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> 382a18da7aSGeorge Spelvin #include <linux/bitops.h> 391da177e4SLinus Torvalds #include <asm/uaccess.h> 401da177e4SLinus Torvalds 41e81e3f4dSEric Paris #include "internal.h" 42c7105365SAl Viro #include "mount.h" 43e81e3f4dSEric Paris 441da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 451da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 461da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 471da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 481da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 491da177e4SLinus Torvalds * 501da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 511da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 521da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 531da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 541da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 551da177e4SLinus Torvalds * the special cases of the former code. 561da177e4SLinus Torvalds * 571da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 581da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 591da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 601da177e4SLinus Torvalds * 611da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 621da177e4SLinus Torvalds * resolution to correspond with current state of the code. 631da177e4SLinus Torvalds * 641da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 651da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 661da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 671da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 681da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 691da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 701da177e4SLinus Torvalds */ 711da177e4SLinus Torvalds 721da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 731da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 741da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 751da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 761da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 771da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 7825985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 791da177e4SLinus Torvalds * 801da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 811da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 821da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 831da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 841da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 851da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 861da177e4SLinus Torvalds * and in the old Linux semantics. 871da177e4SLinus Torvalds */ 881da177e4SLinus Torvalds 891da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 901da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 911da177e4SLinus Torvalds * 921da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 931da177e4SLinus Torvalds */ 941da177e4SLinus Torvalds 951da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 961da177e4SLinus Torvalds * inside the path - always follow. 971da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 981da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 991da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1001da177e4SLinus Torvalds * otherwise - don't follow. 1011da177e4SLinus Torvalds * (applied in that order). 1021da177e4SLinus Torvalds * 1031da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1041da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1051da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1061da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1071da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1081da177e4SLinus Torvalds */ 1091da177e4SLinus Torvalds /* 1101da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 111a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1121da177e4SLinus Torvalds * any extra contention... 1131da177e4SLinus Torvalds */ 1141da177e4SLinus Torvalds 1151da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1161da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1171da177e4SLinus Torvalds * kernel data space before using them.. 1181da177e4SLinus Torvalds * 1191da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1201da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1211da177e4SLinus Torvalds */ 1227950e385SJeff Layton 123fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12491a27b2aSJeff Layton 12551f39a1fSDavid Drysdale struct filename * 12691a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12791a27b2aSJeff Layton { 12894b5d262SAl Viro struct filename *result; 1297950e385SJeff Layton char *kname; 13094b5d262SAl Viro int len; 1311da177e4SLinus Torvalds 1327ac86265SJeff Layton result = audit_reusename(filename); 1337ac86265SJeff Layton if (result) 1347ac86265SJeff Layton return result; 1357ac86265SJeff Layton 1367950e385SJeff Layton result = __getname(); 1373f9f0aa6SLinus Torvalds if (unlikely(!result)) 1384043cde8SEric Paris return ERR_PTR(-ENOMEM); 1391da177e4SLinus Torvalds 1407950e385SJeff Layton /* 1417950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1427950e385SJeff Layton * allocation 1437950e385SJeff Layton */ 144fd2f7cb5SAl Viro kname = (char *)result->iname; 14591a27b2aSJeff Layton result->name = kname; 1467950e385SJeff Layton 14794b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 14891a27b2aSJeff Layton if (unlikely(len < 0)) { 14994b5d262SAl Viro __putname(result); 15094b5d262SAl Viro return ERR_PTR(len); 15191a27b2aSJeff Layton } 1523f9f0aa6SLinus Torvalds 1537950e385SJeff Layton /* 1547950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1557950e385SJeff Layton * separate struct filename so we can dedicate the entire 1567950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1577950e385SJeff Layton * userland. 1587950e385SJeff Layton */ 15994b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 160fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1617950e385SJeff Layton kname = (char *)result; 1627950e385SJeff Layton 163fd2f7cb5SAl Viro /* 164fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 165fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 166fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 167fd2f7cb5SAl Viro */ 168fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 16994b5d262SAl Viro if (unlikely(!result)) { 17094b5d262SAl Viro __putname(kname); 17194b5d262SAl Viro return ERR_PTR(-ENOMEM); 1727950e385SJeff Layton } 1737950e385SJeff Layton result->name = kname; 17494b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17594b5d262SAl Viro if (unlikely(len < 0)) { 17694b5d262SAl Viro __putname(kname); 17794b5d262SAl Viro kfree(result); 17894b5d262SAl Viro return ERR_PTR(len); 17994b5d262SAl Viro } 18094b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18194b5d262SAl Viro __putname(kname); 18294b5d262SAl Viro kfree(result); 18394b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18494b5d262SAl Viro } 1857950e385SJeff Layton } 1867950e385SJeff Layton 18794b5d262SAl Viro result->refcnt = 1; 1883f9f0aa6SLinus Torvalds /* The empty path is special. */ 1893f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1903f9f0aa6SLinus Torvalds if (empty) 1911fa1e7f6SAndy Whitcroft *empty = 1; 19294b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19394b5d262SAl Viro putname(result); 19494b5d262SAl Viro return ERR_PTR(-ENOENT); 1951da177e4SLinus Torvalds } 19694b5d262SAl Viro } 1977950e385SJeff Layton 1987950e385SJeff Layton result->uptr = filename; 199c4ad8f98SLinus Torvalds result->aname = NULL; 2001da177e4SLinus Torvalds audit_getname(result); 2011da177e4SLinus Torvalds return result; 2023f9f0aa6SLinus Torvalds } 2033f9f0aa6SLinus Torvalds 20491a27b2aSJeff Layton struct filename * 20591a27b2aSJeff Layton getname(const char __user * filename) 206f52e0c11SAl Viro { 207f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 208f52e0c11SAl Viro } 209f52e0c11SAl Viro 210c4ad8f98SLinus Torvalds struct filename * 211c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 212c4ad8f98SLinus Torvalds { 213c4ad8f98SLinus Torvalds struct filename *result; 21408518549SPaul Moore int len = strlen(filename) + 1; 215c4ad8f98SLinus Torvalds 216c4ad8f98SLinus Torvalds result = __getname(); 217c4ad8f98SLinus Torvalds if (unlikely(!result)) 218c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 219c4ad8f98SLinus Torvalds 22008518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 221fd2f7cb5SAl Viro result->name = (char *)result->iname; 22208518549SPaul Moore } else if (len <= PATH_MAX) { 22308518549SPaul Moore struct filename *tmp; 22408518549SPaul Moore 22508518549SPaul Moore tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 22608518549SPaul Moore if (unlikely(!tmp)) { 22708518549SPaul Moore __putname(result); 22808518549SPaul Moore return ERR_PTR(-ENOMEM); 22908518549SPaul Moore } 23008518549SPaul Moore tmp->name = (char *)result; 23108518549SPaul Moore result = tmp; 23208518549SPaul Moore } else { 23308518549SPaul Moore __putname(result); 23408518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23508518549SPaul Moore } 23608518549SPaul Moore memcpy((char *)result->name, filename, len); 237c4ad8f98SLinus Torvalds result->uptr = NULL; 238c4ad8f98SLinus Torvalds result->aname = NULL; 23955422d0bSPaul Moore result->refcnt = 1; 240fd3522fdSPaul Moore audit_getname(result); 241c4ad8f98SLinus Torvalds 242c4ad8f98SLinus Torvalds return result; 243c4ad8f98SLinus Torvalds } 244c4ad8f98SLinus Torvalds 24591a27b2aSJeff Layton void putname(struct filename *name) 2461da177e4SLinus Torvalds { 24755422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 24855422d0bSPaul Moore 24955422d0bSPaul Moore if (--name->refcnt > 0) 25055422d0bSPaul Moore return; 25155422d0bSPaul Moore 252fd2f7cb5SAl Viro if (name->name != name->iname) { 25355422d0bSPaul Moore __putname(name->name); 25455422d0bSPaul Moore kfree(name); 25555422d0bSPaul Moore } else 25655422d0bSPaul Moore __putname(name); 2571da177e4SLinus Torvalds } 2581da177e4SLinus Torvalds 259e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 260e77819e5SLinus Torvalds { 26184635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 262e77819e5SLinus Torvalds struct posix_acl *acl; 263e77819e5SLinus Torvalds 264e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2653567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2663567866bSAl Viro if (!acl) 267e77819e5SLinus Torvalds return -EAGAIN; 2683567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 269b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 270e77819e5SLinus Torvalds return -ECHILD; 271206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 272e77819e5SLinus Torvalds } 273e77819e5SLinus Torvalds 2742982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2754e34e719SChristoph Hellwig if (IS_ERR(acl)) 2764e34e719SChristoph Hellwig return PTR_ERR(acl); 277e77819e5SLinus Torvalds if (acl) { 278e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 279e77819e5SLinus Torvalds posix_acl_release(acl); 280e77819e5SLinus Torvalds return error; 281e77819e5SLinus Torvalds } 28284635d68SLinus Torvalds #endif 283e77819e5SLinus Torvalds 284e77819e5SLinus Torvalds return -EAGAIN; 285e77819e5SLinus Torvalds } 286e77819e5SLinus Torvalds 2875909ccaaSLinus Torvalds /* 288948409c7SAndreas Gruenbacher * This does the basic permission checking 2895909ccaaSLinus Torvalds */ 2907e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2915909ccaaSLinus Torvalds { 29226cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2935909ccaaSLinus Torvalds 2948e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2955909ccaaSLinus Torvalds mode >>= 6; 2965909ccaaSLinus Torvalds else { 297e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 2987e40145eSAl Viro int error = check_acl(inode, mask); 2995909ccaaSLinus Torvalds if (error != -EAGAIN) 3005909ccaaSLinus Torvalds return error; 3015909ccaaSLinus Torvalds } 3025909ccaaSLinus Torvalds 3035909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3045909ccaaSLinus Torvalds mode >>= 3; 3055909ccaaSLinus Torvalds } 3065909ccaaSLinus Torvalds 3075909ccaaSLinus Torvalds /* 3085909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3095909ccaaSLinus Torvalds */ 3109c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3115909ccaaSLinus Torvalds return 0; 3125909ccaaSLinus Torvalds return -EACCES; 3135909ccaaSLinus Torvalds } 3141da177e4SLinus Torvalds 3151da177e4SLinus Torvalds /** 3161da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3171da177e4SLinus Torvalds * @inode: inode to check access rights for 3188fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3191da177e4SLinus Torvalds * 3201da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3211da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3221da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 323b74c79e9SNick Piggin * are used for other things. 324b74c79e9SNick Piggin * 325b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 326b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 327b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3281da177e4SLinus Torvalds */ 3292830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3301da177e4SLinus Torvalds { 3315909ccaaSLinus Torvalds int ret; 3321da177e4SLinus Torvalds 3331da177e4SLinus Torvalds /* 334948409c7SAndreas Gruenbacher * Do the basic permission checks. 3351da177e4SLinus Torvalds */ 3367e40145eSAl Viro ret = acl_permission_check(inode, mask); 3375909ccaaSLinus Torvalds if (ret != -EACCES) 3385909ccaaSLinus Torvalds return ret; 3391da177e4SLinus Torvalds 340d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 341d594e7ecSAl Viro /* DACs are overridable for directories */ 34223adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 343d594e7ecSAl Viro return 0; 344d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34523adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34623adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 347d594e7ecSAl Viro return 0; 348d594e7ecSAl Viro return -EACCES; 349d594e7ecSAl Viro } 3501da177e4SLinus Torvalds /* 3511da177e4SLinus Torvalds * Read/write DACs are always overridable. 352d594e7ecSAl Viro * Executable DACs are overridable when there is 353d594e7ecSAl Viro * at least one exec bit set. 3541da177e4SLinus Torvalds */ 355d594e7ecSAl Viro if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 35623adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3571da177e4SLinus Torvalds return 0; 3581da177e4SLinus Torvalds 3591da177e4SLinus Torvalds /* 3601da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3611da177e4SLinus Torvalds */ 3627ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 363d594e7ecSAl Viro if (mask == MAY_READ) 36423adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3651da177e4SLinus Torvalds return 0; 3661da177e4SLinus Torvalds 3671da177e4SLinus Torvalds return -EACCES; 3681da177e4SLinus Torvalds } 3694d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3701da177e4SLinus Torvalds 3713ddcd056SLinus Torvalds /* 3723ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3733ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3743ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3753ddcd056SLinus Torvalds * permission function, use the fast case". 3763ddcd056SLinus Torvalds */ 3773ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3783ddcd056SLinus Torvalds { 3793ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3803ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3813ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3823ddcd056SLinus Torvalds 3833ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3843ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3853ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3863ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3873ddcd056SLinus Torvalds } 3883ddcd056SLinus Torvalds return generic_permission(inode, mask); 3893ddcd056SLinus Torvalds } 3903ddcd056SLinus Torvalds 391cb23beb5SChristoph Hellwig /** 3920bdaea90SDavid Howells * __inode_permission - Check for access rights to a given inode 3930bdaea90SDavid Howells * @inode: Inode to check permission on 3940bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 395cb23beb5SChristoph Hellwig * 3960bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. 397948409c7SAndreas Gruenbacher * 398948409c7SAndreas Gruenbacher * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 3990bdaea90SDavid Howells * 4000bdaea90SDavid Howells * This does not check for a read-only file system. You probably want 4010bdaea90SDavid Howells * inode_permission(). 402cb23beb5SChristoph Hellwig */ 4030bdaea90SDavid Howells int __inode_permission(struct inode *inode, int mask) 4041da177e4SLinus Torvalds { 405e6305c43SAl Viro int retval; 4061da177e4SLinus Torvalds 4073ddcd056SLinus Torvalds if (unlikely(mask & MAY_WRITE)) { 4081da177e4SLinus Torvalds /* 4091da177e4SLinus Torvalds * Nobody gets write access to an immutable file. 4101da177e4SLinus Torvalds */ 4111da177e4SLinus Torvalds if (IS_IMMUTABLE(inode)) 4121da177e4SLinus Torvalds return -EACCES; 4130bd23d09SEric W. Biederman 4140bd23d09SEric W. Biederman /* 4150bd23d09SEric W. Biederman * Updating mtime will likely cause i_uid and i_gid to be 4160bd23d09SEric W. Biederman * written back improperly if their true value is unknown 4170bd23d09SEric W. Biederman * to the vfs. 4180bd23d09SEric W. Biederman */ 4190bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 4200bd23d09SEric W. Biederman return -EACCES; 4211da177e4SLinus Torvalds } 4221da177e4SLinus Torvalds 4233ddcd056SLinus Torvalds retval = do_inode_permission(inode, mask); 4241da177e4SLinus Torvalds if (retval) 4251da177e4SLinus Torvalds return retval; 4261da177e4SLinus Torvalds 42708ce5f16SSerge E. Hallyn retval = devcgroup_inode_permission(inode, mask); 42808ce5f16SSerge E. Hallyn if (retval) 42908ce5f16SSerge E. Hallyn return retval; 43008ce5f16SSerge E. Hallyn 431d09ca739SEric Paris return security_inode_permission(inode, mask); 4321da177e4SLinus Torvalds } 433bd5d0856SMiklos Szeredi EXPORT_SYMBOL(__inode_permission); 4341da177e4SLinus Torvalds 435f4d6ff89SAl Viro /** 4360bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4370bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 43855852635SRandy Dunlap * @inode: Inode to check permission on 4390bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4400bdaea90SDavid Howells * 4410bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4420bdaea90SDavid Howells */ 4430bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4440bdaea90SDavid Howells { 4450bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4460bdaea90SDavid Howells umode_t mode = inode->i_mode; 4470bdaea90SDavid Howells 4480bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 4490bdaea90SDavid Howells if ((sb->s_flags & MS_RDONLY) && 4500bdaea90SDavid Howells (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4510bdaea90SDavid Howells return -EROFS; 4520bdaea90SDavid Howells } 4530bdaea90SDavid Howells return 0; 4540bdaea90SDavid Howells } 4550bdaea90SDavid Howells 4560bdaea90SDavid Howells /** 4570bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4580bdaea90SDavid Howells * @inode: Inode to check permission on 4590bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4600bdaea90SDavid Howells * 4610bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4620bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4630bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4640bdaea90SDavid Howells * 4650bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4660bdaea90SDavid Howells */ 4670bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4680bdaea90SDavid Howells { 4690bdaea90SDavid Howells int retval; 4700bdaea90SDavid Howells 4710bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4720bdaea90SDavid Howells if (retval) 4730bdaea90SDavid Howells return retval; 4740bdaea90SDavid Howells return __inode_permission(inode, mask); 4750bdaea90SDavid Howells } 4764d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4770bdaea90SDavid Howells 4780bdaea90SDavid Howells /** 4795dd784d0SJan Blunck * path_get - get a reference to a path 4805dd784d0SJan Blunck * @path: path to get the reference to 4815dd784d0SJan Blunck * 4825dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4835dd784d0SJan Blunck */ 484dcf787f3SAl Viro void path_get(const struct path *path) 4855dd784d0SJan Blunck { 4865dd784d0SJan Blunck mntget(path->mnt); 4875dd784d0SJan Blunck dget(path->dentry); 4885dd784d0SJan Blunck } 4895dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4905dd784d0SJan Blunck 4915dd784d0SJan Blunck /** 4921d957f9bSJan Blunck * path_put - put a reference to a path 4931d957f9bSJan Blunck * @path: path to put the reference to 4941d957f9bSJan Blunck * 4951d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4961d957f9bSJan Blunck */ 497dcf787f3SAl Viro void path_put(const struct path *path) 4981da177e4SLinus Torvalds { 4991d957f9bSJan Blunck dput(path->dentry); 5001d957f9bSJan Blunck mntput(path->mnt); 5011da177e4SLinus Torvalds } 5021d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5031da177e4SLinus Torvalds 504894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5051f55a6ecSAl Viro struct nameidata { 5061f55a6ecSAl Viro struct path path; 5071f55a6ecSAl Viro struct qstr last; 5081f55a6ecSAl Viro struct path root; 5091f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 5101f55a6ecSAl Viro unsigned int flags; 5119883d185SAl Viro unsigned seq, m_seq; 5121f55a6ecSAl Viro int last_type; 5131f55a6ecSAl Viro unsigned depth; 514756daf26SNeilBrown int total_link_count; 515697fc6caSAl Viro struct saved { 516697fc6caSAl Viro struct path link; 517fceef393SAl Viro struct delayed_call done; 518697fc6caSAl Viro const char *name; 5190450b2d1SAl Viro unsigned seq; 520894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5219883d185SAl Viro struct filename *name; 5229883d185SAl Viro struct nameidata *saved; 523fceef393SAl Viro struct inode *link_inode; 5249883d185SAl Viro unsigned root_seq; 5259883d185SAl Viro int dfd; 5261f55a6ecSAl Viro }; 5271f55a6ecSAl Viro 5289883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 529894bc8c4SAl Viro { 530756daf26SNeilBrown struct nameidata *old = current->nameidata; 531756daf26SNeilBrown p->stack = p->internal; 532c8a53ee5SAl Viro p->dfd = dfd; 533c8a53ee5SAl Viro p->name = name; 534756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5359883d185SAl Viro p->saved = old; 536756daf26SNeilBrown current->nameidata = p; 537894bc8c4SAl Viro } 538894bc8c4SAl Viro 5399883d185SAl Viro static void restore_nameidata(void) 540894bc8c4SAl Viro { 5419883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 542756daf26SNeilBrown 543756daf26SNeilBrown current->nameidata = old; 544756daf26SNeilBrown if (old) 545756daf26SNeilBrown old->total_link_count = now->total_link_count; 546e1a63bbcSAl Viro if (now->stack != now->internal) 547756daf26SNeilBrown kfree(now->stack); 548894bc8c4SAl Viro } 549894bc8c4SAl Viro 550894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 551894bc8c4SAl Viro { 552bc40aee0SAl Viro struct saved *p; 553bc40aee0SAl Viro 554bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 555bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 556bc40aee0SAl Viro GFP_ATOMIC); 557bc40aee0SAl Viro if (unlikely(!p)) 558bc40aee0SAl Viro return -ECHILD; 559bc40aee0SAl Viro } else { 560bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 561894bc8c4SAl Viro GFP_KERNEL); 562894bc8c4SAl Viro if (unlikely(!p)) 563894bc8c4SAl Viro return -ENOMEM; 564bc40aee0SAl Viro } 565894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 566894bc8c4SAl Viro nd->stack = p; 567894bc8c4SAl Viro return 0; 568894bc8c4SAl Viro } 569894bc8c4SAl Viro 570397d425dSEric W. Biederman /** 571397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 572397d425dSEric W. Biederman * @path: nameidate to verify 573397d425dSEric W. Biederman * 574397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 575397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 576397d425dSEric W. Biederman */ 577397d425dSEric W. Biederman static bool path_connected(const struct path *path) 578397d425dSEric W. Biederman { 579397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 580397d425dSEric W. Biederman 581397d425dSEric W. Biederman /* Only bind mounts can have disconnected paths */ 582397d425dSEric W. Biederman if (mnt->mnt_root == mnt->mnt_sb->s_root) 583397d425dSEric W. Biederman return true; 584397d425dSEric W. Biederman 585397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 586397d425dSEric W. Biederman } 587397d425dSEric W. Biederman 588894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 589894bc8c4SAl Viro { 590da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 591894bc8c4SAl Viro return 0; 592894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 593894bc8c4SAl Viro return 0; 594894bc8c4SAl Viro return __nd_alloc_stack(nd); 595894bc8c4SAl Viro } 596894bc8c4SAl Viro 5977973387aSAl Viro static void drop_links(struct nameidata *nd) 5987973387aSAl Viro { 5997973387aSAl Viro int i = nd->depth; 6007973387aSAl Viro while (i--) { 6017973387aSAl Viro struct saved *last = nd->stack + i; 602fceef393SAl Viro do_delayed_call(&last->done); 603fceef393SAl Viro clear_delayed_call(&last->done); 6047973387aSAl Viro } 6057973387aSAl Viro } 6067973387aSAl Viro 6077973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6087973387aSAl Viro { 6097973387aSAl Viro drop_links(nd); 6107973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6117973387aSAl Viro int i; 6127973387aSAl Viro path_put(&nd->path); 6137973387aSAl Viro for (i = 0; i < nd->depth; i++) 6147973387aSAl Viro path_put(&nd->stack[i].link); 615102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 616102b8af2SAl Viro path_put(&nd->root); 617102b8af2SAl Viro nd->root.mnt = NULL; 618102b8af2SAl Viro } 6197973387aSAl Viro } else { 6207973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6217973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6227973387aSAl Viro nd->root.mnt = NULL; 6237973387aSAl Viro rcu_read_unlock(); 6247973387aSAl Viro } 6257973387aSAl Viro nd->depth = 0; 6267973387aSAl Viro } 6277973387aSAl Viro 6287973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6297973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6307973387aSAl Viro struct path *path, unsigned seq) 6317973387aSAl Viro { 6327973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6337973387aSAl Viro if (unlikely(res)) { 6347973387aSAl Viro if (res > 0) 6357973387aSAl Viro path->mnt = NULL; 6367973387aSAl Viro path->dentry = NULL; 6377973387aSAl Viro return false; 6387973387aSAl Viro } 6397973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6407973387aSAl Viro path->dentry = NULL; 6417973387aSAl Viro return false; 6427973387aSAl Viro } 6437973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6447973387aSAl Viro } 6457973387aSAl Viro 6467973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6477973387aSAl Viro { 6487973387aSAl Viro int i; 6497973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6507973387aSAl Viro struct saved *last = nd->stack + i; 6517973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6527973387aSAl Viro drop_links(nd); 6537973387aSAl Viro nd->depth = i + 1; 6547973387aSAl Viro return false; 6557973387aSAl Viro } 6567973387aSAl Viro } 6577973387aSAl Viro return true; 6587973387aSAl Viro } 6597973387aSAl Viro 66019660af7SAl Viro /* 66131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 66219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 66319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 66457e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66931e6b01fSNick Piggin */ 67031e6b01fSNick Piggin 67131e6b01fSNick Piggin /** 67219660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 67319660af7SAl Viro * @nd: nameidata pathwalk data 67419660af7SAl Viro * @dentry: child of nd->path.dentry or NULL 6756e9918b7SAl Viro * @seq: seq number to check dentry against 67639191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67731e6b01fSNick Piggin * 67819660af7SAl Viro * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry 67919660af7SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 68019660af7SAl Viro * @nd or NULL. Must be called from rcu-walk context. 6817973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6827973387aSAl Viro * terminate_walk(). 68331e6b01fSNick Piggin */ 6846e9918b7SAl Viro static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq) 68531e6b01fSNick Piggin { 68631e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68731e6b01fSNick Piggin 68831e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 689e5c832d5SLinus Torvalds 690e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6917973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6927973387aSAl Viro goto out2; 6937973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 6947973387aSAl Viro goto out2; 6957973387aSAl Viro if (unlikely(!lockref_get_not_dead(&parent->d_lockref))) 6967973387aSAl Viro goto out1; 69748a066e7SAl Viro 69815570086SLinus Torvalds /* 69915570086SLinus Torvalds * For a negative lookup, the lookup sequence point is the parents 70015570086SLinus Torvalds * sequence point, and it only needs to revalidate the parent dentry. 70115570086SLinus Torvalds * 70215570086SLinus Torvalds * For a positive lookup, we need to move both the parent and the 70315570086SLinus Torvalds * dentry from the RCU domain to be properly refcounted. And the 70415570086SLinus Torvalds * sequence number in the dentry validates *both* dentry counters, 70515570086SLinus Torvalds * since we checked the sequence number of the parent after we got 70615570086SLinus Torvalds * the child sequence number. So we know the parent must still 70715570086SLinus Torvalds * be valid if the child sequence number is still valid. 70815570086SLinus Torvalds */ 70919660af7SAl Viro if (!dentry) { 710e5c832d5SLinus Torvalds if (read_seqcount_retry(&parent->d_seq, nd->seq)) 711e5c832d5SLinus Torvalds goto out; 71219660af7SAl Viro BUG_ON(nd->inode != parent->d_inode); 71319660af7SAl Viro } else { 714e5c832d5SLinus Torvalds if (!lockref_get_not_dead(&dentry->d_lockref)) 715e5c832d5SLinus Torvalds goto out; 7166e9918b7SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 717e5c832d5SLinus Torvalds goto drop_dentry; 71819660af7SAl Viro } 719e5c832d5SLinus Torvalds 720e5c832d5SLinus Torvalds /* 721e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 722e5c832d5SLinus Torvalds * still valid and get it if required. 723e5c832d5SLinus Torvalds */ 724e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7255a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7265a8d87e8SAl Viro rcu_read_unlock(); 7275a8d87e8SAl Viro dput(dentry); 7285a8d87e8SAl Viro return -ECHILD; 7297973387aSAl Viro } 73031e6b01fSNick Piggin } 73131e6b01fSNick Piggin 7328b61e74fSAl Viro rcu_read_unlock(); 73331e6b01fSNick Piggin return 0; 73419660af7SAl Viro 735e5c832d5SLinus Torvalds drop_dentry: 7368b61e74fSAl Viro rcu_read_unlock(); 737e5c832d5SLinus Torvalds dput(dentry); 738d0d27277SLinus Torvalds goto drop_root_mnt; 7397973387aSAl Viro out2: 7407973387aSAl Viro nd->path.mnt = NULL; 7417973387aSAl Viro out1: 7427973387aSAl Viro nd->path.dentry = NULL; 743e5c832d5SLinus Torvalds out: 7448b61e74fSAl Viro rcu_read_unlock(); 745d0d27277SLinus Torvalds drop_root_mnt: 746d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 747d0d27277SLinus Torvalds nd->root.mnt = NULL; 74831e6b01fSNick Piggin return -ECHILD; 74931e6b01fSNick Piggin } 75031e6b01fSNick Piggin 7517973387aSAl Viro static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq) 7527973387aSAl Viro { 7537973387aSAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 7547973387aSAl Viro drop_links(nd); 7557973387aSAl Viro nd->depth = 0; 7567973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 7577973387aSAl Viro nd->path.mnt = NULL; 7587973387aSAl Viro nd->path.dentry = NULL; 7597973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7607973387aSAl Viro nd->root.mnt = NULL; 7617973387aSAl Viro rcu_read_unlock(); 7627973387aSAl Viro } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) { 7637973387aSAl Viro return 0; 7647973387aSAl Viro } 7657973387aSAl Viro path_put(link); 7667973387aSAl Viro return -ECHILD; 7677973387aSAl Viro } 7687973387aSAl Viro 7694ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 77034286d66SNick Piggin { 7714ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 77234286d66SNick Piggin } 77334286d66SNick Piggin 7749f1fafeeSAl Viro /** 7759f1fafeeSAl Viro * complete_walk - successful completion of path walk 7769f1fafeeSAl Viro * @nd: pointer nameidata 77739159de2SJeff Layton * 7789f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7799f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7809f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7819f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7829f1fafeeSAl Viro * need to drop nd->path. 78339159de2SJeff Layton */ 7849f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78539159de2SJeff Layton { 78616c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78739159de2SJeff Layton int status; 78839159de2SJeff Layton 7899f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7909f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7919f1fafeeSAl Viro nd->root.mnt = NULL; 7926e9918b7SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 79348a066e7SAl Viro return -ECHILD; 79448a066e7SAl Viro } 7959f1fafeeSAl Viro 79616c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 79739159de2SJeff Layton return 0; 79839159de2SJeff Layton 799ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 80016c2cd71SAl Viro return 0; 80116c2cd71SAl Viro 802ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 80339159de2SJeff Layton if (status > 0) 80439159de2SJeff Layton return 0; 80539159de2SJeff Layton 80616c2cd71SAl Viro if (!status) 80739159de2SJeff Layton status = -ESTALE; 80816c2cd71SAl Viro 80939159de2SJeff Layton return status; 81039159de2SJeff Layton } 81139159de2SJeff Layton 81218d8c860SAl Viro static void set_root(struct nameidata *nd) 8132a737871SAl Viro { 81431e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8159e6697e2SAl Viro 8169e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8178f47a016SAl Viro unsigned seq; 818c28cc364SNick Piggin 819c28cc364SNick Piggin do { 820c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 82131e6b01fSNick Piggin nd->root = fs->root; 8228f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 823c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8249e6697e2SAl Viro } else { 8259e6697e2SAl Viro get_fs_root(fs, &nd->root); 8269e6697e2SAl Viro } 82731e6b01fSNick Piggin } 82831e6b01fSNick Piggin 8291d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 830051d3812SIan Kent { 831051d3812SIan Kent dput(path->dentry); 8324ac91378SJan Blunck if (path->mnt != nd->path.mnt) 833051d3812SIan Kent mntput(path->mnt); 834051d3812SIan Kent } 835051d3812SIan Kent 8367b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8377b9337aaSNick Piggin struct nameidata *nd) 838051d3812SIan Kent { 83931e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8404ac91378SJan Blunck dput(nd->path.dentry); 84131e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8424ac91378SJan Blunck mntput(nd->path.mnt); 8439a229683SHuang Shijie } 84431e6b01fSNick Piggin nd->path.mnt = path->mnt; 8454ac91378SJan Blunck nd->path.dentry = path->dentry; 846051d3812SIan Kent } 847051d3812SIan Kent 848248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 849248fb5b9SAl Viro { 850248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 851248fb5b9SAl Viro struct dentry *d; 852248fb5b9SAl Viro nd->path = nd->root; 853248fb5b9SAl Viro d = nd->path.dentry; 854248fb5b9SAl Viro nd->inode = d->d_inode; 855248fb5b9SAl Viro nd->seq = nd->root_seq; 856248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 857248fb5b9SAl Viro return -ECHILD; 858248fb5b9SAl Viro } else { 859248fb5b9SAl Viro path_put(&nd->path); 860248fb5b9SAl Viro nd->path = nd->root; 861248fb5b9SAl Viro path_get(&nd->path); 862248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 863248fb5b9SAl Viro } 864248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 865248fb5b9SAl Viro return 0; 866248fb5b9SAl Viro } 867248fb5b9SAl Viro 868b5fb63c1SChristoph Hellwig /* 8696b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 870b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 871b5fb63c1SChristoph Hellwig */ 8726e77137bSAl Viro void nd_jump_link(struct path *path) 873b5fb63c1SChristoph Hellwig { 8746e77137bSAl Viro struct nameidata *nd = current->nameidata; 875b5fb63c1SChristoph Hellwig path_put(&nd->path); 876b5fb63c1SChristoph Hellwig 877b5fb63c1SChristoph Hellwig nd->path = *path; 878b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 879b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 880b5fb63c1SChristoph Hellwig } 881b5fb63c1SChristoph Hellwig 882b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 883574197e0SAl Viro { 88421c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 885fceef393SAl Viro do_delayed_call(&last->done); 8866548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 887b9ff4429SAl Viro path_put(&last->link); 888574197e0SAl Viro } 889574197e0SAl Viro 890561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 891561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 892800179c9SKees Cook 893800179c9SKees Cook /** 894800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 89555852635SRandy Dunlap * @nd: nameidata pathwalk data 896800179c9SKees Cook * 897800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 898800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 899800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 900800179c9SKees Cook * processes from failing races against path names that may change out 901800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 902800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 903800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 904800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 905800179c9SKees Cook * 906800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 907800179c9SKees Cook */ 908fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 909800179c9SKees Cook { 910800179c9SKees Cook const struct inode *inode; 911800179c9SKees Cook const struct inode *parent; 9122d7f9e2aSSeth Forshee kuid_t puid; 913800179c9SKees Cook 914800179c9SKees Cook if (!sysctl_protected_symlinks) 915800179c9SKees Cook return 0; 916800179c9SKees Cook 917800179c9SKees Cook /* Allowed if owner and follower match. */ 918fceef393SAl Viro inode = nd->link_inode; 91981abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 920800179c9SKees Cook return 0; 921800179c9SKees Cook 922800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 923aa65fa35SAl Viro parent = nd->inode; 924800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 925800179c9SKees Cook return 0; 926800179c9SKees Cook 927800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9282d7f9e2aSSeth Forshee puid = parent->i_uid; 9292d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 930800179c9SKees Cook return 0; 931800179c9SKees Cook 93231956502SAl Viro if (nd->flags & LOOKUP_RCU) 93331956502SAl Viro return -ECHILD; 93431956502SAl Viro 9351cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 936800179c9SKees Cook return -EACCES; 937800179c9SKees Cook } 938800179c9SKees Cook 939800179c9SKees Cook /** 940800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 941800179c9SKees Cook * @inode: the source inode to hardlink from 942800179c9SKees Cook * 943800179c9SKees Cook * Return false if at least one of the following conditions: 944800179c9SKees Cook * - inode is not a regular file 945800179c9SKees Cook * - inode is setuid 946800179c9SKees Cook * - inode is setgid and group-exec 947800179c9SKees Cook * - access failure for read and write 948800179c9SKees Cook * 949800179c9SKees Cook * Otherwise returns true. 950800179c9SKees Cook */ 951800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 952800179c9SKees Cook { 953800179c9SKees Cook umode_t mode = inode->i_mode; 954800179c9SKees Cook 955800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 956800179c9SKees Cook if (!S_ISREG(mode)) 957800179c9SKees Cook return false; 958800179c9SKees Cook 959800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 960800179c9SKees Cook if (mode & S_ISUID) 961800179c9SKees Cook return false; 962800179c9SKees Cook 963800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 964800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 965800179c9SKees Cook return false; 966800179c9SKees Cook 967800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 968800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 969800179c9SKees Cook return false; 970800179c9SKees Cook 971800179c9SKees Cook return true; 972800179c9SKees Cook } 973800179c9SKees Cook 974800179c9SKees Cook /** 975800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 976800179c9SKees Cook * @link: the source to hardlink from 977800179c9SKees Cook * 978800179c9SKees Cook * Block hardlink when all of: 979800179c9SKees Cook * - sysctl_protected_hardlinks enabled 980800179c9SKees Cook * - fsuid does not match inode 981800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 982f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 983800179c9SKees Cook * 984800179c9SKees Cook * Returns 0 if successful, -ve on error. 985800179c9SKees Cook */ 986800179c9SKees Cook static int may_linkat(struct path *link) 987800179c9SKees Cook { 988800179c9SKees Cook struct inode *inode; 989800179c9SKees Cook 990800179c9SKees Cook if (!sysctl_protected_hardlinks) 991800179c9SKees Cook return 0; 992800179c9SKees Cook 993800179c9SKees Cook inode = link->dentry->d_inode; 994800179c9SKees Cook 995800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 996800179c9SKees Cook * otherwise, it must be a safe source. 997800179c9SKees Cook */ 998f2ca3796SDirk Steinmetz if (inode_owner_or_capable(inode) || safe_hardlink_source(inode)) 999800179c9SKees Cook return 0; 1000800179c9SKees Cook 1001a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 1002800179c9SKees Cook return -EPERM; 1003800179c9SKees Cook } 1004800179c9SKees Cook 10053b2e7f75SAl Viro static __always_inline 10063b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10071da177e4SLinus Torvalds { 1008ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10091cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1010fceef393SAl Viro struct inode *inode = nd->link_inode; 10116d7b5aaeSAl Viro int error; 10120a959df5SAl Viro const char *res; 10131da177e4SLinus Torvalds 10148fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10158fa9dd24SNeilBrown touch_atime(&last->link); 10168fa9dd24SNeilBrown cond_resched(); 10178fa9dd24SNeilBrown } else if (atime_needs_update(&last->link, inode)) { 10188fa9dd24SNeilBrown if (unlikely(unlazy_walk(nd, NULL, 0))) 10198fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10208fa9dd24SNeilBrown touch_atime(&last->link); 10218fa9dd24SNeilBrown } 10228fa9dd24SNeilBrown 1023bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1024bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1025bda0be7aSNeilBrown if (unlikely(error)) 10266920a440SAl Viro return ERR_PTR(error); 102736f3b4f6SAl Viro 102886acdca1SAl Viro nd->last_type = LAST_BIND; 1029d4dee48bSAl Viro res = inode->i_link; 1030d4dee48bSAl Viro if (!res) { 1031fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1032fceef393SAl Viro struct delayed_call *); 1033fceef393SAl Viro get = inode->i_op->get_link; 10348c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1035fceef393SAl Viro res = get(NULL, inode, &last->done); 10366b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10378c1b4566SAl Viro if (unlikely(unlazy_walk(nd, NULL, 0))) 10388c1b4566SAl Viro return ERR_PTR(-ECHILD); 1039fceef393SAl Viro res = get(dentry, inode, &last->done); 10408c1b4566SAl Viro } 10416b255391SAl Viro } else { 1042fceef393SAl Viro res = get(dentry, inode, &last->done); 10436b255391SAl Viro } 1044fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1045fab51e8aSAl Viro return res; 10460a959df5SAl Viro } 1047fab51e8aSAl Viro if (*res == '/') { 1048fab51e8aSAl Viro if (!nd->root.mnt) 1049fab51e8aSAl Viro set_root(nd); 1050248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1051fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1052fab51e8aSAl Viro while (unlikely(*++res == '/')) 1053fab51e8aSAl Viro ; 1054fab51e8aSAl Viro } 1055fab51e8aSAl Viro if (!*res) 1056fab51e8aSAl Viro res = NULL; 10570a959df5SAl Viro return res; 10580a959df5SAl Viro } 10596d7b5aaeSAl Viro 1060f015f126SDavid Howells /* 1061f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1062f015f126SDavid Howells * 1063f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1064f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1065f015f126SDavid Howells * Up is towards /. 1066f015f126SDavid Howells * 1067f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1068f015f126SDavid Howells * root. 1069f015f126SDavid Howells */ 1070bab77ebfSAl Viro int follow_up(struct path *path) 10711da177e4SLinus Torvalds { 10720714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10730714a533SAl Viro struct mount *parent; 10741da177e4SLinus Torvalds struct dentry *mountpoint; 107599b7db7bSNick Piggin 107648a066e7SAl Viro read_seqlock_excl(&mount_lock); 10770714a533SAl Viro parent = mnt->mnt_parent; 10783c0a6163SAl Viro if (parent == mnt) { 107948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10801da177e4SLinus Torvalds return 0; 10811da177e4SLinus Torvalds } 10820714a533SAl Viro mntget(&parent->mnt); 1083a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 108448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1085bab77ebfSAl Viro dput(path->dentry); 1086bab77ebfSAl Viro path->dentry = mountpoint; 1087bab77ebfSAl Viro mntput(path->mnt); 10880714a533SAl Viro path->mnt = &parent->mnt; 10891da177e4SLinus Torvalds return 1; 10901da177e4SLinus Torvalds } 10914d359507SAl Viro EXPORT_SYMBOL(follow_up); 10921da177e4SLinus Torvalds 1093b5c84bf6SNick Piggin /* 10949875cf80SDavid Howells * Perform an automount 10959875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10969875cf80SDavid Howells * were called with. 10971da177e4SLinus Torvalds */ 1098756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10999875cf80SDavid Howells bool *need_mntput) 110031e6b01fSNick Piggin { 11019875cf80SDavid Howells struct vfsmount *mnt; 1102ea5b778aSDavid Howells int err; 11039875cf80SDavid Howells 11049875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 11059875cf80SDavid Howells return -EREMOTE; 11069875cf80SDavid Howells 11070ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11080ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11090ec26fd0SMiklos Szeredi * the name. 11100ec26fd0SMiklos Szeredi * 11110ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11125a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11130ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11140ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11150ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11160ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11175a30d8a2SDavid Howells */ 1118756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 1119d94c177bSLinus Torvalds LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11205a30d8a2SDavid Howells path->dentry->d_inode) 11219875cf80SDavid Howells return -EISDIR; 11220ec26fd0SMiklos Szeredi 1123756daf26SNeilBrown nd->total_link_count++; 1124756daf26SNeilBrown if (nd->total_link_count >= 40) 11259875cf80SDavid Howells return -ELOOP; 11269875cf80SDavid Howells 11279875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11289875cf80SDavid Howells if (IS_ERR(mnt)) { 11299875cf80SDavid Howells /* 11309875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11319875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11329875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11339875cf80SDavid Howells * 11349875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11359875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11369875cf80SDavid Howells * the path is inaccessible and we should say so. 11379875cf80SDavid Howells */ 1138756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11399875cf80SDavid Howells return -EREMOTE; 11409875cf80SDavid Howells return PTR_ERR(mnt); 114131e6b01fSNick Piggin } 1142ea5b778aSDavid Howells 11439875cf80SDavid Howells if (!mnt) /* mount collision */ 11449875cf80SDavid Howells return 0; 11459875cf80SDavid Howells 11468aef1884SAl Viro if (!*need_mntput) { 11478aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11488aef1884SAl Viro mntget(path->mnt); 11498aef1884SAl Viro *need_mntput = true; 11508aef1884SAl Viro } 115119a167afSAl Viro err = finish_automount(mnt, path); 1152ea5b778aSDavid Howells 1153ea5b778aSDavid Howells switch (err) { 1154ea5b778aSDavid Howells case -EBUSY: 1155ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 115619a167afSAl Viro return 0; 1157ea5b778aSDavid Howells case 0: 11588aef1884SAl Viro path_put(path); 11599875cf80SDavid Howells path->mnt = mnt; 11609875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11619875cf80SDavid Howells return 0; 116219a167afSAl Viro default: 116319a167afSAl Viro return err; 11649875cf80SDavid Howells } 116519a167afSAl Viro 1166ea5b778aSDavid Howells } 11679875cf80SDavid Howells 11689875cf80SDavid Howells /* 11699875cf80SDavid Howells * Handle a dentry that is managed in some way. 1170cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11719875cf80SDavid Howells * - Flagged as mountpoint 11729875cf80SDavid Howells * - Flagged as automount point 11739875cf80SDavid Howells * 11749875cf80SDavid Howells * This may only be called in refwalk mode. 11759875cf80SDavid Howells * 11769875cf80SDavid Howells * Serialization is taken care of in namespace.c 11779875cf80SDavid Howells */ 1178756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11799875cf80SDavid Howells { 11808aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11819875cf80SDavid Howells unsigned managed; 11829875cf80SDavid Howells bool need_mntput = false; 11838aef1884SAl Viro int ret = 0; 11849875cf80SDavid Howells 11859875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11869875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11879875cf80SDavid Howells * the components of that value change under us */ 11889875cf80SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 11899875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11909875cf80SDavid Howells unlikely(managed != 0)) { 1191cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1192cc53ce53SDavid Howells * being held. */ 1193cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1194cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1195cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 11961aed3e42SAl Viro ret = path->dentry->d_op->d_manage(path->dentry, false); 1197cc53ce53SDavid Howells if (ret < 0) 11988aef1884SAl Viro break; 1199cc53ce53SDavid Howells } 1200cc53ce53SDavid Howells 12019875cf80SDavid Howells /* Transit to a mounted filesystem. */ 12029875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 12039875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12049875cf80SDavid Howells if (mounted) { 12059875cf80SDavid Howells dput(path->dentry); 12069875cf80SDavid Howells if (need_mntput) 1207463ffb2eSAl Viro mntput(path->mnt); 1208463ffb2eSAl Viro path->mnt = mounted; 1209463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12109875cf80SDavid Howells need_mntput = true; 12119875cf80SDavid Howells continue; 1212463ffb2eSAl Viro } 1213463ffb2eSAl Viro 12149875cf80SDavid Howells /* Something is mounted on this dentry in another 12159875cf80SDavid Howells * namespace and/or whatever was mounted there in this 121648a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 121748a066e7SAl Viro * get it */ 12181da177e4SLinus Torvalds } 12199875cf80SDavid Howells 12209875cf80SDavid Howells /* Handle an automount point */ 12219875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1222756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12239875cf80SDavid Howells if (ret < 0) 12248aef1884SAl Viro break; 12259875cf80SDavid Howells continue; 12269875cf80SDavid Howells } 12279875cf80SDavid Howells 12289875cf80SDavid Howells /* We didn't change the current path point */ 12299875cf80SDavid Howells break; 12309875cf80SDavid Howells } 12318aef1884SAl Viro 12328aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12338aef1884SAl Viro mntput(path->mnt); 1234e9742b53SAl Viro if (ret == -EISDIR || !ret) 1235e9742b53SAl Viro ret = 1; 12368402752eSAl Viro if (need_mntput) 12378402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12388402752eSAl Viro if (unlikely(ret < 0)) 12398402752eSAl Viro path_put_conditional(path, nd); 12408402752eSAl Viro return ret; 12411da177e4SLinus Torvalds } 12421da177e4SLinus Torvalds 1243cc53ce53SDavid Howells int follow_down_one(struct path *path) 12441da177e4SLinus Torvalds { 12451da177e4SLinus Torvalds struct vfsmount *mounted; 12461da177e4SLinus Torvalds 12471c755af4SAl Viro mounted = lookup_mnt(path); 12481da177e4SLinus Torvalds if (mounted) { 12499393bd07SAl Viro dput(path->dentry); 12509393bd07SAl Viro mntput(path->mnt); 12519393bd07SAl Viro path->mnt = mounted; 12529393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12531da177e4SLinus Torvalds return 1; 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds return 0; 12561da177e4SLinus Torvalds } 12574d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12581da177e4SLinus Torvalds 1259b8faf035SNeilBrown static inline int managed_dentry_rcu(struct dentry *dentry) 126062a7375eSIan Kent { 1261b8faf035SNeilBrown return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1262b8faf035SNeilBrown dentry->d_op->d_manage(dentry, true) : 0; 126362a7375eSIan Kent } 126462a7375eSIan Kent 12659875cf80SDavid Howells /* 1266287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1267287548e4SAl Viro * we meet a managed dentry that would need blocking. 12689875cf80SDavid Howells */ 12699875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1270254cf582SAl Viro struct inode **inode, unsigned *seqp) 12719875cf80SDavid Howells { 127262a7375eSIan Kent for (;;) { 1273c7105365SAl Viro struct mount *mounted; 127462a7375eSIan Kent /* 127562a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 127662a7375eSIan Kent * that wants to block transit. 127762a7375eSIan Kent */ 1278b8faf035SNeilBrown switch (managed_dentry_rcu(path->dentry)) { 1279b8faf035SNeilBrown case -ECHILD: 1280b8faf035SNeilBrown default: 1281ab90911fSDavid Howells return false; 1282b8faf035SNeilBrown case -EISDIR: 1283b8faf035SNeilBrown return true; 1284b8faf035SNeilBrown case 0: 1285b8faf035SNeilBrown break; 1286b8faf035SNeilBrown } 128762a7375eSIan Kent 128862a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1289b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 129062a7375eSIan Kent 1291474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12929875cf80SDavid Howells if (!mounted) 12939875cf80SDavid Howells break; 1294c7105365SAl Viro path->mnt = &mounted->mnt; 1295c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1296a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1297254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 129859430262SLinus Torvalds /* 129959430262SLinus Torvalds * Update the inode too. We don't need to re-check the 130059430262SLinus Torvalds * dentry sequence number here after this d_inode read, 130159430262SLinus Torvalds * because a mount-point is always pinned. 130259430262SLinus Torvalds */ 130359430262SLinus Torvalds *inode = path->dentry->d_inode; 13049875cf80SDavid Howells } 1305f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1306b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1307287548e4SAl Viro } 1308287548e4SAl Viro 130931e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 131031e6b01fSNick Piggin { 13114023bfc9SAl Viro struct inode *inode = nd->inode; 131231e6b01fSNick Piggin 131331e6b01fSNick Piggin while (1) { 1314aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 131531e6b01fSNick Piggin break; 131631e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 131731e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 131831e6b01fSNick Piggin struct dentry *parent = old->d_parent; 131931e6b01fSNick Piggin unsigned seq; 132031e6b01fSNick Piggin 13214023bfc9SAl Viro inode = parent->d_inode; 132231e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1323aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1324aed434adSAl Viro return -ECHILD; 132531e6b01fSNick Piggin nd->path.dentry = parent; 132631e6b01fSNick Piggin nd->seq = seq; 1327397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1328397d425dSEric W. Biederman return -ENOENT; 132931e6b01fSNick Piggin break; 1330aed434adSAl Viro } else { 1331aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1332aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1333aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1334aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1335aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1336aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1337aed434adSAl Viro return -ECHILD; 1338aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 133931e6b01fSNick Piggin break; 1340aed434adSAl Viro /* we know that mountpoint was pinned */ 1341aed434adSAl Viro nd->path.dentry = mountpoint; 1342aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1343aed434adSAl Viro inode = inode2; 1344aed434adSAl Viro nd->seq = seq; 134531e6b01fSNick Piggin } 1346aed434adSAl Viro } 1347aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1348b37199e6SAl Viro struct mount *mounted; 1349b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1350aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1351aed434adSAl Viro return -ECHILD; 1352b37199e6SAl Viro if (!mounted) 1353b37199e6SAl Viro break; 1354b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1355b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13564023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1357b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1358b37199e6SAl Viro } 13594023bfc9SAl Viro nd->inode = inode; 136031e6b01fSNick Piggin return 0; 136131e6b01fSNick Piggin } 136231e6b01fSNick Piggin 13639875cf80SDavid Howells /* 1364cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1365cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1366cc53ce53SDavid Howells * caller is permitted to proceed or not. 1367cc53ce53SDavid Howells */ 13687cc90cc3SAl Viro int follow_down(struct path *path) 1369cc53ce53SDavid Howells { 1370cc53ce53SDavid Howells unsigned managed; 1371cc53ce53SDavid Howells int ret; 1372cc53ce53SDavid Howells 1373cc53ce53SDavid Howells while (managed = ACCESS_ONCE(path->dentry->d_flags), 1374cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1375cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1376cc53ce53SDavid Howells * being held. 1377cc53ce53SDavid Howells * 1378cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1379cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1380cc53ce53SDavid Howells * other than its daemon the right to mount on its 1381cc53ce53SDavid Howells * superstructure. 1382cc53ce53SDavid Howells * 1383cc53ce53SDavid Howells * The filesystem may sleep at this point. 1384cc53ce53SDavid Howells */ 1385cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1386cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1387cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1388ab90911fSDavid Howells ret = path->dentry->d_op->d_manage( 13891aed3e42SAl Viro path->dentry, false); 1390cc53ce53SDavid Howells if (ret < 0) 1391cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1392cc53ce53SDavid Howells } 1393cc53ce53SDavid Howells 1394cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1395cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1396cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1397cc53ce53SDavid Howells if (!mounted) 1398cc53ce53SDavid Howells break; 1399cc53ce53SDavid Howells dput(path->dentry); 1400cc53ce53SDavid Howells mntput(path->mnt); 1401cc53ce53SDavid Howells path->mnt = mounted; 1402cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1403cc53ce53SDavid Howells continue; 1404cc53ce53SDavid Howells } 1405cc53ce53SDavid Howells 1406cc53ce53SDavid Howells /* Don't handle automount points here */ 1407cc53ce53SDavid Howells break; 1408cc53ce53SDavid Howells } 1409cc53ce53SDavid Howells return 0; 1410cc53ce53SDavid Howells } 14114d359507SAl Viro EXPORT_SYMBOL(follow_down); 1412cc53ce53SDavid Howells 1413cc53ce53SDavid Howells /* 14149875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14159875cf80SDavid Howells */ 14169875cf80SDavid Howells static void follow_mount(struct path *path) 14179875cf80SDavid Howells { 14189875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14199875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14209875cf80SDavid Howells if (!mounted) 14219875cf80SDavid Howells break; 14229875cf80SDavid Howells dput(path->dentry); 14239875cf80SDavid Howells mntput(path->mnt); 14249875cf80SDavid Howells path->mnt = mounted; 14259875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14269875cf80SDavid Howells } 14279875cf80SDavid Howells } 14289875cf80SDavid Howells 1429eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1430eedf265aSEric W. Biederman { 1431eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1432eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1433eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1434eedf265aSEric W. Biederman dput(old); 1435eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1436eedf265aSEric W. Biederman return -ENOENT; 1437eedf265aSEric W. Biederman return 0; 1438eedf265aSEric W. Biederman } 1439eedf265aSEric W. Biederman 1440397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14411da177e4SLinus Torvalds { 14421da177e4SLinus Torvalds while(1) { 14432a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14442a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14451da177e4SLinus Torvalds break; 14461da177e4SLinus Torvalds } 14474ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1448eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1449eedf265aSEric W. Biederman if (ret) 1450eedf265aSEric W. Biederman return ret; 14511da177e4SLinus Torvalds break; 14521da177e4SLinus Torvalds } 14533088dd70SAl Viro if (!follow_up(&nd->path)) 14541da177e4SLinus Torvalds break; 14551da177e4SLinus Torvalds } 145679ed0226SAl Viro follow_mount(&nd->path); 145731e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1458397d425dSEric W. Biederman return 0; 14591da177e4SLinus Torvalds } 14601da177e4SLinus Torvalds 14611da177e4SLinus Torvalds /* 1462bad61189SMiklos Szeredi * This looks up the name in dcache, possibly revalidates the old dentry and 1463bad61189SMiklos Szeredi * allocates a new one if not found or not valid. In the need_lookup argument 1464bad61189SMiklos Szeredi * returns whether i_op->lookup is necessary. 1465baa03890SNick Piggin */ 1466e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1467e3c13928SAl Viro struct dentry *dir, 14686c51e513SAl Viro unsigned int flags) 1469baa03890SNick Piggin { 1470baa03890SNick Piggin struct dentry *dentry; 1471bad61189SMiklos Szeredi int error; 1472baa03890SNick Piggin 1473bad61189SMiklos Szeredi dentry = d_lookup(dir, name); 1474bad61189SMiklos Szeredi if (dentry) { 147539e3c955SJeff Layton if (dentry->d_flags & DCACHE_OP_REVALIDATE) { 1476201f956eSAl Viro error = d_revalidate(dentry, flags); 1477bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 147874ff0ffcSAl Viro if (!error) 14795542aa2fSEric W. Biederman d_invalidate(dentry); 1480bad61189SMiklos Szeredi dput(dentry); 148174ff0ffcSAl Viro return ERR_PTR(error); 1482bad61189SMiklos Szeredi } 1483bad61189SMiklos Szeredi } 1484bad61189SMiklos Szeredi } 1485baa03890SNick Piggin return dentry; 1486baa03890SNick Piggin } 1487baa03890SNick Piggin 1488baa03890SNick Piggin /* 148913a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 149013a2c3beSJ. Bruce Fields * unhashed. 1491bad61189SMiklos Szeredi * 1492bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 149344396f4bSJosef Bacik */ 1494bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 149572bd866aSAl Viro unsigned int flags) 149644396f4bSJosef Bacik { 149744396f4bSJosef Bacik struct dentry *old; 149844396f4bSJosef Bacik 149944396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1500bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1501e188dc02SMiklos Szeredi dput(dentry); 150244396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1503e188dc02SMiklos Szeredi } 150444396f4bSJosef Bacik 150572bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 150644396f4bSJosef Bacik if (unlikely(old)) { 150744396f4bSJosef Bacik dput(dentry); 150844396f4bSJosef Bacik dentry = old; 150944396f4bSJosef Bacik } 151044396f4bSJosef Bacik return dentry; 151144396f4bSJosef Bacik } 151244396f4bSJosef Bacik 1513e3c13928SAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 151472bd866aSAl Viro struct dentry *base, unsigned int flags) 1515a3255546SAl Viro { 15166c51e513SAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 1517a3255546SAl Viro 15186c51e513SAl Viro if (dentry) 1519a3255546SAl Viro return dentry; 1520bad61189SMiklos Szeredi 15216c51e513SAl Viro dentry = d_alloc(base, name); 15226c51e513SAl Viro if (unlikely(!dentry)) 15236c51e513SAl Viro return ERR_PTR(-ENOMEM); 15246c51e513SAl Viro 152572bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1526a3255546SAl Viro } 1527a3255546SAl Viro 1528e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1529254cf582SAl Viro struct path *path, struct inode **inode, 1530254cf582SAl Viro unsigned *seqp) 15311da177e4SLinus Torvalds { 15324ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 153331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15345a18fff2SAl Viro int status = 1; 15359875cf80SDavid Howells int err; 15369875cf80SDavid Howells 15373cac260aSAl Viro /* 1538b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15395d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15405d0f49c1SAl Viro * going to fall back to non-racy lookup. 1541b04f784eSNick Piggin */ 154231e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 154331e6b01fSNick Piggin unsigned seq; 1544766c4cbfSAl Viro bool negative; 1545da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15465d0f49c1SAl Viro if (unlikely(!dentry)) { 15475d0f49c1SAl Viro if (unlazy_walk(nd, NULL, 0)) 15485d0f49c1SAl Viro return -ECHILD; 1549e9742b53SAl Viro return 0; 15505d0f49c1SAl Viro } 15515a18fff2SAl Viro 155212f8ad4bSLinus Torvalds /* 155312f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 155412f8ad4bSLinus Torvalds * the dentry name information from lookup. 155512f8ad4bSLinus Torvalds */ 155663afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1557766c4cbfSAl Viro negative = d_is_negative(dentry); 15585d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 155912f8ad4bSLinus Torvalds return -ECHILD; 156012f8ad4bSLinus Torvalds 156112f8ad4bSLinus Torvalds /* 156212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 156312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 156412f8ad4bSLinus Torvalds * 156512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 156612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 156712f8ad4bSLinus Torvalds */ 15685d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 156931e6b01fSNick Piggin return -ECHILD; 15705a18fff2SAl Viro 1571254cf582SAl Viro *seqp = seq; 15725d0f49c1SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 15734ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15745a18fff2SAl Viro if (unlikely(status <= 0)) { 15755d0f49c1SAl Viro if (unlazy_walk(nd, dentry, seq)) 15765d0f49c1SAl Viro return -ECHILD; 15775d0f49c1SAl Viro if (status == -ECHILD) 15785d0f49c1SAl Viro status = d_revalidate(dentry, nd->flags); 15795d0f49c1SAl Viro } else { 1580daf3761cSTrond Myklebust /* 1581daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1582daf3761cSTrond Myklebust * case that drops it. 1583daf3761cSTrond Myklebust */ 15845d0f49c1SAl Viro if (unlikely(negative)) 1585daf3761cSTrond Myklebust return -ENOENT; 158631e6b01fSNick Piggin path->mnt = mnt; 158731e6b01fSNick Piggin path->dentry = dentry; 1588254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1589e9742b53SAl Viro return 1; 1590254cf582SAl Viro if (unlazy_walk(nd, dentry, seq)) 15915a18fff2SAl Viro return -ECHILD; 15925d0f49c1SAl Viro } 15935a18fff2SAl Viro } else { 1594e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 159581e6f520SAl Viro if (unlikely(!dentry)) 1596e9742b53SAl Viro return 0; 15975d0f49c1SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 15984ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15995d0f49c1SAl Viro } 16005a18fff2SAl Viro if (unlikely(status <= 0)) { 1601e9742b53SAl Viro if (!status) 16025d0f49c1SAl Viro d_invalidate(dentry); 16035a18fff2SAl Viro dput(dentry); 16045a18fff2SAl Viro return status; 16055a18fff2SAl Viro } 1606766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1607766c4cbfSAl Viro dput(dentry); 1608766c4cbfSAl Viro return -ENOENT; 1609766c4cbfSAl Viro } 16105d0f49c1SAl Viro 16111da177e4SLinus Torvalds path->mnt = mnt; 16121da177e4SLinus Torvalds path->dentry = dentry; 1613756daf26SNeilBrown err = follow_managed(path, nd); 1614e9742b53SAl Viro if (likely(err > 0)) 161563afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16168402752eSAl Viro return err; 1617697f514dSMiklos Szeredi } 1618697f514dSMiklos Szeredi 1619697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1620e3c13928SAl Viro static struct dentry *lookup_slow(const struct qstr *name, 1621e3c13928SAl Viro struct dentry *dir, 1622e3c13928SAl Viro unsigned int flags) 1623697f514dSMiklos Szeredi { 162494bdd655SAl Viro struct dentry *dentry = ERR_PTR(-ENOENT), *old; 16251936386eSAl Viro struct inode *inode = dir->d_inode; 1626d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16271936386eSAl Viro 16289902af79SAl Viro inode_lock_shared(inode); 16291936386eSAl Viro /* Don't go there if it's already dead */ 163094bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 163194bdd655SAl Viro goto out; 163294bdd655SAl Viro again: 1633d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 163494bdd655SAl Viro if (IS_ERR(dentry)) 163594bdd655SAl Viro goto out; 163694bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1637949a852eSAl Viro if ((dentry->d_flags & DCACHE_OP_REVALIDATE) && 1638949a852eSAl Viro !(flags & LOOKUP_NO_REVAL)) { 1639949a852eSAl Viro int error = d_revalidate(dentry, flags); 1640949a852eSAl Viro if (unlikely(error <= 0)) { 164194bdd655SAl Viro if (!error) { 1642949a852eSAl Viro d_invalidate(dentry); 1643949a852eSAl Viro dput(dentry); 164494bdd655SAl Viro goto again; 164594bdd655SAl Viro } 164694bdd655SAl Viro dput(dentry); 1647949a852eSAl Viro dentry = ERR_PTR(error); 1648949a852eSAl Viro } 1649949a852eSAl Viro } 165094bdd655SAl Viro } else { 16511936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 165285c7f810SAl Viro d_lookup_done(dentry); 16531936386eSAl Viro if (unlikely(old)) { 16541936386eSAl Viro dput(dentry); 16551936386eSAl Viro dentry = old; 1656949a852eSAl Viro } 1657949a852eSAl Viro } 165894bdd655SAl Viro out: 16599902af79SAl Viro inode_unlock_shared(inode); 1660e3c13928SAl Viro return dentry; 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 166352094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 166452094c8aSAl Viro { 166552094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16664ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 166752094c8aSAl Viro if (err != -ECHILD) 166852094c8aSAl Viro return err; 16696e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 167052094c8aSAl Viro return -ECHILD; 167152094c8aSAl Viro } 16724ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 167352094c8aSAl Viro } 167452094c8aSAl Viro 16759856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16769856fa1bSAl Viro { 16779856fa1bSAl Viro if (type == LAST_DOTDOT) { 16789e6697e2SAl Viro if (!nd->root.mnt) 16799e6697e2SAl Viro set_root(nd); 16809856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 168170291aecSAl Viro return follow_dotdot_rcu(nd); 16829856fa1bSAl Viro } else 1683397d425dSEric W. Biederman return follow_dotdot(nd); 16849856fa1bSAl Viro } 16859856fa1bSAl Viro return 0; 16869856fa1bSAl Viro } 16879856fa1bSAl Viro 1688181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1689181548c0SAl Viro struct inode *inode, unsigned seq) 1690d63ff28fSAl Viro { 1691626de996SAl Viro int error; 16921cf2665bSAl Viro struct saved *last; 1693756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1694626de996SAl Viro path_to_nameidata(link, nd); 1695626de996SAl Viro return -ELOOP; 1696626de996SAl Viro } 1697bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1698cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1699cd179f44SAl Viro mntget(link->mnt); 17007973387aSAl Viro } 1701626de996SAl Viro error = nd_alloc_stack(nd); 1702626de996SAl Viro if (unlikely(error)) { 1703bc40aee0SAl Viro if (error == -ECHILD) { 1704bc40aee0SAl Viro if (unlikely(unlazy_link(nd, link, seq))) 1705bc40aee0SAl Viro return -ECHILD; 1706bc40aee0SAl Viro error = nd_alloc_stack(nd); 1707bc40aee0SAl Viro } 1708bc40aee0SAl Viro if (error) { 1709cd179f44SAl Viro path_put(link); 1710626de996SAl Viro return error; 1711626de996SAl Viro } 1712bc40aee0SAl Viro } 1713626de996SAl Viro 1714ab104923SAl Viro last = nd->stack + nd->depth++; 17151cf2665bSAl Viro last->link = *link; 1716fceef393SAl Viro clear_delayed_call(&last->done); 1717fceef393SAl Viro nd->link_inode = inode; 17180450b2d1SAl Viro last->seq = seq; 1719d63ff28fSAl Viro return 1; 1720d63ff28fSAl Viro } 1721d63ff28fSAl Viro 17223ddcd056SLinus Torvalds /* 17233ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17243ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17253ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17263ddcd056SLinus Torvalds * for the common case. 17273ddcd056SLinus Torvalds */ 1728254cf582SAl Viro static inline int should_follow_link(struct nameidata *nd, struct path *link, 1729181548c0SAl Viro int follow, 1730181548c0SAl Viro struct inode *inode, unsigned seq) 17313ddcd056SLinus Torvalds { 1732d63ff28fSAl Viro if (likely(!d_is_symlink(link->dentry))) 1733d63ff28fSAl Viro return 0; 1734d63ff28fSAl Viro if (!follow) 1735d63ff28fSAl Viro return 0; 1736a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1737a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 1738a7f77542SAl Viro if (read_seqcount_retry(&link->dentry->d_seq, seq)) 1739a7f77542SAl Viro return -ECHILD; 1740a7f77542SAl Viro } 1741181548c0SAl Viro return pick_link(nd, link, inode, seq); 17423ddcd056SLinus Torvalds } 17433ddcd056SLinus Torvalds 17444693a547SAl Viro enum {WALK_GET = 1, WALK_PUT = 2}; 17454693a547SAl Viro 17464693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1747ce57dfc1SAl Viro { 1748caa85634SAl Viro struct path path; 1749ce57dfc1SAl Viro struct inode *inode; 1750254cf582SAl Viro unsigned seq; 1751ce57dfc1SAl Viro int err; 1752ce57dfc1SAl Viro /* 1753ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1754ce57dfc1SAl Viro * to be able to know about the current root directory and 1755ce57dfc1SAl Viro * parent relationships. 1756ce57dfc1SAl Viro */ 17574693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17584693a547SAl Viro err = handle_dots(nd, nd->last_type); 17594693a547SAl Viro if (flags & WALK_PUT) 17604693a547SAl Viro put_link(nd); 17614693a547SAl Viro return err; 17624693a547SAl Viro } 1763254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1764e9742b53SAl Viro if (unlikely(err <= 0)) { 1765697f514dSMiklos Szeredi if (err < 0) 1766f0a9ba70SAl Viro return err; 1767e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1768e3c13928SAl Viro nd->flags); 1769e3c13928SAl Viro if (IS_ERR(path.dentry)) 1770e3c13928SAl Viro return PTR_ERR(path.dentry); 17717500c38aSAl Viro 1772e3c13928SAl Viro path.mnt = nd->path.mnt; 1773e3c13928SAl Viro err = follow_managed(&path, nd); 1774e3c13928SAl Viro if (unlikely(err < 0)) 1775f0a9ba70SAl Viro return err; 1776697f514dSMiklos Szeredi 17777500c38aSAl Viro if (unlikely(d_is_negative(path.dentry))) { 17787500c38aSAl Viro path_to_nameidata(&path, nd); 17797500c38aSAl Viro return -ENOENT; 17807500c38aSAl Viro } 17817500c38aSAl Viro 1782254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1783d4565649SAl Viro inode = d_backing_inode(path.dentry); 1784766c4cbfSAl Viro } 1785697f514dSMiklos Szeredi 17864693a547SAl Viro if (flags & WALK_PUT) 17874693a547SAl Viro put_link(nd); 1788181548c0SAl Viro err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq); 1789d63ff28fSAl Viro if (unlikely(err)) 1790d63ff28fSAl Viro return err; 1791caa85634SAl Viro path_to_nameidata(&path, nd); 1792ce57dfc1SAl Viro nd->inode = inode; 1793254cf582SAl Viro nd->seq = seq; 1794ce57dfc1SAl Viro return 0; 1795ce57dfc1SAl Viro } 1796ce57dfc1SAl Viro 17971da177e4SLinus Torvalds /* 1798bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1799bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1800bfcfaa77SLinus Torvalds * 1801bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1802bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1803bfcfaa77SLinus Torvalds * fast. 1804bfcfaa77SLinus Torvalds * 1805bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1806bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1807bfcfaa77SLinus Torvalds * crossing operation. 1808bfcfaa77SLinus Torvalds * 1809bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1810bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1811bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1812bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1813bfcfaa77SLinus Torvalds */ 1814bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1815bfcfaa77SLinus Torvalds 1816f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1817bfcfaa77SLinus Torvalds 1818468a9428SGeorge Spelvin #ifdef HASH_MIX 1819bfcfaa77SLinus Torvalds 1820468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1821468a9428SGeorge Spelvin 1822468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18232a18da7aSGeorge Spelvin /* 18242a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18252a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18262a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18272a18da7aSGeorge Spelvin * and no temporaries. 18282a18da7aSGeorge Spelvin * 18292a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18302a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18312a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18322a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18332a18da7aSGeorge Spelvin * 18342a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18352a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18362a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18372a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18382a18da7aSGeorge Spelvin * 18392a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18402a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18412a18da7aSGeorge Spelvin * toggling any given output bit. 18422a18da7aSGeorge Spelvin * 18432a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18442a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18452a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18462a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18472a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18482a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18492a18da7aSGeorge Spelvin * Perfect: 8192 258048 18502a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18512a18da7aSGeorge Spelvin */ 18522a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18532a18da7aSGeorge Spelvin ( x ^= (a), \ 18542a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18552a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 18562a18da7aSGeorge Spelvin y *= 9 ) 1857bfcfaa77SLinus Torvalds 18580fed3ac8SGeorge Spelvin /* 18592a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 18602a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 18612a18da7aSGeorge Spelvin * work done before the hash value is used. 18620fed3ac8SGeorge Spelvin */ 18632a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18640fed3ac8SGeorge Spelvin { 18652a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 18662a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 18672a18da7aSGeorge Spelvin return y >> 32; 18680fed3ac8SGeorge Spelvin } 18690fed3ac8SGeorge Spelvin 1870bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1871bfcfaa77SLinus Torvalds 18722a18da7aSGeorge Spelvin /* 18732a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 18742a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18752a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 18762a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 18772a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 18782a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 18792a18da7aSGeorge Spelvin * Perfect: 2048 31744 18802a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 18812a18da7aSGeorge Spelvin */ 18822a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18832a18da7aSGeorge Spelvin ( x ^= (a), \ 18842a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 18852a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 18862a18da7aSGeorge Spelvin y *= 9 ) 1887bfcfaa77SLinus Torvalds 18882a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18890fed3ac8SGeorge Spelvin { 18902a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 18912a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 18920fed3ac8SGeorge Spelvin } 18930fed3ac8SGeorge Spelvin 1894bfcfaa77SLinus Torvalds #endif 1895bfcfaa77SLinus Torvalds 18962a18da7aSGeorge Spelvin /* 18972a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 18982a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 18992a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19002a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19012a18da7aSGeorge Spelvin * finds the delimiter after the name. 19022a18da7aSGeorge Spelvin */ 1903fcfd2fbfSGeorge Spelvin unsigned int full_name_hash(const char *name, unsigned int len) 1904bfcfaa77SLinus Torvalds { 19052a18da7aSGeorge Spelvin unsigned long a, x = 0, y = 0; 1906bfcfaa77SLinus Torvalds 1907bfcfaa77SLinus Torvalds for (;;) { 1908fcfd2fbfSGeorge Spelvin if (!len) 1909fcfd2fbfSGeorge Spelvin goto done; 1910e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1911bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1912bfcfaa77SLinus Torvalds break; 19132a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1914bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1915bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1916bfcfaa77SLinus Torvalds } 19172a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1918bfcfaa77SLinus Torvalds done: 19192a18da7aSGeorge Spelvin return fold_hash(x, y); 1920bfcfaa77SLinus Torvalds } 1921bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1922bfcfaa77SLinus Torvalds 1923fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 1924fcfd2fbfSGeorge Spelvin u64 hashlen_string(const char *name) 1925fcfd2fbfSGeorge Spelvin { 19262a18da7aSGeorge Spelvin unsigned long a = 0, x = 0, y = 0, adata, mask, len; 1927fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1928fcfd2fbfSGeorge Spelvin 1929fcfd2fbfSGeorge Spelvin len = -sizeof(unsigned long); 1930fcfd2fbfSGeorge Spelvin do { 19312a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1932fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 1933fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1934fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1935fcfd2fbfSGeorge Spelvin 1936fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1937fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19382a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1939fcfd2fbfSGeorge Spelvin 19402a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1941fcfd2fbfSGeorge Spelvin } 1942fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1943fcfd2fbfSGeorge Spelvin 1944bfcfaa77SLinus Torvalds /* 1945bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1946d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1947bfcfaa77SLinus Torvalds */ 1948d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 1949bfcfaa77SLinus Torvalds { 19502a18da7aSGeorge Spelvin unsigned long a = 0, b, x = 0, y = 0, adata, bdata, mask, len; 195136126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1952bfcfaa77SLinus Torvalds 1953bfcfaa77SLinus Torvalds len = -sizeof(unsigned long); 1954bfcfaa77SLinus Torvalds do { 19552a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1956bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 1957e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 195836126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 195936126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1960bfcfaa77SLinus Torvalds 196136126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 196236126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 196336126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 19642a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 196536126f8fSLinus Torvalds 19662a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1967bfcfaa77SLinus Torvalds } 1968bfcfaa77SLinus Torvalds 19692a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 1970bfcfaa77SLinus Torvalds 1971fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 1972fcfd2fbfSGeorge Spelvin unsigned int full_name_hash(const char *name, unsigned int len) 19730145acc2SLinus Torvalds { 19740145acc2SLinus Torvalds unsigned long hash = init_name_hash(); 19750145acc2SLinus Torvalds while (len--) 1976fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 19770145acc2SLinus Torvalds return end_name_hash(hash); 19780145acc2SLinus Torvalds } 1979ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 19800145acc2SLinus Torvalds 1981fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 1982f2a031b6SGeorge Spelvin u64 hashlen_string(const char *name) 1983fcfd2fbfSGeorge Spelvin { 1984fcfd2fbfSGeorge Spelvin unsigned long hash = init_name_hash(); 1985fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 1986fcfd2fbfSGeorge Spelvin 1987fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 1988e0ab7af9SGeorge Spelvin while (c) { 1989fcfd2fbfSGeorge Spelvin len++; 1990fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 1991fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 1992e0ab7af9SGeorge Spelvin } 1993fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 1994fcfd2fbfSGeorge Spelvin } 1995f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1996fcfd2fbfSGeorge Spelvin 19973ddcd056SLinus Torvalds /* 1998200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1999200e9ef7SLinus Torvalds * one character. 2000200e9ef7SLinus Torvalds */ 2001d6bb3e90SLinus Torvalds static inline u64 hash_name(const char *name) 2002200e9ef7SLinus Torvalds { 2003200e9ef7SLinus Torvalds unsigned long hash = init_name_hash(); 2004200e9ef7SLinus Torvalds unsigned long len = 0, c; 2005200e9ef7SLinus Torvalds 2006200e9ef7SLinus Torvalds c = (unsigned char)*name; 2007200e9ef7SLinus Torvalds do { 2008200e9ef7SLinus Torvalds len++; 2009200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2010200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2011200e9ef7SLinus Torvalds } while (c && c != '/'); 2012d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2013200e9ef7SLinus Torvalds } 2014200e9ef7SLinus Torvalds 2015bfcfaa77SLinus Torvalds #endif 2016bfcfaa77SLinus Torvalds 2017200e9ef7SLinus Torvalds /* 20181da177e4SLinus Torvalds * Name resolution. 2019ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2020ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20211da177e4SLinus Torvalds * 2022ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2023ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20241da177e4SLinus Torvalds */ 20256de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20261da177e4SLinus Torvalds { 20271da177e4SLinus Torvalds int err; 20281da177e4SLinus Torvalds 20291da177e4SLinus Torvalds while (*name=='/') 20301da177e4SLinus Torvalds name++; 20311da177e4SLinus Torvalds if (!*name) 20329e18f10aSAl Viro return 0; 20331da177e4SLinus Torvalds 20341da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20351da177e4SLinus Torvalds for(;;) { 2036d6bb3e90SLinus Torvalds u64 hash_len; 2037fe479a58SAl Viro int type; 20381da177e4SLinus Torvalds 203952094c8aSAl Viro err = may_lookup(nd); 20401da177e4SLinus Torvalds if (err) 20413595e234SAl Viro return err; 20421da177e4SLinus Torvalds 2043d6bb3e90SLinus Torvalds hash_len = hash_name(name); 20441da177e4SLinus Torvalds 2045fe479a58SAl Viro type = LAST_NORM; 2046d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2047fe479a58SAl Viro case 2: 2048200e9ef7SLinus Torvalds if (name[1] == '.') { 2049fe479a58SAl Viro type = LAST_DOTDOT; 205016c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 205116c2cd71SAl Viro } 2052fe479a58SAl Viro break; 2053fe479a58SAl Viro case 1: 2054fe479a58SAl Viro type = LAST_DOT; 2055fe479a58SAl Viro } 20565a202bcdSAl Viro if (likely(type == LAST_NORM)) { 20575a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 205816c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 20595a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2060a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2061da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 20625a202bcdSAl Viro if (err < 0) 20633595e234SAl Viro return err; 2064d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2065d6bb3e90SLinus Torvalds name = this.name; 20665a202bcdSAl Viro } 20675a202bcdSAl Viro } 2068fe479a58SAl Viro 2069d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2070d6bb3e90SLinus Torvalds nd->last.name = name; 20715f4a6a69SAl Viro nd->last_type = type; 20725f4a6a69SAl Viro 2073d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2074d6bb3e90SLinus Torvalds if (!*name) 2075bdf6cbf1SAl Viro goto OK; 2076200e9ef7SLinus Torvalds /* 2077200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2078200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2079200e9ef7SLinus Torvalds */ 2080200e9ef7SLinus Torvalds do { 2081d6bb3e90SLinus Torvalds name++; 2082d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 20838620c238SAl Viro if (unlikely(!*name)) { 20848620c238SAl Viro OK: 2085368ee9baSAl Viro /* pathname body, done */ 20868620c238SAl Viro if (!nd->depth) 20878620c238SAl Viro return 0; 20888620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2089368ee9baSAl Viro /* trailing symlink, done */ 20908620c238SAl Viro if (!name) 20918620c238SAl Viro return 0; 20928620c238SAl Viro /* last component of nested symlink */ 20934693a547SAl Viro err = walk_component(nd, WALK_GET | WALK_PUT); 20948620c238SAl Viro } else { 20954693a547SAl Viro err = walk_component(nd, WALK_GET); 20968620c238SAl Viro } 2097ce57dfc1SAl Viro if (err < 0) 20983595e234SAl Viro return err; 2099fe479a58SAl Viro 2100ce57dfc1SAl Viro if (err) { 2101626de996SAl Viro const char *s = get_link(nd); 21025a460275SAl Viro 2103a1c83681SViresh Kumar if (IS_ERR(s)) 21043595e234SAl Viro return PTR_ERR(s); 2105172a39a0SAl Viro err = 0; 210612b09578SAl Viro if (unlikely(!s)) { 210712b09578SAl Viro /* jumped */ 2108b9ff4429SAl Viro put_link(nd); 210912b09578SAl Viro } else { 2110071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 211132cd7468SAl Viro name = s; 21129e18f10aSAl Viro continue; 211348c8b0c5SAl Viro } 211431e6b01fSNick Piggin } 211597242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 211697242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 211797242f99SAl Viro if (unlazy_walk(nd, NULL, 0)) 211897242f99SAl Viro return -ECHILD; 211997242f99SAl Viro } 21203595e234SAl Viro return -ENOTDIR; 21215f4a6a69SAl Viro } 2122ce57dfc1SAl Viro } 212397242f99SAl Viro } 21241da177e4SLinus Torvalds 2125c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 212631e6b01fSNick Piggin { 212731e6b01fSNick Piggin int retval = 0; 2128c8a53ee5SAl Viro const char *s = nd->name->name; 212931e6b01fSNick Piggin 213031e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2131980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 213231e6b01fSNick Piggin nd->depth = 0; 21335b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2134b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2135b18825a7SDavid Howells struct inode *inode = root->d_inode; 2136fd2f7cb5SAl Viro if (*s) { 213744b1d530SMiklos Szeredi if (!d_can_lookup(root)) 2138368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21395b6ca027SAl Viro retval = inode_permission(inode, MAY_EXEC); 21405b6ca027SAl Viro if (retval) 2141368ee9baSAl Viro return ERR_PTR(retval); 214273d049a4SAl Viro } 21435b6ca027SAl Viro nd->path = nd->root; 21445b6ca027SAl Viro nd->inode = inode; 21455b6ca027SAl Viro if (flags & LOOKUP_RCU) { 21468b61e74fSAl Viro rcu_read_lock(); 21475b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 21488f47a016SAl Viro nd->root_seq = nd->seq; 214948a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 21505b6ca027SAl Viro } else { 21515b6ca027SAl Viro path_get(&nd->path); 21525b6ca027SAl Viro } 2153368ee9baSAl Viro return s; 21545b6ca027SAl Viro } 21555b6ca027SAl Viro 215631e6b01fSNick Piggin nd->root.mnt = NULL; 2157248fb5b9SAl Viro nd->path.mnt = NULL; 2158248fb5b9SAl Viro nd->path.dentry = NULL; 215931e6b01fSNick Piggin 216048a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2161fd2f7cb5SAl Viro if (*s == '/') { 21629e6697e2SAl Viro if (flags & LOOKUP_RCU) 21638b61e74fSAl Viro rcu_read_lock(); 2164e41f7d4eSAl Viro set_root(nd); 2165248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2166ef55d917SAl Viro return s; 2167ef55d917SAl Viro nd->root.mnt = NULL; 2168ef55d917SAl Viro rcu_read_unlock(); 2169ef55d917SAl Viro return ERR_PTR(-ECHILD); 2170c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2171e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 217231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2173c28cc364SNick Piggin unsigned seq; 217431e6b01fSNick Piggin 21758b61e74fSAl Viro rcu_read_lock(); 217631e6b01fSNick Piggin 2177c28cc364SNick Piggin do { 2178c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 217931e6b01fSNick Piggin nd->path = fs->pwd; 2180ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2181c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2182c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2183e41f7d4eSAl Viro } else { 2184e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2185ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2186e41f7d4eSAl Viro } 2187ef55d917SAl Viro return s; 218831e6b01fSNick Piggin } else { 2189582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2190c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 219131e6b01fSNick Piggin struct dentry *dentry; 219231e6b01fSNick Piggin 21932903ff01SAl Viro if (!f.file) 2194368ee9baSAl Viro return ERR_PTR(-EBADF); 219531e6b01fSNick Piggin 21962903ff01SAl Viro dentry = f.file->f_path.dentry; 219731e6b01fSNick Piggin 2198fd2f7cb5SAl Viro if (*s) { 219944b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 22002903ff01SAl Viro fdput(f); 2201368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2202f52e0c11SAl Viro } 22032903ff01SAl Viro } 22042903ff01SAl Viro 22052903ff01SAl Viro nd->path = f.file->f_path; 2206e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 22078b61e74fSAl Viro rcu_read_lock(); 220834a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 220934a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22105590ff0dSUlrich Drepper } else { 22112903ff01SAl Viro path_get(&nd->path); 221234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22131da177e4SLinus Torvalds } 221434a26b99SAl Viro fdput(f); 2215368ee9baSAl Viro return s; 2216e41f7d4eSAl Viro } 22179b4a9b14SAl Viro } 22189b4a9b14SAl Viro 22193bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 222095fa25d9SAl Viro { 222195fa25d9SAl Viro const char *s; 2222fec2fa24SAl Viro int error = may_follow_link(nd); 2223deb106c6SAl Viro if (unlikely(error)) 22243bdba28bSAl Viro return ERR_PTR(error); 222595fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2226fab51e8aSAl Viro nd->stack[0].name = NULL; 22273b2e7f75SAl Viro s = get_link(nd); 2228deb106c6SAl Viro return s ? s : ""; 222995fa25d9SAl Viro } 223095fa25d9SAl Viro 2231caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2232bd92d7feSAl Viro { 2233bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2234bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2235bd92d7feSAl Viro 2236bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 2237deb106c6SAl Viro return walk_component(nd, 22384693a547SAl Viro nd->flags & LOOKUP_FOLLOW 22394693a547SAl Viro ? nd->depth 22404693a547SAl Viro ? WALK_PUT | WALK_GET 22414693a547SAl Viro : WALK_GET 22424693a547SAl Viro : 0); 2243bd92d7feSAl Viro } 2244bd92d7feSAl Viro 22459b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2246c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 22479b4a9b14SAl Viro { 2248c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2249bd92d7feSAl Viro int err; 225031e6b01fSNick Piggin 2251368ee9baSAl Viro if (IS_ERR(s)) 2252368ee9baSAl Viro return PTR_ERR(s); 22533bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 22543bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 22553bdba28bSAl Viro s = trailing_symlink(nd); 22563bdba28bSAl Viro if (IS_ERR(s)) { 22573bdba28bSAl Viro err = PTR_ERR(s); 22586d7b5aaeSAl Viro break; 2259bd92d7feSAl Viro } 2260bd92d7feSAl Viro } 22619f1fafeeSAl Viro if (!err) 22629f1fafeeSAl Viro err = complete_walk(nd); 2263bd92d7feSAl Viro 2264deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2265deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2266bd23a539SAl Viro err = -ENOTDIR; 2267625b6d10SAl Viro if (!err) { 2268625b6d10SAl Viro *path = nd->path; 2269625b6d10SAl Viro nd->path.mnt = NULL; 2270625b6d10SAl Viro nd->path.dentry = NULL; 2271625b6d10SAl Viro } 2272deb106c6SAl Viro terminate_walk(nd); 2273bd92d7feSAl Viro return err; 227431e6b01fSNick Piggin } 227531e6b01fSNick Piggin 2276625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 22779ad1aaa6SAl Viro struct path *path, struct path *root) 2278873f1eedSJeff Layton { 2279894bc8c4SAl Viro int retval; 22809883d185SAl Viro struct nameidata nd; 2281abc9f5beSAl Viro if (IS_ERR(name)) 2282abc9f5beSAl Viro return PTR_ERR(name); 22839ad1aaa6SAl Viro if (unlikely(root)) { 22849ad1aaa6SAl Viro nd.root = *root; 22859ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 22869ad1aaa6SAl Viro } 22879883d185SAl Viro set_nameidata(&nd, dfd, name); 2288c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2289873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2290c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2291873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2292c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2293873f1eedSJeff Layton 2294873f1eedSJeff Layton if (likely(!retval)) 2295625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 22969883d185SAl Viro restore_nameidata(); 2297e4bd1c1aSAl Viro putname(name); 2298873f1eedSJeff Layton return retval; 2299873f1eedSJeff Layton } 2300873f1eedSJeff Layton 23018bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2302c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2303391172c4SAl Viro struct path *parent) 23048bcb77faSAl Viro { 2305c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2306368ee9baSAl Viro int err; 2307368ee9baSAl Viro if (IS_ERR(s)) 2308368ee9baSAl Viro return PTR_ERR(s); 2309368ee9baSAl Viro err = link_path_walk(s, nd); 23108bcb77faSAl Viro if (!err) 23118bcb77faSAl Viro err = complete_walk(nd); 2312391172c4SAl Viro if (!err) { 2313391172c4SAl Viro *parent = nd->path; 2314391172c4SAl Viro nd->path.mnt = NULL; 2315391172c4SAl Viro nd->path.dentry = NULL; 2316391172c4SAl Viro } 2317deb106c6SAl Viro terminate_walk(nd); 23188bcb77faSAl Viro return err; 23198bcb77faSAl Viro } 23208bcb77faSAl Viro 23215c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2322391172c4SAl Viro unsigned int flags, struct path *parent, 2323391172c4SAl Viro struct qstr *last, int *type) 23248bcb77faSAl Viro { 23258bcb77faSAl Viro int retval; 23269883d185SAl Viro struct nameidata nd; 23278bcb77faSAl Viro 23285c31b6ceSAl Viro if (IS_ERR(name)) 23295c31b6ceSAl Viro return name; 23309883d185SAl Viro set_nameidata(&nd, dfd, name); 2331c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23328bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2333c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23348bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2335c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2336391172c4SAl Viro if (likely(!retval)) { 2337391172c4SAl Viro *last = nd.last; 2338391172c4SAl Viro *type = nd.last_type; 2339391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 23405c31b6ceSAl Viro } else { 23415c31b6ceSAl Viro putname(name); 23425c31b6ceSAl Viro name = ERR_PTR(retval); 2343391172c4SAl Viro } 23449883d185SAl Viro restore_nameidata(); 23455c31b6ceSAl Viro return name; 23468bcb77faSAl Viro } 23478bcb77faSAl Viro 234879714f72SAl Viro /* does lookup, returns the object with parent locked */ 234979714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 23505590ff0dSUlrich Drepper { 23515c31b6ceSAl Viro struct filename *filename; 23525c31b6ceSAl Viro struct dentry *d; 2353391172c4SAl Viro struct qstr last; 2354391172c4SAl Viro int type; 235551689104SPaul Moore 23565c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 23575c31b6ceSAl Viro &last, &type); 235851689104SPaul Moore if (IS_ERR(filename)) 235951689104SPaul Moore return ERR_CAST(filename); 23605c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2361391172c4SAl Viro path_put(path); 23625c31b6ceSAl Viro putname(filename); 23635c31b6ceSAl Viro return ERR_PTR(-EINVAL); 236479714f72SAl Viro } 23655955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2366391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 236779714f72SAl Viro if (IS_ERR(d)) { 23685955102cSAl Viro inode_unlock(path->dentry->d_inode); 2369391172c4SAl Viro path_put(path); 237079714f72SAl Viro } 237151689104SPaul Moore putname(filename); 237279714f72SAl Viro return d; 23735590ff0dSUlrich Drepper } 23745590ff0dSUlrich Drepper 2375d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2376d1811465SAl Viro { 2377abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2378abc9f5beSAl Viro flags, path, NULL); 2379d1811465SAl Viro } 23804d359507SAl Viro EXPORT_SYMBOL(kern_path); 2381d1811465SAl Viro 238216f18200SJosef 'Jeff' Sipek /** 238316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 238416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 238516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 238616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 238716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2388e0a01249SAl Viro * @path: pointer to struct path to fill 238916f18200SJosef 'Jeff' Sipek */ 239016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 239116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2392e0a01249SAl Viro struct path *path) 239316f18200SJosef 'Jeff' Sipek { 23949ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 23959ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2396abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2397abc9f5beSAl Viro flags , path, &root); 239816f18200SJosef 'Jeff' Sipek } 23994d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 240016f18200SJosef 'Jeff' Sipek 2401eead1911SChristoph Hellwig /** 24023c9fe8cdSMiklos Szeredi * lookup_hash - lookup single pathname component on already hashed name 24033c9fe8cdSMiklos Szeredi * @name: name and hash to lookup 24043c9fe8cdSMiklos Szeredi * @base: base directory to lookup from 24053c9fe8cdSMiklos Szeredi * 24063c9fe8cdSMiklos Szeredi * The name must have been verified and hashed (see lookup_one_len()). Using 24073c9fe8cdSMiklos Szeredi * this after just full_name_hash() is unsafe. 24083c9fe8cdSMiklos Szeredi * 24093c9fe8cdSMiklos Szeredi * This function also doesn't check for search permission on base directory. 24103c9fe8cdSMiklos Szeredi * 24113c9fe8cdSMiklos Szeredi * Use lookup_one_len_unlocked() instead, unless you really know what you are 24123c9fe8cdSMiklos Szeredi * doing. 24133c9fe8cdSMiklos Szeredi * 24143c9fe8cdSMiklos Szeredi * Do not hold i_mutex; this helper takes i_mutex if necessary. 24153c9fe8cdSMiklos Szeredi */ 24163c9fe8cdSMiklos Szeredi struct dentry *lookup_hash(const struct qstr *name, struct dentry *base) 24173c9fe8cdSMiklos Szeredi { 24183c9fe8cdSMiklos Szeredi struct dentry *ret; 24193c9fe8cdSMiklos Szeredi 24203c9fe8cdSMiklos Szeredi ret = lookup_dcache(name, base, 0); 24213c9fe8cdSMiklos Szeredi if (!ret) 24223c9fe8cdSMiklos Szeredi ret = lookup_slow(name, base, 0); 24233c9fe8cdSMiklos Szeredi 24243c9fe8cdSMiklos Szeredi return ret; 24253c9fe8cdSMiklos Szeredi } 24263c9fe8cdSMiklos Szeredi EXPORT_SYMBOL(lookup_hash); 24273c9fe8cdSMiklos Szeredi 24283c9fe8cdSMiklos Szeredi /** 2429a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2430eead1911SChristoph Hellwig * @name: pathname component to lookup 2431eead1911SChristoph Hellwig * @base: base directory to lookup from 2432eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2433eead1911SChristoph Hellwig * 2434a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 24359e7543e9SAl Viro * not be called by generic code. 2436bbddca8eSNeilBrown * 2437bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2438eead1911SChristoph Hellwig */ 2439057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2440057f6c01SJames Morris { 2441057f6c01SJames Morris struct qstr this; 24426a96ba54SAl Viro unsigned int c; 2443cda309deSMiklos Szeredi int err; 2444057f6c01SJames Morris 24455955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 24462f9092e1SDavid Woodhouse 24476a96ba54SAl Viro this.name = name; 24486a96ba54SAl Viro this.len = len; 24490145acc2SLinus Torvalds this.hash = full_name_hash(name, len); 24506a96ba54SAl Viro if (!len) 24516a96ba54SAl Viro return ERR_PTR(-EACCES); 24526a96ba54SAl Viro 245321d8a15aSAl Viro if (unlikely(name[0] == '.')) { 245421d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 245521d8a15aSAl Viro return ERR_PTR(-EACCES); 245621d8a15aSAl Viro } 245721d8a15aSAl Viro 24586a96ba54SAl Viro while (len--) { 24596a96ba54SAl Viro c = *(const unsigned char *)name++; 24606a96ba54SAl Viro if (c == '/' || c == '\0') 24616a96ba54SAl Viro return ERR_PTR(-EACCES); 24626a96ba54SAl Viro } 24635a202bcdSAl Viro /* 24645a202bcdSAl Viro * See if the low-level filesystem might want 24655a202bcdSAl Viro * to use its own hash.. 24665a202bcdSAl Viro */ 24675a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2468da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 24695a202bcdSAl Viro if (err < 0) 24705a202bcdSAl Viro return ERR_PTR(err); 24715a202bcdSAl Viro } 2472eead1911SChristoph Hellwig 2473cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2474cda309deSMiklos Szeredi if (err) 2475cda309deSMiklos Szeredi return ERR_PTR(err); 2476cda309deSMiklos Szeredi 247772bd866aSAl Viro return __lookup_hash(&this, base, 0); 2478057f6c01SJames Morris } 24794d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2480057f6c01SJames Morris 2481bbddca8eSNeilBrown /** 2482bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2483bbddca8eSNeilBrown * @name: pathname component to lookup 2484bbddca8eSNeilBrown * @base: base directory to lookup from 2485bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2486bbddca8eSNeilBrown * 2487bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2488bbddca8eSNeilBrown * not be called by generic code. 2489bbddca8eSNeilBrown * 2490bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2491bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2492bbddca8eSNeilBrown */ 2493bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2494bbddca8eSNeilBrown struct dentry *base, int len) 2495bbddca8eSNeilBrown { 2496bbddca8eSNeilBrown struct qstr this; 2497bbddca8eSNeilBrown unsigned int c; 2498bbddca8eSNeilBrown int err; 2499bbddca8eSNeilBrown 2500bbddca8eSNeilBrown this.name = name; 2501bbddca8eSNeilBrown this.len = len; 2502bbddca8eSNeilBrown this.hash = full_name_hash(name, len); 2503bbddca8eSNeilBrown if (!len) 2504bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2505bbddca8eSNeilBrown 2506bbddca8eSNeilBrown if (unlikely(name[0] == '.')) { 2507bbddca8eSNeilBrown if (len < 2 || (len == 2 && name[1] == '.')) 2508bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2509bbddca8eSNeilBrown } 2510bbddca8eSNeilBrown 2511bbddca8eSNeilBrown while (len--) { 2512bbddca8eSNeilBrown c = *(const unsigned char *)name++; 2513bbddca8eSNeilBrown if (c == '/' || c == '\0') 2514bbddca8eSNeilBrown return ERR_PTR(-EACCES); 2515bbddca8eSNeilBrown } 2516bbddca8eSNeilBrown /* 2517bbddca8eSNeilBrown * See if the low-level filesystem might want 2518bbddca8eSNeilBrown * to use its own hash.. 2519bbddca8eSNeilBrown */ 2520bbddca8eSNeilBrown if (base->d_flags & DCACHE_OP_HASH) { 2521bbddca8eSNeilBrown int err = base->d_op->d_hash(base, &this); 2522bbddca8eSNeilBrown if (err < 0) 2523bbddca8eSNeilBrown return ERR_PTR(err); 2524bbddca8eSNeilBrown } 2525bbddca8eSNeilBrown 2526bbddca8eSNeilBrown err = inode_permission(base->d_inode, MAY_EXEC); 2527bbddca8eSNeilBrown if (err) 2528bbddca8eSNeilBrown return ERR_PTR(err); 2529bbddca8eSNeilBrown 25303c9fe8cdSMiklos Szeredi return lookup_hash(&this, base); 2531bbddca8eSNeilBrown } 2532bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2533bbddca8eSNeilBrown 2534eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2535eedf265aSEric W. Biederman int path_pts(struct path *path) 2536eedf265aSEric W. Biederman { 2537eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2538eedf265aSEric W. Biederman * the input path. 2539eedf265aSEric W. Biederman */ 2540eedf265aSEric W. Biederman struct dentry *child, *parent; 2541eedf265aSEric W. Biederman struct qstr this; 2542eedf265aSEric W. Biederman int ret; 2543eedf265aSEric W. Biederman 2544eedf265aSEric W. Biederman ret = path_parent_directory(path); 2545eedf265aSEric W. Biederman if (ret) 2546eedf265aSEric W. Biederman return ret; 2547eedf265aSEric W. Biederman 2548eedf265aSEric W. Biederman parent = path->dentry; 2549eedf265aSEric W. Biederman this.name = "pts"; 2550eedf265aSEric W. Biederman this.len = 3; 2551eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2552eedf265aSEric W. Biederman if (!child) 2553eedf265aSEric W. Biederman return -ENOENT; 2554eedf265aSEric W. Biederman 2555eedf265aSEric W. Biederman path->dentry = child; 2556eedf265aSEric W. Biederman dput(parent); 2557eedf265aSEric W. Biederman follow_mount(path); 2558eedf265aSEric W. Biederman return 0; 2559eedf265aSEric W. Biederman } 2560eedf265aSEric W. Biederman #endif 2561eedf265aSEric W. Biederman 25621fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 25631fa1e7f6SAndy Whitcroft struct path *path, int *empty) 25641da177e4SLinus Torvalds { 2565abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2566abc9f5beSAl Viro flags, path, NULL); 25671da177e4SLinus Torvalds } 2568b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 25691fa1e7f6SAndy Whitcroft 2570873f1eedSJeff Layton /* 2571873f1eedSJeff Layton * NB: most callers don't do anything directly with the reference to the 2572873f1eedSJeff Layton * to struct filename, but the nd->last pointer points into the name string 2573873f1eedSJeff Layton * allocated by getname. So we must hold the reference to it until all 2574873f1eedSJeff Layton * path-walking is complete. 2575873f1eedSJeff Layton */ 2576a2ec4a2dSAl Viro static inline struct filename * 2577f5beed75SAl Viro user_path_parent(int dfd, const char __user *path, 2578f5beed75SAl Viro struct path *parent, 2579f5beed75SAl Viro struct qstr *last, 2580f5beed75SAl Viro int *type, 25819e790bd6SJeff Layton unsigned int flags) 25822ad94ae6SAl Viro { 25839e790bd6SJeff Layton /* only LOOKUP_REVAL is allowed in extra flags */ 25845c31b6ceSAl Viro return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL, 25855c31b6ceSAl Viro parent, last, type); 25862ad94ae6SAl Viro } 25872ad94ae6SAl Viro 25888033426eSJeff Layton /** 2589197df04cSAl Viro * mountpoint_last - look up last component for umount 25908033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 25918033426eSJeff Layton * @path: pointer to container for result 25928033426eSJeff Layton * 25938033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 25948033426eSJeff Layton * need to resolve the path without doing any revalidation. 25958033426eSJeff Layton * 25968033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 25978033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 25988033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 25998033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 26008033426eSJeff Layton * bogus and it doesn't exist. 26018033426eSJeff Layton * 26028033426eSJeff Layton * Returns: 26038033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 26048033426eSJeff Layton * lookup found a negative dentry. The nd->path reference will also be 26058033426eSJeff Layton * put in this case. 26068033426eSJeff Layton * 26078033426eSJeff Layton * 0: if we successfully resolved nd->path and found it to not to be a 26088033426eSJeff Layton * symlink that needs to be followed. "path" will also be populated. 26098033426eSJeff Layton * The nd->path reference will also be put. 26108033426eSJeff Layton * 26118033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 26128033426eSJeff Layton * that needs to be followed. "path" will be populated with the path 26138033426eSJeff Layton * to the link, and nd->path will *not* be put. 26148033426eSJeff Layton */ 26158033426eSJeff Layton static int 2616197df04cSAl Viro mountpoint_last(struct nameidata *nd, struct path *path) 26178033426eSJeff Layton { 26188033426eSJeff Layton int error = 0; 26198033426eSJeff Layton struct dentry *dentry; 26208033426eSJeff Layton struct dentry *dir = nd->path.dentry; 26218033426eSJeff Layton 262235759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 262335759521SAl Viro if (nd->flags & LOOKUP_RCU) { 26246e9918b7SAl Viro if (unlazy_walk(nd, NULL, 0)) 2625deb106c6SAl Viro return -ECHILD; 26268033426eSJeff Layton } 26278033426eSJeff Layton 26288033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 26298033426eSJeff Layton 26308033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 26318033426eSJeff Layton error = handle_dots(nd, nd->last_type); 263235759521SAl Viro if (error) 2633deb106c6SAl Viro return error; 26348033426eSJeff Layton dentry = dget(nd->path.dentry); 2635949a852eSAl Viro } else { 26368033426eSJeff Layton dentry = d_lookup(dir, &nd->last); 26378033426eSJeff Layton if (!dentry) { 26388033426eSJeff Layton /* 2639949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2640949a852eSAl Viro * cache, so that means that this dentry is probably 2641949a852eSAl Viro * a symlink or the path doesn't actually point 2642949a852eSAl Viro * to a mounted dentry. 26438033426eSJeff Layton */ 2644949a852eSAl Viro dentry = lookup_slow(&nd->last, dir, 2645949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2646949a852eSAl Viro if (IS_ERR(dentry)) 2647deb106c6SAl Viro return PTR_ERR(dentry); 26488033426eSJeff Layton } 2649bcceeebaSDave Jones } 2650698934dfSDavid Howells if (d_is_negative(dentry)) { 26518033426eSJeff Layton dput(dentry); 2652deb106c6SAl Viro return -ENOENT; 265335759521SAl Viro } 2654191d7f73SAl Viro if (nd->depth) 2655191d7f73SAl Viro put_link(nd); 26568033426eSJeff Layton path->dentry = dentry; 2657295dc39dSVasily Averin path->mnt = nd->path.mnt; 2658181548c0SAl Viro error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW, 2659181548c0SAl Viro d_backing_inode(dentry), 0); 2660deb106c6SAl Viro if (unlikely(error)) 2661d63ff28fSAl Viro return error; 2662295dc39dSVasily Averin mntget(path->mnt); 26638033426eSJeff Layton follow_mount(path); 2664deb106c6SAl Viro return 0; 26658033426eSJeff Layton } 26668033426eSJeff Layton 26678033426eSJeff Layton /** 2668197df04cSAl Viro * path_mountpoint - look up a path to be umounted 26692a78b857SMasanari Iida * @nd: lookup context 26708033426eSJeff Layton * @flags: lookup flags 2671c8a53ee5SAl Viro * @path: pointer to container for result 26728033426eSJeff Layton * 26738033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2674606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 26758033426eSJeff Layton */ 26768033426eSJeff Layton static int 2677c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 26788033426eSJeff Layton { 2679c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2680368ee9baSAl Viro int err; 2681368ee9baSAl Viro if (IS_ERR(s)) 2682368ee9baSAl Viro return PTR_ERR(s); 26833bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 26843bdba28bSAl Viro (err = mountpoint_last(nd, path)) > 0) { 26853bdba28bSAl Viro s = trailing_symlink(nd); 26863bdba28bSAl Viro if (IS_ERR(s)) { 26873bdba28bSAl Viro err = PTR_ERR(s); 26888033426eSJeff Layton break; 26898033426eSJeff Layton } 26903bdba28bSAl Viro } 2691deb106c6SAl Viro terminate_walk(nd); 26928033426eSJeff Layton return err; 26938033426eSJeff Layton } 26948033426eSJeff Layton 26952d864651SAl Viro static int 2696668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 26972d864651SAl Viro unsigned int flags) 26982d864651SAl Viro { 26999883d185SAl Viro struct nameidata nd; 2700cbaab2dbSAl Viro int error; 2701668696dcSAl Viro if (IS_ERR(name)) 2702668696dcSAl Viro return PTR_ERR(name); 27039883d185SAl Viro set_nameidata(&nd, dfd, name); 2704c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 27052d864651SAl Viro if (unlikely(error == -ECHILD)) 2706c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 27072d864651SAl Viro if (unlikely(error == -ESTALE)) 2708c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 27092d864651SAl Viro if (likely(!error)) 2710668696dcSAl Viro audit_inode(name, path->dentry, 0); 27119883d185SAl Viro restore_nameidata(); 2712668696dcSAl Viro putname(name); 27132d864651SAl Viro return error; 27142d864651SAl Viro } 27152d864651SAl Viro 27168033426eSJeff Layton /** 2717197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 27188033426eSJeff Layton * @dfd: directory file descriptor 27198033426eSJeff Layton * @name: pathname from userland 27208033426eSJeff Layton * @flags: lookup flags 27218033426eSJeff Layton * @path: pointer to container to hold result 27228033426eSJeff Layton * 27238033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27248033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27258033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27268033426eSJeff Layton * and avoid revalidating the last component. 27278033426eSJeff Layton * 27288033426eSJeff Layton * Returns 0 and populates "path" on success. 27298033426eSJeff Layton */ 27308033426eSJeff Layton int 2731197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27328033426eSJeff Layton struct path *path) 27338033426eSJeff Layton { 2734cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27358033426eSJeff Layton } 27368033426eSJeff Layton 27372d864651SAl Viro int 27382d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27392d864651SAl Viro unsigned int flags) 27402d864651SAl Viro { 2741cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27422d864651SAl Viro } 27432d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27442d864651SAl Viro 2745cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27461da177e4SLinus Torvalds { 27478e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2748da9592edSDavid Howells 27498e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27501da177e4SLinus Torvalds return 0; 27518e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27521da177e4SLinus Torvalds return 0; 275323adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27541da177e4SLinus Torvalds } 2755cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27561da177e4SLinus Torvalds 27571da177e4SLinus Torvalds /* 27581da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27591da177e4SLinus Torvalds * whether the type of victim is right. 27601da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27611da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27621da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27631da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27641da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27651da177e4SLinus Torvalds * a. be owner of dir, or 27661da177e4SLinus Torvalds * b. be owner of victim, or 27671da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27681da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27691da177e4SLinus Torvalds * links pointing to it. 27700bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27710bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27720bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27730bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27740bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27751da177e4SLinus Torvalds * nfs_async_unlink(). 27761da177e4SLinus Torvalds */ 2777b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27781da177e4SLinus Torvalds { 277963afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27801da177e4SLinus Torvalds int error; 27811da177e4SLinus Torvalds 2782b18825a7SDavid Howells if (d_is_negative(victim)) 27831da177e4SLinus Torvalds return -ENOENT; 2784b18825a7SDavid Howells BUG_ON(!inode); 27851da177e4SLinus Torvalds 27861da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 27874fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27881da177e4SLinus Torvalds 2789f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27901da177e4SLinus Torvalds if (error) 27911da177e4SLinus Torvalds return error; 27921da177e4SLinus Torvalds if (IS_APPEND(dir)) 27931da177e4SLinus Torvalds return -EPERM; 2794b18825a7SDavid Howells 2795b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27960bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27971da177e4SLinus Torvalds return -EPERM; 27981da177e4SLinus Torvalds if (isdir) { 279944b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28001da177e4SLinus Torvalds return -ENOTDIR; 28011da177e4SLinus Torvalds if (IS_ROOT(victim)) 28021da177e4SLinus Torvalds return -EBUSY; 280344b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28041da177e4SLinus Torvalds return -EISDIR; 28051da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28061da177e4SLinus Torvalds return -ENOENT; 28071da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28081da177e4SLinus Torvalds return -EBUSY; 28091da177e4SLinus Torvalds return 0; 28101da177e4SLinus Torvalds } 28111da177e4SLinus Torvalds 28121da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28131da177e4SLinus Torvalds * dir. 28141da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28151da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28161da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 28171da177e4SLinus Torvalds * 3. We should have write and exec permissions on dir 28181da177e4SLinus Torvalds * 4. We can't do it if dir is immutable (done in permission()) 28191da177e4SLinus Torvalds */ 2820a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28211da177e4SLinus Torvalds { 282214e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28231da177e4SLinus Torvalds if (child->d_inode) 28241da177e4SLinus Torvalds return -EEXIST; 28251da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28261da177e4SLinus Torvalds return -ENOENT; 2827f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28281da177e4SLinus Torvalds } 28291da177e4SLinus Torvalds 28301da177e4SLinus Torvalds /* 28311da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28321da177e4SLinus Torvalds */ 28331da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28341da177e4SLinus Torvalds { 28351da177e4SLinus Torvalds struct dentry *p; 28361da177e4SLinus Torvalds 28371da177e4SLinus Torvalds if (p1 == p2) { 28385955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28391da177e4SLinus Torvalds return NULL; 28401da177e4SLinus Torvalds } 28411da177e4SLinus Torvalds 2842fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28431da177e4SLinus Torvalds 2844e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2845e2761a11SOGAWA Hirofumi if (p) { 28465955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28475955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28481da177e4SLinus Torvalds return p; 28491da177e4SLinus Torvalds } 28501da177e4SLinus Torvalds 2851e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2852e2761a11SOGAWA Hirofumi if (p) { 28535955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28545955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28551da177e4SLinus Torvalds return p; 28561da177e4SLinus Torvalds } 28571da177e4SLinus Torvalds 28585955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28595955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28601da177e4SLinus Torvalds return NULL; 28611da177e4SLinus Torvalds } 28624d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28631da177e4SLinus Torvalds 28641da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28651da177e4SLinus Torvalds { 28665955102cSAl Viro inode_unlock(p1->d_inode); 28671da177e4SLinus Torvalds if (p1 != p2) { 28685955102cSAl Viro inode_unlock(p2->d_inode); 2869fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28701da177e4SLinus Torvalds } 28711da177e4SLinus Torvalds } 28724d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28731da177e4SLinus Torvalds 28744acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2875312b63fbSAl Viro bool want_excl) 28761da177e4SLinus Torvalds { 2877a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28781da177e4SLinus Torvalds if (error) 28791da177e4SLinus Torvalds return error; 28801da177e4SLinus Torvalds 2881acfa4380SAl Viro if (!dir->i_op->create) 28821da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28831da177e4SLinus Torvalds mode &= S_IALLUGO; 28841da177e4SLinus Torvalds mode |= S_IFREG; 28851da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28861da177e4SLinus Torvalds if (error) 28871da177e4SLinus Torvalds return error; 2888312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2889a74574aaSStephen Smalley if (!error) 2890f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28911da177e4SLinus Torvalds return error; 28921da177e4SLinus Torvalds } 28934d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28941da177e4SLinus Torvalds 2895a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2896a2982cc9SEric W. Biederman { 2897a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2898a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2899a2982cc9SEric W. Biederman } 2900a2982cc9SEric W. Biederman 290173d049a4SAl Viro static int may_open(struct path *path, int acc_mode, int flag) 29021da177e4SLinus Torvalds { 29033fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29041da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29051da177e4SLinus Torvalds int error; 29061da177e4SLinus Torvalds 29071da177e4SLinus Torvalds if (!inode) 29081da177e4SLinus Torvalds return -ENOENT; 29091da177e4SLinus Torvalds 2910c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2911c8fe8f30SChristoph Hellwig case S_IFLNK: 29121da177e4SLinus Torvalds return -ELOOP; 2913c8fe8f30SChristoph Hellwig case S_IFDIR: 2914c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29151da177e4SLinus Torvalds return -EISDIR; 2916c8fe8f30SChristoph Hellwig break; 2917c8fe8f30SChristoph Hellwig case S_IFBLK: 2918c8fe8f30SChristoph Hellwig case S_IFCHR: 2919a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29201da177e4SLinus Torvalds return -EACCES; 2921c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2922c8fe8f30SChristoph Hellwig case S_IFIFO: 2923c8fe8f30SChristoph Hellwig case S_IFSOCK: 29241da177e4SLinus Torvalds flag &= ~O_TRUNC; 2925c8fe8f30SChristoph Hellwig break; 29264a3fd211SDave Hansen } 2927b41572e9SDave Hansen 292862fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2929b41572e9SDave Hansen if (error) 2930b41572e9SDave Hansen return error; 29316146f0d5SMimi Zohar 29321da177e4SLinus Torvalds /* 29331da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29341da177e4SLinus Torvalds */ 29351da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29368737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29377715b521SAl Viro return -EPERM; 29381da177e4SLinus Torvalds if (flag & O_TRUNC) 29397715b521SAl Viro return -EPERM; 29401da177e4SLinus Torvalds } 29411da177e4SLinus Torvalds 29421da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29432e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29447715b521SAl Viro return -EPERM; 29451da177e4SLinus Torvalds 2946f3c7691eSJ. Bruce Fields return 0; 29477715b521SAl Viro } 29487715b521SAl Viro 2949e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29507715b521SAl Viro { 2951e1181ee6SJeff Layton struct path *path = &filp->f_path; 29527715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29537715b521SAl Viro int error = get_write_access(inode); 29541da177e4SLinus Torvalds if (error) 29557715b521SAl Viro return error; 29561da177e4SLinus Torvalds /* 29571da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29581da177e4SLinus Torvalds */ 2959d7a06983SJeff Layton error = locks_verify_locked(filp); 2960be6d3e56SKentaro Takeda if (!error) 2961ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29621da177e4SLinus Torvalds if (!error) { 29637715b521SAl Viro error = do_truncate(path->dentry, 0, 2964d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2965e1181ee6SJeff Layton filp); 29661da177e4SLinus Torvalds } 29671da177e4SLinus Torvalds put_write_access(inode); 2968acd0c935SMimi Zohar return error; 29691da177e4SLinus Torvalds } 29701da177e4SLinus Torvalds 2971d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2972d57999e1SDave Hansen { 29738a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29748a5e929dSAl Viro flag--; 2975d57999e1SDave Hansen return flag; 2976d57999e1SDave Hansen } 2977d57999e1SDave Hansen 2978d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2979d18e9008SMiklos Szeredi { 2980d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2981d18e9008SMiklos Szeredi if (error) 2982d18e9008SMiklos Szeredi return error; 2983d18e9008SMiklos Szeredi 2984d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2985d18e9008SMiklos Szeredi if (error) 2986d18e9008SMiklos Szeredi return error; 2987d18e9008SMiklos Szeredi 2988d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2989d18e9008SMiklos Szeredi } 2990d18e9008SMiklos Szeredi 29911acf0af9SDavid Howells /* 29921acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29931acf0af9SDavid Howells * dentry. 29941acf0af9SDavid Howells * 29951acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29961acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29971acf0af9SDavid Howells * 29981acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 29991acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 30001acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 30011acf0af9SDavid Howells * 30021acf0af9SDavid Howells * Returns an error code otherwise. 30031acf0af9SDavid Howells */ 30042675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 300530d90494SAl Viro struct path *path, struct file *file, 3006015c3bbcSMiklos Szeredi const struct open_flags *op, 30071643b43fSAl Viro int open_flag, umode_t mode, 300847237687SAl Viro int *opened) 3009d18e9008SMiklos Szeredi { 3010d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3011d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3012d18e9008SMiklos Szeredi int error; 3013d18e9008SMiklos Szeredi 3014384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3015d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3016d18e9008SMiklos Szeredi 3017d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3018d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3019d18e9008SMiklos Szeredi 302030d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 302130d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30220fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 30230fb1ea09SAl Viro open_to_namei_flags(open_flag), 30240fb1ea09SAl Viro mode, opened); 30256fbd0714SAl Viro d_lookup_done(dentry); 3026384f26e2SAl Viro if (!error) { 3027d18e9008SMiklos Szeredi /* 3028384f26e2SAl Viro * We didn't have the inode before the open, so check open 3029384f26e2SAl Viro * permission here. 3030d18e9008SMiklos Szeredi */ 3031384f26e2SAl Viro int acc_mode = op->acc_mode; 303203da633aSAl Viro if (*opened & FILE_CREATED) { 303303da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 303403da633aSAl Viro fsnotify_create(dir, dentry); 303562fb4a15SAl Viro acc_mode = 0; 303603da633aSAl Viro } 30372675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3038384f26e2SAl Viro if (WARN_ON(error > 0)) 3039384f26e2SAl Viro error = -EINVAL; 3040384f26e2SAl Viro } else if (error > 0) { 3041d18e9008SMiklos Szeredi if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3042d18e9008SMiklos Szeredi error = -EIO; 3043384f26e2SAl Viro } else { 3044d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3045d18e9008SMiklos Szeredi dput(dentry); 3046d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 304710c64ceaSAl Viro } 3048d18e9008SMiklos Szeredi if (*opened & FILE_CREATED) 3049d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3050d18e9008SMiklos Szeredi path->dentry = dentry; 3051d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 30522675a4ebSAl Viro return 1; 3053d18e9008SMiklos Szeredi } 3054d18e9008SMiklos Szeredi } 3055d18e9008SMiklos Szeredi dput(dentry); 3056d18e9008SMiklos Szeredi return error; 3057d18e9008SMiklos Szeredi } 3058d18e9008SMiklos Szeredi 305931e6b01fSNick Piggin /* 30601acf0af9SDavid Howells * Look up and maybe create and open the last component. 3061d58ffd35SMiklos Szeredi * 3062d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 3063d58ffd35SMiklos Szeredi * 30641acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 30651acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 30661acf0af9SDavid Howells * 30671acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 30681acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 30691acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 30701acf0af9SDavid Howells * specified then a negative dentry may be returned. 30711acf0af9SDavid Howells * 30721acf0af9SDavid Howells * An error code is returned otherwise. 30731acf0af9SDavid Howells * 30741acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 30751acf0af9SDavid Howells * cleared otherwise prior to returning. 3076d58ffd35SMiklos Szeredi */ 30772675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 307830d90494SAl Viro struct file *file, 3079d58ffd35SMiklos Szeredi const struct open_flags *op, 308064894cf8SAl Viro bool got_write, int *opened) 3081d58ffd35SMiklos Szeredi { 3082d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 308354ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30841643b43fSAl Viro int open_flag = op->open_flag; 3085d58ffd35SMiklos Szeredi struct dentry *dentry; 30861643b43fSAl Viro int error, create_error = 0; 30871643b43fSAl Viro umode_t mode = op->mode; 30886fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3089d58ffd35SMiklos Szeredi 3090ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3091ce8644fcSAl Viro return -ENOENT; 3092d58ffd35SMiklos Szeredi 309347237687SAl Viro *opened &= ~FILE_CREATED; 30946fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30956fbd0714SAl Viro for (;;) { 30966fbd0714SAl Viro if (!dentry) { 30976fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3098d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 30992675a4ebSAl Viro return PTR_ERR(dentry); 31006fbd0714SAl Viro } 31016fbd0714SAl Viro if (d_in_lookup(dentry)) 31026fbd0714SAl Viro break; 3103d58ffd35SMiklos Szeredi 31046fbd0714SAl Viro if (!(dentry->d_flags & DCACHE_OP_REVALIDATE)) 31056fbd0714SAl Viro break; 31066fbd0714SAl Viro 31076fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31086fbd0714SAl Viro if (likely(error > 0)) 31096fbd0714SAl Viro break; 31106fbd0714SAl Viro if (error) 31116fbd0714SAl Viro goto out_dput; 31126fbd0714SAl Viro d_invalidate(dentry); 31136fbd0714SAl Viro dput(dentry); 31146fbd0714SAl Viro dentry = NULL; 31156fbd0714SAl Viro } 31166fbd0714SAl Viro if (dentry->d_inode) { 3117d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3118d18e9008SMiklos Szeredi goto out_no_open; 31196c51e513SAl Viro } 3120d18e9008SMiklos Szeredi 31211643b43fSAl Viro /* 31221643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31231643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31241643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31251643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31261643b43fSAl Viro * 31271643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31281643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31291643b43fSAl Viro */ 31301643b43fSAl Viro if (open_flag & O_CREAT) { 31311643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31321643b43fSAl Viro mode &= ~current_umask(); 31331643b43fSAl Viro if (unlikely(!got_write)) { 31341643b43fSAl Viro create_error = -EROFS; 31351643b43fSAl Viro open_flag &= ~O_CREAT; 31361643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31371643b43fSAl Viro goto no_open; 31381643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31391643b43fSAl Viro } else { 31401643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31411643b43fSAl Viro if (create_error) { 31421643b43fSAl Viro open_flag &= ~O_CREAT; 31431643b43fSAl Viro if (open_flag & O_EXCL) 31441643b43fSAl Viro goto no_open; 31451643b43fSAl Viro } 31461643b43fSAl Viro } 31471643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31481643b43fSAl Viro unlikely(!got_write)) { 31491643b43fSAl Viro /* 31501643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31511643b43fSAl Viro * back to lookup + open 31521643b43fSAl Viro */ 31531643b43fSAl Viro goto no_open; 3154d18e9008SMiklos Szeredi } 3155d18e9008SMiklos Szeredi 31561643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 31571643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 31581643b43fSAl Viro mode, opened); 31591643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 31601643b43fSAl Viro error = create_error; 31611643b43fSAl Viro return error; 31621643b43fSAl Viro } 316354ef4872SMiklos Szeredi 31641643b43fSAl Viro no_open: 31656fbd0714SAl Viro if (d_in_lookup(dentry)) { 316612fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 316712fa5e24SAl Viro nd->flags); 31686fbd0714SAl Viro d_lookup_done(dentry); 316912fa5e24SAl Viro if (unlikely(res)) { 317012fa5e24SAl Viro if (IS_ERR(res)) { 317112fa5e24SAl Viro error = PTR_ERR(res); 317212fa5e24SAl Viro goto out_dput; 317312fa5e24SAl Viro } 317412fa5e24SAl Viro dput(dentry); 317512fa5e24SAl Viro dentry = res; 317612fa5e24SAl Viro } 317754ef4872SMiklos Szeredi } 317854ef4872SMiklos Szeredi 3179d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 31801643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 318147237687SAl Viro *opened |= FILE_CREATED; 3182ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3183ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3184ce8644fcSAl Viro error = -EACCES; 3185d58ffd35SMiklos Szeredi goto out_dput; 318677d660a8SMiklos Szeredi } 3187ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31881643b43fSAl Viro open_flag & O_EXCL); 3189d58ffd35SMiklos Szeredi if (error) 3190d58ffd35SMiklos Szeredi goto out_dput; 3191ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3192d58ffd35SMiklos Szeredi } 31931643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31941643b43fSAl Viro error = create_error; 3195d58ffd35SMiklos Szeredi goto out_dput; 3196d58ffd35SMiklos Szeredi } 3197d18e9008SMiklos Szeredi out_no_open: 3198d58ffd35SMiklos Szeredi path->dentry = dentry; 3199d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 32002675a4ebSAl Viro return 1; 3201d58ffd35SMiklos Szeredi 3202d58ffd35SMiklos Szeredi out_dput: 3203d58ffd35SMiklos Szeredi dput(dentry); 32042675a4ebSAl Viro return error; 3205d58ffd35SMiklos Szeredi } 3206d58ffd35SMiklos Szeredi 3207d58ffd35SMiklos Szeredi /* 3208fe2d35ffSAl Viro * Handle the last step of open() 320931e6b01fSNick Piggin */ 3210896475d5SAl Viro static int do_last(struct nameidata *nd, 321130d90494SAl Viro struct file *file, const struct open_flags *op, 321276ae2a5aSAl Viro int *opened) 3213fb1cc555SAl Viro { 3214a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3215ca344a89SAl Viro int open_flag = op->open_flag; 321677d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 321764894cf8SAl Viro bool got_write = false; 3218bcda7652SAl Viro int acc_mode = op->acc_mode; 3219254cf582SAl Viro unsigned seq; 3220a1eb3315SMiklos Szeredi struct inode *inode; 322116b1c1cdSMiklos Szeredi struct path save_parent = { .dentry = NULL, .mnt = NULL }; 3222896475d5SAl Viro struct path path; 322316b1c1cdSMiklos Szeredi bool retried = false; 322416c2cd71SAl Viro int error; 3225fb1cc555SAl Viro 3226c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3227c3e380b0SAl Viro nd->flags |= op->intent; 3228c3e380b0SAl Viro 3229bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3230fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3231deb106c6SAl Viro if (unlikely(error)) 32322675a4ebSAl Viro return error; 3233e83db167SMiklos Szeredi goto finish_open; 32341f36f774SAl Viro } 3235a2c36b45SAl Viro 3236ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3237fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3238fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3239fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3240254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3241e9742b53SAl Viro if (likely(error > 0)) 324271574865SMiklos Szeredi goto finish_lookup; 324371574865SMiklos Szeredi 3244ce57dfc1SAl Viro if (error < 0) 3245deb106c6SAl Viro return error; 3246a1eb3315SMiklos Szeredi 324737d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32486583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3249b6183df7SMiklos Szeredi } else { 3250fe2d35ffSAl Viro /* create side of things */ 3251a3fbbde7SAl Viro /* 3252b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3253b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3254b6183df7SMiklos Szeredi * about to look up 3255a3fbbde7SAl Viro */ 32569f1fafeeSAl Viro error = complete_walk(nd); 3257e8bb73dfSAl Viro if (error) 32582675a4ebSAl Viro return error; 3259fe2d35ffSAl Viro 326076ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 32611f36f774SAl Viro /* trailing slashes? */ 3262deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3263deb106c6SAl Viro return -EISDIR; 3264b6183df7SMiklos Szeredi } 32651f36f774SAl Viro 326616b1c1cdSMiklos Szeredi retry_lookup: 32679cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 326864894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 326964894cf8SAl Viro if (!error) 327064894cf8SAl Viro got_write = true; 327164894cf8SAl Viro /* 327264894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 327364894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 327464894cf8SAl Viro * dropping this one anyway. 327564894cf8SAl Viro */ 327664894cf8SAl Viro } 32779cf843e3SAl Viro if (open_flag & O_CREAT) 32785955102cSAl Viro inode_lock(dir->d_inode); 32799cf843e3SAl Viro else 32809cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3281896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 32829cf843e3SAl Viro if (open_flag & O_CREAT) 32835955102cSAl Viro inode_unlock(dir->d_inode); 32849cf843e3SAl Viro else 32859cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3286fb1cc555SAl Viro 32872675a4ebSAl Viro if (error <= 0) { 32882675a4ebSAl Viro if (error) 3289d58ffd35SMiklos Szeredi goto out; 32906c0d46c4SAl Viro 329147237687SAl Viro if ((*opened & FILE_CREATED) || 3292496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 329377d660a8SMiklos Szeredi will_truncate = false; 3294d18e9008SMiklos Szeredi 329576ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3296d18e9008SMiklos Szeredi goto opened; 3297d18e9008SMiklos Szeredi } 3298d18e9008SMiklos Szeredi 329947237687SAl Viro if (*opened & FILE_CREATED) { 33009b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3301ca344a89SAl Viro open_flag &= ~O_TRUNC; 330277d660a8SMiklos Szeredi will_truncate = false; 330362fb4a15SAl Viro acc_mode = 0; 3304896475d5SAl Viro path_to_nameidata(&path, nd); 3305e83db167SMiklos Szeredi goto finish_open_created; 3306fb1cc555SAl Viro } 3307fb1cc555SAl Viro 3308fb1cc555SAl Viro /* 3309d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3310d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3311d18e9008SMiklos Szeredi * necessary...) 3312d18e9008SMiklos Szeredi */ 331364894cf8SAl Viro if (got_write) { 3314d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 331564894cf8SAl Viro got_write = false; 3316d18e9008SMiklos Szeredi } 3317d18e9008SMiklos Szeredi 33186583fe22SAl Viro if (unlikely(d_is_negative(path.dentry))) { 33196583fe22SAl Viro path_to_nameidata(&path, nd); 33206583fe22SAl Viro return -ENOENT; 33216583fe22SAl Viro } 33226583fe22SAl Viro 33236583fe22SAl Viro /* 33246583fe22SAl Viro * create/update audit record if it already exists. 33256583fe22SAl Viro */ 33266583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 33276583fe22SAl Viro 3328deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3329deb106c6SAl Viro path_to_nameidata(&path, nd); 3330deb106c6SAl Viro return -EEXIST; 3331deb106c6SAl Viro } 3332fb1cc555SAl Viro 3333756daf26SNeilBrown error = follow_managed(&path, nd); 3334deb106c6SAl Viro if (unlikely(error < 0)) 3335deb106c6SAl Viro return error; 3336a3fbbde7SAl Viro 3337254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3338d4565649SAl Viro inode = d_backing_inode(path.dentry); 3339766c4cbfSAl Viro finish_lookup: 3340191d7f73SAl Viro if (nd->depth) 3341191d7f73SAl Viro put_link(nd); 3342181548c0SAl Viro error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW, 3343181548c0SAl Viro inode, seq); 3344deb106c6SAl Viro if (unlikely(error)) 3345d63ff28fSAl Viro return error; 3346fb1cc555SAl Viro 3347896475d5SAl Viro if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) { 3348896475d5SAl Viro path_to_nameidata(&path, nd); 334916b1c1cdSMiklos Szeredi } else { 335016b1c1cdSMiklos Szeredi save_parent.dentry = nd->path.dentry; 3351896475d5SAl Viro save_parent.mnt = mntget(path.mnt); 3352896475d5SAl Viro nd->path.dentry = path.dentry; 335316b1c1cdSMiklos Szeredi 335416b1c1cdSMiklos Szeredi } 3355decf3400SMiklos Szeredi nd->inode = inode; 3356254cf582SAl Viro nd->seq = seq; 3357a3fbbde7SAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3358bc77daa7SAl Viro finish_open: 3359a3fbbde7SAl Viro error = complete_walk(nd); 336016b1c1cdSMiklos Szeredi if (error) { 336116b1c1cdSMiklos Szeredi path_put(&save_parent); 33622675a4ebSAl Viro return error; 336316b1c1cdSMiklos Szeredi } 336476ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3365fb1cc555SAl Viro error = -EISDIR; 336644b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 33672675a4ebSAl Viro goto out; 3368af2f5542SMiklos Szeredi error = -ENOTDIR; 336944b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 33702675a4ebSAl Viro goto out; 33714bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 337277d660a8SMiklos Szeredi will_truncate = false; 33736c0d46c4SAl Viro 33740f9d1a10SAl Viro if (will_truncate) { 33750f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33760f9d1a10SAl Viro if (error) 33772675a4ebSAl Viro goto out; 337864894cf8SAl Viro got_write = true; 33790f9d1a10SAl Viro } 3380e83db167SMiklos Szeredi finish_open_created: 3381bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3382ca344a89SAl Viro if (error) 33832675a4ebSAl Viro goto out; 33844aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 33854aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 33864aa7c634SMiklos Szeredi if (!error) { 33874aa7c634SMiklos Szeredi *opened |= FILE_OPENED; 33884aa7c634SMiklos Szeredi } else { 338930d90494SAl Viro if (error == -EOPENSTALE) 3390f60dc3dbSMiklos Szeredi goto stale_open; 3391015c3bbcSMiklos Szeredi goto out; 3392f60dc3dbSMiklos Szeredi } 3393a8277b9bSMiklos Szeredi opened: 33942675a4ebSAl Viro error = open_check_o_direct(file); 3395fe9ec829SAl Viro if (!error) 33963034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3397fe9ec829SAl Viro if (!error && will_truncate) 33982675a4ebSAl Viro error = handle_truncate(file); 3399ca344a89SAl Viro out: 3400fe9ec829SAl Viro if (unlikely(error) && (*opened & FILE_OPENED)) 3401fe9ec829SAl Viro fput(file); 3402c80567c8SAl Viro if (unlikely(error > 0)) { 3403c80567c8SAl Viro WARN_ON(1); 3404c80567c8SAl Viro error = -EINVAL; 3405c80567c8SAl Viro } 340664894cf8SAl Viro if (got_write) 34070f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 340816b1c1cdSMiklos Szeredi path_put(&save_parent); 34092675a4ebSAl Viro return error; 3410fb1cc555SAl Viro 3411f60dc3dbSMiklos Szeredi stale_open: 3412f60dc3dbSMiklos Szeredi /* If no saved parent or already retried then can't retry */ 3413f60dc3dbSMiklos Szeredi if (!save_parent.dentry || retried) 3414f60dc3dbSMiklos Szeredi goto out; 3415f60dc3dbSMiklos Szeredi 3416f60dc3dbSMiklos Szeredi BUG_ON(save_parent.dentry != dir); 3417f60dc3dbSMiklos Szeredi path_put(&nd->path); 3418f60dc3dbSMiklos Szeredi nd->path = save_parent; 3419f60dc3dbSMiklos Szeredi nd->inode = dir->d_inode; 3420f60dc3dbSMiklos Szeredi save_parent.mnt = NULL; 3421f60dc3dbSMiklos Szeredi save_parent.dentry = NULL; 342264894cf8SAl Viro if (got_write) { 3423f60dc3dbSMiklos Szeredi mnt_drop_write(nd->path.mnt); 342464894cf8SAl Viro got_write = false; 3425f60dc3dbSMiklos Szeredi } 3426f60dc3dbSMiklos Szeredi retried = true; 3427f60dc3dbSMiklos Szeredi goto retry_lookup; 3428fb1cc555SAl Viro } 3429fb1cc555SAl Viro 3430c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 343160545d0dSAl Viro const struct open_flags *op, 343260545d0dSAl Viro struct file *file, int *opened) 343360545d0dSAl Viro { 343460545d0dSAl Viro static const struct qstr name = QSTR_INIT("/", 1); 3435625b6d10SAl Viro struct dentry *child; 343660545d0dSAl Viro struct inode *dir; 3437625b6d10SAl Viro struct path path; 3438c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 343960545d0dSAl Viro if (unlikely(error)) 344060545d0dSAl Viro return error; 3441625b6d10SAl Viro error = mnt_want_write(path.mnt); 344260545d0dSAl Viro if (unlikely(error)) 344360545d0dSAl Viro goto out; 3444625b6d10SAl Viro dir = path.dentry->d_inode; 344560545d0dSAl Viro /* we want directory to be writable */ 3446625b6d10SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 344760545d0dSAl Viro if (error) 344860545d0dSAl Viro goto out2; 344960545d0dSAl Viro if (!dir->i_op->tmpfile) { 345060545d0dSAl Viro error = -EOPNOTSUPP; 345160545d0dSAl Viro goto out2; 345260545d0dSAl Viro } 3453625b6d10SAl Viro child = d_alloc(path.dentry, &name); 345460545d0dSAl Viro if (unlikely(!child)) { 345560545d0dSAl Viro error = -ENOMEM; 345660545d0dSAl Viro goto out2; 345760545d0dSAl Viro } 3458625b6d10SAl Viro dput(path.dentry); 3459625b6d10SAl Viro path.dentry = child; 3460625b6d10SAl Viro error = dir->i_op->tmpfile(dir, child, op->mode); 346160545d0dSAl Viro if (error) 346260545d0dSAl Viro goto out2; 3463c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 346469a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 346562fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 346660545d0dSAl Viro if (error) 346760545d0dSAl Viro goto out2; 3468625b6d10SAl Viro file->f_path.mnt = path.mnt; 3469625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 347060545d0dSAl Viro if (error) 347160545d0dSAl Viro goto out2; 347260545d0dSAl Viro error = open_check_o_direct(file); 3473f4e0c30cSAl Viro if (error) { 347460545d0dSAl Viro fput(file); 3475f4e0c30cSAl Viro } else if (!(op->open_flag & O_EXCL)) { 3476f4e0c30cSAl Viro struct inode *inode = file_inode(file); 3477f4e0c30cSAl Viro spin_lock(&inode->i_lock); 3478f4e0c30cSAl Viro inode->i_state |= I_LINKABLE; 3479f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 3480f4e0c30cSAl Viro } 348160545d0dSAl Viro out2: 3482625b6d10SAl Viro mnt_drop_write(path.mnt); 348360545d0dSAl Viro out: 3484625b6d10SAl Viro path_put(&path); 348560545d0dSAl Viro return error; 348660545d0dSAl Viro } 348760545d0dSAl Viro 34886ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34896ac08709SAl Viro { 34906ac08709SAl Viro struct path path; 34916ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34926ac08709SAl Viro if (!error) { 34936ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 34946ac08709SAl Viro error = vfs_open(&path, file, current_cred()); 34956ac08709SAl Viro path_put(&path); 34966ac08709SAl Viro } 34976ac08709SAl Viro return error; 34986ac08709SAl Viro } 34996ac08709SAl Viro 3500c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3501c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35021da177e4SLinus Torvalds { 3503368ee9baSAl Viro const char *s; 350430d90494SAl Viro struct file *file; 350547237687SAl Viro int opened = 0; 350613aab428SAl Viro int error; 350731e6b01fSNick Piggin 350830d90494SAl Viro file = get_empty_filp(); 35091afc99beSAl Viro if (IS_ERR(file)) 35101afc99beSAl Viro return file; 351131e6b01fSNick Piggin 351230d90494SAl Viro file->f_flags = op->open_flag; 351331e6b01fSNick Piggin 3514bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3515c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3516f15133dfSAl Viro goto out2; 351760545d0dSAl Viro } 351860545d0dSAl Viro 35196ac08709SAl Viro if (unlikely(file->f_flags & O_PATH)) { 35206ac08709SAl Viro error = do_o_path(nd, flags, file); 35216ac08709SAl Viro if (!error) 35226ac08709SAl Viro opened |= FILE_OPENED; 35236ac08709SAl Viro goto out2; 35246ac08709SAl Viro } 35256ac08709SAl Viro 3526c8a53ee5SAl Viro s = path_init(nd, flags); 3527368ee9baSAl Viro if (IS_ERR(s)) { 3528368ee9baSAl Viro put_filp(file); 3529368ee9baSAl Viro return ERR_CAST(s); 3530368ee9baSAl Viro } 35313bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 353276ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 353373d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35343bdba28bSAl Viro s = trailing_symlink(nd); 35353bdba28bSAl Viro if (IS_ERR(s)) { 35363bdba28bSAl Viro error = PTR_ERR(s); 35372675a4ebSAl Viro break; 3538806b681cSAl Viro } 35393bdba28bSAl Viro } 3540deb106c6SAl Viro terminate_walk(nd); 3541f15133dfSAl Viro out2: 35422675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 35432675a4ebSAl Viro BUG_ON(!error); 354430d90494SAl Viro put_filp(file); 3545015c3bbcSMiklos Szeredi } 35462675a4ebSAl Viro if (unlikely(error)) { 35472675a4ebSAl Viro if (error == -EOPENSTALE) { 35482675a4ebSAl Viro if (flags & LOOKUP_RCU) 35492675a4ebSAl Viro error = -ECHILD; 35502675a4ebSAl Viro else 35512675a4ebSAl Viro error = -ESTALE; 35522675a4ebSAl Viro } 35532675a4ebSAl Viro file = ERR_PTR(error); 35542675a4ebSAl Viro } 35552675a4ebSAl Viro return file; 3556de459215SKirill Korotaev } 35571da177e4SLinus Torvalds 3558669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3559f9652e10SAl Viro const struct open_flags *op) 356013aab428SAl Viro { 35619883d185SAl Viro struct nameidata nd; 3562f9652e10SAl Viro int flags = op->lookup_flags; 356313aab428SAl Viro struct file *filp; 356413aab428SAl Viro 35659883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3566c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 356713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3568c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 356913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3570c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35719883d185SAl Viro restore_nameidata(); 357213aab428SAl Viro return filp; 357313aab428SAl Viro } 357413aab428SAl Viro 357573d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3576f9652e10SAl Viro const char *name, const struct open_flags *op) 357773d049a4SAl Viro { 35789883d185SAl Viro struct nameidata nd; 357973d049a4SAl Viro struct file *file; 358051689104SPaul Moore struct filename *filename; 3581f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 358273d049a4SAl Viro 358373d049a4SAl Viro nd.root.mnt = mnt; 358473d049a4SAl Viro nd.root.dentry = dentry; 358573d049a4SAl Viro 3586b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 358773d049a4SAl Viro return ERR_PTR(-ELOOP); 358873d049a4SAl Viro 358951689104SPaul Moore filename = getname_kernel(name); 3590a1c83681SViresh Kumar if (IS_ERR(filename)) 359151689104SPaul Moore return ERR_CAST(filename); 359251689104SPaul Moore 35939883d185SAl Viro set_nameidata(&nd, -1, filename); 3594c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 359573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3596c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 359773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3598c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35999883d185SAl Viro restore_nameidata(); 360051689104SPaul Moore putname(filename); 360173d049a4SAl Viro return file; 360273d049a4SAl Viro } 360373d049a4SAl Viro 3604fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 36051ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36061da177e4SLinus Torvalds { 3607c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3608391172c4SAl Viro struct qstr last; 3609391172c4SAl Viro int type; 3610c30dabfeSJan Kara int err2; 36111ac12b4bSJeff Layton int error; 36121ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36131ac12b4bSJeff Layton 36141ac12b4bSJeff Layton /* 36151ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36161ac12b4bSJeff Layton * other flags passed in are ignored! 36171ac12b4bSJeff Layton */ 36181ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36191ac12b4bSJeff Layton 36205c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36215c31b6ceSAl Viro if (IS_ERR(name)) 36225c31b6ceSAl Viro return ERR_CAST(name); 36231da177e4SLinus Torvalds 3624c663e5d8SChristoph Hellwig /* 3625c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3626c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3627c663e5d8SChristoph Hellwig */ 36285c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3629ed75e95dSAl Viro goto out; 3630c663e5d8SChristoph Hellwig 3631c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3632391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3633c663e5d8SChristoph Hellwig /* 3634c663e5d8SChristoph Hellwig * Do the final lookup. 3635c663e5d8SChristoph Hellwig */ 3636391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36375955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3638391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36391da177e4SLinus Torvalds if (IS_ERR(dentry)) 3640a8104a9fSAl Viro goto unlock; 3641c663e5d8SChristoph Hellwig 3642a8104a9fSAl Viro error = -EEXIST; 3643b18825a7SDavid Howells if (d_is_positive(dentry)) 3644a8104a9fSAl Viro goto fail; 3645b18825a7SDavid Howells 3646c663e5d8SChristoph Hellwig /* 3647c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3648c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3649c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3650c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3651c663e5d8SChristoph Hellwig */ 3652391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3653a8104a9fSAl Viro error = -ENOENT; 3654ed75e95dSAl Viro goto fail; 3655e9baf6e5SAl Viro } 3656c30dabfeSJan Kara if (unlikely(err2)) { 3657c30dabfeSJan Kara error = err2; 3658a8104a9fSAl Viro goto fail; 3659c30dabfeSJan Kara } 3660181c37b6SAl Viro putname(name); 3661e9baf6e5SAl Viro return dentry; 36621da177e4SLinus Torvalds fail: 3663a8104a9fSAl Viro dput(dentry); 3664a8104a9fSAl Viro dentry = ERR_PTR(error); 3665a8104a9fSAl Viro unlock: 36665955102cSAl Viro inode_unlock(path->dentry->d_inode); 3667c30dabfeSJan Kara if (!err2) 3668391172c4SAl Viro mnt_drop_write(path->mnt); 3669ed75e95dSAl Viro out: 3670391172c4SAl Viro path_put(path); 3671181c37b6SAl Viro putname(name); 3672ed75e95dSAl Viro return dentry; 3673dae6ad8fSAl Viro } 3674fa14a0b8SAl Viro 3675fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3676fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3677fa14a0b8SAl Viro { 3678181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3679181c37b6SAl Viro path, lookup_flags); 3680fa14a0b8SAl Viro } 3681dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3682dae6ad8fSAl Viro 3683921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3684921a1650SAl Viro { 3685921a1650SAl Viro dput(dentry); 36865955102cSAl Viro inode_unlock(path->dentry->d_inode); 3687a8104a9fSAl Viro mnt_drop_write(path->mnt); 3688921a1650SAl Viro path_put(path); 3689921a1650SAl Viro } 3690921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3691921a1650SAl Viro 3692520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36931ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3694dae6ad8fSAl Viro { 3695181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3696dae6ad8fSAl Viro } 3697dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3698dae6ad8fSAl Viro 36991a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 37001da177e4SLinus Torvalds { 3701a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37021da177e4SLinus Torvalds 37031da177e4SLinus Torvalds if (error) 37041da177e4SLinus Torvalds return error; 37051da177e4SLinus Torvalds 3706975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 37071da177e4SLinus Torvalds return -EPERM; 37081da177e4SLinus Torvalds 3709acfa4380SAl Viro if (!dir->i_op->mknod) 37101da177e4SLinus Torvalds return -EPERM; 37111da177e4SLinus Torvalds 371208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 371308ce5f16SSerge E. Hallyn if (error) 371408ce5f16SSerge E. Hallyn return error; 371508ce5f16SSerge E. Hallyn 37161da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37171da177e4SLinus Torvalds if (error) 37181da177e4SLinus Torvalds return error; 37191da177e4SLinus Torvalds 37201da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3721a74574aaSStephen Smalley if (!error) 3722f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37231da177e4SLinus Torvalds return error; 37241da177e4SLinus Torvalds } 37254d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37261da177e4SLinus Torvalds 3727f69aac00SAl Viro static int may_mknod(umode_t mode) 3728463c3197SDave Hansen { 3729463c3197SDave Hansen switch (mode & S_IFMT) { 3730463c3197SDave Hansen case S_IFREG: 3731463c3197SDave Hansen case S_IFCHR: 3732463c3197SDave Hansen case S_IFBLK: 3733463c3197SDave Hansen case S_IFIFO: 3734463c3197SDave Hansen case S_IFSOCK: 3735463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3736463c3197SDave Hansen return 0; 3737463c3197SDave Hansen case S_IFDIR: 3738463c3197SDave Hansen return -EPERM; 3739463c3197SDave Hansen default: 3740463c3197SDave Hansen return -EINVAL; 3741463c3197SDave Hansen } 3742463c3197SDave Hansen } 3743463c3197SDave Hansen 37448208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 37452e4d0924SHeiko Carstens unsigned, dev) 37461da177e4SLinus Torvalds { 37471da177e4SLinus Torvalds struct dentry *dentry; 3748dae6ad8fSAl Viro struct path path; 3749dae6ad8fSAl Viro int error; 3750972567f1SJeff Layton unsigned int lookup_flags = 0; 37511da177e4SLinus Torvalds 37528e4bfca1SAl Viro error = may_mknod(mode); 37538e4bfca1SAl Viro if (error) 37548e4bfca1SAl Viro return error; 3755972567f1SJeff Layton retry: 3756972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3757dae6ad8fSAl Viro if (IS_ERR(dentry)) 3758dae6ad8fSAl Viro return PTR_ERR(dentry); 37592ad94ae6SAl Viro 3760dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3761ce3b0f8dSAl Viro mode &= ~current_umask(); 3762dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3763be6d3e56SKentaro Takeda if (error) 3764a8104a9fSAl Viro goto out; 37651da177e4SLinus Torvalds switch (mode & S_IFMT) { 37661da177e4SLinus Torvalds case 0: case S_IFREG: 3767312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 376805d1a717SMimi Zohar if (!error) 376905d1a717SMimi Zohar ima_post_path_mknod(dentry); 37701da177e4SLinus Torvalds break; 37711da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3772dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37731da177e4SLinus Torvalds new_decode_dev(dev)); 37741da177e4SLinus Torvalds break; 37751da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3776dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37771da177e4SLinus Torvalds break; 37781da177e4SLinus Torvalds } 3779a8104a9fSAl Viro out: 3780921a1650SAl Viro done_path_create(&path, dentry); 3781972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3782972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3783972567f1SJeff Layton goto retry; 3784972567f1SJeff Layton } 37851da177e4SLinus Torvalds return error; 37861da177e4SLinus Torvalds } 37871da177e4SLinus Torvalds 37888208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37895590ff0dSUlrich Drepper { 37905590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 37915590ff0dSUlrich Drepper } 37925590ff0dSUlrich Drepper 379318bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37941da177e4SLinus Torvalds { 3795a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37968de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37971da177e4SLinus Torvalds 37981da177e4SLinus Torvalds if (error) 37991da177e4SLinus Torvalds return error; 38001da177e4SLinus Torvalds 3801acfa4380SAl Viro if (!dir->i_op->mkdir) 38021da177e4SLinus Torvalds return -EPERM; 38031da177e4SLinus Torvalds 38041da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38051da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38061da177e4SLinus Torvalds if (error) 38071da177e4SLinus Torvalds return error; 38081da177e4SLinus Torvalds 38098de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38108de52778SAl Viro return -EMLINK; 38118de52778SAl Viro 38121da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3813a74574aaSStephen Smalley if (!error) 3814f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38151da177e4SLinus Torvalds return error; 38161da177e4SLinus Torvalds } 38174d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38181da177e4SLinus Torvalds 3819a218d0fdSAl Viro SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38201da177e4SLinus Torvalds { 38216902d925SDave Hansen struct dentry *dentry; 3822dae6ad8fSAl Viro struct path path; 3823dae6ad8fSAl Viro int error; 3824b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38251da177e4SLinus Torvalds 3826b76d8b82SJeff Layton retry: 3827b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38286902d925SDave Hansen if (IS_ERR(dentry)) 3829dae6ad8fSAl Viro return PTR_ERR(dentry); 38306902d925SDave Hansen 3831dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3832ce3b0f8dSAl Viro mode &= ~current_umask(); 3833dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3834a8104a9fSAl Viro if (!error) 3835dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3836921a1650SAl Viro done_path_create(&path, dentry); 3837b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3838b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3839b76d8b82SJeff Layton goto retry; 3840b76d8b82SJeff Layton } 38411da177e4SLinus Torvalds return error; 38421da177e4SLinus Torvalds } 38431da177e4SLinus Torvalds 3844a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38455590ff0dSUlrich Drepper { 38465590ff0dSUlrich Drepper return sys_mkdirat(AT_FDCWD, pathname, mode); 38475590ff0dSUlrich Drepper } 38485590ff0dSUlrich Drepper 38491da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38501da177e4SLinus Torvalds { 38511da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38521da177e4SLinus Torvalds 38531da177e4SLinus Torvalds if (error) 38541da177e4SLinus Torvalds return error; 38551da177e4SLinus Torvalds 3856acfa4380SAl Viro if (!dir->i_op->rmdir) 38571da177e4SLinus Torvalds return -EPERM; 38581da177e4SLinus Torvalds 38591d2ef590SAl Viro dget(dentry); 38605955102cSAl Viro inode_lock(dentry->d_inode); 3861912dbc15SSage Weil 38621da177e4SLinus Torvalds error = -EBUSY; 38637af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3864912dbc15SSage Weil goto out; 3865912dbc15SSage Weil 38661da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3867912dbc15SSage Weil if (error) 3868912dbc15SSage Weil goto out; 3869912dbc15SSage Weil 38703cebde24SSage Weil shrink_dcache_parent(dentry); 38711da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3872912dbc15SSage Weil if (error) 3873912dbc15SSage Weil goto out; 3874912dbc15SSage Weil 38751da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3876d83c49f3SAl Viro dont_mount(dentry); 38778ed936b5SEric W. Biederman detach_mounts(dentry); 38781da177e4SLinus Torvalds 3879912dbc15SSage Weil out: 38805955102cSAl Viro inode_unlock(dentry->d_inode); 38811d2ef590SAl Viro dput(dentry); 3882912dbc15SSage Weil if (!error) 3883912dbc15SSage Weil d_delete(dentry); 38841da177e4SLinus Torvalds return error; 38851da177e4SLinus Torvalds } 38864d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38871da177e4SLinus Torvalds 38885590ff0dSUlrich Drepper static long do_rmdir(int dfd, const char __user *pathname) 38891da177e4SLinus Torvalds { 38901da177e4SLinus Torvalds int error = 0; 389191a27b2aSJeff Layton struct filename *name; 38921da177e4SLinus Torvalds struct dentry *dentry; 3893f5beed75SAl Viro struct path path; 3894f5beed75SAl Viro struct qstr last; 3895f5beed75SAl Viro int type; 3896c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3897c6ee9206SJeff Layton retry: 3898f5beed75SAl Viro name = user_path_parent(dfd, pathname, 3899f5beed75SAl Viro &path, &last, &type, lookup_flags); 390091a27b2aSJeff Layton if (IS_ERR(name)) 390191a27b2aSJeff Layton return PTR_ERR(name); 39021da177e4SLinus Torvalds 3903f5beed75SAl Viro switch (type) { 39041da177e4SLinus Torvalds case LAST_DOTDOT: 39051da177e4SLinus Torvalds error = -ENOTEMPTY; 39061da177e4SLinus Torvalds goto exit1; 39071da177e4SLinus Torvalds case LAST_DOT: 39081da177e4SLinus Torvalds error = -EINVAL; 39091da177e4SLinus Torvalds goto exit1; 39101da177e4SLinus Torvalds case LAST_ROOT: 39111da177e4SLinus Torvalds error = -EBUSY; 39121da177e4SLinus Torvalds goto exit1; 39131da177e4SLinus Torvalds } 39140612d9fbSOGAWA Hirofumi 3915f5beed75SAl Viro error = mnt_want_write(path.mnt); 3916c30dabfeSJan Kara if (error) 3917c30dabfeSJan Kara goto exit1; 39180612d9fbSOGAWA Hirofumi 39195955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3920f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39211da177e4SLinus Torvalds error = PTR_ERR(dentry); 39226902d925SDave Hansen if (IS_ERR(dentry)) 39236902d925SDave Hansen goto exit2; 3924e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3925e6bc45d6STheodore Ts'o error = -ENOENT; 3926e6bc45d6STheodore Ts'o goto exit3; 3927e6bc45d6STheodore Ts'o } 3928f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3929be6d3e56SKentaro Takeda if (error) 3930c30dabfeSJan Kara goto exit3; 3931f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39320622753bSDave Hansen exit3: 39331da177e4SLinus Torvalds dput(dentry); 39346902d925SDave Hansen exit2: 39355955102cSAl Viro inode_unlock(path.dentry->d_inode); 3936f5beed75SAl Viro mnt_drop_write(path.mnt); 39371da177e4SLinus Torvalds exit1: 3938f5beed75SAl Viro path_put(&path); 39391da177e4SLinus Torvalds putname(name); 3940c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3941c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3942c6ee9206SJeff Layton goto retry; 3943c6ee9206SJeff Layton } 39441da177e4SLinus Torvalds return error; 39451da177e4SLinus Torvalds } 39461da177e4SLinus Torvalds 39473cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39485590ff0dSUlrich Drepper { 39495590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39505590ff0dSUlrich Drepper } 39515590ff0dSUlrich Drepper 3952b21996e3SJ. Bruce Fields /** 3953b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3954b21996e3SJ. Bruce Fields * @dir: parent directory 3955b21996e3SJ. Bruce Fields * @dentry: victim 3956b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3957b21996e3SJ. Bruce Fields * 3958b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3959b21996e3SJ. Bruce Fields * 3960b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3961b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3962b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3963b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3964b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3965b21996e3SJ. Bruce Fields * 3966b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3967b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3968b21996e3SJ. Bruce Fields * to be NFS exported. 3969b21996e3SJ. Bruce Fields */ 3970b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39711da177e4SLinus Torvalds { 39729accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39731da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39741da177e4SLinus Torvalds 39751da177e4SLinus Torvalds if (error) 39761da177e4SLinus Torvalds return error; 39771da177e4SLinus Torvalds 3978acfa4380SAl Viro if (!dir->i_op->unlink) 39791da177e4SLinus Torvalds return -EPERM; 39801da177e4SLinus Torvalds 39815955102cSAl Viro inode_lock(target); 39828ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39831da177e4SLinus Torvalds error = -EBUSY; 39841da177e4SLinus Torvalds else { 39851da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3986bec1052eSAl Viro if (!error) { 39875a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39885a14696cSJ. Bruce Fields if (error) 3989b21996e3SJ. Bruce Fields goto out; 39901da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39918ed936b5SEric W. Biederman if (!error) { 3992d83c49f3SAl Viro dont_mount(dentry); 39938ed936b5SEric W. Biederman detach_mounts(dentry); 39948ed936b5SEric W. Biederman } 3995bec1052eSAl Viro } 39961da177e4SLinus Torvalds } 3997b21996e3SJ. Bruce Fields out: 39985955102cSAl Viro inode_unlock(target); 39991da177e4SLinus Torvalds 40001da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40011da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40029accbb97SJ. Bruce Fields fsnotify_link_count(target); 40031da177e4SLinus Torvalds d_delete(dentry); 40041da177e4SLinus Torvalds } 40050eeca283SRobert Love 40061da177e4SLinus Torvalds return error; 40071da177e4SLinus Torvalds } 40084d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40091da177e4SLinus Torvalds 40101da177e4SLinus Torvalds /* 40111da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40121b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40131da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40141da177e4SLinus Torvalds * while waiting on the I/O. 40151da177e4SLinus Torvalds */ 40165590ff0dSUlrich Drepper static long do_unlinkat(int dfd, const char __user *pathname) 40171da177e4SLinus Torvalds { 40182ad94ae6SAl Viro int error; 401991a27b2aSJeff Layton struct filename *name; 40201da177e4SLinus Torvalds struct dentry *dentry; 4021f5beed75SAl Viro struct path path; 4022f5beed75SAl Viro struct qstr last; 4023f5beed75SAl Viro int type; 40241da177e4SLinus Torvalds struct inode *inode = NULL; 4025b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40265d18f813SJeff Layton unsigned int lookup_flags = 0; 40275d18f813SJeff Layton retry: 4028f5beed75SAl Viro name = user_path_parent(dfd, pathname, 4029f5beed75SAl Viro &path, &last, &type, lookup_flags); 403091a27b2aSJeff Layton if (IS_ERR(name)) 403191a27b2aSJeff Layton return PTR_ERR(name); 40322ad94ae6SAl Viro 40331da177e4SLinus Torvalds error = -EISDIR; 4034f5beed75SAl Viro if (type != LAST_NORM) 40351da177e4SLinus Torvalds goto exit1; 40360612d9fbSOGAWA Hirofumi 4037f5beed75SAl Viro error = mnt_want_write(path.mnt); 4038c30dabfeSJan Kara if (error) 4039c30dabfeSJan Kara goto exit1; 4040b21996e3SJ. Bruce Fields retry_deleg: 40415955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4042f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40431da177e4SLinus Torvalds error = PTR_ERR(dentry); 40441da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40451da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4046f5beed75SAl Viro if (last.name[last.len]) 404750338b88STörök Edwin goto slashes; 40481da177e4SLinus Torvalds inode = dentry->d_inode; 4049b18825a7SDavid Howells if (d_is_negative(dentry)) 4050e6bc45d6STheodore Ts'o goto slashes; 40517de9c6eeSAl Viro ihold(inode); 4052f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4053be6d3e56SKentaro Takeda if (error) 4054c30dabfeSJan Kara goto exit2; 4055f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40561da177e4SLinus Torvalds exit2: 40571da177e4SLinus Torvalds dput(dentry); 40581da177e4SLinus Torvalds } 40595955102cSAl Viro inode_unlock(path.dentry->d_inode); 40601da177e4SLinus Torvalds if (inode) 40611da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4062b21996e3SJ. Bruce Fields inode = NULL; 4063b21996e3SJ. Bruce Fields if (delegated_inode) { 40645a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4065b21996e3SJ. Bruce Fields if (!error) 4066b21996e3SJ. Bruce Fields goto retry_deleg; 4067b21996e3SJ. Bruce Fields } 4068f5beed75SAl Viro mnt_drop_write(path.mnt); 40691da177e4SLinus Torvalds exit1: 4070f5beed75SAl Viro path_put(&path); 40711da177e4SLinus Torvalds putname(name); 40725d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40735d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40745d18f813SJeff Layton inode = NULL; 40755d18f813SJeff Layton goto retry; 40765d18f813SJeff Layton } 40771da177e4SLinus Torvalds return error; 40781da177e4SLinus Torvalds 40791da177e4SLinus Torvalds slashes: 4080b18825a7SDavid Howells if (d_is_negative(dentry)) 4081b18825a7SDavid Howells error = -ENOENT; 408244b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4083b18825a7SDavid Howells error = -EISDIR; 4084b18825a7SDavid Howells else 4085b18825a7SDavid Howells error = -ENOTDIR; 40861da177e4SLinus Torvalds goto exit2; 40871da177e4SLinus Torvalds } 40881da177e4SLinus Torvalds 40892e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40905590ff0dSUlrich Drepper { 40915590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40925590ff0dSUlrich Drepper return -EINVAL; 40935590ff0dSUlrich Drepper 40945590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 40955590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 40965590ff0dSUlrich Drepper 40975590ff0dSUlrich Drepper return do_unlinkat(dfd, pathname); 40985590ff0dSUlrich Drepper } 40995590ff0dSUlrich Drepper 41003480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41015590ff0dSUlrich Drepper { 41025590ff0dSUlrich Drepper return do_unlinkat(AT_FDCWD, pathname); 41035590ff0dSUlrich Drepper } 41045590ff0dSUlrich Drepper 4105db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 41061da177e4SLinus Torvalds { 4107a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 41081da177e4SLinus Torvalds 41091da177e4SLinus Torvalds if (error) 41101da177e4SLinus Torvalds return error; 41111da177e4SLinus Torvalds 4112acfa4380SAl Viro if (!dir->i_op->symlink) 41131da177e4SLinus Torvalds return -EPERM; 41141da177e4SLinus Torvalds 41151da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41161da177e4SLinus Torvalds if (error) 41171da177e4SLinus Torvalds return error; 41181da177e4SLinus Torvalds 41191da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4120a74574aaSStephen Smalley if (!error) 4121f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41221da177e4SLinus Torvalds return error; 41231da177e4SLinus Torvalds } 41244d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41251da177e4SLinus Torvalds 41262e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 41272e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 41281da177e4SLinus Torvalds { 41292ad94ae6SAl Viro int error; 413091a27b2aSJeff Layton struct filename *from; 41316902d925SDave Hansen struct dentry *dentry; 4132dae6ad8fSAl Viro struct path path; 4133f46d3567SJeff Layton unsigned int lookup_flags = 0; 41341da177e4SLinus Torvalds 41351da177e4SLinus Torvalds from = getname(oldname); 41361da177e4SLinus Torvalds if (IS_ERR(from)) 41371da177e4SLinus Torvalds return PTR_ERR(from); 4138f46d3567SJeff Layton retry: 4139f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41401da177e4SLinus Torvalds error = PTR_ERR(dentry); 41416902d925SDave Hansen if (IS_ERR(dentry)) 4142dae6ad8fSAl Viro goto out_putname; 41436902d925SDave Hansen 414491a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4145a8104a9fSAl Viro if (!error) 414691a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4147921a1650SAl Viro done_path_create(&path, dentry); 4148f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4149f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4150f46d3567SJeff Layton goto retry; 4151f46d3567SJeff Layton } 41526902d925SDave Hansen out_putname: 41531da177e4SLinus Torvalds putname(from); 41541da177e4SLinus Torvalds return error; 41551da177e4SLinus Torvalds } 41561da177e4SLinus Torvalds 41573480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41585590ff0dSUlrich Drepper { 41595590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 41605590ff0dSUlrich Drepper } 41615590ff0dSUlrich Drepper 4162146a8595SJ. Bruce Fields /** 4163146a8595SJ. Bruce Fields * vfs_link - create a new link 4164146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4165146a8595SJ. Bruce Fields * @dir: new parent 4166146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4167146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4168146a8595SJ. Bruce Fields * 4169146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4170146a8595SJ. Bruce Fields * 4171146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4172146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4173146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4174146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4175146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4176146a8595SJ. Bruce Fields * 4177146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4178146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4179146a8595SJ. Bruce Fields * to be NFS exported. 4180146a8595SJ. Bruce Fields */ 4181146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41821da177e4SLinus Torvalds { 41831da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41848de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41851da177e4SLinus Torvalds int error; 41861da177e4SLinus Torvalds 41871da177e4SLinus Torvalds if (!inode) 41881da177e4SLinus Torvalds return -ENOENT; 41891da177e4SLinus Torvalds 4190a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 41911da177e4SLinus Torvalds if (error) 41921da177e4SLinus Torvalds return error; 41931da177e4SLinus Torvalds 41941da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 41951da177e4SLinus Torvalds return -EXDEV; 41961da177e4SLinus Torvalds 41971da177e4SLinus Torvalds /* 41981da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 41991da177e4SLinus Torvalds */ 42001da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 42011da177e4SLinus Torvalds return -EPERM; 42020bd23d09SEric W. Biederman /* 42030bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 42040bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 42050bd23d09SEric W. Biederman * the vfs. 42060bd23d09SEric W. Biederman */ 42070bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 42080bd23d09SEric W. Biederman return -EPERM; 4209acfa4380SAl Viro if (!dir->i_op->link) 42101da177e4SLinus Torvalds return -EPERM; 42117e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42121da177e4SLinus Torvalds return -EPERM; 42131da177e4SLinus Torvalds 42141da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42151da177e4SLinus Torvalds if (error) 42161da177e4SLinus Torvalds return error; 42171da177e4SLinus Torvalds 42185955102cSAl Viro inode_lock(inode); 4219aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4220f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4221aae8a97dSAneesh Kumar K.V error = -ENOENT; 42228de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42238de52778SAl Viro error = -EMLINK; 4224146a8595SJ. Bruce Fields else { 4225146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4226146a8595SJ. Bruce Fields if (!error) 42271da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4228146a8595SJ. Bruce Fields } 4229f4e0c30cSAl Viro 4230f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4231f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4232f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4233f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4234f4e0c30cSAl Viro } 42355955102cSAl Viro inode_unlock(inode); 4236e31e14ecSStephen Smalley if (!error) 42377e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42381da177e4SLinus Torvalds return error; 42391da177e4SLinus Torvalds } 42404d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42411da177e4SLinus Torvalds 42421da177e4SLinus Torvalds /* 42431da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42441da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42451da177e4SLinus Torvalds * newname. --KAB 42461da177e4SLinus Torvalds * 42471da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42481da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42491da177e4SLinus Torvalds * and other special files. --ADM 42501da177e4SLinus Torvalds */ 42512e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 42522e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 42531da177e4SLinus Torvalds { 42541da177e4SLinus Torvalds struct dentry *new_dentry; 4255dae6ad8fSAl Viro struct path old_path, new_path; 4256146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 425711a7b371SAneesh Kumar K.V int how = 0; 42581da177e4SLinus Torvalds int error; 42591da177e4SLinus Torvalds 426011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4261c04030e1SUlrich Drepper return -EINVAL; 426211a7b371SAneesh Kumar K.V /* 4263f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4264f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4265f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 426611a7b371SAneesh Kumar K.V */ 4267f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4268f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4269f0cc6ffbSLinus Torvalds return -ENOENT; 427011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4271f0cc6ffbSLinus Torvalds } 4272c04030e1SUlrich Drepper 427311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 427411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4275442e31caSJeff Layton retry: 427611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42771da177e4SLinus Torvalds if (error) 42782ad94ae6SAl Viro return error; 42792ad94ae6SAl Viro 4280442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4281442e31caSJeff Layton (how & LOOKUP_REVAL)); 42821da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42836902d925SDave Hansen if (IS_ERR(new_dentry)) 4284dae6ad8fSAl Viro goto out; 4285dae6ad8fSAl Viro 4286dae6ad8fSAl Viro error = -EXDEV; 4287dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4288dae6ad8fSAl Viro goto out_dput; 4289800179c9SKees Cook error = may_linkat(&old_path); 4290800179c9SKees Cook if (unlikely(error)) 4291800179c9SKees Cook goto out_dput; 4292dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4293be6d3e56SKentaro Takeda if (error) 4294a8104a9fSAl Viro goto out_dput; 4295146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 429675c3f29dSDave Hansen out_dput: 4297921a1650SAl Viro done_path_create(&new_path, new_dentry); 4298146a8595SJ. Bruce Fields if (delegated_inode) { 4299146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4300d22e6338SOleg Drokin if (!error) { 4301d22e6338SOleg Drokin path_put(&old_path); 4302146a8595SJ. Bruce Fields goto retry; 4303146a8595SJ. Bruce Fields } 4304d22e6338SOleg Drokin } 4305442e31caSJeff Layton if (retry_estale(error, how)) { 4306d22e6338SOleg Drokin path_put(&old_path); 4307442e31caSJeff Layton how |= LOOKUP_REVAL; 4308442e31caSJeff Layton goto retry; 4309442e31caSJeff Layton } 43101da177e4SLinus Torvalds out: 43112d8f3038SAl Viro path_put(&old_path); 43121da177e4SLinus Torvalds 43131da177e4SLinus Torvalds return error; 43141da177e4SLinus Torvalds } 43151da177e4SLinus Torvalds 43163480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43175590ff0dSUlrich Drepper { 4318c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43195590ff0dSUlrich Drepper } 43205590ff0dSUlrich Drepper 4321bc27027aSMiklos Szeredi /** 4322bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4323bc27027aSMiklos Szeredi * @old_dir: parent of source 4324bc27027aSMiklos Szeredi * @old_dentry: source 4325bc27027aSMiklos Szeredi * @new_dir: parent of destination 4326bc27027aSMiklos Szeredi * @new_dentry: destination 4327bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4328520c8b16SMiklos Szeredi * @flags: rename flags 4329bc27027aSMiklos Szeredi * 4330bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4331bc27027aSMiklos Szeredi * 4332bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4333bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4334bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4335bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4336bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4337bc27027aSMiklos Szeredi * so. 4338bc27027aSMiklos Szeredi * 4339bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4340bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4341bc27027aSMiklos Szeredi * to be NFS exported. 4342bc27027aSMiklos Szeredi * 43431da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43441da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43451da177e4SLinus Torvalds * Problems: 4346d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43471da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43481da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4349a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43501da177e4SLinus Torvalds * story. 43516cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43526cedba89SJ. Bruce Fields * and source (if it is not a directory). 43531b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43541da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43551da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4356a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43571da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43581da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43591da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4360a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43611da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43621da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 43631da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4364e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 43651b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 43661da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4367c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 43681da177e4SLinus Torvalds * locking]. 43691da177e4SLinus Torvalds */ 43701da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 43718e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4372520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 43731da177e4SLinus Torvalds { 43741da177e4SLinus Torvalds int error; 4375bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 437659b0df21SEric Paris const unsigned char *old_name; 4377bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4378bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4379da1ce067SMiklos Szeredi bool new_is_dir = false; 4380da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 43811da177e4SLinus Torvalds 43829409e22aSMiklos Szeredi /* 43839409e22aSMiklos Szeredi * Check source == target. 43849409e22aSMiklos Szeredi * On overlayfs need to look at underlying inodes. 43859409e22aSMiklos Szeredi */ 43869409e22aSMiklos Szeredi if (vfs_select_inode(old_dentry, 0) == vfs_select_inode(new_dentry, 0)) 43871da177e4SLinus Torvalds return 0; 43881da177e4SLinus Torvalds 43891da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 43901da177e4SLinus Torvalds if (error) 43911da177e4SLinus Torvalds return error; 43921da177e4SLinus Torvalds 4393da1ce067SMiklos Szeredi if (!target) { 4394a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4395da1ce067SMiklos Szeredi } else { 4396da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4397da1ce067SMiklos Szeredi 4398da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43991da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4400da1ce067SMiklos Szeredi else 4401da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4402da1ce067SMiklos Szeredi } 44031da177e4SLinus Torvalds if (error) 44041da177e4SLinus Torvalds return error; 44051da177e4SLinus Torvalds 44067177a9c4SMiklos Szeredi if (!old_dir->i_op->rename && !old_dir->i_op->rename2) 44071da177e4SLinus Torvalds return -EPERM; 44081da177e4SLinus Torvalds 4409520c8b16SMiklos Szeredi if (flags && !old_dir->i_op->rename2) 4410520c8b16SMiklos Szeredi return -EINVAL; 4411520c8b16SMiklos Szeredi 4412bc27027aSMiklos Szeredi /* 4413bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4414bc27027aSMiklos Szeredi * we'll need to flip '..'. 4415bc27027aSMiklos Szeredi */ 4416da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4417da1ce067SMiklos Szeredi if (is_dir) { 4418bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4419bc27027aSMiklos Szeredi if (error) 4420bc27027aSMiklos Szeredi return error; 4421bc27027aSMiklos Szeredi } 4422da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4423da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4424da1ce067SMiklos Szeredi if (error) 4425da1ce067SMiklos Szeredi return error; 4426da1ce067SMiklos Szeredi } 4427da1ce067SMiklos Szeredi } 44280eeca283SRobert Love 44290b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44300b3974ebSMiklos Szeredi flags); 4431bc27027aSMiklos Szeredi if (error) 4432bc27027aSMiklos Szeredi return error; 4433bc27027aSMiklos Szeredi 4434bc27027aSMiklos Szeredi old_name = fsnotify_oldname_init(old_dentry->d_name.name); 4435bc27027aSMiklos Szeredi dget(new_dentry); 4436da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4437bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4438bc27027aSMiklos Szeredi else if (target) 44395955102cSAl Viro inode_lock(target); 4440bc27027aSMiklos Szeredi 4441bc27027aSMiklos Szeredi error = -EBUSY; 44427af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4443bc27027aSMiklos Szeredi goto out; 4444bc27027aSMiklos Szeredi 4445da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4446bc27027aSMiklos Szeredi error = -EMLINK; 4447da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4448bc27027aSMiklos Szeredi goto out; 4449da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4450da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4451da1ce067SMiklos Szeredi goto out; 4452da1ce067SMiklos Szeredi } 4453da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4454bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4455da1ce067SMiklos Szeredi if (!is_dir) { 4456bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4457bc27027aSMiklos Szeredi if (error) 4458bc27027aSMiklos Szeredi goto out; 4459da1ce067SMiklos Szeredi } 4460da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4461bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4462bc27027aSMiklos Szeredi if (error) 4463bc27027aSMiklos Szeredi goto out; 4464bc27027aSMiklos Szeredi } 44657177a9c4SMiklos Szeredi if (!old_dir->i_op->rename2) { 4466520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4467520c8b16SMiklos Szeredi new_dir, new_dentry); 4468520c8b16SMiklos Szeredi } else { 44697177a9c4SMiklos Szeredi WARN_ON(old_dir->i_op->rename != NULL); 4470520c8b16SMiklos Szeredi error = old_dir->i_op->rename2(old_dir, old_dentry, 4471520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4472520c8b16SMiklos Szeredi } 4473bc27027aSMiklos Szeredi if (error) 4474bc27027aSMiklos Szeredi goto out; 4475bc27027aSMiklos Szeredi 4476da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44771da177e4SLinus Torvalds if (is_dir) 4478bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4479bc27027aSMiklos Szeredi dont_mount(new_dentry); 44808ed936b5SEric W. Biederman detach_mounts(new_dentry); 4481bc27027aSMiklos Szeredi } 4482da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4483da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4484bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4485da1ce067SMiklos Szeredi else 4486da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4487da1ce067SMiklos Szeredi } 4488bc27027aSMiklos Szeredi out: 4489da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4490bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4491bc27027aSMiklos Szeredi else if (target) 44925955102cSAl Viro inode_unlock(target); 4493bc27027aSMiklos Szeredi dput(new_dentry); 4494da1ce067SMiklos Szeredi if (!error) { 4495123df294SAl Viro fsnotify_move(old_dir, new_dir, old_name, is_dir, 4496da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4497da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4498da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4499da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4500da1ce067SMiklos Szeredi } 4501da1ce067SMiklos Szeredi } 45020eeca283SRobert Love fsnotify_oldname_free(old_name); 45030eeca283SRobert Love 45041da177e4SLinus Torvalds return error; 45051da177e4SLinus Torvalds } 45064d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 45071da177e4SLinus Torvalds 4508520c8b16SMiklos Szeredi SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4509520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname, unsigned int, flags) 45101da177e4SLinus Torvalds { 45111da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 45121da177e4SLinus Torvalds struct dentry *trap; 4513f5beed75SAl Viro struct path old_path, new_path; 4514f5beed75SAl Viro struct qstr old_last, new_last; 4515f5beed75SAl Viro int old_type, new_type; 45168e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 451791a27b2aSJeff Layton struct filename *from; 451891a27b2aSJeff Layton struct filename *to; 4519f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4520c6a94284SJeff Layton bool should_retry = false; 45212ad94ae6SAl Viro int error; 4522520c8b16SMiklos Szeredi 45230d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4524da1ce067SMiklos Szeredi return -EINVAL; 4525da1ce067SMiklos Szeredi 45260d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45270d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4528520c8b16SMiklos Szeredi return -EINVAL; 4529520c8b16SMiklos Szeredi 45300d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45310d7a8555SMiklos Szeredi return -EPERM; 45320d7a8555SMiklos Szeredi 4533f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4534f5beed75SAl Viro target_flags = 0; 4535f5beed75SAl Viro 4536c6a94284SJeff Layton retry: 4537f5beed75SAl Viro from = user_path_parent(olddfd, oldname, 4538f5beed75SAl Viro &old_path, &old_last, &old_type, lookup_flags); 453991a27b2aSJeff Layton if (IS_ERR(from)) { 454091a27b2aSJeff Layton error = PTR_ERR(from); 45411da177e4SLinus Torvalds goto exit; 454291a27b2aSJeff Layton } 45431da177e4SLinus Torvalds 4544f5beed75SAl Viro to = user_path_parent(newdfd, newname, 4545f5beed75SAl Viro &new_path, &new_last, &new_type, lookup_flags); 454691a27b2aSJeff Layton if (IS_ERR(to)) { 454791a27b2aSJeff Layton error = PTR_ERR(to); 45481da177e4SLinus Torvalds goto exit1; 454991a27b2aSJeff Layton } 45501da177e4SLinus Torvalds 45511da177e4SLinus Torvalds error = -EXDEV; 4552f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45531da177e4SLinus Torvalds goto exit2; 45541da177e4SLinus Torvalds 45551da177e4SLinus Torvalds error = -EBUSY; 4556f5beed75SAl Viro if (old_type != LAST_NORM) 45571da177e4SLinus Torvalds goto exit2; 45581da177e4SLinus Torvalds 45590a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45600a7c3937SMiklos Szeredi error = -EEXIST; 4561f5beed75SAl Viro if (new_type != LAST_NORM) 45621da177e4SLinus Torvalds goto exit2; 45631da177e4SLinus Torvalds 4564f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4565c30dabfeSJan Kara if (error) 4566c30dabfeSJan Kara goto exit2; 4567c30dabfeSJan Kara 45688e6d782cSJ. Bruce Fields retry_deleg: 4569f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45701da177e4SLinus Torvalds 4571f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45721da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45731da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45741da177e4SLinus Torvalds goto exit3; 45751da177e4SLinus Torvalds /* source must exist */ 45761da177e4SLinus Torvalds error = -ENOENT; 4577b18825a7SDavid Howells if (d_is_negative(old_dentry)) 45781da177e4SLinus Torvalds goto exit4; 4579f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 45801da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45811da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 45821da177e4SLinus Torvalds goto exit4; 45830a7c3937SMiklos Szeredi error = -EEXIST; 45840a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 45850a7c3937SMiklos Szeredi goto exit5; 4586da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4587da1ce067SMiklos Szeredi error = -ENOENT; 4588da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4589da1ce067SMiklos Szeredi goto exit5; 4590da1ce067SMiklos Szeredi 4591da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4592da1ce067SMiklos Szeredi error = -ENOTDIR; 4593f5beed75SAl Viro if (new_last.name[new_last.len]) 4594da1ce067SMiklos Szeredi goto exit5; 4595da1ce067SMiklos Szeredi } 4596da1ce067SMiklos Szeredi } 45970a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 45980a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 45990a7c3937SMiklos Szeredi error = -ENOTDIR; 4600f5beed75SAl Viro if (old_last.name[old_last.len]) 46010a7c3937SMiklos Szeredi goto exit5; 4602f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 46030a7c3937SMiklos Szeredi goto exit5; 46040a7c3937SMiklos Szeredi } 46050a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 46060a7c3937SMiklos Szeredi error = -EINVAL; 46070a7c3937SMiklos Szeredi if (old_dentry == trap) 46080a7c3937SMiklos Szeredi goto exit5; 46091da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4610da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46111da177e4SLinus Torvalds error = -ENOTEMPTY; 46121da177e4SLinus Torvalds if (new_dentry == trap) 46131da177e4SLinus Torvalds goto exit5; 46141da177e4SLinus Torvalds 4615f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4616f5beed75SAl Viro &new_path, new_dentry, flags); 4617be6d3e56SKentaro Takeda if (error) 4618c30dabfeSJan Kara goto exit5; 4619f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4620f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4621520c8b16SMiklos Szeredi &delegated_inode, flags); 46221da177e4SLinus Torvalds exit5: 46231da177e4SLinus Torvalds dput(new_dentry); 46241da177e4SLinus Torvalds exit4: 46251da177e4SLinus Torvalds dput(old_dentry); 46261da177e4SLinus Torvalds exit3: 4627f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46288e6d782cSJ. Bruce Fields if (delegated_inode) { 46298e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46308e6d782cSJ. Bruce Fields if (!error) 46318e6d782cSJ. Bruce Fields goto retry_deleg; 46328e6d782cSJ. Bruce Fields } 4633f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46341da177e4SLinus Torvalds exit2: 4635c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4636c6a94284SJeff Layton should_retry = true; 4637f5beed75SAl Viro path_put(&new_path); 46382ad94ae6SAl Viro putname(to); 46391da177e4SLinus Torvalds exit1: 4640f5beed75SAl Viro path_put(&old_path); 46411da177e4SLinus Torvalds putname(from); 4642c6a94284SJeff Layton if (should_retry) { 4643c6a94284SJeff Layton should_retry = false; 4644c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4645c6a94284SJeff Layton goto retry; 4646c6a94284SJeff Layton } 46472ad94ae6SAl Viro exit: 46481da177e4SLinus Torvalds return error; 46491da177e4SLinus Torvalds } 46501da177e4SLinus Torvalds 4651520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4652520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4653520c8b16SMiklos Szeredi { 4654520c8b16SMiklos Szeredi return sys_renameat2(olddfd, oldname, newdfd, newname, 0); 4655520c8b16SMiklos Szeredi } 4656520c8b16SMiklos Szeredi 4657a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 46585590ff0dSUlrich Drepper { 4659520c8b16SMiklos Szeredi return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 46605590ff0dSUlrich Drepper } 46615590ff0dSUlrich Drepper 4662787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4663787fb6bcSMiklos Szeredi { 4664787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4665787fb6bcSMiklos Szeredi if (error) 4666787fb6bcSMiklos Szeredi return error; 4667787fb6bcSMiklos Szeredi 4668787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4669787fb6bcSMiklos Szeredi return -EPERM; 4670787fb6bcSMiklos Szeredi 4671787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4672787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4673787fb6bcSMiklos Szeredi } 4674787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4675787fb6bcSMiklos Szeredi 46765d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 46771da177e4SLinus Torvalds { 46785d826c84SAl Viro int len = PTR_ERR(link); 46791da177e4SLinus Torvalds if (IS_ERR(link)) 46801da177e4SLinus Torvalds goto out; 46811da177e4SLinus Torvalds 46821da177e4SLinus Torvalds len = strlen(link); 46831da177e4SLinus Torvalds if (len > (unsigned) buflen) 46841da177e4SLinus Torvalds len = buflen; 46851da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 46861da177e4SLinus Torvalds len = -EFAULT; 46871da177e4SLinus Torvalds out: 46881da177e4SLinus Torvalds return len; 46891da177e4SLinus Torvalds } 46901da177e4SLinus Torvalds 46911da177e4SLinus Torvalds /* 46921da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 46936b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 46946b255391SAl Viro * for any given inode is up to filesystem. 46951da177e4SLinus Torvalds */ 46961da177e4SLinus Torvalds int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen) 46971da177e4SLinus Torvalds { 4698fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46995f2c4179SAl Viro struct inode *inode = d_inode(dentry); 47005f2c4179SAl Viro const char *link = inode->i_link; 4701694a1764SMarcin Slusarz int res; 4702cc314eefSLinus Torvalds 4703d4dee48bSAl Viro if (!link) { 4704fceef393SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4705680baacbSAl Viro if (IS_ERR(link)) 4706680baacbSAl Viro return PTR_ERR(link); 4707d4dee48bSAl Viro } 4708680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4709fceef393SAl Viro do_delayed_call(&done); 4710694a1764SMarcin Slusarz return res; 47111da177e4SLinus Torvalds } 47124d359507SAl Viro EXPORT_SYMBOL(generic_readlink); 47131da177e4SLinus Torvalds 47141da177e4SLinus Torvalds /* get the link contents into pagecache */ 47156b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4716fceef393SAl Viro struct delayed_call *callback) 47171da177e4SLinus Torvalds { 4718ebd09abbSDuane Griffin char *kaddr; 47191da177e4SLinus Torvalds struct page *page; 47206b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47216b255391SAl Viro 4722d3883d4fSAl Viro if (!dentry) { 4723d3883d4fSAl Viro page = find_get_page(mapping, 0); 4724d3883d4fSAl Viro if (!page) 47256b255391SAl Viro return ERR_PTR(-ECHILD); 4726d3883d4fSAl Viro if (!PageUptodate(page)) { 4727d3883d4fSAl Viro put_page(page); 4728d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4729d3883d4fSAl Viro } 4730d3883d4fSAl Viro } else { 4731090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47321da177e4SLinus Torvalds if (IS_ERR(page)) 47336fe6900eSNick Piggin return (char*)page; 4734d3883d4fSAl Viro } 4735fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 473621fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 473721fc61c7SAl Viro kaddr = page_address(page); 47386b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4739ebd09abbSDuane Griffin return kaddr; 47401da177e4SLinus Torvalds } 47411da177e4SLinus Torvalds 47426b255391SAl Viro EXPORT_SYMBOL(page_get_link); 47431da177e4SLinus Torvalds 4744fceef393SAl Viro void page_put_link(void *arg) 47451da177e4SLinus Torvalds { 4746fceef393SAl Viro put_page(arg); 47471da177e4SLinus Torvalds } 47484d359507SAl Viro EXPORT_SYMBOL(page_put_link); 47491da177e4SLinus Torvalds 4750aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4751aa80deabSAl Viro { 4752fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47536b255391SAl Viro int res = readlink_copy(buffer, buflen, 47546b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4755fceef393SAl Viro &done)); 4756fceef393SAl Viro do_delayed_call(&done); 4757aa80deabSAl Viro return res; 4758aa80deabSAl Viro } 4759aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4760aa80deabSAl Viro 476154566b2cSNick Piggin /* 476254566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 476354566b2cSNick Piggin */ 476454566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 47651da177e4SLinus Torvalds { 47661da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 47670adb25d2SKirill Korotaev struct page *page; 4768afddba49SNick Piggin void *fsdata; 4769beb497abSDmitriy Monakhov int err; 477054566b2cSNick Piggin unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE; 477154566b2cSNick Piggin if (nofs) 477254566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 47731da177e4SLinus Torvalds 47747e53cac4SNeilBrown retry: 4775afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 477654566b2cSNick Piggin flags, &page, &fsdata); 47771da177e4SLinus Torvalds if (err) 4778afddba49SNick Piggin goto fail; 4779afddba49SNick Piggin 478021fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4781afddba49SNick Piggin 4782afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4783afddba49SNick Piggin page, fsdata); 47841da177e4SLinus Torvalds if (err < 0) 47851da177e4SLinus Torvalds goto fail; 4786afddba49SNick Piggin if (err < len-1) 4787afddba49SNick Piggin goto retry; 4788afddba49SNick Piggin 47891da177e4SLinus Torvalds mark_inode_dirty(inode); 47901da177e4SLinus Torvalds return 0; 47911da177e4SLinus Torvalds fail: 47921da177e4SLinus Torvalds return err; 47931da177e4SLinus Torvalds } 47944d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 47951da177e4SLinus Torvalds 47960adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 47970adb25d2SKirill Korotaev { 47980adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4799c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48000adb25d2SKirill Korotaev } 48014d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48020adb25d2SKirill Korotaev 480392e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48041da177e4SLinus Torvalds .readlink = generic_readlink, 48056b255391SAl Viro .get_link = page_get_link, 48061da177e4SLinus Torvalds }; 48071da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4808