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); 599102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 600102b8af2SAl Viro path_put(&nd->root); 601102b8af2SAl Viro nd->root.mnt = NULL; 602102b8af2SAl Viro } 6037973387aSAl Viro } else { 6047973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6057973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6067973387aSAl Viro nd->root.mnt = NULL; 6077973387aSAl Viro rcu_read_unlock(); 6087973387aSAl Viro } 6097973387aSAl Viro nd->depth = 0; 6107973387aSAl Viro } 6117973387aSAl Viro 6127973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6137973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6147973387aSAl Viro struct path *path, unsigned seq) 6157973387aSAl Viro { 6167973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6177973387aSAl Viro if (unlikely(res)) { 6187973387aSAl Viro if (res > 0) 6197973387aSAl Viro path->mnt = NULL; 6207973387aSAl Viro path->dentry = NULL; 6217973387aSAl Viro return false; 6227973387aSAl Viro } 6237973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6247973387aSAl Viro path->dentry = NULL; 6257973387aSAl Viro return false; 6267973387aSAl Viro } 6277973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6287973387aSAl Viro } 6297973387aSAl Viro 6307973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6317973387aSAl Viro { 6327973387aSAl Viro int i; 6337973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6347973387aSAl Viro struct saved *last = nd->stack + i; 6357973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6367973387aSAl Viro drop_links(nd); 6377973387aSAl Viro nd->depth = i + 1; 6387973387aSAl Viro return false; 6397973387aSAl Viro } 6407973387aSAl Viro } 6417973387aSAl Viro return true; 6427973387aSAl Viro } 6437973387aSAl Viro 64419660af7SAl Viro /* 64531e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 64619660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 64719660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 64857e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 64919660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 65019660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 65119660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 65219660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 65331e6b01fSNick Piggin */ 65431e6b01fSNick Piggin 65531e6b01fSNick Piggin /** 65619660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 65719660af7SAl Viro * @nd: nameidata pathwalk data 65839191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 65931e6b01fSNick Piggin * 6604675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6614675ac39SAl Viro * for ref-walk mode. 6624675ac39SAl Viro * Must be called from rcu-walk context. 6637973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6647973387aSAl Viro * terminate_walk(). 66531e6b01fSNick Piggin */ 6664675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 66731e6b01fSNick Piggin { 66831e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 66931e6b01fSNick Piggin 67031e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 671e5c832d5SLinus Torvalds 672e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6737973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6747973387aSAl Viro goto out2; 6754675ac39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 6764675ac39SAl Viro goto out1; 6774675ac39SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 6784675ac39SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) 6794675ac39SAl Viro goto out; 6804675ac39SAl Viro } 6814675ac39SAl Viro rcu_read_unlock(); 6824675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6834675ac39SAl Viro return 0; 6844675ac39SAl Viro 6854675ac39SAl Viro out2: 6864675ac39SAl Viro nd->path.mnt = NULL; 6874675ac39SAl Viro nd->path.dentry = NULL; 6884675ac39SAl Viro out1: 6894675ac39SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6904675ac39SAl Viro nd->root.mnt = NULL; 6914675ac39SAl Viro out: 6924675ac39SAl Viro rcu_read_unlock(); 6934675ac39SAl Viro return -ECHILD; 6944675ac39SAl Viro } 6954675ac39SAl Viro 6964675ac39SAl Viro /** 6974675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 6984675ac39SAl Viro * @nd: nameidata pathwalk data 6994675ac39SAl Viro * @dentry: child of nd->path.dentry 7004675ac39SAl Viro * @seq: seq number to check dentry against 7014675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7024675ac39SAl Viro * 7034675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7044675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7054675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7064675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7074675ac39SAl Viro * terminate_walk(). 7084675ac39SAl Viro */ 7094675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7104675ac39SAl Viro { 7114675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7124675ac39SAl Viro 7134675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7144675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7154675ac39SAl Viro goto out2; 7167973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7177973387aSAl Viro goto out2; 7184675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7197973387aSAl Viro goto out1; 72048a066e7SAl Viro 72115570086SLinus Torvalds /* 7224675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7234675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7244675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7254675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7264675ac39SAl Viro * know the parent must still be valid if the child sequence number is 72715570086SLinus Torvalds */ 7284675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 729e5c832d5SLinus Torvalds goto out; 7304675ac39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) { 7314675ac39SAl Viro rcu_read_unlock(); 7324675ac39SAl Viro dput(dentry); 7334675ac39SAl Viro goto drop_root_mnt; 73419660af7SAl Viro } 735e5c832d5SLinus Torvalds /* 736e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 737e5c832d5SLinus Torvalds * still valid and get it if required. 738e5c832d5SLinus Torvalds */ 739e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7405a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7415a8d87e8SAl Viro rcu_read_unlock(); 7425a8d87e8SAl Viro dput(dentry); 7435a8d87e8SAl Viro return -ECHILD; 7447973387aSAl Viro } 74531e6b01fSNick Piggin } 74631e6b01fSNick Piggin 7478b61e74fSAl Viro rcu_read_unlock(); 74831e6b01fSNick Piggin return 0; 74919660af7SAl Viro 7507973387aSAl Viro out2: 7517973387aSAl Viro nd->path.mnt = NULL; 7527973387aSAl Viro out1: 7537973387aSAl Viro nd->path.dentry = NULL; 754e5c832d5SLinus Torvalds out: 7558b61e74fSAl Viro rcu_read_unlock(); 756d0d27277SLinus Torvalds drop_root_mnt: 757d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 758d0d27277SLinus Torvalds nd->root.mnt = NULL; 75931e6b01fSNick Piggin return -ECHILD; 76031e6b01fSNick Piggin } 76131e6b01fSNick Piggin 7624ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76334286d66SNick Piggin { 764a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7654ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 766a89f8337SAl Viro else 767a89f8337SAl Viro return 1; 76834286d66SNick Piggin } 76934286d66SNick Piggin 7709f1fafeeSAl Viro /** 7719f1fafeeSAl Viro * complete_walk - successful completion of path walk 7729f1fafeeSAl Viro * @nd: pointer nameidata 77339159de2SJeff Layton * 7749f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7759f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7769f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7779f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7789f1fafeeSAl Viro * need to drop nd->path. 77939159de2SJeff Layton */ 7809f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78139159de2SJeff Layton { 78216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78339159de2SJeff Layton int status; 78439159de2SJeff Layton 7859f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7869f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7879f1fafeeSAl Viro nd->root.mnt = NULL; 7884675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 78948a066e7SAl Viro return -ECHILD; 79048a066e7SAl Viro } 7919f1fafeeSAl Viro 79216c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 79339159de2SJeff Layton return 0; 79439159de2SJeff Layton 795ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 79616c2cd71SAl Viro return 0; 79716c2cd71SAl Viro 798ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 79939159de2SJeff Layton if (status > 0) 80039159de2SJeff Layton return 0; 80139159de2SJeff Layton 80216c2cd71SAl Viro if (!status) 80339159de2SJeff Layton status = -ESTALE; 80416c2cd71SAl Viro 80539159de2SJeff Layton return status; 80639159de2SJeff Layton } 80739159de2SJeff Layton 80818d8c860SAl Viro static void set_root(struct nameidata *nd) 8092a737871SAl Viro { 81031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8119e6697e2SAl Viro 8129e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8138f47a016SAl Viro unsigned seq; 814c28cc364SNick Piggin 815c28cc364SNick Piggin do { 816c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 81731e6b01fSNick Piggin nd->root = fs->root; 8188f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 819c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8209e6697e2SAl Viro } else { 8219e6697e2SAl Viro get_fs_root(fs, &nd->root); 8229e6697e2SAl Viro } 82331e6b01fSNick Piggin } 82431e6b01fSNick Piggin 8251d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 826051d3812SIan Kent { 827051d3812SIan Kent dput(path->dentry); 8284ac91378SJan Blunck if (path->mnt != nd->path.mnt) 829051d3812SIan Kent mntput(path->mnt); 830051d3812SIan Kent } 831051d3812SIan Kent 8327b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8337b9337aaSNick Piggin struct nameidata *nd) 834051d3812SIan Kent { 83531e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8364ac91378SJan Blunck dput(nd->path.dentry); 83731e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8384ac91378SJan Blunck mntput(nd->path.mnt); 8399a229683SHuang Shijie } 84031e6b01fSNick Piggin nd->path.mnt = path->mnt; 8414ac91378SJan Blunck nd->path.dentry = path->dentry; 842051d3812SIan Kent } 843051d3812SIan Kent 844248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 845248fb5b9SAl Viro { 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 */ 8686e77137bSAl Viro void nd_jump_link(struct path *path) 869b5fb63c1SChristoph Hellwig { 8706e77137bSAl Viro struct nameidata *nd = current->nameidata; 871b5fb63c1SChristoph Hellwig path_put(&nd->path); 872b5fb63c1SChristoph Hellwig 873b5fb63c1SChristoph Hellwig nd->path = *path; 874b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 875b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 876b5fb63c1SChristoph Hellwig } 877b5fb63c1SChristoph Hellwig 878b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 879574197e0SAl Viro { 88021c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 881fceef393SAl Viro do_delayed_call(&last->done); 8826548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 883b9ff4429SAl Viro path_put(&last->link); 884574197e0SAl Viro } 885574197e0SAl Viro 886561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 887561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 88830aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 88930aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 890800179c9SKees Cook 891800179c9SKees Cook /** 892800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 89355852635SRandy Dunlap * @nd: nameidata pathwalk data 894800179c9SKees Cook * 895800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 896800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 897800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 898800179c9SKees Cook * processes from failing races against path names that may change out 899800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 900800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 901800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 902800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 903800179c9SKees Cook * 904800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 905800179c9SKees Cook */ 906fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 907800179c9SKees Cook { 908800179c9SKees Cook const struct inode *inode; 909800179c9SKees Cook const struct inode *parent; 9102d7f9e2aSSeth Forshee kuid_t puid; 911800179c9SKees Cook 912800179c9SKees Cook if (!sysctl_protected_symlinks) 913800179c9SKees Cook return 0; 914800179c9SKees Cook 915800179c9SKees Cook /* Allowed if owner and follower match. */ 916fceef393SAl Viro inode = nd->link_inode; 91781abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 918800179c9SKees Cook return 0; 919800179c9SKees Cook 920800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 921aa65fa35SAl Viro parent = nd->inode; 922800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 923800179c9SKees Cook return 0; 924800179c9SKees Cook 925800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9262d7f9e2aSSeth Forshee puid = parent->i_uid; 9272d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 928800179c9SKees Cook return 0; 929800179c9SKees Cook 93031956502SAl Viro if (nd->flags & LOOKUP_RCU) 93131956502SAl Viro return -ECHILD; 93231956502SAl Viro 933ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 93494b9d9b7SRichard Guy Briggs audit_log_link_denied("follow_link"); 935800179c9SKees Cook return -EACCES; 936800179c9SKees Cook } 937800179c9SKees Cook 938800179c9SKees Cook /** 939800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 940800179c9SKees Cook * @inode: the source inode to hardlink from 941800179c9SKees Cook * 942800179c9SKees Cook * Return false if at least one of the following conditions: 943800179c9SKees Cook * - inode is not a regular file 944800179c9SKees Cook * - inode is setuid 945800179c9SKees Cook * - inode is setgid and group-exec 946800179c9SKees Cook * - access failure for read and write 947800179c9SKees Cook * 948800179c9SKees Cook * Otherwise returns true. 949800179c9SKees Cook */ 950800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 951800179c9SKees Cook { 952800179c9SKees Cook umode_t mode = inode->i_mode; 953800179c9SKees Cook 954800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 955800179c9SKees Cook if (!S_ISREG(mode)) 956800179c9SKees Cook return false; 957800179c9SKees Cook 958800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 959800179c9SKees Cook if (mode & S_ISUID) 960800179c9SKees Cook return false; 961800179c9SKees Cook 962800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 963800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 964800179c9SKees Cook return false; 965800179c9SKees Cook 966800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 967800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 968800179c9SKees Cook return false; 969800179c9SKees Cook 970800179c9SKees Cook return true; 971800179c9SKees Cook } 972800179c9SKees Cook 973800179c9SKees Cook /** 974800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 975800179c9SKees Cook * @link: the source to hardlink from 976800179c9SKees Cook * 977800179c9SKees Cook * Block hardlink when all of: 978800179c9SKees Cook * - sysctl_protected_hardlinks enabled 979800179c9SKees Cook * - fsuid does not match inode 980800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 981f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 982800179c9SKees Cook * 983800179c9SKees Cook * Returns 0 if successful, -ve on error. 984800179c9SKees Cook */ 985800179c9SKees Cook static int may_linkat(struct path *link) 986800179c9SKees Cook { 987593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 988593d1ce8SEric W. Biederman 989593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 990593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 991593d1ce8SEric W. Biederman return -EOVERFLOW; 992800179c9SKees Cook 993800179c9SKees Cook if (!sysctl_protected_hardlinks) 994800179c9SKees Cook return 0; 995800179c9SKees Cook 996800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 997800179c9SKees Cook * otherwise, it must be a safe source. 998800179c9SKees Cook */ 999cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1000800179c9SKees Cook return 0; 1001800179c9SKees Cook 100294b9d9b7SRichard Guy Briggs audit_log_link_denied("linkat"); 1003800179c9SKees Cook return -EPERM; 1004800179c9SKees Cook } 1005800179c9SKees Cook 100630aba665SSalvatore Mesoraca /** 100730aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 100830aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 100930aba665SSalvatore Mesoraca * exist. 101030aba665SSalvatore Mesoraca * @dir: the sticky parent directory 101130aba665SSalvatore Mesoraca * @inode: the inode of the file to open 101230aba665SSalvatore Mesoraca * 101330aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 101430aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 101530aba665SSalvatore Mesoraca * - the file already exists 101630aba665SSalvatore Mesoraca * - we are in a sticky directory 101730aba665SSalvatore Mesoraca * - we don't own the file 101830aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 101930aba665SSalvatore Mesoraca * - the directory is world writable 102030aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 102130aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 102230aba665SSalvatore Mesoraca * be enough. 102330aba665SSalvatore Mesoraca * 102430aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 102530aba665SSalvatore Mesoraca */ 102630aba665SSalvatore Mesoraca static int may_create_in_sticky(struct dentry * const dir, 102730aba665SSalvatore Mesoraca struct inode * const inode) 102830aba665SSalvatore Mesoraca { 102930aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 103030aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 103130aba665SSalvatore Mesoraca likely(!(dir->d_inode->i_mode & S_ISVTX)) || 103230aba665SSalvatore Mesoraca uid_eq(inode->i_uid, dir->d_inode->i_uid) || 103330aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 103430aba665SSalvatore Mesoraca return 0; 103530aba665SSalvatore Mesoraca 103630aba665SSalvatore Mesoraca if (likely(dir->d_inode->i_mode & 0002) || 103730aba665SSalvatore Mesoraca (dir->d_inode->i_mode & 0020 && 103830aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 103930aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 104030aba665SSalvatore Mesoraca return -EACCES; 104130aba665SSalvatore Mesoraca } 104230aba665SSalvatore Mesoraca return 0; 104330aba665SSalvatore Mesoraca } 104430aba665SSalvatore Mesoraca 10453b2e7f75SAl Viro static __always_inline 10463b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10471da177e4SLinus Torvalds { 1048ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10491cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1050fceef393SAl Viro struct inode *inode = nd->link_inode; 10516d7b5aaeSAl Viro int error; 10520a959df5SAl Viro const char *res; 10531da177e4SLinus Torvalds 10548fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10558fa9dd24SNeilBrown touch_atime(&last->link); 10568fa9dd24SNeilBrown cond_resched(); 1057c6718543SMiklos Szeredi } else if (atime_needs_update(&last->link, inode)) { 10584675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10598fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10608fa9dd24SNeilBrown touch_atime(&last->link); 10618fa9dd24SNeilBrown } 10628fa9dd24SNeilBrown 1063bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1064bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1065bda0be7aSNeilBrown if (unlikely(error)) 10666920a440SAl Viro return ERR_PTR(error); 106736f3b4f6SAl Viro 106886acdca1SAl Viro nd->last_type = LAST_BIND; 1069d4dee48bSAl Viro res = inode->i_link; 1070d4dee48bSAl Viro if (!res) { 1071fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1072fceef393SAl Viro struct delayed_call *); 1073fceef393SAl Viro get = inode->i_op->get_link; 10748c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1075fceef393SAl Viro res = get(NULL, inode, &last->done); 10766b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10774675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10788c1b4566SAl Viro return ERR_PTR(-ECHILD); 1079fceef393SAl Viro res = get(dentry, inode, &last->done); 10808c1b4566SAl Viro } 10816b255391SAl Viro } else { 1082fceef393SAl Viro res = get(dentry, inode, &last->done); 10836b255391SAl Viro } 1084fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1085fab51e8aSAl Viro return res; 10860a959df5SAl Viro } 1087fab51e8aSAl Viro if (*res == '/') { 1088fab51e8aSAl Viro if (!nd->root.mnt) 1089fab51e8aSAl Viro set_root(nd); 1090248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1091fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1092fab51e8aSAl Viro while (unlikely(*++res == '/')) 1093fab51e8aSAl Viro ; 1094fab51e8aSAl Viro } 1095fab51e8aSAl Viro if (!*res) 1096fab51e8aSAl Viro res = NULL; 10970a959df5SAl Viro return res; 10980a959df5SAl Viro } 10996d7b5aaeSAl Viro 1100f015f126SDavid Howells /* 1101f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1102f015f126SDavid Howells * 1103f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1104f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1105f015f126SDavid Howells * Up is towards /. 1106f015f126SDavid Howells * 1107f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1108f015f126SDavid Howells * root. 1109f015f126SDavid Howells */ 1110bab77ebfSAl Viro int follow_up(struct path *path) 11111da177e4SLinus Torvalds { 11120714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11130714a533SAl Viro struct mount *parent; 11141da177e4SLinus Torvalds struct dentry *mountpoint; 111599b7db7bSNick Piggin 111648a066e7SAl Viro read_seqlock_excl(&mount_lock); 11170714a533SAl Viro parent = mnt->mnt_parent; 11183c0a6163SAl Viro if (parent == mnt) { 111948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11201da177e4SLinus Torvalds return 0; 11211da177e4SLinus Torvalds } 11220714a533SAl Viro mntget(&parent->mnt); 1123a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 112448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1125bab77ebfSAl Viro dput(path->dentry); 1126bab77ebfSAl Viro path->dentry = mountpoint; 1127bab77ebfSAl Viro mntput(path->mnt); 11280714a533SAl Viro path->mnt = &parent->mnt; 11291da177e4SLinus Torvalds return 1; 11301da177e4SLinus Torvalds } 11314d359507SAl Viro EXPORT_SYMBOL(follow_up); 11321da177e4SLinus Torvalds 1133b5c84bf6SNick Piggin /* 11349875cf80SDavid Howells * Perform an automount 11359875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11369875cf80SDavid Howells * were called with. 11371da177e4SLinus Torvalds */ 1138756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 11399875cf80SDavid Howells bool *need_mntput) 114031e6b01fSNick Piggin { 11419875cf80SDavid Howells struct vfsmount *mnt; 1142ea5b778aSDavid Howells int err; 11439875cf80SDavid Howells 11449875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 11459875cf80SDavid Howells return -EREMOTE; 11469875cf80SDavid Howells 11470ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11480ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11490ec26fd0SMiklos Szeredi * the name. 11500ec26fd0SMiklos Szeredi * 11510ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11525a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11530ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11540ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11550ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11560ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11575a30d8a2SDavid Howells */ 1158756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11595d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11605d38f049SIan Kent path->dentry->d_inode) 11619875cf80SDavid Howells return -EISDIR; 11620ec26fd0SMiklos Szeredi 1163756daf26SNeilBrown nd->total_link_count++; 1164756daf26SNeilBrown if (nd->total_link_count >= 40) 11659875cf80SDavid Howells return -ELOOP; 11669875cf80SDavid Howells 11679875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11689875cf80SDavid Howells if (IS_ERR(mnt)) { 11699875cf80SDavid Howells /* 11709875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11719875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11729875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11739875cf80SDavid Howells * 11749875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11759875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11769875cf80SDavid Howells * the path is inaccessible and we should say so. 11779875cf80SDavid Howells */ 1178756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11799875cf80SDavid Howells return -EREMOTE; 11809875cf80SDavid Howells return PTR_ERR(mnt); 118131e6b01fSNick Piggin } 1182ea5b778aSDavid Howells 11839875cf80SDavid Howells if (!mnt) /* mount collision */ 11849875cf80SDavid Howells return 0; 11859875cf80SDavid Howells 11868aef1884SAl Viro if (!*need_mntput) { 11878aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11888aef1884SAl Viro mntget(path->mnt); 11898aef1884SAl Viro *need_mntput = true; 11908aef1884SAl Viro } 119119a167afSAl Viro err = finish_automount(mnt, path); 1192ea5b778aSDavid Howells 1193ea5b778aSDavid Howells switch (err) { 1194ea5b778aSDavid Howells case -EBUSY: 1195ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 119619a167afSAl Viro return 0; 1197ea5b778aSDavid Howells case 0: 11988aef1884SAl Viro path_put(path); 11999875cf80SDavid Howells path->mnt = mnt; 12009875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 12019875cf80SDavid Howells return 0; 120219a167afSAl Viro default: 120319a167afSAl Viro return err; 12049875cf80SDavid Howells } 120519a167afSAl Viro 1206ea5b778aSDavid Howells } 12079875cf80SDavid Howells 12089875cf80SDavid Howells /* 12099875cf80SDavid Howells * Handle a dentry that is managed in some way. 1210cc53ce53SDavid Howells * - Flagged for transit management (autofs) 12119875cf80SDavid Howells * - Flagged as mountpoint 12129875cf80SDavid Howells * - Flagged as automount point 12139875cf80SDavid Howells * 12149875cf80SDavid Howells * This may only be called in refwalk mode. 12159875cf80SDavid Howells * 12169875cf80SDavid Howells * Serialization is taken care of in namespace.c 12179875cf80SDavid Howells */ 1218756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 12199875cf80SDavid Howells { 12208aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 12219875cf80SDavid Howells unsigned managed; 12229875cf80SDavid Howells bool need_mntput = false; 12238aef1884SAl Viro int ret = 0; 12249875cf80SDavid Howells 12259875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 12269875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 12279875cf80SDavid Howells * the components of that value change under us */ 12286aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 12299875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 12309875cf80SDavid Howells unlikely(managed != 0)) { 1231cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1232cc53ce53SDavid Howells * being held. */ 1233cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1234cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1235cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1236fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1237cc53ce53SDavid Howells if (ret < 0) 12388aef1884SAl Viro break; 1239cc53ce53SDavid Howells } 1240cc53ce53SDavid Howells 12419875cf80SDavid Howells /* Transit to a mounted filesystem. */ 12429875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 12439875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12449875cf80SDavid Howells if (mounted) { 12459875cf80SDavid Howells dput(path->dentry); 12469875cf80SDavid Howells if (need_mntput) 1247463ffb2eSAl Viro mntput(path->mnt); 1248463ffb2eSAl Viro path->mnt = mounted; 1249463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12509875cf80SDavid Howells need_mntput = true; 12519875cf80SDavid Howells continue; 1252463ffb2eSAl Viro } 1253463ffb2eSAl Viro 12549875cf80SDavid Howells /* Something is mounted on this dentry in another 12559875cf80SDavid Howells * namespace and/or whatever was mounted there in this 125648a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 125748a066e7SAl Viro * get it */ 12581da177e4SLinus Torvalds } 12599875cf80SDavid Howells 12609875cf80SDavid Howells /* Handle an automount point */ 12619875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1262756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12639875cf80SDavid Howells if (ret < 0) 12648aef1884SAl Viro break; 12659875cf80SDavid Howells continue; 12669875cf80SDavid Howells } 12679875cf80SDavid Howells 12689875cf80SDavid Howells /* We didn't change the current path point */ 12699875cf80SDavid Howells break; 12709875cf80SDavid Howells } 12718aef1884SAl Viro 12728aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12738aef1884SAl Viro mntput(path->mnt); 1274e9742b53SAl Viro if (ret == -EISDIR || !ret) 1275e9742b53SAl Viro ret = 1; 12768402752eSAl Viro if (need_mntput) 12778402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12788402752eSAl Viro if (unlikely(ret < 0)) 12798402752eSAl Viro path_put_conditional(path, nd); 12808402752eSAl Viro return ret; 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds 1283cc53ce53SDavid Howells int follow_down_one(struct path *path) 12841da177e4SLinus Torvalds { 12851da177e4SLinus Torvalds struct vfsmount *mounted; 12861da177e4SLinus Torvalds 12871c755af4SAl Viro mounted = lookup_mnt(path); 12881da177e4SLinus Torvalds if (mounted) { 12899393bd07SAl Viro dput(path->dentry); 12909393bd07SAl Viro mntput(path->mnt); 12919393bd07SAl Viro path->mnt = mounted; 12929393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12931da177e4SLinus Torvalds return 1; 12941da177e4SLinus Torvalds } 12951da177e4SLinus Torvalds return 0; 12961da177e4SLinus Torvalds } 12974d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12981da177e4SLinus Torvalds 1299fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 130062a7375eSIan Kent { 1301fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1302fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 130362a7375eSIan Kent } 130462a7375eSIan Kent 13059875cf80SDavid Howells /* 1306287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1307287548e4SAl Viro * we meet a managed dentry that would need blocking. 13089875cf80SDavid Howells */ 13099875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1310254cf582SAl Viro struct inode **inode, unsigned *seqp) 13119875cf80SDavid Howells { 131262a7375eSIan Kent for (;;) { 1313c7105365SAl Viro struct mount *mounted; 131462a7375eSIan Kent /* 131562a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 131662a7375eSIan Kent * that wants to block transit. 131762a7375eSIan Kent */ 1318fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1319b8faf035SNeilBrown case -ECHILD: 1320b8faf035SNeilBrown default: 1321ab90911fSDavid Howells return false; 1322b8faf035SNeilBrown case -EISDIR: 1323b8faf035SNeilBrown return true; 1324b8faf035SNeilBrown case 0: 1325b8faf035SNeilBrown break; 1326b8faf035SNeilBrown } 132762a7375eSIan Kent 132862a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1329b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 133062a7375eSIan Kent 1331474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 13329875cf80SDavid Howells if (!mounted) 13339875cf80SDavid Howells break; 1334c7105365SAl Viro path->mnt = &mounted->mnt; 1335c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1336a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1337254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 133859430262SLinus Torvalds /* 133959430262SLinus Torvalds * Update the inode too. We don't need to re-check the 134059430262SLinus Torvalds * dentry sequence number here after this d_inode read, 134159430262SLinus Torvalds * because a mount-point is always pinned. 134259430262SLinus Torvalds */ 134359430262SLinus Torvalds *inode = path->dentry->d_inode; 13449875cf80SDavid Howells } 1345f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1346b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1347287548e4SAl Viro } 1348287548e4SAl Viro 134931e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 135031e6b01fSNick Piggin { 13514023bfc9SAl Viro struct inode *inode = nd->inode; 135231e6b01fSNick Piggin 135331e6b01fSNick Piggin while (1) { 1354aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 135531e6b01fSNick Piggin break; 135631e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 135731e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 135831e6b01fSNick Piggin struct dentry *parent = old->d_parent; 135931e6b01fSNick Piggin unsigned seq; 136031e6b01fSNick Piggin 13614023bfc9SAl Viro inode = parent->d_inode; 136231e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1363aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1364aed434adSAl Viro return -ECHILD; 136531e6b01fSNick Piggin nd->path.dentry = parent; 136631e6b01fSNick Piggin nd->seq = seq; 1367397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1368397d425dSEric W. Biederman return -ENOENT; 136931e6b01fSNick Piggin break; 1370aed434adSAl Viro } else { 1371aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1372aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1373aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1374aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1375aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1376aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1377aed434adSAl Viro return -ECHILD; 1378aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 137931e6b01fSNick Piggin break; 1380aed434adSAl Viro /* we know that mountpoint was pinned */ 1381aed434adSAl Viro nd->path.dentry = mountpoint; 1382aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1383aed434adSAl Viro inode = inode2; 1384aed434adSAl Viro nd->seq = seq; 138531e6b01fSNick Piggin } 1386aed434adSAl Viro } 1387aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1388b37199e6SAl Viro struct mount *mounted; 1389b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1390aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1391aed434adSAl Viro return -ECHILD; 1392b37199e6SAl Viro if (!mounted) 1393b37199e6SAl Viro break; 1394b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1395b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13964023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1397b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1398b37199e6SAl Viro } 13994023bfc9SAl Viro nd->inode = inode; 140031e6b01fSNick Piggin return 0; 140131e6b01fSNick Piggin } 140231e6b01fSNick Piggin 14039875cf80SDavid Howells /* 1404cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1405cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1406cc53ce53SDavid Howells * caller is permitted to proceed or not. 1407cc53ce53SDavid Howells */ 14087cc90cc3SAl Viro int follow_down(struct path *path) 1409cc53ce53SDavid Howells { 1410cc53ce53SDavid Howells unsigned managed; 1411cc53ce53SDavid Howells int ret; 1412cc53ce53SDavid Howells 14136aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1414cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1415cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1416cc53ce53SDavid Howells * being held. 1417cc53ce53SDavid Howells * 1418cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1419cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1420cc53ce53SDavid Howells * other than its daemon the right to mount on its 1421cc53ce53SDavid Howells * superstructure. 1422cc53ce53SDavid Howells * 1423cc53ce53SDavid Howells * The filesystem may sleep at this point. 1424cc53ce53SDavid Howells */ 1425cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1426cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1427cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1428fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1429cc53ce53SDavid Howells if (ret < 0) 1430cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1431cc53ce53SDavid Howells } 1432cc53ce53SDavid Howells 1433cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1434cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1435cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1436cc53ce53SDavid Howells if (!mounted) 1437cc53ce53SDavid Howells break; 1438cc53ce53SDavid Howells dput(path->dentry); 1439cc53ce53SDavid Howells mntput(path->mnt); 1440cc53ce53SDavid Howells path->mnt = mounted; 1441cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1442cc53ce53SDavid Howells continue; 1443cc53ce53SDavid Howells } 1444cc53ce53SDavid Howells 1445cc53ce53SDavid Howells /* Don't handle automount points here */ 1446cc53ce53SDavid Howells break; 1447cc53ce53SDavid Howells } 1448cc53ce53SDavid Howells return 0; 1449cc53ce53SDavid Howells } 14504d359507SAl Viro EXPORT_SYMBOL(follow_down); 1451cc53ce53SDavid Howells 1452cc53ce53SDavid Howells /* 14539875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14549875cf80SDavid Howells */ 14559875cf80SDavid Howells static void follow_mount(struct path *path) 14569875cf80SDavid Howells { 14579875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14589875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14599875cf80SDavid Howells if (!mounted) 14609875cf80SDavid Howells break; 14619875cf80SDavid Howells dput(path->dentry); 14629875cf80SDavid Howells mntput(path->mnt); 14639875cf80SDavid Howells path->mnt = mounted; 14649875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14659875cf80SDavid Howells } 14669875cf80SDavid Howells } 14679875cf80SDavid Howells 1468eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1469eedf265aSEric W. Biederman { 1470eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1471eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1472eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1473eedf265aSEric W. Biederman dput(old); 1474eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1475eedf265aSEric W. Biederman return -ENOENT; 1476eedf265aSEric W. Biederman return 0; 1477eedf265aSEric W. Biederman } 1478eedf265aSEric W. Biederman 1479397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14801da177e4SLinus Torvalds { 14811da177e4SLinus Torvalds while(1) { 1482030c7e0bSDanilo Krummrich if (path_equal(&nd->path, &nd->root)) 14831da177e4SLinus Torvalds break; 14844ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1485eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1486eedf265aSEric W. Biederman if (ret) 1487eedf265aSEric W. Biederman return ret; 14881da177e4SLinus Torvalds break; 14891da177e4SLinus Torvalds } 14903088dd70SAl Viro if (!follow_up(&nd->path)) 14911da177e4SLinus Torvalds break; 14921da177e4SLinus Torvalds } 149379ed0226SAl Viro follow_mount(&nd->path); 149431e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1495397d425dSEric W. Biederman return 0; 14961da177e4SLinus Torvalds } 14971da177e4SLinus Torvalds 14981da177e4SLinus Torvalds /* 1499f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1500f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1501baa03890SNick Piggin */ 1502e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1503e3c13928SAl Viro struct dentry *dir, 15046c51e513SAl Viro unsigned int flags) 1505baa03890SNick Piggin { 1506a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1507bad61189SMiklos Szeredi if (dentry) { 1508a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1509bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 151074ff0ffcSAl Viro if (!error) 15115542aa2fSEric W. Biederman d_invalidate(dentry); 1512bad61189SMiklos Szeredi dput(dentry); 151374ff0ffcSAl Viro return ERR_PTR(error); 1514bad61189SMiklos Szeredi } 1515bad61189SMiklos Szeredi } 1516baa03890SNick Piggin return dentry; 1517baa03890SNick Piggin } 1518baa03890SNick Piggin 1519baa03890SNick Piggin /* 1520a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1521a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1522a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1523a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1524a03ece5fSAl Viro * at all. 152544396f4bSJosef Bacik */ 1526a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1527a03ece5fSAl Viro struct dentry *base, unsigned int flags) 152844396f4bSJosef Bacik { 1529a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 153044396f4bSJosef Bacik struct dentry *old; 1531a03ece5fSAl Viro struct inode *dir = base->d_inode; 1532a03ece5fSAl Viro 1533a03ece5fSAl Viro if (dentry) 1534a03ece5fSAl Viro return dentry; 153544396f4bSJosef Bacik 153644396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1537a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 153844396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1539a03ece5fSAl Viro 1540a03ece5fSAl Viro dentry = d_alloc(base, name); 1541a03ece5fSAl Viro if (unlikely(!dentry)) 1542a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 154344396f4bSJosef Bacik 154472bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 154544396f4bSJosef Bacik if (unlikely(old)) { 154644396f4bSJosef Bacik dput(dentry); 154744396f4bSJosef Bacik dentry = old; 154844396f4bSJosef Bacik } 154944396f4bSJosef Bacik return dentry; 155044396f4bSJosef Bacik } 155144396f4bSJosef Bacik 1552e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1553254cf582SAl Viro struct path *path, struct inode **inode, 1554254cf582SAl Viro unsigned *seqp) 15551da177e4SLinus Torvalds { 15564ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 155731e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15585a18fff2SAl Viro int status = 1; 15599875cf80SDavid Howells int err; 15609875cf80SDavid Howells 15613cac260aSAl Viro /* 1562b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15635d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15645d0f49c1SAl Viro * going to fall back to non-racy lookup. 1565b04f784eSNick Piggin */ 156631e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 156731e6b01fSNick Piggin unsigned seq; 1568766c4cbfSAl Viro bool negative; 1569da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15705d0f49c1SAl Viro if (unlikely(!dentry)) { 15714675ac39SAl Viro if (unlazy_walk(nd)) 15725d0f49c1SAl Viro return -ECHILD; 1573e9742b53SAl Viro return 0; 15745d0f49c1SAl Viro } 15755a18fff2SAl Viro 157612f8ad4bSLinus Torvalds /* 157712f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 157812f8ad4bSLinus Torvalds * the dentry name information from lookup. 157912f8ad4bSLinus Torvalds */ 158063afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1581766c4cbfSAl Viro negative = d_is_negative(dentry); 15825d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 158312f8ad4bSLinus Torvalds return -ECHILD; 158412f8ad4bSLinus Torvalds 158512f8ad4bSLinus Torvalds /* 158612f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 158712f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 158812f8ad4bSLinus Torvalds * 158912f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 159012f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 159112f8ad4bSLinus Torvalds */ 15925d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 159331e6b01fSNick Piggin return -ECHILD; 15945a18fff2SAl Viro 1595254cf582SAl Viro *seqp = seq; 15964ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1597209a7fb2SAl Viro if (likely(status > 0)) { 1598daf3761cSTrond Myklebust /* 1599daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1600daf3761cSTrond Myklebust * case that drops it. 1601daf3761cSTrond Myklebust */ 16025d0f49c1SAl Viro if (unlikely(negative)) 1603daf3761cSTrond Myklebust return -ENOENT; 160431e6b01fSNick Piggin path->mnt = mnt; 160531e6b01fSNick Piggin path->dentry = dentry; 1606254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1607e9742b53SAl Viro return 1; 16085d0f49c1SAl Viro } 16094675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 1610254cf582SAl Viro return -ECHILD; 1611209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1612209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1613209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16145a18fff2SAl Viro } else { 1615e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 161681e6f520SAl Viro if (unlikely(!dentry)) 1617e9742b53SAl Viro return 0; 16184ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16195d0f49c1SAl Viro } 16205a18fff2SAl Viro if (unlikely(status <= 0)) { 1621e9742b53SAl Viro if (!status) 16225d0f49c1SAl Viro d_invalidate(dentry); 16235a18fff2SAl Viro dput(dentry); 16245a18fff2SAl Viro return status; 16255a18fff2SAl Viro } 1626766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1627766c4cbfSAl Viro dput(dentry); 1628766c4cbfSAl Viro return -ENOENT; 1629766c4cbfSAl Viro } 16305d0f49c1SAl Viro 16311da177e4SLinus Torvalds path->mnt = mnt; 16321da177e4SLinus Torvalds path->dentry = dentry; 1633756daf26SNeilBrown err = follow_managed(path, nd); 1634e9742b53SAl Viro if (likely(err > 0)) 163563afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16368402752eSAl Viro return err; 1637697f514dSMiklos Szeredi } 1638697f514dSMiklos Szeredi 1639697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 164088d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1641e3c13928SAl Viro struct dentry *dir, 1642e3c13928SAl Viro unsigned int flags) 1643697f514dSMiklos Szeredi { 164488d8331aSAl Viro struct dentry *dentry, *old; 16451936386eSAl Viro struct inode *inode = dir->d_inode; 1646d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16471936386eSAl Viro 16481936386eSAl Viro /* Don't go there if it's already dead */ 164994bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 165088d8331aSAl Viro return ERR_PTR(-ENOENT); 165194bdd655SAl Viro again: 1652d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 165394bdd655SAl Viro if (IS_ERR(dentry)) 165488d8331aSAl Viro return dentry; 165594bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1656a89f8337SAl Viro if (!(flags & LOOKUP_NO_REVAL)) { 1657949a852eSAl Viro int error = d_revalidate(dentry, flags); 1658949a852eSAl Viro if (unlikely(error <= 0)) { 165994bdd655SAl Viro if (!error) { 1660949a852eSAl Viro d_invalidate(dentry); 1661949a852eSAl Viro dput(dentry); 166294bdd655SAl Viro goto again; 166394bdd655SAl Viro } 166494bdd655SAl Viro dput(dentry); 1665949a852eSAl Viro dentry = ERR_PTR(error); 1666949a852eSAl Viro } 1667949a852eSAl Viro } 166894bdd655SAl Viro } else { 16691936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 167085c7f810SAl Viro d_lookup_done(dentry); 16711936386eSAl Viro if (unlikely(old)) { 16721936386eSAl Viro dput(dentry); 16731936386eSAl Viro dentry = old; 1674949a852eSAl Viro } 1675949a852eSAl Viro } 1676e3c13928SAl Viro return dentry; 16771da177e4SLinus Torvalds } 16781da177e4SLinus Torvalds 167988d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 168088d8331aSAl Viro struct dentry *dir, 168188d8331aSAl Viro unsigned int flags) 168288d8331aSAl Viro { 168388d8331aSAl Viro struct inode *inode = dir->d_inode; 168488d8331aSAl Viro struct dentry *res; 168588d8331aSAl Viro inode_lock_shared(inode); 168688d8331aSAl Viro res = __lookup_slow(name, dir, flags); 168788d8331aSAl Viro inode_unlock_shared(inode); 168888d8331aSAl Viro return res; 168988d8331aSAl Viro } 169088d8331aSAl Viro 169152094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 169252094c8aSAl Viro { 169352094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16944ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 169552094c8aSAl Viro if (err != -ECHILD) 169652094c8aSAl Viro return err; 16974675ac39SAl Viro if (unlazy_walk(nd)) 169852094c8aSAl Viro return -ECHILD; 169952094c8aSAl Viro } 17004ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 170152094c8aSAl Viro } 170252094c8aSAl Viro 17039856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 17049856fa1bSAl Viro { 17059856fa1bSAl Viro if (type == LAST_DOTDOT) { 17069e6697e2SAl Viro if (!nd->root.mnt) 17079e6697e2SAl Viro set_root(nd); 17089856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 170970291aecSAl Viro return follow_dotdot_rcu(nd); 17109856fa1bSAl Viro } else 1711397d425dSEric W. Biederman return follow_dotdot(nd); 17129856fa1bSAl Viro } 17139856fa1bSAl Viro return 0; 17149856fa1bSAl Viro } 17159856fa1bSAl Viro 1716181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1717181548c0SAl Viro struct inode *inode, unsigned seq) 1718d63ff28fSAl Viro { 1719626de996SAl Viro int error; 17201cf2665bSAl Viro struct saved *last; 1721756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1722626de996SAl Viro path_to_nameidata(link, nd); 1723626de996SAl Viro return -ELOOP; 1724626de996SAl Viro } 1725bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1726cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1727cd179f44SAl Viro mntget(link->mnt); 17287973387aSAl Viro } 1729626de996SAl Viro error = nd_alloc_stack(nd); 1730626de996SAl Viro if (unlikely(error)) { 1731bc40aee0SAl Viro if (error == -ECHILD) { 1732ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1733ad1633a1SAl Viro drop_links(nd); 1734ad1633a1SAl Viro nd->depth = 0; 1735ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1736ad1633a1SAl Viro nd->path.mnt = NULL; 1737ad1633a1SAl Viro nd->path.dentry = NULL; 1738ad1633a1SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1739ad1633a1SAl Viro nd->root.mnt = NULL; 1740ad1633a1SAl Viro rcu_read_unlock(); 17414675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1742bc40aee0SAl Viro error = nd_alloc_stack(nd); 1743bc40aee0SAl Viro } 1744bc40aee0SAl Viro if (error) { 1745cd179f44SAl Viro path_put(link); 1746626de996SAl Viro return error; 1747626de996SAl Viro } 1748bc40aee0SAl Viro } 1749626de996SAl Viro 1750ab104923SAl Viro last = nd->stack + nd->depth++; 17511cf2665bSAl Viro last->link = *link; 1752fceef393SAl Viro clear_delayed_call(&last->done); 1753fceef393SAl Viro nd->link_inode = inode; 17540450b2d1SAl Viro last->seq = seq; 1755d63ff28fSAl Viro return 1; 1756d63ff28fSAl Viro } 1757d63ff28fSAl Viro 17588f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 175931d66bcdSAl Viro 17603ddcd056SLinus Torvalds /* 17613ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17623ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17633ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17643ddcd056SLinus Torvalds * for the common case. 17653ddcd056SLinus Torvalds */ 17668f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 17678f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 17683ddcd056SLinus Torvalds { 176931d66bcdSAl Viro if (!(flags & WALK_MORE) && nd->depth) 177031d66bcdSAl Viro put_link(nd); 17718f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 17728f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 17738f64fb1cSAl Viro /* not a symlink or should not follow */ 17748f64fb1cSAl Viro path_to_nameidata(path, nd); 17758f64fb1cSAl Viro nd->inode = inode; 17768f64fb1cSAl Viro nd->seq = seq; 1777d63ff28fSAl Viro return 0; 17788f64fb1cSAl Viro } 1779a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1780a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 17818f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1782a7f77542SAl Viro return -ECHILD; 1783a7f77542SAl Viro } 17848f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 17853ddcd056SLinus Torvalds } 17863ddcd056SLinus Torvalds 17874693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1788ce57dfc1SAl Viro { 1789caa85634SAl Viro struct path path; 1790ce57dfc1SAl Viro struct inode *inode; 1791254cf582SAl Viro unsigned seq; 1792ce57dfc1SAl Viro int err; 1793ce57dfc1SAl Viro /* 1794ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1795ce57dfc1SAl Viro * to be able to know about the current root directory and 1796ce57dfc1SAl Viro * parent relationships. 1797ce57dfc1SAl Viro */ 17984693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17994693a547SAl Viro err = handle_dots(nd, nd->last_type); 18001c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18014693a547SAl Viro put_link(nd); 18024693a547SAl Viro return err; 18034693a547SAl Viro } 1804254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1805e9742b53SAl Viro if (unlikely(err <= 0)) { 1806697f514dSMiklos Szeredi if (err < 0) 1807f0a9ba70SAl Viro return err; 1808e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1809e3c13928SAl Viro nd->flags); 1810e3c13928SAl Viro if (IS_ERR(path.dentry)) 1811e3c13928SAl Viro return PTR_ERR(path.dentry); 18127500c38aSAl Viro 1813e3c13928SAl Viro path.mnt = nd->path.mnt; 1814e3c13928SAl Viro err = follow_managed(&path, nd); 1815e3c13928SAl Viro if (unlikely(err < 0)) 1816f0a9ba70SAl Viro return err; 1817697f514dSMiklos Szeredi 18187500c38aSAl Viro if (unlikely(d_is_negative(path.dentry))) { 18197500c38aSAl Viro path_to_nameidata(&path, nd); 18207500c38aSAl Viro return -ENOENT; 18217500c38aSAl Viro } 18227500c38aSAl Viro 1823254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1824d4565649SAl Viro inode = d_backing_inode(path.dentry); 1825766c4cbfSAl Viro } 1826697f514dSMiklos Szeredi 18278f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1828ce57dfc1SAl Viro } 1829ce57dfc1SAl Viro 18301da177e4SLinus Torvalds /* 1831bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1832bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1833bfcfaa77SLinus Torvalds * 1834bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1835bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1836bfcfaa77SLinus Torvalds * fast. 1837bfcfaa77SLinus Torvalds * 1838bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1839bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1840bfcfaa77SLinus Torvalds * crossing operation. 1841bfcfaa77SLinus Torvalds * 1842bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1843bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1844bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1845bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1846bfcfaa77SLinus Torvalds */ 1847bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1848bfcfaa77SLinus Torvalds 1849f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1850bfcfaa77SLinus Torvalds 1851468a9428SGeorge Spelvin #ifdef HASH_MIX 1852bfcfaa77SLinus Torvalds 1853468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1854468a9428SGeorge Spelvin 1855468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18562a18da7aSGeorge Spelvin /* 18572a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18582a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18592a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18602a18da7aSGeorge Spelvin * and no temporaries. 18612a18da7aSGeorge Spelvin * 18622a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18632a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18642a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18652a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18662a18da7aSGeorge Spelvin * 18672a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18682a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18692a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18702a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18712a18da7aSGeorge Spelvin * 18722a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18732a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18742a18da7aSGeorge Spelvin * toggling any given output bit. 18752a18da7aSGeorge Spelvin * 18762a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18772a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18782a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18792a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18802a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18812a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18822a18da7aSGeorge Spelvin * Perfect: 8192 258048 18832a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18842a18da7aSGeorge Spelvin */ 18852a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18862a18da7aSGeorge Spelvin ( x ^= (a), \ 18872a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18882a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 18892a18da7aSGeorge Spelvin y *= 9 ) 1890bfcfaa77SLinus Torvalds 18910fed3ac8SGeorge Spelvin /* 18922a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 18932a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 18942a18da7aSGeorge Spelvin * work done before the hash value is used. 18950fed3ac8SGeorge Spelvin */ 18962a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18970fed3ac8SGeorge Spelvin { 18982a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 18992a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19002a18da7aSGeorge Spelvin return y >> 32; 19010fed3ac8SGeorge Spelvin } 19020fed3ac8SGeorge Spelvin 1903bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1904bfcfaa77SLinus Torvalds 19052a18da7aSGeorge Spelvin /* 19062a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19072a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19082a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19092a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19102a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19112a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19122a18da7aSGeorge Spelvin * Perfect: 2048 31744 19132a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19142a18da7aSGeorge Spelvin */ 19152a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19162a18da7aSGeorge Spelvin ( x ^= (a), \ 19172a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19182a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19192a18da7aSGeorge Spelvin y *= 9 ) 1920bfcfaa77SLinus Torvalds 19212a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19220fed3ac8SGeorge Spelvin { 19232a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19242a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19250fed3ac8SGeorge Spelvin } 19260fed3ac8SGeorge Spelvin 1927bfcfaa77SLinus Torvalds #endif 1928bfcfaa77SLinus Torvalds 19292a18da7aSGeorge Spelvin /* 19302a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19312a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19322a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19332a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19342a18da7aSGeorge Spelvin * finds the delimiter after the name. 19352a18da7aSGeorge Spelvin */ 19368387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1937bfcfaa77SLinus Torvalds { 19388387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1939bfcfaa77SLinus Torvalds 1940bfcfaa77SLinus Torvalds for (;;) { 1941fcfd2fbfSGeorge Spelvin if (!len) 1942fcfd2fbfSGeorge Spelvin goto done; 1943e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1944bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1945bfcfaa77SLinus Torvalds break; 19462a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1947bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1948bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1949bfcfaa77SLinus Torvalds } 19502a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1951bfcfaa77SLinus Torvalds done: 19522a18da7aSGeorge Spelvin return fold_hash(x, y); 1953bfcfaa77SLinus Torvalds } 1954bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1955bfcfaa77SLinus Torvalds 1956fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19578387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1958fcfd2fbfSGeorge Spelvin { 19598387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19608387ff25SLinus Torvalds unsigned long adata, mask, len; 1961fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1962fcfd2fbfSGeorge Spelvin 19638387ff25SLinus Torvalds len = 0; 19648387ff25SLinus Torvalds goto inside; 19658387ff25SLinus Torvalds 1966fcfd2fbfSGeorge Spelvin do { 19672a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1968fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19698387ff25SLinus Torvalds inside: 1970fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1971fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1972fcfd2fbfSGeorge Spelvin 1973fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1974fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19752a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1976fcfd2fbfSGeorge Spelvin 19772a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1978fcfd2fbfSGeorge Spelvin } 1979fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1980fcfd2fbfSGeorge Spelvin 1981bfcfaa77SLinus Torvalds /* 1982bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1983d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1984bfcfaa77SLinus Torvalds */ 19858387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1986bfcfaa77SLinus Torvalds { 19878387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 19888387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 198936126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1990bfcfaa77SLinus Torvalds 19918387ff25SLinus Torvalds len = 0; 19928387ff25SLinus Torvalds goto inside; 19938387ff25SLinus Torvalds 1994bfcfaa77SLinus Torvalds do { 19952a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1996bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 19978387ff25SLinus Torvalds inside: 1998e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 199936126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 200036126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2001bfcfaa77SLinus Torvalds 200236126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 200336126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 200436126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20052a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 200636126f8fSLinus Torvalds 20072a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2008bfcfaa77SLinus Torvalds } 2009bfcfaa77SLinus Torvalds 20102a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2011bfcfaa77SLinus Torvalds 2012fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20138387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20140145acc2SLinus Torvalds { 20158387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20160145acc2SLinus Torvalds while (len--) 2017fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20180145acc2SLinus Torvalds return end_name_hash(hash); 20190145acc2SLinus Torvalds } 2020ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20210145acc2SLinus Torvalds 2022fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20238387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2024fcfd2fbfSGeorge Spelvin { 20258387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2026fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2027fcfd2fbfSGeorge Spelvin 2028fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2029e0ab7af9SGeorge Spelvin while (c) { 2030fcfd2fbfSGeorge Spelvin len++; 2031fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2032fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2033e0ab7af9SGeorge Spelvin } 2034fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2035fcfd2fbfSGeorge Spelvin } 2036f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2037fcfd2fbfSGeorge Spelvin 20383ddcd056SLinus Torvalds /* 2039200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2040200e9ef7SLinus Torvalds * one character. 2041200e9ef7SLinus Torvalds */ 20428387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2043200e9ef7SLinus Torvalds { 20448387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2045200e9ef7SLinus Torvalds unsigned long len = 0, c; 2046200e9ef7SLinus Torvalds 2047200e9ef7SLinus Torvalds c = (unsigned char)*name; 2048200e9ef7SLinus Torvalds do { 2049200e9ef7SLinus Torvalds len++; 2050200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2051200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2052200e9ef7SLinus Torvalds } while (c && c != '/'); 2053d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2054200e9ef7SLinus Torvalds } 2055200e9ef7SLinus Torvalds 2056bfcfaa77SLinus Torvalds #endif 2057bfcfaa77SLinus Torvalds 2058200e9ef7SLinus Torvalds /* 20591da177e4SLinus Torvalds * Name resolution. 2060ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2061ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20621da177e4SLinus Torvalds * 2063ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2064ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20651da177e4SLinus Torvalds */ 20666de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20671da177e4SLinus Torvalds { 20681da177e4SLinus Torvalds int err; 20691da177e4SLinus Torvalds 20709b5858e9SAl Viro if (IS_ERR(name)) 20719b5858e9SAl Viro return PTR_ERR(name); 20721da177e4SLinus Torvalds while (*name=='/') 20731da177e4SLinus Torvalds name++; 20741da177e4SLinus Torvalds if (!*name) 20759e18f10aSAl Viro return 0; 20761da177e4SLinus Torvalds 20771da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20781da177e4SLinus Torvalds for(;;) { 2079d6bb3e90SLinus Torvalds u64 hash_len; 2080fe479a58SAl Viro int type; 20811da177e4SLinus Torvalds 208252094c8aSAl Viro err = may_lookup(nd); 20831da177e4SLinus Torvalds if (err) 20843595e234SAl Viro return err; 20851da177e4SLinus Torvalds 20868387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 20871da177e4SLinus Torvalds 2088fe479a58SAl Viro type = LAST_NORM; 2089d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2090fe479a58SAl Viro case 2: 2091200e9ef7SLinus Torvalds if (name[1] == '.') { 2092fe479a58SAl Viro type = LAST_DOTDOT; 209316c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 209416c2cd71SAl Viro } 2095fe479a58SAl Viro break; 2096fe479a58SAl Viro case 1: 2097fe479a58SAl Viro type = LAST_DOT; 2098fe479a58SAl Viro } 20995a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21005a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 210116c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21025a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2103a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2104da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21055a202bcdSAl Viro if (err < 0) 21063595e234SAl Viro return err; 2107d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2108d6bb3e90SLinus Torvalds name = this.name; 21095a202bcdSAl Viro } 21105a202bcdSAl Viro } 2111fe479a58SAl Viro 2112d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2113d6bb3e90SLinus Torvalds nd->last.name = name; 21145f4a6a69SAl Viro nd->last_type = type; 21155f4a6a69SAl Viro 2116d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2117d6bb3e90SLinus Torvalds if (!*name) 2118bdf6cbf1SAl Viro goto OK; 2119200e9ef7SLinus Torvalds /* 2120200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2121200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2122200e9ef7SLinus Torvalds */ 2123200e9ef7SLinus Torvalds do { 2124d6bb3e90SLinus Torvalds name++; 2125d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21268620c238SAl Viro if (unlikely(!*name)) { 21278620c238SAl Viro OK: 2128368ee9baSAl Viro /* pathname body, done */ 21298620c238SAl Viro if (!nd->depth) 21308620c238SAl Viro return 0; 21318620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2132368ee9baSAl Viro /* trailing symlink, done */ 21338620c238SAl Viro if (!name) 21348620c238SAl Viro return 0; 21358620c238SAl Viro /* last component of nested symlink */ 21368f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 21371c4ff1a8SAl Viro } else { 21381c4ff1a8SAl Viro /* not the last component */ 21398f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 21408620c238SAl Viro } 2141ce57dfc1SAl Viro if (err < 0) 21423595e234SAl Viro return err; 2143fe479a58SAl Viro 2144ce57dfc1SAl Viro if (err) { 2145626de996SAl Viro const char *s = get_link(nd); 21465a460275SAl Viro 2147a1c83681SViresh Kumar if (IS_ERR(s)) 21483595e234SAl Viro return PTR_ERR(s); 2149172a39a0SAl Viro err = 0; 215012b09578SAl Viro if (unlikely(!s)) { 215112b09578SAl Viro /* jumped */ 2152b9ff4429SAl Viro put_link(nd); 215312b09578SAl Viro } else { 2154071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 215532cd7468SAl Viro name = s; 21569e18f10aSAl Viro continue; 215748c8b0c5SAl Viro } 215831e6b01fSNick Piggin } 215997242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 216097242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21614675ac39SAl Viro if (unlazy_walk(nd)) 216297242f99SAl Viro return -ECHILD; 216397242f99SAl Viro } 21643595e234SAl Viro return -ENOTDIR; 21655f4a6a69SAl Viro } 2166ce57dfc1SAl Viro } 216797242f99SAl Viro } 21681da177e4SLinus Torvalds 2169edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2170c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 217131e6b01fSNick Piggin { 2172c8a53ee5SAl Viro const char *s = nd->name->name; 217331e6b01fSNick Piggin 2174c0eb027eSLinus Torvalds if (!*s) 2175c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2176edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2177edc2b1daSAl Viro rcu_read_lock(); 2178c0eb027eSLinus Torvalds 217931e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2180980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 218131e6b01fSNick Piggin nd->depth = 0; 21825b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2183b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2184b18825a7SDavid Howells struct inode *inode = root->d_inode; 218593893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2186368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21875b6ca027SAl Viro nd->path = nd->root; 21885b6ca027SAl Viro nd->inode = inode; 21895b6ca027SAl Viro if (flags & LOOKUP_RCU) { 21905b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 21918f47a016SAl Viro nd->root_seq = nd->seq; 219248a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 21935b6ca027SAl Viro } else { 21945b6ca027SAl Viro path_get(&nd->path); 21955b6ca027SAl Viro } 2196368ee9baSAl Viro return s; 21975b6ca027SAl Viro } 21985b6ca027SAl Viro 219931e6b01fSNick Piggin nd->root.mnt = NULL; 2200248fb5b9SAl Viro nd->path.mnt = NULL; 2201248fb5b9SAl Viro nd->path.dentry = NULL; 220231e6b01fSNick Piggin 220348a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2204fd2f7cb5SAl Viro if (*s == '/') { 2205e41f7d4eSAl Viro set_root(nd); 2206248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2207ef55d917SAl Viro return s; 2208ef55d917SAl Viro return ERR_PTR(-ECHILD); 2209c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2210e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 221131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2212c28cc364SNick Piggin unsigned seq; 221331e6b01fSNick Piggin 2214c28cc364SNick Piggin do { 2215c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 221631e6b01fSNick Piggin nd->path = fs->pwd; 2217ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2218c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2219c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2220e41f7d4eSAl Viro } else { 2221e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2222ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2223e41f7d4eSAl Viro } 2224ef55d917SAl Viro return s; 222531e6b01fSNick Piggin } else { 2226582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2227c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 222831e6b01fSNick Piggin struct dentry *dentry; 222931e6b01fSNick Piggin 22302903ff01SAl Viro if (!f.file) 2231368ee9baSAl Viro return ERR_PTR(-EBADF); 223231e6b01fSNick Piggin 22332903ff01SAl Viro dentry = f.file->f_path.dentry; 223431e6b01fSNick Piggin 2235edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22362903ff01SAl Viro fdput(f); 2237368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2238f52e0c11SAl Viro } 22392903ff01SAl Viro 22402903ff01SAl Viro nd->path = f.file->f_path; 2241e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 224234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 224334a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22445590ff0dSUlrich Drepper } else { 22452903ff01SAl Viro path_get(&nd->path); 224634a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22471da177e4SLinus Torvalds } 224834a26b99SAl Viro fdput(f); 2249368ee9baSAl Viro return s; 2250e41f7d4eSAl Viro } 22519b4a9b14SAl Viro } 22529b4a9b14SAl Viro 22533bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 225495fa25d9SAl Viro { 225595fa25d9SAl Viro const char *s; 2256fec2fa24SAl Viro int error = may_follow_link(nd); 2257deb106c6SAl Viro if (unlikely(error)) 22583bdba28bSAl Viro return ERR_PTR(error); 225995fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2260fab51e8aSAl Viro nd->stack[0].name = NULL; 22613b2e7f75SAl Viro s = get_link(nd); 2262deb106c6SAl Viro return s ? s : ""; 226395fa25d9SAl Viro } 226495fa25d9SAl Viro 2265caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2266bd92d7feSAl Viro { 2267bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2268bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2269bd92d7feSAl Viro 2270bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 22711c4ff1a8SAl Viro return walk_component(nd, 0); 2272bd92d7feSAl Viro } 2273bd92d7feSAl Viro 22744f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22754f757f3cSAl Viro { 22764f757f3cSAl Viro struct path path = nd->path; 22774f757f3cSAl Viro struct inode *inode = nd->inode; 22784f757f3cSAl Viro unsigned seq = nd->seq; 22794f757f3cSAl Viro int err; 22804f757f3cSAl Viro 22814f757f3cSAl Viro if (nd->flags & LOOKUP_RCU) { 22824f757f3cSAl Viro /* 22834f757f3cSAl Viro * don't bother with unlazy_walk on failure - we are 22844f757f3cSAl Viro * at the very beginning of walk, so we lose nothing 22854f757f3cSAl Viro * if we simply redo everything in non-RCU mode 22864f757f3cSAl Viro */ 22874f757f3cSAl Viro if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq))) 22884f757f3cSAl Viro return -ECHILD; 22894f757f3cSAl Viro } else { 22904f757f3cSAl Viro dget(path.dentry); 22914f757f3cSAl Viro err = follow_managed(&path, nd); 22924f757f3cSAl Viro if (unlikely(err < 0)) 22934f757f3cSAl Viro return err; 22944f757f3cSAl Viro inode = d_backing_inode(path.dentry); 22954f757f3cSAl Viro seq = 0; 22964f757f3cSAl Viro } 22974f757f3cSAl Viro path_to_nameidata(&path, nd); 22984f757f3cSAl Viro nd->inode = inode; 22994f757f3cSAl Viro nd->seq = seq; 23004f757f3cSAl Viro return 0; 23014f757f3cSAl Viro } 23024f757f3cSAl Viro 23039b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2304c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23059b4a9b14SAl Viro { 2306c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2307bd92d7feSAl Viro int err; 230831e6b01fSNick Piggin 23099b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23104f757f3cSAl Viro err = handle_lookup_down(nd); 23115f336e72SAl Viro if (unlikely(err < 0)) 23125f336e72SAl Viro s = ERR_PTR(err); 23134f757f3cSAl Viro } 23144f757f3cSAl Viro 23153bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 23163bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 23173bdba28bSAl Viro s = trailing_symlink(nd); 2318bd92d7feSAl Viro } 23199f1fafeeSAl Viro if (!err) 23209f1fafeeSAl Viro err = complete_walk(nd); 2321bd92d7feSAl Viro 2322deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2323deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2324bd23a539SAl Viro err = -ENOTDIR; 2325625b6d10SAl Viro if (!err) { 2326625b6d10SAl Viro *path = nd->path; 2327625b6d10SAl Viro nd->path.mnt = NULL; 2328625b6d10SAl Viro nd->path.dentry = NULL; 2329625b6d10SAl Viro } 2330deb106c6SAl Viro terminate_walk(nd); 2331bd92d7feSAl Viro return err; 233231e6b01fSNick Piggin } 233331e6b01fSNick Piggin 233431d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23359ad1aaa6SAl Viro struct path *path, struct path *root) 2336873f1eedSJeff Layton { 2337894bc8c4SAl Viro int retval; 23389883d185SAl Viro struct nameidata nd; 2339abc9f5beSAl Viro if (IS_ERR(name)) 2340abc9f5beSAl Viro return PTR_ERR(name); 23419ad1aaa6SAl Viro if (unlikely(root)) { 23429ad1aaa6SAl Viro nd.root = *root; 23439ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23449ad1aaa6SAl Viro } 23459883d185SAl Viro set_nameidata(&nd, dfd, name); 2346c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2347873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2348c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2349873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2350c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2351873f1eedSJeff Layton 2352873f1eedSJeff Layton if (likely(!retval)) 2353625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 23549883d185SAl Viro restore_nameidata(); 2355e4bd1c1aSAl Viro putname(name); 2356873f1eedSJeff Layton return retval; 2357873f1eedSJeff Layton } 2358873f1eedSJeff Layton 23598bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2360c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2361391172c4SAl Viro struct path *parent) 23628bcb77faSAl Viro { 2363c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23649b5858e9SAl Viro int err = link_path_walk(s, nd); 23658bcb77faSAl Viro if (!err) 23668bcb77faSAl Viro err = complete_walk(nd); 2367391172c4SAl Viro if (!err) { 2368391172c4SAl Viro *parent = nd->path; 2369391172c4SAl Viro nd->path.mnt = NULL; 2370391172c4SAl Viro nd->path.dentry = NULL; 2371391172c4SAl Viro } 2372deb106c6SAl Viro terminate_walk(nd); 23738bcb77faSAl Viro return err; 23748bcb77faSAl Viro } 23758bcb77faSAl Viro 23765c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2377391172c4SAl Viro unsigned int flags, struct path *parent, 2378391172c4SAl Viro struct qstr *last, int *type) 23798bcb77faSAl Viro { 23808bcb77faSAl Viro int retval; 23819883d185SAl Viro struct nameidata nd; 23828bcb77faSAl Viro 23835c31b6ceSAl Viro if (IS_ERR(name)) 23845c31b6ceSAl Viro return name; 23859883d185SAl Viro set_nameidata(&nd, dfd, name); 2386c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23878bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2388c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23898bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2390c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2391391172c4SAl Viro if (likely(!retval)) { 2392391172c4SAl Viro *last = nd.last; 2393391172c4SAl Viro *type = nd.last_type; 2394391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 23955c31b6ceSAl Viro } else { 23965c31b6ceSAl Viro putname(name); 23975c31b6ceSAl Viro name = ERR_PTR(retval); 2398391172c4SAl Viro } 23999883d185SAl Viro restore_nameidata(); 24005c31b6ceSAl Viro return name; 24018bcb77faSAl Viro } 24028bcb77faSAl Viro 240379714f72SAl Viro /* does lookup, returns the object with parent locked */ 240479714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24055590ff0dSUlrich Drepper { 24065c31b6ceSAl Viro struct filename *filename; 24075c31b6ceSAl Viro struct dentry *d; 2408391172c4SAl Viro struct qstr last; 2409391172c4SAl Viro int type; 241051689104SPaul Moore 24115c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24125c31b6ceSAl Viro &last, &type); 241351689104SPaul Moore if (IS_ERR(filename)) 241451689104SPaul Moore return ERR_CAST(filename); 24155c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2416391172c4SAl Viro path_put(path); 24175c31b6ceSAl Viro putname(filename); 24185c31b6ceSAl Viro return ERR_PTR(-EINVAL); 241979714f72SAl Viro } 24205955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2421391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 242279714f72SAl Viro if (IS_ERR(d)) { 24235955102cSAl Viro inode_unlock(path->dentry->d_inode); 2424391172c4SAl Viro path_put(path); 242579714f72SAl Viro } 242651689104SPaul Moore putname(filename); 242779714f72SAl Viro return d; 24285590ff0dSUlrich Drepper } 24295590ff0dSUlrich Drepper 2430d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2431d1811465SAl Viro { 2432abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2433abc9f5beSAl Viro flags, path, NULL); 2434d1811465SAl Viro } 24354d359507SAl Viro EXPORT_SYMBOL(kern_path); 2436d1811465SAl Viro 243716f18200SJosef 'Jeff' Sipek /** 243816f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 243916f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 244016f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 244116f18200SJosef 'Jeff' Sipek * @name: pointer to file name 244216f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2443e0a01249SAl Viro * @path: pointer to struct path to fill 244416f18200SJosef 'Jeff' Sipek */ 244516f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 244616f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2447e0a01249SAl Viro struct path *path) 244816f18200SJosef 'Jeff' Sipek { 24499ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24509ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2451abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2452abc9f5beSAl Viro flags , path, &root); 245316f18200SJosef 'Jeff' Sipek } 24544d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 245516f18200SJosef 'Jeff' Sipek 24563c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24573c95f0dcSAl Viro int len, struct qstr *this) 24583c95f0dcSAl Viro { 24593c95f0dcSAl Viro this->name = name; 24603c95f0dcSAl Viro this->len = len; 24613c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24623c95f0dcSAl Viro if (!len) 24633c95f0dcSAl Viro return -EACCES; 24643c95f0dcSAl Viro 24653c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24663c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24673c95f0dcSAl Viro return -EACCES; 24683c95f0dcSAl Viro } 24693c95f0dcSAl Viro 24703c95f0dcSAl Viro while (len--) { 24713c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24723c95f0dcSAl Viro if (c == '/' || c == '\0') 24733c95f0dcSAl Viro return -EACCES; 24743c95f0dcSAl Viro } 24753c95f0dcSAl Viro /* 24763c95f0dcSAl Viro * See if the low-level filesystem might want 24773c95f0dcSAl Viro * to use its own hash.. 24783c95f0dcSAl Viro */ 24793c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 24803c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 24813c95f0dcSAl Viro if (err < 0) 24823c95f0dcSAl Viro return err; 24833c95f0dcSAl Viro } 24843c95f0dcSAl Viro 24853c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 24863c95f0dcSAl Viro } 24873c95f0dcSAl Viro 2488eead1911SChristoph Hellwig /** 24890da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 24900da0b7fdSDavid Howells * @name: pathname component to lookup 24910da0b7fdSDavid Howells * @base: base directory to lookup from 24920da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 24930da0b7fdSDavid Howells * 24940da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 24950da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 24960da0b7fdSDavid Howells * 24970da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 24980da0b7fdSDavid Howells * not be called by generic code. 24990da0b7fdSDavid Howells * 25000da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25010da0b7fdSDavid Howells */ 25020da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25030da0b7fdSDavid Howells { 25040da0b7fdSDavid Howells struct qstr this; 25050da0b7fdSDavid Howells int err; 25060da0b7fdSDavid Howells 25070da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25080da0b7fdSDavid Howells 25090da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25100da0b7fdSDavid Howells if (err) 25110da0b7fdSDavid Howells return ERR_PTR(err); 25120da0b7fdSDavid Howells 25130da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25140da0b7fdSDavid Howells } 25150da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25160da0b7fdSDavid Howells 25170da0b7fdSDavid Howells /** 2518a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2519eead1911SChristoph Hellwig * @name: pathname component to lookup 2520eead1911SChristoph Hellwig * @base: base directory to lookup from 2521eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2522eead1911SChristoph Hellwig * 2523a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25249e7543e9SAl Viro * not be called by generic code. 2525bbddca8eSNeilBrown * 2526bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2527eead1911SChristoph Hellwig */ 2528057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2529057f6c01SJames Morris { 25308613a209SAl Viro struct dentry *dentry; 2531057f6c01SJames Morris struct qstr this; 2532cda309deSMiklos Szeredi int err; 2533057f6c01SJames Morris 25345955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25352f9092e1SDavid Woodhouse 25363c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2537cda309deSMiklos Szeredi if (err) 2538cda309deSMiklos Szeredi return ERR_PTR(err); 2539cda309deSMiklos Szeredi 25408613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25418613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2542057f6c01SJames Morris } 25434d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2544057f6c01SJames Morris 2545bbddca8eSNeilBrown /** 2546bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2547bbddca8eSNeilBrown * @name: pathname component to lookup 2548bbddca8eSNeilBrown * @base: base directory to lookup from 2549bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2550bbddca8eSNeilBrown * 2551bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2552bbddca8eSNeilBrown * not be called by generic code. 2553bbddca8eSNeilBrown * 2554bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2555bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2556bbddca8eSNeilBrown */ 2557bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2558bbddca8eSNeilBrown struct dentry *base, int len) 2559bbddca8eSNeilBrown { 2560bbddca8eSNeilBrown struct qstr this; 2561bbddca8eSNeilBrown int err; 256220d00ee8SLinus Torvalds struct dentry *ret; 2563bbddca8eSNeilBrown 25643c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2565bbddca8eSNeilBrown if (err) 2566bbddca8eSNeilBrown return ERR_PTR(err); 2567bbddca8eSNeilBrown 256820d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 256920d00ee8SLinus Torvalds if (!ret) 257020d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 257120d00ee8SLinus Torvalds return ret; 2572bbddca8eSNeilBrown } 2573bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2574bbddca8eSNeilBrown 2575eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2576eedf265aSEric W. Biederman int path_pts(struct path *path) 2577eedf265aSEric W. Biederman { 2578eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2579eedf265aSEric W. Biederman * the input path. 2580eedf265aSEric W. Biederman */ 2581eedf265aSEric W. Biederman struct dentry *child, *parent; 2582eedf265aSEric W. Biederman struct qstr this; 2583eedf265aSEric W. Biederman int ret; 2584eedf265aSEric W. Biederman 2585eedf265aSEric W. Biederman ret = path_parent_directory(path); 2586eedf265aSEric W. Biederman if (ret) 2587eedf265aSEric W. Biederman return ret; 2588eedf265aSEric W. Biederman 2589eedf265aSEric W. Biederman parent = path->dentry; 2590eedf265aSEric W. Biederman this.name = "pts"; 2591eedf265aSEric W. Biederman this.len = 3; 2592eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2593eedf265aSEric W. Biederman if (!child) 2594eedf265aSEric W. Biederman return -ENOENT; 2595eedf265aSEric W. Biederman 2596eedf265aSEric W. Biederman path->dentry = child; 2597eedf265aSEric W. Biederman dput(parent); 2598eedf265aSEric W. Biederman follow_mount(path); 2599eedf265aSEric W. Biederman return 0; 2600eedf265aSEric W. Biederman } 2601eedf265aSEric W. Biederman #endif 2602eedf265aSEric W. Biederman 26031fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26041fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26051da177e4SLinus Torvalds { 2606abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2607abc9f5beSAl Viro flags, path, NULL); 26081da177e4SLinus Torvalds } 2609b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26101fa1e7f6SAndy Whitcroft 26118033426eSJeff Layton /** 2612197df04cSAl Viro * mountpoint_last - look up last component for umount 26138033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 26148033426eSJeff Layton * 26158033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 26168033426eSJeff Layton * need to resolve the path without doing any revalidation. 26178033426eSJeff Layton * 26188033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 26198033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 26208033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 26218033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 26228033426eSJeff Layton * bogus and it doesn't exist. 26238033426eSJeff Layton * 26248033426eSJeff Layton * Returns: 26258033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2626ba8f4613SAl Viro * lookup found a negative dentry. 26278033426eSJeff Layton * 2628ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2629ba8f4613SAl Viro * symlink that needs to be followed. 26308033426eSJeff Layton * 26318033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2632ba8f4613SAl Viro * that needs to be followed. 26338033426eSJeff Layton */ 26348033426eSJeff Layton static int 2635ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 26368033426eSJeff Layton { 26378033426eSJeff Layton int error = 0; 26388033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2639ba8f4613SAl Viro struct path path; 26408033426eSJeff Layton 264135759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 264235759521SAl Viro if (nd->flags & LOOKUP_RCU) { 26434675ac39SAl Viro if (unlazy_walk(nd)) 2644deb106c6SAl Viro return -ECHILD; 26458033426eSJeff Layton } 26468033426eSJeff Layton 26478033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 26488033426eSJeff Layton 26498033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 26508033426eSJeff Layton error = handle_dots(nd, nd->last_type); 265135759521SAl Viro if (error) 2652deb106c6SAl Viro return error; 2653ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2654949a852eSAl Viro } else { 2655ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2656ba8f4613SAl Viro if (!path.dentry) { 26578033426eSJeff Layton /* 2658949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2659949a852eSAl Viro * cache, so that means that this dentry is probably 2660949a852eSAl Viro * a symlink or the path doesn't actually point 2661949a852eSAl Viro * to a mounted dentry. 26628033426eSJeff Layton */ 2663ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2664949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2665ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2666ba8f4613SAl Viro return PTR_ERR(path.dentry); 26678033426eSJeff Layton } 2668bcceeebaSDave Jones } 2669ba8f4613SAl Viro if (d_is_negative(path.dentry)) { 2670ba8f4613SAl Viro dput(path.dentry); 2671deb106c6SAl Viro return -ENOENT; 267235759521SAl Viro } 2673ba8f4613SAl Viro path.mnt = nd->path.mnt; 26748f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 26758033426eSJeff Layton } 26768033426eSJeff Layton 26778033426eSJeff Layton /** 2678197df04cSAl Viro * path_mountpoint - look up a path to be umounted 26792a78b857SMasanari Iida * @nd: lookup context 26808033426eSJeff Layton * @flags: lookup flags 2681c8a53ee5SAl Viro * @path: pointer to container for result 26828033426eSJeff Layton * 26838033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2684606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 26858033426eSJeff Layton */ 26868033426eSJeff Layton static int 2687c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 26888033426eSJeff Layton { 2689c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2690368ee9baSAl Viro int err; 26919b5858e9SAl Viro 26923bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2693ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 26943bdba28bSAl Viro s = trailing_symlink(nd); 26953bdba28bSAl Viro } 2696ba8f4613SAl Viro if (!err) { 2697ba8f4613SAl Viro *path = nd->path; 2698ba8f4613SAl Viro nd->path.mnt = NULL; 2699ba8f4613SAl Viro nd->path.dentry = NULL; 2700ba8f4613SAl Viro follow_mount(path); 2701ba8f4613SAl Viro } 2702deb106c6SAl Viro terminate_walk(nd); 27038033426eSJeff Layton return err; 27048033426eSJeff Layton } 27058033426eSJeff Layton 27062d864651SAl Viro static int 2707668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 27082d864651SAl Viro unsigned int flags) 27092d864651SAl Viro { 27109883d185SAl Viro struct nameidata nd; 2711cbaab2dbSAl Viro int error; 2712668696dcSAl Viro if (IS_ERR(name)) 2713668696dcSAl Viro return PTR_ERR(name); 27149883d185SAl Viro set_nameidata(&nd, dfd, name); 2715c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 27162d864651SAl Viro if (unlikely(error == -ECHILD)) 2717c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 27182d864651SAl Viro if (unlikely(error == -ESTALE)) 2719c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 27202d864651SAl Viro if (likely(!error)) 272157d46577SRichard Guy Briggs audit_inode(name, path->dentry, flags & LOOKUP_NO_EVAL); 27229883d185SAl Viro restore_nameidata(); 2723668696dcSAl Viro putname(name); 27242d864651SAl Viro return error; 27252d864651SAl Viro } 27262d864651SAl Viro 27278033426eSJeff Layton /** 2728197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 27298033426eSJeff Layton * @dfd: directory file descriptor 27308033426eSJeff Layton * @name: pathname from userland 27318033426eSJeff Layton * @flags: lookup flags 27328033426eSJeff Layton * @path: pointer to container to hold result 27338033426eSJeff Layton * 27348033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27358033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27368033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27378033426eSJeff Layton * and avoid revalidating the last component. 27388033426eSJeff Layton * 27398033426eSJeff Layton * Returns 0 and populates "path" on success. 27408033426eSJeff Layton */ 27418033426eSJeff Layton int 2742197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27438033426eSJeff Layton struct path *path) 27448033426eSJeff Layton { 2745cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27468033426eSJeff Layton } 27478033426eSJeff Layton 27482d864651SAl Viro int 27492d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27502d864651SAl Viro unsigned int flags) 27512d864651SAl Viro { 2752cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27532d864651SAl Viro } 27542d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27552d864651SAl Viro 2756cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27571da177e4SLinus Torvalds { 27588e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2759da9592edSDavid Howells 27608e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27611da177e4SLinus Torvalds return 0; 27628e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27631da177e4SLinus Torvalds return 0; 276423adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27651da177e4SLinus Torvalds } 2766cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27671da177e4SLinus Torvalds 27681da177e4SLinus Torvalds /* 27691da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27701da177e4SLinus Torvalds * whether the type of victim is right. 27711da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27721da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27731da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27741da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27751da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27761da177e4SLinus Torvalds * a. be owner of dir, or 27771da177e4SLinus Torvalds * b. be owner of victim, or 27781da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27791da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27801da177e4SLinus Torvalds * links pointing to it. 27810bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27820bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27830bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27840bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27850bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27861da177e4SLinus Torvalds * nfs_async_unlink(). 27871da177e4SLinus Torvalds */ 2788b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27891da177e4SLinus Torvalds { 279063afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27911da177e4SLinus Torvalds int error; 27921da177e4SLinus Torvalds 2793b18825a7SDavid Howells if (d_is_negative(victim)) 27941da177e4SLinus Torvalds return -ENOENT; 2795b18825a7SDavid Howells BUG_ON(!inode); 27961da177e4SLinus Torvalds 27971da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2798593d1ce8SEric W. Biederman 2799593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2800593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2801593d1ce8SEric W. Biederman return -EOVERFLOW; 2802593d1ce8SEric W. Biederman 28034fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 28041da177e4SLinus Torvalds 2805f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 28061da177e4SLinus Torvalds if (error) 28071da177e4SLinus Torvalds return error; 28081da177e4SLinus Torvalds if (IS_APPEND(dir)) 28091da177e4SLinus Torvalds return -EPERM; 2810b18825a7SDavid Howells 2811b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 28120bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 28131da177e4SLinus Torvalds return -EPERM; 28141da177e4SLinus Torvalds if (isdir) { 281544b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28161da177e4SLinus Torvalds return -ENOTDIR; 28171da177e4SLinus Torvalds if (IS_ROOT(victim)) 28181da177e4SLinus Torvalds return -EBUSY; 281944b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28201da177e4SLinus Torvalds return -EISDIR; 28211da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28221da177e4SLinus Torvalds return -ENOENT; 28231da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28241da177e4SLinus Torvalds return -EBUSY; 28251da177e4SLinus Torvalds return 0; 28261da177e4SLinus Torvalds } 28271da177e4SLinus Torvalds 28281da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28291da177e4SLinus Torvalds * dir. 28301da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28311da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28321da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2833036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2834036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2835036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28361da177e4SLinus Torvalds */ 2837a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28381da177e4SLinus Torvalds { 2839036d5236SEric W. Biederman struct user_namespace *s_user_ns; 284014e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28411da177e4SLinus Torvalds if (child->d_inode) 28421da177e4SLinus Torvalds return -EEXIST; 28431da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28441da177e4SLinus Torvalds return -ENOENT; 2845036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2846036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2847036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2848036d5236SEric W. Biederman return -EOVERFLOW; 2849f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28501da177e4SLinus Torvalds } 28511da177e4SLinus Torvalds 28521da177e4SLinus Torvalds /* 28531da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28541da177e4SLinus Torvalds */ 28551da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28561da177e4SLinus Torvalds { 28571da177e4SLinus Torvalds struct dentry *p; 28581da177e4SLinus Torvalds 28591da177e4SLinus Torvalds if (p1 == p2) { 28605955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28611da177e4SLinus Torvalds return NULL; 28621da177e4SLinus Torvalds } 28631da177e4SLinus Torvalds 2864fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28651da177e4SLinus Torvalds 2866e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2867e2761a11SOGAWA Hirofumi if (p) { 28685955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28695955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28701da177e4SLinus Torvalds return p; 28711da177e4SLinus Torvalds } 28721da177e4SLinus Torvalds 2873e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2874e2761a11SOGAWA Hirofumi if (p) { 28755955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28765955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28771da177e4SLinus Torvalds return p; 28781da177e4SLinus Torvalds } 28791da177e4SLinus Torvalds 28805955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28815955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28821da177e4SLinus Torvalds return NULL; 28831da177e4SLinus Torvalds } 28844d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28851da177e4SLinus Torvalds 28861da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28871da177e4SLinus Torvalds { 28885955102cSAl Viro inode_unlock(p1->d_inode); 28891da177e4SLinus Torvalds if (p1 != p2) { 28905955102cSAl Viro inode_unlock(p2->d_inode); 2891fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28921da177e4SLinus Torvalds } 28931da177e4SLinus Torvalds } 28944d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28951da177e4SLinus Torvalds 28964acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2897312b63fbSAl Viro bool want_excl) 28981da177e4SLinus Torvalds { 2899a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 29001da177e4SLinus Torvalds if (error) 29011da177e4SLinus Torvalds return error; 29021da177e4SLinus Torvalds 2903acfa4380SAl Viro if (!dir->i_op->create) 29041da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 29051da177e4SLinus Torvalds mode &= S_IALLUGO; 29061da177e4SLinus Torvalds mode |= S_IFREG; 29071da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 29081da177e4SLinus Torvalds if (error) 29091da177e4SLinus Torvalds return error; 2910312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2911a74574aaSStephen Smalley if (!error) 2912f38aa942SAmy Griffis fsnotify_create(dir, dentry); 29131da177e4SLinus Torvalds return error; 29141da177e4SLinus Torvalds } 29154d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29161da177e4SLinus Torvalds 29178e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29188e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29198e6c848eSAl Viro void *arg) 29208e6c848eSAl Viro { 29218e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 29228e6c848eSAl Viro int error = may_create(dir, dentry); 29238e6c848eSAl Viro if (error) 29248e6c848eSAl Viro return error; 29258e6c848eSAl Viro 29268e6c848eSAl Viro mode &= S_IALLUGO; 29278e6c848eSAl Viro mode |= S_IFREG; 29288e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29298e6c848eSAl Viro if (error) 29308e6c848eSAl Viro return error; 29318e6c848eSAl Viro error = f(dentry, mode, arg); 29328e6c848eSAl Viro if (!error) 29338e6c848eSAl Viro fsnotify_create(dir, dentry); 29348e6c848eSAl Viro return error; 29358e6c848eSAl Viro } 29368e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29378e6c848eSAl Viro 2938a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2939a2982cc9SEric W. Biederman { 2940a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2941a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2942a2982cc9SEric W. Biederman } 2943a2982cc9SEric W. Biederman 2944f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 29451da177e4SLinus Torvalds { 29463fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29471da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29481da177e4SLinus Torvalds int error; 29491da177e4SLinus Torvalds 29501da177e4SLinus Torvalds if (!inode) 29511da177e4SLinus Torvalds return -ENOENT; 29521da177e4SLinus Torvalds 2953c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2954c8fe8f30SChristoph Hellwig case S_IFLNK: 29551da177e4SLinus Torvalds return -ELOOP; 2956c8fe8f30SChristoph Hellwig case S_IFDIR: 2957c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29581da177e4SLinus Torvalds return -EISDIR; 2959c8fe8f30SChristoph Hellwig break; 2960c8fe8f30SChristoph Hellwig case S_IFBLK: 2961c8fe8f30SChristoph Hellwig case S_IFCHR: 2962a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29631da177e4SLinus Torvalds return -EACCES; 2964c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2965c8fe8f30SChristoph Hellwig case S_IFIFO: 2966c8fe8f30SChristoph Hellwig case S_IFSOCK: 29671da177e4SLinus Torvalds flag &= ~O_TRUNC; 2968c8fe8f30SChristoph Hellwig break; 29694a3fd211SDave Hansen } 2970b41572e9SDave Hansen 297162fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2972b41572e9SDave Hansen if (error) 2973b41572e9SDave Hansen return error; 29746146f0d5SMimi Zohar 29751da177e4SLinus Torvalds /* 29761da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29771da177e4SLinus Torvalds */ 29781da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29798737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29807715b521SAl Viro return -EPERM; 29811da177e4SLinus Torvalds if (flag & O_TRUNC) 29827715b521SAl Viro return -EPERM; 29831da177e4SLinus Torvalds } 29841da177e4SLinus Torvalds 29851da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29862e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29877715b521SAl Viro return -EPERM; 29881da177e4SLinus Torvalds 2989f3c7691eSJ. Bruce Fields return 0; 29907715b521SAl Viro } 29917715b521SAl Viro 2992e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29937715b521SAl Viro { 2994f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29957715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29967715b521SAl Viro int error = get_write_access(inode); 29971da177e4SLinus Torvalds if (error) 29987715b521SAl Viro return error; 29991da177e4SLinus Torvalds /* 30001da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 30011da177e4SLinus Torvalds */ 3002d7a06983SJeff Layton error = locks_verify_locked(filp); 3003be6d3e56SKentaro Takeda if (!error) 3004ea0d3ab2STetsuo Handa error = security_path_truncate(path); 30051da177e4SLinus Torvalds if (!error) { 30067715b521SAl Viro error = do_truncate(path->dentry, 0, 3007d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3008e1181ee6SJeff Layton filp); 30091da177e4SLinus Torvalds } 30101da177e4SLinus Torvalds put_write_access(inode); 3011acd0c935SMimi Zohar return error; 30121da177e4SLinus Torvalds } 30131da177e4SLinus Torvalds 3014d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3015d57999e1SDave Hansen { 30168a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30178a5e929dSAl Viro flag--; 3018d57999e1SDave Hansen return flag; 3019d57999e1SDave Hansen } 3020d57999e1SDave Hansen 3021d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 3022d18e9008SMiklos Szeredi { 30231328c727SSeth Forshee struct user_namespace *s_user_ns; 3024d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3025d18e9008SMiklos Szeredi if (error) 3026d18e9008SMiklos Szeredi return error; 3027d18e9008SMiklos Szeredi 30281328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 30291328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 30301328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 30311328c727SSeth Forshee return -EOVERFLOW; 30321328c727SSeth Forshee 3033d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 3034d18e9008SMiklos Szeredi if (error) 3035d18e9008SMiklos Szeredi return error; 3036d18e9008SMiklos Szeredi 3037d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3038d18e9008SMiklos Szeredi } 3039d18e9008SMiklos Szeredi 30401acf0af9SDavid Howells /* 30411acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30421acf0af9SDavid Howells * dentry. 30431acf0af9SDavid Howells * 30441acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30451acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30461acf0af9SDavid Howells * 304700a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 304800a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 304900a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 30501acf0af9SDavid Howells * 30511acf0af9SDavid Howells * Returns an error code otherwise. 30521acf0af9SDavid Howells */ 30532675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 305430d90494SAl Viro struct path *path, struct file *file, 3055015c3bbcSMiklos Szeredi const struct open_flags *op, 30563ec2eef1SAl Viro int open_flag, umode_t mode) 3057d18e9008SMiklos Szeredi { 3058d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3059d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3060d18e9008SMiklos Szeredi int error; 3061d18e9008SMiklos Szeredi 3062384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3063d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3064d18e9008SMiklos Szeredi 3065d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3066d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3067d18e9008SMiklos Szeredi 306830d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 306930d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30700fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 307144907d79SAl Viro open_to_namei_flags(open_flag), mode); 30726fbd0714SAl Viro d_lookup_done(dentry); 3073384f26e2SAl Viro if (!error) { 307464e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 3075d18e9008SMiklos Szeredi /* 3076384f26e2SAl Viro * We didn't have the inode before the open, so check open 3077384f26e2SAl Viro * permission here. 3078d18e9008SMiklos Szeredi */ 3079384f26e2SAl Viro int acc_mode = op->acc_mode; 308073a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 308103da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 308203da633aSAl Viro fsnotify_create(dir, dentry); 308362fb4a15SAl Viro acc_mode = 0; 308403da633aSAl Viro } 30852675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3086384f26e2SAl Viro if (WARN_ON(error > 0)) 3087384f26e2SAl Viro error = -EINVAL; 308864e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3089d18e9008SMiklos Szeredi error = -EIO; 3090384f26e2SAl Viro } else { 3091d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3092d18e9008SMiklos Szeredi dput(dentry); 3093d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 309410c64ceaSAl Viro } 309573a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 3096d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3097a01e718fSAl Viro if (unlikely(d_is_negative(dentry))) { 3098a01e718fSAl Viro error = -ENOENT; 3099a01e718fSAl Viro } else { 3100d18e9008SMiklos Szeredi path->dentry = dentry; 3101d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 310200a07c15SAl Viro return 0; 3103d18e9008SMiklos Szeredi } 3104d18e9008SMiklos Szeredi } 3105a01e718fSAl Viro } 3106d18e9008SMiklos Szeredi dput(dentry); 3107d18e9008SMiklos Szeredi return error; 3108d18e9008SMiklos Szeredi } 3109d18e9008SMiklos Szeredi 311031e6b01fSNick Piggin /* 31111acf0af9SDavid Howells * Look up and maybe create and open the last component. 3112d58ffd35SMiklos Szeredi * 311300a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3114d58ffd35SMiklos Szeredi * 311500a07c15SAl Viro * Returns 0 on success, that is, if 311600a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 311700a07c15SAl Viro * the file was not completely opened at this time, though lookups and 311800a07c15SAl Viro * creations were performed. 311900a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 312000a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 312100a07c15SAl Viro * hadn't been specified. 31221acf0af9SDavid Howells * 312300a07c15SAl Viro * An error code is returned on failure. 3124d58ffd35SMiklos Szeredi */ 31252675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 312630d90494SAl Viro struct file *file, 3127d58ffd35SMiklos Szeredi const struct open_flags *op, 31283ec2eef1SAl Viro bool got_write) 3129d58ffd35SMiklos Szeredi { 3130d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 313154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31321643b43fSAl Viro int open_flag = op->open_flag; 3133d58ffd35SMiklos Szeredi struct dentry *dentry; 31341643b43fSAl Viro int error, create_error = 0; 31351643b43fSAl Viro umode_t mode = op->mode; 31366fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3137d58ffd35SMiklos Szeredi 3138ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3139ce8644fcSAl Viro return -ENOENT; 3140d58ffd35SMiklos Szeredi 314173a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 31426fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31436fbd0714SAl Viro for (;;) { 31446fbd0714SAl Viro if (!dentry) { 31456fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3146d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 31472675a4ebSAl Viro return PTR_ERR(dentry); 31486fbd0714SAl Viro } 31496fbd0714SAl Viro if (d_in_lookup(dentry)) 31506fbd0714SAl Viro break; 3151d58ffd35SMiklos Szeredi 31526fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31536fbd0714SAl Viro if (likely(error > 0)) 31546fbd0714SAl Viro break; 31556fbd0714SAl Viro if (error) 31566fbd0714SAl Viro goto out_dput; 31576fbd0714SAl Viro d_invalidate(dentry); 31586fbd0714SAl Viro dput(dentry); 31596fbd0714SAl Viro dentry = NULL; 31606fbd0714SAl Viro } 31616fbd0714SAl Viro if (dentry->d_inode) { 3162d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3163d18e9008SMiklos Szeredi goto out_no_open; 31646c51e513SAl Viro } 3165d18e9008SMiklos Szeredi 31661643b43fSAl Viro /* 31671643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31681643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31691643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31701643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31711643b43fSAl Viro * 31721643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31731643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31741643b43fSAl Viro */ 31751643b43fSAl Viro if (open_flag & O_CREAT) { 31761643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31771643b43fSAl Viro mode &= ~current_umask(); 31781643b43fSAl Viro if (unlikely(!got_write)) { 31791643b43fSAl Viro create_error = -EROFS; 31801643b43fSAl Viro open_flag &= ~O_CREAT; 31811643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31821643b43fSAl Viro goto no_open; 31831643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31841643b43fSAl Viro } else { 31851643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31861643b43fSAl Viro if (create_error) { 31871643b43fSAl Viro open_flag &= ~O_CREAT; 31881643b43fSAl Viro if (open_flag & O_EXCL) 31891643b43fSAl Viro goto no_open; 31901643b43fSAl Viro } 31911643b43fSAl Viro } 31921643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31931643b43fSAl Viro unlikely(!got_write)) { 31941643b43fSAl Viro /* 31951643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31961643b43fSAl Viro * back to lookup + open 31971643b43fSAl Viro */ 31981643b43fSAl Viro goto no_open; 3199d18e9008SMiklos Szeredi } 3200d18e9008SMiklos Szeredi 32011643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 32021643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 32033ec2eef1SAl Viro mode); 32041643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 32051643b43fSAl Viro error = create_error; 32061643b43fSAl Viro return error; 32071643b43fSAl Viro } 320854ef4872SMiklos Szeredi 32091643b43fSAl Viro no_open: 32106fbd0714SAl Viro if (d_in_lookup(dentry)) { 321112fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 321212fa5e24SAl Viro nd->flags); 32136fbd0714SAl Viro d_lookup_done(dentry); 321412fa5e24SAl Viro if (unlikely(res)) { 321512fa5e24SAl Viro if (IS_ERR(res)) { 321612fa5e24SAl Viro error = PTR_ERR(res); 321712fa5e24SAl Viro goto out_dput; 321812fa5e24SAl Viro } 321912fa5e24SAl Viro dput(dentry); 322012fa5e24SAl Viro dentry = res; 322112fa5e24SAl Viro } 322254ef4872SMiklos Szeredi } 322354ef4872SMiklos Szeredi 3224d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32251643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 322673a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3227ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3228ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3229ce8644fcSAl Viro error = -EACCES; 3230d58ffd35SMiklos Szeredi goto out_dput; 323164894cf8SAl Viro } 3232ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 32331643b43fSAl Viro open_flag & O_EXCL); 3234d58ffd35SMiklos Szeredi if (error) 3235d58ffd35SMiklos Szeredi goto out_dput; 3236ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3237d58ffd35SMiklos Szeredi } 32381643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32391643b43fSAl Viro error = create_error; 3240d58ffd35SMiklos Szeredi goto out_dput; 3241d58ffd35SMiklos Szeredi } 3242d18e9008SMiklos Szeredi out_no_open: 3243d58ffd35SMiklos Szeredi path->dentry = dentry; 3244d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 324500a07c15SAl Viro return 0; 3246d58ffd35SMiklos Szeredi 3247d58ffd35SMiklos Szeredi out_dput: 3248d58ffd35SMiklos Szeredi dput(dentry); 32492675a4ebSAl Viro return error; 3250d58ffd35SMiklos Szeredi } 3251d58ffd35SMiklos Szeredi 3252d58ffd35SMiklos Szeredi /* 3253fe2d35ffSAl Viro * Handle the last step of open() 325431e6b01fSNick Piggin */ 3255896475d5SAl Viro static int do_last(struct nameidata *nd, 32563ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3257fb1cc555SAl Viro { 3258a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3259ca344a89SAl Viro int open_flag = op->open_flag; 326077d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 326164894cf8SAl Viro bool got_write = false; 3262bcda7652SAl Viro int acc_mode = op->acc_mode; 3263254cf582SAl Viro unsigned seq; 3264a1eb3315SMiklos Szeredi struct inode *inode; 3265896475d5SAl Viro struct path path; 326616c2cd71SAl Viro int error; 3267fb1cc555SAl Viro 3268c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3269c3e380b0SAl Viro nd->flags |= op->intent; 3270c3e380b0SAl Viro 3271bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3272fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3273deb106c6SAl Viro if (unlikely(error)) 32742675a4ebSAl Viro return error; 3275e83db167SMiklos Szeredi goto finish_open; 32761f36f774SAl Viro } 3277a2c36b45SAl Viro 3278ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3279fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3280fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3281fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3282254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3283e9742b53SAl Viro if (likely(error > 0)) 328471574865SMiklos Szeredi goto finish_lookup; 328571574865SMiklos Szeredi 3286ce57dfc1SAl Viro if (error < 0) 3287deb106c6SAl Viro return error; 3288a1eb3315SMiklos Szeredi 328937d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32906583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3291b6183df7SMiklos Szeredi } else { 3292fe2d35ffSAl Viro /* create side of things */ 3293a3fbbde7SAl Viro /* 3294b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3295b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3296b6183df7SMiklos Szeredi * about to look up 3297a3fbbde7SAl Viro */ 32989f1fafeeSAl Viro error = complete_walk(nd); 3299e8bb73dfSAl Viro if (error) 33002675a4ebSAl Viro return error; 3301fe2d35ffSAl Viro 330276ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 33031f36f774SAl Viro /* trailing slashes? */ 3304deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3305deb106c6SAl Viro return -EISDIR; 3306b6183df7SMiklos Szeredi } 33071f36f774SAl Viro 33089cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 330964894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 331064894cf8SAl Viro if (!error) 331164894cf8SAl Viro got_write = true; 331264894cf8SAl Viro /* 331364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 331464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 331564894cf8SAl Viro * dropping this one anyway. 331664894cf8SAl Viro */ 331764894cf8SAl Viro } 33189cf843e3SAl Viro if (open_flag & O_CREAT) 33195955102cSAl Viro inode_lock(dir->d_inode); 33209cf843e3SAl Viro else 33219cf843e3SAl Viro inode_lock_shared(dir->d_inode); 33223ec2eef1SAl Viro error = lookup_open(nd, &path, file, op, got_write); 33239cf843e3SAl Viro if (open_flag & O_CREAT) 33245955102cSAl Viro inode_unlock(dir->d_inode); 33259cf843e3SAl Viro else 33269cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3327fb1cc555SAl Viro 33282675a4ebSAl Viro if (error) 3329d58ffd35SMiklos Szeredi goto out; 33306c0d46c4SAl Viro 333100a07c15SAl Viro if (file->f_mode & FMODE_OPENED) { 333273a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3333496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 333477d660a8SMiklos Szeredi will_truncate = false; 3335d18e9008SMiklos Szeredi 333676ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3337d18e9008SMiklos Szeredi goto opened; 3338d18e9008SMiklos Szeredi } 3339d18e9008SMiklos Szeredi 334073a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 33419b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3342ca344a89SAl Viro open_flag &= ~O_TRUNC; 334377d660a8SMiklos Szeredi will_truncate = false; 334462fb4a15SAl Viro acc_mode = 0; 3345896475d5SAl Viro path_to_nameidata(&path, nd); 3346e83db167SMiklos Szeredi goto finish_open_created; 3347fb1cc555SAl Viro } 3348fb1cc555SAl Viro 3349fb1cc555SAl Viro /* 3350d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3351d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3352d18e9008SMiklos Szeredi * necessary...) 3353d18e9008SMiklos Szeredi */ 335464894cf8SAl Viro if (got_write) { 3355d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 335664894cf8SAl Viro got_write = false; 3357d18e9008SMiklos Szeredi } 3358d18e9008SMiklos Szeredi 3359e6ec03a2SAl Viro error = follow_managed(&path, nd); 3360e6ec03a2SAl Viro if (unlikely(error < 0)) 3361e6ec03a2SAl Viro return error; 3362e6ec03a2SAl Viro 33636583fe22SAl Viro if (unlikely(d_is_negative(path.dentry))) { 33646583fe22SAl Viro path_to_nameidata(&path, nd); 33656583fe22SAl Viro return -ENOENT; 33666583fe22SAl Viro } 33676583fe22SAl Viro 33686583fe22SAl Viro /* 33696583fe22SAl Viro * create/update audit record if it already exists. 33706583fe22SAl Viro */ 33716583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 33726583fe22SAl Viro 3373deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3374deb106c6SAl Viro path_to_nameidata(&path, nd); 3375deb106c6SAl Viro return -EEXIST; 3376deb106c6SAl Viro } 3377fb1cc555SAl Viro 3378254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3379d4565649SAl Viro inode = d_backing_inode(path.dentry); 3380766c4cbfSAl Viro finish_lookup: 33818f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3382deb106c6SAl Viro if (unlikely(error)) 3383d63ff28fSAl Viro return error; 3384bc77daa7SAl Viro finish_open: 33858f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3386a3fbbde7SAl Viro error = complete_walk(nd); 3387fac7d191SAl Viro if (error) 33882675a4ebSAl Viro return error; 338976ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 339030aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3391fb1cc555SAl Viro error = -EISDIR; 339230aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 33932675a4ebSAl Viro goto out; 339430aba665SSalvatore Mesoraca error = may_create_in_sticky(dir, 339530aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 339630aba665SSalvatore Mesoraca if (unlikely(error)) 339730aba665SSalvatore Mesoraca goto out; 339830aba665SSalvatore Mesoraca } 3399af2f5542SMiklos Szeredi error = -ENOTDIR; 340044b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 34012675a4ebSAl Viro goto out; 34024bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 340377d660a8SMiklos Szeredi will_truncate = false; 34046c0d46c4SAl Viro 34050f9d1a10SAl Viro if (will_truncate) { 34060f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 34070f9d1a10SAl Viro if (error) 34082675a4ebSAl Viro goto out; 340964894cf8SAl Viro got_write = true; 34100f9d1a10SAl Viro } 3411e83db167SMiklos Szeredi finish_open_created: 3412bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3413ca344a89SAl Viro if (error) 34142675a4ebSAl Viro goto out; 3415aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3416ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3417fac7d191SAl Viro if (error) 3418015c3bbcSMiklos Szeredi goto out; 3419a8277b9bSMiklos Szeredi opened: 34206035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3421fe9ec829SAl Viro if (!error && will_truncate) 34222675a4ebSAl Viro error = handle_truncate(file); 3423ca344a89SAl Viro out: 3424c80567c8SAl Viro if (unlikely(error > 0)) { 3425c80567c8SAl Viro WARN_ON(1); 3426c80567c8SAl Viro error = -EINVAL; 3427c80567c8SAl Viro } 342864894cf8SAl Viro if (got_write) 34290f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 34302675a4ebSAl Viro return error; 3431fb1cc555SAl Viro } 3432fb1cc555SAl Viro 3433af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3434af7bd4dcSAmir Goldstein { 3435af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3436af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3437af7bd4dcSAmir Goldstein struct inode *inode; 3438af7bd4dcSAmir Goldstein int error; 3439af7bd4dcSAmir Goldstein 3440af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3441af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3442af7bd4dcSAmir Goldstein if (error) 3443af7bd4dcSAmir Goldstein goto out_err; 3444af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3445af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3446af7bd4dcSAmir Goldstein goto out_err; 3447af7bd4dcSAmir Goldstein error = -ENOMEM; 3448cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3449af7bd4dcSAmir Goldstein if (unlikely(!child)) 3450af7bd4dcSAmir Goldstein goto out_err; 3451af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3452af7bd4dcSAmir Goldstein if (error) 3453af7bd4dcSAmir Goldstein goto out_err; 3454af7bd4dcSAmir Goldstein error = -ENOENT; 3455af7bd4dcSAmir Goldstein inode = child->d_inode; 3456af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3457af7bd4dcSAmir Goldstein goto out_err; 3458af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3459af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3460af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3461af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3462af7bd4dcSAmir Goldstein } 3463fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3464af7bd4dcSAmir Goldstein return child; 3465af7bd4dcSAmir Goldstein 3466af7bd4dcSAmir Goldstein out_err: 3467af7bd4dcSAmir Goldstein dput(child); 3468af7bd4dcSAmir Goldstein return ERR_PTR(error); 3469af7bd4dcSAmir Goldstein } 3470af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3471af7bd4dcSAmir Goldstein 3472c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 347360545d0dSAl Viro const struct open_flags *op, 34743ec2eef1SAl Viro struct file *file) 347560545d0dSAl Viro { 3476625b6d10SAl Viro struct dentry *child; 3477625b6d10SAl Viro struct path path; 3478c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 347960545d0dSAl Viro if (unlikely(error)) 348060545d0dSAl Viro return error; 3481625b6d10SAl Viro error = mnt_want_write(path.mnt); 348260545d0dSAl Viro if (unlikely(error)) 348360545d0dSAl Viro goto out; 3484af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3485af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3486684e73beSHirofumi Nakagawa if (IS_ERR(child)) 348760545d0dSAl Viro goto out2; 3488625b6d10SAl Viro dput(path.dentry); 3489625b6d10SAl Viro path.dentry = child; 3490c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 349169a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 349262fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 349360545d0dSAl Viro if (error) 349460545d0dSAl Viro goto out2; 3495625b6d10SAl Viro file->f_path.mnt = path.mnt; 3496be12af3eSAl Viro error = finish_open(file, child, NULL); 349760545d0dSAl Viro out2: 3498625b6d10SAl Viro mnt_drop_write(path.mnt); 349960545d0dSAl Viro out: 3500625b6d10SAl Viro path_put(&path); 350160545d0dSAl Viro return error; 350260545d0dSAl Viro } 350360545d0dSAl Viro 35046ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 35056ac08709SAl Viro { 35066ac08709SAl Viro struct path path; 35076ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 35086ac08709SAl Viro if (!error) { 35096ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3510ae2bb293SAl Viro error = vfs_open(&path, file); 35116ac08709SAl Viro path_put(&path); 35126ac08709SAl Viro } 35136ac08709SAl Viro return error; 35146ac08709SAl Viro } 35156ac08709SAl Viro 3516c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3517c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35181da177e4SLinus Torvalds { 351930d90494SAl Viro struct file *file; 352013aab428SAl Viro int error; 352131e6b01fSNick Piggin 3522ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35231afc99beSAl Viro if (IS_ERR(file)) 35241afc99beSAl Viro return file; 352531e6b01fSNick Piggin 3526bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 35273ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 35285f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 35296ac08709SAl Viro error = do_o_path(nd, flags, file); 35305f336e72SAl Viro } else { 35315f336e72SAl Viro const char *s = path_init(nd, flags); 35323bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 35333ec2eef1SAl Viro (error = do_last(nd, file, op)) > 0) { 353473d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35353bdba28bSAl Viro s = trailing_symlink(nd); 35363bdba28bSAl Viro } 3537deb106c6SAl Viro terminate_walk(nd); 35385f336e72SAl Viro } 35397c1c01ecSAl Viro if (likely(!error)) { 3540aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35417c1c01ecSAl Viro return file; 35427c1c01ecSAl Viro WARN_ON(1); 35437c1c01ecSAl Viro error = -EINVAL; 3544015c3bbcSMiklos Szeredi } 35457c1c01ecSAl Viro fput(file); 35462675a4ebSAl Viro if (error == -EOPENSTALE) { 35472675a4ebSAl Viro if (flags & LOOKUP_RCU) 35482675a4ebSAl Viro error = -ECHILD; 35492675a4ebSAl Viro else 35502675a4ebSAl Viro error = -ESTALE; 35512675a4ebSAl Viro } 35527c1c01ecSAl Viro return ERR_PTR(error); 3553de459215SKirill Korotaev } 35541da177e4SLinus Torvalds 3555669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3556f9652e10SAl Viro const struct open_flags *op) 355713aab428SAl Viro { 35589883d185SAl Viro struct nameidata nd; 3559f9652e10SAl Viro int flags = op->lookup_flags; 356013aab428SAl Viro struct file *filp; 356113aab428SAl Viro 35629883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3563c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 356413aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3565c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 356613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3567c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35689883d185SAl Viro restore_nameidata(); 356913aab428SAl Viro return filp; 357013aab428SAl Viro } 357113aab428SAl Viro 357273d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3573f9652e10SAl Viro const char *name, const struct open_flags *op) 357473d049a4SAl Viro { 35759883d185SAl Viro struct nameidata nd; 357673d049a4SAl Viro struct file *file; 357751689104SPaul Moore struct filename *filename; 3578f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 357973d049a4SAl Viro 358073d049a4SAl Viro nd.root.mnt = mnt; 358173d049a4SAl Viro nd.root.dentry = dentry; 358273d049a4SAl Viro 3583b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 358473d049a4SAl Viro return ERR_PTR(-ELOOP); 358573d049a4SAl Viro 358651689104SPaul Moore filename = getname_kernel(name); 3587a1c83681SViresh Kumar if (IS_ERR(filename)) 358851689104SPaul Moore return ERR_CAST(filename); 358951689104SPaul Moore 35909883d185SAl Viro set_nameidata(&nd, -1, filename); 3591c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 359273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3593c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 359473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3595c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35969883d185SAl Viro restore_nameidata(); 359751689104SPaul Moore putname(filename); 359873d049a4SAl Viro return file; 359973d049a4SAl Viro } 360073d049a4SAl Viro 3601fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 36021ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36031da177e4SLinus Torvalds { 3604c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3605391172c4SAl Viro struct qstr last; 3606391172c4SAl Viro int type; 3607c30dabfeSJan Kara int err2; 36081ac12b4bSJeff Layton int error; 36091ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36101ac12b4bSJeff Layton 36111ac12b4bSJeff Layton /* 36121ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36131ac12b4bSJeff Layton * other flags passed in are ignored! 36141ac12b4bSJeff Layton */ 36151ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36161ac12b4bSJeff Layton 36175c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36185c31b6ceSAl Viro if (IS_ERR(name)) 36195c31b6ceSAl Viro return ERR_CAST(name); 36201da177e4SLinus Torvalds 3621c663e5d8SChristoph Hellwig /* 3622c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3623c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3624c663e5d8SChristoph Hellwig */ 36255c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3626ed75e95dSAl Viro goto out; 3627c663e5d8SChristoph Hellwig 3628c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3629391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3630c663e5d8SChristoph Hellwig /* 3631c663e5d8SChristoph Hellwig * Do the final lookup. 3632c663e5d8SChristoph Hellwig */ 3633391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36345955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3635391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36361da177e4SLinus Torvalds if (IS_ERR(dentry)) 3637a8104a9fSAl Viro goto unlock; 3638c663e5d8SChristoph Hellwig 3639a8104a9fSAl Viro error = -EEXIST; 3640b18825a7SDavid Howells if (d_is_positive(dentry)) 3641a8104a9fSAl Viro goto fail; 3642b18825a7SDavid Howells 3643c663e5d8SChristoph Hellwig /* 3644c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3645c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3646c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3647c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3648c663e5d8SChristoph Hellwig */ 3649391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3650a8104a9fSAl Viro error = -ENOENT; 3651ed75e95dSAl Viro goto fail; 3652e9baf6e5SAl Viro } 3653c30dabfeSJan Kara if (unlikely(err2)) { 3654c30dabfeSJan Kara error = err2; 3655a8104a9fSAl Viro goto fail; 3656c30dabfeSJan Kara } 3657181c37b6SAl Viro putname(name); 3658e9baf6e5SAl Viro return dentry; 36591da177e4SLinus Torvalds fail: 3660a8104a9fSAl Viro dput(dentry); 3661a8104a9fSAl Viro dentry = ERR_PTR(error); 3662a8104a9fSAl Viro unlock: 36635955102cSAl Viro inode_unlock(path->dentry->d_inode); 3664c30dabfeSJan Kara if (!err2) 3665391172c4SAl Viro mnt_drop_write(path->mnt); 3666ed75e95dSAl Viro out: 3667391172c4SAl Viro path_put(path); 3668181c37b6SAl Viro putname(name); 3669ed75e95dSAl Viro return dentry; 3670dae6ad8fSAl Viro } 3671fa14a0b8SAl Viro 3672fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3673fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3674fa14a0b8SAl Viro { 3675181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3676181c37b6SAl Viro path, lookup_flags); 3677fa14a0b8SAl Viro } 3678dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3679dae6ad8fSAl Viro 3680921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3681921a1650SAl Viro { 3682921a1650SAl Viro dput(dentry); 36835955102cSAl Viro inode_unlock(path->dentry->d_inode); 3684a8104a9fSAl Viro mnt_drop_write(path->mnt); 3685921a1650SAl Viro path_put(path); 3686921a1650SAl Viro } 3687921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3688921a1650SAl Viro 3689520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36901ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3691dae6ad8fSAl Viro { 3692181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3693dae6ad8fSAl Viro } 3694dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3695dae6ad8fSAl Viro 36961a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 36971da177e4SLinus Torvalds { 3698a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36991da177e4SLinus Torvalds 37001da177e4SLinus Torvalds if (error) 37011da177e4SLinus Torvalds return error; 37021da177e4SLinus Torvalds 370394f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 37041da177e4SLinus Torvalds return -EPERM; 37051da177e4SLinus Torvalds 3706acfa4380SAl Viro if (!dir->i_op->mknod) 37071da177e4SLinus Torvalds return -EPERM; 37081da177e4SLinus Torvalds 370908ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 371008ce5f16SSerge E. Hallyn if (error) 371108ce5f16SSerge E. Hallyn return error; 371208ce5f16SSerge E. Hallyn 37131da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37141da177e4SLinus Torvalds if (error) 37151da177e4SLinus Torvalds return error; 37161da177e4SLinus Torvalds 37171da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3718a74574aaSStephen Smalley if (!error) 3719f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37201da177e4SLinus Torvalds return error; 37211da177e4SLinus Torvalds } 37224d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37231da177e4SLinus Torvalds 3724f69aac00SAl Viro static int may_mknod(umode_t mode) 3725463c3197SDave Hansen { 3726463c3197SDave Hansen switch (mode & S_IFMT) { 3727463c3197SDave Hansen case S_IFREG: 3728463c3197SDave Hansen case S_IFCHR: 3729463c3197SDave Hansen case S_IFBLK: 3730463c3197SDave Hansen case S_IFIFO: 3731463c3197SDave Hansen case S_IFSOCK: 3732463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3733463c3197SDave Hansen return 0; 3734463c3197SDave Hansen case S_IFDIR: 3735463c3197SDave Hansen return -EPERM; 3736463c3197SDave Hansen default: 3737463c3197SDave Hansen return -EINVAL; 3738463c3197SDave Hansen } 3739463c3197SDave Hansen } 3740463c3197SDave Hansen 374187c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 374287c4e192SDominik Brodowski unsigned int dev) 37431da177e4SLinus Torvalds { 37441da177e4SLinus Torvalds struct dentry *dentry; 3745dae6ad8fSAl Viro struct path path; 3746dae6ad8fSAl Viro int error; 3747972567f1SJeff Layton unsigned int lookup_flags = 0; 37481da177e4SLinus Torvalds 37498e4bfca1SAl Viro error = may_mknod(mode); 37508e4bfca1SAl Viro if (error) 37518e4bfca1SAl Viro return error; 3752972567f1SJeff Layton retry: 3753972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3754dae6ad8fSAl Viro if (IS_ERR(dentry)) 3755dae6ad8fSAl Viro return PTR_ERR(dentry); 37562ad94ae6SAl Viro 3757dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3758ce3b0f8dSAl Viro mode &= ~current_umask(); 3759dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3760be6d3e56SKentaro Takeda if (error) 3761a8104a9fSAl Viro goto out; 37621da177e4SLinus Torvalds switch (mode & S_IFMT) { 37631da177e4SLinus Torvalds case 0: case S_IFREG: 3764312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 376505d1a717SMimi Zohar if (!error) 376605d1a717SMimi Zohar ima_post_path_mknod(dentry); 37671da177e4SLinus Torvalds break; 37681da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3769dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37701da177e4SLinus Torvalds new_decode_dev(dev)); 37711da177e4SLinus Torvalds break; 37721da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3773dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37741da177e4SLinus Torvalds break; 37751da177e4SLinus Torvalds } 3776a8104a9fSAl Viro out: 3777921a1650SAl Viro done_path_create(&path, dentry); 3778972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3779972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3780972567f1SJeff Layton goto retry; 3781972567f1SJeff Layton } 37821da177e4SLinus Torvalds return error; 37831da177e4SLinus Torvalds } 37841da177e4SLinus Torvalds 378587c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 378687c4e192SDominik Brodowski unsigned int, dev) 378787c4e192SDominik Brodowski { 378887c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 378987c4e192SDominik Brodowski } 379087c4e192SDominik Brodowski 37918208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37925590ff0dSUlrich Drepper { 379387c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 37945590ff0dSUlrich Drepper } 37955590ff0dSUlrich Drepper 379618bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37971da177e4SLinus Torvalds { 3798a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37998de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38001da177e4SLinus Torvalds 38011da177e4SLinus Torvalds if (error) 38021da177e4SLinus Torvalds return error; 38031da177e4SLinus Torvalds 3804acfa4380SAl Viro if (!dir->i_op->mkdir) 38051da177e4SLinus Torvalds return -EPERM; 38061da177e4SLinus Torvalds 38071da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38081da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38091da177e4SLinus Torvalds if (error) 38101da177e4SLinus Torvalds return error; 38111da177e4SLinus Torvalds 38128de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38138de52778SAl Viro return -EMLINK; 38148de52778SAl Viro 38151da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3816a74574aaSStephen Smalley if (!error) 3817f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38181da177e4SLinus Torvalds return error; 38191da177e4SLinus Torvalds } 38204d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38211da177e4SLinus Torvalds 38220101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38231da177e4SLinus Torvalds { 38246902d925SDave Hansen struct dentry *dentry; 3825dae6ad8fSAl Viro struct path path; 3826dae6ad8fSAl Viro int error; 3827b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38281da177e4SLinus Torvalds 3829b76d8b82SJeff Layton retry: 3830b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38316902d925SDave Hansen if (IS_ERR(dentry)) 3832dae6ad8fSAl Viro return PTR_ERR(dentry); 38336902d925SDave Hansen 3834dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3835ce3b0f8dSAl Viro mode &= ~current_umask(); 3836dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3837a8104a9fSAl Viro if (!error) 3838dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3839921a1650SAl Viro done_path_create(&path, dentry); 3840b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3841b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3842b76d8b82SJeff Layton goto retry; 3843b76d8b82SJeff Layton } 38441da177e4SLinus Torvalds return error; 38451da177e4SLinus Torvalds } 38461da177e4SLinus Torvalds 38470101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38480101db7aSDominik Brodowski { 38490101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38500101db7aSDominik Brodowski } 38510101db7aSDominik Brodowski 3852a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38535590ff0dSUlrich Drepper { 38540101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38555590ff0dSUlrich Drepper } 38565590ff0dSUlrich Drepper 38571da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38581da177e4SLinus Torvalds { 38591da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38601da177e4SLinus Torvalds 38611da177e4SLinus Torvalds if (error) 38621da177e4SLinus Torvalds return error; 38631da177e4SLinus Torvalds 3864acfa4380SAl Viro if (!dir->i_op->rmdir) 38651da177e4SLinus Torvalds return -EPERM; 38661da177e4SLinus Torvalds 38671d2ef590SAl Viro dget(dentry); 38685955102cSAl Viro inode_lock(dentry->d_inode); 3869912dbc15SSage Weil 38701da177e4SLinus Torvalds error = -EBUSY; 38717af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3872912dbc15SSage Weil goto out; 3873912dbc15SSage Weil 38741da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3875912dbc15SSage Weil if (error) 3876912dbc15SSage Weil goto out; 3877912dbc15SSage Weil 38781da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3879912dbc15SSage Weil if (error) 3880912dbc15SSage Weil goto out; 3881912dbc15SSage Weil 38828767712fSAl Viro shrink_dcache_parent(dentry); 38831da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3884d83c49f3SAl Viro dont_mount(dentry); 38858ed936b5SEric W. Biederman detach_mounts(dentry); 38861da177e4SLinus Torvalds 3887912dbc15SSage Weil out: 38885955102cSAl Viro inode_unlock(dentry->d_inode); 38891d2ef590SAl Viro dput(dentry); 3890912dbc15SSage Weil if (!error) 3891912dbc15SSage Weil d_delete(dentry); 38921da177e4SLinus Torvalds return error; 38931da177e4SLinus Torvalds } 38944d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38951da177e4SLinus Torvalds 3896f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 38971da177e4SLinus Torvalds { 38981da177e4SLinus Torvalds int error = 0; 389991a27b2aSJeff Layton struct filename *name; 39001da177e4SLinus Torvalds struct dentry *dentry; 3901f5beed75SAl Viro struct path path; 3902f5beed75SAl Viro struct qstr last; 3903f5beed75SAl Viro int type; 3904c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3905c6ee9206SJeff Layton retry: 3906c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3907c1d4dd27SAl Viro &path, &last, &type); 390891a27b2aSJeff Layton if (IS_ERR(name)) 390991a27b2aSJeff Layton return PTR_ERR(name); 39101da177e4SLinus Torvalds 3911f5beed75SAl Viro switch (type) { 39121da177e4SLinus Torvalds case LAST_DOTDOT: 39131da177e4SLinus Torvalds error = -ENOTEMPTY; 39141da177e4SLinus Torvalds goto exit1; 39151da177e4SLinus Torvalds case LAST_DOT: 39161da177e4SLinus Torvalds error = -EINVAL; 39171da177e4SLinus Torvalds goto exit1; 39181da177e4SLinus Torvalds case LAST_ROOT: 39191da177e4SLinus Torvalds error = -EBUSY; 39201da177e4SLinus Torvalds goto exit1; 39211da177e4SLinus Torvalds } 39220612d9fbSOGAWA Hirofumi 3923f5beed75SAl Viro error = mnt_want_write(path.mnt); 3924c30dabfeSJan Kara if (error) 3925c30dabfeSJan Kara goto exit1; 39260612d9fbSOGAWA Hirofumi 39275955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3928f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39291da177e4SLinus Torvalds error = PTR_ERR(dentry); 39306902d925SDave Hansen if (IS_ERR(dentry)) 39316902d925SDave Hansen goto exit2; 3932e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3933e6bc45d6STheodore Ts'o error = -ENOENT; 3934e6bc45d6STheodore Ts'o goto exit3; 3935e6bc45d6STheodore Ts'o } 3936f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3937be6d3e56SKentaro Takeda if (error) 3938c30dabfeSJan Kara goto exit3; 3939f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39400622753bSDave Hansen exit3: 39411da177e4SLinus Torvalds dput(dentry); 39426902d925SDave Hansen exit2: 39435955102cSAl Viro inode_unlock(path.dentry->d_inode); 3944f5beed75SAl Viro mnt_drop_write(path.mnt); 39451da177e4SLinus Torvalds exit1: 3946f5beed75SAl Viro path_put(&path); 39471da177e4SLinus Torvalds putname(name); 3948c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3949c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3950c6ee9206SJeff Layton goto retry; 3951c6ee9206SJeff Layton } 39521da177e4SLinus Torvalds return error; 39531da177e4SLinus Torvalds } 39541da177e4SLinus Torvalds 39553cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39565590ff0dSUlrich Drepper { 39575590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39585590ff0dSUlrich Drepper } 39595590ff0dSUlrich Drepper 3960b21996e3SJ. Bruce Fields /** 3961b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3962b21996e3SJ. Bruce Fields * @dir: parent directory 3963b21996e3SJ. Bruce Fields * @dentry: victim 3964b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3965b21996e3SJ. Bruce Fields * 3966b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3967b21996e3SJ. Bruce Fields * 3968b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3969b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3970b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3971b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3972b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3973b21996e3SJ. Bruce Fields * 3974b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3975b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3976b21996e3SJ. Bruce Fields * to be NFS exported. 3977b21996e3SJ. Bruce Fields */ 3978b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39791da177e4SLinus Torvalds { 39809accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39811da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39821da177e4SLinus Torvalds 39831da177e4SLinus Torvalds if (error) 39841da177e4SLinus Torvalds return error; 39851da177e4SLinus Torvalds 3986acfa4380SAl Viro if (!dir->i_op->unlink) 39871da177e4SLinus Torvalds return -EPERM; 39881da177e4SLinus Torvalds 39895955102cSAl Viro inode_lock(target); 39908ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39911da177e4SLinus Torvalds error = -EBUSY; 39921da177e4SLinus Torvalds else { 39931da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3994bec1052eSAl Viro if (!error) { 39955a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39965a14696cSJ. Bruce Fields if (error) 3997b21996e3SJ. Bruce Fields goto out; 39981da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39998ed936b5SEric W. Biederman if (!error) { 4000d83c49f3SAl Viro dont_mount(dentry); 40018ed936b5SEric W. Biederman detach_mounts(dentry); 40028ed936b5SEric W. Biederman } 4003bec1052eSAl Viro } 40041da177e4SLinus Torvalds } 4005b21996e3SJ. Bruce Fields out: 40065955102cSAl Viro inode_unlock(target); 40071da177e4SLinus Torvalds 40081da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40091da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40109accbb97SJ. Bruce Fields fsnotify_link_count(target); 40111da177e4SLinus Torvalds d_delete(dentry); 40121da177e4SLinus Torvalds } 40130eeca283SRobert Love 40141da177e4SLinus Torvalds return error; 40151da177e4SLinus Torvalds } 40164d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40171da177e4SLinus Torvalds 40181da177e4SLinus Torvalds /* 40191da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40201b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40211da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40221da177e4SLinus Torvalds * while waiting on the I/O. 40231da177e4SLinus Torvalds */ 4024da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40251da177e4SLinus Torvalds { 40262ad94ae6SAl Viro int error; 40271da177e4SLinus Torvalds struct dentry *dentry; 4028f5beed75SAl Viro struct path path; 4029f5beed75SAl Viro struct qstr last; 4030f5beed75SAl Viro int type; 40311da177e4SLinus Torvalds struct inode *inode = NULL; 4032b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40335d18f813SJeff Layton unsigned int lookup_flags = 0; 40345d18f813SJeff Layton retry: 4035da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 403691a27b2aSJeff Layton if (IS_ERR(name)) 403791a27b2aSJeff Layton return PTR_ERR(name); 40382ad94ae6SAl Viro 40391da177e4SLinus Torvalds error = -EISDIR; 4040f5beed75SAl Viro if (type != LAST_NORM) 40411da177e4SLinus Torvalds goto exit1; 40420612d9fbSOGAWA Hirofumi 4043f5beed75SAl Viro error = mnt_want_write(path.mnt); 4044c30dabfeSJan Kara if (error) 4045c30dabfeSJan Kara goto exit1; 4046b21996e3SJ. Bruce Fields retry_deleg: 40475955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4048f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40491da177e4SLinus Torvalds error = PTR_ERR(dentry); 40501da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40511da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4052f5beed75SAl Viro if (last.name[last.len]) 405350338b88STörök Edwin goto slashes; 40541da177e4SLinus Torvalds inode = dentry->d_inode; 4055b18825a7SDavid Howells if (d_is_negative(dentry)) 4056e6bc45d6STheodore Ts'o goto slashes; 40577de9c6eeSAl Viro ihold(inode); 4058f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4059be6d3e56SKentaro Takeda if (error) 4060c30dabfeSJan Kara goto exit2; 4061f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40621da177e4SLinus Torvalds exit2: 40631da177e4SLinus Torvalds dput(dentry); 40641da177e4SLinus Torvalds } 40655955102cSAl Viro inode_unlock(path.dentry->d_inode); 40661da177e4SLinus Torvalds if (inode) 40671da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4068b21996e3SJ. Bruce Fields inode = NULL; 4069b21996e3SJ. Bruce Fields if (delegated_inode) { 40705a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4071b21996e3SJ. Bruce Fields if (!error) 4072b21996e3SJ. Bruce Fields goto retry_deleg; 4073b21996e3SJ. Bruce Fields } 4074f5beed75SAl Viro mnt_drop_write(path.mnt); 40751da177e4SLinus Torvalds exit1: 4076f5beed75SAl Viro path_put(&path); 40775d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40785d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40795d18f813SJeff Layton inode = NULL; 40805d18f813SJeff Layton goto retry; 40815d18f813SJeff Layton } 4082da2f1362SChristoph Hellwig putname(name); 40831da177e4SLinus Torvalds return error; 40841da177e4SLinus Torvalds 40851da177e4SLinus Torvalds slashes: 4086b18825a7SDavid Howells if (d_is_negative(dentry)) 4087b18825a7SDavid Howells error = -ENOENT; 408844b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4089b18825a7SDavid Howells error = -EISDIR; 4090b18825a7SDavid Howells else 4091b18825a7SDavid Howells error = -ENOTDIR; 40921da177e4SLinus Torvalds goto exit2; 40931da177e4SLinus Torvalds } 40941da177e4SLinus Torvalds 40952e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40965590ff0dSUlrich Drepper { 40975590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40985590ff0dSUlrich Drepper return -EINVAL; 40995590ff0dSUlrich Drepper 41005590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 41015590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 41025590ff0dSUlrich Drepper 4103da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 41045590ff0dSUlrich Drepper } 41055590ff0dSUlrich Drepper 41063480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41075590ff0dSUlrich Drepper { 4108da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 41095590ff0dSUlrich Drepper } 41105590ff0dSUlrich Drepper 4111db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 41121da177e4SLinus Torvalds { 4113a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 41141da177e4SLinus Torvalds 41151da177e4SLinus Torvalds if (error) 41161da177e4SLinus Torvalds return error; 41171da177e4SLinus Torvalds 4118acfa4380SAl Viro if (!dir->i_op->symlink) 41191da177e4SLinus Torvalds return -EPERM; 41201da177e4SLinus Torvalds 41211da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41221da177e4SLinus Torvalds if (error) 41231da177e4SLinus Torvalds return error; 41241da177e4SLinus Torvalds 41251da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4126a74574aaSStephen Smalley if (!error) 4127f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41281da177e4SLinus Torvalds return error; 41291da177e4SLinus Torvalds } 41304d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41311da177e4SLinus Torvalds 4132b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 4133b724e846SDominik Brodowski const char __user *newname) 41341da177e4SLinus Torvalds { 41352ad94ae6SAl Viro int error; 413691a27b2aSJeff Layton struct filename *from; 41376902d925SDave Hansen struct dentry *dentry; 4138dae6ad8fSAl Viro struct path path; 4139f46d3567SJeff Layton unsigned int lookup_flags = 0; 41401da177e4SLinus Torvalds 41411da177e4SLinus Torvalds from = getname(oldname); 41421da177e4SLinus Torvalds if (IS_ERR(from)) 41431da177e4SLinus Torvalds return PTR_ERR(from); 4144f46d3567SJeff Layton retry: 4145f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41461da177e4SLinus Torvalds error = PTR_ERR(dentry); 41476902d925SDave Hansen if (IS_ERR(dentry)) 4148dae6ad8fSAl Viro goto out_putname; 41496902d925SDave Hansen 415091a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4151a8104a9fSAl Viro if (!error) 415291a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4153921a1650SAl Viro done_path_create(&path, dentry); 4154f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4155f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4156f46d3567SJeff Layton goto retry; 4157f46d3567SJeff Layton } 41586902d925SDave Hansen out_putname: 41591da177e4SLinus Torvalds putname(from); 41601da177e4SLinus Torvalds return error; 41611da177e4SLinus Torvalds } 41621da177e4SLinus Torvalds 4163b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4164b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4165b724e846SDominik Brodowski { 4166b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4167b724e846SDominik Brodowski } 4168b724e846SDominik Brodowski 41693480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41705590ff0dSUlrich Drepper { 4171b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 41725590ff0dSUlrich Drepper } 41735590ff0dSUlrich Drepper 4174146a8595SJ. Bruce Fields /** 4175146a8595SJ. Bruce Fields * vfs_link - create a new link 4176146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4177146a8595SJ. Bruce Fields * @dir: new parent 4178146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4179146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4180146a8595SJ. Bruce Fields * 4181146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4182146a8595SJ. Bruce Fields * 4183146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4184146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4185146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4186146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4187146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4188146a8595SJ. Bruce Fields * 4189146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4190146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4191146a8595SJ. Bruce Fields * to be NFS exported. 4192146a8595SJ. Bruce Fields */ 4193146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41941da177e4SLinus Torvalds { 41951da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41968de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41971da177e4SLinus Torvalds int error; 41981da177e4SLinus Torvalds 41991da177e4SLinus Torvalds if (!inode) 42001da177e4SLinus Torvalds return -ENOENT; 42011da177e4SLinus Torvalds 4202a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 42031da177e4SLinus Torvalds if (error) 42041da177e4SLinus Torvalds return error; 42051da177e4SLinus Torvalds 42061da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 42071da177e4SLinus Torvalds return -EXDEV; 42081da177e4SLinus Torvalds 42091da177e4SLinus Torvalds /* 42101da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 42111da177e4SLinus Torvalds */ 42121da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 42131da177e4SLinus Torvalds return -EPERM; 42140bd23d09SEric W. Biederman /* 42150bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 42160bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 42170bd23d09SEric W. Biederman * the vfs. 42180bd23d09SEric W. Biederman */ 42190bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 42200bd23d09SEric W. Biederman return -EPERM; 4221acfa4380SAl Viro if (!dir->i_op->link) 42221da177e4SLinus Torvalds return -EPERM; 42237e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42241da177e4SLinus Torvalds return -EPERM; 42251da177e4SLinus Torvalds 42261da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42271da177e4SLinus Torvalds if (error) 42281da177e4SLinus Torvalds return error; 42291da177e4SLinus Torvalds 42305955102cSAl Viro inode_lock(inode); 4231aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4232f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4233aae8a97dSAneesh Kumar K.V error = -ENOENT; 42348de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42358de52778SAl Viro error = -EMLINK; 4236146a8595SJ. Bruce Fields else { 4237146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4238146a8595SJ. Bruce Fields if (!error) 42391da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4240146a8595SJ. Bruce Fields } 4241f4e0c30cSAl Viro 4242f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4243f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4244f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4245f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4246f4e0c30cSAl Viro } 42475955102cSAl Viro inode_unlock(inode); 4248e31e14ecSStephen Smalley if (!error) 42497e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42501da177e4SLinus Torvalds return error; 42511da177e4SLinus Torvalds } 42524d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42531da177e4SLinus Torvalds 42541da177e4SLinus Torvalds /* 42551da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42561da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42571da177e4SLinus Torvalds * newname. --KAB 42581da177e4SLinus Torvalds * 42591da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42601da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42611da177e4SLinus Torvalds * and other special files. --ADM 42621da177e4SLinus Torvalds */ 426346ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 426446ea89ebSDominik Brodowski const char __user *newname, int flags) 42651da177e4SLinus Torvalds { 42661da177e4SLinus Torvalds struct dentry *new_dentry; 4267dae6ad8fSAl Viro struct path old_path, new_path; 4268146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 426911a7b371SAneesh Kumar K.V int how = 0; 42701da177e4SLinus Torvalds int error; 42711da177e4SLinus Torvalds 427211a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4273c04030e1SUlrich Drepper return -EINVAL; 427411a7b371SAneesh Kumar K.V /* 4275f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4276f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4277f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 427811a7b371SAneesh Kumar K.V */ 4279f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4280f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4281f0cc6ffbSLinus Torvalds return -ENOENT; 428211a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4283f0cc6ffbSLinus Torvalds } 4284c04030e1SUlrich Drepper 428511a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 428611a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4287442e31caSJeff Layton retry: 428811a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42891da177e4SLinus Torvalds if (error) 42902ad94ae6SAl Viro return error; 42912ad94ae6SAl Viro 4292442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4293442e31caSJeff Layton (how & LOOKUP_REVAL)); 42941da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42956902d925SDave Hansen if (IS_ERR(new_dentry)) 4296dae6ad8fSAl Viro goto out; 4297dae6ad8fSAl Viro 4298dae6ad8fSAl Viro error = -EXDEV; 4299dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4300dae6ad8fSAl Viro goto out_dput; 4301800179c9SKees Cook error = may_linkat(&old_path); 4302800179c9SKees Cook if (unlikely(error)) 4303800179c9SKees Cook goto out_dput; 4304dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4305be6d3e56SKentaro Takeda if (error) 4306a8104a9fSAl Viro goto out_dput; 4307146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 430875c3f29dSDave Hansen out_dput: 4309921a1650SAl Viro done_path_create(&new_path, new_dentry); 4310146a8595SJ. Bruce Fields if (delegated_inode) { 4311146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4312d22e6338SOleg Drokin if (!error) { 4313d22e6338SOleg Drokin path_put(&old_path); 4314146a8595SJ. Bruce Fields goto retry; 4315146a8595SJ. Bruce Fields } 4316d22e6338SOleg Drokin } 4317442e31caSJeff Layton if (retry_estale(error, how)) { 4318d22e6338SOleg Drokin path_put(&old_path); 4319442e31caSJeff Layton how |= LOOKUP_REVAL; 4320442e31caSJeff Layton goto retry; 4321442e31caSJeff Layton } 43221da177e4SLinus Torvalds out: 43232d8f3038SAl Viro path_put(&old_path); 43241da177e4SLinus Torvalds 43251da177e4SLinus Torvalds return error; 43261da177e4SLinus Torvalds } 43271da177e4SLinus Torvalds 432846ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 432946ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 433046ea89ebSDominik Brodowski { 433146ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 433246ea89ebSDominik Brodowski } 433346ea89ebSDominik Brodowski 43343480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43355590ff0dSUlrich Drepper { 433646ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43375590ff0dSUlrich Drepper } 43385590ff0dSUlrich Drepper 4339bc27027aSMiklos Szeredi /** 4340bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4341bc27027aSMiklos Szeredi * @old_dir: parent of source 4342bc27027aSMiklos Szeredi * @old_dentry: source 4343bc27027aSMiklos Szeredi * @new_dir: parent of destination 4344bc27027aSMiklos Szeredi * @new_dentry: destination 4345bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4346520c8b16SMiklos Szeredi * @flags: rename flags 4347bc27027aSMiklos Szeredi * 4348bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4349bc27027aSMiklos Szeredi * 4350bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4351bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4352bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4353bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4354bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4355bc27027aSMiklos Szeredi * so. 4356bc27027aSMiklos Szeredi * 4357bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4358bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4359bc27027aSMiklos Szeredi * to be NFS exported. 4360bc27027aSMiklos Szeredi * 43611da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43621da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43631da177e4SLinus Torvalds * Problems: 43640117d427SMauro Carvalho Chehab * 4365d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43661da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43671da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4368a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43691da177e4SLinus Torvalds * story. 43706cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43716cedba89SJ. Bruce Fields * and source (if it is not a directory). 43721b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43731da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43741da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4375a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43761da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43771da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43781da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4379a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43801da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43811da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 43821da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4383e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 43841b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 43851da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4386c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 43871da177e4SLinus Torvalds * locking]. 43881da177e4SLinus Torvalds */ 43891da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 43908e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4391520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 43921da177e4SLinus Torvalds { 43931da177e4SLinus Torvalds int error; 4394bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4395bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4396bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4397da1ce067SMiklos Szeredi bool new_is_dir = false; 4398da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 439949d31c2fSAl Viro struct name_snapshot old_name; 44001da177e4SLinus Torvalds 44018d3e2936SMiklos Szeredi if (source == target) 44021da177e4SLinus Torvalds return 0; 44031da177e4SLinus Torvalds 44041da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 44051da177e4SLinus Torvalds if (error) 44061da177e4SLinus Torvalds return error; 44071da177e4SLinus Torvalds 4408da1ce067SMiklos Szeredi if (!target) { 4409a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4410da1ce067SMiklos Szeredi } else { 4411da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4412da1ce067SMiklos Szeredi 4413da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44141da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4415da1ce067SMiklos Szeredi else 4416da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4417da1ce067SMiklos Szeredi } 44181da177e4SLinus Torvalds if (error) 44191da177e4SLinus Torvalds return error; 44201da177e4SLinus Torvalds 44212773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 44221da177e4SLinus Torvalds return -EPERM; 44231da177e4SLinus Torvalds 4424bc27027aSMiklos Szeredi /* 4425bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4426bc27027aSMiklos Szeredi * we'll need to flip '..'. 4427bc27027aSMiklos Szeredi */ 4428da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4429da1ce067SMiklos Szeredi if (is_dir) { 4430bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4431bc27027aSMiklos Szeredi if (error) 4432bc27027aSMiklos Szeredi return error; 4433bc27027aSMiklos Szeredi } 4434da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4435da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4436da1ce067SMiklos Szeredi if (error) 4437da1ce067SMiklos Szeredi return error; 4438da1ce067SMiklos Szeredi } 4439da1ce067SMiklos Szeredi } 44400eeca283SRobert Love 44410b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44420b3974ebSMiklos Szeredi flags); 4443bc27027aSMiklos Szeredi if (error) 4444bc27027aSMiklos Szeredi return error; 4445bc27027aSMiklos Szeredi 444649d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4447bc27027aSMiklos Szeredi dget(new_dentry); 4448da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4449bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4450bc27027aSMiklos Szeredi else if (target) 44515955102cSAl Viro inode_lock(target); 4452bc27027aSMiklos Szeredi 4453bc27027aSMiklos Szeredi error = -EBUSY; 44547af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4455bc27027aSMiklos Szeredi goto out; 4456bc27027aSMiklos Szeredi 4457da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4458bc27027aSMiklos Szeredi error = -EMLINK; 4459da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4460bc27027aSMiklos Szeredi goto out; 4461da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4462da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4463da1ce067SMiklos Szeredi goto out; 4464da1ce067SMiklos Szeredi } 4465da1ce067SMiklos Szeredi if (!is_dir) { 4466bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4467bc27027aSMiklos Szeredi if (error) 4468bc27027aSMiklos Szeredi goto out; 4469da1ce067SMiklos Szeredi } 4470da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4471bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4472bc27027aSMiklos Szeredi if (error) 4473bc27027aSMiklos Szeredi goto out; 4474bc27027aSMiklos Szeredi } 4475520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4476520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4477bc27027aSMiklos Szeredi if (error) 4478bc27027aSMiklos Szeredi goto out; 4479bc27027aSMiklos Szeredi 4480da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44818767712fSAl Viro if (is_dir) { 44828767712fSAl Viro shrink_dcache_parent(new_dentry); 4483bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 44848767712fSAl Viro } 4485bc27027aSMiklos Szeredi dont_mount(new_dentry); 44868ed936b5SEric W. Biederman detach_mounts(new_dentry); 4487bc27027aSMiklos Szeredi } 4488da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4489da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4490bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4491da1ce067SMiklos Szeredi else 4492da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4493da1ce067SMiklos Szeredi } 4494bc27027aSMiklos Szeredi out: 4495da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4496bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4497bc27027aSMiklos Szeredi else if (target) 44985955102cSAl Viro inode_unlock(target); 4499bc27027aSMiklos Szeredi dput(new_dentry); 4500da1ce067SMiklos Szeredi if (!error) { 4501f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4502da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4503da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4504f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4505da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4506da1ce067SMiklos Szeredi } 4507da1ce067SMiklos Szeredi } 450849d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 45090eeca283SRobert Love 45101da177e4SLinus Torvalds return error; 45111da177e4SLinus Torvalds } 45124d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 45131da177e4SLinus Torvalds 4514ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4515ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 45161da177e4SLinus Torvalds { 45171da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 45181da177e4SLinus Torvalds struct dentry *trap; 4519f5beed75SAl Viro struct path old_path, new_path; 4520f5beed75SAl Viro struct qstr old_last, new_last; 4521f5beed75SAl Viro int old_type, new_type; 45228e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 452391a27b2aSJeff Layton struct filename *from; 452491a27b2aSJeff Layton struct filename *to; 4525f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4526c6a94284SJeff Layton bool should_retry = false; 45272ad94ae6SAl Viro int error; 4528520c8b16SMiklos Szeredi 45290d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4530da1ce067SMiklos Szeredi return -EINVAL; 4531da1ce067SMiklos Szeredi 45320d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45330d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4534520c8b16SMiklos Szeredi return -EINVAL; 4535520c8b16SMiklos Szeredi 45360d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45370d7a8555SMiklos Szeredi return -EPERM; 45380d7a8555SMiklos Szeredi 4539f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4540f5beed75SAl Viro target_flags = 0; 4541f5beed75SAl Viro 4542c6a94284SJeff Layton retry: 4543c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4544c1d4dd27SAl Viro &old_path, &old_last, &old_type); 454591a27b2aSJeff Layton if (IS_ERR(from)) { 454691a27b2aSJeff Layton error = PTR_ERR(from); 45471da177e4SLinus Torvalds goto exit; 454891a27b2aSJeff Layton } 45491da177e4SLinus Torvalds 4550c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4551c1d4dd27SAl Viro &new_path, &new_last, &new_type); 455291a27b2aSJeff Layton if (IS_ERR(to)) { 455391a27b2aSJeff Layton error = PTR_ERR(to); 45541da177e4SLinus Torvalds goto exit1; 455591a27b2aSJeff Layton } 45561da177e4SLinus Torvalds 45571da177e4SLinus Torvalds error = -EXDEV; 4558f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45591da177e4SLinus Torvalds goto exit2; 45601da177e4SLinus Torvalds 45611da177e4SLinus Torvalds error = -EBUSY; 4562f5beed75SAl Viro if (old_type != LAST_NORM) 45631da177e4SLinus Torvalds goto exit2; 45641da177e4SLinus Torvalds 45650a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45660a7c3937SMiklos Szeredi error = -EEXIST; 4567f5beed75SAl Viro if (new_type != LAST_NORM) 45681da177e4SLinus Torvalds goto exit2; 45691da177e4SLinus Torvalds 4570f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4571c30dabfeSJan Kara if (error) 4572c30dabfeSJan Kara goto exit2; 4573c30dabfeSJan Kara 45748e6d782cSJ. Bruce Fields retry_deleg: 4575f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45761da177e4SLinus Torvalds 4577f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45781da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45791da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45801da177e4SLinus Torvalds goto exit3; 45811da177e4SLinus Torvalds /* source must exist */ 45821da177e4SLinus Torvalds error = -ENOENT; 4583b18825a7SDavid Howells if (d_is_negative(old_dentry)) 45841da177e4SLinus Torvalds goto exit4; 4585f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 45861da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45871da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 45881da177e4SLinus Torvalds goto exit4; 45890a7c3937SMiklos Szeredi error = -EEXIST; 45900a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 45910a7c3937SMiklos Szeredi goto exit5; 4592da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4593da1ce067SMiklos Szeredi error = -ENOENT; 4594da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4595da1ce067SMiklos Szeredi goto exit5; 4596da1ce067SMiklos Szeredi 4597da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4598da1ce067SMiklos Szeredi error = -ENOTDIR; 4599f5beed75SAl Viro if (new_last.name[new_last.len]) 4600da1ce067SMiklos Szeredi goto exit5; 4601da1ce067SMiklos Szeredi } 4602da1ce067SMiklos Szeredi } 46030a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 46040a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 46050a7c3937SMiklos Szeredi error = -ENOTDIR; 4606f5beed75SAl Viro if (old_last.name[old_last.len]) 46070a7c3937SMiklos Szeredi goto exit5; 4608f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 46090a7c3937SMiklos Szeredi goto exit5; 46100a7c3937SMiklos Szeredi } 46110a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 46120a7c3937SMiklos Szeredi error = -EINVAL; 46130a7c3937SMiklos Szeredi if (old_dentry == trap) 46140a7c3937SMiklos Szeredi goto exit5; 46151da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4616da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46171da177e4SLinus Torvalds error = -ENOTEMPTY; 46181da177e4SLinus Torvalds if (new_dentry == trap) 46191da177e4SLinus Torvalds goto exit5; 46201da177e4SLinus Torvalds 4621f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4622f5beed75SAl Viro &new_path, new_dentry, flags); 4623be6d3e56SKentaro Takeda if (error) 4624c30dabfeSJan Kara goto exit5; 4625f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4626f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4627520c8b16SMiklos Szeredi &delegated_inode, flags); 46281da177e4SLinus Torvalds exit5: 46291da177e4SLinus Torvalds dput(new_dentry); 46301da177e4SLinus Torvalds exit4: 46311da177e4SLinus Torvalds dput(old_dentry); 46321da177e4SLinus Torvalds exit3: 4633f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46348e6d782cSJ. Bruce Fields if (delegated_inode) { 46358e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46368e6d782cSJ. Bruce Fields if (!error) 46378e6d782cSJ. Bruce Fields goto retry_deleg; 46388e6d782cSJ. Bruce Fields } 4639f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46401da177e4SLinus Torvalds exit2: 4641c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4642c6a94284SJeff Layton should_retry = true; 4643f5beed75SAl Viro path_put(&new_path); 46442ad94ae6SAl Viro putname(to); 46451da177e4SLinus Torvalds exit1: 4646f5beed75SAl Viro path_put(&old_path); 46471da177e4SLinus Torvalds putname(from); 4648c6a94284SJeff Layton if (should_retry) { 4649c6a94284SJeff Layton should_retry = false; 4650c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4651c6a94284SJeff Layton goto retry; 4652c6a94284SJeff Layton } 46532ad94ae6SAl Viro exit: 46541da177e4SLinus Torvalds return error; 46551da177e4SLinus Torvalds } 46561da177e4SLinus Torvalds 4657ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4658ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4659ee81feb6SDominik Brodowski { 4660ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4661ee81feb6SDominik Brodowski } 4662ee81feb6SDominik Brodowski 4663520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4664520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4665520c8b16SMiklos Szeredi { 4666ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4667520c8b16SMiklos Szeredi } 4668520c8b16SMiklos Szeredi 4669a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 46705590ff0dSUlrich Drepper { 4671ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 46725590ff0dSUlrich Drepper } 46735590ff0dSUlrich Drepper 4674787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4675787fb6bcSMiklos Szeredi { 4676787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4677787fb6bcSMiklos Szeredi if (error) 4678787fb6bcSMiklos Szeredi return error; 4679787fb6bcSMiklos Szeredi 4680787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4681787fb6bcSMiklos Szeredi return -EPERM; 4682787fb6bcSMiklos Szeredi 4683787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4684787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4685787fb6bcSMiklos Szeredi } 4686787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4687787fb6bcSMiklos Szeredi 46885d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 46891da177e4SLinus Torvalds { 46905d826c84SAl Viro int len = PTR_ERR(link); 46911da177e4SLinus Torvalds if (IS_ERR(link)) 46921da177e4SLinus Torvalds goto out; 46931da177e4SLinus Torvalds 46941da177e4SLinus Torvalds len = strlen(link); 46951da177e4SLinus Torvalds if (len > (unsigned) buflen) 46961da177e4SLinus Torvalds len = buflen; 46971da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 46981da177e4SLinus Torvalds len = -EFAULT; 46991da177e4SLinus Torvalds out: 47001da177e4SLinus Torvalds return len; 47011da177e4SLinus Torvalds } 47021da177e4SLinus Torvalds 4703d60874cdSMiklos Szeredi /** 4704fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4705fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4706fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4707fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4708fd4a0edfSMiklos Szeredi * 4709fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4710fd4a0edfSMiklos Szeredi * 4711fd4a0edfSMiklos Szeredi * Does not call security hook. 4712fd4a0edfSMiklos Szeredi */ 4713fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4714fd4a0edfSMiklos Szeredi { 4715fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4716f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4717f2df5da6SAl Viro const char *link; 4718f2df5da6SAl Viro int res; 4719fd4a0edfSMiklos Szeredi 472076fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 472176fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 472276fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 472376fca90eSMiklos Szeredi 472476fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4725fd4a0edfSMiklos Szeredi return -EINVAL; 4726fd4a0edfSMiklos Szeredi 472776fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 472876fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 472976fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 473076fca90eSMiklos Szeredi } 473176fca90eSMiklos Szeredi 4732f2df5da6SAl Viro link = inode->i_link; 4733f2df5da6SAl Viro if (!link) { 4734f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4735f2df5da6SAl Viro if (IS_ERR(link)) 4736f2df5da6SAl Viro return PTR_ERR(link); 4737f2df5da6SAl Viro } 4738f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4739f2df5da6SAl Viro do_delayed_call(&done); 4740f2df5da6SAl Viro return res; 4741fd4a0edfSMiklos Szeredi } 4742fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 47431da177e4SLinus Torvalds 4744d60874cdSMiklos Szeredi /** 4745d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4746d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4747d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4748d60874cdSMiklos Szeredi * 4749d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4750d60874cdSMiklos Szeredi * 4751d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4752d60874cdSMiklos Szeredi * 4753d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4754d60874cdSMiklos Szeredi */ 4755d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4756d60874cdSMiklos Szeredi { 4757d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4758d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4759d60874cdSMiklos Szeredi 4760d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4761d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4762d60874cdSMiklos Szeredi if (!res) 4763d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4764d60874cdSMiklos Szeredi } 4765d60874cdSMiklos Szeredi return res; 4766d60874cdSMiklos Szeredi } 4767d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4768d60874cdSMiklos Szeredi 47691da177e4SLinus Torvalds /* get the link contents into pagecache */ 47706b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4771fceef393SAl Viro struct delayed_call *callback) 47721da177e4SLinus Torvalds { 4773ebd09abbSDuane Griffin char *kaddr; 47741da177e4SLinus Torvalds struct page *page; 47756b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47766b255391SAl Viro 4777d3883d4fSAl Viro if (!dentry) { 4778d3883d4fSAl Viro page = find_get_page(mapping, 0); 4779d3883d4fSAl Viro if (!page) 47806b255391SAl Viro return ERR_PTR(-ECHILD); 4781d3883d4fSAl Viro if (!PageUptodate(page)) { 4782d3883d4fSAl Viro put_page(page); 4783d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4784d3883d4fSAl Viro } 4785d3883d4fSAl Viro } else { 4786090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47871da177e4SLinus Torvalds if (IS_ERR(page)) 47886fe6900eSNick Piggin return (char*)page; 4789d3883d4fSAl Viro } 4790fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 479121fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 479221fc61c7SAl Viro kaddr = page_address(page); 47936b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4794ebd09abbSDuane Griffin return kaddr; 47951da177e4SLinus Torvalds } 47961da177e4SLinus Torvalds 47976b255391SAl Viro EXPORT_SYMBOL(page_get_link); 47981da177e4SLinus Torvalds 4799fceef393SAl Viro void page_put_link(void *arg) 48001da177e4SLinus Torvalds { 4801fceef393SAl Viro put_page(arg); 48021da177e4SLinus Torvalds } 48034d359507SAl Viro EXPORT_SYMBOL(page_put_link); 48041da177e4SLinus Torvalds 4805aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4806aa80deabSAl Viro { 4807fceef393SAl Viro DEFINE_DELAYED_CALL(done); 48086b255391SAl Viro int res = readlink_copy(buffer, buflen, 48096b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4810fceef393SAl Viro &done)); 4811fceef393SAl Viro do_delayed_call(&done); 4812aa80deabSAl Viro return res; 4813aa80deabSAl Viro } 4814aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4815aa80deabSAl Viro 481654566b2cSNick Piggin /* 481754566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 481854566b2cSNick Piggin */ 481954566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48201da177e4SLinus Torvalds { 48211da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48220adb25d2SKirill Korotaev struct page *page; 4823afddba49SNick Piggin void *fsdata; 4824beb497abSDmitriy Monakhov int err; 4825c718a975STetsuo Handa unsigned int flags = 0; 482654566b2cSNick Piggin if (nofs) 482754566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 48281da177e4SLinus Torvalds 48297e53cac4SNeilBrown retry: 4830afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 483154566b2cSNick Piggin flags, &page, &fsdata); 48321da177e4SLinus Torvalds if (err) 4833afddba49SNick Piggin goto fail; 4834afddba49SNick Piggin 483521fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4836afddba49SNick Piggin 4837afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4838afddba49SNick Piggin page, fsdata); 48391da177e4SLinus Torvalds if (err < 0) 48401da177e4SLinus Torvalds goto fail; 4841afddba49SNick Piggin if (err < len-1) 4842afddba49SNick Piggin goto retry; 4843afddba49SNick Piggin 48441da177e4SLinus Torvalds mark_inode_dirty(inode); 48451da177e4SLinus Torvalds return 0; 48461da177e4SLinus Torvalds fail: 48471da177e4SLinus Torvalds return err; 48481da177e4SLinus Torvalds } 48494d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 48501da177e4SLinus Torvalds 48510adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 48520adb25d2SKirill Korotaev { 48530adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4854c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48550adb25d2SKirill Korotaev } 48564d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48570adb25d2SKirill Korotaev 485892e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48596b255391SAl Viro .get_link = page_get_link, 48601da177e4SLinus Torvalds }; 48611da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4862