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; 498caa85634SAl Viro union { 4991f55a6ecSAl Viro struct qstr last; 500caa85634SAl Viro struct path link; 501caa85634SAl Viro }; 5021f55a6ecSAl Viro struct path root; 5031f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5041f55a6ecSAl Viro unsigned int flags; 5051f55a6ecSAl Viro unsigned seq, m_seq; 5061f55a6ecSAl Viro int last_type; 5071f55a6ecSAl Viro unsigned depth; 5085e53084dSAl Viro struct file *base; 509697fc6caSAl Viro struct saved { 510697fc6caSAl Viro struct path link; 511697fc6caSAl Viro void *cookie; 512697fc6caSAl Viro const char *name; 513894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5141f55a6ecSAl Viro }; 5151f55a6ecSAl Viro 516894bc8c4SAl Viro static void set_nameidata(struct nameidata *nd) 517894bc8c4SAl Viro { 518894bc8c4SAl Viro nd->stack = nd->internal; 519894bc8c4SAl Viro } 520894bc8c4SAl Viro 521894bc8c4SAl Viro static void restore_nameidata(struct nameidata *nd) 522894bc8c4SAl Viro { 523894bc8c4SAl Viro if (nd->stack != nd->internal) { 524894bc8c4SAl Viro kfree(nd->stack); 525894bc8c4SAl Viro nd->stack = nd->internal; 526894bc8c4SAl Viro } 527894bc8c4SAl Viro } 528894bc8c4SAl Viro 529894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 530894bc8c4SAl Viro { 531e269f2a7SAl Viro struct saved *p = kmalloc(MAXSYMLINKS * sizeof(struct saved), 532894bc8c4SAl Viro GFP_KERNEL); 533894bc8c4SAl Viro if (unlikely(!p)) 534894bc8c4SAl Viro return -ENOMEM; 535894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 536894bc8c4SAl Viro nd->stack = p; 537894bc8c4SAl Viro return 0; 538894bc8c4SAl Viro } 539894bc8c4SAl Viro 540894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 541894bc8c4SAl Viro { 542da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 543894bc8c4SAl Viro return 0; 544894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 545894bc8c4SAl Viro return 0; 546894bc8c4SAl Viro return __nd_alloc_stack(nd); 547894bc8c4SAl Viro } 548894bc8c4SAl Viro 54919660af7SAl Viro /* 55031e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 55119660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 55219660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 55319660af7SAl Viro * normal reference counts on dentries and vfsmounts to transition to rcu-walk 55419660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 55519660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 55619660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 55719660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 55831e6b01fSNick Piggin */ 55931e6b01fSNick Piggin 56031e6b01fSNick Piggin /** 56119660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 56219660af7SAl Viro * @nd: nameidata pathwalk data 56319660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 56439191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 56531e6b01fSNick Piggin * 56619660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 56719660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 56819660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 56931e6b01fSNick Piggin */ 57019660af7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry) 57131e6b01fSNick Piggin { 57231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 57331e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 57431e6b01fSNick Piggin 57531e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 576e5c832d5SLinus Torvalds 577e5c832d5SLinus Torvalds /* 57848a066e7SAl Viro * After legitimizing the bastards, terminate_walk() 57948a066e7SAl Viro * will do the right thing for non-RCU mode, and all our 58048a066e7SAl Viro * subsequent exit cases should rcu_read_unlock() 58148a066e7SAl Viro * before returning. Do vfsmount first; if dentry 58248a066e7SAl Viro * can't be legitimized, just set nd->path.dentry to NULL 58348a066e7SAl Viro * and rely on dput(NULL) being a no-op. 584e5c832d5SLinus Torvalds */ 58548a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) 586e5c832d5SLinus Torvalds return -ECHILD; 587e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 58815570086SLinus Torvalds 58948a066e7SAl Viro if (!lockref_get_not_dead(&parent->d_lockref)) { 59048a066e7SAl Viro nd->path.dentry = NULL; 591d870b4a1SAl Viro goto out; 59248a066e7SAl Viro } 59348a066e7SAl Viro 59415570086SLinus Torvalds /* 59515570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 59615570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 59715570086SLinus Torvalds * 59815570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 59915570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 60015570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 60115570086SLinus Torvalds * since we checked the sequence number of the parent after we got 60215570086SLinus Torvalds * the child sequence number. So we know the parent must still 60315570086SLinus Torvalds * be valid if the child sequence number is still valid. 60415570086SLinus Torvalds */ 60519660af7SAl Viro if (!dentry) { 606e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 607e5c832d5SLinus Torvalds goto out; 60819660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 60919660af7SAl Viro } else { 610e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 611e5c832d5SLinus Torvalds goto out; 612e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) 613e5c832d5SLinus Torvalds goto drop_dentry; 61419660af7SAl Viro } 615e5c832d5SLinus Torvalds 616e5c832d5SLinus Torvalds /* 617e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 618e5c832d5SLinus Torvalds * still valid and get it if required. 619e5c832d5SLinus Torvalds */ 620e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 621e5c832d5SLinus Torvalds spin_lock(&fs->lock); 622e5c832d5SLinus Torvalds if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry) 623e5c832d5SLinus Torvalds goto unlock_and_drop_dentry; 62431e6b01fSNick Piggin path_get(&nd->root); 62531e6b01fSNick Piggin spin_unlock(&fs->lock); 62631e6b01fSNick Piggin } 62731e6b01fSNick Piggin 6288b61e74fSAl Viro rcu_read_unlock(); 62931e6b01fSNick Piggin return 0; 63019660af7SAl Viro 631e5c832d5SLinus Torvalds unlock_and_drop_dentry: 63231e6b01fSNick Piggin spin_unlock(&fs->lock); 633e5c832d5SLinus Torvalds drop_dentry: 6348b61e74fSAl Viro rcu_read_unlock(); 635e5c832d5SLinus Torvalds dput(dentry); 636d0d27277SLinus Torvalds goto drop_root_mnt; 637e5c832d5SLinus Torvalds out: 6388b61e74fSAl Viro rcu_read_unlock(); 639d0d27277SLinus Torvalds drop_root_mnt: 640d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 641d0d27277SLinus Torvalds nd->root.mnt = NULL; 64231e6b01fSNick Piggin return -ECHILD; 64331e6b01fSNick Piggin } 64431e6b01fSNick Piggin 6454ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 64634286d66SNick Piggin { 6474ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 64834286d66SNick Piggin } 64934286d66SNick Piggin 6509f1fafeeSAl Viro /** 6519f1fafeeSAl Viro * complete_walk - successful completion of path walk 6529f1fafeeSAl Viro * @nd: pointer nameidata 65339159de2SJeff Layton * 6549f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 6559f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 6569f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 6579f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 6589f1fafeeSAl Viro * need to drop nd->path. 65939159de2SJeff Layton */ 6609f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 66139159de2SJeff Layton { 66216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 66339159de2SJeff Layton int status; 66439159de2SJeff Layton 6659f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 6669f1fafeeSAl Viro nd->flags &= ~LOOKUP_RCU; 6679f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6689f1fafeeSAl Viro nd->root.mnt = NULL; 66915570086SLinus Torvalds 67048a066e7SAl Viro if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) { 6718b61e74fSAl Viro rcu_read_unlock(); 67248a066e7SAl Viro return -ECHILD; 67348a066e7SAl Viro } 674e5c832d5SLinus Torvalds if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) { 6758b61e74fSAl Viro rcu_read_unlock(); 67648a066e7SAl Viro mntput(nd->path.mnt); 6779f1fafeeSAl Viro return -ECHILD; 6789f1fafeeSAl Viro } 679e5c832d5SLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, nd->seq)) { 6808b61e74fSAl Viro rcu_read_unlock(); 681e5c832d5SLinus Torvalds dput(dentry); 68248a066e7SAl Viro mntput(nd->path.mnt); 683e5c832d5SLinus Torvalds return -ECHILD; 684e5c832d5SLinus Torvalds } 6858b61e74fSAl Viro rcu_read_unlock(); 6869f1fafeeSAl Viro } 6879f1fafeeSAl Viro 68816c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 68939159de2SJeff Layton return 0; 69039159de2SJeff Layton 691ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 69216c2cd71SAl Viro return 0; 69316c2cd71SAl Viro 694ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 69539159de2SJeff Layton if (status > 0) 69639159de2SJeff Layton return 0; 69739159de2SJeff Layton 69816c2cd71SAl Viro if (!status) 69939159de2SJeff Layton status = -ESTALE; 70016c2cd71SAl Viro 7019f1fafeeSAl Viro path_put(&nd->path); 70239159de2SJeff Layton return status; 70339159de2SJeff Layton } 70439159de2SJeff Layton 7052a737871SAl Viro static __always_inline void set_root(struct nameidata *nd) 7062a737871SAl Viro { 707f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &nd->root); 7082a737871SAl Viro } 7092a737871SAl Viro 7107bd88377SAl Viro static __always_inline unsigned set_root_rcu(struct nameidata *nd) 71131e6b01fSNick Piggin { 71231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 7137bd88377SAl Viro unsigned seq, res; 714c28cc364SNick Piggin 715c28cc364SNick Piggin do { 716c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 71731e6b01fSNick Piggin nd->root = fs->root; 7187bd88377SAl Viro res = __read_seqcount_begin(&nd->root.dentry->d_seq); 719c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 7207bd88377SAl Viro return res; 72131e6b01fSNick Piggin } 72231e6b01fSNick Piggin 7231d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 724051d3812SIan Kent { 725051d3812SIan Kent dput(path->dentry); 7264ac91378SJan Blunck if (path->mnt != nd->path.mnt) 727051d3812SIan Kent mntput(path->mnt); 728051d3812SIan Kent } 729051d3812SIan Kent 7307b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 7317b9337aaSNick Piggin struct nameidata *nd) 732051d3812SIan Kent { 73331e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 7344ac91378SJan Blunck dput(nd->path.dentry); 73531e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 7364ac91378SJan Blunck mntput(nd->path.mnt); 7379a229683SHuang Shijie } 73831e6b01fSNick Piggin nd->path.mnt = path->mnt; 7394ac91378SJan Blunck nd->path.dentry = path->dentry; 740051d3812SIan Kent } 741051d3812SIan Kent 742b5fb63c1SChristoph Hellwig /* 743b5fb63c1SChristoph Hellwig * Helper to directly jump to a known parsed path from ->follow_link, 744b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 745b5fb63c1SChristoph Hellwig */ 746b5fb63c1SChristoph Hellwig void nd_jump_link(struct nameidata *nd, struct path *path) 747b5fb63c1SChristoph Hellwig { 748b5fb63c1SChristoph Hellwig path_put(&nd->path); 749b5fb63c1SChristoph Hellwig 750b5fb63c1SChristoph Hellwig nd->path = *path; 751b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 752b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 753b5fb63c1SChristoph Hellwig } 754b5fb63c1SChristoph Hellwig 755b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 756574197e0SAl Viro { 75721c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 758b9ff4429SAl Viro struct inode *inode = last->link.dentry->d_inode; 759b9ff4429SAl Viro if (last->cookie && inode->i_op->put_link) 760b9ff4429SAl Viro inode->i_op->put_link(last->link.dentry, last->cookie); 761b9ff4429SAl Viro path_put(&last->link); 762574197e0SAl Viro } 763574197e0SAl Viro 764561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 765561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 766800179c9SKees Cook 767800179c9SKees Cook /** 768800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 769800179c9SKees Cook * @link: The path of the symlink 77055852635SRandy Dunlap * @nd: nameidata pathwalk data 771800179c9SKees Cook * 772800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 773800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 774800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 775800179c9SKees Cook * processes from failing races against path names that may change out 776800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 777800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 778800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 779800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 780800179c9SKees Cook * 781800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 782800179c9SKees Cook */ 783800179c9SKees Cook static inline int may_follow_link(struct path *link, struct nameidata *nd) 784800179c9SKees Cook { 785800179c9SKees Cook const struct inode *inode; 786800179c9SKees Cook const struct inode *parent; 787800179c9SKees Cook 788800179c9SKees Cook if (!sysctl_protected_symlinks) 789800179c9SKees Cook return 0; 790800179c9SKees Cook 791800179c9SKees Cook /* Allowed if owner and follower match. */ 792800179c9SKees Cook inode = link->dentry->d_inode; 79381abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 794800179c9SKees Cook return 0; 795800179c9SKees Cook 796800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 797800179c9SKees Cook parent = nd->path.dentry->d_inode; 798800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 799800179c9SKees Cook return 0; 800800179c9SKees Cook 801800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 80281abe27bSEric W. Biederman if (uid_eq(parent->i_uid, inode->i_uid)) 803800179c9SKees Cook return 0; 804800179c9SKees Cook 805ffd8d101SSasha Levin audit_log_link_denied("follow_link", link); 806800179c9SKees Cook path_put_conditional(link, nd); 807800179c9SKees Cook path_put(&nd->path); 808800179c9SKees Cook return -EACCES; 809800179c9SKees Cook } 810800179c9SKees Cook 811800179c9SKees Cook /** 812800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 813800179c9SKees Cook * @inode: the source inode to hardlink from 814800179c9SKees Cook * 815800179c9SKees Cook * Return false if at least one of the following conditions: 816800179c9SKees Cook * - inode is not a regular file 817800179c9SKees Cook * - inode is setuid 818800179c9SKees Cook * - inode is setgid and group-exec 819800179c9SKees Cook * - access failure for read and write 820800179c9SKees Cook * 821800179c9SKees Cook * Otherwise returns true. 822800179c9SKees Cook */ 823800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 824800179c9SKees Cook { 825800179c9SKees Cook umode_t mode = inode->i_mode; 826800179c9SKees Cook 827800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 828800179c9SKees Cook if (!S_ISREG(mode)) 829800179c9SKees Cook return false; 830800179c9SKees Cook 831800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 832800179c9SKees Cook if (mode & S_ISUID) 833800179c9SKees Cook return false; 834800179c9SKees Cook 835800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 836800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 837800179c9SKees Cook return false; 838800179c9SKees Cook 839800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 840800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 841800179c9SKees Cook return false; 842800179c9SKees Cook 843800179c9SKees Cook return true; 844800179c9SKees Cook } 845800179c9SKees Cook 846800179c9SKees Cook /** 847800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 848800179c9SKees Cook * @link: the source to hardlink from 849800179c9SKees Cook * 850800179c9SKees Cook * Block hardlink when all of: 851800179c9SKees Cook * - sysctl_protected_hardlinks enabled 852800179c9SKees Cook * - fsuid does not match inode 853800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 854800179c9SKees Cook * - not CAP_FOWNER 855800179c9SKees Cook * 856800179c9SKees Cook * Returns 0 if successful, -ve on error. 857800179c9SKees Cook */ 858800179c9SKees Cook static int may_linkat(struct path *link) 859800179c9SKees Cook { 860800179c9SKees Cook const struct cred *cred; 861800179c9SKees Cook struct inode *inode; 862800179c9SKees Cook 863800179c9SKees Cook if (!sysctl_protected_hardlinks) 864800179c9SKees Cook return 0; 865800179c9SKees Cook 866800179c9SKees Cook cred = current_cred(); 867800179c9SKees Cook inode = link->dentry->d_inode; 868800179c9SKees Cook 869800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 870800179c9SKees Cook * otherwise, it must be a safe source. 871800179c9SKees Cook */ 87281abe27bSEric W. Biederman if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) || 873800179c9SKees Cook capable(CAP_FOWNER)) 874800179c9SKees Cook return 0; 875800179c9SKees Cook 876a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 877800179c9SKees Cook return -EPERM; 878800179c9SKees Cook } 879800179c9SKees Cook 8803b2e7f75SAl Viro static __always_inline 8813b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 8821da177e4SLinus Torvalds { 8833b2e7f75SAl Viro struct saved *last = nd->stack + nd->depth; 8843b2e7f75SAl Viro struct dentry *dentry = nd->link.dentry; 8850a959df5SAl Viro struct inode *inode = dentry->d_inode; 8866d7b5aaeSAl Viro int error; 8870a959df5SAl Viro const char *res; 8881da177e4SLinus Torvalds 889844a3917SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 890844a3917SAl Viro 8913b2e7f75SAl Viro if (nd->link.mnt == nd->path.mnt) 8923b2e7f75SAl Viro mntget(nd->link.mnt); 8930e794589SAl Viro 894894bc8c4SAl Viro if (unlikely(current->total_link_count >= MAXSYMLINKS)) { 8953b2e7f75SAl Viro path_put(&nd->link); 8963b2e7f75SAl Viro return ERR_PTR(-ELOOP); 8973b2e7f75SAl Viro } 8983b2e7f75SAl Viro 8993b2e7f75SAl Viro last->link = nd->link; 9003b2e7f75SAl Viro last->cookie = NULL; 9016d7b5aaeSAl Viro 902574197e0SAl Viro cond_resched(); 903574197e0SAl Viro current->total_link_count++; 904574197e0SAl Viro 9053b2e7f75SAl Viro touch_atime(&last->link); 906cd4e91d3SAl Viro 90737882db0SNeilBrown error = security_inode_follow_link(dentry); 9080a959df5SAl Viro res = ERR_PTR(error); 9096d7b5aaeSAl Viro if (error) 9100a959df5SAl Viro goto out; 91136f3b4f6SAl Viro 91286acdca1SAl Viro nd->last_type = LAST_BIND; 913d4dee48bSAl Viro res = inode->i_link; 914d4dee48bSAl Viro if (!res) { 9153b2e7f75SAl Viro res = inode->i_op->follow_link(dentry, &last->cookie, nd); 9160a959df5SAl Viro if (IS_ERR(res)) { 9170a959df5SAl Viro out: 9183b2e7f75SAl Viro path_put(&last->link); 9190fd889d5SAl Viro return res; 9200a959df5SAl Viro } 921d4dee48bSAl Viro } 9220fd889d5SAl Viro nd->depth++; 9230a959df5SAl Viro return res; 9240a959df5SAl Viro } 9256d7b5aaeSAl Viro 92631e6b01fSNick Piggin static int follow_up_rcu(struct path *path) 92731e6b01fSNick Piggin { 9280714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9290714a533SAl Viro struct mount *parent; 93031e6b01fSNick Piggin struct dentry *mountpoint; 93131e6b01fSNick Piggin 9320714a533SAl Viro parent = mnt->mnt_parent; 9330714a533SAl Viro if (&parent->mnt == path->mnt) 93431e6b01fSNick Piggin return 0; 935a73324daSAl Viro mountpoint = mnt->mnt_mountpoint; 93631e6b01fSNick Piggin path->dentry = mountpoint; 9370714a533SAl Viro path->mnt = &parent->mnt; 93831e6b01fSNick Piggin return 1; 93931e6b01fSNick Piggin } 94031e6b01fSNick Piggin 941f015f126SDavid Howells /* 942f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 943f015f126SDavid Howells * 944f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 945f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 946f015f126SDavid Howells * Up is towards /. 947f015f126SDavid Howells * 948f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 949f015f126SDavid Howells * root. 950f015f126SDavid Howells */ 951bab77ebfSAl Viro int follow_up(struct path *path) 9521da177e4SLinus Torvalds { 9530714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 9540714a533SAl Viro struct mount *parent; 9551da177e4SLinus Torvalds struct dentry *mountpoint; 95699b7db7bSNick Piggin 95748a066e7SAl Viro read_seqlock_excl(&mount_lock); 9580714a533SAl Viro parent = mnt->mnt_parent; 9593c0a6163SAl Viro if (parent == mnt) { 96048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 9611da177e4SLinus Torvalds return 0; 9621da177e4SLinus Torvalds } 9630714a533SAl Viro mntget(&parent->mnt); 964a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 96548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 966bab77ebfSAl Viro dput(path->dentry); 967bab77ebfSAl Viro path->dentry = mountpoint; 968bab77ebfSAl Viro mntput(path->mnt); 9690714a533SAl Viro path->mnt = &parent->mnt; 9701da177e4SLinus Torvalds return 1; 9711da177e4SLinus Torvalds } 9724d359507SAl Viro EXPORT_SYMBOL(follow_up); 9731da177e4SLinus Torvalds 974b5c84bf6SNick Piggin /* 9759875cf80SDavid Howells * Perform an automount 9769875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 9779875cf80SDavid Howells * were called with. 9781da177e4SLinus Torvalds */ 9799875cf80SDavid Howells static int follow_automount(struct path *path, unsigned flags, 9809875cf80SDavid Howells bool *need_mntput) 98131e6b01fSNick Piggin { 9829875cf80SDavid Howells struct vfsmount *mnt; 983ea5b778aSDavid Howells int err; 9849875cf80SDavid Howells 9859875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 9869875cf80SDavid Howells return -EREMOTE; 9879875cf80SDavid Howells 9880ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 9890ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 9900ec26fd0SMiklos Szeredi * the name. 9910ec26fd0SMiklos Szeredi * 9920ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 9935a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 9940ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 9950ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 9960ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 9970ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 9985a30d8a2SDavid Howells */ 9995a30d8a2SDavid Howells if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1000d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 10015a30d8a2SDavid Howells path->dentry->d_inode) 10029875cf80SDavid Howells return -EISDIR; 10030ec26fd0SMiklos Szeredi 10049875cf80SDavid Howells current->total_link_count++; 10059875cf80SDavid Howells if (current->total_link_count >= 40) 10069875cf80SDavid Howells return -ELOOP; 10079875cf80SDavid Howells 10089875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 10099875cf80SDavid Howells if (IS_ERR(mnt)) { 10109875cf80SDavid Howells /* 10119875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 10129875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 10139875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 10149875cf80SDavid Howells * 10159875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 10169875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 10179875cf80SDavid Howells * the path is inaccessible and we should say so. 10189875cf80SDavid Howells */ 101949084c3bSAl Viro if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT)) 10209875cf80SDavid Howells return -EREMOTE; 10219875cf80SDavid Howells return PTR_ERR(mnt); 102231e6b01fSNick Piggin } 1023ea5b778aSDavid Howells 10249875cf80SDavid Howells if (!mnt) /* mount collision */ 10259875cf80SDavid Howells return 0; 10269875cf80SDavid Howells 10278aef1884SAl Viro if (!*need_mntput) { 10288aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 10298aef1884SAl Viro mntget(path->mnt); 10308aef1884SAl Viro *need_mntput = true; 10318aef1884SAl Viro } 103219a167afSAl Viro err = finish_automount(mnt, path); 1033ea5b778aSDavid Howells 1034ea5b778aSDavid Howells switch (err) { 1035ea5b778aSDavid Howells case -EBUSY: 1036ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 103719a167afSAl Viro return 0; 1038ea5b778aSDavid Howells case 0: 10398aef1884SAl Viro path_put(path); 10409875cf80SDavid Howells path->mnt = mnt; 10419875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 10429875cf80SDavid Howells return 0; 104319a167afSAl Viro default: 104419a167afSAl Viro return err; 10459875cf80SDavid Howells } 104619a167afSAl Viro 1047ea5b778aSDavid Howells } 10489875cf80SDavid Howells 10499875cf80SDavid Howells /* 10509875cf80SDavid Howells * Handle a dentry that is managed in some way. 1051cc53ce53SDavid Howells * - Flagged for transit management (autofs) 10529875cf80SDavid Howells * - Flagged as mountpoint 10539875cf80SDavid Howells * - Flagged as automount point 10549875cf80SDavid Howells * 10559875cf80SDavid Howells * This may only be called in refwalk mode. 10569875cf80SDavid Howells * 10579875cf80SDavid Howells * Serialization is taken care of in namespace.c 10589875cf80SDavid Howells */ 10599875cf80SDavid Howells static int follow_managed(struct path *path, unsigned flags) 10609875cf80SDavid Howells { 10618aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 10629875cf80SDavid Howells unsigned managed; 10639875cf80SDavid Howells bool need_mntput = false; 10648aef1884SAl Viro int ret = 0; 10659875cf80SDavid Howells 10669875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 10679875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 10689875cf80SDavid Howells * the components of that value change under us */ 10699875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 10709875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 10719875cf80SDavid Howells unlikely(managed != 0)) { 1072cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1073cc53ce53SDavid Howells * being held. */ 1074cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1075cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1076cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 10771aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1078cc53ce53SDavid Howells if (ret < 0) 10798aef1884SAl Viro break; 1080cc53ce53SDavid Howells } 1081cc53ce53SDavid Howells 10829875cf80SDavid Howells /* Transit to a mounted filesystem. */ 10839875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 10849875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 10859875cf80SDavid Howells if (mounted) { 10869875cf80SDavid Howells dput(path->dentry); 10879875cf80SDavid Howells if (need_mntput) 1088463ffb2eSAl Viro mntput(path->mnt); 1089463ffb2eSAl Viro path->mnt = mounted; 1090463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 10919875cf80SDavid Howells need_mntput = true; 10929875cf80SDavid Howells continue; 1093463ffb2eSAl Viro } 1094463ffb2eSAl Viro 10959875cf80SDavid Howells /* Something is mounted on this dentry in another 10969875cf80SDavid Howells * namespace and/or whatever was mounted there in this 109748a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 109848a066e7SAl Viro * get it */ 10991da177e4SLinus Torvalds } 11009875cf80SDavid Howells 11019875cf80SDavid Howells /* Handle an automount point */ 11029875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 11039875cf80SDavid Howells ret = follow_automount(path, flags, &need_mntput); 11049875cf80SDavid Howells if (ret < 0) 11058aef1884SAl Viro break; 11069875cf80SDavid Howells continue; 11079875cf80SDavid Howells } 11089875cf80SDavid Howells 11099875cf80SDavid Howells /* We didn't change the current path point */ 11109875cf80SDavid Howells break; 11119875cf80SDavid Howells } 11128aef1884SAl Viro 11138aef1884SAl Viro if (need_mntput && path->mnt == mnt) 11148aef1884SAl Viro mntput(path->mnt); 11158aef1884SAl Viro if (ret == -EISDIR) 11168aef1884SAl Viro ret = 0; 1117a3fbbde7SAl Viro return ret < 0 ? ret : need_mntput; 11181da177e4SLinus Torvalds } 11191da177e4SLinus Torvalds 1120cc53ce53SDavid Howells int follow_down_one(struct path *path) 11211da177e4SLinus Torvalds { 11221da177e4SLinus Torvalds struct vfsmount *mounted; 11231da177e4SLinus Torvalds 11241c755af4SAl Viro mounted = lookup_mnt(path); 11251da177e4SLinus Torvalds if (mounted) { 11269393bd07SAl Viro dput(path->dentry); 11279393bd07SAl Viro mntput(path->mnt); 11289393bd07SAl Viro path->mnt = mounted; 11299393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 11301da177e4SLinus Torvalds return 1; 11311da177e4SLinus Torvalds } 11321da177e4SLinus Torvalds return 0; 11331da177e4SLinus Torvalds } 11344d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 11351da177e4SLinus Torvalds 1136b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 113762a7375eSIan Kent { 1138b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1139b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 114062a7375eSIan Kent } 114162a7375eSIan Kent 11429875cf80SDavid Howells /* 1143287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1144287548e4SAl Viro * we meet a managed dentry that would need blocking. 11459875cf80SDavid Howells */ 11469875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1147287548e4SAl Viro struct inode **inode) 11489875cf80SDavid Howells { 114962a7375eSIan Kent for (;;) { 1150c7105365SAl Viro struct mount *mounted; 115162a7375eSIan Kent /* 115262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 115362a7375eSIan Kent * that wants to block transit. 115462a7375eSIan Kent */ 1155b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1156b8faf035SNeilBrown case -ECHILD: 1157b8faf035SNeilBrown default: 1158ab90911fSDavid Howells return false; 1159b8faf035SNeilBrown case -EISDIR: 1160b8faf035SNeilBrown return true; 1161b8faf035SNeilBrown case 0: 1162b8faf035SNeilBrown break; 1163b8faf035SNeilBrown } 116462a7375eSIan Kent 116562a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1166b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 116762a7375eSIan Kent 1168474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 11699875cf80SDavid Howells if (!mounted) 11709875cf80SDavid Howells break; 1171c7105365SAl Viro path->mnt = &mounted->mnt; 1172c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1173a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 11749875cf80SDavid Howells nd->seq = read_seqcount_begin(&path->dentry->d_seq); 117559430262SLinus Torvalds /* 117659430262SLinus Torvalds * Update the inode too. We don't need to re-check the 117759430262SLinus Torvalds * dentry sequence number here after this d_inode read, 117859430262SLinus Torvalds * because a mount-point is always pinned. 117959430262SLinus Torvalds */ 118059430262SLinus Torvalds *inode = path->dentry->d_inode; 11819875cf80SDavid Howells } 1182f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1183b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1184287548e4SAl Viro } 1185287548e4SAl Viro 118631e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 118731e6b01fSNick Piggin { 11884023bfc9SAl Viro struct inode *inode = nd->inode; 11897bd88377SAl Viro if (!nd->root.mnt) 119031e6b01fSNick Piggin set_root_rcu(nd); 119131e6b01fSNick Piggin 119231e6b01fSNick Piggin while (1) { 119331e6b01fSNick Piggin if (nd->path.dentry == nd->root.dentry && 119431e6b01fSNick Piggin nd->path.mnt == nd->root.mnt) { 119531e6b01fSNick Piggin break; 119631e6b01fSNick Piggin } 119731e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 119831e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 119931e6b01fSNick Piggin struct dentry *parent = old->d_parent; 120031e6b01fSNick Piggin unsigned seq; 120131e6b01fSNick Piggin 12024023bfc9SAl Viro inode = parent->d_inode; 120331e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 120431e6b01fSNick Piggin if (read_seqcount_retry(&old->d_seq, nd->seq)) 1205ef7562d5SAl Viro goto failed; 120631e6b01fSNick Piggin nd->path.dentry = parent; 120731e6b01fSNick Piggin nd->seq = seq; 120831e6b01fSNick Piggin break; 120931e6b01fSNick Piggin } 121031e6b01fSNick Piggin if (!follow_up_rcu(&nd->path)) 121131e6b01fSNick Piggin break; 12124023bfc9SAl Viro inode = nd->path.dentry->d_inode; 121331e6b01fSNick Piggin nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 121431e6b01fSNick Piggin } 1215b37199e6SAl Viro while (d_mountpoint(nd->path.dentry)) { 1216b37199e6SAl Viro struct mount *mounted; 1217b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1218b37199e6SAl Viro if (!mounted) 1219b37199e6SAl Viro break; 1220b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1221b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 12224023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1223b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1224f5be3e29SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1225b37199e6SAl Viro goto failed; 1226b37199e6SAl Viro } 12274023bfc9SAl Viro nd->inode = inode; 122831e6b01fSNick Piggin return 0; 1229ef7562d5SAl Viro 1230ef7562d5SAl Viro failed: 1231ef7562d5SAl Viro return -ECHILD; 123231e6b01fSNick Piggin } 123331e6b01fSNick Piggin 12349875cf80SDavid Howells /* 1235cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1236cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1237cc53ce53SDavid Howells * caller is permitted to proceed or not. 1238cc53ce53SDavid Howells */ 12397cc90cc3SAl Viro int follow_down(struct path *path) 1240cc53ce53SDavid Howells { 1241cc53ce53SDavid Howells unsigned managed; 1242cc53ce53SDavid Howells int ret; 1243cc53ce53SDavid Howells 1244cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1245cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1246cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1247cc53ce53SDavid Howells * being held. 1248cc53ce53SDavid Howells * 1249cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1250cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1251cc53ce53SDavid Howells * other than its daemon the right to mount on its 1252cc53ce53SDavid Howells * superstructure. 1253cc53ce53SDavid Howells * 1254cc53ce53SDavid Howells * The filesystem may sleep at this point. 1255cc53ce53SDavid Howells */ 1256cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1257cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1258cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1259ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 12601aed3e42SAl Viro path->dentry, false); 1261cc53ce53SDavid Howells if (ret < 0) 1262cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1263cc53ce53SDavid Howells } 1264cc53ce53SDavid Howells 1265cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1266cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1267cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1268cc53ce53SDavid Howells if (!mounted) 1269cc53ce53SDavid Howells break; 1270cc53ce53SDavid Howells dput(path->dentry); 1271cc53ce53SDavid Howells mntput(path->mnt); 1272cc53ce53SDavid Howells path->mnt = mounted; 1273cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1274cc53ce53SDavid Howells continue; 1275cc53ce53SDavid Howells } 1276cc53ce53SDavid Howells 1277cc53ce53SDavid Howells /* Don't handle automount points here */ 1278cc53ce53SDavid Howells break; 1279cc53ce53SDavid Howells } 1280cc53ce53SDavid Howells return 0; 1281cc53ce53SDavid Howells } 12824d359507SAl Viro EXPORT_SYMBOL(follow_down); 1283cc53ce53SDavid Howells 1284cc53ce53SDavid Howells /* 12859875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 12869875cf80SDavid Howells */ 12879875cf80SDavid Howells static void follow_mount(struct path *path) 12889875cf80SDavid Howells { 12899875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 12909875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12919875cf80SDavid Howells if (!mounted) 12929875cf80SDavid Howells break; 12939875cf80SDavid Howells dput(path->dentry); 12949875cf80SDavid Howells mntput(path->mnt); 12959875cf80SDavid Howells path->mnt = mounted; 12969875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 12979875cf80SDavid Howells } 12989875cf80SDavid Howells } 12999875cf80SDavid Howells 130031e6b01fSNick Piggin static void follow_dotdot(struct nameidata *nd) 13011da177e4SLinus Torvalds { 13027bd88377SAl Viro if (!nd->root.mnt) 13032a737871SAl Viro set_root(nd); 1304e518ddb7SAndreas Mohr 13051da177e4SLinus Torvalds while(1) { 13064ac91378SJan Blunck struct dentry *old = nd->path.dentry; 13071da177e4SLinus Torvalds 13082a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 13092a737871SAl Viro nd->path.mnt == nd->root.mnt) { 13101da177e4SLinus Torvalds break; 13111da177e4SLinus Torvalds } 13124ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 13133088dd70SAl Viro /* rare case of legitimate dget_parent()... */ 13143088dd70SAl Viro nd->path.dentry = dget_parent(nd->path.dentry); 13151da177e4SLinus Torvalds dput(old); 13161da177e4SLinus Torvalds break; 13171da177e4SLinus Torvalds } 13183088dd70SAl Viro if (!follow_up(&nd->path)) 13191da177e4SLinus Torvalds break; 13201da177e4SLinus Torvalds } 132179ed0226SAl Viro follow_mount(&nd->path); 132231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 13231da177e4SLinus Torvalds } 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds /* 1326bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1327bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1328bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1329bad61189SMiklos Szeredi * 1330bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 1331baa03890SNick Piggin */ 1332bad61189SMiklos Szeredi static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir, 1333201f956eSAl Viro unsigned int flags, bool *need_lookup) 1334baa03890SNick Piggin { 1335baa03890SNick Piggin struct dentry *dentry; 1336bad61189SMiklos Szeredi int error; 1337baa03890SNick Piggin 1338bad61189SMiklos Szeredi *need_lookup = false; 1339bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1340bad61189SMiklos Szeredi if (dentry) { 134139e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1342201f956eSAl Viro error = d_revalidate(dentry, flags); 1343bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 1344bad61189SMiklos Szeredi if (error < 0) { 1345bad61189SMiklos Szeredi dput(dentry); 1346bad61189SMiklos Szeredi return ERR_PTR(error); 13475542aa2fSEric W. Biederman } else { 13485542aa2fSEric W. Biederman d_invalidate(dentry); 1349bad61189SMiklos Szeredi dput(dentry); 1350bad61189SMiklos Szeredi dentry = NULL; 1351bad61189SMiklos Szeredi } 1352bad61189SMiklos Szeredi } 1353bad61189SMiklos Szeredi } 1354bad61189SMiklos Szeredi } 1355baa03890SNick Piggin 1356bad61189SMiklos Szeredi if (!dentry) { 1357bad61189SMiklos Szeredi dentry = d_alloc(dir, name); 1358baa03890SNick Piggin if (unlikely(!dentry)) 1359baa03890SNick Piggin return ERR_PTR(-ENOMEM); 1360baa03890SNick Piggin 1361bad61189SMiklos Szeredi *need_lookup = true; 1362baa03890SNick Piggin } 1363baa03890SNick Piggin return dentry; 1364baa03890SNick Piggin } 1365baa03890SNick Piggin 1366baa03890SNick Piggin /* 136713a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 136813a2c3beSJ. Bruce Fields * unhashed. 1369bad61189SMiklos Szeredi * 1370bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 137144396f4bSJosef Bacik */ 1372bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 137372bd866aSAl Viro unsigned int flags) 137444396f4bSJosef Bacik { 137544396f4bSJosef Bacik struct dentry *old; 137644396f4bSJosef Bacik 137744396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1378bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1379e188dc02SMiklos Szeredi dput(dentry); 138044396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1381e188dc02SMiklos Szeredi } 138244396f4bSJosef Bacik 138372bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 138444396f4bSJosef Bacik if (unlikely(old)) { 138544396f4bSJosef Bacik dput(dentry); 138644396f4bSJosef Bacik dentry = old; 138744396f4bSJosef Bacik } 138844396f4bSJosef Bacik return dentry; 138944396f4bSJosef Bacik } 139044396f4bSJosef Bacik 1391a3255546SAl Viro static struct dentry *__lookup_hash(struct qstr *name, 139272bd866aSAl Viro struct dentry *base, unsigned int flags) 1393a3255546SAl Viro { 1394bad61189SMiklos Szeredi bool need_lookup; 1395a3255546SAl Viro struct dentry *dentry; 1396a3255546SAl Viro 139772bd866aSAl Viro dentry = lookup_dcache(name, base, flags, &need_lookup); 1398bad61189SMiklos Szeredi if (!need_lookup) 1399a3255546SAl Viro return dentry; 1400bad61189SMiklos Szeredi 140172bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1402a3255546SAl Viro } 1403a3255546SAl Viro 140444396f4bSJosef Bacik /* 14051da177e4SLinus Torvalds * It's more convoluted than I'd like it to be, but... it's still fairly 14061da177e4SLinus Torvalds * small and for now I'd prefer to have fast path as straight as possible. 14071da177e4SLinus Torvalds * It _is_ time-critical. 14081da177e4SLinus Torvalds */ 1409e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 141031e6b01fSNick Piggin struct path *path, struct inode **inode) 14111da177e4SLinus Torvalds { 14124ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 141331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 14145a18fff2SAl Viro int need_reval = 1; 14155a18fff2SAl Viro int status = 1; 14169875cf80SDavid Howells int err; 14179875cf80SDavid Howells 14183cac260aSAl Viro /* 1419b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 1420b04f784eSNick Piggin * of a false negative due to a concurrent rename, we're going to 1421b04f784eSNick Piggin * do the non-racy lookup, below. 1422b04f784eSNick Piggin */ 142331e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 142431e6b01fSNick Piggin unsigned seq; 1425766c4cbfSAl Viro bool negative; 1426da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 14275a18fff2SAl Viro if (!dentry) 14285a18fff2SAl Viro goto unlazy; 14295a18fff2SAl Viro 143012f8ad4bSLinus Torvalds /* 143112f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 143212f8ad4bSLinus Torvalds * the dentry name information from lookup. 143312f8ad4bSLinus Torvalds */ 143412f8ad4bSLinus Torvalds *inode = dentry->d_inode; 1435766c4cbfSAl Viro negative = d_is_negative(dentry); 143612f8ad4bSLinus Torvalds if (read_seqcount_retry(&dentry->d_seq, seq)) 143712f8ad4bSLinus Torvalds return -ECHILD; 1438766c4cbfSAl Viro if (negative) 1439766c4cbfSAl Viro return -ENOENT; 144012f8ad4bSLinus Torvalds 144112f8ad4bSLinus Torvalds /* 144212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 144312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 144412f8ad4bSLinus Torvalds * 144512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 144612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 144712f8ad4bSLinus Torvalds */ 144831e6b01fSNick Piggin if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 144931e6b01fSNick Piggin return -ECHILD; 145031e6b01fSNick Piggin nd->seq = seq; 14515a18fff2SAl Viro 145224643087SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) { 14534ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14545a18fff2SAl Viro if (unlikely(status <= 0)) { 14555a18fff2SAl Viro if (status != -ECHILD) 14565a18fff2SAl Viro need_reval = 0; 14575a18fff2SAl Viro goto unlazy; 14585a18fff2SAl Viro } 145924643087SAl Viro } 146031e6b01fSNick Piggin path->mnt = mnt; 146131e6b01fSNick Piggin path->dentry = dentry; 1462b8faf035SNeilBrown if (likely(__follow_mount_rcu(nd, path, inode))) 14639875cf80SDavid Howells return 0; 14645a18fff2SAl Viro unlazy: 146519660af7SAl Viro if (unlazy_walk(nd, dentry)) 14665a18fff2SAl Viro return -ECHILD; 14675a18fff2SAl Viro } else { 1468e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 146924643087SAl Viro } 14705a18fff2SAl Viro 147181e6f520SAl Viro if (unlikely(!dentry)) 147281e6f520SAl Viro goto need_lookup; 14735a18fff2SAl Viro 14745a18fff2SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval) 14754ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 14765a18fff2SAl Viro if (unlikely(status <= 0)) { 14775a18fff2SAl Viro if (status < 0) { 14785a18fff2SAl Viro dput(dentry); 14795a18fff2SAl Viro return status; 14805a18fff2SAl Viro } 14815542aa2fSEric W. Biederman d_invalidate(dentry); 14825a18fff2SAl Viro dput(dentry); 148381e6f520SAl Viro goto need_lookup; 14845a18fff2SAl Viro } 1485697f514dSMiklos Szeredi 1486766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1487766c4cbfSAl Viro dput(dentry); 1488766c4cbfSAl Viro return -ENOENT; 1489766c4cbfSAl Viro } 14901da177e4SLinus Torvalds path->mnt = mnt; 14911da177e4SLinus Torvalds path->dentry = dentry; 14929875cf80SDavid Howells err = follow_managed(path, nd->flags); 149389312214SIan Kent if (unlikely(err < 0)) { 149489312214SIan Kent path_put_conditional(path, nd); 14959875cf80SDavid Howells return err; 149689312214SIan Kent } 1497a3fbbde7SAl Viro if (err) 1498a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 149931e6b01fSNick Piggin *inode = path->dentry->d_inode; 15001da177e4SLinus Torvalds return 0; 150181e6f520SAl Viro 150281e6f520SAl Viro need_lookup: 1503697f514dSMiklos Szeredi return 1; 1504697f514dSMiklos Szeredi } 1505697f514dSMiklos Szeredi 1506697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1507cc2a5271SAl Viro static int lookup_slow(struct nameidata *nd, struct path *path) 1508697f514dSMiklos Szeredi { 1509697f514dSMiklos Szeredi struct dentry *dentry, *parent; 1510697f514dSMiklos Szeredi int err; 1511697f514dSMiklos Szeredi 1512697f514dSMiklos Szeredi parent = nd->path.dentry; 151381e6f520SAl Viro BUG_ON(nd->inode != parent->d_inode); 151481e6f520SAl Viro 151581e6f520SAl Viro mutex_lock(&parent->d_inode->i_mutex); 1516cc2a5271SAl Viro dentry = __lookup_hash(&nd->last, parent, nd->flags); 151781e6f520SAl Viro mutex_unlock(&parent->d_inode->i_mutex); 151881e6f520SAl Viro if (IS_ERR(dentry)) 151981e6f520SAl Viro return PTR_ERR(dentry); 1520697f514dSMiklos Szeredi path->mnt = nd->path.mnt; 1521697f514dSMiklos Szeredi path->dentry = dentry; 1522697f514dSMiklos Szeredi err = follow_managed(path, nd->flags); 1523697f514dSMiklos Szeredi if (unlikely(err < 0)) { 1524697f514dSMiklos Szeredi path_put_conditional(path, nd); 1525697f514dSMiklos Szeredi return err; 1526697f514dSMiklos Szeredi } 1527697f514dSMiklos Szeredi if (err) 1528697f514dSMiklos Szeredi nd->flags |= LOOKUP_JUMPED; 1529697f514dSMiklos Szeredi return 0; 15301da177e4SLinus Torvalds } 15311da177e4SLinus Torvalds 153252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 153352094c8aSAl Viro { 153452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 15354ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 153652094c8aSAl Viro if (err != -ECHILD) 153752094c8aSAl Viro return err; 153819660af7SAl Viro if (unlazy_walk(nd, NULL)) 153952094c8aSAl Viro return -ECHILD; 154052094c8aSAl Viro } 15414ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 154252094c8aSAl Viro } 154352094c8aSAl Viro 15449856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 15459856fa1bSAl Viro { 15469856fa1bSAl Viro if (type == LAST_DOTDOT) { 15479856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 154870291aecSAl Viro return follow_dotdot_rcu(nd); 15499856fa1bSAl Viro } else 15509856fa1bSAl Viro follow_dotdot(nd); 15519856fa1bSAl Viro } 15529856fa1bSAl Viro return 0; 15539856fa1bSAl Viro } 15549856fa1bSAl Viro 1555951361f9SAl Viro static void terminate_walk(struct nameidata *nd) 1556951361f9SAl Viro { 1557951361f9SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1558951361f9SAl Viro path_put(&nd->path); 1559951361f9SAl Viro } else { 1560951361f9SAl Viro nd->flags &= ~LOOKUP_RCU; 15615b6ca027SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1562951361f9SAl Viro nd->root.mnt = NULL; 15638b61e74fSAl Viro rcu_read_unlock(); 1564951361f9SAl Viro } 156515439726SAl Viro while (unlikely(nd->depth)) 156615439726SAl Viro put_link(nd); 1567951361f9SAl Viro } 1568951361f9SAl Viro 15693ddcd056SLinus Torvalds /* 15703ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 15713ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 15723ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 15733ddcd056SLinus Torvalds * for the common case. 15743ddcd056SLinus Torvalds */ 1575b18825a7SDavid Howells static inline int should_follow_link(struct dentry *dentry, int follow) 15763ddcd056SLinus Torvalds { 1577b18825a7SDavid Howells return unlikely(d_is_symlink(dentry)) ? follow : 0; 15783ddcd056SLinus Torvalds } 15793ddcd056SLinus Torvalds 1580caa85634SAl Viro static int walk_component(struct nameidata *nd, int follow) 1581ce57dfc1SAl Viro { 1582caa85634SAl Viro struct path path; 1583ce57dfc1SAl Viro struct inode *inode; 1584ce57dfc1SAl Viro int err; 1585ce57dfc1SAl Viro /* 1586ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1587ce57dfc1SAl Viro * to be able to know about the current root directory and 1588ce57dfc1SAl Viro * parent relationships. 1589ce57dfc1SAl Viro */ 1590f0a9ba70SAl Viro if (unlikely(nd->last_type != LAST_NORM)) 1591f0a9ba70SAl Viro return handle_dots(nd, nd->last_type); 1592caa85634SAl Viro err = lookup_fast(nd, &path, &inode); 1593ce57dfc1SAl Viro if (unlikely(err)) { 1594697f514dSMiklos Szeredi if (err < 0) 1595f0a9ba70SAl Viro return err; 1596697f514dSMiklos Szeredi 1597caa85634SAl Viro err = lookup_slow(nd, &path); 1598697f514dSMiklos Szeredi if (err < 0) 1599f0a9ba70SAl Viro return err; 1600697f514dSMiklos Szeredi 1601caa85634SAl Viro inode = path.dentry->d_inode; 1602697f514dSMiklos Szeredi err = -ENOENT; 1603caa85634SAl Viro if (d_is_negative(path.dentry)) 1604697f514dSMiklos Szeredi goto out_path_put; 1605766c4cbfSAl Viro } 1606697f514dSMiklos Szeredi 1607caa85634SAl Viro if (should_follow_link(path.dentry, follow)) { 160819660af7SAl Viro if (nd->flags & LOOKUP_RCU) { 1609caa85634SAl Viro if (unlikely(nd->path.mnt != path.mnt || 1610caa85634SAl Viro unlazy_walk(nd, path.dentry))) { 1611f0a9ba70SAl Viro return -ECHILD; 161219660af7SAl Viro } 161319660af7SAl Viro } 1614caa85634SAl Viro BUG_ON(inode != path.dentry->d_inode); 1615caa85634SAl Viro nd->link = path; 1616ce57dfc1SAl Viro return 1; 1617ce57dfc1SAl Viro } 1618caa85634SAl Viro path_to_nameidata(&path, nd); 1619ce57dfc1SAl Viro nd->inode = inode; 1620ce57dfc1SAl Viro return 0; 1621697f514dSMiklos Szeredi 1622697f514dSMiklos Szeredi out_path_put: 1623caa85634SAl Viro path_to_nameidata(&path, nd); 1624697f514dSMiklos Szeredi return err; 1625ce57dfc1SAl Viro } 1626ce57dfc1SAl Viro 16271da177e4SLinus Torvalds /* 1628bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1629bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1630bfcfaa77SLinus Torvalds * 1631bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1632bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1633bfcfaa77SLinus Torvalds * fast. 1634bfcfaa77SLinus Torvalds * 1635bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1636bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1637bfcfaa77SLinus Torvalds * crossing operation. 1638bfcfaa77SLinus Torvalds * 1639bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1640bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1641bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1642bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1643bfcfaa77SLinus Torvalds */ 1644bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1645bfcfaa77SLinus Torvalds 1646f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1647bfcfaa77SLinus Torvalds 1648f68e556eSLinus Torvalds #ifdef CONFIG_64BIT 1649bfcfaa77SLinus Torvalds 1650bfcfaa77SLinus Torvalds static inline unsigned int fold_hash(unsigned long hash) 1651bfcfaa77SLinus Torvalds { 165299d263d4SLinus Torvalds return hash_64(hash, 32); 1653bfcfaa77SLinus Torvalds } 1654bfcfaa77SLinus Torvalds 1655bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1656bfcfaa77SLinus Torvalds 1657bfcfaa77SLinus Torvalds #define fold_hash(x) (x) 1658bfcfaa77SLinus Torvalds 1659bfcfaa77SLinus Torvalds #endif 1660bfcfaa77SLinus Torvalds 1661bfcfaa77SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 1662bfcfaa77SLinus Torvalds { 1663bfcfaa77SLinus Torvalds unsigned long a, mask; 1664bfcfaa77SLinus Torvalds unsigned long hash = 0; 1665bfcfaa77SLinus Torvalds 1666bfcfaa77SLinus Torvalds for (;;) { 1667e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1668bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1669bfcfaa77SLinus Torvalds break; 1670bfcfaa77SLinus Torvalds hash += a; 1671f132c5beSAl Viro hash *= 9; 1672bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1673bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1674bfcfaa77SLinus Torvalds if (!len) 1675bfcfaa77SLinus Torvalds goto done; 1676bfcfaa77SLinus Torvalds } 1677a5c21dceSWill Deacon mask = bytemask_from_count(len); 1678bfcfaa77SLinus Torvalds hash += mask & a; 1679bfcfaa77SLinus Torvalds done: 1680bfcfaa77SLinus Torvalds return fold_hash(hash); 1681bfcfaa77SLinus Torvalds } 1682bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1683bfcfaa77SLinus Torvalds 1684bfcfaa77SLinus Torvalds /* 1685bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1686d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1687bfcfaa77SLinus Torvalds */ 1688d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1689bfcfaa77SLinus Torvalds { 169036126f8fSLinus Torvalds unsigned long a, b, adata, bdata, mask, hash, len; 169136126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1692bfcfaa77SLinus Torvalds 1693bfcfaa77SLinus Torvalds hash = a = 0; 1694bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1695bfcfaa77SLinus Torvalds do { 1696bfcfaa77SLinus Torvalds hash = (hash + a) * 9; 1697bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1698e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 169936126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 170036126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1701bfcfaa77SLinus Torvalds 170236126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 170336126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 170436126f8fSLinus Torvalds 170536126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 170636126f8fSLinus Torvalds 170736126f8fSLinus Torvalds hash += a & zero_bytemask(mask); 17089226b5b4SLinus Torvalds len += find_zero(mask); 1709d6bb3e90SLinus Torvalds return hashlen_create(fold_hash(hash), len); 1710bfcfaa77SLinus Torvalds } 1711bfcfaa77SLinus Torvalds 1712bfcfaa77SLinus Torvalds #else 1713bfcfaa77SLinus Torvalds 17140145acc2SLinus Torvalds unsigned int full_name_hash(const unsigned char *name, unsigned int len) 17150145acc2SLinus Torvalds { 17160145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 17170145acc2SLinus Torvalds while (len--) 17180145acc2SLinus Torvalds hash = partial_name_hash(*name++, hash); 17190145acc2SLinus Torvalds return end_name_hash(hash); 17200145acc2SLinus Torvalds } 1721ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 17220145acc2SLinus Torvalds 17233ddcd056SLinus Torvalds /* 1724200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1725200e9ef7SLinus Torvalds * one character. 1726200e9ef7SLinus Torvalds */ 1727d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1728200e9ef7SLinus Torvalds { 1729200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 1730200e9ef7SLinus Torvalds unsigned long len = 0, c; 1731200e9ef7SLinus Torvalds 1732200e9ef7SLinus Torvalds c = (unsigned char)*name; 1733200e9ef7SLinus Torvalds do { 1734200e9ef7SLinus Torvalds len++; 1735200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 1736200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 1737200e9ef7SLinus Torvalds } while (c && c != '/'); 1738d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 1739200e9ef7SLinus Torvalds } 1740200e9ef7SLinus Torvalds 1741bfcfaa77SLinus Torvalds #endif 1742bfcfaa77SLinus Torvalds 1743200e9ef7SLinus Torvalds /* 17441da177e4SLinus Torvalds * Name resolution. 1745ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 1746ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 17471da177e4SLinus Torvalds * 1748ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 1749ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 17501da177e4SLinus Torvalds */ 17516de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 17521da177e4SLinus Torvalds { 17531da177e4SLinus Torvalds int err; 17541da177e4SLinus Torvalds 17551da177e4SLinus Torvalds while (*name=='/') 17561da177e4SLinus Torvalds name++; 17571da177e4SLinus Torvalds if (!*name) 17589e18f10aSAl Viro return 0; 17591da177e4SLinus Torvalds 17601da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 17611da177e4SLinus Torvalds for(;;) { 1762d6bb3e90SLinus Torvalds u64 hash_len; 1763fe479a58SAl Viro int type; 17641da177e4SLinus Torvalds 176552094c8aSAl Viro err = may_lookup(nd); 17661da177e4SLinus Torvalds if (err) 17671da177e4SLinus Torvalds break; 17681da177e4SLinus Torvalds 1769d6bb3e90SLinus Torvalds hash_len = hash_name(name); 17701da177e4SLinus Torvalds 1771fe479a58SAl Viro type = LAST_NORM; 1772d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 1773fe479a58SAl Viro case 2: 1774200e9ef7SLinus Torvalds if (name[1] == '.') { 1775fe479a58SAl Viro type = LAST_DOTDOT; 177616c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 177716c2cd71SAl Viro } 1778fe479a58SAl Viro break; 1779fe479a58SAl Viro case 1: 1780fe479a58SAl Viro type = LAST_DOT; 1781fe479a58SAl Viro } 17825a202bcdSAl Viro if (likely(type == LAST_NORM)) { 17835a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 178416c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 17855a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 1786a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 1787da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 17885a202bcdSAl Viro if (err < 0) 17895a202bcdSAl Viro break; 1790d6bb3e90SLinus Torvalds hash_len = this.hash_len; 1791d6bb3e90SLinus Torvalds name = this.name; 17925a202bcdSAl Viro } 17935a202bcdSAl Viro } 1794fe479a58SAl Viro 1795d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 1796d6bb3e90SLinus Torvalds nd->last.name = name; 17975f4a6a69SAl Viro nd->last_type = type; 17985f4a6a69SAl Viro 1799d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 1800d6bb3e90SLinus Torvalds if (!*name) 1801bdf6cbf1SAl Viro goto OK; 1802200e9ef7SLinus Torvalds /* 1803200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 1804200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 1805200e9ef7SLinus Torvalds */ 1806200e9ef7SLinus Torvalds do { 1807d6bb3e90SLinus Torvalds name++; 1808d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 18098620c238SAl Viro if (unlikely(!*name)) { 18108620c238SAl Viro OK: 18118620c238SAl Viro /* called from path_init(), done */ 18128620c238SAl Viro if (!nd->depth) 18138620c238SAl Viro return 0; 18148620c238SAl Viro name = nd->stack[nd->depth - 1].name; 18158620c238SAl Viro /* called from trailing_symlink(), done */ 18168620c238SAl Viro if (!name) 18178620c238SAl Viro return 0; 18188620c238SAl Viro /* last component of nested symlink */ 1819caa85634SAl Viro err = walk_component(nd, LOOKUP_FOLLOW); 18208620c238SAl Viro put_link(nd); 18218620c238SAl Viro } else { 18228620c238SAl Viro err = walk_component(nd, LOOKUP_FOLLOW); 18238620c238SAl Viro } 1824ce57dfc1SAl Viro if (err < 0) 1825f0a9ba70SAl Viro break; 1826fe479a58SAl Viro 1827ce57dfc1SAl Viro if (err) { 1828d40bcc09SAl Viro const char *s; 1829d40bcc09SAl Viro 1830894bc8c4SAl Viro err = nd_alloc_stack(nd); 1831894bc8c4SAl Viro if (unlikely(err)) { 1832894bc8c4SAl Viro path_to_nameidata(&nd->link, nd); 1833894bc8c4SAl Viro break; 18345a460275SAl Viro } 18355a460275SAl Viro 18363b2e7f75SAl Viro s = get_link(nd); 18375a460275SAl Viro 1838172a39a0SAl Viro if (unlikely(IS_ERR(s))) { 1839172a39a0SAl Viro err = PTR_ERR(s); 18401bc4b813SAl Viro break; 1841172a39a0SAl Viro } 1842172a39a0SAl Viro err = 0; 184312b09578SAl Viro if (unlikely(!s)) { 184412b09578SAl Viro /* jumped */ 1845b9ff4429SAl Viro put_link(nd); 184612b09578SAl Viro } else { 1847172a39a0SAl Viro if (*s == '/') { 1848172a39a0SAl Viro if (!nd->root.mnt) 1849172a39a0SAl Viro set_root(nd); 1850172a39a0SAl Viro path_put(&nd->path); 1851172a39a0SAl Viro nd->path = nd->root; 1852172a39a0SAl Viro path_get(&nd->root); 1853172a39a0SAl Viro nd->flags |= LOOKUP_JUMPED; 18549e18f10aSAl Viro while (unlikely(*++s == '/')) 18559e18f10aSAl Viro ; 1856172a39a0SAl Viro } 1857172a39a0SAl Viro nd->inode = nd->path.dentry->d_inode; 1858071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 18599e18f10aSAl Viro if (!*s) 18609e18f10aSAl Viro goto OK; 186132cd7468SAl Viro name = s; 18629e18f10aSAl Viro continue; 186348c8b0c5SAl Viro } 186431e6b01fSNick Piggin } 186544b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 18663ddcd056SLinus Torvalds err = -ENOTDIR; 18673ddcd056SLinus Torvalds break; 18685f4a6a69SAl Viro } 1869ce57dfc1SAl Viro } 1870951361f9SAl Viro terminate_walk(nd); 18711da177e4SLinus Torvalds return err; 187207681481SAl Viro } 18731da177e4SLinus Torvalds 18746e8a1f87SAl Viro static int path_init(int dfd, const struct filename *name, unsigned int flags, 18755e53084dSAl Viro struct nameidata *nd) 187631e6b01fSNick Piggin { 187731e6b01fSNick Piggin int retval = 0; 1878fd2f7cb5SAl Viro const char *s = name->name; 187931e6b01fSNick Piggin 188031e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 1881980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 188231e6b01fSNick Piggin nd->depth = 0; 18835e53084dSAl Viro nd->base = NULL; 18845b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 1885b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 1886b18825a7SDavid Howells struct inode *inode = root->d_inode; 1887fd2f7cb5SAl Viro if (*s) { 188844b1d530SMiklos Szeredi if (!d_can_lookup(root)) 18895b6ca027SAl Viro return -ENOTDIR; 18905b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 18915b6ca027SAl Viro if (retval) 18925b6ca027SAl Viro return retval; 189373d049a4SAl Viro } 18945b6ca027SAl Viro nd->path = nd->root; 18955b6ca027SAl Viro nd->inode = inode; 18965b6ca027SAl Viro if (flags & LOOKUP_RCU) { 18978b61e74fSAl Viro rcu_read_lock(); 18985b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 189948a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 19005b6ca027SAl Viro } else { 19015b6ca027SAl Viro path_get(&nd->path); 19025b6ca027SAl Viro } 1903d465887fSAl Viro goto done; 19045b6ca027SAl Viro } 19055b6ca027SAl Viro 190631e6b01fSNick Piggin nd->root.mnt = NULL; 190731e6b01fSNick Piggin 190848a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 1909fd2f7cb5SAl Viro if (*s == '/') { 1910e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19118b61e74fSAl Viro rcu_read_lock(); 19127bd88377SAl Viro nd->seq = set_root_rcu(nd); 1913e41f7d4eSAl Viro } else { 1914e41f7d4eSAl Viro set_root(nd); 1915e41f7d4eSAl Viro path_get(&nd->root); 1916e41f7d4eSAl Viro } 191731e6b01fSNick Piggin nd->path = nd->root; 191831e6b01fSNick Piggin } else if (dfd == AT_FDCWD) { 1919e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 192031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 1921c28cc364SNick Piggin unsigned seq; 192231e6b01fSNick Piggin 19238b61e74fSAl Viro rcu_read_lock(); 192431e6b01fSNick Piggin 1925c28cc364SNick Piggin do { 1926c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 192731e6b01fSNick Piggin nd->path = fs->pwd; 1928c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 1929c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 1930e41f7d4eSAl Viro } else { 1931e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 1932e41f7d4eSAl Viro } 193331e6b01fSNick Piggin } else { 1934582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 19352903ff01SAl Viro struct fd f = fdget_raw(dfd); 193631e6b01fSNick Piggin struct dentry *dentry; 193731e6b01fSNick Piggin 19382903ff01SAl Viro if (!f.file) 19392903ff01SAl Viro return -EBADF; 194031e6b01fSNick Piggin 19412903ff01SAl Viro dentry = f.file->f_path.dentry; 194231e6b01fSNick Piggin 1943fd2f7cb5SAl Viro if (*s) { 194444b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 19452903ff01SAl Viro fdput(f); 19462903ff01SAl Viro return -ENOTDIR; 1947f52e0c11SAl Viro } 19482903ff01SAl Viro } 19492903ff01SAl Viro 19502903ff01SAl Viro nd->path = f.file->f_path; 1951e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 19529c225f26SLinus Torvalds if (f.flags & FDPUT_FPUT) 19535e53084dSAl Viro nd->base = f.file; 1954c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 19558b61e74fSAl Viro rcu_read_lock(); 19565590ff0dSUlrich Drepper } else { 19572903ff01SAl Viro path_get(&nd->path); 19582903ff01SAl Viro fdput(f); 19591da177e4SLinus Torvalds } 1960e41f7d4eSAl Viro } 1961e41f7d4eSAl Viro 196231e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 19634023bfc9SAl Viro if (!(flags & LOOKUP_RCU)) 1964d465887fSAl Viro goto done; 19654023bfc9SAl Viro if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq))) 1966d465887fSAl Viro goto done; 19674023bfc9SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 19684023bfc9SAl Viro nd->root.mnt = NULL; 19694023bfc9SAl Viro rcu_read_unlock(); 19704023bfc9SAl Viro return -ECHILD; 1971d465887fSAl Viro done: 1972d465887fSAl Viro current->total_link_count = 0; 1973dc7af8dcSAl Viro return link_path_walk(s, nd); 19749b4a9b14SAl Viro } 19759b4a9b14SAl Viro 1976893b7775SAl Viro static void path_cleanup(struct nameidata *nd) 1977893b7775SAl Viro { 1978893b7775SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 1979893b7775SAl Viro path_put(&nd->root); 1980893b7775SAl Viro nd->root.mnt = NULL; 1981893b7775SAl Viro } 1982893b7775SAl Viro if (unlikely(nd->base)) 1983893b7775SAl Viro fput(nd->base); 19849b4a9b14SAl Viro } 19859b4a9b14SAl Viro 19861d8e03d3SAl Viro static int trailing_symlink(struct nameidata *nd) 198795fa25d9SAl Viro { 198895fa25d9SAl Viro const char *s; 19891d8e03d3SAl Viro int error = may_follow_link(&nd->link, nd); 199095fa25d9SAl Viro if (unlikely(error)) 199195fa25d9SAl Viro return error; 199295fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 19933b2e7f75SAl Viro s = get_link(nd); 19941bc4b813SAl Viro if (unlikely(IS_ERR(s))) { 19951bc4b813SAl Viro terminate_walk(nd); 199695fa25d9SAl Viro return PTR_ERR(s); 19971bc4b813SAl Viro } 19989ea57b72SAl Viro if (unlikely(!s)) 199995fa25d9SAl Viro return 0; 200095fa25d9SAl Viro if (*s == '/') { 200195fa25d9SAl Viro if (!nd->root.mnt) 200295fa25d9SAl Viro set_root(nd); 200395fa25d9SAl Viro path_put(&nd->path); 200495fa25d9SAl Viro nd->path = nd->root; 200595fa25d9SAl Viro path_get(&nd->root); 200695fa25d9SAl Viro nd->flags |= LOOKUP_JUMPED; 200795fa25d9SAl Viro } 200895fa25d9SAl Viro nd->inode = nd->path.dentry->d_inode; 2009939724dfSAl Viro nd->stack[0].name = NULL; 20108eff733aSAl Viro return link_path_walk(s, nd); 201195fa25d9SAl Viro } 201295fa25d9SAl Viro 2013caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2014bd92d7feSAl Viro { 2015f0a9ba70SAl Viro int err; 2016bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2017bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2018bd92d7feSAl Viro 2019bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2020f0a9ba70SAl Viro err = walk_component(nd, nd->flags & LOOKUP_FOLLOW); 2021191d7f73SAl Viro if (nd->depth) 2022191d7f73SAl Viro put_link(nd); 2023f0a9ba70SAl Viro if (err < 0) 2024f0a9ba70SAl Viro terminate_walk(nd); 2025f0a9ba70SAl Viro return err; 2026bd92d7feSAl Viro } 2027bd92d7feSAl Viro 20289b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 20295eb6b495SAl Viro static int path_lookupat(int dfd, const struct filename *name, 20309b4a9b14SAl Viro unsigned int flags, struct nameidata *nd) 20319b4a9b14SAl Viro { 2032bd92d7feSAl Viro int err; 203331e6b01fSNick Piggin 203431e6b01fSNick Piggin /* 203531e6b01fSNick Piggin * Path walking is largely split up into 2 different synchronisation 203631e6b01fSNick Piggin * schemes, rcu-walk and ref-walk (explained in 203731e6b01fSNick Piggin * Documentation/filesystems/path-lookup.txt). These share much of the 203831e6b01fSNick Piggin * path walk code, but some things particularly setup, cleanup, and 203931e6b01fSNick Piggin * following mounts are sufficiently divergent that functions are 204031e6b01fSNick Piggin * duplicated. Typically there is a function foo(), and its RCU 204131e6b01fSNick Piggin * analogue, foo_rcu(). 204231e6b01fSNick Piggin * 204331e6b01fSNick Piggin * -ECHILD is the error number of choice (just to avoid clashes) that 204431e6b01fSNick Piggin * is returned if some aspect of an rcu-walk fails. Such an error must 204531e6b01fSNick Piggin * be handled by restarting a traditional ref-walk (which will always 204631e6b01fSNick Piggin * be able to complete). 204731e6b01fSNick Piggin */ 20486e8a1f87SAl Viro err = path_init(dfd, name, flags, nd); 2049bd92d7feSAl Viro if (!err && !(flags & LOOKUP_PARENT)) { 2050191d7f73SAl Viro while ((err = lookup_last(nd)) > 0) { 20511d8e03d3SAl Viro err = trailing_symlink(nd); 20526d7b5aaeSAl Viro if (err) 20536d7b5aaeSAl Viro break; 2054bd92d7feSAl Viro } 2055bd92d7feSAl Viro } 2056ee0827cdSAl Viro 20579f1fafeeSAl Viro if (!err) 20589f1fafeeSAl Viro err = complete_walk(nd); 2059bd92d7feSAl Viro 2060bd92d7feSAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) { 206144b1d530SMiklos Szeredi if (!d_can_lookup(nd->path.dentry)) { 2062bd92d7feSAl Viro path_put(&nd->path); 2063bd23a539SAl Viro err = -ENOTDIR; 2064bd92d7feSAl Viro } 2065bd92d7feSAl Viro } 206616c2cd71SAl Viro 2067893b7775SAl Viro path_cleanup(nd); 2068bd92d7feSAl Viro return err; 206931e6b01fSNick Piggin } 207031e6b01fSNick Piggin 2071873f1eedSJeff Layton static int filename_lookup(int dfd, struct filename *name, 2072873f1eedSJeff Layton unsigned int flags, struct nameidata *nd) 2073873f1eedSJeff Layton { 2074894bc8c4SAl Viro int retval; 2075894bc8c4SAl Viro 2076894bc8c4SAl Viro set_nameidata(nd); 2077894bc8c4SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd); 2078894bc8c4SAl Viro 2079873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 20805eb6b495SAl Viro retval = path_lookupat(dfd, name, flags, nd); 2081873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 20825eb6b495SAl Viro retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd); 2083873f1eedSJeff Layton 2084873f1eedSJeff Layton if (likely(!retval)) 2085adb5c247SJeff Layton audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT); 2086894bc8c4SAl Viro restore_nameidata(nd); 2087873f1eedSJeff Layton return retval; 2088873f1eedSJeff Layton } 2089873f1eedSJeff Layton 209079714f72SAl Viro /* does lookup, returns the object with parent locked */ 209179714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 20925590ff0dSUlrich Drepper { 209351689104SPaul Moore struct filename *filename = getname_kernel(name); 209479714f72SAl Viro struct nameidata nd; 209579714f72SAl Viro struct dentry *d; 209651689104SPaul Moore int err; 209751689104SPaul Moore 209851689104SPaul Moore if (IS_ERR(filename)) 209951689104SPaul Moore return ERR_CAST(filename); 210051689104SPaul Moore 210151689104SPaul Moore err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd); 210251689104SPaul Moore if (err) { 210351689104SPaul Moore d = ERR_PTR(err); 210451689104SPaul Moore goto out; 210551689104SPaul Moore } 210679714f72SAl Viro if (nd.last_type != LAST_NORM) { 210779714f72SAl Viro path_put(&nd.path); 210851689104SPaul Moore d = ERR_PTR(-EINVAL); 210951689104SPaul Moore goto out; 211079714f72SAl Viro } 211179714f72SAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 21121e0ea001SAl Viro d = __lookup_hash(&nd.last, nd.path.dentry, 0); 211379714f72SAl Viro if (IS_ERR(d)) { 211479714f72SAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 211579714f72SAl Viro path_put(&nd.path); 211651689104SPaul Moore goto out; 211779714f72SAl Viro } 211879714f72SAl Viro *path = nd.path; 211951689104SPaul Moore out: 212051689104SPaul Moore putname(filename); 212179714f72SAl Viro return d; 21225590ff0dSUlrich Drepper } 21235590ff0dSUlrich Drepper 2124d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2125d1811465SAl Viro { 2126d1811465SAl Viro struct nameidata nd; 212774eb8cc5SAl Viro struct filename *filename = getname_kernel(name); 212874eb8cc5SAl Viro int res = PTR_ERR(filename); 212974eb8cc5SAl Viro 213074eb8cc5SAl Viro if (!IS_ERR(filename)) { 213174eb8cc5SAl Viro res = filename_lookup(AT_FDCWD, filename, flags, &nd); 213274eb8cc5SAl Viro putname(filename); 2133d1811465SAl Viro if (!res) 2134d1811465SAl Viro *path = nd.path; 213574eb8cc5SAl Viro } 2136d1811465SAl Viro return res; 2137d1811465SAl Viro } 21384d359507SAl Viro EXPORT_SYMBOL(kern_path); 2139d1811465SAl Viro 214016f18200SJosef 'Jeff' Sipek /** 214116f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 214216f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 214316f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 214416f18200SJosef 'Jeff' Sipek * @name: pointer to file name 214516f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2146e0a01249SAl Viro * @path: pointer to struct path to fill 214716f18200SJosef 'Jeff' Sipek */ 214816f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 214916f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2150e0a01249SAl Viro struct path *path) 215116f18200SJosef 'Jeff' Sipek { 215274eb8cc5SAl Viro struct filename *filename = getname_kernel(name); 215374eb8cc5SAl Viro int err = PTR_ERR(filename); 215474eb8cc5SAl Viro 215574eb8cc5SAl Viro BUG_ON(flags & LOOKUP_PARENT); 215674eb8cc5SAl Viro 215774eb8cc5SAl Viro /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */ 215874eb8cc5SAl Viro if (!IS_ERR(filename)) { 2159e0a01249SAl Viro struct nameidata nd; 2160e0a01249SAl Viro nd.root.dentry = dentry; 2161e0a01249SAl Viro nd.root.mnt = mnt; 216274eb8cc5SAl Viro err = filename_lookup(AT_FDCWD, filename, 216374eb8cc5SAl Viro flags | LOOKUP_ROOT, &nd); 2164e0a01249SAl Viro if (!err) 2165e0a01249SAl Viro *path = nd.path; 216674eb8cc5SAl Viro putname(filename); 216774eb8cc5SAl Viro } 2168e0a01249SAl Viro return err; 216916f18200SJosef 'Jeff' Sipek } 21704d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 217116f18200SJosef 'Jeff' Sipek 2172eead1911SChristoph Hellwig /** 2173a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2174eead1911SChristoph Hellwig * @name: pathname component to lookup 2175eead1911SChristoph Hellwig * @base: base directory to lookup from 2176eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2177eead1911SChristoph Hellwig * 2178a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 21799e7543e9SAl Viro * not be called by generic code. 2180eead1911SChristoph Hellwig */ 2181057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2182057f6c01SJames Morris { 2183057f6c01SJames Morris struct qstr this; 21846a96ba54SAl Viro unsigned int c; 2185cda309deSMiklos Szeredi int err; 2186057f6c01SJames Morris 21872f9092e1SDavid Woodhouse WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex)); 21882f9092e1SDavid Woodhouse 21896a96ba54SAl Viro this.name = name; 21906a96ba54SAl Viro this.len = len; 21910145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 21926a96ba54SAl Viro if (!len) 21936a96ba54SAl Viro return ERR_PTR(-EACCES); 21946a96ba54SAl Viro 219521d8a15aSAl Viro if (unlikely(name[0] == '.')) { 219621d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 219721d8a15aSAl Viro return ERR_PTR(-EACCES); 219821d8a15aSAl Viro } 219921d8a15aSAl Viro 22006a96ba54SAl Viro while (len--) { 22016a96ba54SAl Viro c = *(const unsigned char *)name++; 22026a96ba54SAl Viro if (c == '/' || c == '\0') 22036a96ba54SAl Viro return ERR_PTR(-EACCES); 22046a96ba54SAl Viro } 22055a202bcdSAl Viro /* 22065a202bcdSAl Viro * See if the low-level filesystem might want 22075a202bcdSAl Viro * to use its own hash.. 22085a202bcdSAl Viro */ 22095a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2210da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 22115a202bcdSAl Viro if (err < 0) 22125a202bcdSAl Viro return ERR_PTR(err); 22135a202bcdSAl Viro } 2214eead1911SChristoph Hellwig 2215cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2216cda309deSMiklos Szeredi if (err) 2217cda309deSMiklos Szeredi return ERR_PTR(err); 2218cda309deSMiklos Szeredi 221972bd866aSAl Viro return __lookup_hash(&this, base, 0); 2220057f6c01SJames Morris } 22214d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2222057f6c01SJames Morris 22231fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 22241fa1e7f6SAndy Whitcroft struct path *path, int *empty) 22251da177e4SLinus Torvalds { 22262d8f3038SAl Viro struct nameidata nd; 222791a27b2aSJeff Layton struct filename *tmp = getname_flags(name, flags, empty); 22281da177e4SLinus Torvalds int err = PTR_ERR(tmp); 22291da177e4SLinus Torvalds if (!IS_ERR(tmp)) { 22302d8f3038SAl Viro 22312d8f3038SAl Viro BUG_ON(flags & LOOKUP_PARENT); 22322d8f3038SAl Viro 2233873f1eedSJeff Layton err = filename_lookup(dfd, tmp, flags, &nd); 22341da177e4SLinus Torvalds putname(tmp); 22352d8f3038SAl Viro if (!err) 22362d8f3038SAl Viro *path = nd.path; 22371da177e4SLinus Torvalds } 22381da177e4SLinus Torvalds return err; 22391da177e4SLinus Torvalds } 22401da177e4SLinus Torvalds 22411fa1e7f6SAndy Whitcroft int user_path_at(int dfd, const char __user *name, unsigned flags, 22421fa1e7f6SAndy Whitcroft struct path *path) 22431fa1e7f6SAndy Whitcroft { 2244f7493e5dSLinus Torvalds return user_path_at_empty(dfd, name, flags, path, NULL); 22451fa1e7f6SAndy Whitcroft } 22464d359507SAl Viro EXPORT_SYMBOL(user_path_at); 22471fa1e7f6SAndy Whitcroft 2248873f1eedSJeff Layton /* 2249873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2250873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2251873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2252873f1eedSJeff Layton * path-walking is complete. 2253873f1eedSJeff Layton */ 225491a27b2aSJeff Layton static struct filename * 2255f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2256f5beed75SAl Viro struct path *parent, 2257f5beed75SAl Viro struct qstr *last, 2258f5beed75SAl Viro int *type, 22599e790bd6SJeff Layton unsigned int flags) 22602ad94ae6SAl Viro { 2261f5beed75SAl Viro struct nameidata nd; 226291a27b2aSJeff Layton struct filename *s = getname(path); 22632ad94ae6SAl Viro int error; 22642ad94ae6SAl Viro 22659e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 22669e790bd6SJeff Layton flags &= LOOKUP_REVAL; 22679e790bd6SJeff Layton 22682ad94ae6SAl Viro if (IS_ERR(s)) 226991a27b2aSJeff Layton return s; 22702ad94ae6SAl Viro 2271f5beed75SAl Viro error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd); 227291a27b2aSJeff Layton if (error) { 22732ad94ae6SAl Viro putname(s); 227491a27b2aSJeff Layton return ERR_PTR(error); 227591a27b2aSJeff Layton } 2276f5beed75SAl Viro *parent = nd.path; 2277f5beed75SAl Viro *last = nd.last; 2278f5beed75SAl Viro *type = nd.last_type; 22792ad94ae6SAl Viro 228091a27b2aSJeff Layton return s; 22812ad94ae6SAl Viro } 22822ad94ae6SAl Viro 22838033426eSJeff Layton /** 2284197df04cSAl Viro * mountpoint_last - look up last component for umount 22858033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 22868033426eSJeff Layton * @path: pointer to container for result 22878033426eSJeff Layton * 22888033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 22898033426eSJeff Layton * need to resolve the path without doing any revalidation. 22908033426eSJeff Layton * 22918033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 22928033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 22938033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 22948033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 22958033426eSJeff Layton * bogus and it doesn't exist. 22968033426eSJeff Layton * 22978033426eSJeff Layton * Returns: 22988033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 22998033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 23008033426eSJeff Layton * put in this case. 23018033426eSJeff Layton * 23028033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 23038033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 23048033426eSJeff Layton * The nd->path reference will also be put. 23058033426eSJeff Layton * 23068033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 23078033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 23088033426eSJeff Layton * to the link, and nd->path will *not* be put. 23098033426eSJeff Layton */ 23108033426eSJeff Layton static int 2311197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 23128033426eSJeff Layton { 23138033426eSJeff Layton int error = 0; 23148033426eSJeff Layton struct dentry *dentry; 23158033426eSJeff Layton struct dentry *dir = nd->path.dentry; 23168033426eSJeff Layton 231735759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 231835759521SAl Viro if (nd->flags & LOOKUP_RCU) { 231935759521SAl Viro if (unlazy_walk(nd, NULL)) { 23208033426eSJeff Layton error = -ECHILD; 232135759521SAl Viro goto out; 232235759521SAl Viro } 23238033426eSJeff Layton } 23248033426eSJeff Layton 23258033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 23268033426eSJeff Layton 23278033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 23288033426eSJeff Layton error = handle_dots(nd, nd->last_type); 232935759521SAl Viro if (error) 233035759521SAl Viro goto out; 23318033426eSJeff Layton dentry = dget(nd->path.dentry); 233235759521SAl Viro goto done; 23338033426eSJeff Layton } 23348033426eSJeff Layton 23358033426eSJeff Layton mutex_lock(&dir->d_inode->i_mutex); 23368033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 23378033426eSJeff Layton if (!dentry) { 23388033426eSJeff Layton /* 23398033426eSJeff Layton * No cached dentry. Mounted dentries are pinned in the cache, 23408033426eSJeff Layton * so that means that this dentry is probably a symlink or the 23418033426eSJeff Layton * path doesn't actually point to a mounted dentry. 23428033426eSJeff Layton */ 23438033426eSJeff Layton dentry = d_alloc(dir, &nd->last); 23448033426eSJeff Layton if (!dentry) { 23458033426eSJeff Layton error = -ENOMEM; 2346bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 234735759521SAl Viro goto out; 23488033426eSJeff Layton } 234935759521SAl Viro dentry = lookup_real(dir->d_inode, dentry, nd->flags); 235035759521SAl Viro error = PTR_ERR(dentry); 2351bcceeebaSDave Jones if (IS_ERR(dentry)) { 2352bcceeebaSDave Jones mutex_unlock(&dir->d_inode->i_mutex); 235335759521SAl Viro goto out; 23548033426eSJeff Layton } 2355bcceeebaSDave Jones } 23568033426eSJeff Layton mutex_unlock(&dir->d_inode->i_mutex); 23578033426eSJeff Layton 235835759521SAl Viro done: 2359698934dfSDavid Howells if (d_is_negative(dentry)) { 23608033426eSJeff Layton error = -ENOENT; 23618033426eSJeff Layton dput(dentry); 236235759521SAl Viro goto out; 236335759521SAl Viro } 2364191d7f73SAl Viro if (nd->depth) 2365191d7f73SAl Viro put_link(nd); 23668033426eSJeff Layton path->dentry = dentry; 2367295dc39dSVasily Averin path->mnt = nd->path.mnt; 2368caa85634SAl Viro if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) { 2369caa85634SAl Viro nd->link = *path; 23708033426eSJeff Layton return 1; 2371caa85634SAl Viro } 2372295dc39dSVasily Averin mntget(path->mnt); 23738033426eSJeff Layton follow_mount(path); 237435759521SAl Viro error = 0; 237535759521SAl Viro out: 23768033426eSJeff Layton terminate_walk(nd); 23778033426eSJeff Layton return error; 23788033426eSJeff Layton } 23798033426eSJeff Layton 23808033426eSJeff Layton /** 2381197df04cSAl Viro * path_mountpoint - look up a path to be umounted 23828033426eSJeff Layton * @dfd: directory file descriptor to start walk from 23838033426eSJeff Layton * @name: full pathname to walk 2384606d6fe3SRandy Dunlap * @path: pointer to container for result 23858033426eSJeff Layton * @flags: lookup flags 23868033426eSJeff Layton * 23878033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2388606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 23898033426eSJeff Layton */ 23908033426eSJeff Layton static int 2391668696dcSAl Viro path_mountpoint(int dfd, const struct filename *name, struct path *path, 239246afd6f6SAl Viro struct nameidata *nd, unsigned int flags) 23938033426eSJeff Layton { 239446afd6f6SAl Viro int err = path_init(dfd, name, flags, nd); 23958033426eSJeff Layton if (unlikely(err)) 2396115cbfdcSAl Viro goto out; 23978033426eSJeff Layton 2398191d7f73SAl Viro while ((err = mountpoint_last(nd, path)) > 0) { 23991d8e03d3SAl Viro err = trailing_symlink(nd); 24008033426eSJeff Layton if (err) 24018033426eSJeff Layton break; 24028033426eSJeff Layton } 24038033426eSJeff Layton out: 240446afd6f6SAl Viro path_cleanup(nd); 24058033426eSJeff Layton return err; 24068033426eSJeff Layton } 24078033426eSJeff Layton 24082d864651SAl Viro static int 2409668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 24102d864651SAl Viro unsigned int flags) 24112d864651SAl Viro { 241246afd6f6SAl Viro struct nameidata nd; 2413cbaab2dbSAl Viro int error; 2414668696dcSAl Viro if (IS_ERR(name)) 2415668696dcSAl Viro return PTR_ERR(name); 2416894bc8c4SAl Viro set_nameidata(&nd); 241746afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU); 24182d864651SAl Viro if (unlikely(error == -ECHILD)) 241946afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags); 24202d864651SAl Viro if (unlikely(error == -ESTALE)) 242146afd6f6SAl Viro error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL); 24222d864651SAl Viro if (likely(!error)) 2423668696dcSAl Viro audit_inode(name, path->dentry, 0); 2424894bc8c4SAl Viro restore_nameidata(&nd); 2425668696dcSAl Viro putname(name); 24262d864651SAl Viro return error; 24272d864651SAl Viro } 24282d864651SAl Viro 24298033426eSJeff Layton /** 2430197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 24318033426eSJeff Layton * @dfd: directory file descriptor 24328033426eSJeff Layton * @name: pathname from userland 24338033426eSJeff Layton * @flags: lookup flags 24348033426eSJeff Layton * @path: pointer to container to hold result 24358033426eSJeff Layton * 24368033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 24378033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 24388033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 24398033426eSJeff Layton * and avoid revalidating the last component. 24408033426eSJeff Layton * 24418033426eSJeff Layton * Returns 0 and populates "path" on success. 24428033426eSJeff Layton */ 24438033426eSJeff Layton int 2444197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 24458033426eSJeff Layton struct path *path) 24468033426eSJeff Layton { 2447cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 24488033426eSJeff Layton } 24498033426eSJeff Layton 24502d864651SAl Viro int 24512d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 24522d864651SAl Viro unsigned int flags) 24532d864651SAl Viro { 2454cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 24552d864651SAl Viro } 24562d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 24572d864651SAl Viro 2458cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 24591da177e4SLinus Torvalds { 24608e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2461da9592edSDavid Howells 24628e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 24631da177e4SLinus Torvalds return 0; 24648e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 24651da177e4SLinus Torvalds return 0; 246623adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 24671da177e4SLinus Torvalds } 2468cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 24691da177e4SLinus Torvalds 24701da177e4SLinus Torvalds /* 24711da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 24721da177e4SLinus Torvalds * whether the type of victim is right. 24731da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 24741da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 24751da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 24761da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 24771da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 24781da177e4SLinus Torvalds * a. be owner of dir, or 24791da177e4SLinus Torvalds * b. be owner of victim, or 24801da177e4SLinus Torvalds * c. have CAP_FOWNER capability 24811da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 24821da177e4SLinus Torvalds * links pointing to it. 24831da177e4SLinus Torvalds * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR. 24841da177e4SLinus Torvalds * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR. 24851da177e4SLinus Torvalds * 9. We can't remove a root or mountpoint. 24861da177e4SLinus Torvalds * 10. We don't allow removal of NFS sillyrenamed files; it's handled by 24871da177e4SLinus Torvalds * nfs_async_unlink(). 24881da177e4SLinus Torvalds */ 2489b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 24901da177e4SLinus Torvalds { 2491b18825a7SDavid Howells struct inode *inode = victim->d_inode; 24921da177e4SLinus Torvalds int error; 24931da177e4SLinus Torvalds 2494b18825a7SDavid Howells if (d_is_negative(victim)) 24951da177e4SLinus Torvalds return -ENOENT; 2496b18825a7SDavid Howells BUG_ON(!inode); 24971da177e4SLinus Torvalds 24981da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 24994fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 25001da177e4SLinus Torvalds 2501f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 25021da177e4SLinus Torvalds if (error) 25031da177e4SLinus Torvalds return error; 25041da177e4SLinus Torvalds if (IS_APPEND(dir)) 25051da177e4SLinus Torvalds return -EPERM; 2506b18825a7SDavid Howells 2507b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 2508b18825a7SDavid Howells IS_IMMUTABLE(inode) || IS_SWAPFILE(inode)) 25091da177e4SLinus Torvalds return -EPERM; 25101da177e4SLinus Torvalds if (isdir) { 251144b1d530SMiklos Szeredi if (!d_is_dir(victim)) 25121da177e4SLinus Torvalds return -ENOTDIR; 25131da177e4SLinus Torvalds if (IS_ROOT(victim)) 25141da177e4SLinus Torvalds return -EBUSY; 251544b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 25161da177e4SLinus Torvalds return -EISDIR; 25171da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25181da177e4SLinus Torvalds return -ENOENT; 25191da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 25201da177e4SLinus Torvalds return -EBUSY; 25211da177e4SLinus Torvalds return 0; 25221da177e4SLinus Torvalds } 25231da177e4SLinus Torvalds 25241da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 25251da177e4SLinus Torvalds * dir. 25261da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 25271da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 25281da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 25291da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 25301da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 25311da177e4SLinus Torvalds */ 2532a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 25331da177e4SLinus Torvalds { 253414e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 25351da177e4SLinus Torvalds if (child->d_inode) 25361da177e4SLinus Torvalds return -EEXIST; 25371da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 25381da177e4SLinus Torvalds return -ENOENT; 2539f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 25401da177e4SLinus Torvalds } 25411da177e4SLinus Torvalds 25421da177e4SLinus Torvalds /* 25431da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 25441da177e4SLinus Torvalds */ 25451da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 25461da177e4SLinus Torvalds { 25471da177e4SLinus Torvalds struct dentry *p; 25481da177e4SLinus Torvalds 25491da177e4SLinus Torvalds if (p1 == p2) { 2550f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 25511da177e4SLinus Torvalds return NULL; 25521da177e4SLinus Torvalds } 25531da177e4SLinus Torvalds 2554a11f3a05SArjan van de Ven mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25551da177e4SLinus Torvalds 2556e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2557e2761a11SOGAWA Hirofumi if (p) { 2558f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT); 2559f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD); 25601da177e4SLinus Torvalds return p; 25611da177e4SLinus Torvalds } 25621da177e4SLinus Torvalds 2563e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2564e2761a11SOGAWA Hirofumi if (p) { 2565f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2566f2eace23SIngo Molnar mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD); 25671da177e4SLinus Torvalds return p; 25681da177e4SLinus Torvalds } 25691da177e4SLinus Torvalds 2570f2eace23SIngo Molnar mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT); 2571d1b72cc6SMiklos Szeredi mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2); 25721da177e4SLinus Torvalds return NULL; 25731da177e4SLinus Torvalds } 25744d359507SAl Viro EXPORT_SYMBOL(lock_rename); 25751da177e4SLinus Torvalds 25761da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 25771da177e4SLinus Torvalds { 25781b1dcc1bSJes Sorensen mutex_unlock(&p1->d_inode->i_mutex); 25791da177e4SLinus Torvalds if (p1 != p2) { 25801b1dcc1bSJes Sorensen mutex_unlock(&p2->d_inode->i_mutex); 2581a11f3a05SArjan van de Ven mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex); 25821da177e4SLinus Torvalds } 25831da177e4SLinus Torvalds } 25844d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 25851da177e4SLinus Torvalds 25864acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2587312b63fbSAl Viro bool want_excl) 25881da177e4SLinus Torvalds { 2589a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 25901da177e4SLinus Torvalds if (error) 25911da177e4SLinus Torvalds return error; 25921da177e4SLinus Torvalds 2593acfa4380SAl Viro if (!dir->i_op->create) 25941da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 25951da177e4SLinus Torvalds mode &= S_IALLUGO; 25961da177e4SLinus Torvalds mode |= S_IFREG; 25971da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 25981da177e4SLinus Torvalds if (error) 25991da177e4SLinus Torvalds return error; 2600312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2601a74574aaSStephen Smalley if (!error) 2602f38aa942SAmy Griffis fsnotify_create(dir, dentry); 26031da177e4SLinus Torvalds return error; 26041da177e4SLinus Torvalds } 26054d359507SAl Viro EXPORT_SYMBOL(vfs_create); 26061da177e4SLinus Torvalds 260773d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 26081da177e4SLinus Torvalds { 26093fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 26101da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 26111da177e4SLinus Torvalds int error; 26121da177e4SLinus Torvalds 2613bcda7652SAl Viro /* O_PATH? */ 2614bcda7652SAl Viro if (!acc_mode) 2615bcda7652SAl Viro return 0; 2616bcda7652SAl Viro 26171da177e4SLinus Torvalds if (!inode) 26181da177e4SLinus Torvalds return -ENOENT; 26191da177e4SLinus Torvalds 2620c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2621c8fe8f30SChristoph Hellwig case S_IFLNK: 26221da177e4SLinus Torvalds return -ELOOP; 2623c8fe8f30SChristoph Hellwig case S_IFDIR: 2624c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 26251da177e4SLinus Torvalds return -EISDIR; 2626c8fe8f30SChristoph Hellwig break; 2627c8fe8f30SChristoph Hellwig case S_IFBLK: 2628c8fe8f30SChristoph Hellwig case S_IFCHR: 26293fb64190SChristoph Hellwig if (path->mnt->mnt_flags & MNT_NODEV) 26301da177e4SLinus Torvalds return -EACCES; 2631c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2632c8fe8f30SChristoph Hellwig case S_IFIFO: 2633c8fe8f30SChristoph Hellwig case S_IFSOCK: 26341da177e4SLinus Torvalds flag &= ~O_TRUNC; 2635c8fe8f30SChristoph Hellwig break; 26364a3fd211SDave Hansen } 2637b41572e9SDave Hansen 26383fb64190SChristoph Hellwig error = inode_permission(inode, acc_mode); 2639b41572e9SDave Hansen if (error) 2640b41572e9SDave Hansen return error; 26416146f0d5SMimi Zohar 26421da177e4SLinus Torvalds /* 26431da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 26441da177e4SLinus Torvalds */ 26451da177e4SLinus Torvalds if (IS_APPEND(inode)) { 26468737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 26477715b521SAl Viro return -EPERM; 26481da177e4SLinus Torvalds if (flag & O_TRUNC) 26497715b521SAl Viro return -EPERM; 26501da177e4SLinus Torvalds } 26511da177e4SLinus Torvalds 26521da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 26532e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 26547715b521SAl Viro return -EPERM; 26551da177e4SLinus Torvalds 2656f3c7691eSJ. Bruce Fields return 0; 26577715b521SAl Viro } 26587715b521SAl Viro 2659e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 26607715b521SAl Viro { 2661e1181ee6SJeff Layton struct path *path = &filp->f_path; 26627715b521SAl Viro struct inode *inode = path->dentry->d_inode; 26637715b521SAl Viro int error = get_write_access(inode); 26641da177e4SLinus Torvalds if (error) 26657715b521SAl Viro return error; 26661da177e4SLinus Torvalds /* 26671da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 26681da177e4SLinus Torvalds */ 2669d7a06983SJeff Layton error = locks_verify_locked(filp); 2670be6d3e56SKentaro Takeda if (!error) 2671ea0d3ab2STetsuo Handa error = security_path_truncate(path); 26721da177e4SLinus Torvalds if (!error) { 26737715b521SAl Viro error = do_truncate(path->dentry, 0, 2674d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2675e1181ee6SJeff Layton filp); 26761da177e4SLinus Torvalds } 26771da177e4SLinus Torvalds put_write_access(inode); 2678acd0c935SMimi Zohar return error; 26791da177e4SLinus Torvalds } 26801da177e4SLinus Torvalds 2681d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2682d57999e1SDave Hansen { 26838a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 26848a5e929dSAl Viro flag--; 2685d57999e1SDave Hansen return flag; 2686d57999e1SDave Hansen } 2687d57999e1SDave Hansen 2688d18e9008SMiklos Szeredi static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode) 2689d18e9008SMiklos Szeredi { 2690d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2691d18e9008SMiklos Szeredi if (error) 2692d18e9008SMiklos Szeredi return error; 2693d18e9008SMiklos Szeredi 2694d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2695d18e9008SMiklos Szeredi if (error) 2696d18e9008SMiklos Szeredi return error; 2697d18e9008SMiklos Szeredi 2698d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2699d18e9008SMiklos Szeredi } 2700d18e9008SMiklos Szeredi 27011acf0af9SDavid Howells /* 27021acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 27031acf0af9SDavid Howells * dentry. 27041acf0af9SDavid Howells * 27051acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 27061acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 27071acf0af9SDavid Howells * 27081acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 27091acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 27101acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 27111acf0af9SDavid Howells * 27121acf0af9SDavid Howells * Returns an error code otherwise. 27131acf0af9SDavid Howells */ 27142675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 271530d90494SAl Viro struct path *path, struct file *file, 2716015c3bbcSMiklos Szeredi const struct open_flags *op, 271764894cf8SAl Viro bool got_write, bool need_lookup, 271847237687SAl Viro int *opened) 2719d18e9008SMiklos Szeredi { 2720d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2721d18e9008SMiklos Szeredi unsigned open_flag = open_to_namei_flags(op->open_flag); 2722d18e9008SMiklos Szeredi umode_t mode; 2723d18e9008SMiklos Szeredi int error; 2724d18e9008SMiklos Szeredi int acc_mode; 2725d18e9008SMiklos Szeredi int create_error = 0; 2726d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2727116cc022SMiklos Szeredi bool excl; 2728d18e9008SMiklos Szeredi 2729d18e9008SMiklos Szeredi BUG_ON(dentry->d_inode); 2730d18e9008SMiklos Szeredi 2731d18e9008SMiklos Szeredi /* Don't create child dentry for a dead directory. */ 2732d18e9008SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 27332675a4ebSAl Viro error = -ENOENT; 2734d18e9008SMiklos Szeredi goto out; 2735d18e9008SMiklos Szeredi } 2736d18e9008SMiklos Szeredi 273762b259d8SMiklos Szeredi mode = op->mode; 2738d18e9008SMiklos Szeredi if ((open_flag & O_CREAT) && !IS_POSIXACL(dir)) 2739d18e9008SMiklos Szeredi mode &= ~current_umask(); 2740d18e9008SMiklos Szeredi 2741116cc022SMiklos Szeredi excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT); 2742116cc022SMiklos Szeredi if (excl) 2743d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2744d18e9008SMiklos Szeredi 2745d18e9008SMiklos Szeredi /* 2746d18e9008SMiklos Szeredi * Checking write permission is tricky, bacuse we don't know if we are 2747d18e9008SMiklos Szeredi * going to actually need it: O_CREAT opens should work as long as the 2748d18e9008SMiklos Szeredi * file exists. But checking existence breaks atomicity. The trick is 2749d18e9008SMiklos Szeredi * to check access and if not granted clear O_CREAT from the flags. 2750d18e9008SMiklos Szeredi * 2751d18e9008SMiklos Szeredi * Another problem is returing the "right" error value (e.g. for an 2752d18e9008SMiklos Szeredi * O_EXCL open we want to return EEXIST not EROFS). 2753d18e9008SMiklos Szeredi */ 275464894cf8SAl Viro if (((open_flag & (O_CREAT | O_TRUNC)) || 275564894cf8SAl Viro (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) { 275664894cf8SAl Viro if (!(open_flag & O_CREAT)) { 2757d18e9008SMiklos Szeredi /* 2758d18e9008SMiklos Szeredi * No O_CREATE -> atomicity not a requirement -> fall 2759d18e9008SMiklos Szeredi * back to lookup + open 2760d18e9008SMiklos Szeredi */ 2761d18e9008SMiklos Szeredi goto no_open; 2762d18e9008SMiklos Szeredi } else if (open_flag & (O_EXCL | O_TRUNC)) { 2763d18e9008SMiklos Szeredi /* Fall back and fail with the right error */ 276464894cf8SAl Viro create_error = -EROFS; 2765d18e9008SMiklos Szeredi goto no_open; 2766d18e9008SMiklos Szeredi } else { 2767d18e9008SMiklos Szeredi /* No side effects, safe to clear O_CREAT */ 276864894cf8SAl Viro create_error = -EROFS; 2769d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2770d18e9008SMiklos Szeredi } 2771d18e9008SMiklos Szeredi } 2772d18e9008SMiklos Szeredi 2773d18e9008SMiklos Szeredi if (open_flag & O_CREAT) { 277438227f78SMiklos Szeredi error = may_o_create(&nd->path, dentry, mode); 2775d18e9008SMiklos Szeredi if (error) { 2776d18e9008SMiklos Szeredi create_error = error; 2777d18e9008SMiklos Szeredi if (open_flag & O_EXCL) 2778d18e9008SMiklos Szeredi goto no_open; 2779d18e9008SMiklos Szeredi open_flag &= ~O_CREAT; 2780d18e9008SMiklos Szeredi } 2781d18e9008SMiklos Szeredi } 2782d18e9008SMiklos Szeredi 2783d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2784d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2785d18e9008SMiklos Szeredi 278630d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 278730d90494SAl Viro file->f_path.mnt = nd->path.mnt; 278830d90494SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode, 278947237687SAl Viro opened); 2790d9585277SAl Viro if (error < 0) { 2791d9585277SAl Viro if (create_error && error == -ENOENT) 2792d9585277SAl Viro error = create_error; 2793d18e9008SMiklos Szeredi goto out; 2794d18e9008SMiklos Szeredi } 2795d18e9008SMiklos Szeredi 2796d9585277SAl Viro if (error) { /* returned 1, that is */ 279730d90494SAl Viro if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 27982675a4ebSAl Viro error = -EIO; 2799d18e9008SMiklos Szeredi goto out; 2800d18e9008SMiklos Szeredi } 280130d90494SAl Viro if (file->f_path.dentry) { 2802d18e9008SMiklos Szeredi dput(dentry); 280330d90494SAl Viro dentry = file->f_path.dentry; 2804d18e9008SMiklos Szeredi } 280503da633aSAl Viro if (*opened & FILE_CREATED) 280603da633aSAl Viro fsnotify_create(dir, dentry); 280703da633aSAl Viro if (!dentry->d_inode) { 280803da633aSAl Viro WARN_ON(*opened & FILE_CREATED); 280903da633aSAl Viro if (create_error) { 281062b2ce96SSage Weil error = create_error; 281162b2ce96SSage Weil goto out; 281262b2ce96SSage Weil } 281303da633aSAl Viro } else { 281403da633aSAl Viro if (excl && !(*opened & FILE_CREATED)) { 281503da633aSAl Viro error = -EEXIST; 281603da633aSAl Viro goto out; 281703da633aSAl Viro } 281803da633aSAl Viro } 2819d18e9008SMiklos Szeredi goto looked_up; 2820d18e9008SMiklos Szeredi } 2821d18e9008SMiklos Szeredi 2822d18e9008SMiklos Szeredi /* 2823d18e9008SMiklos Szeredi * We didn't have the inode before the open, so check open permission 2824d18e9008SMiklos Szeredi * here. 2825d18e9008SMiklos Szeredi */ 282603da633aSAl Viro acc_mode = op->acc_mode; 282703da633aSAl Viro if (*opened & FILE_CREATED) { 282803da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 282903da633aSAl Viro fsnotify_create(dir, dentry); 283003da633aSAl Viro acc_mode = MAY_OPEN; 283103da633aSAl Viro } 28322675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 28332675a4ebSAl Viro if (error) 28342675a4ebSAl Viro fput(file); 2835d18e9008SMiklos Szeredi 2836d18e9008SMiklos Szeredi out: 2837d18e9008SMiklos Szeredi dput(dentry); 28382675a4ebSAl Viro return error; 2839d18e9008SMiklos Szeredi 2840d18e9008SMiklos Szeredi no_open: 2841d18e9008SMiklos Szeredi if (need_lookup) { 284272bd866aSAl Viro dentry = lookup_real(dir, dentry, nd->flags); 2843d18e9008SMiklos Szeredi if (IS_ERR(dentry)) 28442675a4ebSAl Viro return PTR_ERR(dentry); 2845d18e9008SMiklos Szeredi 2846d18e9008SMiklos Szeredi if (create_error) { 2847d18e9008SMiklos Szeredi int open_flag = op->open_flag; 2848d18e9008SMiklos Szeredi 28492675a4ebSAl Viro error = create_error; 2850d18e9008SMiklos Szeredi if ((open_flag & O_EXCL)) { 2851d18e9008SMiklos Szeredi if (!dentry->d_inode) 2852d18e9008SMiklos Szeredi goto out; 2853d18e9008SMiklos Szeredi } else if (!dentry->d_inode) { 2854d18e9008SMiklos Szeredi goto out; 2855d18e9008SMiklos Szeredi } else if ((open_flag & O_TRUNC) && 2856e36cb0b8SDavid Howells d_is_reg(dentry)) { 2857d18e9008SMiklos Szeredi goto out; 2858d18e9008SMiklos Szeredi } 2859d18e9008SMiklos Szeredi /* will fail later, go on to get the right error */ 2860d18e9008SMiklos Szeredi } 2861d18e9008SMiklos Szeredi } 2862d18e9008SMiklos Szeredi looked_up: 2863d18e9008SMiklos Szeredi path->dentry = dentry; 2864d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 28652675a4ebSAl Viro return 1; 2866d18e9008SMiklos Szeredi } 2867d18e9008SMiklos Szeredi 286831e6b01fSNick Piggin /* 28691acf0af9SDavid Howells * Look up and maybe create and open the last component. 2870d58ffd35SMiklos Szeredi * 2871d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 2872d58ffd35SMiklos Szeredi * 28731acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 28741acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 28751acf0af9SDavid Howells * 28761acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 28771acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 28781acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 28791acf0af9SDavid Howells * specified then a negative dentry may be returned. 28801acf0af9SDavid Howells * 28811acf0af9SDavid Howells * An error code is returned otherwise. 28821acf0af9SDavid Howells * 28831acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 28841acf0af9SDavid Howells * cleared otherwise prior to returning. 2885d58ffd35SMiklos Szeredi */ 28862675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 288730d90494SAl Viro struct file *file, 2888d58ffd35SMiklos Szeredi const struct open_flags *op, 288964894cf8SAl Viro bool got_write, int *opened) 2890d58ffd35SMiklos Szeredi { 2891d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 289254ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 2893d58ffd35SMiklos Szeredi struct dentry *dentry; 2894d58ffd35SMiklos Szeredi int error; 289554ef4872SMiklos Szeredi bool need_lookup; 2896d58ffd35SMiklos Szeredi 289747237687SAl Viro *opened &= ~FILE_CREATED; 2898201f956eSAl Viro dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup); 2899d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 29002675a4ebSAl Viro return PTR_ERR(dentry); 2901d58ffd35SMiklos Szeredi 2902d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 2903d18e9008SMiklos Szeredi if (!need_lookup && dentry->d_inode) 2904d18e9008SMiklos Szeredi goto out_no_open; 2905d18e9008SMiklos Szeredi 2906d18e9008SMiklos Szeredi if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) { 290764894cf8SAl Viro return atomic_open(nd, dentry, path, file, op, got_write, 290847237687SAl Viro need_lookup, opened); 2909d18e9008SMiklos Szeredi } 2910d18e9008SMiklos Szeredi 291154ef4872SMiklos Szeredi if (need_lookup) { 291254ef4872SMiklos Szeredi BUG_ON(dentry->d_inode); 291354ef4872SMiklos Szeredi 291472bd866aSAl Viro dentry = lookup_real(dir_inode, dentry, nd->flags); 291554ef4872SMiklos Szeredi if (IS_ERR(dentry)) 29162675a4ebSAl Viro return PTR_ERR(dentry); 291754ef4872SMiklos Szeredi } 291854ef4872SMiklos Szeredi 2919d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 2920d58ffd35SMiklos Szeredi if (!dentry->d_inode && (op->open_flag & O_CREAT)) { 2921d58ffd35SMiklos Szeredi umode_t mode = op->mode; 2922d58ffd35SMiklos Szeredi if (!IS_POSIXACL(dir->d_inode)) 2923d58ffd35SMiklos Szeredi mode &= ~current_umask(); 2924d58ffd35SMiklos Szeredi /* 2925d58ffd35SMiklos Szeredi * This write is needed to ensure that a 2926d58ffd35SMiklos Szeredi * rw->ro transition does not occur between 2927d58ffd35SMiklos Szeredi * the time when the file is created and when 2928d58ffd35SMiklos Szeredi * a permanent write count is taken through 2929015c3bbcSMiklos Szeredi * the 'struct file' in finish_open(). 2930d58ffd35SMiklos Szeredi */ 293164894cf8SAl Viro if (!got_write) { 293264894cf8SAl Viro error = -EROFS; 2933d58ffd35SMiklos Szeredi goto out_dput; 293464894cf8SAl Viro } 293547237687SAl Viro *opened |= FILE_CREATED; 2936d58ffd35SMiklos Szeredi error = security_path_mknod(&nd->path, dentry, mode, 0); 2937d58ffd35SMiklos Szeredi if (error) 2938d58ffd35SMiklos Szeredi goto out_dput; 2939312b63fbSAl Viro error = vfs_create(dir->d_inode, dentry, mode, 2940312b63fbSAl Viro nd->flags & LOOKUP_EXCL); 2941d58ffd35SMiklos Szeredi if (error) 2942d58ffd35SMiklos Szeredi goto out_dput; 2943d58ffd35SMiklos Szeredi } 2944d18e9008SMiklos Szeredi out_no_open: 2945d58ffd35SMiklos Szeredi path->dentry = dentry; 2946d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 29472675a4ebSAl Viro return 1; 2948d58ffd35SMiklos Szeredi 2949d58ffd35SMiklos Szeredi out_dput: 2950d58ffd35SMiklos Szeredi dput(dentry); 29512675a4ebSAl Viro return error; 2952d58ffd35SMiklos Szeredi } 2953d58ffd35SMiklos Szeredi 2954d58ffd35SMiklos Szeredi /* 2955fe2d35ffSAl Viro * Handle the last step of open() 295631e6b01fSNick Piggin */ 2957896475d5SAl Viro static int do_last(struct nameidata *nd, 295830d90494SAl Viro struct file *file, const struct open_flags *op, 2959669abf4eSJeff Layton int *opened, struct filename *name) 2960fb1cc555SAl Viro { 2961a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 2962ca344a89SAl Viro int open_flag = op->open_flag; 296377d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 296464894cf8SAl Viro bool got_write = false; 2965bcda7652SAl Viro int acc_mode = op->acc_mode; 2966a1eb3315SMiklos Szeredi struct inode *inode; 296716b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 2968896475d5SAl Viro struct path path; 296916b1c1cdSMiklos Szeredi bool retried = false; 297016c2cd71SAl Viro int error; 2971fb1cc555SAl Viro 2972c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 2973c3e380b0SAl Viro nd->flags |= op->intent; 2974c3e380b0SAl Viro 2975bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 2976fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 297770291aecSAl Viro if (unlikely(error)) { 297870291aecSAl Viro terminate_walk(nd); 29792675a4ebSAl Viro return error; 298070291aecSAl Viro } 2981e83db167SMiklos Szeredi goto finish_open; 29821f36f774SAl Viro } 2983a2c36b45SAl Viro 2984ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 2985fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 2986fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2987fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 2988896475d5SAl Viro error = lookup_fast(nd, &path, &inode); 298971574865SMiklos Szeredi if (likely(!error)) 299071574865SMiklos Szeredi goto finish_lookup; 299171574865SMiklos Szeredi 2992ce57dfc1SAl Viro if (error < 0) 29932675a4ebSAl Viro goto out; 2994a1eb3315SMiklos Szeredi 299537d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 2996b6183df7SMiklos Szeredi } else { 2997fe2d35ffSAl Viro /* create side of things */ 2998a3fbbde7SAl Viro /* 2999b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3000b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3001b6183df7SMiklos Szeredi * about to look up 3002a3fbbde7SAl Viro */ 30039f1fafeeSAl Viro error = complete_walk(nd); 3004191d7f73SAl Viro if (error) { 3005191d7f73SAl Viro if (nd->depth) 3006191d7f73SAl Viro put_link(nd); 30072675a4ebSAl Viro return error; 3008191d7f73SAl Viro } 3009fe2d35ffSAl Viro 301033e2208aSJeff Layton audit_inode(name, dir, LOOKUP_PARENT); 301116c2cd71SAl Viro error = -EISDIR; 30121f36f774SAl Viro /* trailing slashes? */ 301331e6b01fSNick Piggin if (nd->last.name[nd->last.len]) 30142675a4ebSAl Viro goto out; 3015b6183df7SMiklos Szeredi } 30161f36f774SAl Viro 301716b1c1cdSMiklos Szeredi retry_lookup: 301864894cf8SAl Viro if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 301964894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 302064894cf8SAl Viro if (!error) 302164894cf8SAl Viro got_write = true; 302264894cf8SAl Viro /* 302364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 302464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 302564894cf8SAl Viro * dropping this one anyway. 302664894cf8SAl Viro */ 302764894cf8SAl Viro } 3028a1e28038SAl Viro mutex_lock(&dir->d_inode->i_mutex); 3029896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 3030fb1cc555SAl Viro mutex_unlock(&dir->d_inode->i_mutex); 3031fb1cc555SAl Viro 30322675a4ebSAl Viro if (error <= 0) { 30332675a4ebSAl Viro if (error) 3034d58ffd35SMiklos Szeredi goto out; 30356c0d46c4SAl Viro 303647237687SAl Viro if ((*opened & FILE_CREATED) || 3037496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 303877d660a8SMiklos Szeredi will_truncate = false; 3039d18e9008SMiklos Szeredi 3040adb5c247SJeff Layton audit_inode(name, file->f_path.dentry, 0); 3041d18e9008SMiklos Szeredi goto opened; 3042d18e9008SMiklos Szeredi } 3043d18e9008SMiklos Szeredi 304447237687SAl Viro if (*opened & FILE_CREATED) { 30459b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3046ca344a89SAl Viro open_flag &= ~O_TRUNC; 304777d660a8SMiklos Szeredi will_truncate = false; 3048bcda7652SAl Viro acc_mode = MAY_OPEN; 3049896475d5SAl Viro path_to_nameidata(&path, nd); 3050e83db167SMiklos Szeredi goto finish_open_created; 3051fb1cc555SAl Viro } 3052fb1cc555SAl Viro 3053fb1cc555SAl Viro /* 30543134f37eSJeff Layton * create/update audit record if it already exists. 3055fb1cc555SAl Viro */ 3056896475d5SAl Viro if (d_is_positive(path.dentry)) 3057896475d5SAl Viro audit_inode(name, path.dentry, 0); 3058fb1cc555SAl Viro 3059d18e9008SMiklos Szeredi /* 3060d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3061d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3062d18e9008SMiklos Szeredi * necessary...) 3063d18e9008SMiklos Szeredi */ 306464894cf8SAl Viro if (got_write) { 3065d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 306664894cf8SAl Viro got_write = false; 3067d18e9008SMiklos Szeredi } 3068d18e9008SMiklos Szeredi 3069fb1cc555SAl Viro error = -EEXIST; 3070f8310c59SAl Viro if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT)) 3071fb1cc555SAl Viro goto exit_dput; 3072fb1cc555SAl Viro 3073896475d5SAl Viro error = follow_managed(&path, nd->flags); 30749875cf80SDavid Howells if (error < 0) 3075fb1cc555SAl Viro goto exit_dput; 3076fb1cc555SAl Viro 3077a3fbbde7SAl Viro if (error) 3078a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 3079a3fbbde7SAl Viro 3080decf3400SMiklos Szeredi BUG_ON(nd->flags & LOOKUP_RCU); 3081896475d5SAl Viro inode = path.dentry->d_inode; 3082fb1cc555SAl Viro error = -ENOENT; 3083896475d5SAl Viro if (d_is_negative(path.dentry)) { 3084896475d5SAl Viro path_to_nameidata(&path, nd); 30852675a4ebSAl Viro goto out; 308654c33e7fSMiklos Szeredi } 3087766c4cbfSAl Viro finish_lookup: 3088896475d5SAl Viro if (should_follow_link(path.dentry, nd->flags & LOOKUP_FOLLOW)) { 3089d45ea867SMiklos Szeredi if (nd->flags & LOOKUP_RCU) { 3090896475d5SAl Viro if (unlikely(nd->path.mnt != path.mnt || 3091896475d5SAl Viro unlazy_walk(nd, path.dentry))) { 3092d45ea867SMiklos Szeredi error = -ECHILD; 30932675a4ebSAl Viro goto out; 3094d45ea867SMiklos Szeredi } 3095d45ea867SMiklos Szeredi } 3096896475d5SAl Viro BUG_ON(inode != path.dentry->d_inode); 3097191d7f73SAl Viro if (nd->depth) 3098191d7f73SAl Viro put_link(nd); 3099896475d5SAl Viro nd->link = path; 31002675a4ebSAl Viro return 1; 3101d45ea867SMiklos Szeredi } 3102fb1cc555SAl Viro 3103896475d5SAl Viro if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) { 3104896475d5SAl Viro path_to_nameidata(&path, nd); 3105a5cfe2d5SAl Viro error = -ELOOP; 3106a5cfe2d5SAl Viro goto out; 3107a5cfe2d5SAl Viro } 3108a5cfe2d5SAl Viro 3109896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3110896475d5SAl Viro path_to_nameidata(&path, nd); 311116b1c1cdSMiklos Szeredi } else { 311216b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3113896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3114896475d5SAl Viro nd->path.dentry = path.dentry; 311516b1c1cdSMiklos Szeredi 311616b1c1cdSMiklos Szeredi } 3117decf3400SMiklos Szeredi nd->inode = inode; 3118a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3119bc77daa7SAl Viro finish_open: 3120a3fbbde7SAl Viro error = complete_walk(nd); 312116b1c1cdSMiklos Szeredi if (error) { 3122191d7f73SAl Viro if (nd->depth) 3123191d7f73SAl Viro put_link(nd); 312416b1c1cdSMiklos Szeredi path_put(&save_parent); 31252675a4ebSAl Viro return error; 312616b1c1cdSMiklos Szeredi } 3127bc77daa7SAl Viro audit_inode(name, nd->path.dentry, 0); 3128fb1cc555SAl Viro error = -EISDIR; 312944b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 31302675a4ebSAl Viro goto out; 3131af2f5542SMiklos Szeredi error = -ENOTDIR; 313244b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 31332675a4ebSAl Viro goto out; 31344bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 313577d660a8SMiklos Szeredi will_truncate = false; 31366c0d46c4SAl Viro 31370f9d1a10SAl Viro if (will_truncate) { 31380f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 31390f9d1a10SAl Viro if (error) 31402675a4ebSAl Viro goto out; 314164894cf8SAl Viro got_write = true; 31420f9d1a10SAl Viro } 3143e83db167SMiklos Szeredi finish_open_created: 3144bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3145ca344a89SAl Viro if (error) 31462675a4ebSAl Viro goto out; 31474aa7c634SMiklos Szeredi 31484aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 31494aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 31504aa7c634SMiklos Szeredi if (!error) { 31514aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 31524aa7c634SMiklos Szeredi } else { 315330d90494SAl Viro if (error == -EOPENSTALE) 3154f60dc3dbSMiklos Szeredi goto stale_open; 3155015c3bbcSMiklos Szeredi goto out; 3156f60dc3dbSMiklos Szeredi } 3157a8277b9bSMiklos Szeredi opened: 31582675a4ebSAl Viro error = open_check_o_direct(file); 3159015c3bbcSMiklos Szeredi if (error) 3160015c3bbcSMiklos Szeredi goto exit_fput; 31613034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3162aa4caadbSMiklos Szeredi if (error) 3163aa4caadbSMiklos Szeredi goto exit_fput; 3164aa4caadbSMiklos Szeredi 31650f9d1a10SAl Viro if (will_truncate) { 31662675a4ebSAl Viro error = handle_truncate(file); 3167aa4caadbSMiklos Szeredi if (error) 3168aa4caadbSMiklos Szeredi goto exit_fput; 31690f9d1a10SAl Viro } 3170ca344a89SAl Viro out: 317164894cf8SAl Viro if (got_write) 31720f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 317316b1c1cdSMiklos Szeredi path_put(&save_parent); 3174e276ae67SMiklos Szeredi terminate_walk(nd); 31752675a4ebSAl Viro return error; 3176fb1cc555SAl Viro 3177fb1cc555SAl Viro exit_dput: 3178896475d5SAl Viro path_put_conditional(&path, nd); 3179ca344a89SAl Viro goto out; 3180015c3bbcSMiklos Szeredi exit_fput: 31812675a4ebSAl Viro fput(file); 31822675a4ebSAl Viro goto out; 3183015c3bbcSMiklos Szeredi 3184f60dc3dbSMiklos Szeredi stale_open: 3185f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3186f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3187f60dc3dbSMiklos Szeredi goto out; 3188f60dc3dbSMiklos Szeredi 3189f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3190f60dc3dbSMiklos Szeredi path_put(&nd->path); 3191f60dc3dbSMiklos Szeredi nd->path = save_parent; 3192f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3193f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3194f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 319564894cf8SAl Viro if (got_write) { 3196f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 319764894cf8SAl Viro got_write = false; 3198f60dc3dbSMiklos Szeredi } 3199f60dc3dbSMiklos Szeredi retried = true; 3200f60dc3dbSMiklos Szeredi goto retry_lookup; 3201fb1cc555SAl Viro } 3202fb1cc555SAl Viro 320360545d0dSAl Viro static int do_tmpfile(int dfd, struct filename *pathname, 320460545d0dSAl Viro struct nameidata *nd, int flags, 320560545d0dSAl Viro const struct open_flags *op, 320660545d0dSAl Viro struct file *file, int *opened) 320760545d0dSAl Viro { 320860545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 320960545d0dSAl Viro struct dentry *dentry, *child; 321060545d0dSAl Viro struct inode *dir; 32115eb6b495SAl Viro int error = path_lookupat(dfd, pathname, 321260545d0dSAl Viro flags | LOOKUP_DIRECTORY, nd); 321360545d0dSAl Viro if (unlikely(error)) 321460545d0dSAl Viro return error; 321560545d0dSAl Viro error = mnt_want_write(nd->path.mnt); 321660545d0dSAl Viro if (unlikely(error)) 321760545d0dSAl Viro goto out; 321860545d0dSAl Viro /* we want directory to be writable */ 321960545d0dSAl Viro error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC); 322060545d0dSAl Viro if (error) 322160545d0dSAl Viro goto out2; 322260545d0dSAl Viro dentry = nd->path.dentry; 322360545d0dSAl Viro dir = dentry->d_inode; 322460545d0dSAl Viro if (!dir->i_op->tmpfile) { 322560545d0dSAl Viro error = -EOPNOTSUPP; 322660545d0dSAl Viro goto out2; 322760545d0dSAl Viro } 322860545d0dSAl Viro child = d_alloc(dentry, &name); 322960545d0dSAl Viro if (unlikely(!child)) { 323060545d0dSAl Viro error = -ENOMEM; 323160545d0dSAl Viro goto out2; 323260545d0dSAl Viro } 323360545d0dSAl Viro nd->flags &= ~LOOKUP_DIRECTORY; 323460545d0dSAl Viro nd->flags |= op->intent; 323560545d0dSAl Viro dput(nd->path.dentry); 323660545d0dSAl Viro nd->path.dentry = child; 323760545d0dSAl Viro error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode); 323860545d0dSAl Viro if (error) 323960545d0dSAl Viro goto out2; 324060545d0dSAl Viro audit_inode(pathname, nd->path.dentry, 0); 324169a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 324269a91c23SEric Rannaud error = may_open(&nd->path, MAY_OPEN, op->open_flag); 324360545d0dSAl Viro if (error) 324460545d0dSAl Viro goto out2; 324560545d0dSAl Viro file->f_path.mnt = nd->path.mnt; 324660545d0dSAl Viro error = finish_open(file, nd->path.dentry, NULL, opened); 324760545d0dSAl Viro if (error) 324860545d0dSAl Viro goto out2; 324960545d0dSAl Viro error = open_check_o_direct(file); 3250f4e0c30cSAl Viro if (error) { 325160545d0dSAl Viro fput(file); 3252f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3253f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3254f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3255f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3256f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3257f4e0c30cSAl Viro } 325860545d0dSAl Viro out2: 325960545d0dSAl Viro mnt_drop_write(nd->path.mnt); 326060545d0dSAl Viro out: 326160545d0dSAl Viro path_put(&nd->path); 326260545d0dSAl Viro return error; 326360545d0dSAl Viro } 326460545d0dSAl Viro 3265669abf4eSJeff Layton static struct file *path_openat(int dfd, struct filename *pathname, 326673d049a4SAl Viro struct nameidata *nd, const struct open_flags *op, int flags) 32671da177e4SLinus Torvalds { 326830d90494SAl Viro struct file *file; 326947237687SAl Viro int opened = 0; 327013aab428SAl Viro int error; 327131e6b01fSNick Piggin 327230d90494SAl Viro file = get_empty_filp(); 32731afc99beSAl Viro if (IS_ERR(file)) 32741afc99beSAl Viro return file; 327531e6b01fSNick Piggin 327630d90494SAl Viro file->f_flags = op->open_flag; 327731e6b01fSNick Piggin 3278bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 327960545d0dSAl Viro error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened); 3280f15133dfSAl Viro goto out2; 328160545d0dSAl Viro } 328260545d0dSAl Viro 32836e8a1f87SAl Viro error = path_init(dfd, pathname, flags, nd); 328431e6b01fSNick Piggin if (unlikely(error)) 32852675a4ebSAl Viro goto out; 32861da177e4SLinus Torvalds 3287191d7f73SAl Viro while ((error = do_last(nd, file, op, &opened, pathname)) > 0) { 328873d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 32891d8e03d3SAl Viro error = trailing_symlink(nd); 3290c3e380b0SAl Viro if (unlikely(error)) 32912675a4ebSAl Viro break; 3292806b681cSAl Viro } 329310fa8e62SAl Viro out: 3294893b7775SAl Viro path_cleanup(nd); 3295f15133dfSAl Viro out2: 32962675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 32972675a4ebSAl Viro BUG_ON(!error); 329830d90494SAl Viro put_filp(file); 3299015c3bbcSMiklos Szeredi } 33002675a4ebSAl Viro if (unlikely(error)) { 33012675a4ebSAl Viro if (error == -EOPENSTALE) { 33022675a4ebSAl Viro if (flags & LOOKUP_RCU) 33032675a4ebSAl Viro error = -ECHILD; 33042675a4ebSAl Viro else 33052675a4ebSAl Viro error = -ESTALE; 33062675a4ebSAl Viro } 33072675a4ebSAl Viro file = ERR_PTR(error); 33082675a4ebSAl Viro } 33092675a4ebSAl Viro return file; 3310de459215SKirill Korotaev } 33111da177e4SLinus Torvalds 3312669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3313f9652e10SAl Viro const struct open_flags *op) 331413aab428SAl Viro { 331573d049a4SAl Viro struct nameidata nd; 3316f9652e10SAl Viro int flags = op->lookup_flags; 331713aab428SAl Viro struct file *filp; 331813aab428SAl Viro 3319894bc8c4SAl Viro set_nameidata(&nd); 332073d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU); 332113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 332273d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags); 332313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 332473d049a4SAl Viro filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL); 3325894bc8c4SAl Viro restore_nameidata(&nd); 332613aab428SAl Viro return filp; 332713aab428SAl Viro } 332813aab428SAl Viro 332973d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3330f9652e10SAl Viro const char *name, const struct open_flags *op) 333173d049a4SAl Viro { 333273d049a4SAl Viro struct nameidata nd; 333373d049a4SAl Viro struct file *file; 333451689104SPaul Moore struct filename *filename; 3335f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 333673d049a4SAl Viro 333773d049a4SAl Viro nd.root.mnt = mnt; 333873d049a4SAl Viro nd.root.dentry = dentry; 3339894bc8c4SAl Viro set_nameidata(&nd); 334073d049a4SAl Viro 3341b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 334273d049a4SAl Viro return ERR_PTR(-ELOOP); 334373d049a4SAl Viro 334451689104SPaul Moore filename = getname_kernel(name); 334551689104SPaul Moore if (unlikely(IS_ERR(filename))) 334651689104SPaul Moore return ERR_CAST(filename); 334751689104SPaul Moore 334851689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU); 334973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 335051689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags); 335173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 335251689104SPaul Moore file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL); 3353894bc8c4SAl Viro restore_nameidata(&nd); 335451689104SPaul Moore putname(filename); 335573d049a4SAl Viro return file; 335673d049a4SAl Viro } 335773d049a4SAl Viro 3358fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 33591ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 33601da177e4SLinus Torvalds { 3361c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3362ed75e95dSAl Viro struct nameidata nd; 3363c30dabfeSJan Kara int err2; 33641ac12b4bSJeff Layton int error; 33651ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 33661ac12b4bSJeff Layton 33671ac12b4bSJeff Layton /* 33681ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 33691ac12b4bSJeff Layton * other flags passed in are ignored! 33701ac12b4bSJeff Layton */ 33711ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 33721ac12b4bSJeff Layton 3373fa14a0b8SAl Viro error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd); 3374ed75e95dSAl Viro if (error) 3375ed75e95dSAl Viro return ERR_PTR(error); 33761da177e4SLinus Torvalds 3377c663e5d8SChristoph Hellwig /* 3378c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3379c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3380c663e5d8SChristoph Hellwig */ 3381ed75e95dSAl Viro if (nd.last_type != LAST_NORM) 3382ed75e95dSAl Viro goto out; 3383ed75e95dSAl Viro nd.flags &= ~LOOKUP_PARENT; 3384ed75e95dSAl Viro nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL; 3385c663e5d8SChristoph Hellwig 3386c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3387c30dabfeSJan Kara err2 = mnt_want_write(nd.path.mnt); 3388c663e5d8SChristoph Hellwig /* 3389c663e5d8SChristoph Hellwig * Do the final lookup. 3390c663e5d8SChristoph Hellwig */ 3391ed75e95dSAl Viro mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 33926a9f40d6SAl Viro dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags); 33931da177e4SLinus Torvalds if (IS_ERR(dentry)) 3394a8104a9fSAl Viro goto unlock; 3395c663e5d8SChristoph Hellwig 3396a8104a9fSAl Viro error = -EEXIST; 3397b18825a7SDavid Howells if (d_is_positive(dentry)) 3398a8104a9fSAl Viro goto fail; 3399b18825a7SDavid Howells 3400c663e5d8SChristoph Hellwig /* 3401c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3402c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3403c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3404c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3405c663e5d8SChristoph Hellwig */ 3406ed75e95dSAl Viro if (unlikely(!is_dir && nd.last.name[nd.last.len])) { 3407a8104a9fSAl Viro error = -ENOENT; 3408ed75e95dSAl Viro goto fail; 3409e9baf6e5SAl Viro } 3410c30dabfeSJan Kara if (unlikely(err2)) { 3411c30dabfeSJan Kara error = err2; 3412a8104a9fSAl Viro goto fail; 3413c30dabfeSJan Kara } 3414ed75e95dSAl Viro *path = nd.path; 3415e9baf6e5SAl Viro return dentry; 34161da177e4SLinus Torvalds fail: 3417a8104a9fSAl Viro dput(dentry); 3418a8104a9fSAl Viro dentry = ERR_PTR(error); 3419a8104a9fSAl Viro unlock: 3420dae6ad8fSAl Viro mutex_unlock(&nd.path.dentry->d_inode->i_mutex); 3421c30dabfeSJan Kara if (!err2) 3422c30dabfeSJan Kara mnt_drop_write(nd.path.mnt); 3423ed75e95dSAl Viro out: 3424dae6ad8fSAl Viro path_put(&nd.path); 3425ed75e95dSAl Viro return dentry; 3426dae6ad8fSAl Viro } 3427fa14a0b8SAl Viro 3428fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3429fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3430fa14a0b8SAl Viro { 343151689104SPaul Moore struct filename *filename = getname_kernel(pathname); 343251689104SPaul Moore struct dentry *res; 343351689104SPaul Moore 343451689104SPaul Moore if (IS_ERR(filename)) 343551689104SPaul Moore return ERR_CAST(filename); 343651689104SPaul Moore res = filename_create(dfd, filename, path, lookup_flags); 343751689104SPaul Moore putname(filename); 343851689104SPaul Moore return res; 3439fa14a0b8SAl Viro } 3440dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3441dae6ad8fSAl Viro 3442921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3443921a1650SAl Viro { 3444921a1650SAl Viro dput(dentry); 3445921a1650SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 3446a8104a9fSAl Viro mnt_drop_write(path->mnt); 3447921a1650SAl Viro path_put(path); 3448921a1650SAl Viro } 3449921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3450921a1650SAl Viro 34511ac12b4bSJeff Layton struct dentry *user_path_create(int dfd, const char __user *pathname, 34521ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3453dae6ad8fSAl Viro { 345491a27b2aSJeff Layton struct filename *tmp = getname(pathname); 3455dae6ad8fSAl Viro struct dentry *res; 3456dae6ad8fSAl Viro if (IS_ERR(tmp)) 3457dae6ad8fSAl Viro return ERR_CAST(tmp); 3458fa14a0b8SAl Viro res = filename_create(dfd, tmp, path, lookup_flags); 3459dae6ad8fSAl Viro putname(tmp); 3460dae6ad8fSAl Viro return res; 3461dae6ad8fSAl Viro } 3462dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3463dae6ad8fSAl Viro 34641a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 34651da177e4SLinus Torvalds { 3466a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 34671da177e4SLinus Torvalds 34681da177e4SLinus Torvalds if (error) 34691da177e4SLinus Torvalds return error; 34701da177e4SLinus Torvalds 3471975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 34721da177e4SLinus Torvalds return -EPERM; 34731da177e4SLinus Torvalds 3474acfa4380SAl Viro if (!dir->i_op->mknod) 34751da177e4SLinus Torvalds return -EPERM; 34761da177e4SLinus Torvalds 347708ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 347808ce5f16SSerge E. Hallyn if (error) 347908ce5f16SSerge E. Hallyn return error; 348008ce5f16SSerge E. Hallyn 34811da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 34821da177e4SLinus Torvalds if (error) 34831da177e4SLinus Torvalds return error; 34841da177e4SLinus Torvalds 34851da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3486a74574aaSStephen Smalley if (!error) 3487f38aa942SAmy Griffis fsnotify_create(dir, dentry); 34881da177e4SLinus Torvalds return error; 34891da177e4SLinus Torvalds } 34904d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 34911da177e4SLinus Torvalds 3492f69aac00SAl Viro static int may_mknod(umode_t mode) 3493463c3197SDave Hansen { 3494463c3197SDave Hansen switch (mode & S_IFMT) { 3495463c3197SDave Hansen case S_IFREG: 3496463c3197SDave Hansen case S_IFCHR: 3497463c3197SDave Hansen case S_IFBLK: 3498463c3197SDave Hansen case S_IFIFO: 3499463c3197SDave Hansen case S_IFSOCK: 3500463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3501463c3197SDave Hansen return 0; 3502463c3197SDave Hansen case S_IFDIR: 3503463c3197SDave Hansen return -EPERM; 3504463c3197SDave Hansen default: 3505463c3197SDave Hansen return -EINVAL; 3506463c3197SDave Hansen } 3507463c3197SDave Hansen } 3508463c3197SDave Hansen 35098208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 35102e4d0924SHeiko Carstens unsigned, dev) 35111da177e4SLinus Torvalds { 35121da177e4SLinus Torvalds struct dentry *dentry; 3513dae6ad8fSAl Viro struct path path; 3514dae6ad8fSAl Viro int error; 3515972567f1SJeff Layton unsigned int lookup_flags = 0; 35161da177e4SLinus Torvalds 35178e4bfca1SAl Viro error = may_mknod(mode); 35188e4bfca1SAl Viro if (error) 35198e4bfca1SAl Viro return error; 3520972567f1SJeff Layton retry: 3521972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3522dae6ad8fSAl Viro if (IS_ERR(dentry)) 3523dae6ad8fSAl Viro return PTR_ERR(dentry); 35242ad94ae6SAl Viro 3525dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3526ce3b0f8dSAl Viro mode &= ~current_umask(); 3527dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3528be6d3e56SKentaro Takeda if (error) 3529a8104a9fSAl Viro goto out; 35301da177e4SLinus Torvalds switch (mode & S_IFMT) { 35311da177e4SLinus Torvalds case 0: case S_IFREG: 3532312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 35331da177e4SLinus Torvalds break; 35341da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3535dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 35361da177e4SLinus Torvalds new_decode_dev(dev)); 35371da177e4SLinus Torvalds break; 35381da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3539dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 35401da177e4SLinus Torvalds break; 35411da177e4SLinus Torvalds } 3542a8104a9fSAl Viro out: 3543921a1650SAl Viro done_path_create(&path, dentry); 3544972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3545972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3546972567f1SJeff Layton goto retry; 3547972567f1SJeff Layton } 35481da177e4SLinus Torvalds return error; 35491da177e4SLinus Torvalds } 35501da177e4SLinus Torvalds 35518208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 35525590ff0dSUlrich Drepper { 35535590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 35545590ff0dSUlrich Drepper } 35555590ff0dSUlrich Drepper 355618bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 35571da177e4SLinus Torvalds { 3558a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35598de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 35601da177e4SLinus Torvalds 35611da177e4SLinus Torvalds if (error) 35621da177e4SLinus Torvalds return error; 35631da177e4SLinus Torvalds 3564acfa4380SAl Viro if (!dir->i_op->mkdir) 35651da177e4SLinus Torvalds return -EPERM; 35661da177e4SLinus Torvalds 35671da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 35681da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 35691da177e4SLinus Torvalds if (error) 35701da177e4SLinus Torvalds return error; 35711da177e4SLinus Torvalds 35728de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 35738de52778SAl Viro return -EMLINK; 35748de52778SAl Viro 35751da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3576a74574aaSStephen Smalley if (!error) 3577f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 35781da177e4SLinus Torvalds return error; 35791da177e4SLinus Torvalds } 35804d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 35811da177e4SLinus Torvalds 3582a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 35831da177e4SLinus Torvalds { 35846902d925SDave Hansen struct dentry *dentry; 3585dae6ad8fSAl Viro struct path path; 3586dae6ad8fSAl Viro int error; 3587b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 35881da177e4SLinus Torvalds 3589b76d8b82SJeff Layton retry: 3590b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 35916902d925SDave Hansen if (IS_ERR(dentry)) 3592dae6ad8fSAl Viro return PTR_ERR(dentry); 35936902d925SDave Hansen 3594dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3595ce3b0f8dSAl Viro mode &= ~current_umask(); 3596dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3597a8104a9fSAl Viro if (!error) 3598dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3599921a1650SAl Viro done_path_create(&path, dentry); 3600b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3601b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3602b76d8b82SJeff Layton goto retry; 3603b76d8b82SJeff Layton } 36041da177e4SLinus Torvalds return error; 36051da177e4SLinus Torvalds } 36061da177e4SLinus Torvalds 3607a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36085590ff0dSUlrich Drepper { 36095590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 36105590ff0dSUlrich Drepper } 36115590ff0dSUlrich Drepper 36121da177e4SLinus Torvalds /* 3613a71905f0SSage Weil * The dentry_unhash() helper will try to drop the dentry early: we 3614c0d02594SJ. Bruce Fields * should have a usage count of 1 if we're the only user of this 3615a71905f0SSage Weil * dentry, and if that is true (possibly after pruning the dcache), 3616a71905f0SSage Weil * then we drop the dentry now. 36171da177e4SLinus Torvalds * 36181da177e4SLinus Torvalds * A low-level filesystem can, if it choses, legally 36191da177e4SLinus Torvalds * do a 36201da177e4SLinus Torvalds * 36211da177e4SLinus Torvalds * if (!d_unhashed(dentry)) 36221da177e4SLinus Torvalds * return -EBUSY; 36231da177e4SLinus Torvalds * 36241da177e4SLinus Torvalds * if it cannot handle the case of removing a directory 36251da177e4SLinus Torvalds * that is still in use by something else.. 36261da177e4SLinus Torvalds */ 36271da177e4SLinus Torvalds void dentry_unhash(struct dentry *dentry) 36281da177e4SLinus Torvalds { 36291da177e4SLinus Torvalds shrink_dcache_parent(dentry); 36301da177e4SLinus Torvalds spin_lock(&dentry->d_lock); 363198474236SWaiman Long if (dentry->d_lockref.count == 1) 36321da177e4SLinus Torvalds __d_drop(dentry); 36331da177e4SLinus Torvalds spin_unlock(&dentry->d_lock); 36341da177e4SLinus Torvalds } 36354d359507SAl Viro EXPORT_SYMBOL(dentry_unhash); 36361da177e4SLinus Torvalds 36371da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 36381da177e4SLinus Torvalds { 36391da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 36401da177e4SLinus Torvalds 36411da177e4SLinus Torvalds if (error) 36421da177e4SLinus Torvalds return error; 36431da177e4SLinus Torvalds 3644acfa4380SAl Viro if (!dir->i_op->rmdir) 36451da177e4SLinus Torvalds return -EPERM; 36461da177e4SLinus Torvalds 36471d2ef590SAl Viro dget(dentry); 36481b1dcc1bSJes Sorensen mutex_lock(&dentry->d_inode->i_mutex); 3649912dbc15SSage Weil 36501da177e4SLinus Torvalds error = -EBUSY; 36517af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3652912dbc15SSage Weil goto out; 3653912dbc15SSage Weil 36541da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3655912dbc15SSage Weil if (error) 3656912dbc15SSage Weil goto out; 3657912dbc15SSage Weil 36583cebde24SSage Weil shrink_dcache_parent(dentry); 36591da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3660912dbc15SSage Weil if (error) 3661912dbc15SSage Weil goto out; 3662912dbc15SSage Weil 36631da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3664d83c49f3SAl Viro dont_mount(dentry); 36658ed936b5SEric W. Biederman detach_mounts(dentry); 36661da177e4SLinus Torvalds 3667912dbc15SSage Weil out: 3668912dbc15SSage Weil mutex_unlock(&dentry->d_inode->i_mutex); 36691d2ef590SAl Viro dput(dentry); 3670912dbc15SSage Weil if (!error) 3671912dbc15SSage Weil d_delete(dentry); 36721da177e4SLinus Torvalds return error; 36731da177e4SLinus Torvalds } 36744d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 36751da177e4SLinus Torvalds 36765590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 36771da177e4SLinus Torvalds { 36781da177e4SLinus Torvalds int error = 0; 367991a27b2aSJeff Layton struct filename *name; 36801da177e4SLinus Torvalds struct dentry *dentry; 3681f5beed75SAl Viro struct path path; 3682f5beed75SAl Viro struct qstr last; 3683f5beed75SAl Viro int type; 3684c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3685c6ee9206SJeff Layton retry: 3686f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3687f5beed75SAl Viro &path, &last, &type, lookup_flags); 368891a27b2aSJeff Layton if (IS_ERR(name)) 368991a27b2aSJeff Layton return PTR_ERR(name); 36901da177e4SLinus Torvalds 3691f5beed75SAl Viro switch (type) { 36921da177e4SLinus Torvalds case LAST_DOTDOT: 36931da177e4SLinus Torvalds error = -ENOTEMPTY; 36941da177e4SLinus Torvalds goto exit1; 36951da177e4SLinus Torvalds case LAST_DOT: 36961da177e4SLinus Torvalds error = -EINVAL; 36971da177e4SLinus Torvalds goto exit1; 36981da177e4SLinus Torvalds case LAST_ROOT: 36991da177e4SLinus Torvalds error = -EBUSY; 37001da177e4SLinus Torvalds goto exit1; 37011da177e4SLinus Torvalds } 37020612d9fbSOGAWA Hirofumi 3703f5beed75SAl Viro error = mnt_want_write(path.mnt); 3704c30dabfeSJan Kara if (error) 3705c30dabfeSJan Kara goto exit1; 37060612d9fbSOGAWA Hirofumi 3707f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3708f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37091da177e4SLinus Torvalds error = PTR_ERR(dentry); 37106902d925SDave Hansen if (IS_ERR(dentry)) 37116902d925SDave Hansen goto exit2; 3712e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3713e6bc45d6STheodore Ts'o error = -ENOENT; 3714e6bc45d6STheodore Ts'o goto exit3; 3715e6bc45d6STheodore Ts'o } 3716f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3717be6d3e56SKentaro Takeda if (error) 3718c30dabfeSJan Kara goto exit3; 3719f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37200622753bSDave Hansen exit3: 37211da177e4SLinus Torvalds dput(dentry); 37226902d925SDave Hansen exit2: 3723f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 3724f5beed75SAl Viro mnt_drop_write(path.mnt); 37251da177e4SLinus Torvalds exit1: 3726f5beed75SAl Viro path_put(&path); 37271da177e4SLinus Torvalds putname(name); 3728c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3729c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3730c6ee9206SJeff Layton goto retry; 3731c6ee9206SJeff Layton } 37321da177e4SLinus Torvalds return error; 37331da177e4SLinus Torvalds } 37341da177e4SLinus Torvalds 37353cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37365590ff0dSUlrich Drepper { 37375590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 37385590ff0dSUlrich Drepper } 37395590ff0dSUlrich Drepper 3740b21996e3SJ. Bruce Fields /** 3741b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3742b21996e3SJ. Bruce Fields * @dir: parent directory 3743b21996e3SJ. Bruce Fields * @dentry: victim 3744b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3745b21996e3SJ. Bruce Fields * 3746b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3747b21996e3SJ. Bruce Fields * 3748b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3749b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3750b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3751b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3752b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3753b21996e3SJ. Bruce Fields * 3754b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3755b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3756b21996e3SJ. Bruce Fields * to be NFS exported. 3757b21996e3SJ. Bruce Fields */ 3758b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 37591da177e4SLinus Torvalds { 37609accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 37611da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 37621da177e4SLinus Torvalds 37631da177e4SLinus Torvalds if (error) 37641da177e4SLinus Torvalds return error; 37651da177e4SLinus Torvalds 3766acfa4380SAl Viro if (!dir->i_op->unlink) 37671da177e4SLinus Torvalds return -EPERM; 37681da177e4SLinus Torvalds 37699accbb97SJ. Bruce Fields mutex_lock(&target->i_mutex); 37708ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 37711da177e4SLinus Torvalds error = -EBUSY; 37721da177e4SLinus Torvalds else { 37731da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3774bec1052eSAl Viro if (!error) { 37755a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 37765a14696cSJ. Bruce Fields if (error) 3777b21996e3SJ. Bruce Fields goto out; 37781da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 37798ed936b5SEric W. Biederman if (!error) { 3780d83c49f3SAl Viro dont_mount(dentry); 37818ed936b5SEric W. Biederman detach_mounts(dentry); 37828ed936b5SEric W. Biederman } 3783bec1052eSAl Viro } 37841da177e4SLinus Torvalds } 3785b21996e3SJ. Bruce Fields out: 37869accbb97SJ. Bruce Fields mutex_unlock(&target->i_mutex); 37871da177e4SLinus Torvalds 37881da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 37891da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 37909accbb97SJ. Bruce Fields fsnotify_link_count(target); 37911da177e4SLinus Torvalds d_delete(dentry); 37921da177e4SLinus Torvalds } 37930eeca283SRobert Love 37941da177e4SLinus Torvalds return error; 37951da177e4SLinus Torvalds } 37964d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 37971da177e4SLinus Torvalds 37981da177e4SLinus Torvalds /* 37991da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38001b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38011da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38021da177e4SLinus Torvalds * while waiting on the I/O. 38031da177e4SLinus Torvalds */ 38045590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 38051da177e4SLinus Torvalds { 38062ad94ae6SAl Viro int error; 380791a27b2aSJeff Layton struct filename *name; 38081da177e4SLinus Torvalds struct dentry *dentry; 3809f5beed75SAl Viro struct path path; 3810f5beed75SAl Viro struct qstr last; 3811f5beed75SAl Viro int type; 38121da177e4SLinus Torvalds struct inode *inode = NULL; 3813b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38145d18f813SJeff Layton unsigned int lookup_flags = 0; 38155d18f813SJeff Layton retry: 3816f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3817f5beed75SAl Viro &path, &last, &type, lookup_flags); 381891a27b2aSJeff Layton if (IS_ERR(name)) 381991a27b2aSJeff Layton return PTR_ERR(name); 38202ad94ae6SAl Viro 38211da177e4SLinus Torvalds error = -EISDIR; 3822f5beed75SAl Viro if (type != LAST_NORM) 38231da177e4SLinus Torvalds goto exit1; 38240612d9fbSOGAWA Hirofumi 3825f5beed75SAl Viro error = mnt_want_write(path.mnt); 3826c30dabfeSJan Kara if (error) 3827c30dabfeSJan Kara goto exit1; 3828b21996e3SJ. Bruce Fields retry_deleg: 3829f5beed75SAl Viro mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT); 3830f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38311da177e4SLinus Torvalds error = PTR_ERR(dentry); 38321da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38331da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3834f5beed75SAl Viro if (last.name[last.len]) 383550338b88STörök Edwin goto slashes; 38361da177e4SLinus Torvalds inode = dentry->d_inode; 3837b18825a7SDavid Howells if (d_is_negative(dentry)) 3838e6bc45d6STheodore Ts'o goto slashes; 38397de9c6eeSAl Viro ihold(inode); 3840f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3841be6d3e56SKentaro Takeda if (error) 3842c30dabfeSJan Kara goto exit2; 3843f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 38441da177e4SLinus Torvalds exit2: 38451da177e4SLinus Torvalds dput(dentry); 38461da177e4SLinus Torvalds } 3847f5beed75SAl Viro mutex_unlock(&path.dentry->d_inode->i_mutex); 38481da177e4SLinus Torvalds if (inode) 38491da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3850b21996e3SJ. Bruce Fields inode = NULL; 3851b21996e3SJ. Bruce Fields if (delegated_inode) { 38525a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3853b21996e3SJ. Bruce Fields if (!error) 3854b21996e3SJ. Bruce Fields goto retry_deleg; 3855b21996e3SJ. Bruce Fields } 3856f5beed75SAl Viro mnt_drop_write(path.mnt); 38571da177e4SLinus Torvalds exit1: 3858f5beed75SAl Viro path_put(&path); 38591da177e4SLinus Torvalds putname(name); 38605d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 38615d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 38625d18f813SJeff Layton inode = NULL; 38635d18f813SJeff Layton goto retry; 38645d18f813SJeff Layton } 38651da177e4SLinus Torvalds return error; 38661da177e4SLinus Torvalds 38671da177e4SLinus Torvalds slashes: 3868b18825a7SDavid Howells if (d_is_negative(dentry)) 3869b18825a7SDavid Howells error = -ENOENT; 387044b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3871b18825a7SDavid Howells error = -EISDIR; 3872b18825a7SDavid Howells else 3873b18825a7SDavid Howells error = -ENOTDIR; 38741da177e4SLinus Torvalds goto exit2; 38751da177e4SLinus Torvalds } 38761da177e4SLinus Torvalds 38772e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 38785590ff0dSUlrich Drepper { 38795590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 38805590ff0dSUlrich Drepper return -EINVAL; 38815590ff0dSUlrich Drepper 38825590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 38835590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 38845590ff0dSUlrich Drepper 38855590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 38865590ff0dSUlrich Drepper } 38875590ff0dSUlrich Drepper 38883480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 38895590ff0dSUlrich Drepper { 38905590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 38915590ff0dSUlrich Drepper } 38925590ff0dSUlrich Drepper 3893db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 38941da177e4SLinus Torvalds { 3895a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38961da177e4SLinus Torvalds 38971da177e4SLinus Torvalds if (error) 38981da177e4SLinus Torvalds return error; 38991da177e4SLinus Torvalds 3900acfa4380SAl Viro if (!dir->i_op->symlink) 39011da177e4SLinus Torvalds return -EPERM; 39021da177e4SLinus Torvalds 39031da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39041da177e4SLinus Torvalds if (error) 39051da177e4SLinus Torvalds return error; 39061da177e4SLinus Torvalds 39071da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3908a74574aaSStephen Smalley if (!error) 3909f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39101da177e4SLinus Torvalds return error; 39111da177e4SLinus Torvalds } 39124d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39131da177e4SLinus Torvalds 39142e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 39152e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 39161da177e4SLinus Torvalds { 39172ad94ae6SAl Viro int error; 391891a27b2aSJeff Layton struct filename *from; 39196902d925SDave Hansen struct dentry *dentry; 3920dae6ad8fSAl Viro struct path path; 3921f46d3567SJeff Layton unsigned int lookup_flags = 0; 39221da177e4SLinus Torvalds 39231da177e4SLinus Torvalds from = getname(oldname); 39241da177e4SLinus Torvalds if (IS_ERR(from)) 39251da177e4SLinus Torvalds return PTR_ERR(from); 3926f46d3567SJeff Layton retry: 3927f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39281da177e4SLinus Torvalds error = PTR_ERR(dentry); 39296902d925SDave Hansen if (IS_ERR(dentry)) 3930dae6ad8fSAl Viro goto out_putname; 39316902d925SDave Hansen 393291a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3933a8104a9fSAl Viro if (!error) 393491a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3935921a1650SAl Viro done_path_create(&path, dentry); 3936f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3937f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 3938f46d3567SJeff Layton goto retry; 3939f46d3567SJeff Layton } 39406902d925SDave Hansen out_putname: 39411da177e4SLinus Torvalds putname(from); 39421da177e4SLinus Torvalds return error; 39431da177e4SLinus Torvalds } 39441da177e4SLinus Torvalds 39453480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 39465590ff0dSUlrich Drepper { 39475590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 39485590ff0dSUlrich Drepper } 39495590ff0dSUlrich Drepper 3950146a8595SJ. Bruce Fields /** 3951146a8595SJ. Bruce Fields * vfs_link - create a new link 3952146a8595SJ. Bruce Fields * @old_dentry: object to be linked 3953146a8595SJ. Bruce Fields * @dir: new parent 3954146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 3955146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 3956146a8595SJ. Bruce Fields * 3957146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 3958146a8595SJ. Bruce Fields * 3959146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 3960146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 3961146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 3962146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 3963146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 3964146a8595SJ. Bruce Fields * 3965146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3966146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3967146a8595SJ. Bruce Fields * to be NFS exported. 3968146a8595SJ. Bruce Fields */ 3969146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 39701da177e4SLinus Torvalds { 39711da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 39728de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39731da177e4SLinus Torvalds int error; 39741da177e4SLinus Torvalds 39751da177e4SLinus Torvalds if (!inode) 39761da177e4SLinus Torvalds return -ENOENT; 39771da177e4SLinus Torvalds 3978a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 39791da177e4SLinus Torvalds if (error) 39801da177e4SLinus Torvalds return error; 39811da177e4SLinus Torvalds 39821da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 39831da177e4SLinus Torvalds return -EXDEV; 39841da177e4SLinus Torvalds 39851da177e4SLinus Torvalds /* 39861da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 39871da177e4SLinus Torvalds */ 39881da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 39891da177e4SLinus Torvalds return -EPERM; 3990acfa4380SAl Viro if (!dir->i_op->link) 39911da177e4SLinus Torvalds return -EPERM; 39927e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 39931da177e4SLinus Torvalds return -EPERM; 39941da177e4SLinus Torvalds 39951da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 39961da177e4SLinus Torvalds if (error) 39971da177e4SLinus Torvalds return error; 39981da177e4SLinus Torvalds 39997e79eedbSTetsuo Handa mutex_lock(&inode->i_mutex); 4000aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4001f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4002aae8a97dSAneesh Kumar K.V error = -ENOENT; 40038de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40048de52778SAl Viro error = -EMLINK; 4005146a8595SJ. Bruce Fields else { 4006146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4007146a8595SJ. Bruce Fields if (!error) 40081da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4009146a8595SJ. Bruce Fields } 4010f4e0c30cSAl Viro 4011f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4012f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4013f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4014f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4015f4e0c30cSAl Viro } 40167e79eedbSTetsuo Handa mutex_unlock(&inode->i_mutex); 4017e31e14ecSStephen Smalley if (!error) 40187e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40191da177e4SLinus Torvalds return error; 40201da177e4SLinus Torvalds } 40214d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40221da177e4SLinus Torvalds 40231da177e4SLinus Torvalds /* 40241da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 40251da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 40261da177e4SLinus Torvalds * newname. --KAB 40271da177e4SLinus Torvalds * 40281da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 40291da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 40301da177e4SLinus Torvalds * and other special files. --ADM 40311da177e4SLinus Torvalds */ 40322e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 40332e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 40341da177e4SLinus Torvalds { 40351da177e4SLinus Torvalds struct dentry *new_dentry; 4036dae6ad8fSAl Viro struct path old_path, new_path; 4037146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 403811a7b371SAneesh Kumar K.V int how = 0; 40391da177e4SLinus Torvalds int error; 40401da177e4SLinus Torvalds 404111a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4042c04030e1SUlrich Drepper return -EINVAL; 404311a7b371SAneesh Kumar K.V /* 4044f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4045f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4046f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 404711a7b371SAneesh Kumar K.V */ 4048f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4049f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4050f0cc6ffbSLinus Torvalds return -ENOENT; 405111a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4052f0cc6ffbSLinus Torvalds } 4053c04030e1SUlrich Drepper 405411a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 405511a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4056442e31caSJeff Layton retry: 405711a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 40581da177e4SLinus Torvalds if (error) 40592ad94ae6SAl Viro return error; 40602ad94ae6SAl Viro 4061442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4062442e31caSJeff Layton (how & LOOKUP_REVAL)); 40631da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 40646902d925SDave Hansen if (IS_ERR(new_dentry)) 4065dae6ad8fSAl Viro goto out; 4066dae6ad8fSAl Viro 4067dae6ad8fSAl Viro error = -EXDEV; 4068dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4069dae6ad8fSAl Viro goto out_dput; 4070800179c9SKees Cook error = may_linkat(&old_path); 4071800179c9SKees Cook if (unlikely(error)) 4072800179c9SKees Cook goto out_dput; 4073dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4074be6d3e56SKentaro Takeda if (error) 4075a8104a9fSAl Viro goto out_dput; 4076146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 407775c3f29dSDave Hansen out_dput: 4078921a1650SAl Viro done_path_create(&new_path, new_dentry); 4079146a8595SJ. Bruce Fields if (delegated_inode) { 4080146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4081d22e6338SOleg Drokin if (!error) { 4082d22e6338SOleg Drokin path_put(&old_path); 4083146a8595SJ. Bruce Fields goto retry; 4084146a8595SJ. Bruce Fields } 4085d22e6338SOleg Drokin } 4086442e31caSJeff Layton if (retry_estale(error, how)) { 4087d22e6338SOleg Drokin path_put(&old_path); 4088442e31caSJeff Layton how |= LOOKUP_REVAL; 4089442e31caSJeff Layton goto retry; 4090442e31caSJeff Layton } 40911da177e4SLinus Torvalds out: 40922d8f3038SAl Viro path_put(&old_path); 40931da177e4SLinus Torvalds 40941da177e4SLinus Torvalds return error; 40951da177e4SLinus Torvalds } 40961da177e4SLinus Torvalds 40973480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 40985590ff0dSUlrich Drepper { 4099c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41005590ff0dSUlrich Drepper } 41015590ff0dSUlrich Drepper 4102bc27027aSMiklos Szeredi /** 4103bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4104bc27027aSMiklos Szeredi * @old_dir: parent of source 4105bc27027aSMiklos Szeredi * @old_dentry: source 4106bc27027aSMiklos Szeredi * @new_dir: parent of destination 4107bc27027aSMiklos Szeredi * @new_dentry: destination 4108bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4109520c8b16SMiklos Szeredi * @flags: rename flags 4110bc27027aSMiklos Szeredi * 4111bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4112bc27027aSMiklos Szeredi * 4113bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4114bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4115bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4116bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4117bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4118bc27027aSMiklos Szeredi * so. 4119bc27027aSMiklos Szeredi * 4120bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4121bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4122bc27027aSMiklos Szeredi * to be NFS exported. 4123bc27027aSMiklos Szeredi * 41241da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 41251da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 41261da177e4SLinus Torvalds * Problems: 4127d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 41281da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 41291da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4130a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 41311da177e4SLinus Torvalds * story. 41326cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 41336cedba89SJ. Bruce Fields * and source (if it is not a directory). 41341b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 41351da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 41361da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4137a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 41381da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 41391da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 41401da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4141a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 41421da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 41431da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 41441da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4145e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 41461b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 41471da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4148c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 41491da177e4SLinus Torvalds * locking]. 41501da177e4SLinus Torvalds */ 41511da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 41528e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4153520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 41541da177e4SLinus Torvalds { 41551da177e4SLinus Torvalds int error; 4156bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 415759b0df21SEric Paris const unsigned char *old_name; 4158bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4159bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4160da1ce067SMiklos Szeredi bool new_is_dir = false; 4161da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 41621da177e4SLinus Torvalds 4163bc27027aSMiklos Szeredi if (source == target) 41641da177e4SLinus Torvalds return 0; 41651da177e4SLinus Torvalds 41661da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 41671da177e4SLinus Torvalds if (error) 41681da177e4SLinus Torvalds return error; 41691da177e4SLinus Torvalds 4170da1ce067SMiklos Szeredi if (!target) { 4171a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4172da1ce067SMiklos Szeredi } else { 4173da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4174da1ce067SMiklos Szeredi 4175da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 41761da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4177da1ce067SMiklos Szeredi else 4178da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4179da1ce067SMiklos Szeredi } 41801da177e4SLinus Torvalds if (error) 41811da177e4SLinus Torvalds return error; 41821da177e4SLinus Torvalds 41837177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 41841da177e4SLinus Torvalds return -EPERM; 41851da177e4SLinus Torvalds 4186520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4187520c8b16SMiklos Szeredi return -EINVAL; 4188520c8b16SMiklos Szeredi 4189bc27027aSMiklos Szeredi /* 4190bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4191bc27027aSMiklos Szeredi * we'll need to flip '..'. 4192bc27027aSMiklos Szeredi */ 4193da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4194da1ce067SMiklos Szeredi if (is_dir) { 4195bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4196bc27027aSMiklos Szeredi if (error) 4197bc27027aSMiklos Szeredi return error; 4198bc27027aSMiklos Szeredi } 4199da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4200da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4201da1ce067SMiklos Szeredi if (error) 4202da1ce067SMiklos Szeredi return error; 4203da1ce067SMiklos Szeredi } 4204da1ce067SMiklos Szeredi } 42050eeca283SRobert Love 42060b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42070b3974ebSMiklos Szeredi flags); 4208bc27027aSMiklos Szeredi if (error) 4209bc27027aSMiklos Szeredi return error; 4210bc27027aSMiklos Szeredi 4211bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4212bc27027aSMiklos Szeredi dget(new_dentry); 4213da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4214bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4215bc27027aSMiklos Szeredi else if (target) 4216bc27027aSMiklos Szeredi mutex_lock(&target->i_mutex); 4217bc27027aSMiklos Szeredi 4218bc27027aSMiklos Szeredi error = -EBUSY; 42197af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4220bc27027aSMiklos Szeredi goto out; 4221bc27027aSMiklos Szeredi 4222da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4223bc27027aSMiklos Szeredi error = -EMLINK; 4224da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4225bc27027aSMiklos Szeredi goto out; 4226da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4227da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4228da1ce067SMiklos Szeredi goto out; 4229da1ce067SMiklos Szeredi } 4230da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4231bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4232da1ce067SMiklos Szeredi if (!is_dir) { 4233bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4234bc27027aSMiklos Szeredi if (error) 4235bc27027aSMiklos Szeredi goto out; 4236da1ce067SMiklos Szeredi } 4237da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4238bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4239bc27027aSMiklos Szeredi if (error) 4240bc27027aSMiklos Szeredi goto out; 4241bc27027aSMiklos Szeredi } 42427177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4243520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4244520c8b16SMiklos Szeredi new_dir, new_dentry); 4245520c8b16SMiklos Szeredi } else { 42467177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4247520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4248520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4249520c8b16SMiklos Szeredi } 4250bc27027aSMiklos Szeredi if (error) 4251bc27027aSMiklos Szeredi goto out; 4252bc27027aSMiklos Szeredi 4253da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 42541da177e4SLinus Torvalds if (is_dir) 4255bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4256bc27027aSMiklos Szeredi dont_mount(new_dentry); 42578ed936b5SEric W. Biederman detach_mounts(new_dentry); 4258bc27027aSMiklos Szeredi } 4259da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4260da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4261bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4262da1ce067SMiklos Szeredi else 4263da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4264da1ce067SMiklos Szeredi } 4265bc27027aSMiklos Szeredi out: 4266da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4267bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4268bc27027aSMiklos Szeredi else if (target) 4269bc27027aSMiklos Szeredi mutex_unlock(&target->i_mutex); 4270bc27027aSMiklos Szeredi dput(new_dentry); 4271da1ce067SMiklos Szeredi if (!error) { 4272123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4273da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4274da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4275da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4276da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4277da1ce067SMiklos Szeredi } 4278da1ce067SMiklos Szeredi } 42790eeca283SRobert Love fsnotify_oldname_free(old_name); 42800eeca283SRobert Love 42811da177e4SLinus Torvalds return error; 42821da177e4SLinus Torvalds } 42834d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 42841da177e4SLinus Torvalds 4285520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4286520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 42871da177e4SLinus Torvalds { 42881da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 42891da177e4SLinus Torvalds struct dentry *trap; 4290f5beed75SAl Viro struct path old_path, new_path; 4291f5beed75SAl Viro struct qstr old_last, new_last; 4292f5beed75SAl Viro int old_type, new_type; 42938e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 429491a27b2aSJeff Layton struct filename *from; 429591a27b2aSJeff Layton struct filename *to; 4296f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4297c6a94284SJeff Layton bool should_retry = false; 42982ad94ae6SAl Viro int error; 4299520c8b16SMiklos Szeredi 43000d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4301da1ce067SMiklos Szeredi return -EINVAL; 4302da1ce067SMiklos Szeredi 43030d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43040d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4305520c8b16SMiklos Szeredi return -EINVAL; 4306520c8b16SMiklos Szeredi 43070d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43080d7a8555SMiklos Szeredi return -EPERM; 43090d7a8555SMiklos Szeredi 4310f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4311f5beed75SAl Viro target_flags = 0; 4312f5beed75SAl Viro 4313c6a94284SJeff Layton retry: 4314f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4315f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 431691a27b2aSJeff Layton if (IS_ERR(from)) { 431791a27b2aSJeff Layton error = PTR_ERR(from); 43181da177e4SLinus Torvalds goto exit; 431991a27b2aSJeff Layton } 43201da177e4SLinus Torvalds 4321f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4322f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 432391a27b2aSJeff Layton if (IS_ERR(to)) { 432491a27b2aSJeff Layton error = PTR_ERR(to); 43251da177e4SLinus Torvalds goto exit1; 432691a27b2aSJeff Layton } 43271da177e4SLinus Torvalds 43281da177e4SLinus Torvalds error = -EXDEV; 4329f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 43301da177e4SLinus Torvalds goto exit2; 43311da177e4SLinus Torvalds 43321da177e4SLinus Torvalds error = -EBUSY; 4333f5beed75SAl Viro if (old_type != LAST_NORM) 43341da177e4SLinus Torvalds goto exit2; 43351da177e4SLinus Torvalds 43360a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 43370a7c3937SMiklos Szeredi error = -EEXIST; 4338f5beed75SAl Viro if (new_type != LAST_NORM) 43391da177e4SLinus Torvalds goto exit2; 43401da177e4SLinus Torvalds 4341f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4342c30dabfeSJan Kara if (error) 4343c30dabfeSJan Kara goto exit2; 4344c30dabfeSJan Kara 43458e6d782cSJ. Bruce Fields retry_deleg: 4346f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 43471da177e4SLinus Torvalds 4348f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 43491da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 43501da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 43511da177e4SLinus Torvalds goto exit3; 43521da177e4SLinus Torvalds /* source must exist */ 43531da177e4SLinus Torvalds error = -ENOENT; 4354b18825a7SDavid Howells if (d_is_negative(old_dentry)) 43551da177e4SLinus Torvalds goto exit4; 4356f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 43571da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43581da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 43591da177e4SLinus Torvalds goto exit4; 43600a7c3937SMiklos Szeredi error = -EEXIST; 43610a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 43620a7c3937SMiklos Szeredi goto exit5; 4363da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4364da1ce067SMiklos Szeredi error = -ENOENT; 4365da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4366da1ce067SMiklos Szeredi goto exit5; 4367da1ce067SMiklos Szeredi 4368da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4369da1ce067SMiklos Szeredi error = -ENOTDIR; 4370f5beed75SAl Viro if (new_last.name[new_last.len]) 4371da1ce067SMiklos Szeredi goto exit5; 4372da1ce067SMiklos Szeredi } 4373da1ce067SMiklos Szeredi } 43740a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 43750a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 43760a7c3937SMiklos Szeredi error = -ENOTDIR; 4377f5beed75SAl Viro if (old_last.name[old_last.len]) 43780a7c3937SMiklos Szeredi goto exit5; 4379f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 43800a7c3937SMiklos Szeredi goto exit5; 43810a7c3937SMiklos Szeredi } 43820a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 43830a7c3937SMiklos Szeredi error = -EINVAL; 43840a7c3937SMiklos Szeredi if (old_dentry == trap) 43850a7c3937SMiklos Szeredi goto exit5; 43861da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4387da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43881da177e4SLinus Torvalds error = -ENOTEMPTY; 43891da177e4SLinus Torvalds if (new_dentry == trap) 43901da177e4SLinus Torvalds goto exit5; 43911da177e4SLinus Torvalds 4392f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4393f5beed75SAl Viro &new_path, new_dentry, flags); 4394be6d3e56SKentaro Takeda if (error) 4395c30dabfeSJan Kara goto exit5; 4396f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4397f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4398520c8b16SMiklos Szeredi &delegated_inode, flags); 43991da177e4SLinus Torvalds exit5: 44001da177e4SLinus Torvalds dput(new_dentry); 44011da177e4SLinus Torvalds exit4: 44021da177e4SLinus Torvalds dput(old_dentry); 44031da177e4SLinus Torvalds exit3: 4404f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44058e6d782cSJ. Bruce Fields if (delegated_inode) { 44068e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44078e6d782cSJ. Bruce Fields if (!error) 44088e6d782cSJ. Bruce Fields goto retry_deleg; 44098e6d782cSJ. Bruce Fields } 4410f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44111da177e4SLinus Torvalds exit2: 4412c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4413c6a94284SJeff Layton should_retry = true; 4414f5beed75SAl Viro path_put(&new_path); 44152ad94ae6SAl Viro putname(to); 44161da177e4SLinus Torvalds exit1: 4417f5beed75SAl Viro path_put(&old_path); 44181da177e4SLinus Torvalds putname(from); 4419c6a94284SJeff Layton if (should_retry) { 4420c6a94284SJeff Layton should_retry = false; 4421c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4422c6a94284SJeff Layton goto retry; 4423c6a94284SJeff Layton } 44242ad94ae6SAl Viro exit: 44251da177e4SLinus Torvalds return error; 44261da177e4SLinus Torvalds } 44271da177e4SLinus Torvalds 4428520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4429520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4430520c8b16SMiklos Szeredi { 4431520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4432520c8b16SMiklos Szeredi } 4433520c8b16SMiklos Szeredi 4434a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 44355590ff0dSUlrich Drepper { 4436520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44375590ff0dSUlrich Drepper } 44385590ff0dSUlrich Drepper 4439787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4440787fb6bcSMiklos Szeredi { 4441787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4442787fb6bcSMiklos Szeredi if (error) 4443787fb6bcSMiklos Szeredi return error; 4444787fb6bcSMiklos Szeredi 4445787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4446787fb6bcSMiklos Szeredi return -EPERM; 4447787fb6bcSMiklos Szeredi 4448787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4449787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4450787fb6bcSMiklos Szeredi } 4451787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4452787fb6bcSMiklos Szeredi 44535d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 44541da177e4SLinus Torvalds { 44555d826c84SAl Viro int len = PTR_ERR(link); 44561da177e4SLinus Torvalds if (IS_ERR(link)) 44571da177e4SLinus Torvalds goto out; 44581da177e4SLinus Torvalds 44591da177e4SLinus Torvalds len = strlen(link); 44601da177e4SLinus Torvalds if (len > (unsigned) buflen) 44611da177e4SLinus Torvalds len = buflen; 44621da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 44631da177e4SLinus Torvalds len = -EFAULT; 44641da177e4SLinus Torvalds out: 44651da177e4SLinus Torvalds return len; 44661da177e4SLinus Torvalds } 44675d826c84SAl Viro EXPORT_SYMBOL(readlink_copy); 44681da177e4SLinus Torvalds 44691da177e4SLinus Torvalds /* 44701da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 44711da177e4SLinus Torvalds * have ->follow_link() touching nd only in nd_set_link(). Using (or not 44721da177e4SLinus Torvalds * using) it for any given inode is up to filesystem. 44731da177e4SLinus Torvalds */ 44741da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 44751da177e4SLinus Torvalds { 4476cc314eefSLinus Torvalds void *cookie; 4477d4dee48bSAl Viro const char *link = dentry->d_inode->i_link; 4478694a1764SMarcin Slusarz int res; 4479cc314eefSLinus Torvalds 4480d4dee48bSAl Viro if (!link) { 4481d4dee48bSAl Viro link = dentry->d_inode->i_op->follow_link(dentry, &cookie, NULL); 4482680baacbSAl Viro if (IS_ERR(link)) 4483680baacbSAl Viro return PTR_ERR(link); 4484d4dee48bSAl Viro } 4485680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4486680baacbSAl Viro if (cookie && dentry->d_inode->i_op->put_link) 4487680baacbSAl Viro dentry->d_inode->i_op->put_link(dentry, cookie); 4488694a1764SMarcin Slusarz return res; 44891da177e4SLinus Torvalds } 44904d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 44911da177e4SLinus Torvalds 44921da177e4SLinus Torvalds /* get the link contents into pagecache */ 44931da177e4SLinus Torvalds static char *page_getlink(struct dentry * dentry, struct page **ppage) 44941da177e4SLinus Torvalds { 4495ebd09abbSDuane Griffin char *kaddr; 44961da177e4SLinus Torvalds struct page *page; 44971da177e4SLinus Torvalds struct address_space *mapping = dentry->d_inode->i_mapping; 4498090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 44991da177e4SLinus Torvalds if (IS_ERR(page)) 45006fe6900eSNick Piggin return (char*)page; 45011da177e4SLinus Torvalds *ppage = page; 4502ebd09abbSDuane Griffin kaddr = kmap(page); 4503ebd09abbSDuane Griffin nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1); 4504ebd09abbSDuane Griffin return kaddr; 45051da177e4SLinus Torvalds } 45061da177e4SLinus Torvalds 45071da177e4SLinus Torvalds int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 45081da177e4SLinus Torvalds { 45091da177e4SLinus Torvalds struct page *page = NULL; 45105d826c84SAl Viro int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page)); 45111da177e4SLinus Torvalds if (page) { 45121da177e4SLinus Torvalds kunmap(page); 45131da177e4SLinus Torvalds page_cache_release(page); 45141da177e4SLinus Torvalds } 45151da177e4SLinus Torvalds return res; 45161da177e4SLinus Torvalds } 45174d359507SAl Viro EXPORT_SYMBOL(page_readlink); 45181da177e4SLinus Torvalds 4519680baacbSAl Viro const char *page_follow_link_light(struct dentry *dentry, void **cookie, struct nameidata *nd) 45201da177e4SLinus Torvalds { 4521cc314eefSLinus Torvalds struct page *page = NULL; 4522680baacbSAl Viro char *res = page_getlink(dentry, &page); 4523680baacbSAl Viro if (!IS_ERR(res)) 4524680baacbSAl Viro *cookie = page; 4525680baacbSAl Viro return res; 45261da177e4SLinus Torvalds } 45274d359507SAl Viro EXPORT_SYMBOL(page_follow_link_light); 45281da177e4SLinus Torvalds 4529680baacbSAl Viro void page_put_link(struct dentry *dentry, void *cookie) 45301da177e4SLinus Torvalds { 4531cc314eefSLinus Torvalds struct page *page = cookie; 45321da177e4SLinus Torvalds kunmap(page); 45331da177e4SLinus Torvalds page_cache_release(page); 45341da177e4SLinus Torvalds } 45354d359507SAl Viro EXPORT_SYMBOL(page_put_link); 45361da177e4SLinus Torvalds 453754566b2cSNick Piggin /* 453854566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 453954566b2cSNick Piggin */ 454054566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 45411da177e4SLinus Torvalds { 45421da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 45430adb25d2SKirill Korotaev struct page *page; 4544afddba49SNick Piggin void *fsdata; 4545beb497abSDmitriy Monakhov int err; 45461da177e4SLinus Torvalds char *kaddr; 454754566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 454854566b2cSNick Piggin if (nofs) 454954566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 45501da177e4SLinus Torvalds 45517e53cac4SNeilBrown retry: 4552afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 455354566b2cSNick Piggin flags, &page, &fsdata); 45541da177e4SLinus Torvalds if (err) 4555afddba49SNick Piggin goto fail; 4556afddba49SNick Piggin 4557e8e3c3d6SCong Wang kaddr = kmap_atomic(page); 45581da177e4SLinus Torvalds memcpy(kaddr, symname, len-1); 4559e8e3c3d6SCong Wang kunmap_atomic(kaddr); 4560afddba49SNick Piggin 4561afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4562afddba49SNick Piggin page, fsdata); 45631da177e4SLinus Torvalds if (err < 0) 45641da177e4SLinus Torvalds goto fail; 4565afddba49SNick Piggin if (err < len-1) 4566afddba49SNick Piggin goto retry; 4567afddba49SNick Piggin 45681da177e4SLinus Torvalds mark_inode_dirty(inode); 45691da177e4SLinus Torvalds return 0; 45701da177e4SLinus Torvalds fail: 45711da177e4SLinus Torvalds return err; 45721da177e4SLinus Torvalds } 45734d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 45741da177e4SLinus Torvalds 45750adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 45760adb25d2SKirill Korotaev { 45770adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 457854566b2cSNick Piggin !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS)); 45790adb25d2SKirill Korotaev } 45804d359507SAl Viro EXPORT_SYMBOL(page_symlink); 45810adb25d2SKirill Korotaev 458292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 45831da177e4SLinus Torvalds .readlink = generic_readlink, 45841da177e4SLinus Torvalds .follow_link = page_follow_link_light, 45851da177e4SLinus Torvalds .put_link = page_put_link, 45861da177e4SLinus Torvalds }; 45871da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4588