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> 35e77819e5SLinus Torvalds #include <linux/posix_acl.h> 361da177e4SLinus Torvalds #include <asm/uaccess.h> 371da177e4SLinus Torvalds 38e81e3f4dSEric Paris #include "internal.h" 39c7105365SAl Viro #include "mount.h" 40e81e3f4dSEric Paris 411da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 421da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 431da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 441da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 451da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 461da177e4SLinus Torvalds * 471da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 481da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 491da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 501da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 511da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 521da177e4SLinus Torvalds * the special cases of the former code. 531da177e4SLinus Torvalds * 541da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 551da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 561da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 571da177e4SLinus Torvalds * 581da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 591da177e4SLinus Torvalds * resolution to correspond with current state of the code. 601da177e4SLinus Torvalds * 611da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 621da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 631da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 641da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 651da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 661da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 671da177e4SLinus Torvalds */ 681da177e4SLinus Torvalds 691da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 701da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 711da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 721da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 731da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 741da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7525985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 761da177e4SLinus Torvalds * 771da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 781da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 791da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 801da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 811da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 821da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 831da177e4SLinus Torvalds * and in the old Linux semantics. 841da177e4SLinus Torvalds */ 851da177e4SLinus Torvalds 861da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 871da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 881da177e4SLinus Torvalds * 891da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 901da177e4SLinus Torvalds */ 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 931da177e4SLinus Torvalds * inside the path - always follow. 941da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 951da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 961da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 971da177e4SLinus Torvalds * otherwise - don't follow. 981da177e4SLinus Torvalds * (applied in that order). 991da177e4SLinus Torvalds * 1001da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1011da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1021da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1031da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1041da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1051da177e4SLinus Torvalds */ 1061da177e4SLinus Torvalds /* 1071da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 108a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1091da177e4SLinus Torvalds * any extra contention... 1101da177e4SLinus Torvalds */ 1111da177e4SLinus Torvalds 1121da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1131da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1141da177e4SLinus Torvalds * kernel data space before using them.. 1151da177e4SLinus Torvalds * 1161da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1171da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1181da177e4SLinus Torvalds */ 119858119e1SArjan van de Ven static int do_getname(const char __user *filename, char *page) 1201da177e4SLinus Torvalds { 1211da177e4SLinus Torvalds int retval; 1221da177e4SLinus Torvalds unsigned long len = PATH_MAX; 1231da177e4SLinus Torvalds 1241da177e4SLinus Torvalds if (!segment_eq(get_fs(), KERNEL_DS)) { 1251da177e4SLinus Torvalds if ((unsigned long) filename >= TASK_SIZE) 1261da177e4SLinus Torvalds return -EFAULT; 1271da177e4SLinus Torvalds if (TASK_SIZE - (unsigned long) filename < PATH_MAX) 1281da177e4SLinus Torvalds len = TASK_SIZE - (unsigned long) filename; 1291da177e4SLinus Torvalds } 1301da177e4SLinus Torvalds 1311da177e4SLinus Torvalds retval = strncpy_from_user(page, filename, len); 1321da177e4SLinus Torvalds if (retval > 0) { 1331da177e4SLinus Torvalds if (retval < len) 1341da177e4SLinus Torvalds return 0; 1351da177e4SLinus Torvalds return -ENAMETOOLONG; 1361da177e4SLinus Torvalds } else if (!retval) 1371da177e4SLinus Torvalds retval = -ENOENT; 1381da177e4SLinus Torvalds return retval; 1391da177e4SLinus Torvalds } 1401da177e4SLinus Torvalds 1411fa1e7f6SAndy Whitcroft static char *getname_flags(const char __user *filename, int flags, int *empty) 1421da177e4SLinus Torvalds { 1431da177e4SLinus Torvalds char *tmp, *result; 1441da177e4SLinus Torvalds 1451da177e4SLinus Torvalds result = ERR_PTR(-ENOMEM); 1461da177e4SLinus Torvalds tmp = __getname(); 1471da177e4SLinus Torvalds if (tmp) { 1481da177e4SLinus Torvalds int retval = do_getname(filename, tmp); 1491da177e4SLinus Torvalds 1501da177e4SLinus Torvalds result = tmp; 1511da177e4SLinus Torvalds if (retval < 0) { 1521fa1e7f6SAndy Whitcroft if (retval == -ENOENT && empty) 1531fa1e7f6SAndy Whitcroft *empty = 1; 154f52e0c11SAl Viro if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) { 1551da177e4SLinus Torvalds __putname(tmp); 1561da177e4SLinus Torvalds result = ERR_PTR(retval); 1571da177e4SLinus Torvalds } 1581da177e4SLinus Torvalds } 159f52e0c11SAl Viro } 1601da177e4SLinus Torvalds audit_getname(result); 1611da177e4SLinus Torvalds return result; 1621da177e4SLinus Torvalds } 1631da177e4SLinus Torvalds 164f52e0c11SAl Viro char *getname(const char __user * filename) 165f52e0c11SAl Viro { 1661fa1e7f6SAndy Whitcroft return getname_flags(filename, 0, 0); 167f52e0c11SAl Viro } 168f52e0c11SAl Viro 1691da177e4SLinus Torvalds #ifdef CONFIG_AUDITSYSCALL 1701da177e4SLinus Torvalds void putname(const char *name) 1711da177e4SLinus Torvalds { 1725ac3a9c2SAl Viro if (unlikely(!audit_dummy_context())) 1731da177e4SLinus Torvalds audit_putname(name); 1741da177e4SLinus Torvalds else 1751da177e4SLinus Torvalds __putname(name); 1761da177e4SLinus Torvalds } 1771da177e4SLinus Torvalds EXPORT_SYMBOL(putname); 1781da177e4SLinus Torvalds #endif 1791da177e4SLinus Torvalds 180e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 181e77819e5SLinus Torvalds { 18284635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 183e77819e5SLinus Torvalds struct posix_acl *acl; 184e77819e5SLinus Torvalds 185e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 1863567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 1873567866bSAl Viro if (!acl) 188e77819e5SLinus Torvalds return -EAGAIN; 1893567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 1903567866bSAl Viro if (acl == ACL_NOT_CACHED) 191e77819e5SLinus Torvalds return -ECHILD; 192206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 193e77819e5SLinus Torvalds } 194e77819e5SLinus Torvalds 195e77819e5SLinus Torvalds acl = get_cached_acl(inode, ACL_TYPE_ACCESS); 196e77819e5SLinus Torvalds 197e77819e5SLinus Torvalds /* 1984e34e719SChristoph Hellwig * A filesystem can force a ACL callback by just never filling the 1994e34e719SChristoph Hellwig * ACL cache. But normally you'd fill the cache either at inode 2004e34e719SChristoph Hellwig * instantiation time, or on the first ->get_acl call. 201e77819e5SLinus Torvalds * 2024e34e719SChristoph Hellwig * If the filesystem doesn't have a get_acl() function at all, we'll 2034e34e719SChristoph Hellwig * just create the negative cache entry. 204e77819e5SLinus Torvalds */ 205e77819e5SLinus Torvalds if (acl == ACL_NOT_CACHED) { 2064e34e719SChristoph Hellwig if (inode->i_op->get_acl) { 2074e34e719SChristoph Hellwig acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS); 2084e34e719SChristoph Hellwig if (IS_ERR(acl)) 2094e34e719SChristoph Hellwig return PTR_ERR(acl); 2104e34e719SChristoph Hellwig } else { 211e77819e5SLinus Torvalds set_cached_acl(inode, ACL_TYPE_ACCESS, NULL); 212e77819e5SLinus Torvalds return -EAGAIN; 213e77819e5SLinus Torvalds } 2144e34e719SChristoph Hellwig } 215e77819e5SLinus Torvalds 216e77819e5SLinus Torvalds if (acl) { 217e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 218e77819e5SLinus Torvalds posix_acl_release(acl); 219e77819e5SLinus Torvalds return error; 220e77819e5SLinus Torvalds } 22184635d68SLinus Torvalds #endif 222e77819e5SLinus Torvalds 223e77819e5SLinus Torvalds return -EAGAIN; 224e77819e5SLinus Torvalds } 225e77819e5SLinus Torvalds 2265909ccaaSLinus Torvalds /* 227948409c7SAndreas Gruenbacher * This does the basic permission checking 2285909ccaaSLinus Torvalds */ 2297e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2305909ccaaSLinus Torvalds { 23126cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2325909ccaaSLinus Torvalds 233e795b717SSerge E. Hallyn if (current_user_ns() != inode_userns(inode)) 234e795b717SSerge E. Hallyn goto other_perms; 235e795b717SSerge E. Hallyn 23614067ff5SLinus Torvalds if (likely(current_fsuid() == inode->i_uid)) 2375909ccaaSLinus Torvalds mode >>= 6; 2385909ccaaSLinus Torvalds else { 239e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2407e40145eSAl Viro int error = check_acl(inode, mask); 2415909ccaaSLinus Torvalds if (error != -EAGAIN) 2425909ccaaSLinus Torvalds return error; 2435909ccaaSLinus Torvalds } 2445909ccaaSLinus Torvalds 2455909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 2465909ccaaSLinus Torvalds mode >>= 3; 2475909ccaaSLinus Torvalds } 2485909ccaaSLinus Torvalds 249e795b717SSerge E. Hallyn other_perms: 2505909ccaaSLinus Torvalds /* 2515909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 2525909ccaaSLinus Torvalds */ 2539c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 2545909ccaaSLinus Torvalds return 0; 2555909ccaaSLinus Torvalds return -EACCES; 2565909ccaaSLinus Torvalds } 2571da177e4SLinus Torvalds 2581da177e4SLinus Torvalds /** 2591da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 2601da177e4SLinus Torvalds * @inode: inode to check access rights for 2618fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 2621da177e4SLinus Torvalds * 2631da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 2641da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 2651da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 266b74c79e9SNick Piggin * are used for other things. 267b74c79e9SNick Piggin * 268b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 269b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 270b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 2711da177e4SLinus Torvalds */ 2722830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 2731da177e4SLinus Torvalds { 2745909ccaaSLinus Torvalds int ret; 2751da177e4SLinus Torvalds 2761da177e4SLinus Torvalds /* 277948409c7SAndreas Gruenbacher * Do the basic permission checks. 2781da177e4SLinus Torvalds */ 2797e40145eSAl Viro ret = acl_permission_check(inode, mask); 2805909ccaaSLinus Torvalds if (ret != -EACCES) 2815909ccaaSLinus Torvalds return ret; 2821da177e4SLinus Torvalds 283d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 284d594e7ecSAl Viro /* DACs are overridable for directories */ 285d594e7ecSAl Viro if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) 286d594e7ecSAl Viro return 0; 287d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 288d594e7ecSAl Viro if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) 289d594e7ecSAl Viro return 0; 290d594e7ecSAl Viro return -EACCES; 291d594e7ecSAl Viro } 2921da177e4SLinus Torvalds /* 2931da177e4SLinus Torvalds * Read/write DACs are always overridable. 294d594e7ecSAl Viro * Executable DACs are overridable when there is 295d594e7ecSAl Viro * at least one exec bit set. 2961da177e4SLinus Torvalds */ 297d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 298e795b717SSerge E. Hallyn if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE)) 2991da177e4SLinus Torvalds return 0; 3001da177e4SLinus Torvalds 3011da177e4SLinus Torvalds /* 3021da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3031da177e4SLinus Torvalds */ 3047ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 305d594e7ecSAl Viro if (mask == MAY_READ) 306e795b717SSerge E. Hallyn if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH)) 3071da177e4SLinus Torvalds return 0; 3081da177e4SLinus Torvalds 3091da177e4SLinus Torvalds return -EACCES; 3101da177e4SLinus Torvalds } 3111da177e4SLinus Torvalds 3123ddcd056SLinus Torvalds /* 3133ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3143ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3153ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3163ddcd056SLinus Torvalds * permission function, use the fast case". 3173ddcd056SLinus Torvalds */ 3183ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3193ddcd056SLinus Torvalds { 3203ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3213ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3223ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3233ddcd056SLinus Torvalds 3243ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3253ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3263ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3273ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3283ddcd056SLinus Torvalds } 3293ddcd056SLinus Torvalds return generic_permission(inode, mask); 3303ddcd056SLinus Torvalds } 3313ddcd056SLinus Torvalds 332cb23beb5SChristoph Hellwig /** 333cb23beb5SChristoph Hellwig * inode_permission - check for access rights to a given inode 334cb23beb5SChristoph Hellwig * @inode: inode to check permission on 3358fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 336cb23beb5SChristoph Hellwig * 337cb23beb5SChristoph Hellwig * Used to check for read/write/execute permissions on an inode. 338cb23beb5SChristoph Hellwig * We use "fsuid" for this, letting us set arbitrary permissions 339cb23beb5SChristoph Hellwig * for filesystem access without changing the "normal" uids which 340cb23beb5SChristoph Hellwig * are used for other things. 341948409c7SAndreas Gruenbacher * 342948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 343cb23beb5SChristoph Hellwig */ 344f419a2e3SAl Viro int inode_permission(struct inode *inode, int mask) 3451da177e4SLinus Torvalds { 346e6305c43SAl Viro int retval; 3471da177e4SLinus Torvalds 3483ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 34922590e41SMiklos Szeredi umode_t mode = inode->i_mode; 3501da177e4SLinus Torvalds 3511da177e4SLinus Torvalds /* 3521da177e4SLinus Torvalds * Nobody gets write access to a read-only fs. 3531da177e4SLinus Torvalds */ 3541da177e4SLinus Torvalds if (IS_RDONLY(inode) && 3551da177e4SLinus Torvalds (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 3561da177e4SLinus Torvalds return -EROFS; 3571da177e4SLinus Torvalds 3581da177e4SLinus Torvalds /* 3591da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 3601da177e4SLinus Torvalds */ 3611da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 3621da177e4SLinus Torvalds return -EACCES; 3631da177e4SLinus Torvalds } 3641da177e4SLinus Torvalds 3653ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 3661da177e4SLinus Torvalds if (retval) 3671da177e4SLinus Torvalds return retval; 3681da177e4SLinus Torvalds 36908ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 37008ce5f16SSerge E. Hallyn if (retval) 37108ce5f16SSerge E. Hallyn return retval; 37208ce5f16SSerge E. Hallyn 373d09ca739SEric Paris return security_inode_permission(inode, mask); 3741da177e4SLinus Torvalds } 3751da177e4SLinus Torvalds 376f4d6ff89SAl Viro /** 3775dd784d0SJan Blunck * path_get - get a reference to a path 3785dd784d0SJan Blunck * @path: path to get the reference to 3795dd784d0SJan Blunck * 3805dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 3815dd784d0SJan Blunck */ 3825dd784d0SJan Blunck void path_get(struct path *path) 3835dd784d0SJan Blunck { 3845dd784d0SJan Blunck mntget(path->mnt); 3855dd784d0SJan Blunck dget(path->dentry); 3865dd784d0SJan Blunck } 3875dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 3885dd784d0SJan Blunck 3895dd784d0SJan Blunck /** 3901d957f9bSJan Blunck * path_put - put a reference to a path 3911d957f9bSJan Blunck * @path: path to put the reference to 3921d957f9bSJan Blunck * 3931d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 3941d957f9bSJan Blunck */ 3951d957f9bSJan Blunck void path_put(struct path *path) 3961da177e4SLinus Torvalds { 3971d957f9bSJan Blunck dput(path->dentry); 3981d957f9bSJan Blunck mntput(path->mnt); 3991da177e4SLinus Torvalds } 4001d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4011da177e4SLinus Torvalds 40219660af7SAl Viro /* 40331e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 40419660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 40519660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 40619660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 40719660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 40819660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 40919660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 41019660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 41131e6b01fSNick Piggin */ 41231e6b01fSNick Piggin 41331e6b01fSNick Piggin /** 41419660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 41519660af7SAl Viro * @nd: nameidata pathwalk data 41619660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 41739191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 41831e6b01fSNick Piggin * 41919660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 42019660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 42119660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 42231e6b01fSNick Piggin */ 42319660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 42431e6b01fSNick Piggin { 42531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 42631e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 4275b6ca027SAl Viro int want_root = 0; 42831e6b01fSNick Piggin 42931e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 4305b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 4315b6ca027SAl Viro want_root = 1; 43231e6b01fSNick Piggin spin_lock(&fs->lock); 43331e6b01fSNick Piggin if (nd->root.mnt != fs->root.mnt || 43431e6b01fSNick Piggin nd->root.dentry != fs->root.dentry) 43531e6b01fSNick Piggin goto err_root; 43631e6b01fSNick Piggin } 43731e6b01fSNick Piggin spin_lock(&parent->d_lock); 43819660af7SAl Viro if (!dentry) { 43919660af7SAl Viro if (!__d_rcu_to_refcount(parent, nd->seq)) 44019660af7SAl Viro goto err_parent; 44119660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 44219660af7SAl Viro } else { 44394c0d4ecSAl Viro if (dentry->d_parent != parent) 44494c0d4ecSAl Viro goto err_parent; 44531e6b01fSNick Piggin spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); 44631e6b01fSNick Piggin if (!__d_rcu_to_refcount(dentry, nd->seq)) 44719660af7SAl Viro goto err_child; 44831e6b01fSNick Piggin /* 44919660af7SAl Viro * If the sequence check on the child dentry passed, then 45019660af7SAl Viro * the child has not been removed from its parent. This 45119660af7SAl Viro * means the parent dentry must be valid and able to take 45219660af7SAl Viro * a reference at this point. 45331e6b01fSNick Piggin */ 45431e6b01fSNick Piggin BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent); 45531e6b01fSNick Piggin BUG_ON(!parent->d_count); 45631e6b01fSNick Piggin parent->d_count++; 45731e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 45819660af7SAl Viro } 45931e6b01fSNick Piggin spin_unlock(&parent->d_lock); 4605b6ca027SAl Viro if (want_root) { 46131e6b01fSNick Piggin path_get(&nd->root); 46231e6b01fSNick Piggin spin_unlock(&fs->lock); 46331e6b01fSNick Piggin } 46431e6b01fSNick Piggin mntget(nd->path.mnt); 46531e6b01fSNick Piggin 46631e6b01fSNick Piggin rcu_read_unlock(); 46731e6b01fSNick Piggin br_read_unlock(vfsmount_lock); 46831e6b01fSNick Piggin nd->flags &= ~LOOKUP_RCU; 46931e6b01fSNick Piggin return 0; 47019660af7SAl Viro 47119660af7SAl Viro err_child: 47231e6b01fSNick Piggin spin_unlock(&dentry->d_lock); 47319660af7SAl Viro err_parent: 47431e6b01fSNick Piggin spin_unlock(&parent->d_lock); 47531e6b01fSNick Piggin err_root: 4765b6ca027SAl Viro if (want_root) 47731e6b01fSNick Piggin spin_unlock(&fs->lock); 47831e6b01fSNick Piggin return -ECHILD; 47931e6b01fSNick Piggin } 48031e6b01fSNick Piggin 48131e6b01fSNick Piggin /** 482834f2a4aSTrond Myklebust * release_open_intent - free up open intent resources 483834f2a4aSTrond Myklebust * @nd: pointer to nameidata 484834f2a4aSTrond Myklebust */ 485834f2a4aSTrond Myklebust void release_open_intent(struct nameidata *nd) 486834f2a4aSTrond Myklebust { 4872dab5974SLinus Torvalds struct file *file = nd->intent.open.file; 4882dab5974SLinus Torvalds 4892dab5974SLinus Torvalds if (file && !IS_ERR(file)) { 4902dab5974SLinus Torvalds if (file->f_path.dentry == NULL) 4912dab5974SLinus Torvalds put_filp(file); 492834f2a4aSTrond Myklebust else 4932dab5974SLinus Torvalds fput(file); 4942dab5974SLinus Torvalds } 495834f2a4aSTrond Myklebust } 496834f2a4aSTrond Myklebust 497f60aef7eSAl Viro static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd) 49834286d66SNick Piggin { 499f60aef7eSAl Viro return dentry->d_op->d_revalidate(dentry, nd); 50034286d66SNick Piggin } 50134286d66SNick Piggin 5029f1fafeeSAl Viro /** 5039f1fafeeSAl Viro * complete_walk - successful completion of path walk 5049f1fafeeSAl Viro * @nd: pointer nameidata 50539159de2SJeff Layton * 5069f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 5079f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 5089f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 5099f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 5109f1fafeeSAl Viro * need to drop nd->path. 51139159de2SJeff Layton */ 5129f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 51339159de2SJeff Layton { 51416c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 51539159de2SJeff Layton int status; 51639159de2SJeff Layton 5179f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 5189f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 5199f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 5209f1fafeeSAl Viro nd->root.mnt = NULL; 5219f1fafeeSAl Viro spin_lock(&dentry->d_lock); 5229f1fafeeSAl Viro if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) { 5239f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 5249f1fafeeSAl Viro rcu_read_unlock(); 5259f1fafeeSAl Viro br_read_unlock(vfsmount_lock); 5269f1fafeeSAl Viro return -ECHILD; 5279f1fafeeSAl Viro } 5289f1fafeeSAl Viro BUG_ON(nd->inode != dentry->d_inode); 5299f1fafeeSAl Viro spin_unlock(&dentry->d_lock); 5309f1fafeeSAl Viro mntget(nd->path.mnt); 5319f1fafeeSAl Viro rcu_read_unlock(); 5329f1fafeeSAl Viro br_read_unlock(vfsmount_lock); 5339f1fafeeSAl Viro } 5349f1fafeeSAl Viro 53516c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 53639159de2SJeff Layton return 0; 53739159de2SJeff Layton 53816c2cd71SAl Viro if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE))) 53916c2cd71SAl Viro return 0; 54016c2cd71SAl Viro 54116c2cd71SAl Viro if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT))) 54216c2cd71SAl Viro return 0; 54316c2cd71SAl Viro 54416c2cd71SAl Viro /* Note: we do not d_invalidate() */ 54534286d66SNick Piggin status = d_revalidate(dentry, nd); 54639159de2SJeff Layton if (status > 0) 54739159de2SJeff Layton return 0; 54839159de2SJeff Layton 54916c2cd71SAl Viro if (!status) 55039159de2SJeff Layton status = -ESTALE; 55116c2cd71SAl Viro 5529f1fafeeSAl Viro path_put(&nd->path); 55339159de2SJeff Layton return status; 55439159de2SJeff Layton } 55539159de2SJeff Layton 5562a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 5572a737871SAl Viro { 558f7ad3c6bSMiklos Szeredi if (!nd->root.mnt) 559f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 5602a737871SAl Viro } 5612a737871SAl Viro 5626de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 5636de88d72SAl Viro 56431e6b01fSNick Piggin static __always_inline void set_root_rcu(struct nameidata *nd) 56531e6b01fSNick Piggin { 56631e6b01fSNick Piggin if (!nd->root.mnt) { 56731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 568c28cc364SNick Piggin unsigned seq; 569c28cc364SNick Piggin 570c28cc364SNick Piggin do { 571c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 57231e6b01fSNick Piggin nd->root = fs->root; 573c1530019STim Chen nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 574c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 57531e6b01fSNick Piggin } 57631e6b01fSNick Piggin } 57731e6b01fSNick Piggin 578f1662356SArjan van de Ven static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link) 5791da177e4SLinus Torvalds { 58031e6b01fSNick Piggin int ret; 58131e6b01fSNick Piggin 5821da177e4SLinus Torvalds if (IS_ERR(link)) 5831da177e4SLinus Torvalds goto fail; 5841da177e4SLinus Torvalds 5851da177e4SLinus Torvalds if (*link == '/') { 5862a737871SAl Viro set_root(nd); 5871d957f9bSJan Blunck path_put(&nd->path); 5882a737871SAl Viro nd->path = nd->root; 5892a737871SAl Viro path_get(&nd->root); 59016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 5911da177e4SLinus Torvalds } 59231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 593b4091d5fSChristoph Hellwig 59431e6b01fSNick Piggin ret = link_path_walk(link, nd); 59531e6b01fSNick Piggin return ret; 5961da177e4SLinus Torvalds fail: 5971d957f9bSJan Blunck path_put(&nd->path); 5981da177e4SLinus Torvalds return PTR_ERR(link); 5991da177e4SLinus Torvalds } 6001da177e4SLinus Torvalds 6011d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 602051d3812SIan Kent { 603051d3812SIan Kent dput(path->dentry); 6044ac91378SJan Blunck if (path->mnt != nd->path.mnt) 605051d3812SIan Kent mntput(path->mnt); 606051d3812SIan Kent } 607051d3812SIan Kent 6087b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6097b9337aaSNick Piggin struct nameidata *nd) 610051d3812SIan Kent { 61131e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6124ac91378SJan Blunck dput(nd->path.dentry); 61331e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6144ac91378SJan Blunck mntput(nd->path.mnt); 6159a229683SHuang Shijie } 61631e6b01fSNick Piggin nd->path.mnt = path->mnt; 6174ac91378SJan Blunck nd->path.dentry = path->dentry; 618051d3812SIan Kent } 619051d3812SIan Kent 620574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 621574197e0SAl Viro { 622574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 623574197e0SAl Viro if (!IS_ERR(cookie) && inode->i_op->put_link) 624574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 625574197e0SAl Viro path_put(link); 626574197e0SAl Viro } 627574197e0SAl Viro 628def4af30SAl Viro static __always_inline int 629574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 6301da177e4SLinus Torvalds { 6311da177e4SLinus Torvalds int error; 6327b9337aaSNick Piggin struct dentry *dentry = link->dentry; 6331da177e4SLinus Torvalds 634844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 635844a3917SAl Viro 6360e794589SAl Viro if (link->mnt == nd->path.mnt) 6370e794589SAl Viro mntget(link->mnt); 6380e794589SAl Viro 639574197e0SAl Viro if (unlikely(current->total_link_count >= 40)) { 640574197e0SAl Viro *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */ 641574197e0SAl Viro path_put(&nd->path); 642574197e0SAl Viro return -ELOOP; 643574197e0SAl Viro } 644574197e0SAl Viro cond_resched(); 645574197e0SAl Viro current->total_link_count++; 646574197e0SAl Viro 6477b9337aaSNick Piggin touch_atime(link->mnt, dentry); 6481da177e4SLinus Torvalds nd_set_link(nd, NULL); 649cd4e91d3SAl Viro 65036f3b4f6SAl Viro error = security_inode_follow_link(link->dentry, nd); 65136f3b4f6SAl Viro if (error) { 65236f3b4f6SAl Viro *p = ERR_PTR(error); /* no ->put_link(), please */ 65336f3b4f6SAl Viro path_put(&nd->path); 65436f3b4f6SAl Viro return error; 65536f3b4f6SAl Viro } 65636f3b4f6SAl Viro 65786acdca1SAl Viro nd->last_type = LAST_BIND; 658def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 659def4af30SAl Viro error = PTR_ERR(*p); 660def4af30SAl Viro if (!IS_ERR(*p)) { 6611da177e4SLinus Torvalds char *s = nd_get_link(nd); 662cc314eefSLinus Torvalds error = 0; 6631da177e4SLinus Torvalds if (s) 6641da177e4SLinus Torvalds error = __vfs_follow_link(nd, s); 665bcda7652SAl Viro else if (nd->last_type == LAST_BIND) { 66616c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 667b21041d0SAl Viro nd->inode = nd->path.dentry->d_inode; 668b21041d0SAl Viro if (nd->inode->i_op->follow_link) { 669bcda7652SAl Viro /* stepped on a _really_ weird one */ 670bcda7652SAl Viro path_put(&nd->path); 671bcda7652SAl Viro error = -ELOOP; 672bcda7652SAl Viro } 673bcda7652SAl Viro } 6741da177e4SLinus Torvalds } 6751da177e4SLinus Torvalds return error; 6761da177e4SLinus Torvalds } 6771da177e4SLinus Torvalds 67831e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 67931e6b01fSNick Piggin { 6800714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 6810714a533SAl Viro struct mount *parent; 68231e6b01fSNick Piggin struct dentry *mountpoint; 68331e6b01fSNick Piggin 6840714a533SAl Viro parent = mnt->mnt_parent; 6850714a533SAl Viro if (&parent->mnt == path->mnt) 68631e6b01fSNick Piggin return 0; 687a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 68831e6b01fSNick Piggin path->dentry = mountpoint; 6890714a533SAl Viro path->mnt = &parent->mnt; 69031e6b01fSNick Piggin return 1; 69131e6b01fSNick Piggin } 69231e6b01fSNick Piggin 693bab77ebfSAl Viro int follow_up(struct path *path) 6941da177e4SLinus Torvalds { 6950714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 6960714a533SAl Viro struct mount *parent; 6971da177e4SLinus Torvalds struct dentry *mountpoint; 69899b7db7bSNick Piggin 69999b7db7bSNick Piggin br_read_lock(vfsmount_lock); 7000714a533SAl Viro parent = mnt->mnt_parent; 7010714a533SAl Viro if (&parent->mnt == path->mnt) { 70299b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 7031da177e4SLinus Torvalds return 0; 7041da177e4SLinus Torvalds } 7050714a533SAl Viro mntget(&parent->mnt); 706a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 70799b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 708bab77ebfSAl Viro dput(path->dentry); 709bab77ebfSAl Viro path->dentry = mountpoint; 710bab77ebfSAl Viro mntput(path->mnt); 7110714a533SAl Viro path->mnt = &parent->mnt; 7121da177e4SLinus Torvalds return 1; 7131da177e4SLinus Torvalds } 7141da177e4SLinus Torvalds 715b5c84bf6SNick Piggin /* 7169875cf80SDavid Howells * Perform an automount 7179875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 7189875cf80SDavid Howells * were called with. 7191da177e4SLinus Torvalds */ 7209875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 7219875cf80SDavid Howells bool *need_mntput) 72231e6b01fSNick Piggin { 7239875cf80SDavid Howells struct vfsmount *mnt; 724ea5b778aSDavid Howells int err; 7259875cf80SDavid Howells 7269875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 7279875cf80SDavid Howells return -EREMOTE; 7289875cf80SDavid Howells 7290ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 7300ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 7310ec26fd0SMiklos Szeredi * the name. 7320ec26fd0SMiklos Szeredi * 7330ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 7345a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 7350ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 7360ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 7370ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 7380ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 7395a30d8a2SDavid Howells */ 7405a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 741d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 7425a30d8a2SDavid Howells path->dentry->d_inode) 7439875cf80SDavid Howells return -EISDIR; 7440ec26fd0SMiklos Szeredi 7459875cf80SDavid Howells current->total_link_count++; 7469875cf80SDavid Howells if (current->total_link_count >= 40) 7479875cf80SDavid Howells return -ELOOP; 7489875cf80SDavid Howells 7499875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 7509875cf80SDavid Howells if (IS_ERR(mnt)) { 7519875cf80SDavid Howells /* 7529875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 7539875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 7549875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 7559875cf80SDavid Howells * 7569875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 7579875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 7589875cf80SDavid Howells * the path is inaccessible and we should say so. 7599875cf80SDavid Howells */ 76049084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 7619875cf80SDavid Howells return -EREMOTE; 7629875cf80SDavid Howells return PTR_ERR(mnt); 76331e6b01fSNick Piggin } 764ea5b778aSDavid Howells 7659875cf80SDavid Howells if (!mnt) /* mount collision */ 7669875cf80SDavid Howells return 0; 7679875cf80SDavid Howells 7688aef1884SAl Viro if (!*need_mntput) { 7698aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 7708aef1884SAl Viro mntget(path->mnt); 7718aef1884SAl Viro *need_mntput = true; 7728aef1884SAl Viro } 77319a167afSAl Viro err = finish_automount(mnt, path); 774ea5b778aSDavid Howells 775ea5b778aSDavid Howells switch (err) { 776ea5b778aSDavid Howells case -EBUSY: 777ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 77819a167afSAl Viro return 0; 779ea5b778aSDavid Howells case 0: 7808aef1884SAl Viro path_put(path); 7819875cf80SDavid Howells path->mnt = mnt; 7829875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 7839875cf80SDavid Howells return 0; 78419a167afSAl Viro default: 78519a167afSAl Viro return err; 7869875cf80SDavid Howells } 78719a167afSAl Viro 788ea5b778aSDavid Howells } 7899875cf80SDavid Howells 7909875cf80SDavid Howells /* 7919875cf80SDavid Howells * Handle a dentry that is managed in some way. 792cc53ce53SDavid Howells * - Flagged for transit management (autofs) 7939875cf80SDavid Howells * - Flagged as mountpoint 7949875cf80SDavid Howells * - Flagged as automount point 7959875cf80SDavid Howells * 7969875cf80SDavid Howells * This may only be called in refwalk mode. 7979875cf80SDavid Howells * 7989875cf80SDavid Howells * Serialization is taken care of in namespace.c 7999875cf80SDavid Howells */ 8009875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 8019875cf80SDavid Howells { 8028aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 8039875cf80SDavid Howells unsigned managed; 8049875cf80SDavid Howells bool need_mntput = false; 8058aef1884SAl Viro int ret = 0; 8069875cf80SDavid Howells 8079875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 8089875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 8099875cf80SDavid Howells * the components of that value change under us */ 8109875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 8119875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 8129875cf80SDavid Howells unlikely(managed != 0)) { 813cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 814cc53ce53SDavid Howells * being held. */ 815cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 816cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 817cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 8181aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 819cc53ce53SDavid Howells if (ret < 0) 8208aef1884SAl Viro break; 821cc53ce53SDavid Howells } 822cc53ce53SDavid Howells 8239875cf80SDavid Howells /* Transit to a mounted filesystem. */ 8249875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 8259875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 8269875cf80SDavid Howells if (mounted) { 8279875cf80SDavid Howells dput(path->dentry); 8289875cf80SDavid Howells if (need_mntput) 829463ffb2eSAl Viro mntput(path->mnt); 830463ffb2eSAl Viro path->mnt = mounted; 831463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 8329875cf80SDavid Howells need_mntput = true; 8339875cf80SDavid Howells continue; 834463ffb2eSAl Viro } 835463ffb2eSAl Viro 8369875cf80SDavid Howells /* Something is mounted on this dentry in another 8379875cf80SDavid Howells * namespace and/or whatever was mounted there in this 8389875cf80SDavid Howells * namespace got unmounted before we managed to get the 8399875cf80SDavid Howells * vfsmount_lock */ 8401da177e4SLinus Torvalds } 8419875cf80SDavid Howells 8429875cf80SDavid Howells /* Handle an automount point */ 8439875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 8449875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 8459875cf80SDavid Howells if (ret < 0) 8468aef1884SAl Viro break; 8479875cf80SDavid Howells continue; 8489875cf80SDavid Howells } 8499875cf80SDavid Howells 8509875cf80SDavid Howells /* We didn't change the current path point */ 8519875cf80SDavid Howells break; 8529875cf80SDavid Howells } 8538aef1884SAl Viro 8548aef1884SAl Viro if (need_mntput && path->mnt == mnt) 8558aef1884SAl Viro mntput(path->mnt); 8568aef1884SAl Viro if (ret == -EISDIR) 8578aef1884SAl Viro ret = 0; 858a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 8591da177e4SLinus Torvalds } 8601da177e4SLinus Torvalds 861cc53ce53SDavid Howells int follow_down_one(struct path *path) 8621da177e4SLinus Torvalds { 8631da177e4SLinus Torvalds struct vfsmount *mounted; 8641da177e4SLinus Torvalds 8651c755af4SAl Viro mounted = lookup_mnt(path); 8661da177e4SLinus Torvalds if (mounted) { 8679393bd07SAl Viro dput(path->dentry); 8689393bd07SAl Viro mntput(path->mnt); 8699393bd07SAl Viro path->mnt = mounted; 8709393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 8711da177e4SLinus Torvalds return 1; 8721da177e4SLinus Torvalds } 8731da177e4SLinus Torvalds return 0; 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds 87662a7375eSIan Kent static inline bool managed_dentry_might_block(struct dentry *dentry) 87762a7375eSIan Kent { 87862a7375eSIan Kent return (dentry->d_flags & DCACHE_MANAGE_TRANSIT && 87962a7375eSIan Kent dentry->d_op->d_manage(dentry, true) < 0); 88062a7375eSIan Kent } 88162a7375eSIan Kent 8829875cf80SDavid Howells /* 883287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 884287548e4SAl Viro * we meet a managed dentry that would need blocking. 8859875cf80SDavid Howells */ 8869875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 887287548e4SAl Viro struct inode **inode) 8889875cf80SDavid Howells { 88962a7375eSIan Kent for (;;) { 890c7105365SAl Viro struct mount *mounted; 89162a7375eSIan Kent /* 89262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 89362a7375eSIan Kent * that wants to block transit. 89462a7375eSIan Kent */ 895287548e4SAl Viro if (unlikely(managed_dentry_might_block(path->dentry))) 896ab90911fSDavid Howells return false; 89762a7375eSIan Kent 89862a7375eSIan Kent if (!d_mountpoint(path->dentry)) 89962a7375eSIan Kent break; 90062a7375eSIan Kent 9019875cf80SDavid Howells mounted = __lookup_mnt(path->mnt, path->dentry, 1); 9029875cf80SDavid Howells if (!mounted) 9039875cf80SDavid Howells break; 904c7105365SAl Viro path->mnt = &mounted->mnt; 905c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 906a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 9079875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 90859430262SLinus Torvalds /* 90959430262SLinus Torvalds * Update the inode too. We don't need to re-check the 91059430262SLinus Torvalds * dentry sequence number here after this d_inode read, 91159430262SLinus Torvalds * because a mount-point is always pinned. 91259430262SLinus Torvalds */ 91359430262SLinus Torvalds *inode = path->dentry->d_inode; 9149875cf80SDavid Howells } 9159875cf80SDavid Howells return true; 9169875cf80SDavid Howells } 9179875cf80SDavid Howells 918dea39376SAl Viro static void follow_mount_rcu(struct nameidata *nd) 919287548e4SAl Viro { 920dea39376SAl Viro while (d_mountpoint(nd->path.dentry)) { 921c7105365SAl Viro struct mount *mounted; 922dea39376SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1); 923287548e4SAl Viro if (!mounted) 924287548e4SAl Viro break; 925c7105365SAl Viro nd->path.mnt = &mounted->mnt; 926c7105365SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 927dea39376SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 928287548e4SAl Viro } 929287548e4SAl Viro } 930287548e4SAl Viro 93131e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 93231e6b01fSNick Piggin { 93331e6b01fSNick Piggin set_root_rcu(nd); 93431e6b01fSNick Piggin 93531e6b01fSNick Piggin while (1) { 93631e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 93731e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 93831e6b01fSNick Piggin break; 93931e6b01fSNick Piggin } 94031e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 94131e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 94231e6b01fSNick Piggin struct dentry *parent = old->d_parent; 94331e6b01fSNick Piggin unsigned seq; 94431e6b01fSNick Piggin 94531e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 94631e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 947ef7562d5SAl Viro goto failed; 94831e6b01fSNick Piggin nd->path.dentry = parent; 94931e6b01fSNick Piggin nd->seq = seq; 95031e6b01fSNick Piggin break; 95131e6b01fSNick Piggin } 95231e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 95331e6b01fSNick Piggin break; 95431e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 95531e6b01fSNick Piggin } 956dea39376SAl Viro follow_mount_rcu(nd); 957dea39376SAl Viro nd->inode = nd->path.dentry->d_inode; 95831e6b01fSNick Piggin return 0; 959ef7562d5SAl Viro 960ef7562d5SAl Viro failed: 961ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 9625b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 963ef7562d5SAl Viro nd->root.mnt = NULL; 964ef7562d5SAl Viro rcu_read_unlock(); 965ef7562d5SAl Viro br_read_unlock(vfsmount_lock); 966ef7562d5SAl Viro return -ECHILD; 96731e6b01fSNick Piggin } 96831e6b01fSNick Piggin 9699875cf80SDavid Howells /* 970cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 971cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 972cc53ce53SDavid Howells * caller is permitted to proceed or not. 973cc53ce53SDavid Howells */ 9747cc90cc3SAl Viro int follow_down(struct path *path) 975cc53ce53SDavid Howells { 976cc53ce53SDavid Howells unsigned managed; 977cc53ce53SDavid Howells int ret; 978cc53ce53SDavid Howells 979cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 980cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 981cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 982cc53ce53SDavid Howells * being held. 983cc53ce53SDavid Howells * 984cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 985cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 986cc53ce53SDavid Howells * other than its daemon the right to mount on its 987cc53ce53SDavid Howells * superstructure. 988cc53ce53SDavid Howells * 989cc53ce53SDavid Howells * The filesystem may sleep at this point. 990cc53ce53SDavid Howells */ 991cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 992cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 993cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 994ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 9951aed3e42SAl Viro path->dentry, false); 996cc53ce53SDavid Howells if (ret < 0) 997cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 998cc53ce53SDavid Howells } 999cc53ce53SDavid Howells 1000cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1001cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1002cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1003cc53ce53SDavid Howells if (!mounted) 1004cc53ce53SDavid Howells break; 1005cc53ce53SDavid Howells dput(path->dentry); 1006cc53ce53SDavid Howells mntput(path->mnt); 1007cc53ce53SDavid Howells path->mnt = mounted; 1008cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1009cc53ce53SDavid Howells continue; 1010cc53ce53SDavid Howells } 1011cc53ce53SDavid Howells 1012cc53ce53SDavid Howells /* Don't handle automount points here */ 1013cc53ce53SDavid Howells break; 1014cc53ce53SDavid Howells } 1015cc53ce53SDavid Howells return 0; 1016cc53ce53SDavid Howells } 1017cc53ce53SDavid Howells 1018cc53ce53SDavid Howells /* 10199875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 10209875cf80SDavid Howells */ 10219875cf80SDavid Howells static void follow_mount(struct path *path) 10229875cf80SDavid Howells { 10239875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 10249875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10259875cf80SDavid Howells if (!mounted) 10269875cf80SDavid Howells break; 10279875cf80SDavid Howells dput(path->dentry); 10289875cf80SDavid Howells mntput(path->mnt); 10299875cf80SDavid Howells path->mnt = mounted; 10309875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 10319875cf80SDavid Howells } 10329875cf80SDavid Howells } 10339875cf80SDavid Howells 103431e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 10351da177e4SLinus Torvalds { 10362a737871SAl Viro set_root(nd); 1037e518ddb7SAndreas Mohr 10381da177e4SLinus Torvalds while(1) { 10394ac91378SJan Blunck struct dentry *old = nd->path.dentry; 10401da177e4SLinus Torvalds 10412a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 10422a737871SAl Viro nd->path.mnt == nd->root.mnt) { 10431da177e4SLinus Torvalds break; 10441da177e4SLinus Torvalds } 10454ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 10463088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 10473088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 10481da177e4SLinus Torvalds dput(old); 10491da177e4SLinus Torvalds break; 10501da177e4SLinus Torvalds } 10513088dd70SAl Viro if (!follow_up(&nd->path)) 10521da177e4SLinus Torvalds break; 10531da177e4SLinus Torvalds } 105479ed0226SAl Viro follow_mount(&nd->path); 105531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 10561da177e4SLinus Torvalds } 10571da177e4SLinus Torvalds 10581da177e4SLinus Torvalds /* 1059baa03890SNick Piggin * Allocate a dentry with name and parent, and perform a parent 1060baa03890SNick Piggin * directory ->lookup on it. Returns the new dentry, or ERR_PTR 1061baa03890SNick Piggin * on error. parent->d_inode->i_mutex must be held. d_lookup must 1062baa03890SNick Piggin * have verified that no child exists while under i_mutex. 1063baa03890SNick Piggin */ 1064baa03890SNick Piggin static struct dentry *d_alloc_and_lookup(struct dentry *parent, 1065baa03890SNick Piggin struct qstr *name, struct nameidata *nd) 1066baa03890SNick Piggin { 1067baa03890SNick Piggin struct inode *inode = parent->d_inode; 1068baa03890SNick Piggin struct dentry *dentry; 1069baa03890SNick Piggin struct dentry *old; 1070baa03890SNick Piggin 1071baa03890SNick Piggin /* Don't create child dentry for a dead directory. */ 1072baa03890SNick Piggin if (unlikely(IS_DEADDIR(inode))) 1073baa03890SNick Piggin return ERR_PTR(-ENOENT); 1074baa03890SNick Piggin 1075baa03890SNick Piggin dentry = d_alloc(parent, name); 1076baa03890SNick Piggin if (unlikely(!dentry)) 1077baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1078baa03890SNick Piggin 1079baa03890SNick Piggin old = inode->i_op->lookup(inode, dentry, nd); 1080baa03890SNick Piggin if (unlikely(old)) { 1081baa03890SNick Piggin dput(dentry); 1082baa03890SNick Piggin dentry = old; 1083baa03890SNick Piggin } 1084baa03890SNick Piggin return dentry; 1085baa03890SNick Piggin } 1086baa03890SNick Piggin 1087baa03890SNick Piggin /* 108844396f4bSJosef Bacik * We already have a dentry, but require a lookup to be performed on the parent 108944396f4bSJosef Bacik * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error. 109044396f4bSJosef Bacik * parent->d_inode->i_mutex must be held. d_lookup must have verified that no 109144396f4bSJosef Bacik * child exists while under i_mutex. 109244396f4bSJosef Bacik */ 109344396f4bSJosef Bacik static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry, 109444396f4bSJosef Bacik struct nameidata *nd) 109544396f4bSJosef Bacik { 109644396f4bSJosef Bacik struct inode *inode = parent->d_inode; 109744396f4bSJosef Bacik struct dentry *old; 109844396f4bSJosef Bacik 109944396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 110044396f4bSJosef Bacik if (unlikely(IS_DEADDIR(inode))) 110144396f4bSJosef Bacik return ERR_PTR(-ENOENT); 110244396f4bSJosef Bacik 110344396f4bSJosef Bacik old = inode->i_op->lookup(inode, dentry, nd); 110444396f4bSJosef Bacik if (unlikely(old)) { 110544396f4bSJosef Bacik dput(dentry); 110644396f4bSJosef Bacik dentry = old; 110744396f4bSJosef Bacik } 110844396f4bSJosef Bacik return dentry; 110944396f4bSJosef Bacik } 111044396f4bSJosef Bacik 111144396f4bSJosef Bacik /* 11121da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 11131da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 11141da177e4SLinus Torvalds * It _is_ time-critical. 11151da177e4SLinus Torvalds */ 11161da177e4SLinus Torvalds static int do_lookup(struct nameidata *nd, struct qstr *name, 111731e6b01fSNick Piggin struct path *path, struct inode **inode) 11181da177e4SLinus Torvalds { 11194ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 112031e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 11215a18fff2SAl Viro int need_reval = 1; 11225a18fff2SAl Viro int status = 1; 11239875cf80SDavid Howells int err; 11249875cf80SDavid Howells 11253cac260aSAl Viro /* 1126b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1127b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1128b04f784eSNick Piggin * do the non-racy lookup, below. 1129b04f784eSNick Piggin */ 113031e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 113131e6b01fSNick Piggin unsigned seq; 113231e6b01fSNick Piggin *inode = nd->inode; 113331e6b01fSNick Piggin dentry = __d_lookup_rcu(parent, name, &seq, inode); 11345a18fff2SAl Viro if (!dentry) 11355a18fff2SAl Viro goto unlazy; 11365a18fff2SAl Viro 113731e6b01fSNick Piggin /* Memory barrier in read_seqcount_begin of child is enough */ 113831e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 113931e6b01fSNick Piggin return -ECHILD; 114031e6b01fSNick Piggin nd->seq = seq; 11415a18fff2SAl Viro 114224643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 11435a18fff2SAl Viro status = d_revalidate(dentry, nd); 11445a18fff2SAl Viro if (unlikely(status <= 0)) { 11455a18fff2SAl Viro if (status != -ECHILD) 11465a18fff2SAl Viro need_reval = 0; 11475a18fff2SAl Viro goto unlazy; 11485a18fff2SAl Viro } 114924643087SAl Viro } 115044396f4bSJosef Bacik if (unlikely(d_need_lookup(dentry))) 115144396f4bSJosef Bacik goto unlazy; 115231e6b01fSNick Piggin path->mnt = mnt; 115331e6b01fSNick Piggin path->dentry = dentry; 1154d6e9bd25SAl Viro if (unlikely(!__follow_mount_rcu(nd, path, inode))) 1155d6e9bd25SAl Viro goto unlazy; 1156d6e9bd25SAl Viro if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT)) 1157d6e9bd25SAl Viro goto unlazy; 11589875cf80SDavid Howells return 0; 11595a18fff2SAl Viro unlazy: 116019660af7SAl Viro if (unlazy_walk(nd, dentry)) 11615a18fff2SAl Viro return -ECHILD; 11625a18fff2SAl Viro } else { 116331e6b01fSNick Piggin dentry = __d_lookup(parent, name); 116424643087SAl Viro } 11655a18fff2SAl Viro 116644396f4bSJosef Bacik if (dentry && unlikely(d_need_lookup(dentry))) { 116744396f4bSJosef Bacik dput(dentry); 116844396f4bSJosef Bacik dentry = NULL; 116944396f4bSJosef Bacik } 11705a18fff2SAl Viro retry: 11715a18fff2SAl Viro if (unlikely(!dentry)) { 11725a18fff2SAl Viro struct inode *dir = parent->d_inode; 11735a18fff2SAl Viro BUG_ON(nd->inode != dir); 11745a18fff2SAl Viro 11755a18fff2SAl Viro mutex_lock(&dir->i_mutex); 11765a18fff2SAl Viro dentry = d_lookup(parent, name); 11775a18fff2SAl Viro if (likely(!dentry)) { 11785a18fff2SAl Viro dentry = d_alloc_and_lookup(parent, name, nd); 11795a18fff2SAl Viro if (IS_ERR(dentry)) { 11805a18fff2SAl Viro mutex_unlock(&dir->i_mutex); 11815a18fff2SAl Viro return PTR_ERR(dentry); 11825a18fff2SAl Viro } 11835a18fff2SAl Viro /* known good */ 11845a18fff2SAl Viro need_reval = 0; 11855a18fff2SAl Viro status = 1; 118644396f4bSJosef Bacik } else if (unlikely(d_need_lookup(dentry))) { 118744396f4bSJosef Bacik dentry = d_inode_lookup(parent, dentry, nd); 118844396f4bSJosef Bacik if (IS_ERR(dentry)) { 118944396f4bSJosef Bacik mutex_unlock(&dir->i_mutex); 119044396f4bSJosef Bacik return PTR_ERR(dentry); 119144396f4bSJosef Bacik } 119244396f4bSJosef Bacik /* known good */ 119344396f4bSJosef Bacik need_reval = 0; 119444396f4bSJosef Bacik status = 1; 11955a18fff2SAl Viro } 11965a18fff2SAl Viro mutex_unlock(&dir->i_mutex); 11975a18fff2SAl Viro } 11985a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 11995a18fff2SAl Viro status = d_revalidate(dentry, nd); 12005a18fff2SAl Viro if (unlikely(status <= 0)) { 12015a18fff2SAl Viro if (status < 0) { 12025a18fff2SAl Viro dput(dentry); 12035a18fff2SAl Viro return status; 12045a18fff2SAl Viro } 12055a18fff2SAl Viro if (!d_invalidate(dentry)) { 12065a18fff2SAl Viro dput(dentry); 12075a18fff2SAl Viro dentry = NULL; 12085a18fff2SAl Viro need_reval = 1; 12095a18fff2SAl Viro goto retry; 12105a18fff2SAl Viro } 12115a18fff2SAl Viro } 12125a18fff2SAl Viro 12131da177e4SLinus Torvalds path->mnt = mnt; 12141da177e4SLinus Torvalds path->dentry = dentry; 12159875cf80SDavid Howells err = follow_managed(path, nd->flags); 121689312214SIan Kent if (unlikely(err < 0)) { 121789312214SIan Kent path_put_conditional(path, nd); 12189875cf80SDavid Howells return err; 121989312214SIan Kent } 1220a3fbbde7SAl Viro if (err) 1221a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 122231e6b01fSNick Piggin *inode = path->dentry->d_inode; 12231da177e4SLinus Torvalds return 0; 12241da177e4SLinus Torvalds } 12251da177e4SLinus Torvalds 122652094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 122752094c8aSAl Viro { 122852094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 12294ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 123052094c8aSAl Viro if (err != -ECHILD) 123152094c8aSAl Viro return err; 123219660af7SAl Viro if (unlazy_walk(nd, NULL)) 123352094c8aSAl Viro return -ECHILD; 123452094c8aSAl Viro } 12354ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 123652094c8aSAl Viro } 123752094c8aSAl Viro 12389856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 12399856fa1bSAl Viro { 12409856fa1bSAl Viro if (type == LAST_DOTDOT) { 12419856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 12429856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 12439856fa1bSAl Viro return -ECHILD; 12449856fa1bSAl Viro } else 12459856fa1bSAl Viro follow_dotdot(nd); 12469856fa1bSAl Viro } 12479856fa1bSAl Viro return 0; 12489856fa1bSAl Viro } 12499856fa1bSAl Viro 1250951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1251951361f9SAl Viro { 1252951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1253951361f9SAl Viro path_put(&nd->path); 1254951361f9SAl Viro } else { 1255951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 12565b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1257951361f9SAl Viro nd->root.mnt = NULL; 1258951361f9SAl Viro rcu_read_unlock(); 1259951361f9SAl Viro br_read_unlock(vfsmount_lock); 1260951361f9SAl Viro } 1261951361f9SAl Viro } 1262951361f9SAl Viro 12633ddcd056SLinus Torvalds /* 12643ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 12653ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 12663ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 12673ddcd056SLinus Torvalds * for the common case. 12683ddcd056SLinus Torvalds */ 12697813b94aSLinus Torvalds static inline int should_follow_link(struct inode *inode, int follow) 12703ddcd056SLinus Torvalds { 12713ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) { 12723ddcd056SLinus Torvalds if (likely(inode->i_op->follow_link)) 12733ddcd056SLinus Torvalds return follow; 12743ddcd056SLinus Torvalds 12753ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 12763ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 12773ddcd056SLinus Torvalds inode->i_opflags |= IOP_NOFOLLOW; 12783ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 12793ddcd056SLinus Torvalds } 12803ddcd056SLinus Torvalds return 0; 12813ddcd056SLinus Torvalds } 12823ddcd056SLinus Torvalds 1283ce57dfc1SAl Viro static inline int walk_component(struct nameidata *nd, struct path *path, 1284ce57dfc1SAl Viro struct qstr *name, int type, int follow) 1285ce57dfc1SAl Viro { 1286ce57dfc1SAl Viro struct inode *inode; 1287ce57dfc1SAl Viro int err; 1288ce57dfc1SAl Viro /* 1289ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1290ce57dfc1SAl Viro * to be able to know about the current root directory and 1291ce57dfc1SAl Viro * parent relationships. 1292ce57dfc1SAl Viro */ 1293ce57dfc1SAl Viro if (unlikely(type != LAST_NORM)) 1294ce57dfc1SAl Viro return handle_dots(nd, type); 1295ce57dfc1SAl Viro err = do_lookup(nd, name, path, &inode); 1296ce57dfc1SAl Viro if (unlikely(err)) { 1297ce57dfc1SAl Viro terminate_walk(nd); 1298ce57dfc1SAl Viro return err; 1299ce57dfc1SAl Viro } 1300ce57dfc1SAl Viro if (!inode) { 1301ce57dfc1SAl Viro path_to_nameidata(path, nd); 1302ce57dfc1SAl Viro terminate_walk(nd); 1303ce57dfc1SAl Viro return -ENOENT; 1304ce57dfc1SAl Viro } 13057813b94aSLinus Torvalds if (should_follow_link(inode, follow)) { 130619660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 130719660af7SAl Viro if (unlikely(unlazy_walk(nd, path->dentry))) { 130819660af7SAl Viro terminate_walk(nd); 1309ce57dfc1SAl Viro return -ECHILD; 131019660af7SAl Viro } 131119660af7SAl Viro } 1312ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1313ce57dfc1SAl Viro return 1; 1314ce57dfc1SAl Viro } 1315ce57dfc1SAl Viro path_to_nameidata(path, nd); 1316ce57dfc1SAl Viro nd->inode = inode; 1317ce57dfc1SAl Viro return 0; 1318ce57dfc1SAl Viro } 1319ce57dfc1SAl Viro 13201da177e4SLinus Torvalds /* 1321b356379aSAl Viro * This limits recursive symlink follows to 8, while 1322b356379aSAl Viro * limiting consecutive symlinks to 40. 1323b356379aSAl Viro * 1324b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1325b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1326b356379aSAl Viro */ 1327b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1328b356379aSAl Viro { 1329b356379aSAl Viro int res; 1330b356379aSAl Viro 1331b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1332b356379aSAl Viro path_put_conditional(path, nd); 1333b356379aSAl Viro path_put(&nd->path); 1334b356379aSAl Viro return -ELOOP; 1335b356379aSAl Viro } 13361a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1337b356379aSAl Viro 1338b356379aSAl Viro nd->depth++; 1339b356379aSAl Viro current->link_count++; 1340b356379aSAl Viro 1341b356379aSAl Viro do { 1342b356379aSAl Viro struct path link = *path; 1343b356379aSAl Viro void *cookie; 1344574197e0SAl Viro 1345574197e0SAl Viro res = follow_link(&link, nd, &cookie); 1346b356379aSAl Viro if (!res) 1347b356379aSAl Viro res = walk_component(nd, path, &nd->last, 1348b356379aSAl Viro nd->last_type, LOOKUP_FOLLOW); 1349574197e0SAl Viro put_link(nd, &link, cookie); 1350b356379aSAl Viro } while (res > 0); 1351b356379aSAl Viro 1352b356379aSAl Viro current->link_count--; 1353b356379aSAl Viro nd->depth--; 1354b356379aSAl Viro return res; 1355b356379aSAl Viro } 1356b356379aSAl Viro 1357b356379aSAl Viro /* 13583ddcd056SLinus Torvalds * We really don't want to look at inode->i_op->lookup 13593ddcd056SLinus Torvalds * when we don't have to. So we keep a cache bit in 13603ddcd056SLinus Torvalds * the inode ->i_opflags field that says "yes, we can 13613ddcd056SLinus Torvalds * do lookup on this inode". 13623ddcd056SLinus Torvalds */ 13633ddcd056SLinus Torvalds static inline int can_lookup(struct inode *inode) 13643ddcd056SLinus Torvalds { 13653ddcd056SLinus Torvalds if (likely(inode->i_opflags & IOP_LOOKUP)) 13663ddcd056SLinus Torvalds return 1; 13673ddcd056SLinus Torvalds if (likely(!inode->i_op->lookup)) 13683ddcd056SLinus Torvalds return 0; 13693ddcd056SLinus Torvalds 13703ddcd056SLinus Torvalds /* We do this once for the lifetime of the inode */ 13713ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 13723ddcd056SLinus Torvalds inode->i_opflags |= IOP_LOOKUP; 13733ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 13743ddcd056SLinus Torvalds return 1; 13753ddcd056SLinus Torvalds } 13763ddcd056SLinus Torvalds 13773ddcd056SLinus Torvalds /* 13781da177e4SLinus Torvalds * Name resolution. 1379ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1380ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 13811da177e4SLinus Torvalds * 1382ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1383ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 13841da177e4SLinus Torvalds */ 13856de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 13861da177e4SLinus Torvalds { 13871da177e4SLinus Torvalds struct path next; 13881da177e4SLinus Torvalds int err; 13891da177e4SLinus Torvalds 13901da177e4SLinus Torvalds while (*name=='/') 13911da177e4SLinus Torvalds name++; 13921da177e4SLinus Torvalds if (!*name) 1393086e183aSAl Viro return 0; 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 13961da177e4SLinus Torvalds for(;;) { 13971da177e4SLinus Torvalds unsigned long hash; 13981da177e4SLinus Torvalds struct qstr this; 13991da177e4SLinus Torvalds unsigned int c; 1400fe479a58SAl Viro int type; 14011da177e4SLinus Torvalds 140252094c8aSAl Viro err = may_lookup(nd); 14031da177e4SLinus Torvalds if (err) 14041da177e4SLinus Torvalds break; 14051da177e4SLinus Torvalds 14061da177e4SLinus Torvalds this.name = name; 14071da177e4SLinus Torvalds c = *(const unsigned char *)name; 14081da177e4SLinus Torvalds 14091da177e4SLinus Torvalds hash = init_name_hash(); 14101da177e4SLinus Torvalds do { 14111da177e4SLinus Torvalds name++; 14121da177e4SLinus Torvalds hash = partial_name_hash(c, hash); 14131da177e4SLinus Torvalds c = *(const unsigned char *)name; 14141da177e4SLinus Torvalds } while (c && (c != '/')); 14151da177e4SLinus Torvalds this.len = name - (const char *) this.name; 14161da177e4SLinus Torvalds this.hash = end_name_hash(hash); 14171da177e4SLinus Torvalds 1418fe479a58SAl Viro type = LAST_NORM; 1419fe479a58SAl Viro if (this.name[0] == '.') switch (this.len) { 1420fe479a58SAl Viro case 2: 142116c2cd71SAl Viro if (this.name[1] == '.') { 1422fe479a58SAl Viro type = LAST_DOTDOT; 142316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 142416c2cd71SAl Viro } 1425fe479a58SAl Viro break; 1426fe479a58SAl Viro case 1: 1427fe479a58SAl Viro type = LAST_DOT; 1428fe479a58SAl Viro } 14295a202bcdSAl Viro if (likely(type == LAST_NORM)) { 14305a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 143116c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 14325a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 14335a202bcdSAl Viro err = parent->d_op->d_hash(parent, nd->inode, 14345a202bcdSAl Viro &this); 14355a202bcdSAl Viro if (err < 0) 14365a202bcdSAl Viro break; 14375a202bcdSAl Viro } 14385a202bcdSAl Viro } 1439fe479a58SAl Viro 14401da177e4SLinus Torvalds /* remove trailing slashes? */ 14411da177e4SLinus Torvalds if (!c) 14421da177e4SLinus Torvalds goto last_component; 14431da177e4SLinus Torvalds while (*++name == '/'); 14441da177e4SLinus Torvalds if (!*name) 1445b356379aSAl Viro goto last_component; 14461da177e4SLinus Torvalds 1447ce57dfc1SAl Viro err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW); 1448ce57dfc1SAl Viro if (err < 0) 1449ce57dfc1SAl Viro return err; 1450fe479a58SAl Viro 1451ce57dfc1SAl Viro if (err) { 1452b356379aSAl Viro err = nested_symlink(&next, nd); 14531da177e4SLinus Torvalds if (err) 1454a7472babSAl Viro return err; 145531e6b01fSNick Piggin } 14563ddcd056SLinus Torvalds if (can_lookup(nd->inode)) 14571da177e4SLinus Torvalds continue; 14583ddcd056SLinus Torvalds err = -ENOTDIR; 14593ddcd056SLinus Torvalds break; 14601da177e4SLinus Torvalds /* here ends the main loop */ 14611da177e4SLinus Torvalds 14621da177e4SLinus Torvalds last_component: 1463ce57dfc1SAl Viro nd->last = this; 1464ce57dfc1SAl Viro nd->last_type = type; 1465ce57dfc1SAl Viro return 0; 1466ce57dfc1SAl Viro } 1467951361f9SAl Viro terminate_walk(nd); 14681da177e4SLinus Torvalds return err; 14691da177e4SLinus Torvalds } 14701da177e4SLinus Torvalds 147170e9b357SAl Viro static int path_init(int dfd, const char *name, unsigned int flags, 147270e9b357SAl Viro struct nameidata *nd, struct file **fp) 147331e6b01fSNick Piggin { 147431e6b01fSNick Piggin int retval = 0; 147531e6b01fSNick Piggin int fput_needed; 147631e6b01fSNick Piggin struct file *file; 147731e6b01fSNick Piggin 147831e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 147916c2cd71SAl Viro nd->flags = flags | LOOKUP_JUMPED; 148031e6b01fSNick Piggin nd->depth = 0; 14815b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 14825b6ca027SAl Viro struct inode *inode = nd->root.dentry->d_inode; 148373d049a4SAl Viro if (*name) { 14845b6ca027SAl Viro if (!inode->i_op->lookup) 14855b6ca027SAl Viro return -ENOTDIR; 14865b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 14875b6ca027SAl Viro if (retval) 14885b6ca027SAl Viro return retval; 148973d049a4SAl Viro } 14905b6ca027SAl Viro nd->path = nd->root; 14915b6ca027SAl Viro nd->inode = inode; 14925b6ca027SAl Viro if (flags & LOOKUP_RCU) { 14935b6ca027SAl Viro br_read_lock(vfsmount_lock); 14945b6ca027SAl Viro rcu_read_lock(); 14955b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 14965b6ca027SAl Viro } else { 14975b6ca027SAl Viro path_get(&nd->path); 14985b6ca027SAl Viro } 14995b6ca027SAl Viro return 0; 15005b6ca027SAl Viro } 15015b6ca027SAl Viro 150231e6b01fSNick Piggin nd->root.mnt = NULL; 150331e6b01fSNick Piggin 150431e6b01fSNick Piggin if (*name=='/') { 1505e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 150631e6b01fSNick Piggin br_read_lock(vfsmount_lock); 150731e6b01fSNick Piggin rcu_read_lock(); 1508e41f7d4eSAl Viro set_root_rcu(nd); 1509e41f7d4eSAl Viro } else { 1510e41f7d4eSAl Viro set_root(nd); 1511e41f7d4eSAl Viro path_get(&nd->root); 1512e41f7d4eSAl Viro } 151331e6b01fSNick Piggin nd->path = nd->root; 151431e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1515e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 151631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1517c28cc364SNick Piggin unsigned seq; 151831e6b01fSNick Piggin 151931e6b01fSNick Piggin br_read_lock(vfsmount_lock); 152031e6b01fSNick Piggin rcu_read_lock(); 152131e6b01fSNick Piggin 1522c28cc364SNick Piggin do { 1523c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 152431e6b01fSNick Piggin nd->path = fs->pwd; 1525c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1526c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1527e41f7d4eSAl Viro } else { 1528e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1529e41f7d4eSAl Viro } 153031e6b01fSNick Piggin } else { 153131e6b01fSNick Piggin struct dentry *dentry; 153231e6b01fSNick Piggin 15331abf0c71SAl Viro file = fget_raw_light(dfd, &fput_needed); 153431e6b01fSNick Piggin retval = -EBADF; 153531e6b01fSNick Piggin if (!file) 153631e6b01fSNick Piggin goto out_fail; 153731e6b01fSNick Piggin 153831e6b01fSNick Piggin dentry = file->f_path.dentry; 153931e6b01fSNick Piggin 1540f52e0c11SAl Viro if (*name) { 154131e6b01fSNick Piggin retval = -ENOTDIR; 154231e6b01fSNick Piggin if (!S_ISDIR(dentry->d_inode->i_mode)) 154331e6b01fSNick Piggin goto fput_fail; 154431e6b01fSNick Piggin 15454ad5abb3SAl Viro retval = inode_permission(dentry->d_inode, MAY_EXEC); 154631e6b01fSNick Piggin if (retval) 154731e6b01fSNick Piggin goto fput_fail; 1548f52e0c11SAl Viro } 154931e6b01fSNick Piggin 155031e6b01fSNick Piggin nd->path = file->f_path; 1551e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 155231e6b01fSNick Piggin if (fput_needed) 155370e9b357SAl Viro *fp = file; 1554c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 155531e6b01fSNick Piggin br_read_lock(vfsmount_lock); 155631e6b01fSNick Piggin rcu_read_lock(); 15575590ff0dSUlrich Drepper } else { 15585dd784d0SJan Blunck path_get(&file->f_path); 15595590ff0dSUlrich Drepper fput_light(file, fput_needed); 15601da177e4SLinus Torvalds } 1561e41f7d4eSAl Viro } 1562e41f7d4eSAl Viro 156331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 15649b4a9b14SAl Viro return 0; 15652dfdd266SJosef 'Jeff' Sipek 15669b4a9b14SAl Viro fput_fail: 15679b4a9b14SAl Viro fput_light(file, fput_needed); 15689b4a9b14SAl Viro out_fail: 15699b4a9b14SAl Viro return retval; 15709b4a9b14SAl Viro } 15719b4a9b14SAl Viro 1572bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1573bd92d7feSAl Viro { 1574bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1575bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1576bd92d7feSAl Viro 1577bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 1578bd92d7feSAl Viro return walk_component(nd, path, &nd->last, nd->last_type, 1579bd92d7feSAl Viro nd->flags & LOOKUP_FOLLOW); 1580bd92d7feSAl Viro } 1581bd92d7feSAl Viro 15829b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 1583ee0827cdSAl Viro static int path_lookupat(int dfd, const char *name, 15849b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 15859b4a9b14SAl Viro { 158670e9b357SAl Viro struct file *base = NULL; 1587bd92d7feSAl Viro struct path path; 1588bd92d7feSAl Viro int err; 158931e6b01fSNick Piggin 159031e6b01fSNick Piggin /* 159131e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 159231e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 159331e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 159431e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 159531e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 159631e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 159731e6b01fSNick Piggin * analogue, foo_rcu(). 159831e6b01fSNick Piggin * 159931e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 160031e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 160131e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 160231e6b01fSNick Piggin * be able to complete). 160331e6b01fSNick Piggin */ 1604bd92d7feSAl Viro err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base); 1605ee0827cdSAl Viro 1606bd92d7feSAl Viro if (unlikely(err)) 1607bd92d7feSAl Viro return err; 1608ee0827cdSAl Viro 1609ee0827cdSAl Viro current->total_link_count = 0; 1610bd92d7feSAl Viro err = link_path_walk(name, nd); 1611bd92d7feSAl Viro 1612bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 1613bd92d7feSAl Viro err = lookup_last(nd, &path); 1614bd92d7feSAl Viro while (err > 0) { 1615bd92d7feSAl Viro void *cookie; 1616bd92d7feSAl Viro struct path link = path; 1617bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 1618574197e0SAl Viro err = follow_link(&link, nd, &cookie); 1619bd92d7feSAl Viro if (!err) 1620bd92d7feSAl Viro err = lookup_last(nd, &path); 1621574197e0SAl Viro put_link(nd, &link, cookie); 1622bd92d7feSAl Viro } 1623bd92d7feSAl Viro } 1624ee0827cdSAl Viro 16259f1fafeeSAl Viro if (!err) 16269f1fafeeSAl Viro err = complete_walk(nd); 1627bd92d7feSAl Viro 1628bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 1629bd92d7feSAl Viro if (!nd->inode->i_op->lookup) { 1630bd92d7feSAl Viro path_put(&nd->path); 1631bd23a539SAl Viro err = -ENOTDIR; 1632bd92d7feSAl Viro } 1633bd92d7feSAl Viro } 163416c2cd71SAl Viro 163570e9b357SAl Viro if (base) 163670e9b357SAl Viro fput(base); 1637ee0827cdSAl Viro 16385b6ca027SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 163931e6b01fSNick Piggin path_put(&nd->root); 164031e6b01fSNick Piggin nd->root.mnt = NULL; 164131e6b01fSNick Piggin } 1642bd92d7feSAl Viro return err; 164331e6b01fSNick Piggin } 164431e6b01fSNick Piggin 1645ee0827cdSAl Viro static int do_path_lookup(int dfd, const char *name, 1646ee0827cdSAl Viro unsigned int flags, struct nameidata *nd) 1647ee0827cdSAl Viro { 1648ee0827cdSAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 1649ee0827cdSAl Viro if (unlikely(retval == -ECHILD)) 1650ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags, nd); 1651ee0827cdSAl Viro if (unlikely(retval == -ESTALE)) 1652ee0827cdSAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 1653ee0827cdSAl Viro 165431e6b01fSNick Piggin if (likely(!retval)) { 165531e6b01fSNick Piggin if (unlikely(!audit_dummy_context())) { 165631e6b01fSNick Piggin if (nd->path.dentry && nd->inode) 165731e6b01fSNick Piggin audit_inode(name, nd->path.dentry); 165831e6b01fSNick Piggin } 165931e6b01fSNick Piggin } 1660170aa3d0SUlrich Drepper return retval; 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 1663c9c6cac0SAl Viro int kern_path_parent(const char *name, struct nameidata *nd) 16645590ff0dSUlrich Drepper { 1665c9c6cac0SAl Viro return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd); 16665590ff0dSUlrich Drepper } 16675590ff0dSUlrich Drepper 1668d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 1669d1811465SAl Viro { 1670d1811465SAl Viro struct nameidata nd; 1671d1811465SAl Viro int res = do_path_lookup(AT_FDCWD, name, flags, &nd); 1672d1811465SAl Viro if (!res) 1673d1811465SAl Viro *path = nd.path; 1674d1811465SAl Viro return res; 1675d1811465SAl Viro } 1676d1811465SAl Viro 167716f18200SJosef 'Jeff' Sipek /** 167816f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 167916f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 168016f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 168116f18200SJosef 'Jeff' Sipek * @name: pointer to file name 168216f18200SJosef 'Jeff' Sipek * @flags: lookup flags 1683e0a01249SAl Viro * @path: pointer to struct path to fill 168416f18200SJosef 'Jeff' Sipek */ 168516f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 168616f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 1687e0a01249SAl Viro struct path *path) 168816f18200SJosef 'Jeff' Sipek { 1689e0a01249SAl Viro struct nameidata nd; 1690e0a01249SAl Viro int err; 1691e0a01249SAl Viro nd.root.dentry = dentry; 1692e0a01249SAl Viro nd.root.mnt = mnt; 1693e0a01249SAl Viro BUG_ON(flags & LOOKUP_PARENT); 16945b6ca027SAl Viro /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */ 1695e0a01249SAl Viro err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd); 1696e0a01249SAl Viro if (!err) 1697e0a01249SAl Viro *path = nd.path; 1698e0a01249SAl Viro return err; 169916f18200SJosef 'Jeff' Sipek } 170016f18200SJosef 'Jeff' Sipek 1701eead1911SChristoph Hellwig static struct dentry *__lookup_hash(struct qstr *name, 1702eead1911SChristoph Hellwig struct dentry *base, struct nameidata *nd) 17031da177e4SLinus Torvalds { 170481fca444SChristoph Hellwig struct inode *inode = base->d_inode; 17051da177e4SLinus Torvalds struct dentry *dentry; 17061da177e4SLinus Torvalds int err; 17071da177e4SLinus Torvalds 17084ad5abb3SAl Viro err = inode_permission(inode, MAY_EXEC); 170981fca444SChristoph Hellwig if (err) 171081fca444SChristoph Hellwig return ERR_PTR(err); 17111da177e4SLinus Torvalds 17121da177e4SLinus Torvalds /* 1713b04f784eSNick Piggin * Don't bother with __d_lookup: callers are for creat as 1714b04f784eSNick Piggin * well as unlink, so a lot of the time it would cost 1715b04f784eSNick Piggin * a double lookup. 17166e6b1bd1SAl Viro */ 17176e6b1bd1SAl Viro dentry = d_lookup(base, name); 17186e6b1bd1SAl Viro 171944396f4bSJosef Bacik if (dentry && d_need_lookup(dentry)) { 172044396f4bSJosef Bacik /* 172144396f4bSJosef Bacik * __lookup_hash is called with the parent dir's i_mutex already 172244396f4bSJosef Bacik * held, so we are good to go here. 172344396f4bSJosef Bacik */ 172444396f4bSJosef Bacik dentry = d_inode_lookup(base, dentry, nd); 172544396f4bSJosef Bacik if (IS_ERR(dentry)) 172644396f4bSJosef Bacik return dentry; 172744396f4bSJosef Bacik } 172844396f4bSJosef Bacik 1729d2d9e9fbSAl Viro if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) { 1730d2d9e9fbSAl Viro int status = d_revalidate(dentry, nd); 1731d2d9e9fbSAl Viro if (unlikely(status <= 0)) { 1732d2d9e9fbSAl Viro /* 1733d2d9e9fbSAl Viro * The dentry failed validation. 1734d2d9e9fbSAl Viro * If d_revalidate returned 0 attempt to invalidate 1735d2d9e9fbSAl Viro * the dentry otherwise d_revalidate is asking us 1736d2d9e9fbSAl Viro * to return a fail status. 1737d2d9e9fbSAl Viro */ 1738d2d9e9fbSAl Viro if (status < 0) { 1739d2d9e9fbSAl Viro dput(dentry); 1740d2d9e9fbSAl Viro return ERR_PTR(status); 1741d2d9e9fbSAl Viro } else if (!d_invalidate(dentry)) { 1742d2d9e9fbSAl Viro dput(dentry); 1743d2d9e9fbSAl Viro dentry = NULL; 1744d2d9e9fbSAl Viro } 1745d2d9e9fbSAl Viro } 1746d2d9e9fbSAl Viro } 17476e6b1bd1SAl Viro 17481da177e4SLinus Torvalds if (!dentry) 1749baa03890SNick Piggin dentry = d_alloc_and_lookup(base, name, nd); 17505a202bcdSAl Viro 17511da177e4SLinus Torvalds return dentry; 17521da177e4SLinus Torvalds } 17531da177e4SLinus Torvalds 1754057f6c01SJames Morris /* 1755057f6c01SJames Morris * Restricted form of lookup. Doesn't follow links, single-component only, 1756057f6c01SJames Morris * needs parent already locked. Doesn't follow mounts. 1757057f6c01SJames Morris * SMP-safe. 1758057f6c01SJames Morris */ 1759a244e169SAdrian Bunk static struct dentry *lookup_hash(struct nameidata *nd) 17601da177e4SLinus Torvalds { 17614ac91378SJan Blunck return __lookup_hash(&nd->last, nd->path.dentry, nd); 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds 1764eead1911SChristoph Hellwig /** 1765a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 1766eead1911SChristoph Hellwig * @name: pathname component to lookup 1767eead1911SChristoph Hellwig * @base: base directory to lookup from 1768eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 1769eead1911SChristoph Hellwig * 1770a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 1771a6b91919SRandy Dunlap * not be called by generic code. Also note that by using this function the 1772eead1911SChristoph Hellwig * nameidata argument is passed to the filesystem methods and a filesystem 1773eead1911SChristoph Hellwig * using this helper needs to be prepared for that. 1774eead1911SChristoph Hellwig */ 1775057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 1776057f6c01SJames Morris { 1777057f6c01SJames Morris struct qstr this; 17786a96ba54SAl Viro unsigned long hash; 17796a96ba54SAl Viro unsigned int c; 1780057f6c01SJames Morris 17812f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 17822f9092e1SDavid Woodhouse 17836a96ba54SAl Viro this.name = name; 17846a96ba54SAl Viro this.len = len; 17856a96ba54SAl Viro if (!len) 17866a96ba54SAl Viro return ERR_PTR(-EACCES); 17876a96ba54SAl Viro 17886a96ba54SAl Viro hash = init_name_hash(); 17896a96ba54SAl Viro while (len--) { 17906a96ba54SAl Viro c = *(const unsigned char *)name++; 17916a96ba54SAl Viro if (c == '/' || c == '\0') 17926a96ba54SAl Viro return ERR_PTR(-EACCES); 17936a96ba54SAl Viro hash = partial_name_hash(c, hash); 17946a96ba54SAl Viro } 17956a96ba54SAl Viro this.hash = end_name_hash(hash); 17965a202bcdSAl Viro /* 17975a202bcdSAl Viro * See if the low-level filesystem might want 17985a202bcdSAl Viro * to use its own hash.. 17995a202bcdSAl Viro */ 18005a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 18015a202bcdSAl Viro int err = base->d_op->d_hash(base, base->d_inode, &this); 18025a202bcdSAl Viro if (err < 0) 18035a202bcdSAl Viro return ERR_PTR(err); 18045a202bcdSAl Viro } 1805eead1911SChristoph Hellwig 180649705b77SChristoph Hellwig return __lookup_hash(&this, base, NULL); 1807057f6c01SJames Morris } 1808057f6c01SJames Morris 18091fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 18101fa1e7f6SAndy Whitcroft struct path *path, int *empty) 18111da177e4SLinus Torvalds { 18122d8f3038SAl Viro struct nameidata nd; 18131fa1e7f6SAndy Whitcroft char *tmp = getname_flags(name, flags, empty); 18141da177e4SLinus Torvalds int err = PTR_ERR(tmp); 18151da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 18162d8f3038SAl Viro 18172d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 18182d8f3038SAl Viro 18192d8f3038SAl Viro err = do_path_lookup(dfd, tmp, flags, &nd); 18201da177e4SLinus Torvalds putname(tmp); 18212d8f3038SAl Viro if (!err) 18222d8f3038SAl Viro *path = nd.path; 18231da177e4SLinus Torvalds } 18241da177e4SLinus Torvalds return err; 18251da177e4SLinus Torvalds } 18261da177e4SLinus Torvalds 18271fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 18281fa1e7f6SAndy Whitcroft struct path *path) 18291fa1e7f6SAndy Whitcroft { 18301fa1e7f6SAndy Whitcroft return user_path_at_empty(dfd, name, flags, path, 0); 18311fa1e7f6SAndy Whitcroft } 18321fa1e7f6SAndy Whitcroft 18332ad94ae6SAl Viro static int user_path_parent(int dfd, const char __user *path, 18342ad94ae6SAl Viro struct nameidata *nd, char **name) 18352ad94ae6SAl Viro { 18362ad94ae6SAl Viro char *s = getname(path); 18372ad94ae6SAl Viro int error; 18382ad94ae6SAl Viro 18392ad94ae6SAl Viro if (IS_ERR(s)) 18402ad94ae6SAl Viro return PTR_ERR(s); 18412ad94ae6SAl Viro 18422ad94ae6SAl Viro error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd); 18432ad94ae6SAl Viro if (error) 18442ad94ae6SAl Viro putname(s); 18452ad94ae6SAl Viro else 18462ad94ae6SAl Viro *name = s; 18472ad94ae6SAl Viro 18482ad94ae6SAl Viro return error; 18492ad94ae6SAl Viro } 18502ad94ae6SAl Viro 18511da177e4SLinus Torvalds /* 18521da177e4SLinus Torvalds * It's inline, so penalty for filesystems that don't use sticky bit is 18531da177e4SLinus Torvalds * minimal. 18541da177e4SLinus Torvalds */ 18551da177e4SLinus Torvalds static inline int check_sticky(struct inode *dir, struct inode *inode) 18561da177e4SLinus Torvalds { 1857da9592edSDavid Howells uid_t fsuid = current_fsuid(); 1858da9592edSDavid Howells 18591da177e4SLinus Torvalds if (!(dir->i_mode & S_ISVTX)) 18601da177e4SLinus Torvalds return 0; 1861e795b717SSerge E. Hallyn if (current_user_ns() != inode_userns(inode)) 1862e795b717SSerge E. Hallyn goto other_userns; 1863da9592edSDavid Howells if (inode->i_uid == fsuid) 18641da177e4SLinus Torvalds return 0; 1865da9592edSDavid Howells if (dir->i_uid == fsuid) 18661da177e4SLinus Torvalds return 0; 1867e795b717SSerge E. Hallyn 1868e795b717SSerge E. Hallyn other_userns: 1869e795b717SSerge E. Hallyn return !ns_capable(inode_userns(inode), CAP_FOWNER); 18701da177e4SLinus Torvalds } 18711da177e4SLinus Torvalds 18721da177e4SLinus Torvalds /* 18731da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 18741da177e4SLinus Torvalds * whether the type of victim is right. 18751da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 18761da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 18771da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 18781da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 18791da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 18801da177e4SLinus Torvalds * a. be owner of dir, or 18811da177e4SLinus Torvalds * b. be owner of victim, or 18821da177e4SLinus Torvalds * c. have CAP_FOWNER capability 18831da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 18841da177e4SLinus Torvalds * links pointing to it. 18851da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 18861da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 18871da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 18881da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 18891da177e4SLinus Torvalds * nfs_async_unlink(). 18901da177e4SLinus Torvalds */ 1891858119e1SArjan van de Ven static int may_delete(struct inode *dir,struct dentry *victim,int isdir) 18921da177e4SLinus Torvalds { 18931da177e4SLinus Torvalds int error; 18941da177e4SLinus Torvalds 18951da177e4SLinus Torvalds if (!victim->d_inode) 18961da177e4SLinus Torvalds return -ENOENT; 18971da177e4SLinus Torvalds 18981da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 1899cccc6bbaSAl Viro audit_inode_child(victim, dir); 19001da177e4SLinus Torvalds 1901f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 19021da177e4SLinus Torvalds if (error) 19031da177e4SLinus Torvalds return error; 19041da177e4SLinus Torvalds if (IS_APPEND(dir)) 19051da177e4SLinus Torvalds return -EPERM; 19061da177e4SLinus Torvalds if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)|| 1907f9454548SHugh Dickins IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode)) 19081da177e4SLinus Torvalds return -EPERM; 19091da177e4SLinus Torvalds if (isdir) { 19101da177e4SLinus Torvalds if (!S_ISDIR(victim->d_inode->i_mode)) 19111da177e4SLinus Torvalds return -ENOTDIR; 19121da177e4SLinus Torvalds if (IS_ROOT(victim)) 19131da177e4SLinus Torvalds return -EBUSY; 19141da177e4SLinus Torvalds } else if (S_ISDIR(victim->d_inode->i_mode)) 19151da177e4SLinus Torvalds return -EISDIR; 19161da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 19171da177e4SLinus Torvalds return -ENOENT; 19181da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 19191da177e4SLinus Torvalds return -EBUSY; 19201da177e4SLinus Torvalds return 0; 19211da177e4SLinus Torvalds } 19221da177e4SLinus Torvalds 19231da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 19241da177e4SLinus Torvalds * dir. 19251da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 19261da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 19271da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 19281da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 19291da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 19301da177e4SLinus Torvalds */ 1931a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 19321da177e4SLinus Torvalds { 19331da177e4SLinus Torvalds if (child->d_inode) 19341da177e4SLinus Torvalds return -EEXIST; 19351da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 19361da177e4SLinus Torvalds return -ENOENT; 1937f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 19381da177e4SLinus Torvalds } 19391da177e4SLinus Torvalds 19401da177e4SLinus Torvalds /* 19411da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 19421da177e4SLinus Torvalds */ 19431da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 19441da177e4SLinus Torvalds { 19451da177e4SLinus Torvalds struct dentry *p; 19461da177e4SLinus Torvalds 19471da177e4SLinus Torvalds if (p1 == p2) { 1948f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 19491da177e4SLinus Torvalds return NULL; 19501da177e4SLinus Torvalds } 19511da177e4SLinus Torvalds 1952a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19531da177e4SLinus Torvalds 1954e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 1955e2761a11SOGAWA Hirofumi if (p) { 1956f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 1957f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 19581da177e4SLinus Torvalds return p; 19591da177e4SLinus Torvalds } 19601da177e4SLinus Torvalds 1961e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 1962e2761a11SOGAWA Hirofumi if (p) { 1963f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1964f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19651da177e4SLinus Torvalds return p; 19661da177e4SLinus Torvalds } 19671da177e4SLinus Torvalds 1968f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 1969f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 19701da177e4SLinus Torvalds return NULL; 19711da177e4SLinus Torvalds } 19721da177e4SLinus Torvalds 19731da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 19741da177e4SLinus Torvalds { 19751b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 19761da177e4SLinus Torvalds if (p1 != p2) { 19771b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 1978a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 19791da177e4SLinus Torvalds } 19801da177e4SLinus Torvalds } 19811da177e4SLinus Torvalds 19821da177e4SLinus Torvalds int vfs_create(struct inode *dir, struct dentry *dentry, int mode, 19831da177e4SLinus Torvalds struct nameidata *nd) 19841da177e4SLinus Torvalds { 1985a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 19861da177e4SLinus Torvalds 19871da177e4SLinus Torvalds if (error) 19881da177e4SLinus Torvalds return error; 19891da177e4SLinus Torvalds 1990acfa4380SAl Viro if (!dir->i_op->create) 19911da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 19921da177e4SLinus Torvalds mode &= S_IALLUGO; 19931da177e4SLinus Torvalds mode |= S_IFREG; 19941da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 19951da177e4SLinus Torvalds if (error) 19961da177e4SLinus Torvalds return error; 19971da177e4SLinus Torvalds error = dir->i_op->create(dir, dentry, mode, nd); 1998a74574aaSStephen Smalley if (!error) 1999f38aa942SAmy Griffis fsnotify_create(dir, dentry); 20001da177e4SLinus Torvalds return error; 20011da177e4SLinus Torvalds } 20021da177e4SLinus Torvalds 200373d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 20041da177e4SLinus Torvalds { 20053fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 20061da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 20071da177e4SLinus Torvalds int error; 20081da177e4SLinus Torvalds 2009bcda7652SAl Viro /* O_PATH? */ 2010bcda7652SAl Viro if (!acc_mode) 2011bcda7652SAl Viro return 0; 2012bcda7652SAl Viro 20131da177e4SLinus Torvalds if (!inode) 20141da177e4SLinus Torvalds return -ENOENT; 20151da177e4SLinus Torvalds 2016c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2017c8fe8f30SChristoph Hellwig case S_IFLNK: 20181da177e4SLinus Torvalds return -ELOOP; 2019c8fe8f30SChristoph Hellwig case S_IFDIR: 2020c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 20211da177e4SLinus Torvalds return -EISDIR; 2022c8fe8f30SChristoph Hellwig break; 2023c8fe8f30SChristoph Hellwig case S_IFBLK: 2024c8fe8f30SChristoph Hellwig case S_IFCHR: 20253fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 20261da177e4SLinus Torvalds return -EACCES; 2027c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2028c8fe8f30SChristoph Hellwig case S_IFIFO: 2029c8fe8f30SChristoph Hellwig case S_IFSOCK: 20301da177e4SLinus Torvalds flag &= ~O_TRUNC; 2031c8fe8f30SChristoph Hellwig break; 20324a3fd211SDave Hansen } 2033b41572e9SDave Hansen 20343fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2035b41572e9SDave Hansen if (error) 2036b41572e9SDave Hansen return error; 20376146f0d5SMimi Zohar 20381da177e4SLinus Torvalds /* 20391da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 20401da177e4SLinus Torvalds */ 20411da177e4SLinus Torvalds if (IS_APPEND(inode)) { 20428737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 20437715b521SAl Viro return -EPERM; 20441da177e4SLinus Torvalds if (flag & O_TRUNC) 20457715b521SAl Viro return -EPERM; 20461da177e4SLinus Torvalds } 20471da177e4SLinus Torvalds 20481da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 20492e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 20507715b521SAl Viro return -EPERM; 20511da177e4SLinus Torvalds 2052f3c7691eSJ. Bruce Fields return 0; 20537715b521SAl Viro } 20547715b521SAl Viro 2055e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 20567715b521SAl Viro { 2057e1181ee6SJeff Layton struct path *path = &filp->f_path; 20587715b521SAl Viro struct inode *inode = path->dentry->d_inode; 20597715b521SAl Viro int error = get_write_access(inode); 20601da177e4SLinus Torvalds if (error) 20617715b521SAl Viro return error; 20621da177e4SLinus Torvalds /* 20631da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 20641da177e4SLinus Torvalds */ 20651da177e4SLinus Torvalds error = locks_verify_locked(inode); 2066be6d3e56SKentaro Takeda if (!error) 2067ea0d3ab2STetsuo Handa error = security_path_truncate(path); 20681da177e4SLinus Torvalds if (!error) { 20697715b521SAl Viro error = do_truncate(path->dentry, 0, 2070d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2071e1181ee6SJeff Layton filp); 20721da177e4SLinus Torvalds } 20731da177e4SLinus Torvalds put_write_access(inode); 2074acd0c935SMimi Zohar return error; 20751da177e4SLinus Torvalds } 20761da177e4SLinus Torvalds 2077d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2078d57999e1SDave Hansen { 20798a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 20808a5e929dSAl Viro flag--; 2081d57999e1SDave Hansen return flag; 2082d57999e1SDave Hansen } 2083d57999e1SDave Hansen 208431e6b01fSNick Piggin /* 2085fe2d35ffSAl Viro * Handle the last step of open() 208631e6b01fSNick Piggin */ 2087fb1cc555SAl Viro static struct file *do_last(struct nameidata *nd, struct path *path, 2088c3e380b0SAl Viro const struct open_flags *op, const char *pathname) 2089fb1cc555SAl Viro { 2090a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 20916c0d46c4SAl Viro struct dentry *dentry; 2092ca344a89SAl Viro int open_flag = op->open_flag; 20936c0d46c4SAl Viro int will_truncate = open_flag & O_TRUNC; 2094ca344a89SAl Viro int want_write = 0; 2095bcda7652SAl Viro int acc_mode = op->acc_mode; 2096fb1cc555SAl Viro struct file *filp; 209716c2cd71SAl Viro int error; 2098fb1cc555SAl Viro 2099c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2100c3e380b0SAl Viro nd->flags |= op->intent; 2101c3e380b0SAl Viro 21021f36f774SAl Viro switch (nd->last_type) { 21031f36f774SAl Viro case LAST_DOTDOT: 2104176306f5SNeil Brown case LAST_DOT: 2105fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2106fe2d35ffSAl Viro if (error) 2107fe2d35ffSAl Viro return ERR_PTR(error); 21081f36f774SAl Viro /* fallthrough */ 21091f36f774SAl Viro case LAST_ROOT: 21109f1fafeeSAl Viro error = complete_walk(nd); 211116c2cd71SAl Viro if (error) 21129f1fafeeSAl Viro return ERR_PTR(error); 2113fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2114ca344a89SAl Viro if (open_flag & O_CREAT) { 211516c2cd71SAl Viro error = -EISDIR; 21161f36f774SAl Viro goto exit; 2117fe2d35ffSAl Viro } 2118fe2d35ffSAl Viro goto ok; 21191f36f774SAl Viro case LAST_BIND: 21209f1fafeeSAl Viro error = complete_walk(nd); 212116c2cd71SAl Viro if (error) 21229f1fafeeSAl Viro return ERR_PTR(error); 21231f36f774SAl Viro audit_inode(pathname, dir); 21241f36f774SAl Viro goto ok; 21251f36f774SAl Viro } 2126a2c36b45SAl Viro 2127ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2128bcda7652SAl Viro int symlink_ok = 0; 2129fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2130fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2131bcda7652SAl Viro if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW)) 2132bcda7652SAl Viro symlink_ok = 1; 2133fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2134ce57dfc1SAl Viro error = walk_component(nd, path, &nd->last, LAST_NORM, 2135ce57dfc1SAl Viro !symlink_ok); 2136ce57dfc1SAl Viro if (error < 0) 2137fe2d35ffSAl Viro return ERR_PTR(error); 2138ce57dfc1SAl Viro if (error) /* symlink */ 2139fe2d35ffSAl Viro return NULL; 2140fe2d35ffSAl Viro /* sayonara */ 21419f1fafeeSAl Viro error = complete_walk(nd); 21429f1fafeeSAl Viro if (error) 2143fe2d35ffSAl Viro return ERR_PTR(-ECHILD); 2144fe2d35ffSAl Viro 2145fe2d35ffSAl Viro error = -ENOTDIR; 2146fe2d35ffSAl Viro if (nd->flags & LOOKUP_DIRECTORY) { 2147ce57dfc1SAl Viro if (!nd->inode->i_op->lookup) 2148fe2d35ffSAl Viro goto exit; 2149fe2d35ffSAl Viro } 2150fe2d35ffSAl Viro audit_inode(pathname, nd->path.dentry); 2151fe2d35ffSAl Viro goto ok; 2152fe2d35ffSAl Viro } 2153fe2d35ffSAl Viro 2154fe2d35ffSAl Viro /* create side of things */ 2155a3fbbde7SAl Viro /* 2156a3fbbde7SAl Viro * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED has been 2157a3fbbde7SAl Viro * cleared when we got to the last component we are about to look up 2158a3fbbde7SAl Viro */ 21599f1fafeeSAl Viro error = complete_walk(nd); 21609f1fafeeSAl Viro if (error) 21619f1fafeeSAl Viro return ERR_PTR(error); 2162fe2d35ffSAl Viro 2163fe2d35ffSAl Viro audit_inode(pathname, dir); 216416c2cd71SAl Viro error = -EISDIR; 21651f36f774SAl Viro /* trailing slashes? */ 216631e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 21671f36f774SAl Viro goto exit; 21681f36f774SAl Viro 2169a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 2170a1e28038SAl Viro 21716c0d46c4SAl Viro dentry = lookup_hash(nd); 21726c0d46c4SAl Viro error = PTR_ERR(dentry); 21736c0d46c4SAl Viro if (IS_ERR(dentry)) { 2174fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2175fb1cc555SAl Viro goto exit; 2176fb1cc555SAl Viro } 2177fb1cc555SAl Viro 21786c0d46c4SAl Viro path->dentry = dentry; 21796c0d46c4SAl Viro path->mnt = nd->path.mnt; 21806c0d46c4SAl Viro 2181fb1cc555SAl Viro /* Negative dentry, just create the file */ 21826c0d46c4SAl Viro if (!dentry->d_inode) { 21836c0d46c4SAl Viro int mode = op->mode; 21846c0d46c4SAl Viro if (!IS_POSIXACL(dir->d_inode)) 21856c0d46c4SAl Viro mode &= ~current_umask(); 2186fb1cc555SAl Viro /* 2187fb1cc555SAl Viro * This write is needed to ensure that a 21886c0d46c4SAl Viro * rw->ro transition does not occur between 2189fb1cc555SAl Viro * the time when the file is created and when 2190fb1cc555SAl Viro * a permanent write count is taken through 2191fb1cc555SAl Viro * the 'struct file' in nameidata_to_filp(). 2192fb1cc555SAl Viro */ 2193fb1cc555SAl Viro error = mnt_want_write(nd->path.mnt); 2194fb1cc555SAl Viro if (error) 2195fb1cc555SAl Viro goto exit_mutex_unlock; 2196ca344a89SAl Viro want_write = 1; 21979b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 2198ca344a89SAl Viro open_flag &= ~O_TRUNC; 21996c0d46c4SAl Viro will_truncate = 0; 2200bcda7652SAl Viro acc_mode = MAY_OPEN; 22016c0d46c4SAl Viro error = security_path_mknod(&nd->path, dentry, mode, 0); 22026c0d46c4SAl Viro if (error) 22036c0d46c4SAl Viro goto exit_mutex_unlock; 22046c0d46c4SAl Viro error = vfs_create(dir->d_inode, dentry, mode, nd); 22056c0d46c4SAl Viro if (error) 22066c0d46c4SAl Viro goto exit_mutex_unlock; 22076c0d46c4SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 22086c0d46c4SAl Viro dput(nd->path.dentry); 22096c0d46c4SAl Viro nd->path.dentry = dentry; 2210ca344a89SAl Viro goto common; 2211fb1cc555SAl Viro } 2212fb1cc555SAl Viro 2213fb1cc555SAl Viro /* 2214fb1cc555SAl Viro * It already exists. 2215fb1cc555SAl Viro */ 2216fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2217fb1cc555SAl Viro audit_inode(pathname, path->dentry); 2218fb1cc555SAl Viro 2219fb1cc555SAl Viro error = -EEXIST; 2220ca344a89SAl Viro if (open_flag & O_EXCL) 2221fb1cc555SAl Viro goto exit_dput; 2222fb1cc555SAl Viro 22239875cf80SDavid Howells error = follow_managed(path, nd->flags); 22249875cf80SDavid Howells if (error < 0) 2225fb1cc555SAl Viro goto exit_dput; 2226fb1cc555SAl Viro 2227a3fbbde7SAl Viro if (error) 2228a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 2229a3fbbde7SAl Viro 2230fb1cc555SAl Viro error = -ENOENT; 2231fb1cc555SAl Viro if (!path->dentry->d_inode) 2232fb1cc555SAl Viro goto exit_dput; 22339e67f361SAl Viro 22349e67f361SAl Viro if (path->dentry->d_inode->i_op->follow_link) 2235fb1cc555SAl Viro return NULL; 2236fb1cc555SAl Viro 2237fb1cc555SAl Viro path_to_nameidata(path, nd); 223831e6b01fSNick Piggin nd->inode = path->dentry->d_inode; 2239a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 2240a3fbbde7SAl Viro error = complete_walk(nd); 2241a3fbbde7SAl Viro if (error) 2242a3fbbde7SAl Viro goto exit; 2243fb1cc555SAl Viro error = -EISDIR; 224431e6b01fSNick Piggin if (S_ISDIR(nd->inode->i_mode)) 2245fb1cc555SAl Viro goto exit; 224667ee3ad2SAl Viro ok: 22476c0d46c4SAl Viro if (!S_ISREG(nd->inode->i_mode)) 22486c0d46c4SAl Viro will_truncate = 0; 22496c0d46c4SAl Viro 22500f9d1a10SAl Viro if (will_truncate) { 22510f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 22520f9d1a10SAl Viro if (error) 22530f9d1a10SAl Viro goto exit; 2254ca344a89SAl Viro want_write = 1; 22550f9d1a10SAl Viro } 2256ca344a89SAl Viro common: 2257bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 2258ca344a89SAl Viro if (error) 22590f9d1a10SAl Viro goto exit; 22600f9d1a10SAl Viro filp = nameidata_to_filp(nd); 22610f9d1a10SAl Viro if (!IS_ERR(filp)) { 22620f9d1a10SAl Viro error = ima_file_check(filp, op->acc_mode); 22630f9d1a10SAl Viro if (error) { 22640f9d1a10SAl Viro fput(filp); 22650f9d1a10SAl Viro filp = ERR_PTR(error); 22660f9d1a10SAl Viro } 22670f9d1a10SAl Viro } 22680f9d1a10SAl Viro if (!IS_ERR(filp)) { 22690f9d1a10SAl Viro if (will_truncate) { 22700f9d1a10SAl Viro error = handle_truncate(filp); 22710f9d1a10SAl Viro if (error) { 22720f9d1a10SAl Viro fput(filp); 22730f9d1a10SAl Viro filp = ERR_PTR(error); 22740f9d1a10SAl Viro } 22750f9d1a10SAl Viro } 22760f9d1a10SAl Viro } 2277ca344a89SAl Viro out: 2278ca344a89SAl Viro if (want_write) 22790f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 22800f9d1a10SAl Viro path_put(&nd->path); 2281fb1cc555SAl Viro return filp; 2282fb1cc555SAl Viro 2283fb1cc555SAl Viro exit_mutex_unlock: 2284fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2285fb1cc555SAl Viro exit_dput: 2286fb1cc555SAl Viro path_put_conditional(path, nd); 2287fb1cc555SAl Viro exit: 2288ca344a89SAl Viro filp = ERR_PTR(error); 2289ca344a89SAl Viro goto out; 2290fb1cc555SAl Viro } 2291fb1cc555SAl Viro 229213aab428SAl Viro static struct file *path_openat(int dfd, const char *pathname, 229373d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 22941da177e4SLinus Torvalds { 2295fe2d35ffSAl Viro struct file *base = NULL; 22964a3fd211SDave Hansen struct file *filp; 22979850c056SAl Viro struct path path; 229813aab428SAl Viro int error; 229931e6b01fSNick Piggin 230031e6b01fSNick Piggin filp = get_empty_filp(); 230131e6b01fSNick Piggin if (!filp) 230231e6b01fSNick Piggin return ERR_PTR(-ENFILE); 230331e6b01fSNick Piggin 230447c805dcSAl Viro filp->f_flags = op->open_flag; 230573d049a4SAl Viro nd->intent.open.file = filp; 230673d049a4SAl Viro nd->intent.open.flags = open_to_namei_flags(op->open_flag); 230773d049a4SAl Viro nd->intent.open.create_mode = op->mode; 230831e6b01fSNick Piggin 230973d049a4SAl Viro error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base); 231031e6b01fSNick Piggin if (unlikely(error)) 231113aab428SAl Viro goto out_filp; 231231e6b01fSNick Piggin 2313fe2d35ffSAl Viro current->total_link_count = 0; 231473d049a4SAl Viro error = link_path_walk(pathname, nd); 231531e6b01fSNick Piggin if (unlikely(error)) 231631e6b01fSNick Piggin goto out_filp; 23171da177e4SLinus Torvalds 231873d049a4SAl Viro filp = do_last(nd, &path, op, pathname); 2319806b681cSAl Viro while (unlikely(!filp)) { /* trailing symlink */ 23207b9337aaSNick Piggin struct path link = path; 2321def4af30SAl Viro void *cookie; 2322574197e0SAl Viro if (!(nd->flags & LOOKUP_FOLLOW)) { 232373d049a4SAl Viro path_put_conditional(&path, nd); 232473d049a4SAl Viro path_put(&nd->path); 232540b39136SAl Viro filp = ERR_PTR(-ELOOP); 232640b39136SAl Viro break; 232740b39136SAl Viro } 232873d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 232973d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 2330574197e0SAl Viro error = follow_link(&link, nd, &cookie); 2331c3e380b0SAl Viro if (unlikely(error)) 2332f1afe9efSAl Viro filp = ERR_PTR(error); 2333c3e380b0SAl Viro else 233473d049a4SAl Viro filp = do_last(nd, &path, op, pathname); 2335574197e0SAl Viro put_link(nd, &link, cookie); 2336806b681cSAl Viro } 233710fa8e62SAl Viro out: 233873d049a4SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) 233973d049a4SAl Viro path_put(&nd->root); 2340fe2d35ffSAl Viro if (base) 2341fe2d35ffSAl Viro fput(base); 234273d049a4SAl Viro release_open_intent(nd); 234310fa8e62SAl Viro return filp; 23441da177e4SLinus Torvalds 234531e6b01fSNick Piggin out_filp: 234610fa8e62SAl Viro filp = ERR_PTR(error); 234710fa8e62SAl Viro goto out; 2348de459215SKirill Korotaev } 23491da177e4SLinus Torvalds 235013aab428SAl Viro struct file *do_filp_open(int dfd, const char *pathname, 235113aab428SAl Viro const struct open_flags *op, int flags) 235213aab428SAl Viro { 235373d049a4SAl Viro struct nameidata nd; 235413aab428SAl Viro struct file *filp; 235513aab428SAl Viro 235673d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 235713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 235873d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 235913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 236073d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 236113aab428SAl Viro return filp; 236213aab428SAl Viro } 236313aab428SAl Viro 236473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 236573d049a4SAl Viro const char *name, const struct open_flags *op, int flags) 236673d049a4SAl Viro { 236773d049a4SAl Viro struct nameidata nd; 236873d049a4SAl Viro struct file *file; 236973d049a4SAl Viro 237073d049a4SAl Viro nd.root.mnt = mnt; 237173d049a4SAl Viro nd.root.dentry = dentry; 237273d049a4SAl Viro 237373d049a4SAl Viro flags |= LOOKUP_ROOT; 237473d049a4SAl Viro 2375bcda7652SAl Viro if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN) 237673d049a4SAl Viro return ERR_PTR(-ELOOP); 237773d049a4SAl Viro 237873d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU); 237973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 238073d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags); 238173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 238273d049a4SAl Viro file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL); 238373d049a4SAl Viro return file; 238473d049a4SAl Viro } 238573d049a4SAl Viro 2386ed75e95dSAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir) 23871da177e4SLinus Torvalds { 2388c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 2389ed75e95dSAl Viro struct nameidata nd; 2390ed75e95dSAl Viro int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd); 2391ed75e95dSAl Viro if (error) 2392ed75e95dSAl Viro return ERR_PTR(error); 23931da177e4SLinus Torvalds 2394c663e5d8SChristoph Hellwig /* 2395c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 2396c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 2397c663e5d8SChristoph Hellwig */ 2398ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 2399ed75e95dSAl Viro goto out; 2400ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 2401ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 2402ed75e95dSAl Viro nd.intent.open.flags = O_EXCL; 2403c663e5d8SChristoph Hellwig 2404c663e5d8SChristoph Hellwig /* 2405c663e5d8SChristoph Hellwig * Do the final lookup. 2406c663e5d8SChristoph Hellwig */ 2407ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 2408ed75e95dSAl Viro dentry = lookup_hash(&nd); 24091da177e4SLinus Torvalds if (IS_ERR(dentry)) 24101da177e4SLinus Torvalds goto fail; 2411c663e5d8SChristoph Hellwig 2412e9baf6e5SAl Viro if (dentry->d_inode) 2413e9baf6e5SAl Viro goto eexist; 2414c663e5d8SChristoph Hellwig /* 2415c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 2416c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 2417c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 2418c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 2419c663e5d8SChristoph Hellwig */ 2420ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 24211da177e4SLinus Torvalds dput(dentry); 24221da177e4SLinus Torvalds dentry = ERR_PTR(-ENOENT); 2423ed75e95dSAl Viro goto fail; 2424e9baf6e5SAl Viro } 2425ed75e95dSAl Viro *path = nd.path; 2426e9baf6e5SAl Viro return dentry; 2427e9baf6e5SAl Viro eexist: 2428e9baf6e5SAl Viro dput(dentry); 2429e9baf6e5SAl Viro dentry = ERR_PTR(-EEXIST); 24301da177e4SLinus Torvalds fail: 2431dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 2432ed75e95dSAl Viro out: 2433dae6ad8fSAl Viro path_put(&nd.path); 2434ed75e95dSAl Viro return dentry; 2435dae6ad8fSAl Viro } 2436dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 2437dae6ad8fSAl Viro 2438dae6ad8fSAl Viro struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir) 2439dae6ad8fSAl Viro { 2440dae6ad8fSAl Viro char *tmp = getname(pathname); 2441dae6ad8fSAl Viro struct dentry *res; 2442dae6ad8fSAl Viro if (IS_ERR(tmp)) 2443dae6ad8fSAl Viro return ERR_CAST(tmp); 2444dae6ad8fSAl Viro res = kern_path_create(dfd, tmp, path, is_dir); 2445dae6ad8fSAl Viro putname(tmp); 2446dae6ad8fSAl Viro return res; 2447dae6ad8fSAl Viro } 2448dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 2449dae6ad8fSAl Viro 24501da177e4SLinus Torvalds int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 24511da177e4SLinus Torvalds { 2452a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 24531da177e4SLinus Torvalds 24541da177e4SLinus Torvalds if (error) 24551da177e4SLinus Torvalds return error; 24561da177e4SLinus Torvalds 2457e795b717SSerge E. Hallyn if ((S_ISCHR(mode) || S_ISBLK(mode)) && 2458e795b717SSerge E. Hallyn !ns_capable(inode_userns(dir), CAP_MKNOD)) 24591da177e4SLinus Torvalds return -EPERM; 24601da177e4SLinus Torvalds 2461acfa4380SAl Viro if (!dir->i_op->mknod) 24621da177e4SLinus Torvalds return -EPERM; 24631da177e4SLinus Torvalds 246408ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 246508ce5f16SSerge E. Hallyn if (error) 246608ce5f16SSerge E. Hallyn return error; 246708ce5f16SSerge E. Hallyn 24681da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 24691da177e4SLinus Torvalds if (error) 24701da177e4SLinus Torvalds return error; 24711da177e4SLinus Torvalds 24721da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 2473a74574aaSStephen Smalley if (!error) 2474f38aa942SAmy Griffis fsnotify_create(dir, dentry); 24751da177e4SLinus Torvalds return error; 24761da177e4SLinus Torvalds } 24771da177e4SLinus Torvalds 2478463c3197SDave Hansen static int may_mknod(mode_t mode) 2479463c3197SDave Hansen { 2480463c3197SDave Hansen switch (mode & S_IFMT) { 2481463c3197SDave Hansen case S_IFREG: 2482463c3197SDave Hansen case S_IFCHR: 2483463c3197SDave Hansen case S_IFBLK: 2484463c3197SDave Hansen case S_IFIFO: 2485463c3197SDave Hansen case S_IFSOCK: 2486463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 2487463c3197SDave Hansen return 0; 2488463c3197SDave Hansen case S_IFDIR: 2489463c3197SDave Hansen return -EPERM; 2490463c3197SDave Hansen default: 2491463c3197SDave Hansen return -EINVAL; 2492463c3197SDave Hansen } 2493463c3197SDave Hansen } 2494463c3197SDave Hansen 24952e4d0924SHeiko Carstens SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode, 24962e4d0924SHeiko Carstens unsigned, dev) 24971da177e4SLinus Torvalds { 24981da177e4SLinus Torvalds struct dentry *dentry; 2499dae6ad8fSAl Viro struct path path; 2500dae6ad8fSAl Viro int error; 25011da177e4SLinus Torvalds 25021da177e4SLinus Torvalds if (S_ISDIR(mode)) 25031da177e4SLinus Torvalds return -EPERM; 25041da177e4SLinus Torvalds 2505dae6ad8fSAl Viro dentry = user_path_create(dfd, filename, &path, 0); 2506dae6ad8fSAl Viro if (IS_ERR(dentry)) 2507dae6ad8fSAl Viro return PTR_ERR(dentry); 25082ad94ae6SAl Viro 2509dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 2510ce3b0f8dSAl Viro mode &= ~current_umask(); 2511463c3197SDave Hansen error = may_mknod(mode); 2512463c3197SDave Hansen if (error) 2513463c3197SDave Hansen goto out_dput; 2514dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 2515463c3197SDave Hansen if (error) 2516463c3197SDave Hansen goto out_dput; 2517dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 2518be6d3e56SKentaro Takeda if (error) 2519be6d3e56SKentaro Takeda goto out_drop_write; 25201da177e4SLinus Torvalds switch (mode & S_IFMT) { 25211da177e4SLinus Torvalds case 0: case S_IFREG: 2522dae6ad8fSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,NULL); 25231da177e4SLinus Torvalds break; 25241da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 2525dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 25261da177e4SLinus Torvalds new_decode_dev(dev)); 25271da177e4SLinus Torvalds break; 25281da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 2529dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 25301da177e4SLinus Torvalds break; 25311da177e4SLinus Torvalds } 2532be6d3e56SKentaro Takeda out_drop_write: 2533dae6ad8fSAl Viro mnt_drop_write(path.mnt); 2534463c3197SDave Hansen out_dput: 25351da177e4SLinus Torvalds dput(dentry); 2536dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 2537dae6ad8fSAl Viro path_put(&path); 25381da177e4SLinus Torvalds 25391da177e4SLinus Torvalds return error; 25401da177e4SLinus Torvalds } 25411da177e4SLinus Torvalds 25423480b257SHeiko Carstens SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev) 25435590ff0dSUlrich Drepper { 25445590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 25455590ff0dSUlrich Drepper } 25465590ff0dSUlrich Drepper 25471da177e4SLinus Torvalds int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) 25481da177e4SLinus Torvalds { 2549a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25501da177e4SLinus Torvalds 25511da177e4SLinus Torvalds if (error) 25521da177e4SLinus Torvalds return error; 25531da177e4SLinus Torvalds 2554acfa4380SAl Viro if (!dir->i_op->mkdir) 25551da177e4SLinus Torvalds return -EPERM; 25561da177e4SLinus Torvalds 25571da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 25581da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 25591da177e4SLinus Torvalds if (error) 25601da177e4SLinus Torvalds return error; 25611da177e4SLinus Torvalds 25621da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 2563a74574aaSStephen Smalley if (!error) 2564f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 25651da177e4SLinus Torvalds return error; 25661da177e4SLinus Torvalds } 25671da177e4SLinus Torvalds 25682e4d0924SHeiko Carstens SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode) 25691da177e4SLinus Torvalds { 25706902d925SDave Hansen struct dentry *dentry; 2571dae6ad8fSAl Viro struct path path; 2572dae6ad8fSAl Viro int error; 25731da177e4SLinus Torvalds 2574dae6ad8fSAl Viro dentry = user_path_create(dfd, pathname, &path, 1); 25756902d925SDave Hansen if (IS_ERR(dentry)) 2576dae6ad8fSAl Viro return PTR_ERR(dentry); 25776902d925SDave Hansen 2578dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 2579ce3b0f8dSAl Viro mode &= ~current_umask(); 2580dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 2581463c3197SDave Hansen if (error) 2582463c3197SDave Hansen goto out_dput; 2583dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 2584be6d3e56SKentaro Takeda if (error) 2585be6d3e56SKentaro Takeda goto out_drop_write; 2586dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 2587be6d3e56SKentaro Takeda out_drop_write: 2588dae6ad8fSAl Viro mnt_drop_write(path.mnt); 2589463c3197SDave Hansen out_dput: 25901da177e4SLinus Torvalds dput(dentry); 2591dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 2592dae6ad8fSAl Viro path_put(&path); 25931da177e4SLinus Torvalds return error; 25941da177e4SLinus Torvalds } 25951da177e4SLinus Torvalds 25963cdad428SHeiko Carstens SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode) 25975590ff0dSUlrich Drepper { 25985590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 25995590ff0dSUlrich Drepper } 26005590ff0dSUlrich Drepper 26011da177e4SLinus Torvalds /* 2602a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 2603a71905f0SSage Weil * should have a usage count of 2 if we're the only user of this 2604a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 2605a71905f0SSage Weil * then we drop the dentry now. 26061da177e4SLinus Torvalds * 26071da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 26081da177e4SLinus Torvalds * do a 26091da177e4SLinus Torvalds * 26101da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 26111da177e4SLinus Torvalds * return -EBUSY; 26121da177e4SLinus Torvalds * 26131da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 26141da177e4SLinus Torvalds * that is still in use by something else.. 26151da177e4SLinus Torvalds */ 26161da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 26171da177e4SLinus Torvalds { 26181da177e4SLinus Torvalds shrink_dcache_parent(dentry); 26191da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 262064252c75SSage Weil if (dentry->d_count == 1) 26211da177e4SLinus Torvalds __d_drop(dentry); 26221da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 26231da177e4SLinus Torvalds } 26241da177e4SLinus Torvalds 26251da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 26261da177e4SLinus Torvalds { 26271da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 26281da177e4SLinus Torvalds 26291da177e4SLinus Torvalds if (error) 26301da177e4SLinus Torvalds return error; 26311da177e4SLinus Torvalds 2632acfa4380SAl Viro if (!dir->i_op->rmdir) 26331da177e4SLinus Torvalds return -EPERM; 26341da177e4SLinus Torvalds 26351d2ef590SAl Viro dget(dentry); 26361b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 2637912dbc15SSage Weil 26381da177e4SLinus Torvalds error = -EBUSY; 2639912dbc15SSage Weil if (d_mountpoint(dentry)) 2640912dbc15SSage Weil goto out; 2641912dbc15SSage Weil 26421da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 2643912dbc15SSage Weil if (error) 2644912dbc15SSage Weil goto out; 2645912dbc15SSage Weil 26463cebde24SSage Weil shrink_dcache_parent(dentry); 26471da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 2648912dbc15SSage Weil if (error) 2649912dbc15SSage Weil goto out; 2650912dbc15SSage Weil 26511da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 2652d83c49f3SAl Viro dont_mount(dentry); 26531da177e4SLinus Torvalds 2654912dbc15SSage Weil out: 2655912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 26561d2ef590SAl Viro dput(dentry); 2657912dbc15SSage Weil if (!error) 2658912dbc15SSage Weil d_delete(dentry); 26591da177e4SLinus Torvalds return error; 26601da177e4SLinus Torvalds } 26611da177e4SLinus Torvalds 26625590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 26631da177e4SLinus Torvalds { 26641da177e4SLinus Torvalds int error = 0; 26651da177e4SLinus Torvalds char * name; 26661da177e4SLinus Torvalds struct dentry *dentry; 26671da177e4SLinus Torvalds struct nameidata nd; 26681da177e4SLinus Torvalds 26692ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 26701da177e4SLinus Torvalds if (error) 26712ad94ae6SAl Viro return error; 26721da177e4SLinus Torvalds 26731da177e4SLinus Torvalds switch(nd.last_type) { 26741da177e4SLinus Torvalds case LAST_DOTDOT: 26751da177e4SLinus Torvalds error = -ENOTEMPTY; 26761da177e4SLinus Torvalds goto exit1; 26771da177e4SLinus Torvalds case LAST_DOT: 26781da177e4SLinus Torvalds error = -EINVAL; 26791da177e4SLinus Torvalds goto exit1; 26801da177e4SLinus Torvalds case LAST_ROOT: 26811da177e4SLinus Torvalds error = -EBUSY; 26821da177e4SLinus Torvalds goto exit1; 26831da177e4SLinus Torvalds } 26840612d9fbSOGAWA Hirofumi 26850612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 26860612d9fbSOGAWA Hirofumi 26874ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 268849705b77SChristoph Hellwig dentry = lookup_hash(&nd); 26891da177e4SLinus Torvalds error = PTR_ERR(dentry); 26906902d925SDave Hansen if (IS_ERR(dentry)) 26916902d925SDave Hansen goto exit2; 2692e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 2693e6bc45d6STheodore Ts'o error = -ENOENT; 2694e6bc45d6STheodore Ts'o goto exit3; 2695e6bc45d6STheodore Ts'o } 26960622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 26970622753bSDave Hansen if (error) 26980622753bSDave Hansen goto exit3; 2699be6d3e56SKentaro Takeda error = security_path_rmdir(&nd.path, dentry); 2700be6d3e56SKentaro Takeda if (error) 2701be6d3e56SKentaro Takeda goto exit4; 27024ac91378SJan Blunck error = vfs_rmdir(nd.path.dentry->d_inode, dentry); 2703be6d3e56SKentaro Takeda exit4: 27040622753bSDave Hansen mnt_drop_write(nd.path.mnt); 27050622753bSDave Hansen exit3: 27061da177e4SLinus Torvalds dput(dentry); 27076902d925SDave Hansen exit2: 27084ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 27091da177e4SLinus Torvalds exit1: 27101d957f9bSJan Blunck path_put(&nd.path); 27111da177e4SLinus Torvalds putname(name); 27121da177e4SLinus Torvalds return error; 27131da177e4SLinus Torvalds } 27141da177e4SLinus Torvalds 27153cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 27165590ff0dSUlrich Drepper { 27175590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 27185590ff0dSUlrich Drepper } 27195590ff0dSUlrich Drepper 27201da177e4SLinus Torvalds int vfs_unlink(struct inode *dir, struct dentry *dentry) 27211da177e4SLinus Torvalds { 27221da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 27231da177e4SLinus Torvalds 27241da177e4SLinus Torvalds if (error) 27251da177e4SLinus Torvalds return error; 27261da177e4SLinus Torvalds 2727acfa4380SAl Viro if (!dir->i_op->unlink) 27281da177e4SLinus Torvalds return -EPERM; 27291da177e4SLinus Torvalds 27301b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 27311da177e4SLinus Torvalds if (d_mountpoint(dentry)) 27321da177e4SLinus Torvalds error = -EBUSY; 27331da177e4SLinus Torvalds else { 27341da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 2735bec1052eSAl Viro if (!error) { 27361da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 2737bec1052eSAl Viro if (!error) 2738d83c49f3SAl Viro dont_mount(dentry); 2739bec1052eSAl Viro } 27401da177e4SLinus Torvalds } 27411b1dcc1bSJes Sorensen mutex_unlock(&dentry->d_inode->i_mutex); 27421da177e4SLinus Torvalds 27431da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 27441da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 2745ece95912SJan Kara fsnotify_link_count(dentry->d_inode); 27461da177e4SLinus Torvalds d_delete(dentry); 27471da177e4SLinus Torvalds } 27480eeca283SRobert Love 27491da177e4SLinus Torvalds return error; 27501da177e4SLinus Torvalds } 27511da177e4SLinus Torvalds 27521da177e4SLinus Torvalds /* 27531da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 27541b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 27551da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 27561da177e4SLinus Torvalds * while waiting on the I/O. 27571da177e4SLinus Torvalds */ 27585590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 27591da177e4SLinus Torvalds { 27602ad94ae6SAl Viro int error; 27611da177e4SLinus Torvalds char *name; 27621da177e4SLinus Torvalds struct dentry *dentry; 27631da177e4SLinus Torvalds struct nameidata nd; 27641da177e4SLinus Torvalds struct inode *inode = NULL; 27651da177e4SLinus Torvalds 27662ad94ae6SAl Viro error = user_path_parent(dfd, pathname, &nd, &name); 27671da177e4SLinus Torvalds if (error) 27682ad94ae6SAl Viro return error; 27692ad94ae6SAl Viro 27701da177e4SLinus Torvalds error = -EISDIR; 27711da177e4SLinus Torvalds if (nd.last_type != LAST_NORM) 27721da177e4SLinus Torvalds goto exit1; 27730612d9fbSOGAWA Hirofumi 27740612d9fbSOGAWA Hirofumi nd.flags &= ~LOOKUP_PARENT; 27750612d9fbSOGAWA Hirofumi 27764ac91378SJan Blunck mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 277749705b77SChristoph Hellwig dentry = lookup_hash(&nd); 27781da177e4SLinus Torvalds error = PTR_ERR(dentry); 27791da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 27801da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 278150338b88STörök Edwin if (nd.last.name[nd.last.len]) 278250338b88STörök Edwin goto slashes; 27831da177e4SLinus Torvalds inode = dentry->d_inode; 278450338b88STörök Edwin if (!inode) 2785e6bc45d6STheodore Ts'o goto slashes; 27867de9c6eeSAl Viro ihold(inode); 27870622753bSDave Hansen error = mnt_want_write(nd.path.mnt); 27880622753bSDave Hansen if (error) 27890622753bSDave Hansen goto exit2; 2790be6d3e56SKentaro Takeda error = security_path_unlink(&nd.path, dentry); 2791be6d3e56SKentaro Takeda if (error) 2792be6d3e56SKentaro Takeda goto exit3; 27934ac91378SJan Blunck error = vfs_unlink(nd.path.dentry->d_inode, dentry); 2794be6d3e56SKentaro Takeda exit3: 27950622753bSDave Hansen mnt_drop_write(nd.path.mnt); 27961da177e4SLinus Torvalds exit2: 27971da177e4SLinus Torvalds dput(dentry); 27981da177e4SLinus Torvalds } 27994ac91378SJan Blunck mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 28001da177e4SLinus Torvalds if (inode) 28011da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 28021da177e4SLinus Torvalds exit1: 28031d957f9bSJan Blunck path_put(&nd.path); 28041da177e4SLinus Torvalds putname(name); 28051da177e4SLinus Torvalds return error; 28061da177e4SLinus Torvalds 28071da177e4SLinus Torvalds slashes: 28081da177e4SLinus Torvalds error = !dentry->d_inode ? -ENOENT : 28091da177e4SLinus Torvalds S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR; 28101da177e4SLinus Torvalds goto exit2; 28111da177e4SLinus Torvalds } 28121da177e4SLinus Torvalds 28132e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 28145590ff0dSUlrich Drepper { 28155590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 28165590ff0dSUlrich Drepper return -EINVAL; 28175590ff0dSUlrich Drepper 28185590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 28195590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 28205590ff0dSUlrich Drepper 28215590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 28225590ff0dSUlrich Drepper } 28235590ff0dSUlrich Drepper 28243480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 28255590ff0dSUlrich Drepper { 28265590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 28275590ff0dSUlrich Drepper } 28285590ff0dSUlrich Drepper 2829db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 28301da177e4SLinus Torvalds { 2831a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28321da177e4SLinus Torvalds 28331da177e4SLinus Torvalds if (error) 28341da177e4SLinus Torvalds return error; 28351da177e4SLinus Torvalds 2836acfa4380SAl Viro if (!dir->i_op->symlink) 28371da177e4SLinus Torvalds return -EPERM; 28381da177e4SLinus Torvalds 28391da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 28401da177e4SLinus Torvalds if (error) 28411da177e4SLinus Torvalds return error; 28421da177e4SLinus Torvalds 28431da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 2844a74574aaSStephen Smalley if (!error) 2845f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28461da177e4SLinus Torvalds return error; 28471da177e4SLinus Torvalds } 28481da177e4SLinus Torvalds 28492e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 28502e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 28511da177e4SLinus Torvalds { 28522ad94ae6SAl Viro int error; 28531da177e4SLinus Torvalds char *from; 28546902d925SDave Hansen struct dentry *dentry; 2855dae6ad8fSAl Viro struct path path; 28561da177e4SLinus Torvalds 28571da177e4SLinus Torvalds from = getname(oldname); 28581da177e4SLinus Torvalds if (IS_ERR(from)) 28591da177e4SLinus Torvalds return PTR_ERR(from); 28602ad94ae6SAl Viro 2861dae6ad8fSAl Viro dentry = user_path_create(newdfd, newname, &path, 0); 28621da177e4SLinus Torvalds error = PTR_ERR(dentry); 28636902d925SDave Hansen if (IS_ERR(dentry)) 2864dae6ad8fSAl Viro goto out_putname; 28656902d925SDave Hansen 2866dae6ad8fSAl Viro error = mnt_want_write(path.mnt); 286775c3f29dSDave Hansen if (error) 286875c3f29dSDave Hansen goto out_dput; 2869dae6ad8fSAl Viro error = security_path_symlink(&path, dentry, from); 2870be6d3e56SKentaro Takeda if (error) 2871be6d3e56SKentaro Takeda goto out_drop_write; 2872dae6ad8fSAl Viro error = vfs_symlink(path.dentry->d_inode, dentry, from); 2873be6d3e56SKentaro Takeda out_drop_write: 2874dae6ad8fSAl Viro mnt_drop_write(path.mnt); 287575c3f29dSDave Hansen out_dput: 28761da177e4SLinus Torvalds dput(dentry); 2877dae6ad8fSAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 2878dae6ad8fSAl Viro path_put(&path); 28796902d925SDave Hansen out_putname: 28801da177e4SLinus Torvalds putname(from); 28811da177e4SLinus Torvalds return error; 28821da177e4SLinus Torvalds } 28831da177e4SLinus Torvalds 28843480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 28855590ff0dSUlrich Drepper { 28865590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 28875590ff0dSUlrich Drepper } 28885590ff0dSUlrich Drepper 28891da177e4SLinus Torvalds int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry) 28901da177e4SLinus Torvalds { 28911da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 28921da177e4SLinus Torvalds int error; 28931da177e4SLinus Torvalds 28941da177e4SLinus Torvalds if (!inode) 28951da177e4SLinus Torvalds return -ENOENT; 28961da177e4SLinus Torvalds 2897a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 28981da177e4SLinus Torvalds if (error) 28991da177e4SLinus Torvalds return error; 29001da177e4SLinus Torvalds 29011da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 29021da177e4SLinus Torvalds return -EXDEV; 29031da177e4SLinus Torvalds 29041da177e4SLinus Torvalds /* 29051da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 29061da177e4SLinus Torvalds */ 29071da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 29081da177e4SLinus Torvalds return -EPERM; 2909acfa4380SAl Viro if (!dir->i_op->link) 29101da177e4SLinus Torvalds return -EPERM; 29117e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 29121da177e4SLinus Torvalds return -EPERM; 29131da177e4SLinus Torvalds 29141da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 29151da177e4SLinus Torvalds if (error) 29161da177e4SLinus Torvalds return error; 29171da177e4SLinus Torvalds 29187e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 2919aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 2920aae8a97dSAneesh Kumar K.V if (inode->i_nlink == 0) 2921aae8a97dSAneesh Kumar K.V error = -ENOENT; 2922aae8a97dSAneesh Kumar K.V else 29231da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 29247e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 2925e31e14ecSStephen Smalley if (!error) 29267e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 29271da177e4SLinus Torvalds return error; 29281da177e4SLinus Torvalds } 29291da177e4SLinus Torvalds 29301da177e4SLinus Torvalds /* 29311da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 29321da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 29331da177e4SLinus Torvalds * newname. --KAB 29341da177e4SLinus Torvalds * 29351da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 29361da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 29371da177e4SLinus Torvalds * and other special files. --ADM 29381da177e4SLinus Torvalds */ 29392e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 29402e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 29411da177e4SLinus Torvalds { 29421da177e4SLinus Torvalds struct dentry *new_dentry; 2943dae6ad8fSAl Viro struct path old_path, new_path; 294411a7b371SAneesh Kumar K.V int how = 0; 29451da177e4SLinus Torvalds int error; 29461da177e4SLinus Torvalds 294711a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 2948c04030e1SUlrich Drepper return -EINVAL; 294911a7b371SAneesh Kumar K.V /* 295011a7b371SAneesh Kumar K.V * To use null names we require CAP_DAC_READ_SEARCH 295111a7b371SAneesh Kumar K.V * This ensures that not everyone will be able to create 295211a7b371SAneesh Kumar K.V * handlink using the passed filedescriptor. 295311a7b371SAneesh Kumar K.V */ 295411a7b371SAneesh Kumar K.V if (flags & AT_EMPTY_PATH) { 295511a7b371SAneesh Kumar K.V if (!capable(CAP_DAC_READ_SEARCH)) 295611a7b371SAneesh Kumar K.V return -ENOENT; 295711a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 295811a7b371SAneesh Kumar K.V } 2959c04030e1SUlrich Drepper 296011a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 296111a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 296211a7b371SAneesh Kumar K.V 296311a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 29641da177e4SLinus Torvalds if (error) 29652ad94ae6SAl Viro return error; 29662ad94ae6SAl Viro 2967dae6ad8fSAl Viro new_dentry = user_path_create(newdfd, newname, &new_path, 0); 29681da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 29696902d925SDave Hansen if (IS_ERR(new_dentry)) 2970dae6ad8fSAl Viro goto out; 2971dae6ad8fSAl Viro 2972dae6ad8fSAl Viro error = -EXDEV; 2973dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 2974dae6ad8fSAl Viro goto out_dput; 2975dae6ad8fSAl Viro error = mnt_want_write(new_path.mnt); 297675c3f29dSDave Hansen if (error) 297775c3f29dSDave Hansen goto out_dput; 2978dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 2979be6d3e56SKentaro Takeda if (error) 2980be6d3e56SKentaro Takeda goto out_drop_write; 2981dae6ad8fSAl Viro error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry); 2982be6d3e56SKentaro Takeda out_drop_write: 2983dae6ad8fSAl Viro mnt_drop_write(new_path.mnt); 298475c3f29dSDave Hansen out_dput: 29851da177e4SLinus Torvalds dput(new_dentry); 2986dae6ad8fSAl Viro mutex_unlock(&new_path.dentry->d_inode->i_mutex); 2987dae6ad8fSAl Viro path_put(&new_path); 29881da177e4SLinus Torvalds out: 29892d8f3038SAl Viro path_put(&old_path); 29901da177e4SLinus Torvalds 29911da177e4SLinus Torvalds return error; 29921da177e4SLinus Torvalds } 29931da177e4SLinus Torvalds 29943480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 29955590ff0dSUlrich Drepper { 2996c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 29975590ff0dSUlrich Drepper } 29985590ff0dSUlrich Drepper 29991da177e4SLinus Torvalds /* 30001da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 30011da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 30021da177e4SLinus Torvalds * Problems: 30031da177e4SLinus Torvalds * a) we can get into loop creation. Check is done in is_subdir(). 30041da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 30051da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 3006a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 30071da177e4SLinus Torvalds * story. 30081da177e4SLinus Torvalds * c) we have to lock _three_ objects - parents and victim (if it exists). 30091b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 30101da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 30111da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 3012a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 30131da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 30141da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 30151da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 3016a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 30171da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 30181da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 30191da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 3020e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 30211b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 30221da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 3023c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 30241da177e4SLinus Torvalds * locking]. 30251da177e4SLinus Torvalds */ 302675c96f85SAdrian Bunk static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry, 30271da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30281da177e4SLinus Torvalds { 30291da177e4SLinus Torvalds int error = 0; 30309055cba7SSage Weil struct inode *target = new_dentry->d_inode; 30311da177e4SLinus Torvalds 30321da177e4SLinus Torvalds /* 30331da177e4SLinus Torvalds * If we are going to change the parent - check write permissions, 30341da177e4SLinus Torvalds * we'll need to flip '..'. 30351da177e4SLinus Torvalds */ 30361da177e4SLinus Torvalds if (new_dir != old_dir) { 3037f419a2e3SAl Viro error = inode_permission(old_dentry->d_inode, MAY_WRITE); 30381da177e4SLinus Torvalds if (error) 30391da177e4SLinus Torvalds return error; 30401da177e4SLinus Torvalds } 30411da177e4SLinus Torvalds 30421da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30431da177e4SLinus Torvalds if (error) 30441da177e4SLinus Torvalds return error; 30451da177e4SLinus Torvalds 30461d2ef590SAl Viro dget(new_dentry); 3047d83c49f3SAl Viro if (target) 30481b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 30499055cba7SSage Weil 30501da177e4SLinus Torvalds error = -EBUSY; 30519055cba7SSage Weil if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry)) 30529055cba7SSage Weil goto out; 30539055cba7SSage Weil 30543cebde24SSage Weil if (target) 30553cebde24SSage Weil shrink_dcache_parent(new_dentry); 30561da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 30579055cba7SSage Weil if (error) 30589055cba7SSage Weil goto out; 30599055cba7SSage Weil 30601da177e4SLinus Torvalds if (target) { 30611da177e4SLinus Torvalds target->i_flags |= S_DEAD; 3062d83c49f3SAl Viro dont_mount(new_dentry); 3063d83c49f3SAl Viro } 30649055cba7SSage Weil out: 30659055cba7SSage Weil if (target) 30661b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 30671d2ef590SAl Viro dput(new_dentry); 3068e31e14ecSStephen Smalley if (!error) 3069349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 30701da177e4SLinus Torvalds d_move(old_dentry,new_dentry); 30711da177e4SLinus Torvalds return error; 30721da177e4SLinus Torvalds } 30731da177e4SLinus Torvalds 307475c96f85SAdrian Bunk static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry, 30751da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 30761da177e4SLinus Torvalds { 307751892bbbSSage Weil struct inode *target = new_dentry->d_inode; 30781da177e4SLinus Torvalds int error; 30791da177e4SLinus Torvalds 30801da177e4SLinus Torvalds error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry); 30811da177e4SLinus Torvalds if (error) 30821da177e4SLinus Torvalds return error; 30831da177e4SLinus Torvalds 30841da177e4SLinus Torvalds dget(new_dentry); 30851da177e4SLinus Torvalds if (target) 30861b1dcc1bSJes Sorensen mutex_lock(&target->i_mutex); 308751892bbbSSage Weil 30881da177e4SLinus Torvalds error = -EBUSY; 308951892bbbSSage Weil if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry)) 309051892bbbSSage Weil goto out; 309151892bbbSSage Weil 30921da177e4SLinus Torvalds error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry); 309351892bbbSSage Weil if (error) 309451892bbbSSage Weil goto out; 309551892bbbSSage Weil 3096bec1052eSAl Viro if (target) 3097d83c49f3SAl Viro dont_mount(new_dentry); 3098349457ccSMark Fasheh if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) 30991da177e4SLinus Torvalds d_move(old_dentry, new_dentry); 310051892bbbSSage Weil out: 31011da177e4SLinus Torvalds if (target) 31021b1dcc1bSJes Sorensen mutex_unlock(&target->i_mutex); 31031da177e4SLinus Torvalds dput(new_dentry); 31041da177e4SLinus Torvalds return error; 31051da177e4SLinus Torvalds } 31061da177e4SLinus Torvalds 31071da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 31081da177e4SLinus Torvalds struct inode *new_dir, struct dentry *new_dentry) 31091da177e4SLinus Torvalds { 31101da177e4SLinus Torvalds int error; 31111da177e4SLinus Torvalds int is_dir = S_ISDIR(old_dentry->d_inode->i_mode); 311259b0df21SEric Paris const unsigned char *old_name; 31131da177e4SLinus Torvalds 31141da177e4SLinus Torvalds if (old_dentry->d_inode == new_dentry->d_inode) 31151da177e4SLinus Torvalds return 0; 31161da177e4SLinus Torvalds 31171da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 31181da177e4SLinus Torvalds if (error) 31191da177e4SLinus Torvalds return error; 31201da177e4SLinus Torvalds 31211da177e4SLinus Torvalds if (!new_dentry->d_inode) 3122a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 31231da177e4SLinus Torvalds else 31241da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 31251da177e4SLinus Torvalds if (error) 31261da177e4SLinus Torvalds return error; 31271da177e4SLinus Torvalds 3128acfa4380SAl Viro if (!old_dir->i_op->rename) 31291da177e4SLinus Torvalds return -EPERM; 31301da177e4SLinus Torvalds 31310eeca283SRobert Love old_name = fsnotify_oldname_init(old_dentry->d_name.name); 31320eeca283SRobert Love 31331da177e4SLinus Torvalds if (is_dir) 31341da177e4SLinus Torvalds error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry); 31351da177e4SLinus Torvalds else 31361da177e4SLinus Torvalds error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry); 3137123df294SAl Viro if (!error) 3138123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 31395a190ae6SAl Viro new_dentry->d_inode, old_dentry); 31400eeca283SRobert Love fsnotify_oldname_free(old_name); 31410eeca283SRobert Love 31421da177e4SLinus Torvalds return error; 31431da177e4SLinus Torvalds } 31441da177e4SLinus Torvalds 31452e4d0924SHeiko Carstens SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 31462e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 31471da177e4SLinus Torvalds { 31481da177e4SLinus Torvalds struct dentry *old_dir, *new_dir; 31491da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 31501da177e4SLinus Torvalds struct dentry *trap; 31511da177e4SLinus Torvalds struct nameidata oldnd, newnd; 31522ad94ae6SAl Viro char *from; 31532ad94ae6SAl Viro char *to; 31542ad94ae6SAl Viro int error; 31551da177e4SLinus Torvalds 31562ad94ae6SAl Viro error = user_path_parent(olddfd, oldname, &oldnd, &from); 31571da177e4SLinus Torvalds if (error) 31581da177e4SLinus Torvalds goto exit; 31591da177e4SLinus Torvalds 31602ad94ae6SAl Viro error = user_path_parent(newdfd, newname, &newnd, &to); 31611da177e4SLinus Torvalds if (error) 31621da177e4SLinus Torvalds goto exit1; 31631da177e4SLinus Torvalds 31641da177e4SLinus Torvalds error = -EXDEV; 31654ac91378SJan Blunck if (oldnd.path.mnt != newnd.path.mnt) 31661da177e4SLinus Torvalds goto exit2; 31671da177e4SLinus Torvalds 31684ac91378SJan Blunck old_dir = oldnd.path.dentry; 31691da177e4SLinus Torvalds error = -EBUSY; 31701da177e4SLinus Torvalds if (oldnd.last_type != LAST_NORM) 31711da177e4SLinus Torvalds goto exit2; 31721da177e4SLinus Torvalds 31734ac91378SJan Blunck new_dir = newnd.path.dentry; 31741da177e4SLinus Torvalds if (newnd.last_type != LAST_NORM) 31751da177e4SLinus Torvalds goto exit2; 31761da177e4SLinus Torvalds 31770612d9fbSOGAWA Hirofumi oldnd.flags &= ~LOOKUP_PARENT; 31780612d9fbSOGAWA Hirofumi newnd.flags &= ~LOOKUP_PARENT; 31794e9ed2f8SOGAWA Hirofumi newnd.flags |= LOOKUP_RENAME_TARGET; 31800612d9fbSOGAWA Hirofumi 31811da177e4SLinus Torvalds trap = lock_rename(new_dir, old_dir); 31821da177e4SLinus Torvalds 318349705b77SChristoph Hellwig old_dentry = lookup_hash(&oldnd); 31841da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 31851da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 31861da177e4SLinus Torvalds goto exit3; 31871da177e4SLinus Torvalds /* source must exist */ 31881da177e4SLinus Torvalds error = -ENOENT; 31891da177e4SLinus Torvalds if (!old_dentry->d_inode) 31901da177e4SLinus Torvalds goto exit4; 31911da177e4SLinus Torvalds /* unless the source is a directory trailing slashes give -ENOTDIR */ 31921da177e4SLinus Torvalds if (!S_ISDIR(old_dentry->d_inode->i_mode)) { 31931da177e4SLinus Torvalds error = -ENOTDIR; 31941da177e4SLinus Torvalds if (oldnd.last.name[oldnd.last.len]) 31951da177e4SLinus Torvalds goto exit4; 31961da177e4SLinus Torvalds if (newnd.last.name[newnd.last.len]) 31971da177e4SLinus Torvalds goto exit4; 31981da177e4SLinus Torvalds } 31991da177e4SLinus Torvalds /* source should not be ancestor of target */ 32001da177e4SLinus Torvalds error = -EINVAL; 32011da177e4SLinus Torvalds if (old_dentry == trap) 32021da177e4SLinus Torvalds goto exit4; 320349705b77SChristoph Hellwig new_dentry = lookup_hash(&newnd); 32041da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 32051da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 32061da177e4SLinus Torvalds goto exit4; 32071da177e4SLinus Torvalds /* target should not be an ancestor of source */ 32081da177e4SLinus Torvalds error = -ENOTEMPTY; 32091da177e4SLinus Torvalds if (new_dentry == trap) 32101da177e4SLinus Torvalds goto exit5; 32111da177e4SLinus Torvalds 32129079b1ebSDave Hansen error = mnt_want_write(oldnd.path.mnt); 32139079b1ebSDave Hansen if (error) 32149079b1ebSDave Hansen goto exit5; 3215be6d3e56SKentaro Takeda error = security_path_rename(&oldnd.path, old_dentry, 3216be6d3e56SKentaro Takeda &newnd.path, new_dentry); 3217be6d3e56SKentaro Takeda if (error) 3218be6d3e56SKentaro Takeda goto exit6; 32191da177e4SLinus Torvalds error = vfs_rename(old_dir->d_inode, old_dentry, 32201da177e4SLinus Torvalds new_dir->d_inode, new_dentry); 3221be6d3e56SKentaro Takeda exit6: 32229079b1ebSDave Hansen mnt_drop_write(oldnd.path.mnt); 32231da177e4SLinus Torvalds exit5: 32241da177e4SLinus Torvalds dput(new_dentry); 32251da177e4SLinus Torvalds exit4: 32261da177e4SLinus Torvalds dput(old_dentry); 32271da177e4SLinus Torvalds exit3: 32281da177e4SLinus Torvalds unlock_rename(new_dir, old_dir); 32291da177e4SLinus Torvalds exit2: 32301d957f9bSJan Blunck path_put(&newnd.path); 32312ad94ae6SAl Viro putname(to); 32321da177e4SLinus Torvalds exit1: 32331d957f9bSJan Blunck path_put(&oldnd.path); 32341da177e4SLinus Torvalds putname(from); 32352ad94ae6SAl Viro exit: 32361da177e4SLinus Torvalds return error; 32371da177e4SLinus Torvalds } 32381da177e4SLinus Torvalds 3239a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 32405590ff0dSUlrich Drepper { 32415590ff0dSUlrich Drepper return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname); 32425590ff0dSUlrich Drepper } 32435590ff0dSUlrich Drepper 32441da177e4SLinus Torvalds int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link) 32451da177e4SLinus Torvalds { 32461da177e4SLinus Torvalds int len; 32471da177e4SLinus Torvalds 32481da177e4SLinus Torvalds len = PTR_ERR(link); 32491da177e4SLinus Torvalds if (IS_ERR(link)) 32501da177e4SLinus Torvalds goto out; 32511da177e4SLinus Torvalds 32521da177e4SLinus Torvalds len = strlen(link); 32531da177e4SLinus Torvalds if (len > (unsigned) buflen) 32541da177e4SLinus Torvalds len = buflen; 32551da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 32561da177e4SLinus Torvalds len = -EFAULT; 32571da177e4SLinus Torvalds out: 32581da177e4SLinus Torvalds return len; 32591da177e4SLinus Torvalds } 32601da177e4SLinus Torvalds 32611da177e4SLinus Torvalds /* 32621da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 32631da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 32641da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 32651da177e4SLinus Torvalds */ 32661da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 32671da177e4SLinus Torvalds { 32681da177e4SLinus Torvalds struct nameidata nd; 3269cc314eefSLinus Torvalds void *cookie; 3270694a1764SMarcin Slusarz int res; 3271cc314eefSLinus Torvalds 32721da177e4SLinus Torvalds nd.depth = 0; 3273cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 3274694a1764SMarcin Slusarz if (IS_ERR(cookie)) 3275694a1764SMarcin Slusarz return PTR_ERR(cookie); 3276694a1764SMarcin Slusarz 3277694a1764SMarcin Slusarz res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd)); 32781da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 3279cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 3280694a1764SMarcin Slusarz return res; 32811da177e4SLinus Torvalds } 32821da177e4SLinus Torvalds 32831da177e4SLinus Torvalds int vfs_follow_link(struct nameidata *nd, const char *link) 32841da177e4SLinus Torvalds { 32851da177e4SLinus Torvalds return __vfs_follow_link(nd, link); 32861da177e4SLinus Torvalds } 32871da177e4SLinus Torvalds 32881da177e4SLinus Torvalds /* get the link contents into pagecache */ 32891da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 32901da177e4SLinus Torvalds { 3291ebd09abbSDuane Griffin char *kaddr; 32921da177e4SLinus Torvalds struct page *page; 32931da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 3294090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 32951da177e4SLinus Torvalds if (IS_ERR(page)) 32966fe6900eSNick Piggin return (char*)page; 32971da177e4SLinus Torvalds *ppage = page; 3298ebd09abbSDuane Griffin kaddr = kmap(page); 3299ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 3300ebd09abbSDuane Griffin return kaddr; 33011da177e4SLinus Torvalds } 33021da177e4SLinus Torvalds 33031da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 33041da177e4SLinus Torvalds { 33051da177e4SLinus Torvalds struct page *page = NULL; 33061da177e4SLinus Torvalds char *s = page_getlink(dentry, &page); 33071da177e4SLinus Torvalds int res = vfs_readlink(dentry,buffer,buflen,s); 33081da177e4SLinus Torvalds if (page) { 33091da177e4SLinus Torvalds kunmap(page); 33101da177e4SLinus Torvalds page_cache_release(page); 33111da177e4SLinus Torvalds } 33121da177e4SLinus Torvalds return res; 33131da177e4SLinus Torvalds } 33141da177e4SLinus Torvalds 3315cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 33161da177e4SLinus Torvalds { 3317cc314eefSLinus Torvalds struct page *page = NULL; 33181da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 3319cc314eefSLinus Torvalds return page; 33201da177e4SLinus Torvalds } 33211da177e4SLinus Torvalds 3322cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 33231da177e4SLinus Torvalds { 3324cc314eefSLinus Torvalds struct page *page = cookie; 3325cc314eefSLinus Torvalds 3326cc314eefSLinus Torvalds if (page) { 33271da177e4SLinus Torvalds kunmap(page); 33281da177e4SLinus Torvalds page_cache_release(page); 33291da177e4SLinus Torvalds } 33301da177e4SLinus Torvalds } 33311da177e4SLinus Torvalds 333254566b2cSNick Piggin /* 333354566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 333454566b2cSNick Piggin */ 333554566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 33361da177e4SLinus Torvalds { 33371da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 33380adb25d2SKirill Korotaev struct page *page; 3339afddba49SNick Piggin void *fsdata; 3340beb497abSDmitriy Monakhov int err; 33411da177e4SLinus Torvalds char *kaddr; 334254566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 334354566b2cSNick Piggin if (nofs) 334454566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 33451da177e4SLinus Torvalds 33467e53cac4SNeilBrown retry: 3347afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 334854566b2cSNick Piggin flags, &page, &fsdata); 33491da177e4SLinus Torvalds if (err) 3350afddba49SNick Piggin goto fail; 3351afddba49SNick Piggin 33521da177e4SLinus Torvalds kaddr = kmap_atomic(page, KM_USER0); 33531da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 33541da177e4SLinus Torvalds kunmap_atomic(kaddr, KM_USER0); 3355afddba49SNick Piggin 3356afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 3357afddba49SNick Piggin page, fsdata); 33581da177e4SLinus Torvalds if (err < 0) 33591da177e4SLinus Torvalds goto fail; 3360afddba49SNick Piggin if (err < len-1) 3361afddba49SNick Piggin goto retry; 3362afddba49SNick Piggin 33631da177e4SLinus Torvalds mark_inode_dirty(inode); 33641da177e4SLinus Torvalds return 0; 33651da177e4SLinus Torvalds fail: 33661da177e4SLinus Torvalds return err; 33671da177e4SLinus Torvalds } 33681da177e4SLinus Torvalds 33690adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 33700adb25d2SKirill Korotaev { 33710adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 337254566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 33730adb25d2SKirill Korotaev } 33740adb25d2SKirill Korotaev 337592e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 33761da177e4SLinus Torvalds .readlink = generic_readlink, 33771da177e4SLinus Torvalds .follow_link = page_follow_link_light, 33781da177e4SLinus Torvalds .put_link = page_put_link, 33791da177e4SLinus Torvalds }; 33801da177e4SLinus Torvalds 33812d8f3038SAl Viro EXPORT_SYMBOL(user_path_at); 3382cc53ce53SDavid Howells EXPORT_SYMBOL(follow_down_one); 33831da177e4SLinus Torvalds EXPORT_SYMBOL(follow_down); 33841da177e4SLinus Torvalds EXPORT_SYMBOL(follow_up); 33851da177e4SLinus Torvalds EXPORT_SYMBOL(get_write_access); /* binfmt_aout */ 33861da177e4SLinus Torvalds EXPORT_SYMBOL(getname); 33871da177e4SLinus Torvalds EXPORT_SYMBOL(lock_rename); 33881da177e4SLinus Torvalds EXPORT_SYMBOL(lookup_one_len); 33891da177e4SLinus Torvalds EXPORT_SYMBOL(page_follow_link_light); 33901da177e4SLinus Torvalds EXPORT_SYMBOL(page_put_link); 33911da177e4SLinus Torvalds EXPORT_SYMBOL(page_readlink); 33920adb25d2SKirill Korotaev EXPORT_SYMBOL(__page_symlink); 33931da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink); 33941da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 3395d1811465SAl Viro EXPORT_SYMBOL(kern_path); 339616f18200SJosef 'Jeff' Sipek EXPORT_SYMBOL(vfs_path_lookup); 3397f419a2e3SAl Viro EXPORT_SYMBOL(inode_permission); 33981da177e4SLinus Torvalds EXPORT_SYMBOL(unlock_rename); 33991da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_create); 34001da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_follow_link); 34011da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_link); 34021da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mkdir); 34031da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_mknod); 34041da177e4SLinus Torvalds EXPORT_SYMBOL(generic_permission); 34051da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_readlink); 34061da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rename); 34071da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_rmdir); 34081da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_symlink); 34091da177e4SLinus Torvalds EXPORT_SYMBOL(vfs_unlink); 34101da177e4SLinus Torvalds EXPORT_SYMBOL(dentry_unhash); 34111da177e4SLinus Torvalds EXPORT_SYMBOL(generic_readlink); 3412