick Piggin if (!filp) 226431e6b01fSNick Piggin return ERR_PTR(-ENFILE); 226531e6b01fSNick Piggin 226647c805dcSAl Viro filp->f_flags = op->open_flag; 226731e6b01fSNick Piggin nd.intent.open.file = filp; 226847c805dcSAl Viro nd.intent.open.flags = open_to_namei_flags(op->open_flag); 226947c805dcSAl Viro nd.intent.open.create_mode = op->mode; 227031e6b01fSNick Piggin 2271fe2d35ffSAl Viro error = path_init(dfd, pathname, flags | LOOKUP_PARENT, &nd, &base); 227231e6b01fSNick Piggin if (unlikely(error)) 227313aab428SAl Viro goto out_filp; 227431e6b01fSNick Piggin 2275fe2d35ffSAl Viro current->total_link_count = 0; 2276fe2d35ffSAl Viro error = link_path_walk(pathname, &nd); 227731e6b01fSNick Piggin if (unlikely(error)) 227831e6b01fSNick Piggin goto out_filp; 22791da177e4SLinus Torvalds 228047c805dcSAl Viro filp = do_last(&nd, &path, op, pathname); 2281806b681cSAl Viro while (unlikely(!filp)) { /* trailing symlink */ 22827b9337aaSNick Piggin struct path link = path; 22837b9337aaSNick Piggin struct inode *linki = link.dentry->d_inode; 2284def4af30SAl Viro void *cookie; 228540b39136SAl Viro if (!(nd.flags & LOOKUP_FOLLOW) || count++ == 32) { 228640b39136SAl Viro path_put_conditional(&path, &nd); 228740b39136SAl Viro path_put(&nd.path); 228840b39136SAl Viro filp = ERR_PTR(-ELOOP); 228940b39136SAl Viro break; 229040b39136SAl Viro } 2291806b681cSAl Viro /* 2292806b681cSAl Viro * This is subtle. Instead of calling do_follow_link() we do 2293806b681cSAl Viro * the thing by hands. The reason is that this way we have zero 2294806b681cSAl Viro * link_count and path_walk() (called from ->follow_link) 2295806b681cSAl Viro * honoring LOOKUP_PARENT. After that we have the parent and 2296806b681cSAl Viro * last component, i.e. we are in the same situation as after 2297806b681cSAl Viro * the first path_walk(). Well, almost - if the last component 2298806b681cSAl Viro * is normal we get its copy stored in nd->last.name and we will 2299806b681cSAl Viro * have to putname() it when we are done. Procfs-like symlinks 2300806b681cSAl Viro * just set LAST_BIND. 2301806b681cSAl Viro */ 2302806b681cSAl Viro nd.flags |= LOOKUP_PARENT; 2303c3e380b0SAl Viro nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 23047b9337aaSNick Piggin error = __do_follow_link(&link, &nd, &cookie); 2305c3e380b0SAl Viro if (unlikely(error)) 2306f1afe9efSAl Viro filp = ERR_PTR(error); 2307c3e380b0SAl Viro else 230847c805dcSAl Viro filp = do_last(&nd, &path, op, pathname); 2309f1afe9efSAl Viro if (!IS_ERR(cookie) && linki->i_op->put_link) 23107b9337aaSNick Piggin linki->i_op->put_link(link.dentry, &nd, cookie); 23117b9337aaSNick Piggin path_put(&link); 2312806b681cSAl Viro } 231310fa8e62SAl Viro out: 23145b6ca027SAl Viro if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT)) 23152a737871SAl Viro path_put(&nd.root); 2316fe2d35ffSAl Viro if (base) 2317fe2d35ffSAl Viro fput(base); 23182dab5974SLinus Torvalds release_open_intent(&nd); 231910fa8e62SAl Viro return filp; 23201da177e4SLinus Torvalds 232131e6b01fSNick Piggin out_filp: 232210fa8e62SAl Viro filp = ERR_PTR(error); 232310fa8e62SAl Viro goto out; 2324de459215SKirill Korotaev } 23251da177e4SLinus Torvalds 232613aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 232713aab428SAl Viro const struct open_flags *op, int flags) 232813aab428SAl Viro { 232913aab428SAl Viro struct file *filp; 233013aab428SAl Viro 233113aab428SAl Viro filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU); 233213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 233313aab428SAl Viro filp = path_openat(dfd, pathname, op, flags); 233413aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 233513aab428SAl Viro filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL); 233613aab428SAl Viro return filp; 233713aab428SAl Viro } 233813aab428SAl Viro 23391da177e4SLinus Torvalds /** 23401da177e4SLinus Torvalds * lookup_create - lookup a dentry, creating it if it doesn't exist 23411da177e4SLinus Torvalds * @nd: nameidata info 23421da177e4SLinus Torvalds * @is_dir: directory flag 23431da177e4SLinus Torvalds * 23441da177e4SLinus Torvalds * Simple function to lookup and return a dentry and create it 23451da177e4SLinus Torvalds * if it doesn't exist. Is SMP-safe. 2346c663e5d8SChristoph Hellwig * 23474ac91378SJan Blunck * Returns with nd->path.dentry->d_inode->i_mutex locked. 23481da177e4SLinus Torvalds */ 23491da177e4SLinus Torvalds struct dentry *lookup_create(struct nameidata *nd, int is_dir) 23501da177e4SLinus Torvalds { 2351c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 23521da177e4SLinus Torvalds 23534ac91378SJan Blunck mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2354c663e5d8SChristoph Hellwig /* 2355c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2356c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2357c663e5d8SChristoph Hellwig */ 23581da177e4SLinus Torvalds if (nd->last_type != LAST_NORM) 23591da177e4SLinus Torvalds goto fail; 23601da177e4SLinus Torvalds nd->flags &= ~LOOKUP_PARENT; 23613516586aSAl Viro nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2362a634904aSASANO Masahiro nd->intent.open.flags = O_EXCL; 2363c663e5d8SChristoph Hellwig 2364c663e5d8SChristoph Hellwig /* 2365c663e5d8SChristoph Hellwig * Do the final lookup. 2366c663e5d8SChristoph Hellwig */ 236749705b77SChristoph Hellwig dentry = lookup_hash(nd); 23681da177e4SLinus Torvalds if (IS_ERR(dentry)) 23691da177e4SLinus Torvalds goto fail; 2370c663e5d8SChristoph Hellwig 2371e9baf6e5SAl Viro if (dentry->d_inode) 2372e9baf6e5SAl Viro goto eexist; 2373c663e5d8SChristoph Hellwig /* 2374c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 2375c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 2376c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 2377c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 2378c663e5d8SChristoph Hellwig */ 2379e9baf6e5SAl Viro if (unlikely(!is_dir && nd->last.name[nd->last.len])) { 23801da177e4SLinus Torvalds dput(dentry); 23811da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 2382e9baf6e5SAl Viro } 2383e9baf6e5SAl Viro return dentry; 2384e9baf6e5SAl Viro eexist: 2385e9baf6e5SAl Viro dput(dentry); 2386e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 23871da177e4SLinus Torvalds fail: 23881da177e4SLinus Torvalds return dentry; 23891da177e4SLinus Torvalds } 2390f81a0bffSChristoph Hellwig EXPORT_SYMBOL_GPL(lookup_create); 23911da177e4SLinus Torvalds 23921da177e4SLinus Torvalds int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 23931da177e4SLinus Torvalds { 2394a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 23951da177e4SLinus Torvalds 23961da177e4SLinus Torvalds if (error) 23971da177e4SLinus Torvalds return error; 23981da177e4SLinus Torvalds 23991da177e4SLinus Torvalds if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 24001da177e4SLinus Torvalds return -EPERM; 24011da177e4SLinus Torvalds 2402acfa4380SAl Viro if (!dir->i_op->mknod) 24031da177e4SLinus Torvalds return -EPERM; 24041da177e4SLinus Torvalds 240508ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 240608ce5f16SSerge E. Hallyn if (error) 240708ce5f16SSerge E. Hallyn return error; 240808ce5f16SSerge E. Hallyn 24091da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 24101da177e4SLinus Torvalds if (error) 24111da177e4SLinus Torvalds return error; 24121da177e4SLinus Torvalds 24131da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 2414a74574aaSStephen Smalley if (!error) 2415f38aa942SAmy Griffis fsnotify_create(dir, dentry); 24161da177e4SLinus Torvalds return error; 24171da177e4SLinus Torvalds } 24181da177e4SLinus Torvalds 2419463c3197SDave Hansen static int may_mknod(mode_t mode) 2420463c3197SDave Hansen { 2421463c3197SDave Hansen switch (mode & S_IFMT) { 2422463c3197SDave Hansen case S_IFREG: 2423463c3197SDave Hansen case S_IFCHR: 2424463c3197SDave Hansen case S_IFBLK: 2425463c3197SDave Hansen case S_IFIFO: 2426463c3197SDave Hansen case S_IFSOCK: 2427463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 2428463c3197SDave Hansen return 0; 2429463c3197SDave Hansen case S_IFDIR: 2430463c3197SDave Hansen return -EPERM; 2431463c3197SDave Hansen default: 2432463c3197SDave Hansen return -EINVAL; 2433463c3197SDave Hansen } 2434463c3197SDave Hansen } 2435463c3197SDave Hansen 24362e4d0924SHeiko Carstens SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, 24372e4d0924SHeiko Carstens unsigned, dev) 24381da177e4SLinus Torvalds { 24392ad94ae6SAl Viro int error; 24401da177e4SLinus Torvalds char *tmp; 24411da177e4SLinus Torvalds struct dentry *dentry; 24421da177e4SLinus Torvalds struct nameidata nd; 24431da177e4SLinus Torvalds 24441da177e4SLinus Torvalds if (S_ISDIR(mode)) 24451da177e4SLinus Torvalds return -EPERM; 24461da177e4SLinus Torvalds 24472ad94ae6SAl Viro error = user_path_parent(dfd, filename, &nd, &tmp); 24481da177e4SLinus Torvalds if (error) 24492ad94ae6SAl Viro return error; 24502ad94ae6SAl Viro 24511da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 2452463c3197SDave Hansen if (IS_ERR(dentry)) { 24531da177e4SLinus Torvalds error = PTR_ERR(dentry); 2454463c3197SDave Hansen goto out_unlock; 2455463c3197SDave Hansen } 24564ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2457ce3b0f8dSAl Viro mode &= ~current_umask(); 2458463c3197SDave Hansen error = may_mknod(mode); 2459463c3197SDave Hansen if (error) 2460463c3197SDave Hansen goto out_dput; 2461463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2462463c3197SDave Hansen if (error) 2463463c3197SDave Hansen goto out_dput; 2464be6d3e56SKentaro Takeda error = security_path_mknod(&nd.path, dentry, mode, dev); 2465be6d3e56SKentaro Takeda if (error) 2466be6d3e56SKentaro Takeda goto out_drop_write; 24671da177e4SLinus Torvalds switch (mode & S_IFMT) { 24681da177e4SLinus Torvalds case 0: case S_IFREG: 24694ac91378SJan Blunck error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); 24701da177e4SLinus Torvalds break; 24711da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 24724ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, 24731da177e4SLinus Torvalds new_decode_dev(dev)); 24741da177e4SLinus Torvalds break; 24751da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 24764ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); 24771da177e4SLinus Torvalds break; 24781da177e4SLinus Torvalds } 2479be6d3e56SKentaro Takeda out_drop_write: 2480463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2481463c3197SDave Hansen out_dput: 24821da177e4SLinus Torvalds dput(dentry); 2483463c3197SDave Hansen out_unlock: 24844ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 24851d957f9bSJan Blunck path_put(&nd.path); 24861da177e4SLinus Torvalds putname(tmp); 24871da177e4SLinus Torvalds 24881da177e4SLinus Torvalds return error; 24891da177e4SLinus Torvalds } 24901da177e4SLinus Torvalds 24913480b257SHeiko Carstens SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) 24925590ff0dSUlrich Drepper { 24935590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 24945590ff0dSUlrich Drepper } 24955590ff0dSUlrich Drepper 24961da177e4SLinus Torvalds int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 24971da177e4SLinus Torvalds { 2498a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 24991da177e4SLinus Torvalds 25001da177e4SLinus Torvalds if (error) 25011da177e4SLinus Torvalds return error; 25021da177e4SLinus Torvalds 2503acfa4380SAl Viro if (!dir->i_op->mkdir) 25041da177e4SLinus Torvalds return -EPERM; 25051da177e4SLinus Torvalds 25061da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 25071da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 25081da177e4SLinus Torvalds if (error) 25091da177e4SLinus Torvalds return error; 25101da177e4SLinus Torvalds 25111da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 2512a74574aaSStephen Smalley if (!error) 2513f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 25141da177e4SLinus Torvalds return error; 25151da177e4SLinus Torvalds } 25161da177e4SLinus Torvalds 25172e4d0924SHeiko Carstens SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) 25181da177e4SLinus Torvalds { 25191da177e4SLinus Torvalds int error = 0; 25201da177e4SLinus Torvalds char * tmp; 25216902d925SDave Hansen struct dentry *dentry; 25226902d925SDave Hansen struct nameidata nd; 25231da177e4SLinus Torvalds 25242ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &tmp); 25252ad94ae6SAl Viro if (error) 25266902d925SDave Hansen goto out_err; 25271da177e4SLinus Torvalds 25281da177e4SLinus Torvalds dentry = lookup_create(&nd, 1); 25291da177e4SLinus Torvalds error = PTR_ERR(dentry); 25306902d925SDave Hansen if (IS_ERR(dentry)) 25316902d925SDave Hansen goto out_unlock; 25326902d925SDave Hansen 25334ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2534ce3b0f8dSAl Viro mode &= ~current_umask(); 2535463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2536463c3197SDave Hansen if (error) 2537463c3197SDave Hansen goto out_dput; 2538be6d3e56SKentaro Takeda error = security_path_mkdir(&nd.path, dentry, mode); 2539be6d3e56SKentaro Takeda if (error) 2540be6d3e56SKentaro Takeda goto out_drop_write; 25414ac91378SJan Blunck error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); 2542be6d3e56SKentaro Takeda out_drop_write: 2543463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2544463c3197SDave Hansen out_dput: 25451da177e4SLinus Torvalds dput(dentry); 25466902d925SDave Hansen out_unlock: 25474ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 25481d957f9bSJan Blunck path_put(&nd.path); 25491da177e4SLinus Torvalds putname(tmp); 25506902d925SDave Hansen out_err: 25511da177e4SLinus Torvalds return error; 25521da177e4SLinus Torvalds } 25531da177e4SLinus Torvalds 25543cdad428SHeiko Carstens SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) 25555590ff0dSUlrich Drepper { 25565590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 25575590ff0dSUlrich Drepper } 25585590ff0dSUlrich Drepper 25591da177e4SLinus Torvalds /* 25601da177e4SLinus Torvalds * We try to drop the dentry early: we should have 25611da177e4SLinus Torvalds * a usage count of 2 if we're the only user of this 25621da177e4SLinus Torvalds * dentry, and if that is true (possibly after pruning 25631da177e4SLinus Torvalds * the dcache), then we drop the dentry now. 25641da177e4SLinus Torvalds * 25651da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 25661da177e4SLinus Torvalds * do a 25671da177e4SLinus Torvalds * 25681da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 25691da177e4SLinus Torvalds * return -EBUSY; 25701da177e4SLinus Torvalds * 25711da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 25721da177e4SLinus Torvalds * that is still in use by something else.. 25731da177e4SLinus Torvalds */ 25741da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 25751da177e4SLinus Torvalds { 25761da177e4SLinus Torvalds dget(dentry); 25771da177e4SLinus Torvalds shrink_dcache_parent(dentry); 25781da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 2579b7ab39f6SNick Piggin if (dentry->d_count == 2) 25801da177e4SLinus Torvalds __d_drop(dentry); 25811da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 25821da177e4SLinus Torvalds } 25831da177e4SLinus Torvalds 25841da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 25851da177e4SLinus Torvalds { 25861da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 25871da177e4SLinus Torvalds 25881da177e4SLinus Torvalds if (error) 25891da177e4SLinus Torvalds return error; 25901da177e4SLinus Torvalds 2591acfa4380SAl Viro if (!dir->i_op->rmdir) 25921da177e4SLinus Torvalds return -EPERM; 25931da177e4SLinus Torvalds 25941b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 25951da177e4SLinus Torvalds dentry_unhash(dentry); 25961da177e4SLinus Torvalds if (d_mountpoint(dentry)) 25971da177e4SLinus Torvalds error = -EBUSY; 25981da177e4SLinus Torvalds else { 25991da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 26001da177e4SLinus Torvalds if (!error) { 26011da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 2602d83c49f3SAl Viro if (!error) { 26031da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 2604d83c49f3SAl Viro dont_mount(dentry); 2605d83c49f3SAl Viro } 26061da177e4SLinus Torvalds } 26071da177e4SLinus Torvalds } 26081b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 26091da177e4SLinus Torvalds if (!error) { 26101da177e4SLinus Torvalds d_delete(dentry); 26111da177e4SLinus Torvalds } 26121da177e4SLinus Torvalds dput(dentry); 26131da177e4SLinus Torvalds 26141da177e4SLinus Torvalds return error; 26151da177e4SLinus Torvalds } 26161da177e4SLinus Torvalds 26175590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 26181da177e4SLinus Torvalds { 26191da177e4SLinus Torvalds int error = 0; 26201da177e4SLinus Torvalds char * name; 26211da177e4SLinus Torvalds struct dentry *dentry; 26221da177e4SLinus Torvalds struct nameidata nd; 26231da177e4SLinus Torvalds 26242ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 26251da177e4SLinus Torvalds if (error) 26262ad94ae6SAl Viro return error; 26271da177e4SLinus Torvalds 26281da177e4SLinus Torvalds switch(nd.last_type) { 26291da177e4SLinus Torvalds case LAST_DOTDOT: 26301da177e4SLinus Torvalds error = -ENOTEMPTY; 26311da177e4SLinus Torvalds goto exit1; 26321da177e4SLinus Torvalds case LAST_DOT: 26331da177e4SLinus Torvalds error = -EINVAL; 26341da177e4SLinus Torvalds goto exit1; 26351da177e4SLinus Torvalds case LAST_ROOT: 26361da177e4SLinus Torvalds error = -EBUSY; 26371da177e4SLinus Torvalds goto exit1; 26381da177e4SLinus Torvalds } 26390612d9fbSOGAWA Hirofumi 26400612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 26410612d9fbSOGAWA Hirofumi 26424ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 264349705b77SChristoph Hellwig dentry = lookup_hash(&nd); 26441da177e4SLinus Torvalds error = PTR_ERR(dentry); 26456902d925SDave Hansen if (IS_ERR(dentry)) 26466902d925SDave Hansen goto exit2; 26470622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 26480622753bSDave Hansen if (error) 26490622753bSDave Hansen goto exit3; 2650be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 2651be6d3e56SKentaro Takeda if (error) 2652be6d3e56SKentaro Takeda goto exit4; 26534ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 2654be6d3e56SKentaro Takeda exit4: 26550622753bSDave Hansen mnt_drop_write(nd.path.mnt); 26560622753bSDave Hansen exit3: 26571da177e4SLinus Torvalds dput(dentry); 26586902d925SDave Hansen exit2: 26594ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 26601da177e4SLinus Torvalds exit1: 26611d957f9bSJan Blunck path_put(&nd.path); 26621da177e4SLinus Torvalds putname(name); 26631da177e4SLinus Torvalds return error; 26641da177e4SLinus Torvalds } 26651da177e4SLinus Torvalds 26663cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 26675590ff0dSUlrich Drepper { 26685590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 26695590ff0dSUlrich Drepper } 26705590ff0dSUlrich Drepper 26711da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 26721da177e4SLinus Torvalds { 26731da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 26741da177e4SLinus Torvalds 26751da177e4SLinus Torvalds if (error) 26761da177e4SLinus Torvalds return error; 26771da177e4SLinus Torvalds 2678acfa4380SAl Viro if (!dir->i_op->unlink) 26791da177e4SLinus Torvalds return -EPERM; 26801da177e4SLinus Torvalds 26811b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 26821da177e4SLinus Torvalds if (d_mountpoint(dentry)) 26831da177e4SLinus Torvalds error = -EBUSY; 26841da177e4SLinus Torvalds else { 26851da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 2686bec1052eSAl Viro if (!error) { 26871da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 2688bec1052eSAl Viro if (!error) 2689d83c49f3SAl Viro dont_mount(dentry); 2690bec1052eSAl Viro } 26911da177e4SLinus Torvalds } 26921b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 26931da177e4SLinus Torvalds 26941da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 26951da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 2696ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 26971da177e4SLinus Torvalds d_delete(dentry); 26981da177e4SLinus Torvalds } 26990eeca283SRobert Love 27001da177e4SLinus Torvalds return error; 27011da177e4SLinus Torvalds } 27021da177e4SLinus Torvalds 27031da177e4SLinus Torvalds /* 27041da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 27051b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 27061da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 27071da177e4SLinus Torvalds * while waiting on the I/O. 27081da177e4SLinus Torvalds */ 27095590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 27101da177e4SLinus Torvalds { 27112ad94ae6SAl Viro int error; 27121da177e4SLinus Torvalds char *name; 27131da177e4SLinus Torvalds struct dentry *dentry; 27141da177e4SLinus Torvalds struct nameidata nd; 27151da177e4SLinus Torvalds struct inode *inode = NULL; 27161da177e4SLinus Torvalds 27172ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 27181da177e4SLinus Torvalds if (error) 27192ad94ae6SAl Viro return error; 27202ad94ae6SAl Viro 27211da177e4SLinus Torvalds error = -EISDIR; 27221da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 27231da177e4SLinus Torvalds goto exit1; 27240612d9fbSOGAWA Hirofumi 27250612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 27260612d9fbSOGAWA Hirofumi 27274ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 272849705b77SChristoph Hellwig dentry = lookup_hash(&nd); 27291da177e4SLinus Torvalds error = PTR_ERR(dentry); 27301da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 27311da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 27321da177e4SLinus Torvalds if (nd.last.name[nd.last.len]) 27331da177e4SLinus Torvalds goto slashes; 27341da177e4SLinus Torvalds inode = dentry->d_inode; 27351da177e4SLinus Torvalds if (inode) 27367de9c6eeSAl Viro ihold(inode); 27370622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 27380622753bSDave Hansen if (error) 27390622753bSDave Hansen goto exit2; 2740be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 2741be6d3e56SKentaro Takeda if (error) 2742be6d3e56SKentaro Takeda goto exit3; 27434ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 2744be6d3e56SKentaro Takeda exit3: 27450622753bSDave Hansen mnt_drop_write(nd.path.mnt); 27461da177e4SLinus Torvalds exit2: 27471da177e4SLinus Torvalds dput(dentry); 27481da177e4SLinus Torvalds } 27494ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 27501da177e4SLinus Torvalds if (inode) 27511da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 27521da177e4SLinus Torvalds exit1: 27531d957f9bSJan Blunck path_put(&nd.path); 27541da177e4SLinus Torvalds putname(name); 27551da177e4SLinus Torvalds return error; 27561da177e4SLinus Torvalds 27571da177e4SLinus Torvalds slashes: 27581da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 27591da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 27601da177e4SLinus Torvalds goto exit2; 27611da177e4SLinus Torvalds } 27621da177e4SLinus Torvalds 27632e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 27645590ff0dSUlrich Drepper { 27655590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 27665590ff0dSUlrich Drepper return -EINVAL; 27675590ff0dSUlrich Drepper 27685590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 27695590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 27705590ff0dSUlrich Drepper 27715590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 27725590ff0dSUlrich Drepper } 27735590ff0dSUlrich Drepper 27743480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 27755590ff0dSUlrich Drepper { 27765590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 27775590ff0dSUlrich Drepper } 27785590ff0dSUlrich Drepper 2779db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 27801da177e4SLinus Torvalds { 2781a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 27821da177e4SLinus Torvalds 27831da177e4SLinus Torvalds if (error) 27841da177e4SLinus Torvalds return error; 27851da177e4SLinus Torvalds 2786acfa4380SAl Viro if (!dir->i_op->symlink) 27871da177e4SLinus Torvalds return -EPERM; 27881da177e4SLinus Torvalds 27891da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 27901da177e4SLinus Torvalds if (error) 27911da177e4SLinus Torvalds return error; 27921da177e4SLinus Torvalds 27931da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 2794a74574aaSStephen Smalley if (!error) 2795f38aa942SAmy Griffis fsnotify_create(dir, dentry); 27961da177e4SLinus Torvalds return error; 27971da177e4SLinus Torvalds } 27981da177e4SLinus Torvalds 27992e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 28002e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 28011da177e4SLinus Torvalds { 28022ad94ae6SAl Viro int error; 28031da177e4SLinus Torvalds char *from; 28041da177e4SLinus Torvalds char *to; 28056902d925SDave Hansen struct dentry *dentry; 28066902d925SDave Hansen struct nameidata nd; 28071da177e4SLinus Torvalds 28081da177e4SLinus Torvalds from = getname(oldname); 28091da177e4SLinus Torvalds if (IS_ERR(from)) 28101da177e4SLinus Torvalds return PTR_ERR(from); 28112ad94ae6SAl Viro 28122ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 28132ad94ae6SAl Viro if (error) 28146902d925SDave Hansen goto out_putname; 28151da177e4SLinus Torvalds 28161da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 28171da177e4SLinus Torvalds error = PTR_ERR(dentry); 28186902d925SDave Hansen if (IS_ERR(dentry)) 28196902d925SDave Hansen goto out_unlock; 28206902d925SDave Hansen 282175c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 282275c3f29dSDave Hansen if (error) 282375c3f29dSDave Hansen goto out_dput; 2824be6d3e56SKentaro Takeda error = security_path_symlink(&nd.path, dentry, from); 2825be6d3e56SKentaro Takeda if (error) 2826be6d3e56SKentaro Takeda goto out_drop_write; 2827db2e747bSMiklos Szeredi error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); 2828be6d3e56SKentaro Takeda out_drop_write: 282975c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 283075c3f29dSDave Hansen out_dput: 28311da177e4SLinus Torvalds dput(dentry); 28326902d925SDave Hansen out_unlock: 28334ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 28341d957f9bSJan Blunck path_put(&nd.path); 28351da177e4SLinus Torvalds putname(to); 28366902d925SDave Hansen out_putname: 28371da177e4SLinus Torvalds putname(from); 28381da177e4SLinus Torvalds return error; 28391da177e4SLinus Torvalds } 28401da177e4SLinus Torvalds 28413480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 28425590ff0dSUlrich Drepper { 28435590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 28445590ff0dSUlrich Drepper } 28455590ff0dSUlrich Drepper 28461da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 28471da177e4SLinus Torvalds { 28481da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 28491da177e4SLinus Torvalds int error; 28501da177e4SLinus Torvalds 28511da177e4SLinus Torvalds if (!inode) 28521da177e4SLinus Torvalds return -ENOENT; 28531da177e4SLinus Torvalds 2854a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 28551da177e4SLinus Torvalds if (error) 28561da177e4SLinus Torvalds return error; 28571da177e4SLinus Torvalds 28581da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 28591da177e4SLinus Torvalds return -EXDEV; 28601da177e4SLinus Torvalds 28611da177e4SLinus Torvalds /* 28621da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 28631da177e4SLinus Torvalds */ 28641da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 28651da177e4SLinus Torvalds return -EPERM; 2866acfa4380SAl Viro if (!dir->i_op->link) 28671da177e4SLinus Torvalds return -EPERM; 28687e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 28691da177e4SLinus Torvalds return -EPERM; 28701da177e4SLinus Torvalds 28711da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 28721da177e4SLinus Torvalds if (error) 28731da177e4SLinus Torvalds return error; 28741da177e4SLinus Torvalds 28757e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 28761da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 28777e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 2878e31e14ecSStephen Smalley if (!error) 28797e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 28801da177e4SLinus Torvalds return error; 28811da177e4SLinus Torvalds } 28821da177e4SLinus Torvalds 28831da177e4SLinus Torvalds /* 28841da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 28851da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 28861da177e4SLinus Torvalds * newname. --KAB 28871da177e4SLinus Torvalds * 28881da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 28891da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 28901da177e4SLinus Torvalds * and other special files. --ADM 28911da177e4SLinus Torvalds */ 28922e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 28932e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 28941da177e4SLinus Torvalds { 28951da177e4SLinus Torvalds struct dentry *new_dentry; 28962d8f3038SAl Viro struct nameidata nd; 28972d8f3038SAl Viro struct path old_path; 28981da177e4SLinus Torvalds int error; 28991da177e4SLinus Torvalds char *to; 29001da177e4SLinus Torvalds 290145c9b11aSUlrich Drepper if ((flags & ~AT_SYMLINK_FOLLOW) != 0) 2902c04030e1SUlrich Drepper return -EINVAL; 2903c04030e1SUlrich Drepper 29042d8f3038SAl Viro error = user_path_at(olddfd, oldname, 290545c9b11aSUlrich Drepper flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, 29062d8f3038SAl Viro &old_path); 29071da177e4SLinus Torvalds if (error) 29082ad94ae6SAl Viro return error; 29092ad94ae6SAl Viro 29102ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 29111da177e4SLinus Torvalds if (error) 29121da177e4SLinus Torvalds goto out; 29131da177e4SLinus Torvalds error = -EXDEV; 29142d8f3038SAl Viro if (old_path.mnt != nd.path.mnt) 29151da177e4SLinus Torvalds goto out_release; 29161da177e4SLinus Torvalds new_dentry = lookup_create(&nd, 0); 29171da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 29186902d925SDave Hansen if (IS_ERR(new_dentry)) 29196902d925SDave Hansen goto out_unlock; 292075c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 292175c3f29dSDave Hansen if (error) 292275c3f29dSDave Hansen goto out_dput; 2923be6d3e56SKentaro Takeda error = security_path_link(old_path.dentry, &nd.path, new_dentry); 2924be6d3e56SKentaro Takeda if (error) 2925be6d3e56SKentaro Takeda goto out_drop_write; 29262d8f3038SAl Viro error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); 2927be6d3e56SKentaro Takeda out_drop_write: 292875c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 292975c3f29dSDave Hansen out_dput: 29301da177e4SLinus Torvalds dput(new_dentry); 29316902d925SDave Hansen out_unlock: 29324ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 29331da177e4SLinus Torvalds out_release: 29341d957f9bSJan Blunck path_put(&nd.path); 29352ad94ae6SAl Viro putname(to); 29361da177e4SLinus Torvalds out: 29372d8f3038SAl Viro path_put(&old_path); 29381da177e4SLinus Torvalds 29391da177e4SLinus Torvalds return error; 29401da177e4SLinus Torvalds } 29411da177e4SLinus Torvalds 29423480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 29435590ff0dSUlrich Drepper { 2944c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 29455590ff0dSUlrich Drepper } 29465590ff0dSUlrich Drepper 29471da177e4SLinus Torvalds /* 29481da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 29491da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 29501da177e4SLinus Torvalds * Problems: 29511da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 29521da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 29531da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 2954a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 29551da177e4SLinus Torvalds * story. 29561da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 29571b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 29581da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 29591da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 2960a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 29611da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 29621da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 29631da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 2964a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 29651da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 29661da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 29671da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 29681da177e4SLinus Torvalds * d) some filesystems don't support opened-but-unlinked directories, 29691da177e4SLinus Torvalds * either because of layout or because they are not ready to deal with 29701da177e4SLinus Torvalds * all cases correctly. The latter will be fixed (taking this sort of 29711da177e4SLinus Torvalds * stuff into VFS), but the former is not going away. Solution: the same 29721da177e4SLinus Torvalds * trick as in rmdir(). 29731da177e4SLinus Torvalds * e) conversion from fhandle to dentry may come in the wrong moment - when 29741b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 29751da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 2976c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 29771da177e4SLinus Torvalds * locking]. 29781da177e4SLinus Torvalds */ 297975c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 29801da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 29811da177e4SLinus Torvalds { 29821da177e4SLinus Torvalds int error = 0; 29831da177e4SLinus Torvalds struct inode *target; 29841da177e4SLinus Torvalds 29851da177e4SLinus Torvalds /* 29861da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 29871da177e4SLinus Torvalds * we'll need to flip '..'. 29881da177e4SLinus Torvalds */ 29891da177e4SLinus Torvalds if (new_dir != old_dir) { 2990f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 29911da177e4SLinus Torvalds if (error) 29921da177e4SLinus Torvalds return error; 29931da177e4SLinus Torvalds } 29941da177e4SLinus Torvalds 29951da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 29961da177e4SLinus Torvalds if (error) 29971da177e4SLinus Torvalds return error; 29981da177e4SLinus Torvalds 29991da177e4SLinus Torvalds target = new_dentry->d_inode; 3000d83c49f3SAl Viro if (target) 30011b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 30021da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 30031da177e4SLinus Torvalds error = -EBUSY; 3004d83c49f3SAl Viro else { 3005d83c49f3SAl Viro if (target) 3006d83c49f3SAl Viro dentry_unhash(new_dentry); 30071da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 3008d83c49f3SAl Viro } 30091da177e4SLinus Torvalds if (target) { 3010d83c49f3SAl Viro if (!error) { 30111da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3012d83c49f3SAl Viro dont_mount(new_dentry); 3013d83c49f3SAl Viro } 30141b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 30151da177e4SLinus Torvalds if (d_unhashed(new_dentry)) 30161da177e4SLinus Torvalds d_rehash(new_dentry); 30171da177e4SLinus Torvalds dput(new_dentry); 30181da177e4SLinus Torvalds } 3019e31e14ecSStephen Smalley if (!error) 3020349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 30211da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 30221da177e4SLinus Torvalds return error; 30231da177e4SLinus Torvalds } 30241da177e4SLinus Torvalds 302575c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 30261da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30271da177e4SLinus Torvalds { 30281da177e4SLinus Torvalds struct inode *target; 30291da177e4SLinus Torvalds int error; 30301da177e4SLinus Torvalds 30311da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30321da177e4SLinus Torvalds if (error) 30331da177e4SLinus Torvalds return error; 30341da177e4SLinus Torvalds 30351da177e4SLinus Torvalds dget(new_dentry); 30361da177e4SLinus Torvalds target = new_dentry->d_inode; 30371da177e4SLinus Torvalds if (target) 30381b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 30391da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 30401da177e4SLinus Torvalds error = -EBUSY; 30411da177e4SLinus Torvalds else 30421da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 30431da177e4SLinus Torvalds if (!error) { 3044bec1052eSAl Viro if (target) 3045d83c49f3SAl Viro dont_mount(new_dentry); 3046349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 30471da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 30481da177e4SLinus Torvalds } 30491da177e4SLinus Torvalds if (target) 30501b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 30511da177e4SLinus Torvalds dput(new_dentry); 30521da177e4SLinus Torvalds return error; 30531da177e4SLinus Torvalds } 30541da177e4SLinus Torvalds 30551da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 30561da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30571da177e4SLinus Torvalds { 30581da177e4SLinus Torvalds int error; 30591da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 306059b0df21SEric Paris const unsigned char *old_name; 30611da177e4SLinus Torvalds 30621da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 30631da177e4SLinus Torvalds return 0; 30641da177e4SLinus Torvalds 30651da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 30661da177e4SLinus Torvalds if (error) 30671da177e4SLinus Torvalds return error; 30681da177e4SLinus Torvalds 30691da177e4SLinus Torvalds if (!new_dentry->d_inode) 3070a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 30711da177e4SLinus Torvalds else 30721da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 30731da177e4SLinus Torvalds if (error) 30741da177e4SLinus Torvalds return error; 30751da177e4SLinus Torvalds 3076acfa4380SAl Viro if (!old_dir->i_op->rename) 30771da177e4SLinus Torvalds return -EPERM; 30781da177e4SLinus Torvalds 30790eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 30800eeca283SRobert Love 30811da177e4SLinus Torvalds if (is_dir) 30821da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 30831da177e4SLinus Torvalds else 30841da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3085123df294SAl Viro if (!error) 3086123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 30875a190ae6SAl Viro new_dentry->d_inode, old_dentry); 30880eeca283SRobert Love fsnotify_oldname_free(old_name); 30890eeca283SRobert Love 30901da177e4SLinus Torvalds return error; 30911da177e4SLinus Torvalds } 30921da177e4SLinus Torvalds 30932e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 30942e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 30951da177e4SLinus Torvalds { 30961da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 30971da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 30981da177e4SLinus Torvalds struct dentry *trap; 30991da177e4SLinus Torvalds struct nameidata oldnd, newnd; 31002ad94ae6SAl Viro char *from; 31012ad94ae6SAl Viro char *to; 31022ad94ae6SAl Viro int error; 31031da177e4SLinus Torvalds 31042ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 31051da177e4SLinus Torvalds if (error) 31061da177e4SLinus Torvalds goto exit; 31071da177e4SLinus Torvalds 31082ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 31091da177e4SLinus Torvalds if (error) 31101da177e4SLinus Torvalds goto exit1; 31111da177e4SLinus Torvalds 31121da177e4SLinus Torvalds error = -EXDEV; 31134ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 31141da177e4SLinus Torvalds goto exit2; 31151da177e4SLinus Torvalds 31164ac91378SJan Blunck old_dir = oldnd.path.dentry; 31171da177e4SLinus Torvalds error = -EBUSY; 31181da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 31191da177e4SLinus Torvalds goto exit2; 31201da177e4SLinus Torvalds 31214ac91378SJan Blunck new_dir = newnd.path.dentry; 31221da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 31231da177e4SLinus Torvalds goto exit2; 31241da177e4SLinus Torvalds 31250612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 31260612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 31274e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 31280612d9fbSOGAWA Hirofumi 31291da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 31301da177e4SLinus Torvalds 313149705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 31321da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 31331da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 31341da177e4SLinus Torvalds goto exit3; 31351da177e4SLinus Torvalds /* source must exist */ 31361da177e4SLinus Torvalds error = -ENOENT; 31371da177e4SLinus Torvalds if (!old_dentry->d_inode) 31381da177e4SLinus Torvalds goto exit4; 31391da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 31401da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 31411da177e4SLinus Torvalds error = -ENOTDIR; 31421da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 31431da177e4SLinus Torvalds goto exit4; 31441da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 31451da177e4SLinus Torvalds goto exit4; 31461da177e4SLinus Torvalds } 31471da177e4SLinus Torvalds /* source should not be ancestor of target */ 31481da177e4SLinus Torvalds error = -EINVAL; 31491da177e4SLinus Torvalds if (old_dentry == trap) 31501da177e4SLinus Torvalds goto exit4; 315149705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 31521da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 31531da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 31541da177e4SLinus Torvalds goto exit4; 31551da177e4SLinus Torvalds /* target should not be an ancestor of source */ 31561da177e4SLinus Torvalds error = -ENOTEMPTY; 31571da177e4SLinus Torvalds if (new_dentry == trap) 31581da177e4SLinus Torvalds goto exit5; 31591da177e4SLinus Torvalds 31609079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 31619079b1ebSDave Hansen if (error) 31629079b1ebSDave Hansen goto exit5; 3163be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3164be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3165be6d3e56SKentaro Takeda if (error) 3166be6d3e56SKentaro Takeda goto exit6; 31671da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 31681da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 3169be6d3e56SKentaro Takeda exit6: 31709079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 31711da177e4SLinus Torvalds exit5: 31721da177e4SLinus Torvalds dput(new_dentry); 31731da177e4SLinus Torvalds exit4: 31741da177e4SLinus Torvalds dput(old_dentry); 31751da177e4SLinus Torvalds exit3: 31761da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 31771da177e4SLinus Torvalds exit2: 31781d957f9bSJan Blunck path_put(&newnd.path); 31792ad94ae6SAl Viro putname(to); 31801da177e4SLinus Torvalds exit1: 31811d957f9bSJan Blunck path_put(&oldnd.path); 31821da177e4SLinus Torvalds putname(from); 31832ad94ae6SAl Viro exit: 31841da177e4SLinus Torvalds return error; 31851da177e4SLinus Torvalds } 31861da177e4SLinus Torvalds 3187a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 31885590ff0dSUlrich Drepper { 31895590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 31905590ff0dSUlrich Drepper } 31915590ff0dSUlrich Drepper 31921da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 31931da177e4SLinus Torvalds { 31941da177e4SLinus Torvalds int len; 31951da177e4SLinus Torvalds 31961da177e4SLinus Torvalds len = PTR_ERR(link); 31971da177e4SLinus Torvalds if (IS_ERR(link)) 31981da177e4SLinus Torvalds goto out; 31991da177e4SLinus Torvalds 32001da177e4SLinus Torvalds len = strlen(link); 32011da177e4SLinus Torvalds if (len > (unsigned) buflen) 32021da177e4SLinus Torvalds len = buflen; 32031da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 32041da177e4SLinus Torvalds len = -EFAULT; 32051da177e4SLinus Torvalds out: 32061da177e4SLinus Torvalds return len; 32071da177e4SLinus Torvalds } 32081da177e4SLinus Torvalds 32091da177e4SLinus Torvalds /* 32101da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 32111da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 32121da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 32131da177e4SLinus Torvalds */ 32141da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 32151da177e4SLinus Torvalds { 32161da177e4SLinus Torvalds struct nameidata nd; 3217cc314eefSLinus Torvalds void *cookie; 3218694a1764SMarcin Slusarz int res; 3219cc314eefSLinus Torvalds 32201da177e4SLinus Torvalds nd.depth = 0; 3221cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3222694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3223694a1764SMarcin Slusarz return PTR_ERR(cookie); 3224694a1764SMarcin Slusarz 3225694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 32261da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3227cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3228694a1764SMarcin Slusarz return res; 32291da177e4SLinus Torvalds } 32301da177e4SLinus Torvalds 32311da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 32321da177e4SLinus Torvalds { 32331da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 32341da177e4SLinus Torvalds } 32351da177e4SLinus Torvalds 32361da177e4SLinus Torvalds /* get the link contents into pagecache */ 32371da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 32381da177e4SLinus Torvalds { 3239ebd09abbSDuane Griffin char *kaddr; 32401da177e4SLinus Torvalds struct page *page; 32411da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3242090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 32431da177e4SLinus Torvalds if (IS_ERR(page)) 32446fe6900eSNick Piggin return (char*)page; 32451da177e4SLinus Torvalds *ppage = page; 3246ebd09abbSDuane Griffin kaddr = kmap(page); 3247ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3248ebd09abbSDuane Griffin return kaddr; 32491da177e4SLinus Torvalds } 32501da177e4SLinus Torvalds 32511da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 32521da177e4SLinus Torvalds { 32531da177e4SLinus Torvalds struct page *page = NULL; 32541da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 32551da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 32561da177e4SLinus Torvalds if (page) { 32571da177e4SLinus Torvalds kunmap(page); 32581da177e4SLinus Torvalds page_cache_release(page); 32591da177e4SLinus Torvalds } 32601da177e4SLinus Torvalds return res; 32611da177e4SLinus Torvalds } 32621da177e4SLinus Torvalds 3263cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 32641da177e4SLinus Torvalds { 3265cc314eefSLinus Torvalds struct page *page = NULL; 32661da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3267cc314eefSLinus Torvalds return page; 32681da177e4SLinus Torvalds } 32691da177e4SLinus Torvalds 3270cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 32711da177e4SLinus Torvalds { 3272cc314eefSLinus Torvalds struct page *page = cookie; 3273cc314eefSLinus Torvalds 3274cc314eefSLinus Torvalds if (page) { 32751da177e4SLinus Torvalds kunmap(page); 32761da177e4SLinus Torvalds page_cache_release(page); 32771da177e4SLinus Torvalds } 32781da177e4SLinus Torvalds } 32791da177e4SLinus Torvalds 328054566b2cSNick Piggin /* 328154566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 328254566b2cSNick Piggin */ 328354566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 32841da177e4SLinus Torvalds { 32851da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 32860adb25d2SKirill Korotaev struct page *page; 3287afddba49SNick Piggin void *fsdata; 3288beb497abSDmitriy Monakhov int err; 32891da177e4SLinus Torvalds char *kaddr; 329054566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 329154566b2cSNick Piggin if (nofs) 329254566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 32931da177e4SLinus Torvalds 32947e53cac4SNeilBrown retry: 3295afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 329654566b2cSNick Piggin flags, &page, &fsdata); 32971da177e4SLinus Torvalds if (err) 3298afddba49SNick Piggin goto fail; 3299afddba49SNick Piggin 33001da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 33011da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 33021da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 3303afddba49SNick Piggin 3304afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3305afddba49SNick Piggin page, fsdata); 33061da177e4SLinus Torvalds if (err < 0) 33071da177e4SLinus Torvalds goto fail; 3308afddba49SNick Piggin if (err < len-1) 3309afddba49SNick Piggin goto retry; 3310afddba49SNick Piggin 33111da177e4SLinus Torvalds mark_inode_dirty(inode); 33121da177e4SLinus Torvalds return 0; 33131da177e4SLinus Torvalds fail: 33141da177e4SLinus Torvalds return err; 33151da177e4SLinus Torvalds } 33161da177e4SLinus Torvalds 33170adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 33180adb25d2SKirill Korotaev { 33190adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 332054566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 33210adb25d2SKirill Korotaev } 33220adb25d2SKirill Korotaev 332392e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 33241da177e4SLinus Torvalds .readlink = generic_readlink, 33251da177e4SLinus Torvalds .follow_link = page_follow_link_light, 33261da177e4SLinus Torvalds .put_link = page_put_link, 33271da177e4SLinus Torvalds }; 33281da177e4SLinus Torvalds 33292d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3330cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 33311da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 33321da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 33331da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 33341da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 33351da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 33361da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 33371da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 33381da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 33391da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 33400adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 33411da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 33421da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3343c9c6cac0SAl Viro EXPORT_SYMBOL(kern_path_parent); 3344d1811465SAl Viro EXPORT_SYMBOL(kern_path); 334516f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3346f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 33478c744fb8SChristoph Hellwig EXPORT_SYMBOL(file_permission); 33481da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 33491da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 33501da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 33511da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 33521da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 33531da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 33541da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 33551da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 33561da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 33571da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 33581da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 33591da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 33601da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 33611da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3362