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> 18630d9c47SPaul Gortmaker #include <linux/export.h> 1944696908SDavid S. Miller #include <linux/kernel.h> 201da177e4SLinus Torvalds #include <linux/slab.h> 211da177e4SLinus Torvalds #include <linux/fs.h> 221da177e4SLinus Torvalds #include <linux/namei.h> 231da177e4SLinus Torvalds #include <linux/pagemap.h> 240eeca283SRobert Love #include <linux/fsnotify.h> 251da177e4SLinus Torvalds #include <linux/personality.h> 261da177e4SLinus Torvalds #include <linux/security.h> 276146f0d5SMimi Zohar #include <linux/ima.h> 281da177e4SLinus Torvalds #include <linux/syscalls.h> 291da177e4SLinus Torvalds #include <linux/mount.h> 301da177e4SLinus Torvalds #include <linux/audit.h> 3116f7e0feSRandy Dunlap #include <linux/capability.h> 32834f2a4aSTrond Myklebust #include <linux/file.h> 335590ff0dSUlrich Drepper #include <linux/fcntl.h> 3408ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 355ad4e53bSAl Viro #include <linux/fs_struct.h> 36e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3799d263d4SLinus Torvalds #include <linux/hash.h> 381da177e4SLinus Torvalds #include <asm/uaccess.h> 391da177e4SLinus Torvalds 40e81e3f4dSEric Paris #include "internal.h" 41c7105365SAl Viro #include "mount.h" 42e81e3f4dSEric Paris 431da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 441da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 451da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 461da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 471da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 481da177e4SLinus Torvalds * 491da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 501da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 511da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 521da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 531da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 541da177e4SLinus Torvalds * the special cases of the former code. 551da177e4SLinus Torvalds * 561da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 571da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 581da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 611da177e4SLinus Torvalds * resolution to correspond with current state of the code. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 641da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 651da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 661da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 671da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 681da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 691da177e4SLinus Torvalds */ 701da177e4SLinus Torvalds 711da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 721da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 731da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 741da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 751da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 761da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7725985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 781da177e4SLinus Torvalds * 791da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 801da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 811da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 821da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 831da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 841da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 851da177e4SLinus Torvalds * and in the old Linux semantics. 861da177e4SLinus Torvalds */ 871da177e4SLinus Torvalds 881da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 891da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 901da177e4SLinus Torvalds * 911da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 921da177e4SLinus Torvalds */ 931da177e4SLinus Torvalds 941da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 951da177e4SLinus Torvalds * inside the path - always follow. 961da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 971da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 981da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 991da177e4SLinus Torvalds * otherwise - don't follow. 1001da177e4SLinus Torvalds * (applied in that order). 1011da177e4SLinus Torvalds * 1021da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1031da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1041da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1051da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1061da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1071da177e4SLinus Torvalds */ 1081da177e4SLinus Torvalds /* 1091da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 110a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1111da177e4SLinus Torvalds * any extra contention... 1121da177e4SLinus Torvalds */ 1131da177e4SLinus Torvalds 1141da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1151da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1161da177e4SLinus Torvalds * kernel data space before using them.. 1171da177e4SLinus Torvalds * 1181da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1191da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1201da177e4SLinus Torvalds */ 1217950e385SJeff Layton 122fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12391a27b2aSJeff Layton 12451f39a1fSDavid Drysdale struct filename * 12591a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12691a27b2aSJeff Layton { 12794b5d262SAl Viro struct filename *result; 1287950e385SJeff Layton char *kname; 12994b5d262SAl Viro int len; 1301da177e4SLinus Torvalds 1317ac86265SJeff Layton result = audit_reusename(filename); 1327ac86265SJeff Layton if (result) 1337ac86265SJeff Layton return result; 1347ac86265SJeff Layton 1357950e385SJeff Layton result = __getname(); 1363f9f0aa6SLinus Torvalds if (unlikely(!result)) 1374043cde8SEric Paris return ERR_PTR(-ENOMEM); 1381da177e4SLinus Torvalds 1397950e385SJeff Layton /* 1407950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1417950e385SJeff Layton * allocation 1427950e385SJeff Layton */ 143fd2f7cb5SAl Viro kname = (char *)result->iname; 14491a27b2aSJeff Layton result->name = kname; 1457950e385SJeff Layton 14694b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 14791a27b2aSJeff Layton if (unlikely(len < 0)) { 14894b5d262SAl Viro __putname(result); 14994b5d262SAl Viro return ERR_PTR(len); 15091a27b2aSJeff Layton } 1513f9f0aa6SLinus Torvalds 1527950e385SJeff Layton /* 1537950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1547950e385SJeff Layton * separate struct filename so we can dedicate the entire 1557950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1567950e385SJeff Layton * userland. 1577950e385SJeff Layton */ 15894b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 159fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1607950e385SJeff Layton kname = (char *)result; 1617950e385SJeff Layton 162fd2f7cb5SAl Viro /* 163fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 164fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 165fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 166fd2f7cb5SAl Viro */ 167fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 16894b5d262SAl Viro if (unlikely(!result)) { 16994b5d262SAl Viro __putname(kname); 17094b5d262SAl Viro return ERR_PTR(-ENOMEM); 1717950e385SJeff Layton } 1727950e385SJeff Layton result->name = kname; 17394b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17494b5d262SAl Viro if (unlikely(len < 0)) { 17594b5d262SAl Viro __putname(kname); 17694b5d262SAl Viro kfree(result); 17794b5d262SAl Viro return ERR_PTR(len); 17894b5d262SAl Viro } 17994b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18094b5d262SAl Viro __putname(kname); 18194b5d262SAl Viro kfree(result); 18294b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18394b5d262SAl Viro } 1847950e385SJeff Layton } 1857950e385SJeff Layton 18694b5d262SAl Viro result->refcnt = 1; 1873f9f0aa6SLinus Torvalds /* The empty path is special. */ 1883f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1893f9f0aa6SLinus Torvalds if (empty) 1901fa1e7f6SAndy Whitcroft *empty = 1; 19194b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19294b5d262SAl Viro putname(result); 19394b5d262SAl Viro return ERR_PTR(-ENOENT); 1941da177e4SLinus Torvalds } 19594b5d262SAl Viro } 1967950e385SJeff Layton 1977950e385SJeff Layton result->uptr = filename; 198c4ad8f98SLinus Torvalds result->aname = NULL; 1991da177e4SLinus Torvalds audit_getname(result); 2001da177e4SLinus Torvalds return result; 2013f9f0aa6SLinus Torvalds } 2023f9f0aa6SLinus Torvalds 20391a27b2aSJeff Layton struct filename * 20491a27b2aSJeff Layton getname(const char __user * filename) 205f52e0c11SAl Viro { 206f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 207f52e0c11SAl Viro } 208f52e0c11SAl Viro 209c4ad8f98SLinus Torvalds struct filename * 210c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 211c4ad8f98SLinus Torvalds { 212c4ad8f98SLinus Torvalds struct filename *result; 21308518549SPaul Moore int len = strlen(filename) + 1; 214c4ad8f98SLinus Torvalds 215c4ad8f98SLinus Torvalds result = __getname(); 216c4ad8f98SLinus Torvalds if (unlikely(!result)) 217c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 218c4ad8f98SLinus Torvalds 21908518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 220fd2f7cb5SAl Viro result->name = (char *)result->iname; 22108518549SPaul Moore } else if (len <= PATH_MAX) { 22208518549SPaul Moore struct filename *tmp; 22308518549SPaul Moore 22408518549SPaul Moore tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 22508518549SPaul Moore if (unlikely(!tmp)) { 22608518549SPaul Moore __putname(result); 22708518549SPaul Moore return ERR_PTR(-ENOMEM); 22808518549SPaul Moore } 22908518549SPaul Moore tmp->name = (char *)result; 23008518549SPaul Moore result = tmp; 23108518549SPaul Moore } else { 23208518549SPaul Moore __putname(result); 23308518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23408518549SPaul Moore } 23508518549SPaul Moore memcpy((char *)result->name, filename, len); 236c4ad8f98SLinus Torvalds result->uptr = NULL; 237c4ad8f98SLinus Torvalds result->aname = NULL; 23855422d0bSPaul Moore result->refcnt = 1; 239fd3522fdSPaul Moore audit_getname(result); 240c4ad8f98SLinus Torvalds 241c4ad8f98SLinus Torvalds return result; 242c4ad8f98SLinus Torvalds } 243c4ad8f98SLinus Torvalds 24491a27b2aSJeff Layton void putname(struct filename *name) 2451da177e4SLinus Torvalds { 24655422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 24755422d0bSPaul Moore 24855422d0bSPaul Moore if (--name->refcnt > 0) 24955422d0bSPaul Moore return; 25055422d0bSPaul Moore 251fd2f7cb5SAl Viro if (name->name != name->iname) { 25255422d0bSPaul Moore __putname(name->name); 25355422d0bSPaul Moore kfree(name); 25455422d0bSPaul Moore } else 25555422d0bSPaul Moore __putname(name); 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 258e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 259e77819e5SLinus Torvalds { 26084635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 261e77819e5SLinus Torvalds struct posix_acl *acl; 262e77819e5SLinus Torvalds 263e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2643567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2653567866bSAl Viro if (!acl) 266e77819e5SLinus Torvalds return -EAGAIN; 2673567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 2683567866bSAl Viro if (acl == ACL_NOT_CACHED) 269e77819e5SLinus Torvalds return -ECHILD; 270206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 271e77819e5SLinus Torvalds } 272e77819e5SLinus Torvalds 2732982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2744e34e719SChristoph Hellwig if (IS_ERR(acl)) 2754e34e719SChristoph Hellwig return PTR_ERR(acl); 276e77819e5SLinus Torvalds if (acl) { 277e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 278e77819e5SLinus Torvalds posix_acl_release(acl); 279e77819e5SLinus Torvalds return error; 280e77819e5SLinus Torvalds } 28184635d68SLinus Torvalds #endif 282e77819e5SLinus Torvalds 283e77819e5SLinus Torvalds return -EAGAIN; 284e77819e5SLinus Torvalds } 285e77819e5SLinus Torvalds 2865909ccaaSLinus Torvalds /* 287948409c7SAndreas Gruenbacher * This does the basic permission checking 2885909ccaaSLinus Torvalds */ 2897e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2905909ccaaSLinus Torvalds { 29126cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2925909ccaaSLinus Torvalds 2938e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2945909ccaaSLinus Torvalds mode >>= 6; 2955909ccaaSLinus Torvalds else { 296e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2977e40145eSAl Viro int error = check_acl(inode, mask); 2985909ccaaSLinus Torvalds if (error != -EAGAIN) 2995909ccaaSLinus Torvalds return error; 3005909ccaaSLinus Torvalds } 3015909ccaaSLinus Torvalds 3025909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3035909ccaaSLinus Torvalds mode >>= 3; 3045909ccaaSLinus Torvalds } 3055909ccaaSLinus Torvalds 3065909ccaaSLinus Torvalds /* 3075909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3085909ccaaSLinus Torvalds */ 3099c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3105909ccaaSLinus Torvalds return 0; 3115909ccaaSLinus Torvalds return -EACCES; 3125909ccaaSLinus Torvalds } 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds /** 3151da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3161da177e4SLinus Torvalds * @inode: inode to check access rights for 3178fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3181da177e4SLinus Torvalds * 3191da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3201da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3211da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 322b74c79e9SNick Piggin * are used for other things. 323b74c79e9SNick Piggin * 324b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 325b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 326b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3271da177e4SLinus Torvalds */ 3282830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3291da177e4SLinus Torvalds { 3305909ccaaSLinus Torvalds int ret; 3311da177e4SLinus Torvalds 3321da177e4SLinus Torvalds /* 333948409c7SAndreas Gruenbacher * Do the basic permission checks. 3341da177e4SLinus Torvalds */ 3357e40145eSAl Viro ret = acl_permission_check(inode, mask); 3365909ccaaSLinus Torvalds if (ret != -EACCES) 3375909ccaaSLinus Torvalds return ret; 3381da177e4SLinus Torvalds 339d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 340d594e7ecSAl Viro /* DACs are overridable for directories */ 34123adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 342d594e7ecSAl Viro return 0; 343d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34423adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34523adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 346d594e7ecSAl Viro return 0; 347d594e7ecSAl Viro return -EACCES; 348d594e7ecSAl Viro } 3491da177e4SLinus Torvalds /* 3501da177e4SLinus Torvalds * Read/write DACs are always overridable. 351d594e7ecSAl Viro * Executable DACs are overridable when there is 352d594e7ecSAl Viro * at least one exec bit set. 3531da177e4SLinus Torvalds */ 354d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 35523adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3561da177e4SLinus Torvalds return 0; 3571da177e4SLinus Torvalds 3581da177e4SLinus Torvalds /* 3591da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3601da177e4SLinus Torvalds */ 3617ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 362d594e7ecSAl Viro if (mask == MAY_READ) 36323adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3641da177e4SLinus Torvalds return 0; 3651da177e4SLinus Torvalds 3661da177e4SLinus Torvalds return -EACCES; 3671da177e4SLinus Torvalds } 3684d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3691da177e4SLinus Torvalds 3703ddcd056SLinus Torvalds /* 3713ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3723ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3733ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3743ddcd056SLinus Torvalds * permission function, use the fast case". 3753ddcd056SLinus Torvalds */ 3763ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3773ddcd056SLinus Torvalds { 3783ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3793ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3803ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3813ddcd056SLinus Torvalds 3823ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3833ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3843ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3853ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3863ddcd056SLinus Torvalds } 3873ddcd056SLinus Torvalds return generic_permission(inode, mask); 3883ddcd056SLinus Torvalds } 3893ddcd056SLinus Torvalds 390cb23beb5SChristoph Hellwig /** 3910bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3920bdaea90SDavid Howells * @inode: Inode to check permission on 3930bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 394cb23beb5SChristoph Hellwig * 3950bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 396948409c7SAndreas Gruenbacher * 397948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3980bdaea90SDavid Howells * 3990bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 4000bdaea90SDavid Howells * inode_permission(). 401cb23beb5SChristoph Hellwig */ 4020bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 4031da177e4SLinus Torvalds { 404e6305c43SAl Viro int retval; 4051da177e4SLinus Torvalds 4063ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 4071da177e4SLinus Torvalds /* 4081da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4091da177e4SLinus Torvalds */ 4101da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 4111da177e4SLinus Torvalds return -EACCES; 4121da177e4SLinus Torvalds } 4131da177e4SLinus Torvalds 4143ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4151da177e4SLinus Torvalds if (retval) 4161da177e4SLinus Torvalds return retval; 4171da177e4SLinus Torvalds 41808ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 41908ce5f16SSerge E. Hallyn if (retval) 42008ce5f16SSerge E. Hallyn return retval; 42108ce5f16SSerge E. Hallyn 422d09ca739SEric Paris return security_inode_permission(inode, mask); 4231da177e4SLinus Torvalds } 424bd5d0856SMiklos Szeredi EXPORT_SYMBOL(__inode_permission); 4251da177e4SLinus Torvalds 426f4d6ff89SAl Viro /** 4270bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4280bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 42955852635SRandy Dunlap * @inode: Inode to check permission on 4300bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4310bdaea90SDavid Howells * 4320bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4330bdaea90SDavid Howells */ 4340bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4350bdaea90SDavid Howells { 4360bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4370bdaea90SDavid Howells umode_t mode = inode->i_mode; 4380bdaea90SDavid Howells 4390bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4400bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4410bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4420bdaea90SDavid Howells return -EROFS; 4430bdaea90SDavid Howells } 4440bdaea90SDavid Howells return 0; 4450bdaea90SDavid Howells } 4460bdaea90SDavid Howells 4470bdaea90SDavid Howells /** 4480bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4490bdaea90SDavid Howells * @inode: Inode to check permission on 4500bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4510bdaea90SDavid Howells * 4520bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4530bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4540bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4550bdaea90SDavid Howells * 4560bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4570bdaea90SDavid Howells */ 4580bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4590bdaea90SDavid Howells { 4600bdaea90SDavid Howells int retval; 4610bdaea90SDavid Howells 4620bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4630bdaea90SDavid Howells if (retval) 4640bdaea90SDavid Howells return retval; 4650bdaea90SDavid Howells return __inode_permission(inode, mask); 4660bdaea90SDavid Howells } 4674d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4680bdaea90SDavid Howells 4690bdaea90SDavid Howells /** 4705dd784d0SJan Blunck * path_get - get a reference to a path 4715dd784d0SJan Blunck * @path: path to get the reference to 4725dd784d0SJan Blunck * 4735dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4745dd784d0SJan Blunck */ 475dcf787f3SAl Viro void path_get(const struct path *path) 4765dd784d0SJan Blunck { 4775dd784d0SJan Blunck mntget(path->mnt); 4785dd784d0SJan Blunck dget(path->dentry); 4795dd784d0SJan Blunck } 4805dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4815dd784d0SJan Blunck 4825dd784d0SJan Blunck /** 4831d957f9bSJan Blunck * path_put - put a reference to a path 4841d957f9bSJan Blunck * @path: path to put the reference to 4851d957f9bSJan Blunck * 4861d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4871d957f9bSJan Blunck */ 488dcf787f3SAl Viro void path_put(const struct path *path) 4891da177e4SLinus Torvalds { 4901d957f9bSJan Blunck dput(path->dentry); 4911d957f9bSJan Blunck mntput(path->mnt); 4921da177e4SLinus Torvalds } 4931d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4941da177e4SLinus Torvalds 4951f55a6ecSAl Viro struct nameidata { 4961f55a6ecSAl Viro struct path path; 4971f55a6ecSAl Viro struct qstr last; 4981f55a6ecSAl Viro struct path root; 4991f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5001f55a6ecSAl Viro unsigned int flags; 5011f55a6ecSAl Viro unsigned seq, m_seq; 5021f55a6ecSAl Viro int last_type; 5031f55a6ecSAl Viro unsigned depth; 5045e53084dSAl Viro struct file *base; 5051f55a6ecSAl Viro char *saved_names[MAX_NESTED_LINKS + 1]; 5061f55a6ecSAl Viro }; 5071f55a6ecSAl Viro 50819660af7SAl Viro /* 50931e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 51019660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 51119660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 51219660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 51319660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 51419660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 51519660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 51619660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 51731e6b01fSNick Piggin */ 51831e6b01fSNick Piggin 51931e6b01fSNick Piggin /** 52019660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 52119660af7SAl Viro * @nd: nameidata pathwalk data 52219660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 52339191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 52431e6b01fSNick Piggin * 52519660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 52619660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 52719660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 52831e6b01fSNick Piggin */ 52919660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 53031e6b01fSNick Piggin { 53131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 53231e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 53331e6b01fSNick Piggin 53431e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 535e5c832d5SLinus Torvalds 536e5c832d5SLinus Torvalds /* 53748a066e7SAl Viro * After legitimizing the bastards, terminate_walk() 53848a066e7SAl Viro * will do the right thing for non-RCU mode, and all our 53948a066e7SAl Viro * subsequent exit cases should rcu_read_unlock() 54048a066e7SAl Viro * before returning. Do vfsmount first; if dentry 54148a066e7SAl Viro * can't be legitimized, just set nd->path.dentry to NULL 54248a066e7SAl Viro * and rely on dput(NULL) being a no-op. 543e5c832d5SLinus Torvalds */ 54448a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 545e5c832d5SLinus Torvalds return -ECHILD; 546e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 54715570086SLinus Torvalds 54848a066e7SAl Viro if (!lockref_get_not_dead(&parent->d_lockref)) { 54948a066e7SAl Viro nd->path.dentry = NULL; 550d870b4a1SAl Viro goto out; 55148a066e7SAl Viro } 55248a066e7SAl Viro 55315570086SLinus Torvalds /* 55415570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 55515570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 55615570086SLinus Torvalds * 55715570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 55815570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 55915570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 56015570086SLinus Torvalds * since we checked the sequence number of the parent after we got 56115570086SLinus Torvalds * the child sequence number. So we know the parent must still 56215570086SLinus Torvalds * be valid if the child sequence number is still valid. 56315570086SLinus Torvalds */ 56419660af7SAl Viro if (!dentry) { 565e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 566e5c832d5SLinus Torvalds goto out; 56719660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 56819660af7SAl Viro } else { 569e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 570e5c832d5SLinus Torvalds goto out; 571e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 572e5c832d5SLinus Torvalds goto drop_dentry; 57319660af7SAl Viro } 574e5c832d5SLinus Torvalds 575e5c832d5SLinus Torvalds /* 576e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 577e5c832d5SLinus Torvalds * still valid and get it if required. 578e5c832d5SLinus Torvalds */ 579e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 580e5c832d5SLinus Torvalds spin_lock(&fs->lock); 581e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 582e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 58331e6b01fSNick Piggin path_get(&nd->root); 58431e6b01fSNick Piggin spin_unlock(&fs->lock); 58531e6b01fSNick Piggin } 58631e6b01fSNick Piggin 5878b61e74fSAl Viro rcu_read_unlock(); 58831e6b01fSNick Piggin return 0; 58919660af7SAl Viro 590e5c832d5SLinus Torvalds unlock_and_drop_dentry: 59131e6b01fSNick Piggin spin_unlock(&fs->lock); 592e5c832d5SLinus Torvalds drop_dentry: 5938b61e74fSAl Viro rcu_read_unlock(); 594e5c832d5SLinus Torvalds dput(dentry); 595d0d27277SLinus Torvalds goto drop_root_mnt; 596e5c832d5SLinus Torvalds out: 5978b61e74fSAl Viro rcu_read_unlock(); 598d0d27277SLinus Torvalds drop_root_mnt: 599d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 600d0d27277SLinus Torvalds nd->root.mnt = NULL; 60131e6b01fSNick Piggin return -ECHILD; 60231e6b01fSNick Piggin } 60331e6b01fSNick Piggin 6044ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 60534286d66SNick Piggin { 6064ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 60734286d66SNick Piggin } 60834286d66SNick Piggin 6099f1fafeeSAl Viro /** 6109f1fafeeSAl Viro * complete_walk - successful completion of path walk 6119f1fafeeSAl Viro * @nd: pointer nameidata 61239159de2SJeff Layton * 6139f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 6149f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 6159f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 6169f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 6179f1fafeeSAl Viro * need to drop nd->path. 61839159de2SJeff Layton */ 6199f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 62039159de2SJeff Layton { 62116c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 62239159de2SJeff Layton int status; 62339159de2SJeff Layton 6249f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 6259f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 6269f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6279f1fafeeSAl Viro nd->root.mnt = NULL; 62815570086SLinus Torvalds 62948a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { 6308b61e74fSAl Viro rcu_read_unlock(); 63148a066e7SAl Viro return -ECHILD; 63248a066e7SAl Viro } 633e5c832d5SLinus Torvalds if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 6348b61e74fSAl Viro rcu_read_unlock(); 63548a066e7SAl Viro mntput(nd->path.mnt); 6369f1fafeeSAl Viro return -ECHILD; 6379f1fafeeSAl Viro } 638e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 6398b61e74fSAl Viro rcu_read_unlock(); 640e5c832d5SLinus Torvalds dput(dentry); 64148a066e7SAl Viro mntput(nd->path.mnt); 642e5c832d5SLinus Torvalds return -ECHILD; 643e5c832d5SLinus Torvalds } 6448b61e74fSAl Viro rcu_read_unlock(); 6459f1fafeeSAl Viro } 6469f1fafeeSAl Viro 64716c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 64839159de2SJeff Layton return 0; 64939159de2SJeff Layton 650ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 65116c2cd71SAl Viro return 0; 65216c2cd71SAl Viro 653ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 65439159de2SJeff Layton if (status > 0) 65539159de2SJeff Layton return 0; 65639159de2SJeff Layton 65716c2cd71SAl Viro if (!status) 65839159de2SJeff Layton status = -ESTALE; 65916c2cd71SAl Viro 6609f1fafeeSAl Viro path_put(&nd->path); 66139159de2SJeff Layton return status; 66239159de2SJeff Layton } 66339159de2SJeff Layton 6642a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6652a737871SAl Viro { 666f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6672a737871SAl Viro } 6682a737871SAl Viro 6696de88d72SAl Viro static int link_path_walk(const char *, struct nameidata *); 6706de88d72SAl Viro 6717bd88377SAl Viro static __always_inline unsigned set_root_rcu(struct nameidata *nd) 67231e6b01fSNick Piggin { 67331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 6747bd88377SAl Viro unsigned seq, res; 675c28cc364SNick Piggin 676c28cc364SNick Piggin do { 677c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 67831e6b01fSNick Piggin nd->root = fs->root; 6797bd88377SAl Viro res = __read_seqcount_begin(&nd->root.dentry->d_seq); 680c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 6817bd88377SAl Viro return res; 68231e6b01fSNick Piggin } 68331e6b01fSNick Piggin 6841d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 685051d3812SIan Kent { 686051d3812SIan Kent dput(path->dentry); 6874ac91378SJan Blunck if (path->mnt != nd->path.mnt) 688051d3812SIan Kent mntput(path->mnt); 689051d3812SIan Kent } 690051d3812SIan Kent 6917b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 6927b9337aaSNick Piggin struct nameidata *nd) 693051d3812SIan Kent { 69431e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 6954ac91378SJan Blunck dput(nd->path.dentry); 69631e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 6974ac91378SJan Blunck mntput(nd->path.mnt); 6989a229683SHuang Shijie } 69931e6b01fSNick Piggin nd->path.mnt = path->mnt; 7004ac91378SJan Blunck nd->path.dentry = path->dentry; 701051d3812SIan Kent } 702051d3812SIan Kent 703b5fb63c1SChristoph Hellwig /* 704b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 705b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 706b5fb63c1SChristoph Hellwig */ 707b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 708b5fb63c1SChristoph Hellwig { 709b5fb63c1SChristoph Hellwig path_put(&nd->path); 710b5fb63c1SChristoph Hellwig 711b5fb63c1SChristoph Hellwig nd->path = *path; 712b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 713b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 714b5fb63c1SChristoph Hellwig } 715b5fb63c1SChristoph Hellwig 7161f55a6ecSAl Viro void nd_set_link(struct nameidata *nd, char *path) 7171f55a6ecSAl Viro { 7181f55a6ecSAl Viro nd->saved_names[nd->depth] = path; 7191f55a6ecSAl Viro } 7201f55a6ecSAl Viro EXPORT_SYMBOL(nd_set_link); 7211f55a6ecSAl Viro 7221f55a6ecSAl Viro char *nd_get_link(struct nameidata *nd) 7231f55a6ecSAl Viro { 7241f55a6ecSAl Viro return nd->saved_names[nd->depth]; 7251f55a6ecSAl Viro } 7261f55a6ecSAl Viro EXPORT_SYMBOL(nd_get_link); 7271f55a6ecSAl Viro 728574197e0SAl Viro static inline void put_link(struct nameidata *nd, struct path *link, void *cookie) 729574197e0SAl Viro { 730574197e0SAl Viro struct inode *inode = link->dentry->d_inode; 7316d7b5aaeSAl Viro if (inode->i_op->put_link) 732574197e0SAl Viro inode->i_op->put_link(link->dentry, nd, cookie); 733574197e0SAl Viro path_put(link); 734574197e0SAl Viro } 735574197e0SAl Viro 736561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 737561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 738800179c9SKees Cook 739800179c9SKees Cook /** 740800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 741800179c9SKees Cook * @link: The path of the symlink 74255852635SRandy Dunlap * @nd: nameidata pathwalk data 743800179c9SKees Cook * 744800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 745800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 746800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 747800179c9SKees Cook * processes from failing races against path names that may change out 748800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 749800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 750800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 751800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 752800179c9SKees Cook * 753800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 754800179c9SKees Cook */ 755800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 756800179c9SKees Cook { 757800179c9SKees Cook const struct inode *inode; 758800179c9SKees Cook const struct inode *parent; 759800179c9SKees Cook 760800179c9SKees Cook if (!sysctl_protected_symlinks) 761800179c9SKees Cook return 0; 762800179c9SKees Cook 763800179c9SKees Cook /* Allowed if owner and follower match. */ 764800179c9SKees Cook inode = link->dentry->d_inode; 76581abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 766800179c9SKees Cook return 0; 767800179c9SKees Cook 768800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 769800179c9SKees Cook parent = nd->path.dentry->d_inode; 770800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 771800179c9SKees Cook return 0; 772800179c9SKees Cook 773800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 77481abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 775800179c9SKees Cook return 0; 776800179c9SKees Cook 777ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 778800179c9SKees Cook path_put_conditional(link, nd); 779800179c9SKees Cook path_put(&nd->path); 780800179c9SKees Cook return -EACCES; 781800179c9SKees Cook } 782800179c9SKees Cook 783800179c9SKees Cook /** 784800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 785800179c9SKees Cook * @inode: the source inode to hardlink from 786800179c9SKees Cook * 787800179c9SKees Cook * Return false if at least one of the following conditions: 788800179c9SKees Cook * - inode is not a regular file 789800179c9SKees Cook * - inode is setuid 790800179c9SKees Cook * - inode is setgid and group-exec 791800179c9SKees Cook * - access failure for read and write 792800179c9SKees Cook * 793800179c9SKees Cook * Otherwise returns true. 794800179c9SKees Cook */ 795800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 796800179c9SKees Cook { 797800179c9SKees Cook umode_t mode = inode->i_mode; 798800179c9SKees Cook 799800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 800800179c9SKees Cook if (!S_ISREG(mode)) 801800179c9SKees Cook return false; 802800179c9SKees Cook 803800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 804800179c9SKees Cook if (mode & S_ISUID) 805800179c9SKees Cook return false; 806800179c9SKees Cook 807800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 808800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 809800179c9SKees Cook return false; 810800179c9SKees Cook 811800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 812800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 813800179c9SKees Cook return false; 814800179c9SKees Cook 815800179c9SKees Cook return true; 816800179c9SKees Cook } 817800179c9SKees Cook 818800179c9SKees Cook /** 819800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 820800179c9SKees Cook * @link: the source to hardlink from 821800179c9SKees Cook * 822800179c9SKees Cook * Block hardlink when all of: 823800179c9SKees Cook * - sysctl_protected_hardlinks enabled 824800179c9SKees Cook * - fsuid does not match inode 825800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 826800179c9SKees Cook * - not CAP_FOWNER 827800179c9SKees Cook * 828800179c9SKees Cook * Returns 0 if successful, -ve on error. 829800179c9SKees Cook */ 830800179c9SKees Cook static int may_linkat(struct path *link) 831800179c9SKees Cook { 832800179c9SKees Cook const struct cred *cred; 833800179c9SKees Cook struct inode *inode; 834800179c9SKees Cook 835800179c9SKees Cook if (!sysctl_protected_hardlinks) 836800179c9SKees Cook return 0; 837800179c9SKees Cook 838800179c9SKees Cook cred = current_cred(); 839800179c9SKees Cook inode = link->dentry->d_inode; 840800179c9SKees Cook 841800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 842800179c9SKees Cook * otherwise, it must be a safe source. 843800179c9SKees Cook */ 84481abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 845800179c9SKees Cook capable(CAP_FOWNER)) 846800179c9SKees Cook return 0; 847800179c9SKees Cook 848a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 849800179c9SKees Cook return -EPERM; 850800179c9SKees Cook } 851800179c9SKees Cook 852def4af30SAl Viro static __always_inline int 853574197e0SAl Viro follow_link(struct path *link, struct nameidata *nd, void **p) 8541da177e4SLinus Torvalds { 8557b9337aaSNick Piggin struct dentry *dentry = link->dentry; 8566d7b5aaeSAl Viro int error; 8576d7b5aaeSAl Viro char *s; 8581da177e4SLinus Torvalds 859844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 860844a3917SAl Viro 8610e794589SAl Viro if (link->mnt == nd->path.mnt) 8620e794589SAl Viro mntget(link->mnt); 8630e794589SAl Viro 8646d7b5aaeSAl Viro error = -ELOOP; 8656d7b5aaeSAl Viro if (unlikely(current->total_link_count >= 40)) 8666d7b5aaeSAl Viro goto out_put_nd_path; 8676d7b5aaeSAl Viro 868574197e0SAl Viro cond_resched(); 869574197e0SAl Viro current->total_link_count++; 870574197e0SAl Viro 87168ac1234SAl Viro touch_atime(link); 8721da177e4SLinus Torvalds nd_set_link(nd, NULL); 873cd4e91d3SAl Viro 87437882db0SNeilBrown error = security_inode_follow_link(dentry); 8756d7b5aaeSAl Viro if (error) 8766d7b5aaeSAl Viro goto out_put_nd_path; 87736f3b4f6SAl Viro 87886acdca1SAl Viro nd->last_type = LAST_BIND; 879def4af30SAl Viro *p = dentry->d_inode->i_op->follow_link(dentry, nd); 880def4af30SAl Viro error = PTR_ERR(*p); 8816d7b5aaeSAl Viro if (IS_ERR(*p)) 882408ef013SChristoph Hellwig goto out_put_nd_path; 8836d7b5aaeSAl Viro 884cc314eefSLinus Torvalds error = 0; 8856d7b5aaeSAl Viro s = nd_get_link(nd); 8866d7b5aaeSAl Viro if (s) { 887443ed254SAl Viro if (unlikely(IS_ERR(s))) { 888443ed254SAl Viro path_put(&nd->path); 889443ed254SAl Viro put_link(nd, link, *p); 890443ed254SAl Viro return PTR_ERR(s); 891443ed254SAl Viro } 892443ed254SAl Viro if (*s == '/') { 8937bd88377SAl Viro if (!nd->root.mnt) 894443ed254SAl Viro set_root(nd); 895443ed254SAl Viro path_put(&nd->path); 896443ed254SAl Viro nd->path = nd->root; 897443ed254SAl Viro path_get(&nd->root); 898443ed254SAl Viro nd->flags |= LOOKUP_JUMPED; 899443ed254SAl Viro } 900443ed254SAl Viro nd->inode = nd->path.dentry->d_inode; 901443ed254SAl Viro error = link_path_walk(s, nd); 9026d7b5aaeSAl Viro if (unlikely(error)) 9036d7b5aaeSAl Viro put_link(nd, link, *p); 904b5fb63c1SChristoph Hellwig } 9056d7b5aaeSAl Viro 9066d7b5aaeSAl Viro return error; 9076d7b5aaeSAl Viro 9086d7b5aaeSAl Viro out_put_nd_path: 90998f6ef64SArnd Bergmann *p = NULL; 9106d7b5aaeSAl Viro path_put(&nd->path); 9116d7b5aaeSAl Viro path_put(link); 9121da177e4SLinus Torvalds return error; 9131da177e4SLinus Torvalds } 9141da177e4SLinus Torvalds 91531e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 91631e6b01fSNick Piggin { 9170714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9180714a533SAl Viro struct mount *parent; 91931e6b01fSNick Piggin struct dentry *mountpoint; 92031e6b01fSNick Piggin 9210714a533SAl Viro parent = mnt->mnt_parent; 9220714a533SAl Viro if (&parent->mnt == path->mnt) 92331e6b01fSNick Piggin return 0; 924a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 92531e6b01fSNick Piggin path->dentry = mountpoint; 9260714a533SAl Viro path->mnt = &parent->mnt; 92731e6b01fSNick Piggin return 1; 92831e6b01fSNick Piggin } 92931e6b01fSNick Piggin 930f015f126SDavid Howells /* 931f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 932f015f126SDavid Howells * 933f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 934f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 935f015f126SDavid Howells * Up is towards /. 936f015f126SDavid Howells * 937f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 938f015f126SDavid Howells * root. 939f015f126SDavid Howells */ 940bab77ebfSAl Viro int follow_up(struct path *path) 9411da177e4SLinus Torvalds { 9420714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9430714a533SAl Viro struct mount *parent; 9441da177e4SLinus Torvalds struct dentry *mountpoint; 94599b7db7bSNick Piggin 94648a066e7SAl Viro read_seqlock_excl(&mount_lock); 9470714a533SAl Viro parent = mnt->mnt_parent; 9483c0a6163SAl Viro if (parent == mnt) { 94948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 9501da177e4SLinus Torvalds return 0; 9511da177e4SLinus Torvalds } 9520714a533SAl Viro mntget(&parent->mnt); 953a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 95448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 955bab77ebfSAl Viro dput(path->dentry); 956bab77ebfSAl Viro path->dentry = mountpoint; 957bab77ebfSAl Viro mntput(path->mnt); 9580714a533SAl Viro path->mnt = &parent->mnt; 9591da177e4SLinus Torvalds return 1; 9601da177e4SLinus Torvalds } 9614d359507SAl Viro EXPORT_SYMBOL(follow_up); 9621da177e4SLinus Torvalds 963b5c84bf6SNick Piggin /* 9649875cf80SDavid Howells * Perform an automount 9659875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9669875cf80SDavid Howells * were called with. 9671da177e4SLinus Torvalds */ 9689875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 9699875cf80SDavid Howells bool *need_mntput) 97031e6b01fSNick Piggin { 9719875cf80SDavid Howells struct vfsmount *mnt; 972ea5b778aSDavid Howells int err; 9739875cf80SDavid Howells 9749875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9759875cf80SDavid Howells return -EREMOTE; 9769875cf80SDavid Howells 9770ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9780ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9790ec26fd0SMiklos Szeredi * the name. 9800ec26fd0SMiklos Szeredi * 9810ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9825a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9830ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9840ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9850ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9860ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9875a30d8a2SDavid Howells */ 9885a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 989d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 9905a30d8a2SDavid Howells path->dentry->d_inode) 9919875cf80SDavid Howells return -EISDIR; 9920ec26fd0SMiklos Szeredi 9939875cf80SDavid Howells current->total_link_count++; 9949875cf80SDavid Howells if (current->total_link_count >= 40) 9959875cf80SDavid Howells return -ELOOP; 9969875cf80SDavid Howells 9979875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9989875cf80SDavid Howells if (IS_ERR(mnt)) { 9999875cf80SDavid Howells /* 10009875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 10019875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 10029875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 10039875cf80SDavid Howells * 10049875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 10059875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 10069875cf80SDavid Howells * the path is inaccessible and we should say so. 10079875cf80SDavid Howells */ 100849084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 10099875cf80SDavid Howells return -EREMOTE; 10109875cf80SDavid Howells return PTR_ERR(mnt); 101131e6b01fSNick Piggin } 1012ea5b778aSDavid Howells 10139875cf80SDavid Howells if (!mnt) /* mount collision */ 10149875cf80SDavid Howells return 0; 10159875cf80SDavid Howells 10168aef1884SAl Viro if (!*need_mntput) { 10178aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 10188aef1884SAl Viro mntget(path->mnt); 10198aef1884SAl Viro *need_mntput = true; 10208aef1884SAl Viro } 102119a167afSAl Viro err = finish_automount(mnt, path); 1022ea5b778aSDavid Howells 1023ea5b778aSDavid Howells switch (err) { 1024ea5b778aSDavid Howells case -EBUSY: 1025ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 102619a167afSAl Viro return 0; 1027ea5b778aSDavid Howells case 0: 10288aef1884SAl Viro path_put(path); 10299875cf80SDavid Howells path->mnt = mnt; 10309875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 10319875cf80SDavid Howells return 0; 103219a167afSAl Viro default: 103319a167afSAl Viro return err; 10349875cf80SDavid Howells } 103519a167afSAl Viro 1036ea5b778aSDavid Howells } 10379875cf80SDavid Howells 10389875cf80SDavid Howells /* 10399875cf80SDavid Howells * Handle a dentry that is managed in some way. 1040cc53ce53SDavid Howells * - Flagged for transit management (autofs) 10419875cf80SDavid Howells * - Flagged as mountpoint 10429875cf80SDavid Howells * - Flagged as automount point 10439875cf80SDavid Howells * 10449875cf80SDavid Howells * This may only be called in refwalk mode. 10459875cf80SDavid Howells * 10469875cf80SDavid Howells * Serialization is taken care of in namespace.c 10479875cf80SDavid Howells */ 10489875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 10499875cf80SDavid Howells { 10508aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10519875cf80SDavid Howells unsigned managed; 10529875cf80SDavid Howells bool need_mntput = false; 10538aef1884SAl Viro int ret = 0; 10549875cf80SDavid Howells 10559875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10569875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10579875cf80SDavid Howells * the components of that value change under us */ 10589875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10599875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10609875cf80SDavid Howells unlikely(managed != 0)) { 1061cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1062cc53ce53SDavid Howells * being held. */ 1063cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1064cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1065cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10661aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1067cc53ce53SDavid Howells if (ret < 0) 10688aef1884SAl Viro break; 1069cc53ce53SDavid Howells } 1070cc53ce53SDavid Howells 10719875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10729875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10739875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10749875cf80SDavid Howells if (mounted) { 10759875cf80SDavid Howells dput(path->dentry); 10769875cf80SDavid Howells if (need_mntput) 1077463ffb2eSAl Viro mntput(path->mnt); 1078463ffb2eSAl Viro path->mnt = mounted; 1079463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10809875cf80SDavid Howells need_mntput = true; 10819875cf80SDavid Howells continue; 1082463ffb2eSAl Viro } 1083463ffb2eSAl Viro 10849875cf80SDavid Howells /* Something is mounted on this dentry in another 10859875cf80SDavid Howells * namespace and/or whatever was mounted there in this 108648a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 108748a066e7SAl Viro * get it */ 10881da177e4SLinus Torvalds } 10899875cf80SDavid Howells 10909875cf80SDavid Howells /* Handle an automount point */ 10919875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 10929875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 10939875cf80SDavid Howells if (ret < 0) 10948aef1884SAl Viro break; 10959875cf80SDavid Howells continue; 10969875cf80SDavid Howells } 10979875cf80SDavid Howells 10989875cf80SDavid Howells /* We didn't change the current path point */ 10999875cf80SDavid Howells break; 11009875cf80SDavid Howells } 11018aef1884SAl Viro 11028aef1884SAl Viro if (need_mntput && path->mnt == mnt) 11038aef1884SAl Viro mntput(path->mnt); 11048aef1884SAl Viro if (ret == -EISDIR) 11058aef1884SAl Viro ret = 0; 1106a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 11071da177e4SLinus Torvalds } 11081da177e4SLinus Torvalds 1109cc53ce53SDavid Howells int follow_down_one(struct path *path) 11101da177e4SLinus Torvalds { 11111da177e4SLinus Torvalds struct vfsmount *mounted; 11121da177e4SLinus Torvalds 11131c755af4SAl Viro mounted = lookup_mnt(path); 11141da177e4SLinus Torvalds if (mounted) { 11159393bd07SAl Viro dput(path->dentry); 11169393bd07SAl Viro mntput(path->mnt); 11179393bd07SAl Viro path->mnt = mounted; 11189393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 11191da177e4SLinus Torvalds return 1; 11201da177e4SLinus Torvalds } 11211da177e4SLinus Torvalds return 0; 11221da177e4SLinus Torvalds } 11234d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 11241da177e4SLinus Torvalds 1125b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 112662a7375eSIan Kent { 1127b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1128b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 112962a7375eSIan Kent } 113062a7375eSIan Kent 11319875cf80SDavid Howells /* 1132287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1133287548e4SAl Viro * we meet a managed dentry that would need blocking. 11349875cf80SDavid Howells */ 11359875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1136287548e4SAl Viro struct inode **inode) 11379875cf80SDavid Howells { 113862a7375eSIan Kent for (;;) { 1139c7105365SAl Viro struct mount *mounted; 114062a7375eSIan Kent /* 114162a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 114262a7375eSIan Kent * that wants to block transit. 114362a7375eSIan Kent */ 1144b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1145b8faf035SNeilBrown case -ECHILD: 1146b8faf035SNeilBrown default: 1147ab90911fSDavid Howells return false; 1148b8faf035SNeilBrown case -EISDIR: 1149b8faf035SNeilBrown return true; 1150b8faf035SNeilBrown case 0: 1151b8faf035SNeilBrown break; 1152b8faf035SNeilBrown } 115362a7375eSIan Kent 115462a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1155b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 115662a7375eSIan Kent 1157474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 11589875cf80SDavid Howells if (!mounted) 11599875cf80SDavid Howells break; 1160c7105365SAl Viro path->mnt = &mounted->mnt; 1161c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1162a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 11639875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 116459430262SLinus Torvalds /* 116559430262SLinus Torvalds * Update the inode too. We don't need to re-check the 116659430262SLinus Torvalds * dentry sequence number here after this d_inode read, 116759430262SLinus Torvalds * because a mount-point is always pinned. 116859430262SLinus Torvalds */ 116959430262SLinus Torvalds *inode = path->dentry->d_inode; 11709875cf80SDavid Howells } 1171f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1172b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1173287548e4SAl Viro } 1174287548e4SAl Viro 117531e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 117631e6b01fSNick Piggin { 11774023bfc9SAl Viro struct inode *inode = nd->inode; 11787bd88377SAl Viro if (!nd->root.mnt) 117931e6b01fSNick Piggin set_root_rcu(nd); 118031e6b01fSNick Piggin 118131e6b01fSNick Piggin while (1) { 118231e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 118331e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 118431e6b01fSNick Piggin break; 118531e6b01fSNick Piggin } 118631e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 118731e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 118831e6b01fSNick Piggin struct dentry *parent = old->d_parent; 118931e6b01fSNick Piggin unsigned seq; 119031e6b01fSNick Piggin 11914023bfc9SAl Viro inode = parent->d_inode; 119231e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 119331e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1194ef7562d5SAl Viro goto failed; 119531e6b01fSNick Piggin nd->path.dentry = parent; 119631e6b01fSNick Piggin nd->seq = seq; 119731e6b01fSNick Piggin break; 119831e6b01fSNick Piggin } 119931e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 120031e6b01fSNick Piggin break; 12014023bfc9SAl Viro inode = nd->path.dentry->d_inode; 120231e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 120331e6b01fSNick Piggin } 1204b37199e6SAl Viro while (d_mountpoint(nd->path.dentry)) { 1205b37199e6SAl Viro struct mount *mounted; 1206b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1207b37199e6SAl Viro if (!mounted) 1208b37199e6SAl Viro break; 1209b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1210b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 12114023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1212b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1213f5be3e29SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1214b37199e6SAl Viro goto failed; 1215b37199e6SAl Viro } 12164023bfc9SAl Viro nd->inode = inode; 121731e6b01fSNick Piggin return 0; 1218ef7562d5SAl Viro 1219ef7562d5SAl Viro failed: 1220ef7562d5SAl Viro nd->flags &= ~LOOKUP_RCU; 12215b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1222ef7562d5SAl Viro nd->root.mnt = NULL; 12238b61e74fSAl Viro rcu_read_unlock(); 1224ef7562d5SAl Viro return -ECHILD; 122531e6b01fSNick Piggin } 122631e6b01fSNick Piggin 12279875cf80SDavid Howells /* 1228cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1229cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1230cc53ce53SDavid Howells * caller is permitted to proceed or not. 1231cc53ce53SDavid Howells */ 12327cc90cc3SAl Viro int follow_down(struct path *path) 1233cc53ce53SDavid Howells { 1234cc53ce53SDavid Howells unsigned managed; 1235cc53ce53SDavid Howells int ret; 1236cc53ce53SDavid Howells 1237cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1238cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1239cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1240cc53ce53SDavid Howells * being held. 1241cc53ce53SDavid Howells * 1242cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1243cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1244cc53ce53SDavid Howells * other than its daemon the right to mount on its 1245cc53ce53SDavid Howells * superstructure. 1246cc53ce53SDavid Howells * 1247cc53ce53SDavid Howells * The filesystem may sleep at this point. 1248cc53ce53SDavid Howells */ 1249cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1250cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1251cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1252ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 12531aed3e42SAl Viro path->dentry, false); 1254cc53ce53SDavid Howells if (ret < 0) 1255cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1256cc53ce53SDavid Howells } 1257cc53ce53SDavid Howells 1258cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1259cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1260cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1261cc53ce53SDavid Howells if (!mounted) 1262cc53ce53SDavid Howells break; 1263cc53ce53SDavid Howells dput(path->dentry); 1264cc53ce53SDavid Howells mntput(path->mnt); 1265cc53ce53SDavid Howells path->mnt = mounted; 1266cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1267cc53ce53SDavid Howells continue; 1268cc53ce53SDavid Howells } 1269cc53ce53SDavid Howells 1270cc53ce53SDavid Howells /* Don't handle automount points here */ 1271cc53ce53SDavid Howells break; 1272cc53ce53SDavid Howells } 1273cc53ce53SDavid Howells return 0; 1274cc53ce53SDavid Howells } 12754d359507SAl Viro EXPORT_SYMBOL(follow_down); 1276cc53ce53SDavid Howells 1277cc53ce53SDavid Howells /* 12789875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12799875cf80SDavid Howells */ 12809875cf80SDavid Howells static void follow_mount(struct path *path) 12819875cf80SDavid Howells { 12829875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12839875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12849875cf80SDavid Howells if (!mounted) 12859875cf80SDavid Howells break; 12869875cf80SDavid Howells dput(path->dentry); 12879875cf80SDavid Howells mntput(path->mnt); 12889875cf80SDavid Howells path->mnt = mounted; 12899875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12909875cf80SDavid Howells } 12919875cf80SDavid Howells } 12929875cf80SDavid Howells 129331e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 12941da177e4SLinus Torvalds { 12957bd88377SAl Viro if (!nd->root.mnt) 12962a737871SAl Viro set_root(nd); 1297e518ddb7SAndreas Mohr 12981da177e4SLinus Torvalds while(1) { 12994ac91378SJan Blunck struct dentry *old = nd->path.dentry; 13001da177e4SLinus Torvalds 13012a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 13022a737871SAl Viro nd->path.mnt == nd->root.mnt) { 13031da177e4SLinus Torvalds break; 13041da177e4SLinus Torvalds } 13054ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 13063088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 13073088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 13081da177e4SLinus Torvalds dput(old); 13091da177e4SLinus Torvalds break; 13101da177e4SLinus Torvalds } 13113088dd70SAl Viro if (!follow_up(&nd->path)) 13121da177e4SLinus Torvalds break; 13131da177e4SLinus Torvalds } 131479ed0226SAl Viro follow_mount(&nd->path); 131531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 13161da177e4SLinus Torvalds } 13171da177e4SLinus Torvalds 13181da177e4SLinus Torvalds /* 1319bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1320bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1321bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1322bad61189SMiklos Szeredi * 1323bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1324baa03890SNick Piggin */ 1325bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1326201f956eSAl Viro unsigned int flags, bool *need_lookup) 1327baa03890SNick Piggin { 1328baa03890SNick Piggin struct dentry *dentry; 1329bad61189SMiklos Szeredi int error; 1330baa03890SNick Piggin 1331bad61189SMiklos Szeredi *need_lookup = false; 1332bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1333bad61189SMiklos Szeredi if (dentry) { 133439e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1335201f956eSAl Viro error = d_revalidate(dentry, flags); 1336bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1337bad61189SMiklos Szeredi if (error < 0) { 1338bad61189SMiklos Szeredi dput(dentry); 1339bad61189SMiklos Szeredi return ERR_PTR(error); 13405542aa2fSEric W. Biederman } else { 13415542aa2fSEric W. Biederman d_invalidate(dentry); 1342bad61189SMiklos Szeredi dput(dentry); 1343bad61189SMiklos Szeredi dentry = NULL; 1344bad61189SMiklos Szeredi } 1345bad61189SMiklos Szeredi } 1346bad61189SMiklos Szeredi } 1347bad61189SMiklos Szeredi } 1348baa03890SNick Piggin 1349bad61189SMiklos Szeredi if (!dentry) { 1350bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1351baa03890SNick Piggin if (unlikely(!dentry)) 1352baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1353baa03890SNick Piggin 1354bad61189SMiklos Szeredi *need_lookup = true; 1355baa03890SNick Piggin } 1356baa03890SNick Piggin return dentry; 1357baa03890SNick Piggin } 1358baa03890SNick Piggin 1359baa03890SNick Piggin /* 136013a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 136113a2c3beSJ. Bruce Fields * unhashed. 1362bad61189SMiklos Szeredi * 1363bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 136444396f4bSJosef Bacik */ 1365bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 136672bd866aSAl Viro unsigned int flags) 136744396f4bSJosef Bacik { 136844396f4bSJosef Bacik struct dentry *old; 136944396f4bSJosef Bacik 137044396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1371bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1372e188dc02SMiklos Szeredi dput(dentry); 137344396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1374e188dc02SMiklos Szeredi } 137544396f4bSJosef Bacik 137672bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 137744396f4bSJosef Bacik if (unlikely(old)) { 137844396f4bSJosef Bacik dput(dentry); 137944396f4bSJosef Bacik dentry = old; 138044396f4bSJosef Bacik } 138144396f4bSJosef Bacik return dentry; 138244396f4bSJosef Bacik } 138344396f4bSJosef Bacik 1384a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 138572bd866aSAl Viro struct dentry *base, unsigned int flags) 1386a3255546SAl Viro { 1387bad61189SMiklos Szeredi bool need_lookup; 1388a3255546SAl Viro struct dentry *dentry; 1389a3255546SAl Viro 139072bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1391bad61189SMiklos Szeredi if (!need_lookup) 1392a3255546SAl Viro return dentry; 1393bad61189SMiklos Szeredi 139472bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1395a3255546SAl Viro } 1396a3255546SAl Viro 139744396f4bSJosef Bacik /* 13981da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 13991da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 14001da177e4SLinus Torvalds * It _is_ time-critical. 14011da177e4SLinus Torvalds */ 1402e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 140331e6b01fSNick Piggin struct path *path, struct inode **inode) 14041da177e4SLinus Torvalds { 14054ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 140631e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 14075a18fff2SAl Viro int need_reval = 1; 14085a18fff2SAl Viro int status = 1; 14099875cf80SDavid Howells int err; 14109875cf80SDavid Howells 14113cac260aSAl Viro /* 1412b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1413b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1414b04f784eSNick Piggin * do the non-racy lookup, below. 1415b04f784eSNick Piggin */ 141631e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 141731e6b01fSNick Piggin unsigned seq; 1418766c4cbfSAl Viro bool negative; 1419da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 14205a18fff2SAl Viro if (!dentry) 14215a18fff2SAl Viro goto unlazy; 14225a18fff2SAl Viro 142312f8ad4bSLinus Torvalds /* 142412f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 142512f8ad4bSLinus Torvalds * the dentry name information from lookup. 142612f8ad4bSLinus Torvalds */ 142712f8ad4bSLinus Torvalds *inode = dentry->d_inode; 1428766c4cbfSAl Viro negative = d_is_negative(dentry); 142912f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 143012f8ad4bSLinus Torvalds return -ECHILD; 1431766c4cbfSAl Viro if (negative) 1432766c4cbfSAl Viro return -ENOENT; 143312f8ad4bSLinus Torvalds 143412f8ad4bSLinus Torvalds /* 143512f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 143612f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 143712f8ad4bSLinus Torvalds * 143812f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 143912f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 144012f8ad4bSLinus Torvalds */ 144131e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 144231e6b01fSNick Piggin return -ECHILD; 144331e6b01fSNick Piggin nd->seq = seq; 14445a18fff2SAl Viro 144524643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 14464ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14475a18fff2SAl Viro if (unlikely(status <= 0)) { 14485a18fff2SAl Viro if (status != -ECHILD) 14495a18fff2SAl Viro need_reval = 0; 14505a18fff2SAl Viro goto unlazy; 14515a18fff2SAl Viro } 145224643087SAl Viro } 145331e6b01fSNick Piggin path->mnt = mnt; 145431e6b01fSNick Piggin path->dentry = dentry; 1455b8faf035SNeilBrown if (likely(__follow_mount_rcu(nd, path, inode))) 14569875cf80SDavid Howells return 0; 14575a18fff2SAl Viro unlazy: 145819660af7SAl Viro if (unlazy_walk(nd, dentry)) 14595a18fff2SAl Viro return -ECHILD; 14605a18fff2SAl Viro } else { 1461e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 146224643087SAl Viro } 14635a18fff2SAl Viro 146481e6f520SAl Viro if (unlikely(!dentry)) 146581e6f520SAl Viro goto need_lookup; 14665a18fff2SAl Viro 14675a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14684ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14695a18fff2SAl Viro if (unlikely(status <= 0)) { 14705a18fff2SAl Viro if (status < 0) { 14715a18fff2SAl Viro dput(dentry); 14725a18fff2SAl Viro return status; 14735a18fff2SAl Viro } 14745542aa2fSEric W. Biederman d_invalidate(dentry); 14755a18fff2SAl Viro dput(dentry); 147681e6f520SAl Viro goto need_lookup; 14775a18fff2SAl Viro } 1478697f514dSMiklos Szeredi 1479766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1480766c4cbfSAl Viro dput(dentry); 1481766c4cbfSAl Viro return -ENOENT; 1482766c4cbfSAl Viro } 14831da177e4SLinus Torvalds path->mnt = mnt; 14841da177e4SLinus Torvalds path->dentry = dentry; 14859875cf80SDavid Howells err = follow_managed(path, nd->flags); 148689312214SIan Kent if (unlikely(err < 0)) { 148789312214SIan Kent path_put_conditional(path, nd); 14889875cf80SDavid Howells return err; 148989312214SIan Kent } 1490a3fbbde7SAl Viro if (err) 1491a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 149231e6b01fSNick Piggin *inode = path->dentry->d_inode; 14931da177e4SLinus Torvalds return 0; 149481e6f520SAl Viro 149581e6f520SAl Viro need_lookup: 1496697f514dSMiklos Szeredi return 1; 1497697f514dSMiklos Szeredi } 1498697f514dSMiklos Szeredi 1499697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1500cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1501697f514dSMiklos Szeredi { 1502697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1503697f514dSMiklos Szeredi int err; 1504697f514dSMiklos Szeredi 1505697f514dSMiklos Szeredi parent = nd->path.dentry; 150681e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 150781e6f520SAl Viro 150881e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1509cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 151081e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 151181e6f520SAl Viro if (IS_ERR(dentry)) 151281e6f520SAl Viro return PTR_ERR(dentry); 1513697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1514697f514dSMiklos Szeredi path->dentry = dentry; 1515697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1516697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1517697f514dSMiklos Szeredi path_put_conditional(path, nd); 1518697f514dSMiklos Szeredi return err; 1519697f514dSMiklos Szeredi } 1520697f514dSMiklos Szeredi if (err) 1521697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1522697f514dSMiklos Szeredi return 0; 15231da177e4SLinus Torvalds } 15241da177e4SLinus Torvalds 152552094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 152652094c8aSAl Viro { 152752094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 15284ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 152952094c8aSAl Viro if (err != -ECHILD) 153052094c8aSAl Viro return err; 153119660af7SAl Viro if (unlazy_walk(nd, NULL)) 153252094c8aSAl Viro return -ECHILD; 153352094c8aSAl Viro } 15344ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 153552094c8aSAl Viro } 153652094c8aSAl Viro 15379856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 15389856fa1bSAl Viro { 15399856fa1bSAl Viro if (type == LAST_DOTDOT) { 15409856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 15419856fa1bSAl Viro if (follow_dotdot_rcu(nd)) 15429856fa1bSAl Viro return -ECHILD; 15439856fa1bSAl Viro } else 15449856fa1bSAl Viro follow_dotdot(nd); 15459856fa1bSAl Viro } 15469856fa1bSAl Viro return 0; 15479856fa1bSAl Viro } 15489856fa1bSAl Viro 1549951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1550951361f9SAl Viro { 1551951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1552951361f9SAl Viro path_put(&nd->path); 1553951361f9SAl Viro } else { 1554951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 15555b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1556951361f9SAl Viro nd->root.mnt = NULL; 15578b61e74fSAl Viro rcu_read_unlock(); 1558951361f9SAl Viro } 1559951361f9SAl Viro } 1560951361f9SAl Viro 15613ddcd056SLinus Torvalds /* 15623ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15633ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15643ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15653ddcd056SLinus Torvalds * for the common case. 15663ddcd056SLinus Torvalds */ 1567b18825a7SDavid Howells static inline int should_follow_link(struct dentry *dentry, int follow) 15683ddcd056SLinus Torvalds { 1569b18825a7SDavid Howells return unlikely(d_is_symlink(dentry)) ? follow : 0; 15703ddcd056SLinus Torvalds } 15713ddcd056SLinus Torvalds 157234b128f3SAl Viro static int walk_component(struct nameidata *nd, struct path *path, int follow) 1573ce57dfc1SAl Viro { 1574ce57dfc1SAl Viro struct inode *inode; 1575ce57dfc1SAl Viro int err; 1576ce57dfc1SAl Viro /* 1577ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1578ce57dfc1SAl Viro * to be able to know about the current root directory and 1579ce57dfc1SAl Viro * parent relationships. 1580ce57dfc1SAl Viro */ 158121b9b073SAl Viro if (unlikely(nd->last_type != LAST_NORM)) 158221b9b073SAl Viro return handle_dots(nd, nd->last_type); 1583e97cdc87SAl Viro err = lookup_fast(nd, path, &inode); 1584ce57dfc1SAl Viro if (unlikely(err)) { 1585697f514dSMiklos Szeredi if (err < 0) 1586697f514dSMiklos Szeredi goto out_err; 1587697f514dSMiklos Szeredi 1588cc2a5271SAl Viro err = lookup_slow(nd, path); 1589697f514dSMiklos Szeredi if (err < 0) 1590697f514dSMiklos Szeredi goto out_err; 1591697f514dSMiklos Szeredi 1592697f514dSMiklos Szeredi inode = path->dentry->d_inode; 1593697f514dSMiklos Szeredi err = -ENOENT; 1594698934dfSDavid Howells if (d_is_negative(path->dentry)) 1595697f514dSMiklos Szeredi goto out_path_put; 1596766c4cbfSAl Viro } 1597697f514dSMiklos Szeredi 1598b18825a7SDavid Howells if (should_follow_link(path->dentry, follow)) { 159919660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 16003cab989aSAl Viro if (unlikely(nd->path.mnt != path->mnt || 16013cab989aSAl Viro unlazy_walk(nd, path->dentry))) { 1602697f514dSMiklos Szeredi err = -ECHILD; 1603697f514dSMiklos Szeredi goto out_err; 160419660af7SAl Viro } 160519660af7SAl Viro } 1606ce57dfc1SAl Viro BUG_ON(inode != path->dentry->d_inode); 1607ce57dfc1SAl Viro return 1; 1608ce57dfc1SAl Viro } 1609ce57dfc1SAl Viro path_to_nameidata(path, nd); 1610ce57dfc1SAl Viro nd->inode = inode; 1611ce57dfc1SAl Viro return 0; 1612697f514dSMiklos Szeredi 1613697f514dSMiklos Szeredi out_path_put: 1614697f514dSMiklos Szeredi path_to_nameidata(path, nd); 1615697f514dSMiklos Szeredi out_err: 1616697f514dSMiklos Szeredi terminate_walk(nd); 1617697f514dSMiklos Szeredi return err; 1618ce57dfc1SAl Viro } 1619ce57dfc1SAl Viro 16201da177e4SLinus Torvalds /* 1621b356379aSAl Viro * This limits recursive symlink follows to 8, while 1622b356379aSAl Viro * limiting consecutive symlinks to 40. 1623b356379aSAl Viro * 1624b356379aSAl Viro * Without that kind of total limit, nasty chains of consecutive 1625b356379aSAl Viro * symlinks can cause almost arbitrarily long lookups. 1626b356379aSAl Viro */ 1627b356379aSAl Viro static inline int nested_symlink(struct path *path, struct nameidata *nd) 1628b356379aSAl Viro { 1629b356379aSAl Viro int res; 1630b356379aSAl Viro 1631b356379aSAl Viro if (unlikely(current->link_count >= MAX_NESTED_LINKS)) { 1632b356379aSAl Viro path_put_conditional(path, nd); 1633b356379aSAl Viro path_put(&nd->path); 1634b356379aSAl Viro return -ELOOP; 1635b356379aSAl Viro } 16361a4022f8SErez Zadok BUG_ON(nd->depth >= MAX_NESTED_LINKS); 1637b356379aSAl Viro 1638b356379aSAl Viro nd->depth++; 1639b356379aSAl Viro current->link_count++; 1640b356379aSAl Viro 1641b356379aSAl Viro do { 1642b356379aSAl Viro struct path link = *path; 1643b356379aSAl Viro void *cookie; 1644574197e0SAl Viro 1645574197e0SAl Viro res = follow_link(&link, nd, &cookie); 16466d7b5aaeSAl Viro if (res) 16476d7b5aaeSAl Viro break; 164821b9b073SAl Viro res = walk_component(nd, path, LOOKUP_FOLLOW); 1649574197e0SAl Viro put_link(nd, &link, cookie); 1650b356379aSAl Viro } while (res > 0); 1651b356379aSAl Viro 1652b356379aSAl Viro current->link_count--; 1653b356379aSAl Viro nd->depth--; 1654b356379aSAl Viro return res; 1655b356379aSAl Viro } 1656b356379aSAl Viro 1657b356379aSAl Viro /* 1658bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1659bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1660bfcfaa77SLinus Torvalds * 1661bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1662bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1663bfcfaa77SLinus Torvalds * fast. 1664bfcfaa77SLinus Torvalds * 1665bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1666bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1667bfcfaa77SLinus Torvalds * crossing operation. 1668bfcfaa77SLinus Torvalds * 1669bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1670bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1671bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1672bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1673bfcfaa77SLinus Torvalds */ 1674bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1675bfcfaa77SLinus Torvalds 1676f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1677bfcfaa77SLinus Torvalds 1678f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1679bfcfaa77SLinus Torvalds 1680bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1681bfcfaa77SLinus Torvalds { 168299d263d4SLinus Torvalds return hash_64(hash, 32); 1683bfcfaa77SLinus Torvalds } 1684bfcfaa77SLinus Torvalds 1685bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1686bfcfaa77SLinus Torvalds 1687bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1688bfcfaa77SLinus Torvalds 1689bfcfaa77SLinus Torvalds #endif 1690bfcfaa77SLinus Torvalds 1691bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1692bfcfaa77SLinus Torvalds { 1693bfcfaa77SLinus Torvalds unsigned long a, mask; 1694bfcfaa77SLinus Torvalds unsigned long hash = 0; 1695bfcfaa77SLinus Torvalds 1696bfcfaa77SLinus Torvalds for (;;) { 1697e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1698bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1699bfcfaa77SLinus Torvalds break; 1700bfcfaa77SLinus Torvalds hash += a; 1701f132c5beSAl Viro hash *= 9; 1702bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1703bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1704bfcfaa77SLinus Torvalds if (!len) 1705bfcfaa77SLinus Torvalds goto done; 1706bfcfaa77SLinus Torvalds } 1707a5c21dceSWill Deacon mask = bytemask_from_count(len); 1708bfcfaa77SLinus Torvalds hash += mask & a; 1709bfcfaa77SLinus Torvalds done: 1710bfcfaa77SLinus Torvalds return fold_hash(hash); 1711bfcfaa77SLinus Torvalds } 1712bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1713bfcfaa77SLinus Torvalds 1714bfcfaa77SLinus Torvalds /* 1715bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1716d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1717bfcfaa77SLinus Torvalds */ 1718d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1719bfcfaa77SLinus Torvalds { 172036126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 172136126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1722bfcfaa77SLinus Torvalds 1723bfcfaa77SLinus Torvalds hash = a = 0; 1724bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1725bfcfaa77SLinus Torvalds do { 1726bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1727bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1728e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 172936126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 173036126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1731bfcfaa77SLinus Torvalds 173236126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 173336126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 173436126f8fSLinus Torvalds 173536126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 173636126f8fSLinus Torvalds 173736126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 17389226b5b4SLinus Torvalds len += find_zero(mask); 1739d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1740bfcfaa77SLinus Torvalds } 1741bfcfaa77SLinus Torvalds 1742bfcfaa77SLinus Torvalds #else 1743bfcfaa77SLinus Torvalds 17440145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 17450145acc2SLinus Torvalds { 17460145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 17470145acc2SLinus Torvalds while (len--) 17480145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 17490145acc2SLinus Torvalds return end_name_hash(hash); 17500145acc2SLinus Torvalds } 1751ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 17520145acc2SLinus Torvalds 17533ddcd056SLinus Torvalds /* 1754200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1755200e9ef7SLinus Torvalds * one character. 1756200e9ef7SLinus Torvalds */ 1757d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1758200e9ef7SLinus Torvalds { 1759200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1760200e9ef7SLinus Torvalds unsigned long len = 0, c; 1761200e9ef7SLinus Torvalds 1762200e9ef7SLinus Torvalds c = (unsigned char)*name; 1763200e9ef7SLinus Torvalds do { 1764200e9ef7SLinus Torvalds len++; 1765200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1766200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1767200e9ef7SLinus Torvalds } while (c && c != '/'); 1768d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1769200e9ef7SLinus Torvalds } 1770200e9ef7SLinus Torvalds 1771bfcfaa77SLinus Torvalds #endif 1772bfcfaa77SLinus Torvalds 1773200e9ef7SLinus Torvalds /* 17741da177e4SLinus Torvalds * Name resolution. 1775ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1776ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17771da177e4SLinus Torvalds * 1778ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1779ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17801da177e4SLinus Torvalds */ 17816de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17821da177e4SLinus Torvalds { 17831da177e4SLinus Torvalds struct path next; 17841da177e4SLinus Torvalds int err; 17851da177e4SLinus Torvalds 17861da177e4SLinus Torvalds while (*name=='/') 17871da177e4SLinus Torvalds name++; 17881da177e4SLinus Torvalds if (!*name) 1789086e183aSAl Viro return 0; 17901da177e4SLinus Torvalds 17911da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17921da177e4SLinus Torvalds for(;;) { 1793d6bb3e90SLinus Torvalds u64 hash_len; 1794fe479a58SAl Viro int type; 17951da177e4SLinus Torvalds 179652094c8aSAl Viro err = may_lookup(nd); 17971da177e4SLinus Torvalds if (err) 17981da177e4SLinus Torvalds break; 17991da177e4SLinus Torvalds 1800d6bb3e90SLinus Torvalds hash_len = hash_name(name); 18011da177e4SLinus Torvalds 1802fe479a58SAl Viro type = LAST_NORM; 1803d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1804fe479a58SAl Viro case 2: 1805200e9ef7SLinus Torvalds if (name[1] == '.') { 1806fe479a58SAl Viro type = LAST_DOTDOT; 180716c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 180816c2cd71SAl Viro } 1809fe479a58SAl Viro break; 1810fe479a58SAl Viro case 1: 1811fe479a58SAl Viro type = LAST_DOT; 1812fe479a58SAl Viro } 18135a202bcdSAl Viro if (likely(type == LAST_NORM)) { 18145a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 181516c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 18165a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1817a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1818da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 18195a202bcdSAl Viro if (err < 0) 18205a202bcdSAl Viro break; 1821d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1822d6bb3e90SLinus Torvalds name = this.name; 18235a202bcdSAl Viro } 18245a202bcdSAl Viro } 1825fe479a58SAl Viro 1826d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1827d6bb3e90SLinus Torvalds nd->last.name = name; 18285f4a6a69SAl Viro nd->last_type = type; 18295f4a6a69SAl Viro 1830d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1831d6bb3e90SLinus Torvalds if (!*name) 18325f4a6a69SAl Viro return 0; 1833200e9ef7SLinus Torvalds /* 1834200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1835200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1836200e9ef7SLinus Torvalds */ 1837200e9ef7SLinus Torvalds do { 1838d6bb3e90SLinus Torvalds name++; 1839d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 1840d6bb3e90SLinus Torvalds if (!*name) 18415f4a6a69SAl Viro return 0; 18425f4a6a69SAl Viro 184321b9b073SAl Viro err = walk_component(nd, &next, LOOKUP_FOLLOW); 1844ce57dfc1SAl Viro if (err < 0) 1845ce57dfc1SAl Viro return err; 1846fe479a58SAl Viro 1847ce57dfc1SAl Viro if (err) { 1848b356379aSAl Viro err = nested_symlink(&next, nd); 18491da177e4SLinus Torvalds if (err) 1850a7472babSAl Viro return err; 185131e6b01fSNick Piggin } 185244b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 18533ddcd056SLinus Torvalds err = -ENOTDIR; 18543ddcd056SLinus Torvalds break; 18555f4a6a69SAl Viro } 1856ce57dfc1SAl Viro } 1857951361f9SAl Viro terminate_walk(nd); 18581da177e4SLinus Torvalds return err; 18591da177e4SLinus Torvalds } 18601da177e4SLinus Torvalds 18616e8a1f87SAl Viro static int path_init(int dfd, const struct filename *name, unsigned int flags, 18625e53084dSAl Viro struct nameidata *nd) 186331e6b01fSNick Piggin { 186431e6b01fSNick Piggin int retval = 0; 1865fd2f7cb5SAl Viro const char *s = name->name; 186631e6b01fSNick Piggin 186731e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1868980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 186931e6b01fSNick Piggin nd->depth = 0; 18705e53084dSAl Viro nd->base = NULL; 18715b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1872b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1873b18825a7SDavid Howells struct inode *inode = root->d_inode; 1874fd2f7cb5SAl Viro if (*s) { 187544b1d530SMiklos Szeredi if (!d_can_lookup(root)) 18765b6ca027SAl Viro return -ENOTDIR; 18775b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18785b6ca027SAl Viro if (retval) 18795b6ca027SAl Viro return retval; 188073d049a4SAl Viro } 18815b6ca027SAl Viro nd->path = nd->root; 18825b6ca027SAl Viro nd->inode = inode; 18835b6ca027SAl Viro if (flags & LOOKUP_RCU) { 18848b61e74fSAl Viro rcu_read_lock(); 18855b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 188648a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 18875b6ca027SAl Viro } else { 18885b6ca027SAl Viro path_get(&nd->path); 18895b6ca027SAl Viro } 1890d465887fSAl Viro goto done; 18915b6ca027SAl Viro } 18925b6ca027SAl Viro 189331e6b01fSNick Piggin nd->root.mnt = NULL; 189431e6b01fSNick Piggin 189548a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 1896fd2f7cb5SAl Viro if (*s == '/') { 1897e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 18988b61e74fSAl Viro rcu_read_lock(); 18997bd88377SAl Viro nd->seq = set_root_rcu(nd); 1900e41f7d4eSAl Viro } else { 1901e41f7d4eSAl Viro set_root(nd); 1902e41f7d4eSAl Viro path_get(&nd->root); 1903e41f7d4eSAl Viro } 190431e6b01fSNick Piggin nd->path = nd->root; 190531e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1906e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 190731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1908c28cc364SNick Piggin unsigned seq; 190931e6b01fSNick Piggin 19108b61e74fSAl Viro rcu_read_lock(); 191131e6b01fSNick Piggin 1912c28cc364SNick Piggin do { 1913c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 191431e6b01fSNick Piggin nd->path = fs->pwd; 1915c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1916c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1917e41f7d4eSAl Viro } else { 1918e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1919e41f7d4eSAl Viro } 192031e6b01fSNick Piggin } else { 1921582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 19222903ff01SAl Viro struct fd f = fdget_raw(dfd); 192331e6b01fSNick Piggin struct dentry *dentry; 192431e6b01fSNick Piggin 19252903ff01SAl Viro if (!f.file) 19262903ff01SAl Viro return -EBADF; 192731e6b01fSNick Piggin 19282903ff01SAl Viro dentry = f.file->f_path.dentry; 192931e6b01fSNick Piggin 1930fd2f7cb5SAl Viro if (*s) { 193144b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 19322903ff01SAl Viro fdput(f); 19332903ff01SAl Viro return -ENOTDIR; 1934f52e0c11SAl Viro } 19352903ff01SAl Viro } 19362903ff01SAl Viro 19372903ff01SAl Viro nd->path = f.file->f_path; 1938e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19399c225f26SLinus Torvalds if (f.flags & FDPUT_FPUT) 19405e53084dSAl Viro nd->base = f.file; 1941c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 19428b61e74fSAl Viro rcu_read_lock(); 19435590ff0dSUlrich Drepper } else { 19442903ff01SAl Viro path_get(&nd->path); 19452903ff01SAl Viro fdput(f); 19461da177e4SLinus Torvalds } 1947e41f7d4eSAl Viro } 1948e41f7d4eSAl Viro 194931e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 19504023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 1951d465887fSAl Viro goto done; 19524023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 1953d465887fSAl Viro goto done; 19544023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 19554023bfc9SAl Viro nd->root.mnt = NULL; 19564023bfc9SAl Viro rcu_read_unlock(); 19574023bfc9SAl Viro return -ECHILD; 1958d465887fSAl Viro done: 1959d465887fSAl Viro current->total_link_count = 0; 1960fd2f7cb5SAl Viro return link_path_walk(s, nd); 19619b4a9b14SAl Viro } 19629b4a9b14SAl Viro 1963893b7775SAl Viro static void path_cleanup(struct nameidata *nd) 1964893b7775SAl Viro { 1965893b7775SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 1966893b7775SAl Viro path_put(&nd->root); 1967893b7775SAl Viro nd->root.mnt = NULL; 1968893b7775SAl Viro } 1969893b7775SAl Viro if (unlikely(nd->base)) 1970893b7775SAl Viro fput(nd->base); 19719b4a9b14SAl Viro } 19729b4a9b14SAl Viro 1973bd92d7feSAl Viro static inline int lookup_last(struct nameidata *nd, struct path *path) 1974bd92d7feSAl Viro { 1975bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1976bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1977bd92d7feSAl Viro 1978bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 197921b9b073SAl Viro return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW); 1980bd92d7feSAl Viro } 1981bd92d7feSAl Viro 19829b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 19835eb6b495SAl Viro static int path_lookupat(int dfd, const struct filename *name, 19849b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 19859b4a9b14SAl Viro { 1986bd92d7feSAl Viro struct path path; 1987bd92d7feSAl Viro int err; 198831e6b01fSNick Piggin 198931e6b01fSNick Piggin /* 199031e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 199131e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 199231e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 199331e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 199431e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 199531e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 199631e6b01fSNick Piggin * analogue, foo_rcu(). 199731e6b01fSNick Piggin * 199831e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 199931e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 200031e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 200131e6b01fSNick Piggin * be able to complete). 200231e6b01fSNick Piggin */ 20036e8a1f87SAl Viro err = path_init(dfd, name, flags, nd); 2004bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 2005bd92d7feSAl Viro err = lookup_last(nd, &path); 2006bd92d7feSAl Viro while (err > 0) { 2007bd92d7feSAl Viro void *cookie; 2008bd92d7feSAl Viro struct path link = path; 2009800179c9SKees Cook err = may_follow_link(&link, nd); 2010800179c9SKees Cook if (unlikely(err)) 2011800179c9SKees Cook break; 2012bd92d7feSAl Viro nd->flags |= LOOKUP_PARENT; 2013574197e0SAl Viro err = follow_link(&link, nd, &cookie); 20146d7b5aaeSAl Viro if (err) 20156d7b5aaeSAl Viro break; 2016bd92d7feSAl Viro err = lookup_last(nd, &path); 2017574197e0SAl Viro put_link(nd, &link, cookie); 2018bd92d7feSAl Viro } 2019bd92d7feSAl Viro } 2020ee0827cdSAl Viro 20219f1fafeeSAl Viro if (!err) 20229f1fafeeSAl Viro err = complete_walk(nd); 2023bd92d7feSAl Viro 2024bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 202544b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 2026bd92d7feSAl Viro path_put(&nd->path); 2027bd23a539SAl Viro err = -ENOTDIR; 2028bd92d7feSAl Viro } 2029bd92d7feSAl Viro } 203016c2cd71SAl Viro 2031893b7775SAl Viro path_cleanup(nd); 2032bd92d7feSAl Viro return err; 203331e6b01fSNick Piggin } 203431e6b01fSNick Piggin 2035873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 2036873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 2037873f1eedSJeff Layton { 20385eb6b495SAl Viro int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 2039873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 20405eb6b495SAl Viro retval = path_lookupat(dfd, name, flags, nd); 2041873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 20425eb6b495SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 2043873f1eedSJeff Layton 2044873f1eedSJeff Layton if (likely(!retval)) 2045adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2046873f1eedSJeff Layton return retval; 2047873f1eedSJeff Layton } 2048873f1eedSJeff Layton 204979714f72SAl Viro /* does lookup, returns the object with parent locked */ 205079714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 20515590ff0dSUlrich Drepper { 205251689104SPaul Moore struct filename *filename = getname_kernel(name); 205379714f72SAl Viro struct nameidata nd; 205479714f72SAl Viro struct dentry *d; 205551689104SPaul Moore int err; 205651689104SPaul Moore 205751689104SPaul Moore if (IS_ERR(filename)) 205851689104SPaul Moore return ERR_CAST(filename); 205951689104SPaul Moore 206051689104SPaul Moore err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd); 206151689104SPaul Moore if (err) { 206251689104SPaul Moore d = ERR_PTR(err); 206351689104SPaul Moore goto out; 206451689104SPaul Moore } 206579714f72SAl Viro if (nd.last_type != LAST_NORM) { 206679714f72SAl Viro path_put(&nd.path); 206751689104SPaul Moore d = ERR_PTR(-EINVAL); 206851689104SPaul Moore goto out; 206979714f72SAl Viro } 207079714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 20711e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 207279714f72SAl Viro if (IS_ERR(d)) { 207379714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 207479714f72SAl Viro path_put(&nd.path); 207551689104SPaul Moore goto out; 207679714f72SAl Viro } 207779714f72SAl Viro *path = nd.path; 207851689104SPaul Moore out: 207951689104SPaul Moore putname(filename); 208079714f72SAl Viro return d; 20815590ff0dSUlrich Drepper } 20825590ff0dSUlrich Drepper 2083d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2084d1811465SAl Viro { 2085d1811465SAl Viro struct nameidata nd; 208674eb8cc5SAl Viro struct filename *filename = getname_kernel(name); 208774eb8cc5SAl Viro int res = PTR_ERR(filename); 208874eb8cc5SAl Viro 208974eb8cc5SAl Viro if (!IS_ERR(filename)) { 209074eb8cc5SAl Viro res = filename_lookup(AT_FDCWD, filename, flags, &nd); 209174eb8cc5SAl Viro putname(filename); 2092d1811465SAl Viro if (!res) 2093d1811465SAl Viro *path = nd.path; 209474eb8cc5SAl Viro } 2095d1811465SAl Viro return res; 2096d1811465SAl Viro } 20974d359507SAl Viro EXPORT_SYMBOL(kern_path); 2098d1811465SAl Viro 209916f18200SJosef 'Jeff' Sipek /** 210016f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 210116f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 210216f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 210316f18200SJosef 'Jeff' Sipek * @name: pointer to file name 210416f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2105e0a01249SAl Viro * @path: pointer to struct path to fill 210616f18200SJosef 'Jeff' Sipek */ 210716f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 210816f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2109e0a01249SAl Viro struct path *path) 211016f18200SJosef 'Jeff' Sipek { 211174eb8cc5SAl Viro struct filename *filename = getname_kernel(name); 211274eb8cc5SAl Viro int err = PTR_ERR(filename); 211374eb8cc5SAl Viro 211474eb8cc5SAl Viro BUG_ON(flags & LOOKUP_PARENT); 211574eb8cc5SAl Viro 211674eb8cc5SAl Viro /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */ 211774eb8cc5SAl Viro if (!IS_ERR(filename)) { 2118e0a01249SAl Viro struct nameidata nd; 2119e0a01249SAl Viro nd.root.dentry = dentry; 2120e0a01249SAl Viro nd.root.mnt = mnt; 212174eb8cc5SAl Viro err = filename_lookup(AT_FDCWD, filename, 212274eb8cc5SAl Viro flags | LOOKUP_ROOT, &nd); 2123e0a01249SAl Viro if (!err) 2124e0a01249SAl Viro *path = nd.path; 212574eb8cc5SAl Viro putname(filename); 212674eb8cc5SAl Viro } 2127e0a01249SAl Viro return err; 212816f18200SJosef 'Jeff' Sipek } 21294d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 213016f18200SJosef 'Jeff' Sipek 2131eead1911SChristoph Hellwig /** 2132a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2133eead1911SChristoph Hellwig * @name: pathname component to lookup 2134eead1911SChristoph Hellwig * @base: base directory to lookup from 2135eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2136eead1911SChristoph Hellwig * 2137a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 21389e7543e9SAl Viro * not be called by generic code. 2139eead1911SChristoph Hellwig */ 2140057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2141057f6c01SJames Morris { 2142057f6c01SJames Morris struct qstr this; 21436a96ba54SAl Viro unsigned int c; 2144cda309deSMiklos Szeredi int err; 2145057f6c01SJames Morris 21462f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 21472f9092e1SDavid Woodhouse 21486a96ba54SAl Viro this.name = name; 21496a96ba54SAl Viro this.len = len; 21500145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 21516a96ba54SAl Viro if (!len) 21526a96ba54SAl Viro return ERR_PTR(-EACCES); 21536a96ba54SAl Viro 215421d8a15aSAl Viro if (unlikely(name[0] == '.')) { 215521d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 215621d8a15aSAl Viro return ERR_PTR(-EACCES); 215721d8a15aSAl Viro } 215821d8a15aSAl Viro 21596a96ba54SAl Viro while (len--) { 21606a96ba54SAl Viro c = *(const unsigned char *)name++; 21616a96ba54SAl Viro if (c == '/' || c == '\0') 21626a96ba54SAl Viro return ERR_PTR(-EACCES); 21636a96ba54SAl Viro } 21645a202bcdSAl Viro /* 21655a202bcdSAl Viro * See if the low-level filesystem might want 21665a202bcdSAl Viro * to use its own hash.. 21675a202bcdSAl Viro */ 21685a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2169da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 21705a202bcdSAl Viro if (err < 0) 21715a202bcdSAl Viro return ERR_PTR(err); 21725a202bcdSAl Viro } 2173eead1911SChristoph Hellwig 2174cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2175cda309deSMiklos Szeredi if (err) 2176cda309deSMiklos Szeredi return ERR_PTR(err); 2177cda309deSMiklos Szeredi 217872bd866aSAl Viro return __lookup_hash(&this, base, 0); 2179057f6c01SJames Morris } 21804d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2181057f6c01SJames Morris 21821fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 21831fa1e7f6SAndy Whitcroft struct path *path, int *empty) 21841da177e4SLinus Torvalds { 21852d8f3038SAl Viro struct nameidata nd; 218691a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 21871da177e4SLinus Torvalds int err = PTR_ERR(tmp); 21881da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 21892d8f3038SAl Viro 21902d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 21912d8f3038SAl Viro 2192873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 21931da177e4SLinus Torvalds putname(tmp); 21942d8f3038SAl Viro if (!err) 21952d8f3038SAl Viro *path = nd.path; 21961da177e4SLinus Torvalds } 21971da177e4SLinus Torvalds return err; 21981da177e4SLinus Torvalds } 21991da177e4SLinus Torvalds 22001fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 22011fa1e7f6SAndy Whitcroft struct path *path) 22021fa1e7f6SAndy Whitcroft { 2203f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 22041fa1e7f6SAndy Whitcroft } 22054d359507SAl Viro EXPORT_SYMBOL(user_path_at); 22061fa1e7f6SAndy Whitcroft 2207873f1eedSJeff Layton /* 2208873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2209873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2210873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2211873f1eedSJeff Layton * path-walking is complete. 2212873f1eedSJeff Layton */ 221391a27b2aSJeff Layton static struct filename * 2214f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2215f5beed75SAl Viro struct path *parent, 2216f5beed75SAl Viro struct qstr *last, 2217f5beed75SAl Viro int *type, 22189e790bd6SJeff Layton unsigned int flags) 22192ad94ae6SAl Viro { 2220f5beed75SAl Viro struct nameidata nd; 222191a27b2aSJeff Layton struct filename *s = getname(path); 22222ad94ae6SAl Viro int error; 22232ad94ae6SAl Viro 22249e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 22259e790bd6SJeff Layton flags &= LOOKUP_REVAL; 22269e790bd6SJeff Layton 22272ad94ae6SAl Viro if (IS_ERR(s)) 222891a27b2aSJeff Layton return s; 22292ad94ae6SAl Viro 2230f5beed75SAl Viro error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd); 223191a27b2aSJeff Layton if (error) { 22322ad94ae6SAl Viro putname(s); 223391a27b2aSJeff Layton return ERR_PTR(error); 223491a27b2aSJeff Layton } 2235f5beed75SAl Viro *parent = nd.path; 2236f5beed75SAl Viro *last = nd.last; 2237f5beed75SAl Viro *type = nd.last_type; 22382ad94ae6SAl Viro 223991a27b2aSJeff Layton return s; 22402ad94ae6SAl Viro } 22412ad94ae6SAl Viro 22428033426eSJeff Layton /** 2243197df04cSAl Viro * mountpoint_last - look up last component for umount 22448033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 22458033426eSJeff Layton * @path: pointer to container for result 22468033426eSJeff Layton * 22478033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 22488033426eSJeff Layton * need to resolve the path without doing any revalidation. 22498033426eSJeff Layton * 22508033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 22518033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 22528033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 22538033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 22548033426eSJeff Layton * bogus and it doesn't exist. 22558033426eSJeff Layton * 22568033426eSJeff Layton * Returns: 22578033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 22588033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 22598033426eSJeff Layton * put in this case. 22608033426eSJeff Layton * 22618033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 22628033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 22638033426eSJeff Layton * The nd->path reference will also be put. 22648033426eSJeff Layton * 22658033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 22668033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 22678033426eSJeff Layton * to the link, and nd->path will *not* be put. 22688033426eSJeff Layton */ 22698033426eSJeff Layton static int 2270197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 22718033426eSJeff Layton { 22728033426eSJeff Layton int error = 0; 22738033426eSJeff Layton struct dentry *dentry; 22748033426eSJeff Layton struct dentry *dir = nd->path.dentry; 22758033426eSJeff Layton 227635759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 227735759521SAl Viro if (nd->flags & LOOKUP_RCU) { 227835759521SAl Viro if (unlazy_walk(nd, NULL)) { 22798033426eSJeff Layton error = -ECHILD; 228035759521SAl Viro goto out; 228135759521SAl Viro } 22828033426eSJeff Layton } 22838033426eSJeff Layton 22848033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 22858033426eSJeff Layton 22868033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 22878033426eSJeff Layton error = handle_dots(nd, nd->last_type); 228835759521SAl Viro if (error) 228935759521SAl Viro goto out; 22908033426eSJeff Layton dentry = dget(nd->path.dentry); 229135759521SAl Viro goto done; 22928033426eSJeff Layton } 22938033426eSJeff Layton 22948033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 22958033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 22968033426eSJeff Layton if (!dentry) { 22978033426eSJeff Layton /* 22988033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 22998033426eSJeff Layton * so that means that this dentry is probably a symlink or the 23008033426eSJeff Layton * path doesn't actually point to a mounted dentry. 23018033426eSJeff Layton */ 23028033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 23038033426eSJeff Layton if (!dentry) { 23048033426eSJeff Layton error = -ENOMEM; 2305bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 230635759521SAl Viro goto out; 23078033426eSJeff Layton } 230835759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 230935759521SAl Viro error = PTR_ERR(dentry); 2310bcceeebaSDave Jones if (IS_ERR(dentry)) { 2311bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 231235759521SAl Viro goto out; 23138033426eSJeff Layton } 2314bcceeebaSDave Jones } 23158033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 23168033426eSJeff Layton 231735759521SAl Viro done: 2318698934dfSDavid Howells if (d_is_negative(dentry)) { 23198033426eSJeff Layton error = -ENOENT; 23208033426eSJeff Layton dput(dentry); 232135759521SAl Viro goto out; 232235759521SAl Viro } 23238033426eSJeff Layton path->dentry = dentry; 2324295dc39dSVasily Averin path->mnt = nd->path.mnt; 2325b18825a7SDavid Howells if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) 23268033426eSJeff Layton return 1; 2327295dc39dSVasily Averin mntget(path->mnt); 23288033426eSJeff Layton follow_mount(path); 232935759521SAl Viro error = 0; 233035759521SAl Viro out: 23318033426eSJeff Layton terminate_walk(nd); 23328033426eSJeff Layton return error; 23338033426eSJeff Layton } 23348033426eSJeff Layton 23358033426eSJeff Layton /** 2336197df04cSAl Viro * path_mountpoint - look up a path to be umounted 23378033426eSJeff Layton * @dfd: directory file descriptor to start walk from 23388033426eSJeff Layton * @name: full pathname to walk 2339606d6fe3SRandy Dunlap * @path: pointer to container for result 23408033426eSJeff Layton * @flags: lookup flags 23418033426eSJeff Layton * 23428033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2343606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 23448033426eSJeff Layton */ 23458033426eSJeff Layton static int 2346668696dcSAl Viro path_mountpoint(int dfd, const struct filename *name, struct path *path, 2347668696dcSAl Viro unsigned int flags) 23488033426eSJeff Layton { 23498033426eSJeff Layton struct nameidata nd; 23508033426eSJeff Layton int err; 23518033426eSJeff Layton 23526e8a1f87SAl Viro err = path_init(dfd, name, flags, &nd); 23538033426eSJeff Layton if (unlikely(err)) 2354115cbfdcSAl Viro goto out; 23558033426eSJeff Layton 2356197df04cSAl Viro err = mountpoint_last(&nd, path); 23578033426eSJeff Layton while (err > 0) { 23588033426eSJeff Layton void *cookie; 23598033426eSJeff Layton struct path link = *path; 23608033426eSJeff Layton err = may_follow_link(&link, &nd); 23618033426eSJeff Layton if (unlikely(err)) 23628033426eSJeff Layton break; 23638033426eSJeff Layton nd.flags |= LOOKUP_PARENT; 23648033426eSJeff Layton err = follow_link(&link, &nd, &cookie); 23658033426eSJeff Layton if (err) 23668033426eSJeff Layton break; 2367197df04cSAl Viro err = mountpoint_last(&nd, path); 23688033426eSJeff Layton put_link(&nd, &link, cookie); 23698033426eSJeff Layton } 23708033426eSJeff Layton out: 2371893b7775SAl Viro path_cleanup(&nd); 23728033426eSJeff Layton return err; 23738033426eSJeff Layton } 23748033426eSJeff Layton 23752d864651SAl Viro static int 2376668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 23772d864651SAl Viro unsigned int flags) 23782d864651SAl Viro { 2379cbaab2dbSAl Viro int error; 2380668696dcSAl Viro if (IS_ERR(name)) 2381668696dcSAl Viro return PTR_ERR(name); 2382668696dcSAl Viro error = path_mountpoint(dfd, name, path, flags | LOOKUP_RCU); 23832d864651SAl Viro if (unlikely(error == -ECHILD)) 2384668696dcSAl Viro error = path_mountpoint(dfd, name, path, flags); 23852d864651SAl Viro if (unlikely(error == -ESTALE)) 2386668696dcSAl Viro error = path_mountpoint(dfd, name, path, flags | LOOKUP_REVAL); 23872d864651SAl Viro if (likely(!error)) 2388668696dcSAl Viro audit_inode(name, path->dentry, 0); 2389668696dcSAl Viro putname(name); 23902d864651SAl Viro return error; 23912d864651SAl Viro } 23922d864651SAl Viro 23938033426eSJeff Layton /** 2394197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 23958033426eSJeff Layton * @dfd: directory file descriptor 23968033426eSJeff Layton * @name: pathname from userland 23978033426eSJeff Layton * @flags: lookup flags 23988033426eSJeff Layton * @path: pointer to container to hold result 23998033426eSJeff Layton * 24008033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 24018033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 24028033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 24038033426eSJeff Layton * and avoid revalidating the last component. 24048033426eSJeff Layton * 24058033426eSJeff Layton * Returns 0 and populates "path" on success. 24068033426eSJeff Layton */ 24078033426eSJeff Layton int 2408197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 24098033426eSJeff Layton struct path *path) 24108033426eSJeff Layton { 2411cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 24128033426eSJeff Layton } 24138033426eSJeff Layton 24142d864651SAl Viro int 24152d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 24162d864651SAl Viro unsigned int flags) 24172d864651SAl Viro { 2418cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 24192d864651SAl Viro } 24202d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 24212d864651SAl Viro 2422cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 24231da177e4SLinus Torvalds { 24248e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2425da9592edSDavid Howells 24268e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 24271da177e4SLinus Torvalds return 0; 24288e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 24291da177e4SLinus Torvalds return 0; 243023adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 24311da177e4SLinus Torvalds } 2432cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 24331da177e4SLinus Torvalds 24341da177e4SLinus Torvalds /* 24351da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 24361da177e4SLinus Torvalds * whether the type of victim is right. 24371da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 24381da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 24391da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 24401da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 24411da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 24421da177e4SLinus Torvalds * a. be owner of dir, or 24431da177e4SLinus Torvalds * b. be owner of victim, or 24441da177e4SLinus Torvalds * c. have CAP_FOWNER capability 24451da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 24461da177e4SLinus Torvalds * links pointing to it. 24471da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 24481da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 24491da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 24501da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 24511da177e4SLinus Torvalds * nfs_async_unlink(). 24521da177e4SLinus Torvalds */ 2453b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 24541da177e4SLinus Torvalds { 2455b18825a7SDavid Howells struct inode *inode = victim->d_inode; 24561da177e4SLinus Torvalds int error; 24571da177e4SLinus Torvalds 2458b18825a7SDavid Howells if (d_is_negative(victim)) 24591da177e4SLinus Torvalds return -ENOENT; 2460b18825a7SDavid Howells BUG_ON(!inode); 24611da177e4SLinus Torvalds 24621da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24634fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 24641da177e4SLinus Torvalds 2465f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 24661da177e4SLinus Torvalds if (error) 24671da177e4SLinus Torvalds return error; 24681da177e4SLinus Torvalds if (IS_APPEND(dir)) 24691da177e4SLinus Torvalds return -EPERM; 2470b18825a7SDavid Howells 2471b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2472b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 24731da177e4SLinus Torvalds return -EPERM; 24741da177e4SLinus Torvalds if (isdir) { 247544b1d530SMiklos Szeredi if (!d_is_dir(victim)) 24761da177e4SLinus Torvalds return -ENOTDIR; 24771da177e4SLinus Torvalds if (IS_ROOT(victim)) 24781da177e4SLinus Torvalds return -EBUSY; 247944b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 24801da177e4SLinus Torvalds return -EISDIR; 24811da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 24821da177e4SLinus Torvalds return -ENOENT; 24831da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 24841da177e4SLinus Torvalds return -EBUSY; 24851da177e4SLinus Torvalds return 0; 24861da177e4SLinus Torvalds } 24871da177e4SLinus Torvalds 24881da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 24891da177e4SLinus Torvalds * dir. 24901da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 24911da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 24921da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 24931da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 24941da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 24951da177e4SLinus Torvalds */ 2496a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 24971da177e4SLinus Torvalds { 249814e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 24991da177e4SLinus Torvalds if (child->d_inode) 25001da177e4SLinus Torvalds return -EEXIST; 25011da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25021da177e4SLinus Torvalds return -ENOENT; 2503f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 25041da177e4SLinus Torvalds } 25051da177e4SLinus Torvalds 25061da177e4SLinus Torvalds /* 25071da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 25081da177e4SLinus Torvalds */ 25091da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 25101da177e4SLinus Torvalds { 25111da177e4SLinus Torvalds struct dentry *p; 25121da177e4SLinus Torvalds 25131da177e4SLinus Torvalds if (p1 == p2) { 2514f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 25151da177e4SLinus Torvalds return NULL; 25161da177e4SLinus Torvalds } 25171da177e4SLinus Torvalds 2518a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25191da177e4SLinus Torvalds 2520e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2521e2761a11SOGAWA Hirofumi if (p) { 2522f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2523f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 25241da177e4SLinus Torvalds return p; 25251da177e4SLinus Torvalds } 25261da177e4SLinus Torvalds 2527e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2528e2761a11SOGAWA Hirofumi if (p) { 2529f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2530f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 25311da177e4SLinus Torvalds return p; 25321da177e4SLinus Torvalds } 25331da177e4SLinus Torvalds 2534f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2535d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 25361da177e4SLinus Torvalds return NULL; 25371da177e4SLinus Torvalds } 25384d359507SAl Viro EXPORT_SYMBOL(lock_rename); 25391da177e4SLinus Torvalds 25401da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 25411da177e4SLinus Torvalds { 25421b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 25431da177e4SLinus Torvalds if (p1 != p2) { 25441b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2545a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25461da177e4SLinus Torvalds } 25471da177e4SLinus Torvalds } 25484d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 25491da177e4SLinus Torvalds 25504acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2551312b63fbSAl Viro bool want_excl) 25521da177e4SLinus Torvalds { 2553a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25541da177e4SLinus Torvalds if (error) 25551da177e4SLinus Torvalds return error; 25561da177e4SLinus Torvalds 2557acfa4380SAl Viro if (!dir->i_op->create) 25581da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 25591da177e4SLinus Torvalds mode &= S_IALLUGO; 25601da177e4SLinus Torvalds mode |= S_IFREG; 25611da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 25621da177e4SLinus Torvalds if (error) 25631da177e4SLinus Torvalds return error; 2564312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2565a74574aaSStephen Smalley if (!error) 2566f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25671da177e4SLinus Torvalds return error; 25681da177e4SLinus Torvalds } 25694d359507SAl Viro EXPORT_SYMBOL(vfs_create); 25701da177e4SLinus Torvalds 257173d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 25721da177e4SLinus Torvalds { 25733fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 25741da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 25751da177e4SLinus Torvalds int error; 25761da177e4SLinus Torvalds 2577bcda7652SAl Viro /* O_PATH? */ 2578bcda7652SAl Viro if (!acc_mode) 2579bcda7652SAl Viro return 0; 2580bcda7652SAl Viro 25811da177e4SLinus Torvalds if (!inode) 25821da177e4SLinus Torvalds return -ENOENT; 25831da177e4SLinus Torvalds 2584c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2585c8fe8f30SChristoph Hellwig case S_IFLNK: 25861da177e4SLinus Torvalds return -ELOOP; 2587c8fe8f30SChristoph Hellwig case S_IFDIR: 2588c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 25891da177e4SLinus Torvalds return -EISDIR; 2590c8fe8f30SChristoph Hellwig break; 2591c8fe8f30SChristoph Hellwig case S_IFBLK: 2592c8fe8f30SChristoph Hellwig case S_IFCHR: 25933fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 25941da177e4SLinus Torvalds return -EACCES; 2595c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2596c8fe8f30SChristoph Hellwig case S_IFIFO: 2597c8fe8f30SChristoph Hellwig case S_IFSOCK: 25981da177e4SLinus Torvalds flag &= ~O_TRUNC; 2599c8fe8f30SChristoph Hellwig break; 26004a3fd211SDave Hansen } 2601b41572e9SDave Hansen 26023fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2603b41572e9SDave Hansen if (error) 2604b41572e9SDave Hansen return error; 26056146f0d5SMimi Zohar 26061da177e4SLinus Torvalds /* 26071da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 26081da177e4SLinus Torvalds */ 26091da177e4SLinus Torvalds if (IS_APPEND(inode)) { 26108737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 26117715b521SAl Viro return -EPERM; 26121da177e4SLinus Torvalds if (flag & O_TRUNC) 26137715b521SAl Viro return -EPERM; 26141da177e4SLinus Torvalds } 26151da177e4SLinus Torvalds 26161da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 26172e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 26187715b521SAl Viro return -EPERM; 26191da177e4SLinus Torvalds 2620f3c7691eSJ. Bruce Fields return 0; 26217715b521SAl Viro } 26227715b521SAl Viro 2623e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 26247715b521SAl Viro { 2625e1181ee6SJeff Layton struct path *path = &filp->f_path; 26267715b521SAl Viro struct inode *inode = path->dentry->d_inode; 26277715b521SAl Viro int error = get_write_access(inode); 26281da177e4SLinus Torvalds if (error) 26297715b521SAl Viro return error; 26301da177e4SLinus Torvalds /* 26311da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 26321da177e4SLinus Torvalds */ 2633d7a06983SJeff Layton error = locks_verify_locked(filp); 2634be6d3e56SKentaro Takeda if (!error) 2635ea0d3ab2STetsuo Handa error = security_path_truncate(path); 26361da177e4SLinus Torvalds if (!error) { 26377715b521SAl Viro error = do_truncate(path->dentry, 0, 2638d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2639e1181ee6SJeff Layton filp); 26401da177e4SLinus Torvalds } 26411da177e4SLinus Torvalds put_write_access(inode); 2642acd0c935SMimi Zohar return error; 26431da177e4SLinus Torvalds } 26441da177e4SLinus Torvalds 2645d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2646d57999e1SDave Hansen { 26478a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 26488a5e929dSAl Viro flag--; 2649d57999e1SDave Hansen return flag; 2650d57999e1SDave Hansen } 2651d57999e1SDave Hansen 2652d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2653d18e9008SMiklos Szeredi { 2654d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2655d18e9008SMiklos Szeredi if (error) 2656d18e9008SMiklos Szeredi return error; 2657d18e9008SMiklos Szeredi 2658d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2659d18e9008SMiklos Szeredi if (error) 2660d18e9008SMiklos Szeredi return error; 2661d18e9008SMiklos Szeredi 2662d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2663d18e9008SMiklos Szeredi } 2664d18e9008SMiklos Szeredi 26651acf0af9SDavid Howells /* 26661acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 26671acf0af9SDavid Howells * dentry. 26681acf0af9SDavid Howells * 26691acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 26701acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 26711acf0af9SDavid Howells * 26721acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 26731acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 26741acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 26751acf0af9SDavid Howells * 26761acf0af9SDavid Howells * Returns an error code otherwise. 26771acf0af9SDavid Howells */ 26782675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 267930d90494SAl Viro struct path *path, struct file *file, 2680015c3bbcSMiklos Szeredi const struct open_flags *op, 268164894cf8SAl Viro bool got_write, bool need_lookup, 268247237687SAl Viro int *opened) 2683d18e9008SMiklos Szeredi { 2684d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2685d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2686d18e9008SMiklos Szeredi umode_t mode; 2687d18e9008SMiklos Szeredi int error; 2688d18e9008SMiklos Szeredi int acc_mode; 2689d18e9008SMiklos Szeredi int create_error = 0; 2690d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2691116cc022SMiklos Szeredi bool excl; 2692d18e9008SMiklos Szeredi 2693d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2694d18e9008SMiklos Szeredi 2695d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2696d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 26972675a4ebSAl Viro error = -ENOENT; 2698d18e9008SMiklos Szeredi goto out; 2699d18e9008SMiklos Szeredi } 2700d18e9008SMiklos Szeredi 270162b259d8SMiklos Szeredi mode = op->mode; 2702d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2703d18e9008SMiklos Szeredi mode &= ~current_umask(); 2704d18e9008SMiklos Szeredi 2705116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2706116cc022SMiklos Szeredi if (excl) 2707d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2708d18e9008SMiklos Szeredi 2709d18e9008SMiklos Szeredi /* 2710d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2711d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2712d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2713d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2714d18e9008SMiklos Szeredi * 2715d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2716d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2717d18e9008SMiklos Szeredi */ 271864894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 271964894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 272064894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2721d18e9008SMiklos Szeredi /* 2722d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2723d18e9008SMiklos Szeredi * back to lookup + open 2724d18e9008SMiklos Szeredi */ 2725d18e9008SMiklos Szeredi goto no_open; 2726d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2727d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 272864894cf8SAl Viro create_error = -EROFS; 2729d18e9008SMiklos Szeredi goto no_open; 2730d18e9008SMiklos Szeredi } else { 2731d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 273264894cf8SAl Viro create_error = -EROFS; 2733d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2734d18e9008SMiklos Szeredi } 2735d18e9008SMiklos Szeredi } 2736d18e9008SMiklos Szeredi 2737d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 273838227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2739d18e9008SMiklos Szeredi if (error) { 2740d18e9008SMiklos Szeredi create_error = error; 2741d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2742d18e9008SMiklos Szeredi goto no_open; 2743d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2744d18e9008SMiklos Szeredi } 2745d18e9008SMiklos Szeredi } 2746d18e9008SMiklos Szeredi 2747d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2748d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2749d18e9008SMiklos Szeredi 275030d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 275130d90494SAl Viro file->f_path.mnt = nd->path.mnt; 275230d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 275347237687SAl Viro opened); 2754d9585277SAl Viro if (error < 0) { 2755d9585277SAl Viro if (create_error && error == -ENOENT) 2756d9585277SAl Viro error = create_error; 2757d18e9008SMiklos Szeredi goto out; 2758d18e9008SMiklos Szeredi } 2759d18e9008SMiklos Szeredi 2760d9585277SAl Viro if (error) { /* returned 1, that is */ 276130d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 27622675a4ebSAl Viro error = -EIO; 2763d18e9008SMiklos Szeredi goto out; 2764d18e9008SMiklos Szeredi } 276530d90494SAl Viro if (file->f_path.dentry) { 2766d18e9008SMiklos Szeredi dput(dentry); 276730d90494SAl Viro dentry = file->f_path.dentry; 2768d18e9008SMiklos Szeredi } 276903da633aSAl Viro if (*opened & FILE_CREATED) 277003da633aSAl Viro fsnotify_create(dir, dentry); 277103da633aSAl Viro if (!dentry->d_inode) { 277203da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 277303da633aSAl Viro if (create_error) { 277462b2ce96SSage Weil error = create_error; 277562b2ce96SSage Weil goto out; 277662b2ce96SSage Weil } 277703da633aSAl Viro } else { 277803da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 277903da633aSAl Viro error = -EEXIST; 278003da633aSAl Viro goto out; 278103da633aSAl Viro } 278203da633aSAl Viro } 2783d18e9008SMiklos Szeredi goto looked_up; 2784d18e9008SMiklos Szeredi } 2785d18e9008SMiklos Szeredi 2786d18e9008SMiklos Szeredi /* 2787d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2788d18e9008SMiklos Szeredi * here. 2789d18e9008SMiklos Szeredi */ 279003da633aSAl Viro acc_mode = op->acc_mode; 279103da633aSAl Viro if (*opened & FILE_CREATED) { 279203da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 279303da633aSAl Viro fsnotify_create(dir, dentry); 279403da633aSAl Viro acc_mode = MAY_OPEN; 279503da633aSAl Viro } 27962675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 27972675a4ebSAl Viro if (error) 27982675a4ebSAl Viro fput(file); 2799d18e9008SMiklos Szeredi 2800d18e9008SMiklos Szeredi out: 2801d18e9008SMiklos Szeredi dput(dentry); 28022675a4ebSAl Viro return error; 2803d18e9008SMiklos Szeredi 2804d18e9008SMiklos Szeredi no_open: 2805d18e9008SMiklos Szeredi if (need_lookup) { 280672bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2807d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 28082675a4ebSAl Viro return PTR_ERR(dentry); 2809d18e9008SMiklos Szeredi 2810d18e9008SMiklos Szeredi if (create_error) { 2811d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2812d18e9008SMiklos Szeredi 28132675a4ebSAl Viro error = create_error; 2814d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2815d18e9008SMiklos Szeredi if (!dentry->d_inode) 2816d18e9008SMiklos Szeredi goto out; 2817d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2818d18e9008SMiklos Szeredi goto out; 2819d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2820e36cb0b8SDavid Howells d_is_reg(dentry)) { 2821d18e9008SMiklos Szeredi goto out; 2822d18e9008SMiklos Szeredi } 2823d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2824d18e9008SMiklos Szeredi } 2825d18e9008SMiklos Szeredi } 2826d18e9008SMiklos Szeredi looked_up: 2827d18e9008SMiklos Szeredi path->dentry = dentry; 2828d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 28292675a4ebSAl Viro return 1; 2830d18e9008SMiklos Szeredi } 2831d18e9008SMiklos Szeredi 283231e6b01fSNick Piggin /* 28331acf0af9SDavid Howells * Look up and maybe create and open the last component. 2834d58ffd35SMiklos Szeredi * 2835d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2836d58ffd35SMiklos Szeredi * 28371acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 28381acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 28391acf0af9SDavid Howells * 28401acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 28411acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 28421acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 28431acf0af9SDavid Howells * specified then a negative dentry may be returned. 28441acf0af9SDavid Howells * 28451acf0af9SDavid Howells * An error code is returned otherwise. 28461acf0af9SDavid Howells * 28471acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 28481acf0af9SDavid Howells * cleared otherwise prior to returning. 2849d58ffd35SMiklos Szeredi */ 28502675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 285130d90494SAl Viro struct file *file, 2852d58ffd35SMiklos Szeredi const struct open_flags *op, 285364894cf8SAl Viro bool got_write, int *opened) 2854d58ffd35SMiklos Szeredi { 2855d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 285654ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2857d58ffd35SMiklos Szeredi struct dentry *dentry; 2858d58ffd35SMiklos Szeredi int error; 285954ef4872SMiklos Szeredi bool need_lookup; 2860d58ffd35SMiklos Szeredi 286147237687SAl Viro *opened &= ~FILE_CREATED; 2862201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2863d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 28642675a4ebSAl Viro return PTR_ERR(dentry); 2865d58ffd35SMiklos Szeredi 2866d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2867d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2868d18e9008SMiklos Szeredi goto out_no_open; 2869d18e9008SMiklos Szeredi 2870d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 287164894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 287247237687SAl Viro need_lookup, opened); 2873d18e9008SMiklos Szeredi } 2874d18e9008SMiklos Szeredi 287554ef4872SMiklos Szeredi if (need_lookup) { 287654ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 287754ef4872SMiklos Szeredi 287872bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 287954ef4872SMiklos Szeredi if (IS_ERR(dentry)) 28802675a4ebSAl Viro return PTR_ERR(dentry); 288154ef4872SMiklos Szeredi } 288254ef4872SMiklos Szeredi 2883d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2884d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2885d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2886d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2887d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2888d58ffd35SMiklos Szeredi /* 2889d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2890d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2891d58ffd35SMiklos Szeredi * the time when the file is created and when 2892d58ffd35SMiklos Szeredi * a permanent write count is taken through 2893015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2894d58ffd35SMiklos Szeredi */ 289564894cf8SAl Viro if (!got_write) { 289664894cf8SAl Viro error = -EROFS; 2897d58ffd35SMiklos Szeredi goto out_dput; 289864894cf8SAl Viro } 289947237687SAl Viro *opened |= FILE_CREATED; 2900d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2901d58ffd35SMiklos Szeredi if (error) 2902d58ffd35SMiklos Szeredi goto out_dput; 2903312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2904312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2905d58ffd35SMiklos Szeredi if (error) 2906d58ffd35SMiklos Szeredi goto out_dput; 2907d58ffd35SMiklos Szeredi } 2908d18e9008SMiklos Szeredi out_no_open: 2909d58ffd35SMiklos Szeredi path->dentry = dentry; 2910d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 29112675a4ebSAl Viro return 1; 2912d58ffd35SMiklos Szeredi 2913d58ffd35SMiklos Szeredi out_dput: 2914d58ffd35SMiklos Szeredi dput(dentry); 29152675a4ebSAl Viro return error; 2916d58ffd35SMiklos Szeredi } 2917d58ffd35SMiklos Szeredi 2918d58ffd35SMiklos Szeredi /* 2919fe2d35ffSAl Viro * Handle the last step of open() 292031e6b01fSNick Piggin */ 29212675a4ebSAl Viro static int do_last(struct nameidata *nd, struct path *path, 292230d90494SAl Viro struct file *file, const struct open_flags *op, 2923669abf4eSJeff Layton int *opened, struct filename *name) 2924fb1cc555SAl Viro { 2925a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2926ca344a89SAl Viro int open_flag = op->open_flag; 292777d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 292864894cf8SAl Viro bool got_write = false; 2929bcda7652SAl Viro int acc_mode = op->acc_mode; 2930a1eb3315SMiklos Szeredi struct inode *inode; 293116b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 293216b1c1cdSMiklos Szeredi bool retried = false; 293316c2cd71SAl Viro int error; 2934fb1cc555SAl Viro 2935c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2936c3e380b0SAl Viro nd->flags |= op->intent; 2937c3e380b0SAl Viro 2938bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2939fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2940fe2d35ffSAl Viro if (error) 29412675a4ebSAl Viro return error; 2942e83db167SMiklos Szeredi goto finish_open; 29431f36f774SAl Viro } 2944a2c36b45SAl Viro 2945ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2946fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2947fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2948fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2949e97cdc87SAl Viro error = lookup_fast(nd, path, &inode); 295071574865SMiklos Szeredi if (likely(!error)) 295171574865SMiklos Szeredi goto finish_lookup; 295271574865SMiklos Szeredi 2953ce57dfc1SAl Viro if (error < 0) 29542675a4ebSAl Viro goto out; 2955a1eb3315SMiklos Szeredi 295637d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2957b6183df7SMiklos Szeredi } else { 2958fe2d35ffSAl Viro /* create side of things */ 2959a3fbbde7SAl Viro /* 2960b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2961b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2962b6183df7SMiklos Szeredi * about to look up 2963a3fbbde7SAl Viro */ 29649f1fafeeSAl Viro error = complete_walk(nd); 29659f1fafeeSAl Viro if (error) 29662675a4ebSAl Viro return error; 2967fe2d35ffSAl Viro 296833e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 296916c2cd71SAl Viro error = -EISDIR; 29701f36f774SAl Viro /* trailing slashes? */ 297131e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 29722675a4ebSAl Viro goto out; 2973b6183df7SMiklos Szeredi } 29741f36f774SAl Viro 297516b1c1cdSMiklos Szeredi retry_lookup: 297664894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 297764894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 297864894cf8SAl Viro if (!error) 297964894cf8SAl Viro got_write = true; 298064894cf8SAl Viro /* 298164894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 298264894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 298364894cf8SAl Viro * dropping this one anyway. 298464894cf8SAl Viro */ 298564894cf8SAl Viro } 2986a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 298764894cf8SAl Viro error = lookup_open(nd, path, file, op, got_write, opened); 2988fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 2989fb1cc555SAl Viro 29902675a4ebSAl Viro if (error <= 0) { 29912675a4ebSAl Viro if (error) 2992d58ffd35SMiklos Szeredi goto out; 29936c0d46c4SAl Viro 299447237687SAl Viro if ((*opened & FILE_CREATED) || 2995496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 299677d660a8SMiklos Szeredi will_truncate = false; 2997d18e9008SMiklos Szeredi 2998adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 2999d18e9008SMiklos Szeredi goto opened; 3000d18e9008SMiklos Szeredi } 3001d18e9008SMiklos Szeredi 300247237687SAl Viro if (*opened & FILE_CREATED) { 30039b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3004ca344a89SAl Viro open_flag &= ~O_TRUNC; 300577d660a8SMiklos Szeredi will_truncate = false; 3006bcda7652SAl Viro acc_mode = MAY_OPEN; 3007d58ffd35SMiklos Szeredi path_to_nameidata(path, nd); 3008e83db167SMiklos Szeredi goto finish_open_created; 3009fb1cc555SAl Viro } 3010fb1cc555SAl Viro 3011fb1cc555SAl Viro /* 30123134f37eSJeff Layton * create/update audit record if it already exists. 3013fb1cc555SAl Viro */ 3014b18825a7SDavid Howells if (d_is_positive(path->dentry)) 3015adb5c247SJeff Layton audit_inode(name, path->dentry, 0); 3016fb1cc555SAl Viro 3017d18e9008SMiklos Szeredi /* 3018d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3019d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3020d18e9008SMiklos Szeredi * necessary...) 3021d18e9008SMiklos Szeredi */ 302264894cf8SAl Viro if (got_write) { 3023d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 302464894cf8SAl Viro got_write = false; 3025d18e9008SMiklos Szeredi } 3026d18e9008SMiklos Szeredi 3027fb1cc555SAl Viro error = -EEXIST; 3028f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 3029fb1cc555SAl Viro goto exit_dput; 3030fb1cc555SAl Viro 30319875cf80SDavid Howells error = follow_managed(path, nd->flags); 30329875cf80SDavid Howells if (error < 0) 3033fb1cc555SAl Viro goto exit_dput; 3034fb1cc555SAl Viro 3035a3fbbde7SAl Viro if (error) 3036a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 3037a3fbbde7SAl Viro 3038decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 3039decf3400SMiklos Szeredi inode = path->dentry->d_inode; 3040fb1cc555SAl Viro error = -ENOENT; 3041698934dfSDavid Howells if (d_is_negative(path->dentry)) { 304254c33e7fSMiklos Szeredi path_to_nameidata(path, nd); 30432675a4ebSAl Viro goto out; 304454c33e7fSMiklos Szeredi } 3045766c4cbfSAl Viro finish_lookup: 3046a5cfe2d5SAl Viro if (should_follow_link(path->dentry, nd->flags & LOOKUP_FOLLOW)) { 3047d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 30483cab989aSAl Viro if (unlikely(nd->path.mnt != path->mnt || 30493cab989aSAl Viro unlazy_walk(nd, path->dentry))) { 3050d45ea867SMiklos Szeredi error = -ECHILD; 30512675a4ebSAl Viro goto out; 3052d45ea867SMiklos Szeredi } 3053d45ea867SMiklos Szeredi } 3054d45ea867SMiklos Szeredi BUG_ON(inode != path->dentry->d_inode); 30552675a4ebSAl Viro return 1; 3056d45ea867SMiklos Szeredi } 3057fb1cc555SAl Viro 3058a5cfe2d5SAl Viro if (unlikely(d_is_symlink(path->dentry)) && !(open_flag & O_PATH)) { 3059a5cfe2d5SAl Viro path_to_nameidata(path, nd); 3060a5cfe2d5SAl Viro error = -ELOOP; 3061a5cfe2d5SAl Viro goto out; 3062a5cfe2d5SAl Viro } 3063a5cfe2d5SAl Viro 306416b1c1cdSMiklos Szeredi if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) { 3065fb1cc555SAl Viro path_to_nameidata(path, nd); 306616b1c1cdSMiklos Szeredi } else { 306716b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 306816b1c1cdSMiklos Szeredi save_parent.mnt = mntget(path->mnt); 306916b1c1cdSMiklos Szeredi nd->path.dentry = path->dentry; 307016b1c1cdSMiklos Szeredi 307116b1c1cdSMiklos Szeredi } 3072decf3400SMiklos Szeredi nd->inode = inode; 3073a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3074bc77daa7SAl Viro finish_open: 3075a3fbbde7SAl Viro error = complete_walk(nd); 307616b1c1cdSMiklos Szeredi if (error) { 307716b1c1cdSMiklos Szeredi path_put(&save_parent); 30782675a4ebSAl Viro return error; 307916b1c1cdSMiklos Szeredi } 3080bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3081fb1cc555SAl Viro error = -EISDIR; 308244b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 30832675a4ebSAl Viro goto out; 3084af2f5542SMiklos Szeredi error = -ENOTDIR; 308544b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 30862675a4ebSAl Viro goto out; 30874bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 308877d660a8SMiklos Szeredi will_truncate = false; 30896c0d46c4SAl Viro 30900f9d1a10SAl Viro if (will_truncate) { 30910f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 30920f9d1a10SAl Viro if (error) 30932675a4ebSAl Viro goto out; 309464894cf8SAl Viro got_write = true; 30950f9d1a10SAl Viro } 3096e83db167SMiklos Szeredi finish_open_created: 3097bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3098ca344a89SAl Viro if (error) 30992675a4ebSAl Viro goto out; 31004aa7c634SMiklos Szeredi 31014aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31024aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 31034aa7c634SMiklos Szeredi if (!error) { 31044aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31054aa7c634SMiklos Szeredi } else { 310630d90494SAl Viro if (error == -EOPENSTALE) 3107f60dc3dbSMiklos Szeredi goto stale_open; 3108015c3bbcSMiklos Szeredi goto out; 3109f60dc3dbSMiklos Szeredi } 3110a8277b9bSMiklos Szeredi opened: 31112675a4ebSAl Viro error = open_check_o_direct(file); 3112015c3bbcSMiklos Szeredi if (error) 3113015c3bbcSMiklos Szeredi goto exit_fput; 31143034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3115aa4caadbSMiklos Szeredi if (error) 3116aa4caadbSMiklos Szeredi goto exit_fput; 3117aa4caadbSMiklos Szeredi 31180f9d1a10SAl Viro if (will_truncate) { 31192675a4ebSAl Viro error = handle_truncate(file); 3120aa4caadbSMiklos Szeredi if (error) 3121aa4caadbSMiklos Szeredi goto exit_fput; 31220f9d1a10SAl Viro } 3123ca344a89SAl Viro out: 312464894cf8SAl Viro if (got_write) 31250f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 312616b1c1cdSMiklos Szeredi path_put(&save_parent); 3127e276ae67SMiklos Szeredi terminate_walk(nd); 31282675a4ebSAl Viro return error; 3129fb1cc555SAl Viro 3130fb1cc555SAl Viro exit_dput: 3131fb1cc555SAl Viro path_put_conditional(path, nd); 3132ca344a89SAl Viro goto out; 3133015c3bbcSMiklos Szeredi exit_fput: 31342675a4ebSAl Viro fput(file); 31352675a4ebSAl Viro goto out; 3136015c3bbcSMiklos Szeredi 3137f60dc3dbSMiklos Szeredi stale_open: 3138f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3139f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3140f60dc3dbSMiklos Szeredi goto out; 3141f60dc3dbSMiklos Szeredi 3142f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3143f60dc3dbSMiklos Szeredi path_put(&nd->path); 3144f60dc3dbSMiklos Szeredi nd->path = save_parent; 3145f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3146f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3147f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 314864894cf8SAl Viro if (got_write) { 3149f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 315064894cf8SAl Viro got_write = false; 3151f60dc3dbSMiklos Szeredi } 3152f60dc3dbSMiklos Szeredi retried = true; 3153f60dc3dbSMiklos Szeredi goto retry_lookup; 3154fb1cc555SAl Viro } 3155fb1cc555SAl Viro 315660545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 315760545d0dSAl Viro struct nameidata *nd, int flags, 315860545d0dSAl Viro const struct open_flags *op, 315960545d0dSAl Viro struct file *file, int *opened) 316060545d0dSAl Viro { 316160545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 316260545d0dSAl Viro struct dentry *dentry, *child; 316360545d0dSAl Viro struct inode *dir; 31645eb6b495SAl Viro int error = path_lookupat(dfd, pathname, 316560545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 316660545d0dSAl Viro if (unlikely(error)) 316760545d0dSAl Viro return error; 316860545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 316960545d0dSAl Viro if (unlikely(error)) 317060545d0dSAl Viro goto out; 317160545d0dSAl Viro /* we want directory to be writable */ 317260545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 317360545d0dSAl Viro if (error) 317460545d0dSAl Viro goto out2; 317560545d0dSAl Viro dentry = nd->path.dentry; 317660545d0dSAl Viro dir = dentry->d_inode; 317760545d0dSAl Viro if (!dir->i_op->tmpfile) { 317860545d0dSAl Viro error = -EOPNOTSUPP; 317960545d0dSAl Viro goto out2; 318060545d0dSAl Viro } 318160545d0dSAl Viro child = d_alloc(dentry, &name); 318260545d0dSAl Viro if (unlikely(!child)) { 318360545d0dSAl Viro error = -ENOMEM; 318460545d0dSAl Viro goto out2; 318560545d0dSAl Viro } 318660545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 318760545d0dSAl Viro nd->flags |= op->intent; 318860545d0dSAl Viro dput(nd->path.dentry); 318960545d0dSAl Viro nd->path.dentry = child; 319060545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 319160545d0dSAl Viro if (error) 319260545d0dSAl Viro goto out2; 319360545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 319469a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 319569a91c23SEric Rannaud error = may_open(&nd->path, MAY_OPEN, op->open_flag); 319660545d0dSAl Viro if (error) 319760545d0dSAl Viro goto out2; 319860545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 319960545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 320060545d0dSAl Viro if (error) 320160545d0dSAl Viro goto out2; 320260545d0dSAl Viro error = open_check_o_direct(file); 3203f4e0c30cSAl Viro if (error) { 320460545d0dSAl Viro fput(file); 3205f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3206f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3207f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3208f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3209f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3210f4e0c30cSAl Viro } 321160545d0dSAl Viro out2: 321260545d0dSAl Viro mnt_drop_write(nd->path.mnt); 321360545d0dSAl Viro out: 321460545d0dSAl Viro path_put(&nd->path); 321560545d0dSAl Viro return error; 321660545d0dSAl Viro } 321760545d0dSAl Viro 3218669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 321973d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 32201da177e4SLinus Torvalds { 322130d90494SAl Viro struct file *file; 32229850c056SAl Viro struct path path; 322347237687SAl Viro int opened = 0; 322413aab428SAl Viro int error; 322531e6b01fSNick Piggin 322630d90494SAl Viro file = get_empty_filp(); 32271afc99beSAl Viro if (IS_ERR(file)) 32281afc99beSAl Viro return file; 322931e6b01fSNick Piggin 323030d90494SAl Viro file->f_flags = op->open_flag; 323131e6b01fSNick Piggin 3232bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 323360545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 3234f15133dfSAl Viro goto out2; 323560545d0dSAl Viro } 323660545d0dSAl Viro 32376e8a1f87SAl Viro error = path_init(dfd, pathname, flags, nd); 323831e6b01fSNick Piggin if (unlikely(error)) 32392675a4ebSAl Viro goto out; 32401da177e4SLinus Torvalds 32412675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 32422675a4ebSAl Viro while (unlikely(error > 0)) { /* trailing symlink */ 32437b9337aaSNick Piggin struct path link = path; 3244def4af30SAl Viro void *cookie; 3245800179c9SKees Cook error = may_follow_link(&link, nd); 3246800179c9SKees Cook if (unlikely(error)) 3247800179c9SKees Cook break; 324873d049a4SAl Viro nd->flags |= LOOKUP_PARENT; 324973d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3250574197e0SAl Viro error = follow_link(&link, nd, &cookie); 3251c3e380b0SAl Viro if (unlikely(error)) 32522675a4ebSAl Viro break; 32532675a4ebSAl Viro error = do_last(nd, &path, file, op, &opened, pathname); 3254574197e0SAl Viro put_link(nd, &link, cookie); 3255806b681cSAl Viro } 325610fa8e62SAl Viro out: 3257893b7775SAl Viro path_cleanup(nd); 3258f15133dfSAl Viro out2: 32592675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32602675a4ebSAl Viro BUG_ON(!error); 326130d90494SAl Viro put_filp(file); 3262015c3bbcSMiklos Szeredi } 32632675a4ebSAl Viro if (unlikely(error)) { 32642675a4ebSAl Viro if (error == -EOPENSTALE) { 32652675a4ebSAl Viro if (flags & LOOKUP_RCU) 32662675a4ebSAl Viro error = -ECHILD; 32672675a4ebSAl Viro else 32682675a4ebSAl Viro error = -ESTALE; 32692675a4ebSAl Viro } 32702675a4ebSAl Viro file = ERR_PTR(error); 32712675a4ebSAl Viro } 32722675a4ebSAl Viro return file; 3273de459215SKirill Korotaev } 32741da177e4SLinus Torvalds 3275669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3276f9652e10SAl Viro const struct open_flags *op) 327713aab428SAl Viro { 327873d049a4SAl Viro struct nameidata nd; 3279f9652e10SAl Viro int flags = op->lookup_flags; 328013aab428SAl Viro struct file *filp; 328113aab428SAl Viro 328273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 328313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 328473d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 328513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 328673d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 328713aab428SAl Viro return filp; 328813aab428SAl Viro } 328913aab428SAl Viro 329073d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3291f9652e10SAl Viro const char *name, const struct open_flags *op) 329273d049a4SAl Viro { 329373d049a4SAl Viro struct nameidata nd; 329473d049a4SAl Viro struct file *file; 329551689104SPaul Moore struct filename *filename; 3296f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 329773d049a4SAl Viro 329873d049a4SAl Viro nd.root.mnt = mnt; 329973d049a4SAl Viro nd.root.dentry = dentry; 330073d049a4SAl Viro 3301b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 330273d049a4SAl Viro return ERR_PTR(-ELOOP); 330373d049a4SAl Viro 330451689104SPaul Moore filename = getname_kernel(name); 330551689104SPaul Moore if (unlikely(IS_ERR(filename))) 330651689104SPaul Moore return ERR_CAST(filename); 330751689104SPaul Moore 330851689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU); 330973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 331051689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags); 331173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 331251689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL); 331351689104SPaul Moore putname(filename); 331473d049a4SAl Viro return file; 331573d049a4SAl Viro } 331673d049a4SAl Viro 3317fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 33181ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33191da177e4SLinus Torvalds { 3320c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3321ed75e95dSAl Viro struct nameidata nd; 3322c30dabfeSJan Kara int err2; 33231ac12b4bSJeff Layton int error; 33241ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 33251ac12b4bSJeff Layton 33261ac12b4bSJeff Layton /* 33271ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 33281ac12b4bSJeff Layton * other flags passed in are ignored! 33291ac12b4bSJeff Layton */ 33301ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 33311ac12b4bSJeff Layton 3332fa14a0b8SAl Viro error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd); 3333ed75e95dSAl Viro if (error) 3334ed75e95dSAl Viro return ERR_PTR(error); 33351da177e4SLinus Torvalds 3336c663e5d8SChristoph Hellwig /* 3337c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3338c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3339c663e5d8SChristoph Hellwig */ 3340ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3341ed75e95dSAl Viro goto out; 3342ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3343ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3344c663e5d8SChristoph Hellwig 3345c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3346c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3347c663e5d8SChristoph Hellwig /* 3348c663e5d8SChristoph Hellwig * Do the final lookup. 3349c663e5d8SChristoph Hellwig */ 3350ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 33516a9f40d6SAl Viro dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags); 33521da177e4SLinus Torvalds if (IS_ERR(dentry)) 3353a8104a9fSAl Viro goto unlock; 3354c663e5d8SChristoph Hellwig 3355a8104a9fSAl Viro error = -EEXIST; 3356b18825a7SDavid Howells if (d_is_positive(dentry)) 3357a8104a9fSAl Viro goto fail; 3358b18825a7SDavid Howells 3359c663e5d8SChristoph Hellwig /* 3360c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3361c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3362c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3363c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3364c663e5d8SChristoph Hellwig */ 3365ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3366a8104a9fSAl Viro error = -ENOENT; 3367ed75e95dSAl Viro goto fail; 3368e9baf6e5SAl Viro } 3369c30dabfeSJan Kara if (unlikely(err2)) { 3370c30dabfeSJan Kara error = err2; 3371a8104a9fSAl Viro goto fail; 3372c30dabfeSJan Kara } 3373ed75e95dSAl Viro *path = nd.path; 3374e9baf6e5SAl Viro return dentry; 33751da177e4SLinus Torvalds fail: 3376a8104a9fSAl Viro dput(dentry); 3377a8104a9fSAl Viro dentry = ERR_PTR(error); 3378a8104a9fSAl Viro unlock: 3379dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3380c30dabfeSJan Kara if (!err2) 3381c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3382ed75e95dSAl Viro out: 3383dae6ad8fSAl Viro path_put(&nd.path); 3384ed75e95dSAl Viro return dentry; 3385dae6ad8fSAl Viro } 3386fa14a0b8SAl Viro 3387fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3388fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3389fa14a0b8SAl Viro { 339051689104SPaul Moore struct filename *filename = getname_kernel(pathname); 339151689104SPaul Moore struct dentry *res; 339251689104SPaul Moore 339351689104SPaul Moore if (IS_ERR(filename)) 339451689104SPaul Moore return ERR_CAST(filename); 339551689104SPaul Moore res = filename_create(dfd, filename, path, lookup_flags); 339651689104SPaul Moore putname(filename); 339751689104SPaul Moore return res; 3398fa14a0b8SAl Viro } 3399dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3400dae6ad8fSAl Viro 3401921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3402921a1650SAl Viro { 3403921a1650SAl Viro dput(dentry); 3404921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3405a8104a9fSAl Viro mnt_drop_write(path->mnt); 3406921a1650SAl Viro path_put(path); 3407921a1650SAl Viro } 3408921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3409921a1650SAl Viro 34101ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 34111ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3412dae6ad8fSAl Viro { 341391a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3414dae6ad8fSAl Viro struct dentry *res; 3415dae6ad8fSAl Viro if (IS_ERR(tmp)) 3416dae6ad8fSAl Viro return ERR_CAST(tmp); 3417fa14a0b8SAl Viro res = filename_create(dfd, tmp, path, lookup_flags); 3418dae6ad8fSAl Viro putname(tmp); 3419dae6ad8fSAl Viro return res; 3420dae6ad8fSAl Viro } 3421dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3422dae6ad8fSAl Viro 34231a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34241da177e4SLinus Torvalds { 3425a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34261da177e4SLinus Torvalds 34271da177e4SLinus Torvalds if (error) 34281da177e4SLinus Torvalds return error; 34291da177e4SLinus Torvalds 3430975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34311da177e4SLinus Torvalds return -EPERM; 34321da177e4SLinus Torvalds 3433acfa4380SAl Viro if (!dir->i_op->mknod) 34341da177e4SLinus Torvalds return -EPERM; 34351da177e4SLinus Torvalds 343608ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 343708ce5f16SSerge E. Hallyn if (error) 343808ce5f16SSerge E. Hallyn return error; 343908ce5f16SSerge E. Hallyn 34401da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 34411da177e4SLinus Torvalds if (error) 34421da177e4SLinus Torvalds return error; 34431da177e4SLinus Torvalds 34441da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3445a74574aaSStephen Smalley if (!error) 3446f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34471da177e4SLinus Torvalds return error; 34481da177e4SLinus Torvalds } 34494d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 34501da177e4SLinus Torvalds 3451f69aac00SAl Viro static int may_mknod(umode_t mode) 3452463c3197SDave Hansen { 3453463c3197SDave Hansen switch (mode & S_IFMT) { 3454463c3197SDave Hansen case S_IFREG: 3455463c3197SDave Hansen case S_IFCHR: 3456463c3197SDave Hansen case S_IFBLK: 3457463c3197SDave Hansen case S_IFIFO: 3458463c3197SDave Hansen case S_IFSOCK: 3459463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3460463c3197SDave Hansen return 0; 3461463c3197SDave Hansen case S_IFDIR: 3462463c3197SDave Hansen return -EPERM; 3463463c3197SDave Hansen default: 3464463c3197SDave Hansen return -EINVAL; 3465463c3197SDave Hansen } 3466463c3197SDave Hansen } 3467463c3197SDave Hansen 34688208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 34692e4d0924SHeiko Carstens unsigned, dev) 34701da177e4SLinus Torvalds { 34711da177e4SLinus Torvalds struct dentry *dentry; 3472dae6ad8fSAl Viro struct path path; 3473dae6ad8fSAl Viro int error; 3474972567f1SJeff Layton unsigned int lookup_flags = 0; 34751da177e4SLinus Torvalds 34768e4bfca1SAl Viro error = may_mknod(mode); 34778e4bfca1SAl Viro if (error) 34788e4bfca1SAl Viro return error; 3479972567f1SJeff Layton retry: 3480972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3481dae6ad8fSAl Viro if (IS_ERR(dentry)) 3482dae6ad8fSAl Viro return PTR_ERR(dentry); 34832ad94ae6SAl Viro 3484dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3485ce3b0f8dSAl Viro mode &= ~current_umask(); 3486dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3487be6d3e56SKentaro Takeda if (error) 3488a8104a9fSAl Viro goto out; 34891da177e4SLinus Torvalds switch (mode & S_IFMT) { 34901da177e4SLinus Torvalds case 0: case S_IFREG: 3491312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 34921da177e4SLinus Torvalds break; 34931da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3494dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 34951da177e4SLinus Torvalds new_decode_dev(dev)); 34961da177e4SLinus Torvalds break; 34971da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3498dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 34991da177e4SLinus Torvalds break; 35001da177e4SLinus Torvalds } 3501a8104a9fSAl Viro out: 3502921a1650SAl Viro done_path_create(&path, dentry); 3503972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3504972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3505972567f1SJeff Layton goto retry; 3506972567f1SJeff Layton } 35071da177e4SLinus Torvalds return error; 35081da177e4SLinus Torvalds } 35091da177e4SLinus Torvalds 35108208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35115590ff0dSUlrich Drepper { 35125590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35135590ff0dSUlrich Drepper } 35145590ff0dSUlrich Drepper 351518bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35161da177e4SLinus Torvalds { 3517a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35188de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35191da177e4SLinus Torvalds 35201da177e4SLinus Torvalds if (error) 35211da177e4SLinus Torvalds return error; 35221da177e4SLinus Torvalds 3523acfa4380SAl Viro if (!dir->i_op->mkdir) 35241da177e4SLinus Torvalds return -EPERM; 35251da177e4SLinus Torvalds 35261da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35271da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35281da177e4SLinus Torvalds if (error) 35291da177e4SLinus Torvalds return error; 35301da177e4SLinus Torvalds 35318de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 35328de52778SAl Viro return -EMLINK; 35338de52778SAl Viro 35341da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3535a74574aaSStephen Smalley if (!error) 3536f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 35371da177e4SLinus Torvalds return error; 35381da177e4SLinus Torvalds } 35394d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 35401da177e4SLinus Torvalds 3541a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 35421da177e4SLinus Torvalds { 35436902d925SDave Hansen struct dentry *dentry; 3544dae6ad8fSAl Viro struct path path; 3545dae6ad8fSAl Viro int error; 3546b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 35471da177e4SLinus Torvalds 3548b76d8b82SJeff Layton retry: 3549b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 35506902d925SDave Hansen if (IS_ERR(dentry)) 3551dae6ad8fSAl Viro return PTR_ERR(dentry); 35526902d925SDave Hansen 3553dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3554ce3b0f8dSAl Viro mode &= ~current_umask(); 3555dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3556a8104a9fSAl Viro if (!error) 3557dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3558921a1650SAl Viro done_path_create(&path, dentry); 3559b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3560b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3561b76d8b82SJeff Layton goto retry; 3562b76d8b82SJeff Layton } 35631da177e4SLinus Torvalds return error; 35641da177e4SLinus Torvalds } 35651da177e4SLinus Torvalds 3566a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 35675590ff0dSUlrich Drepper { 35685590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 35695590ff0dSUlrich Drepper } 35705590ff0dSUlrich Drepper 35711da177e4SLinus Torvalds /* 3572a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3573c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3574a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3575a71905f0SSage Weil * then we drop the dentry now. 35761da177e4SLinus Torvalds * 35771da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 35781da177e4SLinus Torvalds * do a 35791da177e4SLinus Torvalds * 35801da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 35811da177e4SLinus Torvalds * return -EBUSY; 35821da177e4SLinus Torvalds * 35831da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 35841da177e4SLinus Torvalds * that is still in use by something else.. 35851da177e4SLinus Torvalds */ 35861da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 35871da177e4SLinus Torvalds { 35881da177e4SLinus Torvalds shrink_dcache_parent(dentry); 35891da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 359098474236SWaiman Long if (dentry->d_lockref.count == 1) 35911da177e4SLinus Torvalds __d_drop(dentry); 35921da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 35931da177e4SLinus Torvalds } 35944d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 35951da177e4SLinus Torvalds 35961da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 35971da177e4SLinus Torvalds { 35981da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 35991da177e4SLinus Torvalds 36001da177e4SLinus Torvalds if (error) 36011da177e4SLinus Torvalds return error; 36021da177e4SLinus Torvalds 3603acfa4380SAl Viro if (!dir->i_op->rmdir) 36041da177e4SLinus Torvalds return -EPERM; 36051da177e4SLinus Torvalds 36061d2ef590SAl Viro dget(dentry); 36071b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3608912dbc15SSage Weil 36091da177e4SLinus Torvalds error = -EBUSY; 36107af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3611912dbc15SSage Weil goto out; 3612912dbc15SSage Weil 36131da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3614912dbc15SSage Weil if (error) 3615912dbc15SSage Weil goto out; 3616912dbc15SSage Weil 36173cebde24SSage Weil shrink_dcache_parent(dentry); 36181da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3619912dbc15SSage Weil if (error) 3620912dbc15SSage Weil goto out; 3621912dbc15SSage Weil 36221da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3623d83c49f3SAl Viro dont_mount(dentry); 36248ed936b5SEric W. Biederman detach_mounts(dentry); 36251da177e4SLinus Torvalds 3626912dbc15SSage Weil out: 3627912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36281d2ef590SAl Viro dput(dentry); 3629912dbc15SSage Weil if (!error) 3630912dbc15SSage Weil d_delete(dentry); 36311da177e4SLinus Torvalds return error; 36321da177e4SLinus Torvalds } 36334d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 36341da177e4SLinus Torvalds 36355590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 36361da177e4SLinus Torvalds { 36371da177e4SLinus Torvalds int error = 0; 363891a27b2aSJeff Layton struct filename *name; 36391da177e4SLinus Torvalds struct dentry *dentry; 3640f5beed75SAl Viro struct path path; 3641f5beed75SAl Viro struct qstr last; 3642f5beed75SAl Viro int type; 3643c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3644c6ee9206SJeff Layton retry: 3645f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3646f5beed75SAl Viro &path, &last, &type, lookup_flags); 364791a27b2aSJeff Layton if (IS_ERR(name)) 364891a27b2aSJeff Layton return PTR_ERR(name); 36491da177e4SLinus Torvalds 3650f5beed75SAl Viro switch (type) { 36511da177e4SLinus Torvalds case LAST_DOTDOT: 36521da177e4SLinus Torvalds error = -ENOTEMPTY; 36531da177e4SLinus Torvalds goto exit1; 36541da177e4SLinus Torvalds case LAST_DOT: 36551da177e4SLinus Torvalds error = -EINVAL; 36561da177e4SLinus Torvalds goto exit1; 36571da177e4SLinus Torvalds case LAST_ROOT: 36581da177e4SLinus Torvalds error = -EBUSY; 36591da177e4SLinus Torvalds goto exit1; 36601da177e4SLinus Torvalds } 36610612d9fbSOGAWA Hirofumi 3662f5beed75SAl Viro error = mnt_want_write(path.mnt); 3663c30dabfeSJan Kara if (error) 3664c30dabfeSJan Kara goto exit1; 36650612d9fbSOGAWA Hirofumi 3666f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3667f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 36681da177e4SLinus Torvalds error = PTR_ERR(dentry); 36696902d925SDave Hansen if (IS_ERR(dentry)) 36706902d925SDave Hansen goto exit2; 3671e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3672e6bc45d6STheodore Ts'o error = -ENOENT; 3673e6bc45d6STheodore Ts'o goto exit3; 3674e6bc45d6STheodore Ts'o } 3675f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3676be6d3e56SKentaro Takeda if (error) 3677c30dabfeSJan Kara goto exit3; 3678f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 36790622753bSDave Hansen exit3: 36801da177e4SLinus Torvalds dput(dentry); 36816902d925SDave Hansen exit2: 3682f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3683f5beed75SAl Viro mnt_drop_write(path.mnt); 36841da177e4SLinus Torvalds exit1: 3685f5beed75SAl Viro path_put(&path); 36861da177e4SLinus Torvalds putname(name); 3687c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3688c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3689c6ee9206SJeff Layton goto retry; 3690c6ee9206SJeff Layton } 36911da177e4SLinus Torvalds return error; 36921da177e4SLinus Torvalds } 36931da177e4SLinus Torvalds 36943cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 36955590ff0dSUlrich Drepper { 36965590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 36975590ff0dSUlrich Drepper } 36985590ff0dSUlrich Drepper 3699b21996e3SJ. Bruce Fields /** 3700b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3701b21996e3SJ. Bruce Fields * @dir: parent directory 3702b21996e3SJ. Bruce Fields * @dentry: victim 3703b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3704b21996e3SJ. Bruce Fields * 3705b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3706b21996e3SJ. Bruce Fields * 3707b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3708b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3709b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3710b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3711b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3712b21996e3SJ. Bruce Fields * 3713b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3714b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3715b21996e3SJ. Bruce Fields * to be NFS exported. 3716b21996e3SJ. Bruce Fields */ 3717b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37181da177e4SLinus Torvalds { 37199accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37201da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37211da177e4SLinus Torvalds 37221da177e4SLinus Torvalds if (error) 37231da177e4SLinus Torvalds return error; 37241da177e4SLinus Torvalds 3725acfa4380SAl Viro if (!dir->i_op->unlink) 37261da177e4SLinus Torvalds return -EPERM; 37271da177e4SLinus Torvalds 37289accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37298ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37301da177e4SLinus Torvalds error = -EBUSY; 37311da177e4SLinus Torvalds else { 37321da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3733bec1052eSAl Viro if (!error) { 37345a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 37355a14696cSJ. Bruce Fields if (error) 3736b21996e3SJ. Bruce Fields goto out; 37371da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 37388ed936b5SEric W. Biederman if (!error) { 3739d83c49f3SAl Viro dont_mount(dentry); 37408ed936b5SEric W. Biederman detach_mounts(dentry); 37418ed936b5SEric W. Biederman } 3742bec1052eSAl Viro } 37431da177e4SLinus Torvalds } 3744b21996e3SJ. Bruce Fields out: 37459accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 37461da177e4SLinus Torvalds 37471da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 37481da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 37499accbb97SJ. Bruce Fields fsnotify_link_count(target); 37501da177e4SLinus Torvalds d_delete(dentry); 37511da177e4SLinus Torvalds } 37520eeca283SRobert Love 37531da177e4SLinus Torvalds return error; 37541da177e4SLinus Torvalds } 37554d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 37561da177e4SLinus Torvalds 37571da177e4SLinus Torvalds /* 37581da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 37591b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 37601da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 37611da177e4SLinus Torvalds * while waiting on the I/O. 37621da177e4SLinus Torvalds */ 37635590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 37641da177e4SLinus Torvalds { 37652ad94ae6SAl Viro int error; 376691a27b2aSJeff Layton struct filename *name; 37671da177e4SLinus Torvalds struct dentry *dentry; 3768f5beed75SAl Viro struct path path; 3769f5beed75SAl Viro struct qstr last; 3770f5beed75SAl Viro int type; 37711da177e4SLinus Torvalds struct inode *inode = NULL; 3772b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 37735d18f813SJeff Layton unsigned int lookup_flags = 0; 37745d18f813SJeff Layton retry: 3775f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3776f5beed75SAl Viro &path, &last, &type, lookup_flags); 377791a27b2aSJeff Layton if (IS_ERR(name)) 377891a27b2aSJeff Layton return PTR_ERR(name); 37792ad94ae6SAl Viro 37801da177e4SLinus Torvalds error = -EISDIR; 3781f5beed75SAl Viro if (type != LAST_NORM) 37821da177e4SLinus Torvalds goto exit1; 37830612d9fbSOGAWA Hirofumi 3784f5beed75SAl Viro error = mnt_want_write(path.mnt); 3785c30dabfeSJan Kara if (error) 3786c30dabfeSJan Kara goto exit1; 3787b21996e3SJ. Bruce Fields retry_deleg: 3788f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3789f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37901da177e4SLinus Torvalds error = PTR_ERR(dentry); 37911da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 37921da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3793f5beed75SAl Viro if (last.name[last.len]) 379450338b88STörök Edwin goto slashes; 37951da177e4SLinus Torvalds inode = dentry->d_inode; 3796b18825a7SDavid Howells if (d_is_negative(dentry)) 3797e6bc45d6STheodore Ts'o goto slashes; 37987de9c6eeSAl Viro ihold(inode); 3799f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3800be6d3e56SKentaro Takeda if (error) 3801c30dabfeSJan Kara goto exit2; 3802f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38031da177e4SLinus Torvalds exit2: 38041da177e4SLinus Torvalds dput(dentry); 38051da177e4SLinus Torvalds } 3806f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38071da177e4SLinus Torvalds if (inode) 38081da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3809b21996e3SJ. Bruce Fields inode = NULL; 3810b21996e3SJ. Bruce Fields if (delegated_inode) { 38115a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3812b21996e3SJ. Bruce Fields if (!error) 3813b21996e3SJ. Bruce Fields goto retry_deleg; 3814b21996e3SJ. Bruce Fields } 3815f5beed75SAl Viro mnt_drop_write(path.mnt); 38161da177e4SLinus Torvalds exit1: 3817f5beed75SAl Viro path_put(&path); 38181da177e4SLinus Torvalds putname(name); 38195d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38205d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38215d18f813SJeff Layton inode = NULL; 38225d18f813SJeff Layton goto retry; 38235d18f813SJeff Layton } 38241da177e4SLinus Torvalds return error; 38251da177e4SLinus Torvalds 38261da177e4SLinus Torvalds slashes: 3827b18825a7SDavid Howells if (d_is_negative(dentry)) 3828b18825a7SDavid Howells error = -ENOENT; 382944b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3830b18825a7SDavid Howells error = -EISDIR; 3831b18825a7SDavid Howells else 3832b18825a7SDavid Howells error = -ENOTDIR; 38331da177e4SLinus Torvalds goto exit2; 38341da177e4SLinus Torvalds } 38351da177e4SLinus Torvalds 38362e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 38375590ff0dSUlrich Drepper { 38385590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 38395590ff0dSUlrich Drepper return -EINVAL; 38405590ff0dSUlrich Drepper 38415590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 38425590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 38435590ff0dSUlrich Drepper 38445590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 38455590ff0dSUlrich Drepper } 38465590ff0dSUlrich Drepper 38473480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 38485590ff0dSUlrich Drepper { 38495590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 38505590ff0dSUlrich Drepper } 38515590ff0dSUlrich Drepper 3852db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 38531da177e4SLinus Torvalds { 3854a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38551da177e4SLinus Torvalds 38561da177e4SLinus Torvalds if (error) 38571da177e4SLinus Torvalds return error; 38581da177e4SLinus Torvalds 3859acfa4380SAl Viro if (!dir->i_op->symlink) 38601da177e4SLinus Torvalds return -EPERM; 38611da177e4SLinus Torvalds 38621da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 38631da177e4SLinus Torvalds if (error) 38641da177e4SLinus Torvalds return error; 38651da177e4SLinus Torvalds 38661da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3867a74574aaSStephen Smalley if (!error) 3868f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38691da177e4SLinus Torvalds return error; 38701da177e4SLinus Torvalds } 38714d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 38721da177e4SLinus Torvalds 38732e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 38742e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 38751da177e4SLinus Torvalds { 38762ad94ae6SAl Viro int error; 387791a27b2aSJeff Layton struct filename *from; 38786902d925SDave Hansen struct dentry *dentry; 3879dae6ad8fSAl Viro struct path path; 3880f46d3567SJeff Layton unsigned int lookup_flags = 0; 38811da177e4SLinus Torvalds 38821da177e4SLinus Torvalds from = getname(oldname); 38831da177e4SLinus Torvalds if (IS_ERR(from)) 38841da177e4SLinus Torvalds return PTR_ERR(from); 3885f46d3567SJeff Layton retry: 3886f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 38871da177e4SLinus Torvalds error = PTR_ERR(dentry); 38886902d925SDave Hansen if (IS_ERR(dentry)) 3889dae6ad8fSAl Viro goto out_putname; 38906902d925SDave Hansen 389191a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3892a8104a9fSAl Viro if (!error) 389391a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3894921a1650SAl Viro done_path_create(&path, dentry); 3895f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3896f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3897f46d3567SJeff Layton goto retry; 3898f46d3567SJeff Layton } 38996902d925SDave Hansen out_putname: 39001da177e4SLinus Torvalds putname(from); 39011da177e4SLinus Torvalds return error; 39021da177e4SLinus Torvalds } 39031da177e4SLinus Torvalds 39043480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39055590ff0dSUlrich Drepper { 39065590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39075590ff0dSUlrich Drepper } 39085590ff0dSUlrich Drepper 3909146a8595SJ. Bruce Fields /** 3910146a8595SJ. Bruce Fields * vfs_link - create a new link 3911146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3912146a8595SJ. Bruce Fields * @dir: new parent 3913146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3914146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3915146a8595SJ. Bruce Fields * 3916146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3917146a8595SJ. Bruce Fields * 3918146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3919146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3920146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3921146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3922146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3923146a8595SJ. Bruce Fields * 3924146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3925146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3926146a8595SJ. Bruce Fields * to be NFS exported. 3927146a8595SJ. Bruce Fields */ 3928146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39291da177e4SLinus Torvalds { 39301da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39318de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39321da177e4SLinus Torvalds int error; 39331da177e4SLinus Torvalds 39341da177e4SLinus Torvalds if (!inode) 39351da177e4SLinus Torvalds return -ENOENT; 39361da177e4SLinus Torvalds 3937a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 39381da177e4SLinus Torvalds if (error) 39391da177e4SLinus Torvalds return error; 39401da177e4SLinus Torvalds 39411da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 39421da177e4SLinus Torvalds return -EXDEV; 39431da177e4SLinus Torvalds 39441da177e4SLinus Torvalds /* 39451da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 39461da177e4SLinus Torvalds */ 39471da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 39481da177e4SLinus Torvalds return -EPERM; 3949acfa4380SAl Viro if (!dir->i_op->link) 39501da177e4SLinus Torvalds return -EPERM; 39517e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 39521da177e4SLinus Torvalds return -EPERM; 39531da177e4SLinus Torvalds 39541da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 39551da177e4SLinus Torvalds if (error) 39561da177e4SLinus Torvalds return error; 39571da177e4SLinus Torvalds 39587e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3959aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3960f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3961aae8a97dSAneesh Kumar K.V error = -ENOENT; 39628de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 39638de52778SAl Viro error = -EMLINK; 3964146a8595SJ. Bruce Fields else { 3965146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 3966146a8595SJ. Bruce Fields if (!error) 39671da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 3968146a8595SJ. Bruce Fields } 3969f4e0c30cSAl Viro 3970f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 3971f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3972f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 3973f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3974f4e0c30cSAl Viro } 39757e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3976e31e14ecSStephen Smalley if (!error) 39777e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 39781da177e4SLinus Torvalds return error; 39791da177e4SLinus Torvalds } 39804d359507SAl Viro EXPORT_SYMBOL(vfs_link); 39811da177e4SLinus Torvalds 39821da177e4SLinus Torvalds /* 39831da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 39841da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 39851da177e4SLinus Torvalds * newname. --KAB 39861da177e4SLinus Torvalds * 39871da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 39881da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 39891da177e4SLinus Torvalds * and other special files. --ADM 39901da177e4SLinus Torvalds */ 39912e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 39922e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 39931da177e4SLinus Torvalds { 39941da177e4SLinus Torvalds struct dentry *new_dentry; 3995dae6ad8fSAl Viro struct path old_path, new_path; 3996146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 399711a7b371SAneesh Kumar K.V int how = 0; 39981da177e4SLinus Torvalds int error; 39991da177e4SLinus Torvalds 400011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4001c04030e1SUlrich Drepper return -EINVAL; 400211a7b371SAneesh Kumar K.V /* 4003f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4004f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4005f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 400611a7b371SAneesh Kumar K.V */ 4007f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4008f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4009f0cc6ffbSLinus Torvalds return -ENOENT; 401011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4011f0cc6ffbSLinus Torvalds } 4012c04030e1SUlrich Drepper 401311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 401411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4015442e31caSJeff Layton retry: 401611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40171da177e4SLinus Torvalds if (error) 40182ad94ae6SAl Viro return error; 40192ad94ae6SAl Viro 4020442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4021442e31caSJeff Layton (how & LOOKUP_REVAL)); 40221da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40236902d925SDave Hansen if (IS_ERR(new_dentry)) 4024dae6ad8fSAl Viro goto out; 4025dae6ad8fSAl Viro 4026dae6ad8fSAl Viro error = -EXDEV; 4027dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4028dae6ad8fSAl Viro goto out_dput; 4029800179c9SKees Cook error = may_linkat(&old_path); 4030800179c9SKees Cook if (unlikely(error)) 4031800179c9SKees Cook goto out_dput; 4032dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4033be6d3e56SKentaro Takeda if (error) 4034a8104a9fSAl Viro goto out_dput; 4035146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 403675c3f29dSDave Hansen out_dput: 4037921a1650SAl Viro done_path_create(&new_path, new_dentry); 4038146a8595SJ. Bruce Fields if (delegated_inode) { 4039146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4040d22e6338SOleg Drokin if (!error) { 4041d22e6338SOleg Drokin path_put(&old_path); 4042146a8595SJ. Bruce Fields goto retry; 4043146a8595SJ. Bruce Fields } 4044d22e6338SOleg Drokin } 4045442e31caSJeff Layton if (retry_estale(error, how)) { 4046d22e6338SOleg Drokin path_put(&old_path); 4047442e31caSJeff Layton how |= LOOKUP_REVAL; 4048442e31caSJeff Layton goto retry; 4049442e31caSJeff Layton } 40501da177e4SLinus Torvalds out: 40512d8f3038SAl Viro path_put(&old_path); 40521da177e4SLinus Torvalds 40531da177e4SLinus Torvalds return error; 40541da177e4SLinus Torvalds } 40551da177e4SLinus Torvalds 40563480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 40575590ff0dSUlrich Drepper { 4058c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 40595590ff0dSUlrich Drepper } 40605590ff0dSUlrich Drepper 4061bc27027aSMiklos Szeredi /** 4062bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4063bc27027aSMiklos Szeredi * @old_dir: parent of source 4064bc27027aSMiklos Szeredi * @old_dentry: source 4065bc27027aSMiklos Szeredi * @new_dir: parent of destination 4066bc27027aSMiklos Szeredi * @new_dentry: destination 4067bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4068520c8b16SMiklos Szeredi * @flags: rename flags 4069bc27027aSMiklos Szeredi * 4070bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4071bc27027aSMiklos Szeredi * 4072bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4073bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4074bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4075bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4076bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4077bc27027aSMiklos Szeredi * so. 4078bc27027aSMiklos Szeredi * 4079bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4080bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4081bc27027aSMiklos Szeredi * to be NFS exported. 4082bc27027aSMiklos Szeredi * 40831da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 40841da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 40851da177e4SLinus Torvalds * Problems: 4086d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 40871da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 40881da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4089a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 40901da177e4SLinus Torvalds * story. 40916cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 40926cedba89SJ. Bruce Fields * and source (if it is not a directory). 40931b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 40941da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 40951da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4096a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 40971da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 40981da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 40991da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4100a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41011da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41021da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41031da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4104e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41051b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41061da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4107c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41081da177e4SLinus Torvalds * locking]. 41091da177e4SLinus Torvalds */ 41101da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41118e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4112520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41131da177e4SLinus Torvalds { 41141da177e4SLinus Torvalds int error; 4115bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 411659b0df21SEric Paris const unsigned char *old_name; 4117bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4118bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4119da1ce067SMiklos Szeredi bool new_is_dir = false; 4120da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41211da177e4SLinus Torvalds 4122bc27027aSMiklos Szeredi if (source == target) 41231da177e4SLinus Torvalds return 0; 41241da177e4SLinus Torvalds 41251da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41261da177e4SLinus Torvalds if (error) 41271da177e4SLinus Torvalds return error; 41281da177e4SLinus Torvalds 4129da1ce067SMiklos Szeredi if (!target) { 4130a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4131da1ce067SMiklos Szeredi } else { 4132da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4133da1ce067SMiklos Szeredi 4134da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 41351da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4136da1ce067SMiklos Szeredi else 4137da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4138da1ce067SMiklos Szeredi } 41391da177e4SLinus Torvalds if (error) 41401da177e4SLinus Torvalds return error; 41411da177e4SLinus Torvalds 41427177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 41431da177e4SLinus Torvalds return -EPERM; 41441da177e4SLinus Torvalds 4145520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4146520c8b16SMiklos Szeredi return -EINVAL; 4147520c8b16SMiklos Szeredi 4148bc27027aSMiklos Szeredi /* 4149bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4150bc27027aSMiklos Szeredi * we'll need to flip '..'. 4151bc27027aSMiklos Szeredi */ 4152da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4153da1ce067SMiklos Szeredi if (is_dir) { 4154bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4155bc27027aSMiklos Szeredi if (error) 4156bc27027aSMiklos Szeredi return error; 4157bc27027aSMiklos Szeredi } 4158da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4159da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4160da1ce067SMiklos Szeredi if (error) 4161da1ce067SMiklos Szeredi return error; 4162da1ce067SMiklos Szeredi } 4163da1ce067SMiklos Szeredi } 41640eeca283SRobert Love 41650b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 41660b3974ebSMiklos Szeredi flags); 4167bc27027aSMiklos Szeredi if (error) 4168bc27027aSMiklos Szeredi return error; 4169bc27027aSMiklos Szeredi 4170bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4171bc27027aSMiklos Szeredi dget(new_dentry); 4172da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4173bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4174bc27027aSMiklos Szeredi else if (target) 4175bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4176bc27027aSMiklos Szeredi 4177bc27027aSMiklos Szeredi error = -EBUSY; 41787af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4179bc27027aSMiklos Szeredi goto out; 4180bc27027aSMiklos Szeredi 4181da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4182bc27027aSMiklos Szeredi error = -EMLINK; 4183da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4184bc27027aSMiklos Szeredi goto out; 4185da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4186da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4187da1ce067SMiklos Szeredi goto out; 4188da1ce067SMiklos Szeredi } 4189da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4190bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4191da1ce067SMiklos Szeredi if (!is_dir) { 4192bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4193bc27027aSMiklos Szeredi if (error) 4194bc27027aSMiklos Szeredi goto out; 4195da1ce067SMiklos Szeredi } 4196da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4197bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4198bc27027aSMiklos Szeredi if (error) 4199bc27027aSMiklos Szeredi goto out; 4200bc27027aSMiklos Szeredi } 42017177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4202520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4203520c8b16SMiklos Szeredi new_dir, new_dentry); 4204520c8b16SMiklos Szeredi } else { 42057177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4206520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4207520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4208520c8b16SMiklos Szeredi } 4209bc27027aSMiklos Szeredi if (error) 4210bc27027aSMiklos Szeredi goto out; 4211bc27027aSMiklos Szeredi 4212da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42131da177e4SLinus Torvalds if (is_dir) 4214bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4215bc27027aSMiklos Szeredi dont_mount(new_dentry); 42168ed936b5SEric W. Biederman detach_mounts(new_dentry); 4217bc27027aSMiklos Szeredi } 4218da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4219da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4220bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4221da1ce067SMiklos Szeredi else 4222da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4223da1ce067SMiklos Szeredi } 4224bc27027aSMiklos Szeredi out: 4225da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4226bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4227bc27027aSMiklos Szeredi else if (target) 4228bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4229bc27027aSMiklos Szeredi dput(new_dentry); 4230da1ce067SMiklos Szeredi if (!error) { 4231123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4232da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4233da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4234da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4235da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4236da1ce067SMiklos Szeredi } 4237da1ce067SMiklos Szeredi } 42380eeca283SRobert Love fsnotify_oldname_free(old_name); 42390eeca283SRobert Love 42401da177e4SLinus Torvalds return error; 42411da177e4SLinus Torvalds } 42424d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 42431da177e4SLinus Torvalds 4244520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4245520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 42461da177e4SLinus Torvalds { 42471da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 42481da177e4SLinus Torvalds struct dentry *trap; 4249f5beed75SAl Viro struct path old_path, new_path; 4250f5beed75SAl Viro struct qstr old_last, new_last; 4251f5beed75SAl Viro int old_type, new_type; 42528e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 425391a27b2aSJeff Layton struct filename *from; 425491a27b2aSJeff Layton struct filename *to; 4255f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4256c6a94284SJeff Layton bool should_retry = false; 42572ad94ae6SAl Viro int error; 4258520c8b16SMiklos Szeredi 42590d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4260da1ce067SMiklos Szeredi return -EINVAL; 4261da1ce067SMiklos Szeredi 42620d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 42630d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4264520c8b16SMiklos Szeredi return -EINVAL; 4265520c8b16SMiklos Szeredi 42660d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 42670d7a8555SMiklos Szeredi return -EPERM; 42680d7a8555SMiklos Szeredi 4269f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4270f5beed75SAl Viro target_flags = 0; 4271f5beed75SAl Viro 4272c6a94284SJeff Layton retry: 4273f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4274f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 427591a27b2aSJeff Layton if (IS_ERR(from)) { 427691a27b2aSJeff Layton error = PTR_ERR(from); 42771da177e4SLinus Torvalds goto exit; 427891a27b2aSJeff Layton } 42791da177e4SLinus Torvalds 4280f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4281f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 428291a27b2aSJeff Layton if (IS_ERR(to)) { 428391a27b2aSJeff Layton error = PTR_ERR(to); 42841da177e4SLinus Torvalds goto exit1; 428591a27b2aSJeff Layton } 42861da177e4SLinus Torvalds 42871da177e4SLinus Torvalds error = -EXDEV; 4288f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 42891da177e4SLinus Torvalds goto exit2; 42901da177e4SLinus Torvalds 42911da177e4SLinus Torvalds error = -EBUSY; 4292f5beed75SAl Viro if (old_type != LAST_NORM) 42931da177e4SLinus Torvalds goto exit2; 42941da177e4SLinus Torvalds 42950a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 42960a7c3937SMiklos Szeredi error = -EEXIST; 4297f5beed75SAl Viro if (new_type != LAST_NORM) 42981da177e4SLinus Torvalds goto exit2; 42991da177e4SLinus Torvalds 4300f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4301c30dabfeSJan Kara if (error) 4302c30dabfeSJan Kara goto exit2; 4303c30dabfeSJan Kara 43048e6d782cSJ. Bruce Fields retry_deleg: 4305f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43061da177e4SLinus Torvalds 4307f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43081da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43091da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43101da177e4SLinus Torvalds goto exit3; 43111da177e4SLinus Torvalds /* source must exist */ 43121da177e4SLinus Torvalds error = -ENOENT; 4313b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43141da177e4SLinus Torvalds goto exit4; 4315f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 43161da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43171da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43181da177e4SLinus Torvalds goto exit4; 43190a7c3937SMiklos Szeredi error = -EEXIST; 43200a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43210a7c3937SMiklos Szeredi goto exit5; 4322da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4323da1ce067SMiklos Szeredi error = -ENOENT; 4324da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4325da1ce067SMiklos Szeredi goto exit5; 4326da1ce067SMiklos Szeredi 4327da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4328da1ce067SMiklos Szeredi error = -ENOTDIR; 4329f5beed75SAl Viro if (new_last.name[new_last.len]) 4330da1ce067SMiklos Szeredi goto exit5; 4331da1ce067SMiklos Szeredi } 4332da1ce067SMiklos Szeredi } 43330a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 43340a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 43350a7c3937SMiklos Szeredi error = -ENOTDIR; 4336f5beed75SAl Viro if (old_last.name[old_last.len]) 43370a7c3937SMiklos Szeredi goto exit5; 4338f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 43390a7c3937SMiklos Szeredi goto exit5; 43400a7c3937SMiklos Szeredi } 43410a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 43420a7c3937SMiklos Szeredi error = -EINVAL; 43430a7c3937SMiklos Szeredi if (old_dentry == trap) 43440a7c3937SMiklos Szeredi goto exit5; 43451da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4346da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43471da177e4SLinus Torvalds error = -ENOTEMPTY; 43481da177e4SLinus Torvalds if (new_dentry == trap) 43491da177e4SLinus Torvalds goto exit5; 43501da177e4SLinus Torvalds 4351f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4352f5beed75SAl Viro &new_path, new_dentry, flags); 4353be6d3e56SKentaro Takeda if (error) 4354c30dabfeSJan Kara goto exit5; 4355f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4356f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4357520c8b16SMiklos Szeredi &delegated_inode, flags); 43581da177e4SLinus Torvalds exit5: 43591da177e4SLinus Torvalds dput(new_dentry); 43601da177e4SLinus Torvalds exit4: 43611da177e4SLinus Torvalds dput(old_dentry); 43621da177e4SLinus Torvalds exit3: 4363f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 43648e6d782cSJ. Bruce Fields if (delegated_inode) { 43658e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 43668e6d782cSJ. Bruce Fields if (!error) 43678e6d782cSJ. Bruce Fields goto retry_deleg; 43688e6d782cSJ. Bruce Fields } 4369f5beed75SAl Viro mnt_drop_write(old_path.mnt); 43701da177e4SLinus Torvalds exit2: 4371c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4372c6a94284SJeff Layton should_retry = true; 4373f5beed75SAl Viro path_put(&new_path); 43742ad94ae6SAl Viro putname(to); 43751da177e4SLinus Torvalds exit1: 4376f5beed75SAl Viro path_put(&old_path); 43771da177e4SLinus Torvalds putname(from); 4378c6a94284SJeff Layton if (should_retry) { 4379c6a94284SJeff Layton should_retry = false; 4380c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4381c6a94284SJeff Layton goto retry; 4382c6a94284SJeff Layton } 43832ad94ae6SAl Viro exit: 43841da177e4SLinus Torvalds return error; 43851da177e4SLinus Torvalds } 43861da177e4SLinus Torvalds 4387520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4388520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4389520c8b16SMiklos Szeredi { 4390520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4391520c8b16SMiklos Szeredi } 4392520c8b16SMiklos Szeredi 4393a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 43945590ff0dSUlrich Drepper { 4395520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43965590ff0dSUlrich Drepper } 43975590ff0dSUlrich Drepper 4398787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4399787fb6bcSMiklos Szeredi { 4400787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4401787fb6bcSMiklos Szeredi if (error) 4402787fb6bcSMiklos Szeredi return error; 4403787fb6bcSMiklos Szeredi 4404787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4405787fb6bcSMiklos Szeredi return -EPERM; 4406787fb6bcSMiklos Szeredi 4407787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4408787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4409787fb6bcSMiklos Szeredi } 4410787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4411787fb6bcSMiklos Szeredi 44125d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44131da177e4SLinus Torvalds { 44145d826c84SAl Viro int len = PTR_ERR(link); 44151da177e4SLinus Torvalds if (IS_ERR(link)) 44161da177e4SLinus Torvalds goto out; 44171da177e4SLinus Torvalds 44181da177e4SLinus Torvalds len = strlen(link); 44191da177e4SLinus Torvalds if (len > (unsigned) buflen) 44201da177e4SLinus Torvalds len = buflen; 44211da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44221da177e4SLinus Torvalds len = -EFAULT; 44231da177e4SLinus Torvalds out: 44241da177e4SLinus Torvalds return len; 44251da177e4SLinus Torvalds } 44265d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44271da177e4SLinus Torvalds 44281da177e4SLinus Torvalds /* 44291da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44301da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44311da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 44321da177e4SLinus Torvalds */ 44331da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44341da177e4SLinus Torvalds { 44351da177e4SLinus Torvalds struct nameidata nd; 4436cc314eefSLinus Torvalds void *cookie; 4437694a1764SMarcin Slusarz int res; 4438cc314eefSLinus Torvalds 44391da177e4SLinus Torvalds nd.depth = 0; 4440cc314eefSLinus Torvalds cookie = dentry->d_inode->i_op->follow_link(dentry, &nd); 4441694a1764SMarcin Slusarz if (IS_ERR(cookie)) 4442694a1764SMarcin Slusarz return PTR_ERR(cookie); 4443694a1764SMarcin Slusarz 44445d826c84SAl Viro res = readlink_copy(buffer, buflen, nd_get_link(&nd)); 44451da177e4SLinus Torvalds if (dentry->d_inode->i_op->put_link) 4446cc314eefSLinus Torvalds dentry->d_inode->i_op->put_link(dentry, &nd, cookie); 4447694a1764SMarcin Slusarz return res; 44481da177e4SLinus Torvalds } 44494d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 44501da177e4SLinus Torvalds 44511da177e4SLinus Torvalds /* get the link contents into pagecache */ 44521da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 44531da177e4SLinus Torvalds { 4454ebd09abbSDuane Griffin char *kaddr; 44551da177e4SLinus Torvalds struct page *page; 44561da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4457090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 44581da177e4SLinus Torvalds if (IS_ERR(page)) 44596fe6900eSNick Piggin return (char*)page; 44601da177e4SLinus Torvalds *ppage = page; 4461ebd09abbSDuane Griffin kaddr = kmap(page); 4462ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4463ebd09abbSDuane Griffin return kaddr; 44641da177e4SLinus Torvalds } 44651da177e4SLinus Torvalds 44661da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44671da177e4SLinus Torvalds { 44681da177e4SLinus Torvalds struct page *page = NULL; 44695d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 44701da177e4SLinus Torvalds if (page) { 44711da177e4SLinus Torvalds kunmap(page); 44721da177e4SLinus Torvalds page_cache_release(page); 44731da177e4SLinus Torvalds } 44741da177e4SLinus Torvalds return res; 44751da177e4SLinus Torvalds } 44764d359507SAl Viro EXPORT_SYMBOL(page_readlink); 44771da177e4SLinus Torvalds 4478cc314eefSLinus Torvalds void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd) 44791da177e4SLinus Torvalds { 4480cc314eefSLinus Torvalds struct page *page = NULL; 44811da177e4SLinus Torvalds nd_set_link(nd, page_getlink(dentry, &page)); 4482cc314eefSLinus Torvalds return page; 44831da177e4SLinus Torvalds } 44844d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 44851da177e4SLinus Torvalds 4486cc314eefSLinus Torvalds void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 44871da177e4SLinus Torvalds { 4488cc314eefSLinus Torvalds struct page *page = cookie; 4489cc314eefSLinus Torvalds 4490cc314eefSLinus Torvalds if (page) { 44911da177e4SLinus Torvalds kunmap(page); 44921da177e4SLinus Torvalds page_cache_release(page); 44931da177e4SLinus Torvalds } 44941da177e4SLinus Torvalds } 44954d359507SAl Viro EXPORT_SYMBOL(page_put_link); 44961da177e4SLinus Torvalds 449754566b2cSNick Piggin /* 449854566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 449954566b2cSNick Piggin */ 450054566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45011da177e4SLinus Torvalds { 45021da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45030adb25d2SKirill Korotaev struct page *page; 4504afddba49SNick Piggin void *fsdata; 4505beb497abSDmitriy Monakhov int err; 45061da177e4SLinus Torvalds char *kaddr; 450754566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 450854566b2cSNick Piggin if (nofs) 450954566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45101da177e4SLinus Torvalds 45117e53cac4SNeilBrown retry: 4512afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 451354566b2cSNick Piggin flags, &page, &fsdata); 45141da177e4SLinus Torvalds if (err) 4515afddba49SNick Piggin goto fail; 4516afddba49SNick Piggin 4517e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 45181da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4519e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4520afddba49SNick Piggin 4521afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4522afddba49SNick Piggin page, fsdata); 45231da177e4SLinus Torvalds if (err < 0) 45241da177e4SLinus Torvalds goto fail; 4525afddba49SNick Piggin if (err < len-1) 4526afddba49SNick Piggin goto retry; 4527afddba49SNick Piggin 45281da177e4SLinus Torvalds mark_inode_dirty(inode); 45291da177e4SLinus Torvalds return 0; 45301da177e4SLinus Torvalds fail: 45311da177e4SLinus Torvalds return err; 45321da177e4SLinus Torvalds } 45334d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 45341da177e4SLinus Torvalds 45350adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 45360adb25d2SKirill Korotaev { 45370adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 453854566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 45390adb25d2SKirill Korotaev } 45404d359507SAl Viro EXPORT_SYMBOL(page_symlink); 45410adb25d2SKirill Korotaev 454292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 45431da177e4SLinus Torvalds .readlink = generic_readlink, 45441da177e4SLinus Torvalds .follow_link = page_follow_link_light, 45451da177e4SLinus Torvalds .put_link = page_put_link, 45461da177e4SLinus Torvalds }; 45471da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4548