1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namei.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * Some corrections by tytso. 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 131da177e4SLinus Torvalds * lookup logic. 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 161da177e4SLinus Torvalds */ 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/init.h> 19630d9c47SPaul Gortmaker #include <linux/export.h> 2044696908SDavid S. Miller #include <linux/kernel.h> 211da177e4SLinus Torvalds #include <linux/slab.h> 221da177e4SLinus Torvalds #include <linux/fs.h> 231da177e4SLinus Torvalds #include <linux/namei.h> 241da177e4SLinus Torvalds #include <linux/pagemap.h> 250eeca283SRobert Love #include <linux/fsnotify.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 286146f0d5SMimi Zohar #include <linux/ima.h> 291da177e4SLinus Torvalds #include <linux/syscalls.h> 301da177e4SLinus Torvalds #include <linux/mount.h> 311da177e4SLinus Torvalds #include <linux/audit.h> 3216f7e0feSRandy Dunlap #include <linux/capability.h> 33834f2a4aSTrond Myklebust #include <linux/file.h> 345590ff0dSUlrich Drepper #include <linux/fcntl.h> 3508ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 365ad4e53bSAl Viro #include <linux/fs_struct.h> 37e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3899d263d4SLinus Torvalds #include <linux/hash.h> 392a18da7aSGeorge Spelvin #include <linux/bitops.h> 40aeaa4a79SEric W. Biederman #include <linux/init_task.h> 417c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 421da177e4SLinus Torvalds 43e81e3f4dSEric Paris #include "internal.h" 44c7105365SAl Viro #include "mount.h" 45e81e3f4dSEric Paris 461da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 471da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 481da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 491da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 501da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 511da177e4SLinus Torvalds * 521da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 531da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 541da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 551da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 561da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 571da177e4SLinus Torvalds * the special cases of the former code. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 601da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 611da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 641da177e4SLinus Torvalds * resolution to correspond with current state of the code. 651da177e4SLinus Torvalds * 661da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 671da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 681da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 691da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 701da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 711da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 721da177e4SLinus Torvalds */ 731da177e4SLinus Torvalds 741da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 751da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 761da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 771da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 781da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 791da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8025985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 811da177e4SLinus Torvalds * 821da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 831da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 841da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 851da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 861da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 871da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 881da177e4SLinus Torvalds * and in the old Linux semantics. 891da177e4SLinus Torvalds */ 901da177e4SLinus Torvalds 911da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 921da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 931da177e4SLinus Torvalds * 941da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 951da177e4SLinus Torvalds */ 961da177e4SLinus Torvalds 971da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 981da177e4SLinus Torvalds * inside the path - always follow. 991da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1001da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1011da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1021da177e4SLinus Torvalds * otherwise - don't follow. 1031da177e4SLinus Torvalds * (applied in that order). 1041da177e4SLinus Torvalds * 1051da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1061da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1071da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1081da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1091da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1101da177e4SLinus Torvalds */ 1111da177e4SLinus Torvalds /* 1121da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 113a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1141da177e4SLinus Torvalds * any extra contention... 1151da177e4SLinus Torvalds */ 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1181da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1191da177e4SLinus Torvalds * kernel data space before using them.. 1201da177e4SLinus Torvalds * 1211da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1221da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1231da177e4SLinus Torvalds */ 1247950e385SJeff Layton 125fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12691a27b2aSJeff Layton 12751f39a1fSDavid Drysdale struct filename * 12891a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12991a27b2aSJeff Layton { 13094b5d262SAl Viro struct filename *result; 1317950e385SJeff Layton char *kname; 13294b5d262SAl Viro int len; 1331da177e4SLinus Torvalds 1347ac86265SJeff Layton result = audit_reusename(filename); 1357ac86265SJeff Layton if (result) 1367ac86265SJeff Layton return result; 1377ac86265SJeff Layton 1387950e385SJeff Layton result = __getname(); 1393f9f0aa6SLinus Torvalds if (unlikely(!result)) 1404043cde8SEric Paris return ERR_PTR(-ENOMEM); 1411da177e4SLinus Torvalds 1427950e385SJeff Layton /* 1437950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1447950e385SJeff Layton * allocation 1457950e385SJeff Layton */ 146fd2f7cb5SAl Viro kname = (char *)result->iname; 14791a27b2aSJeff Layton result->name = kname; 1487950e385SJeff Layton 14994b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15091a27b2aSJeff Layton if (unlikely(len < 0)) { 15194b5d262SAl Viro __putname(result); 15294b5d262SAl Viro return ERR_PTR(len); 15391a27b2aSJeff Layton } 1543f9f0aa6SLinus Torvalds 1557950e385SJeff Layton /* 1567950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1577950e385SJeff Layton * separate struct filename so we can dedicate the entire 1587950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1597950e385SJeff Layton * userland. 1607950e385SJeff Layton */ 16194b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 162fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1637950e385SJeff Layton kname = (char *)result; 1647950e385SJeff Layton 165fd2f7cb5SAl Viro /* 166fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 167fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 168fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 169fd2f7cb5SAl Viro */ 170fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17194b5d262SAl Viro if (unlikely(!result)) { 17294b5d262SAl Viro __putname(kname); 17394b5d262SAl Viro return ERR_PTR(-ENOMEM); 1747950e385SJeff Layton } 1757950e385SJeff Layton result->name = kname; 17694b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17794b5d262SAl Viro if (unlikely(len < 0)) { 17894b5d262SAl Viro __putname(kname); 17994b5d262SAl Viro kfree(result); 18094b5d262SAl Viro return ERR_PTR(len); 18194b5d262SAl Viro } 18294b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18394b5d262SAl Viro __putname(kname); 18494b5d262SAl Viro kfree(result); 18594b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18694b5d262SAl Viro } 1877950e385SJeff Layton } 1887950e385SJeff Layton 18994b5d262SAl Viro result->refcnt = 1; 1903f9f0aa6SLinus Torvalds /* The empty path is special. */ 1913f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1923f9f0aa6SLinus Torvalds if (empty) 1931fa1e7f6SAndy Whitcroft *empty = 1; 19494b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19594b5d262SAl Viro putname(result); 19694b5d262SAl Viro return ERR_PTR(-ENOENT); 1971da177e4SLinus Torvalds } 19894b5d262SAl Viro } 1997950e385SJeff Layton 2007950e385SJeff Layton result->uptr = filename; 201c4ad8f98SLinus Torvalds result->aname = NULL; 2021da177e4SLinus Torvalds audit_getname(result); 2031da177e4SLinus Torvalds return result; 2043f9f0aa6SLinus Torvalds } 2053f9f0aa6SLinus Torvalds 20691a27b2aSJeff Layton struct filename * 20791a27b2aSJeff Layton getname(const char __user * filename) 208f52e0c11SAl Viro { 209f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 210f52e0c11SAl Viro } 211f52e0c11SAl Viro 212c4ad8f98SLinus Torvalds struct filename * 213c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 214c4ad8f98SLinus Torvalds { 215c4ad8f98SLinus Torvalds struct filename *result; 21608518549SPaul Moore int len = strlen(filename) + 1; 217c4ad8f98SLinus Torvalds 218c4ad8f98SLinus Torvalds result = __getname(); 219c4ad8f98SLinus Torvalds if (unlikely(!result)) 220c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 221c4ad8f98SLinus Torvalds 22208518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 223fd2f7cb5SAl Viro result->name = (char *)result->iname; 22408518549SPaul Moore } else if (len <= PATH_MAX) { 22508518549SPaul Moore struct filename *tmp; 22608518549SPaul Moore 22708518549SPaul Moore tmp = kmalloc(sizeof(*tmp), GFP_KERNEL); 22808518549SPaul Moore if (unlikely(!tmp)) { 22908518549SPaul Moore __putname(result); 23008518549SPaul Moore return ERR_PTR(-ENOMEM); 23108518549SPaul Moore } 23208518549SPaul Moore tmp->name = (char *)result; 23308518549SPaul Moore result = tmp; 23408518549SPaul Moore } else { 23508518549SPaul Moore __putname(result); 23608518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23708518549SPaul Moore } 23808518549SPaul Moore memcpy((char *)result->name, filename, len); 239c4ad8f98SLinus Torvalds result->uptr = NULL; 240c4ad8f98SLinus Torvalds result->aname = NULL; 24155422d0bSPaul Moore result->refcnt = 1; 242fd3522fdSPaul Moore audit_getname(result); 243c4ad8f98SLinus Torvalds 244c4ad8f98SLinus Torvalds return result; 245c4ad8f98SLinus Torvalds } 246c4ad8f98SLinus Torvalds 24791a27b2aSJeff Layton void putname(struct filename *name) 2481da177e4SLinus Torvalds { 24955422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 25055422d0bSPaul Moore 25155422d0bSPaul Moore if (--name->refcnt > 0) 25255422d0bSPaul Moore return; 25355422d0bSPaul Moore 254fd2f7cb5SAl Viro if (name->name != name->iname) { 25555422d0bSPaul Moore __putname(name->name); 25655422d0bSPaul Moore kfree(name); 25755422d0bSPaul Moore } else 25855422d0bSPaul Moore __putname(name); 2591da177e4SLinus Torvalds } 2601da177e4SLinus Torvalds 261e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 262e77819e5SLinus Torvalds { 26384635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 264e77819e5SLinus Torvalds struct posix_acl *acl; 265e77819e5SLinus Torvalds 266e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2673567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2683567866bSAl Viro if (!acl) 269e77819e5SLinus Torvalds return -EAGAIN; 2703567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 271b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 272e77819e5SLinus Torvalds return -ECHILD; 273206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 274e77819e5SLinus Torvalds } 275e77819e5SLinus Torvalds 2762982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2774e34e719SChristoph Hellwig if (IS_ERR(acl)) 2784e34e719SChristoph Hellwig return PTR_ERR(acl); 279e77819e5SLinus Torvalds if (acl) { 280e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 281e77819e5SLinus Torvalds posix_acl_release(acl); 282e77819e5SLinus Torvalds return error; 283e77819e5SLinus Torvalds } 28484635d68SLinus Torvalds #endif 285e77819e5SLinus Torvalds 286e77819e5SLinus Torvalds return -EAGAIN; 287e77819e5SLinus Torvalds } 288e77819e5SLinus Torvalds 2895909ccaaSLinus Torvalds /* 290948409c7SAndreas Gruenbacher * This does the basic permission checking 2915909ccaaSLinus Torvalds */ 2927e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2935909ccaaSLinus Torvalds { 29426cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2955909ccaaSLinus Torvalds 2968e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2975909ccaaSLinus Torvalds mode >>= 6; 2985909ccaaSLinus Torvalds else { 299e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 3007e40145eSAl Viro int error = check_acl(inode, mask); 3015909ccaaSLinus Torvalds if (error != -EAGAIN) 3025909ccaaSLinus Torvalds return error; 3035909ccaaSLinus Torvalds } 3045909ccaaSLinus Torvalds 3055909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3065909ccaaSLinus Torvalds mode >>= 3; 3075909ccaaSLinus Torvalds } 3085909ccaaSLinus Torvalds 3095909ccaaSLinus Torvalds /* 3105909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3115909ccaaSLinus Torvalds */ 3129c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3135909ccaaSLinus Torvalds return 0; 3145909ccaaSLinus Torvalds return -EACCES; 3155909ccaaSLinus Torvalds } 3161da177e4SLinus Torvalds 3171da177e4SLinus Torvalds /** 3181da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3191da177e4SLinus Torvalds * @inode: inode to check access rights for 3208fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3211da177e4SLinus Torvalds * 3221da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3231da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3241da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 325b74c79e9SNick Piggin * are used for other things. 326b74c79e9SNick Piggin * 327b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 328b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 329b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3301da177e4SLinus Torvalds */ 3312830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3321da177e4SLinus Torvalds { 3335909ccaaSLinus Torvalds int ret; 3341da177e4SLinus Torvalds 3351da177e4SLinus Torvalds /* 336948409c7SAndreas Gruenbacher * Do the basic permission checks. 3371da177e4SLinus Torvalds */ 3387e40145eSAl Viro ret = acl_permission_check(inode, mask); 3395909ccaaSLinus Torvalds if (ret != -EACCES) 3405909ccaaSLinus Torvalds return ret; 3411da177e4SLinus Torvalds 342d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 343d594e7ecSAl Viro /* DACs are overridable for directories */ 344d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34523adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34623adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 347d594e7ecSAl Viro return 0; 34823adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3491da177e4SLinus Torvalds return 0; 3502a4c2242SStephen Smalley return -EACCES; 3512a4c2242SStephen Smalley } 3521da177e4SLinus Torvalds 3531da177e4SLinus Torvalds /* 3541da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3551da177e4SLinus Torvalds */ 3567ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 357d594e7ecSAl Viro if (mask == MAY_READ) 35823adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3591da177e4SLinus Torvalds return 0; 3602a4c2242SStephen Smalley /* 3612a4c2242SStephen Smalley * Read/write DACs are always overridable. 3622a4c2242SStephen Smalley * Executable DACs are overridable when there is 3632a4c2242SStephen Smalley * at least one exec bit set. 3642a4c2242SStephen Smalley */ 3652a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 3662a4c2242SStephen Smalley if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3672a4c2242SStephen Smalley return 0; 3681da177e4SLinus Torvalds 3691da177e4SLinus Torvalds return -EACCES; 3701da177e4SLinus Torvalds } 3714d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3721da177e4SLinus Torvalds 3733ddcd056SLinus Torvalds /* 3743ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3753ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3763ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3773ddcd056SLinus Torvalds * permission function, use the fast case". 3783ddcd056SLinus Torvalds */ 3793ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3803ddcd056SLinus Torvalds { 3813ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3823ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3833ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3843ddcd056SLinus Torvalds 3853ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3863ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3873ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3883ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3893ddcd056SLinus Torvalds } 3903ddcd056SLinus Torvalds return generic_permission(inode, mask); 3913ddcd056SLinus Torvalds } 3923ddcd056SLinus Torvalds 393cb23beb5SChristoph Hellwig /** 3940bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 3950bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 39655852635SRandy Dunlap * @inode: Inode to check permission on 3970bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3980bdaea90SDavid Howells * 3990bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4000bdaea90SDavid Howells */ 4010bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4020bdaea90SDavid Howells { 4030bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4040bdaea90SDavid Howells umode_t mode = inode->i_mode; 4050bdaea90SDavid Howells 4060bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 407bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4080bdaea90SDavid Howells return -EROFS; 4090bdaea90SDavid Howells } 4100bdaea90SDavid Howells return 0; 4110bdaea90SDavid Howells } 4120bdaea90SDavid Howells 4130bdaea90SDavid Howells /** 4140bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4150bdaea90SDavid Howells * @inode: Inode to check permission on 4160bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4170bdaea90SDavid Howells * 4180bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4190bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4200bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4210bdaea90SDavid Howells * 4220bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4230bdaea90SDavid Howells */ 4240bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4250bdaea90SDavid Howells { 4260bdaea90SDavid Howells int retval; 4270bdaea90SDavid Howells 4280bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4290bdaea90SDavid Howells if (retval) 4300bdaea90SDavid Howells return retval; 4314bfd054aSEric Biggers 4324bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 4334bfd054aSEric Biggers /* 4344bfd054aSEric Biggers * Nobody gets write access to an immutable file. 4354bfd054aSEric Biggers */ 4364bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 4374bfd054aSEric Biggers return -EPERM; 4384bfd054aSEric Biggers 4394bfd054aSEric Biggers /* 4404bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 4414bfd054aSEric Biggers * written back improperly if their true value is unknown 4424bfd054aSEric Biggers * to the vfs. 4434bfd054aSEric Biggers */ 4444bfd054aSEric Biggers if (HAS_UNMAPPED_ID(inode)) 4454bfd054aSEric Biggers return -EACCES; 4464bfd054aSEric Biggers } 4474bfd054aSEric Biggers 4484bfd054aSEric Biggers retval = do_inode_permission(inode, mask); 4494bfd054aSEric Biggers if (retval) 4504bfd054aSEric Biggers return retval; 4514bfd054aSEric Biggers 4524bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 4534bfd054aSEric Biggers if (retval) 4544bfd054aSEric Biggers return retval; 4554bfd054aSEric Biggers 4564bfd054aSEric Biggers return security_inode_permission(inode, mask); 4570bdaea90SDavid Howells } 4584d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4590bdaea90SDavid Howells 4600bdaea90SDavid Howells /** 4615dd784d0SJan Blunck * path_get - get a reference to a path 4625dd784d0SJan Blunck * @path: path to get the reference to 4635dd784d0SJan Blunck * 4645dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4655dd784d0SJan Blunck */ 466dcf787f3SAl Viro void path_get(const struct path *path) 4675dd784d0SJan Blunck { 4685dd784d0SJan Blunck mntget(path->mnt); 4695dd784d0SJan Blunck dget(path->dentry); 4705dd784d0SJan Blunck } 4715dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4725dd784d0SJan Blunck 4735dd784d0SJan Blunck /** 4741d957f9bSJan Blunck * path_put - put a reference to a path 4751d957f9bSJan Blunck * @path: path to put the reference to 4761d957f9bSJan Blunck * 4771d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4781d957f9bSJan Blunck */ 479dcf787f3SAl Viro void path_put(const struct path *path) 4801da177e4SLinus Torvalds { 4811d957f9bSJan Blunck dput(path->dentry); 4821d957f9bSJan Blunck mntput(path->mnt); 4831da177e4SLinus Torvalds } 4841d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4851da177e4SLinus Torvalds 486894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4871f55a6ecSAl Viro struct nameidata { 4881f55a6ecSAl Viro struct path path; 4891f55a6ecSAl Viro struct qstr last; 4901f55a6ecSAl Viro struct path root; 4911f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 4921f55a6ecSAl Viro unsigned int flags; 4939883d185SAl Viro unsigned seq, m_seq; 4941f55a6ecSAl Viro int last_type; 4951f55a6ecSAl Viro unsigned depth; 496756daf26SNeilBrown int total_link_count; 497697fc6caSAl Viro struct saved { 498697fc6caSAl Viro struct path link; 499fceef393SAl Viro struct delayed_call done; 500697fc6caSAl Viro const char *name; 5010450b2d1SAl Viro unsigned seq; 502894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5039883d185SAl Viro struct filename *name; 5049883d185SAl Viro struct nameidata *saved; 505fceef393SAl Viro struct inode *link_inode; 5069883d185SAl Viro unsigned root_seq; 5079883d185SAl Viro int dfd; 5083859a271SKees Cook } __randomize_layout; 5091f55a6ecSAl Viro 5109883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 511894bc8c4SAl Viro { 512756daf26SNeilBrown struct nameidata *old = current->nameidata; 513756daf26SNeilBrown p->stack = p->internal; 514c8a53ee5SAl Viro p->dfd = dfd; 515c8a53ee5SAl Viro p->name = name; 516756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5179883d185SAl Viro p->saved = old; 518756daf26SNeilBrown current->nameidata = p; 519894bc8c4SAl Viro } 520894bc8c4SAl Viro 5219883d185SAl Viro static void restore_nameidata(void) 522894bc8c4SAl Viro { 5239883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 524756daf26SNeilBrown 525756daf26SNeilBrown current->nameidata = old; 526756daf26SNeilBrown if (old) 527756daf26SNeilBrown old->total_link_count = now->total_link_count; 528e1a63bbcSAl Viro if (now->stack != now->internal) 529756daf26SNeilBrown kfree(now->stack); 530894bc8c4SAl Viro } 531894bc8c4SAl Viro 532894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 533894bc8c4SAl Viro { 534bc40aee0SAl Viro struct saved *p; 535bc40aee0SAl Viro 536bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 537bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 538bc40aee0SAl Viro GFP_ATOMIC); 539bc40aee0SAl Viro if (unlikely(!p)) 540bc40aee0SAl Viro return -ECHILD; 541bc40aee0SAl Viro } else { 542bc40aee0SAl Viro p= kmalloc(MAXSYMLINKS * sizeof(struct saved), 543894bc8c4SAl Viro GFP_KERNEL); 544894bc8c4SAl Viro if (unlikely(!p)) 545894bc8c4SAl Viro return -ENOMEM; 546bc40aee0SAl Viro } 547894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 548894bc8c4SAl Viro nd->stack = p; 549894bc8c4SAl Viro return 0; 550894bc8c4SAl Viro } 551894bc8c4SAl Viro 552397d425dSEric W. Biederman /** 553397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 554397d425dSEric W. Biederman * @path: nameidate to verify 555397d425dSEric W. Biederman * 556397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 557397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 558397d425dSEric W. Biederman */ 559397d425dSEric W. Biederman static bool path_connected(const struct path *path) 560397d425dSEric W. Biederman { 561397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 562397d425dSEric W. Biederman 563397d425dSEric W. Biederman /* Only bind mounts can have disconnected paths */ 564397d425dSEric W. Biederman if (mnt->mnt_root == mnt->mnt_sb->s_root) 565397d425dSEric W. Biederman return true; 566397d425dSEric W. Biederman 567397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 568397d425dSEric W. Biederman } 569397d425dSEric W. Biederman 570894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 571894bc8c4SAl Viro { 572da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 573894bc8c4SAl Viro return 0; 574894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 575894bc8c4SAl Viro return 0; 576894bc8c4SAl Viro return __nd_alloc_stack(nd); 577894bc8c4SAl Viro } 578894bc8c4SAl Viro 5797973387aSAl Viro static void drop_links(struct nameidata *nd) 5807973387aSAl Viro { 5817973387aSAl Viro int i = nd->depth; 5827973387aSAl Viro while (i--) { 5837973387aSAl Viro struct saved *last = nd->stack + i; 584fceef393SAl Viro do_delayed_call(&last->done); 585fceef393SAl Viro clear_delayed_call(&last->done); 5867973387aSAl Viro } 5877973387aSAl Viro } 5887973387aSAl Viro 5897973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5907973387aSAl Viro { 5917973387aSAl Viro drop_links(nd); 5927973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5937973387aSAl Viro int i; 5947973387aSAl Viro path_put(&nd->path); 5957973387aSAl Viro for (i = 0; i < nd->depth; i++) 5967973387aSAl Viro path_put(&nd->stack[i].link); 597102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 598102b8af2SAl Viro path_put(&nd->root); 599102b8af2SAl Viro nd->root.mnt = NULL; 600102b8af2SAl Viro } 6017973387aSAl Viro } else { 6027973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6037973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6047973387aSAl Viro nd->root.mnt = NULL; 6057973387aSAl Viro rcu_read_unlock(); 6067973387aSAl Viro } 6077973387aSAl Viro nd->depth = 0; 6087973387aSAl Viro } 6097973387aSAl Viro 6107973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6117973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6127973387aSAl Viro struct path *path, unsigned seq) 6137973387aSAl Viro { 6147973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6157973387aSAl Viro if (unlikely(res)) { 6167973387aSAl Viro if (res > 0) 6177973387aSAl Viro path->mnt = NULL; 6187973387aSAl Viro path->dentry = NULL; 6197973387aSAl Viro return false; 6207973387aSAl Viro } 6217973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6227973387aSAl Viro path->dentry = NULL; 6237973387aSAl Viro return false; 6247973387aSAl Viro } 6257973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6267973387aSAl Viro } 6277973387aSAl Viro 6287973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6297973387aSAl Viro { 6307973387aSAl Viro int i; 6317973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6327973387aSAl Viro struct saved *last = nd->stack + i; 6337973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6347973387aSAl Viro drop_links(nd); 6357973387aSAl Viro nd->depth = i + 1; 6367973387aSAl Viro return false; 6377973387aSAl Viro } 6387973387aSAl Viro } 6397973387aSAl Viro return true; 6407973387aSAl Viro } 6417973387aSAl Viro 64219660af7SAl Viro /* 64331e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 64419660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 64519660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 64657e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 64719660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 64819660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 64919660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 65019660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 65131e6b01fSNick Piggin */ 65231e6b01fSNick Piggin 65331e6b01fSNick Piggin /** 65419660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 65519660af7SAl Viro * @nd: nameidata pathwalk data 65639191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 65731e6b01fSNick Piggin * 6584675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6594675ac39SAl Viro * for ref-walk mode. 6604675ac39SAl Viro * Must be called from rcu-walk context. 6617973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6627973387aSAl Viro * terminate_walk(). 66331e6b01fSNick Piggin */ 6644675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 66531e6b01fSNick Piggin { 66631e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 66731e6b01fSNick Piggin 66831e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 669e5c832d5SLinus Torvalds 670e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6717973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6727973387aSAl Viro goto out2; 6734675ac39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 6744675ac39SAl Viro goto out1; 6754675ac39SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 6764675ac39SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) 6774675ac39SAl Viro goto out; 6784675ac39SAl Viro } 6794675ac39SAl Viro rcu_read_unlock(); 6804675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6814675ac39SAl Viro return 0; 6824675ac39SAl Viro 6834675ac39SAl Viro out2: 6844675ac39SAl Viro nd->path.mnt = NULL; 6854675ac39SAl Viro nd->path.dentry = NULL; 6864675ac39SAl Viro out1: 6874675ac39SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6884675ac39SAl Viro nd->root.mnt = NULL; 6894675ac39SAl Viro out: 6904675ac39SAl Viro rcu_read_unlock(); 6914675ac39SAl Viro return -ECHILD; 6924675ac39SAl Viro } 6934675ac39SAl Viro 6944675ac39SAl Viro /** 6954675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 6964675ac39SAl Viro * @nd: nameidata pathwalk data 6974675ac39SAl Viro * @dentry: child of nd->path.dentry 6984675ac39SAl Viro * @seq: seq number to check dentry against 6994675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7004675ac39SAl Viro * 7014675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7024675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7034675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7044675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7054675ac39SAl Viro * terminate_walk(). 7064675ac39SAl Viro */ 7074675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7084675ac39SAl Viro { 7094675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7104675ac39SAl Viro 7114675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7124675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7134675ac39SAl Viro goto out2; 7147973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7157973387aSAl Viro goto out2; 7164675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7177973387aSAl Viro goto out1; 71848a066e7SAl Viro 71915570086SLinus Torvalds /* 7204675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7214675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7224675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7234675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7244675ac39SAl Viro * know the parent must still be valid if the child sequence number is 72515570086SLinus Torvalds */ 7264675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 727e5c832d5SLinus Torvalds goto out; 7284675ac39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) { 7294675ac39SAl Viro rcu_read_unlock(); 7304675ac39SAl Viro dput(dentry); 7314675ac39SAl Viro goto drop_root_mnt; 73219660af7SAl Viro } 733e5c832d5SLinus Torvalds /* 734e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 735e5c832d5SLinus Torvalds * still valid and get it if required. 736e5c832d5SLinus Torvalds */ 737e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7385a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7395a8d87e8SAl Viro rcu_read_unlock(); 7405a8d87e8SAl Viro dput(dentry); 7415a8d87e8SAl Viro return -ECHILD; 7427973387aSAl Viro } 74331e6b01fSNick Piggin } 74431e6b01fSNick Piggin 7458b61e74fSAl Viro rcu_read_unlock(); 74631e6b01fSNick Piggin return 0; 74719660af7SAl Viro 7487973387aSAl Viro out2: 7497973387aSAl Viro nd->path.mnt = NULL; 7507973387aSAl Viro out1: 7517973387aSAl Viro nd->path.dentry = NULL; 752e5c832d5SLinus Torvalds out: 7538b61e74fSAl Viro rcu_read_unlock(); 754d0d27277SLinus Torvalds drop_root_mnt: 755d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 756d0d27277SLinus Torvalds nd->root.mnt = NULL; 75731e6b01fSNick Piggin return -ECHILD; 75831e6b01fSNick Piggin } 75931e6b01fSNick Piggin 7604ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76134286d66SNick Piggin { 762a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7634ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 764a89f8337SAl Viro else 765a89f8337SAl Viro return 1; 76634286d66SNick Piggin } 76734286d66SNick Piggin 7689f1fafeeSAl Viro /** 7699f1fafeeSAl Viro * complete_walk - successful completion of path walk 7709f1fafeeSAl Viro * @nd: pointer nameidata 77139159de2SJeff Layton * 7729f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7739f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7749f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7759f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7769f1fafeeSAl Viro * need to drop nd->path. 77739159de2SJeff Layton */ 7789f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 77939159de2SJeff Layton { 78016c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78139159de2SJeff Layton int status; 78239159de2SJeff Layton 7839f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7849f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7859f1fafeeSAl Viro nd->root.mnt = NULL; 7864675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 78748a066e7SAl Viro return -ECHILD; 78848a066e7SAl Viro } 7899f1fafeeSAl Viro 79016c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 79139159de2SJeff Layton return 0; 79239159de2SJeff Layton 793ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 79416c2cd71SAl Viro return 0; 79516c2cd71SAl Viro 796ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 79739159de2SJeff Layton if (status > 0) 79839159de2SJeff Layton return 0; 79939159de2SJeff Layton 80016c2cd71SAl Viro if (!status) 80139159de2SJeff Layton status = -ESTALE; 80216c2cd71SAl Viro 80339159de2SJeff Layton return status; 80439159de2SJeff Layton } 80539159de2SJeff Layton 80618d8c860SAl Viro static void set_root(struct nameidata *nd) 8072a737871SAl Viro { 80831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8099e6697e2SAl Viro 8109e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8118f47a016SAl Viro unsigned seq; 812c28cc364SNick Piggin 813c28cc364SNick Piggin do { 814c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 81531e6b01fSNick Piggin nd->root = fs->root; 8168f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 817c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8189e6697e2SAl Viro } else { 8199e6697e2SAl Viro get_fs_root(fs, &nd->root); 8209e6697e2SAl Viro } 82131e6b01fSNick Piggin } 82231e6b01fSNick Piggin 8231d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 824051d3812SIan Kent { 825051d3812SIan Kent dput(path->dentry); 8264ac91378SJan Blunck if (path->mnt != nd->path.mnt) 827051d3812SIan Kent mntput(path->mnt); 828051d3812SIan Kent } 829051d3812SIan Kent 8307b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8317b9337aaSNick Piggin struct nameidata *nd) 832051d3812SIan Kent { 83331e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8344ac91378SJan Blunck dput(nd->path.dentry); 83531e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8364ac91378SJan Blunck mntput(nd->path.mnt); 8379a229683SHuang Shijie } 83831e6b01fSNick Piggin nd->path.mnt = path->mnt; 8394ac91378SJan Blunck nd->path.dentry = path->dentry; 840051d3812SIan Kent } 841051d3812SIan Kent 842248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 843248fb5b9SAl Viro { 844248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 845248fb5b9SAl Viro struct dentry *d; 846248fb5b9SAl Viro nd->path = nd->root; 847248fb5b9SAl Viro d = nd->path.dentry; 848248fb5b9SAl Viro nd->inode = d->d_inode; 849248fb5b9SAl Viro nd->seq = nd->root_seq; 850248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 851248fb5b9SAl Viro return -ECHILD; 852248fb5b9SAl Viro } else { 853248fb5b9SAl Viro path_put(&nd->path); 854248fb5b9SAl Viro nd->path = nd->root; 855248fb5b9SAl Viro path_get(&nd->path); 856248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 857248fb5b9SAl Viro } 858248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 859248fb5b9SAl Viro return 0; 860248fb5b9SAl Viro } 861248fb5b9SAl Viro 862b5fb63c1SChristoph Hellwig /* 8636b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 864b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 865b5fb63c1SChristoph Hellwig */ 8666e77137bSAl Viro void nd_jump_link(struct path *path) 867b5fb63c1SChristoph Hellwig { 8686e77137bSAl Viro struct nameidata *nd = current->nameidata; 869b5fb63c1SChristoph Hellwig path_put(&nd->path); 870b5fb63c1SChristoph Hellwig 871b5fb63c1SChristoph Hellwig nd->path = *path; 872b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 873b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 874b5fb63c1SChristoph Hellwig } 875b5fb63c1SChristoph Hellwig 876b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 877574197e0SAl Viro { 87821c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 879fceef393SAl Viro do_delayed_call(&last->done); 8806548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 881b9ff4429SAl Viro path_put(&last->link); 882574197e0SAl Viro } 883574197e0SAl Viro 884561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 885561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 886800179c9SKees Cook 887800179c9SKees Cook /** 888800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 88955852635SRandy Dunlap * @nd: nameidata pathwalk data 890800179c9SKees Cook * 891800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 892800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 893800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 894800179c9SKees Cook * processes from failing races against path names that may change out 895800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 896800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 897800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 898800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 899800179c9SKees Cook * 900800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 901800179c9SKees Cook */ 902fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 903800179c9SKees Cook { 904800179c9SKees Cook const struct inode *inode; 905800179c9SKees Cook const struct inode *parent; 9062d7f9e2aSSeth Forshee kuid_t puid; 907800179c9SKees Cook 908800179c9SKees Cook if (!sysctl_protected_symlinks) 909800179c9SKees Cook return 0; 910800179c9SKees Cook 911800179c9SKees Cook /* Allowed if owner and follower match. */ 912fceef393SAl Viro inode = nd->link_inode; 91381abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 914800179c9SKees Cook return 0; 915800179c9SKees Cook 916800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 917aa65fa35SAl Viro parent = nd->inode; 918800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 919800179c9SKees Cook return 0; 920800179c9SKees Cook 921800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9222d7f9e2aSSeth Forshee puid = parent->i_uid; 9232d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 924800179c9SKees Cook return 0; 925800179c9SKees Cook 92631956502SAl Viro if (nd->flags & LOOKUP_RCU) 92731956502SAl Viro return -ECHILD; 92831956502SAl Viro 9291cf2665bSAl Viro audit_log_link_denied("follow_link", &nd->stack[0].link); 930800179c9SKees Cook return -EACCES; 931800179c9SKees Cook } 932800179c9SKees Cook 933800179c9SKees Cook /** 934800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 935800179c9SKees Cook * @inode: the source inode to hardlink from 936800179c9SKees Cook * 937800179c9SKees Cook * Return false if at least one of the following conditions: 938800179c9SKees Cook * - inode is not a regular file 939800179c9SKees Cook * - inode is setuid 940800179c9SKees Cook * - inode is setgid and group-exec 941800179c9SKees Cook * - access failure for read and write 942800179c9SKees Cook * 943800179c9SKees Cook * Otherwise returns true. 944800179c9SKees Cook */ 945800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 946800179c9SKees Cook { 947800179c9SKees Cook umode_t mode = inode->i_mode; 948800179c9SKees Cook 949800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 950800179c9SKees Cook if (!S_ISREG(mode)) 951800179c9SKees Cook return false; 952800179c9SKees Cook 953800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 954800179c9SKees Cook if (mode & S_ISUID) 955800179c9SKees Cook return false; 956800179c9SKees Cook 957800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 958800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 959800179c9SKees Cook return false; 960800179c9SKees Cook 961800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 962800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 963800179c9SKees Cook return false; 964800179c9SKees Cook 965800179c9SKees Cook return true; 966800179c9SKees Cook } 967800179c9SKees Cook 968800179c9SKees Cook /** 969800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 970800179c9SKees Cook * @link: the source to hardlink from 971800179c9SKees Cook * 972800179c9SKees Cook * Block hardlink when all of: 973800179c9SKees Cook * - sysctl_protected_hardlinks enabled 974800179c9SKees Cook * - fsuid does not match inode 975800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 976f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 977800179c9SKees Cook * 978800179c9SKees Cook * Returns 0 if successful, -ve on error. 979800179c9SKees Cook */ 980800179c9SKees Cook static int may_linkat(struct path *link) 981800179c9SKees Cook { 982800179c9SKees Cook struct inode *inode; 983800179c9SKees Cook 984800179c9SKees Cook if (!sysctl_protected_hardlinks) 985800179c9SKees Cook return 0; 986800179c9SKees Cook 987800179c9SKees Cook inode = link->dentry->d_inode; 988800179c9SKees Cook 989800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 990800179c9SKees Cook * otherwise, it must be a safe source. 991800179c9SKees Cook */ 992cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 993800179c9SKees Cook return 0; 994800179c9SKees Cook 995a51d9eaaSKees Cook audit_log_link_denied("linkat", link); 996800179c9SKees Cook return -EPERM; 997800179c9SKees Cook } 998800179c9SKees Cook 9993b2e7f75SAl Viro static __always_inline 10003b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10011da177e4SLinus Torvalds { 1002ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10031cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1004fceef393SAl Viro struct inode *inode = nd->link_inode; 10056d7b5aaeSAl Viro int error; 10060a959df5SAl Viro const char *res; 10071da177e4SLinus Torvalds 10088fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10098fa9dd24SNeilBrown touch_atime(&last->link); 10108fa9dd24SNeilBrown cond_resched(); 1011598e3c8fSMiklos Szeredi } else if (atime_needs_update_rcu(&last->link, inode)) { 10124675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10138fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10148fa9dd24SNeilBrown touch_atime(&last->link); 10158fa9dd24SNeilBrown } 10168fa9dd24SNeilBrown 1017bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1018bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1019bda0be7aSNeilBrown if (unlikely(error)) 10206920a440SAl Viro return ERR_PTR(error); 102136f3b4f6SAl Viro 102286acdca1SAl Viro nd->last_type = LAST_BIND; 1023d4dee48bSAl Viro res = inode->i_link; 1024d4dee48bSAl Viro if (!res) { 1025fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1026fceef393SAl Viro struct delayed_call *); 1027fceef393SAl Viro get = inode->i_op->get_link; 10288c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1029fceef393SAl Viro res = get(NULL, inode, &last->done); 10306b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10314675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10328c1b4566SAl Viro return ERR_PTR(-ECHILD); 1033fceef393SAl Viro res = get(dentry, inode, &last->done); 10348c1b4566SAl Viro } 10356b255391SAl Viro } else { 1036fceef393SAl Viro res = get(dentry, inode, &last->done); 10376b255391SAl Viro } 1038fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1039fab51e8aSAl Viro return res; 10400a959df5SAl Viro } 1041fab51e8aSAl Viro if (*res == '/') { 1042fab51e8aSAl Viro if (!nd->root.mnt) 1043fab51e8aSAl Viro set_root(nd); 1044248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1045fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1046fab51e8aSAl Viro while (unlikely(*++res == '/')) 1047fab51e8aSAl Viro ; 1048fab51e8aSAl Viro } 1049fab51e8aSAl Viro if (!*res) 1050fab51e8aSAl Viro res = NULL; 10510a959df5SAl Viro return res; 10520a959df5SAl Viro } 10536d7b5aaeSAl Viro 1054f015f126SDavid Howells /* 1055f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1056f015f126SDavid Howells * 1057f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1058f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1059f015f126SDavid Howells * Up is towards /. 1060f015f126SDavid Howells * 1061f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1062f015f126SDavid Howells * root. 1063f015f126SDavid Howells */ 1064bab77ebfSAl Viro int follow_up(struct path *path) 10651da177e4SLinus Torvalds { 10660714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10670714a533SAl Viro struct mount *parent; 10681da177e4SLinus Torvalds struct dentry *mountpoint; 106999b7db7bSNick Piggin 107048a066e7SAl Viro read_seqlock_excl(&mount_lock); 10710714a533SAl Viro parent = mnt->mnt_parent; 10723c0a6163SAl Viro if (parent == mnt) { 107348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10741da177e4SLinus Torvalds return 0; 10751da177e4SLinus Torvalds } 10760714a533SAl Viro mntget(&parent->mnt); 1077a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 107848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1079bab77ebfSAl Viro dput(path->dentry); 1080bab77ebfSAl Viro path->dentry = mountpoint; 1081bab77ebfSAl Viro mntput(path->mnt); 10820714a533SAl Viro path->mnt = &parent->mnt; 10831da177e4SLinus Torvalds return 1; 10841da177e4SLinus Torvalds } 10854d359507SAl Viro EXPORT_SYMBOL(follow_up); 10861da177e4SLinus Torvalds 1087b5c84bf6SNick Piggin /* 10889875cf80SDavid Howells * Perform an automount 10899875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10909875cf80SDavid Howells * were called with. 10911da177e4SLinus Torvalds */ 1092756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 10939875cf80SDavid Howells bool *need_mntput) 109431e6b01fSNick Piggin { 10959875cf80SDavid Howells struct vfsmount *mnt; 1096ea5b778aSDavid Howells int err; 10979875cf80SDavid Howells 10989875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 10999875cf80SDavid Howells return -EREMOTE; 11009875cf80SDavid Howells 11010ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11020ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11030ec26fd0SMiklos Szeredi * the name. 11040ec26fd0SMiklos Szeredi * 11050ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11065a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11070ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11080ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11090ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11100ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11115a30d8a2SDavid Howells */ 1112756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11135d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11145d38f049SIan Kent path->dentry->d_inode) 11159875cf80SDavid Howells return -EISDIR; 11160ec26fd0SMiklos Szeredi 1117756daf26SNeilBrown nd->total_link_count++; 1118756daf26SNeilBrown if (nd->total_link_count >= 40) 11199875cf80SDavid Howells return -ELOOP; 11209875cf80SDavid Howells 11219875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11229875cf80SDavid Howells if (IS_ERR(mnt)) { 11239875cf80SDavid Howells /* 11249875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11259875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11269875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11279875cf80SDavid Howells * 11289875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11299875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11309875cf80SDavid Howells * the path is inaccessible and we should say so. 11319875cf80SDavid Howells */ 1132756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11339875cf80SDavid Howells return -EREMOTE; 11349875cf80SDavid Howells return PTR_ERR(mnt); 113531e6b01fSNick Piggin } 1136ea5b778aSDavid Howells 11379875cf80SDavid Howells if (!mnt) /* mount collision */ 11389875cf80SDavid Howells return 0; 11399875cf80SDavid Howells 11408aef1884SAl Viro if (!*need_mntput) { 11418aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11428aef1884SAl Viro mntget(path->mnt); 11438aef1884SAl Viro *need_mntput = true; 11448aef1884SAl Viro } 114519a167afSAl Viro err = finish_automount(mnt, path); 1146ea5b778aSDavid Howells 1147ea5b778aSDavid Howells switch (err) { 1148ea5b778aSDavid Howells case -EBUSY: 1149ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 115019a167afSAl Viro return 0; 1151ea5b778aSDavid Howells case 0: 11528aef1884SAl Viro path_put(path); 11539875cf80SDavid Howells path->mnt = mnt; 11549875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11559875cf80SDavid Howells return 0; 115619a167afSAl Viro default: 115719a167afSAl Viro return err; 11589875cf80SDavid Howells } 115919a167afSAl Viro 1160ea5b778aSDavid Howells } 11619875cf80SDavid Howells 11629875cf80SDavid Howells /* 11639875cf80SDavid Howells * Handle a dentry that is managed in some way. 1164cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11659875cf80SDavid Howells * - Flagged as mountpoint 11669875cf80SDavid Howells * - Flagged as automount point 11679875cf80SDavid Howells * 11689875cf80SDavid Howells * This may only be called in refwalk mode. 11699875cf80SDavid Howells * 11709875cf80SDavid Howells * Serialization is taken care of in namespace.c 11719875cf80SDavid Howells */ 1172756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11739875cf80SDavid Howells { 11748aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11759875cf80SDavid Howells unsigned managed; 11769875cf80SDavid Howells bool need_mntput = false; 11778aef1884SAl Viro int ret = 0; 11789875cf80SDavid Howells 11799875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11809875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11819875cf80SDavid Howells * the components of that value change under us */ 11826aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 11839875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11849875cf80SDavid Howells unlikely(managed != 0)) { 1185cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1186cc53ce53SDavid Howells * being held. */ 1187cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1188cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1189cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1190fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1191cc53ce53SDavid Howells if (ret < 0) 11928aef1884SAl Viro break; 1193cc53ce53SDavid Howells } 1194cc53ce53SDavid Howells 11959875cf80SDavid Howells /* Transit to a mounted filesystem. */ 11969875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 11979875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11989875cf80SDavid Howells if (mounted) { 11999875cf80SDavid Howells dput(path->dentry); 12009875cf80SDavid Howells if (need_mntput) 1201463ffb2eSAl Viro mntput(path->mnt); 1202463ffb2eSAl Viro path->mnt = mounted; 1203463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12049875cf80SDavid Howells need_mntput = true; 12059875cf80SDavid Howells continue; 1206463ffb2eSAl Viro } 1207463ffb2eSAl Viro 12089875cf80SDavid Howells /* Something is mounted on this dentry in another 12099875cf80SDavid Howells * namespace and/or whatever was mounted there in this 121048a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 121148a066e7SAl Viro * get it */ 12121da177e4SLinus Torvalds } 12139875cf80SDavid Howells 12149875cf80SDavid Howells /* Handle an automount point */ 12159875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1216756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12179875cf80SDavid Howells if (ret < 0) 12188aef1884SAl Viro break; 12199875cf80SDavid Howells continue; 12209875cf80SDavid Howells } 12219875cf80SDavid Howells 12229875cf80SDavid Howells /* We didn't change the current path point */ 12239875cf80SDavid Howells break; 12249875cf80SDavid Howells } 12258aef1884SAl Viro 12268aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12278aef1884SAl Viro mntput(path->mnt); 1228e9742b53SAl Viro if (ret == -EISDIR || !ret) 1229e9742b53SAl Viro ret = 1; 12308402752eSAl Viro if (need_mntput) 12318402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12328402752eSAl Viro if (unlikely(ret < 0)) 12338402752eSAl Viro path_put_conditional(path, nd); 12348402752eSAl Viro return ret; 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds 1237cc53ce53SDavid Howells int follow_down_one(struct path *path) 12381da177e4SLinus Torvalds { 12391da177e4SLinus Torvalds struct vfsmount *mounted; 12401da177e4SLinus Torvalds 12411c755af4SAl Viro mounted = lookup_mnt(path); 12421da177e4SLinus Torvalds if (mounted) { 12439393bd07SAl Viro dput(path->dentry); 12449393bd07SAl Viro mntput(path->mnt); 12459393bd07SAl Viro path->mnt = mounted; 12469393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12471da177e4SLinus Torvalds return 1; 12481da177e4SLinus Torvalds } 12491da177e4SLinus Torvalds return 0; 12501da177e4SLinus Torvalds } 12514d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12521da177e4SLinus Torvalds 1253fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 125462a7375eSIan Kent { 1255fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1256fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 125762a7375eSIan Kent } 125862a7375eSIan Kent 12599875cf80SDavid Howells /* 1260287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1261287548e4SAl Viro * we meet a managed dentry that would need blocking. 12629875cf80SDavid Howells */ 12639875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1264254cf582SAl Viro struct inode **inode, unsigned *seqp) 12659875cf80SDavid Howells { 126662a7375eSIan Kent for (;;) { 1267c7105365SAl Viro struct mount *mounted; 126862a7375eSIan Kent /* 126962a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 127062a7375eSIan Kent * that wants to block transit. 127162a7375eSIan Kent */ 1272fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1273b8faf035SNeilBrown case -ECHILD: 1274b8faf035SNeilBrown default: 1275ab90911fSDavid Howells return false; 1276b8faf035SNeilBrown case -EISDIR: 1277b8faf035SNeilBrown return true; 1278b8faf035SNeilBrown case 0: 1279b8faf035SNeilBrown break; 1280b8faf035SNeilBrown } 128162a7375eSIan Kent 128262a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1283b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 128462a7375eSIan Kent 1285474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12869875cf80SDavid Howells if (!mounted) 12879875cf80SDavid Howells break; 1288c7105365SAl Viro path->mnt = &mounted->mnt; 1289c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1290a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1291254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 129259430262SLinus Torvalds /* 129359430262SLinus Torvalds * Update the inode too. We don't need to re-check the 129459430262SLinus Torvalds * dentry sequence number here after this d_inode read, 129559430262SLinus Torvalds * because a mount-point is always pinned. 129659430262SLinus Torvalds */ 129759430262SLinus Torvalds *inode = path->dentry->d_inode; 12989875cf80SDavid Howells } 1299f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1300b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1301287548e4SAl Viro } 1302287548e4SAl Viro 130331e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 130431e6b01fSNick Piggin { 13054023bfc9SAl Viro struct inode *inode = nd->inode; 130631e6b01fSNick Piggin 130731e6b01fSNick Piggin while (1) { 1308aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 130931e6b01fSNick Piggin break; 131031e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 131131e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 131231e6b01fSNick Piggin struct dentry *parent = old->d_parent; 131331e6b01fSNick Piggin unsigned seq; 131431e6b01fSNick Piggin 13154023bfc9SAl Viro inode = parent->d_inode; 131631e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1317aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1318aed434adSAl Viro return -ECHILD; 131931e6b01fSNick Piggin nd->path.dentry = parent; 132031e6b01fSNick Piggin nd->seq = seq; 1321397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1322397d425dSEric W. Biederman return -ENOENT; 132331e6b01fSNick Piggin break; 1324aed434adSAl Viro } else { 1325aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1326aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1327aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1328aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1329aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1330aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1331aed434adSAl Viro return -ECHILD; 1332aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 133331e6b01fSNick Piggin break; 1334aed434adSAl Viro /* we know that mountpoint was pinned */ 1335aed434adSAl Viro nd->path.dentry = mountpoint; 1336aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1337aed434adSAl Viro inode = inode2; 1338aed434adSAl Viro nd->seq = seq; 133931e6b01fSNick Piggin } 1340aed434adSAl Viro } 1341aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1342b37199e6SAl Viro struct mount *mounted; 1343b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1344aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1345aed434adSAl Viro return -ECHILD; 1346b37199e6SAl Viro if (!mounted) 1347b37199e6SAl Viro break; 1348b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1349b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13504023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1351b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1352b37199e6SAl Viro } 13534023bfc9SAl Viro nd->inode = inode; 135431e6b01fSNick Piggin return 0; 135531e6b01fSNick Piggin } 135631e6b01fSNick Piggin 13579875cf80SDavid Howells /* 1358cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1359cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1360cc53ce53SDavid Howells * caller is permitted to proceed or not. 1361cc53ce53SDavid Howells */ 13627cc90cc3SAl Viro int follow_down(struct path *path) 1363cc53ce53SDavid Howells { 1364cc53ce53SDavid Howells unsigned managed; 1365cc53ce53SDavid Howells int ret; 1366cc53ce53SDavid Howells 13676aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1368cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1369cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1370cc53ce53SDavid Howells * being held. 1371cc53ce53SDavid Howells * 1372cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1373cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1374cc53ce53SDavid Howells * other than its daemon the right to mount on its 1375cc53ce53SDavid Howells * superstructure. 1376cc53ce53SDavid Howells * 1377cc53ce53SDavid Howells * The filesystem may sleep at this point. 1378cc53ce53SDavid Howells */ 1379cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1380cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1381cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1382fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1383cc53ce53SDavid Howells if (ret < 0) 1384cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1385cc53ce53SDavid Howells } 1386cc53ce53SDavid Howells 1387cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1388cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1389cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1390cc53ce53SDavid Howells if (!mounted) 1391cc53ce53SDavid Howells break; 1392cc53ce53SDavid Howells dput(path->dentry); 1393cc53ce53SDavid Howells mntput(path->mnt); 1394cc53ce53SDavid Howells path->mnt = mounted; 1395cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1396cc53ce53SDavid Howells continue; 1397cc53ce53SDavid Howells } 1398cc53ce53SDavid Howells 1399cc53ce53SDavid Howells /* Don't handle automount points here */ 1400cc53ce53SDavid Howells break; 1401cc53ce53SDavid Howells } 1402cc53ce53SDavid Howells return 0; 1403cc53ce53SDavid Howells } 14044d359507SAl Viro EXPORT_SYMBOL(follow_down); 1405cc53ce53SDavid Howells 1406cc53ce53SDavid Howells /* 14079875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14089875cf80SDavid Howells */ 14099875cf80SDavid Howells static void follow_mount(struct path *path) 14109875cf80SDavid Howells { 14119875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14129875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14139875cf80SDavid Howells if (!mounted) 14149875cf80SDavid Howells break; 14159875cf80SDavid Howells dput(path->dentry); 14169875cf80SDavid Howells mntput(path->mnt); 14179875cf80SDavid Howells path->mnt = mounted; 14189875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14199875cf80SDavid Howells } 14209875cf80SDavid Howells } 14219875cf80SDavid Howells 1422eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1423eedf265aSEric W. Biederman { 1424eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1425eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1426eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1427eedf265aSEric W. Biederman dput(old); 1428eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1429eedf265aSEric W. Biederman return -ENOENT; 1430eedf265aSEric W. Biederman return 0; 1431eedf265aSEric W. Biederman } 1432eedf265aSEric W. Biederman 1433397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14341da177e4SLinus Torvalds { 14351da177e4SLinus Torvalds while(1) { 14362a737871SAl Viro if (nd->path.dentry == nd->root.dentry && 14372a737871SAl Viro nd->path.mnt == nd->root.mnt) { 14381da177e4SLinus Torvalds break; 14391da177e4SLinus Torvalds } 14404ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1441eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1442eedf265aSEric W. Biederman if (ret) 1443eedf265aSEric W. Biederman return ret; 14441da177e4SLinus Torvalds break; 14451da177e4SLinus Torvalds } 14463088dd70SAl Viro if (!follow_up(&nd->path)) 14471da177e4SLinus Torvalds break; 14481da177e4SLinus Torvalds } 144979ed0226SAl Viro follow_mount(&nd->path); 145031e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1451397d425dSEric W. Biederman return 0; 14521da177e4SLinus Torvalds } 14531da177e4SLinus Torvalds 14541da177e4SLinus Torvalds /* 1455f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1456f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1457baa03890SNick Piggin */ 1458e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1459e3c13928SAl Viro struct dentry *dir, 14606c51e513SAl Viro unsigned int flags) 1461baa03890SNick Piggin { 1462a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1463bad61189SMiklos Szeredi if (dentry) { 1464a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1465bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 146674ff0ffcSAl Viro if (!error) 14675542aa2fSEric W. Biederman d_invalidate(dentry); 1468bad61189SMiklos Szeredi dput(dentry); 146974ff0ffcSAl Viro return ERR_PTR(error); 1470bad61189SMiklos Szeredi } 1471bad61189SMiklos Szeredi } 1472baa03890SNick Piggin return dentry; 1473baa03890SNick Piggin } 1474baa03890SNick Piggin 1475baa03890SNick Piggin /* 147613a2c3beSJ. Bruce Fields * Call i_op->lookup on the dentry. The dentry must be negative and 147713a2c3beSJ. Bruce Fields * unhashed. 1478bad61189SMiklos Szeredi * 1479bad61189SMiklos Szeredi * dir->d_inode->i_mutex must be held 148044396f4bSJosef Bacik */ 1481bad61189SMiklos Szeredi static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry, 148272bd866aSAl Viro unsigned int flags) 148344396f4bSJosef Bacik { 148444396f4bSJosef Bacik struct dentry *old; 148544396f4bSJosef Bacik 148644396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1487bad61189SMiklos Szeredi if (unlikely(IS_DEADDIR(dir))) { 1488e188dc02SMiklos Szeredi dput(dentry); 148944396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1490e188dc02SMiklos Szeredi } 149144396f4bSJosef Bacik 149272bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 149344396f4bSJosef Bacik if (unlikely(old)) { 149444396f4bSJosef Bacik dput(dentry); 149544396f4bSJosef Bacik dentry = old; 149644396f4bSJosef Bacik } 149744396f4bSJosef Bacik return dentry; 149844396f4bSJosef Bacik } 149944396f4bSJosef Bacik 1500e3c13928SAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 150172bd866aSAl Viro struct dentry *base, unsigned int flags) 1502a3255546SAl Viro { 15036c51e513SAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 1504a3255546SAl Viro 15056c51e513SAl Viro if (dentry) 1506a3255546SAl Viro return dentry; 1507bad61189SMiklos Szeredi 15086c51e513SAl Viro dentry = d_alloc(base, name); 15096c51e513SAl Viro if (unlikely(!dentry)) 15106c51e513SAl Viro return ERR_PTR(-ENOMEM); 15116c51e513SAl Viro 151272bd866aSAl Viro return lookup_real(base->d_inode, dentry, flags); 1513a3255546SAl Viro } 1514a3255546SAl Viro 1515e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1516254cf582SAl Viro struct path *path, struct inode **inode, 1517254cf582SAl Viro unsigned *seqp) 15181da177e4SLinus Torvalds { 15194ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 152031e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15215a18fff2SAl Viro int status = 1; 15229875cf80SDavid Howells int err; 15239875cf80SDavid Howells 15243cac260aSAl Viro /* 1525b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15265d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15275d0f49c1SAl Viro * going to fall back to non-racy lookup. 1528b04f784eSNick Piggin */ 152931e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 153031e6b01fSNick Piggin unsigned seq; 1531766c4cbfSAl Viro bool negative; 1532da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15335d0f49c1SAl Viro if (unlikely(!dentry)) { 15344675ac39SAl Viro if (unlazy_walk(nd)) 15355d0f49c1SAl Viro return -ECHILD; 1536e9742b53SAl Viro return 0; 15375d0f49c1SAl Viro } 15385a18fff2SAl Viro 153912f8ad4bSLinus Torvalds /* 154012f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 154112f8ad4bSLinus Torvalds * the dentry name information from lookup. 154212f8ad4bSLinus Torvalds */ 154363afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1544766c4cbfSAl Viro negative = d_is_negative(dentry); 15455d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 154612f8ad4bSLinus Torvalds return -ECHILD; 154712f8ad4bSLinus Torvalds 154812f8ad4bSLinus Torvalds /* 154912f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 155012f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 155112f8ad4bSLinus Torvalds * 155212f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 155312f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 155412f8ad4bSLinus Torvalds */ 15555d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 155631e6b01fSNick Piggin return -ECHILD; 15575a18fff2SAl Viro 1558254cf582SAl Viro *seqp = seq; 15594ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1560209a7fb2SAl Viro if (likely(status > 0)) { 1561daf3761cSTrond Myklebust /* 1562daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1563daf3761cSTrond Myklebust * case that drops it. 1564daf3761cSTrond Myklebust */ 15655d0f49c1SAl Viro if (unlikely(negative)) 1566daf3761cSTrond Myklebust return -ENOENT; 156731e6b01fSNick Piggin path->mnt = mnt; 156831e6b01fSNick Piggin path->dentry = dentry; 1569254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1570e9742b53SAl Viro return 1; 15715d0f49c1SAl Viro } 15724675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 1573254cf582SAl Viro return -ECHILD; 1574209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1575209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1576209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 15775a18fff2SAl Viro } else { 1578e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 157981e6f520SAl Viro if (unlikely(!dentry)) 1580e9742b53SAl Viro return 0; 15814ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15825d0f49c1SAl Viro } 15835a18fff2SAl Viro if (unlikely(status <= 0)) { 1584e9742b53SAl Viro if (!status) 15855d0f49c1SAl Viro d_invalidate(dentry); 15865a18fff2SAl Viro dput(dentry); 15875a18fff2SAl Viro return status; 15885a18fff2SAl Viro } 1589766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1590766c4cbfSAl Viro dput(dentry); 1591766c4cbfSAl Viro return -ENOENT; 1592766c4cbfSAl Viro } 15935d0f49c1SAl Viro 15941da177e4SLinus Torvalds path->mnt = mnt; 15951da177e4SLinus Torvalds path->dentry = dentry; 1596756daf26SNeilBrown err = follow_managed(path, nd); 1597e9742b53SAl Viro if (likely(err > 0)) 159863afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 15998402752eSAl Viro return err; 1600697f514dSMiklos Szeredi } 1601697f514dSMiklos Szeredi 1602697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 1603e3c13928SAl Viro static struct dentry *lookup_slow(const struct qstr *name, 1604e3c13928SAl Viro struct dentry *dir, 1605e3c13928SAl Viro unsigned int flags) 1606697f514dSMiklos Szeredi { 160794bdd655SAl Viro struct dentry *dentry = ERR_PTR(-ENOENT), *old; 16081936386eSAl Viro struct inode *inode = dir->d_inode; 1609d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16101936386eSAl Viro 16119902af79SAl Viro inode_lock_shared(inode); 16121936386eSAl Viro /* Don't go there if it's already dead */ 161394bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 161494bdd655SAl Viro goto out; 161594bdd655SAl Viro again: 1616d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 161794bdd655SAl Viro if (IS_ERR(dentry)) 161894bdd655SAl Viro goto out; 161994bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1620a89f8337SAl Viro if (!(flags & LOOKUP_NO_REVAL)) { 1621949a852eSAl Viro int error = d_revalidate(dentry, flags); 1622949a852eSAl Viro if (unlikely(error <= 0)) { 162394bdd655SAl Viro if (!error) { 1624949a852eSAl Viro d_invalidate(dentry); 1625949a852eSAl Viro dput(dentry); 162694bdd655SAl Viro goto again; 162794bdd655SAl Viro } 162894bdd655SAl Viro dput(dentry); 1629949a852eSAl Viro dentry = ERR_PTR(error); 1630949a852eSAl Viro } 1631949a852eSAl Viro } 163294bdd655SAl Viro } else { 16331936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 163485c7f810SAl Viro d_lookup_done(dentry); 16351936386eSAl Viro if (unlikely(old)) { 16361936386eSAl Viro dput(dentry); 16371936386eSAl Viro dentry = old; 1638949a852eSAl Viro } 1639949a852eSAl Viro } 164094bdd655SAl Viro out: 16419902af79SAl Viro inode_unlock_shared(inode); 1642e3c13928SAl Viro return dentry; 16431da177e4SLinus Torvalds } 16441da177e4SLinus Torvalds 164552094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 164652094c8aSAl Viro { 164752094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16484ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 164952094c8aSAl Viro if (err != -ECHILD) 165052094c8aSAl Viro return err; 16514675ac39SAl Viro if (unlazy_walk(nd)) 165252094c8aSAl Viro return -ECHILD; 165352094c8aSAl Viro } 16544ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 165552094c8aSAl Viro } 165652094c8aSAl Viro 16579856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16589856fa1bSAl Viro { 16599856fa1bSAl Viro if (type == LAST_DOTDOT) { 16609e6697e2SAl Viro if (!nd->root.mnt) 16619e6697e2SAl Viro set_root(nd); 16629856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 166370291aecSAl Viro return follow_dotdot_rcu(nd); 16649856fa1bSAl Viro } else 1665397d425dSEric W. Biederman return follow_dotdot(nd); 16669856fa1bSAl Viro } 16679856fa1bSAl Viro return 0; 16689856fa1bSAl Viro } 16699856fa1bSAl Viro 1670181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1671181548c0SAl Viro struct inode *inode, unsigned seq) 1672d63ff28fSAl Viro { 1673626de996SAl Viro int error; 16741cf2665bSAl Viro struct saved *last; 1675756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1676626de996SAl Viro path_to_nameidata(link, nd); 1677626de996SAl Viro return -ELOOP; 1678626de996SAl Viro } 1679bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1680cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1681cd179f44SAl Viro mntget(link->mnt); 16827973387aSAl Viro } 1683626de996SAl Viro error = nd_alloc_stack(nd); 1684626de996SAl Viro if (unlikely(error)) { 1685bc40aee0SAl Viro if (error == -ECHILD) { 1686ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1687ad1633a1SAl Viro drop_links(nd); 1688ad1633a1SAl Viro nd->depth = 0; 1689ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1690ad1633a1SAl Viro nd->path.mnt = NULL; 1691ad1633a1SAl Viro nd->path.dentry = NULL; 1692ad1633a1SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1693ad1633a1SAl Viro nd->root.mnt = NULL; 1694ad1633a1SAl Viro rcu_read_unlock(); 16954675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1696bc40aee0SAl Viro error = nd_alloc_stack(nd); 1697bc40aee0SAl Viro } 1698bc40aee0SAl Viro if (error) { 1699cd179f44SAl Viro path_put(link); 1700626de996SAl Viro return error; 1701626de996SAl Viro } 1702bc40aee0SAl Viro } 1703626de996SAl Viro 1704ab104923SAl Viro last = nd->stack + nd->depth++; 17051cf2665bSAl Viro last->link = *link; 1706fceef393SAl Viro clear_delayed_call(&last->done); 1707fceef393SAl Viro nd->link_inode = inode; 17080450b2d1SAl Viro last->seq = seq; 1709d63ff28fSAl Viro return 1; 1710d63ff28fSAl Viro } 1711d63ff28fSAl Viro 17128f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 171331d66bcdSAl Viro 17143ddcd056SLinus Torvalds /* 17153ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17163ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17173ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17183ddcd056SLinus Torvalds * for the common case. 17193ddcd056SLinus Torvalds */ 17208f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 17218f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 17223ddcd056SLinus Torvalds { 172331d66bcdSAl Viro if (!(flags & WALK_MORE) && nd->depth) 172431d66bcdSAl Viro put_link(nd); 17258f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 17268f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 17278f64fb1cSAl Viro /* not a symlink or should not follow */ 17288f64fb1cSAl Viro path_to_nameidata(path, nd); 17298f64fb1cSAl Viro nd->inode = inode; 17308f64fb1cSAl Viro nd->seq = seq; 1731d63ff28fSAl Viro return 0; 17328f64fb1cSAl Viro } 1733a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1734a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 17358f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1736a7f77542SAl Viro return -ECHILD; 1737a7f77542SAl Viro } 17388f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 17393ddcd056SLinus Torvalds } 17403ddcd056SLinus Torvalds 17414693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1742ce57dfc1SAl Viro { 1743caa85634SAl Viro struct path path; 1744ce57dfc1SAl Viro struct inode *inode; 1745254cf582SAl Viro unsigned seq; 1746ce57dfc1SAl Viro int err; 1747ce57dfc1SAl Viro /* 1748ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1749ce57dfc1SAl Viro * to be able to know about the current root directory and 1750ce57dfc1SAl Viro * parent relationships. 1751ce57dfc1SAl Viro */ 17524693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17534693a547SAl Viro err = handle_dots(nd, nd->last_type); 17541c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 17554693a547SAl Viro put_link(nd); 17564693a547SAl Viro return err; 17574693a547SAl Viro } 1758254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1759e9742b53SAl Viro if (unlikely(err <= 0)) { 1760697f514dSMiklos Szeredi if (err < 0) 1761f0a9ba70SAl Viro return err; 1762e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1763e3c13928SAl Viro nd->flags); 1764e3c13928SAl Viro if (IS_ERR(path.dentry)) 1765e3c13928SAl Viro return PTR_ERR(path.dentry); 17667500c38aSAl Viro 1767e3c13928SAl Viro path.mnt = nd->path.mnt; 1768e3c13928SAl Viro err = follow_managed(&path, nd); 1769e3c13928SAl Viro if (unlikely(err < 0)) 1770f0a9ba70SAl Viro return err; 1771697f514dSMiklos Szeredi 17727500c38aSAl Viro if (unlikely(d_is_negative(path.dentry))) { 17737500c38aSAl Viro path_to_nameidata(&path, nd); 17747500c38aSAl Viro return -ENOENT; 17757500c38aSAl Viro } 17767500c38aSAl Viro 1777254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1778d4565649SAl Viro inode = d_backing_inode(path.dentry); 1779766c4cbfSAl Viro } 1780697f514dSMiklos Szeredi 17818f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1782ce57dfc1SAl Viro } 1783ce57dfc1SAl Viro 17841da177e4SLinus Torvalds /* 1785bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1786bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1787bfcfaa77SLinus Torvalds * 1788bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1789bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1790bfcfaa77SLinus Torvalds * fast. 1791bfcfaa77SLinus Torvalds * 1792bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1793bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1794bfcfaa77SLinus Torvalds * crossing operation. 1795bfcfaa77SLinus Torvalds * 1796bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1797bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1798bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1799bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1800bfcfaa77SLinus Torvalds */ 1801bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1802bfcfaa77SLinus Torvalds 1803f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1804bfcfaa77SLinus Torvalds 1805468a9428SGeorge Spelvin #ifdef HASH_MIX 1806bfcfaa77SLinus Torvalds 1807468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1808468a9428SGeorge Spelvin 1809468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18102a18da7aSGeorge Spelvin /* 18112a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18122a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18132a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18142a18da7aSGeorge Spelvin * and no temporaries. 18152a18da7aSGeorge Spelvin * 18162a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18172a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18182a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18192a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18202a18da7aSGeorge Spelvin * 18212a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18222a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18232a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18242a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18252a18da7aSGeorge Spelvin * 18262a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18272a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18282a18da7aSGeorge Spelvin * toggling any given output bit. 18292a18da7aSGeorge Spelvin * 18302a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18312a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18322a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18332a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18342a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18352a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18362a18da7aSGeorge Spelvin * Perfect: 8192 258048 18372a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18382a18da7aSGeorge Spelvin */ 18392a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18402a18da7aSGeorge Spelvin ( x ^= (a), \ 18412a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18422a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 18432a18da7aSGeorge Spelvin y *= 9 ) 1844bfcfaa77SLinus Torvalds 18450fed3ac8SGeorge Spelvin /* 18462a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 18472a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 18482a18da7aSGeorge Spelvin * work done before the hash value is used. 18490fed3ac8SGeorge Spelvin */ 18502a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18510fed3ac8SGeorge Spelvin { 18522a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 18532a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 18542a18da7aSGeorge Spelvin return y >> 32; 18550fed3ac8SGeorge Spelvin } 18560fed3ac8SGeorge Spelvin 1857bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1858bfcfaa77SLinus Torvalds 18592a18da7aSGeorge Spelvin /* 18602a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 18612a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18622a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 18632a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 18642a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 18652a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 18662a18da7aSGeorge Spelvin * Perfect: 2048 31744 18672a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 18682a18da7aSGeorge Spelvin */ 18692a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18702a18da7aSGeorge Spelvin ( x ^= (a), \ 18712a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 18722a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 18732a18da7aSGeorge Spelvin y *= 9 ) 1874bfcfaa77SLinus Torvalds 18752a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18760fed3ac8SGeorge Spelvin { 18772a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 18782a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 18790fed3ac8SGeorge Spelvin } 18800fed3ac8SGeorge Spelvin 1881bfcfaa77SLinus Torvalds #endif 1882bfcfaa77SLinus Torvalds 18832a18da7aSGeorge Spelvin /* 18842a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 18852a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 18862a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 18872a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 18882a18da7aSGeorge Spelvin * finds the delimiter after the name. 18892a18da7aSGeorge Spelvin */ 18908387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1891bfcfaa77SLinus Torvalds { 18928387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1893bfcfaa77SLinus Torvalds 1894bfcfaa77SLinus Torvalds for (;;) { 1895fcfd2fbfSGeorge Spelvin if (!len) 1896fcfd2fbfSGeorge Spelvin goto done; 1897e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1898bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1899bfcfaa77SLinus Torvalds break; 19002a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1901bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1902bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1903bfcfaa77SLinus Torvalds } 19042a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1905bfcfaa77SLinus Torvalds done: 19062a18da7aSGeorge Spelvin return fold_hash(x, y); 1907bfcfaa77SLinus Torvalds } 1908bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1909bfcfaa77SLinus Torvalds 1910fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19118387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1912fcfd2fbfSGeorge Spelvin { 19138387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19148387ff25SLinus Torvalds unsigned long adata, mask, len; 1915fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1916fcfd2fbfSGeorge Spelvin 19178387ff25SLinus Torvalds len = 0; 19188387ff25SLinus Torvalds goto inside; 19198387ff25SLinus Torvalds 1920fcfd2fbfSGeorge Spelvin do { 19212a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1922fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19238387ff25SLinus Torvalds inside: 1924fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1925fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1926fcfd2fbfSGeorge Spelvin 1927fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1928fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19292a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1930fcfd2fbfSGeorge Spelvin 19312a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1932fcfd2fbfSGeorge Spelvin } 1933fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1934fcfd2fbfSGeorge Spelvin 1935bfcfaa77SLinus Torvalds /* 1936bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1937d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1938bfcfaa77SLinus Torvalds */ 19398387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1940bfcfaa77SLinus Torvalds { 19418387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 19428387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 194336126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1944bfcfaa77SLinus Torvalds 19458387ff25SLinus Torvalds len = 0; 19468387ff25SLinus Torvalds goto inside; 19478387ff25SLinus Torvalds 1948bfcfaa77SLinus Torvalds do { 19492a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1950bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 19518387ff25SLinus Torvalds inside: 1952e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 195336126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 195436126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1955bfcfaa77SLinus Torvalds 195636126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 195736126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 195836126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 19592a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 196036126f8fSLinus Torvalds 19612a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1962bfcfaa77SLinus Torvalds } 1963bfcfaa77SLinus Torvalds 19642a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 1965bfcfaa77SLinus Torvalds 1966fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 19678387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 19680145acc2SLinus Torvalds { 19698387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 19700145acc2SLinus Torvalds while (len--) 1971fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 19720145acc2SLinus Torvalds return end_name_hash(hash); 19730145acc2SLinus Torvalds } 1974ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 19750145acc2SLinus Torvalds 1976fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19778387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1978fcfd2fbfSGeorge Spelvin { 19798387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 1980fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 1981fcfd2fbfSGeorge Spelvin 1982fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 1983e0ab7af9SGeorge Spelvin while (c) { 1984fcfd2fbfSGeorge Spelvin len++; 1985fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 1986fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 1987e0ab7af9SGeorge Spelvin } 1988fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 1989fcfd2fbfSGeorge Spelvin } 1990f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1991fcfd2fbfSGeorge Spelvin 19923ddcd056SLinus Torvalds /* 1993200e9ef7SLinus Torvalds * We know there's a real path component here of at least 1994200e9ef7SLinus Torvalds * one character. 1995200e9ef7SLinus Torvalds */ 19968387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1997200e9ef7SLinus Torvalds { 19988387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 1999200e9ef7SLinus Torvalds unsigned long len = 0, c; 2000200e9ef7SLinus Torvalds 2001200e9ef7SLinus Torvalds c = (unsigned char)*name; 2002200e9ef7SLinus Torvalds do { 2003200e9ef7SLinus Torvalds len++; 2004200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2005200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2006200e9ef7SLinus Torvalds } while (c && c != '/'); 2007d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2008200e9ef7SLinus Torvalds } 2009200e9ef7SLinus Torvalds 2010bfcfaa77SLinus Torvalds #endif 2011bfcfaa77SLinus Torvalds 2012200e9ef7SLinus Torvalds /* 20131da177e4SLinus Torvalds * Name resolution. 2014ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2015ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20161da177e4SLinus Torvalds * 2017ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2018ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20191da177e4SLinus Torvalds */ 20206de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20211da177e4SLinus Torvalds { 20221da177e4SLinus Torvalds int err; 20231da177e4SLinus Torvalds 20241da177e4SLinus Torvalds while (*name=='/') 20251da177e4SLinus Torvalds name++; 20261da177e4SLinus Torvalds if (!*name) 20279e18f10aSAl Viro return 0; 20281da177e4SLinus Torvalds 20291da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20301da177e4SLinus Torvalds for(;;) { 2031d6bb3e90SLinus Torvalds u64 hash_len; 2032fe479a58SAl Viro int type; 20331da177e4SLinus Torvalds 203452094c8aSAl Viro err = may_lookup(nd); 20351da177e4SLinus Torvalds if (err) 20363595e234SAl Viro return err; 20371da177e4SLinus Torvalds 20388387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 20391da177e4SLinus Torvalds 2040fe479a58SAl Viro type = LAST_NORM; 2041d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2042fe479a58SAl Viro case 2: 2043200e9ef7SLinus Torvalds if (name[1] == '.') { 2044fe479a58SAl Viro type = LAST_DOTDOT; 204516c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 204616c2cd71SAl Viro } 2047fe479a58SAl Viro break; 2048fe479a58SAl Viro case 1: 2049fe479a58SAl Viro type = LAST_DOT; 2050fe479a58SAl Viro } 20515a202bcdSAl Viro if (likely(type == LAST_NORM)) { 20525a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 205316c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 20545a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2055a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2056da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 20575a202bcdSAl Viro if (err < 0) 20583595e234SAl Viro return err; 2059d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2060d6bb3e90SLinus Torvalds name = this.name; 20615a202bcdSAl Viro } 20625a202bcdSAl Viro } 2063fe479a58SAl Viro 2064d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2065d6bb3e90SLinus Torvalds nd->last.name = name; 20665f4a6a69SAl Viro nd->last_type = type; 20675f4a6a69SAl Viro 2068d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2069d6bb3e90SLinus Torvalds if (!*name) 2070bdf6cbf1SAl Viro goto OK; 2071200e9ef7SLinus Torvalds /* 2072200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2073200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2074200e9ef7SLinus Torvalds */ 2075200e9ef7SLinus Torvalds do { 2076d6bb3e90SLinus Torvalds name++; 2077d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 20788620c238SAl Viro if (unlikely(!*name)) { 20798620c238SAl Viro OK: 2080368ee9baSAl Viro /* pathname body, done */ 20818620c238SAl Viro if (!nd->depth) 20828620c238SAl Viro return 0; 20838620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2084368ee9baSAl Viro /* trailing symlink, done */ 20858620c238SAl Viro if (!name) 20868620c238SAl Viro return 0; 20878620c238SAl Viro /* last component of nested symlink */ 20888f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 20891c4ff1a8SAl Viro } else { 20901c4ff1a8SAl Viro /* not the last component */ 20918f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 20928620c238SAl Viro } 2093ce57dfc1SAl Viro if (err < 0) 20943595e234SAl Viro return err; 2095fe479a58SAl Viro 2096ce57dfc1SAl Viro if (err) { 2097626de996SAl Viro const char *s = get_link(nd); 20985a460275SAl Viro 2099a1c83681SViresh Kumar if (IS_ERR(s)) 21003595e234SAl Viro return PTR_ERR(s); 2101172a39a0SAl Viro err = 0; 210212b09578SAl Viro if (unlikely(!s)) { 210312b09578SAl Viro /* jumped */ 2104b9ff4429SAl Viro put_link(nd); 210512b09578SAl Viro } else { 2106071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 210732cd7468SAl Viro name = s; 21089e18f10aSAl Viro continue; 210948c8b0c5SAl Viro } 211031e6b01fSNick Piggin } 211197242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 211297242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21134675ac39SAl Viro if (unlazy_walk(nd)) 211497242f99SAl Viro return -ECHILD; 211597242f99SAl Viro } 21163595e234SAl Viro return -ENOTDIR; 21175f4a6a69SAl Viro } 2118ce57dfc1SAl Viro } 211997242f99SAl Viro } 21201da177e4SLinus Torvalds 2121c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 212231e6b01fSNick Piggin { 2123c8a53ee5SAl Viro const char *s = nd->name->name; 212431e6b01fSNick Piggin 2125c0eb027eSLinus Torvalds if (!*s) 2126c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2127c0eb027eSLinus Torvalds 212831e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2129980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 213031e6b01fSNick Piggin nd->depth = 0; 21315b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2132b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2133b18825a7SDavid Howells struct inode *inode = root->d_inode; 213493893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2135368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21365b6ca027SAl Viro nd->path = nd->root; 21375b6ca027SAl Viro nd->inode = inode; 21385b6ca027SAl Viro if (flags & LOOKUP_RCU) { 21398b61e74fSAl Viro rcu_read_lock(); 21405b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 21418f47a016SAl Viro nd->root_seq = nd->seq; 214248a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 21435b6ca027SAl Viro } else { 21445b6ca027SAl Viro path_get(&nd->path); 21455b6ca027SAl Viro } 2146368ee9baSAl Viro return s; 21475b6ca027SAl Viro } 21485b6ca027SAl Viro 214931e6b01fSNick Piggin nd->root.mnt = NULL; 2150248fb5b9SAl Viro nd->path.mnt = NULL; 2151248fb5b9SAl Viro nd->path.dentry = NULL; 215231e6b01fSNick Piggin 215348a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2154fd2f7cb5SAl Viro if (*s == '/') { 21559e6697e2SAl Viro if (flags & LOOKUP_RCU) 21568b61e74fSAl Viro rcu_read_lock(); 2157e41f7d4eSAl Viro set_root(nd); 2158248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2159ef55d917SAl Viro return s; 2160ef55d917SAl Viro nd->root.mnt = NULL; 2161ef55d917SAl Viro rcu_read_unlock(); 2162ef55d917SAl Viro return ERR_PTR(-ECHILD); 2163c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2164e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 216531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2166c28cc364SNick Piggin unsigned seq; 216731e6b01fSNick Piggin 21688b61e74fSAl Viro rcu_read_lock(); 216931e6b01fSNick Piggin 2170c28cc364SNick Piggin do { 2171c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 217231e6b01fSNick Piggin nd->path = fs->pwd; 2173ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2174c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2175c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2176e41f7d4eSAl Viro } else { 2177e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2178ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2179e41f7d4eSAl Viro } 2180ef55d917SAl Viro return s; 218131e6b01fSNick Piggin } else { 2182582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2183c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 218431e6b01fSNick Piggin struct dentry *dentry; 218531e6b01fSNick Piggin 21862903ff01SAl Viro if (!f.file) 2187368ee9baSAl Viro return ERR_PTR(-EBADF); 218831e6b01fSNick Piggin 21892903ff01SAl Viro dentry = f.file->f_path.dentry; 219031e6b01fSNick Piggin 2191fd2f7cb5SAl Viro if (*s) { 219244b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 21932903ff01SAl Viro fdput(f); 2194368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2195f52e0c11SAl Viro } 21962903ff01SAl Viro } 21972903ff01SAl Viro 21982903ff01SAl Viro nd->path = f.file->f_path; 2199e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 22008b61e74fSAl Viro rcu_read_lock(); 220134a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 220234a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22035590ff0dSUlrich Drepper } else { 22042903ff01SAl Viro path_get(&nd->path); 220534a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22061da177e4SLinus Torvalds } 220734a26b99SAl Viro fdput(f); 2208368ee9baSAl Viro return s; 2209e41f7d4eSAl Viro } 22109b4a9b14SAl Viro } 22119b4a9b14SAl Viro 22123bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 221395fa25d9SAl Viro { 221495fa25d9SAl Viro const char *s; 2215fec2fa24SAl Viro int error = may_follow_link(nd); 2216deb106c6SAl Viro if (unlikely(error)) 22173bdba28bSAl Viro return ERR_PTR(error); 221895fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2219fab51e8aSAl Viro nd->stack[0].name = NULL; 22203b2e7f75SAl Viro s = get_link(nd); 2221deb106c6SAl Viro return s ? s : ""; 222295fa25d9SAl Viro } 222395fa25d9SAl Viro 2224caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2225bd92d7feSAl Viro { 2226bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2227bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2228bd92d7feSAl Viro 2229bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 22301c4ff1a8SAl Viro return walk_component(nd, 0); 2231bd92d7feSAl Viro } 2232bd92d7feSAl Viro 22334f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22344f757f3cSAl Viro { 22354f757f3cSAl Viro struct path path = nd->path; 22364f757f3cSAl Viro struct inode *inode = nd->inode; 22374f757f3cSAl Viro unsigned seq = nd->seq; 22384f757f3cSAl Viro int err; 22394f757f3cSAl Viro 22404f757f3cSAl Viro if (nd->flags & LOOKUP_RCU) { 22414f757f3cSAl Viro /* 22424f757f3cSAl Viro * don't bother with unlazy_walk on failure - we are 22434f757f3cSAl Viro * at the very beginning of walk, so we lose nothing 22444f757f3cSAl Viro * if we simply redo everything in non-RCU mode 22454f757f3cSAl Viro */ 22464f757f3cSAl Viro if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq))) 22474f757f3cSAl Viro return -ECHILD; 22484f757f3cSAl Viro } else { 22494f757f3cSAl Viro dget(path.dentry); 22504f757f3cSAl Viro err = follow_managed(&path, nd); 22514f757f3cSAl Viro if (unlikely(err < 0)) 22524f757f3cSAl Viro return err; 22534f757f3cSAl Viro inode = d_backing_inode(path.dentry); 22544f757f3cSAl Viro seq = 0; 22554f757f3cSAl Viro } 22564f757f3cSAl Viro path_to_nameidata(&path, nd); 22574f757f3cSAl Viro nd->inode = inode; 22584f757f3cSAl Viro nd->seq = seq; 22594f757f3cSAl Viro return 0; 22604f757f3cSAl Viro } 22614f757f3cSAl Viro 22629b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2263c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 22649b4a9b14SAl Viro { 2265c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2266bd92d7feSAl Viro int err; 226731e6b01fSNick Piggin 2268368ee9baSAl Viro if (IS_ERR(s)) 2269368ee9baSAl Viro return PTR_ERR(s); 22704f757f3cSAl Viro 22714f757f3cSAl Viro if (unlikely(flags & LOOKUP_DOWN)) { 22724f757f3cSAl Viro err = handle_lookup_down(nd); 22734f757f3cSAl Viro if (unlikely(err < 0)) { 22744f757f3cSAl Viro terminate_walk(nd); 22754f757f3cSAl Viro return err; 22764f757f3cSAl Viro } 22774f757f3cSAl Viro } 22784f757f3cSAl Viro 22793bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 22803bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 22813bdba28bSAl Viro s = trailing_symlink(nd); 22823bdba28bSAl Viro if (IS_ERR(s)) { 22833bdba28bSAl Viro err = PTR_ERR(s); 22846d7b5aaeSAl Viro break; 2285bd92d7feSAl Viro } 2286bd92d7feSAl Viro } 22879f1fafeeSAl Viro if (!err) 22889f1fafeeSAl Viro err = complete_walk(nd); 2289bd92d7feSAl Viro 2290deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2291deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2292bd23a539SAl Viro err = -ENOTDIR; 2293625b6d10SAl Viro if (!err) { 2294625b6d10SAl Viro *path = nd->path; 2295625b6d10SAl Viro nd->path.mnt = NULL; 2296625b6d10SAl Viro nd->path.dentry = NULL; 2297625b6d10SAl Viro } 2298deb106c6SAl Viro terminate_walk(nd); 2299bd92d7feSAl Viro return err; 230031e6b01fSNick Piggin } 230131e6b01fSNick Piggin 2302625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 23039ad1aaa6SAl Viro struct path *path, struct path *root) 2304873f1eedSJeff Layton { 2305894bc8c4SAl Viro int retval; 23069883d185SAl Viro struct nameidata nd; 2307abc9f5beSAl Viro if (IS_ERR(name)) 2308abc9f5beSAl Viro return PTR_ERR(name); 23099ad1aaa6SAl Viro if (unlikely(root)) { 23109ad1aaa6SAl Viro nd.root = *root; 23119ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23129ad1aaa6SAl Viro } 23139883d185SAl Viro set_nameidata(&nd, dfd, name); 2314c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2315873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2316c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2317873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2318c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2319873f1eedSJeff Layton 2320873f1eedSJeff Layton if (likely(!retval)) 2321625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 23229883d185SAl Viro restore_nameidata(); 2323e4bd1c1aSAl Viro putname(name); 2324873f1eedSJeff Layton return retval; 2325873f1eedSJeff Layton } 2326873f1eedSJeff Layton 23278bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2328c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2329391172c4SAl Viro struct path *parent) 23308bcb77faSAl Viro { 2331c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2332368ee9baSAl Viro int err; 2333368ee9baSAl Viro if (IS_ERR(s)) 2334368ee9baSAl Viro return PTR_ERR(s); 2335368ee9baSAl Viro err = link_path_walk(s, nd); 23368bcb77faSAl Viro if (!err) 23378bcb77faSAl Viro err = complete_walk(nd); 2338391172c4SAl Viro if (!err) { 2339391172c4SAl Viro *parent = nd->path; 2340391172c4SAl Viro nd->path.mnt = NULL; 2341391172c4SAl Viro nd->path.dentry = NULL; 2342391172c4SAl Viro } 2343deb106c6SAl Viro terminate_walk(nd); 23448bcb77faSAl Viro return err; 23458bcb77faSAl Viro } 23468bcb77faSAl Viro 23475c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2348391172c4SAl Viro unsigned int flags, struct path *parent, 2349391172c4SAl Viro struct qstr *last, int *type) 23508bcb77faSAl Viro { 23518bcb77faSAl Viro int retval; 23529883d185SAl Viro struct nameidata nd; 23538bcb77faSAl Viro 23545c31b6ceSAl Viro if (IS_ERR(name)) 23555c31b6ceSAl Viro return name; 23569883d185SAl Viro set_nameidata(&nd, dfd, name); 2357c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23588bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2359c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23608bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2361c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2362391172c4SAl Viro if (likely(!retval)) { 2363391172c4SAl Viro *last = nd.last; 2364391172c4SAl Viro *type = nd.last_type; 2365391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 23665c31b6ceSAl Viro } else { 23675c31b6ceSAl Viro putname(name); 23685c31b6ceSAl Viro name = ERR_PTR(retval); 2369391172c4SAl Viro } 23709883d185SAl Viro restore_nameidata(); 23715c31b6ceSAl Viro return name; 23728bcb77faSAl Viro } 23738bcb77faSAl Viro 237479714f72SAl Viro /* does lookup, returns the object with parent locked */ 237579714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 23765590ff0dSUlrich Drepper { 23775c31b6ceSAl Viro struct filename *filename; 23785c31b6ceSAl Viro struct dentry *d; 2379391172c4SAl Viro struct qstr last; 2380391172c4SAl Viro int type; 238151689104SPaul Moore 23825c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 23835c31b6ceSAl Viro &last, &type); 238451689104SPaul Moore if (IS_ERR(filename)) 238551689104SPaul Moore return ERR_CAST(filename); 23865c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2387391172c4SAl Viro path_put(path); 23885c31b6ceSAl Viro putname(filename); 23895c31b6ceSAl Viro return ERR_PTR(-EINVAL); 239079714f72SAl Viro } 23915955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2392391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 239379714f72SAl Viro if (IS_ERR(d)) { 23945955102cSAl Viro inode_unlock(path->dentry->d_inode); 2395391172c4SAl Viro path_put(path); 239679714f72SAl Viro } 239751689104SPaul Moore putname(filename); 239879714f72SAl Viro return d; 23995590ff0dSUlrich Drepper } 24005590ff0dSUlrich Drepper 2401d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2402d1811465SAl Viro { 2403abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2404abc9f5beSAl Viro flags, path, NULL); 2405d1811465SAl Viro } 24064d359507SAl Viro EXPORT_SYMBOL(kern_path); 2407d1811465SAl Viro 240816f18200SJosef 'Jeff' Sipek /** 240916f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 241016f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 241116f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 241216f18200SJosef 'Jeff' Sipek * @name: pointer to file name 241316f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2414e0a01249SAl Viro * @path: pointer to struct path to fill 241516f18200SJosef 'Jeff' Sipek */ 241616f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 241716f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2418e0a01249SAl Viro struct path *path) 241916f18200SJosef 'Jeff' Sipek { 24209ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24219ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2422abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2423abc9f5beSAl Viro flags , path, &root); 242416f18200SJosef 'Jeff' Sipek } 24254d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 242616f18200SJosef 'Jeff' Sipek 2427eead1911SChristoph Hellwig /** 2428a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2429eead1911SChristoph Hellwig * @name: pathname component to lookup 2430eead1911SChristoph Hellwig * @base: base directory to lookup from 2431eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2432eead1911SChristoph Hellwig * 2433a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 24349e7543e9SAl Viro * not be called by generic code. 2435bbddca8eSNeilBrown * 2436bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2437eead1911SChristoph Hellwig */ 2438057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2439057f6c01SJames Morris { 2440057f6c01SJames Morris struct qstr this; 24416a96ba54SAl Viro unsigned int c; 2442cda309deSMiklos Szeredi int err; 2443057f6c01SJames Morris 24445955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 24452f9092e1SDavid Woodhouse 24466a96ba54SAl Viro this.name = name; 24476a96ba54SAl Viro this.len = len; 24488387ff25SLinus Torvalds this.hash = full_name_hash(base, name, len); 24496a96ba54SAl Viro if (!len) 24506a96ba54SAl Viro return ERR_PTR(-EACCES); 24516a96ba54SAl Viro 245221d8a15aSAl Viro if (unlikely(name[0] == '.')) { 245321d8a15aSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 245421d8a15aSAl Viro return ERR_PTR(-EACCES); 245521d8a15aSAl Viro } 245621d8a15aSAl Viro 24576a96ba54SAl Viro while (len--) { 24586a96ba54SAl Viro c = *(const unsigned char *)name++; 24596a96ba54SAl Viro if (c == '/' || c == '\0') 24606a96ba54SAl Viro return ERR_PTR(-EACCES); 24616a96ba54SAl Viro } 24625a202bcdSAl Viro /* 24635a202bcdSAl Viro * See if the low-level filesystem might want 24645a202bcdSAl Viro * to use its own hash.. 24655a202bcdSAl Viro */ 24665a202bcdSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 2467da53be12SLinus Torvalds int err = base->d_op->d_hash(base, &this); 24685a202bcdSAl Viro if (err < 0) 24695a202bcdSAl Viro return ERR_PTR(err); 24705a202bcdSAl Viro } 2471eead1911SChristoph Hellwig 2472cda309deSMiklos Szeredi err = inode_permission(base->d_inode, MAY_EXEC); 2473cda309deSMiklos Szeredi if (err) 2474cda309deSMiklos Szeredi return ERR_PTR(err); 2475cda309deSMiklos Szeredi 247672bd866aSAl Viro return __lookup_hash(&this, base, 0); 2477057f6c01SJames Morris } 24784d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2479057f6c01SJames Morris 2480bbddca8eSNeilBrown /** 2481bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2482bbddca8eSNeilBrown * @name: pathname component to lookup 2483bbddca8eSNeilBrown * @base: base directory to lookup from 2484bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2485bbddca8eSNeilBrown * 2486bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2487bbddca8eSNeilBrown * not be called by generic code. 2488bbddca8eSNeilBrown * 2489bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2490bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2491bbddca8eSNeilBrown */ 2492bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2493bbddca8eSNeilBrown struct dentry *base, int len) 2494bbddca8eSNeilBrown { 2495bbddca8eSNeilBrown struct qstr this; 2496bbddca8eSNeilBrown unsigned int c; 2497bbddca8eSNeilBrown int err; 249820d00ee8SLinus Torvalds struct dentry *ret; 2499bbddca8eSNeilBrown 2500bbddca8eSNeilBrown this.name = name; 2501bbddca8eSNeilBrown this.len = len; 25028387ff25SLinus Torvalds this.hash = full_name_hash(base, 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 253020d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 253120d00ee8SLinus Torvalds if (!ret) 253220d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 253320d00ee8SLinus Torvalds return ret; 2534bbddca8eSNeilBrown } 2535bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2536bbddca8eSNeilBrown 2537eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2538eedf265aSEric W. Biederman int path_pts(struct path *path) 2539eedf265aSEric W. Biederman { 2540eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2541eedf265aSEric W. Biederman * the input path. 2542eedf265aSEric W. Biederman */ 2543eedf265aSEric W. Biederman struct dentry *child, *parent; 2544eedf265aSEric W. Biederman struct qstr this; 2545eedf265aSEric W. Biederman int ret; 2546eedf265aSEric W. Biederman 2547eedf265aSEric W. Biederman ret = path_parent_directory(path); 2548eedf265aSEric W. Biederman if (ret) 2549eedf265aSEric W. Biederman return ret; 2550eedf265aSEric W. Biederman 2551eedf265aSEric W. Biederman parent = path->dentry; 2552eedf265aSEric W. Biederman this.name = "pts"; 2553eedf265aSEric W. Biederman this.len = 3; 2554eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2555eedf265aSEric W. Biederman if (!child) 2556eedf265aSEric W. Biederman return -ENOENT; 2557eedf265aSEric W. Biederman 2558eedf265aSEric W. Biederman path->dentry = child; 2559eedf265aSEric W. Biederman dput(parent); 2560eedf265aSEric W. Biederman follow_mount(path); 2561eedf265aSEric W. Biederman return 0; 2562eedf265aSEric W. Biederman } 2563eedf265aSEric W. Biederman #endif 2564eedf265aSEric W. Biederman 25651fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 25661fa1e7f6SAndy Whitcroft struct path *path, int *empty) 25671da177e4SLinus Torvalds { 2568abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2569abc9f5beSAl Viro flags, path, NULL); 25701da177e4SLinus Torvalds } 2571b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 25721fa1e7f6SAndy Whitcroft 25738033426eSJeff Layton /** 2574197df04cSAl Viro * mountpoint_last - look up last component for umount 25758033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 25768033426eSJeff Layton * 25778033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 25788033426eSJeff Layton * need to resolve the path without doing any revalidation. 25798033426eSJeff Layton * 25808033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 25818033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 25828033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 25838033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 25848033426eSJeff Layton * bogus and it doesn't exist. 25858033426eSJeff Layton * 25868033426eSJeff Layton * Returns: 25878033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2588ba8f4613SAl Viro * lookup found a negative dentry. 25898033426eSJeff Layton * 2590ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2591ba8f4613SAl Viro * symlink that needs to be followed. 25928033426eSJeff Layton * 25938033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2594ba8f4613SAl Viro * that needs to be followed. 25958033426eSJeff Layton */ 25968033426eSJeff Layton static int 2597ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 25988033426eSJeff Layton { 25998033426eSJeff Layton int error = 0; 26008033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2601ba8f4613SAl Viro struct path path; 26028033426eSJeff Layton 260335759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 260435759521SAl Viro if (nd->flags & LOOKUP_RCU) { 26054675ac39SAl Viro if (unlazy_walk(nd)) 2606deb106c6SAl Viro return -ECHILD; 26078033426eSJeff Layton } 26088033426eSJeff Layton 26098033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 26108033426eSJeff Layton 26118033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 26128033426eSJeff Layton error = handle_dots(nd, nd->last_type); 261335759521SAl Viro if (error) 2614deb106c6SAl Viro return error; 2615ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2616949a852eSAl Viro } else { 2617ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2618ba8f4613SAl Viro if (!path.dentry) { 26198033426eSJeff Layton /* 2620949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2621949a852eSAl Viro * cache, so that means that this dentry is probably 2622949a852eSAl Viro * a symlink or the path doesn't actually point 2623949a852eSAl Viro * to a mounted dentry. 26248033426eSJeff Layton */ 2625ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2626949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2627ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2628ba8f4613SAl Viro return PTR_ERR(path.dentry); 26298033426eSJeff Layton } 2630bcceeebaSDave Jones } 2631ba8f4613SAl Viro if (d_is_negative(path.dentry)) { 2632ba8f4613SAl Viro dput(path.dentry); 2633deb106c6SAl Viro return -ENOENT; 263435759521SAl Viro } 2635ba8f4613SAl Viro path.mnt = nd->path.mnt; 26368f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 26378033426eSJeff Layton } 26388033426eSJeff Layton 26398033426eSJeff Layton /** 2640197df04cSAl Viro * path_mountpoint - look up a path to be umounted 26412a78b857SMasanari Iida * @nd: lookup context 26428033426eSJeff Layton * @flags: lookup flags 2643c8a53ee5SAl Viro * @path: pointer to container for result 26448033426eSJeff Layton * 26458033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2646606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 26478033426eSJeff Layton */ 26488033426eSJeff Layton static int 2649c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 26508033426eSJeff Layton { 2651c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2652368ee9baSAl Viro int err; 2653368ee9baSAl Viro if (IS_ERR(s)) 2654368ee9baSAl Viro return PTR_ERR(s); 26553bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2656ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 26573bdba28bSAl Viro s = trailing_symlink(nd); 26583bdba28bSAl Viro if (IS_ERR(s)) { 26593bdba28bSAl Viro err = PTR_ERR(s); 26608033426eSJeff Layton break; 26618033426eSJeff Layton } 26623bdba28bSAl Viro } 2663ba8f4613SAl Viro if (!err) { 2664ba8f4613SAl Viro *path = nd->path; 2665ba8f4613SAl Viro nd->path.mnt = NULL; 2666ba8f4613SAl Viro nd->path.dentry = NULL; 2667ba8f4613SAl Viro follow_mount(path); 2668ba8f4613SAl Viro } 2669deb106c6SAl Viro terminate_walk(nd); 26708033426eSJeff Layton return err; 26718033426eSJeff Layton } 26728033426eSJeff Layton 26732d864651SAl Viro static int 2674668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 26752d864651SAl Viro unsigned int flags) 26762d864651SAl Viro { 26779883d185SAl Viro struct nameidata nd; 2678cbaab2dbSAl Viro int error; 2679668696dcSAl Viro if (IS_ERR(name)) 2680668696dcSAl Viro return PTR_ERR(name); 26819883d185SAl Viro set_nameidata(&nd, dfd, name); 2682c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 26832d864651SAl Viro if (unlikely(error == -ECHILD)) 2684c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 26852d864651SAl Viro if (unlikely(error == -ESTALE)) 2686c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 26872d864651SAl Viro if (likely(!error)) 2688668696dcSAl Viro audit_inode(name, path->dentry, 0); 26899883d185SAl Viro restore_nameidata(); 2690668696dcSAl Viro putname(name); 26912d864651SAl Viro return error; 26922d864651SAl Viro } 26932d864651SAl Viro 26948033426eSJeff Layton /** 2695197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 26968033426eSJeff Layton * @dfd: directory file descriptor 26978033426eSJeff Layton * @name: pathname from userland 26988033426eSJeff Layton * @flags: lookup flags 26998033426eSJeff Layton * @path: pointer to container to hold result 27008033426eSJeff Layton * 27018033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27028033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27038033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27048033426eSJeff Layton * and avoid revalidating the last component. 27058033426eSJeff Layton * 27068033426eSJeff Layton * Returns 0 and populates "path" on success. 27078033426eSJeff Layton */ 27088033426eSJeff Layton int 2709197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27108033426eSJeff Layton struct path *path) 27118033426eSJeff Layton { 2712cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27138033426eSJeff Layton } 27148033426eSJeff Layton 27152d864651SAl Viro int 27162d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27172d864651SAl Viro unsigned int flags) 27182d864651SAl Viro { 2719cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27202d864651SAl Viro } 27212d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27222d864651SAl Viro 2723cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27241da177e4SLinus Torvalds { 27258e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2726da9592edSDavid Howells 27278e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27281da177e4SLinus Torvalds return 0; 27298e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27301da177e4SLinus Torvalds return 0; 273123adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27321da177e4SLinus Torvalds } 2733cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27341da177e4SLinus Torvalds 27351da177e4SLinus Torvalds /* 27361da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27371da177e4SLinus Torvalds * whether the type of victim is right. 27381da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27391da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27401da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27411da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27421da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27431da177e4SLinus Torvalds * a. be owner of dir, or 27441da177e4SLinus Torvalds * b. be owner of victim, or 27451da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27461da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27471da177e4SLinus Torvalds * links pointing to it. 27480bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27490bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27500bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27510bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27520bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27531da177e4SLinus Torvalds * nfs_async_unlink(). 27541da177e4SLinus Torvalds */ 2755b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27561da177e4SLinus Torvalds { 275763afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27581da177e4SLinus Torvalds int error; 27591da177e4SLinus Torvalds 2760b18825a7SDavid Howells if (d_is_negative(victim)) 27611da177e4SLinus Torvalds return -ENOENT; 2762b18825a7SDavid Howells BUG_ON(!inode); 27631da177e4SLinus Torvalds 27641da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 27654fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27661da177e4SLinus Torvalds 2767f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27681da177e4SLinus Torvalds if (error) 27691da177e4SLinus Torvalds return error; 27701da177e4SLinus Torvalds if (IS_APPEND(dir)) 27711da177e4SLinus Torvalds return -EPERM; 2772b18825a7SDavid Howells 2773b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27740bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27751da177e4SLinus Torvalds return -EPERM; 27761da177e4SLinus Torvalds if (isdir) { 277744b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27781da177e4SLinus Torvalds return -ENOTDIR; 27791da177e4SLinus Torvalds if (IS_ROOT(victim)) 27801da177e4SLinus Torvalds return -EBUSY; 278144b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 27821da177e4SLinus Torvalds return -EISDIR; 27831da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27841da177e4SLinus Torvalds return -ENOENT; 27851da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27861da177e4SLinus Torvalds return -EBUSY; 27871da177e4SLinus Torvalds return 0; 27881da177e4SLinus Torvalds } 27891da177e4SLinus Torvalds 27901da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27911da177e4SLinus Torvalds * dir. 27921da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27931da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27941da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2795036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2796036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2797036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27981da177e4SLinus Torvalds */ 2799a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28001da177e4SLinus Torvalds { 2801036d5236SEric W. Biederman struct user_namespace *s_user_ns; 280214e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28031da177e4SLinus Torvalds if (child->d_inode) 28041da177e4SLinus Torvalds return -EEXIST; 28051da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28061da177e4SLinus Torvalds return -ENOENT; 2807036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2808036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2809036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2810036d5236SEric W. Biederman return -EOVERFLOW; 2811f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28121da177e4SLinus Torvalds } 28131da177e4SLinus Torvalds 28141da177e4SLinus Torvalds /* 28151da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28161da177e4SLinus Torvalds */ 28171da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28181da177e4SLinus Torvalds { 28191da177e4SLinus Torvalds struct dentry *p; 28201da177e4SLinus Torvalds 28211da177e4SLinus Torvalds if (p1 == p2) { 28225955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28231da177e4SLinus Torvalds return NULL; 28241da177e4SLinus Torvalds } 28251da177e4SLinus Torvalds 2826fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28271da177e4SLinus Torvalds 2828e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2829e2761a11SOGAWA Hirofumi if (p) { 28305955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28315955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28321da177e4SLinus Torvalds return p; 28331da177e4SLinus Torvalds } 28341da177e4SLinus Torvalds 2835e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2836e2761a11SOGAWA Hirofumi if (p) { 28375955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28385955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28391da177e4SLinus Torvalds return p; 28401da177e4SLinus Torvalds } 28411da177e4SLinus Torvalds 28425955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28435955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28441da177e4SLinus Torvalds return NULL; 28451da177e4SLinus Torvalds } 28464d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28471da177e4SLinus Torvalds 28481da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28491da177e4SLinus Torvalds { 28505955102cSAl Viro inode_unlock(p1->d_inode); 28511da177e4SLinus Torvalds if (p1 != p2) { 28525955102cSAl Viro inode_unlock(p2->d_inode); 2853fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28541da177e4SLinus Torvalds } 28551da177e4SLinus Torvalds } 28564d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28571da177e4SLinus Torvalds 28584acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2859312b63fbSAl Viro bool want_excl) 28601da177e4SLinus Torvalds { 2861a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28621da177e4SLinus Torvalds if (error) 28631da177e4SLinus Torvalds return error; 28641da177e4SLinus Torvalds 2865acfa4380SAl Viro if (!dir->i_op->create) 28661da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28671da177e4SLinus Torvalds mode &= S_IALLUGO; 28681da177e4SLinus Torvalds mode |= S_IFREG; 28691da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28701da177e4SLinus Torvalds if (error) 28711da177e4SLinus Torvalds return error; 2872312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2873a74574aaSStephen Smalley if (!error) 2874f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28751da177e4SLinus Torvalds return error; 28761da177e4SLinus Torvalds } 28774d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28781da177e4SLinus Torvalds 28798e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 28808e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 28818e6c848eSAl Viro void *arg) 28828e6c848eSAl Viro { 28838e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 28848e6c848eSAl Viro int error = may_create(dir, dentry); 28858e6c848eSAl Viro if (error) 28868e6c848eSAl Viro return error; 28878e6c848eSAl Viro 28888e6c848eSAl Viro mode &= S_IALLUGO; 28898e6c848eSAl Viro mode |= S_IFREG; 28908e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 28918e6c848eSAl Viro if (error) 28928e6c848eSAl Viro return error; 28938e6c848eSAl Viro error = f(dentry, mode, arg); 28948e6c848eSAl Viro if (!error) 28958e6c848eSAl Viro fsnotify_create(dir, dentry); 28968e6c848eSAl Viro return error; 28978e6c848eSAl Viro } 28988e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 28998e6c848eSAl Viro 2900a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2901a2982cc9SEric W. Biederman { 2902a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2903a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2904a2982cc9SEric W. Biederman } 2905a2982cc9SEric W. Biederman 2906f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 29071da177e4SLinus Torvalds { 29083fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29091da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29101da177e4SLinus Torvalds int error; 29111da177e4SLinus Torvalds 29121da177e4SLinus Torvalds if (!inode) 29131da177e4SLinus Torvalds return -ENOENT; 29141da177e4SLinus Torvalds 2915c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2916c8fe8f30SChristoph Hellwig case S_IFLNK: 29171da177e4SLinus Torvalds return -ELOOP; 2918c8fe8f30SChristoph Hellwig case S_IFDIR: 2919c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29201da177e4SLinus Torvalds return -EISDIR; 2921c8fe8f30SChristoph Hellwig break; 2922c8fe8f30SChristoph Hellwig case S_IFBLK: 2923c8fe8f30SChristoph Hellwig case S_IFCHR: 2924a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29251da177e4SLinus Torvalds return -EACCES; 2926c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2927c8fe8f30SChristoph Hellwig case S_IFIFO: 2928c8fe8f30SChristoph Hellwig case S_IFSOCK: 29291da177e4SLinus Torvalds flag &= ~O_TRUNC; 2930c8fe8f30SChristoph Hellwig break; 29314a3fd211SDave Hansen } 2932b41572e9SDave Hansen 293362fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2934b41572e9SDave Hansen if (error) 2935b41572e9SDave Hansen return error; 29366146f0d5SMimi Zohar 29371da177e4SLinus Torvalds /* 29381da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29391da177e4SLinus Torvalds */ 29401da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29418737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29427715b521SAl Viro return -EPERM; 29431da177e4SLinus Torvalds if (flag & O_TRUNC) 29447715b521SAl Viro return -EPERM; 29451da177e4SLinus Torvalds } 29461da177e4SLinus Torvalds 29471da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29482e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29497715b521SAl Viro return -EPERM; 29501da177e4SLinus Torvalds 2951f3c7691eSJ. Bruce Fields return 0; 29527715b521SAl Viro } 29537715b521SAl Viro 2954e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29557715b521SAl Viro { 2956f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29577715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29587715b521SAl Viro int error = get_write_access(inode); 29591da177e4SLinus Torvalds if (error) 29607715b521SAl Viro return error; 29611da177e4SLinus Torvalds /* 29621da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29631da177e4SLinus Torvalds */ 2964d7a06983SJeff Layton error = locks_verify_locked(filp); 2965be6d3e56SKentaro Takeda if (!error) 2966ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29671da177e4SLinus Torvalds if (!error) { 29687715b521SAl Viro error = do_truncate(path->dentry, 0, 2969d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2970e1181ee6SJeff Layton filp); 29711da177e4SLinus Torvalds } 29721da177e4SLinus Torvalds put_write_access(inode); 2973acd0c935SMimi Zohar return error; 29741da177e4SLinus Torvalds } 29751da177e4SLinus Torvalds 2976d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2977d57999e1SDave Hansen { 29788a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29798a5e929dSAl Viro flag--; 2980d57999e1SDave Hansen return flag; 2981d57999e1SDave Hansen } 2982d57999e1SDave Hansen 2983d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2984d18e9008SMiklos Szeredi { 29851328c727SSeth Forshee struct user_namespace *s_user_ns; 2986d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2987d18e9008SMiklos Szeredi if (error) 2988d18e9008SMiklos Szeredi return error; 2989d18e9008SMiklos Szeredi 29901328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29911328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29921328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29931328c727SSeth Forshee return -EOVERFLOW; 29941328c727SSeth Forshee 2995d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2996d18e9008SMiklos Szeredi if (error) 2997d18e9008SMiklos Szeredi return error; 2998d18e9008SMiklos Szeredi 2999d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3000d18e9008SMiklos Szeredi } 3001d18e9008SMiklos Szeredi 30021acf0af9SDavid Howells /* 30031acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30041acf0af9SDavid Howells * dentry. 30051acf0af9SDavid Howells * 30061acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30071acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30081acf0af9SDavid Howells * 30091acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 30101acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 30111acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 30121acf0af9SDavid Howells * 30131acf0af9SDavid Howells * Returns an error code otherwise. 30141acf0af9SDavid Howells */ 30152675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 301630d90494SAl Viro struct path *path, struct file *file, 3017015c3bbcSMiklos Szeredi const struct open_flags *op, 30181643b43fSAl Viro int open_flag, umode_t mode, 301947237687SAl Viro int *opened) 3020d18e9008SMiklos Szeredi { 3021d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3022d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3023d18e9008SMiklos Szeredi int error; 3024d18e9008SMiklos Szeredi 3025384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3026d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3027d18e9008SMiklos Szeredi 3028d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3029d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3030d18e9008SMiklos Szeredi 303130d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 303230d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30330fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 30340fb1ea09SAl Viro open_to_namei_flags(open_flag), 30350fb1ea09SAl Viro mode, opened); 30366fbd0714SAl Viro d_lookup_done(dentry); 3037384f26e2SAl Viro if (!error) { 3038d18e9008SMiklos Szeredi /* 3039384f26e2SAl Viro * We didn't have the inode before the open, so check open 3040384f26e2SAl Viro * permission here. 3041d18e9008SMiklos Szeredi */ 3042384f26e2SAl Viro int acc_mode = op->acc_mode; 304303da633aSAl Viro if (*opened & FILE_CREATED) { 304403da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 304503da633aSAl Viro fsnotify_create(dir, dentry); 304662fb4a15SAl Viro acc_mode = 0; 304703da633aSAl Viro } 30482675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3049384f26e2SAl Viro if (WARN_ON(error > 0)) 3050384f26e2SAl Viro error = -EINVAL; 3051384f26e2SAl Viro } else if (error > 0) { 3052d18e9008SMiklos Szeredi if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3053d18e9008SMiklos Szeredi error = -EIO; 3054384f26e2SAl Viro } else { 3055d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3056d18e9008SMiklos Szeredi dput(dentry); 3057d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 305810c64ceaSAl Viro } 3059d18e9008SMiklos Szeredi if (*opened & FILE_CREATED) 3060d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3061a01e718fSAl Viro if (unlikely(d_is_negative(dentry))) { 3062a01e718fSAl Viro error = -ENOENT; 3063a01e718fSAl Viro } else { 3064d18e9008SMiklos Szeredi path->dentry = dentry; 3065d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 30662675a4ebSAl Viro return 1; 3067d18e9008SMiklos Szeredi } 3068d18e9008SMiklos Szeredi } 3069a01e718fSAl Viro } 3070d18e9008SMiklos Szeredi dput(dentry); 3071d18e9008SMiklos Szeredi return error; 3072d18e9008SMiklos Szeredi } 3073d18e9008SMiklos Szeredi 307431e6b01fSNick Piggin /* 30751acf0af9SDavid Howells * Look up and maybe create and open the last component. 3076d58ffd35SMiklos Szeredi * 3077d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 3078d58ffd35SMiklos Szeredi * 30791acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 30801acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 30811acf0af9SDavid Howells * 30821acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 30831acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 30841acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 30851acf0af9SDavid Howells * specified then a negative dentry may be returned. 30861acf0af9SDavid Howells * 30871acf0af9SDavid Howells * An error code is returned otherwise. 30881acf0af9SDavid Howells * 30891acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 30901acf0af9SDavid Howells * cleared otherwise prior to returning. 3091d58ffd35SMiklos Szeredi */ 30922675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 309330d90494SAl Viro struct file *file, 3094d58ffd35SMiklos Szeredi const struct open_flags *op, 309564894cf8SAl Viro bool got_write, int *opened) 3096d58ffd35SMiklos Szeredi { 3097d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 309854ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30991643b43fSAl Viro int open_flag = op->open_flag; 3100d58ffd35SMiklos Szeredi struct dentry *dentry; 31011643b43fSAl Viro int error, create_error = 0; 31021643b43fSAl Viro umode_t mode = op->mode; 31036fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3104d58ffd35SMiklos Szeredi 3105ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3106ce8644fcSAl Viro return -ENOENT; 3107d58ffd35SMiklos Szeredi 310847237687SAl Viro *opened &= ~FILE_CREATED; 31096fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31106fbd0714SAl Viro for (;;) { 31116fbd0714SAl Viro if (!dentry) { 31126fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3113d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 31142675a4ebSAl Viro return PTR_ERR(dentry); 31156fbd0714SAl Viro } 31166fbd0714SAl Viro if (d_in_lookup(dentry)) 31176fbd0714SAl Viro break; 3118d58ffd35SMiklos Szeredi 31196fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31206fbd0714SAl Viro if (likely(error > 0)) 31216fbd0714SAl Viro break; 31226fbd0714SAl Viro if (error) 31236fbd0714SAl Viro goto out_dput; 31246fbd0714SAl Viro d_invalidate(dentry); 31256fbd0714SAl Viro dput(dentry); 31266fbd0714SAl Viro dentry = NULL; 31276fbd0714SAl Viro } 31286fbd0714SAl Viro if (dentry->d_inode) { 3129d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3130d18e9008SMiklos Szeredi goto out_no_open; 31316c51e513SAl Viro } 3132d18e9008SMiklos Szeredi 31331643b43fSAl Viro /* 31341643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31351643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31361643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31371643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31381643b43fSAl Viro * 31391643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31401643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31411643b43fSAl Viro */ 31421643b43fSAl Viro if (open_flag & O_CREAT) { 31431643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31441643b43fSAl Viro mode &= ~current_umask(); 31451643b43fSAl Viro if (unlikely(!got_write)) { 31461643b43fSAl Viro create_error = -EROFS; 31471643b43fSAl Viro open_flag &= ~O_CREAT; 31481643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31491643b43fSAl Viro goto no_open; 31501643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31511643b43fSAl Viro } else { 31521643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31531643b43fSAl Viro if (create_error) { 31541643b43fSAl Viro open_flag &= ~O_CREAT; 31551643b43fSAl Viro if (open_flag & O_EXCL) 31561643b43fSAl Viro goto no_open; 31571643b43fSAl Viro } 31581643b43fSAl Viro } 31591643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31601643b43fSAl Viro unlikely(!got_write)) { 31611643b43fSAl Viro /* 31621643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31631643b43fSAl Viro * back to lookup + open 31641643b43fSAl Viro */ 31651643b43fSAl Viro goto no_open; 3166d18e9008SMiklos Szeredi } 3167d18e9008SMiklos Szeredi 31681643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 31691643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 31701643b43fSAl Viro mode, opened); 31711643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 31721643b43fSAl Viro error = create_error; 31731643b43fSAl Viro return error; 31741643b43fSAl Viro } 317554ef4872SMiklos Szeredi 31761643b43fSAl Viro no_open: 31776fbd0714SAl Viro if (d_in_lookup(dentry)) { 317812fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 317912fa5e24SAl Viro nd->flags); 31806fbd0714SAl Viro d_lookup_done(dentry); 318112fa5e24SAl Viro if (unlikely(res)) { 318212fa5e24SAl Viro if (IS_ERR(res)) { 318312fa5e24SAl Viro error = PTR_ERR(res); 318412fa5e24SAl Viro goto out_dput; 318512fa5e24SAl Viro } 318612fa5e24SAl Viro dput(dentry); 318712fa5e24SAl Viro dentry = res; 318812fa5e24SAl Viro } 318954ef4872SMiklos Szeredi } 319054ef4872SMiklos Szeredi 3191d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 31921643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 319347237687SAl Viro *opened |= FILE_CREATED; 3194ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3195ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3196ce8644fcSAl Viro error = -EACCES; 3197d58ffd35SMiklos Szeredi goto out_dput; 319877d660a8SMiklos Szeredi } 3199ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 32001643b43fSAl Viro open_flag & O_EXCL); 3201d58ffd35SMiklos Szeredi if (error) 3202d58ffd35SMiklos Szeredi goto out_dput; 3203ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3204d58ffd35SMiklos Szeredi } 32051643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32061643b43fSAl Viro error = create_error; 3207d58ffd35SMiklos Szeredi goto out_dput; 3208d58ffd35SMiklos Szeredi } 3209d18e9008SMiklos Szeredi out_no_open: 3210d58ffd35SMiklos Szeredi path->dentry = dentry; 3211d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 32122675a4ebSAl Viro return 1; 3213d58ffd35SMiklos Szeredi 3214d58ffd35SMiklos Szeredi out_dput: 3215d58ffd35SMiklos Szeredi dput(dentry); 32162675a4ebSAl Viro return error; 3217d58ffd35SMiklos Szeredi } 3218d58ffd35SMiklos Szeredi 3219d58ffd35SMiklos Szeredi /* 3220fe2d35ffSAl Viro * Handle the last step of open() 322131e6b01fSNick Piggin */ 3222896475d5SAl Viro static int do_last(struct nameidata *nd, 322330d90494SAl Viro struct file *file, const struct open_flags *op, 322476ae2a5aSAl Viro int *opened) 3225fb1cc555SAl Viro { 3226a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3227ca344a89SAl Viro int open_flag = op->open_flag; 322877d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 322964894cf8SAl Viro bool got_write = false; 3230bcda7652SAl Viro int acc_mode = op->acc_mode; 3231254cf582SAl Viro unsigned seq; 3232a1eb3315SMiklos Szeredi struct inode *inode; 3233896475d5SAl Viro struct path path; 323416c2cd71SAl Viro int error; 3235fb1cc555SAl Viro 3236c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3237c3e380b0SAl Viro nd->flags |= op->intent; 3238c3e380b0SAl Viro 3239bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3240fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3241deb106c6SAl Viro if (unlikely(error)) 32422675a4ebSAl Viro return error; 3243e83db167SMiklos Szeredi goto finish_open; 32441f36f774SAl Viro } 3245a2c36b45SAl Viro 3246ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3247fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3248fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3249fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3250254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3251e9742b53SAl Viro if (likely(error > 0)) 325271574865SMiklos Szeredi goto finish_lookup; 325371574865SMiklos Szeredi 3254ce57dfc1SAl Viro if (error < 0) 3255deb106c6SAl Viro return error; 3256a1eb3315SMiklos Szeredi 325737d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32586583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3259b6183df7SMiklos Szeredi } else { 3260fe2d35ffSAl Viro /* create side of things */ 3261a3fbbde7SAl Viro /* 3262b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3263b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3264b6183df7SMiklos Szeredi * about to look up 3265a3fbbde7SAl Viro */ 32669f1fafeeSAl Viro error = complete_walk(nd); 3267e8bb73dfSAl Viro if (error) 32682675a4ebSAl Viro return error; 3269fe2d35ffSAl Viro 327076ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 32711f36f774SAl Viro /* trailing slashes? */ 3272deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3273deb106c6SAl Viro return -EISDIR; 3274b6183df7SMiklos Szeredi } 32751f36f774SAl Viro 32769cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 327764894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 327864894cf8SAl Viro if (!error) 327964894cf8SAl Viro got_write = true; 328064894cf8SAl Viro /* 328164894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 328264894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 328364894cf8SAl Viro * dropping this one anyway. 328464894cf8SAl Viro */ 328564894cf8SAl Viro } 32869cf843e3SAl Viro if (open_flag & O_CREAT) 32875955102cSAl Viro inode_lock(dir->d_inode); 32889cf843e3SAl Viro else 32899cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3290896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 32919cf843e3SAl Viro if (open_flag & O_CREAT) 32925955102cSAl Viro inode_unlock(dir->d_inode); 32939cf843e3SAl Viro else 32949cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3295fb1cc555SAl Viro 32962675a4ebSAl Viro if (error <= 0) { 32972675a4ebSAl Viro if (error) 3298d58ffd35SMiklos Szeredi goto out; 32996c0d46c4SAl Viro 330047237687SAl Viro if ((*opened & FILE_CREATED) || 3301496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 330277d660a8SMiklos Szeredi will_truncate = false; 3303d18e9008SMiklos Szeredi 330476ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3305d18e9008SMiklos Szeredi goto opened; 3306d18e9008SMiklos Szeredi } 3307d18e9008SMiklos Szeredi 330847237687SAl Viro if (*opened & FILE_CREATED) { 33099b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3310ca344a89SAl Viro open_flag &= ~O_TRUNC; 331177d660a8SMiklos Szeredi will_truncate = false; 331262fb4a15SAl Viro acc_mode = 0; 3313896475d5SAl Viro path_to_nameidata(&path, nd); 3314e83db167SMiklos Szeredi goto finish_open_created; 3315fb1cc555SAl Viro } 3316fb1cc555SAl Viro 3317fb1cc555SAl Viro /* 3318d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3319d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3320d18e9008SMiklos Szeredi * necessary...) 3321d18e9008SMiklos Szeredi */ 332264894cf8SAl Viro if (got_write) { 3323d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 332464894cf8SAl Viro got_write = false; 3325d18e9008SMiklos Szeredi } 3326d18e9008SMiklos Szeredi 3327e6ec03a2SAl Viro error = follow_managed(&path, nd); 3328e6ec03a2SAl Viro if (unlikely(error < 0)) 3329e6ec03a2SAl Viro return error; 3330e6ec03a2SAl Viro 33316583fe22SAl Viro if (unlikely(d_is_negative(path.dentry))) { 33326583fe22SAl Viro path_to_nameidata(&path, nd); 33336583fe22SAl Viro return -ENOENT; 33346583fe22SAl Viro } 33356583fe22SAl Viro 33366583fe22SAl Viro /* 33376583fe22SAl Viro * create/update audit record if it already exists. 33386583fe22SAl Viro */ 33396583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 33406583fe22SAl Viro 3341deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3342deb106c6SAl Viro path_to_nameidata(&path, nd); 3343deb106c6SAl Viro return -EEXIST; 3344deb106c6SAl Viro } 3345fb1cc555SAl Viro 3346254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3347d4565649SAl Viro inode = d_backing_inode(path.dentry); 3348766c4cbfSAl Viro finish_lookup: 33498f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3350deb106c6SAl Viro if (unlikely(error)) 3351d63ff28fSAl Viro return error; 3352bc77daa7SAl Viro finish_open: 33538f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3354a3fbbde7SAl Viro error = complete_walk(nd); 3355fac7d191SAl Viro if (error) 33562675a4ebSAl Viro return error; 335776ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3358fb1cc555SAl Viro error = -EISDIR; 335944b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 33602675a4ebSAl Viro goto out; 3361af2f5542SMiklos Szeredi error = -ENOTDIR; 336244b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 33632675a4ebSAl Viro goto out; 33644bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 336577d660a8SMiklos Szeredi will_truncate = false; 33666c0d46c4SAl Viro 33670f9d1a10SAl Viro if (will_truncate) { 33680f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33690f9d1a10SAl Viro if (error) 33702675a4ebSAl Viro goto out; 337164894cf8SAl Viro got_write = true; 33720f9d1a10SAl Viro } 3373e83db167SMiklos Szeredi finish_open_created: 3374bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3375ca344a89SAl Viro if (error) 33762675a4ebSAl Viro goto out; 33774aa7c634SMiklos Szeredi BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */ 33784aa7c634SMiklos Szeredi error = vfs_open(&nd->path, file, current_cred()); 3379fac7d191SAl Viro if (error) 3380015c3bbcSMiklos Szeredi goto out; 3381fac7d191SAl Viro *opened |= FILE_OPENED; 3382a8277b9bSMiklos Szeredi opened: 33832675a4ebSAl Viro error = open_check_o_direct(file); 3384fe9ec829SAl Viro if (!error) 33853034a146SDmitry Kasatkin error = ima_file_check(file, op->acc_mode, *opened); 3386fe9ec829SAl Viro if (!error && will_truncate) 33872675a4ebSAl Viro error = handle_truncate(file); 3388ca344a89SAl Viro out: 3389fe9ec829SAl Viro if (unlikely(error) && (*opened & FILE_OPENED)) 3390fe9ec829SAl Viro fput(file); 3391c80567c8SAl Viro if (unlikely(error > 0)) { 3392c80567c8SAl Viro WARN_ON(1); 3393c80567c8SAl Viro error = -EINVAL; 3394c80567c8SAl Viro } 339564894cf8SAl Viro if (got_write) 33960f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 33972675a4ebSAl Viro return error; 3398fb1cc555SAl Viro } 3399fb1cc555SAl Viro 3400af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3401af7bd4dcSAmir Goldstein { 3402af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3403af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3404af7bd4dcSAmir Goldstein struct inode *inode; 3405af7bd4dcSAmir Goldstein int error; 3406af7bd4dcSAmir Goldstein 3407af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3408af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3409af7bd4dcSAmir Goldstein if (error) 3410af7bd4dcSAmir Goldstein goto out_err; 3411af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3412af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3413af7bd4dcSAmir Goldstein goto out_err; 3414af7bd4dcSAmir Goldstein error = -ENOMEM; 3415cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3416af7bd4dcSAmir Goldstein if (unlikely(!child)) 3417af7bd4dcSAmir Goldstein goto out_err; 3418af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3419af7bd4dcSAmir Goldstein if (error) 3420af7bd4dcSAmir Goldstein goto out_err; 3421af7bd4dcSAmir Goldstein error = -ENOENT; 3422af7bd4dcSAmir Goldstein inode = child->d_inode; 3423af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3424af7bd4dcSAmir Goldstein goto out_err; 3425af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3426af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3427af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3428af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3429af7bd4dcSAmir Goldstein } 3430af7bd4dcSAmir Goldstein return child; 3431af7bd4dcSAmir Goldstein 3432af7bd4dcSAmir Goldstein out_err: 3433af7bd4dcSAmir Goldstein dput(child); 3434af7bd4dcSAmir Goldstein return ERR_PTR(error); 3435af7bd4dcSAmir Goldstein } 3436af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3437af7bd4dcSAmir Goldstein 3438c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 343960545d0dSAl Viro const struct open_flags *op, 344060545d0dSAl Viro struct file *file, int *opened) 344160545d0dSAl Viro { 3442625b6d10SAl Viro struct dentry *child; 3443625b6d10SAl Viro struct path path; 3444c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 344560545d0dSAl Viro if (unlikely(error)) 344660545d0dSAl Viro return error; 3447625b6d10SAl Viro error = mnt_want_write(path.mnt); 344860545d0dSAl Viro if (unlikely(error)) 344960545d0dSAl Viro goto out; 3450af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3451af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3452684e73beSHirofumi Nakagawa if (IS_ERR(child)) 345360545d0dSAl Viro goto out2; 3454625b6d10SAl Viro dput(path.dentry); 3455625b6d10SAl Viro path.dentry = child; 3456c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 345769a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 345862fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 345960545d0dSAl Viro if (error) 346060545d0dSAl Viro goto out2; 3461625b6d10SAl Viro file->f_path.mnt = path.mnt; 3462625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 346360545d0dSAl Viro if (error) 346460545d0dSAl Viro goto out2; 346560545d0dSAl Viro error = open_check_o_direct(file); 3466af7bd4dcSAmir Goldstein if (error) 346760545d0dSAl Viro fput(file); 346860545d0dSAl Viro out2: 3469625b6d10SAl Viro mnt_drop_write(path.mnt); 347060545d0dSAl Viro out: 3471625b6d10SAl Viro path_put(&path); 347260545d0dSAl Viro return error; 347360545d0dSAl Viro } 347460545d0dSAl Viro 34756ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34766ac08709SAl Viro { 34776ac08709SAl Viro struct path path; 34786ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34796ac08709SAl Viro if (!error) { 34806ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 34816ac08709SAl Viro error = vfs_open(&path, file, current_cred()); 34826ac08709SAl Viro path_put(&path); 34836ac08709SAl Viro } 34846ac08709SAl Viro return error; 34856ac08709SAl Viro } 34866ac08709SAl Viro 3487c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3488c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34891da177e4SLinus Torvalds { 3490368ee9baSAl Viro const char *s; 349130d90494SAl Viro struct file *file; 349247237687SAl Viro int opened = 0; 349313aab428SAl Viro int error; 349431e6b01fSNick Piggin 349530d90494SAl Viro file = get_empty_filp(); 34961afc99beSAl Viro if (IS_ERR(file)) 34971afc99beSAl Viro return file; 349831e6b01fSNick Piggin 349930d90494SAl Viro file->f_flags = op->open_flag; 350031e6b01fSNick Piggin 3501bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3502c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3503f15133dfSAl Viro goto out2; 350460545d0dSAl Viro } 350560545d0dSAl Viro 35066ac08709SAl Viro if (unlikely(file->f_flags & O_PATH)) { 35076ac08709SAl Viro error = do_o_path(nd, flags, file); 35086ac08709SAl Viro if (!error) 35096ac08709SAl Viro opened |= FILE_OPENED; 35106ac08709SAl Viro goto out2; 35116ac08709SAl Viro } 35126ac08709SAl Viro 3513c8a53ee5SAl Viro s = path_init(nd, flags); 3514368ee9baSAl Viro if (IS_ERR(s)) { 3515368ee9baSAl Viro put_filp(file); 3516368ee9baSAl Viro return ERR_CAST(s); 3517368ee9baSAl Viro } 35183bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 351976ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 352073d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35213bdba28bSAl Viro s = trailing_symlink(nd); 35223bdba28bSAl Viro if (IS_ERR(s)) { 35233bdba28bSAl Viro error = PTR_ERR(s); 35242675a4ebSAl Viro break; 3525806b681cSAl Viro } 35263bdba28bSAl Viro } 3527deb106c6SAl Viro terminate_walk(nd); 3528f15133dfSAl Viro out2: 35292675a4ebSAl Viro if (!(opened & FILE_OPENED)) { 35302675a4ebSAl Viro BUG_ON(!error); 353130d90494SAl Viro put_filp(file); 3532015c3bbcSMiklos Szeredi } 35332675a4ebSAl Viro if (unlikely(error)) { 35342675a4ebSAl Viro if (error == -EOPENSTALE) { 35352675a4ebSAl Viro if (flags & LOOKUP_RCU) 35362675a4ebSAl Viro error = -ECHILD; 35372675a4ebSAl Viro else 35382675a4ebSAl Viro error = -ESTALE; 35392675a4ebSAl Viro } 35402675a4ebSAl Viro file = ERR_PTR(error); 35412675a4ebSAl Viro } 35422675a4ebSAl Viro return file; 3543de459215SKirill Korotaev } 35441da177e4SLinus Torvalds 3545669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3546f9652e10SAl Viro const struct open_flags *op) 354713aab428SAl Viro { 35489883d185SAl Viro struct nameidata nd; 3549f9652e10SAl Viro int flags = op->lookup_flags; 355013aab428SAl Viro struct file *filp; 355113aab428SAl Viro 35529883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3553c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 355413aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3555c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 355613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3557c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35589883d185SAl Viro restore_nameidata(); 355913aab428SAl Viro return filp; 356013aab428SAl Viro } 356113aab428SAl Viro 356273d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3563f9652e10SAl Viro const char *name, const struct open_flags *op) 356473d049a4SAl Viro { 35659883d185SAl Viro struct nameidata nd; 356673d049a4SAl Viro struct file *file; 356751689104SPaul Moore struct filename *filename; 3568f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 356973d049a4SAl Viro 357073d049a4SAl Viro nd.root.mnt = mnt; 357173d049a4SAl Viro nd.root.dentry = dentry; 357273d049a4SAl Viro 3573b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 357473d049a4SAl Viro return ERR_PTR(-ELOOP); 357573d049a4SAl Viro 357651689104SPaul Moore filename = getname_kernel(name); 3577a1c83681SViresh Kumar if (IS_ERR(filename)) 357851689104SPaul Moore return ERR_CAST(filename); 357951689104SPaul Moore 35809883d185SAl Viro set_nameidata(&nd, -1, filename); 3581c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 358273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3583c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 358473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3585c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35869883d185SAl Viro restore_nameidata(); 358751689104SPaul Moore putname(filename); 358873d049a4SAl Viro return file; 358973d049a4SAl Viro } 359073d049a4SAl Viro 3591fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35921ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35931da177e4SLinus Torvalds { 3594c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3595391172c4SAl Viro struct qstr last; 3596391172c4SAl Viro int type; 3597c30dabfeSJan Kara int err2; 35981ac12b4bSJeff Layton int error; 35991ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36001ac12b4bSJeff Layton 36011ac12b4bSJeff Layton /* 36021ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36031ac12b4bSJeff Layton * other flags passed in are ignored! 36041ac12b4bSJeff Layton */ 36051ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36061ac12b4bSJeff Layton 36075c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36085c31b6ceSAl Viro if (IS_ERR(name)) 36095c31b6ceSAl Viro return ERR_CAST(name); 36101da177e4SLinus Torvalds 3611c663e5d8SChristoph Hellwig /* 3612c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3613c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3614c663e5d8SChristoph Hellwig */ 36155c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3616ed75e95dSAl Viro goto out; 3617c663e5d8SChristoph Hellwig 3618c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3619391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3620c663e5d8SChristoph Hellwig /* 3621c663e5d8SChristoph Hellwig * Do the final lookup. 3622c663e5d8SChristoph Hellwig */ 3623391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36245955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3625391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36261da177e4SLinus Torvalds if (IS_ERR(dentry)) 3627a8104a9fSAl Viro goto unlock; 3628c663e5d8SChristoph Hellwig 3629a8104a9fSAl Viro error = -EEXIST; 3630b18825a7SDavid Howells if (d_is_positive(dentry)) 3631a8104a9fSAl Viro goto fail; 3632b18825a7SDavid Howells 3633c663e5d8SChristoph Hellwig /* 3634c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3635c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3636c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3637c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3638c663e5d8SChristoph Hellwig */ 3639391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3640a8104a9fSAl Viro error = -ENOENT; 3641ed75e95dSAl Viro goto fail; 3642e9baf6e5SAl Viro } 3643c30dabfeSJan Kara if (unlikely(err2)) { 3644c30dabfeSJan Kara error = err2; 3645a8104a9fSAl Viro goto fail; 3646c30dabfeSJan Kara } 3647181c37b6SAl Viro putname(name); 3648e9baf6e5SAl Viro return dentry; 36491da177e4SLinus Torvalds fail: 3650a8104a9fSAl Viro dput(dentry); 3651a8104a9fSAl Viro dentry = ERR_PTR(error); 3652a8104a9fSAl Viro unlock: 36535955102cSAl Viro inode_unlock(path->dentry->d_inode); 3654c30dabfeSJan Kara if (!err2) 3655391172c4SAl Viro mnt_drop_write(path->mnt); 3656ed75e95dSAl Viro out: 3657391172c4SAl Viro path_put(path); 3658181c37b6SAl Viro putname(name); 3659ed75e95dSAl Viro return dentry; 3660dae6ad8fSAl Viro } 3661fa14a0b8SAl Viro 3662fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3663fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3664fa14a0b8SAl Viro { 3665181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3666181c37b6SAl Viro path, lookup_flags); 3667fa14a0b8SAl Viro } 3668dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3669dae6ad8fSAl Viro 3670921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3671921a1650SAl Viro { 3672921a1650SAl Viro dput(dentry); 36735955102cSAl Viro inode_unlock(path->dentry->d_inode); 3674a8104a9fSAl Viro mnt_drop_write(path->mnt); 3675921a1650SAl Viro path_put(path); 3676921a1650SAl Viro } 3677921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3678921a1650SAl Viro 3679520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36801ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3681dae6ad8fSAl Viro { 3682181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3683dae6ad8fSAl Viro } 3684dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3685dae6ad8fSAl Viro 36861a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 36871da177e4SLinus Torvalds { 3688a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36891da177e4SLinus Torvalds 36901da177e4SLinus Torvalds if (error) 36911da177e4SLinus Torvalds return error; 36921da177e4SLinus Torvalds 3693975d6b39SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 36941da177e4SLinus Torvalds return -EPERM; 36951da177e4SLinus Torvalds 3696acfa4380SAl Viro if (!dir->i_op->mknod) 36971da177e4SLinus Torvalds return -EPERM; 36981da177e4SLinus Torvalds 369908ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 370008ce5f16SSerge E. Hallyn if (error) 370108ce5f16SSerge E. Hallyn return error; 370208ce5f16SSerge E. Hallyn 37031da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37041da177e4SLinus Torvalds if (error) 37051da177e4SLinus Torvalds return error; 37061da177e4SLinus Torvalds 37071da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3708a74574aaSStephen Smalley if (!error) 3709f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37101da177e4SLinus Torvalds return error; 37111da177e4SLinus Torvalds } 37124d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37131da177e4SLinus Torvalds 3714f69aac00SAl Viro static int may_mknod(umode_t mode) 3715463c3197SDave Hansen { 3716463c3197SDave Hansen switch (mode & S_IFMT) { 3717463c3197SDave Hansen case S_IFREG: 3718463c3197SDave Hansen case S_IFCHR: 3719463c3197SDave Hansen case S_IFBLK: 3720463c3197SDave Hansen case S_IFIFO: 3721463c3197SDave Hansen case S_IFSOCK: 3722463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3723463c3197SDave Hansen return 0; 3724463c3197SDave Hansen case S_IFDIR: 3725463c3197SDave Hansen return -EPERM; 3726463c3197SDave Hansen default: 3727463c3197SDave Hansen return -EINVAL; 3728463c3197SDave Hansen } 3729463c3197SDave Hansen } 3730463c3197SDave Hansen 37318208a22bSAl Viro SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 37322e4d0924SHeiko Carstens unsigned, dev) 37331da177e4SLinus Torvalds { 37341da177e4SLinus Torvalds struct dentry *dentry; 3735dae6ad8fSAl Viro struct path path; 3736dae6ad8fSAl Viro int error; 3737972567f1SJeff Layton unsigned int lookup_flags = 0; 37381da177e4SLinus Torvalds 37398e4bfca1SAl Viro error = may_mknod(mode); 37408e4bfca1SAl Viro if (error) 37418e4bfca1SAl Viro return error; 3742972567f1SJeff Layton retry: 3743972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3744dae6ad8fSAl Viro if (IS_ERR(dentry)) 3745dae6ad8fSAl Viro return PTR_ERR(dentry); 37462ad94ae6SAl Viro 3747dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3748ce3b0f8dSAl Viro mode &= ~current_umask(); 3749dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3750be6d3e56SKentaro Takeda if (error) 3751a8104a9fSAl Viro goto out; 37521da177e4SLinus Torvalds switch (mode & S_IFMT) { 37531da177e4SLinus Torvalds case 0: case S_IFREG: 3754312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 375505d1a717SMimi Zohar if (!error) 375605d1a717SMimi Zohar ima_post_path_mknod(dentry); 37571da177e4SLinus Torvalds break; 37581da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3759dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37601da177e4SLinus Torvalds new_decode_dev(dev)); 37611da177e4SLinus Torvalds break; 37621da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3763dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37641da177e4SLinus Torvalds break; 37651da177e4SLinus Torvalds } 3766a8104a9fSAl Viro out: 3767921a1650SAl Viro done_path_create(&path, dentry); 3768972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3769972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3770972567f1SJeff Layton goto retry; 3771972567f1SJeff Layton } 37721da177e4SLinus Torvalds return error; 37731da177e4SLinus Torvalds } 37741da177e4SLinus Torvalds 37758208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37765590ff0dSUlrich Drepper { 37775590ff0dSUlrich Drepper return sys_mknodat(AT_FDCWD, filename, mode, dev); 37785590ff0dSUlrich Drepper } 37795590ff0dSUlrich Drepper 378018bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37811da177e4SLinus Torvalds { 3782a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37838de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37841da177e4SLinus Torvalds 37851da177e4SLinus Torvalds if (error) 37861da177e4SLinus Torvalds return error; 37871da177e4SLinus Torvalds 3788acfa4380SAl Viro if (!dir->i_op->mkdir) 37891da177e4SLinus Torvalds return -EPERM; 37901da177e4SLinus Torvalds 37911da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 37921da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 37931da177e4SLinus Torvalds if (error) 37941da177e4SLinus Torvalds return error; 37951da177e4SLinus Torvalds 37968de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 37978de52778SAl Viro return -EMLINK; 37988de52778SAl Viro 37991da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3800a74574aaSStephen Smalley if (!error) 3801f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38021da177e4SLinus Torvalds return error; 38031da177e4SLinus Torvalds } 38044d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38051da177e4SLinus Torvalds 38060101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38071da177e4SLinus Torvalds { 38086902d925SDave Hansen struct dentry *dentry; 3809dae6ad8fSAl Viro struct path path; 3810dae6ad8fSAl Viro int error; 3811b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38121da177e4SLinus Torvalds 3813b76d8b82SJeff Layton retry: 3814b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38156902d925SDave Hansen if (IS_ERR(dentry)) 3816dae6ad8fSAl Viro return PTR_ERR(dentry); 38176902d925SDave Hansen 3818dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3819ce3b0f8dSAl Viro mode &= ~current_umask(); 3820dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3821a8104a9fSAl Viro if (!error) 3822dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3823921a1650SAl Viro done_path_create(&path, dentry); 3824b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3825b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3826b76d8b82SJeff Layton goto retry; 3827b76d8b82SJeff Layton } 38281da177e4SLinus Torvalds return error; 38291da177e4SLinus Torvalds } 38301da177e4SLinus Torvalds 38310101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38320101db7aSDominik Brodowski { 38330101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38340101db7aSDominik Brodowski } 38350101db7aSDominik Brodowski 3836a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38375590ff0dSUlrich Drepper { 38380101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38395590ff0dSUlrich Drepper } 38405590ff0dSUlrich Drepper 38411da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38421da177e4SLinus Torvalds { 38431da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38441da177e4SLinus Torvalds 38451da177e4SLinus Torvalds if (error) 38461da177e4SLinus Torvalds return error; 38471da177e4SLinus Torvalds 3848acfa4380SAl Viro if (!dir->i_op->rmdir) 38491da177e4SLinus Torvalds return -EPERM; 38501da177e4SLinus Torvalds 38511d2ef590SAl Viro dget(dentry); 38525955102cSAl Viro inode_lock(dentry->d_inode); 3853912dbc15SSage Weil 38541da177e4SLinus Torvalds error = -EBUSY; 38557af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3856912dbc15SSage Weil goto out; 3857912dbc15SSage Weil 38581da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3859912dbc15SSage Weil if (error) 3860912dbc15SSage Weil goto out; 3861912dbc15SSage Weil 38623cebde24SSage Weil shrink_dcache_parent(dentry); 38631da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3864912dbc15SSage Weil if (error) 3865912dbc15SSage Weil goto out; 3866912dbc15SSage Weil 38671da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3868d83c49f3SAl Viro dont_mount(dentry); 38698ed936b5SEric W. Biederman detach_mounts(dentry); 38701da177e4SLinus Torvalds 3871912dbc15SSage Weil out: 38725955102cSAl Viro inode_unlock(dentry->d_inode); 38731d2ef590SAl Viro dput(dentry); 3874912dbc15SSage Weil if (!error) 3875912dbc15SSage Weil d_delete(dentry); 38761da177e4SLinus Torvalds return error; 38771da177e4SLinus Torvalds } 38784d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38791da177e4SLinus Torvalds 3880f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 38811da177e4SLinus Torvalds { 38821da177e4SLinus Torvalds int error = 0; 388391a27b2aSJeff Layton struct filename *name; 38841da177e4SLinus Torvalds struct dentry *dentry; 3885f5beed75SAl Viro struct path path; 3886f5beed75SAl Viro struct qstr last; 3887f5beed75SAl Viro int type; 3888c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3889c6ee9206SJeff Layton retry: 3890c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3891c1d4dd27SAl Viro &path, &last, &type); 389291a27b2aSJeff Layton if (IS_ERR(name)) 389391a27b2aSJeff Layton return PTR_ERR(name); 38941da177e4SLinus Torvalds 3895f5beed75SAl Viro switch (type) { 38961da177e4SLinus Torvalds case LAST_DOTDOT: 38971da177e4SLinus Torvalds error = -ENOTEMPTY; 38981da177e4SLinus Torvalds goto exit1; 38991da177e4SLinus Torvalds case LAST_DOT: 39001da177e4SLinus Torvalds error = -EINVAL; 39011da177e4SLinus Torvalds goto exit1; 39021da177e4SLinus Torvalds case LAST_ROOT: 39031da177e4SLinus Torvalds error = -EBUSY; 39041da177e4SLinus Torvalds goto exit1; 39051da177e4SLinus Torvalds } 39060612d9fbSOGAWA Hirofumi 3907f5beed75SAl Viro error = mnt_want_write(path.mnt); 3908c30dabfeSJan Kara if (error) 3909c30dabfeSJan Kara goto exit1; 39100612d9fbSOGAWA Hirofumi 39115955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3912f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39131da177e4SLinus Torvalds error = PTR_ERR(dentry); 39146902d925SDave Hansen if (IS_ERR(dentry)) 39156902d925SDave Hansen goto exit2; 3916e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3917e6bc45d6STheodore Ts'o error = -ENOENT; 3918e6bc45d6STheodore Ts'o goto exit3; 3919e6bc45d6STheodore Ts'o } 3920f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3921be6d3e56SKentaro Takeda if (error) 3922c30dabfeSJan Kara goto exit3; 3923f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39240622753bSDave Hansen exit3: 39251da177e4SLinus Torvalds dput(dentry); 39266902d925SDave Hansen exit2: 39275955102cSAl Viro inode_unlock(path.dentry->d_inode); 3928f5beed75SAl Viro mnt_drop_write(path.mnt); 39291da177e4SLinus Torvalds exit1: 3930f5beed75SAl Viro path_put(&path); 39311da177e4SLinus Torvalds putname(name); 3932c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3933c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3934c6ee9206SJeff Layton goto retry; 3935c6ee9206SJeff Layton } 39361da177e4SLinus Torvalds return error; 39371da177e4SLinus Torvalds } 39381da177e4SLinus Torvalds 39393cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39405590ff0dSUlrich Drepper { 39415590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39425590ff0dSUlrich Drepper } 39435590ff0dSUlrich Drepper 3944b21996e3SJ. Bruce Fields /** 3945b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3946b21996e3SJ. Bruce Fields * @dir: parent directory 3947b21996e3SJ. Bruce Fields * @dentry: victim 3948b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3949b21996e3SJ. Bruce Fields * 3950b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3951b21996e3SJ. Bruce Fields * 3952b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3953b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3954b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3955b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3956b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3957b21996e3SJ. Bruce Fields * 3958b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3959b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3960b21996e3SJ. Bruce Fields * to be NFS exported. 3961b21996e3SJ. Bruce Fields */ 3962b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39631da177e4SLinus Torvalds { 39649accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39651da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39661da177e4SLinus Torvalds 39671da177e4SLinus Torvalds if (error) 39681da177e4SLinus Torvalds return error; 39691da177e4SLinus Torvalds 3970acfa4380SAl Viro if (!dir->i_op->unlink) 39711da177e4SLinus Torvalds return -EPERM; 39721da177e4SLinus Torvalds 39735955102cSAl Viro inode_lock(target); 39748ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39751da177e4SLinus Torvalds error = -EBUSY; 39761da177e4SLinus Torvalds else { 39771da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3978bec1052eSAl Viro if (!error) { 39795a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39805a14696cSJ. Bruce Fields if (error) 3981b21996e3SJ. Bruce Fields goto out; 39821da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39838ed936b5SEric W. Biederman if (!error) { 3984d83c49f3SAl Viro dont_mount(dentry); 39858ed936b5SEric W. Biederman detach_mounts(dentry); 39868ed936b5SEric W. Biederman } 3987bec1052eSAl Viro } 39881da177e4SLinus Torvalds } 3989b21996e3SJ. Bruce Fields out: 39905955102cSAl Viro inode_unlock(target); 39911da177e4SLinus Torvalds 39921da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 39931da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 39949accbb97SJ. Bruce Fields fsnotify_link_count(target); 39951da177e4SLinus Torvalds d_delete(dentry); 39961da177e4SLinus Torvalds } 39970eeca283SRobert Love 39981da177e4SLinus Torvalds return error; 39991da177e4SLinus Torvalds } 40004d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40011da177e4SLinus Torvalds 40021da177e4SLinus Torvalds /* 40031da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40041b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40051da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40061da177e4SLinus Torvalds * while waiting on the I/O. 40071da177e4SLinus Torvalds */ 4008da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40091da177e4SLinus Torvalds { 40102ad94ae6SAl Viro int error; 40111da177e4SLinus Torvalds struct dentry *dentry; 4012f5beed75SAl Viro struct path path; 4013f5beed75SAl Viro struct qstr last; 4014f5beed75SAl Viro int type; 40151da177e4SLinus Torvalds struct inode *inode = NULL; 4016b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40175d18f813SJeff Layton unsigned int lookup_flags = 0; 40185d18f813SJeff Layton retry: 4019da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 402091a27b2aSJeff Layton if (IS_ERR(name)) 402191a27b2aSJeff Layton return PTR_ERR(name); 40222ad94ae6SAl Viro 40231da177e4SLinus Torvalds error = -EISDIR; 4024f5beed75SAl Viro if (type != LAST_NORM) 40251da177e4SLinus Torvalds goto exit1; 40260612d9fbSOGAWA Hirofumi 4027f5beed75SAl Viro error = mnt_want_write(path.mnt); 4028c30dabfeSJan Kara if (error) 4029c30dabfeSJan Kara goto exit1; 4030b21996e3SJ. Bruce Fields retry_deleg: 40315955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4032f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40331da177e4SLinus Torvalds error = PTR_ERR(dentry); 40341da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40351da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4036f5beed75SAl Viro if (last.name[last.len]) 403750338b88STörök Edwin goto slashes; 40381da177e4SLinus Torvalds inode = dentry->d_inode; 4039b18825a7SDavid Howells if (d_is_negative(dentry)) 4040e6bc45d6STheodore Ts'o goto slashes; 40417de9c6eeSAl Viro ihold(inode); 4042f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4043be6d3e56SKentaro Takeda if (error) 4044c30dabfeSJan Kara goto exit2; 4045f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40461da177e4SLinus Torvalds exit2: 40471da177e4SLinus Torvalds dput(dentry); 40481da177e4SLinus Torvalds } 40495955102cSAl Viro inode_unlock(path.dentry->d_inode); 40501da177e4SLinus Torvalds if (inode) 40511da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4052b21996e3SJ. Bruce Fields inode = NULL; 4053b21996e3SJ. Bruce Fields if (delegated_inode) { 40545a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4055b21996e3SJ. Bruce Fields if (!error) 4056b21996e3SJ. Bruce Fields goto retry_deleg; 4057b21996e3SJ. Bruce Fields } 4058f5beed75SAl Viro mnt_drop_write(path.mnt); 40591da177e4SLinus Torvalds exit1: 4060f5beed75SAl Viro path_put(&path); 40615d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40625d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40635d18f813SJeff Layton inode = NULL; 40645d18f813SJeff Layton goto retry; 40655d18f813SJeff Layton } 4066da2f1362SChristoph Hellwig putname(name); 40671da177e4SLinus Torvalds return error; 40681da177e4SLinus Torvalds 40691da177e4SLinus Torvalds slashes: 4070b18825a7SDavid Howells if (d_is_negative(dentry)) 4071b18825a7SDavid Howells error = -ENOENT; 407244b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4073b18825a7SDavid Howells error = -EISDIR; 4074b18825a7SDavid Howells else 4075b18825a7SDavid Howells error = -ENOTDIR; 40761da177e4SLinus Torvalds goto exit2; 40771da177e4SLinus Torvalds } 40781da177e4SLinus Torvalds 40792e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40805590ff0dSUlrich Drepper { 40815590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40825590ff0dSUlrich Drepper return -EINVAL; 40835590ff0dSUlrich Drepper 40845590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 40855590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 40865590ff0dSUlrich Drepper 4087da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 40885590ff0dSUlrich Drepper } 40895590ff0dSUlrich Drepper 40903480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 40915590ff0dSUlrich Drepper { 4092da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 40935590ff0dSUlrich Drepper } 40945590ff0dSUlrich Drepper 4095db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 40961da177e4SLinus Torvalds { 4097a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 40981da177e4SLinus Torvalds 40991da177e4SLinus Torvalds if (error) 41001da177e4SLinus Torvalds return error; 41011da177e4SLinus Torvalds 4102acfa4380SAl Viro if (!dir->i_op->symlink) 41031da177e4SLinus Torvalds return -EPERM; 41041da177e4SLinus Torvalds 41051da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41061da177e4SLinus Torvalds if (error) 41071da177e4SLinus Torvalds return error; 41081da177e4SLinus Torvalds 41091da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4110a74574aaSStephen Smalley if (!error) 4111f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41121da177e4SLinus Torvalds return error; 41131da177e4SLinus Torvalds } 41144d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41151da177e4SLinus Torvalds 41162e4d0924SHeiko Carstens SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 41172e4d0924SHeiko Carstens int, newdfd, const char __user *, newname) 41181da177e4SLinus Torvalds { 41192ad94ae6SAl Viro int error; 412091a27b2aSJeff Layton struct filename *from; 41216902d925SDave Hansen struct dentry *dentry; 4122dae6ad8fSAl Viro struct path path; 4123f46d3567SJeff Layton unsigned int lookup_flags = 0; 41241da177e4SLinus Torvalds 41251da177e4SLinus Torvalds from = getname(oldname); 41261da177e4SLinus Torvalds if (IS_ERR(from)) 41271da177e4SLinus Torvalds return PTR_ERR(from); 4128f46d3567SJeff Layton retry: 4129f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41301da177e4SLinus Torvalds error = PTR_ERR(dentry); 41316902d925SDave Hansen if (IS_ERR(dentry)) 4132dae6ad8fSAl Viro goto out_putname; 41336902d925SDave Hansen 413491a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4135a8104a9fSAl Viro if (!error) 413691a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4137921a1650SAl Viro done_path_create(&path, dentry); 4138f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4139f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4140f46d3567SJeff Layton goto retry; 4141f46d3567SJeff Layton } 41426902d925SDave Hansen out_putname: 41431da177e4SLinus Torvalds putname(from); 41441da177e4SLinus Torvalds return error; 41451da177e4SLinus Torvalds } 41461da177e4SLinus Torvalds 41473480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41485590ff0dSUlrich Drepper { 41495590ff0dSUlrich Drepper return sys_symlinkat(oldname, AT_FDCWD, newname); 41505590ff0dSUlrich Drepper } 41515590ff0dSUlrich Drepper 4152146a8595SJ. Bruce Fields /** 4153146a8595SJ. Bruce Fields * vfs_link - create a new link 4154146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4155146a8595SJ. Bruce Fields * @dir: new parent 4156146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4157146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4158146a8595SJ. Bruce Fields * 4159146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4160146a8595SJ. Bruce Fields * 4161146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4162146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4163146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4164146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4165146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4166146a8595SJ. Bruce Fields * 4167146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4168146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4169146a8595SJ. Bruce Fields * to be NFS exported. 4170146a8595SJ. Bruce Fields */ 4171146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41721da177e4SLinus Torvalds { 41731da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41748de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41751da177e4SLinus Torvalds int error; 41761da177e4SLinus Torvalds 41771da177e4SLinus Torvalds if (!inode) 41781da177e4SLinus Torvalds return -ENOENT; 41791da177e4SLinus Torvalds 4180a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 41811da177e4SLinus Torvalds if (error) 41821da177e4SLinus Torvalds return error; 41831da177e4SLinus Torvalds 41841da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 41851da177e4SLinus Torvalds return -EXDEV; 41861da177e4SLinus Torvalds 41871da177e4SLinus Torvalds /* 41881da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 41891da177e4SLinus Torvalds */ 41901da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 41911da177e4SLinus Torvalds return -EPERM; 41920bd23d09SEric W. Biederman /* 41930bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 41940bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 41950bd23d09SEric W. Biederman * the vfs. 41960bd23d09SEric W. Biederman */ 41970bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 41980bd23d09SEric W. Biederman return -EPERM; 4199acfa4380SAl Viro if (!dir->i_op->link) 42001da177e4SLinus Torvalds return -EPERM; 42017e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42021da177e4SLinus Torvalds return -EPERM; 42031da177e4SLinus Torvalds 42041da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42051da177e4SLinus Torvalds if (error) 42061da177e4SLinus Torvalds return error; 42071da177e4SLinus Torvalds 42085955102cSAl Viro inode_lock(inode); 4209aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4210f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4211aae8a97dSAneesh Kumar K.V error = -ENOENT; 42128de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42138de52778SAl Viro error = -EMLINK; 4214146a8595SJ. Bruce Fields else { 4215146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4216146a8595SJ. Bruce Fields if (!error) 42171da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4218146a8595SJ. Bruce Fields } 4219f4e0c30cSAl Viro 4220f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4221f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4222f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4223f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4224f4e0c30cSAl Viro } 42255955102cSAl Viro inode_unlock(inode); 4226e31e14ecSStephen Smalley if (!error) 42277e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42281da177e4SLinus Torvalds return error; 42291da177e4SLinus Torvalds } 42304d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42311da177e4SLinus Torvalds 42321da177e4SLinus Torvalds /* 42331da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42341da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42351da177e4SLinus Torvalds * newname. --KAB 42361da177e4SLinus Torvalds * 42371da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42381da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42391da177e4SLinus Torvalds * and other special files. --ADM 42401da177e4SLinus Torvalds */ 42412e4d0924SHeiko Carstens SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 42422e4d0924SHeiko Carstens int, newdfd, const char __user *, newname, int, flags) 42431da177e4SLinus Torvalds { 42441da177e4SLinus Torvalds struct dentry *new_dentry; 4245dae6ad8fSAl Viro struct path old_path, new_path; 4246146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 424711a7b371SAneesh Kumar K.V int how = 0; 42481da177e4SLinus Torvalds int error; 42491da177e4SLinus Torvalds 425011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4251c04030e1SUlrich Drepper return -EINVAL; 425211a7b371SAneesh Kumar K.V /* 4253f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4254f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4255f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 425611a7b371SAneesh Kumar K.V */ 4257f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4258f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4259f0cc6ffbSLinus Torvalds return -ENOENT; 426011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4261f0cc6ffbSLinus Torvalds } 4262c04030e1SUlrich Drepper 426311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 426411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4265442e31caSJeff Layton retry: 426611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42671da177e4SLinus Torvalds if (error) 42682ad94ae6SAl Viro return error; 42692ad94ae6SAl Viro 4270442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4271442e31caSJeff Layton (how & LOOKUP_REVAL)); 42721da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42736902d925SDave Hansen if (IS_ERR(new_dentry)) 4274dae6ad8fSAl Viro goto out; 4275dae6ad8fSAl Viro 4276dae6ad8fSAl Viro error = -EXDEV; 4277dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4278dae6ad8fSAl Viro goto out_dput; 4279800179c9SKees Cook error = may_linkat(&old_path); 4280800179c9SKees Cook if (unlikely(error)) 4281800179c9SKees Cook goto out_dput; 4282dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4283be6d3e56SKentaro Takeda if (error) 4284a8104a9fSAl Viro goto out_dput; 4285146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 428675c3f29dSDave Hansen out_dput: 4287921a1650SAl Viro done_path_create(&new_path, new_dentry); 4288146a8595SJ. Bruce Fields if (delegated_inode) { 4289146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4290d22e6338SOleg Drokin if (!error) { 4291d22e6338SOleg Drokin path_put(&old_path); 4292146a8595SJ. Bruce Fields goto retry; 4293146a8595SJ. Bruce Fields } 4294d22e6338SOleg Drokin } 4295442e31caSJeff Layton if (retry_estale(error, how)) { 4296d22e6338SOleg Drokin path_put(&old_path); 4297442e31caSJeff Layton how |= LOOKUP_REVAL; 4298442e31caSJeff Layton goto retry; 4299442e31caSJeff Layton } 43001da177e4SLinus Torvalds out: 43012d8f3038SAl Viro path_put(&old_path); 43021da177e4SLinus Torvalds 43031da177e4SLinus Torvalds return error; 43041da177e4SLinus Torvalds } 43051da177e4SLinus Torvalds 43063480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43075590ff0dSUlrich Drepper { 4308c04030e1SUlrich Drepper return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43095590ff0dSUlrich Drepper } 43105590ff0dSUlrich Drepper 4311bc27027aSMiklos Szeredi /** 4312bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4313bc27027aSMiklos Szeredi * @old_dir: parent of source 4314bc27027aSMiklos Szeredi * @old_dentry: source 4315bc27027aSMiklos Szeredi * @new_dir: parent of destination 4316bc27027aSMiklos Szeredi * @new_dentry: destination 4317bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4318520c8b16SMiklos Szeredi * @flags: rename flags 4319bc27027aSMiklos Szeredi * 4320bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4321bc27027aSMiklos Szeredi * 4322bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4323bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4324bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4325bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4326bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4327bc27027aSMiklos Szeredi * so. 4328bc27027aSMiklos Szeredi * 4329bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4330bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4331bc27027aSMiklos Szeredi * to be NFS exported. 4332bc27027aSMiklos Szeredi * 43331da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43341da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43351da177e4SLinus Torvalds * Problems: 43360117d427SMauro Carvalho Chehab * 4337d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43381da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43391da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4340a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43411da177e4SLinus Torvalds * story. 43426cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43436cedba89SJ. Bruce Fields * and source (if it is not a directory). 43441b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43451da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43461da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4347a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43481da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43491da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43501da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4351a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43521da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43531da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 43541da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4355e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 43561b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 43571da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4358c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 43591da177e4SLinus Torvalds * locking]. 43601da177e4SLinus Torvalds */ 43611da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 43628e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4363520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 43641da177e4SLinus Torvalds { 43651da177e4SLinus Torvalds int error; 4366bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4367bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4368bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4369da1ce067SMiklos Szeredi bool new_is_dir = false; 4370da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 437149d31c2fSAl Viro struct name_snapshot old_name; 43721da177e4SLinus Torvalds 43738d3e2936SMiklos Szeredi if (source == target) 43741da177e4SLinus Torvalds return 0; 43751da177e4SLinus Torvalds 43761da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 43771da177e4SLinus Torvalds if (error) 43781da177e4SLinus Torvalds return error; 43791da177e4SLinus Torvalds 4380da1ce067SMiklos Szeredi if (!target) { 4381a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4382da1ce067SMiklos Szeredi } else { 4383da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4384da1ce067SMiklos Szeredi 4385da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43861da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4387da1ce067SMiklos Szeredi else 4388da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4389da1ce067SMiklos Szeredi } 43901da177e4SLinus Torvalds if (error) 43911da177e4SLinus Torvalds return error; 43921da177e4SLinus Torvalds 43932773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 43941da177e4SLinus Torvalds return -EPERM; 43951da177e4SLinus Torvalds 4396bc27027aSMiklos Szeredi /* 4397bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4398bc27027aSMiklos Szeredi * we'll need to flip '..'. 4399bc27027aSMiklos Szeredi */ 4400da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4401da1ce067SMiklos Szeredi if (is_dir) { 4402bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4403bc27027aSMiklos Szeredi if (error) 4404bc27027aSMiklos Szeredi return error; 4405bc27027aSMiklos Szeredi } 4406da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4407da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4408da1ce067SMiklos Szeredi if (error) 4409da1ce067SMiklos Szeredi return error; 4410da1ce067SMiklos Szeredi } 4411da1ce067SMiklos Szeredi } 44120eeca283SRobert Love 44130b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44140b3974ebSMiklos Szeredi flags); 4415bc27027aSMiklos Szeredi if (error) 4416bc27027aSMiklos Szeredi return error; 4417bc27027aSMiklos Szeredi 441849d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4419bc27027aSMiklos Szeredi dget(new_dentry); 4420da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4421bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4422bc27027aSMiklos Szeredi else if (target) 44235955102cSAl Viro inode_lock(target); 4424bc27027aSMiklos Szeredi 4425bc27027aSMiklos Szeredi error = -EBUSY; 44267af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4427bc27027aSMiklos Szeredi goto out; 4428bc27027aSMiklos Szeredi 4429da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4430bc27027aSMiklos Szeredi error = -EMLINK; 4431da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4432bc27027aSMiklos Szeredi goto out; 4433da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4434da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4435da1ce067SMiklos Szeredi goto out; 4436da1ce067SMiklos Szeredi } 4437da1ce067SMiklos Szeredi if (is_dir && !(flags & RENAME_EXCHANGE) && target) 4438bc27027aSMiklos Szeredi shrink_dcache_parent(new_dentry); 4439da1ce067SMiklos Szeredi if (!is_dir) { 4440bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4441bc27027aSMiklos Szeredi if (error) 4442bc27027aSMiklos Szeredi goto out; 4443da1ce067SMiklos Szeredi } 4444da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4445bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4446bc27027aSMiklos Szeredi if (error) 4447bc27027aSMiklos Szeredi goto out; 4448bc27027aSMiklos Szeredi } 4449520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4450520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4451bc27027aSMiklos Szeredi if (error) 4452bc27027aSMiklos Szeredi goto out; 4453bc27027aSMiklos Szeredi 4454da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44551da177e4SLinus Torvalds if (is_dir) 4456bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 4457bc27027aSMiklos Szeredi dont_mount(new_dentry); 44588ed936b5SEric W. Biederman detach_mounts(new_dentry); 4459bc27027aSMiklos Szeredi } 4460da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4461da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4462bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4463da1ce067SMiklos Szeredi else 4464da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4465da1ce067SMiklos Szeredi } 4466bc27027aSMiklos Szeredi out: 4467da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4468bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4469bc27027aSMiklos Szeredi else if (target) 44705955102cSAl Viro inode_unlock(target); 4471bc27027aSMiklos Szeredi dput(new_dentry); 4472da1ce067SMiklos Szeredi if (!error) { 447349d31c2fSAl Viro fsnotify_move(old_dir, new_dir, old_name.name, is_dir, 4474da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4475da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4476da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4477da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4478da1ce067SMiklos Szeredi } 4479da1ce067SMiklos Szeredi } 448049d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 44810eeca283SRobert Love 44821da177e4SLinus Torvalds return error; 44831da177e4SLinus Torvalds } 44844d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 44851da177e4SLinus Torvalds 4486ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4487ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 44881da177e4SLinus Torvalds { 44891da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 44901da177e4SLinus Torvalds struct dentry *trap; 4491f5beed75SAl Viro struct path old_path, new_path; 4492f5beed75SAl Viro struct qstr old_last, new_last; 4493f5beed75SAl Viro int old_type, new_type; 44948e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 449591a27b2aSJeff Layton struct filename *from; 449691a27b2aSJeff Layton struct filename *to; 4497f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4498c6a94284SJeff Layton bool should_retry = false; 44992ad94ae6SAl Viro int error; 4500520c8b16SMiklos Szeredi 45010d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4502da1ce067SMiklos Szeredi return -EINVAL; 4503da1ce067SMiklos Szeredi 45040d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45050d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4506520c8b16SMiklos Szeredi return -EINVAL; 4507520c8b16SMiklos Szeredi 45080d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45090d7a8555SMiklos Szeredi return -EPERM; 45100d7a8555SMiklos Szeredi 4511f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4512f5beed75SAl Viro target_flags = 0; 4513f5beed75SAl Viro 4514c6a94284SJeff Layton retry: 4515c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4516c1d4dd27SAl Viro &old_path, &old_last, &old_type); 451791a27b2aSJeff Layton if (IS_ERR(from)) { 451891a27b2aSJeff Layton error = PTR_ERR(from); 45191da177e4SLinus Torvalds goto exit; 452091a27b2aSJeff Layton } 45211da177e4SLinus Torvalds 4522c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4523c1d4dd27SAl Viro &new_path, &new_last, &new_type); 452491a27b2aSJeff Layton if (IS_ERR(to)) { 452591a27b2aSJeff Layton error = PTR_ERR(to); 45261da177e4SLinus Torvalds goto exit1; 452791a27b2aSJeff Layton } 45281da177e4SLinus Torvalds 45291da177e4SLinus Torvalds error = -EXDEV; 4530f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45311da177e4SLinus Torvalds goto exit2; 45321da177e4SLinus Torvalds 45331da177e4SLinus Torvalds error = -EBUSY; 4534f5beed75SAl Viro if (old_type != LAST_NORM) 45351da177e4SLinus Torvalds goto exit2; 45361da177e4SLinus Torvalds 45370a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45380a7c3937SMiklos Szeredi error = -EEXIST; 4539f5beed75SAl Viro if (new_type != LAST_NORM) 45401da177e4SLinus Torvalds goto exit2; 45411da177e4SLinus Torvalds 4542f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4543c30dabfeSJan Kara if (error) 4544c30dabfeSJan Kara goto exit2; 4545c30dabfeSJan Kara 45468e6d782cSJ. Bruce Fields retry_deleg: 4547f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45481da177e4SLinus Torvalds 4549f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45501da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45511da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45521da177e4SLinus Torvalds goto exit3; 45531da177e4SLinus Torvalds /* source must exist */ 45541da177e4SLinus Torvalds error = -ENOENT; 4555b18825a7SDavid Howells if (d_is_negative(old_dentry)) 45561da177e4SLinus Torvalds goto exit4; 4557f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 45581da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45591da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 45601da177e4SLinus Torvalds goto exit4; 45610a7c3937SMiklos Szeredi error = -EEXIST; 45620a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 45630a7c3937SMiklos Szeredi goto exit5; 4564da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4565da1ce067SMiklos Szeredi error = -ENOENT; 4566da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4567da1ce067SMiklos Szeredi goto exit5; 4568da1ce067SMiklos Szeredi 4569da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4570da1ce067SMiklos Szeredi error = -ENOTDIR; 4571f5beed75SAl Viro if (new_last.name[new_last.len]) 4572da1ce067SMiklos Szeredi goto exit5; 4573da1ce067SMiklos Szeredi } 4574da1ce067SMiklos Szeredi } 45750a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 45760a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 45770a7c3937SMiklos Szeredi error = -ENOTDIR; 4578f5beed75SAl Viro if (old_last.name[old_last.len]) 45790a7c3937SMiklos Szeredi goto exit5; 4580f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 45810a7c3937SMiklos Szeredi goto exit5; 45820a7c3937SMiklos Szeredi } 45830a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 45840a7c3937SMiklos Szeredi error = -EINVAL; 45850a7c3937SMiklos Szeredi if (old_dentry == trap) 45860a7c3937SMiklos Szeredi goto exit5; 45871da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4588da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 45891da177e4SLinus Torvalds error = -ENOTEMPTY; 45901da177e4SLinus Torvalds if (new_dentry == trap) 45911da177e4SLinus Torvalds goto exit5; 45921da177e4SLinus Torvalds 4593f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4594f5beed75SAl Viro &new_path, new_dentry, flags); 4595be6d3e56SKentaro Takeda if (error) 4596c30dabfeSJan Kara goto exit5; 4597f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4598f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4599520c8b16SMiklos Szeredi &delegated_inode, flags); 46001da177e4SLinus Torvalds exit5: 46011da177e4SLinus Torvalds dput(new_dentry); 46021da177e4SLinus Torvalds exit4: 46031da177e4SLinus Torvalds dput(old_dentry); 46041da177e4SLinus Torvalds exit3: 4605f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46068e6d782cSJ. Bruce Fields if (delegated_inode) { 46078e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46088e6d782cSJ. Bruce Fields if (!error) 46098e6d782cSJ. Bruce Fields goto retry_deleg; 46108e6d782cSJ. Bruce Fields } 4611f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46121da177e4SLinus Torvalds exit2: 4613c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4614c6a94284SJeff Layton should_retry = true; 4615f5beed75SAl Viro path_put(&new_path); 46162ad94ae6SAl Viro putname(to); 46171da177e4SLinus Torvalds exit1: 4618f5beed75SAl Viro path_put(&old_path); 46191da177e4SLinus Torvalds putname(from); 4620c6a94284SJeff Layton if (should_retry) { 4621c6a94284SJeff Layton should_retry = false; 4622c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4623c6a94284SJeff Layton goto retry; 4624c6a94284SJeff Layton } 46252ad94ae6SAl Viro exit: 46261da177e4SLinus Torvalds return error; 46271da177e4SLinus Torvalds } 46281da177e4SLinus Torvalds 4629ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4630ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4631ee81feb6SDominik Brodowski { 4632ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4633ee81feb6SDominik Brodowski } 4634ee81feb6SDominik Brodowski 4635520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4636520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4637520c8b16SMiklos Szeredi { 4638ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4639520c8b16SMiklos Szeredi } 4640520c8b16SMiklos Szeredi 4641a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 46425590ff0dSUlrich Drepper { 4643ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 46445590ff0dSUlrich Drepper } 46455590ff0dSUlrich Drepper 4646787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4647787fb6bcSMiklos Szeredi { 4648787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4649787fb6bcSMiklos Szeredi if (error) 4650787fb6bcSMiklos Szeredi return error; 4651787fb6bcSMiklos Szeredi 4652787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4653787fb6bcSMiklos Szeredi return -EPERM; 4654787fb6bcSMiklos Szeredi 4655787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4656787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4657787fb6bcSMiklos Szeredi } 4658787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4659787fb6bcSMiklos Szeredi 46605d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 46611da177e4SLinus Torvalds { 46625d826c84SAl Viro int len = PTR_ERR(link); 46631da177e4SLinus Torvalds if (IS_ERR(link)) 46641da177e4SLinus Torvalds goto out; 46651da177e4SLinus Torvalds 46661da177e4SLinus Torvalds len = strlen(link); 46671da177e4SLinus Torvalds if (len > (unsigned) buflen) 46681da177e4SLinus Torvalds len = buflen; 46691da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 46701da177e4SLinus Torvalds len = -EFAULT; 46711da177e4SLinus Torvalds out: 46721da177e4SLinus Torvalds return len; 46731da177e4SLinus Torvalds } 46741da177e4SLinus Torvalds 46751da177e4SLinus Torvalds /* 46761da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 46776b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 46786b255391SAl Viro * for any given inode is up to filesystem. 46791da177e4SLinus Torvalds */ 4680d16744ecSMiklos Szeredi static int generic_readlink(struct dentry *dentry, char __user *buffer, 4681d16744ecSMiklos Szeredi int buflen) 46821da177e4SLinus Torvalds { 4683fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46845f2c4179SAl Viro struct inode *inode = d_inode(dentry); 46855f2c4179SAl Viro const char *link = inode->i_link; 4686694a1764SMarcin Slusarz int res; 4687cc314eefSLinus Torvalds 4688d4dee48bSAl Viro if (!link) { 4689fceef393SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4690680baacbSAl Viro if (IS_ERR(link)) 4691680baacbSAl Viro return PTR_ERR(link); 4692d4dee48bSAl Viro } 4693680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4694fceef393SAl Viro do_delayed_call(&done); 4695694a1764SMarcin Slusarz return res; 46961da177e4SLinus Torvalds } 46971da177e4SLinus Torvalds 4698d60874cdSMiklos Szeredi /** 4699fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4700fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4701fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4702fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4703fd4a0edfSMiklos Szeredi * 4704fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4705fd4a0edfSMiklos Szeredi * 4706fd4a0edfSMiklos Szeredi * Does not call security hook. 4707fd4a0edfSMiklos Szeredi */ 4708fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4709fd4a0edfSMiklos Szeredi { 4710fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4711fd4a0edfSMiklos Szeredi 471276fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 471376fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 471476fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 471576fca90eSMiklos Szeredi 471676fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4717fd4a0edfSMiklos Szeredi return -EINVAL; 4718fd4a0edfSMiklos Szeredi 471976fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 472076fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 472176fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 472276fca90eSMiklos Szeredi } 472376fca90eSMiklos Szeredi 472476fca90eSMiklos Szeredi return generic_readlink(dentry, buffer, buflen); 4725fd4a0edfSMiklos Szeredi } 4726fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 47271da177e4SLinus Torvalds 4728d60874cdSMiklos Szeredi /** 4729d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4730d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4731d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4732d60874cdSMiklos Szeredi * 4733d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4734d60874cdSMiklos Szeredi * 4735d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4736d60874cdSMiklos Szeredi * 4737d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4738d60874cdSMiklos Szeredi */ 4739d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4740d60874cdSMiklos Szeredi { 4741d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4742d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4743d60874cdSMiklos Szeredi 4744d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4745d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4746d60874cdSMiklos Szeredi if (!res) 4747d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4748d60874cdSMiklos Szeredi } 4749d60874cdSMiklos Szeredi return res; 4750d60874cdSMiklos Szeredi } 4751d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4752d60874cdSMiklos Szeredi 47531da177e4SLinus Torvalds /* get the link contents into pagecache */ 47546b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4755fceef393SAl Viro struct delayed_call *callback) 47561da177e4SLinus Torvalds { 4757ebd09abbSDuane Griffin char *kaddr; 47581da177e4SLinus Torvalds struct page *page; 47596b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47606b255391SAl Viro 4761d3883d4fSAl Viro if (!dentry) { 4762d3883d4fSAl Viro page = find_get_page(mapping, 0); 4763d3883d4fSAl Viro if (!page) 47646b255391SAl Viro return ERR_PTR(-ECHILD); 4765d3883d4fSAl Viro if (!PageUptodate(page)) { 4766d3883d4fSAl Viro put_page(page); 4767d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4768d3883d4fSAl Viro } 4769d3883d4fSAl Viro } else { 4770090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47711da177e4SLinus Torvalds if (IS_ERR(page)) 47726fe6900eSNick Piggin return (char*)page; 4773d3883d4fSAl Viro } 4774fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 477521fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 477621fc61c7SAl Viro kaddr = page_address(page); 47776b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4778ebd09abbSDuane Griffin return kaddr; 47791da177e4SLinus Torvalds } 47801da177e4SLinus Torvalds 47816b255391SAl Viro EXPORT_SYMBOL(page_get_link); 47821da177e4SLinus Torvalds 4783fceef393SAl Viro void page_put_link(void *arg) 47841da177e4SLinus Torvalds { 4785fceef393SAl Viro put_page(arg); 47861da177e4SLinus Torvalds } 47874d359507SAl Viro EXPORT_SYMBOL(page_put_link); 47881da177e4SLinus Torvalds 4789aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4790aa80deabSAl Viro { 4791fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47926b255391SAl Viro int res = readlink_copy(buffer, buflen, 47936b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4794fceef393SAl Viro &done)); 4795fceef393SAl Viro do_delayed_call(&done); 4796aa80deabSAl Viro return res; 4797aa80deabSAl Viro } 4798aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4799aa80deabSAl Viro 480054566b2cSNick Piggin /* 480154566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 480254566b2cSNick Piggin */ 480354566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48041da177e4SLinus Torvalds { 48051da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48060adb25d2SKirill Korotaev struct page *page; 4807afddba49SNick Piggin void *fsdata; 4808beb497abSDmitriy Monakhov int err; 4809c718a975STetsuo Handa unsigned int flags = 0; 481054566b2cSNick Piggin if (nofs) 481154566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 48121da177e4SLinus Torvalds 48137e53cac4SNeilBrown retry: 4814afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 481554566b2cSNick Piggin flags, &page, &fsdata); 48161da177e4SLinus Torvalds if (err) 4817afddba49SNick Piggin goto fail; 4818afddba49SNick Piggin 481921fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4820afddba49SNick Piggin 4821afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4822afddba49SNick Piggin page, fsdata); 48231da177e4SLinus Torvalds if (err < 0) 48241da177e4SLinus Torvalds goto fail; 4825afddba49SNick Piggin if (err < len-1) 4826afddba49SNick Piggin goto retry; 4827afddba49SNick Piggin 48281da177e4SLinus Torvalds mark_inode_dirty(inode); 48291da177e4SLinus Torvalds return 0; 48301da177e4SLinus Torvalds fail: 48311da177e4SLinus Torvalds return err; 48321da177e4SLinus Torvalds } 48334d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 48341da177e4SLinus Torvalds 48350adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 48360adb25d2SKirill Korotaev { 48370adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4838c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48390adb25d2SKirill Korotaev } 48404d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48410adb25d2SKirill Korotaev 484292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48436b255391SAl Viro .get_link = page_get_link, 48441da177e4SLinus Torvalds }; 48451da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4846