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 { 8704b99d499SAleksa Sarai int error = -ELOOP; 8716e77137bSAl Viro struct nameidata *nd = current->nameidata; 872b5fb63c1SChristoph Hellwig 8734b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 8744b99d499SAleksa Sarai goto err; 8754b99d499SAleksa Sarai 8764b99d499SAleksa Sarai path_put(&nd->path); 877b5fb63c1SChristoph Hellwig nd->path = *path; 878b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 879b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 8801bc82070SAleksa Sarai return 0; 8814b99d499SAleksa Sarai 8824b99d499SAleksa Sarai err: 8834b99d499SAleksa Sarai path_put(path); 8844b99d499SAleksa Sarai return error; 885b5fb63c1SChristoph Hellwig } 886b5fb63c1SChristoph Hellwig 887b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 888574197e0SAl Viro { 88921c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 890fceef393SAl Viro do_delayed_call(&last->done); 8916548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 892b9ff4429SAl Viro path_put(&last->link); 893574197e0SAl Viro } 894574197e0SAl Viro 895561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 896561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 89730aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 89830aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 899800179c9SKees Cook 900800179c9SKees Cook /** 901800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 90255852635SRandy Dunlap * @nd: nameidata pathwalk data 903800179c9SKees Cook * 904800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 905800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 906800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 907800179c9SKees Cook * processes from failing races against path names that may change out 908800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 909800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 910800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 911800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 912800179c9SKees Cook * 913800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 914800179c9SKees Cook */ 915fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 916800179c9SKees Cook { 917800179c9SKees Cook const struct inode *inode; 918800179c9SKees Cook const struct inode *parent; 9192d7f9e2aSSeth Forshee kuid_t puid; 920800179c9SKees Cook 921800179c9SKees Cook if (!sysctl_protected_symlinks) 922800179c9SKees Cook return 0; 923800179c9SKees Cook 924800179c9SKees Cook /* Allowed if owner and follower match. */ 925fceef393SAl Viro inode = nd->link_inode; 92681abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 927800179c9SKees Cook return 0; 928800179c9SKees Cook 929800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 930aa65fa35SAl Viro parent = nd->inode; 931800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 932800179c9SKees Cook return 0; 933800179c9SKees Cook 934800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9352d7f9e2aSSeth Forshee puid = parent->i_uid; 9362d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 937800179c9SKees Cook return 0; 938800179c9SKees Cook 93931956502SAl Viro if (nd->flags & LOOKUP_RCU) 94031956502SAl Viro return -ECHILD; 94131956502SAl Viro 942ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 943245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 944800179c9SKees Cook return -EACCES; 945800179c9SKees Cook } 946800179c9SKees Cook 947800179c9SKees Cook /** 948800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 949800179c9SKees Cook * @inode: the source inode to hardlink from 950800179c9SKees Cook * 951800179c9SKees Cook * Return false if at least one of the following conditions: 952800179c9SKees Cook * - inode is not a regular file 953800179c9SKees Cook * - inode is setuid 954800179c9SKees Cook * - inode is setgid and group-exec 955800179c9SKees Cook * - access failure for read and write 956800179c9SKees Cook * 957800179c9SKees Cook * Otherwise returns true. 958800179c9SKees Cook */ 959800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 960800179c9SKees Cook { 961800179c9SKees Cook umode_t mode = inode->i_mode; 962800179c9SKees Cook 963800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 964800179c9SKees Cook if (!S_ISREG(mode)) 965800179c9SKees Cook return false; 966800179c9SKees Cook 967800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 968800179c9SKees Cook if (mode & S_ISUID) 969800179c9SKees Cook return false; 970800179c9SKees Cook 971800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 972800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 973800179c9SKees Cook return false; 974800179c9SKees Cook 975800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 976800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 977800179c9SKees Cook return false; 978800179c9SKees Cook 979800179c9SKees Cook return true; 980800179c9SKees Cook } 981800179c9SKees Cook 982800179c9SKees Cook /** 983800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 984800179c9SKees Cook * @link: the source to hardlink from 985800179c9SKees Cook * 986800179c9SKees Cook * Block hardlink when all of: 987800179c9SKees Cook * - sysctl_protected_hardlinks enabled 988800179c9SKees Cook * - fsuid does not match inode 989800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 990f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 991800179c9SKees Cook * 992800179c9SKees Cook * Returns 0 if successful, -ve on error. 993800179c9SKees Cook */ 994800179c9SKees Cook static int may_linkat(struct path *link) 995800179c9SKees Cook { 996593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 997593d1ce8SEric W. Biederman 998593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 999593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 1000593d1ce8SEric W. Biederman return -EOVERFLOW; 1001800179c9SKees Cook 1002800179c9SKees Cook if (!sysctl_protected_hardlinks) 1003800179c9SKees Cook return 0; 1004800179c9SKees Cook 1005800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1006800179c9SKees Cook * otherwise, it must be a safe source. 1007800179c9SKees Cook */ 1008cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1009800179c9SKees Cook return 0; 1010800179c9SKees Cook 1011245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1012800179c9SKees Cook return -EPERM; 1013800179c9SKees Cook } 1014800179c9SKees Cook 101530aba665SSalvatore Mesoraca /** 101630aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 101730aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 101830aba665SSalvatore Mesoraca * exist. 101930aba665SSalvatore Mesoraca * @dir: the sticky parent directory 102030aba665SSalvatore Mesoraca * @inode: the inode of the file to open 102130aba665SSalvatore Mesoraca * 102230aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 102330aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 102430aba665SSalvatore Mesoraca * - the file already exists 102530aba665SSalvatore Mesoraca * - we are in a sticky directory 102630aba665SSalvatore Mesoraca * - we don't own the file 102730aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 102830aba665SSalvatore Mesoraca * - the directory is world writable 102930aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 103030aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 103130aba665SSalvatore Mesoraca * be enough. 103230aba665SSalvatore Mesoraca * 103330aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 103430aba665SSalvatore Mesoraca */ 103530aba665SSalvatore Mesoraca static int may_create_in_sticky(struct dentry * const dir, 103630aba665SSalvatore Mesoraca struct inode * const inode) 103730aba665SSalvatore Mesoraca { 103830aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 103930aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 104030aba665SSalvatore Mesoraca likely(!(dir->d_inode->i_mode & S_ISVTX)) || 104130aba665SSalvatore Mesoraca uid_eq(inode->i_uid, dir->d_inode->i_uid) || 104230aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 104330aba665SSalvatore Mesoraca return 0; 104430aba665SSalvatore Mesoraca 104530aba665SSalvatore Mesoraca if (likely(dir->d_inode->i_mode & 0002) || 104630aba665SSalvatore Mesoraca (dir->d_inode->i_mode & 0020 && 104730aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 104830aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1049245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1050245d7369SKees Cook "sticky_create_fifo" : 1051245d7369SKees Cook "sticky_create_regular"; 1052245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 105330aba665SSalvatore Mesoraca return -EACCES; 105430aba665SSalvatore Mesoraca } 105530aba665SSalvatore Mesoraca return 0; 105630aba665SSalvatore Mesoraca } 105730aba665SSalvatore Mesoraca 10583b2e7f75SAl Viro static __always_inline 10593b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10601da177e4SLinus Torvalds { 1061ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10621cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1063fceef393SAl Viro struct inode *inode = nd->link_inode; 10646d7b5aaeSAl Viro int error; 10650a959df5SAl Viro const char *res; 10661da177e4SLinus Torvalds 106727812141SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) 106827812141SAleksa Sarai return ERR_PTR(-ELOOP); 106927812141SAleksa Sarai 10708fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10718fa9dd24SNeilBrown touch_atime(&last->link); 10728fa9dd24SNeilBrown cond_resched(); 1073c6718543SMiklos Szeredi } else if (atime_needs_update(&last->link, inode)) { 10744675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10758fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10768fa9dd24SNeilBrown touch_atime(&last->link); 10778fa9dd24SNeilBrown } 10788fa9dd24SNeilBrown 1079bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1080bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1081bda0be7aSNeilBrown if (unlikely(error)) 10826920a440SAl Viro return ERR_PTR(error); 108336f3b4f6SAl Viro 108486acdca1SAl Viro nd->last_type = LAST_BIND; 10854c4f7c19SEric Biggers res = READ_ONCE(inode->i_link); 1086d4dee48bSAl Viro if (!res) { 1087fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1088fceef393SAl Viro struct delayed_call *); 1089fceef393SAl Viro get = inode->i_op->get_link; 10908c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1091fceef393SAl Viro res = get(NULL, inode, &last->done); 10926b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10934675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10948c1b4566SAl Viro return ERR_PTR(-ECHILD); 1095fceef393SAl Viro res = get(dentry, inode, &last->done); 10968c1b4566SAl Viro } 10976b255391SAl Viro } else { 1098fceef393SAl Viro res = get(dentry, inode, &last->done); 10996b255391SAl Viro } 1100fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1101fab51e8aSAl Viro return res; 11020a959df5SAl Viro } 1103fab51e8aSAl Viro if (*res == '/') { 1104740a1678SAleksa Sarai error = nd_jump_root(nd); 1105740a1678SAleksa Sarai if (unlikely(error)) 1106740a1678SAleksa Sarai return ERR_PTR(error); 1107fab51e8aSAl Viro while (unlikely(*++res == '/')) 1108fab51e8aSAl Viro ; 1109fab51e8aSAl Viro } 1110fab51e8aSAl Viro if (!*res) 1111fab51e8aSAl Viro res = NULL; 11120a959df5SAl Viro return res; 11130a959df5SAl Viro } 11146d7b5aaeSAl Viro 1115f015f126SDavid Howells /* 1116f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1117f015f126SDavid Howells * 1118f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1119f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1120f015f126SDavid Howells * Up is towards /. 1121f015f126SDavid Howells * 1122f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1123f015f126SDavid Howells * root. 1124f015f126SDavid Howells */ 1125bab77ebfSAl Viro int follow_up(struct path *path) 11261da177e4SLinus Torvalds { 11270714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11280714a533SAl Viro struct mount *parent; 11291da177e4SLinus Torvalds struct dentry *mountpoint; 113099b7db7bSNick Piggin 113148a066e7SAl Viro read_seqlock_excl(&mount_lock); 11320714a533SAl Viro parent = mnt->mnt_parent; 11333c0a6163SAl Viro if (parent == mnt) { 113448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11351da177e4SLinus Torvalds return 0; 11361da177e4SLinus Torvalds } 11370714a533SAl Viro mntget(&parent->mnt); 1138a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 113948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1140bab77ebfSAl Viro dput(path->dentry); 1141bab77ebfSAl Viro path->dentry = mountpoint; 1142bab77ebfSAl Viro mntput(path->mnt); 11430714a533SAl Viro path->mnt = &parent->mnt; 11441da177e4SLinus Torvalds return 1; 11451da177e4SLinus Torvalds } 11464d359507SAl Viro EXPORT_SYMBOL(follow_up); 11471da177e4SLinus Torvalds 1148b5c84bf6SNick Piggin /* 11499875cf80SDavid Howells * Perform an automount 11509875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11519875cf80SDavid Howells * were called with. 11521da177e4SLinus Torvalds */ 1153756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 11549875cf80SDavid Howells bool *need_mntput) 115531e6b01fSNick Piggin { 11569875cf80SDavid Howells struct vfsmount *mnt; 1157ea5b778aSDavid Howells int err; 11589875cf80SDavid Howells 11599875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 11609875cf80SDavid Howells return -EREMOTE; 11619875cf80SDavid Howells 11620ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11630ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11640ec26fd0SMiklos Szeredi * the name. 11650ec26fd0SMiklos Szeredi * 11660ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11675a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11680ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11690ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11700ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11710ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11725a30d8a2SDavid Howells */ 1173756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11745d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11755d38f049SIan Kent path->dentry->d_inode) 11769875cf80SDavid Howells return -EISDIR; 11770ec26fd0SMiklos Szeredi 1178756daf26SNeilBrown nd->total_link_count++; 1179756daf26SNeilBrown if (nd->total_link_count >= 40) 11809875cf80SDavid Howells return -ELOOP; 11819875cf80SDavid Howells 11829875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11839875cf80SDavid Howells if (IS_ERR(mnt)) { 11849875cf80SDavid Howells /* 11859875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11869875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11879875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11889875cf80SDavid Howells * 11899875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11909875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11919875cf80SDavid Howells * the path is inaccessible and we should say so. 11929875cf80SDavid Howells */ 1193756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11949875cf80SDavid Howells return -EREMOTE; 11959875cf80SDavid Howells return PTR_ERR(mnt); 119631e6b01fSNick Piggin } 1197ea5b778aSDavid Howells 11989875cf80SDavid Howells if (!mnt) /* mount collision */ 11999875cf80SDavid Howells return 0; 12009875cf80SDavid Howells 12018aef1884SAl Viro if (!*need_mntput) { 12028aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 12038aef1884SAl Viro mntget(path->mnt); 12048aef1884SAl Viro *need_mntput = true; 12058aef1884SAl Viro } 120619a167afSAl Viro err = finish_automount(mnt, path); 1207ea5b778aSDavid Howells 1208ea5b778aSDavid Howells switch (err) { 1209ea5b778aSDavid Howells case -EBUSY: 1210ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 121119a167afSAl Viro return 0; 1212ea5b778aSDavid Howells case 0: 12138aef1884SAl Viro path_put(path); 12149875cf80SDavid Howells path->mnt = mnt; 12159875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 12169875cf80SDavid Howells return 0; 121719a167afSAl Viro default: 121819a167afSAl Viro return err; 12199875cf80SDavid Howells } 122019a167afSAl Viro 1221ea5b778aSDavid Howells } 12229875cf80SDavid Howells 12239875cf80SDavid Howells /* 12249875cf80SDavid Howells * Handle a dentry that is managed in some way. 1225cc53ce53SDavid Howells * - Flagged for transit management (autofs) 12269875cf80SDavid Howells * - Flagged as mountpoint 12279875cf80SDavid Howells * - Flagged as automount point 12289875cf80SDavid Howells * 12299875cf80SDavid Howells * This may only be called in refwalk mode. 1230d41efb52SAl Viro * On success path->dentry is known positive. 12319875cf80SDavid Howells * 12329875cf80SDavid Howells * Serialization is taken care of in namespace.c 12339875cf80SDavid Howells */ 1234756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 12359875cf80SDavid Howells { 12368aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 1237d41efb52SAl Viro unsigned flags; 12389875cf80SDavid Howells bool need_mntput = false; 12398aef1884SAl Viro int ret = 0; 12409875cf80SDavid Howells 12419875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 12429875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 12439875cf80SDavid Howells * the components of that value change under us */ 12442fa6b1e0SAl Viro while (flags = smp_load_acquire(&path->dentry->d_flags), 1245d41efb52SAl Viro unlikely(flags & DCACHE_MANAGED_DENTRY)) { 1246cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1247cc53ce53SDavid Howells * being held. */ 1248d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1249cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1250cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1251fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1252cc53ce53SDavid Howells if (ret < 0) 12538aef1884SAl Viro break; 1254cc53ce53SDavid Howells } 1255cc53ce53SDavid Howells 12569875cf80SDavid Howells /* Transit to a mounted filesystem. */ 1257d41efb52SAl Viro if (flags & DCACHE_MOUNTED) { 12589875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12599875cf80SDavid Howells if (mounted) { 12609875cf80SDavid Howells dput(path->dentry); 12619875cf80SDavid Howells if (need_mntput) 1262463ffb2eSAl Viro mntput(path->mnt); 1263463ffb2eSAl Viro path->mnt = mounted; 1264463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12659875cf80SDavid Howells need_mntput = true; 12669875cf80SDavid Howells continue; 1267463ffb2eSAl Viro } 1268463ffb2eSAl Viro 12699875cf80SDavid Howells /* Something is mounted on this dentry in another 12709875cf80SDavid Howells * namespace and/or whatever was mounted there in this 127148a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 127248a066e7SAl Viro * get it */ 12731da177e4SLinus Torvalds } 12749875cf80SDavid Howells 12759875cf80SDavid Howells /* Handle an automount point */ 1276d41efb52SAl Viro if (flags & DCACHE_NEED_AUTOMOUNT) { 1277756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12789875cf80SDavid Howells if (ret < 0) 12798aef1884SAl Viro break; 12809875cf80SDavid Howells continue; 12819875cf80SDavid Howells } 12829875cf80SDavid Howells 12839875cf80SDavid Howells /* We didn't change the current path point */ 12849875cf80SDavid Howells break; 12859875cf80SDavid Howells } 12868aef1884SAl Viro 12878aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12888aef1884SAl Viro mntput(path->mnt); 12898402752eSAl Viro if (need_mntput) 12908402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 1291d41efb52SAl Viro if (ret == -EISDIR || !ret) 1292d41efb52SAl Viro ret = 1; 1293d41efb52SAl Viro if (ret > 0 && unlikely(d_flags_negative(flags))) 1294d41efb52SAl Viro ret = -ENOENT; 12958402752eSAl Viro if (unlikely(ret < 0)) 12968402752eSAl Viro path_put_conditional(path, nd); 12978402752eSAl Viro return ret; 12981da177e4SLinus Torvalds } 12991da177e4SLinus Torvalds 1300cc53ce53SDavid Howells int follow_down_one(struct path *path) 13011da177e4SLinus Torvalds { 13021da177e4SLinus Torvalds struct vfsmount *mounted; 13031da177e4SLinus Torvalds 13041c755af4SAl Viro mounted = lookup_mnt(path); 13051da177e4SLinus Torvalds if (mounted) { 13069393bd07SAl Viro dput(path->dentry); 13079393bd07SAl Viro mntput(path->mnt); 13089393bd07SAl Viro path->mnt = mounted; 13099393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 13101da177e4SLinus Torvalds return 1; 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds return 0; 13131da177e4SLinus Torvalds } 13144d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 13151da177e4SLinus Torvalds 1316fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 131762a7375eSIan Kent { 1318fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1319fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 132062a7375eSIan Kent } 132162a7375eSIan Kent 13229875cf80SDavid Howells /* 1323287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1324287548e4SAl Viro * we meet a managed dentry that would need blocking. 13259875cf80SDavid Howells */ 13269875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1327254cf582SAl Viro struct inode **inode, unsigned *seqp) 13289875cf80SDavid Howells { 132962a7375eSIan Kent for (;;) { 1330c7105365SAl Viro struct mount *mounted; 133162a7375eSIan Kent /* 133262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 133362a7375eSIan Kent * that wants to block transit. 133462a7375eSIan Kent */ 1335fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1336b8faf035SNeilBrown case -ECHILD: 1337b8faf035SNeilBrown default: 1338ab90911fSDavid Howells return false; 1339b8faf035SNeilBrown case -EISDIR: 1340b8faf035SNeilBrown return true; 1341b8faf035SNeilBrown case 0: 1342b8faf035SNeilBrown break; 1343b8faf035SNeilBrown } 134462a7375eSIan Kent 134562a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1346b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 134762a7375eSIan Kent 1348474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 13499875cf80SDavid Howells if (!mounted) 13509875cf80SDavid Howells break; 1351c7105365SAl Viro path->mnt = &mounted->mnt; 1352c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1353a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1354254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 135559430262SLinus Torvalds /* 135659430262SLinus Torvalds * Update the inode too. We don't need to re-check the 135759430262SLinus Torvalds * dentry sequence number here after this d_inode read, 135859430262SLinus Torvalds * because a mount-point is always pinned. 135959430262SLinus Torvalds */ 136059430262SLinus Torvalds *inode = path->dentry->d_inode; 13619875cf80SDavid Howells } 1362f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1363b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1364287548e4SAl Viro } 1365287548e4SAl Viro 136631e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 136731e6b01fSNick Piggin { 13684023bfc9SAl Viro struct inode *inode = nd->inode; 136931e6b01fSNick Piggin 137031e6b01fSNick Piggin while (1) { 1371aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 137231e6b01fSNick Piggin break; 137331e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 137431e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 137531e6b01fSNick Piggin struct dentry *parent = old->d_parent; 137631e6b01fSNick Piggin unsigned seq; 137731e6b01fSNick Piggin 13784023bfc9SAl Viro inode = parent->d_inode; 137931e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1380aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1381aed434adSAl Viro return -ECHILD; 138231e6b01fSNick Piggin nd->path.dentry = parent; 138331e6b01fSNick Piggin nd->seq = seq; 1384397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 13852b98149cSAleksa Sarai return -ECHILD; 138631e6b01fSNick Piggin break; 1387aed434adSAl Viro } else { 1388aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1389aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1390aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1391aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1392aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1393aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1394aed434adSAl Viro return -ECHILD; 1395aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 139631e6b01fSNick Piggin break; 1397aed434adSAl Viro /* we know that mountpoint was pinned */ 1398aed434adSAl Viro nd->path.dentry = mountpoint; 1399aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1400aed434adSAl Viro inode = inode2; 1401aed434adSAl Viro nd->seq = seq; 140231e6b01fSNick Piggin } 1403aed434adSAl Viro } 1404aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1405b37199e6SAl Viro struct mount *mounted; 1406b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1407aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1408aed434adSAl Viro return -ECHILD; 1409b37199e6SAl Viro if (!mounted) 1410b37199e6SAl Viro break; 1411b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1412b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 14134023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1414b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1415b37199e6SAl Viro } 14164023bfc9SAl Viro nd->inode = inode; 141731e6b01fSNick Piggin return 0; 141831e6b01fSNick Piggin } 141931e6b01fSNick Piggin 14209875cf80SDavid Howells /* 1421cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1422cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1423cc53ce53SDavid Howells * caller is permitted to proceed or not. 1424cc53ce53SDavid Howells */ 14257cc90cc3SAl Viro int follow_down(struct path *path) 1426cc53ce53SDavid Howells { 1427cc53ce53SDavid Howells unsigned managed; 1428cc53ce53SDavid Howells int ret; 1429cc53ce53SDavid Howells 14306aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1431cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1432cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1433cc53ce53SDavid Howells * being held. 1434cc53ce53SDavid Howells * 1435cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1436cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1437cc53ce53SDavid Howells * other than its daemon the right to mount on its 1438cc53ce53SDavid Howells * superstructure. 1439cc53ce53SDavid Howells * 1440cc53ce53SDavid Howells * The filesystem may sleep at this point. 1441cc53ce53SDavid Howells */ 1442cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1443cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1444cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1445fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1446cc53ce53SDavid Howells if (ret < 0) 1447cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1448cc53ce53SDavid Howells } 1449cc53ce53SDavid Howells 1450cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1451cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1452cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1453cc53ce53SDavid Howells if (!mounted) 1454cc53ce53SDavid Howells break; 1455cc53ce53SDavid Howells dput(path->dentry); 1456cc53ce53SDavid Howells mntput(path->mnt); 1457cc53ce53SDavid Howells path->mnt = mounted; 1458cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1459cc53ce53SDavid Howells continue; 1460cc53ce53SDavid Howells } 1461cc53ce53SDavid Howells 1462cc53ce53SDavid Howells /* Don't handle automount points here */ 1463cc53ce53SDavid Howells break; 1464cc53ce53SDavid Howells } 1465cc53ce53SDavid Howells return 0; 1466cc53ce53SDavid Howells } 14674d359507SAl Viro EXPORT_SYMBOL(follow_down); 1468cc53ce53SDavid Howells 1469cc53ce53SDavid Howells /* 14709875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14719875cf80SDavid Howells */ 14729875cf80SDavid Howells static void follow_mount(struct path *path) 14739875cf80SDavid Howells { 14749875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14759875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14769875cf80SDavid Howells if (!mounted) 14779875cf80SDavid Howells break; 14789875cf80SDavid Howells dput(path->dentry); 14799875cf80SDavid Howells mntput(path->mnt); 14809875cf80SDavid Howells path->mnt = mounted; 14819875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14829875cf80SDavid Howells } 14839875cf80SDavid Howells } 14849875cf80SDavid Howells 1485eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1486eedf265aSEric W. Biederman { 1487eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1488eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1489eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1490eedf265aSEric W. Biederman dput(old); 1491eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1492eedf265aSEric W. Biederman return -ENOENT; 1493eedf265aSEric W. Biederman return 0; 1494eedf265aSEric W. Biederman } 1495eedf265aSEric W. Biederman 1496397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14971da177e4SLinus Torvalds { 14981da177e4SLinus Torvalds while(1) { 1499030c7e0bSDanilo Krummrich if (path_equal(&nd->path, &nd->root)) 15001da177e4SLinus Torvalds break; 15014ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1502eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1503eedf265aSEric W. Biederman if (ret) 1504eedf265aSEric W. Biederman return ret; 15051da177e4SLinus Torvalds break; 15061da177e4SLinus Torvalds } 15073088dd70SAl Viro if (!follow_up(&nd->path)) 15081da177e4SLinus Torvalds break; 15091da177e4SLinus Torvalds } 151079ed0226SAl Viro follow_mount(&nd->path); 151131e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1512397d425dSEric W. Biederman return 0; 15131da177e4SLinus Torvalds } 15141da177e4SLinus Torvalds 15151da177e4SLinus Torvalds /* 1516f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1517f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1518baa03890SNick Piggin */ 1519e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1520e3c13928SAl Viro struct dentry *dir, 15216c51e513SAl Viro unsigned int flags) 1522baa03890SNick Piggin { 1523a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1524bad61189SMiklos Szeredi if (dentry) { 1525a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1526bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 152774ff0ffcSAl Viro if (!error) 15285542aa2fSEric W. Biederman d_invalidate(dentry); 1529bad61189SMiklos Szeredi dput(dentry); 153074ff0ffcSAl Viro return ERR_PTR(error); 1531bad61189SMiklos Szeredi } 1532bad61189SMiklos Szeredi } 1533baa03890SNick Piggin return dentry; 1534baa03890SNick Piggin } 1535baa03890SNick Piggin 1536baa03890SNick Piggin /* 1537a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1538a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1539a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1540a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1541a03ece5fSAl Viro * at all. 154244396f4bSJosef Bacik */ 1543a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1544a03ece5fSAl Viro struct dentry *base, unsigned int flags) 154544396f4bSJosef Bacik { 1546a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 154744396f4bSJosef Bacik struct dentry *old; 1548a03ece5fSAl Viro struct inode *dir = base->d_inode; 1549a03ece5fSAl Viro 1550a03ece5fSAl Viro if (dentry) 1551a03ece5fSAl Viro return dentry; 155244396f4bSJosef Bacik 155344396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1554a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 155544396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1556a03ece5fSAl Viro 1557a03ece5fSAl Viro dentry = d_alloc(base, name); 1558a03ece5fSAl Viro if (unlikely(!dentry)) 1559a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 156044396f4bSJosef Bacik 156172bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 156244396f4bSJosef Bacik if (unlikely(old)) { 156344396f4bSJosef Bacik dput(dentry); 156444396f4bSJosef Bacik dentry = old; 156544396f4bSJosef Bacik } 156644396f4bSJosef Bacik return dentry; 156744396f4bSJosef Bacik } 156844396f4bSJosef Bacik 1569e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1570254cf582SAl Viro struct path *path, struct inode **inode, 1571254cf582SAl Viro unsigned *seqp) 15721da177e4SLinus Torvalds { 15734ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 157431e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15755a18fff2SAl Viro int status = 1; 15769875cf80SDavid Howells int err; 15779875cf80SDavid Howells 15783cac260aSAl Viro /* 1579b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15805d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15815d0f49c1SAl Viro * going to fall back to non-racy lookup. 1582b04f784eSNick Piggin */ 158331e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 158431e6b01fSNick Piggin unsigned seq; 1585766c4cbfSAl Viro bool negative; 1586da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15875d0f49c1SAl Viro if (unlikely(!dentry)) { 15884675ac39SAl Viro if (unlazy_walk(nd)) 15895d0f49c1SAl Viro return -ECHILD; 1590e9742b53SAl Viro return 0; 15915d0f49c1SAl Viro } 15925a18fff2SAl Viro 159312f8ad4bSLinus Torvalds /* 159412f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 159512f8ad4bSLinus Torvalds * the dentry name information from lookup. 159612f8ad4bSLinus Torvalds */ 159763afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1598766c4cbfSAl Viro negative = d_is_negative(dentry); 15995d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 160012f8ad4bSLinus Torvalds return -ECHILD; 160112f8ad4bSLinus Torvalds 160212f8ad4bSLinus Torvalds /* 160312f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 160412f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 160512f8ad4bSLinus Torvalds * 160612f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 160712f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 160812f8ad4bSLinus Torvalds */ 16095d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 161031e6b01fSNick Piggin return -ECHILD; 16115a18fff2SAl Viro 1612254cf582SAl Viro *seqp = seq; 16134ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1614209a7fb2SAl Viro if (likely(status > 0)) { 1615daf3761cSTrond Myklebust /* 1616daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1617daf3761cSTrond Myklebust * case that drops it. 1618daf3761cSTrond Myklebust */ 16195d0f49c1SAl Viro if (unlikely(negative)) 1620daf3761cSTrond Myklebust return -ENOENT; 162131e6b01fSNick Piggin path->mnt = mnt; 162231e6b01fSNick Piggin path->dentry = dentry; 1623254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1624e9742b53SAl Viro return 1; 16255d0f49c1SAl Viro } 16264675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 1627254cf582SAl Viro return -ECHILD; 1628209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1629209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1630209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16315a18fff2SAl Viro } else { 1632e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 163381e6f520SAl Viro if (unlikely(!dentry)) 1634e9742b53SAl Viro return 0; 16354ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16365d0f49c1SAl Viro } 16375a18fff2SAl Viro if (unlikely(status <= 0)) { 1638e9742b53SAl Viro if (!status) 16395d0f49c1SAl Viro d_invalidate(dentry); 16405a18fff2SAl Viro dput(dentry); 16415a18fff2SAl Viro return status; 16425a18fff2SAl Viro } 16435d0f49c1SAl Viro 16441da177e4SLinus Torvalds path->mnt = mnt; 16451da177e4SLinus Torvalds path->dentry = dentry; 1646756daf26SNeilBrown err = follow_managed(path, nd); 1647e9742b53SAl Viro if (likely(err > 0)) 164863afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16498402752eSAl Viro return err; 1650697f514dSMiklos Szeredi } 1651697f514dSMiklos Szeredi 1652697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 165388d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1654e3c13928SAl Viro struct dentry *dir, 1655e3c13928SAl Viro unsigned int flags) 1656697f514dSMiklos Szeredi { 165788d8331aSAl Viro struct dentry *dentry, *old; 16581936386eSAl Viro struct inode *inode = dir->d_inode; 1659d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16601936386eSAl Viro 16611936386eSAl Viro /* Don't go there if it's already dead */ 166294bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 166388d8331aSAl Viro return ERR_PTR(-ENOENT); 166494bdd655SAl Viro again: 1665d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 166694bdd655SAl Viro if (IS_ERR(dentry)) 166788d8331aSAl Viro return dentry; 166894bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1669a89f8337SAl Viro if (!(flags & LOOKUP_NO_REVAL)) { 1670949a852eSAl Viro int error = d_revalidate(dentry, flags); 1671949a852eSAl Viro if (unlikely(error <= 0)) { 167294bdd655SAl Viro if (!error) { 1673949a852eSAl Viro d_invalidate(dentry); 1674949a852eSAl Viro dput(dentry); 167594bdd655SAl Viro goto again; 167694bdd655SAl Viro } 167794bdd655SAl Viro dput(dentry); 1678949a852eSAl Viro dentry = ERR_PTR(error); 1679949a852eSAl Viro } 1680949a852eSAl Viro } 168194bdd655SAl Viro } else { 16821936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 168385c7f810SAl Viro d_lookup_done(dentry); 16841936386eSAl Viro if (unlikely(old)) { 16851936386eSAl Viro dput(dentry); 16861936386eSAl Viro dentry = old; 1687949a852eSAl Viro } 1688949a852eSAl Viro } 1689e3c13928SAl Viro return dentry; 16901da177e4SLinus Torvalds } 16911da177e4SLinus Torvalds 169288d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 169388d8331aSAl Viro struct dentry *dir, 169488d8331aSAl Viro unsigned int flags) 169588d8331aSAl Viro { 169688d8331aSAl Viro struct inode *inode = dir->d_inode; 169788d8331aSAl Viro struct dentry *res; 169888d8331aSAl Viro inode_lock_shared(inode); 169988d8331aSAl Viro res = __lookup_slow(name, dir, flags); 170088d8331aSAl Viro inode_unlock_shared(inode); 170188d8331aSAl Viro return res; 170288d8331aSAl Viro } 170388d8331aSAl Viro 170452094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 170552094c8aSAl Viro { 170652094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17074ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 170852094c8aSAl Viro if (err != -ECHILD) 170952094c8aSAl Viro return err; 17104675ac39SAl Viro if (unlazy_walk(nd)) 171152094c8aSAl Viro return -ECHILD; 171252094c8aSAl Viro } 17134ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 171452094c8aSAl Viro } 171552094c8aSAl Viro 17169856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 17179856fa1bSAl Viro { 17189856fa1bSAl Viro if (type == LAST_DOTDOT) { 1719740a1678SAleksa Sarai int error = 0; 1720740a1678SAleksa Sarai 1721740a1678SAleksa Sarai if (!nd->root.mnt) { 1722740a1678SAleksa Sarai error = set_root(nd); 1723740a1678SAleksa Sarai if (error) 1724740a1678SAleksa Sarai return error; 1725740a1678SAleksa Sarai } 17269856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 172770291aecSAl Viro return follow_dotdot_rcu(nd); 17289856fa1bSAl Viro } else 1729397d425dSEric W. Biederman return follow_dotdot(nd); 17309856fa1bSAl Viro } 17319856fa1bSAl Viro return 0; 17329856fa1bSAl Viro } 17339856fa1bSAl Viro 1734181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1735181548c0SAl Viro struct inode *inode, unsigned seq) 1736d63ff28fSAl Viro { 1737626de996SAl Viro int error; 17381cf2665bSAl Viro struct saved *last; 1739756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1740626de996SAl Viro path_to_nameidata(link, nd); 1741626de996SAl Viro return -ELOOP; 1742626de996SAl Viro } 1743bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1744cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1745cd179f44SAl Viro mntget(link->mnt); 17467973387aSAl Viro } 1747626de996SAl Viro error = nd_alloc_stack(nd); 1748626de996SAl Viro if (unlikely(error)) { 1749bc40aee0SAl Viro if (error == -ECHILD) { 1750ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1751ad1633a1SAl Viro drop_links(nd); 1752ad1633a1SAl Viro nd->depth = 0; 1753ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1754ad1633a1SAl Viro nd->path.mnt = NULL; 1755ad1633a1SAl Viro nd->path.dentry = NULL; 1756ad1633a1SAl Viro rcu_read_unlock(); 17574675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1758bc40aee0SAl Viro error = nd_alloc_stack(nd); 1759bc40aee0SAl Viro } 1760bc40aee0SAl Viro if (error) { 1761cd179f44SAl Viro path_put(link); 1762626de996SAl Viro return error; 1763626de996SAl Viro } 1764bc40aee0SAl Viro } 1765626de996SAl Viro 1766ab104923SAl Viro last = nd->stack + nd->depth++; 17671cf2665bSAl Viro last->link = *link; 1768fceef393SAl Viro clear_delayed_call(&last->done); 1769fceef393SAl Viro nd->link_inode = inode; 17700450b2d1SAl Viro last->seq = seq; 1771d63ff28fSAl Viro return 1; 1772d63ff28fSAl Viro } 1773d63ff28fSAl Viro 17748f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 177531d66bcdSAl Viro 17763ddcd056SLinus Torvalds /* 17773ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17783ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17793ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17803ddcd056SLinus Torvalds * for the common case. 17813ddcd056SLinus Torvalds */ 17828f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 17838f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 17843ddcd056SLinus Torvalds { 178531d66bcdSAl Viro if (!(flags & WALK_MORE) && nd->depth) 178631d66bcdSAl Viro put_link(nd); 17878f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 17888f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 17898f64fb1cSAl Viro /* not a symlink or should not follow */ 17908f64fb1cSAl Viro path_to_nameidata(path, nd); 17918f64fb1cSAl Viro nd->inode = inode; 17928f64fb1cSAl Viro nd->seq = seq; 1793d63ff28fSAl Viro return 0; 17948f64fb1cSAl Viro } 1795a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1796a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 17978f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1798a7f77542SAl Viro return -ECHILD; 1799a7f77542SAl Viro } 18008f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 18013ddcd056SLinus Torvalds } 18023ddcd056SLinus Torvalds 18034693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1804ce57dfc1SAl Viro { 1805caa85634SAl Viro struct path path; 1806ce57dfc1SAl Viro struct inode *inode; 1807254cf582SAl Viro unsigned seq; 1808ce57dfc1SAl Viro int err; 1809ce57dfc1SAl Viro /* 1810ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1811ce57dfc1SAl Viro * to be able to know about the current root directory and 1812ce57dfc1SAl Viro * parent relationships. 1813ce57dfc1SAl Viro */ 18144693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18154693a547SAl Viro err = handle_dots(nd, nd->last_type); 18161c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18174693a547SAl Viro put_link(nd); 18184693a547SAl Viro return err; 18194693a547SAl Viro } 1820254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1821e9742b53SAl Viro if (unlikely(err <= 0)) { 1822697f514dSMiklos Szeredi if (err < 0) 1823f0a9ba70SAl Viro return err; 1824e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1825e3c13928SAl Viro nd->flags); 1826e3c13928SAl Viro if (IS_ERR(path.dentry)) 1827e3c13928SAl Viro return PTR_ERR(path.dentry); 18287500c38aSAl Viro 1829e3c13928SAl Viro path.mnt = nd->path.mnt; 1830e3c13928SAl Viro err = follow_managed(&path, nd); 1831e3c13928SAl Viro if (unlikely(err < 0)) 1832f0a9ba70SAl Viro return err; 1833697f514dSMiklos Szeredi 1834254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1835d4565649SAl Viro inode = d_backing_inode(path.dentry); 1836766c4cbfSAl Viro } 1837697f514dSMiklos Szeredi 18388f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1839ce57dfc1SAl Viro } 1840ce57dfc1SAl Viro 18411da177e4SLinus Torvalds /* 1842bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1843bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1844bfcfaa77SLinus Torvalds * 1845bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1846bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1847bfcfaa77SLinus Torvalds * fast. 1848bfcfaa77SLinus Torvalds * 1849bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1850bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1851bfcfaa77SLinus Torvalds * crossing operation. 1852bfcfaa77SLinus Torvalds * 1853bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1854bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1855bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1856bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1857bfcfaa77SLinus Torvalds */ 1858bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1859bfcfaa77SLinus Torvalds 1860f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1861bfcfaa77SLinus Torvalds 1862468a9428SGeorge Spelvin #ifdef HASH_MIX 1863bfcfaa77SLinus Torvalds 1864468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1865468a9428SGeorge Spelvin 1866468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18672a18da7aSGeorge Spelvin /* 18682a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18692a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18702a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18712a18da7aSGeorge Spelvin * and no temporaries. 18722a18da7aSGeorge Spelvin * 18732a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18742a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18752a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18762a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18772a18da7aSGeorge Spelvin * 18782a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18792a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18802a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18812a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18822a18da7aSGeorge Spelvin * 18832a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18842a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18852a18da7aSGeorge Spelvin * toggling any given output bit. 18862a18da7aSGeorge Spelvin * 18872a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18882a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18892a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18902a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18912a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18922a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18932a18da7aSGeorge Spelvin * Perfect: 8192 258048 18942a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18952a18da7aSGeorge Spelvin */ 18962a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18972a18da7aSGeorge Spelvin ( x ^= (a), \ 18982a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18992a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 19002a18da7aSGeorge Spelvin y *= 9 ) 1901bfcfaa77SLinus Torvalds 19020fed3ac8SGeorge Spelvin /* 19032a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 19042a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 19052a18da7aSGeorge Spelvin * work done before the hash value is used. 19060fed3ac8SGeorge Spelvin */ 19072a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19080fed3ac8SGeorge Spelvin { 19092a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 19102a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19112a18da7aSGeorge Spelvin return y >> 32; 19120fed3ac8SGeorge Spelvin } 19130fed3ac8SGeorge Spelvin 1914bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1915bfcfaa77SLinus Torvalds 19162a18da7aSGeorge Spelvin /* 19172a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19182a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19192a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19202a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19212a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19222a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19232a18da7aSGeorge Spelvin * Perfect: 2048 31744 19242a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19252a18da7aSGeorge Spelvin */ 19262a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19272a18da7aSGeorge Spelvin ( x ^= (a), \ 19282a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19292a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19302a18da7aSGeorge Spelvin y *= 9 ) 1931bfcfaa77SLinus Torvalds 19322a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19330fed3ac8SGeorge Spelvin { 19342a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19352a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19360fed3ac8SGeorge Spelvin } 19370fed3ac8SGeorge Spelvin 1938bfcfaa77SLinus Torvalds #endif 1939bfcfaa77SLinus Torvalds 19402a18da7aSGeorge Spelvin /* 19412a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19422a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19432a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19442a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19452a18da7aSGeorge Spelvin * finds the delimiter after the name. 19462a18da7aSGeorge Spelvin */ 19478387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1948bfcfaa77SLinus Torvalds { 19498387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1950bfcfaa77SLinus Torvalds 1951bfcfaa77SLinus Torvalds for (;;) { 1952fcfd2fbfSGeorge Spelvin if (!len) 1953fcfd2fbfSGeorge Spelvin goto done; 1954e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1955bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1956bfcfaa77SLinus Torvalds break; 19572a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1958bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1959bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1960bfcfaa77SLinus Torvalds } 19612a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1962bfcfaa77SLinus Torvalds done: 19632a18da7aSGeorge Spelvin return fold_hash(x, y); 1964bfcfaa77SLinus Torvalds } 1965bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1966bfcfaa77SLinus Torvalds 1967fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19688387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1969fcfd2fbfSGeorge Spelvin { 19708387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19718387ff25SLinus Torvalds unsigned long adata, mask, len; 1972fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1973fcfd2fbfSGeorge Spelvin 19748387ff25SLinus Torvalds len = 0; 19758387ff25SLinus Torvalds goto inside; 19768387ff25SLinus Torvalds 1977fcfd2fbfSGeorge Spelvin do { 19782a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1979fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19808387ff25SLinus Torvalds inside: 1981fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1982fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1983fcfd2fbfSGeorge Spelvin 1984fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1985fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19862a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1987fcfd2fbfSGeorge Spelvin 19882a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1989fcfd2fbfSGeorge Spelvin } 1990fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1991fcfd2fbfSGeorge Spelvin 1992bfcfaa77SLinus Torvalds /* 1993bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1994d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1995bfcfaa77SLinus Torvalds */ 19968387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1997bfcfaa77SLinus Torvalds { 19988387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 19998387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 200036126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2001bfcfaa77SLinus Torvalds 20028387ff25SLinus Torvalds len = 0; 20038387ff25SLinus Torvalds goto inside; 20048387ff25SLinus Torvalds 2005bfcfaa77SLinus Torvalds do { 20062a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2007bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 20088387ff25SLinus Torvalds inside: 2009e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 201036126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 201136126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2012bfcfaa77SLinus Torvalds 201336126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 201436126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 201536126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20162a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 201736126f8fSLinus Torvalds 20182a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2019bfcfaa77SLinus Torvalds } 2020bfcfaa77SLinus Torvalds 20212a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2022bfcfaa77SLinus Torvalds 2023fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20248387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20250145acc2SLinus Torvalds { 20268387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20270145acc2SLinus Torvalds while (len--) 2028fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20290145acc2SLinus Torvalds return end_name_hash(hash); 20300145acc2SLinus Torvalds } 2031ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20320145acc2SLinus Torvalds 2033fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20348387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2035fcfd2fbfSGeorge Spelvin { 20368387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2037fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2038fcfd2fbfSGeorge Spelvin 2039fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2040e0ab7af9SGeorge Spelvin while (c) { 2041fcfd2fbfSGeorge Spelvin len++; 2042fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2043fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2044e0ab7af9SGeorge Spelvin } 2045fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2046fcfd2fbfSGeorge Spelvin } 2047f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2048fcfd2fbfSGeorge Spelvin 20493ddcd056SLinus Torvalds /* 2050200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2051200e9ef7SLinus Torvalds * one character. 2052200e9ef7SLinus Torvalds */ 20538387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2054200e9ef7SLinus Torvalds { 20558387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2056200e9ef7SLinus Torvalds unsigned long len = 0, c; 2057200e9ef7SLinus Torvalds 2058200e9ef7SLinus Torvalds c = (unsigned char)*name; 2059200e9ef7SLinus Torvalds do { 2060200e9ef7SLinus Torvalds len++; 2061200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2062200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2063200e9ef7SLinus Torvalds } while (c && c != '/'); 2064d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2065200e9ef7SLinus Torvalds } 2066200e9ef7SLinus Torvalds 2067bfcfaa77SLinus Torvalds #endif 2068bfcfaa77SLinus Torvalds 2069200e9ef7SLinus Torvalds /* 20701da177e4SLinus Torvalds * Name resolution. 2071ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2072ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20731da177e4SLinus Torvalds * 2074ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2075ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20761da177e4SLinus Torvalds */ 20776de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20781da177e4SLinus Torvalds { 20791da177e4SLinus Torvalds int err; 20801da177e4SLinus Torvalds 20819b5858e9SAl Viro if (IS_ERR(name)) 20829b5858e9SAl Viro return PTR_ERR(name); 20831da177e4SLinus Torvalds while (*name=='/') 20841da177e4SLinus Torvalds name++; 20851da177e4SLinus Torvalds if (!*name) 20869e18f10aSAl Viro return 0; 20871da177e4SLinus Torvalds 20881da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20891da177e4SLinus Torvalds for(;;) { 2090d6bb3e90SLinus Torvalds u64 hash_len; 2091fe479a58SAl Viro int type; 20921da177e4SLinus Torvalds 209352094c8aSAl Viro err = may_lookup(nd); 20941da177e4SLinus Torvalds if (err) 20953595e234SAl Viro return err; 20961da177e4SLinus Torvalds 20978387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 20981da177e4SLinus Torvalds 2099fe479a58SAl Viro type = LAST_NORM; 2100d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2101fe479a58SAl Viro case 2: 2102200e9ef7SLinus Torvalds if (name[1] == '.') { 2103fe479a58SAl Viro type = LAST_DOTDOT; 210416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 210516c2cd71SAl Viro } 2106fe479a58SAl Viro break; 2107fe479a58SAl Viro case 1: 2108fe479a58SAl Viro type = LAST_DOT; 2109fe479a58SAl Viro } 21105a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21115a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 211216c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21135a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2114a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2115da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21165a202bcdSAl Viro if (err < 0) 21173595e234SAl Viro return err; 2118d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2119d6bb3e90SLinus Torvalds name = this.name; 21205a202bcdSAl Viro } 21215a202bcdSAl Viro } 2122fe479a58SAl Viro 2123d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2124d6bb3e90SLinus Torvalds nd->last.name = name; 21255f4a6a69SAl Viro nd->last_type = type; 21265f4a6a69SAl Viro 2127d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2128d6bb3e90SLinus Torvalds if (!*name) 2129bdf6cbf1SAl Viro goto OK; 2130200e9ef7SLinus Torvalds /* 2131200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2132200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2133200e9ef7SLinus Torvalds */ 2134200e9ef7SLinus Torvalds do { 2135d6bb3e90SLinus Torvalds name++; 2136d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21378620c238SAl Viro if (unlikely(!*name)) { 21388620c238SAl Viro OK: 2139368ee9baSAl Viro /* pathname body, done */ 21408620c238SAl Viro if (!nd->depth) 21418620c238SAl Viro return 0; 21428620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2143368ee9baSAl Viro /* trailing symlink, done */ 21448620c238SAl Viro if (!name) 21458620c238SAl Viro return 0; 21468620c238SAl Viro /* last component of nested symlink */ 21478f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 21481c4ff1a8SAl Viro } else { 21491c4ff1a8SAl Viro /* not the last component */ 21508f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 21518620c238SAl Viro } 2152ce57dfc1SAl Viro if (err < 0) 21533595e234SAl Viro return err; 2154fe479a58SAl Viro 2155ce57dfc1SAl Viro if (err) { 2156626de996SAl Viro const char *s = get_link(nd); 21575a460275SAl Viro 2158a1c83681SViresh Kumar if (IS_ERR(s)) 21593595e234SAl Viro return PTR_ERR(s); 2160172a39a0SAl Viro err = 0; 216112b09578SAl Viro if (unlikely(!s)) { 216212b09578SAl Viro /* jumped */ 2163b9ff4429SAl Viro put_link(nd); 216412b09578SAl Viro } else { 2165071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 216632cd7468SAl Viro name = s; 21679e18f10aSAl Viro continue; 216848c8b0c5SAl Viro } 216931e6b01fSNick Piggin } 217097242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 217197242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21724675ac39SAl Viro if (unlazy_walk(nd)) 217397242f99SAl Viro return -ECHILD; 217497242f99SAl Viro } 21753595e234SAl Viro return -ENOTDIR; 21765f4a6a69SAl Viro } 2177ce57dfc1SAl Viro } 217897242f99SAl Viro } 21791da177e4SLinus Torvalds 2180edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2181c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 218231e6b01fSNick Piggin { 2183740a1678SAleksa Sarai int error; 2184c8a53ee5SAl Viro const char *s = nd->name->name; 218531e6b01fSNick Piggin 2186c0eb027eSLinus Torvalds if (!*s) 2187c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2188edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2189edc2b1daSAl Viro rcu_read_lock(); 2190c0eb027eSLinus Torvalds 219131e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2192980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 219331e6b01fSNick Piggin nd->depth = 0; 21945b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2195b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2196b18825a7SDavid Howells struct inode *inode = root->d_inode; 219793893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2198368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21995b6ca027SAl Viro nd->path = nd->root; 22005b6ca027SAl Viro nd->inode = inode; 22015b6ca027SAl Viro if (flags & LOOKUP_RCU) { 22025b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 22038f47a016SAl Viro nd->root_seq = nd->seq; 220448a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 22055b6ca027SAl Viro } else { 22065b6ca027SAl Viro path_get(&nd->path); 22075b6ca027SAl Viro } 2208368ee9baSAl Viro return s; 22095b6ca027SAl Viro } 22105b6ca027SAl Viro 221131e6b01fSNick Piggin nd->root.mnt = NULL; 2212248fb5b9SAl Viro nd->path.mnt = NULL; 2213248fb5b9SAl Viro nd->path.dentry = NULL; 221431e6b01fSNick Piggin 221548a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2216740a1678SAleksa Sarai 2217740a1678SAleksa Sarai /* Figure out the starting path and root (if needed). */ 2218fd2f7cb5SAl Viro if (*s == '/') { 2219740a1678SAleksa Sarai error = nd_jump_root(nd); 2220740a1678SAleksa Sarai if (unlikely(error)) 2221740a1678SAleksa Sarai return ERR_PTR(error); 2222ef55d917SAl Viro return s; 2223c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2224e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 222531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2226c28cc364SNick Piggin unsigned seq; 222731e6b01fSNick Piggin 2228c28cc364SNick Piggin do { 2229c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 223031e6b01fSNick Piggin nd->path = fs->pwd; 2231ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2232c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2233c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2234e41f7d4eSAl Viro } else { 2235e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2236ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2237e41f7d4eSAl Viro } 2238ef55d917SAl Viro return s; 223931e6b01fSNick Piggin } else { 2240582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2241c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 224231e6b01fSNick Piggin struct dentry *dentry; 224331e6b01fSNick Piggin 22442903ff01SAl Viro if (!f.file) 2245368ee9baSAl Viro return ERR_PTR(-EBADF); 224631e6b01fSNick Piggin 22472903ff01SAl Viro dentry = f.file->f_path.dentry; 224831e6b01fSNick Piggin 2249edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22502903ff01SAl Viro fdput(f); 2251368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2252f52e0c11SAl Viro } 22532903ff01SAl Viro 22542903ff01SAl Viro nd->path = f.file->f_path; 2255e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 225634a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 225734a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22585590ff0dSUlrich Drepper } else { 22592903ff01SAl Viro path_get(&nd->path); 226034a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22611da177e4SLinus Torvalds } 226234a26b99SAl Viro fdput(f); 2263368ee9baSAl Viro return s; 2264e41f7d4eSAl Viro } 22659b4a9b14SAl Viro } 22669b4a9b14SAl Viro 22673bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 226895fa25d9SAl Viro { 226995fa25d9SAl Viro const char *s; 2270fec2fa24SAl Viro int error = may_follow_link(nd); 2271deb106c6SAl Viro if (unlikely(error)) 22723bdba28bSAl Viro return ERR_PTR(error); 227395fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2274fab51e8aSAl Viro nd->stack[0].name = NULL; 22753b2e7f75SAl Viro s = get_link(nd); 2276deb106c6SAl Viro return s ? s : ""; 227795fa25d9SAl Viro } 227895fa25d9SAl Viro 2279caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2280bd92d7feSAl Viro { 2281bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2282bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2283bd92d7feSAl Viro 2284bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 22851c4ff1a8SAl Viro return walk_component(nd, 0); 2286bd92d7feSAl Viro } 2287bd92d7feSAl Viro 22884f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22894f757f3cSAl Viro { 22904f757f3cSAl Viro struct path path = nd->path; 22914f757f3cSAl Viro struct inode *inode = nd->inode; 22924f757f3cSAl Viro unsigned seq = nd->seq; 22934f757f3cSAl Viro int err; 22944f757f3cSAl Viro 22954f757f3cSAl Viro if (nd->flags & LOOKUP_RCU) { 22964f757f3cSAl Viro /* 22974f757f3cSAl Viro * don't bother with unlazy_walk on failure - we are 22984f757f3cSAl Viro * at the very beginning of walk, so we lose nothing 22994f757f3cSAl Viro * if we simply redo everything in non-RCU mode 23004f757f3cSAl Viro */ 23014f757f3cSAl Viro if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq))) 23024f757f3cSAl Viro return -ECHILD; 23034f757f3cSAl Viro } else { 23044f757f3cSAl Viro dget(path.dentry); 23054f757f3cSAl Viro err = follow_managed(&path, nd); 23064f757f3cSAl Viro if (unlikely(err < 0)) 23074f757f3cSAl Viro return err; 23084f757f3cSAl Viro inode = d_backing_inode(path.dentry); 23094f757f3cSAl Viro seq = 0; 23104f757f3cSAl Viro } 23114f757f3cSAl Viro path_to_nameidata(&path, nd); 23124f757f3cSAl Viro nd->inode = inode; 23134f757f3cSAl Viro nd->seq = seq; 23144f757f3cSAl Viro return 0; 23154f757f3cSAl Viro } 23164f757f3cSAl Viro 23179b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2318c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23199b4a9b14SAl Viro { 2320c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2321bd92d7feSAl Viro int err; 232231e6b01fSNick Piggin 23239b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23244f757f3cSAl Viro err = handle_lookup_down(nd); 23255f336e72SAl Viro if (unlikely(err < 0)) 23265f336e72SAl Viro s = ERR_PTR(err); 23274f757f3cSAl Viro } 23284f757f3cSAl Viro 23293bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 23303bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 23313bdba28bSAl Viro s = trailing_symlink(nd); 2332bd92d7feSAl Viro } 23339f1fafeeSAl Viro if (!err) 23349f1fafeeSAl Viro err = complete_walk(nd); 2335bd92d7feSAl Viro 2336deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2337deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2338bd23a539SAl Viro err = -ENOTDIR; 2339625b6d10SAl Viro if (!err) { 2340625b6d10SAl Viro *path = nd->path; 2341625b6d10SAl Viro nd->path.mnt = NULL; 2342625b6d10SAl Viro nd->path.dentry = NULL; 2343625b6d10SAl Viro } 2344deb106c6SAl Viro terminate_walk(nd); 2345bd92d7feSAl Viro return err; 234631e6b01fSNick Piggin } 234731e6b01fSNick Piggin 234831d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23499ad1aaa6SAl Viro struct path *path, struct path *root) 2350873f1eedSJeff Layton { 2351894bc8c4SAl Viro int retval; 23529883d185SAl Viro struct nameidata nd; 2353abc9f5beSAl Viro if (IS_ERR(name)) 2354abc9f5beSAl Viro return PTR_ERR(name); 23559ad1aaa6SAl Viro if (unlikely(root)) { 23569ad1aaa6SAl Viro nd.root = *root; 23579ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23589ad1aaa6SAl Viro } 23599883d185SAl Viro set_nameidata(&nd, dfd, name); 2360c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2361873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2362c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2363873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2364c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2365873f1eedSJeff Layton 2366873f1eedSJeff Layton if (likely(!retval)) 2367ff0ebee2SAl Viro audit_inode(name, path->dentry, 0); 23689883d185SAl Viro restore_nameidata(); 2369e4bd1c1aSAl Viro putname(name); 2370873f1eedSJeff Layton return retval; 2371873f1eedSJeff Layton } 2372873f1eedSJeff Layton 23738bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2374c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2375391172c4SAl Viro struct path *parent) 23768bcb77faSAl Viro { 2377c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23789b5858e9SAl Viro int err = link_path_walk(s, nd); 23798bcb77faSAl Viro if (!err) 23808bcb77faSAl Viro err = complete_walk(nd); 2381391172c4SAl Viro if (!err) { 2382391172c4SAl Viro *parent = nd->path; 2383391172c4SAl Viro nd->path.mnt = NULL; 2384391172c4SAl Viro nd->path.dentry = NULL; 2385391172c4SAl Viro } 2386deb106c6SAl Viro terminate_walk(nd); 23878bcb77faSAl Viro return err; 23888bcb77faSAl Viro } 23898bcb77faSAl Viro 23905c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2391391172c4SAl Viro unsigned int flags, struct path *parent, 2392391172c4SAl Viro struct qstr *last, int *type) 23938bcb77faSAl Viro { 23948bcb77faSAl Viro int retval; 23959883d185SAl Viro struct nameidata nd; 23968bcb77faSAl Viro 23975c31b6ceSAl Viro if (IS_ERR(name)) 23985c31b6ceSAl Viro return name; 23999883d185SAl Viro set_nameidata(&nd, dfd, name); 2400c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 24018bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2402c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 24038bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2404c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2405391172c4SAl Viro if (likely(!retval)) { 2406391172c4SAl Viro *last = nd.last; 2407391172c4SAl Viro *type = nd.last_type; 2408c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 24095c31b6ceSAl Viro } else { 24105c31b6ceSAl Viro putname(name); 24115c31b6ceSAl Viro name = ERR_PTR(retval); 2412391172c4SAl Viro } 24139883d185SAl Viro restore_nameidata(); 24145c31b6ceSAl Viro return name; 24158bcb77faSAl Viro } 24168bcb77faSAl Viro 241779714f72SAl Viro /* does lookup, returns the object with parent locked */ 241879714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24195590ff0dSUlrich Drepper { 24205c31b6ceSAl Viro struct filename *filename; 24215c31b6ceSAl Viro struct dentry *d; 2422391172c4SAl Viro struct qstr last; 2423391172c4SAl Viro int type; 242451689104SPaul Moore 24255c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24265c31b6ceSAl Viro &last, &type); 242751689104SPaul Moore if (IS_ERR(filename)) 242851689104SPaul Moore return ERR_CAST(filename); 24295c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2430391172c4SAl Viro path_put(path); 24315c31b6ceSAl Viro putname(filename); 24325c31b6ceSAl Viro return ERR_PTR(-EINVAL); 243379714f72SAl Viro } 24345955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2435391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 243679714f72SAl Viro if (IS_ERR(d)) { 24375955102cSAl Viro inode_unlock(path->dentry->d_inode); 2438391172c4SAl Viro path_put(path); 243979714f72SAl Viro } 244051689104SPaul Moore putname(filename); 244179714f72SAl Viro return d; 24425590ff0dSUlrich Drepper } 24435590ff0dSUlrich Drepper 2444d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2445d1811465SAl Viro { 2446abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2447abc9f5beSAl Viro flags, path, NULL); 2448d1811465SAl Viro } 24494d359507SAl Viro EXPORT_SYMBOL(kern_path); 2450d1811465SAl Viro 245116f18200SJosef 'Jeff' Sipek /** 245216f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 245316f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 245416f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 245516f18200SJosef 'Jeff' Sipek * @name: pointer to file name 245616f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2457e0a01249SAl Viro * @path: pointer to struct path to fill 245816f18200SJosef 'Jeff' Sipek */ 245916f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 246016f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2461e0a01249SAl Viro struct path *path) 246216f18200SJosef 'Jeff' Sipek { 24639ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24649ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2465abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2466abc9f5beSAl Viro flags , path, &root); 246716f18200SJosef 'Jeff' Sipek } 24684d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 246916f18200SJosef 'Jeff' Sipek 24703c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24713c95f0dcSAl Viro int len, struct qstr *this) 24723c95f0dcSAl Viro { 24733c95f0dcSAl Viro this->name = name; 24743c95f0dcSAl Viro this->len = len; 24753c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24763c95f0dcSAl Viro if (!len) 24773c95f0dcSAl Viro return -EACCES; 24783c95f0dcSAl Viro 24793c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24803c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24813c95f0dcSAl Viro return -EACCES; 24823c95f0dcSAl Viro } 24833c95f0dcSAl Viro 24843c95f0dcSAl Viro while (len--) { 24853c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24863c95f0dcSAl Viro if (c == '/' || c == '\0') 24873c95f0dcSAl Viro return -EACCES; 24883c95f0dcSAl Viro } 24893c95f0dcSAl Viro /* 24903c95f0dcSAl Viro * See if the low-level filesystem might want 24913c95f0dcSAl Viro * to use its own hash.. 24923c95f0dcSAl Viro */ 24933c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 24943c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 24953c95f0dcSAl Viro if (err < 0) 24963c95f0dcSAl Viro return err; 24973c95f0dcSAl Viro } 24983c95f0dcSAl Viro 24993c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 25003c95f0dcSAl Viro } 25013c95f0dcSAl Viro 2502eead1911SChristoph Hellwig /** 25030da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 25040da0b7fdSDavid Howells * @name: pathname component to lookup 25050da0b7fdSDavid Howells * @base: base directory to lookup from 25060da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 25070da0b7fdSDavid Howells * 25080da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 25090da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 25100da0b7fdSDavid Howells * 25110da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25120da0b7fdSDavid Howells * not be called by generic code. 25130da0b7fdSDavid Howells * 25140da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25150da0b7fdSDavid Howells */ 25160da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25170da0b7fdSDavid Howells { 25180da0b7fdSDavid Howells struct qstr this; 25190da0b7fdSDavid Howells int err; 25200da0b7fdSDavid Howells 25210da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25220da0b7fdSDavid Howells 25230da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25240da0b7fdSDavid Howells if (err) 25250da0b7fdSDavid Howells return ERR_PTR(err); 25260da0b7fdSDavid Howells 25270da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25280da0b7fdSDavid Howells } 25290da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25300da0b7fdSDavid Howells 25310da0b7fdSDavid Howells /** 2532a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2533eead1911SChristoph Hellwig * @name: pathname component to lookup 2534eead1911SChristoph Hellwig * @base: base directory to lookup from 2535eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2536eead1911SChristoph Hellwig * 2537a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25389e7543e9SAl Viro * not be called by generic code. 2539bbddca8eSNeilBrown * 2540bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2541eead1911SChristoph Hellwig */ 2542057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2543057f6c01SJames Morris { 25448613a209SAl Viro struct dentry *dentry; 2545057f6c01SJames Morris struct qstr this; 2546cda309deSMiklos Szeredi int err; 2547057f6c01SJames Morris 25485955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25492f9092e1SDavid Woodhouse 25503c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2551cda309deSMiklos Szeredi if (err) 2552cda309deSMiklos Szeredi return ERR_PTR(err); 2553cda309deSMiklos Szeredi 25548613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25558613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2556057f6c01SJames Morris } 25574d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2558057f6c01SJames Morris 2559bbddca8eSNeilBrown /** 2560bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2561bbddca8eSNeilBrown * @name: pathname component to lookup 2562bbddca8eSNeilBrown * @base: base directory to lookup from 2563bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2564bbddca8eSNeilBrown * 2565bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2566bbddca8eSNeilBrown * not be called by generic code. 2567bbddca8eSNeilBrown * 2568bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2569bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2570bbddca8eSNeilBrown */ 2571bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2572bbddca8eSNeilBrown struct dentry *base, int len) 2573bbddca8eSNeilBrown { 2574bbddca8eSNeilBrown struct qstr this; 2575bbddca8eSNeilBrown int err; 257620d00ee8SLinus Torvalds struct dentry *ret; 2577bbddca8eSNeilBrown 25783c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2579bbddca8eSNeilBrown if (err) 2580bbddca8eSNeilBrown return ERR_PTR(err); 2581bbddca8eSNeilBrown 258220d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 258320d00ee8SLinus Torvalds if (!ret) 258420d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 258520d00ee8SLinus Torvalds return ret; 2586bbddca8eSNeilBrown } 2587bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2588bbddca8eSNeilBrown 25896c2d4798SAl Viro /* 25906c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 25916c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 25926c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 25936c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 25946c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 25956c2d4798SAl Viro * this one avoids such problems. 25966c2d4798SAl Viro */ 25976c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 25986c2d4798SAl Viro struct dentry *base, int len) 25996c2d4798SAl Viro { 26006c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 26012fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 26026c2d4798SAl Viro dput(ret); 26036c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 26046c2d4798SAl Viro } 26056c2d4798SAl Viro return ret; 26066c2d4798SAl Viro } 26076c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 26086c2d4798SAl Viro 2609eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2610eedf265aSEric W. Biederman int path_pts(struct path *path) 2611eedf265aSEric W. Biederman { 2612eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2613eedf265aSEric W. Biederman * the input path. 2614eedf265aSEric W. Biederman */ 2615eedf265aSEric W. Biederman struct dentry *child, *parent; 2616eedf265aSEric W. Biederman struct qstr this; 2617eedf265aSEric W. Biederman int ret; 2618eedf265aSEric W. Biederman 2619eedf265aSEric W. Biederman ret = path_parent_directory(path); 2620eedf265aSEric W. Biederman if (ret) 2621eedf265aSEric W. Biederman return ret; 2622eedf265aSEric W. Biederman 2623eedf265aSEric W. Biederman parent = path->dentry; 2624eedf265aSEric W. Biederman this.name = "pts"; 2625eedf265aSEric W. Biederman this.len = 3; 2626eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2627eedf265aSEric W. Biederman if (!child) 2628eedf265aSEric W. Biederman return -ENOENT; 2629eedf265aSEric W. Biederman 2630eedf265aSEric W. Biederman path->dentry = child; 2631eedf265aSEric W. Biederman dput(parent); 2632eedf265aSEric W. Biederman follow_mount(path); 2633eedf265aSEric W. Biederman return 0; 2634eedf265aSEric W. Biederman } 2635eedf265aSEric W. Biederman #endif 2636eedf265aSEric W. Biederman 26371fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26381fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26391da177e4SLinus Torvalds { 2640abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2641abc9f5beSAl Viro flags, path, NULL); 26421da177e4SLinus Torvalds } 2643b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26441fa1e7f6SAndy Whitcroft 26458033426eSJeff Layton /** 2646197df04cSAl Viro * mountpoint_last - look up last component for umount 26478033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 26488033426eSJeff Layton * 26498033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 26508033426eSJeff Layton * need to resolve the path without doing any revalidation. 26518033426eSJeff Layton * 26528033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 26538033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 26548033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 26558033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 26568033426eSJeff Layton * bogus and it doesn't exist. 26578033426eSJeff Layton * 26588033426eSJeff Layton * Returns: 26598033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2660ba8f4613SAl Viro * lookup found a negative dentry. 26618033426eSJeff Layton * 2662ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2663ba8f4613SAl Viro * symlink that needs to be followed. 26648033426eSJeff Layton * 26658033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2666ba8f4613SAl Viro * that needs to be followed. 26678033426eSJeff Layton */ 26688033426eSJeff Layton static int 2669ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 26708033426eSJeff Layton { 26718033426eSJeff Layton int error = 0; 26728033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2673ba8f4613SAl Viro struct path path; 26748033426eSJeff Layton 267535759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 267635759521SAl Viro if (nd->flags & LOOKUP_RCU) { 26774675ac39SAl Viro if (unlazy_walk(nd)) 2678deb106c6SAl Viro return -ECHILD; 26798033426eSJeff Layton } 26808033426eSJeff Layton 26818033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 26828033426eSJeff Layton 26838033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 26848033426eSJeff Layton error = handle_dots(nd, nd->last_type); 268535759521SAl Viro if (error) 2686deb106c6SAl Viro return error; 2687ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2688949a852eSAl Viro } else { 2689ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2690ba8f4613SAl Viro if (!path.dentry) { 26918033426eSJeff Layton /* 2692949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2693949a852eSAl Viro * cache, so that means that this dentry is probably 2694949a852eSAl Viro * a symlink or the path doesn't actually point 2695949a852eSAl Viro * to a mounted dentry. 26968033426eSJeff Layton */ 2697ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2698949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2699ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2700ba8f4613SAl Viro return PTR_ERR(path.dentry); 27018033426eSJeff Layton } 2702bcceeebaSDave Jones } 27032fa6b1e0SAl Viro if (d_flags_negative(smp_load_acquire(&path.dentry->d_flags))) { 2704ba8f4613SAl Viro dput(path.dentry); 2705deb106c6SAl Viro return -ENOENT; 270635759521SAl Viro } 2707ba8f4613SAl Viro path.mnt = nd->path.mnt; 27088f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 27098033426eSJeff Layton } 27108033426eSJeff Layton 27118033426eSJeff Layton /** 2712197df04cSAl Viro * path_mountpoint - look up a path to be umounted 27132a78b857SMasanari Iida * @nd: lookup context 27148033426eSJeff Layton * @flags: lookup flags 2715c8a53ee5SAl Viro * @path: pointer to container for result 27168033426eSJeff Layton * 27178033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2718606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 27198033426eSJeff Layton */ 27208033426eSJeff Layton static int 2721c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 27228033426eSJeff Layton { 2723c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2724368ee9baSAl Viro int err; 27259b5858e9SAl Viro 27263bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2727ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 27283bdba28bSAl Viro s = trailing_symlink(nd); 27293bdba28bSAl Viro } 2730ba8f4613SAl Viro if (!err) { 2731ba8f4613SAl Viro *path = nd->path; 2732ba8f4613SAl Viro nd->path.mnt = NULL; 2733ba8f4613SAl Viro nd->path.dentry = NULL; 2734ba8f4613SAl Viro follow_mount(path); 2735ba8f4613SAl Viro } 2736deb106c6SAl Viro terminate_walk(nd); 27378033426eSJeff Layton return err; 27388033426eSJeff Layton } 27398033426eSJeff Layton 27402d864651SAl Viro static int 2741668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 27422d864651SAl Viro unsigned int flags) 27432d864651SAl Viro { 27449883d185SAl Viro struct nameidata nd; 2745cbaab2dbSAl Viro int error; 2746668696dcSAl Viro if (IS_ERR(name)) 2747668696dcSAl Viro return PTR_ERR(name); 27489883d185SAl Viro set_nameidata(&nd, dfd, name); 2749c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 27502d864651SAl Viro if (unlikely(error == -ECHILD)) 2751c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 27522d864651SAl Viro if (unlikely(error == -ESTALE)) 2753c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 27542d864651SAl Viro if (likely(!error)) 2755c9b07eabSAl Viro audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL); 27569883d185SAl Viro restore_nameidata(); 2757668696dcSAl Viro putname(name); 27582d864651SAl Viro return error; 27592d864651SAl Viro } 27602d864651SAl Viro 27618033426eSJeff Layton /** 2762197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 27638033426eSJeff Layton * @dfd: directory file descriptor 27648033426eSJeff Layton * @name: pathname from userland 27658033426eSJeff Layton * @flags: lookup flags 27668033426eSJeff Layton * @path: pointer to container to hold result 27678033426eSJeff Layton * 27688033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27698033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27708033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27718033426eSJeff Layton * and avoid revalidating the last component. 27728033426eSJeff Layton * 27738033426eSJeff Layton * Returns 0 and populates "path" on success. 27748033426eSJeff Layton */ 27758033426eSJeff Layton int 2776197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27778033426eSJeff Layton struct path *path) 27788033426eSJeff Layton { 2779cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27808033426eSJeff Layton } 27818033426eSJeff Layton 27822d864651SAl Viro int 27832d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27842d864651SAl Viro unsigned int flags) 27852d864651SAl Viro { 2786cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27872d864651SAl Viro } 27882d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27892d864651SAl Viro 2790cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27911da177e4SLinus Torvalds { 27928e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2793da9592edSDavid Howells 27948e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27951da177e4SLinus Torvalds return 0; 27968e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27971da177e4SLinus Torvalds return 0; 279823adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27991da177e4SLinus Torvalds } 2800cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 28011da177e4SLinus Torvalds 28021da177e4SLinus Torvalds /* 28031da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 28041da177e4SLinus Torvalds * whether the type of victim is right. 28051da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 28061da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 28071da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 28081da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 28091da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 28101da177e4SLinus Torvalds * a. be owner of dir, or 28111da177e4SLinus Torvalds * b. be owner of victim, or 28121da177e4SLinus Torvalds * c. have CAP_FOWNER capability 28131da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 28141da177e4SLinus Torvalds * links pointing to it. 28150bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 28160bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 28170bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 28180bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 28190bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 28201da177e4SLinus Torvalds * nfs_async_unlink(). 28211da177e4SLinus Torvalds */ 2822b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 28231da177e4SLinus Torvalds { 282463afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 28251da177e4SLinus Torvalds int error; 28261da177e4SLinus Torvalds 2827b18825a7SDavid Howells if (d_is_negative(victim)) 28281da177e4SLinus Torvalds return -ENOENT; 2829b18825a7SDavid Howells BUG_ON(!inode); 28301da177e4SLinus Torvalds 28311da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2832593d1ce8SEric W. Biederman 2833593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2834593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2835593d1ce8SEric W. Biederman return -EOVERFLOW; 2836593d1ce8SEric W. Biederman 28374fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 28381da177e4SLinus Torvalds 2839f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 28401da177e4SLinus Torvalds if (error) 28411da177e4SLinus Torvalds return error; 28421da177e4SLinus Torvalds if (IS_APPEND(dir)) 28431da177e4SLinus Torvalds return -EPERM; 2844b18825a7SDavid Howells 2845b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 28460bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 28471da177e4SLinus Torvalds return -EPERM; 28481da177e4SLinus Torvalds if (isdir) { 284944b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28501da177e4SLinus Torvalds return -ENOTDIR; 28511da177e4SLinus Torvalds if (IS_ROOT(victim)) 28521da177e4SLinus Torvalds return -EBUSY; 285344b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28541da177e4SLinus Torvalds return -EISDIR; 28551da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28561da177e4SLinus Torvalds return -ENOENT; 28571da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28581da177e4SLinus Torvalds return -EBUSY; 28591da177e4SLinus Torvalds return 0; 28601da177e4SLinus Torvalds } 28611da177e4SLinus Torvalds 28621da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28631da177e4SLinus Torvalds * dir. 28641da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28651da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28661da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2867036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2868036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2869036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28701da177e4SLinus Torvalds */ 2871a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28721da177e4SLinus Torvalds { 2873036d5236SEric W. Biederman struct user_namespace *s_user_ns; 287414e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28751da177e4SLinus Torvalds if (child->d_inode) 28761da177e4SLinus Torvalds return -EEXIST; 28771da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28781da177e4SLinus Torvalds return -ENOENT; 2879036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2880036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2881036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2882036d5236SEric W. Biederman return -EOVERFLOW; 2883f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28841da177e4SLinus Torvalds } 28851da177e4SLinus Torvalds 28861da177e4SLinus Torvalds /* 28871da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28881da177e4SLinus Torvalds */ 28891da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28901da177e4SLinus Torvalds { 28911da177e4SLinus Torvalds struct dentry *p; 28921da177e4SLinus Torvalds 28931da177e4SLinus Torvalds if (p1 == p2) { 28945955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28951da177e4SLinus Torvalds return NULL; 28961da177e4SLinus Torvalds } 28971da177e4SLinus Torvalds 2898fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28991da177e4SLinus Torvalds 2900e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2901e2761a11SOGAWA Hirofumi if (p) { 29025955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 29035955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 29041da177e4SLinus Torvalds return p; 29051da177e4SLinus Torvalds } 29061da177e4SLinus Torvalds 2907e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2908e2761a11SOGAWA Hirofumi if (p) { 29095955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29105955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 29111da177e4SLinus Torvalds return p; 29121da177e4SLinus Torvalds } 29131da177e4SLinus Torvalds 29145955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29155955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 29161da177e4SLinus Torvalds return NULL; 29171da177e4SLinus Torvalds } 29184d359507SAl Viro EXPORT_SYMBOL(lock_rename); 29191da177e4SLinus Torvalds 29201da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 29211da177e4SLinus Torvalds { 29225955102cSAl Viro inode_unlock(p1->d_inode); 29231da177e4SLinus Torvalds if (p1 != p2) { 29245955102cSAl Viro inode_unlock(p2->d_inode); 2925fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 29261da177e4SLinus Torvalds } 29271da177e4SLinus Torvalds } 29284d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 29291da177e4SLinus Torvalds 29304acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2931312b63fbSAl Viro bool want_excl) 29321da177e4SLinus Torvalds { 2933a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 29341da177e4SLinus Torvalds if (error) 29351da177e4SLinus Torvalds return error; 29361da177e4SLinus Torvalds 2937acfa4380SAl Viro if (!dir->i_op->create) 29381da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 29391da177e4SLinus Torvalds mode &= S_IALLUGO; 29401da177e4SLinus Torvalds mode |= S_IFREG; 29411da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 29421da177e4SLinus Torvalds if (error) 29431da177e4SLinus Torvalds return error; 2944312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2945a74574aaSStephen Smalley if (!error) 2946f38aa942SAmy Griffis fsnotify_create(dir, dentry); 29471da177e4SLinus Torvalds return error; 29481da177e4SLinus Torvalds } 29494d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29501da177e4SLinus Torvalds 29518e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29528e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29538e6c848eSAl Viro void *arg) 29548e6c848eSAl Viro { 29558e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 29568e6c848eSAl Viro int error = may_create(dir, dentry); 29578e6c848eSAl Viro if (error) 29588e6c848eSAl Viro return error; 29598e6c848eSAl Viro 29608e6c848eSAl Viro mode &= S_IALLUGO; 29618e6c848eSAl Viro mode |= S_IFREG; 29628e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29638e6c848eSAl Viro if (error) 29648e6c848eSAl Viro return error; 29658e6c848eSAl Viro error = f(dentry, mode, arg); 29668e6c848eSAl Viro if (!error) 29678e6c848eSAl Viro fsnotify_create(dir, dentry); 29688e6c848eSAl Viro return error; 29698e6c848eSAl Viro } 29708e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29718e6c848eSAl Viro 2972a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2973a2982cc9SEric W. Biederman { 2974a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2975a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2976a2982cc9SEric W. Biederman } 2977a2982cc9SEric W. Biederman 2978f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 29791da177e4SLinus Torvalds { 29803fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29811da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29821da177e4SLinus Torvalds int error; 29831da177e4SLinus Torvalds 29841da177e4SLinus Torvalds if (!inode) 29851da177e4SLinus Torvalds return -ENOENT; 29861da177e4SLinus Torvalds 2987c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2988c8fe8f30SChristoph Hellwig case S_IFLNK: 29891da177e4SLinus Torvalds return -ELOOP; 2990c8fe8f30SChristoph Hellwig case S_IFDIR: 2991c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29921da177e4SLinus Torvalds return -EISDIR; 2993c8fe8f30SChristoph Hellwig break; 2994c8fe8f30SChristoph Hellwig case S_IFBLK: 2995c8fe8f30SChristoph Hellwig case S_IFCHR: 2996a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29971da177e4SLinus Torvalds return -EACCES; 2998c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2999c8fe8f30SChristoph Hellwig case S_IFIFO: 3000c8fe8f30SChristoph Hellwig case S_IFSOCK: 30011da177e4SLinus Torvalds flag &= ~O_TRUNC; 3002c8fe8f30SChristoph Hellwig break; 30034a3fd211SDave Hansen } 3004b41572e9SDave Hansen 300562fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 3006b41572e9SDave Hansen if (error) 3007b41572e9SDave Hansen return error; 30086146f0d5SMimi Zohar 30091da177e4SLinus Torvalds /* 30101da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 30111da177e4SLinus Torvalds */ 30121da177e4SLinus Torvalds if (IS_APPEND(inode)) { 30138737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 30147715b521SAl Viro return -EPERM; 30151da177e4SLinus Torvalds if (flag & O_TRUNC) 30167715b521SAl Viro return -EPERM; 30171da177e4SLinus Torvalds } 30181da177e4SLinus Torvalds 30191da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 30202e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 30217715b521SAl Viro return -EPERM; 30221da177e4SLinus Torvalds 3023f3c7691eSJ. Bruce Fields return 0; 30247715b521SAl Viro } 30257715b521SAl Viro 3026e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 30277715b521SAl Viro { 3028f0bb5aafSAl Viro const struct path *path = &filp->f_path; 30297715b521SAl Viro struct inode *inode = path->dentry->d_inode; 30307715b521SAl Viro int error = get_write_access(inode); 30311da177e4SLinus Torvalds if (error) 30327715b521SAl Viro return error; 30331da177e4SLinus Torvalds /* 30341da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 30351da177e4SLinus Torvalds */ 3036d7a06983SJeff Layton error = locks_verify_locked(filp); 3037be6d3e56SKentaro Takeda if (!error) 3038ea0d3ab2STetsuo Handa error = security_path_truncate(path); 30391da177e4SLinus Torvalds if (!error) { 30407715b521SAl Viro error = do_truncate(path->dentry, 0, 3041d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3042e1181ee6SJeff Layton filp); 30431da177e4SLinus Torvalds } 30441da177e4SLinus Torvalds put_write_access(inode); 3045acd0c935SMimi Zohar return error; 30461da177e4SLinus Torvalds } 30471da177e4SLinus Torvalds 3048d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3049d57999e1SDave Hansen { 30508a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30518a5e929dSAl Viro flag--; 3052d57999e1SDave Hansen return flag; 3053d57999e1SDave Hansen } 3054d57999e1SDave Hansen 3055d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 3056d18e9008SMiklos Szeredi { 30571328c727SSeth Forshee struct user_namespace *s_user_ns; 3058d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3059d18e9008SMiklos Szeredi if (error) 3060d18e9008SMiklos Szeredi return error; 3061d18e9008SMiklos Szeredi 30621328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 30631328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 30641328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 30651328c727SSeth Forshee return -EOVERFLOW; 30661328c727SSeth Forshee 3067d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 3068d18e9008SMiklos Szeredi if (error) 3069d18e9008SMiklos Szeredi return error; 3070d18e9008SMiklos Szeredi 3071d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3072d18e9008SMiklos Szeredi } 3073d18e9008SMiklos Szeredi 30741acf0af9SDavid Howells /* 30751acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30761acf0af9SDavid Howells * dentry. 30771acf0af9SDavid Howells * 30781acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30791acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30801acf0af9SDavid Howells * 308100a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 308200a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 308300a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 30841acf0af9SDavid Howells * 30851acf0af9SDavid Howells * Returns an error code otherwise. 30861acf0af9SDavid Howells */ 30872675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 308830d90494SAl Viro struct path *path, struct file *file, 3089015c3bbcSMiklos Szeredi const struct open_flags *op, 30903ec2eef1SAl Viro int open_flag, umode_t mode) 3091d18e9008SMiklos Szeredi { 3092d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3093d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3094d18e9008SMiklos Szeredi int error; 3095d18e9008SMiklos Szeredi 3096384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3097d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3098d18e9008SMiklos Szeredi 3099d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3100d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3101d18e9008SMiklos Szeredi 310230d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 310330d90494SAl Viro file->f_path.mnt = nd->path.mnt; 31040fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 310544907d79SAl Viro open_to_namei_flags(open_flag), mode); 31066fbd0714SAl Viro d_lookup_done(dentry); 3107384f26e2SAl Viro if (!error) { 310864e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 3109d18e9008SMiklos Szeredi /* 3110384f26e2SAl Viro * We didn't have the inode before the open, so check open 3111384f26e2SAl Viro * permission here. 3112d18e9008SMiklos Szeredi */ 3113384f26e2SAl Viro int acc_mode = op->acc_mode; 311473a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 311503da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 311603da633aSAl Viro fsnotify_create(dir, dentry); 311762fb4a15SAl Viro acc_mode = 0; 311803da633aSAl Viro } 31192675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3120384f26e2SAl Viro if (WARN_ON(error > 0)) 3121384f26e2SAl Viro error = -EINVAL; 312264e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3123d18e9008SMiklos Szeredi error = -EIO; 3124384f26e2SAl Viro } else { 3125d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3126d18e9008SMiklos Szeredi dput(dentry); 3127d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 312810c64ceaSAl Viro } 312973a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 3130d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3131a01e718fSAl Viro if (unlikely(d_is_negative(dentry))) { 3132a01e718fSAl Viro error = -ENOENT; 3133a01e718fSAl Viro } else { 3134d18e9008SMiklos Szeredi path->dentry = dentry; 3135d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 313600a07c15SAl Viro return 0; 3137d18e9008SMiklos Szeredi } 3138d18e9008SMiklos Szeredi } 3139a01e718fSAl Viro } 3140d18e9008SMiklos Szeredi dput(dentry); 3141d18e9008SMiklos Szeredi return error; 3142d18e9008SMiklos Szeredi } 3143d18e9008SMiklos Szeredi 314431e6b01fSNick Piggin /* 31451acf0af9SDavid Howells * Look up and maybe create and open the last component. 3146d58ffd35SMiklos Szeredi * 314700a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3148d58ffd35SMiklos Szeredi * 314900a07c15SAl Viro * Returns 0 on success, that is, if 315000a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 315100a07c15SAl Viro * the file was not completely opened at this time, though lookups and 315200a07c15SAl Viro * creations were performed. 315300a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 315400a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 315500a07c15SAl Viro * hadn't been specified. 31561acf0af9SDavid Howells * 315700a07c15SAl Viro * An error code is returned on failure. 3158d58ffd35SMiklos Szeredi */ 31592675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 316030d90494SAl Viro struct file *file, 3161d58ffd35SMiklos Szeredi const struct open_flags *op, 31623ec2eef1SAl Viro bool got_write) 3163d58ffd35SMiklos Szeredi { 3164d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 316554ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31661643b43fSAl Viro int open_flag = op->open_flag; 3167d58ffd35SMiklos Szeredi struct dentry *dentry; 31681643b43fSAl Viro int error, create_error = 0; 31691643b43fSAl Viro umode_t mode = op->mode; 31706fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3171d58ffd35SMiklos Szeredi 3172ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3173ce8644fcSAl Viro return -ENOENT; 3174d58ffd35SMiklos Szeredi 317573a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 31766fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31776fbd0714SAl Viro for (;;) { 31786fbd0714SAl Viro if (!dentry) { 31796fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3180d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 31812675a4ebSAl Viro return PTR_ERR(dentry); 31826fbd0714SAl Viro } 31836fbd0714SAl Viro if (d_in_lookup(dentry)) 31846fbd0714SAl Viro break; 3185d58ffd35SMiklos Szeredi 31866fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31876fbd0714SAl Viro if (likely(error > 0)) 31886fbd0714SAl Viro break; 31896fbd0714SAl Viro if (error) 31906fbd0714SAl Viro goto out_dput; 31916fbd0714SAl Viro d_invalidate(dentry); 31926fbd0714SAl Viro dput(dentry); 31936fbd0714SAl Viro dentry = NULL; 31946fbd0714SAl Viro } 31956fbd0714SAl Viro if (dentry->d_inode) { 3196d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3197d18e9008SMiklos Szeredi goto out_no_open; 31986c51e513SAl Viro } 3199d18e9008SMiklos Szeredi 32001643b43fSAl Viro /* 32011643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 32021643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 32031643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 32041643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 32051643b43fSAl Viro * 32061643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 32071643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 32081643b43fSAl Viro */ 32091643b43fSAl Viro if (open_flag & O_CREAT) { 32101643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 32111643b43fSAl Viro mode &= ~current_umask(); 32121643b43fSAl Viro if (unlikely(!got_write)) { 32131643b43fSAl Viro create_error = -EROFS; 32141643b43fSAl Viro open_flag &= ~O_CREAT; 32151643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 32161643b43fSAl Viro goto no_open; 32171643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 32181643b43fSAl Viro } else { 32191643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 32201643b43fSAl Viro if (create_error) { 32211643b43fSAl Viro open_flag &= ~O_CREAT; 32221643b43fSAl Viro if (open_flag & O_EXCL) 32231643b43fSAl Viro goto no_open; 32241643b43fSAl Viro } 32251643b43fSAl Viro } 32261643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 32271643b43fSAl Viro unlikely(!got_write)) { 32281643b43fSAl Viro /* 32291643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 32301643b43fSAl Viro * back to lookup + open 32311643b43fSAl Viro */ 32321643b43fSAl Viro goto no_open; 3233d18e9008SMiklos Szeredi } 3234d18e9008SMiklos Szeredi 32351643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 32361643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 32373ec2eef1SAl Viro mode); 32381643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 32391643b43fSAl Viro error = create_error; 32401643b43fSAl Viro return error; 32411643b43fSAl Viro } 324254ef4872SMiklos Szeredi 32431643b43fSAl Viro no_open: 32446fbd0714SAl Viro if (d_in_lookup(dentry)) { 324512fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 324612fa5e24SAl Viro nd->flags); 32476fbd0714SAl Viro d_lookup_done(dentry); 324812fa5e24SAl Viro if (unlikely(res)) { 324912fa5e24SAl Viro if (IS_ERR(res)) { 325012fa5e24SAl Viro error = PTR_ERR(res); 325112fa5e24SAl Viro goto out_dput; 325212fa5e24SAl Viro } 325312fa5e24SAl Viro dput(dentry); 325412fa5e24SAl Viro dentry = res; 325512fa5e24SAl Viro } 325654ef4872SMiklos Szeredi } 325754ef4872SMiklos Szeredi 3258d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32591643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 326073a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3261ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3262ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3263ce8644fcSAl Viro error = -EACCES; 3264d58ffd35SMiklos Szeredi goto out_dput; 326564894cf8SAl Viro } 3266ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 32671643b43fSAl Viro open_flag & O_EXCL); 3268d58ffd35SMiklos Szeredi if (error) 3269d58ffd35SMiklos Szeredi goto out_dput; 3270ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3271d58ffd35SMiklos Szeredi } 32721643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32731643b43fSAl Viro error = create_error; 3274d58ffd35SMiklos Szeredi goto out_dput; 3275d58ffd35SMiklos Szeredi } 3276d18e9008SMiklos Szeredi out_no_open: 3277d58ffd35SMiklos Szeredi path->dentry = dentry; 3278d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 327900a07c15SAl Viro return 0; 3280d58ffd35SMiklos Szeredi 3281d58ffd35SMiklos Szeredi out_dput: 3282d58ffd35SMiklos Szeredi dput(dentry); 32832675a4ebSAl Viro return error; 3284d58ffd35SMiklos Szeredi } 3285d58ffd35SMiklos Szeredi 3286d58ffd35SMiklos Szeredi /* 3287fe2d35ffSAl Viro * Handle the last step of open() 328831e6b01fSNick Piggin */ 3289896475d5SAl Viro static int do_last(struct nameidata *nd, 32903ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3291fb1cc555SAl Viro { 3292a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3293ca344a89SAl Viro int open_flag = op->open_flag; 329477d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 329564894cf8SAl Viro bool got_write = false; 3296bcda7652SAl Viro int acc_mode = op->acc_mode; 3297254cf582SAl Viro unsigned seq; 3298a1eb3315SMiklos Szeredi struct inode *inode; 3299896475d5SAl Viro struct path path; 330016c2cd71SAl Viro int error; 3301fb1cc555SAl Viro 3302c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3303c3e380b0SAl Viro nd->flags |= op->intent; 3304c3e380b0SAl Viro 3305bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3306fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3307deb106c6SAl Viro if (unlikely(error)) 33082675a4ebSAl Viro return error; 3309e83db167SMiklos Szeredi goto finish_open; 33101f36f774SAl Viro } 3311a2c36b45SAl Viro 3312ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3313fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3314fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3315fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3316254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3317e9742b53SAl Viro if (likely(error > 0)) 331871574865SMiklos Szeredi goto finish_lookup; 331971574865SMiklos Szeredi 3320ce57dfc1SAl Viro if (error < 0) 3321deb106c6SAl Viro return error; 3322a1eb3315SMiklos Szeredi 332337d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 33246583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3325b6183df7SMiklos Szeredi } else { 3326fe2d35ffSAl Viro /* create side of things */ 3327a3fbbde7SAl Viro /* 3328b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3329b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3330b6183df7SMiklos Szeredi * about to look up 3331a3fbbde7SAl Viro */ 33329f1fafeeSAl Viro error = complete_walk(nd); 3333e8bb73dfSAl Viro if (error) 33342675a4ebSAl Viro return error; 3335fe2d35ffSAl Viro 3336c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 33371f36f774SAl Viro /* trailing slashes? */ 3338deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3339deb106c6SAl Viro return -EISDIR; 3340b6183df7SMiklos Szeredi } 33411f36f774SAl Viro 33429cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 334364894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 334464894cf8SAl Viro if (!error) 334564894cf8SAl Viro got_write = true; 334664894cf8SAl Viro /* 334764894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 334864894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 334964894cf8SAl Viro * dropping this one anyway. 335064894cf8SAl Viro */ 335164894cf8SAl Viro } 33529cf843e3SAl Viro if (open_flag & O_CREAT) 33535955102cSAl Viro inode_lock(dir->d_inode); 33549cf843e3SAl Viro else 33559cf843e3SAl Viro inode_lock_shared(dir->d_inode); 33563ec2eef1SAl Viro error = lookup_open(nd, &path, file, op, got_write); 33579cf843e3SAl Viro if (open_flag & O_CREAT) 33585955102cSAl Viro inode_unlock(dir->d_inode); 33599cf843e3SAl Viro else 33609cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3361fb1cc555SAl Viro 33622675a4ebSAl Viro if (error) 3363d58ffd35SMiklos Szeredi goto out; 33646c0d46c4SAl Viro 336500a07c15SAl Viro if (file->f_mode & FMODE_OPENED) { 336673a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3367496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 336877d660a8SMiklos Szeredi will_truncate = false; 3369d18e9008SMiklos Szeredi 337076ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3371d18e9008SMiklos Szeredi goto opened; 3372d18e9008SMiklos Szeredi } 3373d18e9008SMiklos Szeredi 337473a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 33759b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3376ca344a89SAl Viro open_flag &= ~O_TRUNC; 337777d660a8SMiklos Szeredi will_truncate = false; 337862fb4a15SAl Viro acc_mode = 0; 3379896475d5SAl Viro path_to_nameidata(&path, nd); 3380e83db167SMiklos Szeredi goto finish_open_created; 3381fb1cc555SAl Viro } 3382fb1cc555SAl Viro 3383fb1cc555SAl Viro /* 3384d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3385d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3386d18e9008SMiklos Szeredi * necessary...) 3387d18e9008SMiklos Szeredi */ 338864894cf8SAl Viro if (got_write) { 3389d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 339064894cf8SAl Viro got_write = false; 3391d18e9008SMiklos Szeredi } 3392d18e9008SMiklos Szeredi 3393e6ec03a2SAl Viro error = follow_managed(&path, nd); 3394e6ec03a2SAl Viro if (unlikely(error < 0)) 3395e6ec03a2SAl Viro return error; 3396e6ec03a2SAl Viro 33976583fe22SAl Viro /* 33986583fe22SAl Viro * create/update audit record if it already exists. 33996583fe22SAl Viro */ 34006583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 34016583fe22SAl Viro 3402deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3403deb106c6SAl Viro path_to_nameidata(&path, nd); 3404deb106c6SAl Viro return -EEXIST; 3405deb106c6SAl Viro } 3406fb1cc555SAl Viro 3407254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3408d4565649SAl Viro inode = d_backing_inode(path.dentry); 3409766c4cbfSAl Viro finish_lookup: 34108f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3411deb106c6SAl Viro if (unlikely(error)) 3412d63ff28fSAl Viro return error; 3413bc77daa7SAl Viro finish_open: 34148f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3415a3fbbde7SAl Viro error = complete_walk(nd); 3416fac7d191SAl Viro if (error) 34172675a4ebSAl Viro return error; 341876ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 341930aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3420fb1cc555SAl Viro error = -EISDIR; 342130aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 34222675a4ebSAl Viro goto out; 342330aba665SSalvatore Mesoraca error = may_create_in_sticky(dir, 342430aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 342530aba665SSalvatore Mesoraca if (unlikely(error)) 342630aba665SSalvatore Mesoraca goto out; 342730aba665SSalvatore Mesoraca } 3428af2f5542SMiklos Szeredi error = -ENOTDIR; 342944b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 34302675a4ebSAl Viro goto out; 34314bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 343277d660a8SMiklos Szeredi will_truncate = false; 34336c0d46c4SAl Viro 34340f9d1a10SAl Viro if (will_truncate) { 34350f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 34360f9d1a10SAl Viro if (error) 34372675a4ebSAl Viro goto out; 343864894cf8SAl Viro got_write = true; 34390f9d1a10SAl Viro } 3440e83db167SMiklos Szeredi finish_open_created: 3441bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3442ca344a89SAl Viro if (error) 34432675a4ebSAl Viro goto out; 3444aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3445ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3446fac7d191SAl Viro if (error) 3447015c3bbcSMiklos Szeredi goto out; 3448a8277b9bSMiklos Szeredi opened: 34496035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3450fe9ec829SAl Viro if (!error && will_truncate) 34512675a4ebSAl Viro error = handle_truncate(file); 3452ca344a89SAl Viro out: 3453c80567c8SAl Viro if (unlikely(error > 0)) { 3454c80567c8SAl Viro WARN_ON(1); 3455c80567c8SAl Viro error = -EINVAL; 3456c80567c8SAl Viro } 345764894cf8SAl Viro if (got_write) 34580f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 34592675a4ebSAl Viro return error; 3460fb1cc555SAl Viro } 3461fb1cc555SAl Viro 3462af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3463af7bd4dcSAmir Goldstein { 3464af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3465af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3466af7bd4dcSAmir Goldstein struct inode *inode; 3467af7bd4dcSAmir Goldstein int error; 3468af7bd4dcSAmir Goldstein 3469af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3470af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3471af7bd4dcSAmir Goldstein if (error) 3472af7bd4dcSAmir Goldstein goto out_err; 3473af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3474af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3475af7bd4dcSAmir Goldstein goto out_err; 3476af7bd4dcSAmir Goldstein error = -ENOMEM; 3477cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3478af7bd4dcSAmir Goldstein if (unlikely(!child)) 3479af7bd4dcSAmir Goldstein goto out_err; 3480af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3481af7bd4dcSAmir Goldstein if (error) 3482af7bd4dcSAmir Goldstein goto out_err; 3483af7bd4dcSAmir Goldstein error = -ENOENT; 3484af7bd4dcSAmir Goldstein inode = child->d_inode; 3485af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3486af7bd4dcSAmir Goldstein goto out_err; 3487af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3488af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3489af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3490af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3491af7bd4dcSAmir Goldstein } 3492fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3493af7bd4dcSAmir Goldstein return child; 3494af7bd4dcSAmir Goldstein 3495af7bd4dcSAmir Goldstein out_err: 3496af7bd4dcSAmir Goldstein dput(child); 3497af7bd4dcSAmir Goldstein return ERR_PTR(error); 3498af7bd4dcSAmir Goldstein } 3499af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3500af7bd4dcSAmir Goldstein 3501c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 350260545d0dSAl Viro const struct open_flags *op, 35033ec2eef1SAl Viro struct file *file) 350460545d0dSAl Viro { 3505625b6d10SAl Viro struct dentry *child; 3506625b6d10SAl Viro struct path path; 3507c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 350860545d0dSAl Viro if (unlikely(error)) 350960545d0dSAl Viro return error; 3510625b6d10SAl Viro error = mnt_want_write(path.mnt); 351160545d0dSAl Viro if (unlikely(error)) 351260545d0dSAl Viro goto out; 3513af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3514af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3515684e73beSHirofumi Nakagawa if (IS_ERR(child)) 351660545d0dSAl Viro goto out2; 3517625b6d10SAl Viro dput(path.dentry); 3518625b6d10SAl Viro path.dentry = child; 3519c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 352069a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 352162fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 352260545d0dSAl Viro if (error) 352360545d0dSAl Viro goto out2; 3524625b6d10SAl Viro file->f_path.mnt = path.mnt; 3525be12af3eSAl Viro error = finish_open(file, child, NULL); 352660545d0dSAl Viro out2: 3527625b6d10SAl Viro mnt_drop_write(path.mnt); 352860545d0dSAl Viro out: 3529625b6d10SAl Viro path_put(&path); 353060545d0dSAl Viro return error; 353160545d0dSAl Viro } 353260545d0dSAl Viro 35336ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 35346ac08709SAl Viro { 35356ac08709SAl Viro struct path path; 35366ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 35376ac08709SAl Viro if (!error) { 35386ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3539ae2bb293SAl Viro error = vfs_open(&path, file); 35406ac08709SAl Viro path_put(&path); 35416ac08709SAl Viro } 35426ac08709SAl Viro return error; 35436ac08709SAl Viro } 35446ac08709SAl Viro 3545c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3546c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35471da177e4SLinus Torvalds { 354830d90494SAl Viro struct file *file; 354913aab428SAl Viro int error; 355031e6b01fSNick Piggin 3551ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35521afc99beSAl Viro if (IS_ERR(file)) 35531afc99beSAl Viro return file; 355431e6b01fSNick Piggin 3555bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 35563ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 35575f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 35586ac08709SAl Viro error = do_o_path(nd, flags, file); 35595f336e72SAl Viro } else { 35605f336e72SAl Viro const char *s = path_init(nd, flags); 35613bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 35623ec2eef1SAl Viro (error = do_last(nd, file, op)) > 0) { 356373d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35643bdba28bSAl Viro s = trailing_symlink(nd); 35653bdba28bSAl Viro } 3566deb106c6SAl Viro terminate_walk(nd); 35675f336e72SAl Viro } 35687c1c01ecSAl Viro if (likely(!error)) { 3569aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35707c1c01ecSAl Viro return file; 35717c1c01ecSAl Viro WARN_ON(1); 35727c1c01ecSAl Viro error = -EINVAL; 3573015c3bbcSMiklos Szeredi } 35747c1c01ecSAl Viro fput(file); 35752675a4ebSAl Viro if (error == -EOPENSTALE) { 35762675a4ebSAl Viro if (flags & LOOKUP_RCU) 35772675a4ebSAl Viro error = -ECHILD; 35782675a4ebSAl Viro else 35792675a4ebSAl Viro error = -ESTALE; 35802675a4ebSAl Viro } 35817c1c01ecSAl Viro return ERR_PTR(error); 3582de459215SKirill Korotaev } 35831da177e4SLinus Torvalds 3584669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3585f9652e10SAl Viro const struct open_flags *op) 358613aab428SAl Viro { 35879883d185SAl Viro struct nameidata nd; 3588f9652e10SAl Viro int flags = op->lookup_flags; 358913aab428SAl Viro struct file *filp; 359013aab428SAl Viro 35919883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3592c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 359313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3594c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 359513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3596c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35979883d185SAl Viro restore_nameidata(); 359813aab428SAl Viro return filp; 359913aab428SAl Viro } 360013aab428SAl Viro 360173d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3602f9652e10SAl Viro const char *name, const struct open_flags *op) 360373d049a4SAl Viro { 36049883d185SAl Viro struct nameidata nd; 360573d049a4SAl Viro struct file *file; 360651689104SPaul Moore struct filename *filename; 3607f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 360873d049a4SAl Viro 360973d049a4SAl Viro nd.root.mnt = mnt; 361073d049a4SAl Viro nd.root.dentry = dentry; 361173d049a4SAl Viro 3612b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 361373d049a4SAl Viro return ERR_PTR(-ELOOP); 361473d049a4SAl Viro 361551689104SPaul Moore filename = getname_kernel(name); 3616a1c83681SViresh Kumar if (IS_ERR(filename)) 361751689104SPaul Moore return ERR_CAST(filename); 361851689104SPaul Moore 36199883d185SAl Viro set_nameidata(&nd, -1, filename); 3620c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 362173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3622c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 362373d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3624c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 36259883d185SAl Viro restore_nameidata(); 362651689104SPaul Moore putname(filename); 362773d049a4SAl Viro return file; 362873d049a4SAl Viro } 362973d049a4SAl Viro 3630fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 36311ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36321da177e4SLinus Torvalds { 3633c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3634391172c4SAl Viro struct qstr last; 3635391172c4SAl Viro int type; 3636c30dabfeSJan Kara int err2; 36371ac12b4bSJeff Layton int error; 36381ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36391ac12b4bSJeff Layton 36401ac12b4bSJeff Layton /* 36411ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36421ac12b4bSJeff Layton * other flags passed in are ignored! 36431ac12b4bSJeff Layton */ 36441ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36451ac12b4bSJeff Layton 36465c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36475c31b6ceSAl Viro if (IS_ERR(name)) 36485c31b6ceSAl Viro return ERR_CAST(name); 36491da177e4SLinus Torvalds 3650c663e5d8SChristoph Hellwig /* 3651c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3652c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3653c663e5d8SChristoph Hellwig */ 36545c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3655ed75e95dSAl Viro goto out; 3656c663e5d8SChristoph Hellwig 3657c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3658391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3659c663e5d8SChristoph Hellwig /* 3660c663e5d8SChristoph Hellwig * Do the final lookup. 3661c663e5d8SChristoph Hellwig */ 3662391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36635955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3664391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36651da177e4SLinus Torvalds if (IS_ERR(dentry)) 3666a8104a9fSAl Viro goto unlock; 3667c663e5d8SChristoph Hellwig 3668a8104a9fSAl Viro error = -EEXIST; 3669b18825a7SDavid Howells if (d_is_positive(dentry)) 3670a8104a9fSAl Viro goto fail; 3671b18825a7SDavid Howells 3672c663e5d8SChristoph Hellwig /* 3673c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3674c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3675c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3676c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3677c663e5d8SChristoph Hellwig */ 3678391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3679a8104a9fSAl Viro error = -ENOENT; 3680ed75e95dSAl Viro goto fail; 3681e9baf6e5SAl Viro } 3682c30dabfeSJan Kara if (unlikely(err2)) { 3683c30dabfeSJan Kara error = err2; 3684a8104a9fSAl Viro goto fail; 3685c30dabfeSJan Kara } 3686181c37b6SAl Viro putname(name); 3687e9baf6e5SAl Viro return dentry; 36881da177e4SLinus Torvalds fail: 3689a8104a9fSAl Viro dput(dentry); 3690a8104a9fSAl Viro dentry = ERR_PTR(error); 3691a8104a9fSAl Viro unlock: 36925955102cSAl Viro inode_unlock(path->dentry->d_inode); 3693c30dabfeSJan Kara if (!err2) 3694391172c4SAl Viro mnt_drop_write(path->mnt); 3695ed75e95dSAl Viro out: 3696391172c4SAl Viro path_put(path); 3697181c37b6SAl Viro putname(name); 3698ed75e95dSAl Viro return dentry; 3699dae6ad8fSAl Viro } 3700fa14a0b8SAl Viro 3701fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3702fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3703fa14a0b8SAl Viro { 3704181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3705181c37b6SAl Viro path, lookup_flags); 3706fa14a0b8SAl Viro } 3707dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3708dae6ad8fSAl Viro 3709921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3710921a1650SAl Viro { 3711921a1650SAl Viro dput(dentry); 37125955102cSAl Viro inode_unlock(path->dentry->d_inode); 3713a8104a9fSAl Viro mnt_drop_write(path->mnt); 3714921a1650SAl Viro path_put(path); 3715921a1650SAl Viro } 3716921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3717921a1650SAl Viro 3718520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 37191ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3720dae6ad8fSAl Viro { 3721181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3722dae6ad8fSAl Viro } 3723dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3724dae6ad8fSAl Viro 37251a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 37261da177e4SLinus Torvalds { 3727a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37281da177e4SLinus Torvalds 37291da177e4SLinus Torvalds if (error) 37301da177e4SLinus Torvalds return error; 37311da177e4SLinus Torvalds 373294f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 37331da177e4SLinus Torvalds return -EPERM; 37341da177e4SLinus Torvalds 3735acfa4380SAl Viro if (!dir->i_op->mknod) 37361da177e4SLinus Torvalds return -EPERM; 37371da177e4SLinus Torvalds 373808ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 373908ce5f16SSerge E. Hallyn if (error) 374008ce5f16SSerge E. Hallyn return error; 374108ce5f16SSerge E. Hallyn 37421da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37431da177e4SLinus Torvalds if (error) 37441da177e4SLinus Torvalds return error; 37451da177e4SLinus Torvalds 37461da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3747a74574aaSStephen Smalley if (!error) 3748f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37491da177e4SLinus Torvalds return error; 37501da177e4SLinus Torvalds } 37514d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37521da177e4SLinus Torvalds 3753f69aac00SAl Viro static int may_mknod(umode_t mode) 3754463c3197SDave Hansen { 3755463c3197SDave Hansen switch (mode & S_IFMT) { 3756463c3197SDave Hansen case S_IFREG: 3757463c3197SDave Hansen case S_IFCHR: 3758463c3197SDave Hansen case S_IFBLK: 3759463c3197SDave Hansen case S_IFIFO: 3760463c3197SDave Hansen case S_IFSOCK: 3761463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3762463c3197SDave Hansen return 0; 3763463c3197SDave Hansen case S_IFDIR: 3764463c3197SDave Hansen return -EPERM; 3765463c3197SDave Hansen default: 3766463c3197SDave Hansen return -EINVAL; 3767463c3197SDave Hansen } 3768463c3197SDave Hansen } 3769463c3197SDave Hansen 377087c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 377187c4e192SDominik Brodowski unsigned int dev) 37721da177e4SLinus Torvalds { 37731da177e4SLinus Torvalds struct dentry *dentry; 3774dae6ad8fSAl Viro struct path path; 3775dae6ad8fSAl Viro int error; 3776972567f1SJeff Layton unsigned int lookup_flags = 0; 37771da177e4SLinus Torvalds 37788e4bfca1SAl Viro error = may_mknod(mode); 37798e4bfca1SAl Viro if (error) 37808e4bfca1SAl Viro return error; 3781972567f1SJeff Layton retry: 3782972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3783dae6ad8fSAl Viro if (IS_ERR(dentry)) 3784dae6ad8fSAl Viro return PTR_ERR(dentry); 37852ad94ae6SAl Viro 3786dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3787ce3b0f8dSAl Viro mode &= ~current_umask(); 3788dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3789be6d3e56SKentaro Takeda if (error) 3790a8104a9fSAl Viro goto out; 37911da177e4SLinus Torvalds switch (mode & S_IFMT) { 37921da177e4SLinus Torvalds case 0: case S_IFREG: 3793312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 379405d1a717SMimi Zohar if (!error) 379505d1a717SMimi Zohar ima_post_path_mknod(dentry); 37961da177e4SLinus Torvalds break; 37971da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3798dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37991da177e4SLinus Torvalds new_decode_dev(dev)); 38001da177e4SLinus Torvalds break; 38011da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3802dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 38031da177e4SLinus Torvalds break; 38041da177e4SLinus Torvalds } 3805a8104a9fSAl Viro out: 3806921a1650SAl Viro done_path_create(&path, dentry); 3807972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3808972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3809972567f1SJeff Layton goto retry; 3810972567f1SJeff Layton } 38111da177e4SLinus Torvalds return error; 38121da177e4SLinus Torvalds } 38131da177e4SLinus Torvalds 381487c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 381587c4e192SDominik Brodowski unsigned int, dev) 381687c4e192SDominik Brodowski { 381787c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 381887c4e192SDominik Brodowski } 381987c4e192SDominik Brodowski 38208208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 38215590ff0dSUlrich Drepper { 382287c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 38235590ff0dSUlrich Drepper } 38245590ff0dSUlrich Drepper 382518bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 38261da177e4SLinus Torvalds { 3827a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 38288de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38291da177e4SLinus Torvalds 38301da177e4SLinus Torvalds if (error) 38311da177e4SLinus Torvalds return error; 38321da177e4SLinus Torvalds 3833acfa4380SAl Viro if (!dir->i_op->mkdir) 38341da177e4SLinus Torvalds return -EPERM; 38351da177e4SLinus Torvalds 38361da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38371da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38381da177e4SLinus Torvalds if (error) 38391da177e4SLinus Torvalds return error; 38401da177e4SLinus Torvalds 38418de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38428de52778SAl Viro return -EMLINK; 38438de52778SAl Viro 38441da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3845a74574aaSStephen Smalley if (!error) 3846f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38471da177e4SLinus Torvalds return error; 38481da177e4SLinus Torvalds } 38494d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38501da177e4SLinus Torvalds 38510101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38521da177e4SLinus Torvalds { 38536902d925SDave Hansen struct dentry *dentry; 3854dae6ad8fSAl Viro struct path path; 3855dae6ad8fSAl Viro int error; 3856b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38571da177e4SLinus Torvalds 3858b76d8b82SJeff Layton retry: 3859b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38606902d925SDave Hansen if (IS_ERR(dentry)) 3861dae6ad8fSAl Viro return PTR_ERR(dentry); 38626902d925SDave Hansen 3863dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3864ce3b0f8dSAl Viro mode &= ~current_umask(); 3865dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3866a8104a9fSAl Viro if (!error) 3867dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3868921a1650SAl Viro done_path_create(&path, dentry); 3869b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3870b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3871b76d8b82SJeff Layton goto retry; 3872b76d8b82SJeff Layton } 38731da177e4SLinus Torvalds return error; 38741da177e4SLinus Torvalds } 38751da177e4SLinus Torvalds 38760101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38770101db7aSDominik Brodowski { 38780101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38790101db7aSDominik Brodowski } 38800101db7aSDominik Brodowski 3881a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38825590ff0dSUlrich Drepper { 38830101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38845590ff0dSUlrich Drepper } 38855590ff0dSUlrich Drepper 38861da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38871da177e4SLinus Torvalds { 38881da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38891da177e4SLinus Torvalds 38901da177e4SLinus Torvalds if (error) 38911da177e4SLinus Torvalds return error; 38921da177e4SLinus Torvalds 3893acfa4380SAl Viro if (!dir->i_op->rmdir) 38941da177e4SLinus Torvalds return -EPERM; 38951da177e4SLinus Torvalds 38961d2ef590SAl Viro dget(dentry); 38975955102cSAl Viro inode_lock(dentry->d_inode); 3898912dbc15SSage Weil 38991da177e4SLinus Torvalds error = -EBUSY; 39007af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3901912dbc15SSage Weil goto out; 3902912dbc15SSage Weil 39031da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3904912dbc15SSage Weil if (error) 3905912dbc15SSage Weil goto out; 3906912dbc15SSage Weil 39071da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3908912dbc15SSage Weil if (error) 3909912dbc15SSage Weil goto out; 3910912dbc15SSage Weil 39118767712fSAl Viro shrink_dcache_parent(dentry); 39121da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3913d83c49f3SAl Viro dont_mount(dentry); 39148ed936b5SEric W. Biederman detach_mounts(dentry); 3915116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 39161da177e4SLinus Torvalds 3917912dbc15SSage Weil out: 39185955102cSAl Viro inode_unlock(dentry->d_inode); 39191d2ef590SAl Viro dput(dentry); 3920912dbc15SSage Weil if (!error) 3921912dbc15SSage Weil d_delete(dentry); 39221da177e4SLinus Torvalds return error; 39231da177e4SLinus Torvalds } 39244d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 39251da177e4SLinus Torvalds 3926f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 39271da177e4SLinus Torvalds { 39281da177e4SLinus Torvalds int error = 0; 392991a27b2aSJeff Layton struct filename *name; 39301da177e4SLinus Torvalds struct dentry *dentry; 3931f5beed75SAl Viro struct path path; 3932f5beed75SAl Viro struct qstr last; 3933f5beed75SAl Viro int type; 3934c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3935c6ee9206SJeff Layton retry: 3936c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3937c1d4dd27SAl Viro &path, &last, &type); 393891a27b2aSJeff Layton if (IS_ERR(name)) 393991a27b2aSJeff Layton return PTR_ERR(name); 39401da177e4SLinus Torvalds 3941f5beed75SAl Viro switch (type) { 39421da177e4SLinus Torvalds case LAST_DOTDOT: 39431da177e4SLinus Torvalds error = -ENOTEMPTY; 39441da177e4SLinus Torvalds goto exit1; 39451da177e4SLinus Torvalds case LAST_DOT: 39461da177e4SLinus Torvalds error = -EINVAL; 39471da177e4SLinus Torvalds goto exit1; 39481da177e4SLinus Torvalds case LAST_ROOT: 39491da177e4SLinus Torvalds error = -EBUSY; 39501da177e4SLinus Torvalds goto exit1; 39511da177e4SLinus Torvalds } 39520612d9fbSOGAWA Hirofumi 3953f5beed75SAl Viro error = mnt_want_write(path.mnt); 3954c30dabfeSJan Kara if (error) 3955c30dabfeSJan Kara goto exit1; 39560612d9fbSOGAWA Hirofumi 39575955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3958f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39591da177e4SLinus Torvalds error = PTR_ERR(dentry); 39606902d925SDave Hansen if (IS_ERR(dentry)) 39616902d925SDave Hansen goto exit2; 3962e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3963e6bc45d6STheodore Ts'o error = -ENOENT; 3964e6bc45d6STheodore Ts'o goto exit3; 3965e6bc45d6STheodore Ts'o } 3966f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3967be6d3e56SKentaro Takeda if (error) 3968c30dabfeSJan Kara goto exit3; 3969f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39700622753bSDave Hansen exit3: 39711da177e4SLinus Torvalds dput(dentry); 39726902d925SDave Hansen exit2: 39735955102cSAl Viro inode_unlock(path.dentry->d_inode); 3974f5beed75SAl Viro mnt_drop_write(path.mnt); 39751da177e4SLinus Torvalds exit1: 3976f5beed75SAl Viro path_put(&path); 39771da177e4SLinus Torvalds putname(name); 3978c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3979c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3980c6ee9206SJeff Layton goto retry; 3981c6ee9206SJeff Layton } 39821da177e4SLinus Torvalds return error; 39831da177e4SLinus Torvalds } 39841da177e4SLinus Torvalds 39853cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39865590ff0dSUlrich Drepper { 39875590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39885590ff0dSUlrich Drepper } 39895590ff0dSUlrich Drepper 3990b21996e3SJ. Bruce Fields /** 3991b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3992b21996e3SJ. Bruce Fields * @dir: parent directory 3993b21996e3SJ. Bruce Fields * @dentry: victim 3994b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3995b21996e3SJ. Bruce Fields * 3996b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3997b21996e3SJ. Bruce Fields * 3998b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3999b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4000b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4001b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4002b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4003b21996e3SJ. Bruce Fields * 4004b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4005b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4006b21996e3SJ. Bruce Fields * to be NFS exported. 4007b21996e3SJ. Bruce Fields */ 4008b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 40091da177e4SLinus Torvalds { 40109accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 40111da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 40121da177e4SLinus Torvalds 40131da177e4SLinus Torvalds if (error) 40141da177e4SLinus Torvalds return error; 40151da177e4SLinus Torvalds 4016acfa4380SAl Viro if (!dir->i_op->unlink) 40171da177e4SLinus Torvalds return -EPERM; 40181da177e4SLinus Torvalds 40195955102cSAl Viro inode_lock(target); 40208ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 40211da177e4SLinus Torvalds error = -EBUSY; 40221da177e4SLinus Torvalds else { 40231da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4024bec1052eSAl Viro if (!error) { 40255a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 40265a14696cSJ. Bruce Fields if (error) 4027b21996e3SJ. Bruce Fields goto out; 40281da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 40298ed936b5SEric W. Biederman if (!error) { 4030d83c49f3SAl Viro dont_mount(dentry); 40318ed936b5SEric W. Biederman detach_mounts(dentry); 4032116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 40338ed936b5SEric W. Biederman } 4034bec1052eSAl Viro } 40351da177e4SLinus Torvalds } 4036b21996e3SJ. Bruce Fields out: 40375955102cSAl Viro inode_unlock(target); 40381da177e4SLinus Torvalds 40391da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40401da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40419accbb97SJ. Bruce Fields fsnotify_link_count(target); 40421da177e4SLinus Torvalds d_delete(dentry); 40431da177e4SLinus Torvalds } 40440eeca283SRobert Love 40451da177e4SLinus Torvalds return error; 40461da177e4SLinus Torvalds } 40474d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40481da177e4SLinus Torvalds 40491da177e4SLinus Torvalds /* 40501da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40511b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40521da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40531da177e4SLinus Torvalds * while waiting on the I/O. 40541da177e4SLinus Torvalds */ 4055da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40561da177e4SLinus Torvalds { 40572ad94ae6SAl Viro int error; 40581da177e4SLinus Torvalds struct dentry *dentry; 4059f5beed75SAl Viro struct path path; 4060f5beed75SAl Viro struct qstr last; 4061f5beed75SAl Viro int type; 40621da177e4SLinus Torvalds struct inode *inode = NULL; 4063b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40645d18f813SJeff Layton unsigned int lookup_flags = 0; 40655d18f813SJeff Layton retry: 4066da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 406791a27b2aSJeff Layton if (IS_ERR(name)) 406891a27b2aSJeff Layton return PTR_ERR(name); 40692ad94ae6SAl Viro 40701da177e4SLinus Torvalds error = -EISDIR; 4071f5beed75SAl Viro if (type != LAST_NORM) 40721da177e4SLinus Torvalds goto exit1; 40730612d9fbSOGAWA Hirofumi 4074f5beed75SAl Viro error = mnt_want_write(path.mnt); 4075c30dabfeSJan Kara if (error) 4076c30dabfeSJan Kara goto exit1; 4077b21996e3SJ. Bruce Fields retry_deleg: 40785955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4079f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40801da177e4SLinus Torvalds error = PTR_ERR(dentry); 40811da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40821da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4083f5beed75SAl Viro if (last.name[last.len]) 408450338b88STörök Edwin goto slashes; 40851da177e4SLinus Torvalds inode = dentry->d_inode; 4086b18825a7SDavid Howells if (d_is_negative(dentry)) 4087e6bc45d6STheodore Ts'o goto slashes; 40887de9c6eeSAl Viro ihold(inode); 4089f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4090be6d3e56SKentaro Takeda if (error) 4091c30dabfeSJan Kara goto exit2; 4092f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40931da177e4SLinus Torvalds exit2: 40941da177e4SLinus Torvalds dput(dentry); 40951da177e4SLinus Torvalds } 40965955102cSAl Viro inode_unlock(path.dentry->d_inode); 40971da177e4SLinus Torvalds if (inode) 40981da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4099b21996e3SJ. Bruce Fields inode = NULL; 4100b21996e3SJ. Bruce Fields if (delegated_inode) { 41015a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4102b21996e3SJ. Bruce Fields if (!error) 4103b21996e3SJ. Bruce Fields goto retry_deleg; 4104b21996e3SJ. Bruce Fields } 4105f5beed75SAl Viro mnt_drop_write(path.mnt); 41061da177e4SLinus Torvalds exit1: 4107f5beed75SAl Viro path_put(&path); 41085d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 41095d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 41105d18f813SJeff Layton inode = NULL; 41115d18f813SJeff Layton goto retry; 41125d18f813SJeff Layton } 4113da2f1362SChristoph Hellwig putname(name); 41141da177e4SLinus Torvalds return error; 41151da177e4SLinus Torvalds 41161da177e4SLinus Torvalds slashes: 4117b18825a7SDavid Howells if (d_is_negative(dentry)) 4118b18825a7SDavid Howells error = -ENOENT; 411944b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4120b18825a7SDavid Howells error = -EISDIR; 4121b18825a7SDavid Howells else 4122b18825a7SDavid Howells error = -ENOTDIR; 41231da177e4SLinus Torvalds goto exit2; 41241da177e4SLinus Torvalds } 41251da177e4SLinus Torvalds 41262e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 41275590ff0dSUlrich Drepper { 41285590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 41295590ff0dSUlrich Drepper return -EINVAL; 41305590ff0dSUlrich Drepper 41315590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 41325590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 41335590ff0dSUlrich Drepper 4134da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 41355590ff0dSUlrich Drepper } 41365590ff0dSUlrich Drepper 41373480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41385590ff0dSUlrich Drepper { 4139da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 41405590ff0dSUlrich Drepper } 41415590ff0dSUlrich Drepper 4142db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 41431da177e4SLinus Torvalds { 4144a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 41451da177e4SLinus Torvalds 41461da177e4SLinus Torvalds if (error) 41471da177e4SLinus Torvalds return error; 41481da177e4SLinus Torvalds 4149acfa4380SAl Viro if (!dir->i_op->symlink) 41501da177e4SLinus Torvalds return -EPERM; 41511da177e4SLinus Torvalds 41521da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41531da177e4SLinus Torvalds if (error) 41541da177e4SLinus Torvalds return error; 41551da177e4SLinus Torvalds 41561da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4157a74574aaSStephen Smalley if (!error) 4158f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41591da177e4SLinus Torvalds return error; 41601da177e4SLinus Torvalds } 41614d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41621da177e4SLinus Torvalds 4163b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 4164b724e846SDominik Brodowski const char __user *newname) 41651da177e4SLinus Torvalds { 41662ad94ae6SAl Viro int error; 416791a27b2aSJeff Layton struct filename *from; 41686902d925SDave Hansen struct dentry *dentry; 4169dae6ad8fSAl Viro struct path path; 4170f46d3567SJeff Layton unsigned int lookup_flags = 0; 41711da177e4SLinus Torvalds 41721da177e4SLinus Torvalds from = getname(oldname); 41731da177e4SLinus Torvalds if (IS_ERR(from)) 41741da177e4SLinus Torvalds return PTR_ERR(from); 4175f46d3567SJeff Layton retry: 4176f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41771da177e4SLinus Torvalds error = PTR_ERR(dentry); 41786902d925SDave Hansen if (IS_ERR(dentry)) 4179dae6ad8fSAl Viro goto out_putname; 41806902d925SDave Hansen 418191a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4182a8104a9fSAl Viro if (!error) 418391a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4184921a1650SAl Viro done_path_create(&path, dentry); 4185f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4186f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4187f46d3567SJeff Layton goto retry; 4188f46d3567SJeff Layton } 41896902d925SDave Hansen out_putname: 41901da177e4SLinus Torvalds putname(from); 41911da177e4SLinus Torvalds return error; 41921da177e4SLinus Torvalds } 41931da177e4SLinus Torvalds 4194b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4195b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4196b724e846SDominik Brodowski { 4197b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4198b724e846SDominik Brodowski } 4199b724e846SDominik Brodowski 42003480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 42015590ff0dSUlrich Drepper { 4202b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 42035590ff0dSUlrich Drepper } 42045590ff0dSUlrich Drepper 4205146a8595SJ. Bruce Fields /** 4206146a8595SJ. Bruce Fields * vfs_link - create a new link 4207146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4208146a8595SJ. Bruce Fields * @dir: new parent 4209146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4210146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4211146a8595SJ. Bruce Fields * 4212146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4213146a8595SJ. Bruce Fields * 4214146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4215146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4216146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4217146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4218146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4219146a8595SJ. Bruce Fields * 4220146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4221146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4222146a8595SJ. Bruce Fields * to be NFS exported. 4223146a8595SJ. Bruce Fields */ 4224146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 42251da177e4SLinus Torvalds { 42261da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 42278de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 42281da177e4SLinus Torvalds int error; 42291da177e4SLinus Torvalds 42301da177e4SLinus Torvalds if (!inode) 42311da177e4SLinus Torvalds return -ENOENT; 42321da177e4SLinus Torvalds 4233a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 42341da177e4SLinus Torvalds if (error) 42351da177e4SLinus Torvalds return error; 42361da177e4SLinus Torvalds 42371da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 42381da177e4SLinus Torvalds return -EXDEV; 42391da177e4SLinus Torvalds 42401da177e4SLinus Torvalds /* 42411da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 42421da177e4SLinus Torvalds */ 42431da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 42441da177e4SLinus Torvalds return -EPERM; 42450bd23d09SEric W. Biederman /* 42460bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 42470bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 42480bd23d09SEric W. Biederman * the vfs. 42490bd23d09SEric W. Biederman */ 42500bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 42510bd23d09SEric W. Biederman return -EPERM; 4252acfa4380SAl Viro if (!dir->i_op->link) 42531da177e4SLinus Torvalds return -EPERM; 42547e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42551da177e4SLinus Torvalds return -EPERM; 42561da177e4SLinus Torvalds 42571da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42581da177e4SLinus Torvalds if (error) 42591da177e4SLinus Torvalds return error; 42601da177e4SLinus Torvalds 42615955102cSAl Viro inode_lock(inode); 4262aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4263f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4264aae8a97dSAneesh Kumar K.V error = -ENOENT; 42658de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42668de52778SAl Viro error = -EMLINK; 4267146a8595SJ. Bruce Fields else { 4268146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4269146a8595SJ. Bruce Fields if (!error) 42701da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4271146a8595SJ. Bruce Fields } 4272f4e0c30cSAl Viro 4273f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4274f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4275f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4276f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4277f4e0c30cSAl Viro } 42785955102cSAl Viro inode_unlock(inode); 4279e31e14ecSStephen Smalley if (!error) 42807e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42811da177e4SLinus Torvalds return error; 42821da177e4SLinus Torvalds } 42834d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42841da177e4SLinus Torvalds 42851da177e4SLinus Torvalds /* 42861da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42871da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42881da177e4SLinus Torvalds * newname. --KAB 42891da177e4SLinus Torvalds * 42901da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42911da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42921da177e4SLinus Torvalds * and other special files. --ADM 42931da177e4SLinus Torvalds */ 429446ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 429546ea89ebSDominik Brodowski const char __user *newname, int flags) 42961da177e4SLinus Torvalds { 42971da177e4SLinus Torvalds struct dentry *new_dentry; 4298dae6ad8fSAl Viro struct path old_path, new_path; 4299146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 430011a7b371SAneesh Kumar K.V int how = 0; 43011da177e4SLinus Torvalds int error; 43021da177e4SLinus Torvalds 430311a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4304c04030e1SUlrich Drepper return -EINVAL; 430511a7b371SAneesh Kumar K.V /* 4306f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4307f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4308f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 430911a7b371SAneesh Kumar K.V */ 4310f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4311f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4312f0cc6ffbSLinus Torvalds return -ENOENT; 431311a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4314f0cc6ffbSLinus Torvalds } 4315c04030e1SUlrich Drepper 431611a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 431711a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4318442e31caSJeff Layton retry: 431911a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 43201da177e4SLinus Torvalds if (error) 43212ad94ae6SAl Viro return error; 43222ad94ae6SAl Viro 4323442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4324442e31caSJeff Layton (how & LOOKUP_REVAL)); 43251da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43266902d925SDave Hansen if (IS_ERR(new_dentry)) 4327dae6ad8fSAl Viro goto out; 4328dae6ad8fSAl Viro 4329dae6ad8fSAl Viro error = -EXDEV; 4330dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4331dae6ad8fSAl Viro goto out_dput; 4332800179c9SKees Cook error = may_linkat(&old_path); 4333800179c9SKees Cook if (unlikely(error)) 4334800179c9SKees Cook goto out_dput; 4335dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4336be6d3e56SKentaro Takeda if (error) 4337a8104a9fSAl Viro goto out_dput; 4338146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 433975c3f29dSDave Hansen out_dput: 4340921a1650SAl Viro done_path_create(&new_path, new_dentry); 4341146a8595SJ. Bruce Fields if (delegated_inode) { 4342146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4343d22e6338SOleg Drokin if (!error) { 4344d22e6338SOleg Drokin path_put(&old_path); 4345146a8595SJ. Bruce Fields goto retry; 4346146a8595SJ. Bruce Fields } 4347d22e6338SOleg Drokin } 4348442e31caSJeff Layton if (retry_estale(error, how)) { 4349d22e6338SOleg Drokin path_put(&old_path); 4350442e31caSJeff Layton how |= LOOKUP_REVAL; 4351442e31caSJeff Layton goto retry; 4352442e31caSJeff Layton } 43531da177e4SLinus Torvalds out: 43542d8f3038SAl Viro path_put(&old_path); 43551da177e4SLinus Torvalds 43561da177e4SLinus Torvalds return error; 43571da177e4SLinus Torvalds } 43581da177e4SLinus Torvalds 435946ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 436046ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 436146ea89ebSDominik Brodowski { 436246ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 436346ea89ebSDominik Brodowski } 436446ea89ebSDominik Brodowski 43653480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43665590ff0dSUlrich Drepper { 436746ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43685590ff0dSUlrich Drepper } 43695590ff0dSUlrich Drepper 4370bc27027aSMiklos Szeredi /** 4371bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4372bc27027aSMiklos Szeredi * @old_dir: parent of source 4373bc27027aSMiklos Szeredi * @old_dentry: source 4374bc27027aSMiklos Szeredi * @new_dir: parent of destination 4375bc27027aSMiklos Szeredi * @new_dentry: destination 4376bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4377520c8b16SMiklos Szeredi * @flags: rename flags 4378bc27027aSMiklos Szeredi * 4379bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4380bc27027aSMiklos Szeredi * 4381bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4382bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4383bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4384bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4385bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4386bc27027aSMiklos Szeredi * so. 4387bc27027aSMiklos Szeredi * 4388bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4389bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4390bc27027aSMiklos Szeredi * to be NFS exported. 4391bc27027aSMiklos Szeredi * 43921da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43931da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43941da177e4SLinus Torvalds * Problems: 43950117d427SMauro Carvalho Chehab * 4396d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43971da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43981da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4399a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 44001da177e4SLinus Torvalds * story. 44016cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 44026cedba89SJ. Bruce Fields * and source (if it is not a directory). 44031b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 44041da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 44051da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4406a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 44071da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 44081da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 44091da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4410a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 44111da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 44121da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 44131da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4414e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 44151b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 44161da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4417c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 44181da177e4SLinus Torvalds * locking]. 44191da177e4SLinus Torvalds */ 44201da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 44218e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4422520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 44231da177e4SLinus Torvalds { 44241da177e4SLinus Torvalds int error; 4425bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4426bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4427bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4428da1ce067SMiklos Szeredi bool new_is_dir = false; 4429da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 443049d31c2fSAl Viro struct name_snapshot old_name; 44311da177e4SLinus Torvalds 44328d3e2936SMiklos Szeredi if (source == target) 44331da177e4SLinus Torvalds return 0; 44341da177e4SLinus Torvalds 44351da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 44361da177e4SLinus Torvalds if (error) 44371da177e4SLinus Torvalds return error; 44381da177e4SLinus Torvalds 4439da1ce067SMiklos Szeredi if (!target) { 4440a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4441da1ce067SMiklos Szeredi } else { 4442da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4443da1ce067SMiklos Szeredi 4444da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44451da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4446da1ce067SMiklos Szeredi else 4447da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4448da1ce067SMiklos Szeredi } 44491da177e4SLinus Torvalds if (error) 44501da177e4SLinus Torvalds return error; 44511da177e4SLinus Torvalds 44522773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 44531da177e4SLinus Torvalds return -EPERM; 44541da177e4SLinus Torvalds 4455bc27027aSMiklos Szeredi /* 4456bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4457bc27027aSMiklos Szeredi * we'll need to flip '..'. 4458bc27027aSMiklos Szeredi */ 4459da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4460da1ce067SMiklos Szeredi if (is_dir) { 4461bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4462bc27027aSMiklos Szeredi if (error) 4463bc27027aSMiklos Szeredi return error; 4464bc27027aSMiklos Szeredi } 4465da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4466da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4467da1ce067SMiklos Szeredi if (error) 4468da1ce067SMiklos Szeredi return error; 4469da1ce067SMiklos Szeredi } 4470da1ce067SMiklos Szeredi } 44710eeca283SRobert Love 44720b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44730b3974ebSMiklos Szeredi flags); 4474bc27027aSMiklos Szeredi if (error) 4475bc27027aSMiklos Szeredi return error; 4476bc27027aSMiklos Szeredi 447749d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4478bc27027aSMiklos Szeredi dget(new_dentry); 4479da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4480bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4481bc27027aSMiklos Szeredi else if (target) 44825955102cSAl Viro inode_lock(target); 4483bc27027aSMiklos Szeredi 4484bc27027aSMiklos Szeredi error = -EBUSY; 44857af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4486bc27027aSMiklos Szeredi goto out; 4487bc27027aSMiklos Szeredi 4488da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4489bc27027aSMiklos Szeredi error = -EMLINK; 4490da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4491bc27027aSMiklos Szeredi goto out; 4492da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4493da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4494da1ce067SMiklos Szeredi goto out; 4495da1ce067SMiklos Szeredi } 4496da1ce067SMiklos Szeredi if (!is_dir) { 4497bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4498bc27027aSMiklos Szeredi if (error) 4499bc27027aSMiklos Szeredi goto out; 4500da1ce067SMiklos Szeredi } 4501da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4502bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4503bc27027aSMiklos Szeredi if (error) 4504bc27027aSMiklos Szeredi goto out; 4505bc27027aSMiklos Szeredi } 4506520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4507520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4508bc27027aSMiklos Szeredi if (error) 4509bc27027aSMiklos Szeredi goto out; 4510bc27027aSMiklos Szeredi 4511da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 45128767712fSAl Viro if (is_dir) { 45138767712fSAl Viro shrink_dcache_parent(new_dentry); 4514bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 45158767712fSAl Viro } 4516bc27027aSMiklos Szeredi dont_mount(new_dentry); 45178ed936b5SEric W. Biederman detach_mounts(new_dentry); 4518bc27027aSMiklos Szeredi } 4519da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4520da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4521bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4522da1ce067SMiklos Szeredi else 4523da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4524da1ce067SMiklos Szeredi } 4525bc27027aSMiklos Szeredi out: 4526da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4527bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4528bc27027aSMiklos Szeredi else if (target) 45295955102cSAl Viro inode_unlock(target); 4530bc27027aSMiklos Szeredi dput(new_dentry); 4531da1ce067SMiklos Szeredi if (!error) { 4532f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4533da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4534da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4535f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4536da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4537da1ce067SMiklos Szeredi } 4538da1ce067SMiklos Szeredi } 453949d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 45400eeca283SRobert Love 45411da177e4SLinus Torvalds return error; 45421da177e4SLinus Torvalds } 45434d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 45441da177e4SLinus Torvalds 4545ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4546ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 45471da177e4SLinus Torvalds { 45481da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 45491da177e4SLinus Torvalds struct dentry *trap; 4550f5beed75SAl Viro struct path old_path, new_path; 4551f5beed75SAl Viro struct qstr old_last, new_last; 4552f5beed75SAl Viro int old_type, new_type; 45538e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 455491a27b2aSJeff Layton struct filename *from; 455591a27b2aSJeff Layton struct filename *to; 4556f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4557c6a94284SJeff Layton bool should_retry = false; 45582ad94ae6SAl Viro int error; 4559520c8b16SMiklos Szeredi 45600d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4561da1ce067SMiklos Szeredi return -EINVAL; 4562da1ce067SMiklos Szeredi 45630d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45640d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4565520c8b16SMiklos Szeredi return -EINVAL; 4566520c8b16SMiklos Szeredi 45670d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45680d7a8555SMiklos Szeredi return -EPERM; 45690d7a8555SMiklos Szeredi 4570f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4571f5beed75SAl Viro target_flags = 0; 4572f5beed75SAl Viro 4573c6a94284SJeff Layton retry: 4574c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4575c1d4dd27SAl Viro &old_path, &old_last, &old_type); 457691a27b2aSJeff Layton if (IS_ERR(from)) { 457791a27b2aSJeff Layton error = PTR_ERR(from); 45781da177e4SLinus Torvalds goto exit; 457991a27b2aSJeff Layton } 45801da177e4SLinus Torvalds 4581c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4582c1d4dd27SAl Viro &new_path, &new_last, &new_type); 458391a27b2aSJeff Layton if (IS_ERR(to)) { 458491a27b2aSJeff Layton error = PTR_ERR(to); 45851da177e4SLinus Torvalds goto exit1; 458691a27b2aSJeff Layton } 45871da177e4SLinus Torvalds 45881da177e4SLinus Torvalds error = -EXDEV; 4589f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45901da177e4SLinus Torvalds goto exit2; 45911da177e4SLinus Torvalds 45921da177e4SLinus Torvalds error = -EBUSY; 4593f5beed75SAl Viro if (old_type != LAST_NORM) 45941da177e4SLinus Torvalds goto exit2; 45951da177e4SLinus Torvalds 45960a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45970a7c3937SMiklos Szeredi error = -EEXIST; 4598f5beed75SAl Viro if (new_type != LAST_NORM) 45991da177e4SLinus Torvalds goto exit2; 46001da177e4SLinus Torvalds 4601f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4602c30dabfeSJan Kara if (error) 4603c30dabfeSJan Kara goto exit2; 4604c30dabfeSJan Kara 46058e6d782cSJ. Bruce Fields retry_deleg: 4606f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 46071da177e4SLinus Torvalds 4608f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 46091da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 46101da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 46111da177e4SLinus Torvalds goto exit3; 46121da177e4SLinus Torvalds /* source must exist */ 46131da177e4SLinus Torvalds error = -ENOENT; 4614b18825a7SDavid Howells if (d_is_negative(old_dentry)) 46151da177e4SLinus Torvalds goto exit4; 4616f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 46171da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 46181da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 46191da177e4SLinus Torvalds goto exit4; 46200a7c3937SMiklos Szeredi error = -EEXIST; 46210a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 46220a7c3937SMiklos Szeredi goto exit5; 4623da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4624da1ce067SMiklos Szeredi error = -ENOENT; 4625da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4626da1ce067SMiklos Szeredi goto exit5; 4627da1ce067SMiklos Szeredi 4628da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4629da1ce067SMiklos Szeredi error = -ENOTDIR; 4630f5beed75SAl Viro if (new_last.name[new_last.len]) 4631da1ce067SMiklos Szeredi goto exit5; 4632da1ce067SMiklos Szeredi } 4633da1ce067SMiklos Szeredi } 46340a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 46350a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 46360a7c3937SMiklos Szeredi error = -ENOTDIR; 4637f5beed75SAl Viro if (old_last.name[old_last.len]) 46380a7c3937SMiklos Szeredi goto exit5; 4639f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 46400a7c3937SMiklos Szeredi goto exit5; 46410a7c3937SMiklos Szeredi } 46420a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 46430a7c3937SMiklos Szeredi error = -EINVAL; 46440a7c3937SMiklos Szeredi if (old_dentry == trap) 46450a7c3937SMiklos Szeredi goto exit5; 46461da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4647da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46481da177e4SLinus Torvalds error = -ENOTEMPTY; 46491da177e4SLinus Torvalds if (new_dentry == trap) 46501da177e4SLinus Torvalds goto exit5; 46511da177e4SLinus Torvalds 4652f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4653f5beed75SAl Viro &new_path, new_dentry, flags); 4654be6d3e56SKentaro Takeda if (error) 4655c30dabfeSJan Kara goto exit5; 4656f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4657f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4658520c8b16SMiklos Szeredi &delegated_inode, flags); 46591da177e4SLinus Torvalds exit5: 46601da177e4SLinus Torvalds dput(new_dentry); 46611da177e4SLinus Torvalds exit4: 46621da177e4SLinus Torvalds dput(old_dentry); 46631da177e4SLinus Torvalds exit3: 4664f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46658e6d782cSJ. Bruce Fields if (delegated_inode) { 46668e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46678e6d782cSJ. Bruce Fields if (!error) 46688e6d782cSJ. Bruce Fields goto retry_deleg; 46698e6d782cSJ. Bruce Fields } 4670f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46711da177e4SLinus Torvalds exit2: 4672c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4673c6a94284SJeff Layton should_retry = true; 4674f5beed75SAl Viro path_put(&new_path); 46752ad94ae6SAl Viro putname(to); 46761da177e4SLinus Torvalds exit1: 4677f5beed75SAl Viro path_put(&old_path); 46781da177e4SLinus Torvalds putname(from); 4679c6a94284SJeff Layton if (should_retry) { 4680c6a94284SJeff Layton should_retry = false; 4681c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4682c6a94284SJeff Layton goto retry; 4683c6a94284SJeff Layton } 46842ad94ae6SAl Viro exit: 46851da177e4SLinus Torvalds return error; 46861da177e4SLinus Torvalds } 46871da177e4SLinus Torvalds 4688ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4689ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4690ee81feb6SDominik Brodowski { 4691ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4692ee81feb6SDominik Brodowski } 4693ee81feb6SDominik Brodowski 4694520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4695520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4696520c8b16SMiklos Szeredi { 4697ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4698520c8b16SMiklos Szeredi } 4699520c8b16SMiklos Szeredi 4700a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 47015590ff0dSUlrich Drepper { 4702ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 47035590ff0dSUlrich Drepper } 47045590ff0dSUlrich Drepper 4705787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4706787fb6bcSMiklos Szeredi { 4707787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4708787fb6bcSMiklos Szeredi if (error) 4709787fb6bcSMiklos Szeredi return error; 4710787fb6bcSMiklos Szeredi 4711787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4712787fb6bcSMiklos Szeredi return -EPERM; 4713787fb6bcSMiklos Szeredi 4714787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4715787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4716787fb6bcSMiklos Szeredi } 4717787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4718787fb6bcSMiklos Szeredi 47195d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 47201da177e4SLinus Torvalds { 47215d826c84SAl Viro int len = PTR_ERR(link); 47221da177e4SLinus Torvalds if (IS_ERR(link)) 47231da177e4SLinus Torvalds goto out; 47241da177e4SLinus Torvalds 47251da177e4SLinus Torvalds len = strlen(link); 47261da177e4SLinus Torvalds if (len > (unsigned) buflen) 47271da177e4SLinus Torvalds len = buflen; 47281da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 47291da177e4SLinus Torvalds len = -EFAULT; 47301da177e4SLinus Torvalds out: 47311da177e4SLinus Torvalds return len; 47321da177e4SLinus Torvalds } 47331da177e4SLinus Torvalds 4734d60874cdSMiklos Szeredi /** 4735fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4736fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4737fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4738fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4739fd4a0edfSMiklos Szeredi * 4740fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4741fd4a0edfSMiklos Szeredi * 4742fd4a0edfSMiklos Szeredi * Does not call security hook. 4743fd4a0edfSMiklos Szeredi */ 4744fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4745fd4a0edfSMiklos Szeredi { 4746fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4747f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4748f2df5da6SAl Viro const char *link; 4749f2df5da6SAl Viro int res; 4750fd4a0edfSMiklos Szeredi 475176fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 475276fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 475376fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 475476fca90eSMiklos Szeredi 475576fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4756fd4a0edfSMiklos Szeredi return -EINVAL; 4757fd4a0edfSMiklos Szeredi 475876fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 475976fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 476076fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 476176fca90eSMiklos Szeredi } 476276fca90eSMiklos Szeredi 47634c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4764f2df5da6SAl Viro if (!link) { 4765f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4766f2df5da6SAl Viro if (IS_ERR(link)) 4767f2df5da6SAl Viro return PTR_ERR(link); 4768f2df5da6SAl Viro } 4769f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4770f2df5da6SAl Viro do_delayed_call(&done); 4771f2df5da6SAl Viro return res; 4772fd4a0edfSMiklos Szeredi } 4773fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 47741da177e4SLinus Torvalds 4775d60874cdSMiklos Szeredi /** 4776d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4777d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4778d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4779d60874cdSMiklos Szeredi * 4780d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4781d60874cdSMiklos Szeredi * 4782d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4783d60874cdSMiklos Szeredi * 4784d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4785d60874cdSMiklos Szeredi */ 4786d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4787d60874cdSMiklos Szeredi { 4788d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4789d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4790d60874cdSMiklos Szeredi 4791d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4792d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4793d60874cdSMiklos Szeredi if (!res) 4794d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4795d60874cdSMiklos Szeredi } 4796d60874cdSMiklos Szeredi return res; 4797d60874cdSMiklos Szeredi } 4798d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4799d60874cdSMiklos Szeredi 48001da177e4SLinus Torvalds /* get the link contents into pagecache */ 48016b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4802fceef393SAl Viro struct delayed_call *callback) 48031da177e4SLinus Torvalds { 4804ebd09abbSDuane Griffin char *kaddr; 48051da177e4SLinus Torvalds struct page *page; 48066b255391SAl Viro struct address_space *mapping = inode->i_mapping; 48076b255391SAl Viro 4808d3883d4fSAl Viro if (!dentry) { 4809d3883d4fSAl Viro page = find_get_page(mapping, 0); 4810d3883d4fSAl Viro if (!page) 48116b255391SAl Viro return ERR_PTR(-ECHILD); 4812d3883d4fSAl Viro if (!PageUptodate(page)) { 4813d3883d4fSAl Viro put_page(page); 4814d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4815d3883d4fSAl Viro } 4816d3883d4fSAl Viro } else { 4817090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 48181da177e4SLinus Torvalds if (IS_ERR(page)) 48196fe6900eSNick Piggin return (char*)page; 4820d3883d4fSAl Viro } 4821fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 482221fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 482321fc61c7SAl Viro kaddr = page_address(page); 48246b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4825ebd09abbSDuane Griffin return kaddr; 48261da177e4SLinus Torvalds } 48271da177e4SLinus Torvalds 48286b255391SAl Viro EXPORT_SYMBOL(page_get_link); 48291da177e4SLinus Torvalds 4830fceef393SAl Viro void page_put_link(void *arg) 48311da177e4SLinus Torvalds { 4832fceef393SAl Viro put_page(arg); 48331da177e4SLinus Torvalds } 48344d359507SAl Viro EXPORT_SYMBOL(page_put_link); 48351da177e4SLinus Torvalds 4836aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4837aa80deabSAl Viro { 4838fceef393SAl Viro DEFINE_DELAYED_CALL(done); 48396b255391SAl Viro int res = readlink_copy(buffer, buflen, 48406b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4841fceef393SAl Viro &done)); 4842fceef393SAl Viro do_delayed_call(&done); 4843aa80deabSAl Viro return res; 4844aa80deabSAl Viro } 4845aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4846aa80deabSAl Viro 484754566b2cSNick Piggin /* 484854566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 484954566b2cSNick Piggin */ 485054566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48511da177e4SLinus Torvalds { 48521da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48530adb25d2SKirill Korotaev struct page *page; 4854afddba49SNick Piggin void *fsdata; 4855beb497abSDmitriy Monakhov int err; 4856c718a975STetsuo Handa unsigned int flags = 0; 485754566b2cSNick Piggin if (nofs) 485854566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 48591da177e4SLinus Torvalds 48607e53cac4SNeilBrown retry: 4861afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 486254566b2cSNick Piggin flags, &page, &fsdata); 48631da177e4SLinus Torvalds if (err) 4864afddba49SNick Piggin goto fail; 4865afddba49SNick Piggin 486621fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4867afddba49SNick Piggin 4868afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4869afddba49SNick Piggin page, fsdata); 48701da177e4SLinus Torvalds if (err < 0) 48711da177e4SLinus Torvalds goto fail; 4872afddba49SNick Piggin if (err < len-1) 4873afddba49SNick Piggin goto retry; 4874afddba49SNick Piggin 48751da177e4SLinus Torvalds mark_inode_dirty(inode); 48761da177e4SLinus Torvalds return 0; 48771da177e4SLinus Torvalds fail: 48781da177e4SLinus Torvalds return err; 48791da177e4SLinus Torvalds } 48804d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 48811da177e4SLinus Torvalds 48820adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 48830adb25d2SKirill Korotaev { 48840adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4885c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48860adb25d2SKirill Korotaev } 48874d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48880adb25d2SKirill Korotaev 488992e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48906b255391SAl Viro .get_link = page_get_link, 48911da177e4SLinus Torvalds }; 48921da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4893