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 */ 1725909ccaaSLinus Torvalds static int acl_permission_check(struct inode *inode, int mask, 1735909ccaaSLinus Torvalds int (*check_acl)(struct inode *inode, int mask)) 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) { 1835909ccaaSLinus Torvalds int error = check_acl(inode, mask); 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 2051da177e4SLinus Torvalds * 2061da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 2071da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 2081da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 2091da177e4SLinus Torvalds * are used for other things.. 2101da177e4SLinus Torvalds */ 2111da177e4SLinus Torvalds int generic_permission(struct inode *inode, int mask, 2121da177e4SLinus Torvalds int (*check_acl)(struct inode *inode, int mask)) 2131da177e4SLinus Torvalds { 2145909ccaaSLinus Torvalds int ret; 2151da177e4SLinus Torvalds 2161da177e4SLinus Torvalds /* 2175909ccaaSLinus Torvalds * Do the basic POSIX ACL permission checks. 2181da177e4SLinus Torvalds */ 2195909ccaaSLinus Torvalds ret = acl_permission_check(inode, mask, check_acl); 2205909ccaaSLinus Torvalds if (ret != -EACCES) 2215909ccaaSLinus Torvalds return ret; 2221da177e4SLinus Torvalds 2231da177e4SLinus Torvalds /* 2241da177e4SLinus Torvalds * Read/write DACs are always overridable. 2251da177e4SLinus Torvalds * Executable DACs are overridable if at least one exec bit is set. 2261da177e4SLinus Torvalds */ 227f696a365SMiklos Szeredi if (!(mask & MAY_EXEC) || execute_ok(inode)) 2281da177e4SLinus Torvalds if (capable(CAP_DAC_OVERRIDE)) 2291da177e4SLinus Torvalds return 0; 2301da177e4SLinus Torvalds 2311da177e4SLinus Torvalds /* 2321da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 2331da177e4SLinus Torvalds */ 2347ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 2351da177e4SLinus Torvalds if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE))) 2361da177e4SLinus Torvalds if (capable(CAP_DAC_READ_SEARCH)) 2371da177e4SLinus Torvalds return 0; 2381da177e4SLinus Torvalds 2391da177e4SLinus Torvalds return -EACCES; 2401da177e4SLinus Torvalds } 2411da177e4SLinus Torvalds 242cb23beb5SChristoph Hellwig /** 243cb23beb5SChristoph Hellwig * inode_permission - check for access rights to a given inode 244cb23beb5SChristoph Hellwig * @inode: inode to check permission on 245cb23beb5SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 246cb23beb5SChristoph Hellwig * 247cb23beb5SChristoph Hellwig * Used to check for read/write/execute permissions on an inode. 248cb23beb5SChristoph Hellwig * We use "fsuid" for this, letting us set arbitrary permissions 249cb23beb5SChristoph Hellwig * for filesystem access without changing the "normal" uids which 250cb23beb5SChristoph Hellwig * are used for other things. 251cb23beb5SChristoph Hellwig */ 252f419a2e3SAl Viro int inode_permission(struct inode *inode, int mask) 2531da177e4SLinus Torvalds { 254e6305c43SAl Viro int retval; 2551da177e4SLinus Torvalds 2561da177e4SLinus Torvalds if (mask & MAY_WRITE) { 25722590e41SMiklos Szeredi umode_t mode = inode->i_mode; 2581da177e4SLinus Torvalds 2591da177e4SLinus Torvalds /* 2601da177e4SLinus Torvalds * Nobody gets write access to a read-only fs. 2611da177e4SLinus Torvalds */ 2621da177e4SLinus Torvalds if (IS_RDONLY(inode) && 2631da177e4SLinus Torvalds (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 2641da177e4SLinus Torvalds return -EROFS; 2651da177e4SLinus Torvalds 2661da177e4SLinus Torvalds /* 2671da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 2681da177e4SLinus Torvalds */ 2691da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 2701da177e4SLinus Torvalds return -EACCES; 2711da177e4SLinus Torvalds } 2721da177e4SLinus Torvalds 273acfa4380SAl Viro if (inode->i_op->permission) 274b77b0646SAl Viro retval = inode->i_op->permission(inode, mask); 275f696a365SMiklos Szeredi else 2765909ccaaSLinus Torvalds retval = generic_permission(inode, mask, inode->i_op->check_acl); 277f696a365SMiklos Szeredi 2781da177e4SLinus Torvalds if (retval) 2791da177e4SLinus Torvalds return retval; 2801da177e4SLinus Torvalds 28108ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 28208ce5f16SSerge E. Hallyn if (retval) 28308ce5f16SSerge E. Hallyn return retval; 28408ce5f16SSerge E. Hallyn 285d09ca739SEric Paris return security_inode_permission(inode, mask); 2861da177e4SLinus Torvalds } 2871da177e4SLinus Torvalds 288e4543eddSChristoph Hellwig /** 2898c744fb8SChristoph Hellwig * file_permission - check for additional access rights to a given file 2908c744fb8SChristoph Hellwig * @file: file to check access rights for 2918c744fb8SChristoph Hellwig * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 2928c744fb8SChristoph Hellwig * 2938c744fb8SChristoph Hellwig * Used to check for read/write/execute permissions on an already opened 2948c744fb8SChristoph Hellwig * file. 2958c744fb8SChristoph Hellwig * 2968c744fb8SChristoph Hellwig * Note: 2978c744fb8SChristoph Hellwig * Do not use this function in new code. All access checks should 298cb23beb5SChristoph Hellwig * be done using inode_permission(). 2998c744fb8SChristoph Hellwig */ 3008c744fb8SChristoph Hellwig int file_permission(struct file *file, int mask) 3018c744fb8SChristoph Hellwig { 302f419a2e3SAl Viro return inode_permission(file->f_path.dentry->d_inode, mask); 3038c744fb8SChristoph Hellwig } 3048c744fb8SChristoph Hellwig 3051da177e4SLinus Torvalds /* 3061da177e4SLinus Torvalds * get_write_access() gets write permission for a file. 3071da177e4SLinus Torvalds * put_write_access() releases this write permission. 3081da177e4SLinus Torvalds * This is used for regular files. 3091da177e4SLinus Torvalds * We cannot support write (and maybe mmap read-write shared) accesses and 3101da177e4SLinus Torvalds * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode 3111da177e4SLinus Torvalds * can have the following values: 3121da177e4SLinus Torvalds * 0: no writers, no VM_DENYWRITE mappings 3131da177e4SLinus Torvalds * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist 3141da177e4SLinus Torvalds * > 0: (i_writecount) users are writing to the file. 3151da177e4SLinus Torvalds * 3161da177e4SLinus Torvalds * Normally we operate on that counter with atomic_{inc,dec} and it's safe 3171da177e4SLinus Torvalds * except for the cases where we don't hold i_writecount yet. Then we need to 3181da177e4SLinus Torvalds * use {get,deny}_write_access() - these functions check the sign and refuse 3191da177e4SLinus Torvalds * to do the change if sign is wrong. Exclusion between them is provided by 3201da177e4SLinus Torvalds * the inode->i_lock spinlock. 3211da177e4SLinus Torvalds */ 3221da177e4SLinus Torvalds 3231da177e4SLinus Torvalds int get_write_access(struct inode * inode) 3241da177e4SLinus Torvalds { 3251da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3261da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) < 0) { 3271da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3281da177e4SLinus Torvalds return -ETXTBSY; 3291da177e4SLinus Torvalds } 3301da177e4SLinus Torvalds atomic_inc(&inode->i_writecount); 3311da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3321da177e4SLinus Torvalds 3331da177e4SLinus Torvalds return 0; 3341da177e4SLinus Torvalds } 3351da177e4SLinus Torvalds 3361da177e4SLinus Torvalds int deny_write_access(struct file * file) 3371da177e4SLinus Torvalds { 3380f7fc9e4SJosef "Jeff" Sipek struct inode *inode = file->f_path.dentry->d_inode; 3391da177e4SLinus Torvalds 3401da177e4SLinus Torvalds spin_lock(&inode->i_lock); 3411da177e4SLinus Torvalds if (atomic_read(&inode->i_writecount) > 0) { 3421da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3431da177e4SLinus Torvalds return -ETXTBSY; 3441da177e4SLinus Torvalds } 3451da177e4SLinus Torvalds atomic_dec(&inode->i_writecount); 3461da177e4SLinus Torvalds spin_unlock(&inode->i_lock); 3471da177e4SLinus Torvalds 3481da177e4SLinus Torvalds return 0; 3491da177e4SLinus Torvalds } 3501da177e4SLinus Torvalds 3511d957f9bSJan Blunck /** 3525dd784d0SJan Blunck * path_get - get a reference to a path 3535dd784d0SJan Blunck * @path: path to get the reference to 3545dd784d0SJan Blunck * 3555dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3565dd784d0SJan Blunck */ 3575dd784d0SJan Blunck void path_get(struct path *path) 3585dd784d0SJan Blunck { 3595dd784d0SJan Blunck mntget(path->mnt); 3605dd784d0SJan Blunck dget(path->dentry); 3615dd784d0SJan Blunck } 3625dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 3635dd784d0SJan Blunck 3645dd784d0SJan Blunck /** 3651d957f9bSJan Blunck * path_put - put a reference to a path 3661d957f9bSJan Blunck * @path: path to put the reference to 3671d957f9bSJan Blunck * 3681d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 3691d957f9bSJan Blunck */ 3701d957f9bSJan Blunck void path_put(struct path *path) 3711da177e4SLinus Torvalds { 3721d957f9bSJan Blunck dput(path->dentry); 3731d957f9bSJan Blunck mntput(path->mnt); 3741da177e4SLinus Torvalds } 3751d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 3761da177e4SLinus Torvalds 377834f2a4aSTrond Myklebust /** 378834f2a4aSTrond Myklebust * release_open_intent - free up open intent resources 379834f2a4aSTrond Myklebust * @nd: pointer to nameidata 380834f2a4aSTrond Myklebust */ 381834f2a4aSTrond Myklebust void release_open_intent(struct nameidata *nd) 382834f2a4aSTrond Myklebust { 3830f7fc9e4SJosef "Jeff" Sipek if (nd->intent.open.file->f_path.dentry == NULL) 384834f2a4aSTrond Myklebust put_filp(nd->intent.open.file); 385834f2a4aSTrond Myklebust else 386834f2a4aSTrond Myklebust fput(nd->intent.open.file); 387834f2a4aSTrond Myklebust } 388834f2a4aSTrond Myklebust 389bcdc5e01SIan Kent static inline struct dentry * 390bcdc5e01SIan Kent do_revalidate(struct dentry *dentry, struct nameidata *nd) 391bcdc5e01SIan Kent { 392bcdc5e01SIan Kent int status = dentry->d_op->d_revalidate(dentry, nd); 393bcdc5e01SIan Kent if (unlikely(status <= 0)) { 394bcdc5e01SIan Kent /* 395bcdc5e01SIan Kent * The dentry failed validation. 396bcdc5e01SIan Kent * If d_revalidate returned 0 attempt to invalidate 397bcdc5e01SIan Kent * the dentry otherwise d_revalidate is asking us 398bcdc5e01SIan Kent * to return a fail status. 399bcdc5e01SIan Kent */ 400bcdc5e01SIan Kent if (!status) { 401bcdc5e01SIan Kent if (!d_invalidate(dentry)) { 402bcdc5e01SIan Kent dput(dentry); 403bcdc5e01SIan Kent dentry = NULL; 404bcdc5e01SIan Kent } 405bcdc5e01SIan Kent } else { 406bcdc5e01SIan Kent dput(dentry); 407bcdc5e01SIan Kent dentry = ERR_PTR(status); 408bcdc5e01SIan Kent } 409bcdc5e01SIan Kent } 410bcdc5e01SIan Kent return dentry; 411bcdc5e01SIan Kent } 412bcdc5e01SIan Kent 4131da177e4SLinus Torvalds /* 41439159de2SJeff Layton * force_reval_path - force revalidation of a dentry 41539159de2SJeff Layton * 41639159de2SJeff Layton * In some situations the path walking code will trust dentries without 41739159de2SJeff Layton * revalidating them. This causes problems for filesystems that depend on 41839159de2SJeff Layton * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set 41939159de2SJeff Layton * (which indicates that it's possible for the dentry to go stale), force 42039159de2SJeff Layton * a d_revalidate call before proceeding. 42139159de2SJeff Layton * 42239159de2SJeff Layton * Returns 0 if the revalidation was successful. If the revalidation fails, 42339159de2SJeff Layton * either return the error returned by d_revalidate or -ESTALE if the 42439159de2SJeff Layton * revalidation it just returned 0. If d_revalidate returns 0, we attempt to 42539159de2SJeff Layton * invalidate the dentry. It's up to the caller to handle putting references 42639159de2SJeff Layton * to the path if necessary. 42739159de2SJeff Layton */ 42839159de2SJeff Layton static int 42939159de2SJeff Layton force_reval_path(struct path *path, struct nameidata *nd) 43039159de2SJeff Layton { 43139159de2SJeff Layton int status; 43239159de2SJeff Layton struct dentry *dentry = path->dentry; 43339159de2SJeff Layton 43439159de2SJeff Layton /* 43539159de2SJeff Layton * only check on filesystems where it's possible for the dentry to 43639159de2SJeff Layton * become stale. It's assumed that if this flag is set then the 43739159de2SJeff Layton * d_revalidate op will also be defined. 43839159de2SJeff Layton */ 43939159de2SJeff Layton if (!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) 44039159de2SJeff Layton return 0; 44139159de2SJeff Layton 44239159de2SJeff Layton status = dentry->d_op->d_revalidate(dentry, nd); 44339159de2SJeff Layton if (status > 0) 44439159de2SJeff Layton return 0; 44539159de2SJeff Layton 44639159de2SJeff Layton if (!status) { 44739159de2SJeff Layton d_invalidate(dentry); 44839159de2SJeff Layton status = -ESTALE; 44939159de2SJeff Layton } 45039159de2SJeff Layton return status; 45139159de2SJeff Layton } 45239159de2SJeff Layton 45339159de2SJeff Layton /* 454b75b5086SAl Viro * Short-cut version of permission(), for calling on directories 455b75b5086SAl Viro * during pathname resolution. Combines parts of permission() 456b75b5086SAl Viro * and generic_permission(), and tests ONLY for MAY_EXEC permission. 4571da177e4SLinus Torvalds * 4581da177e4SLinus Torvalds * If appropriate, check DAC only. If not appropriate, or 459b75b5086SAl Viro * short-cut DAC fails, then call ->permission() to do more 4601da177e4SLinus Torvalds * complete permission check. 4611da177e4SLinus Torvalds */ 462b75b5086SAl Viro static int exec_permission(struct inode *inode) 4631da177e4SLinus Torvalds { 4645909ccaaSLinus Torvalds int ret; 4651da177e4SLinus Torvalds 466cb9179eaSLinus Torvalds if (inode->i_op->permission) { 4675909ccaaSLinus Torvalds ret = inode->i_op->permission(inode, MAY_EXEC); 468cb9179eaSLinus Torvalds if (!ret) 469cb9179eaSLinus Torvalds goto ok; 470cb9179eaSLinus Torvalds return ret; 471cb9179eaSLinus Torvalds } 4725909ccaaSLinus Torvalds ret = acl_permission_check(inode, MAY_EXEC, inode->i_op->check_acl); 4735909ccaaSLinus Torvalds if (!ret) 4741da177e4SLinus Torvalds goto ok; 4751da177e4SLinus Torvalds 476f1ac9f6bSLinus Torvalds if (capable(CAP_DAC_OVERRIDE) || capable(CAP_DAC_READ_SEARCH)) 4771da177e4SLinus Torvalds goto ok; 4781da177e4SLinus Torvalds 4795909ccaaSLinus Torvalds return ret; 4801da177e4SLinus Torvalds ok: 481b77b0646SAl Viro return security_inode_permission(inode, MAY_EXEC); 4821da177e4SLinus Torvalds } 4831da177e4SLinus Torvalds 4842a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 4852a737871SAl Viro { 486f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 487f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 4882a737871SAl Viro } 4892a737871SAl Viro 4906de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 4916de88d72SAl Viro 492f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 4931da177e4SLinus Torvalds { 4941da177e4SLinus Torvalds if (IS_ERR(link)) 4951da177e4SLinus Torvalds goto fail; 4961da177e4SLinus Torvalds 4971da177e4SLinus Torvalds if (*link == '/') { 4982a737871SAl Viro set_root(nd); 4991d957f9bSJan Blunck path_put(&nd->path); 5002a737871SAl Viro nd->path = nd->root; 5012a737871SAl Viro path_get(&nd->root); 5021da177e4SLinus Torvalds } 503b4091d5fSChristoph Hellwig 504def4af30SAl Viro return link_path_walk(link, nd); 5051da177e4SLinus Torvalds fail: 5061d957f9bSJan Blunck path_put(&nd->path); 5071da177e4SLinus Torvalds return PTR_ERR(link); 5081da177e4SLinus Torvalds } 5091da177e4SLinus Torvalds 5101d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 511051d3812SIan Kent { 512051d3812SIan Kent dput(path->dentry); 5134ac91378SJan Blunck if (path->mnt != nd->path.mnt) 514051d3812SIan Kent mntput(path->mnt); 515051d3812SIan Kent } 516051d3812SIan Kent 517051d3812SIan Kent static inline void path_to_nameidata(struct path *path, struct nameidata *nd) 518051d3812SIan Kent { 5194ac91378SJan Blunck dput(nd->path.dentry); 5209a229683SHuang Shijie if (nd->path.mnt != path->mnt) { 5214ac91378SJan Blunck mntput(nd->path.mnt); 5224ac91378SJan Blunck nd->path.mnt = path->mnt; 5239a229683SHuang Shijie } 5244ac91378SJan Blunck nd->path.dentry = path->dentry; 525051d3812SIan Kent } 526051d3812SIan Kent 527def4af30SAl Viro static __always_inline int 528def4af30SAl Viro __do_follow_link(struct path *path, struct nameidata *nd, void **p) 5291da177e4SLinus Torvalds { 5301da177e4SLinus Torvalds int error; 531cd4e91d3SAl Viro struct dentry *dentry = path->dentry; 5321da177e4SLinus Torvalds 533d671a1cbSAl Viro touch_atime(path->mnt, dentry); 5341da177e4SLinus Torvalds nd_set_link(nd, NULL); 535cd4e91d3SAl Viro 5364ac91378SJan Blunck if (path->mnt != nd->path.mnt) { 537051d3812SIan Kent path_to_nameidata(path, nd); 538051d3812SIan Kent dget(dentry); 539051d3812SIan Kent } 540cd4e91d3SAl Viro mntget(path->mnt); 54186acdca1SAl Viro nd->last_type = LAST_BIND; 542def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 543def4af30SAl Viro error = PTR_ERR(*p); 544def4af30SAl Viro if (!IS_ERR(*p)) { 5451da177e4SLinus Torvalds char *s = nd_get_link(nd); 546cc314eefSLinus Torvalds error = 0; 5471da177e4SLinus Torvalds if (s) 5481da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 54939159de2SJeff Layton else if (nd->last_type == LAST_BIND) { 55039159de2SJeff Layton error = force_reval_path(&nd->path, nd); 55139159de2SJeff Layton if (error) 55239159de2SJeff Layton path_put(&nd->path); 55339159de2SJeff Layton } 5541da177e4SLinus Torvalds } 5551da177e4SLinus Torvalds return error; 5561da177e4SLinus Torvalds } 5571da177e4SLinus Torvalds 5581da177e4SLinus Torvalds /* 5591da177e4SLinus Torvalds * This limits recursive symlink follows to 8, while 5601da177e4SLinus Torvalds * limiting consecutive symlinks to 40. 5611da177e4SLinus Torvalds * 5621da177e4SLinus Torvalds * Without that kind of total limit, nasty chains of consecutive 5631da177e4SLinus Torvalds * symlinks can cause almost arbitrarily long lookups. 5641da177e4SLinus Torvalds */ 56590ebe565SAl Viro static inline int do_follow_link(struct path *path, struct nameidata *nd) 5661da177e4SLinus Torvalds { 567def4af30SAl Viro void *cookie; 5681da177e4SLinus Torvalds int err = -ELOOP; 5691da177e4SLinus Torvalds if (current->link_count >= MAX_NESTED_LINKS) 5701da177e4SLinus Torvalds goto loop; 5711da177e4SLinus Torvalds if (current->total_link_count >= 40) 5721da177e4SLinus Torvalds goto loop; 5731da177e4SLinus Torvalds BUG_ON(nd->depth >= MAX_NESTED_LINKS); 5741da177e4SLinus Torvalds cond_resched(); 57590ebe565SAl Viro err = security_inode_follow_link(path->dentry, nd); 5761da177e4SLinus Torvalds if (err) 5771da177e4SLinus Torvalds goto loop; 5781da177e4SLinus Torvalds current->link_count++; 5791da177e4SLinus Torvalds current->total_link_count++; 5801da177e4SLinus Torvalds nd->depth++; 581def4af30SAl Viro err = __do_follow_link(path, nd, &cookie); 582def4af30SAl Viro if (!IS_ERR(cookie) && path->dentry->d_inode->i_op->put_link) 583def4af30SAl Viro path->dentry->d_inode->i_op->put_link(path->dentry, nd, cookie); 584258fa999SAl Viro path_put(path); 5851da177e4SLinus Torvalds current->link_count--; 5861da177e4SLinus Torvalds nd->depth--; 5871da177e4SLinus Torvalds return err; 5881da177e4SLinus Torvalds loop: 5891d957f9bSJan Blunck path_put_conditional(path, nd); 5901d957f9bSJan Blunck path_put(&nd->path); 5911da177e4SLinus Torvalds return err; 5921da177e4SLinus Torvalds } 5931da177e4SLinus Torvalds 594bab77ebfSAl Viro int follow_up(struct path *path) 5951da177e4SLinus Torvalds { 5961da177e4SLinus Torvalds struct vfsmount *parent; 5971da177e4SLinus Torvalds struct dentry *mountpoint; 5981da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 599bab77ebfSAl Viro parent = path->mnt->mnt_parent; 600bab77ebfSAl Viro if (parent == path->mnt) { 6011da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 6021da177e4SLinus Torvalds return 0; 6031da177e4SLinus Torvalds } 6041da177e4SLinus Torvalds mntget(parent); 605bab77ebfSAl Viro mountpoint = dget(path->mnt->mnt_mountpoint); 6061da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 607bab77ebfSAl Viro dput(path->dentry); 608bab77ebfSAl Viro path->dentry = mountpoint; 609bab77ebfSAl Viro mntput(path->mnt); 610bab77ebfSAl Viro path->mnt = parent; 6111da177e4SLinus Torvalds return 1; 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds 6141da177e4SLinus Torvalds /* no need for dcache_lock, as serialization is taken care in 6151da177e4SLinus Torvalds * namespace.c 6161da177e4SLinus Torvalds */ 617463ffb2eSAl Viro static int __follow_mount(struct path *path) 618463ffb2eSAl Viro { 619463ffb2eSAl Viro int res = 0; 620463ffb2eSAl Viro while (d_mountpoint(path->dentry)) { 6211c755af4SAl Viro struct vfsmount *mounted = lookup_mnt(path); 622463ffb2eSAl Viro if (!mounted) 623463ffb2eSAl Viro break; 624463ffb2eSAl Viro dput(path->dentry); 625463ffb2eSAl Viro if (res) 626463ffb2eSAl Viro mntput(path->mnt); 627463ffb2eSAl Viro path->mnt = mounted; 628463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 629463ffb2eSAl Viro res = 1; 630463ffb2eSAl Viro } 631463ffb2eSAl Viro return res; 632463ffb2eSAl Viro } 633463ffb2eSAl Viro 63479ed0226SAl Viro static void follow_mount(struct path *path) 6351da177e4SLinus Torvalds { 63679ed0226SAl Viro while (d_mountpoint(path->dentry)) { 6371c755af4SAl Viro struct vfsmount *mounted = lookup_mnt(path); 6381da177e4SLinus Torvalds if (!mounted) 6391da177e4SLinus Torvalds break; 64079ed0226SAl Viro dput(path->dentry); 64179ed0226SAl Viro mntput(path->mnt); 64279ed0226SAl Viro path->mnt = mounted; 64379ed0226SAl Viro path->dentry = dget(mounted->mnt_root); 6441da177e4SLinus Torvalds } 6451da177e4SLinus Torvalds } 6461da177e4SLinus Torvalds 6471da177e4SLinus Torvalds /* no need for dcache_lock, as serialization is taken care in 6481da177e4SLinus Torvalds * namespace.c 6491da177e4SLinus Torvalds */ 6509393bd07SAl Viro int follow_down(struct path *path) 6511da177e4SLinus Torvalds { 6521da177e4SLinus Torvalds struct vfsmount *mounted; 6531da177e4SLinus Torvalds 6541c755af4SAl Viro mounted = lookup_mnt(path); 6551da177e4SLinus Torvalds if (mounted) { 6569393bd07SAl Viro dput(path->dentry); 6579393bd07SAl Viro mntput(path->mnt); 6589393bd07SAl Viro path->mnt = mounted; 6599393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 6601da177e4SLinus Torvalds return 1; 6611da177e4SLinus Torvalds } 6621da177e4SLinus Torvalds return 0; 6631da177e4SLinus Torvalds } 6641da177e4SLinus Torvalds 665f1662356SArjan van de Ven static __always_inline void follow_dotdot(struct nameidata *nd) 6661da177e4SLinus Torvalds { 6672a737871SAl Viro set_root(nd); 668e518ddb7SAndreas Mohr 6691da177e4SLinus Torvalds while(1) { 6704ac91378SJan Blunck struct dentry *old = nd->path.dentry; 6711da177e4SLinus Torvalds 6722a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 6732a737871SAl Viro nd->path.mnt == nd->root.mnt) { 6741da177e4SLinus Torvalds break; 6751da177e4SLinus Torvalds } 6764ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 6773088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 6783088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 6791da177e4SLinus Torvalds dput(old); 6801da177e4SLinus Torvalds break; 6811da177e4SLinus Torvalds } 6823088dd70SAl Viro if (!follow_up(&nd->path)) 6831da177e4SLinus Torvalds break; 6841da177e4SLinus Torvalds } 68579ed0226SAl Viro follow_mount(&nd->path); 6861da177e4SLinus Torvalds } 6871da177e4SLinus Torvalds 6881da177e4SLinus Torvalds /* 689baa03890SNick Piggin * Allocate a dentry with name and parent, and perform a parent 690baa03890SNick Piggin * directory ->lookup on it. Returns the new dentry, or ERR_PTR 691baa03890SNick Piggin * on error. parent->d_inode->i_mutex must be held. d_lookup must 692baa03890SNick Piggin * have verified that no child exists while under i_mutex. 693baa03890SNick Piggin */ 694baa03890SNick Piggin static struct dentry *d_alloc_and_lookup(struct dentry *parent, 695baa03890SNick Piggin struct qstr *name, struct nameidata *nd) 696baa03890SNick Piggin { 697baa03890SNick Piggin struct inode *inode = parent->d_inode; 698baa03890SNick Piggin struct dentry *dentry; 699baa03890SNick Piggin struct dentry *old; 700baa03890SNick Piggin 701baa03890SNick Piggin /* Don't create child dentry for a dead directory. */ 702baa03890SNick Piggin if (unlikely(IS_DEADDIR(inode))) 703baa03890SNick Piggin return ERR_PTR(-ENOENT); 704baa03890SNick Piggin 705baa03890SNick Piggin dentry = d_alloc(parent, name); 706baa03890SNick Piggin if (unlikely(!dentry)) 707baa03890SNick Piggin return ERR_PTR(-ENOMEM); 708baa03890SNick Piggin 709baa03890SNick Piggin old = inode->i_op->lookup(inode, dentry, nd); 710baa03890SNick Piggin if (unlikely(old)) { 711baa03890SNick Piggin dput(dentry); 712baa03890SNick Piggin dentry = old; 713baa03890SNick Piggin } 714baa03890SNick Piggin return dentry; 715baa03890SNick Piggin } 716baa03890SNick Piggin 717baa03890SNick Piggin /* 7181da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 7191da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 7201da177e4SLinus Torvalds * It _is_ time-critical. 7211da177e4SLinus Torvalds */ 7221da177e4SLinus Torvalds static int do_lookup(struct nameidata *nd, struct qstr *name, 7231da177e4SLinus Torvalds struct path *path) 7241da177e4SLinus Torvalds { 7254ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 7266e6b1bd1SAl Viro struct dentry *dentry, *parent; 7276e6b1bd1SAl Viro struct inode *dir; 7283cac260aSAl Viro /* 7293cac260aSAl Viro * See if the low-level filesystem might want 7303cac260aSAl Viro * to use its own hash.. 7313cac260aSAl Viro */ 7323cac260aSAl Viro if (nd->path.dentry->d_op && nd->path.dentry->d_op->d_hash) { 7333cac260aSAl Viro int err = nd->path.dentry->d_op->d_hash(nd->path.dentry, name); 7343cac260aSAl Viro if (err < 0) 7353cac260aSAl Viro return err; 7363cac260aSAl Viro } 7371da177e4SLinus Torvalds 738b04f784eSNick Piggin /* 739b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 740b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 741b04f784eSNick Piggin * do the non-racy lookup, below. 742b04f784eSNick Piggin */ 7433cac260aSAl Viro dentry = __d_lookup(nd->path.dentry, name); 7441da177e4SLinus Torvalds if (!dentry) 7451da177e4SLinus Torvalds goto need_lookup; 7462e2e88eaSNick Piggin found: 7471da177e4SLinus Torvalds if (dentry->d_op && dentry->d_op->d_revalidate) 7481da177e4SLinus Torvalds goto need_revalidate; 7491da177e4SLinus Torvalds done: 7501da177e4SLinus Torvalds path->mnt = mnt; 7511da177e4SLinus Torvalds path->dentry = dentry; 752634ee701SAl Viro __follow_mount(path); 7531da177e4SLinus Torvalds return 0; 7541da177e4SLinus Torvalds 7551da177e4SLinus Torvalds need_lookup: 7566e6b1bd1SAl Viro parent = nd->path.dentry; 7576e6b1bd1SAl Viro dir = parent->d_inode; 7586e6b1bd1SAl Viro 7596e6b1bd1SAl Viro mutex_lock(&dir->i_mutex); 7606e6b1bd1SAl Viro /* 7616e6b1bd1SAl Viro * First re-do the cached lookup just in case it was created 762b04f784eSNick Piggin * while we waited for the directory semaphore, or the first 763b04f784eSNick Piggin * lookup failed due to an unrelated rename. 7646e6b1bd1SAl Viro * 765b04f784eSNick Piggin * This could use version numbering or similar to avoid unnecessary 766b04f784eSNick Piggin * cache lookups, but then we'd have to do the first lookup in the 767b04f784eSNick Piggin * non-racy way. However in the common case here, everything should 768b04f784eSNick Piggin * be hot in cache, so would it be a big win? 7696e6b1bd1SAl Viro */ 7706e6b1bd1SAl Viro dentry = d_lookup(parent, name); 771baa03890SNick Piggin if (likely(!dentry)) { 772baa03890SNick Piggin dentry = d_alloc_and_lookup(parent, name, nd); 7736e6b1bd1SAl Viro mutex_unlock(&dir->i_mutex); 7746e6b1bd1SAl Viro if (IS_ERR(dentry)) 7756e6b1bd1SAl Viro goto fail; 7766e6b1bd1SAl Viro goto done; 7776e6b1bd1SAl Viro } 7786e6b1bd1SAl Viro /* 7796e6b1bd1SAl Viro * Uhhuh! Nasty case: the cache was re-populated while 7806e6b1bd1SAl Viro * we waited on the semaphore. Need to revalidate. 7816e6b1bd1SAl Viro */ 7826e6b1bd1SAl Viro mutex_unlock(&dir->i_mutex); 7832e2e88eaSNick Piggin goto found; 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds need_revalidate: 786bcdc5e01SIan Kent dentry = do_revalidate(dentry, nd); 787bcdc5e01SIan Kent if (!dentry) 7881da177e4SLinus Torvalds goto need_lookup; 789bcdc5e01SIan Kent if (IS_ERR(dentry)) 790bcdc5e01SIan Kent goto fail; 791bcdc5e01SIan Kent goto done; 7921da177e4SLinus Torvalds 7931da177e4SLinus Torvalds fail: 7941da177e4SLinus Torvalds return PTR_ERR(dentry); 7951da177e4SLinus Torvalds } 7961da177e4SLinus Torvalds 7971da177e4SLinus Torvalds /* 798ac278a9cSAl Viro * This is a temporary kludge to deal with "automount" symlinks; proper 799ac278a9cSAl Viro * solution is to trigger them on follow_mount(), so that do_lookup() 800ac278a9cSAl Viro * would DTRT. To be killed before 2.6.34-final. 801ac278a9cSAl Viro */ 802ac278a9cSAl Viro static inline int follow_on_final(struct inode *inode, unsigned lookup_flags) 803ac278a9cSAl Viro { 804ac278a9cSAl Viro return inode && unlikely(inode->i_op->follow_link) && 805ac278a9cSAl Viro ((lookup_flags & LOOKUP_FOLLOW) || S_ISDIR(inode->i_mode)); 806ac278a9cSAl Viro } 807ac278a9cSAl Viro 808ac278a9cSAl Viro /* 8091da177e4SLinus Torvalds * Name resolution. 810ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 811ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 8121da177e4SLinus Torvalds * 813ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 814ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 8151da177e4SLinus Torvalds */ 8166de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 8171da177e4SLinus Torvalds { 8181da177e4SLinus Torvalds struct path next; 8191da177e4SLinus Torvalds struct inode *inode; 8201da177e4SLinus Torvalds int err; 8211da177e4SLinus Torvalds unsigned int lookup_flags = nd->flags; 8221da177e4SLinus Torvalds 8231da177e4SLinus Torvalds while (*name=='/') 8241da177e4SLinus Torvalds name++; 8251da177e4SLinus Torvalds if (!*name) 8261da177e4SLinus Torvalds goto return_reval; 8271da177e4SLinus Torvalds 8284ac91378SJan Blunck inode = nd->path.dentry->d_inode; 8291da177e4SLinus Torvalds if (nd->depth) 830f55eab82STrond Myklebust lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE); 8311da177e4SLinus Torvalds 8321da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 8331da177e4SLinus Torvalds for(;;) { 8341da177e4SLinus Torvalds unsigned long hash; 8351da177e4SLinus Torvalds struct qstr this; 8361da177e4SLinus Torvalds unsigned int c; 8371da177e4SLinus Torvalds 838cdce5d6bSTrond Myklebust nd->flags |= LOOKUP_CONTINUE; 839b75b5086SAl Viro err = exec_permission(inode); 8401da177e4SLinus Torvalds if (err) 8411da177e4SLinus Torvalds break; 8421da177e4SLinus Torvalds 8431da177e4SLinus Torvalds this.name = name; 8441da177e4SLinus Torvalds c = *(const unsigned char *)name; 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds hash = init_name_hash(); 8471da177e4SLinus Torvalds do { 8481da177e4SLinus Torvalds name++; 8491da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 8501da177e4SLinus Torvalds c = *(const unsigned char *)name; 8511da177e4SLinus Torvalds } while (c && (c != '/')); 8521da177e4SLinus Torvalds this.len = name - (const char *) this.name; 8531da177e4SLinus Torvalds this.hash = end_name_hash(hash); 8541da177e4SLinus Torvalds 8551da177e4SLinus Torvalds /* remove trailing slashes? */ 8561da177e4SLinus Torvalds if (!c) 8571da177e4SLinus Torvalds goto last_component; 8581da177e4SLinus Torvalds while (*++name == '/'); 8591da177e4SLinus Torvalds if (!*name) 8601da177e4SLinus Torvalds goto last_with_slashes; 8611da177e4SLinus Torvalds 8621da177e4SLinus Torvalds /* 8631da177e4SLinus Torvalds * "." and ".." are special - ".." especially so because it has 8641da177e4SLinus Torvalds * to be able to know about the current root directory and 8651da177e4SLinus Torvalds * parent relationships. 8661da177e4SLinus Torvalds */ 8671da177e4SLinus Torvalds if (this.name[0] == '.') switch (this.len) { 8681da177e4SLinus Torvalds default: 8691da177e4SLinus Torvalds break; 8701da177e4SLinus Torvalds case 2: 8711da177e4SLinus Torvalds if (this.name[1] != '.') 8721da177e4SLinus Torvalds break; 87358c465ebSAl Viro follow_dotdot(nd); 8744ac91378SJan Blunck inode = nd->path.dentry->d_inode; 8751da177e4SLinus Torvalds /* fallthrough */ 8761da177e4SLinus Torvalds case 1: 8771da177e4SLinus Torvalds continue; 8781da177e4SLinus Torvalds } 8791da177e4SLinus Torvalds /* This does the actual lookups.. */ 8801da177e4SLinus Torvalds err = do_lookup(nd, &this, &next); 8811da177e4SLinus Torvalds if (err) 8821da177e4SLinus Torvalds break; 8831da177e4SLinus Torvalds 8841da177e4SLinus Torvalds err = -ENOENT; 8851da177e4SLinus Torvalds inode = next.dentry->d_inode; 8861da177e4SLinus Torvalds if (!inode) 8871da177e4SLinus Torvalds goto out_dput; 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds if (inode->i_op->follow_link) { 89090ebe565SAl Viro err = do_follow_link(&next, nd); 8911da177e4SLinus Torvalds if (err) 8921da177e4SLinus Torvalds goto return_err; 8931da177e4SLinus Torvalds err = -ENOENT; 8944ac91378SJan Blunck inode = nd->path.dentry->d_inode; 8951da177e4SLinus Torvalds if (!inode) 8961da177e4SLinus Torvalds break; 89709dd17d3SMiklos Szeredi } else 89809dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 8991da177e4SLinus Torvalds err = -ENOTDIR; 9001da177e4SLinus Torvalds if (!inode->i_op->lookup) 9011da177e4SLinus Torvalds break; 9021da177e4SLinus Torvalds continue; 9031da177e4SLinus Torvalds /* here ends the main loop */ 9041da177e4SLinus Torvalds 9051da177e4SLinus Torvalds last_with_slashes: 9061da177e4SLinus Torvalds lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 9071da177e4SLinus Torvalds last_component: 908f55eab82STrond Myklebust /* Clear LOOKUP_CONTINUE iff it was previously unset */ 909f55eab82STrond Myklebust nd->flags &= lookup_flags | ~LOOKUP_CONTINUE; 9101da177e4SLinus Torvalds if (lookup_flags & LOOKUP_PARENT) 9111da177e4SLinus Torvalds goto lookup_parent; 9121da177e4SLinus Torvalds if (this.name[0] == '.') switch (this.len) { 9131da177e4SLinus Torvalds default: 9141da177e4SLinus Torvalds break; 9151da177e4SLinus Torvalds case 2: 9161da177e4SLinus Torvalds if (this.name[1] != '.') 9171da177e4SLinus Torvalds break; 91858c465ebSAl Viro follow_dotdot(nd); 9194ac91378SJan Blunck inode = nd->path.dentry->d_inode; 9201da177e4SLinus Torvalds /* fallthrough */ 9211da177e4SLinus Torvalds case 1: 9221da177e4SLinus Torvalds goto return_reval; 9231da177e4SLinus Torvalds } 9241da177e4SLinus Torvalds err = do_lookup(nd, &this, &next); 9251da177e4SLinus Torvalds if (err) 9261da177e4SLinus Torvalds break; 9271da177e4SLinus Torvalds inode = next.dentry->d_inode; 928ac278a9cSAl Viro if (follow_on_final(inode, lookup_flags)) { 92990ebe565SAl Viro err = do_follow_link(&next, nd); 9301da177e4SLinus Torvalds if (err) 9311da177e4SLinus Torvalds goto return_err; 9324ac91378SJan Blunck inode = nd->path.dentry->d_inode; 93309dd17d3SMiklos Szeredi } else 93409dd17d3SMiklos Szeredi path_to_nameidata(&next, nd); 9351da177e4SLinus Torvalds err = -ENOENT; 9361da177e4SLinus Torvalds if (!inode) 9371da177e4SLinus Torvalds break; 9381da177e4SLinus Torvalds if (lookup_flags & LOOKUP_DIRECTORY) { 9391da177e4SLinus Torvalds err = -ENOTDIR; 940acfa4380SAl Viro if (!inode->i_op->lookup) 9411da177e4SLinus Torvalds break; 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds goto return_base; 9441da177e4SLinus Torvalds lookup_parent: 9451da177e4SLinus Torvalds nd->last = this; 9461da177e4SLinus Torvalds nd->last_type = LAST_NORM; 9471da177e4SLinus Torvalds if (this.name[0] != '.') 9481da177e4SLinus Torvalds goto return_base; 9491da177e4SLinus Torvalds if (this.len == 1) 9501da177e4SLinus Torvalds nd->last_type = LAST_DOT; 9511da177e4SLinus Torvalds else if (this.len == 2 && this.name[1] == '.') 9521da177e4SLinus Torvalds nd->last_type = LAST_DOTDOT; 9531da177e4SLinus Torvalds else 9541da177e4SLinus Torvalds goto return_base; 9551da177e4SLinus Torvalds return_reval: 9561da177e4SLinus Torvalds /* 9571da177e4SLinus Torvalds * We bypassed the ordinary revalidation routines. 9581da177e4SLinus Torvalds * We may need to check the cached dentry for staleness. 9591da177e4SLinus Torvalds */ 9604ac91378SJan Blunck if (nd->path.dentry && nd->path.dentry->d_sb && 9614ac91378SJan Blunck (nd->path.dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) { 9621da177e4SLinus Torvalds err = -ESTALE; 9631da177e4SLinus Torvalds /* Note: we do not d_invalidate() */ 9644ac91378SJan Blunck if (!nd->path.dentry->d_op->d_revalidate( 9654ac91378SJan Blunck nd->path.dentry, nd)) 9661da177e4SLinus Torvalds break; 9671da177e4SLinus Torvalds } 9681da177e4SLinus Torvalds return_base: 9691da177e4SLinus Torvalds return 0; 9701da177e4SLinus Torvalds out_dput: 9711d957f9bSJan Blunck path_put_conditional(&next, nd); 9721da177e4SLinus Torvalds break; 9731da177e4SLinus Torvalds } 9741d957f9bSJan Blunck path_put(&nd->path); 9751da177e4SLinus Torvalds return_err: 9761da177e4SLinus Torvalds return err; 9771da177e4SLinus Torvalds } 9781da177e4SLinus Torvalds 979fc9b52cdSHarvey Harrison static int path_walk(const char *name, struct nameidata *nd) 9801da177e4SLinus Torvalds { 9816de88d72SAl Viro struct path save = nd->path; 9826de88d72SAl Viro int result; 9836de88d72SAl Viro 9841da177e4SLinus Torvalds current->total_link_count = 0; 9856de88d72SAl Viro 9866de88d72SAl Viro /* make sure the stuff we saved doesn't go away */ 9876de88d72SAl Viro path_get(&save); 9886de88d72SAl Viro 9896de88d72SAl Viro result = link_path_walk(name, nd); 9906de88d72SAl Viro if (result == -ESTALE) { 9916de88d72SAl Viro /* nd->path had been dropped */ 9926de88d72SAl Viro current->total_link_count = 0; 9936de88d72SAl Viro nd->path = save; 9946de88d72SAl Viro path_get(&nd->path); 9956de88d72SAl Viro nd->flags |= LOOKUP_REVAL; 9966de88d72SAl Viro result = link_path_walk(name, nd); 9976de88d72SAl Viro } 9986de88d72SAl Viro 9996de88d72SAl Viro path_put(&save); 10006de88d72SAl Viro 10016de88d72SAl Viro return result; 10021da177e4SLinus Torvalds } 10031da177e4SLinus Torvalds 10049b4a9b14SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, struct nameidata *nd) 10051da177e4SLinus Torvalds { 1006ea3834d9SPrasanna Meda int retval = 0; 1007170aa3d0SUlrich Drepper int fput_needed; 1008170aa3d0SUlrich Drepper struct file *file; 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds nd->last_type = LAST_ROOT; /* if there are only slashes... */ 10111da177e4SLinus Torvalds nd->flags = flags; 10121da177e4SLinus Torvalds nd->depth = 0; 10132a737871SAl Viro nd->root.mnt = NULL; 10141da177e4SLinus Torvalds 10151da177e4SLinus Torvalds if (*name=='/') { 10162a737871SAl Viro set_root(nd); 10172a737871SAl Viro nd->path = nd->root; 10182a737871SAl Viro path_get(&nd->root); 10195590ff0dSUlrich Drepper } else if (dfd == AT_FDCWD) { 1020f7ad3c6bSMiklos Szeredi get_fs_pwd(current->fs, &nd->path); 10215590ff0dSUlrich Drepper } else { 10225590ff0dSUlrich Drepper struct dentry *dentry; 10235590ff0dSUlrich Drepper 10245590ff0dSUlrich Drepper file = fget_light(dfd, &fput_needed); 10255590ff0dSUlrich Drepper retval = -EBADF; 1026170aa3d0SUlrich Drepper if (!file) 10276d09bb62STrond Myklebust goto out_fail; 10285590ff0dSUlrich Drepper 10290f7fc9e4SJosef "Jeff" Sipek dentry = file->f_path.dentry; 10305590ff0dSUlrich Drepper 10315590ff0dSUlrich Drepper retval = -ENOTDIR; 1032170aa3d0SUlrich Drepper if (!S_ISDIR(dentry->d_inode->i_mode)) 10336d09bb62STrond Myklebust goto fput_fail; 10345590ff0dSUlrich Drepper 10355590ff0dSUlrich Drepper retval = file_permission(file, MAY_EXEC); 1036170aa3d0SUlrich Drepper if (retval) 10376d09bb62STrond Myklebust goto fput_fail; 10385590ff0dSUlrich Drepper 10395dd784d0SJan Blunck nd->path = file->f_path; 10405dd784d0SJan Blunck path_get(&file->f_path); 10415590ff0dSUlrich Drepper 10425590ff0dSUlrich Drepper fput_light(file, fput_needed); 10431da177e4SLinus Torvalds } 10449b4a9b14SAl Viro return 0; 10452dfdd266SJosef 'Jeff' Sipek 10469b4a9b14SAl Viro fput_fail: 10479b4a9b14SAl Viro fput_light(file, fput_needed); 10489b4a9b14SAl Viro out_fail: 10499b4a9b14SAl Viro return retval; 10509b4a9b14SAl Viro } 10519b4a9b14SAl Viro 10529b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 10539b4a9b14SAl Viro static int do_path_lookup(int dfd, const char *name, 10549b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 10559b4a9b14SAl Viro { 10569b4a9b14SAl Viro int retval = path_init(dfd, name, flags, nd); 10579b4a9b14SAl Viro if (!retval) 10582dfdd266SJosef 'Jeff' Sipek retval = path_walk(name, nd); 10594ac91378SJan Blunck if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry && 10604ac91378SJan Blunck nd->path.dentry->d_inode)) 10614ac91378SJan Blunck audit_inode(name, nd->path.dentry); 10622a737871SAl Viro if (nd->root.mnt) { 10632a737871SAl Viro path_put(&nd->root); 10642a737871SAl Viro nd->root.mnt = NULL; 10652a737871SAl Viro } 1066170aa3d0SUlrich Drepper return retval; 10671da177e4SLinus Torvalds } 10681da177e4SLinus Torvalds 1069fc9b52cdSHarvey Harrison int path_lookup(const char *name, unsigned int flags, 10705590ff0dSUlrich Drepper struct nameidata *nd) 10715590ff0dSUlrich Drepper { 10725590ff0dSUlrich Drepper return do_path_lookup(AT_FDCWD, name, flags, nd); 10735590ff0dSUlrich Drepper } 10745590ff0dSUlrich Drepper 1075d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1076d1811465SAl Viro { 1077d1811465SAl Viro struct nameidata nd; 1078d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1079d1811465SAl Viro if (!res) 1080d1811465SAl Viro *path = nd.path; 1081d1811465SAl Viro return res; 1082d1811465SAl Viro } 1083d1811465SAl Viro 108416f18200SJosef 'Jeff' Sipek /** 108516f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 108616f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 108716f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 108816f18200SJosef 'Jeff' Sipek * @name: pointer to file name 108916f18200SJosef 'Jeff' Sipek * @flags: lookup flags 109016f18200SJosef 'Jeff' Sipek * @nd: pointer to nameidata 109116f18200SJosef 'Jeff' Sipek */ 109216f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 109316f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 109416f18200SJosef 'Jeff' Sipek struct nameidata *nd) 109516f18200SJosef 'Jeff' Sipek { 109616f18200SJosef 'Jeff' Sipek int retval; 109716f18200SJosef 'Jeff' Sipek 109816f18200SJosef 'Jeff' Sipek /* same as do_path_lookup */ 109916f18200SJosef 'Jeff' Sipek nd->last_type = LAST_ROOT; 110016f18200SJosef 'Jeff' Sipek nd->flags = flags; 110116f18200SJosef 'Jeff' Sipek nd->depth = 0; 110216f18200SJosef 'Jeff' Sipek 1103c8e7f449SJan Blunck nd->path.dentry = dentry; 1104c8e7f449SJan Blunck nd->path.mnt = mnt; 1105c8e7f449SJan Blunck path_get(&nd->path); 11065b857119SAl Viro nd->root = nd->path; 11075b857119SAl Viro path_get(&nd->root); 110816f18200SJosef 'Jeff' Sipek 110916f18200SJosef 'Jeff' Sipek retval = path_walk(name, nd); 11104ac91378SJan Blunck if (unlikely(!retval && !audit_dummy_context() && nd->path.dentry && 11114ac91378SJan Blunck nd->path.dentry->d_inode)) 11124ac91378SJan Blunck audit_inode(name, nd->path.dentry); 111316f18200SJosef 'Jeff' Sipek 11142a737871SAl Viro path_put(&nd->root); 11152a737871SAl Viro nd->root.mnt = NULL; 111616f18200SJosef 'Jeff' Sipek 11172a737871SAl Viro return retval; 111816f18200SJosef 'Jeff' Sipek } 111916f18200SJosef 'Jeff' Sipek 1120eead1911SChristoph Hellwig static struct dentry *__lookup_hash(struct qstr *name, 1121eead1911SChristoph Hellwig struct dentry *base, struct nameidata *nd) 11221da177e4SLinus Torvalds { 11231da177e4SLinus Torvalds struct dentry *dentry; 11241da177e4SLinus Torvalds struct inode *inode; 11251da177e4SLinus Torvalds int err; 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds inode = base->d_inode; 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds /* 11301da177e4SLinus Torvalds * See if the low-level filesystem might want 11311da177e4SLinus Torvalds * to use its own hash.. 11321da177e4SLinus Torvalds */ 11331da177e4SLinus Torvalds if (base->d_op && base->d_op->d_hash) { 11341da177e4SLinus Torvalds err = base->d_op->d_hash(base, name); 11351da177e4SLinus Torvalds dentry = ERR_PTR(err); 11361da177e4SLinus Torvalds if (err < 0) 11371da177e4SLinus Torvalds goto out; 11381da177e4SLinus Torvalds } 11391da177e4SLinus Torvalds 1140b04f784eSNick Piggin /* 1141b04f784eSNick Piggin * Don't bother with __d_lookup: callers are for creat as 1142b04f784eSNick Piggin * well as unlink, so a lot of the time it would cost 1143b04f784eSNick Piggin * a double lookup. 11446e6b1bd1SAl Viro */ 11456e6b1bd1SAl Viro dentry = d_lookup(base, name); 11466e6b1bd1SAl Viro 11476e6b1bd1SAl Viro if (dentry && dentry->d_op && dentry->d_op->d_revalidate) 11486e6b1bd1SAl Viro dentry = do_revalidate(dentry, nd); 11496e6b1bd1SAl Viro 11501da177e4SLinus Torvalds if (!dentry) 1151baa03890SNick Piggin dentry = d_alloc_and_lookup(base, name, nd); 11521da177e4SLinus Torvalds out: 11531da177e4SLinus Torvalds return dentry; 11541da177e4SLinus Torvalds } 11551da177e4SLinus Torvalds 1156057f6c01SJames Morris /* 1157057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1158057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1159057f6c01SJames Morris * SMP-safe. 1160057f6c01SJames Morris */ 1161a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 11621da177e4SLinus Torvalds { 1163eead1911SChristoph Hellwig int err; 1164eead1911SChristoph Hellwig 1165b75b5086SAl Viro err = exec_permission(nd->path.dentry->d_inode); 1166eead1911SChristoph Hellwig if (err) 1167eead1911SChristoph Hellwig return ERR_PTR(err); 11684ac91378SJan Blunck return __lookup_hash(&nd->last, nd->path.dentry, nd); 11691da177e4SLinus Torvalds } 11701da177e4SLinus Torvalds 1171eead1911SChristoph Hellwig static int __lookup_one_len(const char *name, struct qstr *this, 1172eead1911SChristoph Hellwig struct dentry *base, int len) 11731da177e4SLinus Torvalds { 11741da177e4SLinus Torvalds unsigned long hash; 11751da177e4SLinus Torvalds unsigned int c; 11761da177e4SLinus Torvalds 1177057f6c01SJames Morris this->name = name; 1178057f6c01SJames Morris this->len = len; 11791da177e4SLinus Torvalds if (!len) 1180057f6c01SJames Morris return -EACCES; 11811da177e4SLinus Torvalds 11821da177e4SLinus Torvalds hash = init_name_hash(); 11831da177e4SLinus Torvalds while (len--) { 11841da177e4SLinus Torvalds c = *(const unsigned char *)name++; 11851da177e4SLinus Torvalds if (c == '/' || c == '\0') 1186057f6c01SJames Morris return -EACCES; 11871da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 11881da177e4SLinus Torvalds } 1189057f6c01SJames Morris this->hash = end_name_hash(hash); 1190057f6c01SJames Morris return 0; 1191057f6c01SJames Morris } 11921da177e4SLinus Torvalds 1193eead1911SChristoph Hellwig /** 1194a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1195eead1911SChristoph Hellwig * @name: pathname component to lookup 1196eead1911SChristoph Hellwig * @base: base directory to lookup from 1197eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1198eead1911SChristoph Hellwig * 1199a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1200a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1201eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1202eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1203eead1911SChristoph Hellwig */ 1204057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1205057f6c01SJames Morris { 1206057f6c01SJames Morris int err; 1207057f6c01SJames Morris struct qstr this; 1208057f6c01SJames Morris 12092f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 12102f9092e1SDavid Woodhouse 1211057f6c01SJames Morris err = __lookup_one_len(name, &this, base, len); 1212057f6c01SJames Morris if (err) 1213057f6c01SJames Morris return ERR_PTR(err); 1214eead1911SChristoph Hellwig 1215b75b5086SAl Viro err = exec_permission(base->d_inode); 1216eead1911SChristoph Hellwig if (err) 1217eead1911SChristoph Hellwig return ERR_PTR(err); 121849705b77SChristoph Hellwig return __lookup_hash(&this, base, NULL); 1219057f6c01SJames Morris } 1220057f6c01SJames Morris 12212d8f3038SAl Viro int user_path_at(int dfd, const char __user *name, unsigned flags, 12222d8f3038SAl Viro struct path *path) 12231da177e4SLinus Torvalds { 12242d8f3038SAl Viro struct nameidata nd; 12251da177e4SLinus Torvalds char *tmp = getname(name); 12261da177e4SLinus Torvalds int err = PTR_ERR(tmp); 12271da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 12282d8f3038SAl Viro 12292d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 12302d8f3038SAl Viro 12312d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 12321da177e4SLinus Torvalds putname(tmp); 12332d8f3038SAl Viro if (!err) 12342d8f3038SAl Viro *path = nd.path; 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds return err; 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds 12392ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 12402ad94ae6SAl Viro struct nameidata *nd, char **name) 12412ad94ae6SAl Viro { 12422ad94ae6SAl Viro char *s = getname(path); 12432ad94ae6SAl Viro int error; 12442ad94ae6SAl Viro 12452ad94ae6SAl Viro if (IS_ERR(s)) 12462ad94ae6SAl Viro return PTR_ERR(s); 12472ad94ae6SAl Viro 12482ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 12492ad94ae6SAl Viro if (error) 12502ad94ae6SAl Viro putname(s); 12512ad94ae6SAl Viro else 12522ad94ae6SAl Viro *name = s; 12532ad94ae6SAl Viro 12542ad94ae6SAl Viro return error; 12552ad94ae6SAl Viro } 12562ad94ae6SAl Viro 12571da177e4SLinus Torvalds /* 12581da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 12591da177e4SLinus Torvalds * minimal. 12601da177e4SLinus Torvalds */ 12611da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 12621da177e4SLinus Torvalds { 1263da9592edSDavid Howells uid_t fsuid = current_fsuid(); 1264da9592edSDavid Howells 12651da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 12661da177e4SLinus Torvalds return 0; 1267da9592edSDavid Howells if (inode->i_uid == fsuid) 12681da177e4SLinus Torvalds return 0; 1269da9592edSDavid Howells if (dir->i_uid == fsuid) 12701da177e4SLinus Torvalds return 0; 12711da177e4SLinus Torvalds return !capable(CAP_FOWNER); 12721da177e4SLinus Torvalds } 12731da177e4SLinus Torvalds 12741da177e4SLinus Torvalds /* 12751da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 12761da177e4SLinus Torvalds * whether the type of victim is right. 12771da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 12781da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 12791da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 12801da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 12811da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 12821da177e4SLinus Torvalds * a. be owner of dir, or 12831da177e4SLinus Torvalds * b. be owner of victim, or 12841da177e4SLinus Torvalds * c. have CAP_FOWNER capability 12851da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 12861da177e4SLinus Torvalds * links pointing to it. 12871da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 12881da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 12891da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 12901da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 12911da177e4SLinus Torvalds * nfs_async_unlink(). 12921da177e4SLinus Torvalds */ 1293858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 12941da177e4SLinus Torvalds { 12951da177e4SLinus Torvalds int error; 12961da177e4SLinus Torvalds 12971da177e4SLinus Torvalds if (!victim->d_inode) 12981da177e4SLinus Torvalds return -ENOENT; 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 1301cccc6bbaSAl Viro audit_inode_child(victim, dir); 13021da177e4SLinus Torvalds 1303f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 13041da177e4SLinus Torvalds if (error) 13051da177e4SLinus Torvalds return error; 13061da177e4SLinus Torvalds if (IS_APPEND(dir)) 13071da177e4SLinus Torvalds return -EPERM; 13081da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 1309f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 13101da177e4SLinus Torvalds return -EPERM; 13111da177e4SLinus Torvalds if (isdir) { 13121da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 13131da177e4SLinus Torvalds return -ENOTDIR; 13141da177e4SLinus Torvalds if (IS_ROOT(victim)) 13151da177e4SLinus Torvalds return -EBUSY; 13161da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 13171da177e4SLinus Torvalds return -EISDIR; 13181da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 13191da177e4SLinus Torvalds return -ENOENT; 13201da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 13211da177e4SLinus Torvalds return -EBUSY; 13221da177e4SLinus Torvalds return 0; 13231da177e4SLinus Torvalds } 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 13261da177e4SLinus Torvalds * dir. 13271da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 13281da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 13291da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 13301da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 13311da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 13321da177e4SLinus Torvalds */ 1333a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 13341da177e4SLinus Torvalds { 13351da177e4SLinus Torvalds if (child->d_inode) 13361da177e4SLinus Torvalds return -EEXIST; 13371da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 13381da177e4SLinus Torvalds return -ENOENT; 1339f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 13401da177e4SLinus Torvalds } 13411da177e4SLinus Torvalds 13421da177e4SLinus Torvalds /* 13431da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 13441da177e4SLinus Torvalds */ 13451da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 13461da177e4SLinus Torvalds { 13471da177e4SLinus Torvalds struct dentry *p; 13481da177e4SLinus Torvalds 13491da177e4SLinus Torvalds if (p1 == p2) { 1350f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 13511da177e4SLinus Torvalds return NULL; 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds 1354a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 13551da177e4SLinus Torvalds 1356e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 1357e2761a11SOGAWA Hirofumi if (p) { 1358f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1359f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 13601da177e4SLinus Torvalds return p; 13611da177e4SLinus Torvalds } 13621da177e4SLinus Torvalds 1363e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 1364e2761a11SOGAWA Hirofumi if (p) { 1365f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1366f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 13671da177e4SLinus Torvalds return p; 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 1370f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1371f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 13721da177e4SLinus Torvalds return NULL; 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 13761da177e4SLinus Torvalds { 13771b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 13781da177e4SLinus Torvalds if (p1 != p2) { 13791b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 1380a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 13811da177e4SLinus Torvalds } 13821da177e4SLinus Torvalds } 13831da177e4SLinus Torvalds 13841da177e4SLinus Torvalds int vfs_create(struct inode *dir, struct dentry *dentry, int mode, 13851da177e4SLinus Torvalds struct nameidata *nd) 13861da177e4SLinus Torvalds { 1387a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 13881da177e4SLinus Torvalds 13891da177e4SLinus Torvalds if (error) 13901da177e4SLinus Torvalds return error; 13911da177e4SLinus Torvalds 1392acfa4380SAl Viro if (!dir->i_op->create) 13931da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 13941da177e4SLinus Torvalds mode &= S_IALLUGO; 13951da177e4SLinus Torvalds mode |= S_IFREG; 13961da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 13971da177e4SLinus Torvalds if (error) 13981da177e4SLinus Torvalds return error; 13991da177e4SLinus Torvalds error = dir->i_op->create(dir, dentry, mode, nd); 1400a74574aaSStephen Smalley if (!error) 1401f38aa942SAmy Griffis fsnotify_create(dir, dentry); 14021da177e4SLinus Torvalds return error; 14031da177e4SLinus Torvalds } 14041da177e4SLinus Torvalds 14053fb64190SChristoph Hellwig int may_open(struct path *path, int acc_mode, int flag) 14061da177e4SLinus Torvalds { 14073fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 14081da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 14091da177e4SLinus Torvalds int error; 14101da177e4SLinus Torvalds 14111da177e4SLinus Torvalds if (!inode) 14121da177e4SLinus Torvalds return -ENOENT; 14131da177e4SLinus Torvalds 1414c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 1415c8fe8f30SChristoph Hellwig case S_IFLNK: 14161da177e4SLinus Torvalds return -ELOOP; 1417c8fe8f30SChristoph Hellwig case S_IFDIR: 1418c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 14191da177e4SLinus Torvalds return -EISDIR; 1420c8fe8f30SChristoph Hellwig break; 1421c8fe8f30SChristoph Hellwig case S_IFBLK: 1422c8fe8f30SChristoph Hellwig case S_IFCHR: 14233fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 14241da177e4SLinus Torvalds return -EACCES; 1425c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 1426c8fe8f30SChristoph Hellwig case S_IFIFO: 1427c8fe8f30SChristoph Hellwig case S_IFSOCK: 14281da177e4SLinus Torvalds flag &= ~O_TRUNC; 1429c8fe8f30SChristoph Hellwig break; 14304a3fd211SDave Hansen } 1431b41572e9SDave Hansen 14323fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 1433b41572e9SDave Hansen if (error) 1434b41572e9SDave Hansen return error; 14356146f0d5SMimi Zohar 14361da177e4SLinus Torvalds /* 14371da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 14381da177e4SLinus Torvalds */ 14391da177e4SLinus Torvalds if (IS_APPEND(inode)) { 14408737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 14417715b521SAl Viro return -EPERM; 14421da177e4SLinus Torvalds if (flag & O_TRUNC) 14437715b521SAl Viro return -EPERM; 14441da177e4SLinus Torvalds } 14451da177e4SLinus Torvalds 14461da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 14477715b521SAl Viro if (flag & O_NOATIME && !is_owner_or_cap(inode)) 14487715b521SAl Viro return -EPERM; 14491da177e4SLinus Torvalds 14501da177e4SLinus Torvalds /* 14511da177e4SLinus Torvalds * Ensure there are no outstanding leases on the file. 14521da177e4SLinus Torvalds */ 1453b65a9cfcSAl Viro return break_lease(inode, flag); 14547715b521SAl Viro } 14557715b521SAl Viro 14567715b521SAl Viro static int handle_truncate(struct path *path) 14577715b521SAl Viro { 14587715b521SAl Viro struct inode *inode = path->dentry->d_inode; 14597715b521SAl Viro int error = get_write_access(inode); 14601da177e4SLinus Torvalds if (error) 14617715b521SAl Viro return error; 14621da177e4SLinus Torvalds /* 14631da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 14641da177e4SLinus Torvalds */ 14651da177e4SLinus Torvalds error = locks_verify_locked(inode); 1466be6d3e56SKentaro Takeda if (!error) 1467ea0d3ab2STetsuo Handa error = security_path_truncate(path); 14681da177e4SLinus Torvalds if (!error) { 14697715b521SAl Viro error = do_truncate(path->dentry, 0, 1470d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 1471d139d7ffSMiklos Szeredi NULL); 14721da177e4SLinus Torvalds } 14731da177e4SLinus Torvalds put_write_access(inode); 1474acd0c935SMimi Zohar return error; 14751da177e4SLinus Torvalds } 14761da177e4SLinus Torvalds 1477d57999e1SDave Hansen /* 1478d57999e1SDave Hansen * Be careful about ever adding any more callers of this 1479d57999e1SDave Hansen * function. Its flags must be in the namei format, not 1480d57999e1SDave Hansen * what get passed to sys_open(). 1481d57999e1SDave Hansen */ 1482d57999e1SDave Hansen static int __open_namei_create(struct nameidata *nd, struct path *path, 14838737c930SAl Viro int open_flag, int mode) 1484aab520e2SDave Hansen { 1485aab520e2SDave Hansen int error; 14864ac91378SJan Blunck struct dentry *dir = nd->path.dentry; 1487aab520e2SDave Hansen 1488aab520e2SDave Hansen if (!IS_POSIXACL(dir->d_inode)) 1489ce3b0f8dSAl Viro mode &= ~current_umask(); 1490be6d3e56SKentaro Takeda error = security_path_mknod(&nd->path, path->dentry, mode, 0); 1491be6d3e56SKentaro Takeda if (error) 1492be6d3e56SKentaro Takeda goto out_unlock; 1493aab520e2SDave Hansen error = vfs_create(dir->d_inode, path->dentry, mode, nd); 1494be6d3e56SKentaro Takeda out_unlock: 1495aab520e2SDave Hansen mutex_unlock(&dir->d_inode->i_mutex); 14964ac91378SJan Blunck dput(nd->path.dentry); 14974ac91378SJan Blunck nd->path.dentry = path->dentry; 1498aab520e2SDave Hansen if (error) 1499aab520e2SDave Hansen return error; 1500aab520e2SDave Hansen /* Don't check for write permission, don't truncate */ 15018737c930SAl Viro return may_open(&nd->path, 0, open_flag & ~O_TRUNC); 1502aab520e2SDave Hansen } 1503aab520e2SDave Hansen 15041da177e4SLinus Torvalds /* 1505d57999e1SDave Hansen * Note that while the flag value (low two bits) for sys_open means: 1506d57999e1SDave Hansen * 00 - read-only 1507d57999e1SDave Hansen * 01 - write-only 1508d57999e1SDave Hansen * 10 - read-write 1509d57999e1SDave Hansen * 11 - special 1510d57999e1SDave Hansen * it is changed into 1511d57999e1SDave Hansen * 00 - no permissions needed 1512d57999e1SDave Hansen * 01 - read-permission 1513d57999e1SDave Hansen * 10 - write-permission 1514d57999e1SDave Hansen * 11 - read-write 1515d57999e1SDave Hansen * for the internal routines (ie open_namei()/follow_link() etc) 1516d57999e1SDave Hansen * This is more logical, and also allows the 00 "no perm needed" 1517d57999e1SDave Hansen * to be used for symlinks (where the permissions are checked 1518d57999e1SDave Hansen * later). 1519d57999e1SDave Hansen * 1520d57999e1SDave Hansen */ 1521d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 1522d57999e1SDave Hansen { 1523d57999e1SDave Hansen if ((flag+1) & O_ACCMODE) 1524d57999e1SDave Hansen flag++; 1525d57999e1SDave Hansen return flag; 1526d57999e1SDave Hansen } 1527d57999e1SDave Hansen 15287715b521SAl Viro static int open_will_truncate(int flag, struct inode *inode) 15294a3fd211SDave Hansen { 1530d57999e1SDave Hansen /* 15314a3fd211SDave Hansen * We'll never write to the fs underlying 15324a3fd211SDave Hansen * a device file. 15334a3fd211SDave Hansen */ 15344a3fd211SDave Hansen if (special_file(inode->i_mode)) 15354a3fd211SDave Hansen return 0; 15364a3fd211SDave Hansen return (flag & O_TRUNC); 15374a3fd211SDave Hansen } 15384a3fd211SDave Hansen 1539648fa861SAl Viro static struct file *finish_open(struct nameidata *nd, 15409a66179eSAl Viro int open_flag, int acc_mode) 1541648fa861SAl Viro { 1542648fa861SAl Viro struct file *filp; 1543648fa861SAl Viro int will_truncate; 1544648fa861SAl Viro int error; 1545648fa861SAl Viro 15469a66179eSAl Viro will_truncate = open_will_truncate(open_flag, nd->path.dentry->d_inode); 1547648fa861SAl Viro if (will_truncate) { 1548648fa861SAl Viro error = mnt_want_write(nd->path.mnt); 1549648fa861SAl Viro if (error) 1550648fa861SAl Viro goto exit; 1551648fa861SAl Viro } 1552648fa861SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 1553648fa861SAl Viro if (error) { 1554648fa861SAl Viro if (will_truncate) 1555648fa861SAl Viro mnt_drop_write(nd->path.mnt); 1556648fa861SAl Viro goto exit; 1557648fa861SAl Viro } 1558648fa861SAl Viro filp = nameidata_to_filp(nd); 1559648fa861SAl Viro if (!IS_ERR(filp)) { 1560648fa861SAl Viro error = ima_file_check(filp, acc_mode); 1561648fa861SAl Viro if (error) { 1562648fa861SAl Viro fput(filp); 1563648fa861SAl Viro filp = ERR_PTR(error); 1564648fa861SAl Viro } 1565648fa861SAl Viro } 1566648fa861SAl Viro if (!IS_ERR(filp)) { 1567648fa861SAl Viro if (will_truncate) { 1568648fa861SAl Viro error = handle_truncate(&nd->path); 1569648fa861SAl Viro if (error) { 1570648fa861SAl Viro fput(filp); 1571648fa861SAl Viro filp = ERR_PTR(error); 1572648fa861SAl Viro } 1573648fa861SAl Viro } 1574648fa861SAl Viro } 1575648fa861SAl Viro /* 1576648fa861SAl Viro * It is now safe to drop the mnt write 1577648fa861SAl Viro * because the filp has had a write taken 1578648fa861SAl Viro * on its behalf. 1579648fa861SAl Viro */ 1580648fa861SAl Viro if (will_truncate) 1581648fa861SAl Viro mnt_drop_write(nd->path.mnt); 1582648fa861SAl Viro return filp; 1583648fa861SAl Viro 1584648fa861SAl Viro exit: 1585648fa861SAl Viro if (!IS_ERR(nd->intent.open.file)) 1586648fa861SAl Viro release_open_intent(nd); 1587648fa861SAl Viro path_put(&nd->path); 1588648fa861SAl Viro return ERR_PTR(error); 1589648fa861SAl Viro } 1590648fa861SAl Viro 1591fb1cc555SAl Viro static struct file *do_last(struct nameidata *nd, struct path *path, 15925b369df8SAl Viro int open_flag, int acc_mode, 15933e297b61SAl Viro int mode, const char *pathname) 1594fb1cc555SAl Viro { 1595a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 1596fb1cc555SAl Viro struct file *filp; 15971f36f774SAl Viro int error = -EISDIR; 1598fb1cc555SAl Viro 15991f36f774SAl Viro switch (nd->last_type) { 16001f36f774SAl Viro case LAST_DOTDOT: 16011f36f774SAl Viro follow_dotdot(nd); 16021f36f774SAl Viro dir = nd->path.dentry; 1603176306f5SNeil Brown case LAST_DOT: 16041f36f774SAl Viro if (nd->path.mnt->mnt_sb->s_type->fs_flags & FS_REVAL_DOT) { 16051f36f774SAl Viro if (!dir->d_op->d_revalidate(dir, nd)) { 16061f36f774SAl Viro error = -ESTALE; 1607a2c36b45SAl Viro goto exit; 16081f36f774SAl Viro } 16091f36f774SAl Viro } 16101f36f774SAl Viro /* fallthrough */ 16111f36f774SAl Viro case LAST_ROOT: 16121f36f774SAl Viro if (open_flag & O_CREAT) 16131f36f774SAl Viro goto exit; 16141f36f774SAl Viro /* fallthrough */ 16151f36f774SAl Viro case LAST_BIND: 16161f36f774SAl Viro audit_inode(pathname, dir); 16171f36f774SAl Viro goto ok; 16181f36f774SAl Viro } 1619a2c36b45SAl Viro 16201f36f774SAl Viro /* trailing slashes? */ 16211f36f774SAl Viro if (nd->last.name[nd->last.len]) { 16221f36f774SAl Viro if (open_flag & O_CREAT) 16231f36f774SAl Viro goto exit; 1624002baeecSJan Kara nd->flags |= LOOKUP_DIRECTORY | LOOKUP_FOLLOW; 16251f36f774SAl Viro } 16261f36f774SAl Viro 16271f36f774SAl Viro /* just plain open? */ 16281f36f774SAl Viro if (!(open_flag & O_CREAT)) { 16291f36f774SAl Viro error = do_lookup(nd, &nd->last, path); 16301f36f774SAl Viro if (error) 16311f36f774SAl Viro goto exit; 16321f36f774SAl Viro error = -ENOENT; 16331f36f774SAl Viro if (!path->dentry->d_inode) 16341f36f774SAl Viro goto exit_dput; 16351f36f774SAl Viro if (path->dentry->d_inode->i_op->follow_link) 16361f36f774SAl Viro return NULL; 16371f36f774SAl Viro error = -ENOTDIR; 16383e297b61SAl Viro if (nd->flags & LOOKUP_DIRECTORY) { 16393e297b61SAl Viro if (!path->dentry->d_inode->i_op->lookup) 16401f36f774SAl Viro goto exit_dput; 16413e297b61SAl Viro } 16421f36f774SAl Viro path_to_nameidata(path, nd); 16431f36f774SAl Viro audit_inode(pathname, nd->path.dentry); 16441f36f774SAl Viro goto ok; 16451f36f774SAl Viro } 16461f36f774SAl Viro 16471f36f774SAl Viro /* OK, it's O_CREAT */ 1648a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 1649a1e28038SAl Viro 1650a1e28038SAl Viro path->dentry = lookup_hash(nd); 1651a1e28038SAl Viro path->mnt = nd->path.mnt; 1652a1e28038SAl Viro 1653fb1cc555SAl Viro error = PTR_ERR(path->dentry); 1654fb1cc555SAl Viro if (IS_ERR(path->dentry)) { 1655fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 1656fb1cc555SAl Viro goto exit; 1657fb1cc555SAl Viro } 1658fb1cc555SAl Viro 1659fb1cc555SAl Viro if (IS_ERR(nd->intent.open.file)) { 1660fb1cc555SAl Viro error = PTR_ERR(nd->intent.open.file); 1661fb1cc555SAl Viro goto exit_mutex_unlock; 1662fb1cc555SAl Viro } 1663fb1cc555SAl Viro 1664fb1cc555SAl Viro /* Negative dentry, just create the file */ 1665fb1cc555SAl Viro if (!path->dentry->d_inode) { 1666fb1cc555SAl Viro /* 1667fb1cc555SAl Viro * This write is needed to ensure that a 1668fb1cc555SAl Viro * ro->rw transition does not occur between 1669fb1cc555SAl Viro * the time when the file is created and when 1670fb1cc555SAl Viro * a permanent write count is taken through 1671fb1cc555SAl Viro * the 'struct file' in nameidata_to_filp(). 1672fb1cc555SAl Viro */ 1673fb1cc555SAl Viro error = mnt_want_write(nd->path.mnt); 1674fb1cc555SAl Viro if (error) 1675fb1cc555SAl Viro goto exit_mutex_unlock; 1676fb1cc555SAl Viro error = __open_namei_create(nd, path, open_flag, mode); 1677fb1cc555SAl Viro if (error) { 1678fb1cc555SAl Viro mnt_drop_write(nd->path.mnt); 1679fb1cc555SAl Viro goto exit; 1680fb1cc555SAl Viro } 1681fb1cc555SAl Viro filp = nameidata_to_filp(nd); 1682fb1cc555SAl Viro mnt_drop_write(nd->path.mnt); 1683fb1cc555SAl Viro if (!IS_ERR(filp)) { 1684fb1cc555SAl Viro error = ima_file_check(filp, acc_mode); 1685fb1cc555SAl Viro if (error) { 1686fb1cc555SAl Viro fput(filp); 1687fb1cc555SAl Viro filp = ERR_PTR(error); 1688fb1cc555SAl Viro } 1689fb1cc555SAl Viro } 1690fb1cc555SAl Viro return filp; 1691fb1cc555SAl Viro } 1692fb1cc555SAl Viro 1693fb1cc555SAl Viro /* 1694fb1cc555SAl Viro * It already exists. 1695fb1cc555SAl Viro */ 1696fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 1697fb1cc555SAl Viro audit_inode(pathname, path->dentry); 1698fb1cc555SAl Viro 1699fb1cc555SAl Viro error = -EEXIST; 17005b369df8SAl Viro if (open_flag & O_EXCL) 1701fb1cc555SAl Viro goto exit_dput; 1702fb1cc555SAl Viro 1703fb1cc555SAl Viro if (__follow_mount(path)) { 1704fb1cc555SAl Viro error = -ELOOP; 17055b369df8SAl Viro if (open_flag & O_NOFOLLOW) 1706fb1cc555SAl Viro goto exit_dput; 1707fb1cc555SAl Viro } 1708fb1cc555SAl Viro 1709fb1cc555SAl Viro error = -ENOENT; 1710fb1cc555SAl Viro if (!path->dentry->d_inode) 1711fb1cc555SAl Viro goto exit_dput; 17129e67f361SAl Viro 17139e67f361SAl Viro if (path->dentry->d_inode->i_op->follow_link) 1714fb1cc555SAl Viro return NULL; 1715fb1cc555SAl Viro 1716fb1cc555SAl Viro path_to_nameidata(path, nd); 1717fb1cc555SAl Viro error = -EISDIR; 1718fb1cc555SAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode)) 1719fb1cc555SAl Viro goto exit; 172067ee3ad2SAl Viro ok: 17219a66179eSAl Viro filp = finish_open(nd, open_flag, acc_mode); 1722fb1cc555SAl Viro return filp; 1723fb1cc555SAl Viro 1724fb1cc555SAl Viro exit_mutex_unlock: 1725fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 1726fb1cc555SAl Viro exit_dput: 1727fb1cc555SAl Viro path_put_conditional(path, nd); 1728fb1cc555SAl Viro exit: 1729fb1cc555SAl Viro if (!IS_ERR(nd->intent.open.file)) 1730fb1cc555SAl Viro release_open_intent(nd); 1731fb1cc555SAl Viro path_put(&nd->path); 1732fb1cc555SAl Viro return ERR_PTR(error); 1733fb1cc555SAl Viro } 1734fb1cc555SAl Viro 17354a3fd211SDave Hansen /* 17364a3fd211SDave Hansen * Note that the low bits of the passed in "open_flag" 17374a3fd211SDave Hansen * are not the same as in the local variable "flag". See 17384a3fd211SDave Hansen * open_to_namei_flags() for more details. 17391da177e4SLinus Torvalds */ 1740a70e65dfSChristoph Hellwig struct file *do_filp_open(int dfd, const char *pathname, 17416e8341a1SAl Viro int open_flag, int mode, int acc_mode) 17421da177e4SLinus Torvalds { 17434a3fd211SDave Hansen struct file *filp; 1744a70e65dfSChristoph Hellwig struct nameidata nd; 17456e8341a1SAl Viro int error; 17469850c056SAl Viro struct path path; 17471da177e4SLinus Torvalds int count = 0; 1748d57999e1SDave Hansen int flag = open_to_namei_flags(open_flag); 17499850c056SAl Viro int force_reval = 0; 17501f36f774SAl Viro 17511f36f774SAl Viro if (!(open_flag & O_CREAT)) 17521f36f774SAl Viro mode = 0; 17531da177e4SLinus Torvalds 17546b2f3d1fSChristoph Hellwig /* 17556b2f3d1fSChristoph Hellwig * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only 17566b2f3d1fSChristoph Hellwig * check for O_DSYNC if the need any syncing at all we enforce it's 17576b2f3d1fSChristoph Hellwig * always set instead of having to deal with possibly weird behaviour 17586b2f3d1fSChristoph Hellwig * for malicious applications setting only __O_SYNC. 17596b2f3d1fSChristoph Hellwig */ 17606b2f3d1fSChristoph Hellwig if (open_flag & __O_SYNC) 17616b2f3d1fSChristoph Hellwig open_flag |= O_DSYNC; 17626b2f3d1fSChristoph Hellwig 17636e8341a1SAl Viro if (!acc_mode) 17646d125529SAl Viro acc_mode = MAY_OPEN | ACC_MODE(open_flag); 17651da177e4SLinus Torvalds 1766834f2a4aSTrond Myklebust /* O_TRUNC implies we need access checks for write permissions */ 17674296e2cbSAl Viro if (open_flag & O_TRUNC) 1768834f2a4aSTrond Myklebust acc_mode |= MAY_WRITE; 1769834f2a4aSTrond Myklebust 17701da177e4SLinus Torvalds /* Allow the LSM permission hook to distinguish append 17711da177e4SLinus Torvalds access from general write access. */ 17724296e2cbSAl Viro if (open_flag & O_APPEND) 17731da177e4SLinus Torvalds acc_mode |= MAY_APPEND; 17741da177e4SLinus Torvalds 17751f36f774SAl Viro /* find the parent */ 17769850c056SAl Viro reval: 17779b4a9b14SAl Viro error = path_init(dfd, pathname, LOOKUP_PARENT, &nd); 17781da177e4SLinus Torvalds if (error) 1779a70e65dfSChristoph Hellwig return ERR_PTR(error); 17809850c056SAl Viro if (force_reval) 17819850c056SAl Viro nd.flags |= LOOKUP_REVAL; 17823866248eSAl Viro 17833866248eSAl Viro current->total_link_count = 0; 17843866248eSAl Viro error = link_path_walk(pathname, &nd); 1785654f562cSJ. R. Okajima if (error) { 178610fa8e62SAl Viro filp = ERR_PTR(error); 178710fa8e62SAl Viro goto out; 1788654f562cSJ. R. Okajima } 17891f36f774SAl Viro if (unlikely(!audit_dummy_context()) && (open_flag & O_CREAT)) 17909b4a9b14SAl Viro audit_inode(pathname, nd.path.dentry); 17911da177e4SLinus Torvalds 17921da177e4SLinus Torvalds /* 1793a2c36b45SAl Viro * We have the parent and last component. 17941da177e4SLinus Torvalds */ 17951da177e4SLinus Torvalds 17968737f3a1SAl Viro error = -ENFILE; 17978737f3a1SAl Viro filp = get_empty_filp(); 17988737f3a1SAl Viro if (filp == NULL) 17998737f3a1SAl Viro goto exit_parent; 18008737f3a1SAl Viro nd.intent.open.file = filp; 1801482928d5SAl Viro filp->f_flags = open_flag; 18028737f3a1SAl Viro nd.intent.open.flags = flag; 18038737f3a1SAl Viro nd.intent.open.create_mode = mode; 1804a70e65dfSChristoph Hellwig nd.flags &= ~LOOKUP_PARENT; 18051f36f774SAl Viro nd.flags |= LOOKUP_OPEN; 18061f36f774SAl Viro if (open_flag & O_CREAT) { 18071f36f774SAl Viro nd.flags |= LOOKUP_CREATE; 18084296e2cbSAl Viro if (open_flag & O_EXCL) 18093516586aSAl Viro nd.flags |= LOOKUP_EXCL; 18101f36f774SAl Viro } 18113e297b61SAl Viro if (open_flag & O_DIRECTORY) 18123e297b61SAl Viro nd.flags |= LOOKUP_DIRECTORY; 1813002baeecSJan Kara if (!(open_flag & O_NOFOLLOW)) 1814002baeecSJan Kara nd.flags |= LOOKUP_FOLLOW; 18153e297b61SAl Viro filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname); 1816806b681cSAl Viro while (unlikely(!filp)) { /* trailing symlink */ 1817def4af30SAl Viro struct path holder; 18181f36f774SAl Viro struct inode *inode = path.dentry->d_inode; 1819def4af30SAl Viro void *cookie; 1820806b681cSAl Viro error = -ELOOP; 18211f36f774SAl Viro /* S_ISDIR part is a temporary automount kludge */ 1822002baeecSJan Kara if (!(nd.flags & LOOKUP_FOLLOW) && !S_ISDIR(inode->i_mode)) 18231f36f774SAl Viro goto exit_dput; 18241f36f774SAl Viro if (count++ == 32) 1825806b681cSAl Viro goto exit_dput; 1826806b681cSAl Viro /* 1827806b681cSAl Viro * This is subtle. Instead of calling do_follow_link() we do 1828806b681cSAl Viro * the thing by hands. The reason is that this way we have zero 1829806b681cSAl Viro * link_count and path_walk() (called from ->follow_link) 1830806b681cSAl Viro * honoring LOOKUP_PARENT. After that we have the parent and 1831806b681cSAl Viro * last component, i.e. we are in the same situation as after 1832806b681cSAl Viro * the first path_walk(). Well, almost - if the last component 1833806b681cSAl Viro * is normal we get its copy stored in nd->last.name and we will 1834806b681cSAl Viro * have to putname() it when we are done. Procfs-like symlinks 1835806b681cSAl Viro * just set LAST_BIND. 1836806b681cSAl Viro */ 1837806b681cSAl Viro nd.flags |= LOOKUP_PARENT; 1838806b681cSAl Viro error = security_inode_follow_link(path.dentry, &nd); 1839806b681cSAl Viro if (error) 1840806b681cSAl Viro goto exit_dput; 1841def4af30SAl Viro error = __do_follow_link(&path, &nd, &cookie); 1842def4af30SAl Viro if (unlikely(error)) { 1843806b681cSAl Viro /* nd.path had been dropped */ 1844def4af30SAl Viro if (!IS_ERR(cookie) && inode->i_op->put_link) 1845def4af30SAl Viro inode->i_op->put_link(path.dentry, &nd, cookie); 1846def4af30SAl Viro path_put(&path); 1847a70e65dfSChristoph Hellwig release_open_intent(&nd); 184810fa8e62SAl Viro filp = ERR_PTR(error); 1849806b681cSAl Viro goto out; 1850806b681cSAl Viro } 1851def4af30SAl Viro holder = path; 1852806b681cSAl Viro nd.flags &= ~LOOKUP_PARENT; 18533e297b61SAl Viro filp = do_last(&nd, &path, open_flag, acc_mode, mode, pathname); 1854def4af30SAl Viro if (inode->i_op->put_link) 1855def4af30SAl Viro inode->i_op->put_link(holder.dentry, &nd, cookie); 1856def4af30SAl Viro path_put(&holder); 1857806b681cSAl Viro } 185810fa8e62SAl Viro out: 18592a737871SAl Viro if (nd.root.mnt) 18602a737871SAl Viro path_put(&nd.root); 186110fa8e62SAl Viro if (filp == ERR_PTR(-ESTALE) && !force_reval) { 186210fa8e62SAl Viro force_reval = 1; 186310fa8e62SAl Viro goto reval; 186410fa8e62SAl Viro } 186510fa8e62SAl Viro return filp; 18661da177e4SLinus Torvalds 1867806b681cSAl Viro exit_dput: 1868806b681cSAl Viro path_put_conditional(&path, &nd); 1869806b681cSAl Viro if (!IS_ERR(nd.intent.open.file)) 1870a70e65dfSChristoph Hellwig release_open_intent(&nd); 1871806b681cSAl Viro exit_parent: 1872806b681cSAl Viro path_put(&nd.path); 187310fa8e62SAl Viro filp = ERR_PTR(error); 187410fa8e62SAl Viro goto out; 1875de459215SKirill Korotaev } 18761da177e4SLinus Torvalds 18771da177e4SLinus Torvalds /** 1878a70e65dfSChristoph Hellwig * filp_open - open file and return file pointer 1879a70e65dfSChristoph Hellwig * 1880a70e65dfSChristoph Hellwig * @filename: path to open 1881a70e65dfSChristoph Hellwig * @flags: open flags as per the open(2) second argument 1882a70e65dfSChristoph Hellwig * @mode: mode for the new file if O_CREAT is set, else ignored 1883a70e65dfSChristoph Hellwig * 1884a70e65dfSChristoph Hellwig * This is the helper to open a file from kernelspace if you really 1885a70e65dfSChristoph Hellwig * have to. But in generally you should not do this, so please move 1886a70e65dfSChristoph Hellwig * along, nothing to see here.. 1887a70e65dfSChristoph Hellwig */ 1888a70e65dfSChristoph Hellwig struct file *filp_open(const char *filename, int flags, int mode) 1889a70e65dfSChristoph Hellwig { 18906e8341a1SAl Viro return do_filp_open(AT_FDCWD, filename, flags, mode, 0); 1891a70e65dfSChristoph Hellwig } 1892a70e65dfSChristoph Hellwig EXPORT_SYMBOL(filp_open); 1893a70e65dfSChristoph Hellwig 1894a70e65dfSChristoph Hellwig /** 18951da177e4SLinus Torvalds * lookup_create - lookup a dentry, creating it if it doesn't exist 18961da177e4SLinus Torvalds * @nd: nameidata info 18971da177e4SLinus Torvalds * @is_dir: directory flag 18981da177e4SLinus Torvalds * 18991da177e4SLinus Torvalds * Simple function to lookup and return a dentry and create it 19001da177e4SLinus Torvalds * if it doesn't exist. Is SMP-safe. 1901c663e5d8SChristoph Hellwig * 19024ac91378SJan Blunck * Returns with nd->path.dentry->d_inode->i_mutex locked. 19031da177e4SLinus Torvalds */ 19041da177e4SLinus Torvalds struct dentry *lookup_create(struct nameidata *nd, int is_dir) 19051da177e4SLinus Torvalds { 1906c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 19071da177e4SLinus Torvalds 19084ac91378SJan Blunck mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 1909c663e5d8SChristoph Hellwig /* 1910c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 1911c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 1912c663e5d8SChristoph Hellwig */ 19131da177e4SLinus Torvalds if (nd->last_type != LAST_NORM) 19141da177e4SLinus Torvalds goto fail; 19151da177e4SLinus Torvalds nd->flags &= ~LOOKUP_PARENT; 19163516586aSAl Viro nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL; 1917a634904aSASANO Masahiro nd->intent.open.flags = O_EXCL; 1918c663e5d8SChristoph Hellwig 1919c663e5d8SChristoph Hellwig /* 1920c663e5d8SChristoph Hellwig * Do the final lookup. 1921c663e5d8SChristoph Hellwig */ 192249705b77SChristoph Hellwig dentry = lookup_hash(nd); 19231da177e4SLinus Torvalds if (IS_ERR(dentry)) 19241da177e4SLinus Torvalds goto fail; 1925c663e5d8SChristoph Hellwig 1926e9baf6e5SAl Viro if (dentry->d_inode) 1927e9baf6e5SAl Viro goto eexist; 1928c663e5d8SChristoph Hellwig /* 1929c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 1930c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 1931c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 1932c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 1933c663e5d8SChristoph Hellwig */ 1934e9baf6e5SAl Viro if (unlikely(!is_dir && nd->last.name[nd->last.len])) { 19351da177e4SLinus Torvalds dput(dentry); 19361da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 1937e9baf6e5SAl Viro } 1938e9baf6e5SAl Viro return dentry; 1939e9baf6e5SAl Viro eexist: 1940e9baf6e5SAl Viro dput(dentry); 1941e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 19421da177e4SLinus Torvalds fail: 19431da177e4SLinus Torvalds return dentry; 19441da177e4SLinus Torvalds } 1945f81a0bffSChristoph Hellwig EXPORT_SYMBOL_GPL(lookup_create); 19461da177e4SLinus Torvalds 19471da177e4SLinus Torvalds int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 19481da177e4SLinus Torvalds { 1949a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 19501da177e4SLinus Torvalds 19511da177e4SLinus Torvalds if (error) 19521da177e4SLinus Torvalds return error; 19531da177e4SLinus Torvalds 19541da177e4SLinus Torvalds if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 19551da177e4SLinus Torvalds return -EPERM; 19561da177e4SLinus Torvalds 1957acfa4380SAl Viro if (!dir->i_op->mknod) 19581da177e4SLinus Torvalds return -EPERM; 19591da177e4SLinus Torvalds 196008ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 196108ce5f16SSerge E. Hallyn if (error) 196208ce5f16SSerge E. Hallyn return error; 196308ce5f16SSerge E. Hallyn 19641da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 19651da177e4SLinus Torvalds if (error) 19661da177e4SLinus Torvalds return error; 19671da177e4SLinus Torvalds 19681da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 1969a74574aaSStephen Smalley if (!error) 1970f38aa942SAmy Griffis fsnotify_create(dir, dentry); 19711da177e4SLinus Torvalds return error; 19721da177e4SLinus Torvalds } 19731da177e4SLinus Torvalds 1974463c3197SDave Hansen static int may_mknod(mode_t mode) 1975463c3197SDave Hansen { 1976463c3197SDave Hansen switch (mode & S_IFMT) { 1977463c3197SDave Hansen case S_IFREG: 1978463c3197SDave Hansen case S_IFCHR: 1979463c3197SDave Hansen case S_IFBLK: 1980463c3197SDave Hansen case S_IFIFO: 1981463c3197SDave Hansen case S_IFSOCK: 1982463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 1983463c3197SDave Hansen return 0; 1984463c3197SDave Hansen case S_IFDIR: 1985463c3197SDave Hansen return -EPERM; 1986463c3197SDave Hansen default: 1987463c3197SDave Hansen return -EINVAL; 1988463c3197SDave Hansen } 1989463c3197SDave Hansen } 1990463c3197SDave Hansen 19912e4d0924SHeiko Carstens SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, 19922e4d0924SHeiko Carstens unsigned, dev) 19931da177e4SLinus Torvalds { 19942ad94ae6SAl Viro int error; 19951da177e4SLinus Torvalds char *tmp; 19961da177e4SLinus Torvalds struct dentry *dentry; 19971da177e4SLinus Torvalds struct nameidata nd; 19981da177e4SLinus Torvalds 19991da177e4SLinus Torvalds if (S_ISDIR(mode)) 20001da177e4SLinus Torvalds return -EPERM; 20011da177e4SLinus Torvalds 20022ad94ae6SAl Viro error = user_path_parent(dfd, filename, &nd, &tmp); 20031da177e4SLinus Torvalds if (error) 20042ad94ae6SAl Viro return error; 20052ad94ae6SAl Viro 20061da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 2007463c3197SDave Hansen if (IS_ERR(dentry)) { 20081da177e4SLinus Torvalds error = PTR_ERR(dentry); 2009463c3197SDave Hansen goto out_unlock; 2010463c3197SDave Hansen } 20114ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2012ce3b0f8dSAl Viro mode &= ~current_umask(); 2013463c3197SDave Hansen error = may_mknod(mode); 2014463c3197SDave Hansen if (error) 2015463c3197SDave Hansen goto out_dput; 2016463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2017463c3197SDave Hansen if (error) 2018463c3197SDave Hansen goto out_dput; 2019be6d3e56SKentaro Takeda error = security_path_mknod(&nd.path, dentry, mode, dev); 2020be6d3e56SKentaro Takeda if (error) 2021be6d3e56SKentaro Takeda goto out_drop_write; 20221da177e4SLinus Torvalds switch (mode & S_IFMT) { 20231da177e4SLinus Torvalds case 0: case S_IFREG: 20244ac91378SJan Blunck error = vfs_create(nd.path.dentry->d_inode,dentry,mode,&nd); 20251da177e4SLinus Torvalds break; 20261da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 20274ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode, 20281da177e4SLinus Torvalds new_decode_dev(dev)); 20291da177e4SLinus Torvalds break; 20301da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 20314ac91378SJan Blunck error = vfs_mknod(nd.path.dentry->d_inode,dentry,mode,0); 20321da177e4SLinus Torvalds break; 20331da177e4SLinus Torvalds } 2034be6d3e56SKentaro Takeda out_drop_write: 2035463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2036463c3197SDave Hansen out_dput: 20371da177e4SLinus Torvalds dput(dentry); 2038463c3197SDave Hansen out_unlock: 20394ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 20401d957f9bSJan Blunck path_put(&nd.path); 20411da177e4SLinus Torvalds putname(tmp); 20421da177e4SLinus Torvalds 20431da177e4SLinus Torvalds return error; 20441da177e4SLinus Torvalds } 20451da177e4SLinus Torvalds 20463480b257SHeiko Carstens SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) 20475590ff0dSUlrich Drepper { 20485590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 20495590ff0dSUlrich Drepper } 20505590ff0dSUlrich Drepper 20511da177e4SLinus Torvalds int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 20521da177e4SLinus Torvalds { 2053a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 20541da177e4SLinus Torvalds 20551da177e4SLinus Torvalds if (error) 20561da177e4SLinus Torvalds return error; 20571da177e4SLinus Torvalds 2058acfa4380SAl Viro if (!dir->i_op->mkdir) 20591da177e4SLinus Torvalds return -EPERM; 20601da177e4SLinus Torvalds 20611da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 20621da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 20631da177e4SLinus Torvalds if (error) 20641da177e4SLinus Torvalds return error; 20651da177e4SLinus Torvalds 20661da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 2067a74574aaSStephen Smalley if (!error) 2068f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 20691da177e4SLinus Torvalds return error; 20701da177e4SLinus Torvalds } 20711da177e4SLinus Torvalds 20722e4d0924SHeiko Carstens SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) 20731da177e4SLinus Torvalds { 20741da177e4SLinus Torvalds int error = 0; 20751da177e4SLinus Torvalds char * tmp; 20766902d925SDave Hansen struct dentry *dentry; 20776902d925SDave Hansen struct nameidata nd; 20781da177e4SLinus Torvalds 20792ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &tmp); 20802ad94ae6SAl Viro if (error) 20816902d925SDave Hansen goto out_err; 20821da177e4SLinus Torvalds 20831da177e4SLinus Torvalds dentry = lookup_create(&nd, 1); 20841da177e4SLinus Torvalds error = PTR_ERR(dentry); 20856902d925SDave Hansen if (IS_ERR(dentry)) 20866902d925SDave Hansen goto out_unlock; 20876902d925SDave Hansen 20884ac91378SJan Blunck if (!IS_POSIXACL(nd.path.dentry->d_inode)) 2089ce3b0f8dSAl Viro mode &= ~current_umask(); 2090463c3197SDave Hansen error = mnt_want_write(nd.path.mnt); 2091463c3197SDave Hansen if (error) 2092463c3197SDave Hansen goto out_dput; 2093be6d3e56SKentaro Takeda error = security_path_mkdir(&nd.path, dentry, mode); 2094be6d3e56SKentaro Takeda if (error) 2095be6d3e56SKentaro Takeda goto out_drop_write; 20964ac91378SJan Blunck error = vfs_mkdir(nd.path.dentry->d_inode, dentry, mode); 2097be6d3e56SKentaro Takeda out_drop_write: 2098463c3197SDave Hansen mnt_drop_write(nd.path.mnt); 2099463c3197SDave Hansen out_dput: 21001da177e4SLinus Torvalds dput(dentry); 21016902d925SDave Hansen out_unlock: 21024ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 21031d957f9bSJan Blunck path_put(&nd.path); 21041da177e4SLinus Torvalds putname(tmp); 21056902d925SDave Hansen out_err: 21061da177e4SLinus Torvalds return error; 21071da177e4SLinus Torvalds } 21081da177e4SLinus Torvalds 21093cdad428SHeiko Carstens SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) 21105590ff0dSUlrich Drepper { 21115590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 21125590ff0dSUlrich Drepper } 21135590ff0dSUlrich Drepper 21141da177e4SLinus Torvalds /* 21151da177e4SLinus Torvalds * We try to drop the dentry early: we should have 21161da177e4SLinus Torvalds * a usage count of 2 if we're the only user of this 21171da177e4SLinus Torvalds * dentry, and if that is true (possibly after pruning 21181da177e4SLinus Torvalds * the dcache), then we drop the dentry now. 21191da177e4SLinus Torvalds * 21201da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 21211da177e4SLinus Torvalds * do a 21221da177e4SLinus Torvalds * 21231da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 21241da177e4SLinus Torvalds * return -EBUSY; 21251da177e4SLinus Torvalds * 21261da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 21271da177e4SLinus Torvalds * that is still in use by something else.. 21281da177e4SLinus Torvalds */ 21291da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 21301da177e4SLinus Torvalds { 21311da177e4SLinus Torvalds dget(dentry); 21321da177e4SLinus Torvalds shrink_dcache_parent(dentry); 21331da177e4SLinus Torvalds spin_lock(&dcache_lock); 21341da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 21351da177e4SLinus Torvalds if (atomic_read(&dentry->d_count) == 2) 21361da177e4SLinus Torvalds __d_drop(dentry); 21371da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 21381da177e4SLinus Torvalds spin_unlock(&dcache_lock); 21391da177e4SLinus Torvalds } 21401da177e4SLinus Torvalds 21411da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 21421da177e4SLinus Torvalds { 21431da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 21441da177e4SLinus Torvalds 21451da177e4SLinus Torvalds if (error) 21461da177e4SLinus Torvalds return error; 21471da177e4SLinus Torvalds 2148acfa4380SAl Viro if (!dir->i_op->rmdir) 21491da177e4SLinus Torvalds return -EPERM; 21501da177e4SLinus Torvalds 21511b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 21521da177e4SLinus Torvalds dentry_unhash(dentry); 21531da177e4SLinus Torvalds if (d_mountpoint(dentry)) 21541da177e4SLinus Torvalds error = -EBUSY; 21551da177e4SLinus Torvalds else { 21561da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 21571da177e4SLinus Torvalds if (!error) { 21581da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 2159d83c49f3SAl Viro if (!error) { 21601da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 2161d83c49f3SAl Viro dont_mount(dentry); 2162d83c49f3SAl Viro } 21631da177e4SLinus Torvalds } 21641da177e4SLinus Torvalds } 21651b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 21661da177e4SLinus Torvalds if (!error) { 21671da177e4SLinus Torvalds d_delete(dentry); 21681da177e4SLinus Torvalds } 21691da177e4SLinus Torvalds dput(dentry); 21701da177e4SLinus Torvalds 21711da177e4SLinus Torvalds return error; 21721da177e4SLinus Torvalds } 21731da177e4SLinus Torvalds 21745590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 21751da177e4SLinus Torvalds { 21761da177e4SLinus Torvalds int error = 0; 21771da177e4SLinus Torvalds char * name; 21781da177e4SLinus Torvalds struct dentry *dentry; 21791da177e4SLinus Torvalds struct nameidata nd; 21801da177e4SLinus Torvalds 21812ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 21821da177e4SLinus Torvalds if (error) 21832ad94ae6SAl Viro return error; 21841da177e4SLinus Torvalds 21851da177e4SLinus Torvalds switch(nd.last_type) { 21861da177e4SLinus Torvalds case LAST_DOTDOT: 21871da177e4SLinus Torvalds error = -ENOTEMPTY; 21881da177e4SLinus Torvalds goto exit1; 21891da177e4SLinus Torvalds case LAST_DOT: 21901da177e4SLinus Torvalds error = -EINVAL; 21911da177e4SLinus Torvalds goto exit1; 21921da177e4SLinus Torvalds case LAST_ROOT: 21931da177e4SLinus Torvalds error = -EBUSY; 21941da177e4SLinus Torvalds goto exit1; 21951da177e4SLinus Torvalds } 21960612d9fbSOGAWA Hirofumi 21970612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 21980612d9fbSOGAWA Hirofumi 21994ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 220049705b77SChristoph Hellwig dentry = lookup_hash(&nd); 22011da177e4SLinus Torvalds error = PTR_ERR(dentry); 22026902d925SDave Hansen if (IS_ERR(dentry)) 22036902d925SDave Hansen goto exit2; 22040622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 22050622753bSDave Hansen if (error) 22060622753bSDave Hansen goto exit3; 2207be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 2208be6d3e56SKentaro Takeda if (error) 2209be6d3e56SKentaro Takeda goto exit4; 22104ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 2211be6d3e56SKentaro Takeda exit4: 22120622753bSDave Hansen mnt_drop_write(nd.path.mnt); 22130622753bSDave Hansen exit3: 22141da177e4SLinus Torvalds dput(dentry); 22156902d925SDave Hansen exit2: 22164ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 22171da177e4SLinus Torvalds exit1: 22181d957f9bSJan Blunck path_put(&nd.path); 22191da177e4SLinus Torvalds putname(name); 22201da177e4SLinus Torvalds return error; 22211da177e4SLinus Torvalds } 22221da177e4SLinus Torvalds 22233cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 22245590ff0dSUlrich Drepper { 22255590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 22265590ff0dSUlrich Drepper } 22275590ff0dSUlrich Drepper 22281da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 22291da177e4SLinus Torvalds { 22301da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 22311da177e4SLinus Torvalds 22321da177e4SLinus Torvalds if (error) 22331da177e4SLinus Torvalds return error; 22341da177e4SLinus Torvalds 2235acfa4380SAl Viro if (!dir->i_op->unlink) 22361da177e4SLinus Torvalds return -EPERM; 22371da177e4SLinus Torvalds 22381b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 22391da177e4SLinus Torvalds if (d_mountpoint(dentry)) 22401da177e4SLinus Torvalds error = -EBUSY; 22411da177e4SLinus Torvalds else { 22421da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 2243bec1052eSAl Viro if (!error) { 22441da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 2245bec1052eSAl Viro if (!error) 2246d83c49f3SAl Viro dont_mount(dentry); 2247bec1052eSAl Viro } 22481da177e4SLinus Torvalds } 22491b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 22501da177e4SLinus Torvalds 22511da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 22521da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 2253ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 22541da177e4SLinus Torvalds d_delete(dentry); 22551da177e4SLinus Torvalds } 22560eeca283SRobert Love 22571da177e4SLinus Torvalds return error; 22581da177e4SLinus Torvalds } 22591da177e4SLinus Torvalds 22601da177e4SLinus Torvalds /* 22611da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 22621b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 22631da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 22641da177e4SLinus Torvalds * while waiting on the I/O. 22651da177e4SLinus Torvalds */ 22665590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 22671da177e4SLinus Torvalds { 22682ad94ae6SAl Viro int error; 22691da177e4SLinus Torvalds char *name; 22701da177e4SLinus Torvalds struct dentry *dentry; 22711da177e4SLinus Torvalds struct nameidata nd; 22721da177e4SLinus Torvalds struct inode *inode = NULL; 22731da177e4SLinus Torvalds 22742ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 22751da177e4SLinus Torvalds if (error) 22762ad94ae6SAl Viro return error; 22772ad94ae6SAl Viro 22781da177e4SLinus Torvalds error = -EISDIR; 22791da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 22801da177e4SLinus Torvalds goto exit1; 22810612d9fbSOGAWA Hirofumi 22820612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 22830612d9fbSOGAWA Hirofumi 22844ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 228549705b77SChristoph Hellwig dentry = lookup_hash(&nd); 22861da177e4SLinus Torvalds error = PTR_ERR(dentry); 22871da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 22881da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 22891da177e4SLinus Torvalds if (nd.last.name[nd.last.len]) 22901da177e4SLinus Torvalds goto slashes; 22911da177e4SLinus Torvalds inode = dentry->d_inode; 22921da177e4SLinus Torvalds if (inode) 22931da177e4SLinus Torvalds atomic_inc(&inode->i_count); 22940622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 22950622753bSDave Hansen if (error) 22960622753bSDave Hansen goto exit2; 2297be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 2298be6d3e56SKentaro Takeda if (error) 2299be6d3e56SKentaro Takeda goto exit3; 23004ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 2301be6d3e56SKentaro Takeda exit3: 23020622753bSDave Hansen mnt_drop_write(nd.path.mnt); 23031da177e4SLinus Torvalds exit2: 23041da177e4SLinus Torvalds dput(dentry); 23051da177e4SLinus Torvalds } 23064ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 23071da177e4SLinus Torvalds if (inode) 23081da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 23091da177e4SLinus Torvalds exit1: 23101d957f9bSJan Blunck path_put(&nd.path); 23111da177e4SLinus Torvalds putname(name); 23121da177e4SLinus Torvalds return error; 23131da177e4SLinus Torvalds 23141da177e4SLinus Torvalds slashes: 23151da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 23161da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 23171da177e4SLinus Torvalds goto exit2; 23181da177e4SLinus Torvalds } 23191da177e4SLinus Torvalds 23202e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 23215590ff0dSUlrich Drepper { 23225590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 23235590ff0dSUlrich Drepper return -EINVAL; 23245590ff0dSUlrich Drepper 23255590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 23265590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 23275590ff0dSUlrich Drepper 23285590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 23295590ff0dSUlrich Drepper } 23305590ff0dSUlrich Drepper 23313480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 23325590ff0dSUlrich Drepper { 23335590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 23345590ff0dSUlrich Drepper } 23355590ff0dSUlrich Drepper 2336db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 23371da177e4SLinus Torvalds { 2338a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 23391da177e4SLinus Torvalds 23401da177e4SLinus Torvalds if (error) 23411da177e4SLinus Torvalds return error; 23421da177e4SLinus Torvalds 2343acfa4380SAl Viro if (!dir->i_op->symlink) 23441da177e4SLinus Torvalds return -EPERM; 23451da177e4SLinus Torvalds 23461da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 23471da177e4SLinus Torvalds if (error) 23481da177e4SLinus Torvalds return error; 23491da177e4SLinus Torvalds 23501da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 2351a74574aaSStephen Smalley if (!error) 2352f38aa942SAmy Griffis fsnotify_create(dir, dentry); 23531da177e4SLinus Torvalds return error; 23541da177e4SLinus Torvalds } 23551da177e4SLinus Torvalds 23562e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 23572e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 23581da177e4SLinus Torvalds { 23592ad94ae6SAl Viro int error; 23601da177e4SLinus Torvalds char *from; 23611da177e4SLinus Torvalds char *to; 23626902d925SDave Hansen struct dentry *dentry; 23636902d925SDave Hansen struct nameidata nd; 23641da177e4SLinus Torvalds 23651da177e4SLinus Torvalds from = getname(oldname); 23661da177e4SLinus Torvalds if (IS_ERR(from)) 23671da177e4SLinus Torvalds return PTR_ERR(from); 23682ad94ae6SAl Viro 23692ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 23702ad94ae6SAl Viro if (error) 23716902d925SDave Hansen goto out_putname; 23721da177e4SLinus Torvalds 23731da177e4SLinus Torvalds dentry = lookup_create(&nd, 0); 23741da177e4SLinus Torvalds error = PTR_ERR(dentry); 23756902d925SDave Hansen if (IS_ERR(dentry)) 23766902d925SDave Hansen goto out_unlock; 23776902d925SDave Hansen 237875c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 237975c3f29dSDave Hansen if (error) 238075c3f29dSDave Hansen goto out_dput; 2381be6d3e56SKentaro Takeda error = security_path_symlink(&nd.path, dentry, from); 2382be6d3e56SKentaro Takeda if (error) 2383be6d3e56SKentaro Takeda goto out_drop_write; 2384db2e747bSMiklos Szeredi error = vfs_symlink(nd.path.dentry->d_inode, dentry, from); 2385be6d3e56SKentaro Takeda out_drop_write: 238675c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 238775c3f29dSDave Hansen out_dput: 23881da177e4SLinus Torvalds dput(dentry); 23896902d925SDave Hansen out_unlock: 23904ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 23911d957f9bSJan Blunck path_put(&nd.path); 23921da177e4SLinus Torvalds putname(to); 23936902d925SDave Hansen out_putname: 23941da177e4SLinus Torvalds putname(from); 23951da177e4SLinus Torvalds return error; 23961da177e4SLinus Torvalds } 23971da177e4SLinus Torvalds 23983480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 23995590ff0dSUlrich Drepper { 24005590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 24015590ff0dSUlrich Drepper } 24025590ff0dSUlrich Drepper 24031da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 24041da177e4SLinus Torvalds { 24051da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 24061da177e4SLinus Torvalds int error; 24071da177e4SLinus Torvalds 24081da177e4SLinus Torvalds if (!inode) 24091da177e4SLinus Torvalds return -ENOENT; 24101da177e4SLinus Torvalds 2411a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 24121da177e4SLinus Torvalds if (error) 24131da177e4SLinus Torvalds return error; 24141da177e4SLinus Torvalds 24151da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 24161da177e4SLinus Torvalds return -EXDEV; 24171da177e4SLinus Torvalds 24181da177e4SLinus Torvalds /* 24191da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 24201da177e4SLinus Torvalds */ 24211da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 24221da177e4SLinus Torvalds return -EPERM; 2423acfa4380SAl Viro if (!dir->i_op->link) 24241da177e4SLinus Torvalds return -EPERM; 24257e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 24261da177e4SLinus Torvalds return -EPERM; 24271da177e4SLinus Torvalds 24281da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 24291da177e4SLinus Torvalds if (error) 24301da177e4SLinus Torvalds return error; 24311da177e4SLinus Torvalds 24327e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 24331da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 24347e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 2435e31e14ecSStephen Smalley if (!error) 24367e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 24371da177e4SLinus Torvalds return error; 24381da177e4SLinus Torvalds } 24391da177e4SLinus Torvalds 24401da177e4SLinus Torvalds /* 24411da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 24421da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 24431da177e4SLinus Torvalds * newname. --KAB 24441da177e4SLinus Torvalds * 24451da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 24461da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 24471da177e4SLinus Torvalds * and other special files. --ADM 24481da177e4SLinus Torvalds */ 24492e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 24502e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 24511da177e4SLinus Torvalds { 24521da177e4SLinus Torvalds struct dentry *new_dentry; 24532d8f3038SAl Viro struct nameidata nd; 24542d8f3038SAl Viro struct path old_path; 24551da177e4SLinus Torvalds int error; 24561da177e4SLinus Torvalds char *to; 24571da177e4SLinus Torvalds 245845c9b11aSUlrich Drepper if ((flags & ~AT_SYMLINK_FOLLOW) != 0) 2459c04030e1SUlrich Drepper return -EINVAL; 2460c04030e1SUlrich Drepper 24612d8f3038SAl Viro error = user_path_at(olddfd, oldname, 246245c9b11aSUlrich Drepper flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0, 24632d8f3038SAl Viro &old_path); 24641da177e4SLinus Torvalds if (error) 24652ad94ae6SAl Viro return error; 24662ad94ae6SAl Viro 24672ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &nd, &to); 24681da177e4SLinus Torvalds if (error) 24691da177e4SLinus Torvalds goto out; 24701da177e4SLinus Torvalds error = -EXDEV; 24712d8f3038SAl Viro if (old_path.mnt != nd.path.mnt) 24721da177e4SLinus Torvalds goto out_release; 24731da177e4SLinus Torvalds new_dentry = lookup_create(&nd, 0); 24741da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 24756902d925SDave Hansen if (IS_ERR(new_dentry)) 24766902d925SDave Hansen goto out_unlock; 247775c3f29dSDave Hansen error = mnt_want_write(nd.path.mnt); 247875c3f29dSDave Hansen if (error) 247975c3f29dSDave Hansen goto out_dput; 2480be6d3e56SKentaro Takeda error = security_path_link(old_path.dentry, &nd.path, new_dentry); 2481be6d3e56SKentaro Takeda if (error) 2482be6d3e56SKentaro Takeda goto out_drop_write; 24832d8f3038SAl Viro error = vfs_link(old_path.dentry, nd.path.dentry->d_inode, new_dentry); 2484be6d3e56SKentaro Takeda out_drop_write: 248575c3f29dSDave Hansen mnt_drop_write(nd.path.mnt); 248675c3f29dSDave Hansen out_dput: 24871da177e4SLinus Torvalds dput(new_dentry); 24886902d925SDave Hansen out_unlock: 24894ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 24901da177e4SLinus Torvalds out_release: 24911d957f9bSJan Blunck path_put(&nd.path); 24922ad94ae6SAl Viro putname(to); 24931da177e4SLinus Torvalds out: 24942d8f3038SAl Viro path_put(&old_path); 24951da177e4SLinus Torvalds 24961da177e4SLinus Torvalds return error; 24971da177e4SLinus Torvalds } 24981da177e4SLinus Torvalds 24993480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 25005590ff0dSUlrich Drepper { 2501c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 25025590ff0dSUlrich Drepper } 25035590ff0dSUlrich Drepper 25041da177e4SLinus Torvalds /* 25051da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 25061da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 25071da177e4SLinus Torvalds * Problems: 25081da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 25091da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 25101da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 2511a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 25121da177e4SLinus Torvalds * story. 25131da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 25141b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 25151da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 25161da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 2517a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 25181da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 25191da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 25201da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 2521a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 25221da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 25231da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 25241da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 25251da177e4SLinus Torvalds * d) some filesystems don't support opened-but-unlinked directories, 25261da177e4SLinus Torvalds * either because of layout or because they are not ready to deal with 25271da177e4SLinus Torvalds * all cases correctly. The latter will be fixed (taking this sort of 25281da177e4SLinus Torvalds * stuff into VFS), but the former is not going away. Solution: the same 25291da177e4SLinus Torvalds * trick as in rmdir(). 25301da177e4SLinus Torvalds * e) conversion from fhandle to dentry may come in the wrong moment - when 25311b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 25321da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 2533c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 25341da177e4SLinus Torvalds * locking]. 25351da177e4SLinus Torvalds */ 253675c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 25371da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 25381da177e4SLinus Torvalds { 25391da177e4SLinus Torvalds int error = 0; 25401da177e4SLinus Torvalds struct inode *target; 25411da177e4SLinus Torvalds 25421da177e4SLinus Torvalds /* 25431da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 25441da177e4SLinus Torvalds * we'll need to flip '..'. 25451da177e4SLinus Torvalds */ 25461da177e4SLinus Torvalds if (new_dir != old_dir) { 2547f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 25481da177e4SLinus Torvalds if (error) 25491da177e4SLinus Torvalds return error; 25501da177e4SLinus Torvalds } 25511da177e4SLinus Torvalds 25521da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 25531da177e4SLinus Torvalds if (error) 25541da177e4SLinus Torvalds return error; 25551da177e4SLinus Torvalds 25561da177e4SLinus Torvalds target = new_dentry->d_inode; 2557d83c49f3SAl Viro if (target) 25581b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 25591da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 25601da177e4SLinus Torvalds error = -EBUSY; 2561d83c49f3SAl Viro else { 2562d83c49f3SAl Viro if (target) 2563d83c49f3SAl Viro dentry_unhash(new_dentry); 25641da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 2565d83c49f3SAl Viro } 25661da177e4SLinus Torvalds if (target) { 2567d83c49f3SAl Viro if (!error) { 25681da177e4SLinus Torvalds target->i_flags |= S_DEAD; 2569d83c49f3SAl Viro dont_mount(new_dentry); 2570d83c49f3SAl Viro } 25711b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 25721da177e4SLinus Torvalds if (d_unhashed(new_dentry)) 25731da177e4SLinus Torvalds d_rehash(new_dentry); 25741da177e4SLinus Torvalds dput(new_dentry); 25751da177e4SLinus Torvalds } 2576e31e14ecSStephen Smalley if (!error) 2577349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 25781da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 25791da177e4SLinus Torvalds return error; 25801da177e4SLinus Torvalds } 25811da177e4SLinus Torvalds 258275c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 25831da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 25841da177e4SLinus Torvalds { 25851da177e4SLinus Torvalds struct inode *target; 25861da177e4SLinus Torvalds int error; 25871da177e4SLinus Torvalds 25881da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 25891da177e4SLinus Torvalds if (error) 25901da177e4SLinus Torvalds return error; 25911da177e4SLinus Torvalds 25921da177e4SLinus Torvalds dget(new_dentry); 25931da177e4SLinus Torvalds target = new_dentry->d_inode; 25941da177e4SLinus Torvalds if (target) 25951b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 25961da177e4SLinus Torvalds if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 25971da177e4SLinus Torvalds error = -EBUSY; 25981da177e4SLinus Torvalds else 25991da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 26001da177e4SLinus Torvalds if (!error) { 2601bec1052eSAl Viro if (target) 2602d83c49f3SAl Viro dont_mount(new_dentry); 2603349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 26041da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 26051da177e4SLinus Torvalds } 26061da177e4SLinus Torvalds if (target) 26071b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 26081da177e4SLinus Torvalds dput(new_dentry); 26091da177e4SLinus Torvalds return error; 26101da177e4SLinus Torvalds } 26111da177e4SLinus Torvalds 26121da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 26131da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 26141da177e4SLinus Torvalds { 26151da177e4SLinus Torvalds int error; 26161da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 261759b0df21SEric Paris const unsigned char *old_name; 26181da177e4SLinus Torvalds 26191da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 26201da177e4SLinus Torvalds return 0; 26211da177e4SLinus Torvalds 26221da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 26231da177e4SLinus Torvalds if (error) 26241da177e4SLinus Torvalds return error; 26251da177e4SLinus Torvalds 26261da177e4SLinus Torvalds if (!new_dentry->d_inode) 2627a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 26281da177e4SLinus Torvalds else 26291da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 26301da177e4SLinus Torvalds if (error) 26311da177e4SLinus Torvalds return error; 26321da177e4SLinus Torvalds 2633acfa4380SAl Viro if (!old_dir->i_op->rename) 26341da177e4SLinus Torvalds return -EPERM; 26351da177e4SLinus Torvalds 26360eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 26370eeca283SRobert Love 26381da177e4SLinus Torvalds if (is_dir) 26391da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 26401da177e4SLinus Torvalds else 26411da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 2642123df294SAl Viro if (!error) 2643123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 26445a190ae6SAl Viro new_dentry->d_inode, old_dentry); 26450eeca283SRobert Love fsnotify_oldname_free(old_name); 26460eeca283SRobert Love 26471da177e4SLinus Torvalds return error; 26481da177e4SLinus Torvalds } 26491da177e4SLinus Torvalds 26502e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 26512e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 26521da177e4SLinus Torvalds { 26531da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 26541da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 26551da177e4SLinus Torvalds struct dentry *trap; 26561da177e4SLinus Torvalds struct nameidata oldnd, newnd; 26572ad94ae6SAl Viro char *from; 26582ad94ae6SAl Viro char *to; 26592ad94ae6SAl Viro int error; 26601da177e4SLinus Torvalds 26612ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 26621da177e4SLinus Torvalds if (error) 26631da177e4SLinus Torvalds goto exit; 26641da177e4SLinus Torvalds 26652ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 26661da177e4SLinus Torvalds if (error) 26671da177e4SLinus Torvalds goto exit1; 26681da177e4SLinus Torvalds 26691da177e4SLinus Torvalds error = -EXDEV; 26704ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 26711da177e4SLinus Torvalds goto exit2; 26721da177e4SLinus Torvalds 26734ac91378SJan Blunck old_dir = oldnd.path.dentry; 26741da177e4SLinus Torvalds error = -EBUSY; 26751da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 26761da177e4SLinus Torvalds goto exit2; 26771da177e4SLinus Torvalds 26784ac91378SJan Blunck new_dir = newnd.path.dentry; 26791da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 26801da177e4SLinus Torvalds goto exit2; 26811da177e4SLinus Torvalds 26820612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 26830612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 26844e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 26850612d9fbSOGAWA Hirofumi 26861da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 26871da177e4SLinus Torvalds 268849705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 26891da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 26901da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 26911da177e4SLinus Torvalds goto exit3; 26921da177e4SLinus Torvalds /* source must exist */ 26931da177e4SLinus Torvalds error = -ENOENT; 26941da177e4SLinus Torvalds if (!old_dentry->d_inode) 26951da177e4SLinus Torvalds goto exit4; 26961da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 26971da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 26981da177e4SLinus Torvalds error = -ENOTDIR; 26991da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 27001da177e4SLinus Torvalds goto exit4; 27011da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 27021da177e4SLinus Torvalds goto exit4; 27031da177e4SLinus Torvalds } 27041da177e4SLinus Torvalds /* source should not be ancestor of target */ 27051da177e4SLinus Torvalds error = -EINVAL; 27061da177e4SLinus Torvalds if (old_dentry == trap) 27071da177e4SLinus Torvalds goto exit4; 270849705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 27091da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 27101da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 27111da177e4SLinus Torvalds goto exit4; 27121da177e4SLinus Torvalds /* target should not be an ancestor of source */ 27131da177e4SLinus Torvalds error = -ENOTEMPTY; 27141da177e4SLinus Torvalds if (new_dentry == trap) 27151da177e4SLinus Torvalds goto exit5; 27161da177e4SLinus Torvalds 27179079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 27189079b1ebSDave Hansen if (error) 27199079b1ebSDave Hansen goto exit5; 2720be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 2721be6d3e56SKentaro Takeda &newnd.path, new_dentry); 2722be6d3e56SKentaro Takeda if (error) 2723be6d3e56SKentaro Takeda goto exit6; 27241da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 27251da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 2726be6d3e56SKentaro Takeda exit6: 27279079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 27281da177e4SLinus Torvalds exit5: 27291da177e4SLinus Torvalds dput(new_dentry); 27301da177e4SLinus Torvalds exit4: 27311da177e4SLinus Torvalds dput(old_dentry); 27321da177e4SLinus Torvalds exit3: 27331da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 27341da177e4SLinus Torvalds exit2: 27351d957f9bSJan Blunck path_put(&newnd.path); 27362ad94ae6SAl Viro putname(to); 27371da177e4SLinus Torvalds exit1: 27381d957f9bSJan Blunck path_put(&oldnd.path); 27391da177e4SLinus Torvalds putname(from); 27402ad94ae6SAl Viro exit: 27411da177e4SLinus Torvalds return error; 27421da177e4SLinus Torvalds } 27431da177e4SLinus Torvalds 2744a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 27455590ff0dSUlrich Drepper { 27465590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 27475590ff0dSUlrich Drepper } 27485590ff0dSUlrich Drepper 27491da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 27501da177e4SLinus Torvalds { 27511da177e4SLinus Torvalds int len; 27521da177e4SLinus Torvalds 27531da177e4SLinus Torvalds len = PTR_ERR(link); 27541da177e4SLinus Torvalds if (IS_ERR(link)) 27551da177e4SLinus Torvalds goto out; 27561da177e4SLinus Torvalds 27571da177e4SLinus Torvalds len = strlen(link); 27581da177e4SLinus Torvalds if (len > (unsigned) buflen) 27591da177e4SLinus Torvalds len = buflen; 27601da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 27611da177e4SLinus Torvalds len = -EFAULT; 27621da177e4SLinus Torvalds out: 27631da177e4SLinus Torvalds return len; 27641da177e4SLinus Torvalds } 27651da177e4SLinus Torvalds 27661da177e4SLinus Torvalds /* 27671da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 27681da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 27691da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 27701da177e4SLinus Torvalds */ 27711da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 27721da177e4SLinus Torvalds { 27731da177e4SLinus Torvalds struct nameidata nd; 2774cc314eefSLinus Torvalds void *cookie; 2775694a1764SMarcin Slusarz int res; 2776cc314eefSLinus Torvalds 27771da177e4SLinus Torvalds nd.depth = 0; 2778cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 2779694a1764SMarcin Slusarz if (IS_ERR(cookie)) 2780694a1764SMarcin Slusarz return PTR_ERR(cookie); 2781694a1764SMarcin Slusarz 2782694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 27831da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 2784cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 2785694a1764SMarcin Slusarz return res; 27861da177e4SLinus Torvalds } 27871da177e4SLinus Torvalds 27881da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 27891da177e4SLinus Torvalds { 27901da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 27911da177e4SLinus Torvalds } 27921da177e4SLinus Torvalds 27931da177e4SLinus Torvalds /* get the link contents into pagecache */ 27941da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 27951da177e4SLinus Torvalds { 2796ebd09abbSDuane Griffin char *kaddr; 27971da177e4SLinus Torvalds struct page *page; 27981da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 2799090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 28001da177e4SLinus Torvalds if (IS_ERR(page)) 28016fe6900eSNick Piggin return (char*)page; 28021da177e4SLinus Torvalds *ppage = page; 2803ebd09abbSDuane Griffin kaddr = kmap(page); 2804ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 2805ebd09abbSDuane Griffin return kaddr; 28061da177e4SLinus Torvalds } 28071da177e4SLinus Torvalds 28081da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 28091da177e4SLinus Torvalds { 28101da177e4SLinus Torvalds struct page *page = NULL; 28111da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 28121da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 28131da177e4SLinus Torvalds if (page) { 28141da177e4SLinus Torvalds kunmap(page); 28151da177e4SLinus Torvalds page_cache_release(page); 28161da177e4SLinus Torvalds } 28171da177e4SLinus Torvalds return res; 28181da177e4SLinus Torvalds } 28191da177e4SLinus Torvalds 2820cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 28211da177e4SLinus Torvalds { 2822cc314eefSLinus Torvalds struct page *page = NULL; 28231da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 2824cc314eefSLinus Torvalds return page; 28251da177e4SLinus Torvalds } 28261da177e4SLinus Torvalds 2827cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 28281da177e4SLinus Torvalds { 2829cc314eefSLinus Torvalds struct page *page = cookie; 2830cc314eefSLinus Torvalds 2831cc314eefSLinus Torvalds if (page) { 28321da177e4SLinus Torvalds kunmap(page); 28331da177e4SLinus Torvalds page_cache_release(page); 28341da177e4SLinus Torvalds } 28351da177e4SLinus Torvalds } 28361da177e4SLinus Torvalds 283754566b2cSNick Piggin /* 283854566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 283954566b2cSNick Piggin */ 284054566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 28411da177e4SLinus Torvalds { 28421da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 28430adb25d2SKirill Korotaev struct page *page; 2844afddba49SNick Piggin void *fsdata; 2845beb497abSDmitriy Monakhov int err; 28461da177e4SLinus Torvalds char *kaddr; 284754566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 284854566b2cSNick Piggin if (nofs) 284954566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 28501da177e4SLinus Torvalds 28517e53cac4SNeilBrown retry: 2852afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 285354566b2cSNick Piggin flags, &page, &fsdata); 28541da177e4SLinus Torvalds if (err) 2855afddba49SNick Piggin goto fail; 2856afddba49SNick Piggin 28571da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 28581da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 28591da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 2860afddba49SNick Piggin 2861afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 2862afddba49SNick Piggin page, fsdata); 28631da177e4SLinus Torvalds if (err < 0) 28641da177e4SLinus Torvalds goto fail; 2865afddba49SNick Piggin if (err < len-1) 2866afddba49SNick Piggin goto retry; 2867afddba49SNick Piggin 28681da177e4SLinus Torvalds mark_inode_dirty(inode); 28691da177e4SLinus Torvalds return 0; 28701da177e4SLinus Torvalds fail: 28711da177e4SLinus Torvalds return err; 28721da177e4SLinus Torvalds } 28731da177e4SLinus Torvalds 28740adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 28750adb25d2SKirill Korotaev { 28760adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 287754566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 28780adb25d2SKirill Korotaev } 28790adb25d2SKirill Korotaev 288092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 28811da177e4SLinus Torvalds .readlink = generic_readlink, 28821da177e4SLinus Torvalds .follow_link = page_follow_link_light, 28831da177e4SLinus Torvalds .put_link = page_put_link, 28841da177e4SLinus Torvalds }; 28851da177e4SLinus Torvalds 28862d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 28871da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 28881da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 28891da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 28901da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 28911da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 28921da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 28931da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 28941da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 28951da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 28960adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 28971da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 28981da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 28991da177e4SLinus Torvalds EXPORT_SYMBOL(path_lookup); 2900d1811465SAl Viro EXPORT_SYMBOL(kern_path); 290116f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 2902f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 29038c744fb8SChristoph Hellwig EXPORT_SYMBOL(file_permission); 29041da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 29051da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 29061da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 29071da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 29081da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 29091da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 29101da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 29111da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 29121da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 29131da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 29141da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 29151da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 29161da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 29171da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 2918