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; 4949883d185SAl Viro unsigned seq, m_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 { 644ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 645ee594bffSAl Viro return true; 64684a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 647ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 648ee594bffSAl Viro } 649ee594bffSAl Viro 65019660af7SAl Viro /* 65131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 65319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 65457e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 65519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 65619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 65719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 65819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 65931e6b01fSNick Piggin */ 66031e6b01fSNick Piggin 66131e6b01fSNick Piggin /** 66219660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 66319660af7SAl Viro * @nd: nameidata pathwalk data 66439191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 66531e6b01fSNick Piggin * 6664675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6674675ac39SAl Viro * for ref-walk mode. 6684675ac39SAl Viro * Must be called from rcu-walk context. 6697973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6707973387aSAl Viro * terminate_walk(). 67131e6b01fSNick Piggin */ 6724675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 67331e6b01fSNick Piggin { 67431e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 67531e6b01fSNick Piggin 67631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 677e5c832d5SLinus Torvalds 678e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6797973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6804675ac39SAl Viro goto out1; 68184a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 68284a2bd39SAl Viro goto out; 683ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 6844675ac39SAl Viro goto out; 6854675ac39SAl Viro rcu_read_unlock(); 6864675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6874675ac39SAl Viro return 0; 6884675ac39SAl Viro 68984a2bd39SAl Viro out1: 6904675ac39SAl Viro nd->path.mnt = NULL; 6914675ac39SAl Viro nd->path.dentry = NULL; 6924675ac39SAl Viro out: 6934675ac39SAl Viro rcu_read_unlock(); 6944675ac39SAl Viro return -ECHILD; 6954675ac39SAl Viro } 6964675ac39SAl Viro 6974675ac39SAl Viro /** 6984675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 6994675ac39SAl Viro * @nd: nameidata pathwalk data 7004675ac39SAl Viro * @dentry: child of nd->path.dentry 7014675ac39SAl Viro * @seq: seq number to check dentry against 7024675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7034675ac39SAl Viro * 7044675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7054675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7064675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7074675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7084675ac39SAl Viro * terminate_walk(). 7094675ac39SAl Viro */ 7104675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7114675ac39SAl Viro { 7124675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7134675ac39SAl Viro 7144675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7154675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7164675ac39SAl Viro goto out2; 7177973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7187973387aSAl Viro goto out2; 7194675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7207973387aSAl Viro goto out1; 72148a066e7SAl Viro 72215570086SLinus Torvalds /* 7234675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7244675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7254675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7264675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7274675ac39SAl Viro * know the parent must still be valid if the child sequence number is 72815570086SLinus Torvalds */ 7294675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 730e5c832d5SLinus Torvalds goto out; 73184a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 73284a2bd39SAl Viro goto out_dput; 733e5c832d5SLinus Torvalds /* 734e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 735e5c832d5SLinus Torvalds * still valid and get it if required. 736e5c832d5SLinus Torvalds */ 73784a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 73884a2bd39SAl Viro goto out_dput; 7398b61e74fSAl Viro rcu_read_unlock(); 74031e6b01fSNick Piggin return 0; 74119660af7SAl Viro 7427973387aSAl Viro out2: 7437973387aSAl Viro nd->path.mnt = NULL; 7447973387aSAl Viro out1: 7457973387aSAl Viro nd->path.dentry = NULL; 746e5c832d5SLinus Torvalds out: 7478b61e74fSAl Viro rcu_read_unlock(); 74884a2bd39SAl Viro return -ECHILD; 74984a2bd39SAl Viro out_dput: 75084a2bd39SAl Viro rcu_read_unlock(); 75184a2bd39SAl Viro dput(dentry); 75231e6b01fSNick Piggin return -ECHILD; 75331e6b01fSNick Piggin } 75431e6b01fSNick Piggin 7554ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 75634286d66SNick Piggin { 757a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7584ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 759a89f8337SAl Viro else 760a89f8337SAl Viro return 1; 76134286d66SNick Piggin } 76234286d66SNick Piggin 7639f1fafeeSAl Viro /** 7649f1fafeeSAl Viro * complete_walk - successful completion of path walk 7659f1fafeeSAl Viro * @nd: pointer nameidata 76639159de2SJeff Layton * 7679f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7689f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7699f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7709f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7719f1fafeeSAl Viro * need to drop nd->path. 77239159de2SJeff Layton */ 7739f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 77439159de2SJeff Layton { 77516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 77639159de2SJeff Layton int status; 77739159de2SJeff Layton 7789f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7799f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7809f1fafeeSAl Viro nd->root.mnt = NULL; 7814675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 78248a066e7SAl Viro return -ECHILD; 78348a066e7SAl Viro } 7849f1fafeeSAl Viro 78516c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 78639159de2SJeff Layton return 0; 78739159de2SJeff Layton 788ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 78916c2cd71SAl Viro return 0; 79016c2cd71SAl Viro 791ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 79239159de2SJeff Layton if (status > 0) 79339159de2SJeff Layton return 0; 79439159de2SJeff Layton 79516c2cd71SAl Viro if (!status) 79639159de2SJeff Layton status = -ESTALE; 79716c2cd71SAl Viro 79839159de2SJeff Layton return status; 79939159de2SJeff Layton } 80039159de2SJeff Layton 801740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 8022a737871SAl Viro { 80331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8049e6697e2SAl Viro 8059e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8068f47a016SAl Viro unsigned seq; 807c28cc364SNick Piggin 808c28cc364SNick Piggin do { 809c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 81031e6b01fSNick Piggin nd->root = fs->root; 8118f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 812c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8139e6697e2SAl Viro } else { 8149e6697e2SAl Viro get_fs_root(fs, &nd->root); 81584a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 8169e6697e2SAl Viro } 817740a1678SAleksa Sarai return 0; 81831e6b01fSNick Piggin } 81931e6b01fSNick Piggin 8201d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 821051d3812SIan Kent { 822051d3812SIan Kent dput(path->dentry); 8234ac91378SJan Blunck if (path->mnt != nd->path.mnt) 824051d3812SIan Kent mntput(path->mnt); 825051d3812SIan Kent } 826051d3812SIan Kent 8277b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8287b9337aaSNick Piggin struct nameidata *nd) 829051d3812SIan Kent { 83031e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8314ac91378SJan Blunck dput(nd->path.dentry); 83231e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8334ac91378SJan Blunck mntput(nd->path.mnt); 8349a229683SHuang Shijie } 83531e6b01fSNick Piggin nd->path.mnt = path->mnt; 8364ac91378SJan Blunck nd->path.dentry = path->dentry; 837051d3812SIan Kent } 838051d3812SIan Kent 839248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 840248fb5b9SAl Viro { 841740a1678SAleksa Sarai if (!nd->root.mnt) { 842740a1678SAleksa Sarai int error = set_root(nd); 843740a1678SAleksa Sarai if (error) 844740a1678SAleksa Sarai return error; 845740a1678SAleksa Sarai } 846248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 847248fb5b9SAl Viro struct dentry *d; 848248fb5b9SAl Viro nd->path = nd->root; 849248fb5b9SAl Viro d = nd->path.dentry; 850248fb5b9SAl Viro nd->inode = d->d_inode; 851248fb5b9SAl Viro nd->seq = nd->root_seq; 852248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 853248fb5b9SAl Viro return -ECHILD; 854248fb5b9SAl Viro } else { 855248fb5b9SAl Viro path_put(&nd->path); 856248fb5b9SAl Viro nd->path = nd->root; 857248fb5b9SAl Viro path_get(&nd->path); 858248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 859248fb5b9SAl Viro } 860248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 861248fb5b9SAl Viro return 0; 862248fb5b9SAl Viro } 863248fb5b9SAl Viro 864b5fb63c1SChristoph Hellwig /* 8656b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 866b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 867b5fb63c1SChristoph Hellwig */ 8681bc82070SAleksa Sarai int nd_jump_link(struct path *path) 869b5fb63c1SChristoph Hellwig { 8706e77137bSAl Viro struct nameidata *nd = current->nameidata; 871b5fb63c1SChristoph Hellwig path_put(&nd->path); 872b5fb63c1SChristoph Hellwig 873b5fb63c1SChristoph Hellwig nd->path = *path; 874b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 875b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 8761bc82070SAleksa Sarai return 0; 877b5fb63c1SChristoph Hellwig } 878b5fb63c1SChristoph Hellwig 879b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 880574197e0SAl Viro { 88121c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 882fceef393SAl Viro do_delayed_call(&last->done); 8836548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 884b9ff4429SAl Viro path_put(&last->link); 885574197e0SAl Viro } 886574197e0SAl Viro 887561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 888561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 88930aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 89030aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 891800179c9SKees Cook 892800179c9SKees Cook /** 893800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 89455852635SRandy Dunlap * @nd: nameidata pathwalk data 895800179c9SKees Cook * 896800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 897800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 898800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 899800179c9SKees Cook * processes from failing races against path names that may change out 900800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 901800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 902800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 903800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 904800179c9SKees Cook * 905800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 906800179c9SKees Cook */ 907fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 908800179c9SKees Cook { 909800179c9SKees Cook const struct inode *inode; 910800179c9SKees Cook const struct inode *parent; 9112d7f9e2aSSeth Forshee kuid_t puid; 912800179c9SKees Cook 913800179c9SKees Cook if (!sysctl_protected_symlinks) 914800179c9SKees Cook return 0; 915800179c9SKees Cook 916800179c9SKees Cook /* Allowed if owner and follower match. */ 917fceef393SAl Viro inode = nd->link_inode; 91881abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 919800179c9SKees Cook return 0; 920800179c9SKees Cook 921800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 922aa65fa35SAl Viro parent = nd->inode; 923800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 924800179c9SKees Cook return 0; 925800179c9SKees Cook 926800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9272d7f9e2aSSeth Forshee puid = parent->i_uid; 9282d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 929800179c9SKees Cook return 0; 930800179c9SKees Cook 93131956502SAl Viro if (nd->flags & LOOKUP_RCU) 93231956502SAl Viro return -ECHILD; 93331956502SAl Viro 934ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 935245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 936800179c9SKees Cook return -EACCES; 937800179c9SKees Cook } 938800179c9SKees Cook 939800179c9SKees Cook /** 940800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 941800179c9SKees Cook * @inode: the source inode to hardlink from 942800179c9SKees Cook * 943800179c9SKees Cook * Return false if at least one of the following conditions: 944800179c9SKees Cook * - inode is not a regular file 945800179c9SKees Cook * - inode is setuid 946800179c9SKees Cook * - inode is setgid and group-exec 947800179c9SKees Cook * - access failure for read and write 948800179c9SKees Cook * 949800179c9SKees Cook * Otherwise returns true. 950800179c9SKees Cook */ 951800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 952800179c9SKees Cook { 953800179c9SKees Cook umode_t mode = inode->i_mode; 954800179c9SKees Cook 955800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 956800179c9SKees Cook if (!S_ISREG(mode)) 957800179c9SKees Cook return false; 958800179c9SKees Cook 959800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 960800179c9SKees Cook if (mode & S_ISUID) 961800179c9SKees Cook return false; 962800179c9SKees Cook 963800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 964800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 965800179c9SKees Cook return false; 966800179c9SKees Cook 967800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 968800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 969800179c9SKees Cook return false; 970800179c9SKees Cook 971800179c9SKees Cook return true; 972800179c9SKees Cook } 973800179c9SKees Cook 974800179c9SKees Cook /** 975800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 976800179c9SKees Cook * @link: the source to hardlink from 977800179c9SKees Cook * 978800179c9SKees Cook * Block hardlink when all of: 979800179c9SKees Cook * - sysctl_protected_hardlinks enabled 980800179c9SKees Cook * - fsuid does not match inode 981800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 982f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 983800179c9SKees Cook * 984800179c9SKees Cook * Returns 0 if successful, -ve on error. 985800179c9SKees Cook */ 986800179c9SKees Cook static int may_linkat(struct path *link) 987800179c9SKees Cook { 988593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 989593d1ce8SEric W. Biederman 990593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 991593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 992593d1ce8SEric W. Biederman return -EOVERFLOW; 993800179c9SKees Cook 994800179c9SKees Cook if (!sysctl_protected_hardlinks) 995800179c9SKees Cook return 0; 996800179c9SKees Cook 997800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 998800179c9SKees Cook * otherwise, it must be a safe source. 999800179c9SKees Cook */ 1000cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1001800179c9SKees Cook return 0; 1002800179c9SKees Cook 1003245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1004800179c9SKees Cook return -EPERM; 1005800179c9SKees Cook } 1006800179c9SKees Cook 100730aba665SSalvatore Mesoraca /** 100830aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 100930aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 101030aba665SSalvatore Mesoraca * exist. 101130aba665SSalvatore Mesoraca * @dir: the sticky parent directory 101230aba665SSalvatore Mesoraca * @inode: the inode of the file to open 101330aba665SSalvatore Mesoraca * 101430aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 101530aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 101630aba665SSalvatore Mesoraca * - the file already exists 101730aba665SSalvatore Mesoraca * - we are in a sticky directory 101830aba665SSalvatore Mesoraca * - we don't own the file 101930aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 102030aba665SSalvatore Mesoraca * - the directory is world writable 102130aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 102230aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 102330aba665SSalvatore Mesoraca * be enough. 102430aba665SSalvatore Mesoraca * 102530aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 102630aba665SSalvatore Mesoraca */ 102730aba665SSalvatore Mesoraca static int may_create_in_sticky(struct dentry * const dir, 102830aba665SSalvatore Mesoraca struct inode * const inode) 102930aba665SSalvatore Mesoraca { 103030aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 103130aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 103230aba665SSalvatore Mesoraca likely(!(dir->d_inode->i_mode & S_ISVTX)) || 103330aba665SSalvatore Mesoraca uid_eq(inode->i_uid, dir->d_inode->i_uid) || 103430aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 103530aba665SSalvatore Mesoraca return 0; 103630aba665SSalvatore Mesoraca 103730aba665SSalvatore Mesoraca if (likely(dir->d_inode->i_mode & 0002) || 103830aba665SSalvatore Mesoraca (dir->d_inode->i_mode & 0020 && 103930aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 104030aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1041245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1042245d7369SKees Cook "sticky_create_fifo" : 1043245d7369SKees Cook "sticky_create_regular"; 1044245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 104530aba665SSalvatore Mesoraca return -EACCES; 104630aba665SSalvatore Mesoraca } 104730aba665SSalvatore Mesoraca return 0; 104830aba665SSalvatore Mesoraca } 104930aba665SSalvatore Mesoraca 10503b2e7f75SAl Viro static __always_inline 10513b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10521da177e4SLinus Torvalds { 1053ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10541cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1055fceef393SAl Viro struct inode *inode = nd->link_inode; 10566d7b5aaeSAl Viro int error; 10570a959df5SAl Viro const char *res; 10581da177e4SLinus Torvalds 10598fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10608fa9dd24SNeilBrown touch_atime(&last->link); 10618fa9dd24SNeilBrown cond_resched(); 1062c6718543SMiklos Szeredi } else if (atime_needs_update(&last->link, inode)) { 10634675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10648fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10658fa9dd24SNeilBrown touch_atime(&last->link); 10668fa9dd24SNeilBrown } 10678fa9dd24SNeilBrown 1068bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1069bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1070bda0be7aSNeilBrown if (unlikely(error)) 10716920a440SAl Viro return ERR_PTR(error); 107236f3b4f6SAl Viro 107386acdca1SAl Viro nd->last_type = LAST_BIND; 10744c4f7c19SEric Biggers res = READ_ONCE(inode->i_link); 1075d4dee48bSAl Viro if (!res) { 1076fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1077fceef393SAl Viro struct delayed_call *); 1078fceef393SAl Viro get = inode->i_op->get_link; 10798c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1080fceef393SAl Viro res = get(NULL, inode, &last->done); 10816b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10824675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10838c1b4566SAl Viro return ERR_PTR(-ECHILD); 1084fceef393SAl Viro res = get(dentry, inode, &last->done); 10858c1b4566SAl Viro } 10866b255391SAl Viro } else { 1087fceef393SAl Viro res = get(dentry, inode, &last->done); 10886b255391SAl Viro } 1089fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1090fab51e8aSAl Viro return res; 10910a959df5SAl Viro } 1092fab51e8aSAl Viro if (*res == '/') { 1093740a1678SAleksa Sarai error = nd_jump_root(nd); 1094740a1678SAleksa Sarai if (unlikely(error)) 1095740a1678SAleksa Sarai return ERR_PTR(error); 1096fab51e8aSAl Viro while (unlikely(*++res == '/')) 1097fab51e8aSAl Viro ; 1098fab51e8aSAl Viro } 1099fab51e8aSAl Viro if (!*res) 1100fab51e8aSAl Viro res = NULL; 11010a959df5SAl Viro return res; 11020a959df5SAl Viro } 11036d7b5aaeSAl Viro 1104f015f126SDavid Howells /* 1105f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1106f015f126SDavid Howells * 1107f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1108f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1109f015f126SDavid Howells * Up is towards /. 1110f015f126SDavid Howells * 1111f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1112f015f126SDavid Howells * root. 1113f015f126SDavid Howells */ 1114bab77ebfSAl Viro int follow_up(struct path *path) 11151da177e4SLinus Torvalds { 11160714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11170714a533SAl Viro struct mount *parent; 11181da177e4SLinus Torvalds struct dentry *mountpoint; 111999b7db7bSNick Piggin 112048a066e7SAl Viro read_seqlock_excl(&mount_lock); 11210714a533SAl Viro parent = mnt->mnt_parent; 11223c0a6163SAl Viro if (parent == mnt) { 112348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11241da177e4SLinus Torvalds return 0; 11251da177e4SLinus Torvalds } 11260714a533SAl Viro mntget(&parent->mnt); 1127a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 112848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1129bab77ebfSAl Viro dput(path->dentry); 1130bab77ebfSAl Viro path->dentry = mountpoint; 1131bab77ebfSAl Viro mntput(path->mnt); 11320714a533SAl Viro path->mnt = &parent->mnt; 11331da177e4SLinus Torvalds return 1; 11341da177e4SLinus Torvalds } 11354d359507SAl Viro EXPORT_SYMBOL(follow_up); 11361da177e4SLinus Torvalds 1137b5c84bf6SNick Piggin /* 11389875cf80SDavid Howells * Perform an automount 11399875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11409875cf80SDavid Howells * were called with. 11411da177e4SLinus Torvalds */ 1142756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 11439875cf80SDavid Howells bool *need_mntput) 114431e6b01fSNick Piggin { 11459875cf80SDavid Howells struct vfsmount *mnt; 1146ea5b778aSDavid Howells int err; 11479875cf80SDavid Howells 11489875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 11499875cf80SDavid Howells return -EREMOTE; 11509875cf80SDavid Howells 11510ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11520ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11530ec26fd0SMiklos Szeredi * the name. 11540ec26fd0SMiklos Szeredi * 11550ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11565a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11570ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11580ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11590ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11600ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11615a30d8a2SDavid Howells */ 1162756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11635d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11645d38f049SIan Kent path->dentry->d_inode) 11659875cf80SDavid Howells return -EISDIR; 11660ec26fd0SMiklos Szeredi 1167756daf26SNeilBrown nd->total_link_count++; 1168756daf26SNeilBrown if (nd->total_link_count >= 40) 11699875cf80SDavid Howells return -ELOOP; 11709875cf80SDavid Howells 11719875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11729875cf80SDavid Howells if (IS_ERR(mnt)) { 11739875cf80SDavid Howells /* 11749875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11759875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11769875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11779875cf80SDavid Howells * 11789875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11799875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11809875cf80SDavid Howells * the path is inaccessible and we should say so. 11819875cf80SDavid Howells */ 1182756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11839875cf80SDavid Howells return -EREMOTE; 11849875cf80SDavid Howells return PTR_ERR(mnt); 118531e6b01fSNick Piggin } 1186ea5b778aSDavid Howells 11879875cf80SDavid Howells if (!mnt) /* mount collision */ 11889875cf80SDavid Howells return 0; 11899875cf80SDavid Howells 11908aef1884SAl Viro if (!*need_mntput) { 11918aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11928aef1884SAl Viro mntget(path->mnt); 11938aef1884SAl Viro *need_mntput = true; 11948aef1884SAl Viro } 119519a167afSAl Viro err = finish_automount(mnt, path); 1196ea5b778aSDavid Howells 1197ea5b778aSDavid Howells switch (err) { 1198ea5b778aSDavid Howells case -EBUSY: 1199ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 120019a167afSAl Viro return 0; 1201ea5b778aSDavid Howells case 0: 12028aef1884SAl Viro path_put(path); 12039875cf80SDavid Howells path->mnt = mnt; 12049875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 12059875cf80SDavid Howells return 0; 120619a167afSAl Viro default: 120719a167afSAl Viro return err; 12089875cf80SDavid Howells } 120919a167afSAl Viro 1210ea5b778aSDavid Howells } 12119875cf80SDavid Howells 12129875cf80SDavid Howells /* 12139875cf80SDavid Howells * Handle a dentry that is managed in some way. 1214cc53ce53SDavid Howells * - Flagged for transit management (autofs) 12159875cf80SDavid Howells * - Flagged as mountpoint 12169875cf80SDavid Howells * - Flagged as automount point 12179875cf80SDavid Howells * 12189875cf80SDavid Howells * This may only be called in refwalk mode. 1219d41efb52SAl Viro * On success path->dentry is known positive. 12209875cf80SDavid Howells * 12219875cf80SDavid Howells * Serialization is taken care of in namespace.c 12229875cf80SDavid Howells */ 1223756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 12249875cf80SDavid Howells { 12258aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 1226d41efb52SAl Viro unsigned flags; 12279875cf80SDavid Howells bool need_mntput = false; 12288aef1884SAl Viro int ret = 0; 12299875cf80SDavid Howells 12309875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 12319875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 12329875cf80SDavid Howells * the components of that value change under us */ 12332fa6b1e0SAl Viro while (flags = smp_load_acquire(&path->dentry->d_flags), 1234d41efb52SAl Viro unlikely(flags & DCACHE_MANAGED_DENTRY)) { 1235cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1236cc53ce53SDavid Howells * being held. */ 1237d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1238cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1239cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1240fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1241cc53ce53SDavid Howells if (ret < 0) 12428aef1884SAl Viro break; 1243cc53ce53SDavid Howells } 1244cc53ce53SDavid Howells 12459875cf80SDavid Howells /* Transit to a mounted filesystem. */ 1246d41efb52SAl Viro if (flags & DCACHE_MOUNTED) { 12479875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12489875cf80SDavid Howells if (mounted) { 12499875cf80SDavid Howells dput(path->dentry); 12509875cf80SDavid Howells if (need_mntput) 1251463ffb2eSAl Viro mntput(path->mnt); 1252463ffb2eSAl Viro path->mnt = mounted; 1253463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12549875cf80SDavid Howells need_mntput = true; 12559875cf80SDavid Howells continue; 1256463ffb2eSAl Viro } 1257463ffb2eSAl Viro 12589875cf80SDavid Howells /* Something is mounted on this dentry in another 12599875cf80SDavid Howells * namespace and/or whatever was mounted there in this 126048a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 126148a066e7SAl Viro * get it */ 12621da177e4SLinus Torvalds } 12639875cf80SDavid Howells 12649875cf80SDavid Howells /* Handle an automount point */ 1265d41efb52SAl Viro if (flags & DCACHE_NEED_AUTOMOUNT) { 1266756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12679875cf80SDavid Howells if (ret < 0) 12688aef1884SAl Viro break; 12699875cf80SDavid Howells continue; 12709875cf80SDavid Howells } 12719875cf80SDavid Howells 12729875cf80SDavid Howells /* We didn't change the current path point */ 12739875cf80SDavid Howells break; 12749875cf80SDavid Howells } 12758aef1884SAl Viro 12768aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12778aef1884SAl Viro mntput(path->mnt); 12788402752eSAl Viro if (need_mntput) 12798402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 1280d41efb52SAl Viro if (ret == -EISDIR || !ret) 1281d41efb52SAl Viro ret = 1; 1282d41efb52SAl Viro if (ret > 0 && unlikely(d_flags_negative(flags))) 1283d41efb52SAl Viro ret = -ENOENT; 12848402752eSAl Viro if (unlikely(ret < 0)) 12858402752eSAl Viro path_put_conditional(path, nd); 12868402752eSAl Viro return ret; 12871da177e4SLinus Torvalds } 12881da177e4SLinus Torvalds 1289cc53ce53SDavid Howells int follow_down_one(struct path *path) 12901da177e4SLinus Torvalds { 12911da177e4SLinus Torvalds struct vfsmount *mounted; 12921da177e4SLinus Torvalds 12931c755af4SAl Viro mounted = lookup_mnt(path); 12941da177e4SLinus Torvalds if (mounted) { 12959393bd07SAl Viro dput(path->dentry); 12969393bd07SAl Viro mntput(path->mnt); 12979393bd07SAl Viro path->mnt = mounted; 12989393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12991da177e4SLinus Torvalds return 1; 13001da177e4SLinus Torvalds } 13011da177e4SLinus Torvalds return 0; 13021da177e4SLinus Torvalds } 13034d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 13041da177e4SLinus Torvalds 1305fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 130662a7375eSIan Kent { 1307fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1308fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 130962a7375eSIan Kent } 131062a7375eSIan Kent 13119875cf80SDavid Howells /* 1312287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1313287548e4SAl Viro * we meet a managed dentry that would need blocking. 13149875cf80SDavid Howells */ 13159875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1316254cf582SAl Viro struct inode **inode, unsigned *seqp) 13179875cf80SDavid Howells { 131862a7375eSIan Kent for (;;) { 1319c7105365SAl Viro struct mount *mounted; 132062a7375eSIan Kent /* 132162a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 132262a7375eSIan Kent * that wants to block transit. 132362a7375eSIan Kent */ 1324fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1325b8faf035SNeilBrown case -ECHILD: 1326b8faf035SNeilBrown default: 1327ab90911fSDavid Howells return false; 1328b8faf035SNeilBrown case -EISDIR: 1329b8faf035SNeilBrown return true; 1330b8faf035SNeilBrown case 0: 1331b8faf035SNeilBrown break; 1332b8faf035SNeilBrown } 133362a7375eSIan Kent 133462a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1335b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 133662a7375eSIan Kent 1337474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 13389875cf80SDavid Howells if (!mounted) 13399875cf80SDavid Howells break; 1340c7105365SAl Viro path->mnt = &mounted->mnt; 1341c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1342a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1343254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 134459430262SLinus Torvalds /* 134559430262SLinus Torvalds * Update the inode too. We don't need to re-check the 134659430262SLinus Torvalds * dentry sequence number here after this d_inode read, 134759430262SLinus Torvalds * because a mount-point is always pinned. 134859430262SLinus Torvalds */ 134959430262SLinus Torvalds *inode = path->dentry->d_inode; 13509875cf80SDavid Howells } 1351f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1352b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1353287548e4SAl Viro } 1354287548e4SAl Viro 135531e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 135631e6b01fSNick Piggin { 13574023bfc9SAl Viro struct inode *inode = nd->inode; 135831e6b01fSNick Piggin 135931e6b01fSNick Piggin while (1) { 1360aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 136131e6b01fSNick Piggin break; 136231e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 136331e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 136431e6b01fSNick Piggin struct dentry *parent = old->d_parent; 136531e6b01fSNick Piggin unsigned seq; 136631e6b01fSNick Piggin 13674023bfc9SAl Viro inode = parent->d_inode; 136831e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1369aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1370aed434adSAl Viro return -ECHILD; 137131e6b01fSNick Piggin nd->path.dentry = parent; 137231e6b01fSNick Piggin nd->seq = seq; 1373397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 13742b98149cSAleksa Sarai return -ECHILD; 137531e6b01fSNick Piggin break; 1376aed434adSAl Viro } else { 1377aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1378aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1379aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1380aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1381aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1382aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1383aed434adSAl Viro return -ECHILD; 1384aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 138531e6b01fSNick Piggin break; 1386aed434adSAl Viro /* we know that mountpoint was pinned */ 1387aed434adSAl Viro nd->path.dentry = mountpoint; 1388aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1389aed434adSAl Viro inode = inode2; 1390aed434adSAl Viro nd->seq = seq; 139131e6b01fSNick Piggin } 1392aed434adSAl Viro } 1393aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1394b37199e6SAl Viro struct mount *mounted; 1395b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1396aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1397aed434adSAl Viro return -ECHILD; 1398b37199e6SAl Viro if (!mounted) 1399b37199e6SAl Viro break; 1400b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1401b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 14024023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1403b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1404b37199e6SAl Viro } 14054023bfc9SAl Viro nd->inode = inode; 140631e6b01fSNick Piggin return 0; 140731e6b01fSNick Piggin } 140831e6b01fSNick Piggin 14099875cf80SDavid Howells /* 1410cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1411cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1412cc53ce53SDavid Howells * caller is permitted to proceed or not. 1413cc53ce53SDavid Howells */ 14147cc90cc3SAl Viro int follow_down(struct path *path) 1415cc53ce53SDavid Howells { 1416cc53ce53SDavid Howells unsigned managed; 1417cc53ce53SDavid Howells int ret; 1418cc53ce53SDavid Howells 14196aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1420cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1421cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1422cc53ce53SDavid Howells * being held. 1423cc53ce53SDavid Howells * 1424cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1425cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1426cc53ce53SDavid Howells * other than its daemon the right to mount on its 1427cc53ce53SDavid Howells * superstructure. 1428cc53ce53SDavid Howells * 1429cc53ce53SDavid Howells * The filesystem may sleep at this point. 1430cc53ce53SDavid Howells */ 1431cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1432cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1433cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1434fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1435cc53ce53SDavid Howells if (ret < 0) 1436cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1437cc53ce53SDavid Howells } 1438cc53ce53SDavid Howells 1439cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1440cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1441cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1442cc53ce53SDavid Howells if (!mounted) 1443cc53ce53SDavid Howells break; 1444cc53ce53SDavid Howells dput(path->dentry); 1445cc53ce53SDavid Howells mntput(path->mnt); 1446cc53ce53SDavid Howells path->mnt = mounted; 1447cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1448cc53ce53SDavid Howells continue; 1449cc53ce53SDavid Howells } 1450cc53ce53SDavid Howells 1451cc53ce53SDavid Howells /* Don't handle automount points here */ 1452cc53ce53SDavid Howells break; 1453cc53ce53SDavid Howells } 1454cc53ce53SDavid Howells return 0; 1455cc53ce53SDavid Howells } 14564d359507SAl Viro EXPORT_SYMBOL(follow_down); 1457cc53ce53SDavid Howells 1458cc53ce53SDavid Howells /* 14599875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14609875cf80SDavid Howells */ 14619875cf80SDavid Howells static void follow_mount(struct path *path) 14629875cf80SDavid Howells { 14639875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14649875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14659875cf80SDavid Howells if (!mounted) 14669875cf80SDavid Howells break; 14679875cf80SDavid Howells dput(path->dentry); 14689875cf80SDavid Howells mntput(path->mnt); 14699875cf80SDavid Howells path->mnt = mounted; 14709875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14719875cf80SDavid Howells } 14729875cf80SDavid Howells } 14739875cf80SDavid Howells 1474eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1475eedf265aSEric W. Biederman { 1476eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1477eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1478eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1479eedf265aSEric W. Biederman dput(old); 1480eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1481eedf265aSEric W. Biederman return -ENOENT; 1482eedf265aSEric W. Biederman return 0; 1483eedf265aSEric W. Biederman } 1484eedf265aSEric W. Biederman 1485397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14861da177e4SLinus Torvalds { 14871da177e4SLinus Torvalds while(1) { 1488030c7e0bSDanilo Krummrich if (path_equal(&nd->path, &nd->root)) 14891da177e4SLinus Torvalds break; 14904ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1491eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1492eedf265aSEric W. Biederman if (ret) 1493eedf265aSEric W. Biederman return ret; 14941da177e4SLinus Torvalds break; 14951da177e4SLinus Torvalds } 14963088dd70SAl Viro if (!follow_up(&nd->path)) 14971da177e4SLinus Torvalds break; 14981da177e4SLinus Torvalds } 149979ed0226SAl Viro follow_mount(&nd->path); 150031e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1501397d425dSEric W. Biederman return 0; 15021da177e4SLinus Torvalds } 15031da177e4SLinus Torvalds 15041da177e4SLinus Torvalds /* 1505f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1506f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1507baa03890SNick Piggin */ 1508e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1509e3c13928SAl Viro struct dentry *dir, 15106c51e513SAl Viro unsigned int flags) 1511baa03890SNick Piggin { 1512a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1513bad61189SMiklos Szeredi if (dentry) { 1514a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1515bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 151674ff0ffcSAl Viro if (!error) 15175542aa2fSEric W. Biederman d_invalidate(dentry); 1518bad61189SMiklos Szeredi dput(dentry); 151974ff0ffcSAl Viro return ERR_PTR(error); 1520bad61189SMiklos Szeredi } 1521bad61189SMiklos Szeredi } 1522baa03890SNick Piggin return dentry; 1523baa03890SNick Piggin } 1524baa03890SNick Piggin 1525baa03890SNick Piggin /* 1526a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1527a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1528a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1529a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1530a03ece5fSAl Viro * at all. 153144396f4bSJosef Bacik */ 1532a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1533a03ece5fSAl Viro struct dentry *base, unsigned int flags) 153444396f4bSJosef Bacik { 1535a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 153644396f4bSJosef Bacik struct dentry *old; 1537a03ece5fSAl Viro struct inode *dir = base->d_inode; 1538a03ece5fSAl Viro 1539a03ece5fSAl Viro if (dentry) 1540a03ece5fSAl Viro return dentry; 154144396f4bSJosef Bacik 154244396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1543a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 154444396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1545a03ece5fSAl Viro 1546a03ece5fSAl Viro dentry = d_alloc(base, name); 1547a03ece5fSAl Viro if (unlikely(!dentry)) 1548a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 154944396f4bSJosef Bacik 155072bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 155144396f4bSJosef Bacik if (unlikely(old)) { 155244396f4bSJosef Bacik dput(dentry); 155344396f4bSJosef Bacik dentry = old; 155444396f4bSJosef Bacik } 155544396f4bSJosef Bacik return dentry; 155644396f4bSJosef Bacik } 155744396f4bSJosef Bacik 1558e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1559254cf582SAl Viro struct path *path, struct inode **inode, 1560254cf582SAl Viro unsigned *seqp) 15611da177e4SLinus Torvalds { 15624ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 156331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15645a18fff2SAl Viro int status = 1; 15659875cf80SDavid Howells int err; 15669875cf80SDavid Howells 15673cac260aSAl Viro /* 1568b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15695d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15705d0f49c1SAl Viro * going to fall back to non-racy lookup. 1571b04f784eSNick Piggin */ 157231e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 157331e6b01fSNick Piggin unsigned seq; 1574766c4cbfSAl Viro bool negative; 1575da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15765d0f49c1SAl Viro if (unlikely(!dentry)) { 15774675ac39SAl Viro if (unlazy_walk(nd)) 15785d0f49c1SAl Viro return -ECHILD; 1579e9742b53SAl Viro return 0; 15805d0f49c1SAl Viro } 15815a18fff2SAl Viro 158212f8ad4bSLinus Torvalds /* 158312f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 158412f8ad4bSLinus Torvalds * the dentry name information from lookup. 158512f8ad4bSLinus Torvalds */ 158663afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1587766c4cbfSAl Viro negative = d_is_negative(dentry); 15885d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 158912f8ad4bSLinus Torvalds return -ECHILD; 159012f8ad4bSLinus Torvalds 159112f8ad4bSLinus Torvalds /* 159212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 159312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 159412f8ad4bSLinus Torvalds * 159512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 159612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 159712f8ad4bSLinus Torvalds */ 15985d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 159931e6b01fSNick Piggin return -ECHILD; 16005a18fff2SAl Viro 1601254cf582SAl Viro *seqp = seq; 16024ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1603209a7fb2SAl Viro if (likely(status > 0)) { 1604daf3761cSTrond Myklebust /* 1605daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1606daf3761cSTrond Myklebust * case that drops it. 1607daf3761cSTrond Myklebust */ 16085d0f49c1SAl Viro if (unlikely(negative)) 1609daf3761cSTrond Myklebust return -ENOENT; 161031e6b01fSNick Piggin path->mnt = mnt; 161131e6b01fSNick Piggin path->dentry = dentry; 1612254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1613e9742b53SAl Viro return 1; 16145d0f49c1SAl Viro } 16154675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 1616254cf582SAl Viro return -ECHILD; 1617209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1618209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1619209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16205a18fff2SAl Viro } else { 1621e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 162281e6f520SAl Viro if (unlikely(!dentry)) 1623e9742b53SAl Viro return 0; 16244ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16255d0f49c1SAl Viro } 16265a18fff2SAl Viro if (unlikely(status <= 0)) { 1627e9742b53SAl Viro if (!status) 16285d0f49c1SAl Viro d_invalidate(dentry); 16295a18fff2SAl Viro dput(dentry); 16305a18fff2SAl Viro return status; 16315a18fff2SAl Viro } 16325d0f49c1SAl Viro 16331da177e4SLinus Torvalds path->mnt = mnt; 16341da177e4SLinus Torvalds path->dentry = dentry; 1635756daf26SNeilBrown err = follow_managed(path, nd); 1636e9742b53SAl Viro if (likely(err > 0)) 163763afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16388402752eSAl Viro return err; 1639697f514dSMiklos Szeredi } 1640697f514dSMiklos Szeredi 1641697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 164288d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1643e3c13928SAl Viro struct dentry *dir, 1644e3c13928SAl Viro unsigned int flags) 1645697f514dSMiklos Szeredi { 164688d8331aSAl Viro struct dentry *dentry, *old; 16471936386eSAl Viro struct inode *inode = dir->d_inode; 1648d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16491936386eSAl Viro 16501936386eSAl Viro /* Don't go there if it's already dead */ 165194bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 165288d8331aSAl Viro return ERR_PTR(-ENOENT); 165394bdd655SAl Viro again: 1654d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 165594bdd655SAl Viro if (IS_ERR(dentry)) 165688d8331aSAl Viro return dentry; 165794bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1658a89f8337SAl Viro if (!(flags & LOOKUP_NO_REVAL)) { 1659949a852eSAl Viro int error = d_revalidate(dentry, flags); 1660949a852eSAl Viro if (unlikely(error <= 0)) { 166194bdd655SAl Viro if (!error) { 1662949a852eSAl Viro d_invalidate(dentry); 1663949a852eSAl Viro dput(dentry); 166494bdd655SAl Viro goto again; 166594bdd655SAl Viro } 166694bdd655SAl Viro dput(dentry); 1667949a852eSAl Viro dentry = ERR_PTR(error); 1668949a852eSAl Viro } 1669949a852eSAl Viro } 167094bdd655SAl Viro } else { 16711936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 167285c7f810SAl Viro d_lookup_done(dentry); 16731936386eSAl Viro if (unlikely(old)) { 16741936386eSAl Viro dput(dentry); 16751936386eSAl Viro dentry = old; 1676949a852eSAl Viro } 1677949a852eSAl Viro } 1678e3c13928SAl Viro return dentry; 16791da177e4SLinus Torvalds } 16801da177e4SLinus Torvalds 168188d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 168288d8331aSAl Viro struct dentry *dir, 168388d8331aSAl Viro unsigned int flags) 168488d8331aSAl Viro { 168588d8331aSAl Viro struct inode *inode = dir->d_inode; 168688d8331aSAl Viro struct dentry *res; 168788d8331aSAl Viro inode_lock_shared(inode); 168888d8331aSAl Viro res = __lookup_slow(name, dir, flags); 168988d8331aSAl Viro inode_unlock_shared(inode); 169088d8331aSAl Viro return res; 169188d8331aSAl Viro } 169288d8331aSAl Viro 169352094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 169452094c8aSAl Viro { 169552094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16964ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 169752094c8aSAl Viro if (err != -ECHILD) 169852094c8aSAl Viro return err; 16994675ac39SAl Viro if (unlazy_walk(nd)) 170052094c8aSAl Viro return -ECHILD; 170152094c8aSAl Viro } 17024ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 170352094c8aSAl Viro } 170452094c8aSAl Viro 17059856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 17069856fa1bSAl Viro { 17079856fa1bSAl Viro if (type == LAST_DOTDOT) { 1708740a1678SAleksa Sarai int error = 0; 1709740a1678SAleksa Sarai 1710740a1678SAleksa Sarai if (!nd->root.mnt) { 1711740a1678SAleksa Sarai error = set_root(nd); 1712740a1678SAleksa Sarai if (error) 1713740a1678SAleksa Sarai return error; 1714740a1678SAleksa Sarai } 17159856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 171670291aecSAl Viro return follow_dotdot_rcu(nd); 17179856fa1bSAl Viro } else 1718397d425dSEric W. Biederman return follow_dotdot(nd); 17199856fa1bSAl Viro } 17209856fa1bSAl Viro return 0; 17219856fa1bSAl Viro } 17229856fa1bSAl Viro 1723181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1724181548c0SAl Viro struct inode *inode, unsigned seq) 1725d63ff28fSAl Viro { 1726626de996SAl Viro int error; 17271cf2665bSAl Viro struct saved *last; 1728756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1729626de996SAl Viro path_to_nameidata(link, nd); 1730626de996SAl Viro return -ELOOP; 1731626de996SAl Viro } 1732bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1733cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1734cd179f44SAl Viro mntget(link->mnt); 17357973387aSAl Viro } 1736626de996SAl Viro error = nd_alloc_stack(nd); 1737626de996SAl Viro if (unlikely(error)) { 1738bc40aee0SAl Viro if (error == -ECHILD) { 1739ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1740ad1633a1SAl Viro drop_links(nd); 1741ad1633a1SAl Viro nd->depth = 0; 1742ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1743ad1633a1SAl Viro nd->path.mnt = NULL; 1744ad1633a1SAl Viro nd->path.dentry = NULL; 1745ad1633a1SAl Viro rcu_read_unlock(); 17464675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1747bc40aee0SAl Viro error = nd_alloc_stack(nd); 1748bc40aee0SAl Viro } 1749bc40aee0SAl Viro if (error) { 1750cd179f44SAl Viro path_put(link); 1751626de996SAl Viro return error; 1752626de996SAl Viro } 1753bc40aee0SAl Viro } 1754626de996SAl Viro 1755ab104923SAl Viro last = nd->stack + nd->depth++; 17561cf2665bSAl Viro last->link = *link; 1757fceef393SAl Viro clear_delayed_call(&last->done); 1758fceef393SAl Viro nd->link_inode = inode; 17590450b2d1SAl Viro last->seq = seq; 1760d63ff28fSAl Viro return 1; 1761d63ff28fSAl Viro } 1762d63ff28fSAl Viro 17638f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 176431d66bcdSAl Viro 17653ddcd056SLinus Torvalds /* 17663ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17673ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17683ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17693ddcd056SLinus Torvalds * for the common case. 17703ddcd056SLinus Torvalds */ 17718f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 17728f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 17733ddcd056SLinus Torvalds { 177431d66bcdSAl Viro if (!(flags & WALK_MORE) && nd->depth) 177531d66bcdSAl Viro put_link(nd); 17768f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 17778f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 17788f64fb1cSAl Viro /* not a symlink or should not follow */ 17798f64fb1cSAl Viro path_to_nameidata(path, nd); 17808f64fb1cSAl Viro nd->inode = inode; 17818f64fb1cSAl Viro nd->seq = seq; 1782d63ff28fSAl Viro return 0; 17838f64fb1cSAl Viro } 1784a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1785a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 17868f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1787a7f77542SAl Viro return -ECHILD; 1788a7f77542SAl Viro } 17898f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 17903ddcd056SLinus Torvalds } 17913ddcd056SLinus Torvalds 17924693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1793ce57dfc1SAl Viro { 1794caa85634SAl Viro struct path path; 1795ce57dfc1SAl Viro struct inode *inode; 1796254cf582SAl Viro unsigned seq; 1797ce57dfc1SAl Viro int err; 1798ce57dfc1SAl Viro /* 1799ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1800ce57dfc1SAl Viro * to be able to know about the current root directory and 1801ce57dfc1SAl Viro * parent relationships. 1802ce57dfc1SAl Viro */ 18034693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18044693a547SAl Viro err = handle_dots(nd, nd->last_type); 18051c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18064693a547SAl Viro put_link(nd); 18074693a547SAl Viro return err; 18084693a547SAl Viro } 1809254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1810e9742b53SAl Viro if (unlikely(err <= 0)) { 1811697f514dSMiklos Szeredi if (err < 0) 1812f0a9ba70SAl Viro return err; 1813e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1814e3c13928SAl Viro nd->flags); 1815e3c13928SAl Viro if (IS_ERR(path.dentry)) 1816e3c13928SAl Viro return PTR_ERR(path.dentry); 18177500c38aSAl Viro 1818e3c13928SAl Viro path.mnt = nd->path.mnt; 1819e3c13928SAl Viro err = follow_managed(&path, nd); 1820e3c13928SAl Viro if (unlikely(err < 0)) 1821f0a9ba70SAl Viro return err; 1822697f514dSMiklos Szeredi 1823254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1824d4565649SAl Viro inode = d_backing_inode(path.dentry); 1825766c4cbfSAl Viro } 1826697f514dSMiklos Szeredi 18278f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1828ce57dfc1SAl Viro } 1829ce57dfc1SAl Viro 18301da177e4SLinus Torvalds /* 1831bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1832bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1833bfcfaa77SLinus Torvalds * 1834bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1835bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1836bfcfaa77SLinus Torvalds * fast. 1837bfcfaa77SLinus Torvalds * 1838bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1839bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1840bfcfaa77SLinus Torvalds * crossing operation. 1841bfcfaa77SLinus Torvalds * 1842bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1843bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1844bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1845bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1846bfcfaa77SLinus Torvalds */ 1847bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1848bfcfaa77SLinus Torvalds 1849f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1850bfcfaa77SLinus Torvalds 1851468a9428SGeorge Spelvin #ifdef HASH_MIX 1852bfcfaa77SLinus Torvalds 1853468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1854468a9428SGeorge Spelvin 1855468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18562a18da7aSGeorge Spelvin /* 18572a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18582a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18592a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18602a18da7aSGeorge Spelvin * and no temporaries. 18612a18da7aSGeorge Spelvin * 18622a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18632a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18642a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18652a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18662a18da7aSGeorge Spelvin * 18672a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18682a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18692a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18702a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18712a18da7aSGeorge Spelvin * 18722a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18732a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18742a18da7aSGeorge Spelvin * toggling any given output bit. 18752a18da7aSGeorge Spelvin * 18762a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18772a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18782a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18792a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18802a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18812a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18822a18da7aSGeorge Spelvin * Perfect: 8192 258048 18832a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18842a18da7aSGeorge Spelvin */ 18852a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18862a18da7aSGeorge Spelvin ( x ^= (a), \ 18872a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18882a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 18892a18da7aSGeorge Spelvin y *= 9 ) 1890bfcfaa77SLinus Torvalds 18910fed3ac8SGeorge Spelvin /* 18922a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 18932a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 18942a18da7aSGeorge Spelvin * work done before the hash value is used. 18950fed3ac8SGeorge Spelvin */ 18962a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18970fed3ac8SGeorge Spelvin { 18982a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 18992a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19002a18da7aSGeorge Spelvin return y >> 32; 19010fed3ac8SGeorge Spelvin } 19020fed3ac8SGeorge Spelvin 1903bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1904bfcfaa77SLinus Torvalds 19052a18da7aSGeorge Spelvin /* 19062a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19072a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19082a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19092a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19102a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19112a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19122a18da7aSGeorge Spelvin * Perfect: 2048 31744 19132a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19142a18da7aSGeorge Spelvin */ 19152a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19162a18da7aSGeorge Spelvin ( x ^= (a), \ 19172a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19182a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19192a18da7aSGeorge Spelvin y *= 9 ) 1920bfcfaa77SLinus Torvalds 19212a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19220fed3ac8SGeorge Spelvin { 19232a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19242a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19250fed3ac8SGeorge Spelvin } 19260fed3ac8SGeorge Spelvin 1927bfcfaa77SLinus Torvalds #endif 1928bfcfaa77SLinus Torvalds 19292a18da7aSGeorge Spelvin /* 19302a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19312a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19322a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19332a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19342a18da7aSGeorge Spelvin * finds the delimiter after the name. 19352a18da7aSGeorge Spelvin */ 19368387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1937bfcfaa77SLinus Torvalds { 19388387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1939bfcfaa77SLinus Torvalds 1940bfcfaa77SLinus Torvalds for (;;) { 1941fcfd2fbfSGeorge Spelvin if (!len) 1942fcfd2fbfSGeorge Spelvin goto done; 1943e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1944bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1945bfcfaa77SLinus Torvalds break; 19462a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1947bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1948bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1949bfcfaa77SLinus Torvalds } 19502a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1951bfcfaa77SLinus Torvalds done: 19522a18da7aSGeorge Spelvin return fold_hash(x, y); 1953bfcfaa77SLinus Torvalds } 1954bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1955bfcfaa77SLinus Torvalds 1956fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19578387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1958fcfd2fbfSGeorge Spelvin { 19598387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19608387ff25SLinus Torvalds unsigned long adata, mask, len; 1961fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1962fcfd2fbfSGeorge Spelvin 19638387ff25SLinus Torvalds len = 0; 19648387ff25SLinus Torvalds goto inside; 19658387ff25SLinus Torvalds 1966fcfd2fbfSGeorge Spelvin do { 19672a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1968fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19698387ff25SLinus Torvalds inside: 1970fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1971fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1972fcfd2fbfSGeorge Spelvin 1973fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1974fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19752a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1976fcfd2fbfSGeorge Spelvin 19772a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1978fcfd2fbfSGeorge Spelvin } 1979fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1980fcfd2fbfSGeorge Spelvin 1981bfcfaa77SLinus Torvalds /* 1982bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1983d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1984bfcfaa77SLinus Torvalds */ 19858387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1986bfcfaa77SLinus Torvalds { 19878387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 19888387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 198936126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1990bfcfaa77SLinus Torvalds 19918387ff25SLinus Torvalds len = 0; 19928387ff25SLinus Torvalds goto inside; 19938387ff25SLinus Torvalds 1994bfcfaa77SLinus Torvalds do { 19952a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1996bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 19978387ff25SLinus Torvalds inside: 1998e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 199936126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 200036126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2001bfcfaa77SLinus Torvalds 200236126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 200336126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 200436126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20052a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 200636126f8fSLinus Torvalds 20072a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2008bfcfaa77SLinus Torvalds } 2009bfcfaa77SLinus Torvalds 20102a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2011bfcfaa77SLinus Torvalds 2012fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20138387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20140145acc2SLinus Torvalds { 20158387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20160145acc2SLinus Torvalds while (len--) 2017fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20180145acc2SLinus Torvalds return end_name_hash(hash); 20190145acc2SLinus Torvalds } 2020ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20210145acc2SLinus Torvalds 2022fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20238387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2024fcfd2fbfSGeorge Spelvin { 20258387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2026fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2027fcfd2fbfSGeorge Spelvin 2028fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2029e0ab7af9SGeorge Spelvin while (c) { 2030fcfd2fbfSGeorge Spelvin len++; 2031fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2032fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2033e0ab7af9SGeorge Spelvin } 2034fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2035fcfd2fbfSGeorge Spelvin } 2036f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2037fcfd2fbfSGeorge Spelvin 20383ddcd056SLinus Torvalds /* 2039200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2040200e9ef7SLinus Torvalds * one character. 2041200e9ef7SLinus Torvalds */ 20428387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2043200e9ef7SLinus Torvalds { 20448387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2045200e9ef7SLinus Torvalds unsigned long len = 0, c; 2046200e9ef7SLinus Torvalds 2047200e9ef7SLinus Torvalds c = (unsigned char)*name; 2048200e9ef7SLinus Torvalds do { 2049200e9ef7SLinus Torvalds len++; 2050200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2051200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2052200e9ef7SLinus Torvalds } while (c && c != '/'); 2053d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2054200e9ef7SLinus Torvalds } 2055200e9ef7SLinus Torvalds 2056bfcfaa77SLinus Torvalds #endif 2057bfcfaa77SLinus Torvalds 2058200e9ef7SLinus Torvalds /* 20591da177e4SLinus Torvalds * Name resolution. 2060ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2061ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20621da177e4SLinus Torvalds * 2063ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2064ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20651da177e4SLinus Torvalds */ 20666de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20671da177e4SLinus Torvalds { 20681da177e4SLinus Torvalds int err; 20691da177e4SLinus Torvalds 20709b5858e9SAl Viro if (IS_ERR(name)) 20719b5858e9SAl Viro return PTR_ERR(name); 20721da177e4SLinus Torvalds while (*name=='/') 20731da177e4SLinus Torvalds name++; 20741da177e4SLinus Torvalds if (!*name) 20759e18f10aSAl Viro return 0; 20761da177e4SLinus Torvalds 20771da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20781da177e4SLinus Torvalds for(;;) { 2079d6bb3e90SLinus Torvalds u64 hash_len; 2080fe479a58SAl Viro int type; 20811da177e4SLinus Torvalds 208252094c8aSAl Viro err = may_lookup(nd); 20831da177e4SLinus Torvalds if (err) 20843595e234SAl Viro return err; 20851da177e4SLinus Torvalds 20868387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 20871da177e4SLinus Torvalds 2088fe479a58SAl Viro type = LAST_NORM; 2089d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2090fe479a58SAl Viro case 2: 2091200e9ef7SLinus Torvalds if (name[1] == '.') { 2092fe479a58SAl Viro type = LAST_DOTDOT; 209316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 209416c2cd71SAl Viro } 2095fe479a58SAl Viro break; 2096fe479a58SAl Viro case 1: 2097fe479a58SAl Viro type = LAST_DOT; 2098fe479a58SAl Viro } 20995a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21005a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 210116c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21025a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2103a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2104da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21055a202bcdSAl Viro if (err < 0) 21063595e234SAl Viro return err; 2107d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2108d6bb3e90SLinus Torvalds name = this.name; 21095a202bcdSAl Viro } 21105a202bcdSAl Viro } 2111fe479a58SAl Viro 2112d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2113d6bb3e90SLinus Torvalds nd->last.name = name; 21145f4a6a69SAl Viro nd->last_type = type; 21155f4a6a69SAl Viro 2116d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2117d6bb3e90SLinus Torvalds if (!*name) 2118bdf6cbf1SAl Viro goto OK; 2119200e9ef7SLinus Torvalds /* 2120200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2121200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2122200e9ef7SLinus Torvalds */ 2123200e9ef7SLinus Torvalds do { 2124d6bb3e90SLinus Torvalds name++; 2125d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21268620c238SAl Viro if (unlikely(!*name)) { 21278620c238SAl Viro OK: 2128368ee9baSAl Viro /* pathname body, done */ 21298620c238SAl Viro if (!nd->depth) 21308620c238SAl Viro return 0; 21318620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2132368ee9baSAl Viro /* trailing symlink, done */ 21338620c238SAl Viro if (!name) 21348620c238SAl Viro return 0; 21358620c238SAl Viro /* last component of nested symlink */ 21368f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 21371c4ff1a8SAl Viro } else { 21381c4ff1a8SAl Viro /* not the last component */ 21398f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 21408620c238SAl Viro } 2141ce57dfc1SAl Viro if (err < 0) 21423595e234SAl Viro return err; 2143fe479a58SAl Viro 2144ce57dfc1SAl Viro if (err) { 2145626de996SAl Viro const char *s = get_link(nd); 21465a460275SAl Viro 2147a1c83681SViresh Kumar if (IS_ERR(s)) 21483595e234SAl Viro return PTR_ERR(s); 2149172a39a0SAl Viro err = 0; 215012b09578SAl Viro if (unlikely(!s)) { 215112b09578SAl Viro /* jumped */ 2152b9ff4429SAl Viro put_link(nd); 215312b09578SAl Viro } else { 2154071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 215532cd7468SAl Viro name = s; 21569e18f10aSAl Viro continue; 215748c8b0c5SAl Viro } 215831e6b01fSNick Piggin } 215997242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 216097242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21614675ac39SAl Viro if (unlazy_walk(nd)) 216297242f99SAl Viro return -ECHILD; 216397242f99SAl Viro } 21643595e234SAl Viro return -ENOTDIR; 21655f4a6a69SAl Viro } 2166ce57dfc1SAl Viro } 216797242f99SAl Viro } 21681da177e4SLinus Torvalds 2169edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2170c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 217131e6b01fSNick Piggin { 2172740a1678SAleksa Sarai int error; 2173c8a53ee5SAl Viro const char *s = nd->name->name; 217431e6b01fSNick Piggin 2175c0eb027eSLinus Torvalds if (!*s) 2176c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2177edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2178edc2b1daSAl Viro rcu_read_lock(); 2179c0eb027eSLinus Torvalds 218031e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2181980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 218231e6b01fSNick Piggin nd->depth = 0; 21835b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2184b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2185b18825a7SDavid Howells struct inode *inode = root->d_inode; 218693893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2187368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21885b6ca027SAl Viro nd->path = nd->root; 21895b6ca027SAl Viro nd->inode = inode; 21905b6ca027SAl Viro if (flags & LOOKUP_RCU) { 21915b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 21928f47a016SAl Viro nd->root_seq = nd->seq; 219348a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 21945b6ca027SAl Viro } else { 21955b6ca027SAl Viro path_get(&nd->path); 21965b6ca027SAl Viro } 2197368ee9baSAl Viro return s; 21985b6ca027SAl Viro } 21995b6ca027SAl Viro 220031e6b01fSNick Piggin nd->root.mnt = NULL; 2201248fb5b9SAl Viro nd->path.mnt = NULL; 2202248fb5b9SAl Viro nd->path.dentry = NULL; 220331e6b01fSNick Piggin 220448a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2205740a1678SAleksa Sarai 2206740a1678SAleksa Sarai /* Figure out the starting path and root (if needed). */ 2207fd2f7cb5SAl Viro if (*s == '/') { 2208740a1678SAleksa Sarai error = nd_jump_root(nd); 2209740a1678SAleksa Sarai if (unlikely(error)) 2210740a1678SAleksa Sarai return ERR_PTR(error); 2211ef55d917SAl Viro return s; 2212c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2213e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 221431e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2215c28cc364SNick Piggin unsigned seq; 221631e6b01fSNick Piggin 2217c28cc364SNick Piggin do { 2218c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 221931e6b01fSNick Piggin nd->path = fs->pwd; 2220ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2221c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2222c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2223e41f7d4eSAl Viro } else { 2224e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2225ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2226e41f7d4eSAl Viro } 2227ef55d917SAl Viro return s; 222831e6b01fSNick Piggin } else { 2229582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2230c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 223131e6b01fSNick Piggin struct dentry *dentry; 223231e6b01fSNick Piggin 22332903ff01SAl Viro if (!f.file) 2234368ee9baSAl Viro return ERR_PTR(-EBADF); 223531e6b01fSNick Piggin 22362903ff01SAl Viro dentry = f.file->f_path.dentry; 223731e6b01fSNick Piggin 2238edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22392903ff01SAl Viro fdput(f); 2240368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2241f52e0c11SAl Viro } 22422903ff01SAl Viro 22432903ff01SAl Viro nd->path = f.file->f_path; 2244e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 224534a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 224634a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22475590ff0dSUlrich Drepper } else { 22482903ff01SAl Viro path_get(&nd->path); 224934a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22501da177e4SLinus Torvalds } 225134a26b99SAl Viro fdput(f); 2252368ee9baSAl Viro return s; 2253e41f7d4eSAl Viro } 22549b4a9b14SAl Viro } 22559b4a9b14SAl Viro 22563bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 225795fa25d9SAl Viro { 225895fa25d9SAl Viro const char *s; 2259fec2fa24SAl Viro int error = may_follow_link(nd); 2260deb106c6SAl Viro if (unlikely(error)) 22613bdba28bSAl Viro return ERR_PTR(error); 226295fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2263fab51e8aSAl Viro nd->stack[0].name = NULL; 22643b2e7f75SAl Viro s = get_link(nd); 2265deb106c6SAl Viro return s ? s : ""; 226695fa25d9SAl Viro } 226795fa25d9SAl Viro 2268caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2269bd92d7feSAl Viro { 2270bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2271bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2272bd92d7feSAl Viro 2273bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 22741c4ff1a8SAl Viro return walk_component(nd, 0); 2275bd92d7feSAl Viro } 2276bd92d7feSAl Viro 22774f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22784f757f3cSAl Viro { 22794f757f3cSAl Viro struct path path = nd->path; 22804f757f3cSAl Viro struct inode *inode = nd->inode; 22814f757f3cSAl Viro unsigned seq = nd->seq; 22824f757f3cSAl Viro int err; 22834f757f3cSAl Viro 22844f757f3cSAl Viro if (nd->flags & LOOKUP_RCU) { 22854f757f3cSAl Viro /* 22864f757f3cSAl Viro * don't bother with unlazy_walk on failure - we are 22874f757f3cSAl Viro * at the very beginning of walk, so we lose nothing 22884f757f3cSAl Viro * if we simply redo everything in non-RCU mode 22894f757f3cSAl Viro */ 22904f757f3cSAl Viro if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq))) 22914f757f3cSAl Viro return -ECHILD; 22924f757f3cSAl Viro } else { 22934f757f3cSAl Viro dget(path.dentry); 22944f757f3cSAl Viro err = follow_managed(&path, nd); 22954f757f3cSAl Viro if (unlikely(err < 0)) 22964f757f3cSAl Viro return err; 22974f757f3cSAl Viro inode = d_backing_inode(path.dentry); 22984f757f3cSAl Viro seq = 0; 22994f757f3cSAl Viro } 23004f757f3cSAl Viro path_to_nameidata(&path, nd); 23014f757f3cSAl Viro nd->inode = inode; 23024f757f3cSAl Viro nd->seq = seq; 23034f757f3cSAl Viro return 0; 23044f757f3cSAl Viro } 23054f757f3cSAl Viro 23069b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2307c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23089b4a9b14SAl Viro { 2309c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2310bd92d7feSAl Viro int err; 231131e6b01fSNick Piggin 23129b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23134f757f3cSAl Viro err = handle_lookup_down(nd); 23145f336e72SAl Viro if (unlikely(err < 0)) 23155f336e72SAl Viro s = ERR_PTR(err); 23164f757f3cSAl Viro } 23174f757f3cSAl Viro 23183bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 23193bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 23203bdba28bSAl Viro s = trailing_symlink(nd); 2321bd92d7feSAl Viro } 23229f1fafeeSAl Viro if (!err) 23239f1fafeeSAl Viro err = complete_walk(nd); 2324bd92d7feSAl Viro 2325deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2326deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2327bd23a539SAl Viro err = -ENOTDIR; 2328625b6d10SAl Viro if (!err) { 2329625b6d10SAl Viro *path = nd->path; 2330625b6d10SAl Viro nd->path.mnt = NULL; 2331625b6d10SAl Viro nd->path.dentry = NULL; 2332625b6d10SAl Viro } 2333deb106c6SAl Viro terminate_walk(nd); 2334bd92d7feSAl Viro return err; 233531e6b01fSNick Piggin } 233631e6b01fSNick Piggin 233731d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23389ad1aaa6SAl Viro struct path *path, struct path *root) 2339873f1eedSJeff Layton { 2340894bc8c4SAl Viro int retval; 23419883d185SAl Viro struct nameidata nd; 2342abc9f5beSAl Viro if (IS_ERR(name)) 2343abc9f5beSAl Viro return PTR_ERR(name); 23449ad1aaa6SAl Viro if (unlikely(root)) { 23459ad1aaa6SAl Viro nd.root = *root; 23469ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23479ad1aaa6SAl Viro } 23489883d185SAl Viro set_nameidata(&nd, dfd, name); 2349c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2350873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2351c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2352873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2353c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2354873f1eedSJeff Layton 2355873f1eedSJeff Layton if (likely(!retval)) 2356ff0ebee2SAl Viro audit_inode(name, path->dentry, 0); 23579883d185SAl Viro restore_nameidata(); 2358e4bd1c1aSAl Viro putname(name); 2359873f1eedSJeff Layton return retval; 2360873f1eedSJeff Layton } 2361873f1eedSJeff Layton 23628bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2363c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2364391172c4SAl Viro struct path *parent) 23658bcb77faSAl Viro { 2366c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23679b5858e9SAl Viro int err = link_path_walk(s, nd); 23688bcb77faSAl Viro if (!err) 23698bcb77faSAl Viro err = complete_walk(nd); 2370391172c4SAl Viro if (!err) { 2371391172c4SAl Viro *parent = nd->path; 2372391172c4SAl Viro nd->path.mnt = NULL; 2373391172c4SAl Viro nd->path.dentry = NULL; 2374391172c4SAl Viro } 2375deb106c6SAl Viro terminate_walk(nd); 23768bcb77faSAl Viro return err; 23778bcb77faSAl Viro } 23788bcb77faSAl Viro 23795c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2380391172c4SAl Viro unsigned int flags, struct path *parent, 2381391172c4SAl Viro struct qstr *last, int *type) 23828bcb77faSAl Viro { 23838bcb77faSAl Viro int retval; 23849883d185SAl Viro struct nameidata nd; 23858bcb77faSAl Viro 23865c31b6ceSAl Viro if (IS_ERR(name)) 23875c31b6ceSAl Viro return name; 23889883d185SAl Viro set_nameidata(&nd, dfd, name); 2389c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23908bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2391c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23928bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2393c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2394391172c4SAl Viro if (likely(!retval)) { 2395391172c4SAl Viro *last = nd.last; 2396391172c4SAl Viro *type = nd.last_type; 2397c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 23985c31b6ceSAl Viro } else { 23995c31b6ceSAl Viro putname(name); 24005c31b6ceSAl Viro name = ERR_PTR(retval); 2401391172c4SAl Viro } 24029883d185SAl Viro restore_nameidata(); 24035c31b6ceSAl Viro return name; 24048bcb77faSAl Viro } 24058bcb77faSAl Viro 240679714f72SAl Viro /* does lookup, returns the object with parent locked */ 240779714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24085590ff0dSUlrich Drepper { 24095c31b6ceSAl Viro struct filename *filename; 24105c31b6ceSAl Viro struct dentry *d; 2411391172c4SAl Viro struct qstr last; 2412391172c4SAl Viro int type; 241351689104SPaul Moore 24145c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24155c31b6ceSAl Viro &last, &type); 241651689104SPaul Moore if (IS_ERR(filename)) 241751689104SPaul Moore return ERR_CAST(filename); 24185c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2419391172c4SAl Viro path_put(path); 24205c31b6ceSAl Viro putname(filename); 24215c31b6ceSAl Viro return ERR_PTR(-EINVAL); 242279714f72SAl Viro } 24235955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2424391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 242579714f72SAl Viro if (IS_ERR(d)) { 24265955102cSAl Viro inode_unlock(path->dentry->d_inode); 2427391172c4SAl Viro path_put(path); 242879714f72SAl Viro } 242951689104SPaul Moore putname(filename); 243079714f72SAl Viro return d; 24315590ff0dSUlrich Drepper } 24325590ff0dSUlrich Drepper 2433d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2434d1811465SAl Viro { 2435abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2436abc9f5beSAl Viro flags, path, NULL); 2437d1811465SAl Viro } 24384d359507SAl Viro EXPORT_SYMBOL(kern_path); 2439d1811465SAl Viro 244016f18200SJosef 'Jeff' Sipek /** 244116f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 244216f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 244316f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 244416f18200SJosef 'Jeff' Sipek * @name: pointer to file name 244516f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2446e0a01249SAl Viro * @path: pointer to struct path to fill 244716f18200SJosef 'Jeff' Sipek */ 244816f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 244916f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2450e0a01249SAl Viro struct path *path) 245116f18200SJosef 'Jeff' Sipek { 24529ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24539ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2454abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2455abc9f5beSAl Viro flags , path, &root); 245616f18200SJosef 'Jeff' Sipek } 24574d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 245816f18200SJosef 'Jeff' Sipek 24593c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24603c95f0dcSAl Viro int len, struct qstr *this) 24613c95f0dcSAl Viro { 24623c95f0dcSAl Viro this->name = name; 24633c95f0dcSAl Viro this->len = len; 24643c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24653c95f0dcSAl Viro if (!len) 24663c95f0dcSAl Viro return -EACCES; 24673c95f0dcSAl Viro 24683c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24693c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24703c95f0dcSAl Viro return -EACCES; 24713c95f0dcSAl Viro } 24723c95f0dcSAl Viro 24733c95f0dcSAl Viro while (len--) { 24743c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24753c95f0dcSAl Viro if (c == '/' || c == '\0') 24763c95f0dcSAl Viro return -EACCES; 24773c95f0dcSAl Viro } 24783c95f0dcSAl Viro /* 24793c95f0dcSAl Viro * See if the low-level filesystem might want 24803c95f0dcSAl Viro * to use its own hash.. 24813c95f0dcSAl Viro */ 24823c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 24833c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 24843c95f0dcSAl Viro if (err < 0) 24853c95f0dcSAl Viro return err; 24863c95f0dcSAl Viro } 24873c95f0dcSAl Viro 24883c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 24893c95f0dcSAl Viro } 24903c95f0dcSAl Viro 2491eead1911SChristoph Hellwig /** 24920da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 24930da0b7fdSDavid Howells * @name: pathname component to lookup 24940da0b7fdSDavid Howells * @base: base directory to lookup from 24950da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 24960da0b7fdSDavid Howells * 24970da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 24980da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 24990da0b7fdSDavid Howells * 25000da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25010da0b7fdSDavid Howells * not be called by generic code. 25020da0b7fdSDavid Howells * 25030da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25040da0b7fdSDavid Howells */ 25050da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25060da0b7fdSDavid Howells { 25070da0b7fdSDavid Howells struct qstr this; 25080da0b7fdSDavid Howells int err; 25090da0b7fdSDavid Howells 25100da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25110da0b7fdSDavid Howells 25120da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25130da0b7fdSDavid Howells if (err) 25140da0b7fdSDavid Howells return ERR_PTR(err); 25150da0b7fdSDavid Howells 25160da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25170da0b7fdSDavid Howells } 25180da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25190da0b7fdSDavid Howells 25200da0b7fdSDavid Howells /** 2521a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2522eead1911SChristoph Hellwig * @name: pathname component to lookup 2523eead1911SChristoph Hellwig * @base: base directory to lookup from 2524eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2525eead1911SChristoph Hellwig * 2526a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25279e7543e9SAl Viro * not be called by generic code. 2528bbddca8eSNeilBrown * 2529bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2530eead1911SChristoph Hellwig */ 2531057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2532057f6c01SJames Morris { 25338613a209SAl Viro struct dentry *dentry; 2534057f6c01SJames Morris struct qstr this; 2535cda309deSMiklos Szeredi int err; 2536057f6c01SJames Morris 25375955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25382f9092e1SDavid Woodhouse 25393c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2540cda309deSMiklos Szeredi if (err) 2541cda309deSMiklos Szeredi return ERR_PTR(err); 2542cda309deSMiklos Szeredi 25438613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25448613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2545057f6c01SJames Morris } 25464d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2547057f6c01SJames Morris 2548bbddca8eSNeilBrown /** 2549bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2550bbddca8eSNeilBrown * @name: pathname component to lookup 2551bbddca8eSNeilBrown * @base: base directory to lookup from 2552bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2553bbddca8eSNeilBrown * 2554bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2555bbddca8eSNeilBrown * not be called by generic code. 2556bbddca8eSNeilBrown * 2557bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2558bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2559bbddca8eSNeilBrown */ 2560bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2561bbddca8eSNeilBrown struct dentry *base, int len) 2562bbddca8eSNeilBrown { 2563bbddca8eSNeilBrown struct qstr this; 2564bbddca8eSNeilBrown int err; 256520d00ee8SLinus Torvalds struct dentry *ret; 2566bbddca8eSNeilBrown 25673c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2568bbddca8eSNeilBrown if (err) 2569bbddca8eSNeilBrown return ERR_PTR(err); 2570bbddca8eSNeilBrown 257120d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 257220d00ee8SLinus Torvalds if (!ret) 257320d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 257420d00ee8SLinus Torvalds return ret; 2575bbddca8eSNeilBrown } 2576bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2577bbddca8eSNeilBrown 25786c2d4798SAl Viro /* 25796c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 25806c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 25816c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 25826c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 25836c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 25846c2d4798SAl Viro * this one avoids such problems. 25856c2d4798SAl Viro */ 25866c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 25876c2d4798SAl Viro struct dentry *base, int len) 25886c2d4798SAl Viro { 25896c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 25902fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 25916c2d4798SAl Viro dput(ret); 25926c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 25936c2d4798SAl Viro } 25946c2d4798SAl Viro return ret; 25956c2d4798SAl Viro } 25966c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 25976c2d4798SAl Viro 2598eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2599eedf265aSEric W. Biederman int path_pts(struct path *path) 2600eedf265aSEric W. Biederman { 2601eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2602eedf265aSEric W. Biederman * the input path. 2603eedf265aSEric W. Biederman */ 2604eedf265aSEric W. Biederman struct dentry *child, *parent; 2605eedf265aSEric W. Biederman struct qstr this; 2606eedf265aSEric W. Biederman int ret; 2607eedf265aSEric W. Biederman 2608eedf265aSEric W. Biederman ret = path_parent_directory(path); 2609eedf265aSEric W. Biederman if (ret) 2610eedf265aSEric W. Biederman return ret; 2611eedf265aSEric W. Biederman 2612eedf265aSEric W. Biederman parent = path->dentry; 2613eedf265aSEric W. Biederman this.name = "pts"; 2614eedf265aSEric W. Biederman this.len = 3; 2615eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2616eedf265aSEric W. Biederman if (!child) 2617eedf265aSEric W. Biederman return -ENOENT; 2618eedf265aSEric W. Biederman 2619eedf265aSEric W. Biederman path->dentry = child; 2620eedf265aSEric W. Biederman dput(parent); 2621eedf265aSEric W. Biederman follow_mount(path); 2622eedf265aSEric W. Biederman return 0; 2623eedf265aSEric W. Biederman } 2624eedf265aSEric W. Biederman #endif 2625eedf265aSEric W. Biederman 26261fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26271fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26281da177e4SLinus Torvalds { 2629abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2630abc9f5beSAl Viro flags, path, NULL); 26311da177e4SLinus Torvalds } 2632b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26331fa1e7f6SAndy Whitcroft 26348033426eSJeff Layton /** 2635197df04cSAl Viro * mountpoint_last - look up last component for umount 26368033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 26378033426eSJeff Layton * 26388033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 26398033426eSJeff Layton * need to resolve the path without doing any revalidation. 26408033426eSJeff Layton * 26418033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 26428033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 26438033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 26448033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 26458033426eSJeff Layton * bogus and it doesn't exist. 26468033426eSJeff Layton * 26478033426eSJeff Layton * Returns: 26488033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2649ba8f4613SAl Viro * lookup found a negative dentry. 26508033426eSJeff Layton * 2651ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2652ba8f4613SAl Viro * symlink that needs to be followed. 26538033426eSJeff Layton * 26548033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2655ba8f4613SAl Viro * that needs to be followed. 26568033426eSJeff Layton */ 26578033426eSJeff Layton static int 2658ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 26598033426eSJeff Layton { 26608033426eSJeff Layton int error = 0; 26618033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2662ba8f4613SAl Viro struct path path; 26638033426eSJeff Layton 266435759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 266535759521SAl Viro if (nd->flags & LOOKUP_RCU) { 26664675ac39SAl Viro if (unlazy_walk(nd)) 2667deb106c6SAl Viro return -ECHILD; 26688033426eSJeff Layton } 26698033426eSJeff Layton 26708033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 26718033426eSJeff Layton 26728033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 26738033426eSJeff Layton error = handle_dots(nd, nd->last_type); 267435759521SAl Viro if (error) 2675deb106c6SAl Viro return error; 2676ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2677949a852eSAl Viro } else { 2678ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2679ba8f4613SAl Viro if (!path.dentry) { 26808033426eSJeff Layton /* 2681949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2682949a852eSAl Viro * cache, so that means that this dentry is probably 2683949a852eSAl Viro * a symlink or the path doesn't actually point 2684949a852eSAl Viro * to a mounted dentry. 26858033426eSJeff Layton */ 2686ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2687949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2688ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2689ba8f4613SAl Viro return PTR_ERR(path.dentry); 26908033426eSJeff Layton } 2691bcceeebaSDave Jones } 26922fa6b1e0SAl Viro if (d_flags_negative(smp_load_acquire(&path.dentry->d_flags))) { 2693ba8f4613SAl Viro dput(path.dentry); 2694deb106c6SAl Viro return -ENOENT; 269535759521SAl Viro } 2696ba8f4613SAl Viro path.mnt = nd->path.mnt; 26978f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 26988033426eSJeff Layton } 26998033426eSJeff Layton 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)) && 2716ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 27173bdba28bSAl Viro s = trailing_symlink(nd); 27183bdba28bSAl Viro } 2719ba8f4613SAl Viro if (!err) { 2720ba8f4613SAl Viro *path = nd->path; 2721ba8f4613SAl Viro nd->path.mnt = NULL; 2722ba8f4613SAl Viro nd->path.dentry = NULL; 2723ba8f4613SAl Viro follow_mount(path); 2724ba8f4613SAl Viro } 2725deb106c6SAl Viro terminate_walk(nd); 27268033426eSJeff Layton return err; 27278033426eSJeff Layton } 27288033426eSJeff Layton 27292d864651SAl Viro static int 2730668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 27312d864651SAl Viro unsigned int flags) 27322d864651SAl Viro { 27339883d185SAl Viro struct nameidata nd; 2734cbaab2dbSAl Viro int error; 2735668696dcSAl Viro if (IS_ERR(name)) 2736668696dcSAl Viro return PTR_ERR(name); 27379883d185SAl Viro set_nameidata(&nd, dfd, name); 2738c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 27392d864651SAl Viro if (unlikely(error == -ECHILD)) 2740c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 27412d864651SAl Viro if (unlikely(error == -ESTALE)) 2742c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 27432d864651SAl Viro if (likely(!error)) 2744c9b07eabSAl Viro audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL); 27459883d185SAl Viro restore_nameidata(); 2746668696dcSAl Viro putname(name); 27472d864651SAl Viro return error; 27482d864651SAl Viro } 27492d864651SAl Viro 27508033426eSJeff Layton /** 2751197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 27528033426eSJeff Layton * @dfd: directory file descriptor 27538033426eSJeff Layton * @name: pathname from userland 27548033426eSJeff Layton * @flags: lookup flags 27558033426eSJeff Layton * @path: pointer to container to hold result 27568033426eSJeff Layton * 27578033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27588033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27598033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27608033426eSJeff Layton * and avoid revalidating the last component. 27618033426eSJeff Layton * 27628033426eSJeff Layton * Returns 0 and populates "path" on success. 27638033426eSJeff Layton */ 27648033426eSJeff Layton int 2765197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27668033426eSJeff Layton struct path *path) 27678033426eSJeff Layton { 2768cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27698033426eSJeff Layton } 27708033426eSJeff Layton 27712d864651SAl Viro int 27722d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27732d864651SAl Viro unsigned int flags) 27742d864651SAl Viro { 2775cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27762d864651SAl Viro } 27772d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27782d864651SAl Viro 2779cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27801da177e4SLinus Torvalds { 27818e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2782da9592edSDavid Howells 27838e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27841da177e4SLinus Torvalds return 0; 27858e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27861da177e4SLinus Torvalds return 0; 278723adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27881da177e4SLinus Torvalds } 2789cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27901da177e4SLinus Torvalds 27911da177e4SLinus Torvalds /* 27921da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27931da177e4SLinus Torvalds * whether the type of victim is right. 27941da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27951da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27961da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27971da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27981da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27991da177e4SLinus Torvalds * a. be owner of dir, or 28001da177e4SLinus Torvalds * b. be owner of victim, or 28011da177e4SLinus Torvalds * c. have CAP_FOWNER capability 28021da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 28031da177e4SLinus Torvalds * links pointing to it. 28040bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 28050bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 28060bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 28070bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 28080bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 28091da177e4SLinus Torvalds * nfs_async_unlink(). 28101da177e4SLinus Torvalds */ 2811b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 28121da177e4SLinus Torvalds { 281363afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 28141da177e4SLinus Torvalds int error; 28151da177e4SLinus Torvalds 2816b18825a7SDavid Howells if (d_is_negative(victim)) 28171da177e4SLinus Torvalds return -ENOENT; 2818b18825a7SDavid Howells BUG_ON(!inode); 28191da177e4SLinus Torvalds 28201da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2821593d1ce8SEric W. Biederman 2822593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2823593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2824593d1ce8SEric W. Biederman return -EOVERFLOW; 2825593d1ce8SEric W. Biederman 28264fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 28271da177e4SLinus Torvalds 2828f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 28291da177e4SLinus Torvalds if (error) 28301da177e4SLinus Torvalds return error; 28311da177e4SLinus Torvalds if (IS_APPEND(dir)) 28321da177e4SLinus Torvalds return -EPERM; 2833b18825a7SDavid Howells 2834b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 28350bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 28361da177e4SLinus Torvalds return -EPERM; 28371da177e4SLinus Torvalds if (isdir) { 283844b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28391da177e4SLinus Torvalds return -ENOTDIR; 28401da177e4SLinus Torvalds if (IS_ROOT(victim)) 28411da177e4SLinus Torvalds return -EBUSY; 284244b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28431da177e4SLinus Torvalds return -EISDIR; 28441da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28451da177e4SLinus Torvalds return -ENOENT; 28461da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28471da177e4SLinus Torvalds return -EBUSY; 28481da177e4SLinus Torvalds return 0; 28491da177e4SLinus Torvalds } 28501da177e4SLinus Torvalds 28511da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28521da177e4SLinus Torvalds * dir. 28531da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28541da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28551da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2856036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2857036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2858036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28591da177e4SLinus Torvalds */ 2860a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28611da177e4SLinus Torvalds { 2862036d5236SEric W. Biederman struct user_namespace *s_user_ns; 286314e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28641da177e4SLinus Torvalds if (child->d_inode) 28651da177e4SLinus Torvalds return -EEXIST; 28661da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28671da177e4SLinus Torvalds return -ENOENT; 2868036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2869036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2870036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2871036d5236SEric W. Biederman return -EOVERFLOW; 2872f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28731da177e4SLinus Torvalds } 28741da177e4SLinus Torvalds 28751da177e4SLinus Torvalds /* 28761da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28771da177e4SLinus Torvalds */ 28781da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28791da177e4SLinus Torvalds { 28801da177e4SLinus Torvalds struct dentry *p; 28811da177e4SLinus Torvalds 28821da177e4SLinus Torvalds if (p1 == p2) { 28835955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28841da177e4SLinus Torvalds return NULL; 28851da177e4SLinus Torvalds } 28861da177e4SLinus Torvalds 2887fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28881da177e4SLinus Torvalds 2889e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2890e2761a11SOGAWA Hirofumi if (p) { 28915955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28925955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28931da177e4SLinus Torvalds return p; 28941da177e4SLinus Torvalds } 28951da177e4SLinus Torvalds 2896e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2897e2761a11SOGAWA Hirofumi if (p) { 28985955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28995955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 29001da177e4SLinus Torvalds return p; 29011da177e4SLinus Torvalds } 29021da177e4SLinus Torvalds 29035955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29045955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 29051da177e4SLinus Torvalds return NULL; 29061da177e4SLinus Torvalds } 29074d359507SAl Viro EXPORT_SYMBOL(lock_rename); 29081da177e4SLinus Torvalds 29091da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 29101da177e4SLinus Torvalds { 29115955102cSAl Viro inode_unlock(p1->d_inode); 29121da177e4SLinus Torvalds if (p1 != p2) { 29135955102cSAl Viro inode_unlock(p2->d_inode); 2914fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 29151da177e4SLinus Torvalds } 29161da177e4SLinus Torvalds } 29174d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 29181da177e4SLinus Torvalds 29194acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2920312b63fbSAl Viro bool want_excl) 29211da177e4SLinus Torvalds { 2922a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 29231da177e4SLinus Torvalds if (error) 29241da177e4SLinus Torvalds return error; 29251da177e4SLinus Torvalds 2926acfa4380SAl Viro if (!dir->i_op->create) 29271da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 29281da177e4SLinus Torvalds mode &= S_IALLUGO; 29291da177e4SLinus Torvalds mode |= S_IFREG; 29301da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 29311da177e4SLinus Torvalds if (error) 29321da177e4SLinus Torvalds return error; 2933312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2934a74574aaSStephen Smalley if (!error) 2935f38aa942SAmy Griffis fsnotify_create(dir, dentry); 29361da177e4SLinus Torvalds return error; 29371da177e4SLinus Torvalds } 29384d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29391da177e4SLinus Torvalds 29408e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29418e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29428e6c848eSAl Viro void *arg) 29438e6c848eSAl Viro { 29448e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 29458e6c848eSAl Viro int error = may_create(dir, dentry); 29468e6c848eSAl Viro if (error) 29478e6c848eSAl Viro return error; 29488e6c848eSAl Viro 29498e6c848eSAl Viro mode &= S_IALLUGO; 29508e6c848eSAl Viro mode |= S_IFREG; 29518e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29528e6c848eSAl Viro if (error) 29538e6c848eSAl Viro return error; 29548e6c848eSAl Viro error = f(dentry, mode, arg); 29558e6c848eSAl Viro if (!error) 29568e6c848eSAl Viro fsnotify_create(dir, dentry); 29578e6c848eSAl Viro return error; 29588e6c848eSAl Viro } 29598e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29608e6c848eSAl Viro 2961a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2962a2982cc9SEric W. Biederman { 2963a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2964a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2965a2982cc9SEric W. Biederman } 2966a2982cc9SEric W. Biederman 2967f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 29681da177e4SLinus Torvalds { 29693fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29701da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29711da177e4SLinus Torvalds int error; 29721da177e4SLinus Torvalds 29731da177e4SLinus Torvalds if (!inode) 29741da177e4SLinus Torvalds return -ENOENT; 29751da177e4SLinus Torvalds 2976c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2977c8fe8f30SChristoph Hellwig case S_IFLNK: 29781da177e4SLinus Torvalds return -ELOOP; 2979c8fe8f30SChristoph Hellwig case S_IFDIR: 2980c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29811da177e4SLinus Torvalds return -EISDIR; 2982c8fe8f30SChristoph Hellwig break; 2983c8fe8f30SChristoph Hellwig case S_IFBLK: 2984c8fe8f30SChristoph Hellwig case S_IFCHR: 2985a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29861da177e4SLinus Torvalds return -EACCES; 2987c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2988c8fe8f30SChristoph Hellwig case S_IFIFO: 2989c8fe8f30SChristoph Hellwig case S_IFSOCK: 29901da177e4SLinus Torvalds flag &= ~O_TRUNC; 2991c8fe8f30SChristoph Hellwig break; 29924a3fd211SDave Hansen } 2993b41572e9SDave Hansen 299462fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2995b41572e9SDave Hansen if (error) 2996b41572e9SDave Hansen return error; 29976146f0d5SMimi Zohar 29981da177e4SLinus Torvalds /* 29991da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 30001da177e4SLinus Torvalds */ 30011da177e4SLinus Torvalds if (IS_APPEND(inode)) { 30028737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 30037715b521SAl Viro return -EPERM; 30041da177e4SLinus Torvalds if (flag & O_TRUNC) 30057715b521SAl Viro return -EPERM; 30061da177e4SLinus Torvalds } 30071da177e4SLinus Torvalds 30081da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 30092e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 30107715b521SAl Viro return -EPERM; 30111da177e4SLinus Torvalds 3012f3c7691eSJ. Bruce Fields return 0; 30137715b521SAl Viro } 30147715b521SAl Viro 3015e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 30167715b521SAl Viro { 3017f0bb5aafSAl Viro const struct path *path = &filp->f_path; 30187715b521SAl Viro struct inode *inode = path->dentry->d_inode; 30197715b521SAl Viro int error = get_write_access(inode); 30201da177e4SLinus Torvalds if (error) 30217715b521SAl Viro return error; 30221da177e4SLinus Torvalds /* 30231da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 30241da177e4SLinus Torvalds */ 3025d7a06983SJeff Layton error = locks_verify_locked(filp); 3026be6d3e56SKentaro Takeda if (!error) 3027ea0d3ab2STetsuo Handa error = security_path_truncate(path); 30281da177e4SLinus Torvalds if (!error) { 30297715b521SAl Viro error = do_truncate(path->dentry, 0, 3030d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3031e1181ee6SJeff Layton filp); 30321da177e4SLinus Torvalds } 30331da177e4SLinus Torvalds put_write_access(inode); 3034acd0c935SMimi Zohar return error; 30351da177e4SLinus Torvalds } 30361da177e4SLinus Torvalds 3037d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3038d57999e1SDave Hansen { 30398a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30408a5e929dSAl Viro flag--; 3041d57999e1SDave Hansen return flag; 3042d57999e1SDave Hansen } 3043d57999e1SDave Hansen 3044d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 3045d18e9008SMiklos Szeredi { 30461328c727SSeth Forshee struct user_namespace *s_user_ns; 3047d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3048d18e9008SMiklos Szeredi if (error) 3049d18e9008SMiklos Szeredi return error; 3050d18e9008SMiklos Szeredi 30511328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 30521328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 30531328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 30541328c727SSeth Forshee return -EOVERFLOW; 30551328c727SSeth Forshee 3056d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 3057d18e9008SMiklos Szeredi if (error) 3058d18e9008SMiklos Szeredi return error; 3059d18e9008SMiklos Szeredi 3060d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3061d18e9008SMiklos Szeredi } 3062d18e9008SMiklos Szeredi 30631acf0af9SDavid Howells /* 30641acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30651acf0af9SDavid Howells * dentry. 30661acf0af9SDavid Howells * 30671acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30681acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30691acf0af9SDavid Howells * 307000a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 307100a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 307200a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 30731acf0af9SDavid Howells * 30741acf0af9SDavid Howells * Returns an error code otherwise. 30751acf0af9SDavid Howells */ 30762675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 307730d90494SAl Viro struct path *path, struct file *file, 3078015c3bbcSMiklos Szeredi const struct open_flags *op, 30793ec2eef1SAl Viro int open_flag, umode_t mode) 3080d18e9008SMiklos Szeredi { 3081d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3082d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3083d18e9008SMiklos Szeredi int error; 3084d18e9008SMiklos Szeredi 3085384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3086d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3087d18e9008SMiklos Szeredi 3088d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3089d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3090d18e9008SMiklos Szeredi 309130d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 309230d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30930fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 309444907d79SAl Viro open_to_namei_flags(open_flag), mode); 30956fbd0714SAl Viro d_lookup_done(dentry); 3096384f26e2SAl Viro if (!error) { 309764e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 3098d18e9008SMiklos Szeredi /* 3099384f26e2SAl Viro * We didn't have the inode before the open, so check open 3100384f26e2SAl Viro * permission here. 3101d18e9008SMiklos Szeredi */ 3102384f26e2SAl Viro int acc_mode = op->acc_mode; 310373a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 310403da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 310503da633aSAl Viro fsnotify_create(dir, dentry); 310662fb4a15SAl Viro acc_mode = 0; 310703da633aSAl Viro } 31082675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3109384f26e2SAl Viro if (WARN_ON(error > 0)) 3110384f26e2SAl Viro error = -EINVAL; 311164e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3112d18e9008SMiklos Szeredi error = -EIO; 3113384f26e2SAl Viro } else { 3114d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3115d18e9008SMiklos Szeredi dput(dentry); 3116d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 311710c64ceaSAl Viro } 311873a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 3119d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3120a01e718fSAl Viro if (unlikely(d_is_negative(dentry))) { 3121a01e718fSAl Viro error = -ENOENT; 3122a01e718fSAl Viro } else { 3123d18e9008SMiklos Szeredi path->dentry = dentry; 3124d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 312500a07c15SAl Viro return 0; 3126d18e9008SMiklos Szeredi } 3127d18e9008SMiklos Szeredi } 3128a01e718fSAl Viro } 3129d18e9008SMiklos Szeredi dput(dentry); 3130d18e9008SMiklos Szeredi return error; 3131d18e9008SMiklos Szeredi } 3132d18e9008SMiklos Szeredi 313331e6b01fSNick Piggin /* 31341acf0af9SDavid Howells * Look up and maybe create and open the last component. 3135d58ffd35SMiklos Szeredi * 313600a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3137d58ffd35SMiklos Szeredi * 313800a07c15SAl Viro * Returns 0 on success, that is, if 313900a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 314000a07c15SAl Viro * the file was not completely opened at this time, though lookups and 314100a07c15SAl Viro * creations were performed. 314200a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 314300a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 314400a07c15SAl Viro * hadn't been specified. 31451acf0af9SDavid Howells * 314600a07c15SAl Viro * An error code is returned on failure. 3147d58ffd35SMiklos Szeredi */ 31482675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 314930d90494SAl Viro 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))) 3162ce8644fcSAl Viro return -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)) 31702675a4ebSAl Viro return PTR_ERR(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 */ 3186d18e9008SMiklos Szeredi goto out_no_open; 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) { 32251643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 32263ec2eef1SAl Viro mode); 32271643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 32281643b43fSAl Viro error = create_error; 32291643b43fSAl Viro return error; 32301643b43fSAl Viro } 323154ef4872SMiklos Szeredi 32321643b43fSAl Viro no_open: 32336fbd0714SAl Viro if (d_in_lookup(dentry)) { 323412fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 323512fa5e24SAl Viro nd->flags); 32366fbd0714SAl Viro d_lookup_done(dentry); 323712fa5e24SAl Viro if (unlikely(res)) { 323812fa5e24SAl Viro if (IS_ERR(res)) { 323912fa5e24SAl Viro error = PTR_ERR(res); 324012fa5e24SAl Viro goto out_dput; 324112fa5e24SAl Viro } 324212fa5e24SAl Viro dput(dentry); 324312fa5e24SAl Viro dentry = res; 324412fa5e24SAl Viro } 324554ef4872SMiklos Szeredi } 324654ef4872SMiklos Szeredi 3247d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32481643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 324973a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3250ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3251ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3252ce8644fcSAl Viro error = -EACCES; 3253d58ffd35SMiklos Szeredi goto out_dput; 325464894cf8SAl Viro } 3255ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 32561643b43fSAl Viro open_flag & O_EXCL); 3257d58ffd35SMiklos Szeredi if (error) 3258d58ffd35SMiklos Szeredi goto out_dput; 3259ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3260d58ffd35SMiklos Szeredi } 32611643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32621643b43fSAl Viro error = create_error; 3263d58ffd35SMiklos Szeredi goto out_dput; 3264d58ffd35SMiklos Szeredi } 3265d18e9008SMiklos Szeredi out_no_open: 3266d58ffd35SMiklos Szeredi path->dentry = dentry; 3267d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 326800a07c15SAl Viro return 0; 3269d58ffd35SMiklos Szeredi 3270d58ffd35SMiklos Szeredi out_dput: 3271d58ffd35SMiklos Szeredi dput(dentry); 32722675a4ebSAl Viro return error; 3273d58ffd35SMiklos Szeredi } 3274d58ffd35SMiklos Szeredi 3275d58ffd35SMiklos Szeredi /* 3276fe2d35ffSAl Viro * Handle the last step of open() 327731e6b01fSNick Piggin */ 3278896475d5SAl Viro static int do_last(struct nameidata *nd, 32793ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3280fb1cc555SAl Viro { 3281a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3282ca344a89SAl Viro int open_flag = op->open_flag; 328377d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 328464894cf8SAl Viro bool got_write = false; 3285bcda7652SAl Viro int acc_mode = op->acc_mode; 3286254cf582SAl Viro unsigned seq; 3287a1eb3315SMiklos Szeredi struct inode *inode; 3288896475d5SAl Viro struct path path; 328916c2cd71SAl Viro int error; 3290fb1cc555SAl Viro 3291c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3292c3e380b0SAl Viro nd->flags |= op->intent; 3293c3e380b0SAl Viro 3294bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3295fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3296deb106c6SAl Viro if (unlikely(error)) 32972675a4ebSAl Viro return error; 3298e83db167SMiklos Szeredi goto finish_open; 32991f36f774SAl Viro } 3300a2c36b45SAl Viro 3301ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3302fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3303fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3304fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3305254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3306e9742b53SAl Viro if (likely(error > 0)) 330771574865SMiklos Szeredi goto finish_lookup; 330871574865SMiklos Szeredi 3309ce57dfc1SAl Viro if (error < 0) 3310deb106c6SAl Viro return error; 3311a1eb3315SMiklos 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); 33453ec2eef1SAl Viro error = lookup_open(nd, &path, 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 33512675a4ebSAl Viro if (error) 3352d58ffd35SMiklos Szeredi goto out; 33536c0d46c4SAl Viro 335400a07c15SAl Viro if (file->f_mode & FMODE_OPENED) { 335573a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3356496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 335777d660a8SMiklos Szeredi will_truncate = false; 3358d18e9008SMiklos Szeredi 335976ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3360d18e9008SMiklos Szeredi goto opened; 3361d18e9008SMiklos Szeredi } 3362d18e9008SMiklos Szeredi 336373a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 33649b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3365ca344a89SAl Viro open_flag &= ~O_TRUNC; 336677d660a8SMiklos Szeredi will_truncate = false; 336762fb4a15SAl Viro acc_mode = 0; 3368896475d5SAl Viro path_to_nameidata(&path, nd); 3369e83db167SMiklos Szeredi goto finish_open_created; 3370fb1cc555SAl Viro } 3371fb1cc555SAl Viro 3372fb1cc555SAl Viro /* 3373d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3374d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3375d18e9008SMiklos Szeredi * necessary...) 3376d18e9008SMiklos Szeredi */ 337764894cf8SAl Viro if (got_write) { 3378d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 337964894cf8SAl Viro got_write = false; 3380d18e9008SMiklos Szeredi } 3381d18e9008SMiklos Szeredi 3382e6ec03a2SAl Viro error = follow_managed(&path, nd); 3383e6ec03a2SAl Viro if (unlikely(error < 0)) 3384e6ec03a2SAl Viro return error; 3385e6ec03a2SAl Viro 33866583fe22SAl Viro /* 33876583fe22SAl Viro * create/update audit record if it already exists. 33886583fe22SAl Viro */ 33896583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 33906583fe22SAl Viro 3391deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3392deb106c6SAl Viro path_to_nameidata(&path, nd); 3393deb106c6SAl Viro return -EEXIST; 3394deb106c6SAl Viro } 3395fb1cc555SAl Viro 3396254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3397d4565649SAl Viro inode = d_backing_inode(path.dentry); 3398766c4cbfSAl Viro finish_lookup: 33998f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3400deb106c6SAl Viro if (unlikely(error)) 3401d63ff28fSAl Viro return error; 3402bc77daa7SAl Viro finish_open: 34038f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3404a3fbbde7SAl Viro error = complete_walk(nd); 3405fac7d191SAl Viro if (error) 34062675a4ebSAl Viro return error; 340776ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 340830aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3409fb1cc555SAl Viro error = -EISDIR; 341030aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 34112675a4ebSAl Viro goto out; 341230aba665SSalvatore Mesoraca error = may_create_in_sticky(dir, 341330aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 341430aba665SSalvatore Mesoraca if (unlikely(error)) 341530aba665SSalvatore Mesoraca goto out; 341630aba665SSalvatore Mesoraca } 3417af2f5542SMiklos Szeredi error = -ENOTDIR; 341844b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 34192675a4ebSAl Viro goto out; 34204bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 342177d660a8SMiklos Szeredi will_truncate = false; 34226c0d46c4SAl Viro 34230f9d1a10SAl Viro if (will_truncate) { 34240f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 34250f9d1a10SAl Viro if (error) 34262675a4ebSAl Viro goto out; 342764894cf8SAl Viro got_write = true; 34280f9d1a10SAl Viro } 3429e83db167SMiklos Szeredi finish_open_created: 3430bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3431ca344a89SAl Viro if (error) 34322675a4ebSAl Viro goto out; 3433aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3434ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3435fac7d191SAl Viro if (error) 3436015c3bbcSMiklos Szeredi goto out; 3437a8277b9bSMiklos Szeredi opened: 34386035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3439fe9ec829SAl Viro if (!error && will_truncate) 34402675a4ebSAl Viro error = handle_truncate(file); 3441ca344a89SAl Viro out: 3442c80567c8SAl Viro if (unlikely(error > 0)) { 3443c80567c8SAl Viro WARN_ON(1); 3444c80567c8SAl Viro error = -EINVAL; 3445c80567c8SAl Viro } 344664894cf8SAl Viro if (got_write) 34470f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 34482675a4ebSAl Viro return error; 3449fb1cc555SAl Viro } 3450fb1cc555SAl Viro 3451af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3452af7bd4dcSAmir Goldstein { 3453af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3454af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3455af7bd4dcSAmir Goldstein struct inode *inode; 3456af7bd4dcSAmir Goldstein int error; 3457af7bd4dcSAmir Goldstein 3458af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3459af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3460af7bd4dcSAmir Goldstein if (error) 3461af7bd4dcSAmir Goldstein goto out_err; 3462af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3463af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3464af7bd4dcSAmir Goldstein goto out_err; 3465af7bd4dcSAmir Goldstein error = -ENOMEM; 3466cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3467af7bd4dcSAmir Goldstein if (unlikely(!child)) 3468af7bd4dcSAmir Goldstein goto out_err; 3469af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3470af7bd4dcSAmir Goldstein if (error) 3471af7bd4dcSAmir Goldstein goto out_err; 3472af7bd4dcSAmir Goldstein error = -ENOENT; 3473af7bd4dcSAmir Goldstein inode = child->d_inode; 3474af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3475af7bd4dcSAmir Goldstein goto out_err; 3476af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3477af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3478af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3479af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3480af7bd4dcSAmir Goldstein } 3481fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3482af7bd4dcSAmir Goldstein return child; 3483af7bd4dcSAmir Goldstein 3484af7bd4dcSAmir Goldstein out_err: 3485af7bd4dcSAmir Goldstein dput(child); 3486af7bd4dcSAmir Goldstein return ERR_PTR(error); 3487af7bd4dcSAmir Goldstein } 3488af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3489af7bd4dcSAmir Goldstein 3490c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 349160545d0dSAl Viro const struct open_flags *op, 34923ec2eef1SAl Viro struct file *file) 349360545d0dSAl Viro { 3494625b6d10SAl Viro struct dentry *child; 3495625b6d10SAl Viro struct path path; 3496c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 349760545d0dSAl Viro if (unlikely(error)) 349860545d0dSAl Viro return error; 3499625b6d10SAl Viro error = mnt_want_write(path.mnt); 350060545d0dSAl Viro if (unlikely(error)) 350160545d0dSAl Viro goto out; 3502af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3503af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3504684e73beSHirofumi Nakagawa if (IS_ERR(child)) 350560545d0dSAl Viro goto out2; 3506625b6d10SAl Viro dput(path.dentry); 3507625b6d10SAl Viro path.dentry = child; 3508c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 350969a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 351062fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 351160545d0dSAl Viro if (error) 351260545d0dSAl Viro goto out2; 3513625b6d10SAl Viro file->f_path.mnt = path.mnt; 3514be12af3eSAl Viro error = finish_open(file, child, NULL); 351560545d0dSAl Viro out2: 3516625b6d10SAl Viro mnt_drop_write(path.mnt); 351760545d0dSAl Viro out: 3518625b6d10SAl Viro path_put(&path); 351960545d0dSAl Viro return error; 352060545d0dSAl Viro } 352160545d0dSAl Viro 35226ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 35236ac08709SAl Viro { 35246ac08709SAl Viro struct path path; 35256ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 35266ac08709SAl Viro if (!error) { 35276ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3528ae2bb293SAl Viro error = vfs_open(&path, file); 35296ac08709SAl Viro path_put(&path); 35306ac08709SAl Viro } 35316ac08709SAl Viro return error; 35326ac08709SAl Viro } 35336ac08709SAl Viro 3534c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3535c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35361da177e4SLinus Torvalds { 353730d90494SAl Viro struct file *file; 353813aab428SAl Viro int error; 353931e6b01fSNick Piggin 3540ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35411afc99beSAl Viro if (IS_ERR(file)) 35421afc99beSAl Viro return file; 354331e6b01fSNick Piggin 3544bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 35453ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 35465f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 35476ac08709SAl Viro error = do_o_path(nd, flags, file); 35485f336e72SAl Viro } else { 35495f336e72SAl Viro const char *s = path_init(nd, flags); 35503bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 35513ec2eef1SAl Viro (error = do_last(nd, file, op)) > 0) { 355273d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35533bdba28bSAl Viro s = trailing_symlink(nd); 35543bdba28bSAl Viro } 3555deb106c6SAl Viro terminate_walk(nd); 35565f336e72SAl Viro } 35577c1c01ecSAl Viro if (likely(!error)) { 3558aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35597c1c01ecSAl Viro return file; 35607c1c01ecSAl Viro WARN_ON(1); 35617c1c01ecSAl Viro error = -EINVAL; 3562015c3bbcSMiklos Szeredi } 35637c1c01ecSAl Viro fput(file); 35642675a4ebSAl Viro if (error == -EOPENSTALE) { 35652675a4ebSAl Viro if (flags & LOOKUP_RCU) 35662675a4ebSAl Viro error = -ECHILD; 35672675a4ebSAl Viro else 35682675a4ebSAl Viro error = -ESTALE; 35692675a4ebSAl Viro } 35707c1c01ecSAl Viro return ERR_PTR(error); 3571de459215SKirill Korotaev } 35721da177e4SLinus Torvalds 3573669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3574f9652e10SAl Viro const struct open_flags *op) 357513aab428SAl Viro { 35769883d185SAl Viro struct nameidata nd; 3577f9652e10SAl Viro int flags = op->lookup_flags; 357813aab428SAl Viro struct file *filp; 357913aab428SAl Viro 35809883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3581c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 358213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3583c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 358413aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3585c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35869883d185SAl Viro restore_nameidata(); 358713aab428SAl Viro return filp; 358813aab428SAl Viro } 358913aab428SAl Viro 359073d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3591f9652e10SAl Viro const char *name, const struct open_flags *op) 359273d049a4SAl Viro { 35939883d185SAl Viro struct nameidata nd; 359473d049a4SAl Viro struct file *file; 359551689104SPaul Moore struct filename *filename; 3596f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 359773d049a4SAl Viro 359873d049a4SAl Viro nd.root.mnt = mnt; 359973d049a4SAl Viro nd.root.dentry = dentry; 360073d049a4SAl Viro 3601b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 360273d049a4SAl Viro return ERR_PTR(-ELOOP); 360373d049a4SAl Viro 360451689104SPaul Moore filename = getname_kernel(name); 3605a1c83681SViresh Kumar if (IS_ERR(filename)) 360651689104SPaul Moore return ERR_CAST(filename); 360751689104SPaul Moore 36089883d185SAl Viro set_nameidata(&nd, -1, filename); 3609c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 361073d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3611c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 361273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3613c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 36149883d185SAl Viro restore_nameidata(); 361551689104SPaul Moore putname(filename); 361673d049a4SAl Viro return file; 361773d049a4SAl Viro } 361873d049a4SAl Viro 3619fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 36201ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36211da177e4SLinus Torvalds { 3622c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3623391172c4SAl Viro struct qstr last; 3624391172c4SAl Viro int type; 3625c30dabfeSJan Kara int err2; 36261ac12b4bSJeff Layton int error; 36271ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36281ac12b4bSJeff Layton 36291ac12b4bSJeff Layton /* 36301ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36311ac12b4bSJeff Layton * other flags passed in are ignored! 36321ac12b4bSJeff Layton */ 36331ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36341ac12b4bSJeff Layton 36355c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36365c31b6ceSAl Viro if (IS_ERR(name)) 36375c31b6ceSAl Viro return ERR_CAST(name); 36381da177e4SLinus Torvalds 3639c663e5d8SChristoph Hellwig /* 3640c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3641c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3642c663e5d8SChristoph Hellwig */ 36435c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3644ed75e95dSAl Viro goto out; 3645c663e5d8SChristoph Hellwig 3646c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3647391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3648c663e5d8SChristoph Hellwig /* 3649c663e5d8SChristoph Hellwig * Do the final lookup. 3650c663e5d8SChristoph Hellwig */ 3651391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36525955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3653391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36541da177e4SLinus Torvalds if (IS_ERR(dentry)) 3655a8104a9fSAl Viro goto unlock; 3656c663e5d8SChristoph Hellwig 3657a8104a9fSAl Viro error = -EEXIST; 3658b18825a7SDavid Howells if (d_is_positive(dentry)) 3659a8104a9fSAl Viro goto fail; 3660b18825a7SDavid Howells 3661c663e5d8SChristoph Hellwig /* 3662c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3663c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3664c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3665c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3666c663e5d8SChristoph Hellwig */ 3667391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3668a8104a9fSAl Viro error = -ENOENT; 3669ed75e95dSAl Viro goto fail; 3670e9baf6e5SAl Viro } 3671c30dabfeSJan Kara if (unlikely(err2)) { 3672c30dabfeSJan Kara error = err2; 3673a8104a9fSAl Viro goto fail; 3674c30dabfeSJan Kara } 3675181c37b6SAl Viro putname(name); 3676e9baf6e5SAl Viro return dentry; 36771da177e4SLinus Torvalds fail: 3678a8104a9fSAl Viro dput(dentry); 3679a8104a9fSAl Viro dentry = ERR_PTR(error); 3680a8104a9fSAl Viro unlock: 36815955102cSAl Viro inode_unlock(path->dentry->d_inode); 3682c30dabfeSJan Kara if (!err2) 3683391172c4SAl Viro mnt_drop_write(path->mnt); 3684ed75e95dSAl Viro out: 3685391172c4SAl Viro path_put(path); 3686181c37b6SAl Viro putname(name); 3687ed75e95dSAl Viro return dentry; 3688dae6ad8fSAl Viro } 3689fa14a0b8SAl Viro 3690fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3691fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3692fa14a0b8SAl Viro { 3693181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3694181c37b6SAl Viro path, lookup_flags); 3695fa14a0b8SAl Viro } 3696dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3697dae6ad8fSAl Viro 3698921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3699921a1650SAl Viro { 3700921a1650SAl Viro dput(dentry); 37015955102cSAl Viro inode_unlock(path->dentry->d_inode); 3702a8104a9fSAl Viro mnt_drop_write(path->mnt); 3703921a1650SAl Viro path_put(path); 3704921a1650SAl Viro } 3705921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3706921a1650SAl Viro 3707520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 37081ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3709dae6ad8fSAl Viro { 3710181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3711dae6ad8fSAl Viro } 3712dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3713dae6ad8fSAl Viro 37141a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 37151da177e4SLinus Torvalds { 3716a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37171da177e4SLinus Torvalds 37181da177e4SLinus Torvalds if (error) 37191da177e4SLinus Torvalds return error; 37201da177e4SLinus Torvalds 372194f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 37221da177e4SLinus Torvalds return -EPERM; 37231da177e4SLinus Torvalds 3724acfa4380SAl Viro if (!dir->i_op->mknod) 37251da177e4SLinus Torvalds return -EPERM; 37261da177e4SLinus Torvalds 372708ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 372808ce5f16SSerge E. Hallyn if (error) 372908ce5f16SSerge E. Hallyn return error; 373008ce5f16SSerge E. Hallyn 37311da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37321da177e4SLinus Torvalds if (error) 37331da177e4SLinus Torvalds return error; 37341da177e4SLinus Torvalds 37351da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3736a74574aaSStephen Smalley if (!error) 3737f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37381da177e4SLinus Torvalds return error; 37391da177e4SLinus Torvalds } 37404d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37411da177e4SLinus Torvalds 3742f69aac00SAl Viro static int may_mknod(umode_t mode) 3743463c3197SDave Hansen { 3744463c3197SDave Hansen switch (mode & S_IFMT) { 3745463c3197SDave Hansen case S_IFREG: 3746463c3197SDave Hansen case S_IFCHR: 3747463c3197SDave Hansen case S_IFBLK: 3748463c3197SDave Hansen case S_IFIFO: 3749463c3197SDave Hansen case S_IFSOCK: 3750463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3751463c3197SDave Hansen return 0; 3752463c3197SDave Hansen case S_IFDIR: 3753463c3197SDave Hansen return -EPERM; 3754463c3197SDave Hansen default: 3755463c3197SDave Hansen return -EINVAL; 3756463c3197SDave Hansen } 3757463c3197SDave Hansen } 3758463c3197SDave Hansen 375987c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 376087c4e192SDominik Brodowski unsigned int dev) 37611da177e4SLinus Torvalds { 37621da177e4SLinus Torvalds struct dentry *dentry; 3763dae6ad8fSAl Viro struct path path; 3764dae6ad8fSAl Viro int error; 3765972567f1SJeff Layton unsigned int lookup_flags = 0; 37661da177e4SLinus Torvalds 37678e4bfca1SAl Viro error = may_mknod(mode); 37688e4bfca1SAl Viro if (error) 37698e4bfca1SAl Viro return error; 3770972567f1SJeff Layton retry: 3771972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3772dae6ad8fSAl Viro if (IS_ERR(dentry)) 3773dae6ad8fSAl Viro return PTR_ERR(dentry); 37742ad94ae6SAl Viro 3775dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3776ce3b0f8dSAl Viro mode &= ~current_umask(); 3777dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3778be6d3e56SKentaro Takeda if (error) 3779a8104a9fSAl Viro goto out; 37801da177e4SLinus Torvalds switch (mode & S_IFMT) { 37811da177e4SLinus Torvalds case 0: case S_IFREG: 3782312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 378305d1a717SMimi Zohar if (!error) 378405d1a717SMimi Zohar ima_post_path_mknod(dentry); 37851da177e4SLinus Torvalds break; 37861da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3787dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37881da177e4SLinus Torvalds new_decode_dev(dev)); 37891da177e4SLinus Torvalds break; 37901da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3791dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37921da177e4SLinus Torvalds break; 37931da177e4SLinus Torvalds } 3794a8104a9fSAl Viro out: 3795921a1650SAl Viro done_path_create(&path, dentry); 3796972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3797972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3798972567f1SJeff Layton goto retry; 3799972567f1SJeff Layton } 38001da177e4SLinus Torvalds return error; 38011da177e4SLinus Torvalds } 38021da177e4SLinus Torvalds 380387c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 380487c4e192SDominik Brodowski unsigned int, dev) 380587c4e192SDominik Brodowski { 380687c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 380787c4e192SDominik Brodowski } 380887c4e192SDominik Brodowski 38098208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 38105590ff0dSUlrich Drepper { 381187c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 38125590ff0dSUlrich Drepper } 38135590ff0dSUlrich Drepper 381418bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 38151da177e4SLinus Torvalds { 3816a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38178de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38181da177e4SLinus Torvalds 38191da177e4SLinus Torvalds if (error) 38201da177e4SLinus Torvalds return error; 38211da177e4SLinus Torvalds 3822acfa4380SAl Viro if (!dir->i_op->mkdir) 38231da177e4SLinus Torvalds return -EPERM; 38241da177e4SLinus Torvalds 38251da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38261da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38271da177e4SLinus Torvalds if (error) 38281da177e4SLinus Torvalds return error; 38291da177e4SLinus Torvalds 38308de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38318de52778SAl Viro return -EMLINK; 38328de52778SAl Viro 38331da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3834a74574aaSStephen Smalley if (!error) 3835f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38361da177e4SLinus Torvalds return error; 38371da177e4SLinus Torvalds } 38384d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38391da177e4SLinus Torvalds 38400101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38411da177e4SLinus Torvalds { 38426902d925SDave Hansen struct dentry *dentry; 3843dae6ad8fSAl Viro struct path path; 3844dae6ad8fSAl Viro int error; 3845b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38461da177e4SLinus Torvalds 3847b76d8b82SJeff Layton retry: 3848b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38496902d925SDave Hansen if (IS_ERR(dentry)) 3850dae6ad8fSAl Viro return PTR_ERR(dentry); 38516902d925SDave Hansen 3852dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3853ce3b0f8dSAl Viro mode &= ~current_umask(); 3854dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3855a8104a9fSAl Viro if (!error) 3856dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3857921a1650SAl Viro done_path_create(&path, dentry); 3858b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3859b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3860b76d8b82SJeff Layton goto retry; 3861b76d8b82SJeff Layton } 38621da177e4SLinus Torvalds return error; 38631da177e4SLinus Torvalds } 38641da177e4SLinus Torvalds 38650101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38660101db7aSDominik Brodowski { 38670101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38680101db7aSDominik Brodowski } 38690101db7aSDominik Brodowski 3870a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38715590ff0dSUlrich Drepper { 38720101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38735590ff0dSUlrich Drepper } 38745590ff0dSUlrich Drepper 38751da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38761da177e4SLinus Torvalds { 38771da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38781da177e4SLinus Torvalds 38791da177e4SLinus Torvalds if (error) 38801da177e4SLinus Torvalds return error; 38811da177e4SLinus Torvalds 3882acfa4380SAl Viro if (!dir->i_op->rmdir) 38831da177e4SLinus Torvalds return -EPERM; 38841da177e4SLinus Torvalds 38851d2ef590SAl Viro dget(dentry); 38865955102cSAl Viro inode_lock(dentry->d_inode); 3887912dbc15SSage Weil 38881da177e4SLinus Torvalds error = -EBUSY; 38897af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3890912dbc15SSage Weil goto out; 3891912dbc15SSage Weil 38921da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3893912dbc15SSage Weil if (error) 3894912dbc15SSage Weil goto out; 3895912dbc15SSage Weil 38961da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3897912dbc15SSage Weil if (error) 3898912dbc15SSage Weil goto out; 3899912dbc15SSage Weil 39008767712fSAl Viro shrink_dcache_parent(dentry); 39011da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3902d83c49f3SAl Viro dont_mount(dentry); 39038ed936b5SEric W. Biederman detach_mounts(dentry); 3904116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 39051da177e4SLinus Torvalds 3906912dbc15SSage Weil out: 39075955102cSAl Viro inode_unlock(dentry->d_inode); 39081d2ef590SAl Viro dput(dentry); 3909912dbc15SSage Weil if (!error) 3910912dbc15SSage Weil d_delete(dentry); 39111da177e4SLinus Torvalds return error; 39121da177e4SLinus Torvalds } 39134d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 39141da177e4SLinus Torvalds 3915f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 39161da177e4SLinus Torvalds { 39171da177e4SLinus Torvalds int error = 0; 391891a27b2aSJeff Layton struct filename *name; 39191da177e4SLinus Torvalds struct dentry *dentry; 3920f5beed75SAl Viro struct path path; 3921f5beed75SAl Viro struct qstr last; 3922f5beed75SAl Viro int type; 3923c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3924c6ee9206SJeff Layton retry: 3925c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3926c1d4dd27SAl Viro &path, &last, &type); 392791a27b2aSJeff Layton if (IS_ERR(name)) 392891a27b2aSJeff Layton return PTR_ERR(name); 39291da177e4SLinus Torvalds 3930f5beed75SAl Viro switch (type) { 39311da177e4SLinus Torvalds case LAST_DOTDOT: 39321da177e4SLinus Torvalds error = -ENOTEMPTY; 39331da177e4SLinus Torvalds goto exit1; 39341da177e4SLinus Torvalds case LAST_DOT: 39351da177e4SLinus Torvalds error = -EINVAL; 39361da177e4SLinus Torvalds goto exit1; 39371da177e4SLinus Torvalds case LAST_ROOT: 39381da177e4SLinus Torvalds error = -EBUSY; 39391da177e4SLinus Torvalds goto exit1; 39401da177e4SLinus Torvalds } 39410612d9fbSOGAWA Hirofumi 3942f5beed75SAl Viro error = mnt_want_write(path.mnt); 3943c30dabfeSJan Kara if (error) 3944c30dabfeSJan Kara goto exit1; 39450612d9fbSOGAWA Hirofumi 39465955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3947f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39481da177e4SLinus Torvalds error = PTR_ERR(dentry); 39496902d925SDave Hansen if (IS_ERR(dentry)) 39506902d925SDave Hansen goto exit2; 3951e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3952e6bc45d6STheodore Ts'o error = -ENOENT; 3953e6bc45d6STheodore Ts'o goto exit3; 3954e6bc45d6STheodore Ts'o } 3955f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3956be6d3e56SKentaro Takeda if (error) 3957c30dabfeSJan Kara goto exit3; 3958f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39590622753bSDave Hansen exit3: 39601da177e4SLinus Torvalds dput(dentry); 39616902d925SDave Hansen exit2: 39625955102cSAl Viro inode_unlock(path.dentry->d_inode); 3963f5beed75SAl Viro mnt_drop_write(path.mnt); 39641da177e4SLinus Torvalds exit1: 3965f5beed75SAl Viro path_put(&path); 39661da177e4SLinus Torvalds putname(name); 3967c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3968c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3969c6ee9206SJeff Layton goto retry; 3970c6ee9206SJeff Layton } 39711da177e4SLinus Torvalds return error; 39721da177e4SLinus Torvalds } 39731da177e4SLinus Torvalds 39743cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39755590ff0dSUlrich Drepper { 39765590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39775590ff0dSUlrich Drepper } 39785590ff0dSUlrich Drepper 3979b21996e3SJ. Bruce Fields /** 3980b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3981b21996e3SJ. Bruce Fields * @dir: parent directory 3982b21996e3SJ. Bruce Fields * @dentry: victim 3983b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3984b21996e3SJ. Bruce Fields * 3985b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3986b21996e3SJ. Bruce Fields * 3987b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3988b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3989b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3990b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3991b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3992b21996e3SJ. Bruce Fields * 3993b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3994b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3995b21996e3SJ. Bruce Fields * to be NFS exported. 3996b21996e3SJ. Bruce Fields */ 3997b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39981da177e4SLinus Torvalds { 39999accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 40001da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 40011da177e4SLinus Torvalds 40021da177e4SLinus Torvalds if (error) 40031da177e4SLinus Torvalds return error; 40041da177e4SLinus Torvalds 4005acfa4380SAl Viro if (!dir->i_op->unlink) 40061da177e4SLinus Torvalds return -EPERM; 40071da177e4SLinus Torvalds 40085955102cSAl Viro inode_lock(target); 40098ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 40101da177e4SLinus Torvalds error = -EBUSY; 40111da177e4SLinus Torvalds else { 40121da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4013bec1052eSAl Viro if (!error) { 40145a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 40155a14696cSJ. Bruce Fields if (error) 4016b21996e3SJ. Bruce Fields goto out; 40171da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 40188ed936b5SEric W. Biederman if (!error) { 4019d83c49f3SAl Viro dont_mount(dentry); 40208ed936b5SEric W. Biederman detach_mounts(dentry); 4021116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 40228ed936b5SEric W. Biederman } 4023bec1052eSAl Viro } 40241da177e4SLinus Torvalds } 4025b21996e3SJ. Bruce Fields out: 40265955102cSAl Viro inode_unlock(target); 40271da177e4SLinus Torvalds 40281da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40291da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40309accbb97SJ. Bruce Fields fsnotify_link_count(target); 40311da177e4SLinus Torvalds d_delete(dentry); 40321da177e4SLinus Torvalds } 40330eeca283SRobert Love 40341da177e4SLinus Torvalds return error; 40351da177e4SLinus Torvalds } 40364d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40371da177e4SLinus Torvalds 40381da177e4SLinus Torvalds /* 40391da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40401b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40411da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40421da177e4SLinus Torvalds * while waiting on the I/O. 40431da177e4SLinus Torvalds */ 4044da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40451da177e4SLinus Torvalds { 40462ad94ae6SAl Viro int error; 40471da177e4SLinus Torvalds struct dentry *dentry; 4048f5beed75SAl Viro struct path path; 4049f5beed75SAl Viro struct qstr last; 4050f5beed75SAl Viro int type; 40511da177e4SLinus Torvalds struct inode *inode = NULL; 4052b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40535d18f813SJeff Layton unsigned int lookup_flags = 0; 40545d18f813SJeff Layton retry: 4055da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 405691a27b2aSJeff Layton if (IS_ERR(name)) 405791a27b2aSJeff Layton return PTR_ERR(name); 40582ad94ae6SAl Viro 40591da177e4SLinus Torvalds error = -EISDIR; 4060f5beed75SAl Viro if (type != LAST_NORM) 40611da177e4SLinus Torvalds goto exit1; 40620612d9fbSOGAWA Hirofumi 4063f5beed75SAl Viro error = mnt_want_write(path.mnt); 4064c30dabfeSJan Kara if (error) 4065c30dabfeSJan Kara goto exit1; 4066b21996e3SJ. Bruce Fields retry_deleg: 40675955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4068f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40691da177e4SLinus Torvalds error = PTR_ERR(dentry); 40701da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40711da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4072f5beed75SAl Viro if (last.name[last.len]) 407350338b88STörök Edwin goto slashes; 40741da177e4SLinus Torvalds inode = dentry->d_inode; 4075b18825a7SDavid Howells if (d_is_negative(dentry)) 4076e6bc45d6STheodore Ts'o goto slashes; 40777de9c6eeSAl Viro ihold(inode); 4078f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4079be6d3e56SKentaro Takeda if (error) 4080c30dabfeSJan Kara goto exit2; 4081f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40821da177e4SLinus Torvalds exit2: 40831da177e4SLinus Torvalds dput(dentry); 40841da177e4SLinus Torvalds } 40855955102cSAl Viro inode_unlock(path.dentry->d_inode); 40861da177e4SLinus Torvalds if (inode) 40871da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4088b21996e3SJ. Bruce Fields inode = NULL; 4089b21996e3SJ. Bruce Fields if (delegated_inode) { 40905a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4091b21996e3SJ. Bruce Fields if (!error) 4092b21996e3SJ. Bruce Fields goto retry_deleg; 4093b21996e3SJ. Bruce Fields } 4094f5beed75SAl Viro mnt_drop_write(path.mnt); 40951da177e4SLinus Torvalds exit1: 4096f5beed75SAl Viro path_put(&path); 40975d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40985d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40995d18f813SJeff Layton inode = NULL; 41005d18f813SJeff Layton goto retry; 41015d18f813SJeff Layton } 4102da2f1362SChristoph Hellwig putname(name); 41031da177e4SLinus Torvalds return error; 41041da177e4SLinus Torvalds 41051da177e4SLinus Torvalds slashes: 4106b18825a7SDavid Howells if (d_is_negative(dentry)) 4107b18825a7SDavid Howells error = -ENOENT; 410844b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4109b18825a7SDavid Howells error = -EISDIR; 4110b18825a7SDavid Howells else 4111b18825a7SDavid Howells error = -ENOTDIR; 41121da177e4SLinus Torvalds goto exit2; 41131da177e4SLinus Torvalds } 41141da177e4SLinus Torvalds 41152e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 41165590ff0dSUlrich Drepper { 41175590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 41185590ff0dSUlrich Drepper return -EINVAL; 41195590ff0dSUlrich Drepper 41205590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 41215590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 41225590ff0dSUlrich Drepper 4123da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 41245590ff0dSUlrich Drepper } 41255590ff0dSUlrich Drepper 41263480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41275590ff0dSUlrich Drepper { 4128da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 41295590ff0dSUlrich Drepper } 41305590ff0dSUlrich Drepper 4131db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 41321da177e4SLinus Torvalds { 4133a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 41341da177e4SLinus Torvalds 41351da177e4SLinus Torvalds if (error) 41361da177e4SLinus Torvalds return error; 41371da177e4SLinus Torvalds 4138acfa4380SAl Viro if (!dir->i_op->symlink) 41391da177e4SLinus Torvalds return -EPERM; 41401da177e4SLinus Torvalds 41411da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41421da177e4SLinus Torvalds if (error) 41431da177e4SLinus Torvalds return error; 41441da177e4SLinus Torvalds 41451da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4146a74574aaSStephen Smalley if (!error) 4147f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41481da177e4SLinus Torvalds return error; 41491da177e4SLinus Torvalds } 41504d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41511da177e4SLinus Torvalds 4152b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 4153b724e846SDominik Brodowski const char __user *newname) 41541da177e4SLinus Torvalds { 41552ad94ae6SAl Viro int error; 415691a27b2aSJeff Layton struct filename *from; 41576902d925SDave Hansen struct dentry *dentry; 4158dae6ad8fSAl Viro struct path path; 4159f46d3567SJeff Layton unsigned int lookup_flags = 0; 41601da177e4SLinus Torvalds 41611da177e4SLinus Torvalds from = getname(oldname); 41621da177e4SLinus Torvalds if (IS_ERR(from)) 41631da177e4SLinus Torvalds return PTR_ERR(from); 4164f46d3567SJeff Layton retry: 4165f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41661da177e4SLinus Torvalds error = PTR_ERR(dentry); 41676902d925SDave Hansen if (IS_ERR(dentry)) 4168dae6ad8fSAl Viro goto out_putname; 41696902d925SDave Hansen 417091a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4171a8104a9fSAl Viro if (!error) 417291a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4173921a1650SAl Viro done_path_create(&path, dentry); 4174f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4175f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4176f46d3567SJeff Layton goto retry; 4177f46d3567SJeff Layton } 41786902d925SDave Hansen out_putname: 41791da177e4SLinus Torvalds putname(from); 41801da177e4SLinus Torvalds return error; 41811da177e4SLinus Torvalds } 41821da177e4SLinus Torvalds 4183b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4184b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4185b724e846SDominik Brodowski { 4186b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4187b724e846SDominik Brodowski } 4188b724e846SDominik Brodowski 41893480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41905590ff0dSUlrich Drepper { 4191b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 41925590ff0dSUlrich Drepper } 41935590ff0dSUlrich Drepper 4194146a8595SJ. Bruce Fields /** 4195146a8595SJ. Bruce Fields * vfs_link - create a new link 4196146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4197146a8595SJ. Bruce Fields * @dir: new parent 4198146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4199146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4200146a8595SJ. Bruce Fields * 4201146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4202146a8595SJ. Bruce Fields * 4203146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4204146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4205146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4206146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4207146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4208146a8595SJ. Bruce Fields * 4209146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4210146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4211146a8595SJ. Bruce Fields * to be NFS exported. 4212146a8595SJ. Bruce Fields */ 4213146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 42141da177e4SLinus Torvalds { 42151da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 42168de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 42171da177e4SLinus Torvalds int error; 42181da177e4SLinus Torvalds 42191da177e4SLinus Torvalds if (!inode) 42201da177e4SLinus Torvalds return -ENOENT; 42211da177e4SLinus Torvalds 4222a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 42231da177e4SLinus Torvalds if (error) 42241da177e4SLinus Torvalds return error; 42251da177e4SLinus Torvalds 42261da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 42271da177e4SLinus Torvalds return -EXDEV; 42281da177e4SLinus Torvalds 42291da177e4SLinus Torvalds /* 42301da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 42311da177e4SLinus Torvalds */ 42321da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 42331da177e4SLinus Torvalds return -EPERM; 42340bd23d09SEric W. Biederman /* 42350bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 42360bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 42370bd23d09SEric W. Biederman * the vfs. 42380bd23d09SEric W. Biederman */ 42390bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 42400bd23d09SEric W. Biederman return -EPERM; 4241acfa4380SAl Viro if (!dir->i_op->link) 42421da177e4SLinus Torvalds return -EPERM; 42437e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42441da177e4SLinus Torvalds return -EPERM; 42451da177e4SLinus Torvalds 42461da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42471da177e4SLinus Torvalds if (error) 42481da177e4SLinus Torvalds return error; 42491da177e4SLinus Torvalds 42505955102cSAl Viro inode_lock(inode); 4251aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4252f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4253aae8a97dSAneesh Kumar K.V error = -ENOENT; 42548de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42558de52778SAl Viro error = -EMLINK; 4256146a8595SJ. Bruce Fields else { 4257146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4258146a8595SJ. Bruce Fields if (!error) 42591da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4260146a8595SJ. Bruce Fields } 4261f4e0c30cSAl Viro 4262f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4263f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4264f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4265f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4266f4e0c30cSAl Viro } 42675955102cSAl Viro inode_unlock(inode); 4268e31e14ecSStephen Smalley if (!error) 42697e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42701da177e4SLinus Torvalds return error; 42711da177e4SLinus Torvalds } 42724d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42731da177e4SLinus Torvalds 42741da177e4SLinus Torvalds /* 42751da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42761da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42771da177e4SLinus Torvalds * newname. --KAB 42781da177e4SLinus Torvalds * 42791da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42801da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42811da177e4SLinus Torvalds * and other special files. --ADM 42821da177e4SLinus Torvalds */ 428346ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 428446ea89ebSDominik Brodowski const char __user *newname, int flags) 42851da177e4SLinus Torvalds { 42861da177e4SLinus Torvalds struct dentry *new_dentry; 4287dae6ad8fSAl Viro struct path old_path, new_path; 4288146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 428911a7b371SAneesh Kumar K.V int how = 0; 42901da177e4SLinus Torvalds int error; 42911da177e4SLinus Torvalds 429211a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4293c04030e1SUlrich Drepper return -EINVAL; 429411a7b371SAneesh Kumar K.V /* 4295f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4296f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4297f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 429811a7b371SAneesh Kumar K.V */ 4299f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4300f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4301f0cc6ffbSLinus Torvalds return -ENOENT; 430211a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4303f0cc6ffbSLinus Torvalds } 4304c04030e1SUlrich Drepper 430511a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 430611a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4307442e31caSJeff Layton retry: 430811a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 43091da177e4SLinus Torvalds if (error) 43102ad94ae6SAl Viro return error; 43112ad94ae6SAl Viro 4312442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4313442e31caSJeff Layton (how & LOOKUP_REVAL)); 43141da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43156902d925SDave Hansen if (IS_ERR(new_dentry)) 4316dae6ad8fSAl Viro goto out; 4317dae6ad8fSAl Viro 4318dae6ad8fSAl Viro error = -EXDEV; 4319dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4320dae6ad8fSAl Viro goto out_dput; 4321800179c9SKees Cook error = may_linkat(&old_path); 4322800179c9SKees Cook if (unlikely(error)) 4323800179c9SKees Cook goto out_dput; 4324dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4325be6d3e56SKentaro Takeda if (error) 4326a8104a9fSAl Viro goto out_dput; 4327146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 432875c3f29dSDave Hansen out_dput: 4329921a1650SAl Viro done_path_create(&new_path, new_dentry); 4330146a8595SJ. Bruce Fields if (delegated_inode) { 4331146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4332d22e6338SOleg Drokin if (!error) { 4333d22e6338SOleg Drokin path_put(&old_path); 4334146a8595SJ. Bruce Fields goto retry; 4335146a8595SJ. Bruce Fields } 4336d22e6338SOleg Drokin } 4337442e31caSJeff Layton if (retry_estale(error, how)) { 4338d22e6338SOleg Drokin path_put(&old_path); 4339442e31caSJeff Layton how |= LOOKUP_REVAL; 4340442e31caSJeff Layton goto retry; 4341442e31caSJeff Layton } 43421da177e4SLinus Torvalds out: 43432d8f3038SAl Viro path_put(&old_path); 43441da177e4SLinus Torvalds 43451da177e4SLinus Torvalds return error; 43461da177e4SLinus Torvalds } 43471da177e4SLinus Torvalds 434846ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 434946ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 435046ea89ebSDominik Brodowski { 435146ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 435246ea89ebSDominik Brodowski } 435346ea89ebSDominik Brodowski 43543480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43555590ff0dSUlrich Drepper { 435646ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43575590ff0dSUlrich Drepper } 43585590ff0dSUlrich Drepper 4359bc27027aSMiklos Szeredi /** 4360bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4361bc27027aSMiklos Szeredi * @old_dir: parent of source 4362bc27027aSMiklos Szeredi * @old_dentry: source 4363bc27027aSMiklos Szeredi * @new_dir: parent of destination 4364bc27027aSMiklos Szeredi * @new_dentry: destination 4365bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4366520c8b16SMiklos Szeredi * @flags: rename flags 4367bc27027aSMiklos Szeredi * 4368bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4369bc27027aSMiklos Szeredi * 4370bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4371bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4372bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4373bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4374bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4375bc27027aSMiklos Szeredi * so. 4376bc27027aSMiklos Szeredi * 4377bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4378bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4379bc27027aSMiklos Szeredi * to be NFS exported. 4380bc27027aSMiklos Szeredi * 43811da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43821da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43831da177e4SLinus Torvalds * Problems: 43840117d427SMauro Carvalho Chehab * 4385d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43861da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43871da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4388a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43891da177e4SLinus Torvalds * story. 43906cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43916cedba89SJ. Bruce Fields * and source (if it is not a directory). 43921b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43931da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43941da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4395a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43961da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43971da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43981da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4399a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 44001da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 44011da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 44021da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4403e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 44041b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 44051da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4406c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 44071da177e4SLinus Torvalds * locking]. 44081da177e4SLinus Torvalds */ 44091da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 44108e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4411520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 44121da177e4SLinus Torvalds { 44131da177e4SLinus Torvalds int error; 4414bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4415bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4416bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4417da1ce067SMiklos Szeredi bool new_is_dir = false; 4418da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 441949d31c2fSAl Viro struct name_snapshot old_name; 44201da177e4SLinus Torvalds 44218d3e2936SMiklos Szeredi if (source == target) 44221da177e4SLinus Torvalds return 0; 44231da177e4SLinus Torvalds 44241da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 44251da177e4SLinus Torvalds if (error) 44261da177e4SLinus Torvalds return error; 44271da177e4SLinus Torvalds 4428da1ce067SMiklos Szeredi if (!target) { 4429a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4430da1ce067SMiklos Szeredi } else { 4431da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4432da1ce067SMiklos Szeredi 4433da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44341da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4435da1ce067SMiklos Szeredi else 4436da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4437da1ce067SMiklos Szeredi } 44381da177e4SLinus Torvalds if (error) 44391da177e4SLinus Torvalds return error; 44401da177e4SLinus Torvalds 44412773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 44421da177e4SLinus Torvalds return -EPERM; 44431da177e4SLinus Torvalds 4444bc27027aSMiklos Szeredi /* 4445bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4446bc27027aSMiklos Szeredi * we'll need to flip '..'. 4447bc27027aSMiklos Szeredi */ 4448da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4449da1ce067SMiklos Szeredi if (is_dir) { 4450bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4451bc27027aSMiklos Szeredi if (error) 4452bc27027aSMiklos Szeredi return error; 4453bc27027aSMiklos Szeredi } 4454da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4455da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4456da1ce067SMiklos Szeredi if (error) 4457da1ce067SMiklos Szeredi return error; 4458da1ce067SMiklos Szeredi } 4459da1ce067SMiklos Szeredi } 44600eeca283SRobert Love 44610b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44620b3974ebSMiklos Szeredi flags); 4463bc27027aSMiklos Szeredi if (error) 4464bc27027aSMiklos Szeredi return error; 4465bc27027aSMiklos Szeredi 446649d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4467bc27027aSMiklos Szeredi dget(new_dentry); 4468da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4469bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4470bc27027aSMiklos Szeredi else if (target) 44715955102cSAl Viro inode_lock(target); 4472bc27027aSMiklos Szeredi 4473bc27027aSMiklos Szeredi error = -EBUSY; 44747af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4475bc27027aSMiklos Szeredi goto out; 4476bc27027aSMiklos Szeredi 4477da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4478bc27027aSMiklos Szeredi error = -EMLINK; 4479da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4480bc27027aSMiklos Szeredi goto out; 4481da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4482da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4483da1ce067SMiklos Szeredi goto out; 4484da1ce067SMiklos Szeredi } 4485da1ce067SMiklos Szeredi if (!is_dir) { 4486bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4487bc27027aSMiklos Szeredi if (error) 4488bc27027aSMiklos Szeredi goto out; 4489da1ce067SMiklos Szeredi } 4490da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4491bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4492bc27027aSMiklos Szeredi if (error) 4493bc27027aSMiklos Szeredi goto out; 4494bc27027aSMiklos Szeredi } 4495520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4496520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4497bc27027aSMiklos Szeredi if (error) 4498bc27027aSMiklos Szeredi goto out; 4499bc27027aSMiklos Szeredi 4500da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 45018767712fSAl Viro if (is_dir) { 45028767712fSAl Viro shrink_dcache_parent(new_dentry); 4503bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 45048767712fSAl Viro } 4505bc27027aSMiklos Szeredi dont_mount(new_dentry); 45068ed936b5SEric W. Biederman detach_mounts(new_dentry); 4507bc27027aSMiklos Szeredi } 4508da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4509da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4510bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4511da1ce067SMiklos Szeredi else 4512da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4513da1ce067SMiklos Szeredi } 4514bc27027aSMiklos Szeredi out: 4515da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4516bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4517bc27027aSMiklos Szeredi else if (target) 45185955102cSAl Viro inode_unlock(target); 4519bc27027aSMiklos Szeredi dput(new_dentry); 4520da1ce067SMiklos Szeredi if (!error) { 4521f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4522da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4523da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4524f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4525da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4526da1ce067SMiklos Szeredi } 4527da1ce067SMiklos Szeredi } 452849d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 45290eeca283SRobert Love 45301da177e4SLinus Torvalds return error; 45311da177e4SLinus Torvalds } 45324d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 45331da177e4SLinus Torvalds 4534ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4535ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 45361da177e4SLinus Torvalds { 45371da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 45381da177e4SLinus Torvalds struct dentry *trap; 4539f5beed75SAl Viro struct path old_path, new_path; 4540f5beed75SAl Viro struct qstr old_last, new_last; 4541f5beed75SAl Viro int old_type, new_type; 45428e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 454391a27b2aSJeff Layton struct filename *from; 454491a27b2aSJeff Layton struct filename *to; 4545f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4546c6a94284SJeff Layton bool should_retry = false; 45472ad94ae6SAl Viro int error; 4548520c8b16SMiklos Szeredi 45490d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4550da1ce067SMiklos Szeredi return -EINVAL; 4551da1ce067SMiklos Szeredi 45520d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45530d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4554520c8b16SMiklos Szeredi return -EINVAL; 4555520c8b16SMiklos Szeredi 45560d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45570d7a8555SMiklos Szeredi return -EPERM; 45580d7a8555SMiklos Szeredi 4559f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4560f5beed75SAl Viro target_flags = 0; 4561f5beed75SAl Viro 4562c6a94284SJeff Layton retry: 4563c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4564c1d4dd27SAl Viro &old_path, &old_last, &old_type); 456591a27b2aSJeff Layton if (IS_ERR(from)) { 456691a27b2aSJeff Layton error = PTR_ERR(from); 45671da177e4SLinus Torvalds goto exit; 456891a27b2aSJeff Layton } 45691da177e4SLinus Torvalds 4570c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4571c1d4dd27SAl Viro &new_path, &new_last, &new_type); 457291a27b2aSJeff Layton if (IS_ERR(to)) { 457391a27b2aSJeff Layton error = PTR_ERR(to); 45741da177e4SLinus Torvalds goto exit1; 457591a27b2aSJeff Layton } 45761da177e4SLinus Torvalds 45771da177e4SLinus Torvalds error = -EXDEV; 4578f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45791da177e4SLinus Torvalds goto exit2; 45801da177e4SLinus Torvalds 45811da177e4SLinus Torvalds error = -EBUSY; 4582f5beed75SAl Viro if (old_type != LAST_NORM) 45831da177e4SLinus Torvalds goto exit2; 45841da177e4SLinus Torvalds 45850a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45860a7c3937SMiklos Szeredi error = -EEXIST; 4587f5beed75SAl Viro if (new_type != LAST_NORM) 45881da177e4SLinus Torvalds goto exit2; 45891da177e4SLinus Torvalds 4590f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4591c30dabfeSJan Kara if (error) 4592c30dabfeSJan Kara goto exit2; 4593c30dabfeSJan Kara 45948e6d782cSJ. Bruce Fields retry_deleg: 4595f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45961da177e4SLinus Torvalds 4597f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45981da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45991da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 46001da177e4SLinus Torvalds goto exit3; 46011da177e4SLinus Torvalds /* source must exist */ 46021da177e4SLinus Torvalds error = -ENOENT; 4603b18825a7SDavid Howells if (d_is_negative(old_dentry)) 46041da177e4SLinus Torvalds goto exit4; 4605f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 46061da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 46071da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 46081da177e4SLinus Torvalds goto exit4; 46090a7c3937SMiklos Szeredi error = -EEXIST; 46100a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 46110a7c3937SMiklos Szeredi goto exit5; 4612da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4613da1ce067SMiklos Szeredi error = -ENOENT; 4614da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4615da1ce067SMiklos Szeredi goto exit5; 4616da1ce067SMiklos Szeredi 4617da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4618da1ce067SMiklos Szeredi error = -ENOTDIR; 4619f5beed75SAl Viro if (new_last.name[new_last.len]) 4620da1ce067SMiklos Szeredi goto exit5; 4621da1ce067SMiklos Szeredi } 4622da1ce067SMiklos Szeredi } 46230a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 46240a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 46250a7c3937SMiklos Szeredi error = -ENOTDIR; 4626f5beed75SAl Viro if (old_last.name[old_last.len]) 46270a7c3937SMiklos Szeredi goto exit5; 4628f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 46290a7c3937SMiklos Szeredi goto exit5; 46300a7c3937SMiklos Szeredi } 46310a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 46320a7c3937SMiklos Szeredi error = -EINVAL; 46330a7c3937SMiklos Szeredi if (old_dentry == trap) 46340a7c3937SMiklos Szeredi goto exit5; 46351da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4636da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46371da177e4SLinus Torvalds error = -ENOTEMPTY; 46381da177e4SLinus Torvalds if (new_dentry == trap) 46391da177e4SLinus Torvalds goto exit5; 46401da177e4SLinus Torvalds 4641f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4642f5beed75SAl Viro &new_path, new_dentry, flags); 4643be6d3e56SKentaro Takeda if (error) 4644c30dabfeSJan Kara goto exit5; 4645f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4646f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4647520c8b16SMiklos Szeredi &delegated_inode, flags); 46481da177e4SLinus Torvalds exit5: 46491da177e4SLinus Torvalds dput(new_dentry); 46501da177e4SLinus Torvalds exit4: 46511da177e4SLinus Torvalds dput(old_dentry); 46521da177e4SLinus Torvalds exit3: 4653f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46548e6d782cSJ. Bruce Fields if (delegated_inode) { 46558e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46568e6d782cSJ. Bruce Fields if (!error) 46578e6d782cSJ. Bruce Fields goto retry_deleg; 46588e6d782cSJ. Bruce Fields } 4659f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46601da177e4SLinus Torvalds exit2: 4661c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4662c6a94284SJeff Layton should_retry = true; 4663f5beed75SAl Viro path_put(&new_path); 46642ad94ae6SAl Viro putname(to); 46651da177e4SLinus Torvalds exit1: 4666f5beed75SAl Viro path_put(&old_path); 46671da177e4SLinus Torvalds putname(from); 4668c6a94284SJeff Layton if (should_retry) { 4669c6a94284SJeff Layton should_retry = false; 4670c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4671c6a94284SJeff Layton goto retry; 4672c6a94284SJeff Layton } 46732ad94ae6SAl Viro exit: 46741da177e4SLinus Torvalds return error; 46751da177e4SLinus Torvalds } 46761da177e4SLinus Torvalds 4677ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4678ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4679ee81feb6SDominik Brodowski { 4680ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4681ee81feb6SDominik Brodowski } 4682ee81feb6SDominik Brodowski 4683520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4684520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4685520c8b16SMiklos Szeredi { 4686ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4687520c8b16SMiklos Szeredi } 4688520c8b16SMiklos Szeredi 4689a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 46905590ff0dSUlrich Drepper { 4691ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 46925590ff0dSUlrich Drepper } 46935590ff0dSUlrich Drepper 4694787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4695787fb6bcSMiklos Szeredi { 4696787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4697787fb6bcSMiklos Szeredi if (error) 4698787fb6bcSMiklos Szeredi return error; 4699787fb6bcSMiklos Szeredi 4700787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4701787fb6bcSMiklos Szeredi return -EPERM; 4702787fb6bcSMiklos Szeredi 4703787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4704787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4705787fb6bcSMiklos Szeredi } 4706787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4707787fb6bcSMiklos Szeredi 47085d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 47091da177e4SLinus Torvalds { 47105d826c84SAl Viro int len = PTR_ERR(link); 47111da177e4SLinus Torvalds if (IS_ERR(link)) 47121da177e4SLinus Torvalds goto out; 47131da177e4SLinus Torvalds 47141da177e4SLinus Torvalds len = strlen(link); 47151da177e4SLinus Torvalds if (len > (unsigned) buflen) 47161da177e4SLinus Torvalds len = buflen; 47171da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 47181da177e4SLinus Torvalds len = -EFAULT; 47191da177e4SLinus Torvalds out: 47201da177e4SLinus Torvalds return len; 47211da177e4SLinus Torvalds } 47221da177e4SLinus Torvalds 4723d60874cdSMiklos Szeredi /** 4724fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4725fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4726fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4727fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4728fd4a0edfSMiklos Szeredi * 4729fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4730fd4a0edfSMiklos Szeredi * 4731fd4a0edfSMiklos Szeredi * Does not call security hook. 4732fd4a0edfSMiklos Szeredi */ 4733fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4734fd4a0edfSMiklos Szeredi { 4735fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4736f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4737f2df5da6SAl Viro const char *link; 4738f2df5da6SAl Viro int res; 4739fd4a0edfSMiklos Szeredi 474076fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 474176fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 474276fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 474376fca90eSMiklos Szeredi 474476fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4745fd4a0edfSMiklos Szeredi return -EINVAL; 4746fd4a0edfSMiklos Szeredi 474776fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 474876fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 474976fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 475076fca90eSMiklos Szeredi } 475176fca90eSMiklos Szeredi 47524c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4753f2df5da6SAl Viro if (!link) { 4754f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4755f2df5da6SAl Viro if (IS_ERR(link)) 4756f2df5da6SAl Viro return PTR_ERR(link); 4757f2df5da6SAl Viro } 4758f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4759f2df5da6SAl Viro do_delayed_call(&done); 4760f2df5da6SAl Viro return res; 4761fd4a0edfSMiklos Szeredi } 4762fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 47631da177e4SLinus Torvalds 4764d60874cdSMiklos Szeredi /** 4765d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4766d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4767d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4768d60874cdSMiklos Szeredi * 4769d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4770d60874cdSMiklos Szeredi * 4771d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4772d60874cdSMiklos Szeredi * 4773d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4774d60874cdSMiklos Szeredi */ 4775d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4776d60874cdSMiklos Szeredi { 4777d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4778d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4779d60874cdSMiklos Szeredi 4780d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4781d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4782d60874cdSMiklos Szeredi if (!res) 4783d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4784d60874cdSMiklos Szeredi } 4785d60874cdSMiklos Szeredi return res; 4786d60874cdSMiklos Szeredi } 4787d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4788d60874cdSMiklos Szeredi 47891da177e4SLinus Torvalds /* get the link contents into pagecache */ 47906b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4791fceef393SAl Viro struct delayed_call *callback) 47921da177e4SLinus Torvalds { 4793ebd09abbSDuane Griffin char *kaddr; 47941da177e4SLinus Torvalds struct page *page; 47956b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47966b255391SAl Viro 4797d3883d4fSAl Viro if (!dentry) { 4798d3883d4fSAl Viro page = find_get_page(mapping, 0); 4799d3883d4fSAl Viro if (!page) 48006b255391SAl Viro return ERR_PTR(-ECHILD); 4801d3883d4fSAl Viro if (!PageUptodate(page)) { 4802d3883d4fSAl Viro put_page(page); 4803d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4804d3883d4fSAl Viro } 4805d3883d4fSAl Viro } else { 4806090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 48071da177e4SLinus Torvalds if (IS_ERR(page)) 48086fe6900eSNick Piggin return (char*)page; 4809d3883d4fSAl Viro } 4810fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 481121fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 481221fc61c7SAl Viro kaddr = page_address(page); 48136b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4814ebd09abbSDuane Griffin return kaddr; 48151da177e4SLinus Torvalds } 48161da177e4SLinus Torvalds 48176b255391SAl Viro EXPORT_SYMBOL(page_get_link); 48181da177e4SLinus Torvalds 4819fceef393SAl Viro void page_put_link(void *arg) 48201da177e4SLinus Torvalds { 4821fceef393SAl Viro put_page(arg); 48221da177e4SLinus Torvalds } 48234d359507SAl Viro EXPORT_SYMBOL(page_put_link); 48241da177e4SLinus Torvalds 4825aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4826aa80deabSAl Viro { 4827fceef393SAl Viro DEFINE_DELAYED_CALL(done); 48286b255391SAl Viro int res = readlink_copy(buffer, buflen, 48296b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4830fceef393SAl Viro &done)); 4831fceef393SAl Viro do_delayed_call(&done); 4832aa80deabSAl Viro return res; 4833aa80deabSAl Viro } 4834aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4835aa80deabSAl Viro 483654566b2cSNick Piggin /* 483754566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 483854566b2cSNick Piggin */ 483954566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48401da177e4SLinus Torvalds { 48411da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48420adb25d2SKirill Korotaev struct page *page; 4843afddba49SNick Piggin void *fsdata; 4844beb497abSDmitriy Monakhov int err; 4845c718a975STetsuo Handa unsigned int flags = 0; 484654566b2cSNick Piggin if (nofs) 484754566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 48481da177e4SLinus Torvalds 48497e53cac4SNeilBrown retry: 4850afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 485154566b2cSNick Piggin flags, &page, &fsdata); 48521da177e4SLinus Torvalds if (err) 4853afddba49SNick Piggin goto fail; 4854afddba49SNick Piggin 485521fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4856afddba49SNick Piggin 4857afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4858afddba49SNick Piggin page, fsdata); 48591da177e4SLinus Torvalds if (err < 0) 48601da177e4SLinus Torvalds goto fail; 4861afddba49SNick Piggin if (err < len-1) 4862afddba49SNick Piggin goto retry; 4863afddba49SNick Piggin 48641da177e4SLinus Torvalds mark_inode_dirty(inode); 48651da177e4SLinus Torvalds return 0; 48661da177e4SLinus Torvalds fail: 48671da177e4SLinus Torvalds return err; 48681da177e4SLinus Torvalds } 48694d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 48701da177e4SLinus Torvalds 48710adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 48720adb25d2SKirill Korotaev { 48730adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4874c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48750adb25d2SKirill Korotaev } 48764d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48770adb25d2SKirill Korotaev 487892e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48796b255391SAl Viro .get_link = page_get_link, 48801da177e4SLinus Torvalds }; 48811da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4882