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 495894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4961f55a6ecSAl Viro struct nameidata { 4971f55a6ecSAl Viro struct path path; 4981f55a6ecSAl Viro struct qstr last; 4991f55a6ecSAl Viro struct path root; 5001f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5011f55a6ecSAl Viro unsigned int flags; 5021f55a6ecSAl Viro unsigned seq, m_seq; 5031f55a6ecSAl Viro int last_type; 5041f55a6ecSAl Viro unsigned depth; 505756daf26SNeilBrown int total_link_count; 506697fc6caSAl Viro struct saved { 507697fc6caSAl Viro struct path link; 508697fc6caSAl Viro void *cookie; 509697fc6caSAl Viro const char *name; 510237d8b32SAl Viro struct inode *inode; 511894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5121f55a6ecSAl Viro }; 5131f55a6ecSAl Viro 514756daf26SNeilBrown static struct nameidata *set_nameidata(struct nameidata *p) 515894bc8c4SAl Viro { 516756daf26SNeilBrown struct nameidata *old = current->nameidata; 517756daf26SNeilBrown p->stack = p->internal; 518756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 519756daf26SNeilBrown current->nameidata = p; 520756daf26SNeilBrown return old; 521894bc8c4SAl Viro } 522894bc8c4SAl Viro 523756daf26SNeilBrown static void restore_nameidata(struct nameidata *old) 524894bc8c4SAl Viro { 525756daf26SNeilBrown struct nameidata *now = current->nameidata; 526756daf26SNeilBrown 527756daf26SNeilBrown current->nameidata = old; 528756daf26SNeilBrown if (old) 529756daf26SNeilBrown old->total_link_count = now->total_link_count; 530756daf26SNeilBrown if (now->stack != now->internal) { 531756daf26SNeilBrown kfree(now->stack); 532756daf26SNeilBrown now->stack = now->internal; 533894bc8c4SAl Viro } 534894bc8c4SAl Viro } 535894bc8c4SAl Viro 536894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 537894bc8c4SAl Viro { 538e269f2a7SAl Viro struct saved *p = kmalloc(MAXSYMLINKS * sizeof(struct saved), 539894bc8c4SAl Viro GFP_KERNEL); 540894bc8c4SAl Viro if (unlikely(!p)) 541894bc8c4SAl Viro return -ENOMEM; 542894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 543894bc8c4SAl Viro nd->stack = p; 544894bc8c4SAl Viro return 0; 545894bc8c4SAl Viro } 546894bc8c4SAl Viro 547894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 548894bc8c4SAl Viro { 549da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 550894bc8c4SAl Viro return 0; 551894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 552894bc8c4SAl Viro return 0; 553894bc8c4SAl Viro return __nd_alloc_stack(nd); 554894bc8c4SAl Viro } 555894bc8c4SAl Viro 55619660af7SAl Viro /* 55731e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 55819660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 55919660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 56019660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 56119660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 56219660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 56319660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 56419660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 56531e6b01fSNick Piggin */ 56631e6b01fSNick Piggin 56731e6b01fSNick Piggin /** 56819660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 56919660af7SAl Viro * @nd: nameidata pathwalk data 57019660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 5716e9918b7SAl Viro * @seq: seq number to check dentry against 57239191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 57331e6b01fSNick Piggin * 57419660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 57519660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 57619660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 57731e6b01fSNick Piggin */ 5786e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 57931e6b01fSNick Piggin { 58031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 58131e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 58231e6b01fSNick Piggin 58331e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 584e5c832d5SLinus Torvalds 585e5c832d5SLinus Torvalds /* 58648a066e7SAl Viro * After legitimizing the bastards, terminate_walk() 58748a066e7SAl Viro * will do the right thing for non-RCU mode, and all our 58848a066e7SAl Viro * subsequent exit cases should rcu_read_unlock() 58948a066e7SAl Viro * before returning. Do vfsmount first; if dentry 59048a066e7SAl Viro * can't be legitimized, just set nd->path.dentry to NULL 59148a066e7SAl Viro * and rely on dput(NULL) being a no-op. 592e5c832d5SLinus Torvalds */ 59348a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 594e5c832d5SLinus Torvalds return -ECHILD; 595e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 59615570086SLinus Torvalds 59748a066e7SAl Viro if (!lockref_get_not_dead(&parent->d_lockref)) { 59848a066e7SAl Viro nd->path.dentry = NULL; 599d870b4a1SAl Viro goto out; 60048a066e7SAl Viro } 60148a066e7SAl Viro 60215570086SLinus Torvalds /* 60315570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 60415570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 60515570086SLinus Torvalds * 60615570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 60715570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 60815570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 60915570086SLinus Torvalds * since we checked the sequence number of the parent after we got 61015570086SLinus Torvalds * the child sequence number. So we know the parent must still 61115570086SLinus Torvalds * be valid if the child sequence number is still valid. 61215570086SLinus Torvalds */ 61319660af7SAl Viro if (!dentry) { 614e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 615e5c832d5SLinus Torvalds goto out; 61619660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 61719660af7SAl Viro } else { 618e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 619e5c832d5SLinus Torvalds goto out; 6206e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 621e5c832d5SLinus Torvalds goto drop_dentry; 62219660af7SAl Viro } 623e5c832d5SLinus Torvalds 624e5c832d5SLinus Torvalds /* 625e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 626e5c832d5SLinus Torvalds * still valid and get it if required. 627e5c832d5SLinus Torvalds */ 628e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 629e5c832d5SLinus Torvalds spin_lock(&fs->lock); 630e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 631e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 63231e6b01fSNick Piggin path_get(&nd->root); 63331e6b01fSNick Piggin spin_unlock(&fs->lock); 63431e6b01fSNick Piggin } 63531e6b01fSNick Piggin 6368b61e74fSAl Viro rcu_read_unlock(); 63731e6b01fSNick Piggin return 0; 63819660af7SAl Viro 639e5c832d5SLinus Torvalds unlock_and_drop_dentry: 64031e6b01fSNick Piggin spin_unlock(&fs->lock); 641e5c832d5SLinus Torvalds drop_dentry: 6428b61e74fSAl Viro rcu_read_unlock(); 643e5c832d5SLinus Torvalds dput(dentry); 644d0d27277SLinus Torvalds goto drop_root_mnt; 645e5c832d5SLinus Torvalds out: 6468b61e74fSAl Viro rcu_read_unlock(); 647d0d27277SLinus Torvalds drop_root_mnt: 648d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 649d0d27277SLinus Torvalds nd->root.mnt = NULL; 65031e6b01fSNick Piggin return -ECHILD; 65131e6b01fSNick Piggin } 65231e6b01fSNick Piggin 6534ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 65434286d66SNick Piggin { 6554ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 65634286d66SNick Piggin } 65734286d66SNick Piggin 6589f1fafeeSAl Viro /** 6599f1fafeeSAl Viro * complete_walk - successful completion of path walk 6609f1fafeeSAl Viro * @nd: pointer nameidata 66139159de2SJeff Layton * 6629f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 6639f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 6649f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 6659f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 6669f1fafeeSAl Viro * need to drop nd->path. 66739159de2SJeff Layton */ 6689f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 66939159de2SJeff Layton { 67016c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 67139159de2SJeff Layton int status; 67239159de2SJeff Layton 6739f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 6749f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6759f1fafeeSAl Viro nd->root.mnt = NULL; 6766e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 67748a066e7SAl Viro return -ECHILD; 67848a066e7SAl Viro } 6799f1fafeeSAl Viro 68016c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 68139159de2SJeff Layton return 0; 68239159de2SJeff Layton 683ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 68416c2cd71SAl Viro return 0; 68516c2cd71SAl Viro 686ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 68739159de2SJeff Layton if (status > 0) 68839159de2SJeff Layton return 0; 68939159de2SJeff Layton 69016c2cd71SAl Viro if (!status) 69139159de2SJeff Layton status = -ESTALE; 69216c2cd71SAl Viro 69339159de2SJeff Layton return status; 69439159de2SJeff Layton } 69539159de2SJeff Layton 6962a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 6972a737871SAl Viro { 698f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 6992a737871SAl Viro } 7002a737871SAl Viro 7017bd88377SAl Viro static __always_inline unsigned set_root_rcu(struct nameidata *nd) 70231e6b01fSNick Piggin { 70331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 7047bd88377SAl Viro unsigned seq, res; 705c28cc364SNick Piggin 706c28cc364SNick Piggin do { 707c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 70831e6b01fSNick Piggin nd->root = fs->root; 7097bd88377SAl Viro res = __read_seqcount_begin(&nd->root.dentry->d_seq); 710c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 7117bd88377SAl Viro return res; 71231e6b01fSNick Piggin } 71331e6b01fSNick Piggin 7141d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 715051d3812SIan Kent { 716051d3812SIan Kent dput(path->dentry); 7174ac91378SJan Blunck if (path->mnt != nd->path.mnt) 718051d3812SIan Kent mntput(path->mnt); 719051d3812SIan Kent } 720051d3812SIan Kent 7217b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 7227b9337aaSNick Piggin struct nameidata *nd) 723051d3812SIan Kent { 72431e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 7254ac91378SJan Blunck dput(nd->path.dentry); 72631e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 7274ac91378SJan Blunck mntput(nd->path.mnt); 7289a229683SHuang Shijie } 72931e6b01fSNick Piggin nd->path.mnt = path->mnt; 7304ac91378SJan Blunck nd->path.dentry = path->dentry; 731051d3812SIan Kent } 732051d3812SIan Kent 733b5fb63c1SChristoph Hellwig /* 734b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 735b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 736b5fb63c1SChristoph Hellwig */ 7376e77137bSAl Viro void nd_jump_link(struct path *path) 738b5fb63c1SChristoph Hellwig { 7396e77137bSAl Viro struct nameidata *nd = current->nameidata; 740b5fb63c1SChristoph Hellwig path_put(&nd->path); 741b5fb63c1SChristoph Hellwig 742b5fb63c1SChristoph Hellwig nd->path = *path; 743b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 744b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 745b5fb63c1SChristoph Hellwig } 746b5fb63c1SChristoph Hellwig 747b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 748574197e0SAl Viro { 74921c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 750237d8b32SAl Viro struct inode *inode = last->inode; 751b9ff4429SAl Viro if (last->cookie && inode->i_op->put_link) 752b9ff4429SAl Viro inode->i_op->put_link(last->link.dentry, last->cookie); 753b9ff4429SAl Viro path_put(&last->link); 754574197e0SAl Viro } 755574197e0SAl Viro 756561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 757561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 758800179c9SKees Cook 759800179c9SKees Cook /** 760800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 76155852635SRandy Dunlap * @nd: nameidata pathwalk data 762800179c9SKees Cook * 763800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 764800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 765800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 766800179c9SKees Cook * processes from failing races against path names that may change out 767800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 768800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 769800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 770800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 771800179c9SKees Cook * 772800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 773800179c9SKees Cook */ 774fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 775800179c9SKees Cook { 776800179c9SKees Cook const struct inode *inode; 777800179c9SKees Cook const struct inode *parent; 778800179c9SKees Cook 779800179c9SKees Cook if (!sysctl_protected_symlinks) 780800179c9SKees Cook return 0; 781800179c9SKees Cook 782800179c9SKees Cook /* Allowed if owner and follower match. */ 783237d8b32SAl Viro inode = nd->stack[0].inode; 78481abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 785800179c9SKees Cook return 0; 786800179c9SKees Cook 787800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 788800179c9SKees Cook parent = nd->path.dentry->d_inode; 789800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 790800179c9SKees Cook return 0; 791800179c9SKees Cook 792800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 79381abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 794800179c9SKees Cook return 0; 795800179c9SKees Cook 7961cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 797800179c9SKees Cook return -EACCES; 798800179c9SKees Cook } 799800179c9SKees Cook 800800179c9SKees Cook /** 801800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 802800179c9SKees Cook * @inode: the source inode to hardlink from 803800179c9SKees Cook * 804800179c9SKees Cook * Return false if at least one of the following conditions: 805800179c9SKees Cook * - inode is not a regular file 806800179c9SKees Cook * - inode is setuid 807800179c9SKees Cook * - inode is setgid and group-exec 808800179c9SKees Cook * - access failure for read and write 809800179c9SKees Cook * 810800179c9SKees Cook * Otherwise returns true. 811800179c9SKees Cook */ 812800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 813800179c9SKees Cook { 814800179c9SKees Cook umode_t mode = inode->i_mode; 815800179c9SKees Cook 816800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 817800179c9SKees Cook if (!S_ISREG(mode)) 818800179c9SKees Cook return false; 819800179c9SKees Cook 820800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 821800179c9SKees Cook if (mode & S_ISUID) 822800179c9SKees Cook return false; 823800179c9SKees Cook 824800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 825800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 826800179c9SKees Cook return false; 827800179c9SKees Cook 828800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 829800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 830800179c9SKees Cook return false; 831800179c9SKees Cook 832800179c9SKees Cook return true; 833800179c9SKees Cook } 834800179c9SKees Cook 835800179c9SKees Cook /** 836800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 837800179c9SKees Cook * @link: the source to hardlink from 838800179c9SKees Cook * 839800179c9SKees Cook * Block hardlink when all of: 840800179c9SKees Cook * - sysctl_protected_hardlinks enabled 841800179c9SKees Cook * - fsuid does not match inode 842800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 843800179c9SKees Cook * - not CAP_FOWNER 844800179c9SKees Cook * 845800179c9SKees Cook * Returns 0 if successful, -ve on error. 846800179c9SKees Cook */ 847800179c9SKees Cook static int may_linkat(struct path *link) 848800179c9SKees Cook { 849800179c9SKees Cook const struct cred *cred; 850800179c9SKees Cook struct inode *inode; 851800179c9SKees Cook 852800179c9SKees Cook if (!sysctl_protected_hardlinks) 853800179c9SKees Cook return 0; 854800179c9SKees Cook 855800179c9SKees Cook cred = current_cred(); 856800179c9SKees Cook inode = link->dentry->d_inode; 857800179c9SKees Cook 858800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 859800179c9SKees Cook * otherwise, it must be a safe source. 860800179c9SKees Cook */ 86181abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 862800179c9SKees Cook capable(CAP_FOWNER)) 863800179c9SKees Cook return 0; 864800179c9SKees Cook 865a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 866800179c9SKees Cook return -EPERM; 867800179c9SKees Cook } 868800179c9SKees Cook 8693b2e7f75SAl Viro static __always_inline 8703b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 8711da177e4SLinus Torvalds { 872ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 8731cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 874237d8b32SAl Viro struct inode *inode = last->inode; 8756d7b5aaeSAl Viro int error; 8760a959df5SAl Viro const char *res; 8771da177e4SLinus Torvalds 878844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 879844a3917SAl Viro 880574197e0SAl Viro cond_resched(); 881574197e0SAl Viro 8823b2e7f75SAl Viro touch_atime(&last->link); 883cd4e91d3SAl Viro 88437882db0SNeilBrown error = security_inode_follow_link(dentry); 8856d7b5aaeSAl Viro if (error) 8866920a440SAl Viro return ERR_PTR(error); 88736f3b4f6SAl Viro 88886acdca1SAl Viro nd->last_type = LAST_BIND; 889d4dee48bSAl Viro res = inode->i_link; 890d4dee48bSAl Viro if (!res) { 8916e77137bSAl Viro res = inode->i_op->follow_link(dentry, &last->cookie); 892fab51e8aSAl Viro if (IS_ERR_OR_NULL(res)) { 8936920a440SAl Viro last->cookie = NULL; 894fab51e8aSAl Viro return res; 8950a959df5SAl Viro } 896fab51e8aSAl Viro } 897fab51e8aSAl Viro if (*res == '/') { 898fab51e8aSAl Viro if (!nd->root.mnt) 899fab51e8aSAl Viro set_root(nd); 900fab51e8aSAl Viro path_put(&nd->path); 901fab51e8aSAl Viro nd->path = nd->root; 902fab51e8aSAl Viro path_get(&nd->root); 903fab51e8aSAl Viro nd->inode = nd->path.dentry->d_inode; 904fab51e8aSAl Viro nd->flags |= LOOKUP_JUMPED; 905fab51e8aSAl Viro while (unlikely(*++res == '/')) 906fab51e8aSAl Viro ; 907fab51e8aSAl Viro } 908fab51e8aSAl Viro if (!*res) 909fab51e8aSAl Viro res = NULL; 9100a959df5SAl Viro return res; 9110a959df5SAl Viro } 9126d7b5aaeSAl Viro 91331e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 91431e6b01fSNick Piggin { 9150714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9160714a533SAl Viro struct mount *parent; 91731e6b01fSNick Piggin struct dentry *mountpoint; 91831e6b01fSNick Piggin 9190714a533SAl Viro parent = mnt->mnt_parent; 9200714a533SAl Viro if (&parent->mnt == path->mnt) 92131e6b01fSNick Piggin return 0; 922a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 92331e6b01fSNick Piggin path->dentry = mountpoint; 9240714a533SAl Viro path->mnt = &parent->mnt; 92531e6b01fSNick Piggin return 1; 92631e6b01fSNick Piggin } 92731e6b01fSNick Piggin 928f015f126SDavid Howells /* 929f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 930f015f126SDavid Howells * 931f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 932f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 933f015f126SDavid Howells * Up is towards /. 934f015f126SDavid Howells * 935f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 936f015f126SDavid Howells * root. 937f015f126SDavid Howells */ 938bab77ebfSAl Viro int follow_up(struct path *path) 9391da177e4SLinus Torvalds { 9400714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9410714a533SAl Viro struct mount *parent; 9421da177e4SLinus Torvalds struct dentry *mountpoint; 94399b7db7bSNick Piggin 94448a066e7SAl Viro read_seqlock_excl(&mount_lock); 9450714a533SAl Viro parent = mnt->mnt_parent; 9463c0a6163SAl Viro if (parent == mnt) { 94748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 9481da177e4SLinus Torvalds return 0; 9491da177e4SLinus Torvalds } 9500714a533SAl Viro mntget(&parent->mnt); 951a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 95248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 953bab77ebfSAl Viro dput(path->dentry); 954bab77ebfSAl Viro path->dentry = mountpoint; 955bab77ebfSAl Viro mntput(path->mnt); 9560714a533SAl Viro path->mnt = &parent->mnt; 9571da177e4SLinus Torvalds return 1; 9581da177e4SLinus Torvalds } 9594d359507SAl Viro EXPORT_SYMBOL(follow_up); 9601da177e4SLinus Torvalds 961b5c84bf6SNick Piggin /* 9629875cf80SDavid Howells * Perform an automount 9639875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9649875cf80SDavid Howells * were called with. 9651da177e4SLinus Torvalds */ 966756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 9679875cf80SDavid Howells bool *need_mntput) 96831e6b01fSNick Piggin { 9699875cf80SDavid Howells struct vfsmount *mnt; 970ea5b778aSDavid Howells int err; 9719875cf80SDavid Howells 9729875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9739875cf80SDavid Howells return -EREMOTE; 9749875cf80SDavid Howells 9750ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9760ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9770ec26fd0SMiklos Szeredi * the name. 9780ec26fd0SMiklos Szeredi * 9790ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9805a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9810ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9820ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9830ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9840ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9855a30d8a2SDavid Howells */ 986756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 987d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 9885a30d8a2SDavid Howells path->dentry->d_inode) 9899875cf80SDavid Howells return -EISDIR; 9900ec26fd0SMiklos Szeredi 991756daf26SNeilBrown nd->total_link_count++; 992756daf26SNeilBrown if (nd->total_link_count >= 40) 9939875cf80SDavid Howells return -ELOOP; 9949875cf80SDavid Howells 9959875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 9969875cf80SDavid Howells if (IS_ERR(mnt)) { 9979875cf80SDavid Howells /* 9989875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 9999875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 10009875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 10019875cf80SDavid Howells * 10029875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 10039875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 10049875cf80SDavid Howells * the path is inaccessible and we should say so. 10059875cf80SDavid Howells */ 1006756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 10079875cf80SDavid Howells return -EREMOTE; 10089875cf80SDavid Howells return PTR_ERR(mnt); 100931e6b01fSNick Piggin } 1010ea5b778aSDavid Howells 10119875cf80SDavid Howells if (!mnt) /* mount collision */ 10129875cf80SDavid Howells return 0; 10139875cf80SDavid Howells 10148aef1884SAl Viro if (!*need_mntput) { 10158aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 10168aef1884SAl Viro mntget(path->mnt); 10178aef1884SAl Viro *need_mntput = true; 10188aef1884SAl Viro } 101919a167afSAl Viro err = finish_automount(mnt, path); 1020ea5b778aSDavid Howells 1021ea5b778aSDavid Howells switch (err) { 1022ea5b778aSDavid Howells case -EBUSY: 1023ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 102419a167afSAl Viro return 0; 1025ea5b778aSDavid Howells case 0: 10268aef1884SAl Viro path_put(path); 10279875cf80SDavid Howells path->mnt = mnt; 10289875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 10299875cf80SDavid Howells return 0; 103019a167afSAl Viro default: 103119a167afSAl Viro return err; 10329875cf80SDavid Howells } 103319a167afSAl Viro 1034ea5b778aSDavid Howells } 10359875cf80SDavid Howells 10369875cf80SDavid Howells /* 10379875cf80SDavid Howells * Handle a dentry that is managed in some way. 1038cc53ce53SDavid Howells * - Flagged for transit management (autofs) 10399875cf80SDavid Howells * - Flagged as mountpoint 10409875cf80SDavid Howells * - Flagged as automount point 10419875cf80SDavid Howells * 10429875cf80SDavid Howells * This may only be called in refwalk mode. 10439875cf80SDavid Howells * 10449875cf80SDavid Howells * Serialization is taken care of in namespace.c 10459875cf80SDavid Howells */ 1046756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 10479875cf80SDavid Howells { 10488aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10499875cf80SDavid Howells unsigned managed; 10509875cf80SDavid Howells bool need_mntput = false; 10518aef1884SAl Viro int ret = 0; 10529875cf80SDavid Howells 10539875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10549875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10559875cf80SDavid Howells * the components of that value change under us */ 10569875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10579875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10589875cf80SDavid Howells unlikely(managed != 0)) { 1059cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1060cc53ce53SDavid Howells * being held. */ 1061cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1062cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1063cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10641aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1065cc53ce53SDavid Howells if (ret < 0) 10668aef1884SAl Viro break; 1067cc53ce53SDavid Howells } 1068cc53ce53SDavid Howells 10699875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10709875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10719875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10729875cf80SDavid Howells if (mounted) { 10739875cf80SDavid Howells dput(path->dentry); 10749875cf80SDavid Howells if (need_mntput) 1075463ffb2eSAl Viro mntput(path->mnt); 1076463ffb2eSAl Viro path->mnt = mounted; 1077463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10789875cf80SDavid Howells need_mntput = true; 10799875cf80SDavid Howells continue; 1080463ffb2eSAl Viro } 1081463ffb2eSAl Viro 10829875cf80SDavid Howells /* Something is mounted on this dentry in another 10839875cf80SDavid Howells * namespace and/or whatever was mounted there in this 108448a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 108548a066e7SAl Viro * get it */ 10861da177e4SLinus Torvalds } 10879875cf80SDavid Howells 10889875cf80SDavid Howells /* Handle an automount point */ 10899875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1090756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 10919875cf80SDavid Howells if (ret < 0) 10928aef1884SAl Viro break; 10939875cf80SDavid Howells continue; 10949875cf80SDavid Howells } 10959875cf80SDavid Howells 10969875cf80SDavid Howells /* We didn't change the current path point */ 10979875cf80SDavid Howells break; 10989875cf80SDavid Howells } 10998aef1884SAl Viro 11008aef1884SAl Viro if (need_mntput && path->mnt == mnt) 11018aef1884SAl Viro mntput(path->mnt); 11028aef1884SAl Viro if (ret == -EISDIR) 11038aef1884SAl Viro ret = 0; 11048402752eSAl Viro if (need_mntput) 11058402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 11068402752eSAl Viro if (unlikely(ret < 0)) 11078402752eSAl Viro path_put_conditional(path, nd); 11088402752eSAl Viro return ret; 11091da177e4SLinus Torvalds } 11101da177e4SLinus Torvalds 1111cc53ce53SDavid Howells int follow_down_one(struct path *path) 11121da177e4SLinus Torvalds { 11131da177e4SLinus Torvalds struct vfsmount *mounted; 11141da177e4SLinus Torvalds 11151c755af4SAl Viro mounted = lookup_mnt(path); 11161da177e4SLinus Torvalds if (mounted) { 11179393bd07SAl Viro dput(path->dentry); 11189393bd07SAl Viro mntput(path->mnt); 11199393bd07SAl Viro path->mnt = mounted; 11209393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 11211da177e4SLinus Torvalds return 1; 11221da177e4SLinus Torvalds } 11231da177e4SLinus Torvalds return 0; 11241da177e4SLinus Torvalds } 11254d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 11261da177e4SLinus Torvalds 1127b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 112862a7375eSIan Kent { 1129b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1130b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 113162a7375eSIan Kent } 113262a7375eSIan Kent 11339875cf80SDavid Howells /* 1134287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1135287548e4SAl Viro * we meet a managed dentry that would need blocking. 11369875cf80SDavid Howells */ 11379875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1138254cf582SAl Viro struct inode **inode, unsigned *seqp) 11399875cf80SDavid Howells { 114062a7375eSIan Kent for (;;) { 1141c7105365SAl Viro struct mount *mounted; 114262a7375eSIan Kent /* 114362a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 114462a7375eSIan Kent * that wants to block transit. 114562a7375eSIan Kent */ 1146b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1147b8faf035SNeilBrown case -ECHILD: 1148b8faf035SNeilBrown default: 1149ab90911fSDavid Howells return false; 1150b8faf035SNeilBrown case -EISDIR: 1151b8faf035SNeilBrown return true; 1152b8faf035SNeilBrown case 0: 1153b8faf035SNeilBrown break; 1154b8faf035SNeilBrown } 115562a7375eSIan Kent 115662a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1157b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 115862a7375eSIan Kent 1159474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 11609875cf80SDavid Howells if (!mounted) 11619875cf80SDavid Howells break; 1162c7105365SAl Viro path->mnt = &mounted->mnt; 1163c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1164a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1165254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 116659430262SLinus Torvalds /* 116759430262SLinus Torvalds * Update the inode too. We don't need to re-check the 116859430262SLinus Torvalds * dentry sequence number here after this d_inode read, 116959430262SLinus Torvalds * because a mount-point is always pinned. 117059430262SLinus Torvalds */ 117159430262SLinus Torvalds *inode = path->dentry->d_inode; 11729875cf80SDavid Howells } 1173f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1174b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1175287548e4SAl Viro } 1176287548e4SAl Viro 117731e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 117831e6b01fSNick Piggin { 11794023bfc9SAl Viro struct inode *inode = nd->inode; 11807bd88377SAl Viro if (!nd->root.mnt) 118131e6b01fSNick Piggin set_root_rcu(nd); 118231e6b01fSNick Piggin 118331e6b01fSNick Piggin while (1) { 118431e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 118531e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 118631e6b01fSNick Piggin break; 118731e6b01fSNick Piggin } 118831e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 118931e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 119031e6b01fSNick Piggin struct dentry *parent = old->d_parent; 119131e6b01fSNick Piggin unsigned seq; 119231e6b01fSNick Piggin 11934023bfc9SAl Viro inode = parent->d_inode; 119431e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 119531e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1196ef7562d5SAl Viro goto failed; 119731e6b01fSNick Piggin nd->path.dentry = parent; 119831e6b01fSNick Piggin nd->seq = seq; 119931e6b01fSNick Piggin break; 120031e6b01fSNick Piggin } 120131e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 120231e6b01fSNick Piggin break; 12034023bfc9SAl Viro inode = nd->path.dentry->d_inode; 120431e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 120531e6b01fSNick Piggin } 1206b37199e6SAl Viro while (d_mountpoint(nd->path.dentry)) { 1207b37199e6SAl Viro struct mount *mounted; 1208b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1209b37199e6SAl Viro if (!mounted) 1210b37199e6SAl Viro break; 1211b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1212b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 12134023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1214b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1215f5be3e29SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1216b37199e6SAl Viro goto failed; 1217b37199e6SAl Viro } 12184023bfc9SAl Viro nd->inode = inode; 121931e6b01fSNick Piggin return 0; 1220ef7562d5SAl Viro 1221ef7562d5SAl Viro failed: 1222ef7562d5SAl Viro return -ECHILD; 122331e6b01fSNick Piggin } 122431e6b01fSNick Piggin 12259875cf80SDavid Howells /* 1226cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1227cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1228cc53ce53SDavid Howells * caller is permitted to proceed or not. 1229cc53ce53SDavid Howells */ 12307cc90cc3SAl Viro int follow_down(struct path *path) 1231cc53ce53SDavid Howells { 1232cc53ce53SDavid Howells unsigned managed; 1233cc53ce53SDavid Howells int ret; 1234cc53ce53SDavid Howells 1235cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1236cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1237cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1238cc53ce53SDavid Howells * being held. 1239cc53ce53SDavid Howells * 1240cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1241cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1242cc53ce53SDavid Howells * other than its daemon the right to mount on its 1243cc53ce53SDavid Howells * superstructure. 1244cc53ce53SDavid Howells * 1245cc53ce53SDavid Howells * The filesystem may sleep at this point. 1246cc53ce53SDavid Howells */ 1247cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1248cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1249cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1250ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 12511aed3e42SAl Viro path->dentry, false); 1252cc53ce53SDavid Howells if (ret < 0) 1253cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1254cc53ce53SDavid Howells } 1255cc53ce53SDavid Howells 1256cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1257cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1258cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1259cc53ce53SDavid Howells if (!mounted) 1260cc53ce53SDavid Howells break; 1261cc53ce53SDavid Howells dput(path->dentry); 1262cc53ce53SDavid Howells mntput(path->mnt); 1263cc53ce53SDavid Howells path->mnt = mounted; 1264cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1265cc53ce53SDavid Howells continue; 1266cc53ce53SDavid Howells } 1267cc53ce53SDavid Howells 1268cc53ce53SDavid Howells /* Don't handle automount points here */ 1269cc53ce53SDavid Howells break; 1270cc53ce53SDavid Howells } 1271cc53ce53SDavid Howells return 0; 1272cc53ce53SDavid Howells } 12734d359507SAl Viro EXPORT_SYMBOL(follow_down); 1274cc53ce53SDavid Howells 1275cc53ce53SDavid Howells /* 12769875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12779875cf80SDavid Howells */ 12789875cf80SDavid Howells static void follow_mount(struct path *path) 12799875cf80SDavid Howells { 12809875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12819875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12829875cf80SDavid Howells if (!mounted) 12839875cf80SDavid Howells break; 12849875cf80SDavid Howells dput(path->dentry); 12859875cf80SDavid Howells mntput(path->mnt); 12869875cf80SDavid Howells path->mnt = mounted; 12879875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12889875cf80SDavid Howells } 12899875cf80SDavid Howells } 12909875cf80SDavid Howells 129131e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 12921da177e4SLinus Torvalds { 12937bd88377SAl Viro if (!nd->root.mnt) 12942a737871SAl Viro set_root(nd); 1295e518ddb7SAndreas Mohr 12961da177e4SLinus Torvalds while(1) { 12974ac91378SJan Blunck struct dentry *old = nd->path.dentry; 12981da177e4SLinus Torvalds 12992a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 13002a737871SAl Viro nd->path.mnt == nd->root.mnt) { 13011da177e4SLinus Torvalds break; 13021da177e4SLinus Torvalds } 13034ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 13043088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 13053088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 13061da177e4SLinus Torvalds dput(old); 13071da177e4SLinus Torvalds break; 13081da177e4SLinus Torvalds } 13093088dd70SAl Viro if (!follow_up(&nd->path)) 13101da177e4SLinus Torvalds break; 13111da177e4SLinus Torvalds } 131279ed0226SAl Viro follow_mount(&nd->path); 131331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 13141da177e4SLinus Torvalds } 13151da177e4SLinus Torvalds 13161da177e4SLinus Torvalds /* 1317bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1318bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1319bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1320bad61189SMiklos Szeredi * 1321bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1322baa03890SNick Piggin */ 1323bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1324201f956eSAl Viro unsigned int flags, bool *need_lookup) 1325baa03890SNick Piggin { 1326baa03890SNick Piggin struct dentry *dentry; 1327bad61189SMiklos Szeredi int error; 1328baa03890SNick Piggin 1329bad61189SMiklos Szeredi *need_lookup = false; 1330bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1331bad61189SMiklos Szeredi if (dentry) { 133239e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1333201f956eSAl Viro error = d_revalidate(dentry, flags); 1334bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1335bad61189SMiklos Szeredi if (error < 0) { 1336bad61189SMiklos Szeredi dput(dentry); 1337bad61189SMiklos Szeredi return ERR_PTR(error); 13385542aa2fSEric W. Biederman } else { 13395542aa2fSEric W. Biederman d_invalidate(dentry); 1340bad61189SMiklos Szeredi dput(dentry); 1341bad61189SMiklos Szeredi dentry = NULL; 1342bad61189SMiklos Szeredi } 1343bad61189SMiklos Szeredi } 1344bad61189SMiklos Szeredi } 1345bad61189SMiklos Szeredi } 1346baa03890SNick Piggin 1347bad61189SMiklos Szeredi if (!dentry) { 1348bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1349baa03890SNick Piggin if (unlikely(!dentry)) 1350baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1351baa03890SNick Piggin 1352bad61189SMiklos Szeredi *need_lookup = true; 1353baa03890SNick Piggin } 1354baa03890SNick Piggin return dentry; 1355baa03890SNick Piggin } 1356baa03890SNick Piggin 1357baa03890SNick Piggin /* 135813a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 135913a2c3beSJ. Bruce Fields * unhashed. 1360bad61189SMiklos Szeredi * 1361bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 136244396f4bSJosef Bacik */ 1363bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 136472bd866aSAl Viro unsigned int flags) 136544396f4bSJosef Bacik { 136644396f4bSJosef Bacik struct dentry *old; 136744396f4bSJosef Bacik 136844396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1369bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1370e188dc02SMiklos Szeredi dput(dentry); 137144396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1372e188dc02SMiklos Szeredi } 137344396f4bSJosef Bacik 137472bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 137544396f4bSJosef Bacik if (unlikely(old)) { 137644396f4bSJosef Bacik dput(dentry); 137744396f4bSJosef Bacik dentry = old; 137844396f4bSJosef Bacik } 137944396f4bSJosef Bacik return dentry; 138044396f4bSJosef Bacik } 138144396f4bSJosef Bacik 1382a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 138372bd866aSAl Viro struct dentry *base, unsigned int flags) 1384a3255546SAl Viro { 1385bad61189SMiklos Szeredi bool need_lookup; 1386a3255546SAl Viro struct dentry *dentry; 1387a3255546SAl Viro 138872bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1389bad61189SMiklos Szeredi if (!need_lookup) 1390a3255546SAl Viro return dentry; 1391bad61189SMiklos Szeredi 139272bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1393a3255546SAl Viro } 1394a3255546SAl Viro 139544396f4bSJosef Bacik /* 13961da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 13971da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 13981da177e4SLinus Torvalds * It _is_ time-critical. 13991da177e4SLinus Torvalds */ 1400e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1401254cf582SAl Viro struct path *path, struct inode **inode, 1402254cf582SAl Viro unsigned *seqp) 14031da177e4SLinus Torvalds { 14044ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 140531e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 14065a18fff2SAl Viro int need_reval = 1; 14075a18fff2SAl Viro int status = 1; 14089875cf80SDavid Howells int err; 14099875cf80SDavid Howells 14103cac260aSAl Viro /* 1411b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1412b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1413b04f784eSNick Piggin * do the non-racy lookup, below. 1414b04f784eSNick Piggin */ 141531e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 141631e6b01fSNick Piggin unsigned seq; 1417766c4cbfSAl Viro bool negative; 1418da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 14195a18fff2SAl Viro if (!dentry) 14205a18fff2SAl Viro goto unlazy; 14215a18fff2SAl Viro 142212f8ad4bSLinus Torvalds /* 142312f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 142412f8ad4bSLinus Torvalds * the dentry name information from lookup. 142512f8ad4bSLinus Torvalds */ 142663afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1427766c4cbfSAl Viro negative = d_is_negative(dentry); 142812f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 142912f8ad4bSLinus Torvalds return -ECHILD; 1430766c4cbfSAl Viro if (negative) 1431766c4cbfSAl Viro return -ENOENT; 143212f8ad4bSLinus Torvalds 143312f8ad4bSLinus Torvalds /* 143412f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 143512f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 143612f8ad4bSLinus Torvalds * 143712f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 143812f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 143912f8ad4bSLinus Torvalds */ 144031e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 144131e6b01fSNick Piggin return -ECHILD; 14425a18fff2SAl Viro 1443254cf582SAl Viro *seqp = seq; 144424643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 14454ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14465a18fff2SAl Viro if (unlikely(status <= 0)) { 14475a18fff2SAl Viro if (status != -ECHILD) 14485a18fff2SAl Viro need_reval = 0; 14495a18fff2SAl Viro goto unlazy; 14505a18fff2SAl Viro } 145124643087SAl Viro } 145231e6b01fSNick Piggin path->mnt = mnt; 145331e6b01fSNick Piggin path->dentry = dentry; 1454254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 14559875cf80SDavid Howells return 0; 14565a18fff2SAl Viro unlazy: 1457254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 14585a18fff2SAl Viro return -ECHILD; 14595a18fff2SAl Viro } else { 1460e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 146124643087SAl Viro } 14625a18fff2SAl Viro 146381e6f520SAl Viro if (unlikely(!dentry)) 146481e6f520SAl Viro goto need_lookup; 14655a18fff2SAl Viro 14665a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14674ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14685a18fff2SAl Viro if (unlikely(status <= 0)) { 14695a18fff2SAl Viro if (status < 0) { 14705a18fff2SAl Viro dput(dentry); 14715a18fff2SAl Viro return status; 14725a18fff2SAl Viro } 14735542aa2fSEric W. Biederman d_invalidate(dentry); 14745a18fff2SAl Viro dput(dentry); 147581e6f520SAl Viro goto need_lookup; 14765a18fff2SAl Viro } 1477697f514dSMiklos Szeredi 1478766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1479766c4cbfSAl Viro dput(dentry); 1480766c4cbfSAl Viro return -ENOENT; 1481766c4cbfSAl Viro } 14821da177e4SLinus Torvalds path->mnt = mnt; 14831da177e4SLinus Torvalds path->dentry = dentry; 1484756daf26SNeilBrown err = follow_managed(path, nd); 14858402752eSAl Viro if (likely(!err)) 148663afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 14878402752eSAl Viro return err; 148881e6f520SAl Viro 148981e6f520SAl Viro need_lookup: 1490697f514dSMiklos Szeredi return 1; 1491697f514dSMiklos Szeredi } 1492697f514dSMiklos Szeredi 1493697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1494cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1495697f514dSMiklos Szeredi { 1496697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1497697f514dSMiklos Szeredi 1498697f514dSMiklos Szeredi parent = nd->path.dentry; 149981e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 150081e6f520SAl Viro 150181e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1502cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 150381e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 150481e6f520SAl Viro if (IS_ERR(dentry)) 150581e6f520SAl Viro return PTR_ERR(dentry); 1506697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1507697f514dSMiklos Szeredi path->dentry = dentry; 15088402752eSAl Viro return follow_managed(path, nd); 15091da177e4SLinus Torvalds } 15101da177e4SLinus Torvalds 151152094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 151252094c8aSAl Viro { 151352094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 15144ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 151552094c8aSAl Viro if (err != -ECHILD) 151652094c8aSAl Viro return err; 15176e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 151852094c8aSAl Viro return -ECHILD; 151952094c8aSAl Viro } 15204ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 152152094c8aSAl Viro } 152252094c8aSAl Viro 15239856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 15249856fa1bSAl Viro { 15259856fa1bSAl Viro if (type == LAST_DOTDOT) { 15269856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 152770291aecSAl Viro return follow_dotdot_rcu(nd); 15289856fa1bSAl Viro } else 15299856fa1bSAl Viro follow_dotdot(nd); 15309856fa1bSAl Viro } 15319856fa1bSAl Viro return 0; 15329856fa1bSAl Viro } 15339856fa1bSAl Viro 1534951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1535951361f9SAl Viro { 1536951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1537951361f9SAl Viro path_put(&nd->path); 1538951361f9SAl Viro } else { 1539951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 15405b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1541951361f9SAl Viro nd->root.mnt = NULL; 15428b61e74fSAl Viro rcu_read_unlock(); 1543951361f9SAl Viro } 154415439726SAl Viro while (unlikely(nd->depth)) 154515439726SAl Viro put_link(nd); 1546951361f9SAl Viro } 1547951361f9SAl Viro 1548181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1549181548c0SAl Viro struct inode *inode, unsigned seq) 1550d63ff28fSAl Viro { 1551626de996SAl Viro int error; 15521cf2665bSAl Viro struct saved *last; 1553756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1554626de996SAl Viro path_to_nameidata(link, nd); 1555626de996SAl Viro return -ELOOP; 1556626de996SAl Viro } 1557d63ff28fSAl Viro if (nd->flags & LOOKUP_RCU) { 1558d63ff28fSAl Viro if (unlikely(nd->path.mnt != link->mnt || 1559254cf582SAl Viro unlazy_walk(nd, link->dentry, seq))) { 1560d63ff28fSAl Viro return -ECHILD; 1561d63ff28fSAl Viro } 1562d63ff28fSAl Viro } 1563cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1564cd179f44SAl Viro mntget(link->mnt); 1565626de996SAl Viro error = nd_alloc_stack(nd); 1566626de996SAl Viro if (unlikely(error)) { 1567cd179f44SAl Viro path_put(link); 1568626de996SAl Viro return error; 1569626de996SAl Viro } 1570626de996SAl Viro 1571ab104923SAl Viro last = nd->stack + nd->depth++; 15721cf2665bSAl Viro last->link = *link; 1573ab104923SAl Viro last->cookie = NULL; 1574181548c0SAl Viro last->inode = inode; 1575d63ff28fSAl Viro return 1; 1576d63ff28fSAl Viro } 1577d63ff28fSAl Viro 15783ddcd056SLinus Torvalds /* 15793ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15803ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15813ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15823ddcd056SLinus Torvalds * for the common case. 15833ddcd056SLinus Torvalds */ 1584254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1585181548c0SAl Viro int follow, 1586181548c0SAl Viro struct inode *inode, unsigned seq) 15873ddcd056SLinus Torvalds { 1588d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1589d63ff28fSAl Viro return 0; 1590d63ff28fSAl Viro if (!follow) 1591d63ff28fSAl Viro return 0; 1592181548c0SAl Viro return pick_link(nd, link, inode, seq); 15933ddcd056SLinus Torvalds } 15943ddcd056SLinus Torvalds 15954693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 15964693a547SAl Viro 15974693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1598ce57dfc1SAl Viro { 1599caa85634SAl Viro struct path path; 1600ce57dfc1SAl Viro struct inode *inode; 1601254cf582SAl Viro unsigned seq; 1602ce57dfc1SAl Viro int err; 1603ce57dfc1SAl Viro /* 1604ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1605ce57dfc1SAl Viro * to be able to know about the current root directory and 1606ce57dfc1SAl Viro * parent relationships. 1607ce57dfc1SAl Viro */ 16084693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 16094693a547SAl Viro err = handle_dots(nd, nd->last_type); 16104693a547SAl Viro if (flags & WALK_PUT) 16114693a547SAl Viro put_link(nd); 16124693a547SAl Viro return err; 16134693a547SAl Viro } 1614254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1615ce57dfc1SAl Viro if (unlikely(err)) { 1616697f514dSMiklos Szeredi if (err < 0) 1617f0a9ba70SAl Viro return err; 1618697f514dSMiklos Szeredi 1619caa85634SAl Viro err = lookup_slow(nd, &path); 1620697f514dSMiklos Szeredi if (err < 0) 1621f0a9ba70SAl Viro return err; 1622697f514dSMiklos Szeredi 162363afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 1624254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1625697f514dSMiklos Szeredi err = -ENOENT; 1626caa85634SAl Viro if (d_is_negative(path.dentry)) 1627697f514dSMiklos Szeredi goto out_path_put; 1628766c4cbfSAl Viro } 1629697f514dSMiklos Szeredi 16304693a547SAl Viro if (flags & WALK_PUT) 16314693a547SAl Viro put_link(nd); 1632181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1633d63ff28fSAl Viro if (unlikely(err)) 1634d63ff28fSAl Viro return err; 1635caa85634SAl Viro path_to_nameidata(&path, nd); 1636ce57dfc1SAl Viro nd->inode = inode; 1637254cf582SAl Viro nd->seq = seq; 1638ce57dfc1SAl Viro return 0; 1639697f514dSMiklos Szeredi 1640697f514dSMiklos Szeredi out_path_put: 1641caa85634SAl Viro path_to_nameidata(&path, nd); 1642697f514dSMiklos Szeredi return err; 1643ce57dfc1SAl Viro } 1644ce57dfc1SAl Viro 16451da177e4SLinus Torvalds /* 1646bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1647bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1648bfcfaa77SLinus Torvalds * 1649bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1650bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1651bfcfaa77SLinus Torvalds * fast. 1652bfcfaa77SLinus Torvalds * 1653bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1654bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1655bfcfaa77SLinus Torvalds * crossing operation. 1656bfcfaa77SLinus Torvalds * 1657bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1658bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1659bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1660bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1661bfcfaa77SLinus Torvalds */ 1662bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1663bfcfaa77SLinus Torvalds 1664f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1665bfcfaa77SLinus Torvalds 1666f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1667bfcfaa77SLinus Torvalds 1668bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1669bfcfaa77SLinus Torvalds { 167099d263d4SLinus Torvalds return hash_64(hash, 32); 1671bfcfaa77SLinus Torvalds } 1672bfcfaa77SLinus Torvalds 1673bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1674bfcfaa77SLinus Torvalds 1675bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1676bfcfaa77SLinus Torvalds 1677bfcfaa77SLinus Torvalds #endif 1678bfcfaa77SLinus Torvalds 1679bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1680bfcfaa77SLinus Torvalds { 1681bfcfaa77SLinus Torvalds unsigned long a, mask; 1682bfcfaa77SLinus Torvalds unsigned long hash = 0; 1683bfcfaa77SLinus Torvalds 1684bfcfaa77SLinus Torvalds for (;;) { 1685e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1686bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1687bfcfaa77SLinus Torvalds break; 1688bfcfaa77SLinus Torvalds hash += a; 1689f132c5beSAl Viro hash *= 9; 1690bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1691bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1692bfcfaa77SLinus Torvalds if (!len) 1693bfcfaa77SLinus Torvalds goto done; 1694bfcfaa77SLinus Torvalds } 1695a5c21dceSWill Deacon mask = bytemask_from_count(len); 1696bfcfaa77SLinus Torvalds hash += mask & a; 1697bfcfaa77SLinus Torvalds done: 1698bfcfaa77SLinus Torvalds return fold_hash(hash); 1699bfcfaa77SLinus Torvalds } 1700bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1701bfcfaa77SLinus Torvalds 1702bfcfaa77SLinus Torvalds /* 1703bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1704d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1705bfcfaa77SLinus Torvalds */ 1706d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1707bfcfaa77SLinus Torvalds { 170836126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 170936126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1710bfcfaa77SLinus Torvalds 1711bfcfaa77SLinus Torvalds hash = a = 0; 1712bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1713bfcfaa77SLinus Torvalds do { 1714bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1715bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1716e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 171736126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 171836126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1719bfcfaa77SLinus Torvalds 172036126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 172136126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 172236126f8fSLinus Torvalds 172336126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 172436126f8fSLinus Torvalds 172536126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 17269226b5b4SLinus Torvalds len += find_zero(mask); 1727d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1728bfcfaa77SLinus Torvalds } 1729bfcfaa77SLinus Torvalds 1730bfcfaa77SLinus Torvalds #else 1731bfcfaa77SLinus Torvalds 17320145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 17330145acc2SLinus Torvalds { 17340145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 17350145acc2SLinus Torvalds while (len--) 17360145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 17370145acc2SLinus Torvalds return end_name_hash(hash); 17380145acc2SLinus Torvalds } 1739ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 17400145acc2SLinus Torvalds 17413ddcd056SLinus Torvalds /* 1742200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1743200e9ef7SLinus Torvalds * one character. 1744200e9ef7SLinus Torvalds */ 1745d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1746200e9ef7SLinus Torvalds { 1747200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1748200e9ef7SLinus Torvalds unsigned long len = 0, c; 1749200e9ef7SLinus Torvalds 1750200e9ef7SLinus Torvalds c = (unsigned char)*name; 1751200e9ef7SLinus Torvalds do { 1752200e9ef7SLinus Torvalds len++; 1753200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1754200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1755200e9ef7SLinus Torvalds } while (c && c != '/'); 1756d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1757200e9ef7SLinus Torvalds } 1758200e9ef7SLinus Torvalds 1759bfcfaa77SLinus Torvalds #endif 1760bfcfaa77SLinus Torvalds 1761200e9ef7SLinus Torvalds /* 17621da177e4SLinus Torvalds * Name resolution. 1763ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1764ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17651da177e4SLinus Torvalds * 1766ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1767ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17681da177e4SLinus Torvalds */ 17696de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17701da177e4SLinus Torvalds { 17711da177e4SLinus Torvalds int err; 17721da177e4SLinus Torvalds 17731da177e4SLinus Torvalds while (*name=='/') 17741da177e4SLinus Torvalds name++; 17751da177e4SLinus Torvalds if (!*name) 17769e18f10aSAl Viro return 0; 17771da177e4SLinus Torvalds 17781da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17791da177e4SLinus Torvalds for(;;) { 1780d6bb3e90SLinus Torvalds u64 hash_len; 1781fe479a58SAl Viro int type; 17821da177e4SLinus Torvalds 178352094c8aSAl Viro err = may_lookup(nd); 17841da177e4SLinus Torvalds if (err) 17853595e234SAl Viro return err; 17861da177e4SLinus Torvalds 1787d6bb3e90SLinus Torvalds hash_len = hash_name(name); 17881da177e4SLinus Torvalds 1789fe479a58SAl Viro type = LAST_NORM; 1790d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1791fe479a58SAl Viro case 2: 1792200e9ef7SLinus Torvalds if (name[1] == '.') { 1793fe479a58SAl Viro type = LAST_DOTDOT; 179416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 179516c2cd71SAl Viro } 1796fe479a58SAl Viro break; 1797fe479a58SAl Viro case 1: 1798fe479a58SAl Viro type = LAST_DOT; 1799fe479a58SAl Viro } 18005a202bcdSAl Viro if (likely(type == LAST_NORM)) { 18015a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 180216c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 18035a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1804a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1805da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 18065a202bcdSAl Viro if (err < 0) 18073595e234SAl Viro return err; 1808d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1809d6bb3e90SLinus Torvalds name = this.name; 18105a202bcdSAl Viro } 18115a202bcdSAl Viro } 1812fe479a58SAl Viro 1813d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1814d6bb3e90SLinus Torvalds nd->last.name = name; 18155f4a6a69SAl Viro nd->last_type = type; 18165f4a6a69SAl Viro 1817d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1818d6bb3e90SLinus Torvalds if (!*name) 1819bdf6cbf1SAl Viro goto OK; 1820200e9ef7SLinus Torvalds /* 1821200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1822200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1823200e9ef7SLinus Torvalds */ 1824200e9ef7SLinus Torvalds do { 1825d6bb3e90SLinus Torvalds name++; 1826d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 18278620c238SAl Viro if (unlikely(!*name)) { 18288620c238SAl Viro OK: 1829368ee9baSAl Viro /* pathname body, done */ 18308620c238SAl Viro if (!nd->depth) 18318620c238SAl Viro return 0; 18328620c238SAl Viro name = nd->stack[nd->depth - 1].name; 1833368ee9baSAl Viro /* trailing symlink, done */ 18348620c238SAl Viro if (!name) 18358620c238SAl Viro return 0; 18368620c238SAl Viro /* last component of nested symlink */ 18374693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 18388620c238SAl Viro } else { 18394693a547SAl Viro err = walk_component(nd, WALK_GET); 18408620c238SAl Viro } 1841ce57dfc1SAl Viro if (err < 0) 18423595e234SAl Viro return err; 1843fe479a58SAl Viro 1844ce57dfc1SAl Viro if (err) { 1845626de996SAl Viro const char *s = get_link(nd); 18465a460275SAl Viro 18473595e234SAl Viro if (unlikely(IS_ERR(s))) 18483595e234SAl Viro return PTR_ERR(s); 1849172a39a0SAl Viro err = 0; 185012b09578SAl Viro if (unlikely(!s)) { 185112b09578SAl Viro /* jumped */ 1852b9ff4429SAl Viro put_link(nd); 185312b09578SAl Viro } else { 1854071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 185532cd7468SAl Viro name = s; 18569e18f10aSAl Viro continue; 185748c8b0c5SAl Viro } 185831e6b01fSNick Piggin } 18593595e234SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) 18603595e234SAl Viro return -ENOTDIR; 18615f4a6a69SAl Viro } 1862ce57dfc1SAl Viro } 18631da177e4SLinus Torvalds 1864368ee9baSAl Viro static const char *path_init(int dfd, const struct filename *name, 1865368ee9baSAl Viro unsigned int flags, struct nameidata *nd) 186631e6b01fSNick Piggin { 186731e6b01fSNick Piggin int retval = 0; 1868fd2f7cb5SAl Viro const char *s = name->name; 186931e6b01fSNick Piggin 187031e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1871980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 187231e6b01fSNick Piggin nd->depth = 0; 1873368ee9baSAl Viro nd->total_link_count = 0; 18745b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1875b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1876b18825a7SDavid Howells struct inode *inode = root->d_inode; 1877fd2f7cb5SAl Viro if (*s) { 187844b1d530SMiklos Szeredi if (!d_can_lookup(root)) 1879368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 18805b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18815b6ca027SAl Viro if (retval) 1882368ee9baSAl Viro return ERR_PTR(retval); 188373d049a4SAl Viro } 18845b6ca027SAl Viro nd->path = nd->root; 18855b6ca027SAl Viro nd->inode = inode; 18865b6ca027SAl Viro if (flags & LOOKUP_RCU) { 18878b61e74fSAl Viro rcu_read_lock(); 18885b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 188948a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 18905b6ca027SAl Viro } else { 18915b6ca027SAl Viro path_get(&nd->path); 18925b6ca027SAl Viro } 1893368ee9baSAl Viro return s; 18945b6ca027SAl Viro } 18955b6ca027SAl Viro 189631e6b01fSNick Piggin nd->root.mnt = NULL; 189731e6b01fSNick Piggin 189848a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 1899fd2f7cb5SAl Viro if (*s == '/') { 1900e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19018b61e74fSAl Viro rcu_read_lock(); 19027bd88377SAl Viro nd->seq = set_root_rcu(nd); 1903e41f7d4eSAl Viro } else { 1904e41f7d4eSAl Viro set_root(nd); 1905e41f7d4eSAl Viro path_get(&nd->root); 1906e41f7d4eSAl Viro } 190731e6b01fSNick Piggin nd->path = nd->root; 190831e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1909e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 191031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1911c28cc364SNick Piggin unsigned seq; 191231e6b01fSNick Piggin 19138b61e74fSAl Viro rcu_read_lock(); 191431e6b01fSNick Piggin 1915c28cc364SNick Piggin do { 1916c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 191731e6b01fSNick Piggin nd->path = fs->pwd; 1918c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1919c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1920e41f7d4eSAl Viro } else { 1921e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1922e41f7d4eSAl Viro } 192331e6b01fSNick Piggin } else { 1924582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 19252903ff01SAl Viro struct fd f = fdget_raw(dfd); 192631e6b01fSNick Piggin struct dentry *dentry; 192731e6b01fSNick Piggin 19282903ff01SAl Viro if (!f.file) 1929368ee9baSAl Viro return ERR_PTR(-EBADF); 193031e6b01fSNick Piggin 19312903ff01SAl Viro dentry = f.file->f_path.dentry; 193231e6b01fSNick Piggin 1933fd2f7cb5SAl Viro if (*s) { 193444b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 19352903ff01SAl Viro fdput(f); 1936368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 1937f52e0c11SAl Viro } 19382903ff01SAl Viro } 19392903ff01SAl Viro 19402903ff01SAl Viro nd->path = f.file->f_path; 1941e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19428b61e74fSAl Viro rcu_read_lock(); 194334a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 194434a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 19455590ff0dSUlrich Drepper } else { 19462903ff01SAl Viro path_get(&nd->path); 194734a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 19481da177e4SLinus Torvalds } 194934a26b99SAl Viro fdput(f); 1950368ee9baSAl Viro return s; 1951e41f7d4eSAl Viro } 1952e41f7d4eSAl Viro 195331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 19544023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 1955368ee9baSAl Viro return s; 19564023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 1957368ee9baSAl Viro return s; 19584023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 19594023bfc9SAl Viro nd->root.mnt = NULL; 19604023bfc9SAl Viro rcu_read_unlock(); 1961368ee9baSAl Viro return ERR_PTR(-ECHILD); 19629b4a9b14SAl Viro } 19639b4a9b14SAl Viro 1964893b7775SAl Viro static void path_cleanup(struct nameidata *nd) 1965893b7775SAl Viro { 1966893b7775SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 1967893b7775SAl Viro path_put(&nd->root); 1968893b7775SAl Viro nd->root.mnt = NULL; 1969893b7775SAl Viro } 19709b4a9b14SAl Viro } 19719b4a9b14SAl Viro 19723bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 197395fa25d9SAl Viro { 197495fa25d9SAl Viro const char *s; 1975fec2fa24SAl Viro int error = may_follow_link(nd); 1976deb106c6SAl Viro if (unlikely(error)) 19773bdba28bSAl Viro return ERR_PTR(error); 197895fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 1979fab51e8aSAl Viro nd->stack[0].name = NULL; 19803b2e7f75SAl Viro s = get_link(nd); 1981deb106c6SAl Viro return s ? s : ""; 198295fa25d9SAl Viro } 198395fa25d9SAl Viro 1984caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 1985bd92d7feSAl Viro { 1986bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 1987bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 1988bd92d7feSAl Viro 1989bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 1990deb106c6SAl Viro return walk_component(nd, 19914693a547SAl Viro nd->flags & LOOKUP_FOLLOW 19924693a547SAl Viro ? nd->depth 19934693a547SAl Viro ? WALK_PUT | WALK_GET 19944693a547SAl Viro : WALK_GET 19954693a547SAl Viro : 0); 1996bd92d7feSAl Viro } 1997bd92d7feSAl Viro 19989b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 19995eb6b495SAl Viro static int path_lookupat(int dfd, const struct filename *name, 20009b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 20019b4a9b14SAl Viro { 2002368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2003bd92d7feSAl Viro int err; 200431e6b01fSNick Piggin 2005368ee9baSAl Viro if (IS_ERR(s)) 2006368ee9baSAl Viro return PTR_ERR(s); 20073bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 20083bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 20093bdba28bSAl Viro s = trailing_symlink(nd); 20103bdba28bSAl Viro if (IS_ERR(s)) { 20113bdba28bSAl Viro err = PTR_ERR(s); 20126d7b5aaeSAl Viro break; 2013bd92d7feSAl Viro } 2014bd92d7feSAl Viro } 20159f1fafeeSAl Viro if (!err) 20169f1fafeeSAl Viro err = complete_walk(nd); 2017bd92d7feSAl Viro 2018deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2019deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2020bd23a539SAl Viro err = -ENOTDIR; 2021deb106c6SAl Viro if (err) 2022deb106c6SAl Viro terminate_walk(nd); 202316c2cd71SAl Viro 2024893b7775SAl Viro path_cleanup(nd); 2025bd92d7feSAl Viro return err; 202631e6b01fSNick Piggin } 202731e6b01fSNick Piggin 2028873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 2029873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 2030873f1eedSJeff Layton { 2031894bc8c4SAl Viro int retval; 2032756daf26SNeilBrown struct nameidata *saved_nd = set_nameidata(nd); 2033894bc8c4SAl Viro 2034894bc8c4SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 2035873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 20365eb6b495SAl Viro retval = path_lookupat(dfd, name, flags, nd); 2037873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 20385eb6b495SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 2039873f1eedSJeff Layton 2040873f1eedSJeff Layton if (likely(!retval)) 2041adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2042756daf26SNeilBrown restore_nameidata(saved_nd); 2043873f1eedSJeff Layton return retval; 2044873f1eedSJeff Layton } 2045873f1eedSJeff Layton 20468bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 20478bcb77faSAl Viro static int path_parentat(int dfd, const struct filename *name, 20488bcb77faSAl Viro unsigned int flags, struct nameidata *nd) 20498bcb77faSAl Viro { 2050368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2051368ee9baSAl Viro int err; 2052368ee9baSAl Viro if (IS_ERR(s)) 2053368ee9baSAl Viro return PTR_ERR(s); 2054368ee9baSAl Viro err = link_path_walk(s, nd); 20558bcb77faSAl Viro if (!err) 20568bcb77faSAl Viro err = complete_walk(nd); 2057deb106c6SAl Viro if (err) 2058deb106c6SAl Viro terminate_walk(nd); 20598bcb77faSAl Viro path_cleanup(nd); 20608bcb77faSAl Viro return err; 20618bcb77faSAl Viro } 20628bcb77faSAl Viro 20638bcb77faSAl Viro static int filename_parentat(int dfd, struct filename *name, 20648bcb77faSAl Viro unsigned int flags, struct nameidata *nd) 20658bcb77faSAl Viro { 20668bcb77faSAl Viro int retval; 20678bcb77faSAl Viro struct nameidata *saved_nd = set_nameidata(nd); 20688bcb77faSAl Viro 20698bcb77faSAl Viro retval = path_parentat(dfd, name, flags | LOOKUP_RCU, nd); 20708bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 20718bcb77faSAl Viro retval = path_parentat(dfd, name, flags, nd); 20728bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 20738bcb77faSAl Viro retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, nd); 20748bcb77faSAl Viro 20758bcb77faSAl Viro if (likely(!retval)) 20768bcb77faSAl Viro audit_inode(name, nd->path.dentry, LOOKUP_PARENT); 20778bcb77faSAl Viro restore_nameidata(saved_nd); 20788bcb77faSAl Viro return retval; 20798bcb77faSAl Viro } 20808bcb77faSAl Viro 208179714f72SAl Viro /* does lookup, returns the object with parent locked */ 208279714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 20835590ff0dSUlrich Drepper { 208451689104SPaul Moore struct filename *filename = getname_kernel(name); 208579714f72SAl Viro struct nameidata nd; 208679714f72SAl Viro struct dentry *d; 208751689104SPaul Moore int err; 208851689104SPaul Moore 208951689104SPaul Moore if (IS_ERR(filename)) 209051689104SPaul Moore return ERR_CAST(filename); 209151689104SPaul Moore 20928bcb77faSAl Viro err = filename_parentat(AT_FDCWD, filename, 0, &nd); 209351689104SPaul Moore if (err) { 209451689104SPaul Moore d = ERR_PTR(err); 209551689104SPaul Moore goto out; 209651689104SPaul Moore } 209779714f72SAl Viro if (nd.last_type != LAST_NORM) { 209879714f72SAl Viro path_put(&nd.path); 209951689104SPaul Moore d = ERR_PTR(-EINVAL); 210051689104SPaul Moore goto out; 210179714f72SAl Viro } 210279714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 21031e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 210479714f72SAl Viro if (IS_ERR(d)) { 210579714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 210679714f72SAl Viro path_put(&nd.path); 210751689104SPaul Moore goto out; 210879714f72SAl Viro } 210979714f72SAl Viro *path = nd.path; 211051689104SPaul Moore out: 211151689104SPaul Moore putname(filename); 211279714f72SAl Viro return d; 21135590ff0dSUlrich Drepper } 21145590ff0dSUlrich Drepper 2115d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2116d1811465SAl Viro { 2117d1811465SAl Viro struct nameidata nd; 211874eb8cc5SAl Viro struct filename *filename = getname_kernel(name); 211974eb8cc5SAl Viro int res = PTR_ERR(filename); 212074eb8cc5SAl Viro 212174eb8cc5SAl Viro if (!IS_ERR(filename)) { 212274eb8cc5SAl Viro res = filename_lookup(AT_FDCWD, filename, flags, &nd); 212374eb8cc5SAl Viro putname(filename); 2124d1811465SAl Viro if (!res) 2125d1811465SAl Viro *path = nd.path; 212674eb8cc5SAl Viro } 2127d1811465SAl Viro return res; 2128d1811465SAl Viro } 21294d359507SAl Viro EXPORT_SYMBOL(kern_path); 2130d1811465SAl Viro 213116f18200SJosef 'Jeff' Sipek /** 213216f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 213316f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 213416f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 213516f18200SJosef 'Jeff' Sipek * @name: pointer to file name 213616f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2137e0a01249SAl Viro * @path: pointer to struct path to fill 213816f18200SJosef 'Jeff' Sipek */ 213916f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 214016f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2141e0a01249SAl Viro struct path *path) 214216f18200SJosef 'Jeff' Sipek { 214374eb8cc5SAl Viro struct filename *filename = getname_kernel(name); 214474eb8cc5SAl Viro int err = PTR_ERR(filename); 214574eb8cc5SAl Viro 214674eb8cc5SAl Viro BUG_ON(flags & LOOKUP_PARENT); 214774eb8cc5SAl Viro 214874eb8cc5SAl Viro /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */ 214974eb8cc5SAl Viro if (!IS_ERR(filename)) { 2150e0a01249SAl Viro struct nameidata nd; 2151e0a01249SAl Viro nd.root.dentry = dentry; 2152e0a01249SAl Viro nd.root.mnt = mnt; 215374eb8cc5SAl Viro err = filename_lookup(AT_FDCWD, filename, 215474eb8cc5SAl Viro flags | LOOKUP_ROOT, &nd); 2155e0a01249SAl Viro if (!err) 2156e0a01249SAl Viro *path = nd.path; 215774eb8cc5SAl Viro putname(filename); 215874eb8cc5SAl Viro } 2159e0a01249SAl Viro return err; 216016f18200SJosef 'Jeff' Sipek } 21614d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 216216f18200SJosef 'Jeff' Sipek 2163eead1911SChristoph Hellwig /** 2164a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2165eead1911SChristoph Hellwig * @name: pathname component to lookup 2166eead1911SChristoph Hellwig * @base: base directory to lookup from 2167eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2168eead1911SChristoph Hellwig * 2169a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 21709e7543e9SAl Viro * not be called by generic code. 2171eead1911SChristoph Hellwig */ 2172057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2173057f6c01SJames Morris { 2174057f6c01SJames Morris struct qstr this; 21756a96ba54SAl Viro unsigned int c; 2176cda309deSMiklos Szeredi int err; 2177057f6c01SJames Morris 21782f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 21792f9092e1SDavid Woodhouse 21806a96ba54SAl Viro this.name = name; 21816a96ba54SAl Viro this.len = len; 21820145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 21836a96ba54SAl Viro if (!len) 21846a96ba54SAl Viro return ERR_PTR(-EACCES); 21856a96ba54SAl Viro 218621d8a15aSAl Viro if (unlikely(name[0] == '.')) { 218721d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 218821d8a15aSAl Viro return ERR_PTR(-EACCES); 218921d8a15aSAl Viro } 219021d8a15aSAl Viro 21916a96ba54SAl Viro while (len--) { 21926a96ba54SAl Viro c = *(const unsigned char *)name++; 21936a96ba54SAl Viro if (c == '/' || c == '\0') 21946a96ba54SAl Viro return ERR_PTR(-EACCES); 21956a96ba54SAl Viro } 21965a202bcdSAl Viro /* 21975a202bcdSAl Viro * See if the low-level filesystem might want 21985a202bcdSAl Viro * to use its own hash.. 21995a202bcdSAl Viro */ 22005a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2201da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 22025a202bcdSAl Viro if (err < 0) 22035a202bcdSAl Viro return ERR_PTR(err); 22045a202bcdSAl Viro } 2205eead1911SChristoph Hellwig 2206cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2207cda309deSMiklos Szeredi if (err) 2208cda309deSMiklos Szeredi return ERR_PTR(err); 2209cda309deSMiklos Szeredi 221072bd866aSAl Viro return __lookup_hash(&this, base, 0); 2211057f6c01SJames Morris } 22124d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2213057f6c01SJames Morris 22141fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 22151fa1e7f6SAndy Whitcroft struct path *path, int *empty) 22161da177e4SLinus Torvalds { 22172d8f3038SAl Viro struct nameidata nd; 221891a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 22191da177e4SLinus Torvalds int err = PTR_ERR(tmp); 22201da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 22212d8f3038SAl Viro 22222d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 22232d8f3038SAl Viro 2224873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 22251da177e4SLinus Torvalds putname(tmp); 22262d8f3038SAl Viro if (!err) 22272d8f3038SAl Viro *path = nd.path; 22281da177e4SLinus Torvalds } 22291da177e4SLinus Torvalds return err; 22301da177e4SLinus Torvalds } 22311da177e4SLinus Torvalds 22321fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 22331fa1e7f6SAndy Whitcroft struct path *path) 22341fa1e7f6SAndy Whitcroft { 2235f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 22361fa1e7f6SAndy Whitcroft } 22374d359507SAl Viro EXPORT_SYMBOL(user_path_at); 22381fa1e7f6SAndy Whitcroft 2239873f1eedSJeff Layton /* 2240873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2241873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2242873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2243873f1eedSJeff Layton * path-walking is complete. 2244873f1eedSJeff Layton */ 224591a27b2aSJeff Layton static struct filename * 2246f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2247f5beed75SAl Viro struct path *parent, 2248f5beed75SAl Viro struct qstr *last, 2249f5beed75SAl Viro int *type, 22509e790bd6SJeff Layton unsigned int flags) 22512ad94ae6SAl Viro { 2252f5beed75SAl Viro struct nameidata nd; 225391a27b2aSJeff Layton struct filename *s = getname(path); 22542ad94ae6SAl Viro int error; 22552ad94ae6SAl Viro 22569e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 22579e790bd6SJeff Layton flags &= LOOKUP_REVAL; 22589e790bd6SJeff Layton 22592ad94ae6SAl Viro if (IS_ERR(s)) 226091a27b2aSJeff Layton return s; 22612ad94ae6SAl Viro 22628bcb77faSAl Viro error = filename_parentat(dfd, s, flags, &nd); 226391a27b2aSJeff Layton if (error) { 22642ad94ae6SAl Viro putname(s); 226591a27b2aSJeff Layton return ERR_PTR(error); 226691a27b2aSJeff Layton } 2267f5beed75SAl Viro *parent = nd.path; 2268f5beed75SAl Viro *last = nd.last; 2269f5beed75SAl Viro *type = nd.last_type; 22702ad94ae6SAl Viro 227191a27b2aSJeff Layton return s; 22722ad94ae6SAl Viro } 22732ad94ae6SAl Viro 22748033426eSJeff Layton /** 2275197df04cSAl Viro * mountpoint_last - look up last component for umount 22768033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 22778033426eSJeff Layton * @path: pointer to container for result 22788033426eSJeff Layton * 22798033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 22808033426eSJeff Layton * need to resolve the path without doing any revalidation. 22818033426eSJeff Layton * 22828033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 22838033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 22848033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 22858033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 22868033426eSJeff Layton * bogus and it doesn't exist. 22878033426eSJeff Layton * 22888033426eSJeff Layton * Returns: 22898033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 22908033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 22918033426eSJeff Layton * put in this case. 22928033426eSJeff Layton * 22938033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 22948033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 22958033426eSJeff Layton * The nd->path reference will also be put. 22968033426eSJeff Layton * 22978033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 22988033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 22998033426eSJeff Layton * to the link, and nd->path will *not* be put. 23008033426eSJeff Layton */ 23018033426eSJeff Layton static int 2302197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 23038033426eSJeff Layton { 23048033426eSJeff Layton int error = 0; 23058033426eSJeff Layton struct dentry *dentry; 23068033426eSJeff Layton struct dentry *dir = nd->path.dentry; 23078033426eSJeff Layton 230835759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 230935759521SAl Viro if (nd->flags & LOOKUP_RCU) { 23106e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2311deb106c6SAl Viro return -ECHILD; 23128033426eSJeff Layton } 23138033426eSJeff Layton 23148033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 23158033426eSJeff Layton 23168033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 23178033426eSJeff Layton error = handle_dots(nd, nd->last_type); 231835759521SAl Viro if (error) 2319deb106c6SAl Viro return error; 23208033426eSJeff Layton dentry = dget(nd->path.dentry); 232135759521SAl Viro goto done; 23228033426eSJeff Layton } 23238033426eSJeff Layton 23248033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 23258033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 23268033426eSJeff Layton if (!dentry) { 23278033426eSJeff Layton /* 23288033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 23298033426eSJeff Layton * so that means that this dentry is probably a symlink or the 23308033426eSJeff Layton * path doesn't actually point to a mounted dentry. 23318033426eSJeff Layton */ 23328033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 23338033426eSJeff Layton if (!dentry) { 2334bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2335deb106c6SAl Viro return -ENOMEM; 23368033426eSJeff Layton } 233735759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 2338bcceeebaSDave Jones if (IS_ERR(dentry)) { 2339bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 2340deb106c6SAl Viro return PTR_ERR(dentry); 23418033426eSJeff Layton } 2342bcceeebaSDave Jones } 23438033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 23448033426eSJeff Layton 234535759521SAl Viro done: 2346698934dfSDavid Howells if (d_is_negative(dentry)) { 23478033426eSJeff Layton dput(dentry); 2348deb106c6SAl Viro return -ENOENT; 234935759521SAl Viro } 2350191d7f73SAl Viro if (nd->depth) 2351191d7f73SAl Viro put_link(nd); 23528033426eSJeff Layton path->dentry = dentry; 2353295dc39dSVasily Averin path->mnt = nd->path.mnt; 2354181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2355181548c0SAl Viro d_backing_inode(dentry), 0); 2356deb106c6SAl Viro if (unlikely(error)) 2357d63ff28fSAl Viro return error; 2358295dc39dSVasily Averin mntget(path->mnt); 23598033426eSJeff Layton follow_mount(path); 2360deb106c6SAl Viro return 0; 23618033426eSJeff Layton } 23628033426eSJeff Layton 23638033426eSJeff Layton /** 2364197df04cSAl Viro * path_mountpoint - look up a path to be umounted 23658033426eSJeff Layton * @dfd: directory file descriptor to start walk from 23668033426eSJeff Layton * @name: full pathname to walk 2367606d6fe3SRandy Dunlap * @path: pointer to container for result 23688033426eSJeff Layton * @flags: lookup flags 23698033426eSJeff Layton * 23708033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2371606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 23728033426eSJeff Layton */ 23738033426eSJeff Layton static int 2374668696dcSAl Viro path_mountpoint(int dfd, const struct filename *name, struct path *path, 237546afd6f6SAl Viro struct nameidata *nd, unsigned int flags) 23768033426eSJeff Layton { 2377368ee9baSAl Viro const char *s = path_init(dfd, name, flags, nd); 2378368ee9baSAl Viro int err; 2379368ee9baSAl Viro if (IS_ERR(s)) 2380368ee9baSAl Viro return PTR_ERR(s); 23813bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 23823bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 23833bdba28bSAl Viro s = trailing_symlink(nd); 23843bdba28bSAl Viro if (IS_ERR(s)) { 23853bdba28bSAl Viro err = PTR_ERR(s); 23868033426eSJeff Layton break; 23878033426eSJeff Layton } 23883bdba28bSAl Viro } 2389deb106c6SAl Viro terminate_walk(nd); 239046afd6f6SAl Viro path_cleanup(nd); 23918033426eSJeff Layton return err; 23928033426eSJeff Layton } 23938033426eSJeff Layton 23942d864651SAl Viro static int 2395668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 23962d864651SAl Viro unsigned int flags) 23972d864651SAl Viro { 2398756daf26SNeilBrown struct nameidata nd, *saved; 2399cbaab2dbSAl Viro int error; 2400668696dcSAl Viro if (IS_ERR(name)) 2401668696dcSAl Viro return PTR_ERR(name); 2402756daf26SNeilBrown saved = set_nameidata(&nd); 240346afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU); 24042d864651SAl Viro if (unlikely(error == -ECHILD)) 240546afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags); 24062d864651SAl Viro if (unlikely(error == -ESTALE)) 240746afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL); 24082d864651SAl Viro if (likely(!error)) 2409668696dcSAl Viro audit_inode(name, path->dentry, 0); 2410756daf26SNeilBrown restore_nameidata(saved); 2411668696dcSAl Viro putname(name); 24122d864651SAl Viro return error; 24132d864651SAl Viro } 24142d864651SAl Viro 24158033426eSJeff Layton /** 2416197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 24178033426eSJeff Layton * @dfd: directory file descriptor 24188033426eSJeff Layton * @name: pathname from userland 24198033426eSJeff Layton * @flags: lookup flags 24208033426eSJeff Layton * @path: pointer to container to hold result 24218033426eSJeff Layton * 24228033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 24238033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 24248033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 24258033426eSJeff Layton * and avoid revalidating the last component. 24268033426eSJeff Layton * 24278033426eSJeff Layton * Returns 0 and populates "path" on success. 24288033426eSJeff Layton */ 24298033426eSJeff Layton int 2430197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 24318033426eSJeff Layton struct path *path) 24328033426eSJeff Layton { 2433cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 24348033426eSJeff Layton } 24358033426eSJeff Layton 24362d864651SAl Viro int 24372d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 24382d864651SAl Viro unsigned int flags) 24392d864651SAl Viro { 2440cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 24412d864651SAl Viro } 24422d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 24432d864651SAl Viro 2444cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 24451da177e4SLinus Torvalds { 24468e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2447da9592edSDavid Howells 24488e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 24491da177e4SLinus Torvalds return 0; 24508e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 24511da177e4SLinus Torvalds return 0; 245223adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 24531da177e4SLinus Torvalds } 2454cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 24551da177e4SLinus Torvalds 24561da177e4SLinus Torvalds /* 24571da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 24581da177e4SLinus Torvalds * whether the type of victim is right. 24591da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 24601da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 24611da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 24621da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 24631da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 24641da177e4SLinus Torvalds * a. be owner of dir, or 24651da177e4SLinus Torvalds * b. be owner of victim, or 24661da177e4SLinus Torvalds * c. have CAP_FOWNER capability 24671da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 24681da177e4SLinus Torvalds * links pointing to it. 24691da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 24701da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 24711da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 24721da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 24731da177e4SLinus Torvalds * nfs_async_unlink(). 24741da177e4SLinus Torvalds */ 2475b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 24761da177e4SLinus Torvalds { 247763afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 24781da177e4SLinus Torvalds int error; 24791da177e4SLinus Torvalds 2480b18825a7SDavid Howells if (d_is_negative(victim)) 24811da177e4SLinus Torvalds return -ENOENT; 2482b18825a7SDavid Howells BUG_ON(!inode); 24831da177e4SLinus Torvalds 24841da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24854fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 24861da177e4SLinus Torvalds 2487f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 24881da177e4SLinus Torvalds if (error) 24891da177e4SLinus Torvalds return error; 24901da177e4SLinus Torvalds if (IS_APPEND(dir)) 24911da177e4SLinus Torvalds return -EPERM; 2492b18825a7SDavid Howells 2493b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2494b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 24951da177e4SLinus Torvalds return -EPERM; 24961da177e4SLinus Torvalds if (isdir) { 249744b1d530SMiklos Szeredi if (!d_is_dir(victim)) 24981da177e4SLinus Torvalds return -ENOTDIR; 24991da177e4SLinus Torvalds if (IS_ROOT(victim)) 25001da177e4SLinus Torvalds return -EBUSY; 250144b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 25021da177e4SLinus Torvalds return -EISDIR; 25031da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25041da177e4SLinus Torvalds return -ENOENT; 25051da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 25061da177e4SLinus Torvalds return -EBUSY; 25071da177e4SLinus Torvalds return 0; 25081da177e4SLinus Torvalds } 25091da177e4SLinus Torvalds 25101da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 25111da177e4SLinus Torvalds * dir. 25121da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 25131da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 25141da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 25151da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 25161da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 25171da177e4SLinus Torvalds */ 2518a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 25191da177e4SLinus Torvalds { 252014e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 25211da177e4SLinus Torvalds if (child->d_inode) 25221da177e4SLinus Torvalds return -EEXIST; 25231da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25241da177e4SLinus Torvalds return -ENOENT; 2525f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 25261da177e4SLinus Torvalds } 25271da177e4SLinus Torvalds 25281da177e4SLinus Torvalds /* 25291da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 25301da177e4SLinus Torvalds */ 25311da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 25321da177e4SLinus Torvalds { 25331da177e4SLinus Torvalds struct dentry *p; 25341da177e4SLinus Torvalds 25351da177e4SLinus Torvalds if (p1 == p2) { 2536f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 25371da177e4SLinus Torvalds return NULL; 25381da177e4SLinus Torvalds } 25391da177e4SLinus Torvalds 2540a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25411da177e4SLinus Torvalds 2542e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2543e2761a11SOGAWA Hirofumi if (p) { 2544f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2545f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 25461da177e4SLinus Torvalds return p; 25471da177e4SLinus Torvalds } 25481da177e4SLinus Torvalds 2549e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2550e2761a11SOGAWA Hirofumi if (p) { 2551f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2552f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 25531da177e4SLinus Torvalds return p; 25541da177e4SLinus Torvalds } 25551da177e4SLinus Torvalds 2556f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2557d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 25581da177e4SLinus Torvalds return NULL; 25591da177e4SLinus Torvalds } 25604d359507SAl Viro EXPORT_SYMBOL(lock_rename); 25611da177e4SLinus Torvalds 25621da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 25631da177e4SLinus Torvalds { 25641b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 25651da177e4SLinus Torvalds if (p1 != p2) { 25661b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2567a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25681da177e4SLinus Torvalds } 25691da177e4SLinus Torvalds } 25704d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 25711da177e4SLinus Torvalds 25724acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2573312b63fbSAl Viro bool want_excl) 25741da177e4SLinus Torvalds { 2575a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25761da177e4SLinus Torvalds if (error) 25771da177e4SLinus Torvalds return error; 25781da177e4SLinus Torvalds 2579acfa4380SAl Viro if (!dir->i_op->create) 25801da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 25811da177e4SLinus Torvalds mode &= S_IALLUGO; 25821da177e4SLinus Torvalds mode |= S_IFREG; 25831da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 25841da177e4SLinus Torvalds if (error) 25851da177e4SLinus Torvalds return error; 2586312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2587a74574aaSStephen Smalley if (!error) 2588f38aa942SAmy Griffis fsnotify_create(dir, dentry); 25891da177e4SLinus Torvalds return error; 25901da177e4SLinus Torvalds } 25914d359507SAl Viro EXPORT_SYMBOL(vfs_create); 25921da177e4SLinus Torvalds 259373d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 25941da177e4SLinus Torvalds { 25953fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 25961da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 25971da177e4SLinus Torvalds int error; 25981da177e4SLinus Torvalds 2599bcda7652SAl Viro /* O_PATH? */ 2600bcda7652SAl Viro if (!acc_mode) 2601bcda7652SAl Viro return 0; 2602bcda7652SAl Viro 26031da177e4SLinus Torvalds if (!inode) 26041da177e4SLinus Torvalds return -ENOENT; 26051da177e4SLinus Torvalds 2606c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2607c8fe8f30SChristoph Hellwig case S_IFLNK: 26081da177e4SLinus Torvalds return -ELOOP; 2609c8fe8f30SChristoph Hellwig case S_IFDIR: 2610c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 26111da177e4SLinus Torvalds return -EISDIR; 2612c8fe8f30SChristoph Hellwig break; 2613c8fe8f30SChristoph Hellwig case S_IFBLK: 2614c8fe8f30SChristoph Hellwig case S_IFCHR: 26153fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 26161da177e4SLinus Torvalds return -EACCES; 2617c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2618c8fe8f30SChristoph Hellwig case S_IFIFO: 2619c8fe8f30SChristoph Hellwig case S_IFSOCK: 26201da177e4SLinus Torvalds flag &= ~O_TRUNC; 2621c8fe8f30SChristoph Hellwig break; 26224a3fd211SDave Hansen } 2623b41572e9SDave Hansen 26243fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2625b41572e9SDave Hansen if (error) 2626b41572e9SDave Hansen return error; 26276146f0d5SMimi Zohar 26281da177e4SLinus Torvalds /* 26291da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 26301da177e4SLinus Torvalds */ 26311da177e4SLinus Torvalds if (IS_APPEND(inode)) { 26328737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 26337715b521SAl Viro return -EPERM; 26341da177e4SLinus Torvalds if (flag & O_TRUNC) 26357715b521SAl Viro return -EPERM; 26361da177e4SLinus Torvalds } 26371da177e4SLinus Torvalds 26381da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 26392e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 26407715b521SAl Viro return -EPERM; 26411da177e4SLinus Torvalds 2642f3c7691eSJ. Bruce Fields return 0; 26437715b521SAl Viro } 26447715b521SAl Viro 2645e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 26467715b521SAl Viro { 2647e1181ee6SJeff Layton struct path *path = &filp->f_path; 26487715b521SAl Viro struct inode *inode = path->dentry->d_inode; 26497715b521SAl Viro int error = get_write_access(inode); 26501da177e4SLinus Torvalds if (error) 26517715b521SAl Viro return error; 26521da177e4SLinus Torvalds /* 26531da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 26541da177e4SLinus Torvalds */ 2655d7a06983SJeff Layton error = locks_verify_locked(filp); 2656be6d3e56SKentaro Takeda if (!error) 2657ea0d3ab2STetsuo Handa error = security_path_truncate(path); 26581da177e4SLinus Torvalds if (!error) { 26597715b521SAl Viro error = do_truncate(path->dentry, 0, 2660d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2661e1181ee6SJeff Layton filp); 26621da177e4SLinus Torvalds } 26631da177e4SLinus Torvalds put_write_access(inode); 2664acd0c935SMimi Zohar return error; 26651da177e4SLinus Torvalds } 26661da177e4SLinus Torvalds 2667d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2668d57999e1SDave Hansen { 26698a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 26708a5e929dSAl Viro flag--; 2671d57999e1SDave Hansen return flag; 2672d57999e1SDave Hansen } 2673d57999e1SDave Hansen 2674d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2675d18e9008SMiklos Szeredi { 2676d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2677d18e9008SMiklos Szeredi if (error) 2678d18e9008SMiklos Szeredi return error; 2679d18e9008SMiklos Szeredi 2680d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2681d18e9008SMiklos Szeredi if (error) 2682d18e9008SMiklos Szeredi return error; 2683d18e9008SMiklos Szeredi 2684d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2685d18e9008SMiklos Szeredi } 2686d18e9008SMiklos Szeredi 26871acf0af9SDavid Howells /* 26881acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 26891acf0af9SDavid Howells * dentry. 26901acf0af9SDavid Howells * 26911acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 26921acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 26931acf0af9SDavid Howells * 26941acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 26951acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 26961acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 26971acf0af9SDavid Howells * 26981acf0af9SDavid Howells * Returns an error code otherwise. 26991acf0af9SDavid Howells */ 27002675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 270130d90494SAl Viro struct path *path, struct file *file, 2702015c3bbcSMiklos Szeredi const struct open_flags *op, 270364894cf8SAl Viro bool got_write, bool need_lookup, 270447237687SAl Viro int *opened) 2705d18e9008SMiklos Szeredi { 2706d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2707d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2708d18e9008SMiklos Szeredi umode_t mode; 2709d18e9008SMiklos Szeredi int error; 2710d18e9008SMiklos Szeredi int acc_mode; 2711d18e9008SMiklos Szeredi int create_error = 0; 2712d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2713116cc022SMiklos Szeredi bool excl; 2714d18e9008SMiklos Szeredi 2715d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2716d18e9008SMiklos Szeredi 2717d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2718d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 27192675a4ebSAl Viro error = -ENOENT; 2720d18e9008SMiklos Szeredi goto out; 2721d18e9008SMiklos Szeredi } 2722d18e9008SMiklos Szeredi 272362b259d8SMiklos Szeredi mode = op->mode; 2724d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2725d18e9008SMiklos Szeredi mode &= ~current_umask(); 2726d18e9008SMiklos Szeredi 2727116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2728116cc022SMiklos Szeredi if (excl) 2729d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2730d18e9008SMiklos Szeredi 2731d18e9008SMiklos Szeredi /* 2732d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2733d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2734d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2735d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2736d18e9008SMiklos Szeredi * 2737d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2738d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2739d18e9008SMiklos Szeredi */ 274064894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 274164894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 274264894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2743d18e9008SMiklos Szeredi /* 2744d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2745d18e9008SMiklos Szeredi * back to lookup + open 2746d18e9008SMiklos Szeredi */ 2747d18e9008SMiklos Szeredi goto no_open; 2748d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2749d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 275064894cf8SAl Viro create_error = -EROFS; 2751d18e9008SMiklos Szeredi goto no_open; 2752d18e9008SMiklos Szeredi } else { 2753d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 275464894cf8SAl Viro create_error = -EROFS; 2755d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2756d18e9008SMiklos Szeredi } 2757d18e9008SMiklos Szeredi } 2758d18e9008SMiklos Szeredi 2759d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 276038227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2761d18e9008SMiklos Szeredi if (error) { 2762d18e9008SMiklos Szeredi create_error = error; 2763d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2764d18e9008SMiklos Szeredi goto no_open; 2765d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2766d18e9008SMiklos Szeredi } 2767d18e9008SMiklos Szeredi } 2768d18e9008SMiklos Szeredi 2769d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2770d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2771d18e9008SMiklos Szeredi 277230d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 277330d90494SAl Viro file->f_path.mnt = nd->path.mnt; 277430d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 277547237687SAl Viro opened); 2776d9585277SAl Viro if (error < 0) { 2777d9585277SAl Viro if (create_error && error == -ENOENT) 2778d9585277SAl Viro error = create_error; 2779d18e9008SMiklos Szeredi goto out; 2780d18e9008SMiklos Szeredi } 2781d18e9008SMiklos Szeredi 2782d9585277SAl Viro if (error) { /* returned 1, that is */ 278330d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 27842675a4ebSAl Viro error = -EIO; 2785d18e9008SMiklos Szeredi goto out; 2786d18e9008SMiklos Szeredi } 278730d90494SAl Viro if (file->f_path.dentry) { 2788d18e9008SMiklos Szeredi dput(dentry); 278930d90494SAl Viro dentry = file->f_path.dentry; 2790d18e9008SMiklos Szeredi } 279103da633aSAl Viro if (*opened & FILE_CREATED) 279203da633aSAl Viro fsnotify_create(dir, dentry); 279303da633aSAl Viro if (!dentry->d_inode) { 279403da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 279503da633aSAl Viro if (create_error) { 279662b2ce96SSage Weil error = create_error; 279762b2ce96SSage Weil goto out; 279862b2ce96SSage Weil } 279903da633aSAl Viro } else { 280003da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 280103da633aSAl Viro error = -EEXIST; 280203da633aSAl Viro goto out; 280303da633aSAl Viro } 280403da633aSAl Viro } 2805d18e9008SMiklos Szeredi goto looked_up; 2806d18e9008SMiklos Szeredi } 2807d18e9008SMiklos Szeredi 2808d18e9008SMiklos Szeredi /* 2809d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2810d18e9008SMiklos Szeredi * here. 2811d18e9008SMiklos Szeredi */ 281203da633aSAl Viro acc_mode = op->acc_mode; 281303da633aSAl Viro if (*opened & FILE_CREATED) { 281403da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 281503da633aSAl Viro fsnotify_create(dir, dentry); 281603da633aSAl Viro acc_mode = MAY_OPEN; 281703da633aSAl Viro } 28182675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 28192675a4ebSAl Viro if (error) 28202675a4ebSAl Viro fput(file); 2821d18e9008SMiklos Szeredi 2822d18e9008SMiklos Szeredi out: 2823d18e9008SMiklos Szeredi dput(dentry); 28242675a4ebSAl Viro return error; 2825d18e9008SMiklos Szeredi 2826d18e9008SMiklos Szeredi no_open: 2827d18e9008SMiklos Szeredi if (need_lookup) { 282872bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2829d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 28302675a4ebSAl Viro return PTR_ERR(dentry); 2831d18e9008SMiklos Szeredi 2832d18e9008SMiklos Szeredi if (create_error) { 2833d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2834d18e9008SMiklos Szeredi 28352675a4ebSAl Viro error = create_error; 2836d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2837d18e9008SMiklos Szeredi if (!dentry->d_inode) 2838d18e9008SMiklos Szeredi goto out; 2839d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2840d18e9008SMiklos Szeredi goto out; 2841d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2842e36cb0b8SDavid Howells d_is_reg(dentry)) { 2843d18e9008SMiklos Szeredi goto out; 2844d18e9008SMiklos Szeredi } 2845d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2846d18e9008SMiklos Szeredi } 2847d18e9008SMiklos Szeredi } 2848d18e9008SMiklos Szeredi looked_up: 2849d18e9008SMiklos Szeredi path->dentry = dentry; 2850d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 28512675a4ebSAl Viro return 1; 2852d18e9008SMiklos Szeredi } 2853d18e9008SMiklos Szeredi 285431e6b01fSNick Piggin /* 28551acf0af9SDavid Howells * Look up and maybe create and open the last component. 2856d58ffd35SMiklos Szeredi * 2857d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2858d58ffd35SMiklos Szeredi * 28591acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 28601acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 28611acf0af9SDavid Howells * 28621acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 28631acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 28641acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 28651acf0af9SDavid Howells * specified then a negative dentry may be returned. 28661acf0af9SDavid Howells * 28671acf0af9SDavid Howells * An error code is returned otherwise. 28681acf0af9SDavid Howells * 28691acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 28701acf0af9SDavid Howells * cleared otherwise prior to returning. 2871d58ffd35SMiklos Szeredi */ 28722675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 287330d90494SAl Viro struct file *file, 2874d58ffd35SMiklos Szeredi const struct open_flags *op, 287564894cf8SAl Viro bool got_write, int *opened) 2876d58ffd35SMiklos Szeredi { 2877d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 287854ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2879d58ffd35SMiklos Szeredi struct dentry *dentry; 2880d58ffd35SMiklos Szeredi int error; 288154ef4872SMiklos Szeredi bool need_lookup; 2882d58ffd35SMiklos Szeredi 288347237687SAl Viro *opened &= ~FILE_CREATED; 2884201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2885d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 28862675a4ebSAl Viro return PTR_ERR(dentry); 2887d58ffd35SMiklos Szeredi 2888d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2889d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2890d18e9008SMiklos Szeredi goto out_no_open; 2891d18e9008SMiklos Szeredi 2892d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 289364894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 289447237687SAl Viro need_lookup, opened); 2895d18e9008SMiklos Szeredi } 2896d18e9008SMiklos Szeredi 289754ef4872SMiklos Szeredi if (need_lookup) { 289854ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 289954ef4872SMiklos Szeredi 290072bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 290154ef4872SMiklos Szeredi if (IS_ERR(dentry)) 29022675a4ebSAl Viro return PTR_ERR(dentry); 290354ef4872SMiklos Szeredi } 290454ef4872SMiklos Szeredi 2905d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2906d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2907d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2908d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2909d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2910d58ffd35SMiklos Szeredi /* 2911d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2912d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2913d58ffd35SMiklos Szeredi * the time when the file is created and when 2914d58ffd35SMiklos Szeredi * a permanent write count is taken through 2915015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2916d58ffd35SMiklos Szeredi */ 291764894cf8SAl Viro if (!got_write) { 291864894cf8SAl Viro error = -EROFS; 2919d58ffd35SMiklos Szeredi goto out_dput; 292064894cf8SAl Viro } 292147237687SAl Viro *opened |= FILE_CREATED; 2922d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2923d58ffd35SMiklos Szeredi if (error) 2924d58ffd35SMiklos Szeredi goto out_dput; 2925312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2926312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2927d58ffd35SMiklos Szeredi if (error) 2928d58ffd35SMiklos Szeredi goto out_dput; 2929d58ffd35SMiklos Szeredi } 2930d18e9008SMiklos Szeredi out_no_open: 2931d58ffd35SMiklos Szeredi path->dentry = dentry; 2932d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 29332675a4ebSAl Viro return 1; 2934d58ffd35SMiklos Szeredi 2935d58ffd35SMiklos Szeredi out_dput: 2936d58ffd35SMiklos Szeredi dput(dentry); 29372675a4ebSAl Viro return error; 2938d58ffd35SMiklos Szeredi } 2939d58ffd35SMiklos Szeredi 2940d58ffd35SMiklos Szeredi /* 2941fe2d35ffSAl Viro * Handle the last step of open() 294231e6b01fSNick Piggin */ 2943896475d5SAl Viro static int do_last(struct nameidata *nd, 294430d90494SAl Viro struct file *file, const struct open_flags *op, 2945669abf4eSJeff Layton int *opened, struct filename *name) 2946fb1cc555SAl Viro { 2947a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2948ca344a89SAl Viro int open_flag = op->open_flag; 294977d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 295064894cf8SAl Viro bool got_write = false; 2951bcda7652SAl Viro int acc_mode = op->acc_mode; 2952254cf582SAl Viro unsigned seq; 2953a1eb3315SMiklos Szeredi struct inode *inode; 295416b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 2955896475d5SAl Viro struct path path; 295616b1c1cdSMiklos Szeredi bool retried = false; 295716c2cd71SAl Viro int error; 2958fb1cc555SAl Viro 2959c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2960c3e380b0SAl Viro nd->flags |= op->intent; 2961c3e380b0SAl Viro 2962bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2963fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 2964deb106c6SAl Viro if (unlikely(error)) 29652675a4ebSAl Viro return error; 2966e83db167SMiklos Szeredi goto finish_open; 29671f36f774SAl Viro } 2968a2c36b45SAl Viro 2969ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2970fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2971fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2972fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2973254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 297471574865SMiklos Szeredi if (likely(!error)) 297571574865SMiklos Szeredi goto finish_lookup; 297671574865SMiklos Szeredi 2977ce57dfc1SAl Viro if (error < 0) 2978deb106c6SAl Viro return error; 2979a1eb3315SMiklos Szeredi 298037d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2981b6183df7SMiklos Szeredi } else { 2982fe2d35ffSAl Viro /* create side of things */ 2983a3fbbde7SAl Viro /* 2984b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 2985b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 2986b6183df7SMiklos Szeredi * about to look up 2987a3fbbde7SAl Viro */ 29889f1fafeeSAl Viro error = complete_walk(nd); 2989e8bb73dfSAl Viro if (error) 29902675a4ebSAl Viro return error; 2991fe2d35ffSAl Viro 299233e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 29931f36f774SAl Viro /* trailing slashes? */ 2994deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 2995deb106c6SAl Viro return -EISDIR; 2996b6183df7SMiklos Szeredi } 29971f36f774SAl Viro 299816b1c1cdSMiklos Szeredi retry_lookup: 299964894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 300064894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 300164894cf8SAl Viro if (!error) 300264894cf8SAl Viro got_write = true; 300364894cf8SAl Viro /* 300464894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 300564894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 300664894cf8SAl Viro * dropping this one anyway. 300764894cf8SAl Viro */ 300864894cf8SAl Viro } 3009a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 3010896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 3011fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 3012fb1cc555SAl Viro 30132675a4ebSAl Viro if (error <= 0) { 30142675a4ebSAl Viro if (error) 3015d58ffd35SMiklos Szeredi goto out; 30166c0d46c4SAl Viro 301747237687SAl Viro if ((*opened & FILE_CREATED) || 3018496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 301977d660a8SMiklos Szeredi will_truncate = false; 3020d18e9008SMiklos Szeredi 3021adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 3022d18e9008SMiklos Szeredi goto opened; 3023d18e9008SMiklos Szeredi } 3024d18e9008SMiklos Szeredi 302547237687SAl Viro if (*opened & FILE_CREATED) { 30269b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3027ca344a89SAl Viro open_flag &= ~O_TRUNC; 302877d660a8SMiklos Szeredi will_truncate = false; 3029bcda7652SAl Viro acc_mode = MAY_OPEN; 3030896475d5SAl Viro path_to_nameidata(&path, nd); 3031e83db167SMiklos Szeredi goto finish_open_created; 3032fb1cc555SAl Viro } 3033fb1cc555SAl Viro 3034fb1cc555SAl Viro /* 30353134f37eSJeff Layton * create/update audit record if it already exists. 3036fb1cc555SAl Viro */ 3037896475d5SAl Viro if (d_is_positive(path.dentry)) 3038896475d5SAl Viro audit_inode(name, path.dentry, 0); 3039fb1cc555SAl Viro 3040d18e9008SMiklos Szeredi /* 3041d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3042d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3043d18e9008SMiklos Szeredi * necessary...) 3044d18e9008SMiklos Szeredi */ 304564894cf8SAl Viro if (got_write) { 3046d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 304764894cf8SAl Viro got_write = false; 3048d18e9008SMiklos Szeredi } 3049d18e9008SMiklos Szeredi 3050deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3051deb106c6SAl Viro path_to_nameidata(&path, nd); 3052deb106c6SAl Viro return -EEXIST; 3053deb106c6SAl Viro } 3054fb1cc555SAl Viro 3055756daf26SNeilBrown error = follow_managed(&path, nd); 3056deb106c6SAl Viro if (unlikely(error < 0)) 3057deb106c6SAl Viro return error; 3058a3fbbde7SAl Viro 3059decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 306063afdfc7SDavid Howells inode = d_backing_inode(path.dentry); 3061254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3062deb106c6SAl Viro if (unlikely(d_is_negative(path.dentry))) { 3063896475d5SAl Viro path_to_nameidata(&path, nd); 3064deb106c6SAl Viro return -ENOENT; 306554c33e7fSMiklos Szeredi } 3066766c4cbfSAl Viro finish_lookup: 3067191d7f73SAl Viro if (nd->depth) 3068191d7f73SAl Viro put_link(nd); 3069181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3070181548c0SAl Viro inode, seq); 3071deb106c6SAl Viro if (unlikely(error)) 3072d63ff28fSAl Viro return error; 3073fb1cc555SAl Viro 3074896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3075896475d5SAl Viro path_to_nameidata(&path, nd); 3076deb106c6SAl Viro return -ELOOP; 3077a5cfe2d5SAl Viro } 3078a5cfe2d5SAl Viro 3079896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3080896475d5SAl Viro path_to_nameidata(&path, nd); 308116b1c1cdSMiklos Szeredi } else { 308216b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3083896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3084896475d5SAl Viro nd->path.dentry = path.dentry; 308516b1c1cdSMiklos Szeredi 308616b1c1cdSMiklos Szeredi } 3087decf3400SMiklos Szeredi nd->inode = inode; 3088254cf582SAl Viro nd->seq = seq; 3089a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3090bc77daa7SAl Viro finish_open: 3091a3fbbde7SAl Viro error = complete_walk(nd); 309216b1c1cdSMiklos Szeredi if (error) { 309316b1c1cdSMiklos Szeredi path_put(&save_parent); 30942675a4ebSAl Viro return error; 309516b1c1cdSMiklos Szeredi } 3096bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3097fb1cc555SAl Viro error = -EISDIR; 309844b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 30992675a4ebSAl Viro goto out; 3100af2f5542SMiklos Szeredi error = -ENOTDIR; 310144b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 31022675a4ebSAl Viro goto out; 31034bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 310477d660a8SMiklos Szeredi will_truncate = false; 31056c0d46c4SAl Viro 31060f9d1a10SAl Viro if (will_truncate) { 31070f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 31080f9d1a10SAl Viro if (error) 31092675a4ebSAl Viro goto out; 311064894cf8SAl Viro got_write = true; 31110f9d1a10SAl Viro } 3112e83db167SMiklos Szeredi finish_open_created: 3113bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3114ca344a89SAl Viro if (error) 31152675a4ebSAl Viro goto out; 31164aa7c634SMiklos Szeredi 31174aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31184aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 31194aa7c634SMiklos Szeredi if (!error) { 31204aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31214aa7c634SMiklos Szeredi } else { 312230d90494SAl Viro if (error == -EOPENSTALE) 3123f60dc3dbSMiklos Szeredi goto stale_open; 3124015c3bbcSMiklos Szeredi goto out; 3125f60dc3dbSMiklos Szeredi } 3126a8277b9bSMiklos Szeredi opened: 31272675a4ebSAl Viro error = open_check_o_direct(file); 3128015c3bbcSMiklos Szeredi if (error) 3129015c3bbcSMiklos Szeredi goto exit_fput; 31303034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3131aa4caadbSMiklos Szeredi if (error) 3132aa4caadbSMiklos Szeredi goto exit_fput; 3133aa4caadbSMiklos Szeredi 31340f9d1a10SAl Viro if (will_truncate) { 31352675a4ebSAl Viro error = handle_truncate(file); 3136aa4caadbSMiklos Szeredi if (error) 3137aa4caadbSMiklos Szeredi goto exit_fput; 31380f9d1a10SAl Viro } 3139ca344a89SAl Viro out: 314064894cf8SAl Viro if (got_write) 31410f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 314216b1c1cdSMiklos Szeredi path_put(&save_parent); 31432675a4ebSAl Viro return error; 3144fb1cc555SAl Viro 3145015c3bbcSMiklos Szeredi exit_fput: 31462675a4ebSAl Viro fput(file); 31472675a4ebSAl Viro goto out; 3148015c3bbcSMiklos Szeredi 3149f60dc3dbSMiklos Szeredi stale_open: 3150f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3151f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3152f60dc3dbSMiklos Szeredi goto out; 3153f60dc3dbSMiklos Szeredi 3154f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3155f60dc3dbSMiklos Szeredi path_put(&nd->path); 3156f60dc3dbSMiklos Szeredi nd->path = save_parent; 3157f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3158f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3159f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 316064894cf8SAl Viro if (got_write) { 3161f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 316264894cf8SAl Viro got_write = false; 3163f60dc3dbSMiklos Szeredi } 3164f60dc3dbSMiklos Szeredi retried = true; 3165f60dc3dbSMiklos Szeredi goto retry_lookup; 3166fb1cc555SAl Viro } 3167fb1cc555SAl Viro 316860545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 316960545d0dSAl Viro struct nameidata *nd, int flags, 317060545d0dSAl Viro const struct open_flags *op, 317160545d0dSAl Viro struct file *file, int *opened) 317260545d0dSAl Viro { 317360545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 317460545d0dSAl Viro struct dentry *dentry, *child; 317560545d0dSAl Viro struct inode *dir; 31765eb6b495SAl Viro int error = path_lookupat(dfd, pathname, 317760545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 317860545d0dSAl Viro if (unlikely(error)) 317960545d0dSAl Viro return error; 318060545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 318160545d0dSAl Viro if (unlikely(error)) 318260545d0dSAl Viro goto out; 318360545d0dSAl Viro /* we want directory to be writable */ 318460545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 318560545d0dSAl Viro if (error) 318660545d0dSAl Viro goto out2; 318760545d0dSAl Viro dentry = nd->path.dentry; 318860545d0dSAl Viro dir = dentry->d_inode; 318960545d0dSAl Viro if (!dir->i_op->tmpfile) { 319060545d0dSAl Viro error = -EOPNOTSUPP; 319160545d0dSAl Viro goto out2; 319260545d0dSAl Viro } 319360545d0dSAl Viro child = d_alloc(dentry, &name); 319460545d0dSAl Viro if (unlikely(!child)) { 319560545d0dSAl Viro error = -ENOMEM; 319660545d0dSAl Viro goto out2; 319760545d0dSAl Viro } 319860545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 319960545d0dSAl Viro nd->flags |= op->intent; 320060545d0dSAl Viro dput(nd->path.dentry); 320160545d0dSAl Viro nd->path.dentry = child; 320260545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 320360545d0dSAl Viro if (error) 320460545d0dSAl Viro goto out2; 320560545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 320669a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 320769a91c23SEric Rannaud error = may_open(&nd->path, MAY_OPEN, op->open_flag); 320860545d0dSAl Viro if (error) 320960545d0dSAl Viro goto out2; 321060545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 321160545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 321260545d0dSAl Viro if (error) 321360545d0dSAl Viro goto out2; 321460545d0dSAl Viro error = open_check_o_direct(file); 3215f4e0c30cSAl Viro if (error) { 321660545d0dSAl Viro fput(file); 3217f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3218f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3219f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3220f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3221f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3222f4e0c30cSAl Viro } 322360545d0dSAl Viro out2: 322460545d0dSAl Viro mnt_drop_write(nd->path.mnt); 322560545d0dSAl Viro out: 322660545d0dSAl Viro path_put(&nd->path); 322760545d0dSAl Viro return error; 322860545d0dSAl Viro } 322960545d0dSAl Viro 3230669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 323173d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 32321da177e4SLinus Torvalds { 3233368ee9baSAl Viro const char *s; 323430d90494SAl Viro struct file *file; 323547237687SAl Viro int opened = 0; 323613aab428SAl Viro int error; 323731e6b01fSNick Piggin 323830d90494SAl Viro file = get_empty_filp(); 32391afc99beSAl Viro if (IS_ERR(file)) 32401afc99beSAl Viro return file; 324131e6b01fSNick Piggin 324230d90494SAl Viro file->f_flags = op->open_flag; 324331e6b01fSNick Piggin 3244bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 324560545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 3246f15133dfSAl Viro goto out2; 324760545d0dSAl Viro } 324860545d0dSAl Viro 3249368ee9baSAl Viro s = path_init(dfd, pathname, flags, nd); 3250368ee9baSAl Viro if (IS_ERR(s)) { 3251368ee9baSAl Viro put_filp(file); 3252368ee9baSAl Viro return ERR_CAST(s); 3253368ee9baSAl Viro } 32543bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 32553bdba28bSAl Viro (error = do_last(nd, file, op, &opened, pathname)) > 0) { 325673d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 32573bdba28bSAl Viro s = trailing_symlink(nd); 32583bdba28bSAl Viro if (IS_ERR(s)) { 32593bdba28bSAl Viro error = PTR_ERR(s); 32602675a4ebSAl Viro break; 3261806b681cSAl Viro } 32623bdba28bSAl Viro } 3263deb106c6SAl Viro terminate_walk(nd); 3264893b7775SAl Viro path_cleanup(nd); 3265f15133dfSAl Viro out2: 32662675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32672675a4ebSAl Viro BUG_ON(!error); 326830d90494SAl Viro put_filp(file); 3269015c3bbcSMiklos Szeredi } 32702675a4ebSAl Viro if (unlikely(error)) { 32712675a4ebSAl Viro if (error == -EOPENSTALE) { 32722675a4ebSAl Viro if (flags & LOOKUP_RCU) 32732675a4ebSAl Viro error = -ECHILD; 32742675a4ebSAl Viro else 32752675a4ebSAl Viro error = -ESTALE; 32762675a4ebSAl Viro } 32772675a4ebSAl Viro file = ERR_PTR(error); 32782675a4ebSAl Viro } 32792675a4ebSAl Viro return file; 3280de459215SKirill Korotaev } 32811da177e4SLinus Torvalds 3282669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3283f9652e10SAl Viro const struct open_flags *op) 328413aab428SAl Viro { 3285756daf26SNeilBrown struct nameidata nd, *saved_nd = set_nameidata(&nd); 3286f9652e10SAl Viro int flags = op->lookup_flags; 328713aab428SAl Viro struct file *filp; 328813aab428SAl Viro 328973d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 329013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 329173d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 329213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 329373d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 3294756daf26SNeilBrown restore_nameidata(saved_nd); 329513aab428SAl Viro return filp; 329613aab428SAl Viro } 329713aab428SAl Viro 329873d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3299f9652e10SAl Viro const char *name, const struct open_flags *op) 330073d049a4SAl Viro { 3301756daf26SNeilBrown struct nameidata nd, *saved_nd; 330273d049a4SAl Viro struct file *file; 330351689104SPaul Moore struct filename *filename; 3304f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 330573d049a4SAl Viro 330673d049a4SAl Viro nd.root.mnt = mnt; 330773d049a4SAl Viro nd.root.dentry = dentry; 330873d049a4SAl Viro 3309b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 331073d049a4SAl Viro return ERR_PTR(-ELOOP); 331173d049a4SAl Viro 331251689104SPaul Moore filename = getname_kernel(name); 331351689104SPaul Moore if (unlikely(IS_ERR(filename))) 331451689104SPaul Moore return ERR_CAST(filename); 331551689104SPaul Moore 3316756daf26SNeilBrown saved_nd = set_nameidata(&nd); 331751689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU); 331873d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 331951689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags); 332073d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 332151689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL); 3322756daf26SNeilBrown restore_nameidata(saved_nd); 332351689104SPaul Moore putname(filename); 332473d049a4SAl Viro return file; 332573d049a4SAl Viro } 332673d049a4SAl Viro 3327fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 33281ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33291da177e4SLinus Torvalds { 3330c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3331ed75e95dSAl Viro struct nameidata nd; 3332c30dabfeSJan Kara int err2; 33331ac12b4bSJeff Layton int error; 33341ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 33351ac12b4bSJeff Layton 33361ac12b4bSJeff Layton /* 33371ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 33381ac12b4bSJeff Layton * other flags passed in are ignored! 33391ac12b4bSJeff Layton */ 33401ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 33411ac12b4bSJeff Layton 33428bcb77faSAl Viro error = filename_parentat(dfd, name, lookup_flags, &nd); 3343ed75e95dSAl Viro if (error) 3344ed75e95dSAl Viro return ERR_PTR(error); 33451da177e4SLinus Torvalds 3346c663e5d8SChristoph Hellwig /* 3347c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3348c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3349c663e5d8SChristoph Hellwig */ 3350ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3351ed75e95dSAl Viro goto out; 3352ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3353ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3354c663e5d8SChristoph Hellwig 3355c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3356c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3357c663e5d8SChristoph Hellwig /* 3358c663e5d8SChristoph Hellwig * Do the final lookup. 3359c663e5d8SChristoph Hellwig */ 3360ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 33616a9f40d6SAl Viro dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags); 33621da177e4SLinus Torvalds if (IS_ERR(dentry)) 3363a8104a9fSAl Viro goto unlock; 3364c663e5d8SChristoph Hellwig 3365a8104a9fSAl Viro error = -EEXIST; 3366b18825a7SDavid Howells if (d_is_positive(dentry)) 3367a8104a9fSAl Viro goto fail; 3368b18825a7SDavid Howells 3369c663e5d8SChristoph Hellwig /* 3370c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3371c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3372c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3373c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3374c663e5d8SChristoph Hellwig */ 3375ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3376a8104a9fSAl Viro error = -ENOENT; 3377ed75e95dSAl Viro goto fail; 3378e9baf6e5SAl Viro } 3379c30dabfeSJan Kara if (unlikely(err2)) { 3380c30dabfeSJan Kara error = err2; 3381a8104a9fSAl Viro goto fail; 3382c30dabfeSJan Kara } 3383ed75e95dSAl Viro *path = nd.path; 3384e9baf6e5SAl Viro return dentry; 33851da177e4SLinus Torvalds fail: 3386a8104a9fSAl Viro dput(dentry); 3387a8104a9fSAl Viro dentry = ERR_PTR(error); 3388a8104a9fSAl Viro unlock: 3389dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3390c30dabfeSJan Kara if (!err2) 3391c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3392ed75e95dSAl Viro out: 3393dae6ad8fSAl Viro path_put(&nd.path); 3394ed75e95dSAl Viro return dentry; 3395dae6ad8fSAl Viro } 3396fa14a0b8SAl Viro 3397fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3398fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3399fa14a0b8SAl Viro { 340051689104SPaul Moore struct filename *filename = getname_kernel(pathname); 340151689104SPaul Moore struct dentry *res; 340251689104SPaul Moore 340351689104SPaul Moore if (IS_ERR(filename)) 340451689104SPaul Moore return ERR_CAST(filename); 340551689104SPaul Moore res = filename_create(dfd, filename, path, lookup_flags); 340651689104SPaul Moore putname(filename); 340751689104SPaul Moore return res; 3408fa14a0b8SAl Viro } 3409dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3410dae6ad8fSAl Viro 3411921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3412921a1650SAl Viro { 3413921a1650SAl Viro dput(dentry); 3414921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3415a8104a9fSAl Viro mnt_drop_write(path->mnt); 3416921a1650SAl Viro path_put(path); 3417921a1650SAl Viro } 3418921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3419921a1650SAl Viro 34201ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 34211ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3422dae6ad8fSAl Viro { 342391a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3424dae6ad8fSAl Viro struct dentry *res; 3425dae6ad8fSAl Viro if (IS_ERR(tmp)) 3426dae6ad8fSAl Viro return ERR_CAST(tmp); 3427fa14a0b8SAl Viro res = filename_create(dfd, tmp, path, lookup_flags); 3428dae6ad8fSAl Viro putname(tmp); 3429dae6ad8fSAl Viro return res; 3430dae6ad8fSAl Viro } 3431dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3432dae6ad8fSAl Viro 34331a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34341da177e4SLinus Torvalds { 3435a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34361da177e4SLinus Torvalds 34371da177e4SLinus Torvalds if (error) 34381da177e4SLinus Torvalds return error; 34391da177e4SLinus Torvalds 3440975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34411da177e4SLinus Torvalds return -EPERM; 34421da177e4SLinus Torvalds 3443acfa4380SAl Viro if (!dir->i_op->mknod) 34441da177e4SLinus Torvalds return -EPERM; 34451da177e4SLinus Torvalds 344608ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 344708ce5f16SSerge E. Hallyn if (error) 344808ce5f16SSerge E. Hallyn return error; 344908ce5f16SSerge E. Hallyn 34501da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 34511da177e4SLinus Torvalds if (error) 34521da177e4SLinus Torvalds return error; 34531da177e4SLinus Torvalds 34541da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3455a74574aaSStephen Smalley if (!error) 3456f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34571da177e4SLinus Torvalds return error; 34581da177e4SLinus Torvalds } 34594d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 34601da177e4SLinus Torvalds 3461f69aac00SAl Viro static int may_mknod(umode_t mode) 3462463c3197SDave Hansen { 3463463c3197SDave Hansen switch (mode & S_IFMT) { 3464463c3197SDave Hansen case S_IFREG: 3465463c3197SDave Hansen case S_IFCHR: 3466463c3197SDave Hansen case S_IFBLK: 3467463c3197SDave Hansen case S_IFIFO: 3468463c3197SDave Hansen case S_IFSOCK: 3469463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3470463c3197SDave Hansen return 0; 3471463c3197SDave Hansen case S_IFDIR: 3472463c3197SDave Hansen return -EPERM; 3473463c3197SDave Hansen default: 3474463c3197SDave Hansen return -EINVAL; 3475463c3197SDave Hansen } 3476463c3197SDave Hansen } 3477463c3197SDave Hansen 34788208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 34792e4d0924SHeiko Carstens unsigned, dev) 34801da177e4SLinus Torvalds { 34811da177e4SLinus Torvalds struct dentry *dentry; 3482dae6ad8fSAl Viro struct path path; 3483dae6ad8fSAl Viro int error; 3484972567f1SJeff Layton unsigned int lookup_flags = 0; 34851da177e4SLinus Torvalds 34868e4bfca1SAl Viro error = may_mknod(mode); 34878e4bfca1SAl Viro if (error) 34888e4bfca1SAl Viro return error; 3489972567f1SJeff Layton retry: 3490972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3491dae6ad8fSAl Viro if (IS_ERR(dentry)) 3492dae6ad8fSAl Viro return PTR_ERR(dentry); 34932ad94ae6SAl Viro 3494dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3495ce3b0f8dSAl Viro mode &= ~current_umask(); 3496dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3497be6d3e56SKentaro Takeda if (error) 3498a8104a9fSAl Viro goto out; 34991da177e4SLinus Torvalds switch (mode & S_IFMT) { 35001da177e4SLinus Torvalds case 0: case S_IFREG: 3501312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 35021da177e4SLinus Torvalds break; 35031da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3504dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 35051da177e4SLinus Torvalds new_decode_dev(dev)); 35061da177e4SLinus Torvalds break; 35071da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3508dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 35091da177e4SLinus Torvalds break; 35101da177e4SLinus Torvalds } 3511a8104a9fSAl Viro out: 3512921a1650SAl Viro done_path_create(&path, dentry); 3513972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3514972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3515972567f1SJeff Layton goto retry; 3516972567f1SJeff Layton } 35171da177e4SLinus Torvalds return error; 35181da177e4SLinus Torvalds } 35191da177e4SLinus Torvalds 35208208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35215590ff0dSUlrich Drepper { 35225590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35235590ff0dSUlrich Drepper } 35245590ff0dSUlrich Drepper 352518bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35261da177e4SLinus Torvalds { 3527a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35288de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35291da177e4SLinus Torvalds 35301da177e4SLinus Torvalds if (error) 35311da177e4SLinus Torvalds return error; 35321da177e4SLinus Torvalds 3533acfa4380SAl Viro if (!dir->i_op->mkdir) 35341da177e4SLinus Torvalds return -EPERM; 35351da177e4SLinus Torvalds 35361da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35371da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35381da177e4SLinus Torvalds if (error) 35391da177e4SLinus Torvalds return error; 35401da177e4SLinus Torvalds 35418de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 35428de52778SAl Viro return -EMLINK; 35438de52778SAl Viro 35441da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3545a74574aaSStephen Smalley if (!error) 3546f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 35471da177e4SLinus Torvalds return error; 35481da177e4SLinus Torvalds } 35494d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 35501da177e4SLinus Torvalds 3551a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 35521da177e4SLinus Torvalds { 35536902d925SDave Hansen struct dentry *dentry; 3554dae6ad8fSAl Viro struct path path; 3555dae6ad8fSAl Viro int error; 3556b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 35571da177e4SLinus Torvalds 3558b76d8b82SJeff Layton retry: 3559b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 35606902d925SDave Hansen if (IS_ERR(dentry)) 3561dae6ad8fSAl Viro return PTR_ERR(dentry); 35626902d925SDave Hansen 3563dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3564ce3b0f8dSAl Viro mode &= ~current_umask(); 3565dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3566a8104a9fSAl Viro if (!error) 3567dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3568921a1650SAl Viro done_path_create(&path, dentry); 3569b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3570b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3571b76d8b82SJeff Layton goto retry; 3572b76d8b82SJeff Layton } 35731da177e4SLinus Torvalds return error; 35741da177e4SLinus Torvalds } 35751da177e4SLinus Torvalds 3576a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 35775590ff0dSUlrich Drepper { 35785590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 35795590ff0dSUlrich Drepper } 35805590ff0dSUlrich Drepper 35811da177e4SLinus Torvalds /* 3582a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3583c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3584a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3585a71905f0SSage Weil * then we drop the dentry now. 35861da177e4SLinus Torvalds * 35871da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 35881da177e4SLinus Torvalds * do a 35891da177e4SLinus Torvalds * 35901da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 35911da177e4SLinus Torvalds * return -EBUSY; 35921da177e4SLinus Torvalds * 35931da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 35941da177e4SLinus Torvalds * that is still in use by something else.. 35951da177e4SLinus Torvalds */ 35961da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 35971da177e4SLinus Torvalds { 35981da177e4SLinus Torvalds shrink_dcache_parent(dentry); 35991da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 360098474236SWaiman Long if (dentry->d_lockref.count == 1) 36011da177e4SLinus Torvalds __d_drop(dentry); 36021da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 36031da177e4SLinus Torvalds } 36044d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 36051da177e4SLinus Torvalds 36061da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36071da177e4SLinus Torvalds { 36081da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36091da177e4SLinus Torvalds 36101da177e4SLinus Torvalds if (error) 36111da177e4SLinus Torvalds return error; 36121da177e4SLinus Torvalds 3613acfa4380SAl Viro if (!dir->i_op->rmdir) 36141da177e4SLinus Torvalds return -EPERM; 36151da177e4SLinus Torvalds 36161d2ef590SAl Viro dget(dentry); 36171b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3618912dbc15SSage Weil 36191da177e4SLinus Torvalds error = -EBUSY; 36207af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3621912dbc15SSage Weil goto out; 3622912dbc15SSage Weil 36231da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3624912dbc15SSage Weil if (error) 3625912dbc15SSage Weil goto out; 3626912dbc15SSage Weil 36273cebde24SSage Weil shrink_dcache_parent(dentry); 36281da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3629912dbc15SSage Weil if (error) 3630912dbc15SSage Weil goto out; 3631912dbc15SSage Weil 36321da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3633d83c49f3SAl Viro dont_mount(dentry); 36348ed936b5SEric W. Biederman detach_mounts(dentry); 36351da177e4SLinus Torvalds 3636912dbc15SSage Weil out: 3637912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36381d2ef590SAl Viro dput(dentry); 3639912dbc15SSage Weil if (!error) 3640912dbc15SSage Weil d_delete(dentry); 36411da177e4SLinus Torvalds return error; 36421da177e4SLinus Torvalds } 36434d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 36441da177e4SLinus Torvalds 36455590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 36461da177e4SLinus Torvalds { 36471da177e4SLinus Torvalds int error = 0; 364891a27b2aSJeff Layton struct filename *name; 36491da177e4SLinus Torvalds struct dentry *dentry; 3650f5beed75SAl Viro struct path path; 3651f5beed75SAl Viro struct qstr last; 3652f5beed75SAl Viro int type; 3653c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3654c6ee9206SJeff Layton retry: 3655f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3656f5beed75SAl Viro &path, &last, &type, lookup_flags); 365791a27b2aSJeff Layton if (IS_ERR(name)) 365891a27b2aSJeff Layton return PTR_ERR(name); 36591da177e4SLinus Torvalds 3660f5beed75SAl Viro switch (type) { 36611da177e4SLinus Torvalds case LAST_DOTDOT: 36621da177e4SLinus Torvalds error = -ENOTEMPTY; 36631da177e4SLinus Torvalds goto exit1; 36641da177e4SLinus Torvalds case LAST_DOT: 36651da177e4SLinus Torvalds error = -EINVAL; 36661da177e4SLinus Torvalds goto exit1; 36671da177e4SLinus Torvalds case LAST_ROOT: 36681da177e4SLinus Torvalds error = -EBUSY; 36691da177e4SLinus Torvalds goto exit1; 36701da177e4SLinus Torvalds } 36710612d9fbSOGAWA Hirofumi 3672f5beed75SAl Viro error = mnt_want_write(path.mnt); 3673c30dabfeSJan Kara if (error) 3674c30dabfeSJan Kara goto exit1; 36750612d9fbSOGAWA Hirofumi 3676f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3677f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 36781da177e4SLinus Torvalds error = PTR_ERR(dentry); 36796902d925SDave Hansen if (IS_ERR(dentry)) 36806902d925SDave Hansen goto exit2; 3681e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3682e6bc45d6STheodore Ts'o error = -ENOENT; 3683e6bc45d6STheodore Ts'o goto exit3; 3684e6bc45d6STheodore Ts'o } 3685f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3686be6d3e56SKentaro Takeda if (error) 3687c30dabfeSJan Kara goto exit3; 3688f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 36890622753bSDave Hansen exit3: 36901da177e4SLinus Torvalds dput(dentry); 36916902d925SDave Hansen exit2: 3692f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3693f5beed75SAl Viro mnt_drop_write(path.mnt); 36941da177e4SLinus Torvalds exit1: 3695f5beed75SAl Viro path_put(&path); 36961da177e4SLinus Torvalds putname(name); 3697c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3698c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3699c6ee9206SJeff Layton goto retry; 3700c6ee9206SJeff Layton } 37011da177e4SLinus Torvalds return error; 37021da177e4SLinus Torvalds } 37031da177e4SLinus Torvalds 37043cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37055590ff0dSUlrich Drepper { 37065590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 37075590ff0dSUlrich Drepper } 37085590ff0dSUlrich Drepper 3709b21996e3SJ. Bruce Fields /** 3710b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3711b21996e3SJ. Bruce Fields * @dir: parent directory 3712b21996e3SJ. Bruce Fields * @dentry: victim 3713b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3714b21996e3SJ. Bruce Fields * 3715b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3716b21996e3SJ. Bruce Fields * 3717b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3718b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3719b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3720b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3721b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3722b21996e3SJ. Bruce Fields * 3723b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3724b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3725b21996e3SJ. Bruce Fields * to be NFS exported. 3726b21996e3SJ. Bruce Fields */ 3727b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37281da177e4SLinus Torvalds { 37299accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37301da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37311da177e4SLinus Torvalds 37321da177e4SLinus Torvalds if (error) 37331da177e4SLinus Torvalds return error; 37341da177e4SLinus Torvalds 3735acfa4380SAl Viro if (!dir->i_op->unlink) 37361da177e4SLinus Torvalds return -EPERM; 37371da177e4SLinus Torvalds 37389accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37398ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37401da177e4SLinus Torvalds error = -EBUSY; 37411da177e4SLinus Torvalds else { 37421da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3743bec1052eSAl Viro if (!error) { 37445a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 37455a14696cSJ. Bruce Fields if (error) 3746b21996e3SJ. Bruce Fields goto out; 37471da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 37488ed936b5SEric W. Biederman if (!error) { 3749d83c49f3SAl Viro dont_mount(dentry); 37508ed936b5SEric W. Biederman detach_mounts(dentry); 37518ed936b5SEric W. Biederman } 3752bec1052eSAl Viro } 37531da177e4SLinus Torvalds } 3754b21996e3SJ. Bruce Fields out: 37559accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 37561da177e4SLinus Torvalds 37571da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 37581da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 37599accbb97SJ. Bruce Fields fsnotify_link_count(target); 37601da177e4SLinus Torvalds d_delete(dentry); 37611da177e4SLinus Torvalds } 37620eeca283SRobert Love 37631da177e4SLinus Torvalds return error; 37641da177e4SLinus Torvalds } 37654d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 37661da177e4SLinus Torvalds 37671da177e4SLinus Torvalds /* 37681da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 37691b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 37701da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 37711da177e4SLinus Torvalds * while waiting on the I/O. 37721da177e4SLinus Torvalds */ 37735590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 37741da177e4SLinus Torvalds { 37752ad94ae6SAl Viro int error; 377691a27b2aSJeff Layton struct filename *name; 37771da177e4SLinus Torvalds struct dentry *dentry; 3778f5beed75SAl Viro struct path path; 3779f5beed75SAl Viro struct qstr last; 3780f5beed75SAl Viro int type; 37811da177e4SLinus Torvalds struct inode *inode = NULL; 3782b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 37835d18f813SJeff Layton unsigned int lookup_flags = 0; 37845d18f813SJeff Layton retry: 3785f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3786f5beed75SAl Viro &path, &last, &type, lookup_flags); 378791a27b2aSJeff Layton if (IS_ERR(name)) 378891a27b2aSJeff Layton return PTR_ERR(name); 37892ad94ae6SAl Viro 37901da177e4SLinus Torvalds error = -EISDIR; 3791f5beed75SAl Viro if (type != LAST_NORM) 37921da177e4SLinus Torvalds goto exit1; 37930612d9fbSOGAWA Hirofumi 3794f5beed75SAl Viro error = mnt_want_write(path.mnt); 3795c30dabfeSJan Kara if (error) 3796c30dabfeSJan Kara goto exit1; 3797b21996e3SJ. Bruce Fields retry_deleg: 3798f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3799f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38001da177e4SLinus Torvalds error = PTR_ERR(dentry); 38011da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38021da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3803f5beed75SAl Viro if (last.name[last.len]) 380450338b88STörök Edwin goto slashes; 38051da177e4SLinus Torvalds inode = dentry->d_inode; 3806b18825a7SDavid Howells if (d_is_negative(dentry)) 3807e6bc45d6STheodore Ts'o goto slashes; 38087de9c6eeSAl Viro ihold(inode); 3809f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3810be6d3e56SKentaro Takeda if (error) 3811c30dabfeSJan Kara goto exit2; 3812f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38131da177e4SLinus Torvalds exit2: 38141da177e4SLinus Torvalds dput(dentry); 38151da177e4SLinus Torvalds } 3816f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38171da177e4SLinus Torvalds if (inode) 38181da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3819b21996e3SJ. Bruce Fields inode = NULL; 3820b21996e3SJ. Bruce Fields if (delegated_inode) { 38215a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3822b21996e3SJ. Bruce Fields if (!error) 3823b21996e3SJ. Bruce Fields goto retry_deleg; 3824b21996e3SJ. Bruce Fields } 3825f5beed75SAl Viro mnt_drop_write(path.mnt); 38261da177e4SLinus Torvalds exit1: 3827f5beed75SAl Viro path_put(&path); 38281da177e4SLinus Torvalds putname(name); 38295d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38305d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38315d18f813SJeff Layton inode = NULL; 38325d18f813SJeff Layton goto retry; 38335d18f813SJeff Layton } 38341da177e4SLinus Torvalds return error; 38351da177e4SLinus Torvalds 38361da177e4SLinus Torvalds slashes: 3837b18825a7SDavid Howells if (d_is_negative(dentry)) 3838b18825a7SDavid Howells error = -ENOENT; 383944b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3840b18825a7SDavid Howells error = -EISDIR; 3841b18825a7SDavid Howells else 3842b18825a7SDavid Howells error = -ENOTDIR; 38431da177e4SLinus Torvalds goto exit2; 38441da177e4SLinus Torvalds } 38451da177e4SLinus Torvalds 38462e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 38475590ff0dSUlrich Drepper { 38485590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 38495590ff0dSUlrich Drepper return -EINVAL; 38505590ff0dSUlrich Drepper 38515590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 38525590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 38535590ff0dSUlrich Drepper 38545590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 38555590ff0dSUlrich Drepper } 38565590ff0dSUlrich Drepper 38573480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 38585590ff0dSUlrich Drepper { 38595590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 38605590ff0dSUlrich Drepper } 38615590ff0dSUlrich Drepper 3862db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 38631da177e4SLinus Torvalds { 3864a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38651da177e4SLinus Torvalds 38661da177e4SLinus Torvalds if (error) 38671da177e4SLinus Torvalds return error; 38681da177e4SLinus Torvalds 3869acfa4380SAl Viro if (!dir->i_op->symlink) 38701da177e4SLinus Torvalds return -EPERM; 38711da177e4SLinus Torvalds 38721da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 38731da177e4SLinus Torvalds if (error) 38741da177e4SLinus Torvalds return error; 38751da177e4SLinus Torvalds 38761da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3877a74574aaSStephen Smalley if (!error) 3878f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38791da177e4SLinus Torvalds return error; 38801da177e4SLinus Torvalds } 38814d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 38821da177e4SLinus Torvalds 38832e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 38842e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 38851da177e4SLinus Torvalds { 38862ad94ae6SAl Viro int error; 388791a27b2aSJeff Layton struct filename *from; 38886902d925SDave Hansen struct dentry *dentry; 3889dae6ad8fSAl Viro struct path path; 3890f46d3567SJeff Layton unsigned int lookup_flags = 0; 38911da177e4SLinus Torvalds 38921da177e4SLinus Torvalds from = getname(oldname); 38931da177e4SLinus Torvalds if (IS_ERR(from)) 38941da177e4SLinus Torvalds return PTR_ERR(from); 3895f46d3567SJeff Layton retry: 3896f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 38971da177e4SLinus Torvalds error = PTR_ERR(dentry); 38986902d925SDave Hansen if (IS_ERR(dentry)) 3899dae6ad8fSAl Viro goto out_putname; 39006902d925SDave Hansen 390191a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3902a8104a9fSAl Viro if (!error) 390391a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3904921a1650SAl Viro done_path_create(&path, dentry); 3905f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3906f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3907f46d3567SJeff Layton goto retry; 3908f46d3567SJeff Layton } 39096902d925SDave Hansen out_putname: 39101da177e4SLinus Torvalds putname(from); 39111da177e4SLinus Torvalds return error; 39121da177e4SLinus Torvalds } 39131da177e4SLinus Torvalds 39143480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39155590ff0dSUlrich Drepper { 39165590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39175590ff0dSUlrich Drepper } 39185590ff0dSUlrich Drepper 3919146a8595SJ. Bruce Fields /** 3920146a8595SJ. Bruce Fields * vfs_link - create a new link 3921146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3922146a8595SJ. Bruce Fields * @dir: new parent 3923146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3924146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3925146a8595SJ. Bruce Fields * 3926146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3927146a8595SJ. Bruce Fields * 3928146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3929146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3930146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3931146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3932146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3933146a8595SJ. Bruce Fields * 3934146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3935146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3936146a8595SJ. Bruce Fields * to be NFS exported. 3937146a8595SJ. Bruce Fields */ 3938146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39391da177e4SLinus Torvalds { 39401da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39418de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39421da177e4SLinus Torvalds int error; 39431da177e4SLinus Torvalds 39441da177e4SLinus Torvalds if (!inode) 39451da177e4SLinus Torvalds return -ENOENT; 39461da177e4SLinus Torvalds 3947a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 39481da177e4SLinus Torvalds if (error) 39491da177e4SLinus Torvalds return error; 39501da177e4SLinus Torvalds 39511da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 39521da177e4SLinus Torvalds return -EXDEV; 39531da177e4SLinus Torvalds 39541da177e4SLinus Torvalds /* 39551da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 39561da177e4SLinus Torvalds */ 39571da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 39581da177e4SLinus Torvalds return -EPERM; 3959acfa4380SAl Viro if (!dir->i_op->link) 39601da177e4SLinus Torvalds return -EPERM; 39617e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 39621da177e4SLinus Torvalds return -EPERM; 39631da177e4SLinus Torvalds 39641da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 39651da177e4SLinus Torvalds if (error) 39661da177e4SLinus Torvalds return error; 39671da177e4SLinus Torvalds 39687e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 3969aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 3970f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 3971aae8a97dSAneesh Kumar K.V error = -ENOENT; 39728de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 39738de52778SAl Viro error = -EMLINK; 3974146a8595SJ. Bruce Fields else { 3975146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 3976146a8595SJ. Bruce Fields if (!error) 39771da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 3978146a8595SJ. Bruce Fields } 3979f4e0c30cSAl Viro 3980f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 3981f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3982f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 3983f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3984f4e0c30cSAl Viro } 39857e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 3986e31e14ecSStephen Smalley if (!error) 39877e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 39881da177e4SLinus Torvalds return error; 39891da177e4SLinus Torvalds } 39904d359507SAl Viro EXPORT_SYMBOL(vfs_link); 39911da177e4SLinus Torvalds 39921da177e4SLinus Torvalds /* 39931da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 39941da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 39951da177e4SLinus Torvalds * newname. --KAB 39961da177e4SLinus Torvalds * 39971da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 39981da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 39991da177e4SLinus Torvalds * and other special files. --ADM 40001da177e4SLinus Torvalds */ 40012e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 40022e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 40031da177e4SLinus Torvalds { 40041da177e4SLinus Torvalds struct dentry *new_dentry; 4005dae6ad8fSAl Viro struct path old_path, new_path; 4006146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 400711a7b371SAneesh Kumar K.V int how = 0; 40081da177e4SLinus Torvalds int error; 40091da177e4SLinus Torvalds 401011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4011c04030e1SUlrich Drepper return -EINVAL; 401211a7b371SAneesh Kumar K.V /* 4013f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4014f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4015f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 401611a7b371SAneesh Kumar K.V */ 4017f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4018f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4019f0cc6ffbSLinus Torvalds return -ENOENT; 402011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4021f0cc6ffbSLinus Torvalds } 4022c04030e1SUlrich Drepper 402311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 402411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4025442e31caSJeff Layton retry: 402611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40271da177e4SLinus Torvalds if (error) 40282ad94ae6SAl Viro return error; 40292ad94ae6SAl Viro 4030442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4031442e31caSJeff Layton (how & LOOKUP_REVAL)); 40321da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40336902d925SDave Hansen if (IS_ERR(new_dentry)) 4034dae6ad8fSAl Viro goto out; 4035dae6ad8fSAl Viro 4036dae6ad8fSAl Viro error = -EXDEV; 4037dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4038dae6ad8fSAl Viro goto out_dput; 4039800179c9SKees Cook error = may_linkat(&old_path); 4040800179c9SKees Cook if (unlikely(error)) 4041800179c9SKees Cook goto out_dput; 4042dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4043be6d3e56SKentaro Takeda if (error) 4044a8104a9fSAl Viro goto out_dput; 4045146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 404675c3f29dSDave Hansen out_dput: 4047921a1650SAl Viro done_path_create(&new_path, new_dentry); 4048146a8595SJ. Bruce Fields if (delegated_inode) { 4049146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4050d22e6338SOleg Drokin if (!error) { 4051d22e6338SOleg Drokin path_put(&old_path); 4052146a8595SJ. Bruce Fields goto retry; 4053146a8595SJ. Bruce Fields } 4054d22e6338SOleg Drokin } 4055442e31caSJeff Layton if (retry_estale(error, how)) { 4056d22e6338SOleg Drokin path_put(&old_path); 4057442e31caSJeff Layton how |= LOOKUP_REVAL; 4058442e31caSJeff Layton goto retry; 4059442e31caSJeff Layton } 40601da177e4SLinus Torvalds out: 40612d8f3038SAl Viro path_put(&old_path); 40621da177e4SLinus Torvalds 40631da177e4SLinus Torvalds return error; 40641da177e4SLinus Torvalds } 40651da177e4SLinus Torvalds 40663480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 40675590ff0dSUlrich Drepper { 4068c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 40695590ff0dSUlrich Drepper } 40705590ff0dSUlrich Drepper 4071bc27027aSMiklos Szeredi /** 4072bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4073bc27027aSMiklos Szeredi * @old_dir: parent of source 4074bc27027aSMiklos Szeredi * @old_dentry: source 4075bc27027aSMiklos Szeredi * @new_dir: parent of destination 4076bc27027aSMiklos Szeredi * @new_dentry: destination 4077bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4078520c8b16SMiklos Szeredi * @flags: rename flags 4079bc27027aSMiklos Szeredi * 4080bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4081bc27027aSMiklos Szeredi * 4082bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4083bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4084bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4085bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4086bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4087bc27027aSMiklos Szeredi * so. 4088bc27027aSMiklos Szeredi * 4089bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4090bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4091bc27027aSMiklos Szeredi * to be NFS exported. 4092bc27027aSMiklos Szeredi * 40931da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 40941da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 40951da177e4SLinus Torvalds * Problems: 4096d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 40971da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 40981da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4099a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 41001da177e4SLinus Torvalds * story. 41016cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 41026cedba89SJ. Bruce Fields * and source (if it is not a directory). 41031b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 41041da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 41051da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4106a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 41071da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 41081da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 41091da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4110a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41111da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41121da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41131da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4114e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41151b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41161da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4117c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41181da177e4SLinus Torvalds * locking]. 41191da177e4SLinus Torvalds */ 41201da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41218e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4122520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41231da177e4SLinus Torvalds { 41241da177e4SLinus Torvalds int error; 4125bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 412659b0df21SEric Paris const unsigned char *old_name; 4127bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4128bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4129da1ce067SMiklos Szeredi bool new_is_dir = false; 4130da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41311da177e4SLinus Torvalds 4132bc27027aSMiklos Szeredi if (source == target) 41331da177e4SLinus Torvalds return 0; 41341da177e4SLinus Torvalds 41351da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41361da177e4SLinus Torvalds if (error) 41371da177e4SLinus Torvalds return error; 41381da177e4SLinus Torvalds 4139da1ce067SMiklos Szeredi if (!target) { 4140a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4141da1ce067SMiklos Szeredi } else { 4142da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4143da1ce067SMiklos Szeredi 4144da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 41451da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4146da1ce067SMiklos Szeredi else 4147da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4148da1ce067SMiklos Szeredi } 41491da177e4SLinus Torvalds if (error) 41501da177e4SLinus Torvalds return error; 41511da177e4SLinus Torvalds 41527177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 41531da177e4SLinus Torvalds return -EPERM; 41541da177e4SLinus Torvalds 4155520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4156520c8b16SMiklos Szeredi return -EINVAL; 4157520c8b16SMiklos Szeredi 4158bc27027aSMiklos Szeredi /* 4159bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4160bc27027aSMiklos Szeredi * we'll need to flip '..'. 4161bc27027aSMiklos Szeredi */ 4162da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4163da1ce067SMiklos Szeredi if (is_dir) { 4164bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4165bc27027aSMiklos Szeredi if (error) 4166bc27027aSMiklos Szeredi return error; 4167bc27027aSMiklos Szeredi } 4168da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4169da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4170da1ce067SMiklos Szeredi if (error) 4171da1ce067SMiklos Szeredi return error; 4172da1ce067SMiklos Szeredi } 4173da1ce067SMiklos Szeredi } 41740eeca283SRobert Love 41750b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 41760b3974ebSMiklos Szeredi flags); 4177bc27027aSMiklos Szeredi if (error) 4178bc27027aSMiklos Szeredi return error; 4179bc27027aSMiklos Szeredi 4180bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4181bc27027aSMiklos Szeredi dget(new_dentry); 4182da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4183bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4184bc27027aSMiklos Szeredi else if (target) 4185bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4186bc27027aSMiklos Szeredi 4187bc27027aSMiklos Szeredi error = -EBUSY; 41887af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4189bc27027aSMiklos Szeredi goto out; 4190bc27027aSMiklos Szeredi 4191da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4192bc27027aSMiklos Szeredi error = -EMLINK; 4193da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4194bc27027aSMiklos Szeredi goto out; 4195da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4196da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4197da1ce067SMiklos Szeredi goto out; 4198da1ce067SMiklos Szeredi } 4199da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4200bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4201da1ce067SMiklos Szeredi if (!is_dir) { 4202bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4203bc27027aSMiklos Szeredi if (error) 4204bc27027aSMiklos Szeredi goto out; 4205da1ce067SMiklos Szeredi } 4206da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4207bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4208bc27027aSMiklos Szeredi if (error) 4209bc27027aSMiklos Szeredi goto out; 4210bc27027aSMiklos Szeredi } 42117177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4212520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4213520c8b16SMiklos Szeredi new_dir, new_dentry); 4214520c8b16SMiklos Szeredi } else { 42157177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4216520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4217520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4218520c8b16SMiklos Szeredi } 4219bc27027aSMiklos Szeredi if (error) 4220bc27027aSMiklos Szeredi goto out; 4221bc27027aSMiklos Szeredi 4222da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42231da177e4SLinus Torvalds if (is_dir) 4224bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4225bc27027aSMiklos Szeredi dont_mount(new_dentry); 42268ed936b5SEric W. Biederman detach_mounts(new_dentry); 4227bc27027aSMiklos Szeredi } 4228da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4229da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4230bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4231da1ce067SMiklos Szeredi else 4232da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4233da1ce067SMiklos Szeredi } 4234bc27027aSMiklos Szeredi out: 4235da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4236bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4237bc27027aSMiklos Szeredi else if (target) 4238bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4239bc27027aSMiklos Szeredi dput(new_dentry); 4240da1ce067SMiklos Szeredi if (!error) { 4241123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4242da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4243da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4244da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4245da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4246da1ce067SMiklos Szeredi } 4247da1ce067SMiklos Szeredi } 42480eeca283SRobert Love fsnotify_oldname_free(old_name); 42490eeca283SRobert Love 42501da177e4SLinus Torvalds return error; 42511da177e4SLinus Torvalds } 42524d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 42531da177e4SLinus Torvalds 4254520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4255520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 42561da177e4SLinus Torvalds { 42571da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 42581da177e4SLinus Torvalds struct dentry *trap; 4259f5beed75SAl Viro struct path old_path, new_path; 4260f5beed75SAl Viro struct qstr old_last, new_last; 4261f5beed75SAl Viro int old_type, new_type; 42628e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 426391a27b2aSJeff Layton struct filename *from; 426491a27b2aSJeff Layton struct filename *to; 4265f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4266c6a94284SJeff Layton bool should_retry = false; 42672ad94ae6SAl Viro int error; 4268520c8b16SMiklos Szeredi 42690d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4270da1ce067SMiklos Szeredi return -EINVAL; 4271da1ce067SMiklos Szeredi 42720d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 42730d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4274520c8b16SMiklos Szeredi return -EINVAL; 4275520c8b16SMiklos Szeredi 42760d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 42770d7a8555SMiklos Szeredi return -EPERM; 42780d7a8555SMiklos Szeredi 4279f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4280f5beed75SAl Viro target_flags = 0; 4281f5beed75SAl Viro 4282c6a94284SJeff Layton retry: 4283f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4284f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 428591a27b2aSJeff Layton if (IS_ERR(from)) { 428691a27b2aSJeff Layton error = PTR_ERR(from); 42871da177e4SLinus Torvalds goto exit; 428891a27b2aSJeff Layton } 42891da177e4SLinus Torvalds 4290f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4291f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 429291a27b2aSJeff Layton if (IS_ERR(to)) { 429391a27b2aSJeff Layton error = PTR_ERR(to); 42941da177e4SLinus Torvalds goto exit1; 429591a27b2aSJeff Layton } 42961da177e4SLinus Torvalds 42971da177e4SLinus Torvalds error = -EXDEV; 4298f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 42991da177e4SLinus Torvalds goto exit2; 43001da177e4SLinus Torvalds 43011da177e4SLinus Torvalds error = -EBUSY; 4302f5beed75SAl Viro if (old_type != LAST_NORM) 43031da177e4SLinus Torvalds goto exit2; 43041da177e4SLinus Torvalds 43050a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43060a7c3937SMiklos Szeredi error = -EEXIST; 4307f5beed75SAl Viro if (new_type != LAST_NORM) 43081da177e4SLinus Torvalds goto exit2; 43091da177e4SLinus Torvalds 4310f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4311c30dabfeSJan Kara if (error) 4312c30dabfeSJan Kara goto exit2; 4313c30dabfeSJan Kara 43148e6d782cSJ. Bruce Fields retry_deleg: 4315f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43161da177e4SLinus Torvalds 4317f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43181da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43191da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43201da177e4SLinus Torvalds goto exit3; 43211da177e4SLinus Torvalds /* source must exist */ 43221da177e4SLinus Torvalds error = -ENOENT; 4323b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43241da177e4SLinus Torvalds goto exit4; 4325f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 43261da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43271da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43281da177e4SLinus Torvalds goto exit4; 43290a7c3937SMiklos Szeredi error = -EEXIST; 43300a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43310a7c3937SMiklos Szeredi goto exit5; 4332da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4333da1ce067SMiklos Szeredi error = -ENOENT; 4334da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4335da1ce067SMiklos Szeredi goto exit5; 4336da1ce067SMiklos Szeredi 4337da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4338da1ce067SMiklos Szeredi error = -ENOTDIR; 4339f5beed75SAl Viro if (new_last.name[new_last.len]) 4340da1ce067SMiklos Szeredi goto exit5; 4341da1ce067SMiklos Szeredi } 4342da1ce067SMiklos Szeredi } 43430a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 43440a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 43450a7c3937SMiklos Szeredi error = -ENOTDIR; 4346f5beed75SAl Viro if (old_last.name[old_last.len]) 43470a7c3937SMiklos Szeredi goto exit5; 4348f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 43490a7c3937SMiklos Szeredi goto exit5; 43500a7c3937SMiklos Szeredi } 43510a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 43520a7c3937SMiklos Szeredi error = -EINVAL; 43530a7c3937SMiklos Szeredi if (old_dentry == trap) 43540a7c3937SMiklos Szeredi goto exit5; 43551da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4356da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43571da177e4SLinus Torvalds error = -ENOTEMPTY; 43581da177e4SLinus Torvalds if (new_dentry == trap) 43591da177e4SLinus Torvalds goto exit5; 43601da177e4SLinus Torvalds 4361f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4362f5beed75SAl Viro &new_path, new_dentry, flags); 4363be6d3e56SKentaro Takeda if (error) 4364c30dabfeSJan Kara goto exit5; 4365f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4366f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4367520c8b16SMiklos Szeredi &delegated_inode, flags); 43681da177e4SLinus Torvalds exit5: 43691da177e4SLinus Torvalds dput(new_dentry); 43701da177e4SLinus Torvalds exit4: 43711da177e4SLinus Torvalds dput(old_dentry); 43721da177e4SLinus Torvalds exit3: 4373f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 43748e6d782cSJ. Bruce Fields if (delegated_inode) { 43758e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 43768e6d782cSJ. Bruce Fields if (!error) 43778e6d782cSJ. Bruce Fields goto retry_deleg; 43788e6d782cSJ. Bruce Fields } 4379f5beed75SAl Viro mnt_drop_write(old_path.mnt); 43801da177e4SLinus Torvalds exit2: 4381c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4382c6a94284SJeff Layton should_retry = true; 4383f5beed75SAl Viro path_put(&new_path); 43842ad94ae6SAl Viro putname(to); 43851da177e4SLinus Torvalds exit1: 4386f5beed75SAl Viro path_put(&old_path); 43871da177e4SLinus Torvalds putname(from); 4388c6a94284SJeff Layton if (should_retry) { 4389c6a94284SJeff Layton should_retry = false; 4390c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4391c6a94284SJeff Layton goto retry; 4392c6a94284SJeff Layton } 43932ad94ae6SAl Viro exit: 43941da177e4SLinus Torvalds return error; 43951da177e4SLinus Torvalds } 43961da177e4SLinus Torvalds 4397520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4398520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4399520c8b16SMiklos Szeredi { 4400520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4401520c8b16SMiklos Szeredi } 4402520c8b16SMiklos Szeredi 4403a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 44045590ff0dSUlrich Drepper { 4405520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44065590ff0dSUlrich Drepper } 44075590ff0dSUlrich Drepper 4408787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4409787fb6bcSMiklos Szeredi { 4410787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4411787fb6bcSMiklos Szeredi if (error) 4412787fb6bcSMiklos Szeredi return error; 4413787fb6bcSMiklos Szeredi 4414787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4415787fb6bcSMiklos Szeredi return -EPERM; 4416787fb6bcSMiklos Szeredi 4417787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4418787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4419787fb6bcSMiklos Szeredi } 4420787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4421787fb6bcSMiklos Szeredi 44225d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44231da177e4SLinus Torvalds { 44245d826c84SAl Viro int len = PTR_ERR(link); 44251da177e4SLinus Torvalds if (IS_ERR(link)) 44261da177e4SLinus Torvalds goto out; 44271da177e4SLinus Torvalds 44281da177e4SLinus Torvalds len = strlen(link); 44291da177e4SLinus Torvalds if (len > (unsigned) buflen) 44301da177e4SLinus Torvalds len = buflen; 44311da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44321da177e4SLinus Torvalds len = -EFAULT; 44331da177e4SLinus Torvalds out: 44341da177e4SLinus Torvalds return len; 44351da177e4SLinus Torvalds } 44365d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44371da177e4SLinus Torvalds 44381da177e4SLinus Torvalds /* 44391da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44401da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44411da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 44421da177e4SLinus Torvalds */ 44431da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44441da177e4SLinus Torvalds { 4445cc314eefSLinus Torvalds void *cookie; 4446d4dee48bSAl Viro const char *link = dentry->d_inode->i_link; 4447694a1764SMarcin Slusarz int res; 4448cc314eefSLinus Torvalds 4449d4dee48bSAl Viro if (!link) { 44506e77137bSAl Viro link = dentry->d_inode->i_op->follow_link(dentry, &cookie); 4451680baacbSAl Viro if (IS_ERR(link)) 4452680baacbSAl Viro return PTR_ERR(link); 4453d4dee48bSAl Viro } 4454680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 44556e77137bSAl Viro if (dentry->d_inode->i_op->put_link) 4456680baacbSAl Viro dentry->d_inode->i_op->put_link(dentry, cookie); 4457694a1764SMarcin Slusarz return res; 44581da177e4SLinus Torvalds } 44594d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 44601da177e4SLinus Torvalds 44611da177e4SLinus Torvalds /* get the link contents into pagecache */ 44621da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 44631da177e4SLinus Torvalds { 4464ebd09abbSDuane Griffin char *kaddr; 44651da177e4SLinus Torvalds struct page *page; 44661da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4467090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 44681da177e4SLinus Torvalds if (IS_ERR(page)) 44696fe6900eSNick Piggin return (char*)page; 44701da177e4SLinus Torvalds *ppage = page; 4471ebd09abbSDuane Griffin kaddr = kmap(page); 4472ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4473ebd09abbSDuane Griffin return kaddr; 44741da177e4SLinus Torvalds } 44751da177e4SLinus Torvalds 44761da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44771da177e4SLinus Torvalds { 44781da177e4SLinus Torvalds struct page *page = NULL; 44795d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 44801da177e4SLinus Torvalds if (page) { 44811da177e4SLinus Torvalds kunmap(page); 44821da177e4SLinus Torvalds page_cache_release(page); 44831da177e4SLinus Torvalds } 44841da177e4SLinus Torvalds return res; 44851da177e4SLinus Torvalds } 44864d359507SAl Viro EXPORT_SYMBOL(page_readlink); 44871da177e4SLinus Torvalds 44886e77137bSAl Viro const char *page_follow_link_light(struct dentry *dentry, void **cookie) 44891da177e4SLinus Torvalds { 4490cc314eefSLinus Torvalds struct page *page = NULL; 4491680baacbSAl Viro char *res = page_getlink(dentry, &page); 4492680baacbSAl Viro if (!IS_ERR(res)) 4493680baacbSAl Viro *cookie = page; 4494680baacbSAl Viro return res; 44951da177e4SLinus Torvalds } 44964d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 44971da177e4SLinus Torvalds 4498680baacbSAl Viro void page_put_link(struct dentry *dentry, void *cookie) 44991da177e4SLinus Torvalds { 4500cc314eefSLinus Torvalds struct page *page = cookie; 45011da177e4SLinus Torvalds kunmap(page); 45021da177e4SLinus Torvalds page_cache_release(page); 45031da177e4SLinus Torvalds } 45044d359507SAl Viro EXPORT_SYMBOL(page_put_link); 45051da177e4SLinus Torvalds 450654566b2cSNick Piggin /* 450754566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 450854566b2cSNick Piggin */ 450954566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45101da177e4SLinus Torvalds { 45111da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45120adb25d2SKirill Korotaev struct page *page; 4513afddba49SNick Piggin void *fsdata; 4514beb497abSDmitriy Monakhov int err; 45151da177e4SLinus Torvalds char *kaddr; 451654566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 451754566b2cSNick Piggin if (nofs) 451854566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45191da177e4SLinus Torvalds 45207e53cac4SNeilBrown retry: 4521afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 452254566b2cSNick Piggin flags, &page, &fsdata); 45231da177e4SLinus Torvalds if (err) 4524afddba49SNick Piggin goto fail; 4525afddba49SNick Piggin 4526e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 45271da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4528e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4529afddba49SNick Piggin 4530afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4531afddba49SNick Piggin page, fsdata); 45321da177e4SLinus Torvalds if (err < 0) 45331da177e4SLinus Torvalds goto fail; 4534afddba49SNick Piggin if (err < len-1) 4535afddba49SNick Piggin goto retry; 4536afddba49SNick Piggin 45371da177e4SLinus Torvalds mark_inode_dirty(inode); 45381da177e4SLinus Torvalds return 0; 45391da177e4SLinus Torvalds fail: 45401da177e4SLinus Torvalds return err; 45411da177e4SLinus Torvalds } 45424d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 45431da177e4SLinus Torvalds 45440adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 45450adb25d2SKirill Korotaev { 45460adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 454754566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 45480adb25d2SKirill Korotaev } 45494d359507SAl Viro EXPORT_SYMBOL(page_symlink); 45500adb25d2SKirill Korotaev 455192e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 45521da177e4SLinus Torvalds .readlink = generic_readlink, 45531da177e4SLinus Torvalds .follow_link = page_follow_link_light, 45541da177e4SLinus Torvalds .put_link = page_put_link, 45551da177e4SLinus Torvalds }; 45561da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4557