1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namei.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * Some corrections by tytso. 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 131da177e4SLinus Torvalds * lookup logic. 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 161da177e4SLinus Torvalds */ 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/init.h> 19630d9c47SPaul Gortmaker #include <linux/export.h> 2044696908SDavid S. Miller #include <linux/kernel.h> 211da177e4SLinus Torvalds #include <linux/slab.h> 221da177e4SLinus Torvalds #include <linux/fs.h> 231da177e4SLinus Torvalds #include <linux/namei.h> 241da177e4SLinus Torvalds #include <linux/pagemap.h> 250eeca283SRobert Love #include <linux/fsnotify.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 286146f0d5SMimi Zohar #include <linux/ima.h> 291da177e4SLinus Torvalds #include <linux/syscalls.h> 301da177e4SLinus Torvalds #include <linux/mount.h> 311da177e4SLinus Torvalds #include <linux/audit.h> 3216f7e0feSRandy Dunlap #include <linux/capability.h> 33834f2a4aSTrond Myklebust #include <linux/file.h> 345590ff0dSUlrich Drepper #include <linux/fcntl.h> 3508ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 365ad4e53bSAl Viro #include <linux/fs_struct.h> 37e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3899d263d4SLinus Torvalds #include <linux/hash.h> 392a18da7aSGeorge Spelvin #include <linux/bitops.h> 40aeaa4a79SEric W. Biederman #include <linux/init_task.h> 417c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 421da177e4SLinus Torvalds 43e81e3f4dSEric Paris #include "internal.h" 44c7105365SAl Viro #include "mount.h" 45e81e3f4dSEric Paris 461da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 471da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 481da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 491da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 501da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 511da177e4SLinus Torvalds * 521da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 531da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 541da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 551da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 561da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 571da177e4SLinus Torvalds * the special cases of the former code. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 601da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 611da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 641da177e4SLinus Torvalds * resolution to correspond with current state of the code. 651da177e4SLinus Torvalds * 661da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 671da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 681da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 691da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 701da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 711da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 721da177e4SLinus Torvalds */ 731da177e4SLinus Torvalds 741da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 751da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 761da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 771da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 781da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 791da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8025985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 811da177e4SLinus Torvalds * 821da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 831da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 841da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 851da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 861da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 871da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 881da177e4SLinus Torvalds * and in the old Linux semantics. 891da177e4SLinus Torvalds */ 901da177e4SLinus Torvalds 911da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 921da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 931da177e4SLinus Torvalds * 941da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 951da177e4SLinus Torvalds */ 961da177e4SLinus Torvalds 971da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 981da177e4SLinus Torvalds * inside the path - always follow. 991da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1001da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1011da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1021da177e4SLinus Torvalds * otherwise - don't follow. 1031da177e4SLinus Torvalds * (applied in that order). 1041da177e4SLinus Torvalds * 1051da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1061da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1071da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1081da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1091da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1101da177e4SLinus Torvalds */ 1111da177e4SLinus Torvalds /* 1121da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 113a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1141da177e4SLinus Torvalds * any extra contention... 1151da177e4SLinus Torvalds */ 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1181da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1191da177e4SLinus Torvalds * kernel data space before using them.. 1201da177e4SLinus Torvalds * 1211da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1221da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1231da177e4SLinus Torvalds */ 1247950e385SJeff Layton 125fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12691a27b2aSJeff Layton 12751f39a1fSDavid Drysdale struct filename * 12891a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12991a27b2aSJeff Layton { 13094b5d262SAl Viro struct filename *result; 1317950e385SJeff Layton char *kname; 13294b5d262SAl Viro int len; 1331da177e4SLinus Torvalds 1347ac86265SJeff Layton result = audit_reusename(filename); 1357ac86265SJeff Layton if (result) 1367ac86265SJeff Layton return result; 1377ac86265SJeff Layton 1387950e385SJeff Layton result = __getname(); 1393f9f0aa6SLinus Torvalds if (unlikely(!result)) 1404043cde8SEric Paris return ERR_PTR(-ENOMEM); 1411da177e4SLinus Torvalds 1427950e385SJeff Layton /* 1437950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1447950e385SJeff Layton * allocation 1457950e385SJeff Layton */ 146fd2f7cb5SAl Viro kname = (char *)result->iname; 14791a27b2aSJeff Layton result->name = kname; 1487950e385SJeff Layton 14994b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15091a27b2aSJeff Layton if (unlikely(len < 0)) { 15194b5d262SAl Viro __putname(result); 15294b5d262SAl Viro return ERR_PTR(len); 15391a27b2aSJeff Layton } 1543f9f0aa6SLinus Torvalds 1557950e385SJeff Layton /* 1567950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1577950e385SJeff Layton * separate struct filename so we can dedicate the entire 1587950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1597950e385SJeff Layton * userland. 1607950e385SJeff Layton */ 16194b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 162fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1637950e385SJeff Layton kname = (char *)result; 1647950e385SJeff Layton 165fd2f7cb5SAl Viro /* 166fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 167fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 168fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 169fd2f7cb5SAl Viro */ 170fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17194b5d262SAl Viro if (unlikely(!result)) { 17294b5d262SAl Viro __putname(kname); 17394b5d262SAl Viro return ERR_PTR(-ENOMEM); 1747950e385SJeff Layton } 1757950e385SJeff Layton result->name = kname; 17694b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17794b5d262SAl Viro if (unlikely(len < 0)) { 17894b5d262SAl Viro __putname(kname); 17994b5d262SAl Viro kfree(result); 18094b5d262SAl Viro return ERR_PTR(len); 18194b5d262SAl Viro } 18294b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18394b5d262SAl Viro __putname(kname); 18494b5d262SAl Viro kfree(result); 18594b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18694b5d262SAl Viro } 1877950e385SJeff Layton } 1887950e385SJeff Layton 18994b5d262SAl Viro result->refcnt = 1; 1903f9f0aa6SLinus Torvalds /* The empty path is special. */ 1913f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1923f9f0aa6SLinus Torvalds if (empty) 1931fa1e7f6SAndy Whitcroft *empty = 1; 19494b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19594b5d262SAl Viro putname(result); 19694b5d262SAl Viro return ERR_PTR(-ENOENT); 1971da177e4SLinus Torvalds } 19894b5d262SAl Viro } 1997950e385SJeff Layton 2007950e385SJeff Layton result->uptr = filename; 201c4ad8f98SLinus Torvalds result->aname = NULL; 2021da177e4SLinus Torvalds audit_getname(result); 2031da177e4SLinus Torvalds return result; 2043f9f0aa6SLinus Torvalds } 2053f9f0aa6SLinus Torvalds 20691a27b2aSJeff Layton struct filename * 20791a27b2aSJeff Layton getname(const char __user * filename) 208f52e0c11SAl Viro { 209f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 210f52e0c11SAl Viro } 211f52e0c11SAl Viro 212c4ad8f98SLinus Torvalds struct filename * 213c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 214c4ad8f98SLinus Torvalds { 215c4ad8f98SLinus Torvalds struct filename *result; 21608518549SPaul Moore int len = strlen(filename) + 1; 217c4ad8f98SLinus Torvalds 218c4ad8f98SLinus Torvalds result = __getname(); 219c4ad8f98SLinus Torvalds if (unlikely(!result)) 220c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 221c4ad8f98SLinus Torvalds 22208518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 223fd2f7cb5SAl Viro result->name = (char *)result->iname; 22408518549SPaul Moore } else if (len <= PATH_MAX) { 22530ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 22608518549SPaul Moore struct filename *tmp; 22708518549SPaul Moore 22830ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 22908518549SPaul Moore if (unlikely(!tmp)) { 23008518549SPaul Moore __putname(result); 23108518549SPaul Moore return ERR_PTR(-ENOMEM); 23208518549SPaul Moore } 23308518549SPaul Moore tmp->name = (char *)result; 23408518549SPaul Moore result = tmp; 23508518549SPaul Moore } else { 23608518549SPaul Moore __putname(result); 23708518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23808518549SPaul Moore } 23908518549SPaul Moore memcpy((char *)result->name, filename, len); 240c4ad8f98SLinus Torvalds result->uptr = NULL; 241c4ad8f98SLinus Torvalds result->aname = NULL; 24255422d0bSPaul Moore result->refcnt = 1; 243fd3522fdSPaul Moore audit_getname(result); 244c4ad8f98SLinus Torvalds 245c4ad8f98SLinus Torvalds return result; 246c4ad8f98SLinus Torvalds } 247c4ad8f98SLinus Torvalds 24891a27b2aSJeff Layton void putname(struct filename *name) 2491da177e4SLinus Torvalds { 25055422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 25155422d0bSPaul Moore 25255422d0bSPaul Moore if (--name->refcnt > 0) 25355422d0bSPaul Moore return; 25455422d0bSPaul Moore 255fd2f7cb5SAl Viro if (name->name != name->iname) { 25655422d0bSPaul Moore __putname(name->name); 25755422d0bSPaul Moore kfree(name); 25855422d0bSPaul Moore } else 25955422d0bSPaul Moore __putname(name); 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds 262e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 263e77819e5SLinus Torvalds { 26484635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 265e77819e5SLinus Torvalds struct posix_acl *acl; 266e77819e5SLinus Torvalds 267e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2683567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2693567866bSAl Viro if (!acl) 270e77819e5SLinus Torvalds return -EAGAIN; 2713567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 272b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 273e77819e5SLinus Torvalds return -ECHILD; 274206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 275e77819e5SLinus Torvalds } 276e77819e5SLinus Torvalds 2772982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2784e34e719SChristoph Hellwig if (IS_ERR(acl)) 2794e34e719SChristoph Hellwig return PTR_ERR(acl); 280e77819e5SLinus Torvalds if (acl) { 281e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 282e77819e5SLinus Torvalds posix_acl_release(acl); 283e77819e5SLinus Torvalds return error; 284e77819e5SLinus Torvalds } 28584635d68SLinus Torvalds #endif 286e77819e5SLinus Torvalds 287e77819e5SLinus Torvalds return -EAGAIN; 288e77819e5SLinus Torvalds } 289e77819e5SLinus Torvalds 2905909ccaaSLinus Torvalds /* 291948409c7SAndreas Gruenbacher * This does the basic permission checking 2925909ccaaSLinus Torvalds */ 2937e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2945909ccaaSLinus Torvalds { 29526cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2965909ccaaSLinus Torvalds 2978e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2985909ccaaSLinus Torvalds mode >>= 6; 2995909ccaaSLinus Torvalds else { 300e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 3017e40145eSAl Viro int error = check_acl(inode, mask); 3025909ccaaSLinus Torvalds if (error != -EAGAIN) 3035909ccaaSLinus Torvalds return error; 3045909ccaaSLinus Torvalds } 3055909ccaaSLinus Torvalds 3065909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3075909ccaaSLinus Torvalds mode >>= 3; 3085909ccaaSLinus Torvalds } 3095909ccaaSLinus Torvalds 3105909ccaaSLinus Torvalds /* 3115909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3125909ccaaSLinus Torvalds */ 3139c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3145909ccaaSLinus Torvalds return 0; 3155909ccaaSLinus Torvalds return -EACCES; 3165909ccaaSLinus Torvalds } 3171da177e4SLinus Torvalds 3181da177e4SLinus Torvalds /** 3191da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3201da177e4SLinus Torvalds * @inode: inode to check access rights for 3218fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3221da177e4SLinus Torvalds * 3231da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3241da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3251da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 326b74c79e9SNick Piggin * are used for other things. 327b74c79e9SNick Piggin * 328b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 329b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 330b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3311da177e4SLinus Torvalds */ 3322830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3331da177e4SLinus Torvalds { 3345909ccaaSLinus Torvalds int ret; 3351da177e4SLinus Torvalds 3361da177e4SLinus Torvalds /* 337948409c7SAndreas Gruenbacher * Do the basic permission checks. 3381da177e4SLinus Torvalds */ 3397e40145eSAl Viro ret = acl_permission_check(inode, mask); 3405909ccaaSLinus Torvalds if (ret != -EACCES) 3415909ccaaSLinus Torvalds return ret; 3421da177e4SLinus Torvalds 343d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 344d594e7ecSAl Viro /* DACs are overridable for directories */ 345d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34623adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34723adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 348d594e7ecSAl Viro return 0; 34923adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3501da177e4SLinus Torvalds return 0; 3512a4c2242SStephen Smalley return -EACCES; 3522a4c2242SStephen Smalley } 3531da177e4SLinus Torvalds 3541da177e4SLinus Torvalds /* 3551da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3561da177e4SLinus Torvalds */ 3577ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 358d594e7ecSAl Viro if (mask == MAY_READ) 35923adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3601da177e4SLinus Torvalds return 0; 3612a4c2242SStephen Smalley /* 3622a4c2242SStephen Smalley * Read/write DACs are always overridable. 3632a4c2242SStephen Smalley * Executable DACs are overridable when there is 3642a4c2242SStephen Smalley * at least one exec bit set. 3652a4c2242SStephen Smalley */ 3662a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 3672a4c2242SStephen Smalley if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3682a4c2242SStephen Smalley return 0; 3691da177e4SLinus Torvalds 3701da177e4SLinus Torvalds return -EACCES; 3711da177e4SLinus Torvalds } 3724d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3731da177e4SLinus Torvalds 3743ddcd056SLinus Torvalds /* 3753ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3763ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3773ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3783ddcd056SLinus Torvalds * permission function, use the fast case". 3793ddcd056SLinus Torvalds */ 3803ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3813ddcd056SLinus Torvalds { 3823ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3833ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3843ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3853ddcd056SLinus Torvalds 3863ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3873ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3883ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3893ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3903ddcd056SLinus Torvalds } 3913ddcd056SLinus Torvalds return generic_permission(inode, mask); 3923ddcd056SLinus Torvalds } 3933ddcd056SLinus Torvalds 394cb23beb5SChristoph Hellwig /** 3950bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 3960bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 39755852635SRandy Dunlap * @inode: Inode to check permission on 3980bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3990bdaea90SDavid Howells * 4000bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4010bdaea90SDavid Howells */ 4020bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4030bdaea90SDavid Howells { 4040bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4050bdaea90SDavid Howells umode_t mode = inode->i_mode; 4060bdaea90SDavid Howells 4070bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 408bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4090bdaea90SDavid Howells return -EROFS; 4100bdaea90SDavid Howells } 4110bdaea90SDavid Howells return 0; 4120bdaea90SDavid Howells } 4130bdaea90SDavid Howells 4140bdaea90SDavid Howells /** 4150bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4160bdaea90SDavid Howells * @inode: Inode to check permission on 4170bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4180bdaea90SDavid Howells * 4190bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4200bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4210bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4220bdaea90SDavid Howells * 4230bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4240bdaea90SDavid Howells */ 4250bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4260bdaea90SDavid Howells { 4270bdaea90SDavid Howells int retval; 4280bdaea90SDavid Howells 4290bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4300bdaea90SDavid Howells if (retval) 4310bdaea90SDavid Howells return retval; 4324bfd054aSEric Biggers 4334bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 4344bfd054aSEric Biggers /* 4354bfd054aSEric Biggers * Nobody gets write access to an immutable file. 4364bfd054aSEric Biggers */ 4374bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 4384bfd054aSEric Biggers return -EPERM; 4394bfd054aSEric Biggers 4404bfd054aSEric Biggers /* 4414bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 4424bfd054aSEric Biggers * written back improperly if their true value is unknown 4434bfd054aSEric Biggers * to the vfs. 4444bfd054aSEric Biggers */ 4454bfd054aSEric Biggers if (HAS_UNMAPPED_ID(inode)) 4464bfd054aSEric Biggers return -EACCES; 4474bfd054aSEric Biggers } 4484bfd054aSEric Biggers 4494bfd054aSEric Biggers retval = do_inode_permission(inode, mask); 4504bfd054aSEric Biggers if (retval) 4514bfd054aSEric Biggers return retval; 4524bfd054aSEric Biggers 4534bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 4544bfd054aSEric Biggers if (retval) 4554bfd054aSEric Biggers return retval; 4564bfd054aSEric Biggers 4574bfd054aSEric Biggers return security_inode_permission(inode, mask); 4580bdaea90SDavid Howells } 4594d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4600bdaea90SDavid Howells 4610bdaea90SDavid Howells /** 4625dd784d0SJan Blunck * path_get - get a reference to a path 4635dd784d0SJan Blunck * @path: path to get the reference to 4645dd784d0SJan Blunck * 4655dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4665dd784d0SJan Blunck */ 467dcf787f3SAl Viro void path_get(const struct path *path) 4685dd784d0SJan Blunck { 4695dd784d0SJan Blunck mntget(path->mnt); 4705dd784d0SJan Blunck dget(path->dentry); 4715dd784d0SJan Blunck } 4725dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4735dd784d0SJan Blunck 4745dd784d0SJan Blunck /** 4751d957f9bSJan Blunck * path_put - put a reference to a path 4761d957f9bSJan Blunck * @path: path to put the reference to 4771d957f9bSJan Blunck * 4781d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4791d957f9bSJan Blunck */ 480dcf787f3SAl Viro void path_put(const struct path *path) 4811da177e4SLinus Torvalds { 4821d957f9bSJan Blunck dput(path->dentry); 4831d957f9bSJan Blunck mntput(path->mnt); 4841da177e4SLinus Torvalds } 4851d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4861da177e4SLinus Torvalds 487894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4881f55a6ecSAl Viro struct nameidata { 4891f55a6ecSAl Viro struct path path; 4901f55a6ecSAl Viro struct qstr last; 4911f55a6ecSAl Viro struct path root; 4921f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 4931f55a6ecSAl Viro unsigned int flags; 4949883d185SAl Viro unsigned seq, m_seq; 4951f55a6ecSAl Viro int last_type; 4961f55a6ecSAl Viro unsigned depth; 497756daf26SNeilBrown int total_link_count; 498697fc6caSAl Viro struct saved { 499697fc6caSAl Viro struct path link; 500fceef393SAl Viro struct delayed_call done; 501697fc6caSAl Viro const char *name; 5020450b2d1SAl Viro unsigned seq; 503894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5049883d185SAl Viro struct filename *name; 5059883d185SAl Viro struct nameidata *saved; 506fceef393SAl Viro struct inode *link_inode; 5079883d185SAl Viro unsigned root_seq; 5089883d185SAl Viro int dfd; 5093859a271SKees Cook } __randomize_layout; 5101f55a6ecSAl Viro 5119883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 512894bc8c4SAl Viro { 513756daf26SNeilBrown struct nameidata *old = current->nameidata; 514756daf26SNeilBrown p->stack = p->internal; 515c8a53ee5SAl Viro p->dfd = dfd; 516c8a53ee5SAl Viro p->name = name; 517756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5189883d185SAl Viro p->saved = old; 519756daf26SNeilBrown current->nameidata = p; 520894bc8c4SAl Viro } 521894bc8c4SAl Viro 5229883d185SAl Viro static void restore_nameidata(void) 523894bc8c4SAl Viro { 5249883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 525756daf26SNeilBrown 526756daf26SNeilBrown current->nameidata = old; 527756daf26SNeilBrown if (old) 528756daf26SNeilBrown old->total_link_count = now->total_link_count; 529e1a63bbcSAl Viro if (now->stack != now->internal) 530756daf26SNeilBrown kfree(now->stack); 531894bc8c4SAl Viro } 532894bc8c4SAl Viro 533894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 534894bc8c4SAl Viro { 535bc40aee0SAl Viro struct saved *p; 536bc40aee0SAl Viro 537bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 5386da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 539bc40aee0SAl Viro GFP_ATOMIC); 540bc40aee0SAl Viro if (unlikely(!p)) 541bc40aee0SAl Viro return -ECHILD; 542bc40aee0SAl Viro } else { 5436da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 544894bc8c4SAl Viro GFP_KERNEL); 545894bc8c4SAl Viro if (unlikely(!p)) 546894bc8c4SAl Viro return -ENOMEM; 547bc40aee0SAl Viro } 548894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 549894bc8c4SAl Viro nd->stack = p; 550894bc8c4SAl Viro return 0; 551894bc8c4SAl Viro } 552894bc8c4SAl Viro 553397d425dSEric W. Biederman /** 554397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 555397d425dSEric W. Biederman * @path: nameidate to verify 556397d425dSEric W. Biederman * 557397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 558397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 559397d425dSEric W. Biederman */ 560397d425dSEric W. Biederman static bool path_connected(const struct path *path) 561397d425dSEric W. Biederman { 562397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 56395dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 564397d425dSEric W. Biederman 56595dd7758SEric W. Biederman /* Bind mounts and multi-root filesystems can have disconnected paths */ 56695dd7758SEric W. Biederman if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root)) 567397d425dSEric W. Biederman return true; 568397d425dSEric W. Biederman 569397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 570397d425dSEric W. Biederman } 571397d425dSEric W. Biederman 572894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 573894bc8c4SAl Viro { 574da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 575894bc8c4SAl Viro return 0; 576894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 577894bc8c4SAl Viro return 0; 578894bc8c4SAl Viro return __nd_alloc_stack(nd); 579894bc8c4SAl Viro } 580894bc8c4SAl Viro 5817973387aSAl Viro static void drop_links(struct nameidata *nd) 5827973387aSAl Viro { 5837973387aSAl Viro int i = nd->depth; 5847973387aSAl Viro while (i--) { 5857973387aSAl Viro struct saved *last = nd->stack + i; 586fceef393SAl Viro do_delayed_call(&last->done); 587fceef393SAl Viro clear_delayed_call(&last->done); 5887973387aSAl Viro } 5897973387aSAl Viro } 5907973387aSAl Viro 5917973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5927973387aSAl Viro { 5937973387aSAl Viro drop_links(nd); 5947973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5957973387aSAl Viro int i; 5967973387aSAl Viro path_put(&nd->path); 5977973387aSAl Viro for (i = 0; i < nd->depth; i++) 5987973387aSAl Viro path_put(&nd->stack[i].link); 59984a2bd39SAl Viro if (nd->flags & LOOKUP_ROOT_GRABBED) { 600102b8af2SAl Viro path_put(&nd->root); 60184a2bd39SAl Viro nd->flags &= ~LOOKUP_ROOT_GRABBED; 602102b8af2SAl Viro } 6037973387aSAl Viro } else { 6047973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6057973387aSAl Viro rcu_read_unlock(); 6067973387aSAl Viro } 6077973387aSAl Viro nd->depth = 0; 6087973387aSAl Viro } 6097973387aSAl Viro 6107973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6117973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6127973387aSAl Viro struct path *path, unsigned seq) 6137973387aSAl Viro { 6147973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6157973387aSAl Viro if (unlikely(res)) { 6167973387aSAl Viro if (res > 0) 6177973387aSAl Viro path->mnt = NULL; 6187973387aSAl Viro path->dentry = NULL; 6197973387aSAl Viro return false; 6207973387aSAl Viro } 6217973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6227973387aSAl Viro path->dentry = NULL; 6237973387aSAl Viro return false; 6247973387aSAl Viro } 6257973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6267973387aSAl Viro } 6277973387aSAl Viro 6287973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6297973387aSAl Viro { 6307973387aSAl Viro int i; 6317973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6327973387aSAl Viro struct saved *last = nd->stack + i; 6337973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6347973387aSAl Viro drop_links(nd); 6357973387aSAl Viro nd->depth = i + 1; 6367973387aSAl Viro return false; 6377973387aSAl Viro } 6387973387aSAl Viro } 6397973387aSAl Viro return true; 6407973387aSAl Viro } 6417973387aSAl Viro 642ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 643ee594bffSAl Viro { 644ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 645ee594bffSAl Viro return true; 64684a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 647ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 648ee594bffSAl Viro } 649ee594bffSAl Viro 65019660af7SAl Viro /* 65131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 65319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 65457e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 65519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 65619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 65719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 65819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 65931e6b01fSNick Piggin */ 66031e6b01fSNick Piggin 66131e6b01fSNick Piggin /** 66219660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 66319660af7SAl Viro * @nd: nameidata pathwalk data 66439191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 66531e6b01fSNick Piggin * 6664675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6674675ac39SAl Viro * for ref-walk mode. 6684675ac39SAl Viro * Must be called from rcu-walk context. 6697973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6707973387aSAl Viro * terminate_walk(). 67131e6b01fSNick Piggin */ 6724675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 67331e6b01fSNick Piggin { 67431e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 67531e6b01fSNick Piggin 67631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 677e5c832d5SLinus Torvalds 678e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6797973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6804675ac39SAl Viro goto out1; 68184a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 68284a2bd39SAl Viro goto out; 683ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 6844675ac39SAl Viro goto out; 6854675ac39SAl Viro rcu_read_unlock(); 6864675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6874675ac39SAl Viro return 0; 6884675ac39SAl Viro 68984a2bd39SAl Viro out1: 6904675ac39SAl Viro nd->path.mnt = NULL; 6914675ac39SAl Viro nd->path.dentry = NULL; 6924675ac39SAl Viro out: 6934675ac39SAl Viro rcu_read_unlock(); 6944675ac39SAl Viro return -ECHILD; 6954675ac39SAl Viro } 6964675ac39SAl Viro 6974675ac39SAl Viro /** 6984675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 6994675ac39SAl Viro * @nd: nameidata pathwalk data 7004675ac39SAl Viro * @dentry: child of nd->path.dentry 7014675ac39SAl Viro * @seq: seq number to check dentry against 7024675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7034675ac39SAl Viro * 7044675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7054675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7064675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7074675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7084675ac39SAl Viro * terminate_walk(). 7094675ac39SAl Viro */ 7104675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7114675ac39SAl Viro { 7124675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7134675ac39SAl Viro 7144675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7154675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7164675ac39SAl Viro goto out2; 7177973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7187973387aSAl Viro goto out2; 7194675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7207973387aSAl Viro goto out1; 72148a066e7SAl Viro 72215570086SLinus Torvalds /* 7234675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7244675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7254675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7264675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7274675ac39SAl Viro * know the parent must still be valid if the child sequence number is 72815570086SLinus Torvalds */ 7294675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 730e5c832d5SLinus Torvalds goto out; 73184a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 73284a2bd39SAl Viro goto out_dput; 733e5c832d5SLinus Torvalds /* 734e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 735e5c832d5SLinus Torvalds * still valid and get it if required. 736e5c832d5SLinus Torvalds */ 73784a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 73884a2bd39SAl Viro goto out_dput; 7398b61e74fSAl Viro rcu_read_unlock(); 74031e6b01fSNick Piggin return 0; 74119660af7SAl Viro 7427973387aSAl Viro out2: 7437973387aSAl Viro nd->path.mnt = NULL; 7447973387aSAl Viro out1: 7457973387aSAl Viro nd->path.dentry = NULL; 746e5c832d5SLinus Torvalds out: 7478b61e74fSAl Viro rcu_read_unlock(); 74884a2bd39SAl Viro return -ECHILD; 74984a2bd39SAl Viro out_dput: 75084a2bd39SAl Viro rcu_read_unlock(); 75184a2bd39SAl Viro dput(dentry); 75231e6b01fSNick Piggin return -ECHILD; 75331e6b01fSNick Piggin } 75431e6b01fSNick Piggin 7554ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 75634286d66SNick Piggin { 757a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7584ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 759a89f8337SAl Viro else 760a89f8337SAl Viro return 1; 76134286d66SNick Piggin } 76234286d66SNick Piggin 7639f1fafeeSAl Viro /** 7649f1fafeeSAl Viro * complete_walk - successful completion of path walk 7659f1fafeeSAl Viro * @nd: pointer nameidata 76639159de2SJeff Layton * 7679f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7689f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7699f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7709f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7719f1fafeeSAl Viro * need to drop nd->path. 77239159de2SJeff Layton */ 7739f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 77439159de2SJeff Layton { 77516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 77639159de2SJeff Layton int status; 77739159de2SJeff Layton 7789f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7799f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7809f1fafeeSAl Viro nd->root.mnt = NULL; 7814675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 78248a066e7SAl Viro return -ECHILD; 78348a066e7SAl Viro } 7849f1fafeeSAl Viro 78516c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 78639159de2SJeff Layton return 0; 78739159de2SJeff Layton 788ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 78916c2cd71SAl Viro return 0; 79016c2cd71SAl Viro 791ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 79239159de2SJeff Layton if (status > 0) 79339159de2SJeff Layton return 0; 79439159de2SJeff Layton 79516c2cd71SAl Viro if (!status) 79639159de2SJeff Layton status = -ESTALE; 79716c2cd71SAl Viro 79839159de2SJeff Layton return status; 79939159de2SJeff Layton } 80039159de2SJeff Layton 80118d8c860SAl Viro static void set_root(struct nameidata *nd) 8022a737871SAl Viro { 80331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8049e6697e2SAl Viro 8059e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8068f47a016SAl Viro unsigned seq; 807c28cc364SNick Piggin 808c28cc364SNick Piggin do { 809c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 81031e6b01fSNick Piggin nd->root = fs->root; 8118f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 812c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8139e6697e2SAl Viro } else { 8149e6697e2SAl Viro get_fs_root(fs, &nd->root); 81584a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 8169e6697e2SAl Viro } 81731e6b01fSNick Piggin } 81831e6b01fSNick Piggin 8191d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 820051d3812SIan Kent { 821051d3812SIan Kent dput(path->dentry); 8224ac91378SJan Blunck if (path->mnt != nd->path.mnt) 823051d3812SIan Kent mntput(path->mnt); 824051d3812SIan Kent } 825051d3812SIan Kent 8267b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8277b9337aaSNick Piggin struct nameidata *nd) 828051d3812SIan Kent { 82931e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8304ac91378SJan Blunck dput(nd->path.dentry); 83131e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8324ac91378SJan Blunck mntput(nd->path.mnt); 8339a229683SHuang Shijie } 83431e6b01fSNick Piggin nd->path.mnt = path->mnt; 8354ac91378SJan Blunck nd->path.dentry = path->dentry; 836051d3812SIan Kent } 837051d3812SIan Kent 838248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 839248fb5b9SAl Viro { 840248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 841248fb5b9SAl Viro struct dentry *d; 842248fb5b9SAl Viro nd->path = nd->root; 843248fb5b9SAl Viro d = nd->path.dentry; 844248fb5b9SAl Viro nd->inode = d->d_inode; 845248fb5b9SAl Viro nd->seq = nd->root_seq; 846248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 847248fb5b9SAl Viro return -ECHILD; 848248fb5b9SAl Viro } else { 849248fb5b9SAl Viro path_put(&nd->path); 850248fb5b9SAl Viro nd->path = nd->root; 851248fb5b9SAl Viro path_get(&nd->path); 852248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 853248fb5b9SAl Viro } 854248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 855248fb5b9SAl Viro return 0; 856248fb5b9SAl Viro } 857248fb5b9SAl Viro 858b5fb63c1SChristoph Hellwig /* 8596b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 860b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 861b5fb63c1SChristoph Hellwig */ 8626e77137bSAl Viro void nd_jump_link(struct path *path) 863b5fb63c1SChristoph Hellwig { 8646e77137bSAl Viro struct nameidata *nd = current->nameidata; 865b5fb63c1SChristoph Hellwig path_put(&nd->path); 866b5fb63c1SChristoph Hellwig 867b5fb63c1SChristoph Hellwig nd->path = *path; 868b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 869b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 870b5fb63c1SChristoph Hellwig } 871b5fb63c1SChristoph Hellwig 872b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 873574197e0SAl Viro { 87421c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 875fceef393SAl Viro do_delayed_call(&last->done); 8766548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 877b9ff4429SAl Viro path_put(&last->link); 878574197e0SAl Viro } 879574197e0SAl Viro 880561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 881561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 88230aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 88330aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 884800179c9SKees Cook 885800179c9SKees Cook /** 886800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 88755852635SRandy Dunlap * @nd: nameidata pathwalk data 888800179c9SKees Cook * 889800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 890800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 891800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 892800179c9SKees Cook * processes from failing races against path names that may change out 893800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 894800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 895800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 896800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 897800179c9SKees Cook * 898800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 899800179c9SKees Cook */ 900fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 901800179c9SKees Cook { 902800179c9SKees Cook const struct inode *inode; 903800179c9SKees Cook const struct inode *parent; 9042d7f9e2aSSeth Forshee kuid_t puid; 905800179c9SKees Cook 906800179c9SKees Cook if (!sysctl_protected_symlinks) 907800179c9SKees Cook return 0; 908800179c9SKees Cook 909800179c9SKees Cook /* Allowed if owner and follower match. */ 910fceef393SAl Viro inode = nd->link_inode; 91181abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 912800179c9SKees Cook return 0; 913800179c9SKees Cook 914800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 915aa65fa35SAl Viro parent = nd->inode; 916800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 917800179c9SKees Cook return 0; 918800179c9SKees Cook 919800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9202d7f9e2aSSeth Forshee puid = parent->i_uid; 9212d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 922800179c9SKees Cook return 0; 923800179c9SKees Cook 92431956502SAl Viro if (nd->flags & LOOKUP_RCU) 92531956502SAl Viro return -ECHILD; 92631956502SAl Viro 927ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 92894b9d9b7SRichard Guy Briggs audit_log_link_denied("follow_link"); 929800179c9SKees Cook return -EACCES; 930800179c9SKees Cook } 931800179c9SKees Cook 932800179c9SKees Cook /** 933800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 934800179c9SKees Cook * @inode: the source inode to hardlink from 935800179c9SKees Cook * 936800179c9SKees Cook * Return false if at least one of the following conditions: 937800179c9SKees Cook * - inode is not a regular file 938800179c9SKees Cook * - inode is setuid 939800179c9SKees Cook * - inode is setgid and group-exec 940800179c9SKees Cook * - access failure for read and write 941800179c9SKees Cook * 942800179c9SKees Cook * Otherwise returns true. 943800179c9SKees Cook */ 944800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 945800179c9SKees Cook { 946800179c9SKees Cook umode_t mode = inode->i_mode; 947800179c9SKees Cook 948800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 949800179c9SKees Cook if (!S_ISREG(mode)) 950800179c9SKees Cook return false; 951800179c9SKees Cook 952800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 953800179c9SKees Cook if (mode & S_ISUID) 954800179c9SKees Cook return false; 955800179c9SKees Cook 956800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 957800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 958800179c9SKees Cook return false; 959800179c9SKees Cook 960800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 961800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 962800179c9SKees Cook return false; 963800179c9SKees Cook 964800179c9SKees Cook return true; 965800179c9SKees Cook } 966800179c9SKees Cook 967800179c9SKees Cook /** 968800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 969800179c9SKees Cook * @link: the source to hardlink from 970800179c9SKees Cook * 971800179c9SKees Cook * Block hardlink when all of: 972800179c9SKees Cook * - sysctl_protected_hardlinks enabled 973800179c9SKees Cook * - fsuid does not match inode 974800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 975f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 976800179c9SKees Cook * 977800179c9SKees Cook * Returns 0 if successful, -ve on error. 978800179c9SKees Cook */ 979800179c9SKees Cook static int may_linkat(struct path *link) 980800179c9SKees Cook { 981593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 982593d1ce8SEric W. Biederman 983593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 984593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 985593d1ce8SEric W. Biederman return -EOVERFLOW; 986800179c9SKees Cook 987800179c9SKees Cook if (!sysctl_protected_hardlinks) 988800179c9SKees Cook return 0; 989800179c9SKees Cook 990800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 991800179c9SKees Cook * otherwise, it must be a safe source. 992800179c9SKees Cook */ 993cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 994800179c9SKees Cook return 0; 995800179c9SKees Cook 99694b9d9b7SRichard Guy Briggs audit_log_link_denied("linkat"); 997800179c9SKees Cook return -EPERM; 998800179c9SKees Cook } 999800179c9SKees Cook 100030aba665SSalvatore Mesoraca /** 100130aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 100230aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 100330aba665SSalvatore Mesoraca * exist. 100430aba665SSalvatore Mesoraca * @dir: the sticky parent directory 100530aba665SSalvatore Mesoraca * @inode: the inode of the file to open 100630aba665SSalvatore Mesoraca * 100730aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 100830aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 100930aba665SSalvatore Mesoraca * - the file already exists 101030aba665SSalvatore Mesoraca * - we are in a sticky directory 101130aba665SSalvatore Mesoraca * - we don't own the file 101230aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 101330aba665SSalvatore Mesoraca * - the directory is world writable 101430aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 101530aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 101630aba665SSalvatore Mesoraca * be enough. 101730aba665SSalvatore Mesoraca * 101830aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 101930aba665SSalvatore Mesoraca */ 102030aba665SSalvatore Mesoraca static int may_create_in_sticky(struct dentry * const dir, 102130aba665SSalvatore Mesoraca struct inode * const inode) 102230aba665SSalvatore Mesoraca { 102330aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 102430aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 102530aba665SSalvatore Mesoraca likely(!(dir->d_inode->i_mode & S_ISVTX)) || 102630aba665SSalvatore Mesoraca uid_eq(inode->i_uid, dir->d_inode->i_uid) || 102730aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 102830aba665SSalvatore Mesoraca return 0; 102930aba665SSalvatore Mesoraca 103030aba665SSalvatore Mesoraca if (likely(dir->d_inode->i_mode & 0002) || 103130aba665SSalvatore Mesoraca (dir->d_inode->i_mode & 0020 && 103230aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 103330aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 103430aba665SSalvatore Mesoraca return -EACCES; 103530aba665SSalvatore Mesoraca } 103630aba665SSalvatore Mesoraca return 0; 103730aba665SSalvatore Mesoraca } 103830aba665SSalvatore Mesoraca 10393b2e7f75SAl Viro static __always_inline 10403b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10411da177e4SLinus Torvalds { 1042ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10431cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1044fceef393SAl Viro struct inode *inode = nd->link_inode; 10456d7b5aaeSAl Viro int error; 10460a959df5SAl Viro const char *res; 10471da177e4SLinus Torvalds 10488fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10498fa9dd24SNeilBrown touch_atime(&last->link); 10508fa9dd24SNeilBrown cond_resched(); 1051c6718543SMiklos Szeredi } else if (atime_needs_update(&last->link, inode)) { 10524675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10538fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10548fa9dd24SNeilBrown touch_atime(&last->link); 10558fa9dd24SNeilBrown } 10568fa9dd24SNeilBrown 1057bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1058bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1059bda0be7aSNeilBrown if (unlikely(error)) 10606920a440SAl Viro return ERR_PTR(error); 106136f3b4f6SAl Viro 106286acdca1SAl Viro nd->last_type = LAST_BIND; 10634c4f7c19SEric Biggers res = READ_ONCE(inode->i_link); 1064d4dee48bSAl Viro if (!res) { 1065fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1066fceef393SAl Viro struct delayed_call *); 1067fceef393SAl Viro get = inode->i_op->get_link; 10688c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1069fceef393SAl Viro res = get(NULL, inode, &last->done); 10706b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10714675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10728c1b4566SAl Viro return ERR_PTR(-ECHILD); 1073fceef393SAl Viro res = get(dentry, inode, &last->done); 10748c1b4566SAl Viro } 10756b255391SAl Viro } else { 1076fceef393SAl Viro res = get(dentry, inode, &last->done); 10776b255391SAl Viro } 1078fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1079fab51e8aSAl Viro return res; 10800a959df5SAl Viro } 1081fab51e8aSAl Viro if (*res == '/') { 1082fab51e8aSAl Viro if (!nd->root.mnt) 1083fab51e8aSAl Viro set_root(nd); 1084248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1085fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1086fab51e8aSAl Viro while (unlikely(*++res == '/')) 1087fab51e8aSAl Viro ; 1088fab51e8aSAl Viro } 1089fab51e8aSAl Viro if (!*res) 1090fab51e8aSAl Viro res = NULL; 10910a959df5SAl Viro return res; 10920a959df5SAl Viro } 10936d7b5aaeSAl Viro 1094f015f126SDavid Howells /* 1095f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1096f015f126SDavid Howells * 1097f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1098f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1099f015f126SDavid Howells * Up is towards /. 1100f015f126SDavid Howells * 1101f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1102f015f126SDavid Howells * root. 1103f015f126SDavid Howells */ 1104bab77ebfSAl Viro int follow_up(struct path *path) 11051da177e4SLinus Torvalds { 11060714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11070714a533SAl Viro struct mount *parent; 11081da177e4SLinus Torvalds struct dentry *mountpoint; 110999b7db7bSNick Piggin 111048a066e7SAl Viro read_seqlock_excl(&mount_lock); 11110714a533SAl Viro parent = mnt->mnt_parent; 11123c0a6163SAl Viro if (parent == mnt) { 111348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11141da177e4SLinus Torvalds return 0; 11151da177e4SLinus Torvalds } 11160714a533SAl Viro mntget(&parent->mnt); 1117a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 111848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1119bab77ebfSAl Viro dput(path->dentry); 1120bab77ebfSAl Viro path->dentry = mountpoint; 1121bab77ebfSAl Viro mntput(path->mnt); 11220714a533SAl Viro path->mnt = &parent->mnt; 11231da177e4SLinus Torvalds return 1; 11241da177e4SLinus Torvalds } 11254d359507SAl Viro EXPORT_SYMBOL(follow_up); 11261da177e4SLinus Torvalds 1127b5c84bf6SNick Piggin /* 11289875cf80SDavid Howells * Perform an automount 11299875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11309875cf80SDavid Howells * were called with. 11311da177e4SLinus Torvalds */ 1132756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 11339875cf80SDavid Howells bool *need_mntput) 113431e6b01fSNick Piggin { 11359875cf80SDavid Howells struct vfsmount *mnt; 1136ea5b778aSDavid Howells int err; 11379875cf80SDavid Howells 11389875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 11399875cf80SDavid Howells return -EREMOTE; 11409875cf80SDavid Howells 11410ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11420ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11430ec26fd0SMiklos Szeredi * the name. 11440ec26fd0SMiklos Szeredi * 11450ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11465a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11470ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11480ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11490ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11500ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11515a30d8a2SDavid Howells */ 1152756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11535d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11545d38f049SIan Kent path->dentry->d_inode) 11559875cf80SDavid Howells return -EISDIR; 11560ec26fd0SMiklos Szeredi 1157756daf26SNeilBrown nd->total_link_count++; 1158756daf26SNeilBrown if (nd->total_link_count >= 40) 11599875cf80SDavid Howells return -ELOOP; 11609875cf80SDavid Howells 11619875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11629875cf80SDavid Howells if (IS_ERR(mnt)) { 11639875cf80SDavid Howells /* 11649875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11659875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11669875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11679875cf80SDavid Howells * 11689875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11699875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11709875cf80SDavid Howells * the path is inaccessible and we should say so. 11719875cf80SDavid Howells */ 1172756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11739875cf80SDavid Howells return -EREMOTE; 11749875cf80SDavid Howells return PTR_ERR(mnt); 117531e6b01fSNick Piggin } 1176ea5b778aSDavid Howells 11779875cf80SDavid Howells if (!mnt) /* mount collision */ 11789875cf80SDavid Howells return 0; 11799875cf80SDavid Howells 11808aef1884SAl Viro if (!*need_mntput) { 11818aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11828aef1884SAl Viro mntget(path->mnt); 11838aef1884SAl Viro *need_mntput = true; 11848aef1884SAl Viro } 118519a167afSAl Viro err = finish_automount(mnt, path); 1186ea5b778aSDavid Howells 1187ea5b778aSDavid Howells switch (err) { 1188ea5b778aSDavid Howells case -EBUSY: 1189ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 119019a167afSAl Viro return 0; 1191ea5b778aSDavid Howells case 0: 11928aef1884SAl Viro path_put(path); 11939875cf80SDavid Howells path->mnt = mnt; 11949875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11959875cf80SDavid Howells return 0; 119619a167afSAl Viro default: 119719a167afSAl Viro return err; 11989875cf80SDavid Howells } 119919a167afSAl Viro 1200ea5b778aSDavid Howells } 12019875cf80SDavid Howells 12029875cf80SDavid Howells /* 12039875cf80SDavid Howells * Handle a dentry that is managed in some way. 1204cc53ce53SDavid Howells * - Flagged for transit management (autofs) 12059875cf80SDavid Howells * - Flagged as mountpoint 12069875cf80SDavid Howells * - Flagged as automount point 12079875cf80SDavid Howells * 12089875cf80SDavid Howells * This may only be called in refwalk mode. 1209d41efb52SAl Viro * On success path->dentry is known positive. 12109875cf80SDavid Howells * 12119875cf80SDavid Howells * Serialization is taken care of in namespace.c 12129875cf80SDavid Howells */ 1213756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 12149875cf80SDavid Howells { 12158aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 1216d41efb52SAl Viro unsigned flags; 12179875cf80SDavid Howells bool need_mntput = false; 12188aef1884SAl Viro int ret = 0; 12199875cf80SDavid Howells 12209875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 12219875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 12229875cf80SDavid Howells * the components of that value change under us */ 1223d41efb52SAl Viro while (flags = READ_ONCE(path->dentry->d_flags), 1224d41efb52SAl Viro unlikely(flags & DCACHE_MANAGED_DENTRY)) { 1225cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1226cc53ce53SDavid Howells * being held. */ 1227d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1228cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1229cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1230fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1231cc53ce53SDavid Howells if (ret < 0) 12328aef1884SAl Viro break; 1233cc53ce53SDavid Howells } 1234cc53ce53SDavid Howells 12359875cf80SDavid Howells /* Transit to a mounted filesystem. */ 1236d41efb52SAl Viro if (flags & DCACHE_MOUNTED) { 12379875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12389875cf80SDavid Howells if (mounted) { 12399875cf80SDavid Howells dput(path->dentry); 12409875cf80SDavid Howells if (need_mntput) 1241463ffb2eSAl Viro mntput(path->mnt); 1242463ffb2eSAl Viro path->mnt = mounted; 1243463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12449875cf80SDavid Howells need_mntput = true; 12459875cf80SDavid Howells continue; 1246463ffb2eSAl Viro } 1247463ffb2eSAl Viro 12489875cf80SDavid Howells /* Something is mounted on this dentry in another 12499875cf80SDavid Howells * namespace and/or whatever was mounted there in this 125048a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 125148a066e7SAl Viro * get it */ 12521da177e4SLinus Torvalds } 12539875cf80SDavid Howells 12549875cf80SDavid Howells /* Handle an automount point */ 1255d41efb52SAl Viro if (flags & DCACHE_NEED_AUTOMOUNT) { 1256756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12579875cf80SDavid Howells if (ret < 0) 12588aef1884SAl Viro break; 12599875cf80SDavid Howells continue; 12609875cf80SDavid Howells } 12619875cf80SDavid Howells 12629875cf80SDavid Howells /* We didn't change the current path point */ 12639875cf80SDavid Howells break; 12649875cf80SDavid Howells } 12658aef1884SAl Viro 12668aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12678aef1884SAl Viro mntput(path->mnt); 12688402752eSAl Viro if (need_mntput) 12698402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 1270d41efb52SAl Viro if (ret == -EISDIR || !ret) 1271d41efb52SAl Viro ret = 1; 1272d41efb52SAl Viro if (ret > 0 && unlikely(d_flags_negative(flags))) 1273d41efb52SAl Viro ret = -ENOENT; 12748402752eSAl Viro if (unlikely(ret < 0)) 12758402752eSAl Viro path_put_conditional(path, nd); 12768402752eSAl Viro return ret; 12771da177e4SLinus Torvalds } 12781da177e4SLinus Torvalds 1279cc53ce53SDavid Howells int follow_down_one(struct path *path) 12801da177e4SLinus Torvalds { 12811da177e4SLinus Torvalds struct vfsmount *mounted; 12821da177e4SLinus Torvalds 12831c755af4SAl Viro mounted = lookup_mnt(path); 12841da177e4SLinus Torvalds if (mounted) { 12859393bd07SAl Viro dput(path->dentry); 12869393bd07SAl Viro mntput(path->mnt); 12879393bd07SAl Viro path->mnt = mounted; 12889393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12891da177e4SLinus Torvalds return 1; 12901da177e4SLinus Torvalds } 12911da177e4SLinus Torvalds return 0; 12921da177e4SLinus Torvalds } 12934d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12941da177e4SLinus Torvalds 1295fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 129662a7375eSIan Kent { 1297fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1298fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 129962a7375eSIan Kent } 130062a7375eSIan Kent 13019875cf80SDavid Howells /* 1302287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1303287548e4SAl Viro * we meet a managed dentry that would need blocking. 13049875cf80SDavid Howells */ 13059875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1306254cf582SAl Viro struct inode **inode, unsigned *seqp) 13079875cf80SDavid Howells { 130862a7375eSIan Kent for (;;) { 1309c7105365SAl Viro struct mount *mounted; 131062a7375eSIan Kent /* 131162a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 131262a7375eSIan Kent * that wants to block transit. 131362a7375eSIan Kent */ 1314fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1315b8faf035SNeilBrown case -ECHILD: 1316b8faf035SNeilBrown default: 1317ab90911fSDavid Howells return false; 1318b8faf035SNeilBrown case -EISDIR: 1319b8faf035SNeilBrown return true; 1320b8faf035SNeilBrown case 0: 1321b8faf035SNeilBrown break; 1322b8faf035SNeilBrown } 132362a7375eSIan Kent 132462a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1325b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 132662a7375eSIan Kent 1327474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 13289875cf80SDavid Howells if (!mounted) 13299875cf80SDavid Howells break; 1330c7105365SAl Viro path->mnt = &mounted->mnt; 1331c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1332a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1333254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 133459430262SLinus Torvalds /* 133559430262SLinus Torvalds * Update the inode too. We don't need to re-check the 133659430262SLinus Torvalds * dentry sequence number here after this d_inode read, 133759430262SLinus Torvalds * because a mount-point is always pinned. 133859430262SLinus Torvalds */ 133959430262SLinus Torvalds *inode = path->dentry->d_inode; 13409875cf80SDavid Howells } 1341f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1342b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1343287548e4SAl Viro } 1344287548e4SAl Viro 134531e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 134631e6b01fSNick Piggin { 13474023bfc9SAl Viro struct inode *inode = nd->inode; 134831e6b01fSNick Piggin 134931e6b01fSNick Piggin while (1) { 1350aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 135131e6b01fSNick Piggin break; 135231e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 135331e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 135431e6b01fSNick Piggin struct dentry *parent = old->d_parent; 135531e6b01fSNick Piggin unsigned seq; 135631e6b01fSNick Piggin 13574023bfc9SAl Viro inode = parent->d_inode; 135831e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1359aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1360aed434adSAl Viro return -ECHILD; 136131e6b01fSNick Piggin nd->path.dentry = parent; 136231e6b01fSNick Piggin nd->seq = seq; 1363397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1364397d425dSEric W. Biederman return -ENOENT; 136531e6b01fSNick Piggin break; 1366aed434adSAl Viro } else { 1367aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1368aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1369aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1370aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1371aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1372aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1373aed434adSAl Viro return -ECHILD; 1374aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 137531e6b01fSNick Piggin break; 1376aed434adSAl Viro /* we know that mountpoint was pinned */ 1377aed434adSAl Viro nd->path.dentry = mountpoint; 1378aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1379aed434adSAl Viro inode = inode2; 1380aed434adSAl Viro nd->seq = seq; 138131e6b01fSNick Piggin } 1382aed434adSAl Viro } 1383aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1384b37199e6SAl Viro struct mount *mounted; 1385b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1386aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1387aed434adSAl Viro return -ECHILD; 1388b37199e6SAl Viro if (!mounted) 1389b37199e6SAl Viro break; 1390b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1391b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13924023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1393b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1394b37199e6SAl Viro } 13954023bfc9SAl Viro nd->inode = inode; 139631e6b01fSNick Piggin return 0; 139731e6b01fSNick Piggin } 139831e6b01fSNick Piggin 13999875cf80SDavid Howells /* 1400cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1401cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1402cc53ce53SDavid Howells * caller is permitted to proceed or not. 1403cc53ce53SDavid Howells */ 14047cc90cc3SAl Viro int follow_down(struct path *path) 1405cc53ce53SDavid Howells { 1406cc53ce53SDavid Howells unsigned managed; 1407cc53ce53SDavid Howells int ret; 1408cc53ce53SDavid Howells 14096aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1410cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1411cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1412cc53ce53SDavid Howells * being held. 1413cc53ce53SDavid Howells * 1414cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1415cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1416cc53ce53SDavid Howells * other than its daemon the right to mount on its 1417cc53ce53SDavid Howells * superstructure. 1418cc53ce53SDavid Howells * 1419cc53ce53SDavid Howells * The filesystem may sleep at this point. 1420cc53ce53SDavid Howells */ 1421cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1422cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1423cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1424fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1425cc53ce53SDavid Howells if (ret < 0) 1426cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1427cc53ce53SDavid Howells } 1428cc53ce53SDavid Howells 1429cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1430cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1431cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1432cc53ce53SDavid Howells if (!mounted) 1433cc53ce53SDavid Howells break; 1434cc53ce53SDavid Howells dput(path->dentry); 1435cc53ce53SDavid Howells mntput(path->mnt); 1436cc53ce53SDavid Howells path->mnt = mounted; 1437cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1438cc53ce53SDavid Howells continue; 1439cc53ce53SDavid Howells } 1440cc53ce53SDavid Howells 1441cc53ce53SDavid Howells /* Don't handle automount points here */ 1442cc53ce53SDavid Howells break; 1443cc53ce53SDavid Howells } 1444cc53ce53SDavid Howells return 0; 1445cc53ce53SDavid Howells } 14464d359507SAl Viro EXPORT_SYMBOL(follow_down); 1447cc53ce53SDavid Howells 1448cc53ce53SDavid Howells /* 14499875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14509875cf80SDavid Howells */ 14519875cf80SDavid Howells static void follow_mount(struct path *path) 14529875cf80SDavid Howells { 14539875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14549875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14559875cf80SDavid Howells if (!mounted) 14569875cf80SDavid Howells break; 14579875cf80SDavid Howells dput(path->dentry); 14589875cf80SDavid Howells mntput(path->mnt); 14599875cf80SDavid Howells path->mnt = mounted; 14609875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14619875cf80SDavid Howells } 14629875cf80SDavid Howells } 14639875cf80SDavid Howells 1464eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1465eedf265aSEric W. Biederman { 1466eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1467eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1468eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1469eedf265aSEric W. Biederman dput(old); 1470eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1471eedf265aSEric W. Biederman return -ENOENT; 1472eedf265aSEric W. Biederman return 0; 1473eedf265aSEric W. Biederman } 1474eedf265aSEric W. Biederman 1475397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14761da177e4SLinus Torvalds { 14771da177e4SLinus Torvalds while(1) { 1478030c7e0bSDanilo Krummrich if (path_equal(&nd->path, &nd->root)) 14791da177e4SLinus Torvalds break; 14804ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1481eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1482eedf265aSEric W. Biederman if (ret) 1483eedf265aSEric W. Biederman return ret; 14841da177e4SLinus Torvalds break; 14851da177e4SLinus Torvalds } 14863088dd70SAl Viro if (!follow_up(&nd->path)) 14871da177e4SLinus Torvalds break; 14881da177e4SLinus Torvalds } 148979ed0226SAl Viro follow_mount(&nd->path); 149031e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1491397d425dSEric W. Biederman return 0; 14921da177e4SLinus Torvalds } 14931da177e4SLinus Torvalds 14941da177e4SLinus Torvalds /* 1495f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1496f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1497baa03890SNick Piggin */ 1498e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1499e3c13928SAl Viro struct dentry *dir, 15006c51e513SAl Viro unsigned int flags) 1501baa03890SNick Piggin { 1502a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1503bad61189SMiklos Szeredi if (dentry) { 1504a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1505bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 150674ff0ffcSAl Viro if (!error) 15075542aa2fSEric W. Biederman d_invalidate(dentry); 1508bad61189SMiklos Szeredi dput(dentry); 150974ff0ffcSAl Viro return ERR_PTR(error); 1510bad61189SMiklos Szeredi } 1511bad61189SMiklos Szeredi } 1512baa03890SNick Piggin return dentry; 1513baa03890SNick Piggin } 1514baa03890SNick Piggin 1515baa03890SNick Piggin /* 1516a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1517a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1518a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1519a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1520a03ece5fSAl Viro * at all. 152144396f4bSJosef Bacik */ 1522a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1523a03ece5fSAl Viro struct dentry *base, unsigned int flags) 152444396f4bSJosef Bacik { 1525a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 152644396f4bSJosef Bacik struct dentry *old; 1527a03ece5fSAl Viro struct inode *dir = base->d_inode; 1528a03ece5fSAl Viro 1529a03ece5fSAl Viro if (dentry) 1530a03ece5fSAl Viro return dentry; 153144396f4bSJosef Bacik 153244396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1533a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 153444396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1535a03ece5fSAl Viro 1536a03ece5fSAl Viro dentry = d_alloc(base, name); 1537a03ece5fSAl Viro if (unlikely(!dentry)) 1538a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 153944396f4bSJosef Bacik 154072bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 154144396f4bSJosef Bacik if (unlikely(old)) { 154244396f4bSJosef Bacik dput(dentry); 154344396f4bSJosef Bacik dentry = old; 154444396f4bSJosef Bacik } 154544396f4bSJosef Bacik return dentry; 154644396f4bSJosef Bacik } 154744396f4bSJosef Bacik 1548e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1549254cf582SAl Viro struct path *path, struct inode **inode, 1550254cf582SAl Viro unsigned *seqp) 15511da177e4SLinus Torvalds { 15524ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 155331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15545a18fff2SAl Viro int status = 1; 15559875cf80SDavid Howells int err; 15569875cf80SDavid Howells 15573cac260aSAl Viro /* 1558b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15595d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15605d0f49c1SAl Viro * going to fall back to non-racy lookup. 1561b04f784eSNick Piggin */ 156231e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 156331e6b01fSNick Piggin unsigned seq; 1564766c4cbfSAl Viro bool negative; 1565da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15665d0f49c1SAl Viro if (unlikely(!dentry)) { 15674675ac39SAl Viro if (unlazy_walk(nd)) 15685d0f49c1SAl Viro return -ECHILD; 1569e9742b53SAl Viro return 0; 15705d0f49c1SAl Viro } 15715a18fff2SAl Viro 157212f8ad4bSLinus Torvalds /* 157312f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 157412f8ad4bSLinus Torvalds * the dentry name information from lookup. 157512f8ad4bSLinus Torvalds */ 157663afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1577766c4cbfSAl Viro negative = d_is_negative(dentry); 15785d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 157912f8ad4bSLinus Torvalds return -ECHILD; 158012f8ad4bSLinus Torvalds 158112f8ad4bSLinus Torvalds /* 158212f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 158312f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 158412f8ad4bSLinus Torvalds * 158512f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 158612f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 158712f8ad4bSLinus Torvalds */ 15885d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 158931e6b01fSNick Piggin return -ECHILD; 15905a18fff2SAl Viro 1591254cf582SAl Viro *seqp = seq; 15924ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1593209a7fb2SAl Viro if (likely(status > 0)) { 1594daf3761cSTrond Myklebust /* 1595daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1596daf3761cSTrond Myklebust * case that drops it. 1597daf3761cSTrond Myklebust */ 15985d0f49c1SAl Viro if (unlikely(negative)) 1599daf3761cSTrond Myklebust return -ENOENT; 160031e6b01fSNick Piggin path->mnt = mnt; 160131e6b01fSNick Piggin path->dentry = dentry; 1602254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1603e9742b53SAl Viro return 1; 16045d0f49c1SAl Viro } 16054675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 1606254cf582SAl Viro return -ECHILD; 1607209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1608209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1609209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16105a18fff2SAl Viro } else { 1611e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 161281e6f520SAl Viro if (unlikely(!dentry)) 1613e9742b53SAl Viro return 0; 16144ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16155d0f49c1SAl Viro } 16165a18fff2SAl Viro if (unlikely(status <= 0)) { 1617e9742b53SAl Viro if (!status) 16185d0f49c1SAl Viro d_invalidate(dentry); 16195a18fff2SAl Viro dput(dentry); 16205a18fff2SAl Viro return status; 16215a18fff2SAl Viro } 16225d0f49c1SAl Viro 16231da177e4SLinus Torvalds path->mnt = mnt; 16241da177e4SLinus Torvalds path->dentry = dentry; 1625756daf26SNeilBrown err = follow_managed(path, nd); 1626e9742b53SAl Viro if (likely(err > 0)) 162763afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 16288402752eSAl Viro return err; 1629697f514dSMiklos Szeredi } 1630697f514dSMiklos Szeredi 1631697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 163288d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1633e3c13928SAl Viro struct dentry *dir, 1634e3c13928SAl Viro unsigned int flags) 1635697f514dSMiklos Szeredi { 163688d8331aSAl Viro struct dentry *dentry, *old; 16371936386eSAl Viro struct inode *inode = dir->d_inode; 1638d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16391936386eSAl Viro 16401936386eSAl Viro /* Don't go there if it's already dead */ 164194bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 164288d8331aSAl Viro return ERR_PTR(-ENOENT); 164394bdd655SAl Viro again: 1644d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 164594bdd655SAl Viro if (IS_ERR(dentry)) 164688d8331aSAl Viro return dentry; 164794bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1648a89f8337SAl Viro if (!(flags & LOOKUP_NO_REVAL)) { 1649949a852eSAl Viro int error = d_revalidate(dentry, flags); 1650949a852eSAl Viro if (unlikely(error <= 0)) { 165194bdd655SAl Viro if (!error) { 1652949a852eSAl Viro d_invalidate(dentry); 1653949a852eSAl Viro dput(dentry); 165494bdd655SAl Viro goto again; 165594bdd655SAl Viro } 165694bdd655SAl Viro dput(dentry); 1657949a852eSAl Viro dentry = ERR_PTR(error); 1658949a852eSAl Viro } 1659949a852eSAl Viro } 166094bdd655SAl Viro } else { 16611936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 166285c7f810SAl Viro d_lookup_done(dentry); 16631936386eSAl Viro if (unlikely(old)) { 16641936386eSAl Viro dput(dentry); 16651936386eSAl Viro dentry = old; 1666949a852eSAl Viro } 1667949a852eSAl Viro } 1668e3c13928SAl Viro return dentry; 16691da177e4SLinus Torvalds } 16701da177e4SLinus Torvalds 167188d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 167288d8331aSAl Viro struct dentry *dir, 167388d8331aSAl Viro unsigned int flags) 167488d8331aSAl Viro { 167588d8331aSAl Viro struct inode *inode = dir->d_inode; 167688d8331aSAl Viro struct dentry *res; 167788d8331aSAl Viro inode_lock_shared(inode); 167888d8331aSAl Viro res = __lookup_slow(name, dir, flags); 167988d8331aSAl Viro inode_unlock_shared(inode); 168088d8331aSAl Viro return res; 168188d8331aSAl Viro } 168288d8331aSAl Viro 168352094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 168452094c8aSAl Viro { 168552094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16864ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 168752094c8aSAl Viro if (err != -ECHILD) 168852094c8aSAl Viro return err; 16894675ac39SAl Viro if (unlazy_walk(nd)) 169052094c8aSAl Viro return -ECHILD; 169152094c8aSAl Viro } 16924ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 169352094c8aSAl Viro } 169452094c8aSAl Viro 16959856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16969856fa1bSAl Viro { 16979856fa1bSAl Viro if (type == LAST_DOTDOT) { 16989e6697e2SAl Viro if (!nd->root.mnt) 16999e6697e2SAl Viro set_root(nd); 17009856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 170170291aecSAl Viro return follow_dotdot_rcu(nd); 17029856fa1bSAl Viro } else 1703397d425dSEric W. Biederman return follow_dotdot(nd); 17049856fa1bSAl Viro } 17059856fa1bSAl Viro return 0; 17069856fa1bSAl Viro } 17079856fa1bSAl Viro 1708181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1709181548c0SAl Viro struct inode *inode, unsigned seq) 1710d63ff28fSAl Viro { 1711626de996SAl Viro int error; 17121cf2665bSAl Viro struct saved *last; 1713756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1714626de996SAl Viro path_to_nameidata(link, nd); 1715626de996SAl Viro return -ELOOP; 1716626de996SAl Viro } 1717bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1718cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1719cd179f44SAl Viro mntget(link->mnt); 17207973387aSAl Viro } 1721626de996SAl Viro error = nd_alloc_stack(nd); 1722626de996SAl Viro if (unlikely(error)) { 1723bc40aee0SAl Viro if (error == -ECHILD) { 1724ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1725ad1633a1SAl Viro drop_links(nd); 1726ad1633a1SAl Viro nd->depth = 0; 1727ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1728ad1633a1SAl Viro nd->path.mnt = NULL; 1729ad1633a1SAl Viro nd->path.dentry = NULL; 1730ad1633a1SAl Viro rcu_read_unlock(); 17314675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1732bc40aee0SAl Viro error = nd_alloc_stack(nd); 1733bc40aee0SAl Viro } 1734bc40aee0SAl Viro if (error) { 1735cd179f44SAl Viro path_put(link); 1736626de996SAl Viro return error; 1737626de996SAl Viro } 1738bc40aee0SAl Viro } 1739626de996SAl Viro 1740ab104923SAl Viro last = nd->stack + nd->depth++; 17411cf2665bSAl Viro last->link = *link; 1742fceef393SAl Viro clear_delayed_call(&last->done); 1743fceef393SAl Viro nd->link_inode = inode; 17440450b2d1SAl Viro last->seq = seq; 1745d63ff28fSAl Viro return 1; 1746d63ff28fSAl Viro } 1747d63ff28fSAl Viro 17488f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 174931d66bcdSAl Viro 17503ddcd056SLinus Torvalds /* 17513ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17523ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17533ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17543ddcd056SLinus Torvalds * for the common case. 17553ddcd056SLinus Torvalds */ 17568f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 17578f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 17583ddcd056SLinus Torvalds { 175931d66bcdSAl Viro if (!(flags & WALK_MORE) && nd->depth) 176031d66bcdSAl Viro put_link(nd); 17618f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 17628f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 17638f64fb1cSAl Viro /* not a symlink or should not follow */ 17648f64fb1cSAl Viro path_to_nameidata(path, nd); 17658f64fb1cSAl Viro nd->inode = inode; 17668f64fb1cSAl Viro nd->seq = seq; 1767d63ff28fSAl Viro return 0; 17688f64fb1cSAl Viro } 1769a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1770a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 17718f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1772a7f77542SAl Viro return -ECHILD; 1773a7f77542SAl Viro } 17748f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 17753ddcd056SLinus Torvalds } 17763ddcd056SLinus Torvalds 17774693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1778ce57dfc1SAl Viro { 1779caa85634SAl Viro struct path path; 1780ce57dfc1SAl Viro struct inode *inode; 1781254cf582SAl Viro unsigned seq; 1782ce57dfc1SAl Viro int err; 1783ce57dfc1SAl Viro /* 1784ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1785ce57dfc1SAl Viro * to be able to know about the current root directory and 1786ce57dfc1SAl Viro * parent relationships. 1787ce57dfc1SAl Viro */ 17884693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17894693a547SAl Viro err = handle_dots(nd, nd->last_type); 17901c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 17914693a547SAl Viro put_link(nd); 17924693a547SAl Viro return err; 17934693a547SAl Viro } 1794254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1795e9742b53SAl Viro if (unlikely(err <= 0)) { 1796697f514dSMiklos Szeredi if (err < 0) 1797f0a9ba70SAl Viro return err; 1798e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1799e3c13928SAl Viro nd->flags); 1800e3c13928SAl Viro if (IS_ERR(path.dentry)) 1801e3c13928SAl Viro return PTR_ERR(path.dentry); 18027500c38aSAl Viro 1803e3c13928SAl Viro path.mnt = nd->path.mnt; 1804e3c13928SAl Viro err = follow_managed(&path, nd); 1805e3c13928SAl Viro if (unlikely(err < 0)) 1806f0a9ba70SAl Viro return err; 1807697f514dSMiklos Szeredi 1808254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1809d4565649SAl Viro inode = d_backing_inode(path.dentry); 1810766c4cbfSAl Viro } 1811697f514dSMiklos Szeredi 18128f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1813ce57dfc1SAl Viro } 1814ce57dfc1SAl Viro 18151da177e4SLinus Torvalds /* 1816bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1817bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1818bfcfaa77SLinus Torvalds * 1819bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1820bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1821bfcfaa77SLinus Torvalds * fast. 1822bfcfaa77SLinus Torvalds * 1823bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1824bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1825bfcfaa77SLinus Torvalds * crossing operation. 1826bfcfaa77SLinus Torvalds * 1827bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1828bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1829bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1830bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1831bfcfaa77SLinus Torvalds */ 1832bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1833bfcfaa77SLinus Torvalds 1834f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1835bfcfaa77SLinus Torvalds 1836468a9428SGeorge Spelvin #ifdef HASH_MIX 1837bfcfaa77SLinus Torvalds 1838468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1839468a9428SGeorge Spelvin 1840468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18412a18da7aSGeorge Spelvin /* 18422a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18432a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18442a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18452a18da7aSGeorge Spelvin * and no temporaries. 18462a18da7aSGeorge Spelvin * 18472a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18482a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18492a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18502a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18512a18da7aSGeorge Spelvin * 18522a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18532a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18542a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18552a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18562a18da7aSGeorge Spelvin * 18572a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18582a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18592a18da7aSGeorge Spelvin * toggling any given output bit. 18602a18da7aSGeorge Spelvin * 18612a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18622a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18632a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18642a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18652a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18662a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18672a18da7aSGeorge Spelvin * Perfect: 8192 258048 18682a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18692a18da7aSGeorge Spelvin */ 18702a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18712a18da7aSGeorge Spelvin ( x ^= (a), \ 18722a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18732a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 18742a18da7aSGeorge Spelvin y *= 9 ) 1875bfcfaa77SLinus Torvalds 18760fed3ac8SGeorge Spelvin /* 18772a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 18782a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 18792a18da7aSGeorge Spelvin * work done before the hash value is used. 18800fed3ac8SGeorge Spelvin */ 18812a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18820fed3ac8SGeorge Spelvin { 18832a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 18842a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 18852a18da7aSGeorge Spelvin return y >> 32; 18860fed3ac8SGeorge Spelvin } 18870fed3ac8SGeorge Spelvin 1888bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1889bfcfaa77SLinus Torvalds 18902a18da7aSGeorge Spelvin /* 18912a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 18922a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18932a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 18942a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 18952a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 18962a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 18972a18da7aSGeorge Spelvin * Perfect: 2048 31744 18982a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 18992a18da7aSGeorge Spelvin */ 19002a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19012a18da7aSGeorge Spelvin ( x ^= (a), \ 19022a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19032a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19042a18da7aSGeorge Spelvin y *= 9 ) 1905bfcfaa77SLinus Torvalds 19062a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19070fed3ac8SGeorge Spelvin { 19082a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19092a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19100fed3ac8SGeorge Spelvin } 19110fed3ac8SGeorge Spelvin 1912bfcfaa77SLinus Torvalds #endif 1913bfcfaa77SLinus Torvalds 19142a18da7aSGeorge Spelvin /* 19152a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19162a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19172a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19182a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19192a18da7aSGeorge Spelvin * finds the delimiter after the name. 19202a18da7aSGeorge Spelvin */ 19218387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1922bfcfaa77SLinus Torvalds { 19238387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1924bfcfaa77SLinus Torvalds 1925bfcfaa77SLinus Torvalds for (;;) { 1926fcfd2fbfSGeorge Spelvin if (!len) 1927fcfd2fbfSGeorge Spelvin goto done; 1928e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1929bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1930bfcfaa77SLinus Torvalds break; 19312a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1932bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1933bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1934bfcfaa77SLinus Torvalds } 19352a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1936bfcfaa77SLinus Torvalds done: 19372a18da7aSGeorge Spelvin return fold_hash(x, y); 1938bfcfaa77SLinus Torvalds } 1939bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1940bfcfaa77SLinus Torvalds 1941fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19428387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1943fcfd2fbfSGeorge Spelvin { 19448387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19458387ff25SLinus Torvalds unsigned long adata, mask, len; 1946fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1947fcfd2fbfSGeorge Spelvin 19488387ff25SLinus Torvalds len = 0; 19498387ff25SLinus Torvalds goto inside; 19508387ff25SLinus Torvalds 1951fcfd2fbfSGeorge Spelvin do { 19522a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1953fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19548387ff25SLinus Torvalds inside: 1955fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1956fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1957fcfd2fbfSGeorge Spelvin 1958fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1959fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19602a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1961fcfd2fbfSGeorge Spelvin 19622a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1963fcfd2fbfSGeorge Spelvin } 1964fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1965fcfd2fbfSGeorge Spelvin 1966bfcfaa77SLinus Torvalds /* 1967bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1968d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1969bfcfaa77SLinus Torvalds */ 19708387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1971bfcfaa77SLinus Torvalds { 19728387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 19738387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 197436126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1975bfcfaa77SLinus Torvalds 19768387ff25SLinus Torvalds len = 0; 19778387ff25SLinus Torvalds goto inside; 19788387ff25SLinus Torvalds 1979bfcfaa77SLinus Torvalds do { 19802a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1981bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 19828387ff25SLinus Torvalds inside: 1983e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 198436126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 198536126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1986bfcfaa77SLinus Torvalds 198736126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 198836126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 198936126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 19902a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 199136126f8fSLinus Torvalds 19922a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1993bfcfaa77SLinus Torvalds } 1994bfcfaa77SLinus Torvalds 19952a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 1996bfcfaa77SLinus Torvalds 1997fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 19988387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 19990145acc2SLinus Torvalds { 20008387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20010145acc2SLinus Torvalds while (len--) 2002fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20030145acc2SLinus Torvalds return end_name_hash(hash); 20040145acc2SLinus Torvalds } 2005ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20060145acc2SLinus Torvalds 2007fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20088387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2009fcfd2fbfSGeorge Spelvin { 20108387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2011fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2012fcfd2fbfSGeorge Spelvin 2013fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2014e0ab7af9SGeorge Spelvin while (c) { 2015fcfd2fbfSGeorge Spelvin len++; 2016fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2017fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2018e0ab7af9SGeorge Spelvin } 2019fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2020fcfd2fbfSGeorge Spelvin } 2021f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2022fcfd2fbfSGeorge Spelvin 20233ddcd056SLinus Torvalds /* 2024200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2025200e9ef7SLinus Torvalds * one character. 2026200e9ef7SLinus Torvalds */ 20278387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2028200e9ef7SLinus Torvalds { 20298387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2030200e9ef7SLinus Torvalds unsigned long len = 0, c; 2031200e9ef7SLinus Torvalds 2032200e9ef7SLinus Torvalds c = (unsigned char)*name; 2033200e9ef7SLinus Torvalds do { 2034200e9ef7SLinus Torvalds len++; 2035200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2036200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2037200e9ef7SLinus Torvalds } while (c && c != '/'); 2038d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2039200e9ef7SLinus Torvalds } 2040200e9ef7SLinus Torvalds 2041bfcfaa77SLinus Torvalds #endif 2042bfcfaa77SLinus Torvalds 2043200e9ef7SLinus Torvalds /* 20441da177e4SLinus Torvalds * Name resolution. 2045ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2046ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20471da177e4SLinus Torvalds * 2048ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2049ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20501da177e4SLinus Torvalds */ 20516de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20521da177e4SLinus Torvalds { 20531da177e4SLinus Torvalds int err; 20541da177e4SLinus Torvalds 20559b5858e9SAl Viro if (IS_ERR(name)) 20569b5858e9SAl Viro return PTR_ERR(name); 20571da177e4SLinus Torvalds while (*name=='/') 20581da177e4SLinus Torvalds name++; 20591da177e4SLinus Torvalds if (!*name) 20609e18f10aSAl Viro return 0; 20611da177e4SLinus Torvalds 20621da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20631da177e4SLinus Torvalds for(;;) { 2064d6bb3e90SLinus Torvalds u64 hash_len; 2065fe479a58SAl Viro int type; 20661da177e4SLinus Torvalds 206752094c8aSAl Viro err = may_lookup(nd); 20681da177e4SLinus Torvalds if (err) 20693595e234SAl Viro return err; 20701da177e4SLinus Torvalds 20718387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 20721da177e4SLinus Torvalds 2073fe479a58SAl Viro type = LAST_NORM; 2074d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2075fe479a58SAl Viro case 2: 2076200e9ef7SLinus Torvalds if (name[1] == '.') { 2077fe479a58SAl Viro type = LAST_DOTDOT; 207816c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 207916c2cd71SAl Viro } 2080fe479a58SAl Viro break; 2081fe479a58SAl Viro case 1: 2082fe479a58SAl Viro type = LAST_DOT; 2083fe479a58SAl Viro } 20845a202bcdSAl Viro if (likely(type == LAST_NORM)) { 20855a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 208616c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 20875a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2088a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2089da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 20905a202bcdSAl Viro if (err < 0) 20913595e234SAl Viro return err; 2092d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2093d6bb3e90SLinus Torvalds name = this.name; 20945a202bcdSAl Viro } 20955a202bcdSAl Viro } 2096fe479a58SAl Viro 2097d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2098d6bb3e90SLinus Torvalds nd->last.name = name; 20995f4a6a69SAl Viro nd->last_type = type; 21005f4a6a69SAl Viro 2101d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2102d6bb3e90SLinus Torvalds if (!*name) 2103bdf6cbf1SAl Viro goto OK; 2104200e9ef7SLinus Torvalds /* 2105200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2106200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2107200e9ef7SLinus Torvalds */ 2108200e9ef7SLinus Torvalds do { 2109d6bb3e90SLinus Torvalds name++; 2110d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21118620c238SAl Viro if (unlikely(!*name)) { 21128620c238SAl Viro OK: 2113368ee9baSAl Viro /* pathname body, done */ 21148620c238SAl Viro if (!nd->depth) 21158620c238SAl Viro return 0; 21168620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2117368ee9baSAl Viro /* trailing symlink, done */ 21188620c238SAl Viro if (!name) 21198620c238SAl Viro return 0; 21208620c238SAl Viro /* last component of nested symlink */ 21218f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 21221c4ff1a8SAl Viro } else { 21231c4ff1a8SAl Viro /* not the last component */ 21248f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 21258620c238SAl Viro } 2126ce57dfc1SAl Viro if (err < 0) 21273595e234SAl Viro return err; 2128fe479a58SAl Viro 2129ce57dfc1SAl Viro if (err) { 2130626de996SAl Viro const char *s = get_link(nd); 21315a460275SAl Viro 2132a1c83681SViresh Kumar if (IS_ERR(s)) 21333595e234SAl Viro return PTR_ERR(s); 2134172a39a0SAl Viro err = 0; 213512b09578SAl Viro if (unlikely(!s)) { 213612b09578SAl Viro /* jumped */ 2137b9ff4429SAl Viro put_link(nd); 213812b09578SAl Viro } else { 2139071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 214032cd7468SAl Viro name = s; 21419e18f10aSAl Viro continue; 214248c8b0c5SAl Viro } 214331e6b01fSNick Piggin } 214497242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 214597242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21464675ac39SAl Viro if (unlazy_walk(nd)) 214797242f99SAl Viro return -ECHILD; 214897242f99SAl Viro } 21493595e234SAl Viro return -ENOTDIR; 21505f4a6a69SAl Viro } 2151ce57dfc1SAl Viro } 215297242f99SAl Viro } 21531da177e4SLinus Torvalds 2154edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2155c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 215631e6b01fSNick Piggin { 2157c8a53ee5SAl Viro const char *s = nd->name->name; 215831e6b01fSNick Piggin 2159c0eb027eSLinus Torvalds if (!*s) 2160c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2161edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2162edc2b1daSAl Viro rcu_read_lock(); 2163c0eb027eSLinus Torvalds 216431e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2165980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 216631e6b01fSNick Piggin nd->depth = 0; 21675b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2168b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2169b18825a7SDavid Howells struct inode *inode = root->d_inode; 217093893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2171368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21725b6ca027SAl Viro nd->path = nd->root; 21735b6ca027SAl Viro nd->inode = inode; 21745b6ca027SAl Viro if (flags & LOOKUP_RCU) { 21755b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 21768f47a016SAl Viro nd->root_seq = nd->seq; 217748a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 21785b6ca027SAl Viro } else { 21795b6ca027SAl Viro path_get(&nd->path); 21805b6ca027SAl Viro } 2181368ee9baSAl Viro return s; 21825b6ca027SAl Viro } 21835b6ca027SAl Viro 218431e6b01fSNick Piggin nd->root.mnt = NULL; 2185248fb5b9SAl Viro nd->path.mnt = NULL; 2186248fb5b9SAl Viro nd->path.dentry = NULL; 218731e6b01fSNick Piggin 218848a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2189fd2f7cb5SAl Viro if (*s == '/') { 2190e41f7d4eSAl Viro set_root(nd); 2191248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2192ef55d917SAl Viro return s; 2193ef55d917SAl Viro return ERR_PTR(-ECHILD); 2194c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2195e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 219631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2197c28cc364SNick Piggin unsigned seq; 219831e6b01fSNick Piggin 2199c28cc364SNick Piggin do { 2200c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 220131e6b01fSNick Piggin nd->path = fs->pwd; 2202ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2203c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2204c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2205e41f7d4eSAl Viro } else { 2206e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2207ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2208e41f7d4eSAl Viro } 2209ef55d917SAl Viro return s; 221031e6b01fSNick Piggin } else { 2211582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2212c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 221331e6b01fSNick Piggin struct dentry *dentry; 221431e6b01fSNick Piggin 22152903ff01SAl Viro if (!f.file) 2216368ee9baSAl Viro return ERR_PTR(-EBADF); 221731e6b01fSNick Piggin 22182903ff01SAl Viro dentry = f.file->f_path.dentry; 221931e6b01fSNick Piggin 2220edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22212903ff01SAl Viro fdput(f); 2222368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2223f52e0c11SAl Viro } 22242903ff01SAl Viro 22252903ff01SAl Viro nd->path = f.file->f_path; 2226e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 222734a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 222834a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22295590ff0dSUlrich Drepper } else { 22302903ff01SAl Viro path_get(&nd->path); 223134a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22321da177e4SLinus Torvalds } 223334a26b99SAl Viro fdput(f); 2234368ee9baSAl Viro return s; 2235e41f7d4eSAl Viro } 22369b4a9b14SAl Viro } 22379b4a9b14SAl Viro 22383bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 223995fa25d9SAl Viro { 224095fa25d9SAl Viro const char *s; 2241fec2fa24SAl Viro int error = may_follow_link(nd); 2242deb106c6SAl Viro if (unlikely(error)) 22433bdba28bSAl Viro return ERR_PTR(error); 224495fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2245fab51e8aSAl Viro nd->stack[0].name = NULL; 22463b2e7f75SAl Viro s = get_link(nd); 2247deb106c6SAl Viro return s ? s : ""; 224895fa25d9SAl Viro } 224995fa25d9SAl Viro 2250caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2251bd92d7feSAl Viro { 2252bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2253bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2254bd92d7feSAl Viro 2255bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 22561c4ff1a8SAl Viro return walk_component(nd, 0); 2257bd92d7feSAl Viro } 2258bd92d7feSAl Viro 22594f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22604f757f3cSAl Viro { 22614f757f3cSAl Viro struct path path = nd->path; 22624f757f3cSAl Viro struct inode *inode = nd->inode; 22634f757f3cSAl Viro unsigned seq = nd->seq; 22644f757f3cSAl Viro int err; 22654f757f3cSAl Viro 22664f757f3cSAl Viro if (nd->flags & LOOKUP_RCU) { 22674f757f3cSAl Viro /* 22684f757f3cSAl Viro * don't bother with unlazy_walk on failure - we are 22694f757f3cSAl Viro * at the very beginning of walk, so we lose nothing 22704f757f3cSAl Viro * if we simply redo everything in non-RCU mode 22714f757f3cSAl Viro */ 22724f757f3cSAl Viro if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq))) 22734f757f3cSAl Viro return -ECHILD; 22744f757f3cSAl Viro } else { 22754f757f3cSAl Viro dget(path.dentry); 22764f757f3cSAl Viro err = follow_managed(&path, nd); 22774f757f3cSAl Viro if (unlikely(err < 0)) 22784f757f3cSAl Viro return err; 22794f757f3cSAl Viro inode = d_backing_inode(path.dentry); 22804f757f3cSAl Viro seq = 0; 22814f757f3cSAl Viro } 22824f757f3cSAl Viro path_to_nameidata(&path, nd); 22834f757f3cSAl Viro nd->inode = inode; 22844f757f3cSAl Viro nd->seq = seq; 22854f757f3cSAl Viro return 0; 22864f757f3cSAl Viro } 22874f757f3cSAl Viro 22889b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2289c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 22909b4a9b14SAl Viro { 2291c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2292bd92d7feSAl Viro int err; 229331e6b01fSNick Piggin 22949b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 22954f757f3cSAl Viro err = handle_lookup_down(nd); 22965f336e72SAl Viro if (unlikely(err < 0)) 22975f336e72SAl Viro s = ERR_PTR(err); 22984f757f3cSAl Viro } 22994f757f3cSAl Viro 23003bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 23013bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 23023bdba28bSAl Viro s = trailing_symlink(nd); 2303bd92d7feSAl Viro } 23049f1fafeeSAl Viro if (!err) 23059f1fafeeSAl Viro err = complete_walk(nd); 2306bd92d7feSAl Viro 2307deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2308deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2309bd23a539SAl Viro err = -ENOTDIR; 2310625b6d10SAl Viro if (!err) { 2311625b6d10SAl Viro *path = nd->path; 2312625b6d10SAl Viro nd->path.mnt = NULL; 2313625b6d10SAl Viro nd->path.dentry = NULL; 2314625b6d10SAl Viro } 2315deb106c6SAl Viro terminate_walk(nd); 2316bd92d7feSAl Viro return err; 231731e6b01fSNick Piggin } 231831e6b01fSNick Piggin 231931d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23209ad1aaa6SAl Viro struct path *path, struct path *root) 2321873f1eedSJeff Layton { 2322894bc8c4SAl Viro int retval; 23239883d185SAl Viro struct nameidata nd; 2324abc9f5beSAl Viro if (IS_ERR(name)) 2325abc9f5beSAl Viro return PTR_ERR(name); 23269ad1aaa6SAl Viro if (unlikely(root)) { 23279ad1aaa6SAl Viro nd.root = *root; 23289ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23299ad1aaa6SAl Viro } 23309883d185SAl Viro set_nameidata(&nd, dfd, name); 2331c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2332873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2333c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2334873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2335c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2336873f1eedSJeff Layton 2337873f1eedSJeff Layton if (likely(!retval)) 2338ff0ebee2SAl Viro audit_inode(name, path->dentry, 0); 23399883d185SAl Viro restore_nameidata(); 2340e4bd1c1aSAl Viro putname(name); 2341873f1eedSJeff Layton return retval; 2342873f1eedSJeff Layton } 2343873f1eedSJeff Layton 23448bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2345c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2346391172c4SAl Viro struct path *parent) 23478bcb77faSAl Viro { 2348c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23499b5858e9SAl Viro int err = link_path_walk(s, nd); 23508bcb77faSAl Viro if (!err) 23518bcb77faSAl Viro err = complete_walk(nd); 2352391172c4SAl Viro if (!err) { 2353391172c4SAl Viro *parent = nd->path; 2354391172c4SAl Viro nd->path.mnt = NULL; 2355391172c4SAl Viro nd->path.dentry = NULL; 2356391172c4SAl Viro } 2357deb106c6SAl Viro terminate_walk(nd); 23588bcb77faSAl Viro return err; 23598bcb77faSAl Viro } 23608bcb77faSAl Viro 23615c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2362391172c4SAl Viro unsigned int flags, struct path *parent, 2363391172c4SAl Viro struct qstr *last, int *type) 23648bcb77faSAl Viro { 23658bcb77faSAl Viro int retval; 23669883d185SAl Viro struct nameidata nd; 23678bcb77faSAl Viro 23685c31b6ceSAl Viro if (IS_ERR(name)) 23695c31b6ceSAl Viro return name; 23709883d185SAl Viro set_nameidata(&nd, dfd, name); 2371c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23728bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2373c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23748bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2375c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2376391172c4SAl Viro if (likely(!retval)) { 2377391172c4SAl Viro *last = nd.last; 2378391172c4SAl Viro *type = nd.last_type; 2379c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 23805c31b6ceSAl Viro } else { 23815c31b6ceSAl Viro putname(name); 23825c31b6ceSAl Viro name = ERR_PTR(retval); 2383391172c4SAl Viro } 23849883d185SAl Viro restore_nameidata(); 23855c31b6ceSAl Viro return name; 23868bcb77faSAl Viro } 23878bcb77faSAl Viro 238879714f72SAl Viro /* does lookup, returns the object with parent locked */ 238979714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 23905590ff0dSUlrich Drepper { 23915c31b6ceSAl Viro struct filename *filename; 23925c31b6ceSAl Viro struct dentry *d; 2393391172c4SAl Viro struct qstr last; 2394391172c4SAl Viro int type; 239551689104SPaul Moore 23965c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 23975c31b6ceSAl Viro &last, &type); 239851689104SPaul Moore if (IS_ERR(filename)) 239951689104SPaul Moore return ERR_CAST(filename); 24005c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2401391172c4SAl Viro path_put(path); 24025c31b6ceSAl Viro putname(filename); 24035c31b6ceSAl Viro return ERR_PTR(-EINVAL); 240479714f72SAl Viro } 24055955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2406391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 240779714f72SAl Viro if (IS_ERR(d)) { 24085955102cSAl Viro inode_unlock(path->dentry->d_inode); 2409391172c4SAl Viro path_put(path); 241079714f72SAl Viro } 241151689104SPaul Moore putname(filename); 241279714f72SAl Viro return d; 24135590ff0dSUlrich Drepper } 24145590ff0dSUlrich Drepper 2415d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2416d1811465SAl Viro { 2417abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2418abc9f5beSAl Viro flags, path, NULL); 2419d1811465SAl Viro } 24204d359507SAl Viro EXPORT_SYMBOL(kern_path); 2421d1811465SAl Viro 242216f18200SJosef 'Jeff' Sipek /** 242316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 242416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 242516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 242616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 242716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2428e0a01249SAl Viro * @path: pointer to struct path to fill 242916f18200SJosef 'Jeff' Sipek */ 243016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 243116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2432e0a01249SAl Viro struct path *path) 243316f18200SJosef 'Jeff' Sipek { 24349ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24359ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2436abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2437abc9f5beSAl Viro flags , path, &root); 243816f18200SJosef 'Jeff' Sipek } 24394d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 244016f18200SJosef 'Jeff' Sipek 24413c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24423c95f0dcSAl Viro int len, struct qstr *this) 24433c95f0dcSAl Viro { 24443c95f0dcSAl Viro this->name = name; 24453c95f0dcSAl Viro this->len = len; 24463c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24473c95f0dcSAl Viro if (!len) 24483c95f0dcSAl Viro return -EACCES; 24493c95f0dcSAl Viro 24503c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24513c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24523c95f0dcSAl Viro return -EACCES; 24533c95f0dcSAl Viro } 24543c95f0dcSAl Viro 24553c95f0dcSAl Viro while (len--) { 24563c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24573c95f0dcSAl Viro if (c == '/' || c == '\0') 24583c95f0dcSAl Viro return -EACCES; 24593c95f0dcSAl Viro } 24603c95f0dcSAl Viro /* 24613c95f0dcSAl Viro * See if the low-level filesystem might want 24623c95f0dcSAl Viro * to use its own hash.. 24633c95f0dcSAl Viro */ 24643c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 24653c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 24663c95f0dcSAl Viro if (err < 0) 24673c95f0dcSAl Viro return err; 24683c95f0dcSAl Viro } 24693c95f0dcSAl Viro 24703c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 24713c95f0dcSAl Viro } 24723c95f0dcSAl Viro 2473eead1911SChristoph Hellwig /** 24740da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 24750da0b7fdSDavid Howells * @name: pathname component to lookup 24760da0b7fdSDavid Howells * @base: base directory to lookup from 24770da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 24780da0b7fdSDavid Howells * 24790da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 24800da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 24810da0b7fdSDavid Howells * 24820da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 24830da0b7fdSDavid Howells * not be called by generic code. 24840da0b7fdSDavid Howells * 24850da0b7fdSDavid Howells * The caller must hold base->i_mutex. 24860da0b7fdSDavid Howells */ 24870da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 24880da0b7fdSDavid Howells { 24890da0b7fdSDavid Howells struct qstr this; 24900da0b7fdSDavid Howells int err; 24910da0b7fdSDavid Howells 24920da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 24930da0b7fdSDavid Howells 24940da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 24950da0b7fdSDavid Howells if (err) 24960da0b7fdSDavid Howells return ERR_PTR(err); 24970da0b7fdSDavid Howells 24980da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 24990da0b7fdSDavid Howells } 25000da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25010da0b7fdSDavid Howells 25020da0b7fdSDavid Howells /** 2503a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2504eead1911SChristoph Hellwig * @name: pathname component to lookup 2505eead1911SChristoph Hellwig * @base: base directory to lookup from 2506eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2507eead1911SChristoph Hellwig * 2508a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25099e7543e9SAl Viro * not be called by generic code. 2510bbddca8eSNeilBrown * 2511bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2512eead1911SChristoph Hellwig */ 2513057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2514057f6c01SJames Morris { 25158613a209SAl Viro struct dentry *dentry; 2516057f6c01SJames Morris struct qstr this; 2517cda309deSMiklos Szeredi int err; 2518057f6c01SJames Morris 25195955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25202f9092e1SDavid Woodhouse 25213c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2522cda309deSMiklos Szeredi if (err) 2523cda309deSMiklos Szeredi return ERR_PTR(err); 2524cda309deSMiklos Szeredi 25258613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25268613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2527057f6c01SJames Morris } 25284d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2529057f6c01SJames Morris 2530bbddca8eSNeilBrown /** 2531bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2532bbddca8eSNeilBrown * @name: pathname component to lookup 2533bbddca8eSNeilBrown * @base: base directory to lookup from 2534bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2535bbddca8eSNeilBrown * 2536bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2537bbddca8eSNeilBrown * not be called by generic code. 2538bbddca8eSNeilBrown * 2539bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2540bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2541bbddca8eSNeilBrown */ 2542bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2543bbddca8eSNeilBrown struct dentry *base, int len) 2544bbddca8eSNeilBrown { 2545bbddca8eSNeilBrown struct qstr this; 2546bbddca8eSNeilBrown int err; 254720d00ee8SLinus Torvalds struct dentry *ret; 2548bbddca8eSNeilBrown 25493c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2550bbddca8eSNeilBrown if (err) 2551bbddca8eSNeilBrown return ERR_PTR(err); 2552bbddca8eSNeilBrown 255320d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 255420d00ee8SLinus Torvalds if (!ret) 255520d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 255620d00ee8SLinus Torvalds return ret; 2557bbddca8eSNeilBrown } 2558bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2559bbddca8eSNeilBrown 2560eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2561eedf265aSEric W. Biederman int path_pts(struct path *path) 2562eedf265aSEric W. Biederman { 2563eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2564eedf265aSEric W. Biederman * the input path. 2565eedf265aSEric W. Biederman */ 2566eedf265aSEric W. Biederman struct dentry *child, *parent; 2567eedf265aSEric W. Biederman struct qstr this; 2568eedf265aSEric W. Biederman int ret; 2569eedf265aSEric W. Biederman 2570eedf265aSEric W. Biederman ret = path_parent_directory(path); 2571eedf265aSEric W. Biederman if (ret) 2572eedf265aSEric W. Biederman return ret; 2573eedf265aSEric W. Biederman 2574eedf265aSEric W. Biederman parent = path->dentry; 2575eedf265aSEric W. Biederman this.name = "pts"; 2576eedf265aSEric W. Biederman this.len = 3; 2577eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2578eedf265aSEric W. Biederman if (!child) 2579eedf265aSEric W. Biederman return -ENOENT; 2580eedf265aSEric W. Biederman 2581eedf265aSEric W. Biederman path->dentry = child; 2582eedf265aSEric W. Biederman dput(parent); 2583eedf265aSEric W. Biederman follow_mount(path); 2584eedf265aSEric W. Biederman return 0; 2585eedf265aSEric W. Biederman } 2586eedf265aSEric W. Biederman #endif 2587eedf265aSEric W. Biederman 25881fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 25891fa1e7f6SAndy Whitcroft struct path *path, int *empty) 25901da177e4SLinus Torvalds { 2591abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2592abc9f5beSAl Viro flags, path, NULL); 25931da177e4SLinus Torvalds } 2594b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 25951fa1e7f6SAndy Whitcroft 25968033426eSJeff Layton /** 2597197df04cSAl Viro * mountpoint_last - look up last component for umount 25988033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 25998033426eSJeff Layton * 26008033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 26018033426eSJeff Layton * need to resolve the path without doing any revalidation. 26028033426eSJeff Layton * 26038033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 26048033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 26058033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 26068033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 26078033426eSJeff Layton * bogus and it doesn't exist. 26088033426eSJeff Layton * 26098033426eSJeff Layton * Returns: 26108033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2611ba8f4613SAl Viro * lookup found a negative dentry. 26128033426eSJeff Layton * 2613ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2614ba8f4613SAl Viro * symlink that needs to be followed. 26158033426eSJeff Layton * 26168033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2617ba8f4613SAl Viro * that needs to be followed. 26188033426eSJeff Layton */ 26198033426eSJeff Layton static int 2620ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 26218033426eSJeff Layton { 26228033426eSJeff Layton int error = 0; 26238033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2624ba8f4613SAl Viro struct path path; 26258033426eSJeff Layton 262635759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 262735759521SAl Viro if (nd->flags & LOOKUP_RCU) { 26284675ac39SAl Viro if (unlazy_walk(nd)) 2629deb106c6SAl Viro return -ECHILD; 26308033426eSJeff Layton } 26318033426eSJeff Layton 26328033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 26338033426eSJeff Layton 26348033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 26358033426eSJeff Layton error = handle_dots(nd, nd->last_type); 263635759521SAl Viro if (error) 2637deb106c6SAl Viro return error; 2638ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2639949a852eSAl Viro } else { 2640ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2641ba8f4613SAl Viro if (!path.dentry) { 26428033426eSJeff Layton /* 2643949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2644949a852eSAl Viro * cache, so that means that this dentry is probably 2645949a852eSAl Viro * a symlink or the path doesn't actually point 2646949a852eSAl Viro * to a mounted dentry. 26478033426eSJeff Layton */ 2648ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2649949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2650ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2651ba8f4613SAl Viro return PTR_ERR(path.dentry); 26528033426eSJeff Layton } 2653bcceeebaSDave Jones } 2654ba8f4613SAl Viro if (d_is_negative(path.dentry)) { 2655ba8f4613SAl Viro dput(path.dentry); 2656deb106c6SAl Viro return -ENOENT; 265735759521SAl Viro } 2658ba8f4613SAl Viro path.mnt = nd->path.mnt; 26598f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 26608033426eSJeff Layton } 26618033426eSJeff Layton 26628033426eSJeff Layton /** 2663197df04cSAl Viro * path_mountpoint - look up a path to be umounted 26642a78b857SMasanari Iida * @nd: lookup context 26658033426eSJeff Layton * @flags: lookup flags 2666c8a53ee5SAl Viro * @path: pointer to container for result 26678033426eSJeff Layton * 26688033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2669606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 26708033426eSJeff Layton */ 26718033426eSJeff Layton static int 2672c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 26738033426eSJeff Layton { 2674c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2675368ee9baSAl Viro int err; 26769b5858e9SAl Viro 26773bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2678ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 26793bdba28bSAl Viro s = trailing_symlink(nd); 26803bdba28bSAl Viro } 2681ba8f4613SAl Viro if (!err) { 2682ba8f4613SAl Viro *path = nd->path; 2683ba8f4613SAl Viro nd->path.mnt = NULL; 2684ba8f4613SAl Viro nd->path.dentry = NULL; 2685ba8f4613SAl Viro follow_mount(path); 2686ba8f4613SAl Viro } 2687deb106c6SAl Viro terminate_walk(nd); 26888033426eSJeff Layton return err; 26898033426eSJeff Layton } 26908033426eSJeff Layton 26912d864651SAl Viro static int 2692668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 26932d864651SAl Viro unsigned int flags) 26942d864651SAl Viro { 26959883d185SAl Viro struct nameidata nd; 2696cbaab2dbSAl Viro int error; 2697668696dcSAl Viro if (IS_ERR(name)) 2698668696dcSAl Viro return PTR_ERR(name); 26999883d185SAl Viro set_nameidata(&nd, dfd, name); 2700c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 27012d864651SAl Viro if (unlikely(error == -ECHILD)) 2702c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 27032d864651SAl Viro if (unlikely(error == -ESTALE)) 2704c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 27052d864651SAl Viro if (likely(!error)) 2706c9b07eabSAl Viro audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL); 27079883d185SAl Viro restore_nameidata(); 2708668696dcSAl Viro putname(name); 27092d864651SAl Viro return error; 27102d864651SAl Viro } 27112d864651SAl Viro 27128033426eSJeff Layton /** 2713197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 27148033426eSJeff Layton * @dfd: directory file descriptor 27158033426eSJeff Layton * @name: pathname from userland 27168033426eSJeff Layton * @flags: lookup flags 27178033426eSJeff Layton * @path: pointer to container to hold result 27188033426eSJeff Layton * 27198033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27208033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27218033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27228033426eSJeff Layton * and avoid revalidating the last component. 27238033426eSJeff Layton * 27248033426eSJeff Layton * Returns 0 and populates "path" on success. 27258033426eSJeff Layton */ 27268033426eSJeff Layton int 2727197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27288033426eSJeff Layton struct path *path) 27298033426eSJeff Layton { 2730cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27318033426eSJeff Layton } 27328033426eSJeff Layton 27332d864651SAl Viro int 27342d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27352d864651SAl Viro unsigned int flags) 27362d864651SAl Viro { 2737cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27382d864651SAl Viro } 27392d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27402d864651SAl Viro 2741cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27421da177e4SLinus Torvalds { 27438e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2744da9592edSDavid Howells 27458e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27461da177e4SLinus Torvalds return 0; 27478e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27481da177e4SLinus Torvalds return 0; 274923adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27501da177e4SLinus Torvalds } 2751cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27521da177e4SLinus Torvalds 27531da177e4SLinus Torvalds /* 27541da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27551da177e4SLinus Torvalds * whether the type of victim is right. 27561da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27571da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27581da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27591da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27601da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27611da177e4SLinus Torvalds * a. be owner of dir, or 27621da177e4SLinus Torvalds * b. be owner of victim, or 27631da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27641da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27651da177e4SLinus Torvalds * links pointing to it. 27660bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27670bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27680bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27690bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27700bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27711da177e4SLinus Torvalds * nfs_async_unlink(). 27721da177e4SLinus Torvalds */ 2773b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27741da177e4SLinus Torvalds { 277563afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27761da177e4SLinus Torvalds int error; 27771da177e4SLinus Torvalds 2778b18825a7SDavid Howells if (d_is_negative(victim)) 27791da177e4SLinus Torvalds return -ENOENT; 2780b18825a7SDavid Howells BUG_ON(!inode); 27811da177e4SLinus Torvalds 27821da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2783593d1ce8SEric W. Biederman 2784593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2785593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2786593d1ce8SEric W. Biederman return -EOVERFLOW; 2787593d1ce8SEric W. Biederman 27884fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27891da177e4SLinus Torvalds 2790f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27911da177e4SLinus Torvalds if (error) 27921da177e4SLinus Torvalds return error; 27931da177e4SLinus Torvalds if (IS_APPEND(dir)) 27941da177e4SLinus Torvalds return -EPERM; 2795b18825a7SDavid Howells 2796b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27970bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27981da177e4SLinus Torvalds return -EPERM; 27991da177e4SLinus Torvalds if (isdir) { 280044b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28011da177e4SLinus Torvalds return -ENOTDIR; 28021da177e4SLinus Torvalds if (IS_ROOT(victim)) 28031da177e4SLinus Torvalds return -EBUSY; 280444b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28051da177e4SLinus Torvalds return -EISDIR; 28061da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28071da177e4SLinus Torvalds return -ENOENT; 28081da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28091da177e4SLinus Torvalds return -EBUSY; 28101da177e4SLinus Torvalds return 0; 28111da177e4SLinus Torvalds } 28121da177e4SLinus Torvalds 28131da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28141da177e4SLinus Torvalds * dir. 28151da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28161da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28171da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2818036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2819036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2820036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28211da177e4SLinus Torvalds */ 2822a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28231da177e4SLinus Torvalds { 2824036d5236SEric W. Biederman struct user_namespace *s_user_ns; 282514e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28261da177e4SLinus Torvalds if (child->d_inode) 28271da177e4SLinus Torvalds return -EEXIST; 28281da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28291da177e4SLinus Torvalds return -ENOENT; 2830036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2831036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2832036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2833036d5236SEric W. Biederman return -EOVERFLOW; 2834f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28351da177e4SLinus Torvalds } 28361da177e4SLinus Torvalds 28371da177e4SLinus Torvalds /* 28381da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28391da177e4SLinus Torvalds */ 28401da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28411da177e4SLinus Torvalds { 28421da177e4SLinus Torvalds struct dentry *p; 28431da177e4SLinus Torvalds 28441da177e4SLinus Torvalds if (p1 == p2) { 28455955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28461da177e4SLinus Torvalds return NULL; 28471da177e4SLinus Torvalds } 28481da177e4SLinus Torvalds 2849fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28501da177e4SLinus Torvalds 2851e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2852e2761a11SOGAWA Hirofumi if (p) { 28535955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28545955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28551da177e4SLinus Torvalds return p; 28561da177e4SLinus Torvalds } 28571da177e4SLinus Torvalds 2858e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2859e2761a11SOGAWA Hirofumi if (p) { 28605955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28615955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28621da177e4SLinus Torvalds return p; 28631da177e4SLinus Torvalds } 28641da177e4SLinus Torvalds 28655955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28665955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28671da177e4SLinus Torvalds return NULL; 28681da177e4SLinus Torvalds } 28694d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28701da177e4SLinus Torvalds 28711da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28721da177e4SLinus Torvalds { 28735955102cSAl Viro inode_unlock(p1->d_inode); 28741da177e4SLinus Torvalds if (p1 != p2) { 28755955102cSAl Viro inode_unlock(p2->d_inode); 2876fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28771da177e4SLinus Torvalds } 28781da177e4SLinus Torvalds } 28794d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28801da177e4SLinus Torvalds 28814acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2882312b63fbSAl Viro bool want_excl) 28831da177e4SLinus Torvalds { 2884a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28851da177e4SLinus Torvalds if (error) 28861da177e4SLinus Torvalds return error; 28871da177e4SLinus Torvalds 2888acfa4380SAl Viro if (!dir->i_op->create) 28891da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28901da177e4SLinus Torvalds mode &= S_IALLUGO; 28911da177e4SLinus Torvalds mode |= S_IFREG; 28921da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28931da177e4SLinus Torvalds if (error) 28941da177e4SLinus Torvalds return error; 2895312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2896a74574aaSStephen Smalley if (!error) 2897f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28981da177e4SLinus Torvalds return error; 28991da177e4SLinus Torvalds } 29004d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29011da177e4SLinus Torvalds 29028e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29038e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29048e6c848eSAl Viro void *arg) 29058e6c848eSAl Viro { 29068e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 29078e6c848eSAl Viro int error = may_create(dir, dentry); 29088e6c848eSAl Viro if (error) 29098e6c848eSAl Viro return error; 29108e6c848eSAl Viro 29118e6c848eSAl Viro mode &= S_IALLUGO; 29128e6c848eSAl Viro mode |= S_IFREG; 29138e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29148e6c848eSAl Viro if (error) 29158e6c848eSAl Viro return error; 29168e6c848eSAl Viro error = f(dentry, mode, arg); 29178e6c848eSAl Viro if (!error) 29188e6c848eSAl Viro fsnotify_create(dir, dentry); 29198e6c848eSAl Viro return error; 29208e6c848eSAl Viro } 29218e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29228e6c848eSAl Viro 2923a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2924a2982cc9SEric W. Biederman { 2925a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2926a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2927a2982cc9SEric W. Biederman } 2928a2982cc9SEric W. Biederman 2929f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 29301da177e4SLinus Torvalds { 29313fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29321da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29331da177e4SLinus Torvalds int error; 29341da177e4SLinus Torvalds 29351da177e4SLinus Torvalds if (!inode) 29361da177e4SLinus Torvalds return -ENOENT; 29371da177e4SLinus Torvalds 2938c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2939c8fe8f30SChristoph Hellwig case S_IFLNK: 29401da177e4SLinus Torvalds return -ELOOP; 2941c8fe8f30SChristoph Hellwig case S_IFDIR: 2942c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29431da177e4SLinus Torvalds return -EISDIR; 2944c8fe8f30SChristoph Hellwig break; 2945c8fe8f30SChristoph Hellwig case S_IFBLK: 2946c8fe8f30SChristoph Hellwig case S_IFCHR: 2947a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29481da177e4SLinus Torvalds return -EACCES; 2949c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2950c8fe8f30SChristoph Hellwig case S_IFIFO: 2951c8fe8f30SChristoph Hellwig case S_IFSOCK: 29521da177e4SLinus Torvalds flag &= ~O_TRUNC; 2953c8fe8f30SChristoph Hellwig break; 29544a3fd211SDave Hansen } 2955b41572e9SDave Hansen 295662fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2957b41572e9SDave Hansen if (error) 2958b41572e9SDave Hansen return error; 29596146f0d5SMimi Zohar 29601da177e4SLinus Torvalds /* 29611da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29621da177e4SLinus Torvalds */ 29631da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29648737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29657715b521SAl Viro return -EPERM; 29661da177e4SLinus Torvalds if (flag & O_TRUNC) 29677715b521SAl Viro return -EPERM; 29681da177e4SLinus Torvalds } 29691da177e4SLinus Torvalds 29701da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29712e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29727715b521SAl Viro return -EPERM; 29731da177e4SLinus Torvalds 2974f3c7691eSJ. Bruce Fields return 0; 29757715b521SAl Viro } 29767715b521SAl Viro 2977e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29787715b521SAl Viro { 2979f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29807715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29817715b521SAl Viro int error = get_write_access(inode); 29821da177e4SLinus Torvalds if (error) 29837715b521SAl Viro return error; 29841da177e4SLinus Torvalds /* 29851da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29861da177e4SLinus Torvalds */ 2987d7a06983SJeff Layton error = locks_verify_locked(filp); 2988be6d3e56SKentaro Takeda if (!error) 2989ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29901da177e4SLinus Torvalds if (!error) { 29917715b521SAl Viro error = do_truncate(path->dentry, 0, 2992d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2993e1181ee6SJeff Layton filp); 29941da177e4SLinus Torvalds } 29951da177e4SLinus Torvalds put_write_access(inode); 2996acd0c935SMimi Zohar return error; 29971da177e4SLinus Torvalds } 29981da177e4SLinus Torvalds 2999d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3000d57999e1SDave Hansen { 30018a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30028a5e929dSAl Viro flag--; 3003d57999e1SDave Hansen return flag; 3004d57999e1SDave Hansen } 3005d57999e1SDave Hansen 3006d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 3007d18e9008SMiklos Szeredi { 30081328c727SSeth Forshee struct user_namespace *s_user_ns; 3009d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3010d18e9008SMiklos Szeredi if (error) 3011d18e9008SMiklos Szeredi return error; 3012d18e9008SMiklos Szeredi 30131328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 30141328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 30151328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 30161328c727SSeth Forshee return -EOVERFLOW; 30171328c727SSeth Forshee 3018d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 3019d18e9008SMiklos Szeredi if (error) 3020d18e9008SMiklos Szeredi return error; 3021d18e9008SMiklos Szeredi 3022d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3023d18e9008SMiklos Szeredi } 3024d18e9008SMiklos Szeredi 30251acf0af9SDavid Howells /* 30261acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30271acf0af9SDavid Howells * dentry. 30281acf0af9SDavid Howells * 30291acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30301acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30311acf0af9SDavid Howells * 303200a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 303300a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 303400a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 30351acf0af9SDavid Howells * 30361acf0af9SDavid Howells * Returns an error code otherwise. 30371acf0af9SDavid Howells */ 30382675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 303930d90494SAl Viro struct path *path, struct file *file, 3040015c3bbcSMiklos Szeredi const struct open_flags *op, 30413ec2eef1SAl Viro int open_flag, umode_t mode) 3042d18e9008SMiklos Szeredi { 3043d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3044d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3045d18e9008SMiklos Szeredi int error; 3046d18e9008SMiklos Szeredi 3047384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3048d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3049d18e9008SMiklos Szeredi 3050d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3051d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3052d18e9008SMiklos Szeredi 305330d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 305430d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30550fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 305644907d79SAl Viro open_to_namei_flags(open_flag), mode); 30576fbd0714SAl Viro d_lookup_done(dentry); 3058384f26e2SAl Viro if (!error) { 305964e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 3060d18e9008SMiklos Szeredi /* 3061384f26e2SAl Viro * We didn't have the inode before the open, so check open 3062384f26e2SAl Viro * permission here. 3063d18e9008SMiklos Szeredi */ 3064384f26e2SAl Viro int acc_mode = op->acc_mode; 306573a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 306603da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 306703da633aSAl Viro fsnotify_create(dir, dentry); 306862fb4a15SAl Viro acc_mode = 0; 306903da633aSAl Viro } 30702675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3071384f26e2SAl Viro if (WARN_ON(error > 0)) 3072384f26e2SAl Viro error = -EINVAL; 307364e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3074d18e9008SMiklos Szeredi error = -EIO; 3075384f26e2SAl Viro } else { 3076d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3077d18e9008SMiklos Szeredi dput(dentry); 3078d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 307910c64ceaSAl Viro } 308073a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 3081d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3082a01e718fSAl Viro if (unlikely(d_is_negative(dentry))) { 3083a01e718fSAl Viro error = -ENOENT; 3084a01e718fSAl Viro } else { 3085d18e9008SMiklos Szeredi path->dentry = dentry; 3086d18e9008SMiklos Szeredi path->mnt = nd->path.mnt; 308700a07c15SAl Viro return 0; 3088d18e9008SMiklos Szeredi } 3089d18e9008SMiklos Szeredi } 3090a01e718fSAl Viro } 3091d18e9008SMiklos Szeredi dput(dentry); 3092d18e9008SMiklos Szeredi return error; 3093d18e9008SMiklos Szeredi } 3094d18e9008SMiklos Szeredi 309531e6b01fSNick Piggin /* 30961acf0af9SDavid Howells * Look up and maybe create and open the last component. 3097d58ffd35SMiklos Szeredi * 309800a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3099d58ffd35SMiklos Szeredi * 310000a07c15SAl Viro * Returns 0 on success, that is, if 310100a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 310200a07c15SAl Viro * the file was not completely opened at this time, though lookups and 310300a07c15SAl Viro * creations were performed. 310400a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 310500a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 310600a07c15SAl Viro * hadn't been specified. 31071acf0af9SDavid Howells * 310800a07c15SAl Viro * An error code is returned on failure. 3109d58ffd35SMiklos Szeredi */ 31102675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 311130d90494SAl Viro struct file *file, 3112d58ffd35SMiklos Szeredi const struct open_flags *op, 31133ec2eef1SAl Viro bool got_write) 3114d58ffd35SMiklos Szeredi { 3115d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 311654ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31171643b43fSAl Viro int open_flag = op->open_flag; 3118d58ffd35SMiklos Szeredi struct dentry *dentry; 31191643b43fSAl Viro int error, create_error = 0; 31201643b43fSAl Viro umode_t mode = op->mode; 31216fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3122d58ffd35SMiklos Szeredi 3123ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3124ce8644fcSAl Viro return -ENOENT; 3125d58ffd35SMiklos Szeredi 312673a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 31276fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31286fbd0714SAl Viro for (;;) { 31296fbd0714SAl Viro if (!dentry) { 31306fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3131d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 31322675a4ebSAl Viro return PTR_ERR(dentry); 31336fbd0714SAl Viro } 31346fbd0714SAl Viro if (d_in_lookup(dentry)) 31356fbd0714SAl Viro break; 3136d58ffd35SMiklos Szeredi 31376fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31386fbd0714SAl Viro if (likely(error > 0)) 31396fbd0714SAl Viro break; 31406fbd0714SAl Viro if (error) 31416fbd0714SAl Viro goto out_dput; 31426fbd0714SAl Viro d_invalidate(dentry); 31436fbd0714SAl Viro dput(dentry); 31446fbd0714SAl Viro dentry = NULL; 31456fbd0714SAl Viro } 31466fbd0714SAl Viro if (dentry->d_inode) { 3147d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3148d18e9008SMiklos Szeredi goto out_no_open; 31496c51e513SAl Viro } 3150d18e9008SMiklos Szeredi 31511643b43fSAl Viro /* 31521643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31531643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31541643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31551643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31561643b43fSAl Viro * 31571643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31581643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31591643b43fSAl Viro */ 31601643b43fSAl Viro if (open_flag & O_CREAT) { 31611643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31621643b43fSAl Viro mode &= ~current_umask(); 31631643b43fSAl Viro if (unlikely(!got_write)) { 31641643b43fSAl Viro create_error = -EROFS; 31651643b43fSAl Viro open_flag &= ~O_CREAT; 31661643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31671643b43fSAl Viro goto no_open; 31681643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31691643b43fSAl Viro } else { 31701643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31711643b43fSAl Viro if (create_error) { 31721643b43fSAl Viro open_flag &= ~O_CREAT; 31731643b43fSAl Viro if (open_flag & O_EXCL) 31741643b43fSAl Viro goto no_open; 31751643b43fSAl Viro } 31761643b43fSAl Viro } 31771643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31781643b43fSAl Viro unlikely(!got_write)) { 31791643b43fSAl Viro /* 31801643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31811643b43fSAl Viro * back to lookup + open 31821643b43fSAl Viro */ 31831643b43fSAl Viro goto no_open; 3184d18e9008SMiklos Szeredi } 3185d18e9008SMiklos Szeredi 31861643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 31871643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 31883ec2eef1SAl Viro mode); 31891643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 31901643b43fSAl Viro error = create_error; 31911643b43fSAl Viro return error; 31921643b43fSAl Viro } 319354ef4872SMiklos Szeredi 31941643b43fSAl Viro no_open: 31956fbd0714SAl Viro if (d_in_lookup(dentry)) { 319612fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 319712fa5e24SAl Viro nd->flags); 31986fbd0714SAl Viro d_lookup_done(dentry); 319912fa5e24SAl Viro if (unlikely(res)) { 320012fa5e24SAl Viro if (IS_ERR(res)) { 320112fa5e24SAl Viro error = PTR_ERR(res); 320212fa5e24SAl Viro goto out_dput; 320312fa5e24SAl Viro } 320412fa5e24SAl Viro dput(dentry); 320512fa5e24SAl Viro dentry = res; 320612fa5e24SAl Viro } 320754ef4872SMiklos Szeredi } 320854ef4872SMiklos Szeredi 3209d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32101643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 321173a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3212ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3213ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3214ce8644fcSAl Viro error = -EACCES; 3215d58ffd35SMiklos Szeredi goto out_dput; 321664894cf8SAl Viro } 3217ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 32181643b43fSAl Viro open_flag & O_EXCL); 3219d58ffd35SMiklos Szeredi if (error) 3220d58ffd35SMiklos Szeredi goto out_dput; 3221ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3222d58ffd35SMiklos Szeredi } 32231643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32241643b43fSAl Viro error = create_error; 3225d58ffd35SMiklos Szeredi goto out_dput; 3226d58ffd35SMiklos Szeredi } 3227d18e9008SMiklos Szeredi out_no_open: 3228d58ffd35SMiklos Szeredi path->dentry = dentry; 3229d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 323000a07c15SAl Viro return 0; 3231d58ffd35SMiklos Szeredi 3232d58ffd35SMiklos Szeredi out_dput: 3233d58ffd35SMiklos Szeredi dput(dentry); 32342675a4ebSAl Viro return error; 3235d58ffd35SMiklos Szeredi } 3236d58ffd35SMiklos Szeredi 3237d58ffd35SMiklos Szeredi /* 3238fe2d35ffSAl Viro * Handle the last step of open() 323931e6b01fSNick Piggin */ 3240896475d5SAl Viro static int do_last(struct nameidata *nd, 32413ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3242fb1cc555SAl Viro { 3243a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3244ca344a89SAl Viro int open_flag = op->open_flag; 324577d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 324664894cf8SAl Viro bool got_write = false; 3247bcda7652SAl Viro int acc_mode = op->acc_mode; 3248254cf582SAl Viro unsigned seq; 3249a1eb3315SMiklos Szeredi struct inode *inode; 3250896475d5SAl Viro struct path path; 325116c2cd71SAl Viro int error; 3252fb1cc555SAl Viro 3253c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3254c3e380b0SAl Viro nd->flags |= op->intent; 3255c3e380b0SAl Viro 3256bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3257fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3258deb106c6SAl Viro if (unlikely(error)) 32592675a4ebSAl Viro return error; 3260e83db167SMiklos Szeredi goto finish_open; 32611f36f774SAl Viro } 3262a2c36b45SAl Viro 3263ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3264fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3265fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3266fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3267254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3268e9742b53SAl Viro if (likely(error > 0)) 326971574865SMiklos Szeredi goto finish_lookup; 327071574865SMiklos Szeredi 3271ce57dfc1SAl Viro if (error < 0) 3272deb106c6SAl Viro return error; 3273a1eb3315SMiklos Szeredi 327437d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32756583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3276b6183df7SMiklos Szeredi } else { 3277fe2d35ffSAl Viro /* create side of things */ 3278a3fbbde7SAl Viro /* 3279b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3280b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3281b6183df7SMiklos Szeredi * about to look up 3282a3fbbde7SAl Viro */ 32839f1fafeeSAl Viro error = complete_walk(nd); 3284e8bb73dfSAl Viro if (error) 32852675a4ebSAl Viro return error; 3286fe2d35ffSAl Viro 3287c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 32881f36f774SAl Viro /* trailing slashes? */ 3289deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3290deb106c6SAl Viro return -EISDIR; 3291b6183df7SMiklos Szeredi } 32921f36f774SAl Viro 32939cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 329464894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 329564894cf8SAl Viro if (!error) 329664894cf8SAl Viro got_write = true; 329764894cf8SAl Viro /* 329864894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 329964894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 330064894cf8SAl Viro * dropping this one anyway. 330164894cf8SAl Viro */ 330264894cf8SAl Viro } 33039cf843e3SAl Viro if (open_flag & O_CREAT) 33045955102cSAl Viro inode_lock(dir->d_inode); 33059cf843e3SAl Viro else 33069cf843e3SAl Viro inode_lock_shared(dir->d_inode); 33073ec2eef1SAl Viro error = lookup_open(nd, &path, file, op, got_write); 33089cf843e3SAl Viro if (open_flag & O_CREAT) 33095955102cSAl Viro inode_unlock(dir->d_inode); 33109cf843e3SAl Viro else 33119cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3312fb1cc555SAl Viro 33132675a4ebSAl Viro if (error) 3314d58ffd35SMiklos Szeredi goto out; 33156c0d46c4SAl Viro 331600a07c15SAl Viro if (file->f_mode & FMODE_OPENED) { 331773a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3318496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 331977d660a8SMiklos Szeredi will_truncate = false; 3320d18e9008SMiklos Szeredi 332176ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3322d18e9008SMiklos Szeredi goto opened; 3323d18e9008SMiklos Szeredi } 3324d18e9008SMiklos Szeredi 332573a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 33269b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3327ca344a89SAl Viro open_flag &= ~O_TRUNC; 332877d660a8SMiklos Szeredi will_truncate = false; 332962fb4a15SAl Viro acc_mode = 0; 3330896475d5SAl Viro path_to_nameidata(&path, nd); 3331e83db167SMiklos Szeredi goto finish_open_created; 3332fb1cc555SAl Viro } 3333fb1cc555SAl Viro 3334fb1cc555SAl Viro /* 3335d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3336d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3337d18e9008SMiklos Szeredi * necessary...) 3338d18e9008SMiklos Szeredi */ 333964894cf8SAl Viro if (got_write) { 3340d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 334164894cf8SAl Viro got_write = false; 3342d18e9008SMiklos Szeredi } 3343d18e9008SMiklos Szeredi 3344e6ec03a2SAl Viro error = follow_managed(&path, nd); 3345e6ec03a2SAl Viro if (unlikely(error < 0)) 3346e6ec03a2SAl Viro return error; 3347e6ec03a2SAl Viro 33486583fe22SAl Viro /* 33496583fe22SAl Viro * create/update audit record if it already exists. 33506583fe22SAl Viro */ 33516583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 33526583fe22SAl Viro 3353deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3354deb106c6SAl Viro path_to_nameidata(&path, nd); 3355deb106c6SAl Viro return -EEXIST; 3356deb106c6SAl Viro } 3357fb1cc555SAl Viro 3358254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3359d4565649SAl Viro inode = d_backing_inode(path.dentry); 3360766c4cbfSAl Viro finish_lookup: 33618f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3362deb106c6SAl Viro if (unlikely(error)) 3363d63ff28fSAl Viro return error; 3364bc77daa7SAl Viro finish_open: 33658f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3366a3fbbde7SAl Viro error = complete_walk(nd); 3367fac7d191SAl Viro if (error) 33682675a4ebSAl Viro return error; 336976ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 337030aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3371fb1cc555SAl Viro error = -EISDIR; 337230aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 33732675a4ebSAl Viro goto out; 337430aba665SSalvatore Mesoraca error = may_create_in_sticky(dir, 337530aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 337630aba665SSalvatore Mesoraca if (unlikely(error)) 337730aba665SSalvatore Mesoraca goto out; 337830aba665SSalvatore Mesoraca } 3379af2f5542SMiklos Szeredi error = -ENOTDIR; 338044b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 33812675a4ebSAl Viro goto out; 33824bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 338377d660a8SMiklos Szeredi will_truncate = false; 33846c0d46c4SAl Viro 33850f9d1a10SAl Viro if (will_truncate) { 33860f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33870f9d1a10SAl Viro if (error) 33882675a4ebSAl Viro goto out; 338964894cf8SAl Viro got_write = true; 33900f9d1a10SAl Viro } 3391e83db167SMiklos Szeredi finish_open_created: 3392bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3393ca344a89SAl Viro if (error) 33942675a4ebSAl Viro goto out; 3395aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3396ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3397fac7d191SAl Viro if (error) 3398015c3bbcSMiklos Szeredi goto out; 3399a8277b9bSMiklos Szeredi opened: 34006035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3401fe9ec829SAl Viro if (!error && will_truncate) 34022675a4ebSAl Viro error = handle_truncate(file); 3403ca344a89SAl Viro out: 3404c80567c8SAl Viro if (unlikely(error > 0)) { 3405c80567c8SAl Viro WARN_ON(1); 3406c80567c8SAl Viro error = -EINVAL; 3407c80567c8SAl Viro } 340864894cf8SAl Viro if (got_write) 34090f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 34102675a4ebSAl Viro return error; 3411fb1cc555SAl Viro } 3412fb1cc555SAl Viro 3413af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3414af7bd4dcSAmir Goldstein { 3415af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3416af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3417af7bd4dcSAmir Goldstein struct inode *inode; 3418af7bd4dcSAmir Goldstein int error; 3419af7bd4dcSAmir Goldstein 3420af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3421af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3422af7bd4dcSAmir Goldstein if (error) 3423af7bd4dcSAmir Goldstein goto out_err; 3424af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3425af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3426af7bd4dcSAmir Goldstein goto out_err; 3427af7bd4dcSAmir Goldstein error = -ENOMEM; 3428cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3429af7bd4dcSAmir Goldstein if (unlikely(!child)) 3430af7bd4dcSAmir Goldstein goto out_err; 3431af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3432af7bd4dcSAmir Goldstein if (error) 3433af7bd4dcSAmir Goldstein goto out_err; 3434af7bd4dcSAmir Goldstein error = -ENOENT; 3435af7bd4dcSAmir Goldstein inode = child->d_inode; 3436af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3437af7bd4dcSAmir Goldstein goto out_err; 3438af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3439af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3440af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3441af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3442af7bd4dcSAmir Goldstein } 3443fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3444af7bd4dcSAmir Goldstein return child; 3445af7bd4dcSAmir Goldstein 3446af7bd4dcSAmir Goldstein out_err: 3447af7bd4dcSAmir Goldstein dput(child); 3448af7bd4dcSAmir Goldstein return ERR_PTR(error); 3449af7bd4dcSAmir Goldstein } 3450af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3451af7bd4dcSAmir Goldstein 3452c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 345360545d0dSAl Viro const struct open_flags *op, 34543ec2eef1SAl Viro struct file *file) 345560545d0dSAl Viro { 3456625b6d10SAl Viro struct dentry *child; 3457625b6d10SAl Viro struct path path; 3458c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 345960545d0dSAl Viro if (unlikely(error)) 346060545d0dSAl Viro return error; 3461625b6d10SAl Viro error = mnt_want_write(path.mnt); 346260545d0dSAl Viro if (unlikely(error)) 346360545d0dSAl Viro goto out; 3464af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3465af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3466684e73beSHirofumi Nakagawa if (IS_ERR(child)) 346760545d0dSAl Viro goto out2; 3468625b6d10SAl Viro dput(path.dentry); 3469625b6d10SAl Viro path.dentry = child; 3470c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 347169a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 347262fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 347360545d0dSAl Viro if (error) 347460545d0dSAl Viro goto out2; 3475625b6d10SAl Viro file->f_path.mnt = path.mnt; 3476be12af3eSAl Viro error = finish_open(file, child, NULL); 347760545d0dSAl Viro out2: 3478625b6d10SAl Viro mnt_drop_write(path.mnt); 347960545d0dSAl Viro out: 3480625b6d10SAl Viro path_put(&path); 348160545d0dSAl Viro return error; 348260545d0dSAl Viro } 348360545d0dSAl Viro 34846ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34856ac08709SAl Viro { 34866ac08709SAl Viro struct path path; 34876ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34886ac08709SAl Viro if (!error) { 34896ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3490ae2bb293SAl Viro error = vfs_open(&path, file); 34916ac08709SAl Viro path_put(&path); 34926ac08709SAl Viro } 34936ac08709SAl Viro return error; 34946ac08709SAl Viro } 34956ac08709SAl Viro 3496c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3497c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34981da177e4SLinus Torvalds { 349930d90494SAl Viro struct file *file; 350013aab428SAl Viro int error; 350131e6b01fSNick Piggin 3502ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35031afc99beSAl Viro if (IS_ERR(file)) 35041afc99beSAl Viro return file; 350531e6b01fSNick Piggin 3506bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 35073ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 35085f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 35096ac08709SAl Viro error = do_o_path(nd, flags, file); 35105f336e72SAl Viro } else { 35115f336e72SAl Viro const char *s = path_init(nd, flags); 35123bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 35133ec2eef1SAl Viro (error = do_last(nd, file, op)) > 0) { 351473d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35153bdba28bSAl Viro s = trailing_symlink(nd); 35163bdba28bSAl Viro } 3517deb106c6SAl Viro terminate_walk(nd); 35185f336e72SAl Viro } 35197c1c01ecSAl Viro if (likely(!error)) { 3520aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35217c1c01ecSAl Viro return file; 35227c1c01ecSAl Viro WARN_ON(1); 35237c1c01ecSAl Viro error = -EINVAL; 3524015c3bbcSMiklos Szeredi } 35257c1c01ecSAl Viro fput(file); 35262675a4ebSAl Viro if (error == -EOPENSTALE) { 35272675a4ebSAl Viro if (flags & LOOKUP_RCU) 35282675a4ebSAl Viro error = -ECHILD; 35292675a4ebSAl Viro else 35302675a4ebSAl Viro error = -ESTALE; 35312675a4ebSAl Viro } 35327c1c01ecSAl Viro return ERR_PTR(error); 3533de459215SKirill Korotaev } 35341da177e4SLinus Torvalds 3535669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3536f9652e10SAl Viro const struct open_flags *op) 353713aab428SAl Viro { 35389883d185SAl Viro struct nameidata nd; 3539f9652e10SAl Viro int flags = op->lookup_flags; 354013aab428SAl Viro struct file *filp; 354113aab428SAl Viro 35429883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3543c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 354413aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3545c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 354613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3547c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35489883d185SAl Viro restore_nameidata(); 354913aab428SAl Viro return filp; 355013aab428SAl Viro } 355113aab428SAl Viro 355273d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3553f9652e10SAl Viro const char *name, const struct open_flags *op) 355473d049a4SAl Viro { 35559883d185SAl Viro struct nameidata nd; 355673d049a4SAl Viro struct file *file; 355751689104SPaul Moore struct filename *filename; 3558f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 355973d049a4SAl Viro 356073d049a4SAl Viro nd.root.mnt = mnt; 356173d049a4SAl Viro nd.root.dentry = dentry; 356273d049a4SAl Viro 3563b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 356473d049a4SAl Viro return ERR_PTR(-ELOOP); 356573d049a4SAl Viro 356651689104SPaul Moore filename = getname_kernel(name); 3567a1c83681SViresh Kumar if (IS_ERR(filename)) 356851689104SPaul Moore return ERR_CAST(filename); 356951689104SPaul Moore 35709883d185SAl Viro set_nameidata(&nd, -1, filename); 3571c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 357273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3573c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 357473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3575c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35769883d185SAl Viro restore_nameidata(); 357751689104SPaul Moore putname(filename); 357873d049a4SAl Viro return file; 357973d049a4SAl Viro } 358073d049a4SAl Viro 3581fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35821ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35831da177e4SLinus Torvalds { 3584c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3585391172c4SAl Viro struct qstr last; 3586391172c4SAl Viro int type; 3587c30dabfeSJan Kara int err2; 35881ac12b4bSJeff Layton int error; 35891ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 35901ac12b4bSJeff Layton 35911ac12b4bSJeff Layton /* 35921ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 35931ac12b4bSJeff Layton * other flags passed in are ignored! 35941ac12b4bSJeff Layton */ 35951ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 35961ac12b4bSJeff Layton 35975c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 35985c31b6ceSAl Viro if (IS_ERR(name)) 35995c31b6ceSAl Viro return ERR_CAST(name); 36001da177e4SLinus Torvalds 3601c663e5d8SChristoph Hellwig /* 3602c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3603c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3604c663e5d8SChristoph Hellwig */ 36055c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3606ed75e95dSAl Viro goto out; 3607c663e5d8SChristoph Hellwig 3608c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3609391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3610c663e5d8SChristoph Hellwig /* 3611c663e5d8SChristoph Hellwig * Do the final lookup. 3612c663e5d8SChristoph Hellwig */ 3613391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36145955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3615391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36161da177e4SLinus Torvalds if (IS_ERR(dentry)) 3617a8104a9fSAl Viro goto unlock; 3618c663e5d8SChristoph Hellwig 3619a8104a9fSAl Viro error = -EEXIST; 3620b18825a7SDavid Howells if (d_is_positive(dentry)) 3621a8104a9fSAl Viro goto fail; 3622b18825a7SDavid Howells 3623c663e5d8SChristoph Hellwig /* 3624c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3625c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3626c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3627c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3628c663e5d8SChristoph Hellwig */ 3629391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3630a8104a9fSAl Viro error = -ENOENT; 3631ed75e95dSAl Viro goto fail; 3632e9baf6e5SAl Viro } 3633c30dabfeSJan Kara if (unlikely(err2)) { 3634c30dabfeSJan Kara error = err2; 3635a8104a9fSAl Viro goto fail; 3636c30dabfeSJan Kara } 3637181c37b6SAl Viro putname(name); 3638e9baf6e5SAl Viro return dentry; 36391da177e4SLinus Torvalds fail: 3640a8104a9fSAl Viro dput(dentry); 3641a8104a9fSAl Viro dentry = ERR_PTR(error); 3642a8104a9fSAl Viro unlock: 36435955102cSAl Viro inode_unlock(path->dentry->d_inode); 3644c30dabfeSJan Kara if (!err2) 3645391172c4SAl Viro mnt_drop_write(path->mnt); 3646ed75e95dSAl Viro out: 3647391172c4SAl Viro path_put(path); 3648181c37b6SAl Viro putname(name); 3649ed75e95dSAl Viro return dentry; 3650dae6ad8fSAl Viro } 3651fa14a0b8SAl Viro 3652fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3653fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3654fa14a0b8SAl Viro { 3655181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3656181c37b6SAl Viro path, lookup_flags); 3657fa14a0b8SAl Viro } 3658dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3659dae6ad8fSAl Viro 3660921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3661921a1650SAl Viro { 3662921a1650SAl Viro dput(dentry); 36635955102cSAl Viro inode_unlock(path->dentry->d_inode); 3664a8104a9fSAl Viro mnt_drop_write(path->mnt); 3665921a1650SAl Viro path_put(path); 3666921a1650SAl Viro } 3667921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3668921a1650SAl Viro 3669520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36701ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3671dae6ad8fSAl Viro { 3672181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3673dae6ad8fSAl Viro } 3674dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3675dae6ad8fSAl Viro 36761a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 36771da177e4SLinus Torvalds { 3678a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36791da177e4SLinus Torvalds 36801da177e4SLinus Torvalds if (error) 36811da177e4SLinus Torvalds return error; 36821da177e4SLinus Torvalds 368394f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 36841da177e4SLinus Torvalds return -EPERM; 36851da177e4SLinus Torvalds 3686acfa4380SAl Viro if (!dir->i_op->mknod) 36871da177e4SLinus Torvalds return -EPERM; 36881da177e4SLinus Torvalds 368908ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 369008ce5f16SSerge E. Hallyn if (error) 369108ce5f16SSerge E. Hallyn return error; 369208ce5f16SSerge E. Hallyn 36931da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 36941da177e4SLinus Torvalds if (error) 36951da177e4SLinus Torvalds return error; 36961da177e4SLinus Torvalds 36971da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3698a74574aaSStephen Smalley if (!error) 3699f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37001da177e4SLinus Torvalds return error; 37011da177e4SLinus Torvalds } 37024d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37031da177e4SLinus Torvalds 3704f69aac00SAl Viro static int may_mknod(umode_t mode) 3705463c3197SDave Hansen { 3706463c3197SDave Hansen switch (mode & S_IFMT) { 3707463c3197SDave Hansen case S_IFREG: 3708463c3197SDave Hansen case S_IFCHR: 3709463c3197SDave Hansen case S_IFBLK: 3710463c3197SDave Hansen case S_IFIFO: 3711463c3197SDave Hansen case S_IFSOCK: 3712463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3713463c3197SDave Hansen return 0; 3714463c3197SDave Hansen case S_IFDIR: 3715463c3197SDave Hansen return -EPERM; 3716463c3197SDave Hansen default: 3717463c3197SDave Hansen return -EINVAL; 3718463c3197SDave Hansen } 3719463c3197SDave Hansen } 3720463c3197SDave Hansen 372187c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 372287c4e192SDominik Brodowski unsigned int dev) 37231da177e4SLinus Torvalds { 37241da177e4SLinus Torvalds struct dentry *dentry; 3725dae6ad8fSAl Viro struct path path; 3726dae6ad8fSAl Viro int error; 3727972567f1SJeff Layton unsigned int lookup_flags = 0; 37281da177e4SLinus Torvalds 37298e4bfca1SAl Viro error = may_mknod(mode); 37308e4bfca1SAl Viro if (error) 37318e4bfca1SAl Viro return error; 3732972567f1SJeff Layton retry: 3733972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3734dae6ad8fSAl Viro if (IS_ERR(dentry)) 3735dae6ad8fSAl Viro return PTR_ERR(dentry); 37362ad94ae6SAl Viro 3737dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3738ce3b0f8dSAl Viro mode &= ~current_umask(); 3739dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3740be6d3e56SKentaro Takeda if (error) 3741a8104a9fSAl Viro goto out; 37421da177e4SLinus Torvalds switch (mode & S_IFMT) { 37431da177e4SLinus Torvalds case 0: case S_IFREG: 3744312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 374505d1a717SMimi Zohar if (!error) 374605d1a717SMimi Zohar ima_post_path_mknod(dentry); 37471da177e4SLinus Torvalds break; 37481da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3749dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37501da177e4SLinus Torvalds new_decode_dev(dev)); 37511da177e4SLinus Torvalds break; 37521da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3753dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37541da177e4SLinus Torvalds break; 37551da177e4SLinus Torvalds } 3756a8104a9fSAl Viro out: 3757921a1650SAl Viro done_path_create(&path, dentry); 3758972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3759972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3760972567f1SJeff Layton goto retry; 3761972567f1SJeff Layton } 37621da177e4SLinus Torvalds return error; 37631da177e4SLinus Torvalds } 37641da177e4SLinus Torvalds 376587c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 376687c4e192SDominik Brodowski unsigned int, dev) 376787c4e192SDominik Brodowski { 376887c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 376987c4e192SDominik Brodowski } 377087c4e192SDominik Brodowski 37718208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37725590ff0dSUlrich Drepper { 377387c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 37745590ff0dSUlrich Drepper } 37755590ff0dSUlrich Drepper 377618bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37771da177e4SLinus Torvalds { 3778a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37798de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37801da177e4SLinus Torvalds 37811da177e4SLinus Torvalds if (error) 37821da177e4SLinus Torvalds return error; 37831da177e4SLinus Torvalds 3784acfa4380SAl Viro if (!dir->i_op->mkdir) 37851da177e4SLinus Torvalds return -EPERM; 37861da177e4SLinus Torvalds 37871da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 37881da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 37891da177e4SLinus Torvalds if (error) 37901da177e4SLinus Torvalds return error; 37911da177e4SLinus Torvalds 37928de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 37938de52778SAl Viro return -EMLINK; 37948de52778SAl Viro 37951da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3796a74574aaSStephen Smalley if (!error) 3797f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 37981da177e4SLinus Torvalds return error; 37991da177e4SLinus Torvalds } 38004d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38011da177e4SLinus Torvalds 38020101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38031da177e4SLinus Torvalds { 38046902d925SDave Hansen struct dentry *dentry; 3805dae6ad8fSAl Viro struct path path; 3806dae6ad8fSAl Viro int error; 3807b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38081da177e4SLinus Torvalds 3809b76d8b82SJeff Layton retry: 3810b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38116902d925SDave Hansen if (IS_ERR(dentry)) 3812dae6ad8fSAl Viro return PTR_ERR(dentry); 38136902d925SDave Hansen 3814dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3815ce3b0f8dSAl Viro mode &= ~current_umask(); 3816dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3817a8104a9fSAl Viro if (!error) 3818dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3819921a1650SAl Viro done_path_create(&path, dentry); 3820b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3821b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3822b76d8b82SJeff Layton goto retry; 3823b76d8b82SJeff Layton } 38241da177e4SLinus Torvalds return error; 38251da177e4SLinus Torvalds } 38261da177e4SLinus Torvalds 38270101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38280101db7aSDominik Brodowski { 38290101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38300101db7aSDominik Brodowski } 38310101db7aSDominik Brodowski 3832a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38335590ff0dSUlrich Drepper { 38340101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38355590ff0dSUlrich Drepper } 38365590ff0dSUlrich Drepper 38371da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38381da177e4SLinus Torvalds { 38391da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38401da177e4SLinus Torvalds 38411da177e4SLinus Torvalds if (error) 38421da177e4SLinus Torvalds return error; 38431da177e4SLinus Torvalds 3844acfa4380SAl Viro if (!dir->i_op->rmdir) 38451da177e4SLinus Torvalds return -EPERM; 38461da177e4SLinus Torvalds 38471d2ef590SAl Viro dget(dentry); 38485955102cSAl Viro inode_lock(dentry->d_inode); 3849912dbc15SSage Weil 38501da177e4SLinus Torvalds error = -EBUSY; 38517af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3852912dbc15SSage Weil goto out; 3853912dbc15SSage Weil 38541da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3855912dbc15SSage Weil if (error) 3856912dbc15SSage Weil goto out; 3857912dbc15SSage Weil 38581da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3859912dbc15SSage Weil if (error) 3860912dbc15SSage Weil goto out; 3861912dbc15SSage Weil 38628767712fSAl Viro shrink_dcache_parent(dentry); 38631da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3864d83c49f3SAl Viro dont_mount(dentry); 38658ed936b5SEric W. Biederman detach_mounts(dentry); 3866116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 38671da177e4SLinus Torvalds 3868912dbc15SSage Weil out: 38695955102cSAl Viro inode_unlock(dentry->d_inode); 38701d2ef590SAl Viro dput(dentry); 3871912dbc15SSage Weil if (!error) 3872912dbc15SSage Weil d_delete(dentry); 38731da177e4SLinus Torvalds return error; 38741da177e4SLinus Torvalds } 38754d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38761da177e4SLinus Torvalds 3877f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 38781da177e4SLinus Torvalds { 38791da177e4SLinus Torvalds int error = 0; 388091a27b2aSJeff Layton struct filename *name; 38811da177e4SLinus Torvalds struct dentry *dentry; 3882f5beed75SAl Viro struct path path; 3883f5beed75SAl Viro struct qstr last; 3884f5beed75SAl Viro int type; 3885c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3886c6ee9206SJeff Layton retry: 3887c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3888c1d4dd27SAl Viro &path, &last, &type); 388991a27b2aSJeff Layton if (IS_ERR(name)) 389091a27b2aSJeff Layton return PTR_ERR(name); 38911da177e4SLinus Torvalds 3892f5beed75SAl Viro switch (type) { 38931da177e4SLinus Torvalds case LAST_DOTDOT: 38941da177e4SLinus Torvalds error = -ENOTEMPTY; 38951da177e4SLinus Torvalds goto exit1; 38961da177e4SLinus Torvalds case LAST_DOT: 38971da177e4SLinus Torvalds error = -EINVAL; 38981da177e4SLinus Torvalds goto exit1; 38991da177e4SLinus Torvalds case LAST_ROOT: 39001da177e4SLinus Torvalds error = -EBUSY; 39011da177e4SLinus Torvalds goto exit1; 39021da177e4SLinus Torvalds } 39030612d9fbSOGAWA Hirofumi 3904f5beed75SAl Viro error = mnt_want_write(path.mnt); 3905c30dabfeSJan Kara if (error) 3906c30dabfeSJan Kara goto exit1; 39070612d9fbSOGAWA Hirofumi 39085955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3909f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39101da177e4SLinus Torvalds error = PTR_ERR(dentry); 39116902d925SDave Hansen if (IS_ERR(dentry)) 39126902d925SDave Hansen goto exit2; 3913e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3914e6bc45d6STheodore Ts'o error = -ENOENT; 3915e6bc45d6STheodore Ts'o goto exit3; 3916e6bc45d6STheodore Ts'o } 3917f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3918be6d3e56SKentaro Takeda if (error) 3919c30dabfeSJan Kara goto exit3; 3920f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39210622753bSDave Hansen exit3: 39221da177e4SLinus Torvalds dput(dentry); 39236902d925SDave Hansen exit2: 39245955102cSAl Viro inode_unlock(path.dentry->d_inode); 3925f5beed75SAl Viro mnt_drop_write(path.mnt); 39261da177e4SLinus Torvalds exit1: 3927f5beed75SAl Viro path_put(&path); 39281da177e4SLinus Torvalds putname(name); 3929c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3930c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3931c6ee9206SJeff Layton goto retry; 3932c6ee9206SJeff Layton } 39331da177e4SLinus Torvalds return error; 39341da177e4SLinus Torvalds } 39351da177e4SLinus Torvalds 39363cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39375590ff0dSUlrich Drepper { 39385590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39395590ff0dSUlrich Drepper } 39405590ff0dSUlrich Drepper 3941b21996e3SJ. Bruce Fields /** 3942b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3943b21996e3SJ. Bruce Fields * @dir: parent directory 3944b21996e3SJ. Bruce Fields * @dentry: victim 3945b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3946b21996e3SJ. Bruce Fields * 3947b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3948b21996e3SJ. Bruce Fields * 3949b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3950b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3951b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3952b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3953b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3954b21996e3SJ. Bruce Fields * 3955b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3956b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3957b21996e3SJ. Bruce Fields * to be NFS exported. 3958b21996e3SJ. Bruce Fields */ 3959b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39601da177e4SLinus Torvalds { 39619accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39621da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39631da177e4SLinus Torvalds 39641da177e4SLinus Torvalds if (error) 39651da177e4SLinus Torvalds return error; 39661da177e4SLinus Torvalds 3967acfa4380SAl Viro if (!dir->i_op->unlink) 39681da177e4SLinus Torvalds return -EPERM; 39691da177e4SLinus Torvalds 39705955102cSAl Viro inode_lock(target); 39718ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39721da177e4SLinus Torvalds error = -EBUSY; 39731da177e4SLinus Torvalds else { 39741da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3975bec1052eSAl Viro if (!error) { 39765a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39775a14696cSJ. Bruce Fields if (error) 3978b21996e3SJ. Bruce Fields goto out; 39791da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39808ed936b5SEric W. Biederman if (!error) { 3981d83c49f3SAl Viro dont_mount(dentry); 39828ed936b5SEric W. Biederman detach_mounts(dentry); 3983116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 39848ed936b5SEric W. Biederman } 3985bec1052eSAl Viro } 39861da177e4SLinus Torvalds } 3987b21996e3SJ. Bruce Fields out: 39885955102cSAl Viro inode_unlock(target); 39891da177e4SLinus Torvalds 39901da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 39911da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 39929accbb97SJ. Bruce Fields fsnotify_link_count(target); 39931da177e4SLinus Torvalds d_delete(dentry); 39941da177e4SLinus Torvalds } 39950eeca283SRobert Love 39961da177e4SLinus Torvalds return error; 39971da177e4SLinus Torvalds } 39984d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 39991da177e4SLinus Torvalds 40001da177e4SLinus Torvalds /* 40011da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40021b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40031da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40041da177e4SLinus Torvalds * while waiting on the I/O. 40051da177e4SLinus Torvalds */ 4006da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40071da177e4SLinus Torvalds { 40082ad94ae6SAl Viro int error; 40091da177e4SLinus Torvalds struct dentry *dentry; 4010f5beed75SAl Viro struct path path; 4011f5beed75SAl Viro struct qstr last; 4012f5beed75SAl Viro int type; 40131da177e4SLinus Torvalds struct inode *inode = NULL; 4014b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40155d18f813SJeff Layton unsigned int lookup_flags = 0; 40165d18f813SJeff Layton retry: 4017da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 401891a27b2aSJeff Layton if (IS_ERR(name)) 401991a27b2aSJeff Layton return PTR_ERR(name); 40202ad94ae6SAl Viro 40211da177e4SLinus Torvalds error = -EISDIR; 4022f5beed75SAl Viro if (type != LAST_NORM) 40231da177e4SLinus Torvalds goto exit1; 40240612d9fbSOGAWA Hirofumi 4025f5beed75SAl Viro error = mnt_want_write(path.mnt); 4026c30dabfeSJan Kara if (error) 4027c30dabfeSJan Kara goto exit1; 4028b21996e3SJ. Bruce Fields retry_deleg: 40295955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4030f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40311da177e4SLinus Torvalds error = PTR_ERR(dentry); 40321da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40331da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4034f5beed75SAl Viro if (last.name[last.len]) 403550338b88STörök Edwin goto slashes; 40361da177e4SLinus Torvalds inode = dentry->d_inode; 4037b18825a7SDavid Howells if (d_is_negative(dentry)) 4038e6bc45d6STheodore Ts'o goto slashes; 40397de9c6eeSAl Viro ihold(inode); 4040f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4041be6d3e56SKentaro Takeda if (error) 4042c30dabfeSJan Kara goto exit2; 4043f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40441da177e4SLinus Torvalds exit2: 40451da177e4SLinus Torvalds dput(dentry); 40461da177e4SLinus Torvalds } 40475955102cSAl Viro inode_unlock(path.dentry->d_inode); 40481da177e4SLinus Torvalds if (inode) 40491da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4050b21996e3SJ. Bruce Fields inode = NULL; 4051b21996e3SJ. Bruce Fields if (delegated_inode) { 40525a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4053b21996e3SJ. Bruce Fields if (!error) 4054b21996e3SJ. Bruce Fields goto retry_deleg; 4055b21996e3SJ. Bruce Fields } 4056f5beed75SAl Viro mnt_drop_write(path.mnt); 40571da177e4SLinus Torvalds exit1: 4058f5beed75SAl Viro path_put(&path); 40595d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40605d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40615d18f813SJeff Layton inode = NULL; 40625d18f813SJeff Layton goto retry; 40635d18f813SJeff Layton } 4064da2f1362SChristoph Hellwig putname(name); 40651da177e4SLinus Torvalds return error; 40661da177e4SLinus Torvalds 40671da177e4SLinus Torvalds slashes: 4068b18825a7SDavid Howells if (d_is_negative(dentry)) 4069b18825a7SDavid Howells error = -ENOENT; 407044b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4071b18825a7SDavid Howells error = -EISDIR; 4072b18825a7SDavid Howells else 4073b18825a7SDavid Howells error = -ENOTDIR; 40741da177e4SLinus Torvalds goto exit2; 40751da177e4SLinus Torvalds } 40761da177e4SLinus Torvalds 40772e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40785590ff0dSUlrich Drepper { 40795590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40805590ff0dSUlrich Drepper return -EINVAL; 40815590ff0dSUlrich Drepper 40825590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 40835590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 40845590ff0dSUlrich Drepper 4085da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 40865590ff0dSUlrich Drepper } 40875590ff0dSUlrich Drepper 40883480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 40895590ff0dSUlrich Drepper { 4090da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 40915590ff0dSUlrich Drepper } 40925590ff0dSUlrich Drepper 4093db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 40941da177e4SLinus Torvalds { 4095a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 40961da177e4SLinus Torvalds 40971da177e4SLinus Torvalds if (error) 40981da177e4SLinus Torvalds return error; 40991da177e4SLinus Torvalds 4100acfa4380SAl Viro if (!dir->i_op->symlink) 41011da177e4SLinus Torvalds return -EPERM; 41021da177e4SLinus Torvalds 41031da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41041da177e4SLinus Torvalds if (error) 41051da177e4SLinus Torvalds return error; 41061da177e4SLinus Torvalds 41071da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4108a74574aaSStephen Smalley if (!error) 4109f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41101da177e4SLinus Torvalds return error; 41111da177e4SLinus Torvalds } 41124d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41131da177e4SLinus Torvalds 4114b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 4115b724e846SDominik Brodowski const char __user *newname) 41161da177e4SLinus Torvalds { 41172ad94ae6SAl Viro int error; 411891a27b2aSJeff Layton struct filename *from; 41196902d925SDave Hansen struct dentry *dentry; 4120dae6ad8fSAl Viro struct path path; 4121f46d3567SJeff Layton unsigned int lookup_flags = 0; 41221da177e4SLinus Torvalds 41231da177e4SLinus Torvalds from = getname(oldname); 41241da177e4SLinus Torvalds if (IS_ERR(from)) 41251da177e4SLinus Torvalds return PTR_ERR(from); 4126f46d3567SJeff Layton retry: 4127f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41281da177e4SLinus Torvalds error = PTR_ERR(dentry); 41296902d925SDave Hansen if (IS_ERR(dentry)) 4130dae6ad8fSAl Viro goto out_putname; 41316902d925SDave Hansen 413291a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4133a8104a9fSAl Viro if (!error) 413491a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4135921a1650SAl Viro done_path_create(&path, dentry); 4136f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4137f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4138f46d3567SJeff Layton goto retry; 4139f46d3567SJeff Layton } 41406902d925SDave Hansen out_putname: 41411da177e4SLinus Torvalds putname(from); 41421da177e4SLinus Torvalds return error; 41431da177e4SLinus Torvalds } 41441da177e4SLinus Torvalds 4145b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4146b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4147b724e846SDominik Brodowski { 4148b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4149b724e846SDominik Brodowski } 4150b724e846SDominik Brodowski 41513480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41525590ff0dSUlrich Drepper { 4153b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 41545590ff0dSUlrich Drepper } 41555590ff0dSUlrich Drepper 4156146a8595SJ. Bruce Fields /** 4157146a8595SJ. Bruce Fields * vfs_link - create a new link 4158146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4159146a8595SJ. Bruce Fields * @dir: new parent 4160146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4161146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4162146a8595SJ. Bruce Fields * 4163146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4164146a8595SJ. Bruce Fields * 4165146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4166146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4167146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4168146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4169146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4170146a8595SJ. Bruce Fields * 4171146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4172146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4173146a8595SJ. Bruce Fields * to be NFS exported. 4174146a8595SJ. Bruce Fields */ 4175146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41761da177e4SLinus Torvalds { 41771da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41788de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41791da177e4SLinus Torvalds int error; 41801da177e4SLinus Torvalds 41811da177e4SLinus Torvalds if (!inode) 41821da177e4SLinus Torvalds return -ENOENT; 41831da177e4SLinus Torvalds 4184a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 41851da177e4SLinus Torvalds if (error) 41861da177e4SLinus Torvalds return error; 41871da177e4SLinus Torvalds 41881da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 41891da177e4SLinus Torvalds return -EXDEV; 41901da177e4SLinus Torvalds 41911da177e4SLinus Torvalds /* 41921da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 41931da177e4SLinus Torvalds */ 41941da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 41951da177e4SLinus Torvalds return -EPERM; 41960bd23d09SEric W. Biederman /* 41970bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 41980bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 41990bd23d09SEric W. Biederman * the vfs. 42000bd23d09SEric W. Biederman */ 42010bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 42020bd23d09SEric W. Biederman return -EPERM; 4203acfa4380SAl Viro if (!dir->i_op->link) 42041da177e4SLinus Torvalds return -EPERM; 42057e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42061da177e4SLinus Torvalds return -EPERM; 42071da177e4SLinus Torvalds 42081da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42091da177e4SLinus Torvalds if (error) 42101da177e4SLinus Torvalds return error; 42111da177e4SLinus Torvalds 42125955102cSAl Viro inode_lock(inode); 4213aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4214f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4215aae8a97dSAneesh Kumar K.V error = -ENOENT; 42168de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42178de52778SAl Viro error = -EMLINK; 4218146a8595SJ. Bruce Fields else { 4219146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4220146a8595SJ. Bruce Fields if (!error) 42211da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4222146a8595SJ. Bruce Fields } 4223f4e0c30cSAl Viro 4224f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4225f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4226f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4227f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4228f4e0c30cSAl Viro } 42295955102cSAl Viro inode_unlock(inode); 4230e31e14ecSStephen Smalley if (!error) 42317e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42321da177e4SLinus Torvalds return error; 42331da177e4SLinus Torvalds } 42344d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42351da177e4SLinus Torvalds 42361da177e4SLinus Torvalds /* 42371da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42381da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42391da177e4SLinus Torvalds * newname. --KAB 42401da177e4SLinus Torvalds * 42411da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42421da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42431da177e4SLinus Torvalds * and other special files. --ADM 42441da177e4SLinus Torvalds */ 424546ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 424646ea89ebSDominik Brodowski const char __user *newname, int flags) 42471da177e4SLinus Torvalds { 42481da177e4SLinus Torvalds struct dentry *new_dentry; 4249dae6ad8fSAl Viro struct path old_path, new_path; 4250146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 425111a7b371SAneesh Kumar K.V int how = 0; 42521da177e4SLinus Torvalds int error; 42531da177e4SLinus Torvalds 425411a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4255c04030e1SUlrich Drepper return -EINVAL; 425611a7b371SAneesh Kumar K.V /* 4257f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4258f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4259f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 426011a7b371SAneesh Kumar K.V */ 4261f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4262f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4263f0cc6ffbSLinus Torvalds return -ENOENT; 426411a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4265f0cc6ffbSLinus Torvalds } 4266c04030e1SUlrich Drepper 426711a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 426811a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4269442e31caSJeff Layton retry: 427011a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42711da177e4SLinus Torvalds if (error) 42722ad94ae6SAl Viro return error; 42732ad94ae6SAl Viro 4274442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4275442e31caSJeff Layton (how & LOOKUP_REVAL)); 42761da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42776902d925SDave Hansen if (IS_ERR(new_dentry)) 4278dae6ad8fSAl Viro goto out; 4279dae6ad8fSAl Viro 4280dae6ad8fSAl Viro error = -EXDEV; 4281dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4282dae6ad8fSAl Viro goto out_dput; 4283800179c9SKees Cook error = may_linkat(&old_path); 4284800179c9SKees Cook if (unlikely(error)) 4285800179c9SKees Cook goto out_dput; 4286dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4287be6d3e56SKentaro Takeda if (error) 4288a8104a9fSAl Viro goto out_dput; 4289146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 429075c3f29dSDave Hansen out_dput: 4291921a1650SAl Viro done_path_create(&new_path, new_dentry); 4292146a8595SJ. Bruce Fields if (delegated_inode) { 4293146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4294d22e6338SOleg Drokin if (!error) { 4295d22e6338SOleg Drokin path_put(&old_path); 4296146a8595SJ. Bruce Fields goto retry; 4297146a8595SJ. Bruce Fields } 4298d22e6338SOleg Drokin } 4299442e31caSJeff Layton if (retry_estale(error, how)) { 4300d22e6338SOleg Drokin path_put(&old_path); 4301442e31caSJeff Layton how |= LOOKUP_REVAL; 4302442e31caSJeff Layton goto retry; 4303442e31caSJeff Layton } 43041da177e4SLinus Torvalds out: 43052d8f3038SAl Viro path_put(&old_path); 43061da177e4SLinus Torvalds 43071da177e4SLinus Torvalds return error; 43081da177e4SLinus Torvalds } 43091da177e4SLinus Torvalds 431046ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 431146ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 431246ea89ebSDominik Brodowski { 431346ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 431446ea89ebSDominik Brodowski } 431546ea89ebSDominik Brodowski 43163480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43175590ff0dSUlrich Drepper { 431846ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43195590ff0dSUlrich Drepper } 43205590ff0dSUlrich Drepper 4321bc27027aSMiklos Szeredi /** 4322bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4323bc27027aSMiklos Szeredi * @old_dir: parent of source 4324bc27027aSMiklos Szeredi * @old_dentry: source 4325bc27027aSMiklos Szeredi * @new_dir: parent of destination 4326bc27027aSMiklos Szeredi * @new_dentry: destination 4327bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4328520c8b16SMiklos Szeredi * @flags: rename flags 4329bc27027aSMiklos Szeredi * 4330bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4331bc27027aSMiklos Szeredi * 4332bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4333bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4334bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4335bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4336bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4337bc27027aSMiklos Szeredi * so. 4338bc27027aSMiklos Szeredi * 4339bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4340bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4341bc27027aSMiklos Szeredi * to be NFS exported. 4342bc27027aSMiklos Szeredi * 43431da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43441da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43451da177e4SLinus Torvalds * Problems: 43460117d427SMauro Carvalho Chehab * 4347d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43481da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43491da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4350a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43511da177e4SLinus Torvalds * story. 43526cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43536cedba89SJ. Bruce Fields * and source (if it is not a directory). 43541b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43551da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43561da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4357a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43581da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43591da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43601da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4361a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43621da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43631da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 43641da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4365e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 43661b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 43671da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4368c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 43691da177e4SLinus Torvalds * locking]. 43701da177e4SLinus Torvalds */ 43711da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 43728e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4373520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 43741da177e4SLinus Torvalds { 43751da177e4SLinus Torvalds int error; 4376bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4377bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4378bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4379da1ce067SMiklos Szeredi bool new_is_dir = false; 4380da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 438149d31c2fSAl Viro struct name_snapshot old_name; 43821da177e4SLinus Torvalds 43838d3e2936SMiklos Szeredi if (source == target) 43841da177e4SLinus Torvalds return 0; 43851da177e4SLinus Torvalds 43861da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 43871da177e4SLinus Torvalds if (error) 43881da177e4SLinus Torvalds return error; 43891da177e4SLinus Torvalds 4390da1ce067SMiklos Szeredi if (!target) { 4391a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4392da1ce067SMiklos Szeredi } else { 4393da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4394da1ce067SMiklos Szeredi 4395da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43961da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4397da1ce067SMiklos Szeredi else 4398da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4399da1ce067SMiklos Szeredi } 44001da177e4SLinus Torvalds if (error) 44011da177e4SLinus Torvalds return error; 44021da177e4SLinus Torvalds 44032773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 44041da177e4SLinus Torvalds return -EPERM; 44051da177e4SLinus Torvalds 4406bc27027aSMiklos Szeredi /* 4407bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4408bc27027aSMiklos Szeredi * we'll need to flip '..'. 4409bc27027aSMiklos Szeredi */ 4410da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4411da1ce067SMiklos Szeredi if (is_dir) { 4412bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4413bc27027aSMiklos Szeredi if (error) 4414bc27027aSMiklos Szeredi return error; 4415bc27027aSMiklos Szeredi } 4416da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4417da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4418da1ce067SMiklos Szeredi if (error) 4419da1ce067SMiklos Szeredi return error; 4420da1ce067SMiklos Szeredi } 4421da1ce067SMiklos Szeredi } 44220eeca283SRobert Love 44230b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44240b3974ebSMiklos Szeredi flags); 4425bc27027aSMiklos Szeredi if (error) 4426bc27027aSMiklos Szeredi return error; 4427bc27027aSMiklos Szeredi 442849d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4429bc27027aSMiklos Szeredi dget(new_dentry); 4430da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4431bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4432bc27027aSMiklos Szeredi else if (target) 44335955102cSAl Viro inode_lock(target); 4434bc27027aSMiklos Szeredi 4435bc27027aSMiklos Szeredi error = -EBUSY; 44367af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4437bc27027aSMiklos Szeredi goto out; 4438bc27027aSMiklos Szeredi 4439da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4440bc27027aSMiklos Szeredi error = -EMLINK; 4441da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4442bc27027aSMiklos Szeredi goto out; 4443da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4444da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4445da1ce067SMiklos Szeredi goto out; 4446da1ce067SMiklos Szeredi } 4447da1ce067SMiklos Szeredi if (!is_dir) { 4448bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4449bc27027aSMiklos Szeredi if (error) 4450bc27027aSMiklos Szeredi goto out; 4451da1ce067SMiklos Szeredi } 4452da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4453bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4454bc27027aSMiklos Szeredi if (error) 4455bc27027aSMiklos Szeredi goto out; 4456bc27027aSMiklos Szeredi } 4457520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4458520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4459bc27027aSMiklos Szeredi if (error) 4460bc27027aSMiklos Szeredi goto out; 4461bc27027aSMiklos Szeredi 4462da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44638767712fSAl Viro if (is_dir) { 44648767712fSAl Viro shrink_dcache_parent(new_dentry); 4465bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 44668767712fSAl Viro } 4467bc27027aSMiklos Szeredi dont_mount(new_dentry); 44688ed936b5SEric W. Biederman detach_mounts(new_dentry); 4469bc27027aSMiklos Szeredi } 4470da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4471da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4472bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4473da1ce067SMiklos Szeredi else 4474da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4475da1ce067SMiklos Szeredi } 4476bc27027aSMiklos Szeredi out: 4477da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4478bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4479bc27027aSMiklos Szeredi else if (target) 44805955102cSAl Viro inode_unlock(target); 4481bc27027aSMiklos Szeredi dput(new_dentry); 4482da1ce067SMiklos Szeredi if (!error) { 4483f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4484da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4485da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4486f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4487da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4488da1ce067SMiklos Szeredi } 4489da1ce067SMiklos Szeredi } 449049d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 44910eeca283SRobert Love 44921da177e4SLinus Torvalds return error; 44931da177e4SLinus Torvalds } 44944d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 44951da177e4SLinus Torvalds 4496ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4497ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 44981da177e4SLinus Torvalds { 44991da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 45001da177e4SLinus Torvalds struct dentry *trap; 4501f5beed75SAl Viro struct path old_path, new_path; 4502f5beed75SAl Viro struct qstr old_last, new_last; 4503f5beed75SAl Viro int old_type, new_type; 45048e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 450591a27b2aSJeff Layton struct filename *from; 450691a27b2aSJeff Layton struct filename *to; 4507f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4508c6a94284SJeff Layton bool should_retry = false; 45092ad94ae6SAl Viro int error; 4510520c8b16SMiklos Szeredi 45110d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4512da1ce067SMiklos Szeredi return -EINVAL; 4513da1ce067SMiklos Szeredi 45140d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45150d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4516520c8b16SMiklos Szeredi return -EINVAL; 4517520c8b16SMiklos Szeredi 45180d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45190d7a8555SMiklos Szeredi return -EPERM; 45200d7a8555SMiklos Szeredi 4521f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4522f5beed75SAl Viro target_flags = 0; 4523f5beed75SAl Viro 4524c6a94284SJeff Layton retry: 4525c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4526c1d4dd27SAl Viro &old_path, &old_last, &old_type); 452791a27b2aSJeff Layton if (IS_ERR(from)) { 452891a27b2aSJeff Layton error = PTR_ERR(from); 45291da177e4SLinus Torvalds goto exit; 453091a27b2aSJeff Layton } 45311da177e4SLinus Torvalds 4532c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4533c1d4dd27SAl Viro &new_path, &new_last, &new_type); 453491a27b2aSJeff Layton if (IS_ERR(to)) { 453591a27b2aSJeff Layton error = PTR_ERR(to); 45361da177e4SLinus Torvalds goto exit1; 453791a27b2aSJeff Layton } 45381da177e4SLinus Torvalds 45391da177e4SLinus Torvalds error = -EXDEV; 4540f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45411da177e4SLinus Torvalds goto exit2; 45421da177e4SLinus Torvalds 45431da177e4SLinus Torvalds error = -EBUSY; 4544f5beed75SAl Viro if (old_type != LAST_NORM) 45451da177e4SLinus Torvalds goto exit2; 45461da177e4SLinus Torvalds 45470a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45480a7c3937SMiklos Szeredi error = -EEXIST; 4549f5beed75SAl Viro if (new_type != LAST_NORM) 45501da177e4SLinus Torvalds goto exit2; 45511da177e4SLinus Torvalds 4552f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4553c30dabfeSJan Kara if (error) 4554c30dabfeSJan Kara goto exit2; 4555c30dabfeSJan Kara 45568e6d782cSJ. Bruce Fields retry_deleg: 4557f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45581da177e4SLinus Torvalds 4559f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45601da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45611da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45621da177e4SLinus Torvalds goto exit3; 45631da177e4SLinus Torvalds /* source must exist */ 45641da177e4SLinus Torvalds error = -ENOENT; 4565b18825a7SDavid Howells if (d_is_negative(old_dentry)) 45661da177e4SLinus Torvalds goto exit4; 4567f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 45681da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45691da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 45701da177e4SLinus Torvalds goto exit4; 45710a7c3937SMiklos Szeredi error = -EEXIST; 45720a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 45730a7c3937SMiklos Szeredi goto exit5; 4574da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4575da1ce067SMiklos Szeredi error = -ENOENT; 4576da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4577da1ce067SMiklos Szeredi goto exit5; 4578da1ce067SMiklos Szeredi 4579da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4580da1ce067SMiklos Szeredi error = -ENOTDIR; 4581f5beed75SAl Viro if (new_last.name[new_last.len]) 4582da1ce067SMiklos Szeredi goto exit5; 4583da1ce067SMiklos Szeredi } 4584da1ce067SMiklos Szeredi } 45850a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 45860a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 45870a7c3937SMiklos Szeredi error = -ENOTDIR; 4588f5beed75SAl Viro if (old_last.name[old_last.len]) 45890a7c3937SMiklos Szeredi goto exit5; 4590f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 45910a7c3937SMiklos Szeredi goto exit5; 45920a7c3937SMiklos Szeredi } 45930a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 45940a7c3937SMiklos Szeredi error = -EINVAL; 45950a7c3937SMiklos Szeredi if (old_dentry == trap) 45960a7c3937SMiklos Szeredi goto exit5; 45971da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4598da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 45991da177e4SLinus Torvalds error = -ENOTEMPTY; 46001da177e4SLinus Torvalds if (new_dentry == trap) 46011da177e4SLinus Torvalds goto exit5; 46021da177e4SLinus Torvalds 4603f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4604f5beed75SAl Viro &new_path, new_dentry, flags); 4605be6d3e56SKentaro Takeda if (error) 4606c30dabfeSJan Kara goto exit5; 4607f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4608f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4609520c8b16SMiklos Szeredi &delegated_inode, flags); 46101da177e4SLinus Torvalds exit5: 46111da177e4SLinus Torvalds dput(new_dentry); 46121da177e4SLinus Torvalds exit4: 46131da177e4SLinus Torvalds dput(old_dentry); 46141da177e4SLinus Torvalds exit3: 4615f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46168e6d782cSJ. Bruce Fields if (delegated_inode) { 46178e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46188e6d782cSJ. Bruce Fields if (!error) 46198e6d782cSJ. Bruce Fields goto retry_deleg; 46208e6d782cSJ. Bruce Fields } 4621f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46221da177e4SLinus Torvalds exit2: 4623c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4624c6a94284SJeff Layton should_retry = true; 4625f5beed75SAl Viro path_put(&new_path); 46262ad94ae6SAl Viro putname(to); 46271da177e4SLinus Torvalds exit1: 4628f5beed75SAl Viro path_put(&old_path); 46291da177e4SLinus Torvalds putname(from); 4630c6a94284SJeff Layton if (should_retry) { 4631c6a94284SJeff Layton should_retry = false; 4632c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4633c6a94284SJeff Layton goto retry; 4634c6a94284SJeff Layton } 46352ad94ae6SAl Viro exit: 46361da177e4SLinus Torvalds return error; 46371da177e4SLinus Torvalds } 46381da177e4SLinus Torvalds 4639ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4640ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4641ee81feb6SDominik Brodowski { 4642ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4643ee81feb6SDominik Brodowski } 4644ee81feb6SDominik Brodowski 4645520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4646520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4647520c8b16SMiklos Szeredi { 4648ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4649520c8b16SMiklos Szeredi } 4650520c8b16SMiklos Szeredi 4651a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 46525590ff0dSUlrich Drepper { 4653ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 46545590ff0dSUlrich Drepper } 46555590ff0dSUlrich Drepper 4656787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4657787fb6bcSMiklos Szeredi { 4658787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4659787fb6bcSMiklos Szeredi if (error) 4660787fb6bcSMiklos Szeredi return error; 4661787fb6bcSMiklos Szeredi 4662787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4663787fb6bcSMiklos Szeredi return -EPERM; 4664787fb6bcSMiklos Szeredi 4665787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4666787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4667787fb6bcSMiklos Szeredi } 4668787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4669787fb6bcSMiklos Szeredi 46705d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 46711da177e4SLinus Torvalds { 46725d826c84SAl Viro int len = PTR_ERR(link); 46731da177e4SLinus Torvalds if (IS_ERR(link)) 46741da177e4SLinus Torvalds goto out; 46751da177e4SLinus Torvalds 46761da177e4SLinus Torvalds len = strlen(link); 46771da177e4SLinus Torvalds if (len > (unsigned) buflen) 46781da177e4SLinus Torvalds len = buflen; 46791da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 46801da177e4SLinus Torvalds len = -EFAULT; 46811da177e4SLinus Torvalds out: 46821da177e4SLinus Torvalds return len; 46831da177e4SLinus Torvalds } 46841da177e4SLinus Torvalds 4685d60874cdSMiklos Szeredi /** 4686fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4687fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4688fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4689fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4690fd4a0edfSMiklos Szeredi * 4691fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4692fd4a0edfSMiklos Szeredi * 4693fd4a0edfSMiklos Szeredi * Does not call security hook. 4694fd4a0edfSMiklos Szeredi */ 4695fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4696fd4a0edfSMiklos Szeredi { 4697fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4698f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4699f2df5da6SAl Viro const char *link; 4700f2df5da6SAl Viro int res; 4701fd4a0edfSMiklos Szeredi 470276fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 470376fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 470476fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 470576fca90eSMiklos Szeredi 470676fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4707fd4a0edfSMiklos Szeredi return -EINVAL; 4708fd4a0edfSMiklos Szeredi 470976fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 471076fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 471176fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 471276fca90eSMiklos Szeredi } 471376fca90eSMiklos Szeredi 47144c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4715f2df5da6SAl Viro if (!link) { 4716f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4717f2df5da6SAl Viro if (IS_ERR(link)) 4718f2df5da6SAl Viro return PTR_ERR(link); 4719f2df5da6SAl Viro } 4720f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4721f2df5da6SAl Viro do_delayed_call(&done); 4722f2df5da6SAl Viro return res; 4723fd4a0edfSMiklos Szeredi } 4724fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 47251da177e4SLinus Torvalds 4726d60874cdSMiklos Szeredi /** 4727d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4728d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4729d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4730d60874cdSMiklos Szeredi * 4731d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4732d60874cdSMiklos Szeredi * 4733d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4734d60874cdSMiklos Szeredi * 4735d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4736d60874cdSMiklos Szeredi */ 4737d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4738d60874cdSMiklos Szeredi { 4739d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4740d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4741d60874cdSMiklos Szeredi 4742d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4743d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4744d60874cdSMiklos Szeredi if (!res) 4745d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4746d60874cdSMiklos Szeredi } 4747d60874cdSMiklos Szeredi return res; 4748d60874cdSMiklos Szeredi } 4749d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4750d60874cdSMiklos Szeredi 47511da177e4SLinus Torvalds /* get the link contents into pagecache */ 47526b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4753fceef393SAl Viro struct delayed_call *callback) 47541da177e4SLinus Torvalds { 4755ebd09abbSDuane Griffin char *kaddr; 47561da177e4SLinus Torvalds struct page *page; 47576b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47586b255391SAl Viro 4759d3883d4fSAl Viro if (!dentry) { 4760d3883d4fSAl Viro page = find_get_page(mapping, 0); 4761d3883d4fSAl Viro if (!page) 47626b255391SAl Viro return ERR_PTR(-ECHILD); 4763d3883d4fSAl Viro if (!PageUptodate(page)) { 4764d3883d4fSAl Viro put_page(page); 4765d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4766d3883d4fSAl Viro } 4767d3883d4fSAl Viro } else { 4768090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47691da177e4SLinus Torvalds if (IS_ERR(page)) 47706fe6900eSNick Piggin return (char*)page; 4771d3883d4fSAl Viro } 4772fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 477321fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 477421fc61c7SAl Viro kaddr = page_address(page); 47756b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4776ebd09abbSDuane Griffin return kaddr; 47771da177e4SLinus Torvalds } 47781da177e4SLinus Torvalds 47796b255391SAl Viro EXPORT_SYMBOL(page_get_link); 47801da177e4SLinus Torvalds 4781fceef393SAl Viro void page_put_link(void *arg) 47821da177e4SLinus Torvalds { 4783fceef393SAl Viro put_page(arg); 47841da177e4SLinus Torvalds } 47854d359507SAl Viro EXPORT_SYMBOL(page_put_link); 47861da177e4SLinus Torvalds 4787aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4788aa80deabSAl Viro { 4789fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47906b255391SAl Viro int res = readlink_copy(buffer, buflen, 47916b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4792fceef393SAl Viro &done)); 4793fceef393SAl Viro do_delayed_call(&done); 4794aa80deabSAl Viro return res; 4795aa80deabSAl Viro } 4796aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4797aa80deabSAl Viro 479854566b2cSNick Piggin /* 479954566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 480054566b2cSNick Piggin */ 480154566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48021da177e4SLinus Torvalds { 48031da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48040adb25d2SKirill Korotaev struct page *page; 4805afddba49SNick Piggin void *fsdata; 4806beb497abSDmitriy Monakhov int err; 4807c718a975STetsuo Handa unsigned int flags = 0; 480854566b2cSNick Piggin if (nofs) 480954566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 48101da177e4SLinus Torvalds 48117e53cac4SNeilBrown retry: 4812afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 481354566b2cSNick Piggin flags, &page, &fsdata); 48141da177e4SLinus Torvalds if (err) 4815afddba49SNick Piggin goto fail; 4816afddba49SNick Piggin 481721fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4818afddba49SNick Piggin 4819afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4820afddba49SNick Piggin page, fsdata); 48211da177e4SLinus Torvalds if (err < 0) 48221da177e4SLinus Torvalds goto fail; 4823afddba49SNick Piggin if (err < len-1) 4824afddba49SNick Piggin goto retry; 4825afddba49SNick Piggin 48261da177e4SLinus Torvalds mark_inode_dirty(inode); 48271da177e4SLinus Torvalds return 0; 48281da177e4SLinus Torvalds fail: 48291da177e4SLinus Torvalds return err; 48301da177e4SLinus Torvalds } 48314d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 48321da177e4SLinus Torvalds 48330adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 48340adb25d2SKirill Korotaev { 48350adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4836c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48370adb25d2SKirill Korotaev } 48384d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48390adb25d2SKirill Korotaev 484092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48416b255391SAl Viro .get_link = page_get_link, 48421da177e4SLinus Torvalds }; 48431da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4844