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) { 22530ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 22608518549SPaul Moore struct filename *tmp; 22708518549SPaul Moore 22830ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 22908518549SPaul Moore if (unlikely(!tmp)) { 23008518549SPaul Moore __putname(result); 23108518549SPaul Moore return ERR_PTR(-ENOMEM); 23208518549SPaul Moore } 23308518549SPaul Moore tmp->name = (char *)result; 23408518549SPaul Moore result = tmp; 23508518549SPaul Moore } else { 23608518549SPaul Moore __putname(result); 23708518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23808518549SPaul Moore } 23908518549SPaul Moore memcpy((char *)result->name, filename, len); 240c4ad8f98SLinus Torvalds result->uptr = NULL; 241c4ad8f98SLinus Torvalds result->aname = NULL; 24255422d0bSPaul Moore result->refcnt = 1; 243fd3522fdSPaul Moore audit_getname(result); 244c4ad8f98SLinus Torvalds 245c4ad8f98SLinus Torvalds return result; 246c4ad8f98SLinus Torvalds } 247c4ad8f98SLinus Torvalds 24891a27b2aSJeff Layton void putname(struct filename *name) 2491da177e4SLinus Torvalds { 25055422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 25155422d0bSPaul Moore 25255422d0bSPaul Moore if (--name->refcnt > 0) 25355422d0bSPaul Moore return; 25455422d0bSPaul Moore 255fd2f7cb5SAl Viro if (name->name != name->iname) { 25655422d0bSPaul Moore __putname(name->name); 25755422d0bSPaul Moore kfree(name); 25855422d0bSPaul Moore } else 25955422d0bSPaul Moore __putname(name); 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds 262e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 263e77819e5SLinus Torvalds { 26484635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 265e77819e5SLinus Torvalds struct posix_acl *acl; 266e77819e5SLinus Torvalds 267e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2683567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2693567866bSAl Viro if (!acl) 270e77819e5SLinus Torvalds return -EAGAIN; 2713567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 272b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 273e77819e5SLinus Torvalds return -ECHILD; 274206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 275e77819e5SLinus Torvalds } 276e77819e5SLinus Torvalds 2772982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2784e34e719SChristoph Hellwig if (IS_ERR(acl)) 2794e34e719SChristoph Hellwig return PTR_ERR(acl); 280e77819e5SLinus Torvalds if (acl) { 281e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 282e77819e5SLinus Torvalds posix_acl_release(acl); 283e77819e5SLinus Torvalds return error; 284e77819e5SLinus Torvalds } 28584635d68SLinus Torvalds #endif 286e77819e5SLinus Torvalds 287e77819e5SLinus Torvalds return -EAGAIN; 288e77819e5SLinus Torvalds } 289e77819e5SLinus Torvalds 2905909ccaaSLinus Torvalds /* 291948409c7SAndreas Gruenbacher * This does the basic permission checking 2925909ccaaSLinus Torvalds */ 2937e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2945909ccaaSLinus Torvalds { 29526cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2965909ccaaSLinus Torvalds 2978e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2985909ccaaSLinus Torvalds mode >>= 6; 2995909ccaaSLinus Torvalds else { 300e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 3017e40145eSAl Viro int error = check_acl(inode, mask); 3025909ccaaSLinus Torvalds if (error != -EAGAIN) 3035909ccaaSLinus Torvalds return error; 3045909ccaaSLinus Torvalds } 3055909ccaaSLinus Torvalds 3065909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3075909ccaaSLinus Torvalds mode >>= 3; 3085909ccaaSLinus Torvalds } 3095909ccaaSLinus Torvalds 3105909ccaaSLinus Torvalds /* 3115909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3125909ccaaSLinus Torvalds */ 3139c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3145909ccaaSLinus Torvalds return 0; 3155909ccaaSLinus Torvalds return -EACCES; 3165909ccaaSLinus Torvalds } 3171da177e4SLinus Torvalds 3181da177e4SLinus Torvalds /** 3191da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3201da177e4SLinus Torvalds * @inode: inode to check access rights for 3218fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3221da177e4SLinus Torvalds * 3231da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3241da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3251da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 326b74c79e9SNick Piggin * are used for other things. 327b74c79e9SNick Piggin * 328b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 329b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 330b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3311da177e4SLinus Torvalds */ 3322830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3331da177e4SLinus Torvalds { 3345909ccaaSLinus Torvalds int ret; 3351da177e4SLinus Torvalds 3361da177e4SLinus Torvalds /* 337948409c7SAndreas Gruenbacher * Do the basic permission checks. 3381da177e4SLinus Torvalds */ 3397e40145eSAl Viro ret = acl_permission_check(inode, mask); 3405909ccaaSLinus Torvalds if (ret != -EACCES) 3415909ccaaSLinus Torvalds return ret; 3421da177e4SLinus Torvalds 343d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 344d594e7ecSAl Viro /* DACs are overridable for directories */ 345d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34623adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34723adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 348d594e7ecSAl Viro return 0; 34923adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3501da177e4SLinus Torvalds return 0; 3512a4c2242SStephen Smalley return -EACCES; 3522a4c2242SStephen Smalley } 3531da177e4SLinus Torvalds 3541da177e4SLinus Torvalds /* 3551da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3561da177e4SLinus Torvalds */ 3577ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 358d594e7ecSAl Viro if (mask == MAY_READ) 35923adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3601da177e4SLinus Torvalds return 0; 3612a4c2242SStephen Smalley /* 3622a4c2242SStephen Smalley * Read/write DACs are always overridable. 3632a4c2242SStephen Smalley * Executable DACs are overridable when there is 3642a4c2242SStephen Smalley * at least one exec bit set. 3652a4c2242SStephen Smalley */ 3662a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 3672a4c2242SStephen Smalley if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3682a4c2242SStephen Smalley return 0; 3691da177e4SLinus Torvalds 3701da177e4SLinus Torvalds return -EACCES; 3711da177e4SLinus Torvalds } 3724d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3731da177e4SLinus Torvalds 3743ddcd056SLinus Torvalds /* 3753ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3763ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3773ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3783ddcd056SLinus Torvalds * permission function, use the fast case". 3793ddcd056SLinus Torvalds */ 3803ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3813ddcd056SLinus Torvalds { 3823ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3833ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3843ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3853ddcd056SLinus Torvalds 3863ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3873ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3883ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3893ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3903ddcd056SLinus Torvalds } 3913ddcd056SLinus Torvalds return generic_permission(inode, mask); 3923ddcd056SLinus Torvalds } 3933ddcd056SLinus Torvalds 394cb23beb5SChristoph Hellwig /** 3950bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 3960bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 39755852635SRandy Dunlap * @inode: Inode to check permission on 3980bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3990bdaea90SDavid Howells * 4000bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4010bdaea90SDavid Howells */ 4020bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4030bdaea90SDavid Howells { 4040bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4050bdaea90SDavid Howells umode_t mode = inode->i_mode; 4060bdaea90SDavid Howells 4070bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 408bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4090bdaea90SDavid Howells return -EROFS; 4100bdaea90SDavid Howells } 4110bdaea90SDavid Howells return 0; 4120bdaea90SDavid Howells } 4130bdaea90SDavid Howells 4140bdaea90SDavid Howells /** 4150bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4160bdaea90SDavid Howells * @inode: Inode to check permission on 4170bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4180bdaea90SDavid Howells * 4190bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4200bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4210bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4220bdaea90SDavid Howells * 4230bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4240bdaea90SDavid Howells */ 4250bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4260bdaea90SDavid Howells { 4270bdaea90SDavid Howells int retval; 4280bdaea90SDavid Howells 4290bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4300bdaea90SDavid Howells if (retval) 4310bdaea90SDavid Howells return retval; 4324bfd054aSEric Biggers 4334bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 4344bfd054aSEric Biggers /* 4354bfd054aSEric Biggers * Nobody gets write access to an immutable file. 4364bfd054aSEric Biggers */ 4374bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 4384bfd054aSEric Biggers return -EPERM; 4394bfd054aSEric Biggers 4404bfd054aSEric Biggers /* 4414bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 4424bfd054aSEric Biggers * written back improperly if their true value is unknown 4434bfd054aSEric Biggers * to the vfs. 4444bfd054aSEric Biggers */ 4454bfd054aSEric Biggers if (HAS_UNMAPPED_ID(inode)) 4464bfd054aSEric Biggers return -EACCES; 4474bfd054aSEric Biggers } 4484bfd054aSEric Biggers 4494bfd054aSEric Biggers retval = do_inode_permission(inode, mask); 4504bfd054aSEric Biggers if (retval) 4514bfd054aSEric Biggers return retval; 4524bfd054aSEric Biggers 4534bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 4544bfd054aSEric Biggers if (retval) 4554bfd054aSEric Biggers return retval; 4564bfd054aSEric Biggers 4574bfd054aSEric Biggers return security_inode_permission(inode, mask); 4580bdaea90SDavid Howells } 4594d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4600bdaea90SDavid Howells 4610bdaea90SDavid Howells /** 4625dd784d0SJan Blunck * path_get - get a reference to a path 4635dd784d0SJan Blunck * @path: path to get the reference to 4645dd784d0SJan Blunck * 4655dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4665dd784d0SJan Blunck */ 467dcf787f3SAl Viro void path_get(const struct path *path) 4685dd784d0SJan Blunck { 4695dd784d0SJan Blunck mntget(path->mnt); 4705dd784d0SJan Blunck dget(path->dentry); 4715dd784d0SJan Blunck } 4725dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4735dd784d0SJan Blunck 4745dd784d0SJan Blunck /** 4751d957f9bSJan Blunck * path_put - put a reference to a path 4761d957f9bSJan Blunck * @path: path to put the reference to 4771d957f9bSJan Blunck * 4781d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4791d957f9bSJan Blunck */ 480dcf787f3SAl Viro void path_put(const struct path *path) 4811da177e4SLinus Torvalds { 4821d957f9bSJan Blunck dput(path->dentry); 4831d957f9bSJan Blunck mntput(path->mnt); 4841da177e4SLinus Torvalds } 4851d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4861da177e4SLinus Torvalds 487894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4881f55a6ecSAl Viro struct nameidata { 4891f55a6ecSAl Viro struct path path; 4901f55a6ecSAl Viro struct qstr last; 4911f55a6ecSAl Viro struct path root; 4921f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 4931f55a6ecSAl Viro unsigned int flags; 494ab87f9a5SAleksa Sarai unsigned seq, m_seq, r_seq; 4951f55a6ecSAl Viro int last_type; 4961f55a6ecSAl Viro unsigned depth; 497756daf26SNeilBrown int total_link_count; 498697fc6caSAl Viro struct saved { 499697fc6caSAl Viro struct path link; 500fceef393SAl Viro struct delayed_call done; 501697fc6caSAl Viro const char *name; 5020450b2d1SAl Viro unsigned seq; 503894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5049883d185SAl Viro struct filename *name; 5059883d185SAl Viro struct nameidata *saved; 506fceef393SAl Viro struct inode *link_inode; 5079883d185SAl Viro unsigned root_seq; 5089883d185SAl Viro int dfd; 5093859a271SKees Cook } __randomize_layout; 5101f55a6ecSAl Viro 5119883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 512894bc8c4SAl Viro { 513756daf26SNeilBrown struct nameidata *old = current->nameidata; 514756daf26SNeilBrown p->stack = p->internal; 515c8a53ee5SAl Viro p->dfd = dfd; 516c8a53ee5SAl Viro p->name = name; 517756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5189883d185SAl Viro p->saved = old; 519756daf26SNeilBrown current->nameidata = p; 520894bc8c4SAl Viro } 521894bc8c4SAl Viro 5229883d185SAl Viro static void restore_nameidata(void) 523894bc8c4SAl Viro { 5249883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 525756daf26SNeilBrown 526756daf26SNeilBrown current->nameidata = old; 527756daf26SNeilBrown if (old) 528756daf26SNeilBrown old->total_link_count = now->total_link_count; 529e1a63bbcSAl Viro if (now->stack != now->internal) 530756daf26SNeilBrown kfree(now->stack); 531894bc8c4SAl Viro } 532894bc8c4SAl Viro 533894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 534894bc8c4SAl Viro { 535bc40aee0SAl Viro struct saved *p; 536bc40aee0SAl Viro 537bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 5386da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 539bc40aee0SAl Viro GFP_ATOMIC); 540bc40aee0SAl Viro if (unlikely(!p)) 541bc40aee0SAl Viro return -ECHILD; 542bc40aee0SAl Viro } else { 5436da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 544894bc8c4SAl Viro GFP_KERNEL); 545894bc8c4SAl Viro if (unlikely(!p)) 546894bc8c4SAl Viro return -ENOMEM; 547bc40aee0SAl Viro } 548894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 549894bc8c4SAl Viro nd->stack = p; 550894bc8c4SAl Viro return 0; 551894bc8c4SAl Viro } 552894bc8c4SAl Viro 553397d425dSEric W. Biederman /** 554397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 555397d425dSEric W. Biederman * @path: nameidate to verify 556397d425dSEric W. Biederman * 557397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 558397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 559397d425dSEric W. Biederman */ 560397d425dSEric W. Biederman static bool path_connected(const struct path *path) 561397d425dSEric W. Biederman { 562397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 56395dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 564397d425dSEric W. Biederman 56595dd7758SEric W. Biederman /* Bind mounts and multi-root filesystems can have disconnected paths */ 56695dd7758SEric W. Biederman if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root)) 567397d425dSEric W. Biederman return true; 568397d425dSEric W. Biederman 569397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 570397d425dSEric W. Biederman } 571397d425dSEric W. Biederman 572894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 573894bc8c4SAl Viro { 574da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 575894bc8c4SAl Viro return 0; 576894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 577894bc8c4SAl Viro return 0; 578894bc8c4SAl Viro return __nd_alloc_stack(nd); 579894bc8c4SAl Viro } 580894bc8c4SAl Viro 5817973387aSAl Viro static void drop_links(struct nameidata *nd) 5827973387aSAl Viro { 5837973387aSAl Viro int i = nd->depth; 5847973387aSAl Viro while (i--) { 5857973387aSAl Viro struct saved *last = nd->stack + i; 586fceef393SAl Viro do_delayed_call(&last->done); 587fceef393SAl Viro clear_delayed_call(&last->done); 5887973387aSAl Viro } 5897973387aSAl Viro } 5907973387aSAl Viro 5917973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5927973387aSAl Viro { 5937973387aSAl Viro drop_links(nd); 5947973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5957973387aSAl Viro int i; 5967973387aSAl Viro path_put(&nd->path); 5977973387aSAl Viro for (i = 0; i < nd->depth; i++) 5987973387aSAl Viro path_put(&nd->stack[i].link); 59984a2bd39SAl Viro if (nd->flags & LOOKUP_ROOT_GRABBED) { 600102b8af2SAl Viro path_put(&nd->root); 60184a2bd39SAl Viro nd->flags &= ~LOOKUP_ROOT_GRABBED; 602102b8af2SAl Viro } 6037973387aSAl Viro } else { 6047973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 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 642ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 643ee594bffSAl Viro { 644adb21d2bSAleksa Sarai /* 645adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 646adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 647adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 648adb21d2bSAleksa Sarai */ 649adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 650adb21d2bSAleksa Sarai return false; 651adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 652ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 653ee594bffSAl Viro return true; 65484a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 655ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 656ee594bffSAl Viro } 657ee594bffSAl Viro 65819660af7SAl Viro /* 65931e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 66019660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 66119660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 66257e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66319660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66419660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66519660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66619660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66731e6b01fSNick Piggin */ 66831e6b01fSNick Piggin 66931e6b01fSNick Piggin /** 67019660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 67119660af7SAl Viro * @nd: nameidata pathwalk data 67239191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67331e6b01fSNick Piggin * 6744675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6754675ac39SAl Viro * for ref-walk mode. 6764675ac39SAl Viro * Must be called from rcu-walk context. 6777973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6787973387aSAl Viro * terminate_walk(). 67931e6b01fSNick Piggin */ 6804675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 68131e6b01fSNick Piggin { 68231e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68331e6b01fSNick Piggin 68431e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 685e5c832d5SLinus Torvalds 686e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6877973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6884675ac39SAl Viro goto out1; 68984a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 69084a2bd39SAl Viro goto out; 691ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 6924675ac39SAl Viro goto out; 6934675ac39SAl Viro rcu_read_unlock(); 6944675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6954675ac39SAl Viro return 0; 6964675ac39SAl Viro 69784a2bd39SAl Viro out1: 6984675ac39SAl Viro nd->path.mnt = NULL; 6994675ac39SAl Viro nd->path.dentry = NULL; 7004675ac39SAl Viro out: 7014675ac39SAl Viro rcu_read_unlock(); 7024675ac39SAl Viro return -ECHILD; 7034675ac39SAl Viro } 7044675ac39SAl Viro 7054675ac39SAl Viro /** 7064675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7074675ac39SAl Viro * @nd: nameidata pathwalk data 7084675ac39SAl Viro * @dentry: child of nd->path.dentry 7094675ac39SAl Viro * @seq: seq number to check dentry against 7104675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7114675ac39SAl Viro * 7124675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7134675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7144675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7154675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7164675ac39SAl Viro * terminate_walk(). 7174675ac39SAl Viro */ 7184675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7194675ac39SAl Viro { 7204675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7214675ac39SAl Viro 7224675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7234675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7244675ac39SAl Viro goto out2; 7257973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7267973387aSAl Viro goto out2; 7274675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7287973387aSAl Viro goto out1; 72948a066e7SAl Viro 73015570086SLinus Torvalds /* 7314675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7324675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7334675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7344675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7354675ac39SAl Viro * know the parent must still be valid if the child sequence number is 73615570086SLinus Torvalds */ 7374675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 738e5c832d5SLinus Torvalds goto out; 73984a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 74084a2bd39SAl Viro goto out_dput; 741e5c832d5SLinus Torvalds /* 742e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 743e5c832d5SLinus Torvalds * still valid and get it if required. 744e5c832d5SLinus Torvalds */ 74584a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 74684a2bd39SAl Viro goto out_dput; 7478b61e74fSAl Viro rcu_read_unlock(); 74831e6b01fSNick Piggin return 0; 74919660af7SAl Viro 7507973387aSAl Viro out2: 7517973387aSAl Viro nd->path.mnt = NULL; 7527973387aSAl Viro out1: 7537973387aSAl Viro nd->path.dentry = NULL; 754e5c832d5SLinus Torvalds out: 7558b61e74fSAl Viro rcu_read_unlock(); 75684a2bd39SAl Viro return -ECHILD; 75784a2bd39SAl Viro out_dput: 75884a2bd39SAl Viro rcu_read_unlock(); 75984a2bd39SAl Viro dput(dentry); 76031e6b01fSNick Piggin return -ECHILD; 76131e6b01fSNick Piggin } 76231e6b01fSNick Piggin 7634ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76434286d66SNick Piggin { 765a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7664ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 767a89f8337SAl Viro else 768a89f8337SAl Viro return 1; 76934286d66SNick Piggin } 77034286d66SNick Piggin 7719f1fafeeSAl Viro /** 7729f1fafeeSAl Viro * complete_walk - successful completion of path walk 7739f1fafeeSAl Viro * @nd: pointer nameidata 77439159de2SJeff Layton * 7759f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7769f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7779f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7789f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7799f1fafeeSAl Viro * need to drop nd->path. 78039159de2SJeff Layton */ 7819f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78239159de2SJeff Layton { 78316c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78439159de2SJeff Layton int status; 78539159de2SJeff Layton 7869f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 787adb21d2bSAleksa Sarai /* 788adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 789adb21d2bSAleksa Sarai * externally-managed nd->root. 790adb21d2bSAleksa Sarai */ 791adb21d2bSAleksa Sarai if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) 7929f1fafeeSAl Viro nd->root.mnt = NULL; 7934675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 79448a066e7SAl Viro return -ECHILD; 79548a066e7SAl Viro } 7969f1fafeeSAl Viro 797adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 798adb21d2bSAleksa Sarai /* 799adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 800adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 801adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 802adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 803adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 804adb21d2bSAleksa Sarai * 805adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 806adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 807adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 808adb21d2bSAleksa Sarai * requested root to userspace. 809adb21d2bSAleksa Sarai * 810adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 811adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 812adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 813adb21d2bSAleksa Sarai */ 814adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 815adb21d2bSAleksa Sarai return -EXDEV; 816adb21d2bSAleksa Sarai } 817adb21d2bSAleksa Sarai 81816c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 81939159de2SJeff Layton return 0; 82039159de2SJeff Layton 821ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 82216c2cd71SAl Viro return 0; 82316c2cd71SAl Viro 824ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 82539159de2SJeff Layton if (status > 0) 82639159de2SJeff Layton return 0; 82739159de2SJeff Layton 82816c2cd71SAl Viro if (!status) 82939159de2SJeff Layton status = -ESTALE; 83016c2cd71SAl Viro 83139159de2SJeff Layton return status; 83239159de2SJeff Layton } 83339159de2SJeff Layton 834740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 8352a737871SAl Viro { 83631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8379e6697e2SAl Viro 838adb21d2bSAleksa Sarai /* 839adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 840adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 841adb21d2bSAleksa Sarai * from the dirfd. 842adb21d2bSAleksa Sarai */ 843adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 844adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 845adb21d2bSAleksa Sarai 8469e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8478f47a016SAl Viro unsigned seq; 848c28cc364SNick Piggin 849c28cc364SNick Piggin do { 850c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 85131e6b01fSNick Piggin nd->root = fs->root; 8528f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 853c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8549e6697e2SAl Viro } else { 8559e6697e2SAl Viro get_fs_root(fs, &nd->root); 85684a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 8579e6697e2SAl Viro } 858740a1678SAleksa Sarai return 0; 85931e6b01fSNick Piggin } 86031e6b01fSNick Piggin 8611d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 862051d3812SIan Kent { 863051d3812SIan Kent dput(path->dentry); 8644ac91378SJan Blunck if (path->mnt != nd->path.mnt) 865051d3812SIan Kent mntput(path->mnt); 866051d3812SIan Kent } 867051d3812SIan Kent 8687b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8697b9337aaSNick Piggin struct nameidata *nd) 870051d3812SIan Kent { 87131e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8724ac91378SJan Blunck dput(nd->path.dentry); 87331e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8744ac91378SJan Blunck mntput(nd->path.mnt); 8759a229683SHuang Shijie } 87631e6b01fSNick Piggin nd->path.mnt = path->mnt; 8774ac91378SJan Blunck nd->path.dentry = path->dentry; 878051d3812SIan Kent } 879051d3812SIan Kent 880248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 881248fb5b9SAl Viro { 882adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 883adb21d2bSAleksa Sarai return -EXDEV; 88472ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 88572ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 88672ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 88772ba2929SAleksa Sarai return -EXDEV; 88872ba2929SAleksa Sarai } 889740a1678SAleksa Sarai if (!nd->root.mnt) { 890740a1678SAleksa Sarai int error = set_root(nd); 891740a1678SAleksa Sarai if (error) 892740a1678SAleksa Sarai return error; 893740a1678SAleksa Sarai } 894248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 895248fb5b9SAl Viro struct dentry *d; 896248fb5b9SAl Viro nd->path = nd->root; 897248fb5b9SAl Viro d = nd->path.dentry; 898248fb5b9SAl Viro nd->inode = d->d_inode; 899248fb5b9SAl Viro nd->seq = nd->root_seq; 900248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 901248fb5b9SAl Viro return -ECHILD; 902248fb5b9SAl Viro } else { 903248fb5b9SAl Viro path_put(&nd->path); 904248fb5b9SAl Viro nd->path = nd->root; 905248fb5b9SAl Viro path_get(&nd->path); 906248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 907248fb5b9SAl Viro } 908248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 909248fb5b9SAl Viro return 0; 910248fb5b9SAl Viro } 911248fb5b9SAl Viro 912b5fb63c1SChristoph Hellwig /* 9136b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 914b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 915b5fb63c1SChristoph Hellwig */ 9161bc82070SAleksa Sarai int nd_jump_link(struct path *path) 917b5fb63c1SChristoph Hellwig { 9184b99d499SAleksa Sarai int error = -ELOOP; 9196e77137bSAl Viro struct nameidata *nd = current->nameidata; 920b5fb63c1SChristoph Hellwig 9214b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9224b99d499SAleksa Sarai goto err; 9234b99d499SAleksa Sarai 92472ba2929SAleksa Sarai error = -EXDEV; 92572ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 92672ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 92772ba2929SAleksa Sarai goto err; 92872ba2929SAleksa Sarai } 929adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 930adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 931adb21d2bSAleksa Sarai goto err; 93272ba2929SAleksa Sarai 9334b99d499SAleksa Sarai path_put(&nd->path); 934b5fb63c1SChristoph Hellwig nd->path = *path; 935b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 936b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 9371bc82070SAleksa Sarai return 0; 9384b99d499SAleksa Sarai 9394b99d499SAleksa Sarai err: 9404b99d499SAleksa Sarai path_put(path); 9414b99d499SAleksa Sarai return error; 942b5fb63c1SChristoph Hellwig } 943b5fb63c1SChristoph Hellwig 944b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 945574197e0SAl Viro { 94621c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 947fceef393SAl Viro do_delayed_call(&last->done); 9486548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 949b9ff4429SAl Viro path_put(&last->link); 950574197e0SAl Viro } 951574197e0SAl Viro 952561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 953561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 95430aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 95530aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 956800179c9SKees Cook 957800179c9SKees Cook /** 958800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 95955852635SRandy Dunlap * @nd: nameidata pathwalk data 960800179c9SKees Cook * 961800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 962800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 963800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 964800179c9SKees Cook * processes from failing races against path names that may change out 965800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 966800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 967800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 968800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 969800179c9SKees Cook * 970800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 971800179c9SKees Cook */ 972fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 973800179c9SKees Cook { 974800179c9SKees Cook const struct inode *inode; 975800179c9SKees Cook const struct inode *parent; 9762d7f9e2aSSeth Forshee kuid_t puid; 977800179c9SKees Cook 978800179c9SKees Cook if (!sysctl_protected_symlinks) 979800179c9SKees Cook return 0; 980800179c9SKees Cook 981800179c9SKees Cook /* Allowed if owner and follower match. */ 982fceef393SAl Viro inode = nd->link_inode; 98381abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 984800179c9SKees Cook return 0; 985800179c9SKees Cook 986800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 987aa65fa35SAl Viro parent = nd->inode; 988800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 989800179c9SKees Cook return 0; 990800179c9SKees Cook 991800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9922d7f9e2aSSeth Forshee puid = parent->i_uid; 9932d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 994800179c9SKees Cook return 0; 995800179c9SKees Cook 99631956502SAl Viro if (nd->flags & LOOKUP_RCU) 99731956502SAl Viro return -ECHILD; 99831956502SAl Viro 999ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1000245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1001800179c9SKees Cook return -EACCES; 1002800179c9SKees Cook } 1003800179c9SKees Cook 1004800179c9SKees Cook /** 1005800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1006800179c9SKees Cook * @inode: the source inode to hardlink from 1007800179c9SKees Cook * 1008800179c9SKees Cook * Return false if at least one of the following conditions: 1009800179c9SKees Cook * - inode is not a regular file 1010800179c9SKees Cook * - inode is setuid 1011800179c9SKees Cook * - inode is setgid and group-exec 1012800179c9SKees Cook * - access failure for read and write 1013800179c9SKees Cook * 1014800179c9SKees Cook * Otherwise returns true. 1015800179c9SKees Cook */ 1016800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 1017800179c9SKees Cook { 1018800179c9SKees Cook umode_t mode = inode->i_mode; 1019800179c9SKees Cook 1020800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1021800179c9SKees Cook if (!S_ISREG(mode)) 1022800179c9SKees Cook return false; 1023800179c9SKees Cook 1024800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1025800179c9SKees Cook if (mode & S_ISUID) 1026800179c9SKees Cook return false; 1027800179c9SKees Cook 1028800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1029800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1030800179c9SKees Cook return false; 1031800179c9SKees Cook 1032800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1033800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 1034800179c9SKees Cook return false; 1035800179c9SKees Cook 1036800179c9SKees Cook return true; 1037800179c9SKees Cook } 1038800179c9SKees Cook 1039800179c9SKees Cook /** 1040800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1041800179c9SKees Cook * @link: the source to hardlink from 1042800179c9SKees Cook * 1043800179c9SKees Cook * Block hardlink when all of: 1044800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1045800179c9SKees Cook * - fsuid does not match inode 1046800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1047f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1048800179c9SKees Cook * 1049800179c9SKees Cook * Returns 0 if successful, -ve on error. 1050800179c9SKees Cook */ 1051800179c9SKees Cook static int may_linkat(struct path *link) 1052800179c9SKees Cook { 1053593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1054593d1ce8SEric W. Biederman 1055593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1056593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 1057593d1ce8SEric W. Biederman return -EOVERFLOW; 1058800179c9SKees Cook 1059800179c9SKees Cook if (!sysctl_protected_hardlinks) 1060800179c9SKees Cook return 0; 1061800179c9SKees Cook 1062800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1063800179c9SKees Cook * otherwise, it must be a safe source. 1064800179c9SKees Cook */ 1065cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1066800179c9SKees Cook return 0; 1067800179c9SKees Cook 1068245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1069800179c9SKees Cook return -EPERM; 1070800179c9SKees Cook } 1071800179c9SKees Cook 107230aba665SSalvatore Mesoraca /** 107330aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 107430aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 107530aba665SSalvatore Mesoraca * exist. 1076d0cb5018SAl Viro * @dir_mode: mode bits of directory 1077d0cb5018SAl Viro * @dir_uid: owner of directory 107830aba665SSalvatore Mesoraca * @inode: the inode of the file to open 107930aba665SSalvatore Mesoraca * 108030aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 108130aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 108230aba665SSalvatore Mesoraca * - the file already exists 108330aba665SSalvatore Mesoraca * - we are in a sticky directory 108430aba665SSalvatore Mesoraca * - we don't own the file 108530aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 108630aba665SSalvatore Mesoraca * - the directory is world writable 108730aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 108830aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 108930aba665SSalvatore Mesoraca * be enough. 109030aba665SSalvatore Mesoraca * 109130aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 109230aba665SSalvatore Mesoraca */ 1093d0cb5018SAl Viro static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid, 109430aba665SSalvatore Mesoraca struct inode * const inode) 109530aba665SSalvatore Mesoraca { 109630aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 109730aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1098d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1099d0cb5018SAl Viro uid_eq(inode->i_uid, dir_uid) || 110030aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 110130aba665SSalvatore Mesoraca return 0; 110230aba665SSalvatore Mesoraca 1103d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1104d0cb5018SAl Viro (dir_mode & 0020 && 110530aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 110630aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1107245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1108245d7369SKees Cook "sticky_create_fifo" : 1109245d7369SKees Cook "sticky_create_regular"; 1110245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 111130aba665SSalvatore Mesoraca return -EACCES; 111230aba665SSalvatore Mesoraca } 111330aba665SSalvatore Mesoraca return 0; 111430aba665SSalvatore Mesoraca } 111530aba665SSalvatore Mesoraca 11163b2e7f75SAl Viro static __always_inline 11173b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 11181da177e4SLinus Torvalds { 1119ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 11201cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1121fceef393SAl Viro struct inode *inode = nd->link_inode; 11226d7b5aaeSAl Viro int error; 11230a959df5SAl Viro const char *res; 11241da177e4SLinus Torvalds 112527812141SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) 112627812141SAleksa Sarai return ERR_PTR(-ELOOP); 112727812141SAleksa Sarai 11288fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 11298fa9dd24SNeilBrown touch_atime(&last->link); 11308fa9dd24SNeilBrown cond_resched(); 1131c6718543SMiklos Szeredi } else if (atime_needs_update(&last->link, inode)) { 11324675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 11338fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 11348fa9dd24SNeilBrown touch_atime(&last->link); 11358fa9dd24SNeilBrown } 11368fa9dd24SNeilBrown 1137bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1138bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1139bda0be7aSNeilBrown if (unlikely(error)) 11406920a440SAl Viro return ERR_PTR(error); 114136f3b4f6SAl Viro 114286acdca1SAl Viro nd->last_type = LAST_BIND; 11434c4f7c19SEric Biggers res = READ_ONCE(inode->i_link); 1144d4dee48bSAl Viro if (!res) { 1145fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1146fceef393SAl Viro struct delayed_call *); 1147fceef393SAl Viro get = inode->i_op->get_link; 11488c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1149fceef393SAl Viro res = get(NULL, inode, &last->done); 11506b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 11514675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 11528c1b4566SAl Viro return ERR_PTR(-ECHILD); 1153fceef393SAl Viro res = get(dentry, inode, &last->done); 11548c1b4566SAl Viro } 11556b255391SAl Viro } else { 1156fceef393SAl Viro res = get(dentry, inode, &last->done); 11576b255391SAl Viro } 1158fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1159fab51e8aSAl Viro return res; 11600a959df5SAl Viro } 1161fab51e8aSAl Viro if (*res == '/') { 1162740a1678SAleksa Sarai error = nd_jump_root(nd); 1163740a1678SAleksa Sarai if (unlikely(error)) 1164740a1678SAleksa Sarai return ERR_PTR(error); 1165fab51e8aSAl Viro while (unlikely(*++res == '/')) 1166fab51e8aSAl Viro ; 1167fab51e8aSAl Viro } 1168fab51e8aSAl Viro if (!*res) 1169fab51e8aSAl Viro res = NULL; 11700a959df5SAl Viro return res; 11710a959df5SAl Viro } 11726d7b5aaeSAl Viro 1173f015f126SDavid Howells /* 1174f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1175f015f126SDavid Howells * 1176f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1177f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1178f015f126SDavid Howells * Up is towards /. 1179f015f126SDavid Howells * 1180f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1181f015f126SDavid Howells * root. 1182f015f126SDavid Howells */ 1183bab77ebfSAl Viro int follow_up(struct path *path) 11841da177e4SLinus Torvalds { 11850714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11860714a533SAl Viro struct mount *parent; 11871da177e4SLinus Torvalds struct dentry *mountpoint; 118899b7db7bSNick Piggin 118948a066e7SAl Viro read_seqlock_excl(&mount_lock); 11900714a533SAl Viro parent = mnt->mnt_parent; 11913c0a6163SAl Viro if (parent == mnt) { 119248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11931da177e4SLinus Torvalds return 0; 11941da177e4SLinus Torvalds } 11950714a533SAl Viro mntget(&parent->mnt); 1196a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 119748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1198bab77ebfSAl Viro dput(path->dentry); 1199bab77ebfSAl Viro path->dentry = mountpoint; 1200bab77ebfSAl Viro mntput(path->mnt); 12010714a533SAl Viro path->mnt = &parent->mnt; 12021da177e4SLinus Torvalds return 1; 12031da177e4SLinus Torvalds } 12044d359507SAl Viro EXPORT_SYMBOL(follow_up); 12051da177e4SLinus Torvalds 1206b5c84bf6SNick Piggin /* 12079875cf80SDavid Howells * Perform an automount 12089875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 12099875cf80SDavid Howells * were called with. 12101da177e4SLinus Torvalds */ 12111c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 121231e6b01fSNick Piggin { 121325e195aaSAl Viro struct dentry *dentry = path->dentry; 12149875cf80SDavid Howells 12150ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 12160ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 12170ec26fd0SMiklos Szeredi * the name. 12180ec26fd0SMiklos Szeredi * 12190ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 12205a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 12210ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 12220ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 12230ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 12240ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 12255a30d8a2SDavid Howells */ 12261c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 12275d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 122825e195aaSAl Viro dentry->d_inode) 12299875cf80SDavid Howells return -EISDIR; 12300ec26fd0SMiklos Szeredi 12311c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 12329875cf80SDavid Howells return -ELOOP; 12339875cf80SDavid Howells 123425e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1235ea5b778aSDavid Howells } 12369875cf80SDavid Howells 12379875cf80SDavid Howells /* 12389875cf80SDavid Howells * Handle a dentry that is managed in some way. 1239cc53ce53SDavid Howells * - Flagged for transit management (autofs) 12409875cf80SDavid Howells * - Flagged as mountpoint 12419875cf80SDavid Howells * - Flagged as automount point 12429875cf80SDavid Howells * 12439875cf80SDavid Howells * This may only be called in refwalk mode. 1244d41efb52SAl Viro * On success path->dentry is known positive. 12459875cf80SDavid Howells * 12469875cf80SDavid Howells * Serialization is taken care of in namespace.c 12479875cf80SDavid Howells */ 1248756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 12499875cf80SDavid Howells { 12508aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 1251d41efb52SAl Viro unsigned flags; 12529875cf80SDavid Howells bool need_mntput = false; 12538aef1884SAl Viro int ret = 0; 12549875cf80SDavid Howells 12559875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 12569875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 12579875cf80SDavid Howells * the components of that value change under us */ 12582fa6b1e0SAl Viro while (flags = smp_load_acquire(&path->dentry->d_flags), 1259d41efb52SAl Viro unlikely(flags & DCACHE_MANAGED_DENTRY)) { 1260cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1261cc53ce53SDavid Howells * being held. */ 1262d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1263cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1264cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1265fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1266508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1267cc53ce53SDavid Howells if (ret < 0) 12688aef1884SAl Viro break; 1269cc53ce53SDavid Howells } 1270cc53ce53SDavid Howells 12719875cf80SDavid Howells /* Transit to a mounted filesystem. */ 1272d41efb52SAl Viro if (flags & DCACHE_MOUNTED) { 12739875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12749875cf80SDavid Howells if (mounted) { 12759875cf80SDavid Howells dput(path->dentry); 12769875cf80SDavid Howells if (need_mntput) 1277463ffb2eSAl Viro mntput(path->mnt); 1278463ffb2eSAl Viro path->mnt = mounted; 1279463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12809875cf80SDavid Howells need_mntput = true; 12819875cf80SDavid Howells continue; 1282463ffb2eSAl Viro } 1283463ffb2eSAl Viro 12849875cf80SDavid Howells /* Something is mounted on this dentry in another 12859875cf80SDavid Howells * namespace and/or whatever was mounted there in this 128648a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 128748a066e7SAl Viro * get it */ 12881da177e4SLinus Torvalds } 12899875cf80SDavid Howells 12909875cf80SDavid Howells /* Handle an automount point */ 1291d41efb52SAl Viro if (flags & DCACHE_NEED_AUTOMOUNT) { 12921c9f5e06SAl Viro ret = follow_automount(path, &nd->total_link_count, 12931c9f5e06SAl Viro nd->flags); 12949875cf80SDavid Howells if (ret < 0) 12958aef1884SAl Viro break; 12969875cf80SDavid Howells continue; 12979875cf80SDavid Howells } 12989875cf80SDavid Howells 12999875cf80SDavid Howells /* We didn't change the current path point */ 13009875cf80SDavid Howells break; 13019875cf80SDavid Howells } 13028aef1884SAl Viro 130372ba2929SAleksa Sarai if (need_mntput) { 130472ba2929SAleksa Sarai if (path->mnt == mnt) 13058aef1884SAl Viro mntput(path->mnt); 130672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 130772ba2929SAleksa Sarai ret = -EXDEV; 130872ba2929SAleksa Sarai else 13098402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 131072ba2929SAleksa Sarai } 1311d41efb52SAl Viro if (ret == -EISDIR || !ret) 1312d41efb52SAl Viro ret = 1; 1313d41efb52SAl Viro if (ret > 0 && unlikely(d_flags_negative(flags))) 1314d41efb52SAl Viro ret = -ENOENT; 13158402752eSAl Viro if (unlikely(ret < 0)) 13168402752eSAl Viro path_put_conditional(path, nd); 13178402752eSAl Viro return ret; 13181da177e4SLinus Torvalds } 13191da177e4SLinus Torvalds 1320cc53ce53SDavid Howells int follow_down_one(struct path *path) 13211da177e4SLinus Torvalds { 13221da177e4SLinus Torvalds struct vfsmount *mounted; 13231da177e4SLinus Torvalds 13241c755af4SAl Viro mounted = lookup_mnt(path); 13251da177e4SLinus Torvalds if (mounted) { 13269393bd07SAl Viro dput(path->dentry); 13279393bd07SAl Viro mntput(path->mnt); 13289393bd07SAl Viro path->mnt = mounted; 13299393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 13301da177e4SLinus Torvalds return 1; 13311da177e4SLinus Torvalds } 13321da177e4SLinus Torvalds return 0; 13331da177e4SLinus Torvalds } 13344d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 13351da177e4SLinus Torvalds 1336fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 133762a7375eSIan Kent { 1338fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1339fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 134062a7375eSIan Kent } 134162a7375eSIan Kent 13429875cf80SDavid Howells /* 1343287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1344287548e4SAl Viro * we meet a managed dentry that would need blocking. 13459875cf80SDavid Howells */ 13469875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1347254cf582SAl Viro struct inode **inode, unsigned *seqp) 13489875cf80SDavid Howells { 134962a7375eSIan Kent for (;;) { 1350c7105365SAl Viro struct mount *mounted; 135162a7375eSIan Kent /* 135262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 135362a7375eSIan Kent * that wants to block transit. 135462a7375eSIan Kent */ 1355fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1356b8faf035SNeilBrown case -ECHILD: 1357b8faf035SNeilBrown default: 1358ab90911fSDavid Howells return false; 1359b8faf035SNeilBrown case -EISDIR: 1360b8faf035SNeilBrown return true; 1361b8faf035SNeilBrown case 0: 1362b8faf035SNeilBrown break; 1363b8faf035SNeilBrown } 136462a7375eSIan Kent 136562a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1366b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 136762a7375eSIan Kent 1368474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 13699875cf80SDavid Howells if (!mounted) 13709875cf80SDavid Howells break; 137172ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 137272ba2929SAleksa Sarai return false; 1373c7105365SAl Viro path->mnt = &mounted->mnt; 1374c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1375a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1376254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 137759430262SLinus Torvalds /* 137859430262SLinus Torvalds * Update the inode too. We don't need to re-check the 137959430262SLinus Torvalds * dentry sequence number here after this d_inode read, 138059430262SLinus Torvalds * because a mount-point is always pinned. 138159430262SLinus Torvalds */ 138259430262SLinus Torvalds *inode = path->dentry->d_inode; 13839875cf80SDavid Howells } 1384f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1385b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1386287548e4SAl Viro } 1387287548e4SAl Viro 1388db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1389db3c9adeSAl Viro struct path *path, struct inode **inode, 1390db3c9adeSAl Viro unsigned int *seqp) 1391bd7c4b50SAl Viro { 1392db3c9adeSAl Viro int ret; 1393bd7c4b50SAl Viro 1394db3c9adeSAl Viro path->mnt = nd->path.mnt; 1395db3c9adeSAl Viro path->dentry = dentry; 1396c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1397c153007bSAl Viro unsigned int seq = *seqp; 1398c153007bSAl Viro if (unlikely(!*inode)) 1399c153007bSAl Viro return -ENOENT; 1400c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1401c153007bSAl Viro return 1; 1402c153007bSAl Viro if (unlazy_child(nd, dentry, seq)) 1403c153007bSAl Viro return -ECHILD; 1404c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1405c153007bSAl Viro path->mnt = nd->path.mnt; 1406c153007bSAl Viro path->dentry = dentry; 1407c153007bSAl Viro } 1408db3c9adeSAl Viro ret = follow_managed(path, nd); 1409bd7c4b50SAl Viro if (likely(ret >= 0)) { 1410bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1411bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1412bd7c4b50SAl Viro } 1413bd7c4b50SAl Viro return ret; 1414bd7c4b50SAl Viro } 1415bd7c4b50SAl Viro 141631e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 141731e6b01fSNick Piggin { 14184023bfc9SAl Viro struct inode *inode = nd->inode; 141931e6b01fSNick Piggin 142031e6b01fSNick Piggin while (1) { 1421adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1422adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1423adb21d2bSAleksa Sarai return -ECHILD; 142431e6b01fSNick Piggin break; 1425adb21d2bSAleksa Sarai } 142631e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 142731e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 142831e6b01fSNick Piggin struct dentry *parent = old->d_parent; 142931e6b01fSNick Piggin unsigned seq; 143031e6b01fSNick Piggin 14314023bfc9SAl Viro inode = parent->d_inode; 143231e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1433aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1434aed434adSAl Viro return -ECHILD; 143531e6b01fSNick Piggin nd->path.dentry = parent; 143631e6b01fSNick Piggin nd->seq = seq; 1437397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 14382b98149cSAleksa Sarai return -ECHILD; 143931e6b01fSNick Piggin break; 1440aed434adSAl Viro } else { 1441aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1442aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1443aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1444aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1445aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1446aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1447aed434adSAl Viro return -ECHILD; 1448aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 144931e6b01fSNick Piggin break; 145072ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 145172ba2929SAleksa Sarai return -ECHILD; 1452aed434adSAl Viro /* we know that mountpoint was pinned */ 1453aed434adSAl Viro nd->path.dentry = mountpoint; 1454aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1455aed434adSAl Viro inode = inode2; 1456aed434adSAl Viro nd->seq = seq; 145731e6b01fSNick Piggin } 1458aed434adSAl Viro } 1459aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1460b37199e6SAl Viro struct mount *mounted; 1461b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1462aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1463aed434adSAl Viro return -ECHILD; 1464b37199e6SAl Viro if (!mounted) 1465b37199e6SAl Viro break; 146672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 146772ba2929SAleksa Sarai return -ECHILD; 1468b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1469b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 14704023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1471b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1472b37199e6SAl Viro } 14734023bfc9SAl Viro nd->inode = inode; 147431e6b01fSNick Piggin return 0; 147531e6b01fSNick Piggin } 147631e6b01fSNick Piggin 14779875cf80SDavid Howells /* 1478cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1479cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1480cc53ce53SDavid Howells * caller is permitted to proceed or not. 1481cc53ce53SDavid Howells */ 14827cc90cc3SAl Viro int follow_down(struct path *path) 1483cc53ce53SDavid Howells { 1484cc53ce53SDavid Howells unsigned managed; 1485cc53ce53SDavid Howells int ret; 1486cc53ce53SDavid Howells 14876aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1488cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1489cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1490cc53ce53SDavid Howells * being held. 1491cc53ce53SDavid Howells * 1492cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1493cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1494cc53ce53SDavid Howells * other than its daemon the right to mount on its 1495cc53ce53SDavid Howells * superstructure. 1496cc53ce53SDavid Howells * 1497cc53ce53SDavid Howells * The filesystem may sleep at this point. 1498cc53ce53SDavid Howells */ 1499cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1500cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1501cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1502fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1503cc53ce53SDavid Howells if (ret < 0) 1504cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1505cc53ce53SDavid Howells } 1506cc53ce53SDavid Howells 1507cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1508cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1509cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1510cc53ce53SDavid Howells if (!mounted) 1511cc53ce53SDavid Howells break; 1512cc53ce53SDavid Howells dput(path->dentry); 1513cc53ce53SDavid Howells mntput(path->mnt); 1514cc53ce53SDavid Howells path->mnt = mounted; 1515cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1516cc53ce53SDavid Howells continue; 1517cc53ce53SDavid Howells } 1518cc53ce53SDavid Howells 1519cc53ce53SDavid Howells /* Don't handle automount points here */ 1520cc53ce53SDavid Howells break; 1521cc53ce53SDavid Howells } 1522cc53ce53SDavid Howells return 0; 1523cc53ce53SDavid Howells } 15244d359507SAl Viro EXPORT_SYMBOL(follow_down); 1525cc53ce53SDavid Howells 1526cc53ce53SDavid Howells /* 15279875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 15289875cf80SDavid Howells */ 15299875cf80SDavid Howells static void follow_mount(struct path *path) 15309875cf80SDavid Howells { 15319875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 15329875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 15339875cf80SDavid Howells if (!mounted) 15349875cf80SDavid Howells break; 15359875cf80SDavid Howells dput(path->dentry); 15369875cf80SDavid Howells mntput(path->mnt); 15379875cf80SDavid Howells path->mnt = mounted; 15389875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 15399875cf80SDavid Howells } 15409875cf80SDavid Howells } 15419875cf80SDavid Howells 1542eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1543eedf265aSEric W. Biederman { 1544eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1545eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1546eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1547eedf265aSEric W. Biederman dput(old); 1548eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1549eedf265aSEric W. Biederman return -ENOENT; 1550eedf265aSEric W. Biederman return 0; 1551eedf265aSEric W. Biederman } 1552eedf265aSEric W. Biederman 1553397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 15541da177e4SLinus Torvalds { 15551da177e4SLinus Torvalds while (1) { 1556adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1557adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1558adb21d2bSAleksa Sarai return -EXDEV; 15591da177e4SLinus Torvalds break; 1560adb21d2bSAleksa Sarai } 15614ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1562eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1563eedf265aSEric W. Biederman if (ret) 1564eedf265aSEric W. Biederman return ret; 15651da177e4SLinus Torvalds break; 15661da177e4SLinus Torvalds } 15673088dd70SAl Viro if (!follow_up(&nd->path)) 15681da177e4SLinus Torvalds break; 156972ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 157072ba2929SAleksa Sarai return -EXDEV; 15711da177e4SLinus Torvalds } 157279ed0226SAl Viro follow_mount(&nd->path); 157331e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1574397d425dSEric W. Biederman return 0; 15751da177e4SLinus Torvalds } 15761da177e4SLinus Torvalds 15771da177e4SLinus Torvalds /* 1578f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1579f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1580baa03890SNick Piggin */ 1581e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1582e3c13928SAl Viro struct dentry *dir, 15836c51e513SAl Viro unsigned int flags) 1584baa03890SNick Piggin { 1585a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1586bad61189SMiklos Szeredi if (dentry) { 1587a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1588bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 158974ff0ffcSAl Viro if (!error) 15905542aa2fSEric W. Biederman d_invalidate(dentry); 1591bad61189SMiklos Szeredi dput(dentry); 159274ff0ffcSAl Viro return ERR_PTR(error); 1593bad61189SMiklos Szeredi } 1594bad61189SMiklos Szeredi } 1595baa03890SNick Piggin return dentry; 1596baa03890SNick Piggin } 1597baa03890SNick Piggin 1598baa03890SNick Piggin /* 1599a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1600a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1601a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1602a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1603a03ece5fSAl Viro * at all. 160444396f4bSJosef Bacik */ 1605a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1606a03ece5fSAl Viro struct dentry *base, unsigned int flags) 160744396f4bSJosef Bacik { 1608a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 160944396f4bSJosef Bacik struct dentry *old; 1610a03ece5fSAl Viro struct inode *dir = base->d_inode; 1611a03ece5fSAl Viro 1612a03ece5fSAl Viro if (dentry) 1613a03ece5fSAl Viro return dentry; 161444396f4bSJosef Bacik 161544396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1616a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 161744396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1618a03ece5fSAl Viro 1619a03ece5fSAl Viro dentry = d_alloc(base, name); 1620a03ece5fSAl Viro if (unlikely(!dentry)) 1621a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 162244396f4bSJosef Bacik 162372bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 162444396f4bSJosef Bacik if (unlikely(old)) { 162544396f4bSJosef Bacik dput(dentry); 162644396f4bSJosef Bacik dentry = old; 162744396f4bSJosef Bacik } 162844396f4bSJosef Bacik return dentry; 162944396f4bSJosef Bacik } 163044396f4bSJosef Bacik 163120e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 163220e34357SAl Viro struct inode **inode, 1633254cf582SAl Viro unsigned *seqp) 16341da177e4SLinus Torvalds { 163531e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16365a18fff2SAl Viro int status = 1; 16379875cf80SDavid Howells 16383cac260aSAl Viro /* 1639b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16405d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16415d0f49c1SAl Viro * going to fall back to non-racy lookup. 1642b04f784eSNick Piggin */ 164331e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 164431e6b01fSNick Piggin unsigned seq; 1645da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 16465d0f49c1SAl Viro if (unlikely(!dentry)) { 16474675ac39SAl Viro if (unlazy_walk(nd)) 164820e34357SAl Viro return ERR_PTR(-ECHILD); 164920e34357SAl Viro return NULL; 16505d0f49c1SAl Viro } 16515a18fff2SAl Viro 165212f8ad4bSLinus Torvalds /* 165312f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 165412f8ad4bSLinus Torvalds * the dentry name information from lookup. 165512f8ad4bSLinus Torvalds */ 165663afdfc7SDavid Howells *inode = d_backing_inode(dentry); 16575d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 165820e34357SAl Viro return ERR_PTR(-ECHILD); 165912f8ad4bSLinus Torvalds 166012f8ad4bSLinus Torvalds /* 166112f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 166212f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 166312f8ad4bSLinus Torvalds * 166412f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 166512f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 166612f8ad4bSLinus Torvalds */ 16675d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 166820e34357SAl Viro return ERR_PTR(-ECHILD); 16695a18fff2SAl Viro 1670254cf582SAl Viro *seqp = seq; 16714ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1672c153007bSAl Viro if (likely(status > 0)) 167320e34357SAl Viro return dentry; 16744675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 167520e34357SAl Viro return ERR_PTR(-ECHILD); 1676209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1677209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1678209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16795a18fff2SAl Viro } else { 1680e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 168181e6f520SAl Viro if (unlikely(!dentry)) 168220e34357SAl Viro return NULL; 16834ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16845d0f49c1SAl Viro } 16855a18fff2SAl Viro if (unlikely(status <= 0)) { 1686e9742b53SAl Viro if (!status) 16875d0f49c1SAl Viro d_invalidate(dentry); 16885a18fff2SAl Viro dput(dentry); 168920e34357SAl Viro return ERR_PTR(status); 16905a18fff2SAl Viro } 169120e34357SAl Viro return dentry; 1692697f514dSMiklos Szeredi } 1693697f514dSMiklos Szeredi 1694697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 169588d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1696e3c13928SAl Viro struct dentry *dir, 1697e3c13928SAl Viro unsigned int flags) 1698697f514dSMiklos Szeredi { 169988d8331aSAl Viro struct dentry *dentry, *old; 17001936386eSAl Viro struct inode *inode = dir->d_inode; 1701d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 17021936386eSAl Viro 17031936386eSAl Viro /* Don't go there if it's already dead */ 170494bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 170588d8331aSAl Viro return ERR_PTR(-ENOENT); 170694bdd655SAl Viro again: 1707d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 170894bdd655SAl Viro if (IS_ERR(dentry)) 170988d8331aSAl Viro return dentry; 171094bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1711949a852eSAl Viro int error = d_revalidate(dentry, flags); 1712949a852eSAl Viro if (unlikely(error <= 0)) { 171394bdd655SAl Viro if (!error) { 1714949a852eSAl Viro d_invalidate(dentry); 1715949a852eSAl Viro dput(dentry); 171694bdd655SAl Viro goto again; 171794bdd655SAl Viro } 171894bdd655SAl Viro dput(dentry); 1719949a852eSAl Viro dentry = ERR_PTR(error); 1720949a852eSAl Viro } 172194bdd655SAl Viro } else { 17221936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 172385c7f810SAl Viro d_lookup_done(dentry); 17241936386eSAl Viro if (unlikely(old)) { 17251936386eSAl Viro dput(dentry); 17261936386eSAl Viro dentry = old; 1727949a852eSAl Viro } 1728949a852eSAl Viro } 1729e3c13928SAl Viro return dentry; 17301da177e4SLinus Torvalds } 17311da177e4SLinus Torvalds 173288d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 173388d8331aSAl Viro struct dentry *dir, 173488d8331aSAl Viro unsigned int flags) 173588d8331aSAl Viro { 173688d8331aSAl Viro struct inode *inode = dir->d_inode; 173788d8331aSAl Viro struct dentry *res; 173888d8331aSAl Viro inode_lock_shared(inode); 173988d8331aSAl Viro res = __lookup_slow(name, dir, flags); 174088d8331aSAl Viro inode_unlock_shared(inode); 174188d8331aSAl Viro return res; 174288d8331aSAl Viro } 174388d8331aSAl Viro 174452094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 174552094c8aSAl Viro { 174652094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17474ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 174852094c8aSAl Viro if (err != -ECHILD) 174952094c8aSAl Viro return err; 17504675ac39SAl Viro if (unlazy_walk(nd)) 175152094c8aSAl Viro return -ECHILD; 175252094c8aSAl Viro } 17534ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 175452094c8aSAl Viro } 175552094c8aSAl Viro 17569856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 17579856fa1bSAl Viro { 17589856fa1bSAl Viro if (type == LAST_DOTDOT) { 1759740a1678SAleksa Sarai int error = 0; 1760740a1678SAleksa Sarai 1761740a1678SAleksa Sarai if (!nd->root.mnt) { 1762740a1678SAleksa Sarai error = set_root(nd); 1763740a1678SAleksa Sarai if (error) 1764740a1678SAleksa Sarai return error; 1765740a1678SAleksa Sarai } 1766ab87f9a5SAleksa Sarai if (nd->flags & LOOKUP_RCU) 1767ab87f9a5SAleksa Sarai error = follow_dotdot_rcu(nd); 1768ab87f9a5SAleksa Sarai else 1769ab87f9a5SAleksa Sarai error = follow_dotdot(nd); 1770ab87f9a5SAleksa Sarai if (error) 1771ab87f9a5SAleksa Sarai return error; 1772ab87f9a5SAleksa Sarai 1773ab87f9a5SAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1774ab87f9a5SAleksa Sarai /* 1775ab87f9a5SAleksa Sarai * If there was a racing rename or mount along our 1776ab87f9a5SAleksa Sarai * path, then we can't be sure that ".." hasn't jumped 1777ab87f9a5SAleksa Sarai * above nd->root (and so userspace should retry or use 1778ab87f9a5SAleksa Sarai * some fallback). 1779ab87f9a5SAleksa Sarai */ 1780ab87f9a5SAleksa Sarai smp_rmb(); 1781ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 1782ab87f9a5SAleksa Sarai return -EAGAIN; 1783ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 1784ab87f9a5SAleksa Sarai return -EAGAIN; 1785ab87f9a5SAleksa Sarai } 17869856fa1bSAl Viro } 17879856fa1bSAl Viro return 0; 17889856fa1bSAl Viro } 17899856fa1bSAl Viro 1790181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1791181548c0SAl Viro struct inode *inode, unsigned seq) 1792d63ff28fSAl Viro { 1793626de996SAl Viro int error; 17941cf2665bSAl Viro struct saved *last; 1795756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1796626de996SAl Viro path_to_nameidata(link, nd); 1797626de996SAl Viro return -ELOOP; 1798626de996SAl Viro } 1799bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1800cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1801cd179f44SAl Viro mntget(link->mnt); 18027973387aSAl Viro } 1803626de996SAl Viro error = nd_alloc_stack(nd); 1804626de996SAl Viro if (unlikely(error)) { 1805bc40aee0SAl Viro if (error == -ECHILD) { 1806ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1807ad1633a1SAl Viro drop_links(nd); 1808ad1633a1SAl Viro nd->depth = 0; 1809ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1810ad1633a1SAl Viro nd->path.mnt = NULL; 1811ad1633a1SAl Viro nd->path.dentry = NULL; 1812ad1633a1SAl Viro rcu_read_unlock(); 18134675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1814bc40aee0SAl Viro error = nd_alloc_stack(nd); 1815bc40aee0SAl Viro } 1816bc40aee0SAl Viro if (error) { 1817cd179f44SAl Viro path_put(link); 1818626de996SAl Viro return error; 1819626de996SAl Viro } 1820bc40aee0SAl Viro } 1821626de996SAl Viro 1822ab104923SAl Viro last = nd->stack + nd->depth++; 18231cf2665bSAl Viro last->link = *link; 1824fceef393SAl Viro clear_delayed_call(&last->done); 1825fceef393SAl Viro nd->link_inode = inode; 18260450b2d1SAl Viro last->seq = seq; 1827d63ff28fSAl Viro return 1; 1828d63ff28fSAl Viro } 1829d63ff28fSAl Viro 18308f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 183131d66bcdSAl Viro 18323ddcd056SLinus Torvalds /* 18333ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18343ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18353ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18363ddcd056SLinus Torvalds * for the common case. 18373ddcd056SLinus Torvalds */ 18388f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 18398f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 18403ddcd056SLinus Torvalds { 18418f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 18428f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 18438f64fb1cSAl Viro /* not a symlink or should not follow */ 18448f64fb1cSAl Viro path_to_nameidata(path, nd); 18458f64fb1cSAl Viro nd->inode = inode; 18468f64fb1cSAl Viro nd->seq = seq; 1847d63ff28fSAl Viro return 0; 18488f64fb1cSAl Viro } 1849a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1850a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 18518f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1852a7f77542SAl Viro return -ECHILD; 1853a7f77542SAl Viro } 18548f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 18553ddcd056SLinus Torvalds } 18563ddcd056SLinus Torvalds 18574693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1858ce57dfc1SAl Viro { 1859caa85634SAl Viro struct path path; 1860db3c9adeSAl Viro struct dentry *dentry; 1861ce57dfc1SAl Viro struct inode *inode; 1862254cf582SAl Viro unsigned seq; 1863ce57dfc1SAl Viro int err; 1864ce57dfc1SAl Viro /* 1865ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1866ce57dfc1SAl Viro * to be able to know about the current root directory and 1867ce57dfc1SAl Viro * parent relationships. 1868ce57dfc1SAl Viro */ 18694693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18701c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18714693a547SAl Viro put_link(nd); 187256676ec3SAl Viro err = handle_dots(nd, nd->last_type); 18734693a547SAl Viro return err; 18744693a547SAl Viro } 187520e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 187620e34357SAl Viro if (IS_ERR(dentry)) 187720e34357SAl Viro return PTR_ERR(dentry); 187820e34357SAl Viro if (unlikely(!dentry)) { 1879db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1880db3c9adeSAl Viro if (IS_ERR(dentry)) 1881db3c9adeSAl Viro return PTR_ERR(dentry); 188220e34357SAl Viro } 188356676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 188456676ec3SAl Viro put_link(nd); 18857500c38aSAl Viro 1886db3c9adeSAl Viro err = handle_mounts(nd, dentry, &path, &inode, &seq); 1887e3c13928SAl Viro if (unlikely(err < 0)) 1888f0a9ba70SAl Viro return err; 18898f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1890ce57dfc1SAl Viro } 1891ce57dfc1SAl Viro 18921da177e4SLinus Torvalds /* 1893bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1894bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1895bfcfaa77SLinus Torvalds * 1896bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1897bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1898bfcfaa77SLinus Torvalds * fast. 1899bfcfaa77SLinus Torvalds * 1900bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1901bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1902bfcfaa77SLinus Torvalds * crossing operation. 1903bfcfaa77SLinus Torvalds * 1904bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1905bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1906bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1907bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1908bfcfaa77SLinus Torvalds */ 1909bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1910bfcfaa77SLinus Torvalds 1911f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1912bfcfaa77SLinus Torvalds 1913468a9428SGeorge Spelvin #ifdef HASH_MIX 1914bfcfaa77SLinus Torvalds 1915468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1916468a9428SGeorge Spelvin 1917468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 19182a18da7aSGeorge Spelvin /* 19192a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 19202a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 19212a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 19222a18da7aSGeorge Spelvin * and no temporaries. 19232a18da7aSGeorge Spelvin * 19242a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 19252a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 19262a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 19272a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 19282a18da7aSGeorge Spelvin * 19292a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 19302a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 19312a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 19322a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 19332a18da7aSGeorge Spelvin * 19342a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 19352a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 19362a18da7aSGeorge Spelvin * toggling any given output bit. 19372a18da7aSGeorge Spelvin * 19382a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 19392a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19402a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 19412a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 19422a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 19432a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 19442a18da7aSGeorge Spelvin * Perfect: 8192 258048 19452a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 19462a18da7aSGeorge Spelvin */ 19472a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19482a18da7aSGeorge Spelvin ( x ^= (a), \ 19492a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 19502a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 19512a18da7aSGeorge Spelvin y *= 9 ) 1952bfcfaa77SLinus Torvalds 19530fed3ac8SGeorge Spelvin /* 19542a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 19552a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 19562a18da7aSGeorge Spelvin * work done before the hash value is used. 19570fed3ac8SGeorge Spelvin */ 19582a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19590fed3ac8SGeorge Spelvin { 19602a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 19612a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19622a18da7aSGeorge Spelvin return y >> 32; 19630fed3ac8SGeorge Spelvin } 19640fed3ac8SGeorge Spelvin 1965bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1966bfcfaa77SLinus Torvalds 19672a18da7aSGeorge Spelvin /* 19682a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19692a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19702a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19712a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19722a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19732a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19742a18da7aSGeorge Spelvin * Perfect: 2048 31744 19752a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19762a18da7aSGeorge Spelvin */ 19772a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19782a18da7aSGeorge Spelvin ( x ^= (a), \ 19792a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19802a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19812a18da7aSGeorge Spelvin y *= 9 ) 1982bfcfaa77SLinus Torvalds 19832a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19840fed3ac8SGeorge Spelvin { 19852a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19862a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19870fed3ac8SGeorge Spelvin } 19880fed3ac8SGeorge Spelvin 1989bfcfaa77SLinus Torvalds #endif 1990bfcfaa77SLinus Torvalds 19912a18da7aSGeorge Spelvin /* 19922a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19932a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19942a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19952a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19962a18da7aSGeorge Spelvin * finds the delimiter after the name. 19972a18da7aSGeorge Spelvin */ 19988387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1999bfcfaa77SLinus Torvalds { 20008387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2001bfcfaa77SLinus Torvalds 2002bfcfaa77SLinus Torvalds for (;;) { 2003fcfd2fbfSGeorge Spelvin if (!len) 2004fcfd2fbfSGeorge Spelvin goto done; 2005e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2006bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2007bfcfaa77SLinus Torvalds break; 20082a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2009bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2010bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2011bfcfaa77SLinus Torvalds } 20122a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2013bfcfaa77SLinus Torvalds done: 20142a18da7aSGeorge Spelvin return fold_hash(x, y); 2015bfcfaa77SLinus Torvalds } 2016bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2017bfcfaa77SLinus Torvalds 2018fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20198387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2020fcfd2fbfSGeorge Spelvin { 20218387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 20228387ff25SLinus Torvalds unsigned long adata, mask, len; 2023fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2024fcfd2fbfSGeorge Spelvin 20258387ff25SLinus Torvalds len = 0; 20268387ff25SLinus Torvalds goto inside; 20278387ff25SLinus Torvalds 2028fcfd2fbfSGeorge Spelvin do { 20292a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2030fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 20318387ff25SLinus Torvalds inside: 2032fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2033fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2034fcfd2fbfSGeorge Spelvin 2035fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2036fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 20372a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2038fcfd2fbfSGeorge Spelvin 20392a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2040fcfd2fbfSGeorge Spelvin } 2041fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2042fcfd2fbfSGeorge Spelvin 2043bfcfaa77SLinus Torvalds /* 2044bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2045d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2046bfcfaa77SLinus Torvalds */ 20478387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2048bfcfaa77SLinus Torvalds { 20498387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 20508387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 205136126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2052bfcfaa77SLinus Torvalds 20538387ff25SLinus Torvalds len = 0; 20548387ff25SLinus Torvalds goto inside; 20558387ff25SLinus Torvalds 2056bfcfaa77SLinus Torvalds do { 20572a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2058bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 20598387ff25SLinus Torvalds inside: 2060e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 206136126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 206236126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2063bfcfaa77SLinus Torvalds 206436126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 206536126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 206636126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20672a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 206836126f8fSLinus Torvalds 20692a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2070bfcfaa77SLinus Torvalds } 2071bfcfaa77SLinus Torvalds 20722a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2073bfcfaa77SLinus Torvalds 2074fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20758387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20760145acc2SLinus Torvalds { 20778387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20780145acc2SLinus Torvalds while (len--) 2079fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20800145acc2SLinus Torvalds return end_name_hash(hash); 20810145acc2SLinus Torvalds } 2082ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20830145acc2SLinus Torvalds 2084fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20858387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2086fcfd2fbfSGeorge Spelvin { 20878387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2088fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2089fcfd2fbfSGeorge Spelvin 2090fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2091e0ab7af9SGeorge Spelvin while (c) { 2092fcfd2fbfSGeorge Spelvin len++; 2093fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2094fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2095e0ab7af9SGeorge Spelvin } 2096fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2097fcfd2fbfSGeorge Spelvin } 2098f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2099fcfd2fbfSGeorge Spelvin 21003ddcd056SLinus Torvalds /* 2101200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2102200e9ef7SLinus Torvalds * one character. 2103200e9ef7SLinus Torvalds */ 21048387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2105200e9ef7SLinus Torvalds { 21068387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2107200e9ef7SLinus Torvalds unsigned long len = 0, c; 2108200e9ef7SLinus Torvalds 2109200e9ef7SLinus Torvalds c = (unsigned char)*name; 2110200e9ef7SLinus Torvalds do { 2111200e9ef7SLinus Torvalds len++; 2112200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2113200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2114200e9ef7SLinus Torvalds } while (c && c != '/'); 2115d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2116200e9ef7SLinus Torvalds } 2117200e9ef7SLinus Torvalds 2118bfcfaa77SLinus Torvalds #endif 2119bfcfaa77SLinus Torvalds 2120200e9ef7SLinus Torvalds /* 21211da177e4SLinus Torvalds * Name resolution. 2122ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2123ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 21241da177e4SLinus Torvalds * 2125ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2126ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 21271da177e4SLinus Torvalds */ 21286de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 21291da177e4SLinus Torvalds { 21301da177e4SLinus Torvalds int err; 21311da177e4SLinus Torvalds 21329b5858e9SAl Viro if (IS_ERR(name)) 21339b5858e9SAl Viro return PTR_ERR(name); 21341da177e4SLinus Torvalds while (*name=='/') 21351da177e4SLinus Torvalds name++; 21361da177e4SLinus Torvalds if (!*name) 21379e18f10aSAl Viro return 0; 21381da177e4SLinus Torvalds 21391da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 21401da177e4SLinus Torvalds for(;;) { 2141d6bb3e90SLinus Torvalds u64 hash_len; 2142fe479a58SAl Viro int type; 21431da177e4SLinus Torvalds 214452094c8aSAl Viro err = may_lookup(nd); 21451da177e4SLinus Torvalds if (err) 21463595e234SAl Viro return err; 21471da177e4SLinus Torvalds 21488387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 21491da177e4SLinus Torvalds 2150fe479a58SAl Viro type = LAST_NORM; 2151d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2152fe479a58SAl Viro case 2: 2153200e9ef7SLinus Torvalds if (name[1] == '.') { 2154fe479a58SAl Viro type = LAST_DOTDOT; 215516c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 215616c2cd71SAl Viro } 2157fe479a58SAl Viro break; 2158fe479a58SAl Viro case 1: 2159fe479a58SAl Viro type = LAST_DOT; 2160fe479a58SAl Viro } 21615a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21625a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 216316c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21645a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2165a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2166da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21675a202bcdSAl Viro if (err < 0) 21683595e234SAl Viro return err; 2169d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2170d6bb3e90SLinus Torvalds name = this.name; 21715a202bcdSAl Viro } 21725a202bcdSAl Viro } 2173fe479a58SAl Viro 2174d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2175d6bb3e90SLinus Torvalds nd->last.name = name; 21765f4a6a69SAl Viro nd->last_type = type; 21775f4a6a69SAl Viro 2178d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2179d6bb3e90SLinus Torvalds if (!*name) 2180bdf6cbf1SAl Viro goto OK; 2181200e9ef7SLinus Torvalds /* 2182200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2183200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2184200e9ef7SLinus Torvalds */ 2185200e9ef7SLinus Torvalds do { 2186d6bb3e90SLinus Torvalds name++; 2187d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21888620c238SAl Viro if (unlikely(!*name)) { 21898620c238SAl Viro OK: 2190368ee9baSAl Viro /* pathname body, done */ 21918620c238SAl Viro if (!nd->depth) 21928620c238SAl Viro return 0; 21938620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2194368ee9baSAl Viro /* trailing symlink, done */ 21958620c238SAl Viro if (!name) 21968620c238SAl Viro return 0; 21978620c238SAl Viro /* last component of nested symlink */ 21988f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 21991c4ff1a8SAl Viro } else { 22001c4ff1a8SAl Viro /* not the last component */ 22018f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 22028620c238SAl Viro } 2203ce57dfc1SAl Viro if (err < 0) 22043595e234SAl Viro return err; 2205fe479a58SAl Viro 2206ce57dfc1SAl Viro if (err) { 2207626de996SAl Viro const char *s = get_link(nd); 22085a460275SAl Viro 2209a1c83681SViresh Kumar if (IS_ERR(s)) 22103595e234SAl Viro return PTR_ERR(s); 2211172a39a0SAl Viro err = 0; 221212b09578SAl Viro if (unlikely(!s)) { 221312b09578SAl Viro /* jumped */ 2214b9ff4429SAl Viro put_link(nd); 221512b09578SAl Viro } else { 2216071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 221732cd7468SAl Viro name = s; 22189e18f10aSAl Viro continue; 221948c8b0c5SAl Viro } 222031e6b01fSNick Piggin } 222197242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 222297242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 22234675ac39SAl Viro if (unlazy_walk(nd)) 222497242f99SAl Viro return -ECHILD; 222597242f99SAl Viro } 22263595e234SAl Viro return -ENOTDIR; 22275f4a6a69SAl Viro } 2228ce57dfc1SAl Viro } 222997242f99SAl Viro } 22301da177e4SLinus Torvalds 2231edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2232c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 223331e6b01fSNick Piggin { 2234740a1678SAleksa Sarai int error; 2235c8a53ee5SAl Viro const char *s = nd->name->name; 223631e6b01fSNick Piggin 2237c0eb027eSLinus Torvalds if (!*s) 2238c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2239edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2240edc2b1daSAl Viro rcu_read_lock(); 2241c0eb027eSLinus Torvalds 224231e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2243980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 224431e6b01fSNick Piggin nd->depth = 0; 2245ab87f9a5SAleksa Sarai 2246ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2247ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2248ab87f9a5SAleksa Sarai smp_rmb(); 2249ab87f9a5SAleksa Sarai 22505b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2251b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2252b18825a7SDavid Howells struct inode *inode = root->d_inode; 225393893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2254368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 22555b6ca027SAl Viro nd->path = nd->root; 22565b6ca027SAl Viro nd->inode = inode; 22575b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2258ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22598f47a016SAl Viro nd->root_seq = nd->seq; 22605b6ca027SAl Viro } else { 22615b6ca027SAl Viro path_get(&nd->path); 22625b6ca027SAl Viro } 2263368ee9baSAl Viro return s; 22645b6ca027SAl Viro } 22655b6ca027SAl Viro 226631e6b01fSNick Piggin nd->root.mnt = NULL; 2267248fb5b9SAl Viro nd->path.mnt = NULL; 2268248fb5b9SAl Viro nd->path.dentry = NULL; 226931e6b01fSNick Piggin 22708db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 22718db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2272740a1678SAleksa Sarai error = nd_jump_root(nd); 2273740a1678SAleksa Sarai if (unlikely(error)) 2274740a1678SAleksa Sarai return ERR_PTR(error); 2275ef55d917SAl Viro return s; 22768db52c7eSAleksa Sarai } 22778db52c7eSAleksa Sarai 22788db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 22798db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2280e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 228131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2282c28cc364SNick Piggin unsigned seq; 228331e6b01fSNick Piggin 2284c28cc364SNick Piggin do { 2285c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 228631e6b01fSNick Piggin nd->path = fs->pwd; 2287ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2288c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2289c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2290e41f7d4eSAl Viro } else { 2291e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2292ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2293e41f7d4eSAl Viro } 229431e6b01fSNick Piggin } else { 2295582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2296c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 229731e6b01fSNick Piggin struct dentry *dentry; 229831e6b01fSNick Piggin 22992903ff01SAl Viro if (!f.file) 2300368ee9baSAl Viro return ERR_PTR(-EBADF); 230131e6b01fSNick Piggin 23022903ff01SAl Viro dentry = f.file->f_path.dentry; 230331e6b01fSNick Piggin 2304edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 23052903ff01SAl Viro fdput(f); 2306368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2307f52e0c11SAl Viro } 23082903ff01SAl Viro 23092903ff01SAl Viro nd->path = f.file->f_path; 2310e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 231134a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 231234a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23135590ff0dSUlrich Drepper } else { 23142903ff01SAl Viro path_get(&nd->path); 231534a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 23161da177e4SLinus Torvalds } 231734a26b99SAl Viro fdput(f); 2318e41f7d4eSAl Viro } 23198db52c7eSAleksa Sarai 2320adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2321adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2322adb21d2bSAleksa Sarai nd->root = nd->path; 2323adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2324adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2325adb21d2bSAleksa Sarai } else { 2326adb21d2bSAleksa Sarai path_get(&nd->root); 2327adb21d2bSAleksa Sarai nd->flags |= LOOKUP_ROOT_GRABBED; 2328adb21d2bSAleksa Sarai } 2329adb21d2bSAleksa Sarai } 2330adb21d2bSAleksa Sarai return s; 23319b4a9b14SAl Viro } 23329b4a9b14SAl Viro 23333bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 233495fa25d9SAl Viro { 233595fa25d9SAl Viro const char *s; 2336fec2fa24SAl Viro int error = may_follow_link(nd); 2337deb106c6SAl Viro if (unlikely(error)) 23383bdba28bSAl Viro return ERR_PTR(error); 233995fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2340fab51e8aSAl Viro nd->stack[0].name = NULL; 23413b2e7f75SAl Viro s = get_link(nd); 2342deb106c6SAl Viro return s ? s : ""; 234395fa25d9SAl Viro } 234495fa25d9SAl Viro 2345caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2346bd92d7feSAl Viro { 2347bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2348bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2349bd92d7feSAl Viro 2350bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 23511c4ff1a8SAl Viro return walk_component(nd, 0); 2352bd92d7feSAl Viro } 2353bd92d7feSAl Viro 23544f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 23554f757f3cSAl Viro { 2356db3c9adeSAl Viro struct path path; 23574f757f3cSAl Viro struct inode *inode = nd->inode; 23584f757f3cSAl Viro unsigned seq = nd->seq; 23594f757f3cSAl Viro int err; 23604f757f3cSAl Viro 2361c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2362db3c9adeSAl Viro dget(nd->path.dentry); 2363db3c9adeSAl Viro err = handle_mounts(nd, nd->path.dentry, &path, &inode, &seq); 23644f757f3cSAl Viro if (unlikely(err < 0)) 23654f757f3cSAl Viro return err; 23664f757f3cSAl Viro path_to_nameidata(&path, nd); 23674f757f3cSAl Viro nd->inode = inode; 23684f757f3cSAl Viro nd->seq = seq; 23694f757f3cSAl Viro return 0; 23704f757f3cSAl Viro } 23714f757f3cSAl Viro 23729b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2373c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23749b4a9b14SAl Viro { 2375c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2376bd92d7feSAl Viro int err; 237731e6b01fSNick Piggin 23789b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23794f757f3cSAl Viro err = handle_lookup_down(nd); 23805f336e72SAl Viro if (unlikely(err < 0)) 23815f336e72SAl Viro s = ERR_PTR(err); 23824f757f3cSAl Viro } 23834f757f3cSAl Viro 23843bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 23853bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 23863bdba28bSAl Viro s = trailing_symlink(nd); 2387bd92d7feSAl Viro } 23889f1fafeeSAl Viro if (!err) 23899f1fafeeSAl Viro err = complete_walk(nd); 2390bd92d7feSAl Viro 2391deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2392deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2393bd23a539SAl Viro err = -ENOTDIR; 2394625b6d10SAl Viro if (!err) { 2395625b6d10SAl Viro *path = nd->path; 2396625b6d10SAl Viro nd->path.mnt = NULL; 2397625b6d10SAl Viro nd->path.dentry = NULL; 2398625b6d10SAl Viro } 2399deb106c6SAl Viro terminate_walk(nd); 2400bd92d7feSAl Viro return err; 240131e6b01fSNick Piggin } 240231e6b01fSNick Piggin 240331d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 24049ad1aaa6SAl Viro struct path *path, struct path *root) 2405873f1eedSJeff Layton { 2406894bc8c4SAl Viro int retval; 24079883d185SAl Viro struct nameidata nd; 2408abc9f5beSAl Viro if (IS_ERR(name)) 2409abc9f5beSAl Viro return PTR_ERR(name); 24109ad1aaa6SAl Viro if (unlikely(root)) { 24119ad1aaa6SAl Viro nd.root = *root; 24129ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 24139ad1aaa6SAl Viro } 24149883d185SAl Viro set_nameidata(&nd, dfd, name); 2415c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2416873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2417c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2418873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2419c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2420873f1eedSJeff Layton 2421873f1eedSJeff Layton if (likely(!retval)) 2422ff0ebee2SAl Viro audit_inode(name, path->dentry, 0); 24239883d185SAl Viro restore_nameidata(); 2424e4bd1c1aSAl Viro putname(name); 2425873f1eedSJeff Layton return retval; 2426873f1eedSJeff Layton } 2427873f1eedSJeff Layton 24288bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2429c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2430391172c4SAl Viro struct path *parent) 24318bcb77faSAl Viro { 2432c8a53ee5SAl Viro const char *s = path_init(nd, flags); 24339b5858e9SAl Viro int err = link_path_walk(s, nd); 24348bcb77faSAl Viro if (!err) 24358bcb77faSAl Viro err = complete_walk(nd); 2436391172c4SAl Viro if (!err) { 2437391172c4SAl Viro *parent = nd->path; 2438391172c4SAl Viro nd->path.mnt = NULL; 2439391172c4SAl Viro nd->path.dentry = NULL; 2440391172c4SAl Viro } 2441deb106c6SAl Viro terminate_walk(nd); 24428bcb77faSAl Viro return err; 24438bcb77faSAl Viro } 24448bcb77faSAl Viro 24455c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2446391172c4SAl Viro unsigned int flags, struct path *parent, 2447391172c4SAl Viro struct qstr *last, int *type) 24488bcb77faSAl Viro { 24498bcb77faSAl Viro int retval; 24509883d185SAl Viro struct nameidata nd; 24518bcb77faSAl Viro 24525c31b6ceSAl Viro if (IS_ERR(name)) 24535c31b6ceSAl Viro return name; 24549883d185SAl Viro set_nameidata(&nd, dfd, name); 2455c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 24568bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2457c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 24588bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2459c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2460391172c4SAl Viro if (likely(!retval)) { 2461391172c4SAl Viro *last = nd.last; 2462391172c4SAl Viro *type = nd.last_type; 2463c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 24645c31b6ceSAl Viro } else { 24655c31b6ceSAl Viro putname(name); 24665c31b6ceSAl Viro name = ERR_PTR(retval); 2467391172c4SAl Viro } 24689883d185SAl Viro restore_nameidata(); 24695c31b6ceSAl Viro return name; 24708bcb77faSAl Viro } 24718bcb77faSAl Viro 247279714f72SAl Viro /* does lookup, returns the object with parent locked */ 247379714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24745590ff0dSUlrich Drepper { 24755c31b6ceSAl Viro struct filename *filename; 24765c31b6ceSAl Viro struct dentry *d; 2477391172c4SAl Viro struct qstr last; 2478391172c4SAl Viro int type; 247951689104SPaul Moore 24805c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24815c31b6ceSAl Viro &last, &type); 248251689104SPaul Moore if (IS_ERR(filename)) 248351689104SPaul Moore return ERR_CAST(filename); 24845c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2485391172c4SAl Viro path_put(path); 24865c31b6ceSAl Viro putname(filename); 24875c31b6ceSAl Viro return ERR_PTR(-EINVAL); 248879714f72SAl Viro } 24895955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2490391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 249179714f72SAl Viro if (IS_ERR(d)) { 24925955102cSAl Viro inode_unlock(path->dentry->d_inode); 2493391172c4SAl Viro path_put(path); 249479714f72SAl Viro } 249551689104SPaul Moore putname(filename); 249679714f72SAl Viro return d; 24975590ff0dSUlrich Drepper } 24985590ff0dSUlrich Drepper 2499d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2500d1811465SAl Viro { 2501abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2502abc9f5beSAl Viro flags, path, NULL); 2503d1811465SAl Viro } 25044d359507SAl Viro EXPORT_SYMBOL(kern_path); 2505d1811465SAl Viro 250616f18200SJosef 'Jeff' Sipek /** 250716f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 250816f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 250916f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 251016f18200SJosef 'Jeff' Sipek * @name: pointer to file name 251116f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2512e0a01249SAl Viro * @path: pointer to struct path to fill 251316f18200SJosef 'Jeff' Sipek */ 251416f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 251516f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2516e0a01249SAl Viro struct path *path) 251716f18200SJosef 'Jeff' Sipek { 25189ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 25199ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2520abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2521abc9f5beSAl Viro flags , path, &root); 252216f18200SJosef 'Jeff' Sipek } 25234d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 252416f18200SJosef 'Jeff' Sipek 25253c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 25263c95f0dcSAl Viro int len, struct qstr *this) 25273c95f0dcSAl Viro { 25283c95f0dcSAl Viro this->name = name; 25293c95f0dcSAl Viro this->len = len; 25303c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 25313c95f0dcSAl Viro if (!len) 25323c95f0dcSAl Viro return -EACCES; 25333c95f0dcSAl Viro 25343c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 25353c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 25363c95f0dcSAl Viro return -EACCES; 25373c95f0dcSAl Viro } 25383c95f0dcSAl Viro 25393c95f0dcSAl Viro while (len--) { 25403c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 25413c95f0dcSAl Viro if (c == '/' || c == '\0') 25423c95f0dcSAl Viro return -EACCES; 25433c95f0dcSAl Viro } 25443c95f0dcSAl Viro /* 25453c95f0dcSAl Viro * See if the low-level filesystem might want 25463c95f0dcSAl Viro * to use its own hash.. 25473c95f0dcSAl Viro */ 25483c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 25493c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 25503c95f0dcSAl Viro if (err < 0) 25513c95f0dcSAl Viro return err; 25523c95f0dcSAl Viro } 25533c95f0dcSAl Viro 25543c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 25553c95f0dcSAl Viro } 25563c95f0dcSAl Viro 2557eead1911SChristoph Hellwig /** 25580da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 25590da0b7fdSDavid Howells * @name: pathname component to lookup 25600da0b7fdSDavid Howells * @base: base directory to lookup from 25610da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 25620da0b7fdSDavid Howells * 25630da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 25640da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 25650da0b7fdSDavid Howells * 25660da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25670da0b7fdSDavid Howells * not be called by generic code. 25680da0b7fdSDavid Howells * 25690da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25700da0b7fdSDavid Howells */ 25710da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25720da0b7fdSDavid Howells { 25730da0b7fdSDavid Howells struct qstr this; 25740da0b7fdSDavid Howells int err; 25750da0b7fdSDavid Howells 25760da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25770da0b7fdSDavid Howells 25780da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25790da0b7fdSDavid Howells if (err) 25800da0b7fdSDavid Howells return ERR_PTR(err); 25810da0b7fdSDavid Howells 25820da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25830da0b7fdSDavid Howells } 25840da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25850da0b7fdSDavid Howells 25860da0b7fdSDavid Howells /** 2587a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2588eead1911SChristoph Hellwig * @name: pathname component to lookup 2589eead1911SChristoph Hellwig * @base: base directory to lookup from 2590eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2591eead1911SChristoph Hellwig * 2592a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25939e7543e9SAl Viro * not be called by generic code. 2594bbddca8eSNeilBrown * 2595bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2596eead1911SChristoph Hellwig */ 2597057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2598057f6c01SJames Morris { 25998613a209SAl Viro struct dentry *dentry; 2600057f6c01SJames Morris struct qstr this; 2601cda309deSMiklos Szeredi int err; 2602057f6c01SJames Morris 26035955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26042f9092e1SDavid Woodhouse 26053c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2606cda309deSMiklos Szeredi if (err) 2607cda309deSMiklos Szeredi return ERR_PTR(err); 2608cda309deSMiklos Szeredi 26098613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 26108613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2611057f6c01SJames Morris } 26124d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2613057f6c01SJames Morris 2614bbddca8eSNeilBrown /** 2615bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2616bbddca8eSNeilBrown * @name: pathname component to lookup 2617bbddca8eSNeilBrown * @base: base directory to lookup from 2618bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2619bbddca8eSNeilBrown * 2620bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2621bbddca8eSNeilBrown * not be called by generic code. 2622bbddca8eSNeilBrown * 2623bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2624bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2625bbddca8eSNeilBrown */ 2626bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2627bbddca8eSNeilBrown struct dentry *base, int len) 2628bbddca8eSNeilBrown { 2629bbddca8eSNeilBrown struct qstr this; 2630bbddca8eSNeilBrown int err; 263120d00ee8SLinus Torvalds struct dentry *ret; 2632bbddca8eSNeilBrown 26333c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2634bbddca8eSNeilBrown if (err) 2635bbddca8eSNeilBrown return ERR_PTR(err); 2636bbddca8eSNeilBrown 263720d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 263820d00ee8SLinus Torvalds if (!ret) 263920d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 264020d00ee8SLinus Torvalds return ret; 2641bbddca8eSNeilBrown } 2642bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2643bbddca8eSNeilBrown 26446c2d4798SAl Viro /* 26456c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 26466c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 26476c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 26486c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 26496c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 26506c2d4798SAl Viro * this one avoids such problems. 26516c2d4798SAl Viro */ 26526c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 26536c2d4798SAl Viro struct dentry *base, int len) 26546c2d4798SAl Viro { 26556c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 26562fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 26576c2d4798SAl Viro dput(ret); 26586c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 26596c2d4798SAl Viro } 26606c2d4798SAl Viro return ret; 26616c2d4798SAl Viro } 26626c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 26636c2d4798SAl Viro 2664eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2665eedf265aSEric W. Biederman int path_pts(struct path *path) 2666eedf265aSEric W. Biederman { 2667eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2668eedf265aSEric W. Biederman * the input path. 2669eedf265aSEric W. Biederman */ 2670eedf265aSEric W. Biederman struct dentry *child, *parent; 2671eedf265aSEric W. Biederman struct qstr this; 2672eedf265aSEric W. Biederman int ret; 2673eedf265aSEric W. Biederman 2674eedf265aSEric W. Biederman ret = path_parent_directory(path); 2675eedf265aSEric W. Biederman if (ret) 2676eedf265aSEric W. Biederman return ret; 2677eedf265aSEric W. Biederman 2678eedf265aSEric W. Biederman parent = path->dentry; 2679eedf265aSEric W. Biederman this.name = "pts"; 2680eedf265aSEric W. Biederman this.len = 3; 2681eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2682eedf265aSEric W. Biederman if (!child) 2683eedf265aSEric W. Biederman return -ENOENT; 2684eedf265aSEric W. Biederman 2685eedf265aSEric W. Biederman path->dentry = child; 2686eedf265aSEric W. Biederman dput(parent); 2687eedf265aSEric W. Biederman follow_mount(path); 2688eedf265aSEric W. Biederman return 0; 2689eedf265aSEric W. Biederman } 2690eedf265aSEric W. Biederman #endif 2691eedf265aSEric W. Biederman 26921fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26931fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26941da177e4SLinus Torvalds { 2695abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2696abc9f5beSAl Viro flags, path, NULL); 26971da177e4SLinus Torvalds } 2698b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26991fa1e7f6SAndy Whitcroft 27008033426eSJeff Layton /** 2701197df04cSAl Viro * path_mountpoint - look up a path to be umounted 27022a78b857SMasanari Iida * @nd: lookup context 27038033426eSJeff Layton * @flags: lookup flags 2704c8a53ee5SAl Viro * @path: pointer to container for result 27058033426eSJeff Layton * 27068033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2707606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 27088033426eSJeff Layton */ 27098033426eSJeff Layton static int 2710c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 27118033426eSJeff Layton { 2712c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2713368ee9baSAl Viro int err; 27149b5858e9SAl Viro 27153bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2716c64cd6e3SAl Viro (err = lookup_last(nd)) > 0) { 27173bdba28bSAl Viro s = trailing_symlink(nd); 27183bdba28bSAl Viro } 2719c64cd6e3SAl Viro if (!err && (nd->flags & LOOKUP_RCU)) 2720c64cd6e3SAl Viro err = unlazy_walk(nd); 2721c64cd6e3SAl Viro if (!err) 2722c64cd6e3SAl Viro err = handle_lookup_down(nd); 2723ba8f4613SAl Viro if (!err) { 2724ba8f4613SAl Viro *path = nd->path; 2725ba8f4613SAl Viro nd->path.mnt = NULL; 2726ba8f4613SAl Viro nd->path.dentry = NULL; 2727ba8f4613SAl Viro } 2728deb106c6SAl Viro terminate_walk(nd); 27298033426eSJeff Layton return err; 27308033426eSJeff Layton } 27318033426eSJeff Layton 27322d864651SAl Viro static int 2733668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 27342d864651SAl Viro unsigned int flags) 27352d864651SAl Viro { 27369883d185SAl Viro struct nameidata nd; 2737cbaab2dbSAl Viro int error; 2738668696dcSAl Viro if (IS_ERR(name)) 2739668696dcSAl Viro return PTR_ERR(name); 27409883d185SAl Viro set_nameidata(&nd, dfd, name); 2741c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 27422d864651SAl Viro if (unlikely(error == -ECHILD)) 2743c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 27442d864651SAl Viro if (unlikely(error == -ESTALE)) 2745c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 27462d864651SAl Viro if (likely(!error)) 2747c9b07eabSAl Viro audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL); 27489883d185SAl Viro restore_nameidata(); 2749668696dcSAl Viro putname(name); 27502d864651SAl Viro return error; 27512d864651SAl Viro } 27522d864651SAl Viro 27538033426eSJeff Layton /** 2754197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 27558033426eSJeff Layton * @dfd: directory file descriptor 27568033426eSJeff Layton * @name: pathname from userland 27578033426eSJeff Layton * @flags: lookup flags 27588033426eSJeff Layton * @path: pointer to container to hold result 27598033426eSJeff Layton * 27608033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27618033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27628033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27638033426eSJeff Layton * and avoid revalidating the last component. 27648033426eSJeff Layton * 27658033426eSJeff Layton * Returns 0 and populates "path" on success. 27668033426eSJeff Layton */ 27678033426eSJeff Layton int 2768197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27698033426eSJeff Layton struct path *path) 27708033426eSJeff Layton { 2771cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27728033426eSJeff Layton } 27738033426eSJeff Layton 27742d864651SAl Viro int 27752d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27762d864651SAl Viro unsigned int flags) 27772d864651SAl Viro { 2778cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27792d864651SAl Viro } 27802d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27812d864651SAl Viro 2782cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27831da177e4SLinus Torvalds { 27848e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2785da9592edSDavid Howells 27868e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27871da177e4SLinus Torvalds return 0; 27888e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27891da177e4SLinus Torvalds return 0; 279023adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27911da177e4SLinus Torvalds } 2792cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27931da177e4SLinus Torvalds 27941da177e4SLinus Torvalds /* 27951da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27961da177e4SLinus Torvalds * whether the type of victim is right. 27971da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27981da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27991da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 28001da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 28011da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 28021da177e4SLinus Torvalds * a. be owner of dir, or 28031da177e4SLinus Torvalds * b. be owner of victim, or 28041da177e4SLinus Torvalds * c. have CAP_FOWNER capability 28051da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 28061da177e4SLinus Torvalds * links pointing to it. 28070bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 28080bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 28090bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 28100bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 28110bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 28121da177e4SLinus Torvalds * nfs_async_unlink(). 28131da177e4SLinus Torvalds */ 2814b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 28151da177e4SLinus Torvalds { 281663afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 28171da177e4SLinus Torvalds int error; 28181da177e4SLinus Torvalds 2819b18825a7SDavid Howells if (d_is_negative(victim)) 28201da177e4SLinus Torvalds return -ENOENT; 2821b18825a7SDavid Howells BUG_ON(!inode); 28221da177e4SLinus Torvalds 28231da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2824593d1ce8SEric W. Biederman 2825593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2826593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2827593d1ce8SEric W. Biederman return -EOVERFLOW; 2828593d1ce8SEric W. Biederman 28294fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 28301da177e4SLinus Torvalds 2831f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 28321da177e4SLinus Torvalds if (error) 28331da177e4SLinus Torvalds return error; 28341da177e4SLinus Torvalds if (IS_APPEND(dir)) 28351da177e4SLinus Torvalds return -EPERM; 2836b18825a7SDavid Howells 2837b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 28380bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 28391da177e4SLinus Torvalds return -EPERM; 28401da177e4SLinus Torvalds if (isdir) { 284144b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28421da177e4SLinus Torvalds return -ENOTDIR; 28431da177e4SLinus Torvalds if (IS_ROOT(victim)) 28441da177e4SLinus Torvalds return -EBUSY; 284544b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28461da177e4SLinus Torvalds return -EISDIR; 28471da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28481da177e4SLinus Torvalds return -ENOENT; 28491da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28501da177e4SLinus Torvalds return -EBUSY; 28511da177e4SLinus Torvalds return 0; 28521da177e4SLinus Torvalds } 28531da177e4SLinus Torvalds 28541da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28551da177e4SLinus Torvalds * dir. 28561da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28571da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28581da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2859036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2860036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2861036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28621da177e4SLinus Torvalds */ 2863a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28641da177e4SLinus Torvalds { 2865036d5236SEric W. Biederman struct user_namespace *s_user_ns; 286614e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28671da177e4SLinus Torvalds if (child->d_inode) 28681da177e4SLinus Torvalds return -EEXIST; 28691da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28701da177e4SLinus Torvalds return -ENOENT; 2871036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2872036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2873036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2874036d5236SEric W. Biederman return -EOVERFLOW; 2875f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28761da177e4SLinus Torvalds } 28771da177e4SLinus Torvalds 28781da177e4SLinus Torvalds /* 28791da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28801da177e4SLinus Torvalds */ 28811da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28821da177e4SLinus Torvalds { 28831da177e4SLinus Torvalds struct dentry *p; 28841da177e4SLinus Torvalds 28851da177e4SLinus Torvalds if (p1 == p2) { 28865955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28871da177e4SLinus Torvalds return NULL; 28881da177e4SLinus Torvalds } 28891da177e4SLinus Torvalds 2890fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28911da177e4SLinus Torvalds 2892e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2893e2761a11SOGAWA Hirofumi if (p) { 28945955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28955955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28961da177e4SLinus Torvalds return p; 28971da177e4SLinus Torvalds } 28981da177e4SLinus Torvalds 2899e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2900e2761a11SOGAWA Hirofumi if (p) { 29015955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29025955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 29031da177e4SLinus Torvalds return p; 29041da177e4SLinus Torvalds } 29051da177e4SLinus Torvalds 29065955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29075955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 29081da177e4SLinus Torvalds return NULL; 29091da177e4SLinus Torvalds } 29104d359507SAl Viro EXPORT_SYMBOL(lock_rename); 29111da177e4SLinus Torvalds 29121da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 29131da177e4SLinus Torvalds { 29145955102cSAl Viro inode_unlock(p1->d_inode); 29151da177e4SLinus Torvalds if (p1 != p2) { 29165955102cSAl Viro inode_unlock(p2->d_inode); 2917fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 29181da177e4SLinus Torvalds } 29191da177e4SLinus Torvalds } 29204d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 29211da177e4SLinus Torvalds 29224acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2923312b63fbSAl Viro bool want_excl) 29241da177e4SLinus Torvalds { 2925a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 29261da177e4SLinus Torvalds if (error) 29271da177e4SLinus Torvalds return error; 29281da177e4SLinus Torvalds 2929acfa4380SAl Viro if (!dir->i_op->create) 29301da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 29311da177e4SLinus Torvalds mode &= S_IALLUGO; 29321da177e4SLinus Torvalds mode |= S_IFREG; 29331da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 29341da177e4SLinus Torvalds if (error) 29351da177e4SLinus Torvalds return error; 2936312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2937a74574aaSStephen Smalley if (!error) 2938f38aa942SAmy Griffis fsnotify_create(dir, dentry); 29391da177e4SLinus Torvalds return error; 29401da177e4SLinus Torvalds } 29414d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29421da177e4SLinus Torvalds 29438e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29448e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29458e6c848eSAl Viro void *arg) 29468e6c848eSAl Viro { 29478e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 29488e6c848eSAl Viro int error = may_create(dir, dentry); 29498e6c848eSAl Viro if (error) 29508e6c848eSAl Viro return error; 29518e6c848eSAl Viro 29528e6c848eSAl Viro mode &= S_IALLUGO; 29538e6c848eSAl Viro mode |= S_IFREG; 29548e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29558e6c848eSAl Viro if (error) 29568e6c848eSAl Viro return error; 29578e6c848eSAl Viro error = f(dentry, mode, arg); 29588e6c848eSAl Viro if (!error) 29598e6c848eSAl Viro fsnotify_create(dir, dentry); 29608e6c848eSAl Viro return error; 29618e6c848eSAl Viro } 29628e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29638e6c848eSAl Viro 2964a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2965a2982cc9SEric W. Biederman { 2966a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2967a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2968a2982cc9SEric W. Biederman } 2969a2982cc9SEric W. Biederman 2970f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 29711da177e4SLinus Torvalds { 29723fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29731da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29741da177e4SLinus Torvalds int error; 29751da177e4SLinus Torvalds 29761da177e4SLinus Torvalds if (!inode) 29771da177e4SLinus Torvalds return -ENOENT; 29781da177e4SLinus Torvalds 2979c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2980c8fe8f30SChristoph Hellwig case S_IFLNK: 29811da177e4SLinus Torvalds return -ELOOP; 2982c8fe8f30SChristoph Hellwig case S_IFDIR: 2983c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29841da177e4SLinus Torvalds return -EISDIR; 2985c8fe8f30SChristoph Hellwig break; 2986c8fe8f30SChristoph Hellwig case S_IFBLK: 2987c8fe8f30SChristoph Hellwig case S_IFCHR: 2988a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29891da177e4SLinus Torvalds return -EACCES; 2990c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2991c8fe8f30SChristoph Hellwig case S_IFIFO: 2992c8fe8f30SChristoph Hellwig case S_IFSOCK: 29931da177e4SLinus Torvalds flag &= ~O_TRUNC; 2994c8fe8f30SChristoph Hellwig break; 29954a3fd211SDave Hansen } 2996b41572e9SDave Hansen 299762fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2998b41572e9SDave Hansen if (error) 2999b41572e9SDave Hansen return error; 30006146f0d5SMimi Zohar 30011da177e4SLinus Torvalds /* 30021da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 30031da177e4SLinus Torvalds */ 30041da177e4SLinus Torvalds if (IS_APPEND(inode)) { 30058737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 30067715b521SAl Viro return -EPERM; 30071da177e4SLinus Torvalds if (flag & O_TRUNC) 30087715b521SAl Viro return -EPERM; 30091da177e4SLinus Torvalds } 30101da177e4SLinus Torvalds 30111da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 30122e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 30137715b521SAl Viro return -EPERM; 30141da177e4SLinus Torvalds 3015f3c7691eSJ. Bruce Fields return 0; 30167715b521SAl Viro } 30177715b521SAl Viro 3018e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 30197715b521SAl Viro { 3020f0bb5aafSAl Viro const struct path *path = &filp->f_path; 30217715b521SAl Viro struct inode *inode = path->dentry->d_inode; 30227715b521SAl Viro int error = get_write_access(inode); 30231da177e4SLinus Torvalds if (error) 30247715b521SAl Viro return error; 30251da177e4SLinus Torvalds /* 30261da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 30271da177e4SLinus Torvalds */ 3028d7a06983SJeff Layton error = locks_verify_locked(filp); 3029be6d3e56SKentaro Takeda if (!error) 3030ea0d3ab2STetsuo Handa error = security_path_truncate(path); 30311da177e4SLinus Torvalds if (!error) { 30327715b521SAl Viro error = do_truncate(path->dentry, 0, 3033d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3034e1181ee6SJeff Layton filp); 30351da177e4SLinus Torvalds } 30361da177e4SLinus Torvalds put_write_access(inode); 3037acd0c935SMimi Zohar return error; 30381da177e4SLinus Torvalds } 30391da177e4SLinus Torvalds 3040d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3041d57999e1SDave Hansen { 30428a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30438a5e929dSAl Viro flag--; 3044d57999e1SDave Hansen return flag; 3045d57999e1SDave Hansen } 3046d57999e1SDave Hansen 3047d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 3048d18e9008SMiklos Szeredi { 30491328c727SSeth Forshee struct user_namespace *s_user_ns; 3050d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3051d18e9008SMiklos Szeredi if (error) 3052d18e9008SMiklos Szeredi return error; 3053d18e9008SMiklos Szeredi 30541328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 30551328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 30561328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 30571328c727SSeth Forshee return -EOVERFLOW; 30581328c727SSeth Forshee 3059d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 3060d18e9008SMiklos Szeredi if (error) 3061d18e9008SMiklos Szeredi return error; 3062d18e9008SMiklos Szeredi 3063d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3064d18e9008SMiklos Szeredi } 3065d18e9008SMiklos Szeredi 30661acf0af9SDavid Howells /* 30671acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30681acf0af9SDavid Howells * dentry. 30691acf0af9SDavid Howells * 30701acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30711acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30721acf0af9SDavid Howells * 307300a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 307400a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 307500a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 30761acf0af9SDavid Howells * 30771acf0af9SDavid Howells * Returns an error code otherwise. 30781acf0af9SDavid Howells */ 3079239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3080239eb983SAl Viro struct file *file, 3081015c3bbcSMiklos Szeredi const struct open_flags *op, 30823ec2eef1SAl Viro int open_flag, umode_t mode) 3083d18e9008SMiklos Szeredi { 3084d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3085d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3086d18e9008SMiklos Szeredi int error; 3087d18e9008SMiklos Szeredi 3088384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3089d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3090d18e9008SMiklos Szeredi 3091d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3092d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3093d18e9008SMiklos Szeredi 309430d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 309530d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30960fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 309744907d79SAl Viro open_to_namei_flags(open_flag), mode); 30986fbd0714SAl Viro d_lookup_done(dentry); 3099384f26e2SAl Viro if (!error) { 310064e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 3101d18e9008SMiklos Szeredi /* 3102384f26e2SAl Viro * We didn't have the inode before the open, so check open 3103384f26e2SAl Viro * permission here. 3104d18e9008SMiklos Szeredi */ 3105384f26e2SAl Viro int acc_mode = op->acc_mode; 310673a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 310703da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 310803da633aSAl Viro fsnotify_create(dir, dentry); 310962fb4a15SAl Viro acc_mode = 0; 311003da633aSAl Viro } 31112675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3112384f26e2SAl Viro if (WARN_ON(error > 0)) 3113384f26e2SAl Viro error = -EINVAL; 311464e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3115d18e9008SMiklos Szeredi error = -EIO; 3116384f26e2SAl Viro } else { 3117d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3118d18e9008SMiklos Szeredi dput(dentry); 3119d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 312010c64ceaSAl Viro } 312173a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 3122d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3123239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3124a01e718fSAl Viro error = -ENOENT; 3125d18e9008SMiklos Szeredi } 3126d18e9008SMiklos Szeredi } 3127239eb983SAl Viro if (error) { 3128d18e9008SMiklos Szeredi dput(dentry); 3129239eb983SAl Viro dentry = ERR_PTR(error); 3130239eb983SAl Viro } 3131239eb983SAl Viro return dentry; 3132d18e9008SMiklos Szeredi } 3133d18e9008SMiklos Szeredi 313431e6b01fSNick Piggin /* 31351acf0af9SDavid Howells * Look up and maybe create and open the last component. 3136d58ffd35SMiklos Szeredi * 313700a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3138d58ffd35SMiklos Szeredi * 313900a07c15SAl Viro * Returns 0 on success, that is, if 314000a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 314100a07c15SAl Viro * the file was not completely opened at this time, though lookups and 314200a07c15SAl Viro * creations were performed. 314300a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 314400a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 314500a07c15SAl Viro * hadn't been specified. 31461acf0af9SDavid Howells * 314700a07c15SAl Viro * An error code is returned on failure. 3148d58ffd35SMiklos Szeredi */ 3149da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3150d58ffd35SMiklos Szeredi const struct open_flags *op, 31513ec2eef1SAl Viro bool got_write) 3152d58ffd35SMiklos Szeredi { 3153d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 315454ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31551643b43fSAl Viro int open_flag = op->open_flag; 3156d58ffd35SMiklos Szeredi struct dentry *dentry; 31571643b43fSAl Viro int error, create_error = 0; 31581643b43fSAl Viro umode_t mode = op->mode; 31596fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3160d58ffd35SMiklos Szeredi 3161ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3162da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3163d58ffd35SMiklos Szeredi 316473a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 31656fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31666fbd0714SAl Viro for (;;) { 31676fbd0714SAl Viro if (!dentry) { 31686fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3169d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3170da5ebf5aSAl Viro return dentry; 31716fbd0714SAl Viro } 31726fbd0714SAl Viro if (d_in_lookup(dentry)) 31736fbd0714SAl Viro break; 3174d58ffd35SMiklos Szeredi 31756fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31766fbd0714SAl Viro if (likely(error > 0)) 31776fbd0714SAl Viro break; 31786fbd0714SAl Viro if (error) 31796fbd0714SAl Viro goto out_dput; 31806fbd0714SAl Viro d_invalidate(dentry); 31816fbd0714SAl Viro dput(dentry); 31826fbd0714SAl Viro dentry = NULL; 31836fbd0714SAl Viro } 31846fbd0714SAl Viro if (dentry->d_inode) { 3185d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3186da5ebf5aSAl Viro return dentry; 31876c51e513SAl Viro } 3188d18e9008SMiklos Szeredi 31891643b43fSAl Viro /* 31901643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31911643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31921643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31931643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31941643b43fSAl Viro * 31951643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31961643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31971643b43fSAl Viro */ 31981643b43fSAl Viro if (open_flag & O_CREAT) { 31991643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 32001643b43fSAl Viro mode &= ~current_umask(); 32011643b43fSAl Viro if (unlikely(!got_write)) { 32021643b43fSAl Viro create_error = -EROFS; 32031643b43fSAl Viro open_flag &= ~O_CREAT; 32041643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 32051643b43fSAl Viro goto no_open; 32061643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 32071643b43fSAl Viro } else { 32081643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 32091643b43fSAl Viro if (create_error) { 32101643b43fSAl Viro open_flag &= ~O_CREAT; 32111643b43fSAl Viro if (open_flag & O_EXCL) 32121643b43fSAl Viro goto no_open; 32131643b43fSAl Viro } 32141643b43fSAl Viro } 32151643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 32161643b43fSAl Viro unlikely(!got_write)) { 32171643b43fSAl Viro /* 32181643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 32191643b43fSAl Viro * back to lookup + open 32201643b43fSAl Viro */ 32211643b43fSAl Viro goto no_open; 3222d18e9008SMiklos Szeredi } 3223d18e9008SMiklos Szeredi 32241643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3225239eb983SAl Viro dentry = atomic_open(nd, dentry, file, op, open_flag, mode); 3226da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3227da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3228da5ebf5aSAl Viro return dentry; 3229239eb983SAl Viro } 323054ef4872SMiklos Szeredi 32311643b43fSAl Viro no_open: 32326fbd0714SAl Viro if (d_in_lookup(dentry)) { 323312fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 323412fa5e24SAl Viro nd->flags); 32356fbd0714SAl Viro d_lookup_done(dentry); 323612fa5e24SAl Viro if (unlikely(res)) { 323712fa5e24SAl Viro if (IS_ERR(res)) { 323812fa5e24SAl Viro error = PTR_ERR(res); 323912fa5e24SAl Viro goto out_dput; 324012fa5e24SAl Viro } 324112fa5e24SAl Viro dput(dentry); 324212fa5e24SAl Viro dentry = res; 324312fa5e24SAl Viro } 324454ef4872SMiklos Szeredi } 324554ef4872SMiklos Szeredi 3246d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32471643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 324873a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3249ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3250ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3251ce8644fcSAl Viro error = -EACCES; 3252d58ffd35SMiklos Szeredi goto out_dput; 325364894cf8SAl Viro } 3254ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 32551643b43fSAl Viro open_flag & O_EXCL); 3256d58ffd35SMiklos Szeredi if (error) 3257d58ffd35SMiklos Szeredi goto out_dput; 3258ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3259d58ffd35SMiklos Szeredi } 32601643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32611643b43fSAl Viro error = create_error; 3262d58ffd35SMiklos Szeredi goto out_dput; 3263d58ffd35SMiklos Szeredi } 3264da5ebf5aSAl Viro return dentry; 3265d58ffd35SMiklos Szeredi 3266d58ffd35SMiklos Szeredi out_dput: 3267d58ffd35SMiklos Szeredi dput(dentry); 3268da5ebf5aSAl Viro return ERR_PTR(error); 3269d58ffd35SMiklos Szeredi } 3270d58ffd35SMiklos Szeredi 3271d58ffd35SMiklos Szeredi /* 3272fe2d35ffSAl Viro * Handle the last step of open() 327331e6b01fSNick Piggin */ 3274896475d5SAl Viro static int do_last(struct nameidata *nd, 32753ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3276fb1cc555SAl Viro { 3277a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 32786404674aSAl Viro kuid_t dir_uid = nd->inode->i_uid; 32796404674aSAl Viro umode_t dir_mode = nd->inode->i_mode; 3280ca344a89SAl Viro int open_flag = op->open_flag; 328177d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 328264894cf8SAl Viro bool got_write = false; 3283bcda7652SAl Viro int acc_mode = op->acc_mode; 3284254cf582SAl Viro unsigned seq; 3285a1eb3315SMiklos Szeredi struct inode *inode; 3286896475d5SAl Viro struct path path; 3287da5ebf5aSAl Viro struct dentry *dentry; 328816c2cd71SAl Viro int error; 3289fb1cc555SAl Viro 3290c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3291c3e380b0SAl Viro nd->flags |= op->intent; 3292c3e380b0SAl Viro 3293bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 329456676ec3SAl Viro if (nd->depth) 329556676ec3SAl Viro put_link(nd); 3296fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3297deb106c6SAl Viro if (unlikely(error)) 32982675a4ebSAl Viro return error; 3299e83db167SMiklos Szeredi goto finish_open; 33001f36f774SAl Viro } 3301a2c36b45SAl Viro 3302ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3303fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3304fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3305fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 330620e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 330720e34357SAl Viro if (IS_ERR(dentry)) 330820e34357SAl Viro return PTR_ERR(dentry); 330920e34357SAl Viro if (likely(dentry)) 331071574865SMiklos Szeredi goto finish_lookup; 331171574865SMiklos Szeredi 331237d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 33136583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3314b6183df7SMiklos Szeredi } else { 3315fe2d35ffSAl Viro /* create side of things */ 3316a3fbbde7SAl Viro /* 3317b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3318b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3319b6183df7SMiklos Szeredi * about to look up 3320a3fbbde7SAl Viro */ 33219f1fafeeSAl Viro error = complete_walk(nd); 3322e8bb73dfSAl Viro if (error) 33232675a4ebSAl Viro return error; 3324fe2d35ffSAl Viro 3325c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 33261f36f774SAl Viro /* trailing slashes? */ 3327deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3328deb106c6SAl Viro return -EISDIR; 3329b6183df7SMiklos Szeredi } 33301f36f774SAl Viro 33319cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 333264894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 333364894cf8SAl Viro if (!error) 333464894cf8SAl Viro got_write = true; 333564894cf8SAl Viro /* 333664894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 333764894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 333864894cf8SAl Viro * dropping this one anyway. 333964894cf8SAl Viro */ 334064894cf8SAl Viro } 33419cf843e3SAl Viro if (open_flag & O_CREAT) 33425955102cSAl Viro inode_lock(dir->d_inode); 33439cf843e3SAl Viro else 33449cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3345da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 33469cf843e3SAl Viro if (open_flag & O_CREAT) 33475955102cSAl Viro inode_unlock(dir->d_inode); 33489cf843e3SAl Viro else 33499cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3350fb1cc555SAl Viro 3351da5ebf5aSAl Viro if (IS_ERR(dentry)) { 3352da5ebf5aSAl Viro error = PTR_ERR(dentry); 3353d58ffd35SMiklos Szeredi goto out; 3354da5ebf5aSAl Viro } 33556c0d46c4SAl Viro 335600a07c15SAl Viro if (file->f_mode & FMODE_OPENED) { 335773a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3358496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 335977d660a8SMiklos Szeredi will_truncate = false; 3360d18e9008SMiklos Szeredi 336176ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3362da5ebf5aSAl Viro dput(dentry); 3363d18e9008SMiklos Szeredi goto opened; 3364d18e9008SMiklos Szeredi } 3365d18e9008SMiklos Szeredi 336673a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 33679b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3368ca344a89SAl Viro open_flag &= ~O_TRUNC; 336977d660a8SMiklos Szeredi will_truncate = false; 337062fb4a15SAl Viro acc_mode = 0; 3371e73cabffSAl Viro dput(nd->path.dentry); 3372e73cabffSAl Viro nd->path.dentry = dentry; 3373e83db167SMiklos Szeredi goto finish_open_created; 3374fb1cc555SAl Viro } 3375fb1cc555SAl Viro 3376fb1cc555SAl Viro /* 3377d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3378d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3379d18e9008SMiklos Szeredi * necessary...) 3380d18e9008SMiklos Szeredi */ 338164894cf8SAl Viro if (got_write) { 3382d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 338364894cf8SAl Viro got_write = false; 3384d18e9008SMiklos Szeredi } 3385d18e9008SMiklos Szeredi 338620e34357SAl Viro finish_lookup: 338756676ec3SAl Viro if (nd->depth) 338856676ec3SAl Viro put_link(nd); 3389db3c9adeSAl Viro error = handle_mounts(nd, dentry, &path, &inode, &seq); 3390e6ec03a2SAl Viro if (unlikely(error < 0)) 3391e6ec03a2SAl Viro return error; 33928f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3393deb106c6SAl Viro if (unlikely(error)) 3394d63ff28fSAl Viro return error; 339531d1726dSAl Viro 339631d1726dSAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 339731d1726dSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 339831d1726dSAl Viro return -EEXIST; 339931d1726dSAl Viro } 3400bc77daa7SAl Viro finish_open: 34018f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3402a3fbbde7SAl Viro error = complete_walk(nd); 3403fac7d191SAl Viro if (error) 34042675a4ebSAl Viro return error; 340576ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 340630aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3407fb1cc555SAl Viro error = -EISDIR; 340830aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 34092675a4ebSAl Viro goto out; 3410d0cb5018SAl Viro error = may_create_in_sticky(dir_mode, dir_uid, 341130aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 341230aba665SSalvatore Mesoraca if (unlikely(error)) 341330aba665SSalvatore Mesoraca goto out; 341430aba665SSalvatore Mesoraca } 3415af2f5542SMiklos Szeredi error = -ENOTDIR; 341644b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 34172675a4ebSAl Viro goto out; 34184bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 341977d660a8SMiklos Szeredi will_truncate = false; 34206c0d46c4SAl Viro 34210f9d1a10SAl Viro if (will_truncate) { 34220f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 34230f9d1a10SAl Viro if (error) 34242675a4ebSAl Viro goto out; 342564894cf8SAl Viro got_write = true; 34260f9d1a10SAl Viro } 3427e83db167SMiklos Szeredi finish_open_created: 3428bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3429ca344a89SAl Viro if (error) 34302675a4ebSAl Viro goto out; 3431aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3432ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3433fac7d191SAl Viro if (error) 3434015c3bbcSMiklos Szeredi goto out; 3435a8277b9bSMiklos Szeredi opened: 34366035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3437fe9ec829SAl Viro if (!error && will_truncate) 34382675a4ebSAl Viro error = handle_truncate(file); 3439ca344a89SAl Viro out: 3440c80567c8SAl Viro if (unlikely(error > 0)) { 3441c80567c8SAl Viro WARN_ON(1); 3442c80567c8SAl Viro error = -EINVAL; 3443c80567c8SAl Viro } 344464894cf8SAl Viro if (got_write) 34450f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 34462675a4ebSAl Viro return error; 3447fb1cc555SAl Viro } 3448fb1cc555SAl Viro 3449af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3450af7bd4dcSAmir Goldstein { 3451af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3452af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3453af7bd4dcSAmir Goldstein struct inode *inode; 3454af7bd4dcSAmir Goldstein int error; 3455af7bd4dcSAmir Goldstein 3456af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3457af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3458af7bd4dcSAmir Goldstein if (error) 3459af7bd4dcSAmir Goldstein goto out_err; 3460af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3461af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3462af7bd4dcSAmir Goldstein goto out_err; 3463af7bd4dcSAmir Goldstein error = -ENOMEM; 3464cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3465af7bd4dcSAmir Goldstein if (unlikely(!child)) 3466af7bd4dcSAmir Goldstein goto out_err; 3467af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3468af7bd4dcSAmir Goldstein if (error) 3469af7bd4dcSAmir Goldstein goto out_err; 3470af7bd4dcSAmir Goldstein error = -ENOENT; 3471af7bd4dcSAmir Goldstein inode = child->d_inode; 3472af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3473af7bd4dcSAmir Goldstein goto out_err; 3474af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3475af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3476af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3477af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3478af7bd4dcSAmir Goldstein } 3479fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3480af7bd4dcSAmir Goldstein return child; 3481af7bd4dcSAmir Goldstein 3482af7bd4dcSAmir Goldstein out_err: 3483af7bd4dcSAmir Goldstein dput(child); 3484af7bd4dcSAmir Goldstein return ERR_PTR(error); 3485af7bd4dcSAmir Goldstein } 3486af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3487af7bd4dcSAmir Goldstein 3488c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 348960545d0dSAl Viro const struct open_flags *op, 34903ec2eef1SAl Viro struct file *file) 349160545d0dSAl Viro { 3492625b6d10SAl Viro struct dentry *child; 3493625b6d10SAl Viro struct path path; 3494c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 349560545d0dSAl Viro if (unlikely(error)) 349660545d0dSAl Viro return error; 3497625b6d10SAl Viro error = mnt_want_write(path.mnt); 349860545d0dSAl Viro if (unlikely(error)) 349960545d0dSAl Viro goto out; 3500af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3501af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3502684e73beSHirofumi Nakagawa if (IS_ERR(child)) 350360545d0dSAl Viro goto out2; 3504625b6d10SAl Viro dput(path.dentry); 3505625b6d10SAl Viro path.dentry = child; 3506c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 350769a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 350862fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 350960545d0dSAl Viro if (error) 351060545d0dSAl Viro goto out2; 3511625b6d10SAl Viro file->f_path.mnt = path.mnt; 3512be12af3eSAl Viro error = finish_open(file, child, NULL); 351360545d0dSAl Viro out2: 3514625b6d10SAl Viro mnt_drop_write(path.mnt); 351560545d0dSAl Viro out: 3516625b6d10SAl Viro path_put(&path); 351760545d0dSAl Viro return error; 351860545d0dSAl Viro } 351960545d0dSAl Viro 35206ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 35216ac08709SAl Viro { 35226ac08709SAl Viro struct path path; 35236ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 35246ac08709SAl Viro if (!error) { 35256ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3526ae2bb293SAl Viro error = vfs_open(&path, file); 35276ac08709SAl Viro path_put(&path); 35286ac08709SAl Viro } 35296ac08709SAl Viro return error; 35306ac08709SAl Viro } 35316ac08709SAl Viro 3532c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3533c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35341da177e4SLinus Torvalds { 353530d90494SAl Viro struct file *file; 353613aab428SAl Viro int error; 353731e6b01fSNick Piggin 3538ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35391afc99beSAl Viro if (IS_ERR(file)) 35401afc99beSAl Viro return file; 354131e6b01fSNick Piggin 3542bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 35433ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 35445f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 35456ac08709SAl Viro error = do_o_path(nd, flags, file); 35465f336e72SAl Viro } else { 35475f336e72SAl Viro const char *s = path_init(nd, flags); 35483bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 35493ec2eef1SAl Viro (error = do_last(nd, file, op)) > 0) { 355073d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35513bdba28bSAl Viro s = trailing_symlink(nd); 35523bdba28bSAl Viro } 3553deb106c6SAl Viro terminate_walk(nd); 35545f336e72SAl Viro } 35557c1c01ecSAl Viro if (likely(!error)) { 3556aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35577c1c01ecSAl Viro return file; 35587c1c01ecSAl Viro WARN_ON(1); 35597c1c01ecSAl Viro error = -EINVAL; 3560015c3bbcSMiklos Szeredi } 35617c1c01ecSAl Viro fput(file); 35622675a4ebSAl Viro if (error == -EOPENSTALE) { 35632675a4ebSAl Viro if (flags & LOOKUP_RCU) 35642675a4ebSAl Viro error = -ECHILD; 35652675a4ebSAl Viro else 35662675a4ebSAl Viro error = -ESTALE; 35672675a4ebSAl Viro } 35687c1c01ecSAl Viro return ERR_PTR(error); 3569de459215SKirill Korotaev } 35701da177e4SLinus Torvalds 3571669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3572f9652e10SAl Viro const struct open_flags *op) 357313aab428SAl Viro { 35749883d185SAl Viro struct nameidata nd; 3575f9652e10SAl Viro int flags = op->lookup_flags; 357613aab428SAl Viro struct file *filp; 357713aab428SAl Viro 35789883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3579c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 358013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3581c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 358213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3583c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35849883d185SAl Viro restore_nameidata(); 358513aab428SAl Viro return filp; 358613aab428SAl Viro } 358713aab428SAl Viro 358873d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3589f9652e10SAl Viro const char *name, const struct open_flags *op) 359073d049a4SAl Viro { 35919883d185SAl Viro struct nameidata nd; 359273d049a4SAl Viro struct file *file; 359351689104SPaul Moore struct filename *filename; 3594f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 359573d049a4SAl Viro 359673d049a4SAl Viro nd.root.mnt = mnt; 359773d049a4SAl Viro nd.root.dentry = dentry; 359873d049a4SAl Viro 3599b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 360073d049a4SAl Viro return ERR_PTR(-ELOOP); 360173d049a4SAl Viro 360251689104SPaul Moore filename = getname_kernel(name); 3603a1c83681SViresh Kumar if (IS_ERR(filename)) 360451689104SPaul Moore return ERR_CAST(filename); 360551689104SPaul Moore 36069883d185SAl Viro set_nameidata(&nd, -1, filename); 3607c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 360873d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3609c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 361073d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3611c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 36129883d185SAl Viro restore_nameidata(); 361351689104SPaul Moore putname(filename); 361473d049a4SAl Viro return file; 361573d049a4SAl Viro } 361673d049a4SAl Viro 3617fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 36181ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36191da177e4SLinus Torvalds { 3620c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3621391172c4SAl Viro struct qstr last; 3622391172c4SAl Viro int type; 3623c30dabfeSJan Kara int err2; 36241ac12b4bSJeff Layton int error; 36251ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36261ac12b4bSJeff Layton 36271ac12b4bSJeff Layton /* 36281ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36291ac12b4bSJeff Layton * other flags passed in are ignored! 36301ac12b4bSJeff Layton */ 36311ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36321ac12b4bSJeff Layton 36335c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36345c31b6ceSAl Viro if (IS_ERR(name)) 36355c31b6ceSAl Viro return ERR_CAST(name); 36361da177e4SLinus Torvalds 3637c663e5d8SChristoph Hellwig /* 3638c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3639c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3640c663e5d8SChristoph Hellwig */ 36415c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3642ed75e95dSAl Viro goto out; 3643c663e5d8SChristoph Hellwig 3644c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3645391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3646c663e5d8SChristoph Hellwig /* 3647c663e5d8SChristoph Hellwig * Do the final lookup. 3648c663e5d8SChristoph Hellwig */ 3649391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36505955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3651391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36521da177e4SLinus Torvalds if (IS_ERR(dentry)) 3653a8104a9fSAl Viro goto unlock; 3654c663e5d8SChristoph Hellwig 3655a8104a9fSAl Viro error = -EEXIST; 3656b18825a7SDavid Howells if (d_is_positive(dentry)) 3657a8104a9fSAl Viro goto fail; 3658b18825a7SDavid Howells 3659c663e5d8SChristoph Hellwig /* 3660c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3661c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3662c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3663c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3664c663e5d8SChristoph Hellwig */ 3665391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3666a8104a9fSAl Viro error = -ENOENT; 3667ed75e95dSAl Viro goto fail; 3668e9baf6e5SAl Viro } 3669c30dabfeSJan Kara if (unlikely(err2)) { 3670c30dabfeSJan Kara error = err2; 3671a8104a9fSAl Viro goto fail; 3672c30dabfeSJan Kara } 3673181c37b6SAl Viro putname(name); 3674e9baf6e5SAl Viro return dentry; 36751da177e4SLinus Torvalds fail: 3676a8104a9fSAl Viro dput(dentry); 3677a8104a9fSAl Viro dentry = ERR_PTR(error); 3678a8104a9fSAl Viro unlock: 36795955102cSAl Viro inode_unlock(path->dentry->d_inode); 3680c30dabfeSJan Kara if (!err2) 3681391172c4SAl Viro mnt_drop_write(path->mnt); 3682ed75e95dSAl Viro out: 3683391172c4SAl Viro path_put(path); 3684181c37b6SAl Viro putname(name); 3685ed75e95dSAl Viro return dentry; 3686dae6ad8fSAl Viro } 3687fa14a0b8SAl Viro 3688fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3689fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3690fa14a0b8SAl Viro { 3691181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3692181c37b6SAl Viro path, lookup_flags); 3693fa14a0b8SAl Viro } 3694dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3695dae6ad8fSAl Viro 3696921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3697921a1650SAl Viro { 3698921a1650SAl Viro dput(dentry); 36995955102cSAl Viro inode_unlock(path->dentry->d_inode); 3700a8104a9fSAl Viro mnt_drop_write(path->mnt); 3701921a1650SAl Viro path_put(path); 3702921a1650SAl Viro } 3703921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3704921a1650SAl Viro 3705520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 37061ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3707dae6ad8fSAl Viro { 3708181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3709dae6ad8fSAl Viro } 3710dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3711dae6ad8fSAl Viro 37121a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 37131da177e4SLinus Torvalds { 3714a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37151da177e4SLinus Torvalds 37161da177e4SLinus Torvalds if (error) 37171da177e4SLinus Torvalds return error; 37181da177e4SLinus Torvalds 371994f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 37201da177e4SLinus Torvalds return -EPERM; 37211da177e4SLinus Torvalds 3722acfa4380SAl Viro if (!dir->i_op->mknod) 37231da177e4SLinus Torvalds return -EPERM; 37241da177e4SLinus Torvalds 372508ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 372608ce5f16SSerge E. Hallyn if (error) 372708ce5f16SSerge E. Hallyn return error; 372808ce5f16SSerge E. Hallyn 37291da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37301da177e4SLinus Torvalds if (error) 37311da177e4SLinus Torvalds return error; 37321da177e4SLinus Torvalds 37331da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3734a74574aaSStephen Smalley if (!error) 3735f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37361da177e4SLinus Torvalds return error; 37371da177e4SLinus Torvalds } 37384d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37391da177e4SLinus Torvalds 3740f69aac00SAl Viro static int may_mknod(umode_t mode) 3741463c3197SDave Hansen { 3742463c3197SDave Hansen switch (mode & S_IFMT) { 3743463c3197SDave Hansen case S_IFREG: 3744463c3197SDave Hansen case S_IFCHR: 3745463c3197SDave Hansen case S_IFBLK: 3746463c3197SDave Hansen case S_IFIFO: 3747463c3197SDave Hansen case S_IFSOCK: 3748463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3749463c3197SDave Hansen return 0; 3750463c3197SDave Hansen case S_IFDIR: 3751463c3197SDave Hansen return -EPERM; 3752463c3197SDave Hansen default: 3753463c3197SDave Hansen return -EINVAL; 3754463c3197SDave Hansen } 3755463c3197SDave Hansen } 3756463c3197SDave Hansen 375787c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 375887c4e192SDominik Brodowski unsigned int dev) 37591da177e4SLinus Torvalds { 37601da177e4SLinus Torvalds struct dentry *dentry; 3761dae6ad8fSAl Viro struct path path; 3762dae6ad8fSAl Viro int error; 3763972567f1SJeff Layton unsigned int lookup_flags = 0; 37641da177e4SLinus Torvalds 37658e4bfca1SAl Viro error = may_mknod(mode); 37668e4bfca1SAl Viro if (error) 37678e4bfca1SAl Viro return error; 3768972567f1SJeff Layton retry: 3769972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3770dae6ad8fSAl Viro if (IS_ERR(dentry)) 3771dae6ad8fSAl Viro return PTR_ERR(dentry); 37722ad94ae6SAl Viro 3773dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3774ce3b0f8dSAl Viro mode &= ~current_umask(); 3775dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3776be6d3e56SKentaro Takeda if (error) 3777a8104a9fSAl Viro goto out; 37781da177e4SLinus Torvalds switch (mode & S_IFMT) { 37791da177e4SLinus Torvalds case 0: case S_IFREG: 3780312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 378105d1a717SMimi Zohar if (!error) 378205d1a717SMimi Zohar ima_post_path_mknod(dentry); 37831da177e4SLinus Torvalds break; 37841da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3785dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37861da177e4SLinus Torvalds new_decode_dev(dev)); 37871da177e4SLinus Torvalds break; 37881da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3789dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37901da177e4SLinus Torvalds break; 37911da177e4SLinus Torvalds } 3792a8104a9fSAl Viro out: 3793921a1650SAl Viro done_path_create(&path, dentry); 3794972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3795972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3796972567f1SJeff Layton goto retry; 3797972567f1SJeff Layton } 37981da177e4SLinus Torvalds return error; 37991da177e4SLinus Torvalds } 38001da177e4SLinus Torvalds 380187c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 380287c4e192SDominik Brodowski unsigned int, dev) 380387c4e192SDominik Brodowski { 380487c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 380587c4e192SDominik Brodowski } 380687c4e192SDominik Brodowski 38078208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 38085590ff0dSUlrich Drepper { 380987c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 38105590ff0dSUlrich Drepper } 38115590ff0dSUlrich Drepper 381218bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 38131da177e4SLinus Torvalds { 3814a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38158de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38161da177e4SLinus Torvalds 38171da177e4SLinus Torvalds if (error) 38181da177e4SLinus Torvalds return error; 38191da177e4SLinus Torvalds 3820acfa4380SAl Viro if (!dir->i_op->mkdir) 38211da177e4SLinus Torvalds return -EPERM; 38221da177e4SLinus Torvalds 38231da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38241da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38251da177e4SLinus Torvalds if (error) 38261da177e4SLinus Torvalds return error; 38271da177e4SLinus Torvalds 38288de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38298de52778SAl Viro return -EMLINK; 38308de52778SAl Viro 38311da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3832a74574aaSStephen Smalley if (!error) 3833f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38341da177e4SLinus Torvalds return error; 38351da177e4SLinus Torvalds } 38364d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38371da177e4SLinus Torvalds 38380101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38391da177e4SLinus Torvalds { 38406902d925SDave Hansen struct dentry *dentry; 3841dae6ad8fSAl Viro struct path path; 3842dae6ad8fSAl Viro int error; 3843b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38441da177e4SLinus Torvalds 3845b76d8b82SJeff Layton retry: 3846b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38476902d925SDave Hansen if (IS_ERR(dentry)) 3848dae6ad8fSAl Viro return PTR_ERR(dentry); 38496902d925SDave Hansen 3850dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3851ce3b0f8dSAl Viro mode &= ~current_umask(); 3852dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3853a8104a9fSAl Viro if (!error) 3854dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3855921a1650SAl Viro done_path_create(&path, dentry); 3856b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3857b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3858b76d8b82SJeff Layton goto retry; 3859b76d8b82SJeff Layton } 38601da177e4SLinus Torvalds return error; 38611da177e4SLinus Torvalds } 38621da177e4SLinus Torvalds 38630101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38640101db7aSDominik Brodowski { 38650101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38660101db7aSDominik Brodowski } 38670101db7aSDominik Brodowski 3868a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38695590ff0dSUlrich Drepper { 38700101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38715590ff0dSUlrich Drepper } 38725590ff0dSUlrich Drepper 38731da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38741da177e4SLinus Torvalds { 38751da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38761da177e4SLinus Torvalds 38771da177e4SLinus Torvalds if (error) 38781da177e4SLinus Torvalds return error; 38791da177e4SLinus Torvalds 3880acfa4380SAl Viro if (!dir->i_op->rmdir) 38811da177e4SLinus Torvalds return -EPERM; 38821da177e4SLinus Torvalds 38831d2ef590SAl Viro dget(dentry); 38845955102cSAl Viro inode_lock(dentry->d_inode); 3885912dbc15SSage Weil 38861da177e4SLinus Torvalds error = -EBUSY; 38877af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3888912dbc15SSage Weil goto out; 3889912dbc15SSage Weil 38901da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3891912dbc15SSage Weil if (error) 3892912dbc15SSage Weil goto out; 3893912dbc15SSage Weil 38941da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3895912dbc15SSage Weil if (error) 3896912dbc15SSage Weil goto out; 3897912dbc15SSage Weil 38988767712fSAl Viro shrink_dcache_parent(dentry); 38991da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3900d83c49f3SAl Viro dont_mount(dentry); 39018ed936b5SEric W. Biederman detach_mounts(dentry); 3902116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 39031da177e4SLinus Torvalds 3904912dbc15SSage Weil out: 39055955102cSAl Viro inode_unlock(dentry->d_inode); 39061d2ef590SAl Viro dput(dentry); 3907912dbc15SSage Weil if (!error) 3908912dbc15SSage Weil d_delete(dentry); 39091da177e4SLinus Torvalds return error; 39101da177e4SLinus Torvalds } 39114d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 39121da177e4SLinus Torvalds 3913f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 39141da177e4SLinus Torvalds { 39151da177e4SLinus Torvalds int error = 0; 391691a27b2aSJeff Layton struct filename *name; 39171da177e4SLinus Torvalds struct dentry *dentry; 3918f5beed75SAl Viro struct path path; 3919f5beed75SAl Viro struct qstr last; 3920f5beed75SAl Viro int type; 3921c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3922c6ee9206SJeff Layton retry: 3923c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3924c1d4dd27SAl Viro &path, &last, &type); 392591a27b2aSJeff Layton if (IS_ERR(name)) 392691a27b2aSJeff Layton return PTR_ERR(name); 39271da177e4SLinus Torvalds 3928f5beed75SAl Viro switch (type) { 39291da177e4SLinus Torvalds case LAST_DOTDOT: 39301da177e4SLinus Torvalds error = -ENOTEMPTY; 39311da177e4SLinus Torvalds goto exit1; 39321da177e4SLinus Torvalds case LAST_DOT: 39331da177e4SLinus Torvalds error = -EINVAL; 39341da177e4SLinus Torvalds goto exit1; 39351da177e4SLinus Torvalds case LAST_ROOT: 39361da177e4SLinus Torvalds error = -EBUSY; 39371da177e4SLinus Torvalds goto exit1; 39381da177e4SLinus Torvalds } 39390612d9fbSOGAWA Hirofumi 3940f5beed75SAl Viro error = mnt_want_write(path.mnt); 3941c30dabfeSJan Kara if (error) 3942c30dabfeSJan Kara goto exit1; 39430612d9fbSOGAWA Hirofumi 39445955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3945f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39461da177e4SLinus Torvalds error = PTR_ERR(dentry); 39476902d925SDave Hansen if (IS_ERR(dentry)) 39486902d925SDave Hansen goto exit2; 3949e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3950e6bc45d6STheodore Ts'o error = -ENOENT; 3951e6bc45d6STheodore Ts'o goto exit3; 3952e6bc45d6STheodore Ts'o } 3953f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3954be6d3e56SKentaro Takeda if (error) 3955c30dabfeSJan Kara goto exit3; 3956f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39570622753bSDave Hansen exit3: 39581da177e4SLinus Torvalds dput(dentry); 39596902d925SDave Hansen exit2: 39605955102cSAl Viro inode_unlock(path.dentry->d_inode); 3961f5beed75SAl Viro mnt_drop_write(path.mnt); 39621da177e4SLinus Torvalds exit1: 3963f5beed75SAl Viro path_put(&path); 39641da177e4SLinus Torvalds putname(name); 3965c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3966c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3967c6ee9206SJeff Layton goto retry; 3968c6ee9206SJeff Layton } 39691da177e4SLinus Torvalds return error; 39701da177e4SLinus Torvalds } 39711da177e4SLinus Torvalds 39723cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39735590ff0dSUlrich Drepper { 39745590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39755590ff0dSUlrich Drepper } 39765590ff0dSUlrich Drepper 3977b21996e3SJ. Bruce Fields /** 3978b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3979b21996e3SJ. Bruce Fields * @dir: parent directory 3980b21996e3SJ. Bruce Fields * @dentry: victim 3981b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3982b21996e3SJ. Bruce Fields * 3983b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3984b21996e3SJ. Bruce Fields * 3985b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3986b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3987b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3988b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3989b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3990b21996e3SJ. Bruce Fields * 3991b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3992b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3993b21996e3SJ. Bruce Fields * to be NFS exported. 3994b21996e3SJ. Bruce Fields */ 3995b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39961da177e4SLinus Torvalds { 39979accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39981da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39991da177e4SLinus Torvalds 40001da177e4SLinus Torvalds if (error) 40011da177e4SLinus Torvalds return error; 40021da177e4SLinus Torvalds 4003acfa4380SAl Viro if (!dir->i_op->unlink) 40041da177e4SLinus Torvalds return -EPERM; 40051da177e4SLinus Torvalds 40065955102cSAl Viro inode_lock(target); 40078ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 40081da177e4SLinus Torvalds error = -EBUSY; 40091da177e4SLinus Torvalds else { 40101da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4011bec1052eSAl Viro if (!error) { 40125a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 40135a14696cSJ. Bruce Fields if (error) 4014b21996e3SJ. Bruce Fields goto out; 40151da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 40168ed936b5SEric W. Biederman if (!error) { 4017d83c49f3SAl Viro dont_mount(dentry); 40188ed936b5SEric W. Biederman detach_mounts(dentry); 4019116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 40208ed936b5SEric W. Biederman } 4021bec1052eSAl Viro } 40221da177e4SLinus Torvalds } 4023b21996e3SJ. Bruce Fields out: 40245955102cSAl Viro inode_unlock(target); 40251da177e4SLinus Torvalds 40261da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40271da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40289accbb97SJ. Bruce Fields fsnotify_link_count(target); 40291da177e4SLinus Torvalds d_delete(dentry); 40301da177e4SLinus Torvalds } 40310eeca283SRobert Love 40321da177e4SLinus Torvalds return error; 40331da177e4SLinus Torvalds } 40344d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40351da177e4SLinus Torvalds 40361da177e4SLinus Torvalds /* 40371da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40381b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40391da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40401da177e4SLinus Torvalds * while waiting on the I/O. 40411da177e4SLinus Torvalds */ 4042da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40431da177e4SLinus Torvalds { 40442ad94ae6SAl Viro int error; 40451da177e4SLinus Torvalds struct dentry *dentry; 4046f5beed75SAl Viro struct path path; 4047f5beed75SAl Viro struct qstr last; 4048f5beed75SAl Viro int type; 40491da177e4SLinus Torvalds struct inode *inode = NULL; 4050b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40515d18f813SJeff Layton unsigned int lookup_flags = 0; 40525d18f813SJeff Layton retry: 4053da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 405491a27b2aSJeff Layton if (IS_ERR(name)) 405591a27b2aSJeff Layton return PTR_ERR(name); 40562ad94ae6SAl Viro 40571da177e4SLinus Torvalds error = -EISDIR; 4058f5beed75SAl Viro if (type != LAST_NORM) 40591da177e4SLinus Torvalds goto exit1; 40600612d9fbSOGAWA Hirofumi 4061f5beed75SAl Viro error = mnt_want_write(path.mnt); 4062c30dabfeSJan Kara if (error) 4063c30dabfeSJan Kara goto exit1; 4064b21996e3SJ. Bruce Fields retry_deleg: 40655955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4066f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40671da177e4SLinus Torvalds error = PTR_ERR(dentry); 40681da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40691da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4070f5beed75SAl Viro if (last.name[last.len]) 407150338b88STörök Edwin goto slashes; 40721da177e4SLinus Torvalds inode = dentry->d_inode; 4073b18825a7SDavid Howells if (d_is_negative(dentry)) 4074e6bc45d6STheodore Ts'o goto slashes; 40757de9c6eeSAl Viro ihold(inode); 4076f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4077be6d3e56SKentaro Takeda if (error) 4078c30dabfeSJan Kara goto exit2; 4079f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40801da177e4SLinus Torvalds exit2: 40811da177e4SLinus Torvalds dput(dentry); 40821da177e4SLinus Torvalds } 40835955102cSAl Viro inode_unlock(path.dentry->d_inode); 40841da177e4SLinus Torvalds if (inode) 40851da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4086b21996e3SJ. Bruce Fields inode = NULL; 4087b21996e3SJ. Bruce Fields if (delegated_inode) { 40885a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4089b21996e3SJ. Bruce Fields if (!error) 4090b21996e3SJ. Bruce Fields goto retry_deleg; 4091b21996e3SJ. Bruce Fields } 4092f5beed75SAl Viro mnt_drop_write(path.mnt); 40931da177e4SLinus Torvalds exit1: 4094f5beed75SAl Viro path_put(&path); 40955d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40965d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40975d18f813SJeff Layton inode = NULL; 40985d18f813SJeff Layton goto retry; 40995d18f813SJeff Layton } 4100da2f1362SChristoph Hellwig putname(name); 41011da177e4SLinus Torvalds return error; 41021da177e4SLinus Torvalds 41031da177e4SLinus Torvalds slashes: 4104b18825a7SDavid Howells if (d_is_negative(dentry)) 4105b18825a7SDavid Howells error = -ENOENT; 410644b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4107b18825a7SDavid Howells error = -EISDIR; 4108b18825a7SDavid Howells else 4109b18825a7SDavid Howells error = -ENOTDIR; 41101da177e4SLinus Torvalds goto exit2; 41111da177e4SLinus Torvalds } 41121da177e4SLinus Torvalds 41132e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 41145590ff0dSUlrich Drepper { 41155590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 41165590ff0dSUlrich Drepper return -EINVAL; 41175590ff0dSUlrich Drepper 41185590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 41195590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 41205590ff0dSUlrich Drepper 4121da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 41225590ff0dSUlrich Drepper } 41235590ff0dSUlrich Drepper 41243480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41255590ff0dSUlrich Drepper { 4126da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 41275590ff0dSUlrich Drepper } 41285590ff0dSUlrich Drepper 4129db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 41301da177e4SLinus Torvalds { 4131a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 41321da177e4SLinus Torvalds 41331da177e4SLinus Torvalds if (error) 41341da177e4SLinus Torvalds return error; 41351da177e4SLinus Torvalds 4136acfa4380SAl Viro if (!dir->i_op->symlink) 41371da177e4SLinus Torvalds return -EPERM; 41381da177e4SLinus Torvalds 41391da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41401da177e4SLinus Torvalds if (error) 41411da177e4SLinus Torvalds return error; 41421da177e4SLinus Torvalds 41431da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4144a74574aaSStephen Smalley if (!error) 4145f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41461da177e4SLinus Torvalds return error; 41471da177e4SLinus Torvalds } 41484d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41491da177e4SLinus Torvalds 4150b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 4151b724e846SDominik Brodowski const char __user *newname) 41521da177e4SLinus Torvalds { 41532ad94ae6SAl Viro int error; 415491a27b2aSJeff Layton struct filename *from; 41556902d925SDave Hansen struct dentry *dentry; 4156dae6ad8fSAl Viro struct path path; 4157f46d3567SJeff Layton unsigned int lookup_flags = 0; 41581da177e4SLinus Torvalds 41591da177e4SLinus Torvalds from = getname(oldname); 41601da177e4SLinus Torvalds if (IS_ERR(from)) 41611da177e4SLinus Torvalds return PTR_ERR(from); 4162f46d3567SJeff Layton retry: 4163f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41641da177e4SLinus Torvalds error = PTR_ERR(dentry); 41656902d925SDave Hansen if (IS_ERR(dentry)) 4166dae6ad8fSAl Viro goto out_putname; 41676902d925SDave Hansen 416891a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4169a8104a9fSAl Viro if (!error) 417091a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4171921a1650SAl Viro done_path_create(&path, dentry); 4172f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4173f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4174f46d3567SJeff Layton goto retry; 4175f46d3567SJeff Layton } 41766902d925SDave Hansen out_putname: 41771da177e4SLinus Torvalds putname(from); 41781da177e4SLinus Torvalds return error; 41791da177e4SLinus Torvalds } 41801da177e4SLinus Torvalds 4181b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4182b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4183b724e846SDominik Brodowski { 4184b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4185b724e846SDominik Brodowski } 4186b724e846SDominik Brodowski 41873480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41885590ff0dSUlrich Drepper { 4189b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 41905590ff0dSUlrich Drepper } 41915590ff0dSUlrich Drepper 4192146a8595SJ. Bruce Fields /** 4193146a8595SJ. Bruce Fields * vfs_link - create a new link 4194146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4195146a8595SJ. Bruce Fields * @dir: new parent 4196146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4197146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4198146a8595SJ. Bruce Fields * 4199146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4200146a8595SJ. Bruce Fields * 4201146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4202146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4203146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4204146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4205146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4206146a8595SJ. Bruce Fields * 4207146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4208146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4209146a8595SJ. Bruce Fields * to be NFS exported. 4210146a8595SJ. Bruce Fields */ 4211146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 42121da177e4SLinus Torvalds { 42131da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 42148de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 42151da177e4SLinus Torvalds int error; 42161da177e4SLinus Torvalds 42171da177e4SLinus Torvalds if (!inode) 42181da177e4SLinus Torvalds return -ENOENT; 42191da177e4SLinus Torvalds 4220a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 42211da177e4SLinus Torvalds if (error) 42221da177e4SLinus Torvalds return error; 42231da177e4SLinus Torvalds 42241da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 42251da177e4SLinus Torvalds return -EXDEV; 42261da177e4SLinus Torvalds 42271da177e4SLinus Torvalds /* 42281da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 42291da177e4SLinus Torvalds */ 42301da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 42311da177e4SLinus Torvalds return -EPERM; 42320bd23d09SEric W. Biederman /* 42330bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 42340bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 42350bd23d09SEric W. Biederman * the vfs. 42360bd23d09SEric W. Biederman */ 42370bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 42380bd23d09SEric W. Biederman return -EPERM; 4239acfa4380SAl Viro if (!dir->i_op->link) 42401da177e4SLinus Torvalds return -EPERM; 42417e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42421da177e4SLinus Torvalds return -EPERM; 42431da177e4SLinus Torvalds 42441da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42451da177e4SLinus Torvalds if (error) 42461da177e4SLinus Torvalds return error; 42471da177e4SLinus Torvalds 42485955102cSAl Viro inode_lock(inode); 4249aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4250f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4251aae8a97dSAneesh Kumar K.V error = -ENOENT; 42528de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42538de52778SAl Viro error = -EMLINK; 4254146a8595SJ. Bruce Fields else { 4255146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4256146a8595SJ. Bruce Fields if (!error) 42571da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4258146a8595SJ. Bruce Fields } 4259f4e0c30cSAl Viro 4260f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4261f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4262f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4263f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4264f4e0c30cSAl Viro } 42655955102cSAl Viro inode_unlock(inode); 4266e31e14ecSStephen Smalley if (!error) 42677e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42681da177e4SLinus Torvalds return error; 42691da177e4SLinus Torvalds } 42704d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42711da177e4SLinus Torvalds 42721da177e4SLinus Torvalds /* 42731da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42741da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42751da177e4SLinus Torvalds * newname. --KAB 42761da177e4SLinus Torvalds * 42771da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42781da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42791da177e4SLinus Torvalds * and other special files. --ADM 42801da177e4SLinus Torvalds */ 428146ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 428246ea89ebSDominik Brodowski const char __user *newname, int flags) 42831da177e4SLinus Torvalds { 42841da177e4SLinus Torvalds struct dentry *new_dentry; 4285dae6ad8fSAl Viro struct path old_path, new_path; 4286146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 428711a7b371SAneesh Kumar K.V int how = 0; 42881da177e4SLinus Torvalds int error; 42891da177e4SLinus Torvalds 429011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4291c04030e1SUlrich Drepper return -EINVAL; 429211a7b371SAneesh Kumar K.V /* 4293f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4294f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4295f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 429611a7b371SAneesh Kumar K.V */ 4297f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4298f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4299f0cc6ffbSLinus Torvalds return -ENOENT; 430011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4301f0cc6ffbSLinus Torvalds } 4302c04030e1SUlrich Drepper 430311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 430411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4305442e31caSJeff Layton retry: 430611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 43071da177e4SLinus Torvalds if (error) 43082ad94ae6SAl Viro return error; 43092ad94ae6SAl Viro 4310442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4311442e31caSJeff Layton (how & LOOKUP_REVAL)); 43121da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43136902d925SDave Hansen if (IS_ERR(new_dentry)) 4314dae6ad8fSAl Viro goto out; 4315dae6ad8fSAl Viro 4316dae6ad8fSAl Viro error = -EXDEV; 4317dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4318dae6ad8fSAl Viro goto out_dput; 4319800179c9SKees Cook error = may_linkat(&old_path); 4320800179c9SKees Cook if (unlikely(error)) 4321800179c9SKees Cook goto out_dput; 4322dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4323be6d3e56SKentaro Takeda if (error) 4324a8104a9fSAl Viro goto out_dput; 4325146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 432675c3f29dSDave Hansen out_dput: 4327921a1650SAl Viro done_path_create(&new_path, new_dentry); 4328146a8595SJ. Bruce Fields if (delegated_inode) { 4329146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4330d22e6338SOleg Drokin if (!error) { 4331d22e6338SOleg Drokin path_put(&old_path); 4332146a8595SJ. Bruce Fields goto retry; 4333146a8595SJ. Bruce Fields } 4334d22e6338SOleg Drokin } 4335442e31caSJeff Layton if (retry_estale(error, how)) { 4336d22e6338SOleg Drokin path_put(&old_path); 4337442e31caSJeff Layton how |= LOOKUP_REVAL; 4338442e31caSJeff Layton goto retry; 4339442e31caSJeff Layton } 43401da177e4SLinus Torvalds out: 43412d8f3038SAl Viro path_put(&old_path); 43421da177e4SLinus Torvalds 43431da177e4SLinus Torvalds return error; 43441da177e4SLinus Torvalds } 43451da177e4SLinus Torvalds 434646ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 434746ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 434846ea89ebSDominik Brodowski { 434946ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 435046ea89ebSDominik Brodowski } 435146ea89ebSDominik Brodowski 43523480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43535590ff0dSUlrich Drepper { 435446ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43555590ff0dSUlrich Drepper } 43565590ff0dSUlrich Drepper 4357bc27027aSMiklos Szeredi /** 4358bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4359bc27027aSMiklos Szeredi * @old_dir: parent of source 4360bc27027aSMiklos Szeredi * @old_dentry: source 4361bc27027aSMiklos Szeredi * @new_dir: parent of destination 4362bc27027aSMiklos Szeredi * @new_dentry: destination 4363bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4364520c8b16SMiklos Szeredi * @flags: rename flags 4365bc27027aSMiklos Szeredi * 4366bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4367bc27027aSMiklos Szeredi * 4368bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4369bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4370bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4371bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4372bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4373bc27027aSMiklos Szeredi * so. 4374bc27027aSMiklos Szeredi * 4375bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4376bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4377bc27027aSMiklos Szeredi * to be NFS exported. 4378bc27027aSMiklos Szeredi * 43791da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43801da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43811da177e4SLinus Torvalds * Problems: 43820117d427SMauro Carvalho Chehab * 4383d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43841da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43851da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4386a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43871da177e4SLinus Torvalds * story. 43886cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43896cedba89SJ. Bruce Fields * and source (if it is not a directory). 43901b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43911da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43921da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4393a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43941da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43951da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43961da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4397a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43981da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43991da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 44001da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4401e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 44021b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 44031da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4404c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 44051da177e4SLinus Torvalds * locking]. 44061da177e4SLinus Torvalds */ 44071da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 44088e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4409520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 44101da177e4SLinus Torvalds { 44111da177e4SLinus Torvalds int error; 4412bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4413bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4414bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4415da1ce067SMiklos Szeredi bool new_is_dir = false; 4416da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 441749d31c2fSAl Viro struct name_snapshot old_name; 44181da177e4SLinus Torvalds 44198d3e2936SMiklos Szeredi if (source == target) 44201da177e4SLinus Torvalds return 0; 44211da177e4SLinus Torvalds 44221da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 44231da177e4SLinus Torvalds if (error) 44241da177e4SLinus Torvalds return error; 44251da177e4SLinus Torvalds 4426da1ce067SMiklos Szeredi if (!target) { 4427a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4428da1ce067SMiklos Szeredi } else { 4429da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4430da1ce067SMiklos Szeredi 4431da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44321da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4433da1ce067SMiklos Szeredi else 4434da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4435da1ce067SMiklos Szeredi } 44361da177e4SLinus Torvalds if (error) 44371da177e4SLinus Torvalds return error; 44381da177e4SLinus Torvalds 44392773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 44401da177e4SLinus Torvalds return -EPERM; 44411da177e4SLinus Torvalds 4442bc27027aSMiklos Szeredi /* 4443bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4444bc27027aSMiklos Szeredi * we'll need to flip '..'. 4445bc27027aSMiklos Szeredi */ 4446da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4447da1ce067SMiklos Szeredi if (is_dir) { 4448bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4449bc27027aSMiklos Szeredi if (error) 4450bc27027aSMiklos Szeredi return error; 4451bc27027aSMiklos Szeredi } 4452da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4453da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4454da1ce067SMiklos Szeredi if (error) 4455da1ce067SMiklos Szeredi return error; 4456da1ce067SMiklos Szeredi } 4457da1ce067SMiklos Szeredi } 44580eeca283SRobert Love 44590b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44600b3974ebSMiklos Szeredi flags); 4461bc27027aSMiklos Szeredi if (error) 4462bc27027aSMiklos Szeredi return error; 4463bc27027aSMiklos Szeredi 446449d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4465bc27027aSMiklos Szeredi dget(new_dentry); 4466da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4467bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4468bc27027aSMiklos Szeredi else if (target) 44695955102cSAl Viro inode_lock(target); 4470bc27027aSMiklos Szeredi 4471bc27027aSMiklos Szeredi error = -EBUSY; 44727af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4473bc27027aSMiklos Szeredi goto out; 4474bc27027aSMiklos Szeredi 4475da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4476bc27027aSMiklos Szeredi error = -EMLINK; 4477da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4478bc27027aSMiklos Szeredi goto out; 4479da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4480da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4481da1ce067SMiklos Szeredi goto out; 4482da1ce067SMiklos Szeredi } 4483da1ce067SMiklos Szeredi if (!is_dir) { 4484bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4485bc27027aSMiklos Szeredi if (error) 4486bc27027aSMiklos Szeredi goto out; 4487da1ce067SMiklos Szeredi } 4488da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4489bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4490bc27027aSMiklos Szeredi if (error) 4491bc27027aSMiklos Szeredi goto out; 4492bc27027aSMiklos Szeredi } 4493520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4494520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4495bc27027aSMiklos Szeredi if (error) 4496bc27027aSMiklos Szeredi goto out; 4497bc27027aSMiklos Szeredi 4498da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44998767712fSAl Viro if (is_dir) { 45008767712fSAl Viro shrink_dcache_parent(new_dentry); 4501bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 45028767712fSAl Viro } 4503bc27027aSMiklos Szeredi dont_mount(new_dentry); 45048ed936b5SEric W. Biederman detach_mounts(new_dentry); 4505bc27027aSMiklos Szeredi } 4506da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4507da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4508bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4509da1ce067SMiklos Szeredi else 4510da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4511da1ce067SMiklos Szeredi } 4512bc27027aSMiklos Szeredi out: 4513da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4514bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4515bc27027aSMiklos Szeredi else if (target) 45165955102cSAl Viro inode_unlock(target); 4517bc27027aSMiklos Szeredi dput(new_dentry); 4518da1ce067SMiklos Szeredi if (!error) { 4519f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4520da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4521da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4522f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4523da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4524da1ce067SMiklos Szeredi } 4525da1ce067SMiklos Szeredi } 452649d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 45270eeca283SRobert Love 45281da177e4SLinus Torvalds return error; 45291da177e4SLinus Torvalds } 45304d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 45311da177e4SLinus Torvalds 4532ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4533ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 45341da177e4SLinus Torvalds { 45351da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 45361da177e4SLinus Torvalds struct dentry *trap; 4537f5beed75SAl Viro struct path old_path, new_path; 4538f5beed75SAl Viro struct qstr old_last, new_last; 4539f5beed75SAl Viro int old_type, new_type; 45408e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 454191a27b2aSJeff Layton struct filename *from; 454291a27b2aSJeff Layton struct filename *to; 4543f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4544c6a94284SJeff Layton bool should_retry = false; 45452ad94ae6SAl Viro int error; 4546520c8b16SMiklos Szeredi 45470d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4548da1ce067SMiklos Szeredi return -EINVAL; 4549da1ce067SMiklos Szeredi 45500d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45510d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4552520c8b16SMiklos Szeredi return -EINVAL; 4553520c8b16SMiklos Szeredi 45540d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45550d7a8555SMiklos Szeredi return -EPERM; 45560d7a8555SMiklos Szeredi 4557f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4558f5beed75SAl Viro target_flags = 0; 4559f5beed75SAl Viro 4560c6a94284SJeff Layton retry: 4561c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4562c1d4dd27SAl Viro &old_path, &old_last, &old_type); 456391a27b2aSJeff Layton if (IS_ERR(from)) { 456491a27b2aSJeff Layton error = PTR_ERR(from); 45651da177e4SLinus Torvalds goto exit; 456691a27b2aSJeff Layton } 45671da177e4SLinus Torvalds 4568c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4569c1d4dd27SAl Viro &new_path, &new_last, &new_type); 457091a27b2aSJeff Layton if (IS_ERR(to)) { 457191a27b2aSJeff Layton error = PTR_ERR(to); 45721da177e4SLinus Torvalds goto exit1; 457391a27b2aSJeff Layton } 45741da177e4SLinus Torvalds 45751da177e4SLinus Torvalds error = -EXDEV; 4576f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45771da177e4SLinus Torvalds goto exit2; 45781da177e4SLinus Torvalds 45791da177e4SLinus Torvalds error = -EBUSY; 4580f5beed75SAl Viro if (old_type != LAST_NORM) 45811da177e4SLinus Torvalds goto exit2; 45821da177e4SLinus Torvalds 45830a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45840a7c3937SMiklos Szeredi error = -EEXIST; 4585f5beed75SAl Viro if (new_type != LAST_NORM) 45861da177e4SLinus Torvalds goto exit2; 45871da177e4SLinus Torvalds 4588f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4589c30dabfeSJan Kara if (error) 4590c30dabfeSJan Kara goto exit2; 4591c30dabfeSJan Kara 45928e6d782cSJ. Bruce Fields retry_deleg: 4593f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45941da177e4SLinus Torvalds 4595f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45961da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45971da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45981da177e4SLinus Torvalds goto exit3; 45991da177e4SLinus Torvalds /* source must exist */ 46001da177e4SLinus Torvalds error = -ENOENT; 4601b18825a7SDavid Howells if (d_is_negative(old_dentry)) 46021da177e4SLinus Torvalds goto exit4; 4603f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 46041da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 46051da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 46061da177e4SLinus Torvalds goto exit4; 46070a7c3937SMiklos Szeredi error = -EEXIST; 46080a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 46090a7c3937SMiklos Szeredi goto exit5; 4610da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4611da1ce067SMiklos Szeredi error = -ENOENT; 4612da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4613da1ce067SMiklos Szeredi goto exit5; 4614da1ce067SMiklos Szeredi 4615da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4616da1ce067SMiklos Szeredi error = -ENOTDIR; 4617f5beed75SAl Viro if (new_last.name[new_last.len]) 4618da1ce067SMiklos Szeredi goto exit5; 4619da1ce067SMiklos Szeredi } 4620da1ce067SMiklos Szeredi } 46210a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 46220a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 46230a7c3937SMiklos Szeredi error = -ENOTDIR; 4624f5beed75SAl Viro if (old_last.name[old_last.len]) 46250a7c3937SMiklos Szeredi goto exit5; 4626f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 46270a7c3937SMiklos Szeredi goto exit5; 46280a7c3937SMiklos Szeredi } 46290a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 46300a7c3937SMiklos Szeredi error = -EINVAL; 46310a7c3937SMiklos Szeredi if (old_dentry == trap) 46320a7c3937SMiklos Szeredi goto exit5; 46331da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4634da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46351da177e4SLinus Torvalds error = -ENOTEMPTY; 46361da177e4SLinus Torvalds if (new_dentry == trap) 46371da177e4SLinus Torvalds goto exit5; 46381da177e4SLinus Torvalds 4639f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4640f5beed75SAl Viro &new_path, new_dentry, flags); 4641be6d3e56SKentaro Takeda if (error) 4642c30dabfeSJan Kara goto exit5; 4643f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4644f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4645520c8b16SMiklos Szeredi &delegated_inode, flags); 46461da177e4SLinus Torvalds exit5: 46471da177e4SLinus Torvalds dput(new_dentry); 46481da177e4SLinus Torvalds exit4: 46491da177e4SLinus Torvalds dput(old_dentry); 46501da177e4SLinus Torvalds exit3: 4651f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46528e6d782cSJ. Bruce Fields if (delegated_inode) { 46538e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46548e6d782cSJ. Bruce Fields if (!error) 46558e6d782cSJ. Bruce Fields goto retry_deleg; 46568e6d782cSJ. Bruce Fields } 4657f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46581da177e4SLinus Torvalds exit2: 4659c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4660c6a94284SJeff Layton should_retry = true; 4661f5beed75SAl Viro path_put(&new_path); 46622ad94ae6SAl Viro putname(to); 46631da177e4SLinus Torvalds exit1: 4664f5beed75SAl Viro path_put(&old_path); 46651da177e4SLinus Torvalds putname(from); 4666c6a94284SJeff Layton if (should_retry) { 4667c6a94284SJeff Layton should_retry = false; 4668c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4669c6a94284SJeff Layton goto retry; 4670c6a94284SJeff Layton } 46712ad94ae6SAl Viro exit: 46721da177e4SLinus Torvalds return error; 46731da177e4SLinus Torvalds } 46741da177e4SLinus Torvalds 4675ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4676ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4677ee81feb6SDominik Brodowski { 4678ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4679ee81feb6SDominik Brodowski } 4680ee81feb6SDominik Brodowski 4681520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4682520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4683520c8b16SMiklos Szeredi { 4684ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4685520c8b16SMiklos Szeredi } 4686520c8b16SMiklos Szeredi 4687a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 46885590ff0dSUlrich Drepper { 4689ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 46905590ff0dSUlrich Drepper } 46915590ff0dSUlrich Drepper 4692787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4693787fb6bcSMiklos Szeredi { 4694787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4695787fb6bcSMiklos Szeredi if (error) 4696787fb6bcSMiklos Szeredi return error; 4697787fb6bcSMiklos Szeredi 4698787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4699787fb6bcSMiklos Szeredi return -EPERM; 4700787fb6bcSMiklos Szeredi 4701787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4702787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4703787fb6bcSMiklos Szeredi } 4704787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4705787fb6bcSMiklos Szeredi 47065d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 47071da177e4SLinus Torvalds { 47085d826c84SAl Viro int len = PTR_ERR(link); 47091da177e4SLinus Torvalds if (IS_ERR(link)) 47101da177e4SLinus Torvalds goto out; 47111da177e4SLinus Torvalds 47121da177e4SLinus Torvalds len = strlen(link); 47131da177e4SLinus Torvalds if (len > (unsigned) buflen) 47141da177e4SLinus Torvalds len = buflen; 47151da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 47161da177e4SLinus Torvalds len = -EFAULT; 47171da177e4SLinus Torvalds out: 47181da177e4SLinus Torvalds return len; 47191da177e4SLinus Torvalds } 47201da177e4SLinus Torvalds 4721d60874cdSMiklos Szeredi /** 4722fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4723fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4724fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4725fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4726fd4a0edfSMiklos Szeredi * 4727fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4728fd4a0edfSMiklos Szeredi * 4729fd4a0edfSMiklos Szeredi * Does not call security hook. 4730fd4a0edfSMiklos Szeredi */ 4731fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4732fd4a0edfSMiklos Szeredi { 4733fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4734f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4735f2df5da6SAl Viro const char *link; 4736f2df5da6SAl Viro int res; 4737fd4a0edfSMiklos Szeredi 473876fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 473976fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 474076fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 474176fca90eSMiklos Szeredi 474276fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4743fd4a0edfSMiklos Szeredi return -EINVAL; 4744fd4a0edfSMiklos Szeredi 474576fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 474676fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 474776fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 474876fca90eSMiklos Szeredi } 474976fca90eSMiklos Szeredi 47504c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4751f2df5da6SAl Viro if (!link) { 4752f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4753f2df5da6SAl Viro if (IS_ERR(link)) 4754f2df5da6SAl Viro return PTR_ERR(link); 4755f2df5da6SAl Viro } 4756f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4757f2df5da6SAl Viro do_delayed_call(&done); 4758f2df5da6SAl Viro return res; 4759fd4a0edfSMiklos Szeredi } 4760fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 47611da177e4SLinus Torvalds 4762d60874cdSMiklos Szeredi /** 4763d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4764d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4765d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4766d60874cdSMiklos Szeredi * 4767d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4768d60874cdSMiklos Szeredi * 4769d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4770d60874cdSMiklos Szeredi * 4771d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4772d60874cdSMiklos Szeredi */ 4773d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4774d60874cdSMiklos Szeredi { 4775d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4776d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4777d60874cdSMiklos Szeredi 4778d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4779d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4780d60874cdSMiklos Szeredi if (!res) 4781d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4782d60874cdSMiklos Szeredi } 4783d60874cdSMiklos Szeredi return res; 4784d60874cdSMiklos Szeredi } 4785d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4786d60874cdSMiklos Szeredi 47871da177e4SLinus Torvalds /* get the link contents into pagecache */ 47886b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4789fceef393SAl Viro struct delayed_call *callback) 47901da177e4SLinus Torvalds { 4791ebd09abbSDuane Griffin char *kaddr; 47921da177e4SLinus Torvalds struct page *page; 47936b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47946b255391SAl Viro 4795d3883d4fSAl Viro if (!dentry) { 4796d3883d4fSAl Viro page = find_get_page(mapping, 0); 4797d3883d4fSAl Viro if (!page) 47986b255391SAl Viro return ERR_PTR(-ECHILD); 4799d3883d4fSAl Viro if (!PageUptodate(page)) { 4800d3883d4fSAl Viro put_page(page); 4801d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4802d3883d4fSAl Viro } 4803d3883d4fSAl Viro } else { 4804090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 48051da177e4SLinus Torvalds if (IS_ERR(page)) 48066fe6900eSNick Piggin return (char*)page; 4807d3883d4fSAl Viro } 4808fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 480921fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 481021fc61c7SAl Viro kaddr = page_address(page); 48116b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4812ebd09abbSDuane Griffin return kaddr; 48131da177e4SLinus Torvalds } 48141da177e4SLinus Torvalds 48156b255391SAl Viro EXPORT_SYMBOL(page_get_link); 48161da177e4SLinus Torvalds 4817fceef393SAl Viro void page_put_link(void *arg) 48181da177e4SLinus Torvalds { 4819fceef393SAl Viro put_page(arg); 48201da177e4SLinus Torvalds } 48214d359507SAl Viro EXPORT_SYMBOL(page_put_link); 48221da177e4SLinus Torvalds 4823aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4824aa80deabSAl Viro { 4825fceef393SAl Viro DEFINE_DELAYED_CALL(done); 48266b255391SAl Viro int res = readlink_copy(buffer, buflen, 48276b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4828fceef393SAl Viro &done)); 4829fceef393SAl Viro do_delayed_call(&done); 4830aa80deabSAl Viro return res; 4831aa80deabSAl Viro } 4832aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4833aa80deabSAl Viro 483454566b2cSNick Piggin /* 483554566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 483654566b2cSNick Piggin */ 483754566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48381da177e4SLinus Torvalds { 48391da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48400adb25d2SKirill Korotaev struct page *page; 4841afddba49SNick Piggin void *fsdata; 4842beb497abSDmitriy Monakhov int err; 4843c718a975STetsuo Handa unsigned int flags = 0; 484454566b2cSNick Piggin if (nofs) 484554566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 48461da177e4SLinus Torvalds 48477e53cac4SNeilBrown retry: 4848afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 484954566b2cSNick Piggin flags, &page, &fsdata); 48501da177e4SLinus Torvalds if (err) 4851afddba49SNick Piggin goto fail; 4852afddba49SNick Piggin 485321fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4854afddba49SNick Piggin 4855afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4856afddba49SNick Piggin page, fsdata); 48571da177e4SLinus Torvalds if (err < 0) 48581da177e4SLinus Torvalds goto fail; 4859afddba49SNick Piggin if (err < len-1) 4860afddba49SNick Piggin goto retry; 4861afddba49SNick Piggin 48621da177e4SLinus Torvalds mark_inode_dirty(inode); 48631da177e4SLinus Torvalds return 0; 48641da177e4SLinus Torvalds fail: 48651da177e4SLinus Torvalds return err; 48661da177e4SLinus Torvalds } 48674d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 48681da177e4SLinus Torvalds 48690adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 48700adb25d2SKirill Korotaev { 48710adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4872c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48730adb25d2SKirill Korotaev } 48744d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48750adb25d2SKirill Korotaev 487692e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48776b255391SAl Viro .get_link = page_get_link, 48781da177e4SLinus Torvalds }; 48791da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4880