11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namei.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 51da177e4SLinus Torvalds */ 61da177e4SLinus Torvalds 71da177e4SLinus Torvalds /* 81da177e4SLinus Torvalds * Some corrections by tytso. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 121da177e4SLinus Torvalds * lookup logic. 131da177e4SLinus Torvalds */ 141da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 151da177e4SLinus Torvalds */ 161da177e4SLinus Torvalds 171da177e4SLinus Torvalds #include <linux/init.h> 181da177e4SLinus Torvalds #include <linux/module.h> 191da177e4SLinus Torvalds #include <linux/slab.h> 201da177e4SLinus Torvalds #include <linux/fs.h> 211da177e4SLinus Torvalds #include <linux/namei.h> 221da177e4SLinus Torvalds #include <linux/pagemap.h> 230eeca283SRobert Love #include <linux/fsnotify.h> 241da177e4SLinus Torvalds #include <linux/personality.h> 251da177e4SLinus Torvalds #include <linux/security.h> 266146f0d5SMimi Zohar #include <linux/ima.h> 271da177e4SLinus Torvalds #include <linux/syscalls.h> 281da177e4SLinus Torvalds #include <linux/mount.h> 291da177e4SLinus Torvalds #include <linux/audit.h> 3016f7e0feSRandy Dunlap #include <linux/capability.h> 31834f2a4aSTrond Myklebust #include <linux/file.h> 325590ff0dSUlrich Drepper #include <linux/fcntl.h> 3308ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 345ad4e53bSAl Viro #include <linux/fs_struct.h> 351da177e4SLinus Torvalds #include <asm/uaccess.h> 361da177e4SLinus Torvalds 37e81e3f4dSEric Paris #include "internal.h" 38e81e3f4dSEric Paris 391da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 401da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 411da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 421da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 431da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 441da177e4SLinus Torvalds * 451da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 461da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 471da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 481da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 491da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 501da177e4SLinus Torvalds * the special cases of the former code. 511da177e4SLinus Torvalds * 521da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 531da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 541da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 551da177e4SLinus Torvalds * 561da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 571da177e4SLinus Torvalds * resolution to correspond with current state of the code. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 601da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 611da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 621da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 631da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 641da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 651da177e4SLinus Torvalds */ 661da177e4SLinus Torvalds 671da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 681da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 691da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 701da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 711da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 721da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 731da177e4SLinus Torvalds * the name is a symlink pointing to a non-existant name. 741da177e4SLinus Torvalds * 751da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 761da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 771da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 781da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 791da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 801da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 811da177e4SLinus Torvalds * and in the old Linux semantics. 821da177e4SLinus Torvalds */ 831da177e4SLinus Torvalds 841da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 851da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 861da177e4SLinus Torvalds * 871da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 881da177e4SLinus Torvalds */ 891da177e4SLinus Torvalds 901da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 911da177e4SLinus Torvalds * inside the path - always follow. 921da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 931da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 941da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 951da177e4SLinus Torvalds * otherwise - don't follow. 961da177e4SLinus Torvalds * (applied in that order). 971da177e4SLinus Torvalds * 981da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 991da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1001da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1011da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1021da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1031da177e4SLinus Torvalds */ 1041da177e4SLinus Torvalds /* 1051da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 106a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1071da177e4SLinus Torvalds * any extra contention... 1081da177e4SLinus Torvalds */ 1091da177e4SLinus Torvalds 1101da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1111da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1121da177e4SLinus Torvalds * kernel data space before using them.. 1131da177e4SLinus Torvalds * 1141da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1151da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1161da177e4SLinus Torvalds */ 117858119e1SArjan van de Ven static int do_getname(const char __user *filename, char *page) 1181da177e4SLinus Torvalds { 1191da177e4SLinus Torvalds int retval; 1201da177e4SLinus Torvalds unsigned long len = PATH_MAX; 1211da177e4SLinus Torvalds 1221da177e4SLinus Torvalds if (!segment_eq(get_fs(), KERNEL_DS)) { 1231da177e4SLinus Torvalds if ((unsigned long) filename >= TASK_SIZE) 1241da177e4SLinus Torvalds return -EFAULT; 1251da177e4SLinus Torvalds if (TASK_SIZE - (unsigned long) filename < PATH_MAX) 1261da177e4SLinus Torvalds len = TASK_SIZE - (unsigned long) filename; 1271da177e4SLinus Torvalds } 1281da177e4SLinus Torvalds 1291da177e4SLinus Torvalds retval = strncpy_from_user(page, filename, len); 1301da177e4SLinus Torvalds if (retval > 0) { 1311da177e4SLinus Torvalds if (retval < len) 1321da177e4SLinus Torvalds return 0; 1331da177e4SLinus Torvalds return -ENAMETOOLONG; 1341da177e4SLinus Torvalds } else if (!retval) 1351da177e4SLinus Torvalds retval = -ENOENT; 1361da177e4SLinus Torvalds return retval; 1371da177e4SLinus Torvalds } 1381da177e4SLinus Torvalds 1391da177e4SLinus Torvalds char * getname(const char __user * filename) 1401da177e4SLinus Torvalds { 1411da177e4SLinus Torvalds char *tmp, *result; 1421da177e4SLinus Torvalds 1431da177e4SLinus Torvalds result = ERR_PTR(-ENOMEM); 1441da177e4SLinus Torvalds tmp = __getname(); 1451da177e4SLinus Torvalds if (tmp) { 1461da177e4SLinus Torvalds int retval = do_getname(filename, tmp); 1471da177e4SLinus Torvalds 1481da177e4SLinus Torvalds result = tmp; 1491da177e4SLinus Torvalds if (retval < 0) { 1501da177e4SLinus Torvalds __putname(tmp); 1511da177e4SLinus Torvalds result = ERR_PTR(retval); 1521da177e4SLinus Torvalds } 1531da177e4SLinus Torvalds } 1541da177e4SLinus Torvalds audit_getname(result); 1551da177e4SLinus Torvalds return result; 1561da177e4SLinus Torvalds } 1571da177e4SLinus Torvalds 1581da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 1591da177e4SLinus Torvalds void putname(const char *name) 1601da177e4SLinus Torvalds { 1615ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 1621da177e4SLinus Torvalds audit_putname(name); 1631da177e4SLinus Torvalds else 1641da177e4SLinus Torvalds __putname(name); 1651da177e4SLinus Torvalds } 1661da177e4SLinus Torvalds EXPORT_SYMBOL(putname); 1671da177e4SLinus Torvalds #endif 1681da177e4SLinus Torvalds 1695909ccaaSLinus Torvalds /* 1705909ccaaSLinus Torvalds * This does basic POSIX ACL permission checking 1715909ccaaSLinus Torvalds */ 172b74c79e9SNick Piggin static int acl_permission_check(struct inode *inode, int mask, unsigned int flags, 173b74c79e9SNick Piggin int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) 1745909ccaaSLinus Torvalds { 1755909ccaaSLinus Torvalds umode_t mode = inode->i_mode; 1765909ccaaSLinus Torvalds 1775909ccaaSLinus Torvalds mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 1785909ccaaSLinus Torvalds 1795909ccaaSLinus Torvalds if (current_fsuid() == inode->i_uid) 1805909ccaaSLinus Torvalds mode >>= 6; 1815909ccaaSLinus Torvalds else { 1825909ccaaSLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) { 183b74c79e9SNick Piggin int error = check_acl(inode, mask, flags); 1845909ccaaSLinus Torvalds if (error != -EAGAIN) 1855909ccaaSLinus Torvalds return error; 1865909ccaaSLinus Torvalds } 1875909ccaaSLinus Torvalds 1885909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 1895909ccaaSLinus Torvalds mode >>= 3; 1905909ccaaSLinus Torvalds } 1915909ccaaSLinus Torvalds 1925909ccaaSLinus Torvalds /* 1935909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 1945909ccaaSLinus Torvalds */ 1955909ccaaSLinus Torvalds if ((mask & ~mode) == 0) 1965909ccaaSLinus Torvalds return 0; 1975909ccaaSLinus Torvalds return -EACCES; 1985909ccaaSLinus Torvalds } 1991da177e4SLinus Torvalds 2001da177e4SLinus Torvalds /** 2011da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 2021da177e4SLinus Torvalds * @inode: inode to check access rights for 2031da177e4SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 2041da177e4SLinus Torvalds * @check_acl: optional callback to check for Posix ACLs 20539191628SRandy Dunlap * @flags: IPERM_FLAG_ flags. 2061da177e4SLinus Torvalds * 2071da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 2081da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 2091da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 210b74c79e9SNick Piggin * are used for other things. 211b74c79e9SNick Piggin * 212b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 213b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 214b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 2151da177e4SLinus Torvalds */ 216b74c79e9SNick Piggin int generic_permission(struct inode *inode, int mask, unsigned int flags, 217b74c79e9SNick Piggin int (*check_acl)(struct inode *inode, int mask, unsigned int flags)) 2181da177e4SLinus Torvalds { 2195909ccaaSLinus Torvalds int ret; 2201da177e4SLinus Torvalds 2211da177e4SLinus Torvalds /* 2225909ccaaSLinus Torvalds * Do the basic POSIX ACL permission checks. 2231da177e4SLinus Torvalds */ 224b74c79e9SNick Piggin ret = acl_permission_check(inode, mask, flags, check_acl); 2255909ccaaSLinus Torvalds if (ret != -EACCES) 2265909ccaaSLinus Torvalds return ret; 2271da177e4SLinus Torvalds 2281da177e4SLinus Torvalds /* 2291da177e4SLinus Torvalds * Read/write DACs are always overridable. 2301da177e4SLinus Torvalds * Executable DACs are overridable if at least one exec bit is set. 2311da177e4SLinus Torvalds */ 232f696a365SMiklos Szeredi if (!(mask & MAY_EXEC) || execute_ok(inode)) 2331da177e4SLinus Torvalds if (capable(CAP_DAC_OVERRIDE)) 2341da177e4SLinus Torvalds return 0; 2351da177e4SLinus Torvalds 2361da177e4SLinus Torvalds /* 2371da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 2381da177e4SLinus Torvalds */ 2397ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 2401da177e4SLinus Torvalds if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) 2411da177e4SLinus Torvalds if (capable(CAP_DAC_READ_SEARCH)) 2421da177e4SLinus Torvalds return 0; 2431da177e4SLinus Torvalds 2441da177e4SLinus Torvalds return -EACCES; 2451da177e4SLinus Torvalds } 2461da177e4SLinus Torvalds 247cb23beb5SChristoph Hellwig /** 248cb23beb5SChristoph Hellwig * inode_permission - check for access rights to a given inode 249cb23beb5SChristoph Hellwig * @inode: inode to check permission on 250cb23beb5SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 251cb23beb5SChristoph Hellwig * 252cb23beb5SChristoph Hellwig * Used to check for read/write/execute permissions on an inode. 253cb23beb5SChristoph Hellwig * We use "fsuid" for this, letting us set arbitrary permissions 254cb23beb5SChristoph Hellwig * for filesystem access without changing the "normal" uids which 255cb23beb5SChristoph Hellwig * are used for other things. 256cb23beb5SChristoph Hellwig */ 257f419a2e3SAl Viro int inode_permission(struct inode *inode, int mask) 2581da177e4SLinus Torvalds { 259e6305c43SAl Viro int retval; 2601da177e4SLinus Torvalds 2611da177e4SLinus Torvalds if (mask & MAY_WRITE) { 26222590e41SMiklos Szeredi umode_t mode = inode->i_mode; 2631da177e4SLinus Torvalds 2641da177e4SLinus Torvalds /* 2651da177e4SLinus Torvalds * Nobody gets write access to a read-only fs. 2661da177e4SLinus Torvalds */ 2671da177e4SLinus Torvalds if (IS_RDONLY(inode) && 2681da177e4SLinus Torvalds (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 2691da177e4SLinus Torvalds return -EROFS; 2701da177e4SLinus Torvalds 2711da177e4SLinus Torvalds /* 2721da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 2731da177e4SLinus Torvalds */ 2741da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 2751da177e4SLinus Torvalds return -EACCES; 2761da177e4SLinus Torvalds } 2771da177e4SLinus Torvalds 278acfa4380SAl Viro if (inode->i_op->permission) 279b74c79e9SNick Piggin retval = inode->i_op->permission(inode, mask, 0); 280f696a365SMiklos Szeredi else 281b74c79e9SNick Piggin retval = generic_permission(inode, mask, 0, 282b74c79e9SNick Piggin inode->i_op->check_acl); 283f696a365SMiklos Szeredi 2841da177e4SLinus Torvalds if (retval) 2851da177e4SLinus Torvalds return retval; 2861da177e4SLinus Torvalds 28708ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 28808ce5f16SSerge E. Hallyn if (retval) 28908ce5f16SSerge E. Hallyn return retval; 29008ce5f16SSerge E. Hallyn 291d09ca739SEric Paris return security_inode_permission(inode, mask); 2921da177e4SLinus Torvalds } 2931da177e4SLinus Torvalds 294e4543eddSChristoph Hellwig /** 2958c744fb8SChristoph Hellwig * file_permission - check for additional access rights to a given file 2968c744fb8SChristoph Hellwig * @file: file to check access rights for 2978c744fb8SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 2988c744fb8SChristoph Hellwig * 2998c744fb8SChristoph Hellwig * Used to check for read/write/execute permissions on an already opened 3008c744fb8SChristoph Hellwig * file. 3018c744fb8SChristoph Hellwig * 3028c744fb8SChristoph Hellwig * Note: 3038c744fb8SChristoph Hellwig * Do not use this function in new code. All access checks should 304cb23beb5SChristoph Hellwig * be done using inode_permission(). 3058c744fb8SChristoph Hellwig */ 3068c744fb8SChristoph Hellwig int file_permission(struct file *file, int mask) 3078c744fb8SChristoph Hellwig { 308f419a2e3SAl Viro return inode_permission(file->f_path.dentry->d_inode, mask); 3098c744fb8SChristoph Hellwig } 3108c744fb8SChristoph Hellwig 3111da177e4SLinus Torvalds /* 3121da177e4SLinus Torvalds * get_write_access() gets write permission for a file. 3131da177e4SLinus Torvalds * put_write_access() releases this write permission. 3141da177e4SLinus Torvalds * This is used for regular files. 3151da177e4SLinus Torvalds * We cannot support write (and maybe mmap read-write shared) accesses and 3161da177e4SLinus Torvalds * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode 3171da177e4SLinus Torvalds * can have the following values: 3181da177e4SLinus Torvalds * 0: no writers, no VM_DENYWRITE mappings 3191da177e4SLinus Torvalds * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist 3201da177e4SLinus Torvalds * > 0: (i_writecount) users are writing to the file. 3211da177e4SLinus Torvalds * 3221da177e4SLinus Torvalds * Normally we operate on that counter with atomic_{inc,dec} and it's safe 3231da177e4SLinus Torvalds * except for the cases where we don't hold i_writecount yet. Then we need to 3241da177e4SLinus Torvalds * use {get,deny}_write_access() - these functions check the sign and refuse 3251da177e4SLinus Torvalds * to do the change if sign is wrong. Exclusion between them is provided by 3261da177e4SLinus Torvalds * the inode->i_lock spinlock. 3271da177e4SLinus Torvalds */ 3281da177e4SLinus Torvalds 3291da177e4SLinus Torvalds int get_write_access(struct inode * inode) 3301da177e4SLinus Torvalds { 3311da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3321da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) < 0) { 3331da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3341da177e4SLinus Torvalds return -ETXTBSY; 3351da177e4SLinus Torvalds } 3361da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 3371da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3381da177e4SLinus Torvalds 3391da177e4SLinus Torvalds return 0; 3401da177e4SLinus Torvalds } 3411da177e4SLinus Torvalds 3421da177e4SLinus Torvalds int deny_write_access(struct file * file) 3431da177e4SLinus Torvalds { 3440f7fc9e4SJosef "Jeff" Sipek struct inode *inode = file->f_path.dentry->d_inode; 3451da177e4SLinus Torvalds 3461da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3471da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) > 0) { 3481da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3491da177e4SLinus Torvalds return -ETXTBSY; 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds atomic_dec(&inode->i_writecount); 3521da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3531da177e4SLinus Torvalds 3541da177e4SLinus Torvalds return 0; 3551da177e4SLinus Torvalds } 3561da177e4SLinus Torvalds 3571d957f9bSJan Blunck /** 3585dd784d0SJan Blunck * path_get - get a reference to a path 3595dd784d0SJan Blunck * @path: path to get the reference to 3605dd784d0SJan Blunck * 3615dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3625dd784d0SJan Blunck */ 3635dd784d0SJan Blunck void path_get(struct path *path) 3645dd784d0SJan Blunck { 3655dd784d0SJan Blunck mntget(path->mnt); 3665dd784d0SJan Blunck dget(path->dentry); 3675dd784d0SJan Blunck } 3685dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 3695dd784d0SJan Blunck 3705dd784d0SJan Blunck /** 3711d957f9bSJan Blunck * path_put - put a reference to a path 3721d957f9bSJan Blunck * @path: path to put the reference to 3731d957f9bSJan Blunck * 3741d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 3751d957f9bSJan Blunck */ 3761d957f9bSJan Blunck void path_put(struct path *path) 3771da177e4SLinus Torvalds { 3781d957f9bSJan Blunck dput(path->dentry); 3791d957f9bSJan Blunck mntput(path->mnt); 3801da177e4SLinus Torvalds } 3811d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 3821da177e4SLinus Torvalds 383834f2a4aSTrond Myklebust /** 38431e6b01fSNick Piggin * nameidata_drop_rcu - drop this nameidata out of rcu-walk 38531e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 38639191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 38731e6b01fSNick Piggin * 38831e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 38931e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt 39031e6b01fSNick Piggin * to drop out of rcu-walk mode and take normal reference counts on dentries 39131e6b01fSNick Piggin * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take 39231e6b01fSNick Piggin * refcounts at the last known good point before rcu-walk got stuck, so 39331e6b01fSNick Piggin * ref-walk may continue from there. If this is not successful (eg. a seqcount 39431e6b01fSNick Piggin * has changed), then failure is returned and path walk restarts from the 39531e6b01fSNick Piggin * beginning in ref-walk mode. 39631e6b01fSNick Piggin * 39731e6b01fSNick Piggin * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into 39831e6b01fSNick Piggin * ref-walk. Must be called from rcu-walk context. 39931e6b01fSNick Piggin */ 40031e6b01fSNick Piggin static int nameidata_drop_rcu(struct nameidata *nd) 40131e6b01fSNick Piggin { 40231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 40331e6b01fSNick Piggin struct dentry *dentry = nd->path.dentry; 40431e6b01fSNick Piggin 40531e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 40631e6b01fSNick Piggin if (nd->root.mnt) { 40731e6b01fSNick Piggin spin_lock(&fs->lock); 40831e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 40931e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 41031e6b01fSNick Piggin goto err_root; 41131e6b01fSNick Piggin } 41231e6b01fSNick Piggin spin_lock(&dentry->d_lock); 41331e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 41431e6b01fSNick Piggin goto err; 41531e6b01fSNick Piggin BUG_ON(nd->inode != dentry->d_inode); 41631e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 41731e6b01fSNick Piggin if (nd->root.mnt) { 41831e6b01fSNick Piggin path_get(&nd->root); 41931e6b01fSNick Piggin spin_unlock(&fs->lock); 42031e6b01fSNick Piggin } 42131e6b01fSNick Piggin mntget(nd->path.mnt); 42231e6b01fSNick Piggin 42331e6b01fSNick Piggin rcu_read_unlock(); 42431e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 42531e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 42631e6b01fSNick Piggin return 0; 42731e6b01fSNick Piggin err: 42831e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 42931e6b01fSNick Piggin err_root: 43031e6b01fSNick Piggin if (nd->root.mnt) 43131e6b01fSNick Piggin spin_unlock(&fs->lock); 43231e6b01fSNick Piggin return -ECHILD; 43331e6b01fSNick Piggin } 43431e6b01fSNick Piggin 43531e6b01fSNick Piggin /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ 43631e6b01fSNick Piggin static inline int nameidata_drop_rcu_maybe(struct nameidata *nd) 43731e6b01fSNick Piggin { 43831e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) 43931e6b01fSNick Piggin return nameidata_drop_rcu(nd); 44031e6b01fSNick Piggin return 0; 44131e6b01fSNick Piggin } 44231e6b01fSNick Piggin 44331e6b01fSNick Piggin /** 44431e6b01fSNick Piggin * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk 44531e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 44631e6b01fSNick Piggin * @dentry: dentry to drop 44739191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 44831e6b01fSNick Piggin * 44931e6b01fSNick Piggin * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root, 45031e6b01fSNick Piggin * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on 45131e6b01fSNick Piggin * @nd. Must be called from rcu-walk context. 45231e6b01fSNick Piggin */ 45331e6b01fSNick Piggin static int nameidata_dentry_drop_rcu(struct nameidata *nd, struct dentry *dentry) 45431e6b01fSNick Piggin { 45531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 45631e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 45731e6b01fSNick Piggin 45831e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 45931e6b01fSNick Piggin if (nd->root.mnt) { 46031e6b01fSNick Piggin spin_lock(&fs->lock); 46131e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 46231e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 46331e6b01fSNick Piggin goto err_root; 46431e6b01fSNick Piggin } 46531e6b01fSNick Piggin spin_lock(&parent->d_lock); 46631e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 46731e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 46831e6b01fSNick Piggin goto err; 46931e6b01fSNick Piggin /* 47031e6b01fSNick Piggin * If the sequence check on the child dentry passed, then the child has 47131e6b01fSNick Piggin * not been removed from its parent. This means the parent dentry must 47231e6b01fSNick Piggin * be valid and able to take a reference at this point. 47331e6b01fSNick Piggin */ 47431e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 47531e6b01fSNick Piggin BUG_ON(!parent->d_count); 47631e6b01fSNick Piggin parent->d_count++; 47731e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 47831e6b01fSNick Piggin spin_unlock(&parent->d_lock); 47931e6b01fSNick Piggin if (nd->root.mnt) { 48031e6b01fSNick Piggin path_get(&nd->root); 48131e6b01fSNick Piggin spin_unlock(&fs->lock); 48231e6b01fSNick Piggin } 48331e6b01fSNick Piggin mntget(nd->path.mnt); 48431e6b01fSNick Piggin 48531e6b01fSNick Piggin rcu_read_unlock(); 48631e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 48731e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 48831e6b01fSNick Piggin return 0; 48931e6b01fSNick Piggin err: 49031e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 49131e6b01fSNick Piggin spin_unlock(&parent->d_lock); 49231e6b01fSNick Piggin err_root: 49331e6b01fSNick Piggin if (nd->root.mnt) 49431e6b01fSNick Piggin spin_unlock(&fs->lock); 49531e6b01fSNick Piggin return -ECHILD; 49631e6b01fSNick Piggin } 49731e6b01fSNick Piggin 49831e6b01fSNick Piggin /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */ 49931e6b01fSNick Piggin static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata *nd, struct dentry *dentry) 50031e6b01fSNick Piggin { 501a7472babSAl Viro if (nd->flags & LOOKUP_RCU) { 502a7472babSAl Viro if (unlikely(nameidata_dentry_drop_rcu(nd, dentry))) { 503a7472babSAl Viro nd->flags &= ~LOOKUP_RCU; 504a7472babSAl Viro nd->root.mnt = NULL; 505a7472babSAl Viro rcu_read_unlock(); 506a7472babSAl Viro br_read_unlock(vfsmount_lock); 507a7472babSAl Viro return -ECHILD; 508a7472babSAl Viro } 509a7472babSAl Viro } 51031e6b01fSNick Piggin return 0; 51131e6b01fSNick Piggin } 51231e6b01fSNick Piggin 51331e6b01fSNick Piggin /** 51431e6b01fSNick Piggin * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk 51531e6b01fSNick Piggin * @nd: nameidata pathwalk data to drop 51639191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 51731e6b01fSNick Piggin * 51831e6b01fSNick Piggin * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk. 51931e6b01fSNick Piggin * nd->path should be the final element of the lookup, so nd->root is discarded. 52031e6b01fSNick Piggin * Must be called from rcu-walk context. 52131e6b01fSNick Piggin */ 52231e6b01fSNick Piggin static int nameidata_drop_rcu_last(struct nameidata *nd) 52331e6b01fSNick Piggin { 52431e6b01fSNick Piggin struct dentry *dentry = nd->path.dentry; 52531e6b01fSNick Piggin 52631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 52731e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 52831e6b01fSNick Piggin nd->root.mnt = NULL; 52931e6b01fSNick Piggin spin_lock(&dentry->d_lock); 53031e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 53131e6b01fSNick Piggin goto err_unlock; 53231e6b01fSNick Piggin BUG_ON(nd->inode != dentry->d_inode); 53331e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 53431e6b01fSNick Piggin 53531e6b01fSNick Piggin mntget(nd->path.mnt); 53631e6b01fSNick Piggin 53731e6b01fSNick Piggin rcu_read_unlock(); 53831e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 53931e6b01fSNick Piggin 54031e6b01fSNick Piggin return 0; 54131e6b01fSNick Piggin 54231e6b01fSNick Piggin err_unlock: 54331e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 54431e6b01fSNick Piggin rcu_read_unlock(); 54531e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 54631e6b01fSNick Piggin return -ECHILD; 54731e6b01fSNick Piggin } 54831e6b01fSNick Piggin 54931e6b01fSNick Piggin /** 550834f2a4aSTrond Myklebust * release_open_intent - free up open intent resources 551834f2a4aSTrond Myklebust * @nd: pointer to nameidata 552834f2a4aSTrond Myklebust */ 553834f2a4aSTrond Myklebust void release_open_intent(struct nameidata *nd) 554834f2a4aSTrond Myklebust { 5552dab5974SLinus Torvalds struct file *file = nd->intent.open.file; 5562dab5974SLinus Torvalds 5572dab5974SLinus Torvalds if (file && !IS_ERR(file)) { 5582dab5974SLinus Torvalds if (file->f_path.dentry == NULL) 5592dab5974SLinus Torvalds put_filp(file); 560834f2a4aSTrond Myklebust else 5612dab5974SLinus Torvalds fput(file); 5622dab5974SLinus Torvalds } 563834f2a4aSTrond Myklebust } 564834f2a4aSTrond Myklebust 565f60aef7eSAl Viro static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) 56634286d66SNick Piggin { 567f60aef7eSAl Viro return dentry->d_op->d_revalidate(dentry, nd); 56834286d66SNick Piggin } 56934286d66SNick Piggin 570f5e1c1c1SAl Viro static struct dentry * 571bcdc5e01SIan Kent do_revalidate(struct dentry *dentry, struct nameidata *nd) 572bcdc5e01SIan Kent { 573f5e1c1c1SAl Viro int status = d_revalidate(dentry, nd); 574bcdc5e01SIan Kent if (unlikely(status <= 0)) { 575bcdc5e01SIan Kent /* 576bcdc5e01SIan Kent * The dentry failed validation. 577bcdc5e01SIan Kent * If d_revalidate returned 0 attempt to invalidate 578bcdc5e01SIan Kent * the dentry otherwise d_revalidate is asking us 579bcdc5e01SIan Kent * to return a fail status. 580bcdc5e01SIan Kent */ 58134286d66SNick Piggin if (status < 0) { 58234286d66SNick Piggin dput(dentry); 58334286d66SNick Piggin dentry = ERR_PTR(status); 584f5e1c1c1SAl Viro } else if (!d_invalidate(dentry)) { 585bcdc5e01SIan Kent dput(dentry); 586bcdc5e01SIan Kent dentry = NULL; 587bcdc5e01SIan Kent } 588bcdc5e01SIan Kent } 589f5e1c1c1SAl Viro return dentry; 590f5e1c1c1SAl Viro } 591f5e1c1c1SAl Viro 592f5e1c1c1SAl Viro static inline struct dentry * 593f5e1c1c1SAl Viro do_revalidate_rcu(struct dentry *dentry, struct nameidata *nd) 594f5e1c1c1SAl Viro { 595f60aef7eSAl Viro int status = d_revalidate(dentry, nd); 596f5e1c1c1SAl Viro if (likely(status > 0)) 597f5e1c1c1SAl Viro return dentry; 598f5e1c1c1SAl Viro if (status == -ECHILD) { 599f5e1c1c1SAl Viro if (nameidata_dentry_drop_rcu(nd, dentry)) 600f5e1c1c1SAl Viro return ERR_PTR(-ECHILD); 601f5e1c1c1SAl Viro return do_revalidate(dentry, nd); 602f5e1c1c1SAl Viro } 603f5e1c1c1SAl Viro if (status < 0) 604f5e1c1c1SAl Viro return ERR_PTR(status); 605f5e1c1c1SAl Viro /* Don't d_invalidate in rcu-walk mode */ 606f5e1c1c1SAl Viro if (nameidata_dentry_drop_rcu(nd, dentry)) 607f5e1c1c1SAl Viro return ERR_PTR(-ECHILD); 608f5e1c1c1SAl Viro if (!d_invalidate(dentry)) { 609f5e1c1c1SAl Viro dput(dentry); 610f5e1c1c1SAl Viro dentry = NULL; 611bcdc5e01SIan Kent } 612bcdc5e01SIan Kent return dentry; 613bcdc5e01SIan Kent } 614bcdc5e01SIan Kent 6151da177e4SLinus Torvalds /* 61616c2cd71SAl Viro * handle_reval_path - force revalidation of a dentry 61739159de2SJeff Layton * 61839159de2SJeff Layton * In some situations the path walking code will trust dentries without 61939159de2SJeff Layton * revalidating them. This causes problems for filesystems that depend on 62039159de2SJeff Layton * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set 62139159de2SJeff Layton * (which indicates that it's possible for the dentry to go stale), force 62239159de2SJeff Layton * a d_revalidate call before proceeding. 62339159de2SJeff Layton * 62439159de2SJeff Layton * Returns 0 if the revalidation was successful. If the revalidation fails, 62539159de2SJeff Layton * either return the error returned by d_revalidate or -ESTALE if the 62639159de2SJeff Layton * revalidation it just returned 0. If d_revalidate returns 0, we attempt to 62739159de2SJeff Layton * invalidate the dentry. It's up to the caller to handle putting references 62839159de2SJeff Layton * to the path if necessary. 62939159de2SJeff Layton */ 63016c2cd71SAl Viro static inline int handle_reval_path(struct nameidata *nd) 63139159de2SJeff Layton { 63216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 63339159de2SJeff Layton int status; 63439159de2SJeff Layton 63516c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 63639159de2SJeff Layton return 0; 63739159de2SJeff Layton 63816c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 63916c2cd71SAl Viro return 0; 64016c2cd71SAl Viro 64116c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 64216c2cd71SAl Viro return 0; 64316c2cd71SAl Viro 64416c2cd71SAl Viro /* Note: we do not d_invalidate() */ 64534286d66SNick Piggin status = d_revalidate(dentry, nd); 64639159de2SJeff Layton if (status > 0) 64739159de2SJeff Layton return 0; 64839159de2SJeff Layton 64916c2cd71SAl Viro if (!status) 65039159de2SJeff Layton status = -ESTALE; 65116c2cd71SAl Viro 65239159de2SJeff Layton return status; 65339159de2SJeff Layton } 65439159de2SJeff Layton 65539159de2SJeff Layton /* 656b75b5086SAl Viro * Short-cut version of permission(), for calling on directories 657b75b5086SAl Viro * during pathname resolution. Combines parts of permission() 658b75b5086SAl Viro * and generic_permission(), and tests ONLY for MAY_EXEC permission. 6591da177e4SLinus Torvalds * 6601da177e4SLinus Torvalds * If appropriate, check DAC only. If not appropriate, or 661b75b5086SAl Viro * short-cut DAC fails, then call ->permission() to do more 6621da177e4SLinus Torvalds * complete permission check. 6631da177e4SLinus Torvalds */ 664b74c79e9SNick Piggin static inline int exec_permission(struct inode *inode, unsigned int flags) 6651da177e4SLinus Torvalds { 6665909ccaaSLinus Torvalds int ret; 6671da177e4SLinus Torvalds 668cb9179eaSLinus Torvalds if (inode->i_op->permission) { 669b74c79e9SNick Piggin ret = inode->i_op->permission(inode, MAY_EXEC, flags); 670b74c79e9SNick Piggin } else { 671b74c79e9SNick Piggin ret = acl_permission_check(inode, MAY_EXEC, flags, 672b74c79e9SNick Piggin inode->i_op->check_acl); 673cb9179eaSLinus Torvalds } 674b74c79e9SNick Piggin if (likely(!ret)) 6751da177e4SLinus Torvalds goto ok; 676b74c79e9SNick Piggin if (ret == -ECHILD) 67731e6b01fSNick Piggin return ret; 6781da177e4SLinus Torvalds 679f1ac9f6bSLinus Torvalds if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) 6801da177e4SLinus Torvalds goto ok; 6811da177e4SLinus Torvalds 6825909ccaaSLinus Torvalds return ret; 6831da177e4SLinus Torvalds ok: 684b74c79e9SNick Piggin return security_inode_exec_permission(inode, flags); 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds 6872a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6882a737871SAl Viro { 689f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 690f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6912a737871SAl Viro } 6922a737871SAl Viro 6936de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6946de88d72SAl Viro 69531e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 69631e6b01fSNick Piggin { 69731e6b01fSNick Piggin if (!nd->root.mnt) { 69831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 699c28cc364SNick Piggin unsigned seq; 700c28cc364SNick Piggin 701c28cc364SNick Piggin do { 702c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 70331e6b01fSNick Piggin nd->root = fs->root; 704c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 70531e6b01fSNick Piggin } 70631e6b01fSNick Piggin } 70731e6b01fSNick Piggin 708f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 7091da177e4SLinus Torvalds { 71031e6b01fSNick Piggin int ret; 71131e6b01fSNick Piggin 7121da177e4SLinus Torvalds if (IS_ERR(link)) 7131da177e4SLinus Torvalds goto fail; 7141da177e4SLinus Torvalds 7151da177e4SLinus Torvalds if (*link == '/') { 7162a737871SAl Viro set_root(nd); 7171d957f9bSJan Blunck path_put(&nd->path); 7182a737871SAl Viro nd->path = nd->root; 7192a737871SAl Viro path_get(&nd->root); 72016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 7211da177e4SLinus Torvalds } 72231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 723b4091d5fSChristoph Hellwig 72431e6b01fSNick Piggin ret = link_path_walk(link, nd); 72531e6b01fSNick Piggin return ret; 7261da177e4SLinus Torvalds fail: 7271d957f9bSJan Blunck path_put(&nd->path); 7281da177e4SLinus Torvalds return PTR_ERR(link); 7291da177e4SLinus Torvalds } 7301da177e4SLinus Torvalds 7311d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 732051d3812SIan Kent { 733051d3812SIan Kent dput(path->dentry); 7344ac91378SJan Blunck if (path->mnt != nd->path.mnt) 735051d3812SIan Kent mntput(path->mnt); 736051d3812SIan Kent } 737051d3812SIan Kent 7387b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 7397b9337aaSNick Piggin struct nameidata *nd) 740051d3812SIan Kent { 74131e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 7424ac91378SJan Blunck dput(nd->path.dentry); 74331e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 7444ac91378SJan Blunck mntput(nd->path.mnt); 7459a229683SHuang Shijie } 74631e6b01fSNick Piggin nd->path.mnt = path->mnt; 7474ac91378SJan Blunck nd->path.dentry = path->dentry; 748051d3812SIan Kent } 749051d3812SIan Kent 750def4af30SAl Viro static __always_inline int 7517b9337aaSNick Piggin __do_follow_link(const struct path *link, struct nameidata *nd, void **p) 7521da177e4SLinus Torvalds { 7531da177e4SLinus Torvalds int error; 7547b9337aaSNick Piggin struct dentry *dentry = link->dentry; 7551da177e4SLinus Torvalds 756844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 757844a3917SAl Viro 7587b9337aaSNick Piggin touch_atime(link->mnt, dentry); 7591da177e4SLinus Torvalds nd_set_link(nd, NULL); 760cd4e91d3SAl Viro 76187556ef1SDavid Howells if (link->mnt == nd->path.mnt) 7627b9337aaSNick Piggin mntget(link->mnt); 76331e6b01fSNick Piggin 76436f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 76536f3b4f6SAl Viro if (error) { 76636f3b4f6SAl Viro *p = ERR_PTR(error); /* no ->put_link(), please */ 76736f3b4f6SAl Viro path_put(&nd->path); 76836f3b4f6SAl Viro return error; 76936f3b4f6SAl Viro } 77036f3b4f6SAl Viro 77186acdca1SAl Viro nd->last_type = LAST_BIND; 772def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 773def4af30SAl Viro error = PTR_ERR(*p); 774def4af30SAl Viro if (!IS_ERR(*p)) { 7751da177e4SLinus Torvalds char *s = nd_get_link(nd); 776cc314eefSLinus Torvalds error = 0; 7771da177e4SLinus Torvalds if (s) 7781da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 77916c2cd71SAl Viro else if (nd->last_type == LAST_BIND) 78016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 7811da177e4SLinus Torvalds } 7821da177e4SLinus Torvalds return error; 7831da177e4SLinus Torvalds } 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds /* 7861da177e4SLinus Torvalds * This limits recursive symlink follows to 8, while 7871da177e4SLinus Torvalds * limiting consecutive symlinks to 40. 7881da177e4SLinus Torvalds * 7891da177e4SLinus Torvalds * Without that kind of total limit, nasty chains of consecutive 7901da177e4SLinus Torvalds * symlinks can cause almost arbitrarily long lookups. 7911da177e4SLinus Torvalds */ 7923abb17e8SLinus Torvalds static inline int do_follow_link(struct inode *inode, struct path *path, struct nameidata *nd) 7931da177e4SLinus Torvalds { 794def4af30SAl Viro void *cookie; 7951da177e4SLinus Torvalds int err = -ELOOP; 796844a3917SAl Viro 797844a3917SAl Viro /* We drop rcu-walk here */ 798844a3917SAl Viro if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) 799844a3917SAl Viro return -ECHILD; 8003abb17e8SLinus Torvalds BUG_ON(inode != path->dentry->d_inode); 801844a3917SAl Viro 8021da177e4SLinus Torvalds if (current->link_count >= MAX_NESTED_LINKS) 8031da177e4SLinus Torvalds goto loop; 8041da177e4SLinus Torvalds if (current->total_link_count >= 40) 8051da177e4SLinus Torvalds goto loop; 8061da177e4SLinus Torvalds BUG_ON(nd->depth >= MAX_NESTED_LINKS); 8071da177e4SLinus Torvalds cond_resched(); 8081da177e4SLinus Torvalds current->link_count++; 8091da177e4SLinus Torvalds current->total_link_count++; 8101da177e4SLinus Torvalds nd->depth++; 811def4af30SAl Viro err = __do_follow_link(path, nd, &cookie); 812def4af30SAl Viro if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) 813def4af30SAl Viro path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); 814258fa999SAl Viro path_put(path); 8151da177e4SLinus Torvalds current->link_count--; 8161da177e4SLinus Torvalds nd->depth--; 8171da177e4SLinus Torvalds return err; 8181da177e4SLinus Torvalds loop: 8191d957f9bSJan Blunck path_put_conditional(path, nd); 8201d957f9bSJan Blunck path_put(&nd->path); 8211da177e4SLinus Torvalds return err; 8221da177e4SLinus Torvalds } 8231da177e4SLinus Torvalds 82431e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 82531e6b01fSNick Piggin { 82631e6b01fSNick Piggin struct vfsmount *parent; 82731e6b01fSNick Piggin struct dentry *mountpoint; 82831e6b01fSNick Piggin 82931e6b01fSNick Piggin parent = path->mnt->mnt_parent; 83031e6b01fSNick Piggin if (parent == path->mnt) 83131e6b01fSNick Piggin return 0; 83231e6b01fSNick Piggin mountpoint = path->mnt->mnt_mountpoint; 83331e6b01fSNick Piggin path->dentry = mountpoint; 83431e6b01fSNick Piggin path->mnt = parent; 83531e6b01fSNick Piggin return 1; 83631e6b01fSNick Piggin } 83731e6b01fSNick Piggin 838bab77ebfSAl Viro int follow_up(struct path *path) 8391da177e4SLinus Torvalds { 8401da177e4SLinus Torvalds struct vfsmount *parent; 8411da177e4SLinus Torvalds struct dentry *mountpoint; 84299b7db7bSNick Piggin 84399b7db7bSNick Piggin br_read_lock(vfsmount_lock); 844bab77ebfSAl Viro parent = path->mnt->mnt_parent; 845bab77ebfSAl Viro if (parent == path->mnt) { 84699b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 8471da177e4SLinus Torvalds return 0; 8481da177e4SLinus Torvalds } 8491da177e4SLinus Torvalds mntget(parent); 850bab77ebfSAl Viro mountpoint = dget(path->mnt->mnt_mountpoint); 85199b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 852bab77ebfSAl Viro dput(path->dentry); 853bab77ebfSAl Viro path->dentry = mountpoint; 854bab77ebfSAl Viro mntput(path->mnt); 855bab77ebfSAl Viro path->mnt = parent; 8561da177e4SLinus Torvalds return 1; 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds 859b5c84bf6SNick Piggin /* 8609875cf80SDavid Howells * Perform an automount 8619875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 8629875cf80SDavid Howells * were called with. 8631da177e4SLinus Torvalds */ 8649875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 8659875cf80SDavid Howells bool *need_mntput) 86631e6b01fSNick Piggin { 8679875cf80SDavid Howells struct vfsmount *mnt; 868ea5b778aSDavid Howells int err; 8699875cf80SDavid Howells 8709875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 8719875cf80SDavid Howells return -EREMOTE; 8729875cf80SDavid Howells 8736f45b656SDavid Howells /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT 8746f45b656SDavid Howells * and this is the terminal part of the path. 8756f45b656SDavid Howells */ 8766f45b656SDavid Howells if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_CONTINUE)) 8776f45b656SDavid Howells return -EISDIR; /* we actually want to stop here */ 8786f45b656SDavid Howells 8799875cf80SDavid Howells /* We want to mount if someone is trying to open/create a file of any 8809875cf80SDavid Howells * type under the mountpoint, wants to traverse through the mountpoint 8819875cf80SDavid Howells * or wants to open the mounted directory. 8829875cf80SDavid Howells * 8839875cf80SDavid Howells * We don't want to mount if someone's just doing a stat and they've 8849875cf80SDavid Howells * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and 8859875cf80SDavid Howells * appended a '/' to the name. 8869875cf80SDavid Howells */ 8879875cf80SDavid Howells if (!(flags & LOOKUP_FOLLOW) && 8889875cf80SDavid Howells !(flags & (LOOKUP_CONTINUE | LOOKUP_DIRECTORY | 8899875cf80SDavid Howells LOOKUP_OPEN | LOOKUP_CREATE))) 8909875cf80SDavid Howells return -EISDIR; 8919875cf80SDavid Howells 8929875cf80SDavid Howells current->total_link_count++; 8939875cf80SDavid Howells if (current->total_link_count >= 40) 8949875cf80SDavid Howells return -ELOOP; 8959875cf80SDavid Howells 8969875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 8979875cf80SDavid Howells if (IS_ERR(mnt)) { 8989875cf80SDavid Howells /* 8999875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 9009875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 9019875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 9029875cf80SDavid Howells * 9039875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 9049875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 9059875cf80SDavid Howells * the path is inaccessible and we should say so. 9069875cf80SDavid Howells */ 9079875cf80SDavid Howells if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_CONTINUE)) 9089875cf80SDavid Howells return -EREMOTE; 9099875cf80SDavid Howells return PTR_ERR(mnt); 91031e6b01fSNick Piggin } 911ea5b778aSDavid Howells 9129875cf80SDavid Howells if (!mnt) /* mount collision */ 9139875cf80SDavid Howells return 0; 9149875cf80SDavid Howells 91519a167afSAl Viro err = finish_automount(mnt, path); 916ea5b778aSDavid Howells 917ea5b778aSDavid Howells switch (err) { 918ea5b778aSDavid Howells case -EBUSY: 919ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 92019a167afSAl Viro return 0; 921ea5b778aSDavid Howells case 0: 922463ffb2eSAl Viro dput(path->dentry); 9239875cf80SDavid Howells if (*need_mntput) 9249875cf80SDavid Howells mntput(path->mnt); 9259875cf80SDavid Howells path->mnt = mnt; 9269875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 9279875cf80SDavid Howells *need_mntput = true; 9289875cf80SDavid Howells return 0; 92919a167afSAl Viro default: 93019a167afSAl Viro return err; 9319875cf80SDavid Howells } 93219a167afSAl Viro 933ea5b778aSDavid Howells } 9349875cf80SDavid Howells 9359875cf80SDavid Howells /* 9369875cf80SDavid Howells * Handle a dentry that is managed in some way. 937cc53ce53SDavid Howells * - Flagged for transit management (autofs) 9389875cf80SDavid Howells * - Flagged as mountpoint 9399875cf80SDavid Howells * - Flagged as automount point 9409875cf80SDavid Howells * 9419875cf80SDavid Howells * This may only be called in refwalk mode. 9429875cf80SDavid Howells * 9439875cf80SDavid Howells * Serialization is taken care of in namespace.c 9449875cf80SDavid Howells */ 9459875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 9469875cf80SDavid Howells { 9479875cf80SDavid Howells unsigned managed; 9489875cf80SDavid Howells bool need_mntput = false; 9499875cf80SDavid Howells int ret; 9509875cf80SDavid Howells 9519875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 9529875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 9539875cf80SDavid Howells * the components of that value change under us */ 9549875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 9559875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 9569875cf80SDavid Howells unlikely(managed != 0)) { 957cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 958cc53ce53SDavid Howells * being held. */ 959cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 960cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 961cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 962ab90911fSDavid Howells ret = path->dentry->d_op->d_manage(path->dentry, 963ab90911fSDavid Howells false, false); 964cc53ce53SDavid Howells if (ret < 0) 965cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 966cc53ce53SDavid Howells } 967cc53ce53SDavid Howells 9689875cf80SDavid Howells /* Transit to a mounted filesystem. */ 9699875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 9709875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 9719875cf80SDavid Howells if (mounted) { 9729875cf80SDavid Howells dput(path->dentry); 9739875cf80SDavid Howells if (need_mntput) 974463ffb2eSAl Viro mntput(path->mnt); 975463ffb2eSAl Viro path->mnt = mounted; 976463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 9779875cf80SDavid Howells need_mntput = true; 9789875cf80SDavid Howells continue; 979463ffb2eSAl Viro } 980463ffb2eSAl Viro 9819875cf80SDavid Howells /* Something is mounted on this dentry in another 9829875cf80SDavid Howells * namespace and/or whatever was mounted there in this 9839875cf80SDavid Howells * namespace got unmounted before we managed to get the 9849875cf80SDavid Howells * vfsmount_lock */ 9851da177e4SLinus Torvalds } 9869875cf80SDavid Howells 9879875cf80SDavid Howells /* Handle an automount point */ 9889875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 9899875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 9909875cf80SDavid Howells if (ret < 0) 9919875cf80SDavid Howells return ret == -EISDIR ? 0 : ret; 9929875cf80SDavid Howells continue; 9939875cf80SDavid Howells } 9949875cf80SDavid Howells 9959875cf80SDavid Howells /* We didn't change the current path point */ 9969875cf80SDavid Howells break; 9979875cf80SDavid Howells } 9989875cf80SDavid Howells return 0; 9991da177e4SLinus Torvalds } 10001da177e4SLinus Torvalds 1001cc53ce53SDavid Howells int follow_down_one(struct path *path) 10021da177e4SLinus Torvalds { 10031da177e4SLinus Torvalds struct vfsmount *mounted; 10041da177e4SLinus Torvalds 10051c755af4SAl Viro mounted = lookup_mnt(path); 10061da177e4SLinus Torvalds if (mounted) { 10079393bd07SAl Viro dput(path->dentry); 10089393bd07SAl Viro mntput(path->mnt); 10099393bd07SAl Viro path->mnt = mounted; 10109393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 10111da177e4SLinus Torvalds return 1; 10121da177e4SLinus Torvalds } 10131da177e4SLinus Torvalds return 0; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds 10169875cf80SDavid Howells /* 10179875cf80SDavid Howells * Skip to top of mountpoint pile in rcuwalk mode. We abort the rcu-walk if we 1018cc53ce53SDavid Howells * meet a managed dentry and we're not walking to "..". True is returned to 10199875cf80SDavid Howells * continue, false to abort. 10209875cf80SDavid Howells */ 10219875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 10229875cf80SDavid Howells struct inode **inode, bool reverse_transit) 10239875cf80SDavid Howells { 10249875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 10259875cf80SDavid Howells struct vfsmount *mounted; 1026ab90911fSDavid Howells if (unlikely(path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) && 1027ab90911fSDavid Howells !reverse_transit && 1028ab90911fSDavid Howells path->dentry->d_op->d_manage(path->dentry, false, true) < 0) 1029ab90911fSDavid Howells return false; 10309875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 10319875cf80SDavid Howells if (!mounted) 10329875cf80SDavid Howells break; 10339875cf80SDavid Howells path->mnt = mounted; 10349875cf80SDavid Howells path->dentry = mounted->mnt_root; 10359875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 10369875cf80SDavid Howells *inode = path->dentry->d_inode; 10379875cf80SDavid Howells } 10389875cf80SDavid Howells 10399875cf80SDavid Howells if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 10409875cf80SDavid Howells return reverse_transit; 10419875cf80SDavid Howells return true; 10429875cf80SDavid Howells } 10439875cf80SDavid Howells 104431e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 104531e6b01fSNick Piggin { 104631e6b01fSNick Piggin struct inode *inode = nd->inode; 104731e6b01fSNick Piggin 104831e6b01fSNick Piggin set_root_rcu(nd); 104931e6b01fSNick Piggin 105031e6b01fSNick Piggin while (1) { 105131e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 105231e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 105331e6b01fSNick Piggin break; 105431e6b01fSNick Piggin } 105531e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 105631e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 105731e6b01fSNick Piggin struct dentry *parent = old->d_parent; 105831e6b01fSNick Piggin unsigned seq; 105931e6b01fSNick Piggin 106031e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 106131e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1062ef7562d5SAl Viro goto failed; 106331e6b01fSNick Piggin inode = parent->d_inode; 106431e6b01fSNick Piggin nd->path.dentry = parent; 106531e6b01fSNick Piggin nd->seq = seq; 106631e6b01fSNick Piggin break; 106731e6b01fSNick Piggin } 106831e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 106931e6b01fSNick Piggin break; 107031e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 107131e6b01fSNick Piggin inode = nd->path.dentry->d_inode; 107231e6b01fSNick Piggin } 10739875cf80SDavid Howells __follow_mount_rcu(nd, &nd->path, &inode, true); 107431e6b01fSNick Piggin nd->inode = inode; 107531e6b01fSNick Piggin return 0; 1076ef7562d5SAl Viro 1077ef7562d5SAl Viro failed: 1078ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 1079ef7562d5SAl Viro nd->root.mnt = NULL; 1080ef7562d5SAl Viro rcu_read_unlock(); 1081ef7562d5SAl Viro br_read_unlock(vfsmount_lock); 1082ef7562d5SAl Viro return -ECHILD; 108331e6b01fSNick Piggin } 108431e6b01fSNick Piggin 10859875cf80SDavid Howells /* 1086cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1087cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1088cc53ce53SDavid Howells * caller is permitted to proceed or not. 1089cc53ce53SDavid Howells * 1090cc53ce53SDavid Howells * Care must be taken as namespace_sem may be held (indicated by mounting_here 1091cc53ce53SDavid Howells * being true). 1092cc53ce53SDavid Howells */ 1093cc53ce53SDavid Howells int follow_down(struct path *path, bool mounting_here) 1094cc53ce53SDavid Howells { 1095cc53ce53SDavid Howells unsigned managed; 1096cc53ce53SDavid Howells int ret; 1097cc53ce53SDavid Howells 1098cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1099cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1100cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1101cc53ce53SDavid Howells * being held. 1102cc53ce53SDavid Howells * 1103cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1104cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1105cc53ce53SDavid Howells * other than its daemon the right to mount on its 1106cc53ce53SDavid Howells * superstructure. 1107cc53ce53SDavid Howells * 1108cc53ce53SDavid Howells * The filesystem may sleep at this point. 1109cc53ce53SDavid Howells */ 1110cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1111cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1112cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1113ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 1114ab90911fSDavid Howells path->dentry, mounting_here, false); 1115cc53ce53SDavid Howells if (ret < 0) 1116cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1117cc53ce53SDavid Howells } 1118cc53ce53SDavid Howells 1119cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1120cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1121cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1122cc53ce53SDavid Howells if (!mounted) 1123cc53ce53SDavid Howells break; 1124cc53ce53SDavid Howells dput(path->dentry); 1125cc53ce53SDavid Howells mntput(path->mnt); 1126cc53ce53SDavid Howells path->mnt = mounted; 1127cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1128cc53ce53SDavid Howells continue; 1129cc53ce53SDavid Howells } 1130cc53ce53SDavid Howells 1131cc53ce53SDavid Howells /* Don't handle automount points here */ 1132cc53ce53SDavid Howells break; 1133cc53ce53SDavid Howells } 1134cc53ce53SDavid Howells return 0; 1135cc53ce53SDavid Howells } 1136cc53ce53SDavid Howells 1137cc53ce53SDavid Howells /* 11389875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 11399875cf80SDavid Howells */ 11409875cf80SDavid Howells static void follow_mount(struct path *path) 11419875cf80SDavid Howells { 11429875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 11439875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11449875cf80SDavid Howells if (!mounted) 11459875cf80SDavid Howells break; 11469875cf80SDavid Howells dput(path->dentry); 11479875cf80SDavid Howells mntput(path->mnt); 11489875cf80SDavid Howells path->mnt = mounted; 11499875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 11509875cf80SDavid Howells } 11519875cf80SDavid Howells } 11529875cf80SDavid Howells 115331e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 11541da177e4SLinus Torvalds { 11552a737871SAl Viro set_root(nd); 1156e518ddb7SAndreas Mohr 11571da177e4SLinus Torvalds while(1) { 11584ac91378SJan Blunck struct dentry *old = nd->path.dentry; 11591da177e4SLinus Torvalds 11602a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 11612a737871SAl Viro nd->path.mnt == nd->root.mnt) { 11621da177e4SLinus Torvalds break; 11631da177e4SLinus Torvalds } 11644ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 11653088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 11663088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 11671da177e4SLinus Torvalds dput(old); 11681da177e4SLinus Torvalds break; 11691da177e4SLinus Torvalds } 11703088dd70SAl Viro if (!follow_up(&nd->path)) 11711da177e4SLinus Torvalds break; 11721da177e4SLinus Torvalds } 117379ed0226SAl Viro follow_mount(&nd->path); 117431e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 11751da177e4SLinus Torvalds } 11761da177e4SLinus Torvalds 11771da177e4SLinus Torvalds /* 1178baa03890SNick Piggin * Allocate a dentry with name and parent, and perform a parent 1179baa03890SNick Piggin * directory ->lookup on it. Returns the new dentry, or ERR_PTR 1180baa03890SNick Piggin * on error. parent->d_inode->i_mutex must be held. d_lookup must 1181baa03890SNick Piggin * have verified that no child exists while under i_mutex. 1182baa03890SNick Piggin */ 1183baa03890SNick Piggin static struct dentry *d_alloc_and_lookup(struct dentry *parent, 1184baa03890SNick Piggin struct qstr *name, struct nameidata *nd) 1185baa03890SNick Piggin { 1186baa03890SNick Piggin struct inode *inode = parent->d_inode; 1187baa03890SNick Piggin struct dentry *dentry; 1188baa03890SNick Piggin struct dentry *old; 1189baa03890SNick Piggin 1190baa03890SNick Piggin /* Don't create child dentry for a dead directory. */ 1191baa03890SNick Piggin if (unlikely(IS_DEADDIR(inode))) 1192baa03890SNick Piggin return ERR_PTR(-ENOENT); 1193baa03890SNick Piggin 1194baa03890SNick Piggin dentry = d_alloc(parent, name); 1195baa03890SNick Piggin if (unlikely(!dentry)) 1196baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1197baa03890SNick Piggin 1198baa03890SNick Piggin old = inode->i_op->lookup(inode, dentry, nd); 1199baa03890SNick Piggin if (unlikely(old)) { 1200baa03890SNick Piggin dput(dentry); 1201baa03890SNick Piggin dentry = old; 1202baa03890SNick Piggin } 1203baa03890SNick Piggin return dentry; 1204baa03890SNick Piggin } 1205baa03890SNick Piggin 1206baa03890SNick Piggin /* 12071da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 12081da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 12091da177e4SLinus Torvalds * It _is_ time-critical. 12101da177e4SLinus Torvalds */ 12111da177e4SLinus Torvalds static int do_lookup(struct nameidata *nd, struct qstr *name, 121231e6b01fSNick Piggin struct path *path, struct inode **inode) 12131da177e4SLinus Torvalds { 12144ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 121531e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 12166e6b1bd1SAl Viro struct inode *dir; 12179875cf80SDavid Howells int err; 12189875cf80SDavid Howells 12193cac260aSAl Viro /* 1220b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1221b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1222b04f784eSNick Piggin * do the non-racy lookup, below. 1223b04f784eSNick Piggin */ 122431e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 122531e6b01fSNick Piggin unsigned seq; 122631e6b01fSNick Piggin 122731e6b01fSNick Piggin *inode = nd->inode; 122831e6b01fSNick Piggin dentry = __d_lookup_rcu(parent, name, &seq, inode); 122931e6b01fSNick Piggin if (!dentry) { 123031e6b01fSNick Piggin if (nameidata_drop_rcu(nd)) 123131e6b01fSNick Piggin return -ECHILD; 123231e6b01fSNick Piggin goto need_lookup; 123331e6b01fSNick Piggin } 123431e6b01fSNick Piggin /* Memory barrier in read_seqcount_begin of child is enough */ 123531e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 123631e6b01fSNick Piggin return -ECHILD; 123731e6b01fSNick Piggin 123831e6b01fSNick Piggin nd->seq = seq; 123924643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 1240f5e1c1c1SAl Viro dentry = do_revalidate_rcu(dentry, nd); 124124643087SAl Viro if (!dentry) 124224643087SAl Viro goto need_lookup; 124324643087SAl Viro if (IS_ERR(dentry)) 124424643087SAl Viro goto fail; 124524643087SAl Viro if (!(nd->flags & LOOKUP_RCU)) 124624643087SAl Viro goto done; 124724643087SAl Viro } 124831e6b01fSNick Piggin path->mnt = mnt; 124931e6b01fSNick Piggin path->dentry = dentry; 12509875cf80SDavid Howells if (likely(__follow_mount_rcu(nd, path, inode, false))) 12519875cf80SDavid Howells return 0; 12529875cf80SDavid Howells if (nameidata_drop_rcu(nd)) 12539875cf80SDavid Howells return -ECHILD; 12549875cf80SDavid Howells /* fallthru */ 12559875cf80SDavid Howells } 125631e6b01fSNick Piggin dentry = __d_lookup(parent, name); 12571da177e4SLinus Torvalds if (!dentry) 12581da177e4SLinus Torvalds goto need_lookup; 12592e2e88eaSNick Piggin found: 126024643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 126124643087SAl Viro dentry = do_revalidate(dentry, nd); 126224643087SAl Viro if (!dentry) 126324643087SAl Viro goto need_lookup; 126424643087SAl Viro if (IS_ERR(dentry)) 126524643087SAl Viro goto fail; 126624643087SAl Viro } 12671da177e4SLinus Torvalds done: 12681da177e4SLinus Torvalds path->mnt = mnt; 12691da177e4SLinus Torvalds path->dentry = dentry; 12709875cf80SDavid Howells err = follow_managed(path, nd->flags); 127189312214SIan Kent if (unlikely(err < 0)) { 127289312214SIan Kent path_put_conditional(path, nd); 12739875cf80SDavid Howells return err; 127489312214SIan Kent } 127531e6b01fSNick Piggin *inode = path->dentry->d_inode; 12761da177e4SLinus Torvalds return 0; 12771da177e4SLinus Torvalds 12781da177e4SLinus Torvalds need_lookup: 12796e6b1bd1SAl Viro dir = parent->d_inode; 128031e6b01fSNick Piggin BUG_ON(nd->inode != dir); 12816e6b1bd1SAl Viro 12826e6b1bd1SAl Viro mutex_lock(&dir->i_mutex); 12836e6b1bd1SAl Viro /* 12846e6b1bd1SAl Viro * First re-do the cached lookup just in case it was created 1285b04f784eSNick Piggin * while we waited for the directory semaphore, or the first 1286b04f784eSNick Piggin * lookup failed due to an unrelated rename. 12876e6b1bd1SAl Viro * 1288b04f784eSNick Piggin * This could use version numbering or similar to avoid unnecessary 1289b04f784eSNick Piggin * cache lookups, but then we'd have to do the first lookup in the 1290b04f784eSNick Piggin * non-racy way. However in the common case here, everything should 1291b04f784eSNick Piggin * be hot in cache, so would it be a big win? 12926e6b1bd1SAl Viro */ 12936e6b1bd1SAl Viro dentry = d_lookup(parent, name); 1294baa03890SNick Piggin if (likely(!dentry)) { 1295baa03890SNick Piggin dentry = d_alloc_and_lookup(parent, name, nd); 12966e6b1bd1SAl Viro mutex_unlock(&dir->i_mutex); 12976e6b1bd1SAl Viro if (IS_ERR(dentry)) 12986e6b1bd1SAl Viro goto fail; 12996e6b1bd1SAl Viro goto done; 13006e6b1bd1SAl Viro } 13016e6b1bd1SAl Viro /* 13026e6b1bd1SAl Viro * Uhhuh! Nasty case: the cache was re-populated while 13036e6b1bd1SAl Viro * we waited on the semaphore. Need to revalidate. 13046e6b1bd1SAl Viro */ 13056e6b1bd1SAl Viro mutex_unlock(&dir->i_mutex); 13062e2e88eaSNick Piggin goto found; 13071da177e4SLinus Torvalds 13081da177e4SLinus Torvalds fail: 13091da177e4SLinus Torvalds return PTR_ERR(dentry); 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds 131252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 131352094c8aSAl Viro { 131452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 131552094c8aSAl Viro int err = exec_permission(nd->inode, IPERM_FLAG_RCU); 131652094c8aSAl Viro if (err != -ECHILD) 131752094c8aSAl Viro return err; 131852094c8aSAl Viro if (nameidata_drop_rcu(nd)) 131952094c8aSAl Viro return -ECHILD; 132052094c8aSAl Viro } 132152094c8aSAl Viro return exec_permission(nd->inode, 0); 132252094c8aSAl Viro } 132352094c8aSAl Viro 13249856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 13259856fa1bSAl Viro { 13269856fa1bSAl Viro if (type == LAST_DOTDOT) { 13279856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 13289856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 13299856fa1bSAl Viro return -ECHILD; 13309856fa1bSAl Viro } else 13319856fa1bSAl Viro follow_dotdot(nd); 13329856fa1bSAl Viro } 13339856fa1bSAl Viro return 0; 13349856fa1bSAl Viro } 13359856fa1bSAl Viro 1336951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1337951361f9SAl Viro { 1338951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1339951361f9SAl Viro path_put(&nd->path); 1340951361f9SAl Viro } else { 1341951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 1342951361f9SAl Viro nd->root.mnt = NULL; 1343951361f9SAl Viro rcu_read_unlock(); 1344951361f9SAl Viro br_read_unlock(vfsmount_lock); 1345951361f9SAl Viro } 1346951361f9SAl Viro } 1347951361f9SAl Viro 13481da177e4SLinus Torvalds /* 13491da177e4SLinus Torvalds * Name resolution. 1350ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1351ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 13521da177e4SLinus Torvalds * 1353ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1354ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 13551da177e4SLinus Torvalds */ 13566de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 13571da177e4SLinus Torvalds { 13581da177e4SLinus Torvalds struct path next; 13591da177e4SLinus Torvalds int err; 13601da177e4SLinus Torvalds unsigned int lookup_flags = nd->flags; 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds while (*name=='/') 13631da177e4SLinus Torvalds name++; 13641da177e4SLinus Torvalds if (!*name) 1365086e183aSAl Viro return 0; 13661da177e4SLinus Torvalds 13671da177e4SLinus Torvalds if (nd->depth) 1368f55eab82STrond Myklebust lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 13711da177e4SLinus Torvalds for(;;) { 137231e6b01fSNick Piggin struct inode *inode; 13731da177e4SLinus Torvalds unsigned long hash; 13741da177e4SLinus Torvalds struct qstr this; 13751da177e4SLinus Torvalds unsigned int c; 1376fe479a58SAl Viro int type; 13771da177e4SLinus Torvalds 1378cdce5d6bSTrond Myklebust nd->flags |= LOOKUP_CONTINUE; 137952094c8aSAl Viro 138052094c8aSAl Viro err = may_lookup(nd); 13811da177e4SLinus Torvalds if (err) 13821da177e4SLinus Torvalds break; 13831da177e4SLinus Torvalds 13841da177e4SLinus Torvalds this.name = name; 13851da177e4SLinus Torvalds c = *(const unsigned char *)name; 13861da177e4SLinus Torvalds 13871da177e4SLinus Torvalds hash = init_name_hash(); 13881da177e4SLinus Torvalds do { 13891da177e4SLinus Torvalds name++; 13901da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 13911da177e4SLinus Torvalds c = *(const unsigned char *)name; 13921da177e4SLinus Torvalds } while (c && (c != '/')); 13931da177e4SLinus Torvalds this.len = name - (const char *) this.name; 13941da177e4SLinus Torvalds this.hash = end_name_hash(hash); 13951da177e4SLinus Torvalds 1396fe479a58SAl Viro type = LAST_NORM; 1397fe479a58SAl Viro if (this.name[0] == '.') switch (this.len) { 1398fe479a58SAl Viro case 2: 139916c2cd71SAl Viro if (this.name[1] == '.') { 1400fe479a58SAl Viro type = LAST_DOTDOT; 140116c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 140216c2cd71SAl Viro } 1403fe479a58SAl Viro break; 1404fe479a58SAl Viro case 1: 1405fe479a58SAl Viro type = LAST_DOT; 1406fe479a58SAl Viro } 14075a202bcdSAl Viro if (likely(type == LAST_NORM)) { 14085a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 140916c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 14105a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 14115a202bcdSAl Viro err = parent->d_op->d_hash(parent, nd->inode, 14125a202bcdSAl Viro &this); 14135a202bcdSAl Viro if (err < 0) 14145a202bcdSAl Viro break; 14155a202bcdSAl Viro } 14165a202bcdSAl Viro } 1417fe479a58SAl Viro 14181da177e4SLinus Torvalds /* remove trailing slashes? */ 14191da177e4SLinus Torvalds if (!c) 14201da177e4SLinus Torvalds goto last_component; 14211da177e4SLinus Torvalds while (*++name == '/'); 14221da177e4SLinus Torvalds if (!*name) 14231da177e4SLinus Torvalds goto last_with_slashes; 14241da177e4SLinus Torvalds 14251da177e4SLinus Torvalds /* 14261da177e4SLinus Torvalds * "." and ".." are special - ".." especially so because it has 14271da177e4SLinus Torvalds * to be able to know about the current root directory and 14281da177e4SLinus Torvalds * parent relationships. 14291da177e4SLinus Torvalds */ 1430fe479a58SAl Viro if (unlikely(type != LAST_NORM)) { 1431ef7562d5SAl Viro if (handle_dots(nd, type)) 1432ef7562d5SAl Viro return -ECHILD; 14331da177e4SLinus Torvalds continue; 14341da177e4SLinus Torvalds } 1435fe479a58SAl Viro 14361da177e4SLinus Torvalds /* This does the actual lookups.. */ 143731e6b01fSNick Piggin err = do_lookup(nd, &this, &next, &inode); 14381da177e4SLinus Torvalds if (err) 14391da177e4SLinus Torvalds break; 14401da177e4SLinus Torvalds 14417bc055d1SAl Viro if (inode && inode->i_op->follow_link) { 14423abb17e8SLinus Torvalds err = do_follow_link(inode, &next, nd); 14431da177e4SLinus Torvalds if (err) 1444a7472babSAl Viro return err; 144531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 144631e6b01fSNick Piggin } else { 144709dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 144831e6b01fSNick Piggin nd->inode = inode; 144931e6b01fSNick Piggin } 14507bc055d1SAl Viro err = -ENOENT; 14517bc055d1SAl Viro if (!nd->inode) 14527bc055d1SAl Viro break; 14531da177e4SLinus Torvalds err = -ENOTDIR; 145431e6b01fSNick Piggin if (!nd->inode->i_op->lookup) 14551da177e4SLinus Torvalds break; 14561da177e4SLinus Torvalds continue; 14571da177e4SLinus Torvalds /* here ends the main loop */ 14581da177e4SLinus Torvalds 14591da177e4SLinus Torvalds last_with_slashes: 14601da177e4SLinus Torvalds lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 14611da177e4SLinus Torvalds last_component: 1462f55eab82STrond Myklebust /* Clear LOOKUP_CONTINUE iff it was previously unset */ 1463f55eab82STrond Myklebust nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; 14641da177e4SLinus Torvalds if (lookup_flags & LOOKUP_PARENT) 14651da177e4SLinus Torvalds goto lookup_parent; 1466ef7562d5SAl Viro if (unlikely(type != LAST_NORM)) 1467ef7562d5SAl Viro return handle_dots(nd, type); 146831e6b01fSNick Piggin err = do_lookup(nd, &this, &next, &inode); 14691da177e4SLinus Torvalds if (err) 14701da177e4SLinus Torvalds break; 1471db372915SDavid Howells if (inode && unlikely(inode->i_op->follow_link) && 1472db372915SDavid Howells (lookup_flags & LOOKUP_FOLLOW)) { 14733abb17e8SLinus Torvalds err = do_follow_link(inode, &next, nd); 14741da177e4SLinus Torvalds if (err) 1475a7472babSAl Viro return err; 147631e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 147731e6b01fSNick Piggin } else { 147809dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 147931e6b01fSNick Piggin nd->inode = inode; 148031e6b01fSNick Piggin } 14811da177e4SLinus Torvalds err = -ENOENT; 148231e6b01fSNick Piggin if (!nd->inode) 14831da177e4SLinus Torvalds break; 14841da177e4SLinus Torvalds if (lookup_flags & LOOKUP_DIRECTORY) { 14851da177e4SLinus Torvalds err = -ENOTDIR; 148631e6b01fSNick Piggin if (!nd->inode->i_op->lookup) 14871da177e4SLinus Torvalds break; 14881da177e4SLinus Torvalds } 1489086e183aSAl Viro return 0; 14901da177e4SLinus Torvalds lookup_parent: 14911da177e4SLinus Torvalds nd->last = this; 1492fe479a58SAl Viro nd->last_type = type; 14931da177e4SLinus Torvalds return 0; 14941da177e4SLinus Torvalds } 1495951361f9SAl Viro terminate_walk(nd); 14961da177e4SLinus Torvalds return err; 14971da177e4SLinus Torvalds } 14981da177e4SLinus Torvalds 149970e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 150070e9b357SAl Viro struct nameidata *nd, struct file **fp) 150131e6b01fSNick Piggin { 150231e6b01fSNick Piggin int retval = 0; 150331e6b01fSNick Piggin int fput_needed; 150431e6b01fSNick Piggin struct file *file; 150531e6b01fSNick Piggin 150631e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 150716c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 150831e6b01fSNick Piggin nd->depth = 0; 150931e6b01fSNick Piggin nd->root.mnt = NULL; 151031e6b01fSNick Piggin 151131e6b01fSNick Piggin if (*name=='/') { 1512e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 151331e6b01fSNick Piggin br_read_lock(vfsmount_lock); 151431e6b01fSNick Piggin rcu_read_lock(); 1515e41f7d4eSAl Viro set_root_rcu(nd); 1516e41f7d4eSAl Viro } else { 1517e41f7d4eSAl Viro set_root(nd); 1518e41f7d4eSAl Viro path_get(&nd->root); 1519e41f7d4eSAl Viro } 152031e6b01fSNick Piggin nd->path = nd->root; 152131e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1522e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 152331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1524c28cc364SNick Piggin unsigned seq; 152531e6b01fSNick Piggin 152631e6b01fSNick Piggin br_read_lock(vfsmount_lock); 152731e6b01fSNick Piggin rcu_read_lock(); 152831e6b01fSNick Piggin 1529c28cc364SNick Piggin do { 1530c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 153131e6b01fSNick Piggin nd->path = fs->pwd; 1532c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1533c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1534e41f7d4eSAl Viro } else { 1535e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1536e41f7d4eSAl Viro } 153731e6b01fSNick Piggin } else { 153831e6b01fSNick Piggin struct dentry *dentry; 153931e6b01fSNick Piggin 154031e6b01fSNick Piggin file = fget_light(dfd, &fput_needed); 154131e6b01fSNick Piggin retval = -EBADF; 154231e6b01fSNick Piggin if (!file) 154331e6b01fSNick Piggin goto out_fail; 154431e6b01fSNick Piggin 154531e6b01fSNick Piggin dentry = file->f_path.dentry; 154631e6b01fSNick Piggin 154731e6b01fSNick Piggin retval = -ENOTDIR; 154831e6b01fSNick Piggin if (!S_ISDIR(dentry->d_inode->i_mode)) 154931e6b01fSNick Piggin goto fput_fail; 155031e6b01fSNick Piggin 155131e6b01fSNick Piggin retval = file_permission(file, MAY_EXEC); 155231e6b01fSNick Piggin if (retval) 155331e6b01fSNick Piggin goto fput_fail; 155431e6b01fSNick Piggin 155531e6b01fSNick Piggin nd->path = file->f_path; 1556e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 155731e6b01fSNick Piggin if (fput_needed) 155870e9b357SAl Viro *fp = file; 1559c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 156031e6b01fSNick Piggin br_read_lock(vfsmount_lock); 156131e6b01fSNick Piggin rcu_read_lock(); 15625590ff0dSUlrich Drepper } else { 15635dd784d0SJan Blunck path_get(&file->f_path); 15645590ff0dSUlrich Drepper fput_light(file, fput_needed); 15651da177e4SLinus Torvalds } 1566e41f7d4eSAl Viro } 1567e41f7d4eSAl Viro 156831e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 15699b4a9b14SAl Viro return 0; 15702dfdd266SJosef 'Jeff' Sipek 15719b4a9b14SAl Viro fput_fail: 15729b4a9b14SAl Viro fput_light(file, fput_needed); 15739b4a9b14SAl Viro out_fail: 15749b4a9b14SAl Viro return retval; 15759b4a9b14SAl Viro } 15769b4a9b14SAl Viro 15779b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1578ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 15799b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 15809b4a9b14SAl Viro { 158170e9b357SAl Viro struct file *base = NULL; 158231e6b01fSNick Piggin int retval; 158331e6b01fSNick Piggin 158431e6b01fSNick Piggin /* 158531e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 158631e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 158731e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 158831e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 158931e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 159031e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 159131e6b01fSNick Piggin * analogue, foo_rcu(). 159231e6b01fSNick Piggin * 159331e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 159431e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 159531e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 159631e6b01fSNick Piggin * be able to complete). 159731e6b01fSNick Piggin */ 159870e9b357SAl Viro retval = path_init(dfd, name, flags, nd, &base); 1599ee0827cdSAl Viro 160031e6b01fSNick Piggin if (unlikely(retval)) 160131e6b01fSNick Piggin return retval; 1602ee0827cdSAl Viro 1603ee0827cdSAl Viro current->total_link_count = 0; 1604ee0827cdSAl Viro retval = link_path_walk(name, nd); 1605ee0827cdSAl Viro 1606ee0827cdSAl Viro if (nd->flags & LOOKUP_RCU) { 16074455ca62SAl Viro /* went all way through without dropping RCU */ 16084455ca62SAl Viro BUG_ON(retval); 1609086e183aSAl Viro if (nameidata_drop_rcu_last(nd)) 1610086e183aSAl Viro retval = -ECHILD; 1611086e183aSAl Viro } 161231e6b01fSNick Piggin 161316c2cd71SAl Viro if (!retval) 161416c2cd71SAl Viro retval = handle_reval_path(nd); 161516c2cd71SAl Viro 161670e9b357SAl Viro if (base) 161770e9b357SAl Viro fput(base); 1618ee0827cdSAl Viro 161931e6b01fSNick Piggin if (nd->root.mnt) { 162031e6b01fSNick Piggin path_put(&nd->root); 162131e6b01fSNick Piggin nd->root.mnt = NULL; 162231e6b01fSNick Piggin } 1623ee0827cdSAl Viro return retval; 162431e6b01fSNick Piggin } 162531e6b01fSNick Piggin 1626ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1627ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1628ee0827cdSAl Viro { 1629ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1630ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1631ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1632ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1633ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1634ee0827cdSAl Viro 163531e6b01fSNick Piggin if (likely(!retval)) { 163631e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 163731e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 163831e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 163931e6b01fSNick Piggin } 164031e6b01fSNick Piggin } 1641170aa3d0SUlrich Drepper return retval; 16421da177e4SLinus Torvalds } 16431da177e4SLinus Torvalds 1644c9c6cac0SAl Viro int kern_path_parent(const char *name, struct nameidata *nd) 16455590ff0dSUlrich Drepper { 1646c9c6cac0SAl Viro return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); 16475590ff0dSUlrich Drepper } 16485590ff0dSUlrich Drepper 1649d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1650d1811465SAl Viro { 1651d1811465SAl Viro struct nameidata nd; 1652d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1653d1811465SAl Viro if (!res) 1654d1811465SAl Viro *path = nd.path; 1655d1811465SAl Viro return res; 1656d1811465SAl Viro } 1657d1811465SAl Viro 165816f18200SJosef 'Jeff' Sipek /** 165916f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 166016f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 166116f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 166216f18200SJosef 'Jeff' Sipek * @name: pointer to file name 166316f18200SJosef 'Jeff' Sipek * @flags: lookup flags 166416f18200SJosef 'Jeff' Sipek * @nd: pointer to nameidata 166516f18200SJosef 'Jeff' Sipek */ 166616f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 166716f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 166816f18200SJosef 'Jeff' Sipek struct nameidata *nd) 166916f18200SJosef 'Jeff' Sipek { 1670ee0827cdSAl Viro int result; 167116f18200SJosef 'Jeff' Sipek 167216f18200SJosef 'Jeff' Sipek /* same as do_path_lookup */ 167316f18200SJosef 'Jeff' Sipek nd->last_type = LAST_ROOT; 167416c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 167516f18200SJosef 'Jeff' Sipek nd->depth = 0; 167616f18200SJosef 'Jeff' Sipek 1677c8e7f449SJan Blunck nd->path.dentry = dentry; 1678c8e7f449SJan Blunck nd->path.mnt = mnt; 1679c8e7f449SJan Blunck path_get(&nd->path); 16805b857119SAl Viro nd->root = nd->path; 16815b857119SAl Viro path_get(&nd->root); 168231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 168316f18200SJosef 'Jeff' Sipek 1684ee0827cdSAl Viro current->total_link_count = 0; 1685ee0827cdSAl Viro 1686ee0827cdSAl Viro result = link_path_walk(name, nd); 168716c2cd71SAl Viro if (!result) 168816c2cd71SAl Viro result = handle_reval_path(nd); 1689ee0827cdSAl Viro if (result == -ESTALE) { 1690ee0827cdSAl Viro /* nd->path had been dropped */ 1691ee0827cdSAl Viro current->total_link_count = 0; 1692ee0827cdSAl Viro nd->path.dentry = dentry; 1693ee0827cdSAl Viro nd->path.mnt = mnt; 1694ee0827cdSAl Viro nd->inode = dentry->d_inode; 1695ee0827cdSAl Viro path_get(&nd->path); 169616c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_REVAL; 169716c2cd71SAl Viro 1698ee0827cdSAl Viro result = link_path_walk(name, nd); 169916c2cd71SAl Viro if (!result) 170016c2cd71SAl Viro result = handle_reval_path(nd); 1701ee0827cdSAl Viro } 1702ee0827cdSAl Viro if (unlikely(!result && !audit_dummy_context() && nd->path.dentry && 170331e6b01fSNick Piggin nd->inode)) 17044ac91378SJan Blunck audit_inode(name, nd->path.dentry); 170516f18200SJosef 'Jeff' Sipek 17062a737871SAl Viro path_put(&nd->root); 17072a737871SAl Viro nd->root.mnt = NULL; 170816f18200SJosef 'Jeff' Sipek 1709ee0827cdSAl Viro return result; 171016f18200SJosef 'Jeff' Sipek } 171116f18200SJosef 'Jeff' Sipek 1712eead1911SChristoph Hellwig static struct dentry *__lookup_hash(struct qstr *name, 1713eead1911SChristoph Hellwig struct dentry *base, struct nameidata *nd) 17141da177e4SLinus Torvalds { 171581fca444SChristoph Hellwig struct inode *inode = base->d_inode; 17161da177e4SLinus Torvalds struct dentry *dentry; 17171da177e4SLinus Torvalds int err; 17181da177e4SLinus Torvalds 1719b74c79e9SNick Piggin err = exec_permission(inode, 0); 172081fca444SChristoph Hellwig if (err) 172181fca444SChristoph Hellwig return ERR_PTR(err); 17221da177e4SLinus Torvalds 17231da177e4SLinus Torvalds /* 1724b04f784eSNick Piggin * Don't bother with __d_lookup: callers are for creat as 1725b04f784eSNick Piggin * well as unlink, so a lot of the time it would cost 1726b04f784eSNick Piggin * a double lookup. 17276e6b1bd1SAl Viro */ 17286e6b1bd1SAl Viro dentry = d_lookup(base, name); 17296e6b1bd1SAl Viro 1730fb045adbSNick Piggin if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) 17316e6b1bd1SAl Viro dentry = do_revalidate(dentry, nd); 17326e6b1bd1SAl Viro 17331da177e4SLinus Torvalds if (!dentry) 1734baa03890SNick Piggin dentry = d_alloc_and_lookup(base, name, nd); 17355a202bcdSAl Viro 17361da177e4SLinus Torvalds return dentry; 17371da177e4SLinus Torvalds } 17381da177e4SLinus Torvalds 1739057f6c01SJames Morris /* 1740057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1741057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1742057f6c01SJames Morris * SMP-safe. 1743057f6c01SJames Morris */ 1744a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 17451da177e4SLinus Torvalds { 17464ac91378SJan Blunck return __lookup_hash(&nd->last, nd->path.dentry, nd); 17471da177e4SLinus Torvalds } 17481da177e4SLinus Torvalds 1749eead1911SChristoph Hellwig /** 1750a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1751eead1911SChristoph Hellwig * @name: pathname component to lookup 1752eead1911SChristoph Hellwig * @base: base directory to lookup from 1753eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1754eead1911SChristoph Hellwig * 1755a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1756a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1757eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1758eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1759eead1911SChristoph Hellwig */ 1760057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1761057f6c01SJames Morris { 1762057f6c01SJames Morris struct qstr this; 17636a96ba54SAl Viro unsigned long hash; 17646a96ba54SAl Viro unsigned int c; 1765057f6c01SJames Morris 17662f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 17672f9092e1SDavid Woodhouse 17686a96ba54SAl Viro this.name = name; 17696a96ba54SAl Viro this.len = len; 17706a96ba54SAl Viro if (!len) 17716a96ba54SAl Viro return ERR_PTR(-EACCES); 17726a96ba54SAl Viro 17736a96ba54SAl Viro hash = init_name_hash(); 17746a96ba54SAl Viro while (len--) { 17756a96ba54SAl Viro c = *(const unsigned char *)name++; 17766a96ba54SAl Viro if (c == '/' || c == '\0') 17776a96ba54SAl Viro return ERR_PTR(-EACCES); 17786a96ba54SAl Viro hash = partial_name_hash(c, hash); 17796a96ba54SAl Viro } 17806a96ba54SAl Viro this.hash = end_name_hash(hash); 17815a202bcdSAl Viro /* 17825a202bcdSAl Viro * See if the low-level filesystem might want 17835a202bcdSAl Viro * to use its own hash.. 17845a202bcdSAl Viro */ 17855a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 17865a202bcdSAl Viro int err = base->d_op->d_hash(base, base->d_inode, &this); 17875a202bcdSAl Viro if (err < 0) 17885a202bcdSAl Viro return ERR_PTR(err); 17895a202bcdSAl Viro } 1790eead1911SChristoph Hellwig 179149705b77SChristoph Hellwig return __lookup_hash(&this, base, NULL); 1792057f6c01SJames Morris } 1793057f6c01SJames Morris 17942d8f3038SAl Viro int user_path_at(int dfd, const char __user *name, unsigned flags, 17952d8f3038SAl Viro struct path *path) 17961da177e4SLinus Torvalds { 17972d8f3038SAl Viro struct nameidata nd; 17981da177e4SLinus Torvalds char *tmp = getname(name); 17991da177e4SLinus Torvalds int err = PTR_ERR(tmp); 18001da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 18012d8f3038SAl Viro 18022d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 18032d8f3038SAl Viro 18042d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 18051da177e4SLinus Torvalds putname(tmp); 18062d8f3038SAl Viro if (!err) 18072d8f3038SAl Viro *path = nd.path; 18081da177e4SLinus Torvalds } 18091da177e4SLinus Torvalds return err; 18101da177e4SLinus Torvalds } 18111da177e4SLinus Torvalds 18122ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 18132ad94ae6SAl Viro struct nameidata *nd, char **name) 18142ad94ae6SAl Viro { 18152ad94ae6SAl Viro char *s = getname(path); 18162ad94ae6SAl Viro int error; 18172ad94ae6SAl Viro 18182ad94ae6SAl Viro if (IS_ERR(s)) 18192ad94ae6SAl Viro return PTR_ERR(s); 18202ad94ae6SAl Viro 18212ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 18222ad94ae6SAl Viro if (error) 18232ad94ae6SAl Viro putname(s); 18242ad94ae6SAl Viro else 18252ad94ae6SAl Viro *name = s; 18262ad94ae6SAl Viro 18272ad94ae6SAl Viro return error; 18282ad94ae6SAl Viro } 18292ad94ae6SAl Viro 18301da177e4SLinus Torvalds /* 18311da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 18321da177e4SLinus Torvalds * minimal. 18331da177e4SLinus Torvalds */ 18341da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 18351da177e4SLinus Torvalds { 1836da9592edSDavid Howells uid_t fsuid = current_fsuid(); 1837da9592edSDavid Howells 18381da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 18391da177e4SLinus Torvalds return 0; 1840da9592edSDavid Howells if (inode->i_uid == fsuid) 18411da177e4SLinus Torvalds return 0; 1842da9592edSDavid Howells if (dir->i_uid == fsuid) 18431da177e4SLinus Torvalds return 0; 18441da177e4SLinus Torvalds return !capable(CAP_FOWNER); 18451da177e4SLinus Torvalds } 18461da177e4SLinus Torvalds 18471da177e4SLinus Torvalds /* 18481da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 18491da177e4SLinus Torvalds * whether the type of victim is right. 18501da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 18511da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 18521da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 18531da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 18541da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 18551da177e4SLinus Torvalds * a. be owner of dir, or 18561da177e4SLinus Torvalds * b. be owner of victim, or 18571da177e4SLinus Torvalds * c. have CAP_FOWNER capability 18581da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 18591da177e4SLinus Torvalds * links pointing to it. 18601da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 18611da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 18621da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 18631da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 18641da177e4SLinus Torvalds * nfs_async_unlink(). 18651da177e4SLinus Torvalds */ 1866858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 18671da177e4SLinus Torvalds { 18681da177e4SLinus Torvalds int error; 18691da177e4SLinus Torvalds 18701da177e4SLinus Torvalds if (!victim->d_inode) 18711da177e4SLinus Torvalds return -ENOENT; 18721da177e4SLinus Torvalds 18731da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 1874cccc6bbaSAl Viro audit_inode_child(victim, dir); 18751da177e4SLinus Torvalds 1876f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 18771da177e4SLinus Torvalds if (error) 18781da177e4SLinus Torvalds return error; 18791da177e4SLinus Torvalds if (IS_APPEND(dir)) 18801da177e4SLinus Torvalds return -EPERM; 18811da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 1882f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 18831da177e4SLinus Torvalds return -EPERM; 18841da177e4SLinus Torvalds if (isdir) { 18851da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 18861da177e4SLinus Torvalds return -ENOTDIR; 18871da177e4SLinus Torvalds if (IS_ROOT(victim)) 18881da177e4SLinus Torvalds return -EBUSY; 18891da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 18901da177e4SLinus Torvalds return -EISDIR; 18911da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 18921da177e4SLinus Torvalds return -ENOENT; 18931da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 18941da177e4SLinus Torvalds return -EBUSY; 18951da177e4SLinus Torvalds return 0; 18961da177e4SLinus Torvalds } 18971da177e4SLinus Torvalds 18981da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 18991da177e4SLinus Torvalds * dir. 19001da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 19011da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 19021da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 19031da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 19041da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 19051da177e4SLinus Torvalds */ 1906a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 19071da177e4SLinus Torvalds { 19081da177e4SLinus Torvalds if (child->d_inode) 19091da177e4SLinus Torvalds return -EEXIST; 19101da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 19111da177e4SLinus Torvalds return -ENOENT; 1912f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 19131da177e4SLinus Torvalds } 19141da177e4SLinus Torvalds 19151da177e4SLinus Torvalds /* 19161da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 19171da177e4SLinus Torvalds */ 19181da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 19191da177e4SLinus Torvalds { 19201da177e4SLinus Torvalds struct dentry *p; 19211da177e4SLinus Torvalds 19221da177e4SLinus Torvalds if (p1 == p2) { 1923f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 19241da177e4SLinus Torvalds return NULL; 19251da177e4SLinus Torvalds } 19261da177e4SLinus Torvalds 1927a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19281da177e4SLinus Torvalds 1929e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 1930e2761a11SOGAWA Hirofumi if (p) { 1931f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1932f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 19331da177e4SLinus Torvalds return p; 19341da177e4SLinus Torvalds } 19351da177e4SLinus Torvalds 1936e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 1937e2761a11SOGAWA Hirofumi if (p) { 1938f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1939f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19401da177e4SLinus Torvalds return p; 19411da177e4SLinus Torvalds } 19421da177e4SLinus Torvalds 1943f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1944f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19451da177e4SLinus Torvalds return NULL; 19461da177e4SLinus Torvalds } 19471da177e4SLinus Torvalds 19481da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 19491da177e4SLinus Torvalds { 19501b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 19511da177e4SLinus Torvalds if (p1 != p2) { 19521b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 1953a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19541da177e4SLinus Torvalds } 19551da177e4SLinus Torvalds } 19561da177e4SLinus Torvalds 19571da177e4SLinus Torvalds int vfs_create(struct inode *dir, struct dentry *dentry, int mode, 19581da177e4SLinus Torvalds struct nameidata *nd) 19591da177e4SLinus Torvalds { 1960a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 19611da177e4SLinus Torvalds 19621da177e4SLinus Torvalds if (error) 19631da177e4SLinus Torvalds return error; 19641da177e4SLinus Torvalds 1965acfa4380SAl Viro if (!dir->i_op->create) 19661da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 19671da177e4SLinus Torvalds mode &= S_IALLUGO; 19681da177e4SLinus Torvalds mode |= S_IFREG; 19691da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 19701da177e4SLinus Torvalds if (error) 19711da177e4SLinus Torvalds return error; 19721da177e4SLinus Torvalds error = dir->i_op->create(dir, dentry, mode, nd); 1973a74574aaSStephen Smalley if (!error) 1974f38aa942SAmy Griffis fsnotify_create(dir, dentry); 19751da177e4SLinus Torvalds return error; 19761da177e4SLinus Torvalds } 19771da177e4SLinus Torvalds 19783fb64190SChristoph Hellwig int may_open(struct path *path, int acc_mode, int flag) 19791da177e4SLinus Torvalds { 19803fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 19811da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 19821da177e4SLinus Torvalds int error; 19831da177e4SLinus Torvalds 19841da177e4SLinus Torvalds if (!inode) 19851da177e4SLinus Torvalds return -ENOENT; 19861da177e4SLinus Torvalds 1987c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 1988c8fe8f30SChristoph Hellwig case S_IFLNK: 19891da177e4SLinus Torvalds return -ELOOP; 1990c8fe8f30SChristoph Hellwig case S_IFDIR: 1991c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 19921da177e4SLinus Torvalds return -EISDIR; 1993c8fe8f30SChristoph Hellwig break; 1994c8fe8f30SChristoph Hellwig case S_IFBLK: 1995c8fe8f30SChristoph Hellwig case S_IFCHR: 19963fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 19971da177e4SLinus Torvalds return -EACCES; 1998c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 1999c8fe8f30SChristoph Hellwig case S_IFIFO: 2000c8fe8f30SChristoph Hellwig case S_IFSOCK: 20011da177e4SLinus Torvalds flag &= ~O_TRUNC; 2002c8fe8f30SChristoph Hellwig break; 20034a3fd211SDave Hansen } 2004b41572e9SDave Hansen 20053fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2006b41572e9SDave Hansen if (error) 2007b41572e9SDave Hansen return error; 20086146f0d5SMimi Zohar 20091da177e4SLinus Torvalds /* 20101da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 20111da177e4SLinus Torvalds */ 20121da177e4SLinus Torvalds if (IS_APPEND(inode)) { 20138737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 20147715b521SAl Viro return -EPERM; 20151da177e4SLinus Torvalds if (flag & O_TRUNC) 20167715b521SAl Viro return -EPERM; 20171da177e4SLinus Torvalds } 20181da177e4SLinus Torvalds 20191da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 20207715b521SAl Viro if (flag & O_NOATIME && !is_owner_or_cap(inode)) 20217715b521SAl Viro return -EPERM; 20221da177e4SLinus Torvalds 20231da177e4SLinus Torvalds /* 20241da177e4SLinus Torvalds * Ensure there are no outstanding leases on the file. 20251da177e4SLinus Torvalds */ 2026b65a9cfcSAl Viro return break_lease(inode, flag); 20277715b521SAl Viro } 20287715b521SAl Viro 2029e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 20307715b521SAl Viro { 2031e1181ee6SJeff Layton struct path *path = &filp->f_path; 20327715b521SAl Viro struct inode *inode = path->dentry->d_inode; 20337715b521SAl Viro int error = get_write_access(inode); 20341da177e4SLinus Torvalds if (error) 20357715b521SAl Viro return error; 20361da177e4SLinus Torvalds /* 20371da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 20381da177e4SLinus Torvalds */ 20391da177e4SLinus Torvalds error = locks_verify_locked(inode); 2040be6d3e56SKentaro Takeda if (!error) 2041ea0d3ab2STetsuo Handa error = security_path_truncate(path); 20421da177e4SLinus Torvalds if (!error) { 20437715b521SAl Viro error = do_truncate(path->dentry, 0, 2044d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2045e1181ee6SJeff Layton filp); 20461da177e4SLinus Torvalds } 20471da177e4SLinus Torvalds put_write_access(inode); 2048acd0c935SMimi Zohar return error; 20491da177e4SLinus Torvalds } 20501da177e4SLinus Torvalds 2051d57999e1SDave Hansen /* 2052d57999e1SDave Hansen * Be careful about ever adding any more callers of this 2053d57999e1SDave Hansen * function. Its flags must be in the namei format, not 2054d57999e1SDave Hansen * what get passed to sys_open(). 2055d57999e1SDave Hansen */ 2056d57999e1SDave Hansen static int __open_namei_create(struct nameidata *nd, struct path *path, 20578737c930SAl Viro int open_flag, int mode) 2058aab520e2SDave Hansen { 2059aab520e2SDave Hansen int error; 20604ac91378SJan Blunck struct dentry *dir = nd->path.dentry; 2061aab520e2SDave Hansen 2062aab520e2SDave Hansen if (!IS_POSIXACL(dir->d_inode)) 2063ce3b0f8dSAl Viro mode &= ~current_umask(); 2064be6d3e56SKentaro Takeda error = security_path_mknod(&nd->path, path->dentry, mode, 0); 2065be6d3e56SKentaro Takeda if (error) 2066be6d3e56SKentaro Takeda goto out_unlock; 2067aab520e2SDave Hansen error = vfs_create(dir->d_inode, path->dentry, mode, nd); 2068be6d3e56SKentaro Takeda out_unlock: 2069aab520e2SDave Hansen mutex_unlock(&dir->d_inode->i_mutex); 20704ac91378SJan Blunck dput(nd->path.dentry); 20714ac91378SJan Blunck nd->path.dentry = path->dentry; 207231e6b01fSNick Piggin 2073aab520e2SDave Hansen if (error) 2074aab520e2SDave Hansen return error; 2075aab520e2SDave Hansen /* Don't check for write permission, don't truncate */ 20768737c930SAl Viro return may_open(&nd->path, 0, open_flag & ~O_TRUNC); 2077aab520e2SDave Hansen } 2078aab520e2SDave Hansen 20791da177e4SLinus Torvalds /* 2080d57999e1SDave Hansen * Note that while the flag value (low two bits) for sys_open means: 2081d57999e1SDave Hansen * 00 - read-only 2082d57999e1SDave Hansen * 01 - write-only 2083d57999e1SDave Hansen * 10 - read-write 2084d57999e1SDave Hansen * 11 - special 2085d57999e1SDave Hansen * it is changed into 2086d57999e1SDave Hansen * 00 - no permissions needed 2087d57999e1SDave Hansen * 01 - read-permission 2088d57999e1SDave Hansen * 10 - write-permission 2089d57999e1SDave Hansen * 11 - read-write 2090d57999e1SDave Hansen * for the internal routines (ie open_namei()/follow_link() etc) 2091d57999e1SDave Hansen * This is more logical, and also allows the 00 "no perm needed" 2092d57999e1SDave Hansen * to be used for symlinks (where the permissions are checked 2093d57999e1SDave Hansen * later). 2094d57999e1SDave Hansen * 2095d57999e1SDave Hansen */ 2096d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2097d57999e1SDave Hansen { 2098d57999e1SDave Hansen if ((flag+1) & O_ACCMODE) 2099d57999e1SDave Hansen flag++; 2100d57999e1SDave Hansen return flag; 2101d57999e1SDave Hansen } 2102d57999e1SDave Hansen 21037715b521SAl Viro static int open_will_truncate(int flag, struct inode *inode) 21044a3fd211SDave Hansen { 2105d57999e1SDave Hansen /* 21064a3fd211SDave Hansen * We'll never write to the fs underlying 21074a3fd211SDave Hansen * a device file. 21084a3fd211SDave Hansen */ 21094a3fd211SDave Hansen if (special_file(inode->i_mode)) 21104a3fd211SDave Hansen return 0; 21114a3fd211SDave Hansen return (flag & O_TRUNC); 21124a3fd211SDave Hansen } 21134a3fd211SDave Hansen 2114648fa861SAl Viro static struct file *finish_open(struct nameidata *nd, 21159a66179eSAl Viro int open_flag, int acc_mode) 2116648fa861SAl Viro { 2117648fa861SAl Viro struct file *filp; 2118648fa861SAl Viro int will_truncate; 2119648fa861SAl Viro int error; 2120648fa861SAl Viro 21219a66179eSAl Viro will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode); 2122648fa861SAl Viro if (will_truncate) { 2123648fa861SAl Viro error = mnt_want_write(nd->path.mnt); 2124648fa861SAl Viro if (error) 2125648fa861SAl Viro goto exit; 2126648fa861SAl Viro } 2127648fa861SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 2128648fa861SAl Viro if (error) { 2129648fa861SAl Viro if (will_truncate) 2130648fa861SAl Viro mnt_drop_write(nd->path.mnt); 2131648fa861SAl Viro goto exit; 2132648fa861SAl Viro } 2133648fa861SAl Viro filp = nameidata_to_filp(nd); 2134648fa861SAl Viro if (!IS_ERR(filp)) { 2135648fa861SAl Viro error = ima_file_check(filp, acc_mode); 2136648fa861SAl Viro if (error) { 2137648fa861SAl Viro fput(filp); 2138648fa861SAl Viro filp = ERR_PTR(error); 2139648fa861SAl Viro } 2140648fa861SAl Viro } 2141648fa861SAl Viro if (!IS_ERR(filp)) { 2142648fa861SAl Viro if (will_truncate) { 2143e1181ee6SJeff Layton error = handle_truncate(filp); 2144648fa861SAl Viro if (error) { 2145648fa861SAl Viro fput(filp); 2146648fa861SAl Viro filp = ERR_PTR(error); 2147648fa861SAl Viro } 2148648fa861SAl Viro } 2149648fa861SAl Viro } 2150648fa861SAl Viro /* 2151648fa861SAl Viro * It is now safe to drop the mnt write 2152648fa861SAl Viro * because the filp has had a write taken 2153648fa861SAl Viro * on its behalf. 2154648fa861SAl Viro */ 2155648fa861SAl Viro if (will_truncate) 2156648fa861SAl Viro mnt_drop_write(nd->path.mnt); 2157d893f1bcSAl Viro path_put(&nd->path); 2158648fa861SAl Viro return filp; 2159648fa861SAl Viro 2160648fa861SAl Viro exit: 2161648fa861SAl Viro path_put(&nd->path); 2162648fa861SAl Viro return ERR_PTR(error); 2163648fa861SAl Viro } 2164648fa861SAl Viro 216531e6b01fSNick Piggin /* 2166fe2d35ffSAl Viro * Handle the last step of open() 216731e6b01fSNick Piggin */ 2168fb1cc555SAl Viro static struct file *do_last(struct nameidata *nd, struct path *path, 2169c3e380b0SAl Viro const struct open_flags *op, const char *pathname) 2170fb1cc555SAl Viro { 2171a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2172fb1cc555SAl Viro struct file *filp; 2173fe2d35ffSAl Viro struct inode *inode; 217416c2cd71SAl Viro int error; 2175fb1cc555SAl Viro 2176c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2177c3e380b0SAl Viro nd->flags |= op->intent; 2178c3e380b0SAl Viro 21791f36f774SAl Viro switch (nd->last_type) { 21801f36f774SAl Viro case LAST_DOTDOT: 2181176306f5SNeil Brown case LAST_DOT: 2182fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2183fe2d35ffSAl Viro if (error) 2184fe2d35ffSAl Viro return ERR_PTR(error); 21851f36f774SAl Viro /* fallthrough */ 21861f36f774SAl Viro case LAST_ROOT: 2187fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2188fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2189fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2190fe2d35ffSAl Viro } 219116c2cd71SAl Viro error = handle_reval_path(nd); 219216c2cd71SAl Viro if (error) 219316c2cd71SAl Viro goto exit; 2194fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2195fe2d35ffSAl Viro if (op->open_flag & O_CREAT) { 219616c2cd71SAl Viro error = -EISDIR; 21971f36f774SAl Viro goto exit; 2198fe2d35ffSAl Viro } 2199fe2d35ffSAl Viro goto ok; 22001f36f774SAl Viro case LAST_BIND: 2201fe2d35ffSAl Viro /* can't be RCU mode here */ 220216c2cd71SAl Viro error = handle_reval_path(nd); 220316c2cd71SAl Viro if (error) 220416c2cd71SAl Viro goto exit; 22051f36f774SAl Viro audit_inode(pathname, dir); 22061f36f774SAl Viro goto ok; 22071f36f774SAl Viro } 2208a2c36b45SAl Viro 2209fe2d35ffSAl Viro if (!(op->open_flag & O_CREAT)) { 2210fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2211fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2212fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2213fe2d35ffSAl Viro error = do_lookup(nd, &nd->last, path, &inode); 2214fe2d35ffSAl Viro if (error) { 2215fe2d35ffSAl Viro terminate_walk(nd); 2216fe2d35ffSAl Viro return ERR_PTR(error); 2217fe2d35ffSAl Viro } 2218fe2d35ffSAl Viro if (!inode) { 2219fe2d35ffSAl Viro path_to_nameidata(path, nd); 2220fe2d35ffSAl Viro terminate_walk(nd); 2221fe2d35ffSAl Viro return ERR_PTR(-ENOENT); 2222fe2d35ffSAl Viro } 2223fe2d35ffSAl Viro if (unlikely(inode->i_op->follow_link)) { 2224fe2d35ffSAl Viro /* We drop rcu-walk here */ 2225fe2d35ffSAl Viro if (nameidata_dentry_drop_rcu_maybe(nd, path->dentry)) 2226fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2227fe2d35ffSAl Viro return NULL; 2228fe2d35ffSAl Viro } 2229fe2d35ffSAl Viro path_to_nameidata(path, nd); 2230fe2d35ffSAl Viro nd->inode = inode; 2231fe2d35ffSAl Viro /* sayonara */ 2232fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2233fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2234fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2235fe2d35ffSAl Viro } 2236fe2d35ffSAl Viro 2237fe2d35ffSAl Viro error = -ENOTDIR; 2238fe2d35ffSAl Viro if (nd->flags & LOOKUP_DIRECTORY) { 2239fe2d35ffSAl Viro if (!inode->i_op->lookup) 2240fe2d35ffSAl Viro goto exit; 2241fe2d35ffSAl Viro } 2242fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2243fe2d35ffSAl Viro goto ok; 2244fe2d35ffSAl Viro } 2245fe2d35ffSAl Viro 2246fe2d35ffSAl Viro /* create side of things */ 2247fe2d35ffSAl Viro 2248fe2d35ffSAl Viro if (nd->flags & LOOKUP_RCU) { 2249fe2d35ffSAl Viro if (nameidata_drop_rcu_last(nd)) 2250fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2251fe2d35ffSAl Viro } 2252fe2d35ffSAl Viro 2253fe2d35ffSAl Viro audit_inode(pathname, dir); 225416c2cd71SAl Viro error = -EISDIR; 22551f36f774SAl Viro /* trailing slashes? */ 225631e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 22571f36f774SAl Viro goto exit; 22581f36f774SAl Viro 2259a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 2260a1e28038SAl Viro 2261a1e28038SAl Viro path->dentry = lookup_hash(nd); 2262a1e28038SAl Viro path->mnt = nd->path.mnt; 2263a1e28038SAl Viro 2264fb1cc555SAl Viro error = PTR_ERR(path->dentry); 2265fb1cc555SAl Viro if (IS_ERR(path->dentry)) { 2266fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2267fb1cc555SAl Viro goto exit; 2268fb1cc555SAl Viro } 2269fb1cc555SAl Viro 2270fb1cc555SAl Viro if (IS_ERR(nd->intent.open.file)) { 2271fb1cc555SAl Viro error = PTR_ERR(nd->intent.open.file); 2272fb1cc555SAl Viro goto exit_mutex_unlock; 2273fb1cc555SAl Viro } 2274fb1cc555SAl Viro 2275fb1cc555SAl Viro /* Negative dentry, just create the file */ 2276fb1cc555SAl Viro if (!path->dentry->d_inode) { 2277fb1cc555SAl Viro /* 2278fb1cc555SAl Viro * This write is needed to ensure that a 2279fb1cc555SAl Viro * ro->rw transition does not occur between 2280fb1cc555SAl Viro * the time when the file is created and when 2281fb1cc555SAl Viro * a permanent write count is taken through 2282fb1cc555SAl Viro * the 'struct file' in nameidata_to_filp(). 2283fb1cc555SAl Viro */ 2284fb1cc555SAl Viro error = mnt_want_write(nd->path.mnt); 2285fb1cc555SAl Viro if (error) 2286fb1cc555SAl Viro goto exit_mutex_unlock; 2287c3e380b0SAl Viro error = __open_namei_create(nd, path, op->open_flag, op->mode); 2288fb1cc555SAl Viro if (error) { 2289fb1cc555SAl Viro mnt_drop_write(nd->path.mnt); 2290fb1cc555SAl Viro goto exit; 2291fb1cc555SAl Viro } 2292fb1cc555SAl Viro filp = nameidata_to_filp(nd); 2293fb1cc555SAl Viro mnt_drop_write(nd->path.mnt); 2294d893f1bcSAl Viro path_put(&nd->path); 2295fb1cc555SAl Viro if (!IS_ERR(filp)) { 2296c3e380b0SAl Viro error = ima_file_check(filp, op->acc_mode); 2297fb1cc555SAl Viro if (error) { 2298fb1cc555SAl Viro fput(filp); 2299fb1cc555SAl Viro filp = ERR_PTR(error); 2300fb1cc555SAl Viro } 2301fb1cc555SAl Viro } 2302fb1cc555SAl Viro return filp; 2303fb1cc555SAl Viro } 2304fb1cc555SAl Viro 2305fb1cc555SAl Viro /* 2306fb1cc555SAl Viro * It already exists. 2307fb1cc555SAl Viro */ 2308fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2309fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2310fb1cc555SAl Viro 2311fb1cc555SAl Viro error = -EEXIST; 2312c3e380b0SAl Viro if (op->open_flag & O_EXCL) 2313fb1cc555SAl Viro goto exit_dput; 2314fb1cc555SAl Viro 23159875cf80SDavid Howells error = follow_managed(path, nd->flags); 23169875cf80SDavid Howells if (error < 0) 2317fb1cc555SAl Viro goto exit_dput; 2318fb1cc555SAl Viro 2319fb1cc555SAl Viro error = -ENOENT; 2320fb1cc555SAl Viro if (!path->dentry->d_inode) 2321fb1cc555SAl Viro goto exit_dput; 23229e67f361SAl Viro 23239e67f361SAl Viro if (path->dentry->d_inode->i_op->follow_link) 2324fb1cc555SAl Viro return NULL; 2325fb1cc555SAl Viro 2326fb1cc555SAl Viro path_to_nameidata(path, nd); 232731e6b01fSNick Piggin nd->inode = path->dentry->d_inode; 2328fb1cc555SAl Viro error = -EISDIR; 232931e6b01fSNick Piggin if (S_ISDIR(nd->inode->i_mode)) 2330fb1cc555SAl Viro goto exit; 233167ee3ad2SAl Viro ok: 2332c3e380b0SAl Viro filp = finish_open(nd, op->open_flag, op->acc_mode); 2333fb1cc555SAl Viro return filp; 2334fb1cc555SAl Viro 2335fb1cc555SAl Viro exit_mutex_unlock: 2336fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2337fb1cc555SAl Viro exit_dput: 2338fb1cc555SAl Viro path_put_conditional(path, nd); 2339fb1cc555SAl Viro exit: 2340fb1cc555SAl Viro path_put(&nd->path); 2341fb1cc555SAl Viro return ERR_PTR(error); 2342fb1cc555SAl Viro } 2343fb1cc555SAl Viro 234413aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 234547c805dcSAl Viro const struct open_flags *op, int flags) 23461da177e4SLinus Torvalds { 2347fe2d35ffSAl Viro struct file *base = NULL; 23484a3fd211SDave Hansen struct file *filp; 2349a70e65dfSChristoph Hellwig struct nameidata nd; 23509850c056SAl Viro struct path path; 23511da177e4SLinus Torvalds int count = 0; 235213aab428SAl Viro int error; 235331e6b01fSNick Piggin 235431e6b01fSNick Piggin filp = get_empty_filp(); 235531e6b01fSNick Piggin if (!filp) 235631e6b01fSNick Piggin return ERR_PTR(-ENFILE); 235731e6b01fSNick Piggin 235847c805dcSAl Viro filp->f_flags = op->open_flag; 235931e6b01fSNick Piggin nd.intent.open.file = filp; 236047c805dcSAl Viro nd.intent.open.flags = open_to_namei_flags(op->open_flag); 236147c805dcSAl Viro nd.intent.open.create_mode = op->mode; 236231e6b01fSNick Piggin 2363fe2d35ffSAl Viro error = path_init(dfd, pathname, flags | LOOKUP_PARENT, &nd, &base); 236431e6b01fSNick Piggin if (unlikely(error)) 236513aab428SAl Viro goto out_filp; 236631e6b01fSNick Piggin 2367fe2d35ffSAl Viro current->total_link_count = 0; 2368fe2d35ffSAl Viro error = link_path_walk(pathname, &nd); 236931e6b01fSNick Piggin if (unlikely(error)) 237031e6b01fSNick Piggin goto out_filp; 23711da177e4SLinus Torvalds 237247c805dcSAl Viro filp = do_last(&nd, &path, op, pathname); 2373806b681cSAl Viro while (unlikely(!filp)) { /* trailing symlink */ 23747b9337aaSNick Piggin struct path link = path; 23757b9337aaSNick Piggin struct inode *linki = link.dentry->d_inode; 2376def4af30SAl Viro void *cookie; 2377806b681cSAl Viro error = -ELOOP; 2378db372915SDavid Howells if (!(nd.flags & LOOKUP_FOLLOW)) 23791f36f774SAl Viro goto exit_dput; 23801f36f774SAl Viro if (count++ == 32) 2381806b681cSAl Viro goto exit_dput; 2382806b681cSAl Viro /* 2383806b681cSAl Viro * This is subtle. Instead of calling do_follow_link() we do 2384806b681cSAl Viro * the thing by hands. The reason is that this way we have zero 2385806b681cSAl Viro * link_count and path_walk() (called from ->follow_link) 2386806b681cSAl Viro * honoring LOOKUP_PARENT. After that we have the parent and 2387806b681cSAl Viro * last component, i.e. we are in the same situation as after 2388806b681cSAl Viro * the first path_walk(). Well, almost - if the last component 2389806b681cSAl Viro * is normal we get its copy stored in nd->last.name and we will 2390806b681cSAl Viro * have to putname() it when we are done. Procfs-like symlinks 2391806b681cSAl Viro * just set LAST_BIND. 2392806b681cSAl Viro */ 2393806b681cSAl Viro nd.flags |= LOOKUP_PARENT; 2394c3e380b0SAl Viro nd.flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 23957b9337aaSNick Piggin error = __do_follow_link(&link, &nd, &cookie); 2396c3e380b0SAl Viro if (unlikely(error)) 2397f1afe9efSAl Viro filp = ERR_PTR(error); 2398c3e380b0SAl Viro else 239947c805dcSAl Viro filp = do_last(&nd, &path, op, pathname); 2400f1afe9efSAl Viro if (!IS_ERR(cookie) && linki->i_op->put_link) 24017b9337aaSNick Piggin linki->i_op->put_link(link.dentry, &nd, cookie); 24027b9337aaSNick Piggin path_put(&link); 2403806b681cSAl Viro } 240410fa8e62SAl Viro out: 24052a737871SAl Viro if (nd.root.mnt) 24062a737871SAl Viro path_put(&nd.root); 2407fe2d35ffSAl Viro if (base) 2408fe2d35ffSAl Viro fput(base); 24092dab5974SLinus Torvalds release_open_intent(&nd); 241010fa8e62SAl Viro return filp; 24111da177e4SLinus Torvalds 2412806b681cSAl Viro exit_dput: 2413806b681cSAl Viro path_put_conditional(&path, &nd); 241431e6b01fSNick Piggin path_put(&nd.path); 241531e6b01fSNick Piggin out_filp: 241610fa8e62SAl Viro filp = ERR_PTR(error); 241710fa8e62SAl Viro goto out; 2418de459215SKirill Korotaev } 24191da177e4SLinus Torvalds 242013aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 242113aab428SAl Viro const struct open_flags *op, int flags) 242213aab428SAl Viro { 242313aab428SAl Viro struct file *filp; 242413aab428SAl Viro 242513aab428SAl Viro filp = path_openat(dfd, pathname, op, flags | LOOKUP_RCU); 242613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 242713aab428SAl Viro filp = path_openat(dfd, pathname, op, flags); 242813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 242913aab428SAl Viro filp = path_openat(dfd, pathname, op, flags | LOOKUP_REVAL); 243013aab428SAl Viro return filp; 243113aab428SAl Viro } 243213aab428SAl Viro 24331da177e4SLinus Torvalds /** 24341da177e4SLinus Torvalds * lookup_create - lookup a dentry, creating it if it doesn't exist 24351da177e4SLinus Torvalds * @nd: nameidata info 24361da177e4SLinus Torvalds * @is_dir: directory flag 24371da177e4SLinus Torvalds * 24381da177e4SLinus Torvalds * Simple function to lookup and return a dentry and create it 24391da177e4SLinus Torvalds * if it doesn't exist. Is SMP-safe. 2440c663e5d8SChristoph Hellwig * 24414ac91378SJan Blunck * Returns with nd->path.dentry->d_inode->i_mutex locked. 24421da177e4SLinus Torvalds */ 24431da177e4SLinus Torvalds struct dentry *lookup_create(struct nameidata *nd, int is_dir) 24441da177e4SLinus Torvalds { 2445c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 24461da177e4SLinus Torvalds 24474ac91378SJan Blunck mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2448c663e5d8SChristoph Hellwig /* 2449c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2450c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2451c663e5d8SChristoph Hellwig */ 24521da177e4SLinus Torvalds if (nd->last_type != LAST_NORM) 24531da177e4SLinus Torvalds goto fail; 24541da177e4SLinus Torvalds nd->flags &= ~LOOKUP_PARENT; 24553516586aSAl Viro nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2456a634904aSASANO Masahiro nd->intent.open.flags = O_EXCL; 2457c663e5d8SChristoph Hellwig 2458c663e5d8SChristoph Hellwig /* 2459c663e5d8SChristoph Hellwig * Do the final lookup. 2460c663e5d8SChristoph Hellwig */ 246149705b77SChristoph Hellwig dentry = lookup_hash(nd); 24621da177e4SLinus Torvalds if (IS_ERR(dentry)) 24631da177e4SLinus Torvalds goto fail; 2464c663e5d8SChristoph Hellwig 2465e9baf6e5SAl Viro if (dentry->d_inode) 2466e9baf6e5SAl Viro goto eexist; 2467c663e5d8SChristoph Hellwig /* 2468c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 2469c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 2470c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 2471c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 2472c663e5d8SChristoph Hellwig */ 2473e9baf6e5SAl Viro if (unlikely(!is_dir && nd->last.name[nd->last.len])) { 24741da177e4SLinus Torvalds dput(dentry); 24751da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 2476e9baf6e5SAl Viro } 2477e9baf6e5SAl Viro return dentry; 2478e9baf6e5SAl Viro eexist: 2479e9baf6e5SAl Viro dput(dentry); 2480e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 24811da177e4SLinus Torvalds fail: 24821da177e4SLinus Torvalds return dentry; 24831da177e4SLinus Torvalds } 2484f81a0bffSChristoph Hellwig EXPORT_SYMBOL_GPL(lookup_create); 24851da177e4SLinus Torvalds 24861da177e4SLinus Torvalds int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 24871da177e4SLinus Torvalds { 2488a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 24891da177e4SLinus Torvalds 24901da177e4SLinus Torvalds if (error) 24911da177e4SLinus Torvalds return error; 24921da177e4SLinus Torvalds 24931da177e4SLinus Torvalds if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 24941da177e4SLinus Torvalds return -EPERM; 24951da177e4SLinus Torvalds 2496acfa4380SAl Viro if (!dir->i_op->mknod) 24971da177e4SLinus Torvalds return -EPERM; 24981da177e4SLinus Torvalds 249908ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 250008ce5f16SSerge E. Hallyn if (error) 250108ce5f16SSerge E. Hallyn return error; 250208ce5f16SSerge E. Hallyn 25031da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 25041da177e4SLinus Torvalds if (error) 25051da177e4SLinus Torvalds return error; 25061da177e4SLinus Torvalds 25071da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 2508a74574aaSStephen Smalley if (!error) 2509f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25101da177e4SLinus Torvalds return error; 25111da177e4SLinus Torvalds } 25121da177e4SLinus Torvalds 2513463c3197SDave Hansen static int may_mknod(mode_t mode) 2514463c3197SDave Hansen { 2515463c3197SDave Hansen switch (mode & S_IFMT) { 2516463c3197SDave Hansen case S_IFREG: 2517463c3197SDave Hansen case S_IFCHR: 2518463c3197SDave Hansen case S_IFBLK: 2519463c3197SDave Hansen case S_IFIFO: 2520463c3197SDave Hansen case S_IFSOCK: 2521463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 2522463c3197SDave Hansen return 0; 2523463c3197SDave Hansen case S_IFDIR: 2524463c3197SDave Hansen return -EPERM; 2525463c3197SDave Hansen default: 2526463c3197SDave Hansen return -EINVAL; 2527463c3197SDave Hansen } 2528463c3197SDave Hansen } 2529463c3197SDave Hansen 25302e4d0924SHeiko Carstens SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, 25312e4d0924SHeiko Carstens unsigned, dev) 25321da177e4SLinus Torvalds { 25332ad94ae6SAl Viro int error; 25341da177e4SLinus Torvalds char *tmp; 25351da177e4SLinus Torvalds struct dentry *dentry; 25361da177e4SLinus Torvalds struct nameidata nd; 25371da177e4SLinus Torvalds 25381da177e4SLinus Torvalds if (S_ISDIR(mode)) 25391da177e4SLinus Torvalds return -EPERM; 25401da177e4SLinus Torvalds 25412ad94ae6SAl Viro error = user_path_parent(dfd, filename, &nd, &tmp); 25421da177e4SLinus Torvalds if (error) 25432ad94ae6SAl Viro return error; 25442ad94ae6SAl Viro 25451da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 2546463c3197SDave Hansen if (IS_ERR(dentry)) { 25471da177e4SLinus Torvalds error = PTR_ERR(dentry); 2548463c3197SDave Hansen goto out_unlock; 2549463c3197SDave Hansen } 25504ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2551ce3b0f8dSAl Viro mode &= ~current_umask(); 2552463c3197SDave Hansen error = may_mknod(mode); 2553463c3197SDave Hansen if (error) 2554463c3197SDave Hansen goto out_dput; 2555463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2556463c3197SDave Hansen if (error) 2557463c3197SDave Hansen goto out_dput; 2558be6d3e56SKentaro Takeda error = security_path_mknod(&nd.path, dentry, mode, dev); 2559be6d3e56SKentaro Takeda if (error) 2560be6d3e56SKentaro Takeda goto out_drop_write; 25611da177e4SLinus Torvalds switch (mode & S_IFMT) { 25621da177e4SLinus Torvalds case 0: case S_IFREG: 25634ac91378SJan Blunck error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); 25641da177e4SLinus Torvalds break; 25651da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 25664ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, 25671da177e4SLinus Torvalds new_decode_dev(dev)); 25681da177e4SLinus Torvalds break; 25691da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 25704ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); 25711da177e4SLinus Torvalds break; 25721da177e4SLinus Torvalds } 2573be6d3e56SKentaro Takeda out_drop_write: 2574463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2575463c3197SDave Hansen out_dput: 25761da177e4SLinus Torvalds dput(dentry); 2577463c3197SDave Hansen out_unlock: 25784ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 25791d957f9bSJan Blunck path_put(&nd.path); 25801da177e4SLinus Torvalds putname(tmp); 25811da177e4SLinus Torvalds 25821da177e4SLinus Torvalds return error; 25831da177e4SLinus Torvalds } 25841da177e4SLinus Torvalds 25853480b257SHeiko Carstens SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) 25865590ff0dSUlrich Drepper { 25875590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 25885590ff0dSUlrich Drepper } 25895590ff0dSUlrich Drepper 25901da177e4SLinus Torvalds int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 25911da177e4SLinus Torvalds { 2592a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25931da177e4SLinus Torvalds 25941da177e4SLinus Torvalds if (error) 25951da177e4SLinus Torvalds return error; 25961da177e4SLinus Torvalds 2597acfa4380SAl Viro if (!dir->i_op->mkdir) 25981da177e4SLinus Torvalds return -EPERM; 25991da177e4SLinus Torvalds 26001da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 26011da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 26021da177e4SLinus Torvalds if (error) 26031da177e4SLinus Torvalds return error; 26041da177e4SLinus Torvalds 26051da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 2606a74574aaSStephen Smalley if (!error) 2607f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 26081da177e4SLinus Torvalds return error; 26091da177e4SLinus Torvalds } 26101da177e4SLinus Torvalds 26112e4d0924SHeiko Carstens SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) 26121da177e4SLinus Torvalds { 26131da177e4SLinus Torvalds int error = 0; 26141da177e4SLinus Torvalds char * tmp; 26156902d925SDave Hansen struct dentry *dentry; 26166902d925SDave Hansen struct nameidata nd; 26171da177e4SLinus Torvalds 26182ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &tmp); 26192ad94ae6SAl Viro if (error) 26206902d925SDave Hansen goto out_err; 26211da177e4SLinus Torvalds 26221da177e4SLinus Torvalds dentry = lookup_create(&nd, 1); 26231da177e4SLinus Torvalds error = PTR_ERR(dentry); 26246902d925SDave Hansen if (IS_ERR(dentry)) 26256902d925SDave Hansen goto out_unlock; 26266902d925SDave Hansen 26274ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2628ce3b0f8dSAl Viro mode &= ~current_umask(); 2629463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2630463c3197SDave Hansen if (error) 2631463c3197SDave Hansen goto out_dput; 2632be6d3e56SKentaro Takeda error = security_path_mkdir(&nd.path, dentry, mode); 2633be6d3e56SKentaro Takeda if (error) 2634be6d3e56SKentaro Takeda goto out_drop_write; 26354ac91378SJan Blunck error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); 2636be6d3e56SKentaro Takeda out_drop_write: 2637463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2638463c3197SDave Hansen out_dput: 26391da177e4SLinus Torvalds dput(dentry); 26406902d925SDave Hansen out_unlock: 26414ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 26421d957f9bSJan Blunck path_put(&nd.path); 26431da177e4SLinus Torvalds putname(tmp); 26446902d925SDave Hansen out_err: 26451da177e4SLinus Torvalds return error; 26461da177e4SLinus Torvalds } 26471da177e4SLinus Torvalds 26483cdad428SHeiko Carstens SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) 26495590ff0dSUlrich Drepper { 26505590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 26515590ff0dSUlrich Drepper } 26525590ff0dSUlrich Drepper 26531da177e4SLinus Torvalds /* 26541da177e4SLinus Torvalds * We try to drop the dentry early: we should have 26551da177e4SLinus Torvalds * a usage count of 2 if we're the only user of this 26561da177e4SLinus Torvalds * dentry, and if that is true (possibly after pruning 26571da177e4SLinus Torvalds * the dcache), then we drop the dentry now. 26581da177e4SLinus Torvalds * 26591da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 26601da177e4SLinus Torvalds * do a 26611da177e4SLinus Torvalds * 26621da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 26631da177e4SLinus Torvalds * return -EBUSY; 26641da177e4SLinus Torvalds * 26651da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 26661da177e4SLinus Torvalds * that is still in use by something else.. 26671da177e4SLinus Torvalds */ 26681da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 26691da177e4SLinus Torvalds { 26701da177e4SLinus Torvalds dget(dentry); 26711da177e4SLinus Torvalds shrink_dcache_parent(dentry); 26721da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 2673b7ab39f6SNick Piggin if (dentry->d_count == 2) 26741da177e4SLinus Torvalds __d_drop(dentry); 26751da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 26761da177e4SLinus Torvalds } 26771da177e4SLinus Torvalds 26781da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 26791da177e4SLinus Torvalds { 26801da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 26811da177e4SLinus Torvalds 26821da177e4SLinus Torvalds if (error) 26831da177e4SLinus Torvalds return error; 26841da177e4SLinus Torvalds 2685acfa4380SAl Viro if (!dir->i_op->rmdir) 26861da177e4SLinus Torvalds return -EPERM; 26871da177e4SLinus Torvalds 26881b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 26891da177e4SLinus Torvalds dentry_unhash(dentry); 26901da177e4SLinus Torvalds if (d_mountpoint(dentry)) 26911da177e4SLinus Torvalds error = -EBUSY; 26921da177e4SLinus Torvalds else { 26931da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 26941da177e4SLinus Torvalds if (!error) { 26951da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 2696d83c49f3SAl Viro if (!error) { 26971da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 2698d83c49f3SAl Viro dont_mount(dentry); 2699d83c49f3SAl Viro } 27001da177e4SLinus Torvalds } 27011da177e4SLinus Torvalds } 27021b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 27031da177e4SLinus Torvalds if (!error) { 27041da177e4SLinus Torvalds d_delete(dentry); 27051da177e4SLinus Torvalds } 27061da177e4SLinus Torvalds dput(dentry); 27071da177e4SLinus Torvalds 27081da177e4SLinus Torvalds return error; 27091da177e4SLinus Torvalds } 27101da177e4SLinus Torvalds 27115590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 27121da177e4SLinus Torvalds { 27131da177e4SLinus Torvalds int error = 0; 27141da177e4SLinus Torvalds char * name; 27151da177e4SLinus Torvalds struct dentry *dentry; 27161da177e4SLinus Torvalds struct nameidata nd; 27171da177e4SLinus Torvalds 27182ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 27191da177e4SLinus Torvalds if (error) 27202ad94ae6SAl Viro return error; 27211da177e4SLinus Torvalds 27221da177e4SLinus Torvalds switch(nd.last_type) { 27231da177e4SLinus Torvalds case LAST_DOTDOT: 27241da177e4SLinus Torvalds error = -ENOTEMPTY; 27251da177e4SLinus Torvalds goto exit1; 27261da177e4SLinus Torvalds case LAST_DOT: 27271da177e4SLinus Torvalds error = -EINVAL; 27281da177e4SLinus Torvalds goto exit1; 27291da177e4SLinus Torvalds case LAST_ROOT: 27301da177e4SLinus Torvalds error = -EBUSY; 27311da177e4SLinus Torvalds goto exit1; 27321da177e4SLinus Torvalds } 27330612d9fbSOGAWA Hirofumi 27340612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 27350612d9fbSOGAWA Hirofumi 27364ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 273749705b77SChristoph Hellwig dentry = lookup_hash(&nd); 27381da177e4SLinus Torvalds error = PTR_ERR(dentry); 27396902d925SDave Hansen if (IS_ERR(dentry)) 27406902d925SDave Hansen goto exit2; 27410622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 27420622753bSDave Hansen if (error) 27430622753bSDave Hansen goto exit3; 2744be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 2745be6d3e56SKentaro Takeda if (error) 2746be6d3e56SKentaro Takeda goto exit4; 27474ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 2748be6d3e56SKentaro Takeda exit4: 27490622753bSDave Hansen mnt_drop_write(nd.path.mnt); 27500622753bSDave Hansen exit3: 27511da177e4SLinus Torvalds dput(dentry); 27526902d925SDave Hansen exit2: 27534ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 27541da177e4SLinus Torvalds exit1: 27551d957f9bSJan Blunck path_put(&nd.path); 27561da177e4SLinus Torvalds putname(name); 27571da177e4SLinus Torvalds return error; 27581da177e4SLinus Torvalds } 27591da177e4SLinus Torvalds 27603cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 27615590ff0dSUlrich Drepper { 27625590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 27635590ff0dSUlrich Drepper } 27645590ff0dSUlrich Drepper 27651da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 27661da177e4SLinus Torvalds { 27671da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 27681da177e4SLinus Torvalds 27691da177e4SLinus Torvalds if (error) 27701da177e4SLinus Torvalds return error; 27711da177e4SLinus Torvalds 2772acfa4380SAl Viro if (!dir->i_op->unlink) 27731da177e4SLinus Torvalds return -EPERM; 27741da177e4SLinus Torvalds 27751b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 27761da177e4SLinus Torvalds if (d_mountpoint(dentry)) 27771da177e4SLinus Torvalds error = -EBUSY; 27781da177e4SLinus Torvalds else { 27791da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 2780bec1052eSAl Viro if (!error) { 27811da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 2782bec1052eSAl Viro if (!error) 2783d83c49f3SAl Viro dont_mount(dentry); 2784bec1052eSAl Viro } 27851da177e4SLinus Torvalds } 27861b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 27871da177e4SLinus Torvalds 27881da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 27891da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 2790ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 27911da177e4SLinus Torvalds d_delete(dentry); 27921da177e4SLinus Torvalds } 27930eeca283SRobert Love 27941da177e4SLinus Torvalds return error; 27951da177e4SLinus Torvalds } 27961da177e4SLinus Torvalds 27971da177e4SLinus Torvalds /* 27981da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 27991b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 28001da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 28011da177e4SLinus Torvalds * while waiting on the I/O. 28021da177e4SLinus Torvalds */ 28035590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 28041da177e4SLinus Torvalds { 28052ad94ae6SAl Viro int error; 28061da177e4SLinus Torvalds char *name; 28071da177e4SLinus Torvalds struct dentry *dentry; 28081da177e4SLinus Torvalds struct nameidata nd; 28091da177e4SLinus Torvalds struct inode *inode = NULL; 28101da177e4SLinus Torvalds 28112ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 28121da177e4SLinus Torvalds if (error) 28132ad94ae6SAl Viro return error; 28142ad94ae6SAl Viro 28151da177e4SLinus Torvalds error = -EISDIR; 28161da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 28171da177e4SLinus Torvalds goto exit1; 28180612d9fbSOGAWA Hirofumi 28190612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 28200612d9fbSOGAWA Hirofumi 28214ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 282249705b77SChristoph Hellwig dentry = lookup_hash(&nd); 28231da177e4SLinus Torvalds error = PTR_ERR(dentry); 28241da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 28251da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 28261da177e4SLinus Torvalds if (nd.last.name[nd.last.len]) 28271da177e4SLinus Torvalds goto slashes; 28281da177e4SLinus Torvalds inode = dentry->d_inode; 28291da177e4SLinus Torvalds if (inode) 28307de9c6eeSAl Viro ihold(inode); 28310622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 28320622753bSDave Hansen if (error) 28330622753bSDave Hansen goto exit2; 2834be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 2835be6d3e56SKentaro Takeda if (error) 2836be6d3e56SKentaro Takeda goto exit3; 28374ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 2838be6d3e56SKentaro Takeda exit3: 28390622753bSDave Hansen mnt_drop_write(nd.path.mnt); 28401da177e4SLinus Torvalds exit2: 28411da177e4SLinus Torvalds dput(dentry); 28421da177e4SLinus Torvalds } 28434ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 28441da177e4SLinus Torvalds if (inode) 28451da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 28461da177e4SLinus Torvalds exit1: 28471d957f9bSJan Blunck path_put(&nd.path); 28481da177e4SLinus Torvalds putname(name); 28491da177e4SLinus Torvalds return error; 28501da177e4SLinus Torvalds 28511da177e4SLinus Torvalds slashes: 28521da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 28531da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 28541da177e4SLinus Torvalds goto exit2; 28551da177e4SLinus Torvalds } 28561da177e4SLinus Torvalds 28572e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 28585590ff0dSUlrich Drepper { 28595590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 28605590ff0dSUlrich Drepper return -EINVAL; 28615590ff0dSUlrich Drepper 28625590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 28635590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 28645590ff0dSUlrich Drepper 28655590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 28665590ff0dSUlrich Drepper } 28675590ff0dSUlrich Drepper 28683480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 28695590ff0dSUlrich Drepper { 28705590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 28715590ff0dSUlrich Drepper } 28725590ff0dSUlrich Drepper 2873db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 28741da177e4SLinus Torvalds { 2875a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28761da177e4SLinus Torvalds 28771da177e4SLinus Torvalds if (error) 28781da177e4SLinus Torvalds return error; 28791da177e4SLinus Torvalds 2880acfa4380SAl Viro if (!dir->i_op->symlink) 28811da177e4SLinus Torvalds return -EPERM; 28821da177e4SLinus Torvalds 28831da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 28841da177e4SLinus Torvalds if (error) 28851da177e4SLinus Torvalds return error; 28861da177e4SLinus Torvalds 28871da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 2888a74574aaSStephen Smalley if (!error) 2889f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28901da177e4SLinus Torvalds return error; 28911da177e4SLinus Torvalds } 28921da177e4SLinus Torvalds 28932e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 28942e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 28951da177e4SLinus Torvalds { 28962ad94ae6SAl Viro int error; 28971da177e4SLinus Torvalds char *from; 28981da177e4SLinus Torvalds char *to; 28996902d925SDave Hansen struct dentry *dentry; 29006902d925SDave Hansen struct nameidata nd; 29011da177e4SLinus Torvalds 29021da177e4SLinus Torvalds from = getname(oldname); 29031da177e4SLinus Torvalds if (IS_ERR(from)) 29041da177e4SLinus Torvalds return PTR_ERR(from); 29052ad94ae6SAl Viro 29062ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 29072ad94ae6SAl Viro if (error) 29086902d925SDave Hansen goto out_putname; 29091da177e4SLinus Torvalds 29101da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 29111da177e4SLinus Torvalds error = PTR_ERR(dentry); 29126902d925SDave Hansen if (IS_ERR(dentry)) 29136902d925SDave Hansen goto out_unlock; 29146902d925SDave Hansen 291575c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 291675c3f29dSDave Hansen if (error) 291775c3f29dSDave Hansen goto out_dput; 2918be6d3e56SKentaro Takeda error = security_path_symlink(&nd.path, dentry, from); 2919be6d3e56SKentaro Takeda if (error) 2920be6d3e56SKentaro Takeda goto out_drop_write; 2921db2e747bSMiklos Szeredi error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); 2922be6d3e56SKentaro Takeda out_drop_write: 292375c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 292475c3f29dSDave Hansen out_dput: 29251da177e4SLinus Torvalds dput(dentry); 29266902d925SDave Hansen out_unlock: 29274ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 29281d957f9bSJan Blunck path_put(&nd.path); 29291da177e4SLinus Torvalds putname(to); 29306902d925SDave Hansen out_putname: 29311da177e4SLinus Torvalds putname(from); 29321da177e4SLinus Torvalds return error; 29331da177e4SLinus Torvalds } 29341da177e4SLinus Torvalds 29353480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 29365590ff0dSUlrich Drepper { 29375590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 29385590ff0dSUlrich Drepper } 29395590ff0dSUlrich Drepper 29401da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 29411da177e4SLinus Torvalds { 29421da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 29431da177e4SLinus Torvalds int error; 29441da177e4SLinus Torvalds 29451da177e4SLinus Torvalds if (!inode) 29461da177e4SLinus Torvalds return -ENOENT; 29471da177e4SLinus Torvalds 2948a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 29491da177e4SLinus Torvalds if (error) 29501da177e4SLinus Torvalds return error; 29511da177e4SLinus Torvalds 29521da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 29531da177e4SLinus Torvalds return -EXDEV; 29541da177e4SLinus Torvalds 29551da177e4SLinus Torvalds /* 29561da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 29571da177e4SLinus Torvalds */ 29581da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 29591da177e4SLinus Torvalds return -EPERM; 2960acfa4380SAl Viro if (!dir->i_op->link) 29611da177e4SLinus Torvalds return -EPERM; 29627e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 29631da177e4SLinus Torvalds return -EPERM; 29641da177e4SLinus Torvalds 29651da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 29661da177e4SLinus Torvalds if (error) 29671da177e4SLinus Torvalds return error; 29681da177e4SLinus Torvalds 29697e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 29701da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 29717e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 2972e31e14ecSStephen Smalley if (!error) 29737e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 29741da177e4SLinus Torvalds return error; 29751da177e4SLinus Torvalds } 29761da177e4SLinus Torvalds 29771da177e4SLinus Torvalds /* 29781da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 29791da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 29801da177e4SLinus Torvalds * newname. --KAB 29811da177e4SLinus Torvalds * 29821da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 29831da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 29841da177e4SLinus Torvalds * and other special files. --ADM 29851da177e4SLinus Torvalds */ 29862e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 29872e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 29881da177e4SLinus Torvalds { 29891da177e4SLinus Torvalds struct dentry *new_dentry; 29902d8f3038SAl Viro struct nameidata nd; 29912d8f3038SAl Viro struct path old_path; 29921da177e4SLinus Torvalds int error; 29931da177e4SLinus Torvalds char *to; 29941da177e4SLinus Torvalds 299545c9b11aSUlrich Drepper if ((flags & ~AT_SYMLINK_FOLLOW) != 0) 2996c04030e1SUlrich Drepper return -EINVAL; 2997c04030e1SUlrich Drepper 29982d8f3038SAl Viro error = user_path_at(olddfd, oldname, 299945c9b11aSUlrich Drepper flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, 30002d8f3038SAl Viro &old_path); 30011da177e4SLinus Torvalds if (error) 30022ad94ae6SAl Viro return error; 30032ad94ae6SAl Viro 30042ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 30051da177e4SLinus Torvalds if (error) 30061da177e4SLinus Torvalds goto out; 30071da177e4SLinus Torvalds error = -EXDEV; 30082d8f3038SAl Viro if (old_path.mnt != nd.path.mnt) 30091da177e4SLinus Torvalds goto out_release; 30101da177e4SLinus Torvalds new_dentry = lookup_create(&nd, 0); 30111da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 30126902d925SDave Hansen if (IS_ERR(new_dentry)) 30136902d925SDave Hansen goto out_unlock; 301475c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 301575c3f29dSDave Hansen if (error) 301675c3f29dSDave Hansen goto out_dput; 3017be6d3e56SKentaro Takeda error = security_path_link(old_path.dentry, &nd.path, new_dentry); 3018be6d3e56SKentaro Takeda if (error) 3019be6d3e56SKentaro Takeda goto out_drop_write; 30202d8f3038SAl Viro error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); 3021be6d3e56SKentaro Takeda out_drop_write: 302275c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 302375c3f29dSDave Hansen out_dput: 30241da177e4SLinus Torvalds dput(new_dentry); 30256902d925SDave Hansen out_unlock: 30264ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 30271da177e4SLinus Torvalds out_release: 30281d957f9bSJan Blunck path_put(&nd.path); 30292ad94ae6SAl Viro putname(to); 30301da177e4SLinus Torvalds out: 30312d8f3038SAl Viro path_put(&old_path); 30321da177e4SLinus Torvalds 30331da177e4SLinus Torvalds return error; 30341da177e4SLinus Torvalds } 30351da177e4SLinus Torvalds 30363480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 30375590ff0dSUlrich Drepper { 3038c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 30395590ff0dSUlrich Drepper } 30405590ff0dSUlrich Drepper 30411da177e4SLinus Torvalds /* 30421da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 30431da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 30441da177e4SLinus Torvalds * Problems: 30451da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 30461da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 30471da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3048a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 30491da177e4SLinus Torvalds * story. 30501da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 30511b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 30521da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 30531da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3054a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 30551da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 30561da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 30571da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3058a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 30591da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 30601da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 30611da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 30621da177e4SLinus Torvalds * d) some filesystems don't support opened-but-unlinked directories, 30631da177e4SLinus Torvalds * either because of layout or because they are not ready to deal with 30641da177e4SLinus Torvalds * all cases correctly. The latter will be fixed (taking this sort of 30651da177e4SLinus Torvalds * stuff into VFS), but the former is not going away. Solution: the same 30661da177e4SLinus Torvalds * trick as in rmdir(). 30671da177e4SLinus Torvalds * e) conversion from fhandle to dentry may come in the wrong moment - when 30681b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 30691da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3070c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 30711da177e4SLinus Torvalds * locking]. 30721da177e4SLinus Torvalds */ 307375c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 30741da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30751da177e4SLinus Torvalds { 30761da177e4SLinus Torvalds int error = 0; 30771da177e4SLinus Torvalds struct inode *target; 30781da177e4SLinus Torvalds 30791da177e4SLinus Torvalds /* 30801da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 30811da177e4SLinus Torvalds * we'll need to flip '..'. 30821da177e4SLinus Torvalds */ 30831da177e4SLinus Torvalds if (new_dir != old_dir) { 3084f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 30851da177e4SLinus Torvalds if (error) 30861da177e4SLinus Torvalds return error; 30871da177e4SLinus Torvalds } 30881da177e4SLinus Torvalds 30891da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30901da177e4SLinus Torvalds if (error) 30911da177e4SLinus Torvalds return error; 30921da177e4SLinus Torvalds 30931da177e4SLinus Torvalds target = new_dentry->d_inode; 3094d83c49f3SAl Viro if (target) 30951b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 30961da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 30971da177e4SLinus Torvalds error = -EBUSY; 3098d83c49f3SAl Viro else { 3099d83c49f3SAl Viro if (target) 3100d83c49f3SAl Viro dentry_unhash(new_dentry); 31011da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 3102d83c49f3SAl Viro } 31031da177e4SLinus Torvalds if (target) { 3104d83c49f3SAl Viro if (!error) { 31051da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3106d83c49f3SAl Viro dont_mount(new_dentry); 3107d83c49f3SAl Viro } 31081b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 31091da177e4SLinus Torvalds if (d_unhashed(new_dentry)) 31101da177e4SLinus Torvalds d_rehash(new_dentry); 31111da177e4SLinus Torvalds dput(new_dentry); 31121da177e4SLinus Torvalds } 3113e31e14ecSStephen Smalley if (!error) 3114349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 31151da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 31161da177e4SLinus Torvalds return error; 31171da177e4SLinus Torvalds } 31181da177e4SLinus Torvalds 311975c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 31201da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 31211da177e4SLinus Torvalds { 31221da177e4SLinus Torvalds struct inode *target; 31231da177e4SLinus Torvalds int error; 31241da177e4SLinus Torvalds 31251da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 31261da177e4SLinus Torvalds if (error) 31271da177e4SLinus Torvalds return error; 31281da177e4SLinus Torvalds 31291da177e4SLinus Torvalds dget(new_dentry); 31301da177e4SLinus Torvalds target = new_dentry->d_inode; 31311da177e4SLinus Torvalds if (target) 31321b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 31331da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 31341da177e4SLinus Torvalds error = -EBUSY; 31351da177e4SLinus Torvalds else 31361da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 31371da177e4SLinus Torvalds if (!error) { 3138bec1052eSAl Viro if (target) 3139d83c49f3SAl Viro dont_mount(new_dentry); 3140349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 31411da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 31421da177e4SLinus Torvalds } 31431da177e4SLinus Torvalds if (target) 31441b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 31451da177e4SLinus Torvalds dput(new_dentry); 31461da177e4SLinus Torvalds return error; 31471da177e4SLinus Torvalds } 31481da177e4SLinus Torvalds 31491da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 31501da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 31511da177e4SLinus Torvalds { 31521da177e4SLinus Torvalds int error; 31531da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 315459b0df21SEric Paris const unsigned char *old_name; 31551da177e4SLinus Torvalds 31561da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 31571da177e4SLinus Torvalds return 0; 31581da177e4SLinus Torvalds 31591da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 31601da177e4SLinus Torvalds if (error) 31611da177e4SLinus Torvalds return error; 31621da177e4SLinus Torvalds 31631da177e4SLinus Torvalds if (!new_dentry->d_inode) 3164a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 31651da177e4SLinus Torvalds else 31661da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 31671da177e4SLinus Torvalds if (error) 31681da177e4SLinus Torvalds return error; 31691da177e4SLinus Torvalds 3170acfa4380SAl Viro if (!old_dir->i_op->rename) 31711da177e4SLinus Torvalds return -EPERM; 31721da177e4SLinus Torvalds 31730eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 31740eeca283SRobert Love 31751da177e4SLinus Torvalds if (is_dir) 31761da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 31771da177e4SLinus Torvalds else 31781da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3179123df294SAl Viro if (!error) 3180123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 31815a190ae6SAl Viro new_dentry->d_inode, old_dentry); 31820eeca283SRobert Love fsnotify_oldname_free(old_name); 31830eeca283SRobert Love 31841da177e4SLinus Torvalds return error; 31851da177e4SLinus Torvalds } 31861da177e4SLinus Torvalds 31872e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 31882e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 31891da177e4SLinus Torvalds { 31901da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 31911da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 31921da177e4SLinus Torvalds struct dentry *trap; 31931da177e4SLinus Torvalds struct nameidata oldnd, newnd; 31942ad94ae6SAl Viro char *from; 31952ad94ae6SAl Viro char *to; 31962ad94ae6SAl Viro int error; 31971da177e4SLinus Torvalds 31982ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 31991da177e4SLinus Torvalds if (error) 32001da177e4SLinus Torvalds goto exit; 32011da177e4SLinus Torvalds 32022ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 32031da177e4SLinus Torvalds if (error) 32041da177e4SLinus Torvalds goto exit1; 32051da177e4SLinus Torvalds 32061da177e4SLinus Torvalds error = -EXDEV; 32074ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 32081da177e4SLinus Torvalds goto exit2; 32091da177e4SLinus Torvalds 32104ac91378SJan Blunck old_dir = oldnd.path.dentry; 32111da177e4SLinus Torvalds error = -EBUSY; 32121da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 32131da177e4SLinus Torvalds goto exit2; 32141da177e4SLinus Torvalds 32154ac91378SJan Blunck new_dir = newnd.path.dentry; 32161da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 32171da177e4SLinus Torvalds goto exit2; 32181da177e4SLinus Torvalds 32190612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 32200612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 32214e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 32220612d9fbSOGAWA Hirofumi 32231da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 32241da177e4SLinus Torvalds 322549705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 32261da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 32271da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 32281da177e4SLinus Torvalds goto exit3; 32291da177e4SLinus Torvalds /* source must exist */ 32301da177e4SLinus Torvalds error = -ENOENT; 32311da177e4SLinus Torvalds if (!old_dentry->d_inode) 32321da177e4SLinus Torvalds goto exit4; 32331da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 32341da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 32351da177e4SLinus Torvalds error = -ENOTDIR; 32361da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 32371da177e4SLinus Torvalds goto exit4; 32381da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 32391da177e4SLinus Torvalds goto exit4; 32401da177e4SLinus Torvalds } 32411da177e4SLinus Torvalds /* source should not be ancestor of target */ 32421da177e4SLinus Torvalds error = -EINVAL; 32431da177e4SLinus Torvalds if (old_dentry == trap) 32441da177e4SLinus Torvalds goto exit4; 324549705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 32461da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 32471da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 32481da177e4SLinus Torvalds goto exit4; 32491da177e4SLinus Torvalds /* target should not be an ancestor of source */ 32501da177e4SLinus Torvalds error = -ENOTEMPTY; 32511da177e4SLinus Torvalds if (new_dentry == trap) 32521da177e4SLinus Torvalds goto exit5; 32531da177e4SLinus Torvalds 32549079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 32559079b1ebSDave Hansen if (error) 32569079b1ebSDave Hansen goto exit5; 3257be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3258be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3259be6d3e56SKentaro Takeda if (error) 3260be6d3e56SKentaro Takeda goto exit6; 32611da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 32621da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 3263be6d3e56SKentaro Takeda exit6: 32649079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 32651da177e4SLinus Torvalds exit5: 32661da177e4SLinus Torvalds dput(new_dentry); 32671da177e4SLinus Torvalds exit4: 32681da177e4SLinus Torvalds dput(old_dentry); 32691da177e4SLinus Torvalds exit3: 32701da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 32711da177e4SLinus Torvalds exit2: 32721d957f9bSJan Blunck path_put(&newnd.path); 32732ad94ae6SAl Viro putname(to); 32741da177e4SLinus Torvalds exit1: 32751d957f9bSJan Blunck path_put(&oldnd.path); 32761da177e4SLinus Torvalds putname(from); 32772ad94ae6SAl Viro exit: 32781da177e4SLinus Torvalds return error; 32791da177e4SLinus Torvalds } 32801da177e4SLinus Torvalds 3281a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 32825590ff0dSUlrich Drepper { 32835590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 32845590ff0dSUlrich Drepper } 32855590ff0dSUlrich Drepper 32861da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 32871da177e4SLinus Torvalds { 32881da177e4SLinus Torvalds int len; 32891da177e4SLinus Torvalds 32901da177e4SLinus Torvalds len = PTR_ERR(link); 32911da177e4SLinus Torvalds if (IS_ERR(link)) 32921da177e4SLinus Torvalds goto out; 32931da177e4SLinus Torvalds 32941da177e4SLinus Torvalds len = strlen(link); 32951da177e4SLinus Torvalds if (len > (unsigned) buflen) 32961da177e4SLinus Torvalds len = buflen; 32971da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 32981da177e4SLinus Torvalds len = -EFAULT; 32991da177e4SLinus Torvalds out: 33001da177e4SLinus Torvalds return len; 33011da177e4SLinus Torvalds } 33021da177e4SLinus Torvalds 33031da177e4SLinus Torvalds /* 33041da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 33051da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 33061da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 33071da177e4SLinus Torvalds */ 33081da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 33091da177e4SLinus Torvalds { 33101da177e4SLinus Torvalds struct nameidata nd; 3311cc314eefSLinus Torvalds void *cookie; 3312694a1764SMarcin Slusarz int res; 3313cc314eefSLinus Torvalds 33141da177e4SLinus Torvalds nd.depth = 0; 3315cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3316694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3317694a1764SMarcin Slusarz return PTR_ERR(cookie); 3318694a1764SMarcin Slusarz 3319694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 33201da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3321cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3322694a1764SMarcin Slusarz return res; 33231da177e4SLinus Torvalds } 33241da177e4SLinus Torvalds 33251da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 33261da177e4SLinus Torvalds { 33271da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 33281da177e4SLinus Torvalds } 33291da177e4SLinus Torvalds 33301da177e4SLinus Torvalds /* get the link contents into pagecache */ 33311da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 33321da177e4SLinus Torvalds { 3333ebd09abbSDuane Griffin char *kaddr; 33341da177e4SLinus Torvalds struct page *page; 33351da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3336090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 33371da177e4SLinus Torvalds if (IS_ERR(page)) 33386fe6900eSNick Piggin return (char*)page; 33391da177e4SLinus Torvalds *ppage = page; 3340ebd09abbSDuane Griffin kaddr = kmap(page); 3341ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3342ebd09abbSDuane Griffin return kaddr; 33431da177e4SLinus Torvalds } 33441da177e4SLinus Torvalds 33451da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 33461da177e4SLinus Torvalds { 33471da177e4SLinus Torvalds struct page *page = NULL; 33481da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 33491da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 33501da177e4SLinus Torvalds if (page) { 33511da177e4SLinus Torvalds kunmap(page); 33521da177e4SLinus Torvalds page_cache_release(page); 33531da177e4SLinus Torvalds } 33541da177e4SLinus Torvalds return res; 33551da177e4SLinus Torvalds } 33561da177e4SLinus Torvalds 3357cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 33581da177e4SLinus Torvalds { 3359cc314eefSLinus Torvalds struct page *page = NULL; 33601da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3361cc314eefSLinus Torvalds return page; 33621da177e4SLinus Torvalds } 33631da177e4SLinus Torvalds 3364cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 33651da177e4SLinus Torvalds { 3366cc314eefSLinus Torvalds struct page *page = cookie; 3367cc314eefSLinus Torvalds 3368cc314eefSLinus Torvalds if (page) { 33691da177e4SLinus Torvalds kunmap(page); 33701da177e4SLinus Torvalds page_cache_release(page); 33711da177e4SLinus Torvalds } 33721da177e4SLinus Torvalds } 33731da177e4SLinus Torvalds 337454566b2cSNick Piggin /* 337554566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 337654566b2cSNick Piggin */ 337754566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 33781da177e4SLinus Torvalds { 33791da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 33800adb25d2SKirill Korotaev struct page *page; 3381afddba49SNick Piggin void *fsdata; 3382beb497abSDmitriy Monakhov int err; 33831da177e4SLinus Torvalds char *kaddr; 338454566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 338554566b2cSNick Piggin if (nofs) 338654566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 33871da177e4SLinus Torvalds 33887e53cac4SNeilBrown retry: 3389afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 339054566b2cSNick Piggin flags, &page, &fsdata); 33911da177e4SLinus Torvalds if (err) 3392afddba49SNick Piggin goto fail; 3393afddba49SNick Piggin 33941da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 33951da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 33961da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 3397afddba49SNick Piggin 3398afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3399afddba49SNick Piggin page, fsdata); 34001da177e4SLinus Torvalds if (err < 0) 34011da177e4SLinus Torvalds goto fail; 3402afddba49SNick Piggin if (err < len-1) 3403afddba49SNick Piggin goto retry; 3404afddba49SNick Piggin 34051da177e4SLinus Torvalds mark_inode_dirty(inode); 34061da177e4SLinus Torvalds return 0; 34071da177e4SLinus Torvalds fail: 34081da177e4SLinus Torvalds return err; 34091da177e4SLinus Torvalds } 34101da177e4SLinus Torvalds 34110adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 34120adb25d2SKirill Korotaev { 34130adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 341454566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 34150adb25d2SKirill Korotaev } 34160adb25d2SKirill Korotaev 341792e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 34181da177e4SLinus Torvalds .readlink = generic_readlink, 34191da177e4SLinus Torvalds .follow_link = page_follow_link_light, 34201da177e4SLinus Torvalds .put_link = page_put_link, 34211da177e4SLinus Torvalds }; 34221da177e4SLinus Torvalds 34232d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3424cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 34251da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 34261da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 34271da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 34281da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 34291da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 34301da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 34311da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 34321da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 34331da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 34340adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 34351da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 34361da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3437c9c6cac0SAl Viro EXPORT_SYMBOL(kern_path_parent); 3438d1811465SAl Viro EXPORT_SYMBOL(kern_path); 343916f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3440f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 34418c744fb8SChristoph Hellwig EXPORT_SYMBOL(file_permission); 34421da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 34431da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 34441da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 34451da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 34461da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 34471da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 34481da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 34491da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 34501da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 34511da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 34521da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 34531da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 34541da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 34551da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3456