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> 421c949843SRasmus Villemoes #include <linux/build_bug.h> 431da177e4SLinus Torvalds 44e81e3f4dSEric Paris #include "internal.h" 45c7105365SAl Viro #include "mount.h" 46e81e3f4dSEric Paris 471da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 481da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 491da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 501da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 511da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 521da177e4SLinus Torvalds * 531da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 541da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 551da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 561da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 571da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 581da177e4SLinus Torvalds * the special cases of the former code. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 611da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 621da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 631da177e4SLinus Torvalds * 641da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 651da177e4SLinus Torvalds * resolution to correspond with current state of the code. 661da177e4SLinus Torvalds * 671da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 681da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 691da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 701da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 711da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 721da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 731da177e4SLinus Torvalds */ 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 761da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 771da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 781da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 791da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 801da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8125985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 821da177e4SLinus Torvalds * 831da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 841da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 851da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 861da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 871da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 881da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 891da177e4SLinus Torvalds * and in the old Linux semantics. 901da177e4SLinus Torvalds */ 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 931da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 941da177e4SLinus Torvalds * 951da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 961da177e4SLinus Torvalds */ 971da177e4SLinus Torvalds 981da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 991da177e4SLinus Torvalds * inside the path - always follow. 1001da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1011da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1021da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1031da177e4SLinus Torvalds * otherwise - don't follow. 1041da177e4SLinus Torvalds * (applied in that order). 1051da177e4SLinus Torvalds * 1061da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1071da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1081da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1091da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1101da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1111da177e4SLinus Torvalds */ 1121da177e4SLinus Torvalds /* 1131da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 114a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1151da177e4SLinus Torvalds * any extra contention... 1161da177e4SLinus Torvalds */ 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1191da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1201da177e4SLinus Torvalds * kernel data space before using them.. 1211da177e4SLinus Torvalds * 1221da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1231da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1241da177e4SLinus Torvalds */ 1257950e385SJeff Layton 126fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12791a27b2aSJeff Layton 12851f39a1fSDavid Drysdale struct filename * 12991a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13091a27b2aSJeff Layton { 13194b5d262SAl Viro struct filename *result; 1327950e385SJeff Layton char *kname; 13394b5d262SAl Viro int len; 1341c949843SRasmus Villemoes BUILD_BUG_ON(offsetof(struct filename, iname) % sizeof(long) != 0); 1351da177e4SLinus Torvalds 1367ac86265SJeff Layton result = audit_reusename(filename); 1377ac86265SJeff Layton if (result) 1387ac86265SJeff Layton return result; 1397ac86265SJeff Layton 1407950e385SJeff Layton result = __getname(); 1413f9f0aa6SLinus Torvalds if (unlikely(!result)) 1424043cde8SEric Paris return ERR_PTR(-ENOMEM); 1431da177e4SLinus Torvalds 1447950e385SJeff Layton /* 1457950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1467950e385SJeff Layton * allocation 1477950e385SJeff Layton */ 148fd2f7cb5SAl Viro kname = (char *)result->iname; 14991a27b2aSJeff Layton result->name = kname; 1507950e385SJeff Layton 15194b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15291a27b2aSJeff Layton if (unlikely(len < 0)) { 15394b5d262SAl Viro __putname(result); 15494b5d262SAl Viro return ERR_PTR(len); 15591a27b2aSJeff Layton } 1563f9f0aa6SLinus Torvalds 1577950e385SJeff Layton /* 1587950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1597950e385SJeff Layton * separate struct filename so we can dedicate the entire 1607950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1617950e385SJeff Layton * userland. 1627950e385SJeff Layton */ 16394b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 164fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1657950e385SJeff Layton kname = (char *)result; 1667950e385SJeff Layton 167fd2f7cb5SAl Viro /* 168fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 169fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 170fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 171fd2f7cb5SAl Viro */ 172fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17394b5d262SAl Viro if (unlikely(!result)) { 17494b5d262SAl Viro __putname(kname); 17594b5d262SAl Viro return ERR_PTR(-ENOMEM); 1767950e385SJeff Layton } 1777950e385SJeff Layton result->name = kname; 17894b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17994b5d262SAl Viro if (unlikely(len < 0)) { 18094b5d262SAl Viro __putname(kname); 18194b5d262SAl Viro kfree(result); 18294b5d262SAl Viro return ERR_PTR(len); 18394b5d262SAl Viro } 18494b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18594b5d262SAl Viro __putname(kname); 18694b5d262SAl Viro kfree(result); 18794b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18894b5d262SAl Viro } 1897950e385SJeff Layton } 1907950e385SJeff Layton 19194b5d262SAl Viro result->refcnt = 1; 1923f9f0aa6SLinus Torvalds /* The empty path is special. */ 1933f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1943f9f0aa6SLinus Torvalds if (empty) 1951fa1e7f6SAndy Whitcroft *empty = 1; 19694b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19794b5d262SAl Viro putname(result); 19894b5d262SAl Viro return ERR_PTR(-ENOENT); 1991da177e4SLinus Torvalds } 20094b5d262SAl Viro } 2017950e385SJeff Layton 2027950e385SJeff Layton result->uptr = filename; 203c4ad8f98SLinus Torvalds result->aname = NULL; 2041da177e4SLinus Torvalds audit_getname(result); 2051da177e4SLinus Torvalds return result; 2063f9f0aa6SLinus Torvalds } 2073f9f0aa6SLinus Torvalds 20891a27b2aSJeff Layton struct filename * 20991a27b2aSJeff Layton getname(const char __user * filename) 210f52e0c11SAl Viro { 211f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 212f52e0c11SAl Viro } 213f52e0c11SAl Viro 214c4ad8f98SLinus Torvalds struct filename * 215c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 216c4ad8f98SLinus Torvalds { 217c4ad8f98SLinus Torvalds struct filename *result; 21808518549SPaul Moore int len = strlen(filename) + 1; 219c4ad8f98SLinus Torvalds 220c4ad8f98SLinus Torvalds result = __getname(); 221c4ad8f98SLinus Torvalds if (unlikely(!result)) 222c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 223c4ad8f98SLinus Torvalds 22408518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 225fd2f7cb5SAl Viro result->name = (char *)result->iname; 22608518549SPaul Moore } else if (len <= PATH_MAX) { 22730ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 22808518549SPaul Moore struct filename *tmp; 22908518549SPaul Moore 23030ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 23108518549SPaul Moore if (unlikely(!tmp)) { 23208518549SPaul Moore __putname(result); 23308518549SPaul Moore return ERR_PTR(-ENOMEM); 23408518549SPaul Moore } 23508518549SPaul Moore tmp->name = (char *)result; 23608518549SPaul Moore result = tmp; 23708518549SPaul Moore } else { 23808518549SPaul Moore __putname(result); 23908518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 24008518549SPaul Moore } 24108518549SPaul Moore memcpy((char *)result->name, filename, len); 242c4ad8f98SLinus Torvalds result->uptr = NULL; 243c4ad8f98SLinus Torvalds result->aname = NULL; 24455422d0bSPaul Moore result->refcnt = 1; 245fd3522fdSPaul Moore audit_getname(result); 246c4ad8f98SLinus Torvalds 247c4ad8f98SLinus Torvalds return result; 248c4ad8f98SLinus Torvalds } 249c4ad8f98SLinus Torvalds 25091a27b2aSJeff Layton void putname(struct filename *name) 2511da177e4SLinus Torvalds { 25255422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 25355422d0bSPaul Moore 25455422d0bSPaul Moore if (--name->refcnt > 0) 25555422d0bSPaul Moore return; 25655422d0bSPaul Moore 257fd2f7cb5SAl Viro if (name->name != name->iname) { 25855422d0bSPaul Moore __putname(name->name); 25955422d0bSPaul Moore kfree(name); 26055422d0bSPaul Moore } else 26155422d0bSPaul Moore __putname(name); 2621da177e4SLinus Torvalds } 2631da177e4SLinus Torvalds 264e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 265e77819e5SLinus Torvalds { 26684635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 267e77819e5SLinus Torvalds struct posix_acl *acl; 268e77819e5SLinus Torvalds 269e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2703567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2713567866bSAl Viro if (!acl) 272e77819e5SLinus Torvalds return -EAGAIN; 2733567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 274b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 275e77819e5SLinus Torvalds return -ECHILD; 276206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 277e77819e5SLinus Torvalds } 278e77819e5SLinus Torvalds 2792982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2804e34e719SChristoph Hellwig if (IS_ERR(acl)) 2814e34e719SChristoph Hellwig return PTR_ERR(acl); 282e77819e5SLinus Torvalds if (acl) { 283e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 284e77819e5SLinus Torvalds posix_acl_release(acl); 285e77819e5SLinus Torvalds return error; 286e77819e5SLinus Torvalds } 28784635d68SLinus Torvalds #endif 288e77819e5SLinus Torvalds 289e77819e5SLinus Torvalds return -EAGAIN; 290e77819e5SLinus Torvalds } 291e77819e5SLinus Torvalds 2925909ccaaSLinus Torvalds /* 293948409c7SAndreas Gruenbacher * This does the basic permission checking 2945909ccaaSLinus Torvalds */ 2957e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2965909ccaaSLinus Torvalds { 29726cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2985909ccaaSLinus Torvalds 2998e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 3005909ccaaSLinus Torvalds mode >>= 6; 3015909ccaaSLinus Torvalds else { 302e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 3037e40145eSAl Viro int error = check_acl(inode, mask); 3045909ccaaSLinus Torvalds if (error != -EAGAIN) 3055909ccaaSLinus Torvalds return error; 3065909ccaaSLinus Torvalds } 3075909ccaaSLinus Torvalds 3085909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3095909ccaaSLinus Torvalds mode >>= 3; 3105909ccaaSLinus Torvalds } 3115909ccaaSLinus Torvalds 3125909ccaaSLinus Torvalds /* 3135909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3145909ccaaSLinus Torvalds */ 3159c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3165909ccaaSLinus Torvalds return 0; 3175909ccaaSLinus Torvalds return -EACCES; 3185909ccaaSLinus Torvalds } 3191da177e4SLinus Torvalds 3201da177e4SLinus Torvalds /** 3211da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3221da177e4SLinus Torvalds * @inode: inode to check access rights for 3238fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3241da177e4SLinus Torvalds * 3251da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3261da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3271da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 328b74c79e9SNick Piggin * are used for other things. 329b74c79e9SNick Piggin * 330b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 331b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 332b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3331da177e4SLinus Torvalds */ 3342830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3351da177e4SLinus Torvalds { 3365909ccaaSLinus Torvalds int ret; 3371da177e4SLinus Torvalds 3381da177e4SLinus Torvalds /* 339948409c7SAndreas Gruenbacher * Do the basic permission checks. 3401da177e4SLinus Torvalds */ 3417e40145eSAl Viro ret = acl_permission_check(inode, mask); 3425909ccaaSLinus Torvalds if (ret != -EACCES) 3435909ccaaSLinus Torvalds return ret; 3441da177e4SLinus Torvalds 345d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 346d594e7ecSAl Viro /* DACs are overridable for directories */ 347d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34823adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34923adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 350d594e7ecSAl Viro return 0; 35123adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3521da177e4SLinus Torvalds return 0; 3532a4c2242SStephen Smalley return -EACCES; 3542a4c2242SStephen Smalley } 3551da177e4SLinus Torvalds 3561da177e4SLinus Torvalds /* 3571da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3581da177e4SLinus Torvalds */ 3597ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 360d594e7ecSAl Viro if (mask == MAY_READ) 36123adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3621da177e4SLinus Torvalds return 0; 3632a4c2242SStephen Smalley /* 3642a4c2242SStephen Smalley * Read/write DACs are always overridable. 3652a4c2242SStephen Smalley * Executable DACs are overridable when there is 3662a4c2242SStephen Smalley * at least one exec bit set. 3672a4c2242SStephen Smalley */ 3682a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 3692a4c2242SStephen Smalley if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3702a4c2242SStephen Smalley return 0; 3711da177e4SLinus Torvalds 3721da177e4SLinus Torvalds return -EACCES; 3731da177e4SLinus Torvalds } 3744d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3751da177e4SLinus Torvalds 3763ddcd056SLinus Torvalds /* 3773ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3783ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3793ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3803ddcd056SLinus Torvalds * permission function, use the fast case". 3813ddcd056SLinus Torvalds */ 3823ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3833ddcd056SLinus Torvalds { 3843ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3853ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3863ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3873ddcd056SLinus Torvalds 3883ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3893ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3903ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3913ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3923ddcd056SLinus Torvalds } 3933ddcd056SLinus Torvalds return generic_permission(inode, mask); 3943ddcd056SLinus Torvalds } 3953ddcd056SLinus Torvalds 396cb23beb5SChristoph Hellwig /** 3970bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 3980bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 39955852635SRandy Dunlap * @inode: Inode to check permission on 4000bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4010bdaea90SDavid Howells * 4020bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4030bdaea90SDavid Howells */ 4040bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4050bdaea90SDavid Howells { 4060bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4070bdaea90SDavid Howells umode_t mode = inode->i_mode; 4080bdaea90SDavid Howells 4090bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 410bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4110bdaea90SDavid Howells return -EROFS; 4120bdaea90SDavid Howells } 4130bdaea90SDavid Howells return 0; 4140bdaea90SDavid Howells } 4150bdaea90SDavid Howells 4160bdaea90SDavid Howells /** 4170bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4180bdaea90SDavid Howells * @inode: Inode to check permission on 4190bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4200bdaea90SDavid Howells * 4210bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4220bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4230bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4240bdaea90SDavid Howells * 4250bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4260bdaea90SDavid Howells */ 4270bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4280bdaea90SDavid Howells { 4290bdaea90SDavid Howells int retval; 4300bdaea90SDavid Howells 4310bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4320bdaea90SDavid Howells if (retval) 4330bdaea90SDavid Howells return retval; 4344bfd054aSEric Biggers 4354bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 4364bfd054aSEric Biggers /* 4374bfd054aSEric Biggers * Nobody gets write access to an immutable file. 4384bfd054aSEric Biggers */ 4394bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 4404bfd054aSEric Biggers return -EPERM; 4414bfd054aSEric Biggers 4424bfd054aSEric Biggers /* 4434bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 4444bfd054aSEric Biggers * written back improperly if their true value is unknown 4454bfd054aSEric Biggers * to the vfs. 4464bfd054aSEric Biggers */ 4474bfd054aSEric Biggers if (HAS_UNMAPPED_ID(inode)) 4484bfd054aSEric Biggers return -EACCES; 4494bfd054aSEric Biggers } 4504bfd054aSEric Biggers 4514bfd054aSEric Biggers retval = do_inode_permission(inode, mask); 4524bfd054aSEric Biggers if (retval) 4534bfd054aSEric Biggers return retval; 4544bfd054aSEric Biggers 4554bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 4564bfd054aSEric Biggers if (retval) 4574bfd054aSEric Biggers return retval; 4584bfd054aSEric Biggers 4594bfd054aSEric Biggers return security_inode_permission(inode, mask); 4600bdaea90SDavid Howells } 4614d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4620bdaea90SDavid Howells 4630bdaea90SDavid Howells /** 4645dd784d0SJan Blunck * path_get - get a reference to a path 4655dd784d0SJan Blunck * @path: path to get the reference to 4665dd784d0SJan Blunck * 4675dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4685dd784d0SJan Blunck */ 469dcf787f3SAl Viro void path_get(const struct path *path) 4705dd784d0SJan Blunck { 4715dd784d0SJan Blunck mntget(path->mnt); 4725dd784d0SJan Blunck dget(path->dentry); 4735dd784d0SJan Blunck } 4745dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4755dd784d0SJan Blunck 4765dd784d0SJan Blunck /** 4771d957f9bSJan Blunck * path_put - put a reference to a path 4781d957f9bSJan Blunck * @path: path to put the reference to 4791d957f9bSJan Blunck * 4801d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4811d957f9bSJan Blunck */ 482dcf787f3SAl Viro void path_put(const struct path *path) 4831da177e4SLinus Torvalds { 4841d957f9bSJan Blunck dput(path->dentry); 4851d957f9bSJan Blunck mntput(path->mnt); 4861da177e4SLinus Torvalds } 4871d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4881da177e4SLinus Torvalds 489894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4901f55a6ecSAl Viro struct nameidata { 4911f55a6ecSAl Viro struct path path; 4921f55a6ecSAl Viro struct qstr last; 4931f55a6ecSAl Viro struct path root; 4941f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 4951f55a6ecSAl Viro unsigned int flags; 4969883d185SAl Viro unsigned seq, m_seq; 4971f55a6ecSAl Viro int last_type; 4981f55a6ecSAl Viro unsigned depth; 499756daf26SNeilBrown int total_link_count; 500697fc6caSAl Viro struct saved { 501697fc6caSAl Viro struct path link; 502fceef393SAl Viro struct delayed_call done; 503697fc6caSAl Viro const char *name; 5040450b2d1SAl Viro unsigned seq; 505894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5069883d185SAl Viro struct filename *name; 5079883d185SAl Viro struct nameidata *saved; 508fceef393SAl Viro struct inode *link_inode; 5099883d185SAl Viro unsigned root_seq; 5109883d185SAl Viro int dfd; 5113859a271SKees Cook } __randomize_layout; 5121f55a6ecSAl Viro 5139883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 514894bc8c4SAl Viro { 515756daf26SNeilBrown struct nameidata *old = current->nameidata; 516756daf26SNeilBrown p->stack = p->internal; 517c8a53ee5SAl Viro p->dfd = dfd; 518c8a53ee5SAl Viro p->name = name; 519756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5209883d185SAl Viro p->saved = old; 521756daf26SNeilBrown current->nameidata = p; 522894bc8c4SAl Viro } 523894bc8c4SAl Viro 5249883d185SAl Viro static void restore_nameidata(void) 525894bc8c4SAl Viro { 5269883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 527756daf26SNeilBrown 528756daf26SNeilBrown current->nameidata = old; 529756daf26SNeilBrown if (old) 530756daf26SNeilBrown old->total_link_count = now->total_link_count; 531e1a63bbcSAl Viro if (now->stack != now->internal) 532756daf26SNeilBrown kfree(now->stack); 533894bc8c4SAl Viro } 534894bc8c4SAl Viro 535894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 536894bc8c4SAl Viro { 537bc40aee0SAl Viro struct saved *p; 538bc40aee0SAl Viro 539bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 5406da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 541bc40aee0SAl Viro GFP_ATOMIC); 542bc40aee0SAl Viro if (unlikely(!p)) 543bc40aee0SAl Viro return -ECHILD; 544bc40aee0SAl Viro } else { 5456da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 546894bc8c4SAl Viro GFP_KERNEL); 547894bc8c4SAl Viro if (unlikely(!p)) 548894bc8c4SAl Viro return -ENOMEM; 549bc40aee0SAl Viro } 550894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 551894bc8c4SAl Viro nd->stack = p; 552894bc8c4SAl Viro return 0; 553894bc8c4SAl Viro } 554894bc8c4SAl Viro 555397d425dSEric W. Biederman /** 556397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 557397d425dSEric W. Biederman * @path: nameidate to verify 558397d425dSEric W. Biederman * 559397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 560397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 561397d425dSEric W. Biederman */ 562397d425dSEric W. Biederman static bool path_connected(const struct path *path) 563397d425dSEric W. Biederman { 564397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 56595dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 566397d425dSEric W. Biederman 56795dd7758SEric W. Biederman /* Bind mounts and multi-root filesystems can have disconnected paths */ 56895dd7758SEric W. Biederman if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root)) 569397d425dSEric W. Biederman return true; 570397d425dSEric W. Biederman 571397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 572397d425dSEric W. Biederman } 573397d425dSEric W. Biederman 574894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 575894bc8c4SAl Viro { 576da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 577894bc8c4SAl Viro return 0; 578894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 579894bc8c4SAl Viro return 0; 580894bc8c4SAl Viro return __nd_alloc_stack(nd); 581894bc8c4SAl Viro } 582894bc8c4SAl Viro 5837973387aSAl Viro static void drop_links(struct nameidata *nd) 5847973387aSAl Viro { 5857973387aSAl Viro int i = nd->depth; 5867973387aSAl Viro while (i--) { 5877973387aSAl Viro struct saved *last = nd->stack + i; 588fceef393SAl Viro do_delayed_call(&last->done); 589fceef393SAl Viro clear_delayed_call(&last->done); 5907973387aSAl Viro } 5917973387aSAl Viro } 5927973387aSAl Viro 5937973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5947973387aSAl Viro { 5957973387aSAl Viro drop_links(nd); 5967973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5977973387aSAl Viro int i; 5987973387aSAl Viro path_put(&nd->path); 5997973387aSAl Viro for (i = 0; i < nd->depth; i++) 6007973387aSAl Viro path_put(&nd->stack[i].link); 601102b8af2SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 602102b8af2SAl Viro path_put(&nd->root); 603102b8af2SAl Viro nd->root.mnt = NULL; 604102b8af2SAl Viro } 6057973387aSAl Viro } else { 6067973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6077973387aSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6087973387aSAl Viro nd->root.mnt = NULL; 6097973387aSAl Viro rcu_read_unlock(); 6107973387aSAl Viro } 6117973387aSAl Viro nd->depth = 0; 6127973387aSAl Viro } 6137973387aSAl Viro 6147973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6157973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6167973387aSAl Viro struct path *path, unsigned seq) 6177973387aSAl Viro { 6187973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6197973387aSAl Viro if (unlikely(res)) { 6207973387aSAl Viro if (res > 0) 6217973387aSAl Viro path->mnt = NULL; 6227973387aSAl Viro path->dentry = NULL; 6237973387aSAl Viro return false; 6247973387aSAl Viro } 6257973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6267973387aSAl Viro path->dentry = NULL; 6277973387aSAl Viro return false; 6287973387aSAl Viro } 6297973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6307973387aSAl Viro } 6317973387aSAl Viro 6327973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6337973387aSAl Viro { 6347973387aSAl Viro int i; 6357973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6367973387aSAl Viro struct saved *last = nd->stack + i; 6377973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6387973387aSAl Viro drop_links(nd); 6397973387aSAl Viro nd->depth = i + 1; 6407973387aSAl Viro return false; 6417973387aSAl Viro } 6427973387aSAl Viro } 6437973387aSAl Viro return true; 6447973387aSAl Viro } 6457973387aSAl Viro 64619660af7SAl Viro /* 64731e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 64819660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 64919660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 65057e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 65119660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 65219660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 65319660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 65419660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 65531e6b01fSNick Piggin */ 65631e6b01fSNick Piggin 65731e6b01fSNick Piggin /** 65819660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 65919660af7SAl Viro * @nd: nameidata pathwalk data 66039191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 66131e6b01fSNick Piggin * 6624675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6634675ac39SAl Viro * for ref-walk mode. 6644675ac39SAl Viro * Must be called from rcu-walk context. 6657973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6667973387aSAl Viro * terminate_walk(). 66731e6b01fSNick Piggin */ 6684675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 66931e6b01fSNick Piggin { 67031e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 67131e6b01fSNick Piggin 67231e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 673e5c832d5SLinus Torvalds 674e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6757973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6767973387aSAl Viro goto out2; 6774675ac39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 6784675ac39SAl Viro goto out1; 6794675ac39SAl Viro if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 6804675ac39SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) 6814675ac39SAl Viro goto out; 6824675ac39SAl Viro } 6834675ac39SAl Viro rcu_read_unlock(); 6844675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6854675ac39SAl Viro return 0; 6864675ac39SAl Viro 6874675ac39SAl Viro out2: 6884675ac39SAl Viro nd->path.mnt = NULL; 6894675ac39SAl Viro nd->path.dentry = NULL; 6904675ac39SAl Viro out1: 6914675ac39SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 6924675ac39SAl Viro nd->root.mnt = NULL; 6934675ac39SAl Viro out: 6944675ac39SAl Viro rcu_read_unlock(); 6954675ac39SAl Viro return -ECHILD; 6964675ac39SAl Viro } 6974675ac39SAl Viro 6984675ac39SAl Viro /** 6994675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7004675ac39SAl Viro * @nd: nameidata pathwalk data 7014675ac39SAl Viro * @dentry: child of nd->path.dentry 7024675ac39SAl Viro * @seq: seq number to check dentry against 7034675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7044675ac39SAl Viro * 7054675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7064675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7074675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7084675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7094675ac39SAl Viro * terminate_walk(). 7104675ac39SAl Viro */ 7114675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7124675ac39SAl Viro { 7134675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7144675ac39SAl Viro 7154675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7164675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7174675ac39SAl Viro goto out2; 7187973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7197973387aSAl Viro goto out2; 7204675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7217973387aSAl Viro goto out1; 72248a066e7SAl Viro 72315570086SLinus Torvalds /* 7244675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7254675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7264675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7274675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7284675ac39SAl Viro * know the parent must still be valid if the child sequence number is 72915570086SLinus Torvalds */ 7304675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 731e5c832d5SLinus Torvalds goto out; 7324675ac39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) { 7334675ac39SAl Viro rcu_read_unlock(); 7344675ac39SAl Viro dput(dentry); 7354675ac39SAl Viro goto drop_root_mnt; 73619660af7SAl Viro } 737e5c832d5SLinus Torvalds /* 738e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 739e5c832d5SLinus Torvalds * still valid and get it if required. 740e5c832d5SLinus Torvalds */ 741e5c832d5SLinus Torvalds if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) { 7425a8d87e8SAl Viro if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) { 7435a8d87e8SAl Viro rcu_read_unlock(); 7445a8d87e8SAl Viro dput(dentry); 7455a8d87e8SAl Viro return -ECHILD; 7467973387aSAl Viro } 74731e6b01fSNick Piggin } 74831e6b01fSNick Piggin 7498b61e74fSAl Viro rcu_read_unlock(); 75031e6b01fSNick Piggin return 0; 75119660af7SAl Viro 7527973387aSAl Viro out2: 7537973387aSAl Viro nd->path.mnt = NULL; 7547973387aSAl Viro out1: 7557973387aSAl Viro nd->path.dentry = NULL; 756e5c832d5SLinus Torvalds out: 7578b61e74fSAl Viro rcu_read_unlock(); 758d0d27277SLinus Torvalds drop_root_mnt: 759d0d27277SLinus Torvalds if (!(nd->flags & LOOKUP_ROOT)) 760d0d27277SLinus Torvalds nd->root.mnt = NULL; 76131e6b01fSNick Piggin return -ECHILD; 76231e6b01fSNick Piggin } 76331e6b01fSNick Piggin 7644ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76534286d66SNick Piggin { 766a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7674ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 768a89f8337SAl Viro else 769a89f8337SAl Viro return 1; 77034286d66SNick Piggin } 77134286d66SNick Piggin 7729f1fafeeSAl Viro /** 7739f1fafeeSAl Viro * complete_walk - successful completion of path walk 7749f1fafeeSAl Viro * @nd: pointer nameidata 77539159de2SJeff Layton * 7769f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7779f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7789f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7799f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7809f1fafeeSAl Viro * need to drop nd->path. 78139159de2SJeff Layton */ 7829f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78339159de2SJeff Layton { 78416c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78539159de2SJeff Layton int status; 78639159de2SJeff Layton 7879f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 7889f1fafeeSAl Viro if (!(nd->flags & LOOKUP_ROOT)) 7899f1fafeeSAl Viro nd->root.mnt = NULL; 7904675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 79148a066e7SAl Viro return -ECHILD; 79248a066e7SAl Viro } 7939f1fafeeSAl Viro 79416c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 79539159de2SJeff Layton return 0; 79639159de2SJeff Layton 797ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 79816c2cd71SAl Viro return 0; 79916c2cd71SAl Viro 800ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 80139159de2SJeff Layton if (status > 0) 80239159de2SJeff Layton return 0; 80339159de2SJeff Layton 80416c2cd71SAl Viro if (!status) 80539159de2SJeff Layton status = -ESTALE; 80616c2cd71SAl Viro 80739159de2SJeff Layton return status; 80839159de2SJeff Layton } 80939159de2SJeff Layton 81018d8c860SAl Viro static void set_root(struct nameidata *nd) 8112a737871SAl Viro { 81231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8139e6697e2SAl Viro 8149e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8158f47a016SAl Viro unsigned seq; 816c28cc364SNick Piggin 817c28cc364SNick Piggin do { 818c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 81931e6b01fSNick Piggin nd->root = fs->root; 8208f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 821c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8229e6697e2SAl Viro } else { 8239e6697e2SAl Viro get_fs_root(fs, &nd->root); 8249e6697e2SAl Viro } 82531e6b01fSNick Piggin } 82631e6b01fSNick Piggin 8271d957f9bSJan Blunck static void path_put_conditional(struct path *path, struct nameidata *nd) 828051d3812SIan Kent { 829051d3812SIan Kent dput(path->dentry); 8304ac91378SJan Blunck if (path->mnt != nd->path.mnt) 831051d3812SIan Kent mntput(path->mnt); 832051d3812SIan Kent } 833051d3812SIan Kent 8347b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8357b9337aaSNick Piggin struct nameidata *nd) 836051d3812SIan Kent { 83731e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8384ac91378SJan Blunck dput(nd->path.dentry); 83931e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8404ac91378SJan Blunck mntput(nd->path.mnt); 8419a229683SHuang Shijie } 84231e6b01fSNick Piggin nd->path.mnt = path->mnt; 8434ac91378SJan Blunck nd->path.dentry = path->dentry; 844051d3812SIan Kent } 845051d3812SIan Kent 846248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 847248fb5b9SAl Viro { 848248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 849248fb5b9SAl Viro struct dentry *d; 850248fb5b9SAl Viro nd->path = nd->root; 851248fb5b9SAl Viro d = nd->path.dentry; 852248fb5b9SAl Viro nd->inode = d->d_inode; 853248fb5b9SAl Viro nd->seq = nd->root_seq; 854248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 855248fb5b9SAl Viro return -ECHILD; 856248fb5b9SAl Viro } else { 857248fb5b9SAl Viro path_put(&nd->path); 858248fb5b9SAl Viro nd->path = nd->root; 859248fb5b9SAl Viro path_get(&nd->path); 860248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 861248fb5b9SAl Viro } 862248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 863248fb5b9SAl Viro return 0; 864248fb5b9SAl Viro } 865248fb5b9SAl Viro 866b5fb63c1SChristoph Hellwig /* 8676b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 868b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 869b5fb63c1SChristoph Hellwig */ 8706e77137bSAl Viro void nd_jump_link(struct path *path) 871b5fb63c1SChristoph Hellwig { 8726e77137bSAl Viro struct nameidata *nd = current->nameidata; 873b5fb63c1SChristoph Hellwig path_put(&nd->path); 874b5fb63c1SChristoph Hellwig 875b5fb63c1SChristoph Hellwig nd->path = *path; 876b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 877b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 878b5fb63c1SChristoph Hellwig } 879b5fb63c1SChristoph Hellwig 880b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 881574197e0SAl Viro { 88221c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 883fceef393SAl Viro do_delayed_call(&last->done); 8846548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 885b9ff4429SAl Viro path_put(&last->link); 886574197e0SAl Viro } 887574197e0SAl Viro 888561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 889561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 890800179c9SKees Cook 891800179c9SKees Cook /** 892800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 89355852635SRandy Dunlap * @nd: nameidata pathwalk data 894800179c9SKees Cook * 895800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 896800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 897800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 898800179c9SKees Cook * processes from failing races against path names that may change out 899800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 900800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 901800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 902800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 903800179c9SKees Cook * 904800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 905800179c9SKees Cook */ 906fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 907800179c9SKees Cook { 908800179c9SKees Cook const struct inode *inode; 909800179c9SKees Cook const struct inode *parent; 9102d7f9e2aSSeth Forshee kuid_t puid; 911800179c9SKees Cook 912800179c9SKees Cook if (!sysctl_protected_symlinks) 913800179c9SKees Cook return 0; 914800179c9SKees Cook 915800179c9SKees Cook /* Allowed if owner and follower match. */ 916fceef393SAl Viro inode = nd->link_inode; 91781abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 918800179c9SKees Cook return 0; 919800179c9SKees Cook 920800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 921aa65fa35SAl Viro parent = nd->inode; 922800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 923800179c9SKees Cook return 0; 924800179c9SKees Cook 925800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9262d7f9e2aSSeth Forshee puid = parent->i_uid; 9272d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 928800179c9SKees Cook return 0; 929800179c9SKees Cook 93031956502SAl Viro if (nd->flags & LOOKUP_RCU) 93131956502SAl Viro return -ECHILD; 93231956502SAl Viro 933ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 93494b9d9b7SRichard Guy Briggs audit_log_link_denied("follow_link"); 935800179c9SKees Cook return -EACCES; 936800179c9SKees Cook } 937800179c9SKees Cook 938800179c9SKees Cook /** 939800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 940800179c9SKees Cook * @inode: the source inode to hardlink from 941800179c9SKees Cook * 942800179c9SKees Cook * Return false if at least one of the following conditions: 943800179c9SKees Cook * - inode is not a regular file 944800179c9SKees Cook * - inode is setuid 945800179c9SKees Cook * - inode is setgid and group-exec 946800179c9SKees Cook * - access failure for read and write 947800179c9SKees Cook * 948800179c9SKees Cook * Otherwise returns true. 949800179c9SKees Cook */ 950800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 951800179c9SKees Cook { 952800179c9SKees Cook umode_t mode = inode->i_mode; 953800179c9SKees Cook 954800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 955800179c9SKees Cook if (!S_ISREG(mode)) 956800179c9SKees Cook return false; 957800179c9SKees Cook 958800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 959800179c9SKees Cook if (mode & S_ISUID) 960800179c9SKees Cook return false; 961800179c9SKees Cook 962800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 963800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 964800179c9SKees Cook return false; 965800179c9SKees Cook 966800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 967800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 968800179c9SKees Cook return false; 969800179c9SKees Cook 970800179c9SKees Cook return true; 971800179c9SKees Cook } 972800179c9SKees Cook 973800179c9SKees Cook /** 974800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 975800179c9SKees Cook * @link: the source to hardlink from 976800179c9SKees Cook * 977800179c9SKees Cook * Block hardlink when all of: 978800179c9SKees Cook * - sysctl_protected_hardlinks enabled 979800179c9SKees Cook * - fsuid does not match inode 980800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 981f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 982800179c9SKees Cook * 983800179c9SKees Cook * Returns 0 if successful, -ve on error. 984800179c9SKees Cook */ 985800179c9SKees Cook static int may_linkat(struct path *link) 986800179c9SKees Cook { 987593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 988593d1ce8SEric W. Biederman 989593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 990593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 991593d1ce8SEric W. Biederman return -EOVERFLOW; 992800179c9SKees Cook 993800179c9SKees Cook if (!sysctl_protected_hardlinks) 994800179c9SKees Cook return 0; 995800179c9SKees Cook 996800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 997800179c9SKees Cook * otherwise, it must be a safe source. 998800179c9SKees Cook */ 999cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1000800179c9SKees Cook return 0; 1001800179c9SKees Cook 100294b9d9b7SRichard Guy Briggs audit_log_link_denied("linkat"); 1003800179c9SKees Cook return -EPERM; 1004800179c9SKees Cook } 1005800179c9SKees Cook 10063b2e7f75SAl Viro static __always_inline 10073b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 10081da177e4SLinus Torvalds { 1009ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 10101cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1011fceef393SAl Viro struct inode *inode = nd->link_inode; 10126d7b5aaeSAl Viro int error; 10130a959df5SAl Viro const char *res; 10141da177e4SLinus Torvalds 10158fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 10168fa9dd24SNeilBrown touch_atime(&last->link); 10178fa9dd24SNeilBrown cond_resched(); 1018598e3c8fSMiklos Szeredi } else if (atime_needs_update_rcu(&last->link, inode)) { 10194675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10208fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 10218fa9dd24SNeilBrown touch_atime(&last->link); 10228fa9dd24SNeilBrown } 10238fa9dd24SNeilBrown 1024bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1025bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1026bda0be7aSNeilBrown if (unlikely(error)) 10276920a440SAl Viro return ERR_PTR(error); 102836f3b4f6SAl Viro 102986acdca1SAl Viro nd->last_type = LAST_BIND; 1030d4dee48bSAl Viro res = inode->i_link; 1031d4dee48bSAl Viro if (!res) { 1032fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1033fceef393SAl Viro struct delayed_call *); 1034fceef393SAl Viro get = inode->i_op->get_link; 10358c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1036fceef393SAl Viro res = get(NULL, inode, &last->done); 10376b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 10384675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 10398c1b4566SAl Viro return ERR_PTR(-ECHILD); 1040fceef393SAl Viro res = get(dentry, inode, &last->done); 10418c1b4566SAl Viro } 10426b255391SAl Viro } else { 1043fceef393SAl Viro res = get(dentry, inode, &last->done); 10446b255391SAl Viro } 1045fceef393SAl Viro if (IS_ERR_OR_NULL(res)) 1046fab51e8aSAl Viro return res; 10470a959df5SAl Viro } 1048fab51e8aSAl Viro if (*res == '/') { 1049fab51e8aSAl Viro if (!nd->root.mnt) 1050fab51e8aSAl Viro set_root(nd); 1051248fb5b9SAl Viro if (unlikely(nd_jump_root(nd))) 1052fab51e8aSAl Viro return ERR_PTR(-ECHILD); 1053fab51e8aSAl Viro while (unlikely(*++res == '/')) 1054fab51e8aSAl Viro ; 1055fab51e8aSAl Viro } 1056fab51e8aSAl Viro if (!*res) 1057fab51e8aSAl Viro res = NULL; 10580a959df5SAl Viro return res; 10590a959df5SAl Viro } 10606d7b5aaeSAl Viro 1061f015f126SDavid Howells /* 1062f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1063f015f126SDavid Howells * 1064f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1065f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1066f015f126SDavid Howells * Up is towards /. 1067f015f126SDavid Howells * 1068f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1069f015f126SDavid Howells * root. 1070f015f126SDavid Howells */ 1071bab77ebfSAl Viro int follow_up(struct path *path) 10721da177e4SLinus Torvalds { 10730714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 10740714a533SAl Viro struct mount *parent; 10751da177e4SLinus Torvalds struct dentry *mountpoint; 107699b7db7bSNick Piggin 107748a066e7SAl Viro read_seqlock_excl(&mount_lock); 10780714a533SAl Viro parent = mnt->mnt_parent; 10793c0a6163SAl Viro if (parent == mnt) { 108048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 10811da177e4SLinus Torvalds return 0; 10821da177e4SLinus Torvalds } 10830714a533SAl Viro mntget(&parent->mnt); 1084a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 108548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1086bab77ebfSAl Viro dput(path->dentry); 1087bab77ebfSAl Viro path->dentry = mountpoint; 1088bab77ebfSAl Viro mntput(path->mnt); 10890714a533SAl Viro path->mnt = &parent->mnt; 10901da177e4SLinus Torvalds return 1; 10911da177e4SLinus Torvalds } 10924d359507SAl Viro EXPORT_SYMBOL(follow_up); 10931da177e4SLinus Torvalds 1094b5c84bf6SNick Piggin /* 10959875cf80SDavid Howells * Perform an automount 10969875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 10979875cf80SDavid Howells * were called with. 10981da177e4SLinus Torvalds */ 1099756daf26SNeilBrown static int follow_automount(struct path *path, struct nameidata *nd, 11009875cf80SDavid Howells bool *need_mntput) 110131e6b01fSNick Piggin { 11029875cf80SDavid Howells struct vfsmount *mnt; 1103ea5b778aSDavid Howells int err; 11049875cf80SDavid Howells 11059875cf80SDavid Howells if (!path->dentry->d_op || !path->dentry->d_op->d_automount) 11069875cf80SDavid Howells return -EREMOTE; 11079875cf80SDavid Howells 11080ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11090ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11100ec26fd0SMiklos Szeredi * the name. 11110ec26fd0SMiklos Szeredi * 11120ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11135a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11140ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11150ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11160ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11170ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11185a30d8a2SDavid Howells */ 1119756daf26SNeilBrown if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11205d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 11215d38f049SIan Kent path->dentry->d_inode) 11229875cf80SDavid Howells return -EISDIR; 11230ec26fd0SMiklos Szeredi 1124756daf26SNeilBrown nd->total_link_count++; 1125756daf26SNeilBrown if (nd->total_link_count >= 40) 11269875cf80SDavid Howells return -ELOOP; 11279875cf80SDavid Howells 11289875cf80SDavid Howells mnt = path->dentry->d_op->d_automount(path); 11299875cf80SDavid Howells if (IS_ERR(mnt)) { 11309875cf80SDavid Howells /* 11319875cf80SDavid Howells * The filesystem is allowed to return -EISDIR here to indicate 11329875cf80SDavid Howells * it doesn't want to automount. For instance, autofs would do 11339875cf80SDavid Howells * this so that its userspace daemon can mount on this dentry. 11349875cf80SDavid Howells * 11359875cf80SDavid Howells * However, we can only permit this if it's a terminal point in 11369875cf80SDavid Howells * the path being looked up; if it wasn't then the remainder of 11379875cf80SDavid Howells * the path is inaccessible and we should say so. 11389875cf80SDavid Howells */ 1139756daf26SNeilBrown if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT)) 11409875cf80SDavid Howells return -EREMOTE; 11419875cf80SDavid Howells return PTR_ERR(mnt); 114231e6b01fSNick Piggin } 1143ea5b778aSDavid Howells 11449875cf80SDavid Howells if (!mnt) /* mount collision */ 11459875cf80SDavid Howells return 0; 11469875cf80SDavid Howells 11478aef1884SAl Viro if (!*need_mntput) { 11488aef1884SAl Viro /* lock_mount() may release path->mnt on error */ 11498aef1884SAl Viro mntget(path->mnt); 11508aef1884SAl Viro *need_mntput = true; 11518aef1884SAl Viro } 115219a167afSAl Viro err = finish_automount(mnt, path); 1153ea5b778aSDavid Howells 1154ea5b778aSDavid Howells switch (err) { 1155ea5b778aSDavid Howells case -EBUSY: 1156ea5b778aSDavid Howells /* Someone else made a mount here whilst we were busy */ 115719a167afSAl Viro return 0; 1158ea5b778aSDavid Howells case 0: 11598aef1884SAl Viro path_put(path); 11609875cf80SDavid Howells path->mnt = mnt; 11619875cf80SDavid Howells path->dentry = dget(mnt->mnt_root); 11629875cf80SDavid Howells return 0; 116319a167afSAl Viro default: 116419a167afSAl Viro return err; 11659875cf80SDavid Howells } 116619a167afSAl Viro 1167ea5b778aSDavid Howells } 11689875cf80SDavid Howells 11699875cf80SDavid Howells /* 11709875cf80SDavid Howells * Handle a dentry that is managed in some way. 1171cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11729875cf80SDavid Howells * - Flagged as mountpoint 11739875cf80SDavid Howells * - Flagged as automount point 11749875cf80SDavid Howells * 11759875cf80SDavid Howells * This may only be called in refwalk mode. 11769875cf80SDavid Howells * 11779875cf80SDavid Howells * Serialization is taken care of in namespace.c 11789875cf80SDavid Howells */ 1179756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11809875cf80SDavid Howells { 11818aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 11829875cf80SDavid Howells unsigned managed; 11839875cf80SDavid Howells bool need_mntput = false; 11848aef1884SAl Viro int ret = 0; 11859875cf80SDavid Howells 11869875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11879875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11889875cf80SDavid Howells * the components of that value change under us */ 11896aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 11909875cf80SDavid Howells managed &= DCACHE_MANAGED_DENTRY, 11919875cf80SDavid Howells unlikely(managed != 0)) { 1192cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1193cc53ce53SDavid Howells * being held. */ 1194cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1195cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1196cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1197fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1198cc53ce53SDavid Howells if (ret < 0) 11998aef1884SAl Viro break; 1200cc53ce53SDavid Howells } 1201cc53ce53SDavid Howells 12029875cf80SDavid Howells /* Transit to a mounted filesystem. */ 12039875cf80SDavid Howells if (managed & DCACHE_MOUNTED) { 12049875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12059875cf80SDavid Howells if (mounted) { 12069875cf80SDavid Howells dput(path->dentry); 12079875cf80SDavid Howells if (need_mntput) 1208463ffb2eSAl Viro mntput(path->mnt); 1209463ffb2eSAl Viro path->mnt = mounted; 1210463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12119875cf80SDavid Howells need_mntput = true; 12129875cf80SDavid Howells continue; 1213463ffb2eSAl Viro } 1214463ffb2eSAl Viro 12159875cf80SDavid Howells /* Something is mounted on this dentry in another 12169875cf80SDavid Howells * namespace and/or whatever was mounted there in this 121748a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 121848a066e7SAl Viro * get it */ 12191da177e4SLinus Torvalds } 12209875cf80SDavid Howells 12219875cf80SDavid Howells /* Handle an automount point */ 12229875cf80SDavid Howells if (managed & DCACHE_NEED_AUTOMOUNT) { 1223756daf26SNeilBrown ret = follow_automount(path, nd, &need_mntput); 12249875cf80SDavid Howells if (ret < 0) 12258aef1884SAl Viro break; 12269875cf80SDavid Howells continue; 12279875cf80SDavid Howells } 12289875cf80SDavid Howells 12299875cf80SDavid Howells /* We didn't change the current path point */ 12309875cf80SDavid Howells break; 12319875cf80SDavid Howells } 12328aef1884SAl Viro 12338aef1884SAl Viro if (need_mntput && path->mnt == mnt) 12348aef1884SAl Viro mntput(path->mnt); 1235e9742b53SAl Viro if (ret == -EISDIR || !ret) 1236e9742b53SAl Viro ret = 1; 12378402752eSAl Viro if (need_mntput) 12388402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 12398402752eSAl Viro if (unlikely(ret < 0)) 12408402752eSAl Viro path_put_conditional(path, nd); 12418402752eSAl Viro return ret; 12421da177e4SLinus Torvalds } 12431da177e4SLinus Torvalds 1244cc53ce53SDavid Howells int follow_down_one(struct path *path) 12451da177e4SLinus Torvalds { 12461da177e4SLinus Torvalds struct vfsmount *mounted; 12471da177e4SLinus Torvalds 12481c755af4SAl Viro mounted = lookup_mnt(path); 12491da177e4SLinus Torvalds if (mounted) { 12509393bd07SAl Viro dput(path->dentry); 12519393bd07SAl Viro mntput(path->mnt); 12529393bd07SAl Viro path->mnt = mounted; 12539393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12541da177e4SLinus Torvalds return 1; 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds return 0; 12571da177e4SLinus Torvalds } 12584d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12591da177e4SLinus Torvalds 1260fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 126162a7375eSIan Kent { 1262fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1263fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 126462a7375eSIan Kent } 126562a7375eSIan Kent 12669875cf80SDavid Howells /* 1267287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1268287548e4SAl Viro * we meet a managed dentry that would need blocking. 12699875cf80SDavid Howells */ 12709875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1271254cf582SAl Viro struct inode **inode, unsigned *seqp) 12729875cf80SDavid Howells { 127362a7375eSIan Kent for (;;) { 1274c7105365SAl Viro struct mount *mounted; 127562a7375eSIan Kent /* 127662a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 127762a7375eSIan Kent * that wants to block transit. 127862a7375eSIan Kent */ 1279fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1280b8faf035SNeilBrown case -ECHILD: 1281b8faf035SNeilBrown default: 1282ab90911fSDavid Howells return false; 1283b8faf035SNeilBrown case -EISDIR: 1284b8faf035SNeilBrown return true; 1285b8faf035SNeilBrown case 0: 1286b8faf035SNeilBrown break; 1287b8faf035SNeilBrown } 128862a7375eSIan Kent 128962a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1290b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 129162a7375eSIan Kent 1292474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 12939875cf80SDavid Howells if (!mounted) 12949875cf80SDavid Howells break; 1295c7105365SAl Viro path->mnt = &mounted->mnt; 1296c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1297a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1298254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 129959430262SLinus Torvalds /* 130059430262SLinus Torvalds * Update the inode too. We don't need to re-check the 130159430262SLinus Torvalds * dentry sequence number here after this d_inode read, 130259430262SLinus Torvalds * because a mount-point is always pinned. 130359430262SLinus Torvalds */ 130459430262SLinus Torvalds *inode = path->dentry->d_inode; 13059875cf80SDavid Howells } 1306f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1307b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1308287548e4SAl Viro } 1309287548e4SAl Viro 131031e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 131131e6b01fSNick Piggin { 13124023bfc9SAl Viro struct inode *inode = nd->inode; 131331e6b01fSNick Piggin 131431e6b01fSNick Piggin while (1) { 1315aed434adSAl Viro if (path_equal(&nd->path, &nd->root)) 131631e6b01fSNick Piggin break; 131731e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 131831e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 131931e6b01fSNick Piggin struct dentry *parent = old->d_parent; 132031e6b01fSNick Piggin unsigned seq; 132131e6b01fSNick Piggin 13224023bfc9SAl Viro inode = parent->d_inode; 132331e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1324aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1325aed434adSAl Viro return -ECHILD; 132631e6b01fSNick Piggin nd->path.dentry = parent; 132731e6b01fSNick Piggin nd->seq = seq; 1328397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 1329397d425dSEric W. Biederman return -ENOENT; 133031e6b01fSNick Piggin break; 1331aed434adSAl Viro } else { 1332aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1333aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1334aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1335aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1336aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1337aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1338aed434adSAl Viro return -ECHILD; 1339aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 134031e6b01fSNick Piggin break; 1341aed434adSAl Viro /* we know that mountpoint was pinned */ 1342aed434adSAl Viro nd->path.dentry = mountpoint; 1343aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1344aed434adSAl Viro inode = inode2; 1345aed434adSAl Viro nd->seq = seq; 134631e6b01fSNick Piggin } 1347aed434adSAl Viro } 1348aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1349b37199e6SAl Viro struct mount *mounted; 1350b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1351aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1352aed434adSAl Viro return -ECHILD; 1353b37199e6SAl Viro if (!mounted) 1354b37199e6SAl Viro break; 1355b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1356b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 13574023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1358b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1359b37199e6SAl Viro } 13604023bfc9SAl Viro nd->inode = inode; 136131e6b01fSNick Piggin return 0; 136231e6b01fSNick Piggin } 136331e6b01fSNick Piggin 13649875cf80SDavid Howells /* 1365cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1366cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1367cc53ce53SDavid Howells * caller is permitted to proceed or not. 1368cc53ce53SDavid Howells */ 13697cc90cc3SAl Viro int follow_down(struct path *path) 1370cc53ce53SDavid Howells { 1371cc53ce53SDavid Howells unsigned managed; 1372cc53ce53SDavid Howells int ret; 1373cc53ce53SDavid Howells 13746aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1375cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1376cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1377cc53ce53SDavid Howells * being held. 1378cc53ce53SDavid Howells * 1379cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1380cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1381cc53ce53SDavid Howells * other than its daemon the right to mount on its 1382cc53ce53SDavid Howells * superstructure. 1383cc53ce53SDavid Howells * 1384cc53ce53SDavid Howells * The filesystem may sleep at this point. 1385cc53ce53SDavid Howells */ 1386cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1387cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1388cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1389fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1390cc53ce53SDavid Howells if (ret < 0) 1391cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1392cc53ce53SDavid Howells } 1393cc53ce53SDavid Howells 1394cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1395cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1396cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1397cc53ce53SDavid Howells if (!mounted) 1398cc53ce53SDavid Howells break; 1399cc53ce53SDavid Howells dput(path->dentry); 1400cc53ce53SDavid Howells mntput(path->mnt); 1401cc53ce53SDavid Howells path->mnt = mounted; 1402cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1403cc53ce53SDavid Howells continue; 1404cc53ce53SDavid Howells } 1405cc53ce53SDavid Howells 1406cc53ce53SDavid Howells /* Don't handle automount points here */ 1407cc53ce53SDavid Howells break; 1408cc53ce53SDavid Howells } 1409cc53ce53SDavid Howells return 0; 1410cc53ce53SDavid Howells } 14114d359507SAl Viro EXPORT_SYMBOL(follow_down); 1412cc53ce53SDavid Howells 1413cc53ce53SDavid Howells /* 14149875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14159875cf80SDavid Howells */ 14169875cf80SDavid Howells static void follow_mount(struct path *path) 14179875cf80SDavid Howells { 14189875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14199875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14209875cf80SDavid Howells if (!mounted) 14219875cf80SDavid Howells break; 14229875cf80SDavid Howells dput(path->dentry); 14239875cf80SDavid Howells mntput(path->mnt); 14249875cf80SDavid Howells path->mnt = mounted; 14259875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14269875cf80SDavid Howells } 14279875cf80SDavid Howells } 14289875cf80SDavid Howells 1429eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1430eedf265aSEric W. Biederman { 1431eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1432eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1433eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1434eedf265aSEric W. Biederman dput(old); 1435eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1436eedf265aSEric W. Biederman return -ENOENT; 1437eedf265aSEric W. Biederman return 0; 1438eedf265aSEric W. Biederman } 1439eedf265aSEric W. Biederman 1440397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14411da177e4SLinus Torvalds { 14421da177e4SLinus Torvalds while(1) { 1443030c7e0bSDanilo Krummrich if (path_equal(&nd->path, &nd->root)) 14441da177e4SLinus Torvalds break; 14454ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1446eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1447eedf265aSEric W. Biederman if (ret) 1448eedf265aSEric W. Biederman return ret; 14491da177e4SLinus Torvalds break; 14501da177e4SLinus Torvalds } 14513088dd70SAl Viro if (!follow_up(&nd->path)) 14521da177e4SLinus Torvalds break; 14531da177e4SLinus Torvalds } 145479ed0226SAl Viro follow_mount(&nd->path); 145531e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1456397d425dSEric W. Biederman return 0; 14571da177e4SLinus Torvalds } 14581da177e4SLinus Torvalds 14591da177e4SLinus Torvalds /* 1460f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1461f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1462baa03890SNick Piggin */ 1463e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1464e3c13928SAl Viro struct dentry *dir, 14656c51e513SAl Viro unsigned int flags) 1466baa03890SNick Piggin { 1467a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1468bad61189SMiklos Szeredi if (dentry) { 1469a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1470bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 147174ff0ffcSAl Viro if (!error) 14725542aa2fSEric W. Biederman d_invalidate(dentry); 1473bad61189SMiklos Szeredi dput(dentry); 147474ff0ffcSAl Viro return ERR_PTR(error); 1475bad61189SMiklos Szeredi } 1476bad61189SMiklos Szeredi } 1477baa03890SNick Piggin return dentry; 1478baa03890SNick Piggin } 1479baa03890SNick Piggin 1480baa03890SNick Piggin /* 1481a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1482a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1483a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1484a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1485a03ece5fSAl Viro * at all. 148644396f4bSJosef Bacik */ 1487a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1488a03ece5fSAl Viro struct dentry *base, unsigned int flags) 148944396f4bSJosef Bacik { 1490a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 149144396f4bSJosef Bacik struct dentry *old; 1492a03ece5fSAl Viro struct inode *dir = base->d_inode; 1493a03ece5fSAl Viro 1494a03ece5fSAl Viro if (dentry) 1495a03ece5fSAl Viro return dentry; 149644396f4bSJosef Bacik 149744396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1498a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 149944396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1500a03ece5fSAl Viro 1501a03ece5fSAl Viro dentry = d_alloc(base, name); 1502a03ece5fSAl Viro if (unlikely(!dentry)) 1503a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 150444396f4bSJosef Bacik 150572bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 150644396f4bSJosef Bacik if (unlikely(old)) { 150744396f4bSJosef Bacik dput(dentry); 150844396f4bSJosef Bacik dentry = old; 150944396f4bSJosef Bacik } 151044396f4bSJosef Bacik return dentry; 151144396f4bSJosef Bacik } 151244396f4bSJosef Bacik 1513e97cdc87SAl Viro static int lookup_fast(struct nameidata *nd, 1514254cf582SAl Viro struct path *path, struct inode **inode, 1515254cf582SAl Viro unsigned *seqp) 15161da177e4SLinus Torvalds { 15174ac91378SJan Blunck struct vfsmount *mnt = nd->path.mnt; 151831e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15195a18fff2SAl Viro int status = 1; 15209875cf80SDavid Howells int err; 15219875cf80SDavid Howells 15223cac260aSAl Viro /* 1523b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15245d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15255d0f49c1SAl Viro * going to fall back to non-racy lookup. 1526b04f784eSNick Piggin */ 152731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 152831e6b01fSNick Piggin unsigned seq; 1529766c4cbfSAl Viro bool negative; 1530da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15315d0f49c1SAl Viro if (unlikely(!dentry)) { 15324675ac39SAl Viro if (unlazy_walk(nd)) 15335d0f49c1SAl Viro return -ECHILD; 1534e9742b53SAl Viro return 0; 15355d0f49c1SAl Viro } 15365a18fff2SAl Viro 153712f8ad4bSLinus Torvalds /* 153812f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 153912f8ad4bSLinus Torvalds * the dentry name information from lookup. 154012f8ad4bSLinus Torvalds */ 154163afdfc7SDavid Howells *inode = d_backing_inode(dentry); 1542766c4cbfSAl Viro negative = d_is_negative(dentry); 15435d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 154412f8ad4bSLinus Torvalds return -ECHILD; 154512f8ad4bSLinus Torvalds 154612f8ad4bSLinus Torvalds /* 154712f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 154812f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 154912f8ad4bSLinus Torvalds * 155012f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 155112f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 155212f8ad4bSLinus Torvalds */ 15535d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 155431e6b01fSNick Piggin return -ECHILD; 15555a18fff2SAl Viro 1556254cf582SAl Viro *seqp = seq; 15574ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1558209a7fb2SAl Viro if (likely(status > 0)) { 1559daf3761cSTrond Myklebust /* 1560daf3761cSTrond Myklebust * Note: do negative dentry check after revalidation in 1561daf3761cSTrond Myklebust * case that drops it. 1562daf3761cSTrond Myklebust */ 15635d0f49c1SAl Viro if (unlikely(negative)) 1564daf3761cSTrond Myklebust return -ENOENT; 156531e6b01fSNick Piggin path->mnt = mnt; 156631e6b01fSNick Piggin path->dentry = dentry; 1567254cf582SAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1568e9742b53SAl Viro return 1; 15695d0f49c1SAl Viro } 15704675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 1571254cf582SAl Viro return -ECHILD; 1572209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1573209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1574209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 15755a18fff2SAl Viro } else { 1576e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 157781e6f520SAl Viro if (unlikely(!dentry)) 1578e9742b53SAl Viro return 0; 15794ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15805d0f49c1SAl Viro } 15815a18fff2SAl Viro if (unlikely(status <= 0)) { 1582e9742b53SAl Viro if (!status) 15835d0f49c1SAl Viro d_invalidate(dentry); 15845a18fff2SAl Viro dput(dentry); 15855a18fff2SAl Viro return status; 15865a18fff2SAl Viro } 1587766c4cbfSAl Viro if (unlikely(d_is_negative(dentry))) { 1588766c4cbfSAl Viro dput(dentry); 1589766c4cbfSAl Viro return -ENOENT; 1590766c4cbfSAl Viro } 15915d0f49c1SAl Viro 15921da177e4SLinus Torvalds path->mnt = mnt; 15931da177e4SLinus Torvalds path->dentry = dentry; 1594756daf26SNeilBrown err = follow_managed(path, nd); 1595e9742b53SAl Viro if (likely(err > 0)) 159663afdfc7SDavid Howells *inode = d_backing_inode(path->dentry); 15978402752eSAl Viro return err; 1598697f514dSMiklos Szeredi } 1599697f514dSMiklos Szeredi 1600697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 160188d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1602e3c13928SAl Viro struct dentry *dir, 1603e3c13928SAl Viro unsigned int flags) 1604697f514dSMiklos Szeredi { 160588d8331aSAl Viro struct dentry *dentry, *old; 16061936386eSAl Viro struct inode *inode = dir->d_inode; 1607d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16081936386eSAl Viro 16091936386eSAl Viro /* Don't go there if it's already dead */ 161094bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 161188d8331aSAl Viro return ERR_PTR(-ENOENT); 161294bdd655SAl Viro again: 1613d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 161494bdd655SAl Viro if (IS_ERR(dentry)) 161588d8331aSAl Viro return dentry; 161694bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1617a89f8337SAl Viro if (!(flags & LOOKUP_NO_REVAL)) { 1618949a852eSAl Viro int error = d_revalidate(dentry, flags); 1619949a852eSAl Viro if (unlikely(error <= 0)) { 162094bdd655SAl Viro if (!error) { 1621949a852eSAl Viro d_invalidate(dentry); 1622949a852eSAl Viro dput(dentry); 162394bdd655SAl Viro goto again; 162494bdd655SAl Viro } 162594bdd655SAl Viro dput(dentry); 1626949a852eSAl Viro dentry = ERR_PTR(error); 1627949a852eSAl Viro } 1628949a852eSAl Viro } 162994bdd655SAl Viro } else { 16301936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 163185c7f810SAl Viro d_lookup_done(dentry); 16321936386eSAl Viro if (unlikely(old)) { 16331936386eSAl Viro dput(dentry); 16341936386eSAl Viro dentry = old; 1635949a852eSAl Viro } 1636949a852eSAl Viro } 1637e3c13928SAl Viro return dentry; 16381da177e4SLinus Torvalds } 16391da177e4SLinus Torvalds 164088d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 164188d8331aSAl Viro struct dentry *dir, 164288d8331aSAl Viro unsigned int flags) 164388d8331aSAl Viro { 164488d8331aSAl Viro struct inode *inode = dir->d_inode; 164588d8331aSAl Viro struct dentry *res; 164688d8331aSAl Viro inode_lock_shared(inode); 164788d8331aSAl Viro res = __lookup_slow(name, dir, flags); 164888d8331aSAl Viro inode_unlock_shared(inode); 164988d8331aSAl Viro return res; 165088d8331aSAl Viro } 165188d8331aSAl Viro 165252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 165352094c8aSAl Viro { 165452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16554ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 165652094c8aSAl Viro if (err != -ECHILD) 165752094c8aSAl Viro return err; 16584675ac39SAl Viro if (unlazy_walk(nd)) 165952094c8aSAl Viro return -ECHILD; 166052094c8aSAl Viro } 16614ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 166252094c8aSAl Viro } 166352094c8aSAl Viro 16649856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16659856fa1bSAl Viro { 16669856fa1bSAl Viro if (type == LAST_DOTDOT) { 16679e6697e2SAl Viro if (!nd->root.mnt) 16689e6697e2SAl Viro set_root(nd); 16699856fa1bSAl Viro if (nd->flags & LOOKUP_RCU) { 167070291aecSAl Viro return follow_dotdot_rcu(nd); 16719856fa1bSAl Viro } else 1672397d425dSEric W. Biederman return follow_dotdot(nd); 16739856fa1bSAl Viro } 16749856fa1bSAl Viro return 0; 16759856fa1bSAl Viro } 16769856fa1bSAl Viro 1677181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1678181548c0SAl Viro struct inode *inode, unsigned seq) 1679d63ff28fSAl Viro { 1680626de996SAl Viro int error; 16811cf2665bSAl Viro struct saved *last; 1682756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1683626de996SAl Viro path_to_nameidata(link, nd); 1684626de996SAl Viro return -ELOOP; 1685626de996SAl Viro } 1686bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1687cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1688cd179f44SAl Viro mntget(link->mnt); 16897973387aSAl Viro } 1690626de996SAl Viro error = nd_alloc_stack(nd); 1691626de996SAl Viro if (unlikely(error)) { 1692bc40aee0SAl Viro if (error == -ECHILD) { 1693ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1694ad1633a1SAl Viro drop_links(nd); 1695ad1633a1SAl Viro nd->depth = 0; 1696ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1697ad1633a1SAl Viro nd->path.mnt = NULL; 1698ad1633a1SAl Viro nd->path.dentry = NULL; 1699ad1633a1SAl Viro if (!(nd->flags & LOOKUP_ROOT)) 1700ad1633a1SAl Viro nd->root.mnt = NULL; 1701ad1633a1SAl Viro rcu_read_unlock(); 17024675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1703bc40aee0SAl Viro error = nd_alloc_stack(nd); 1704bc40aee0SAl Viro } 1705bc40aee0SAl Viro if (error) { 1706cd179f44SAl Viro path_put(link); 1707626de996SAl Viro return error; 1708626de996SAl Viro } 1709bc40aee0SAl Viro } 1710626de996SAl Viro 1711ab104923SAl Viro last = nd->stack + nd->depth++; 17121cf2665bSAl Viro last->link = *link; 1713fceef393SAl Viro clear_delayed_call(&last->done); 1714fceef393SAl Viro nd->link_inode = inode; 17150450b2d1SAl Viro last->seq = seq; 1716d63ff28fSAl Viro return 1; 1717d63ff28fSAl Viro } 1718d63ff28fSAl Viro 17198f64fb1cSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2}; 172031d66bcdSAl Viro 17213ddcd056SLinus Torvalds /* 17223ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17233ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17243ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17253ddcd056SLinus Torvalds * for the common case. 17263ddcd056SLinus Torvalds */ 17278f64fb1cSAl Viro static inline int step_into(struct nameidata *nd, struct path *path, 17288f64fb1cSAl Viro int flags, struct inode *inode, unsigned seq) 17293ddcd056SLinus Torvalds { 173031d66bcdSAl Viro if (!(flags & WALK_MORE) && nd->depth) 173131d66bcdSAl Viro put_link(nd); 17328f64fb1cSAl Viro if (likely(!d_is_symlink(path->dentry)) || 17338f64fb1cSAl Viro !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) { 17348f64fb1cSAl Viro /* not a symlink or should not follow */ 17358f64fb1cSAl Viro path_to_nameidata(path, nd); 17368f64fb1cSAl Viro nd->inode = inode; 17378f64fb1cSAl Viro nd->seq = seq; 1738d63ff28fSAl Viro return 0; 17398f64fb1cSAl Viro } 1740a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1741a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 17428f64fb1cSAl Viro if (read_seqcount_retry(&path->dentry->d_seq, seq)) 1743a7f77542SAl Viro return -ECHILD; 1744a7f77542SAl Viro } 17458f64fb1cSAl Viro return pick_link(nd, path, inode, seq); 17463ddcd056SLinus Torvalds } 17473ddcd056SLinus Torvalds 17484693a547SAl Viro static int walk_component(struct nameidata *nd, int flags) 1749ce57dfc1SAl Viro { 1750caa85634SAl Viro struct path path; 1751ce57dfc1SAl Viro struct inode *inode; 1752254cf582SAl Viro unsigned seq; 1753ce57dfc1SAl Viro int err; 1754ce57dfc1SAl Viro /* 1755ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1756ce57dfc1SAl Viro * to be able to know about the current root directory and 1757ce57dfc1SAl Viro * parent relationships. 1758ce57dfc1SAl Viro */ 17594693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 17604693a547SAl Viro err = handle_dots(nd, nd->last_type); 17611c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 17624693a547SAl Viro put_link(nd); 17634693a547SAl Viro return err; 17644693a547SAl Viro } 1765254cf582SAl Viro err = lookup_fast(nd, &path, &inode, &seq); 1766e9742b53SAl Viro if (unlikely(err <= 0)) { 1767697f514dSMiklos Szeredi if (err < 0) 1768f0a9ba70SAl Viro return err; 1769e3c13928SAl Viro path.dentry = lookup_slow(&nd->last, nd->path.dentry, 1770e3c13928SAl Viro nd->flags); 1771e3c13928SAl Viro if (IS_ERR(path.dentry)) 1772e3c13928SAl Viro return PTR_ERR(path.dentry); 17737500c38aSAl Viro 1774e3c13928SAl Viro path.mnt = nd->path.mnt; 1775e3c13928SAl Viro err = follow_managed(&path, nd); 1776e3c13928SAl Viro if (unlikely(err < 0)) 1777f0a9ba70SAl Viro return err; 1778697f514dSMiklos Szeredi 17797500c38aSAl Viro if (unlikely(d_is_negative(path.dentry))) { 17807500c38aSAl Viro path_to_nameidata(&path, nd); 17817500c38aSAl Viro return -ENOENT; 17827500c38aSAl Viro } 17837500c38aSAl Viro 1784254cf582SAl Viro seq = 0; /* we are already out of RCU mode */ 1785d4565649SAl Viro inode = d_backing_inode(path.dentry); 1786766c4cbfSAl Viro } 1787697f514dSMiklos Szeredi 17888f64fb1cSAl Viro return step_into(nd, &path, flags, inode, seq); 1789ce57dfc1SAl Viro } 1790ce57dfc1SAl Viro 17911da177e4SLinus Torvalds /* 1792bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1793bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1794bfcfaa77SLinus Torvalds * 1795bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1796bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1797bfcfaa77SLinus Torvalds * fast. 1798bfcfaa77SLinus Torvalds * 1799bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1800bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1801bfcfaa77SLinus Torvalds * crossing operation. 1802bfcfaa77SLinus Torvalds * 1803bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1804bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1805bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1806bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1807bfcfaa77SLinus Torvalds */ 1808bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1809bfcfaa77SLinus Torvalds 1810f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1811bfcfaa77SLinus Torvalds 1812468a9428SGeorge Spelvin #ifdef HASH_MIX 1813bfcfaa77SLinus Torvalds 1814468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1815468a9428SGeorge Spelvin 1816468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18172a18da7aSGeorge Spelvin /* 18182a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18192a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18202a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18212a18da7aSGeorge Spelvin * and no temporaries. 18222a18da7aSGeorge Spelvin * 18232a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18242a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18252a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18262a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18272a18da7aSGeorge Spelvin * 18282a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18292a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18302a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18312a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18322a18da7aSGeorge Spelvin * 18332a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18342a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18352a18da7aSGeorge Spelvin * toggling any given output bit. 18362a18da7aSGeorge Spelvin * 18372a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18382a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18392a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 18402a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 18412a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 18422a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 18432a18da7aSGeorge Spelvin * Perfect: 8192 258048 18442a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 18452a18da7aSGeorge Spelvin */ 18462a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18472a18da7aSGeorge Spelvin ( x ^= (a), \ 18482a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 18492a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 18502a18da7aSGeorge Spelvin y *= 9 ) 1851bfcfaa77SLinus Torvalds 18520fed3ac8SGeorge Spelvin /* 18532a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 18542a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 18552a18da7aSGeorge Spelvin * work done before the hash value is used. 18560fed3ac8SGeorge Spelvin */ 18572a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18580fed3ac8SGeorge Spelvin { 18592a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 18602a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 18612a18da7aSGeorge Spelvin return y >> 32; 18620fed3ac8SGeorge Spelvin } 18630fed3ac8SGeorge Spelvin 1864bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1865bfcfaa77SLinus Torvalds 18662a18da7aSGeorge Spelvin /* 18672a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 18682a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 18692a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 18702a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 18712a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 18722a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 18732a18da7aSGeorge Spelvin * Perfect: 2048 31744 18742a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 18752a18da7aSGeorge Spelvin */ 18762a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 18772a18da7aSGeorge Spelvin ( x ^= (a), \ 18782a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 18792a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 18802a18da7aSGeorge Spelvin y *= 9 ) 1881bfcfaa77SLinus Torvalds 18822a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 18830fed3ac8SGeorge Spelvin { 18842a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 18852a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 18860fed3ac8SGeorge Spelvin } 18870fed3ac8SGeorge Spelvin 1888bfcfaa77SLinus Torvalds #endif 1889bfcfaa77SLinus Torvalds 18902a18da7aSGeorge Spelvin /* 18912a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 18922a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 18932a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 18942a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 18952a18da7aSGeorge Spelvin * finds the delimiter after the name. 18962a18da7aSGeorge Spelvin */ 18978387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1898bfcfaa77SLinus Torvalds { 18998387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1900bfcfaa77SLinus Torvalds 1901bfcfaa77SLinus Torvalds for (;;) { 1902fcfd2fbfSGeorge Spelvin if (!len) 1903fcfd2fbfSGeorge Spelvin goto done; 1904e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1905bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1906bfcfaa77SLinus Torvalds break; 19072a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1908bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1909bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1910bfcfaa77SLinus Torvalds } 19112a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1912bfcfaa77SLinus Torvalds done: 19132a18da7aSGeorge Spelvin return fold_hash(x, y); 1914bfcfaa77SLinus Torvalds } 1915bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1916bfcfaa77SLinus Torvalds 1917fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19188387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1919fcfd2fbfSGeorge Spelvin { 19208387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19218387ff25SLinus Torvalds unsigned long adata, mask, len; 1922fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1923fcfd2fbfSGeorge Spelvin 19248387ff25SLinus Torvalds len = 0; 19258387ff25SLinus Torvalds goto inside; 19268387ff25SLinus Torvalds 1927fcfd2fbfSGeorge Spelvin do { 19282a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1929fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19308387ff25SLinus Torvalds inside: 1931fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1932fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1933fcfd2fbfSGeorge Spelvin 1934fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1935fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19362a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1937fcfd2fbfSGeorge Spelvin 19382a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1939fcfd2fbfSGeorge Spelvin } 1940fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1941fcfd2fbfSGeorge Spelvin 1942bfcfaa77SLinus Torvalds /* 1943bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 1944d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 1945bfcfaa77SLinus Torvalds */ 19468387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 1947bfcfaa77SLinus Torvalds { 19488387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 19498387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 195036126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1951bfcfaa77SLinus Torvalds 19528387ff25SLinus Torvalds len = 0; 19538387ff25SLinus Torvalds goto inside; 19548387ff25SLinus Torvalds 1955bfcfaa77SLinus Torvalds do { 19562a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1957bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 19588387ff25SLinus Torvalds inside: 1959e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 196036126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 196136126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 1962bfcfaa77SLinus Torvalds 196336126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 196436126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 196536126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 19662a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 196736126f8fSLinus Torvalds 19682a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 1969bfcfaa77SLinus Torvalds } 1970bfcfaa77SLinus Torvalds 19712a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 1972bfcfaa77SLinus Torvalds 1973fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 19748387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 19750145acc2SLinus Torvalds { 19768387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 19770145acc2SLinus Torvalds while (len--) 1978fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 19790145acc2SLinus Torvalds return end_name_hash(hash); 19800145acc2SLinus Torvalds } 1981ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 19820145acc2SLinus Torvalds 1983fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19848387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1985fcfd2fbfSGeorge Spelvin { 19868387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 1987fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 1988fcfd2fbfSGeorge Spelvin 1989fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 1990e0ab7af9SGeorge Spelvin while (c) { 1991fcfd2fbfSGeorge Spelvin len++; 1992fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 1993fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 1994e0ab7af9SGeorge Spelvin } 1995fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 1996fcfd2fbfSGeorge Spelvin } 1997f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 1998fcfd2fbfSGeorge Spelvin 19993ddcd056SLinus Torvalds /* 2000200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2001200e9ef7SLinus Torvalds * one character. 2002200e9ef7SLinus Torvalds */ 20038387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2004200e9ef7SLinus Torvalds { 20058387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2006200e9ef7SLinus Torvalds unsigned long len = 0, c; 2007200e9ef7SLinus Torvalds 2008200e9ef7SLinus Torvalds c = (unsigned char)*name; 2009200e9ef7SLinus Torvalds do { 2010200e9ef7SLinus Torvalds len++; 2011200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2012200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2013200e9ef7SLinus Torvalds } while (c && c != '/'); 2014d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2015200e9ef7SLinus Torvalds } 2016200e9ef7SLinus Torvalds 2017bfcfaa77SLinus Torvalds #endif 2018bfcfaa77SLinus Torvalds 2019200e9ef7SLinus Torvalds /* 20201da177e4SLinus Torvalds * Name resolution. 2021ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2022ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20231da177e4SLinus Torvalds * 2024ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2025ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20261da177e4SLinus Torvalds */ 20276de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20281da177e4SLinus Torvalds { 20291da177e4SLinus Torvalds int err; 20301da177e4SLinus Torvalds 20311da177e4SLinus Torvalds while (*name=='/') 20321da177e4SLinus Torvalds name++; 20331da177e4SLinus Torvalds if (!*name) 20349e18f10aSAl Viro return 0; 20351da177e4SLinus Torvalds 20361da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 20371da177e4SLinus Torvalds for(;;) { 2038d6bb3e90SLinus Torvalds u64 hash_len; 2039fe479a58SAl Viro int type; 20401da177e4SLinus Torvalds 204152094c8aSAl Viro err = may_lookup(nd); 20421da177e4SLinus Torvalds if (err) 20433595e234SAl Viro return err; 20441da177e4SLinus Torvalds 20458387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 20461da177e4SLinus Torvalds 2047fe479a58SAl Viro type = LAST_NORM; 2048d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2049fe479a58SAl Viro case 2: 2050200e9ef7SLinus Torvalds if (name[1] == '.') { 2051fe479a58SAl Viro type = LAST_DOTDOT; 205216c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 205316c2cd71SAl Viro } 2054fe479a58SAl Viro break; 2055fe479a58SAl Viro case 1: 2056fe479a58SAl Viro type = LAST_DOT; 2057fe479a58SAl Viro } 20585a202bcdSAl Viro if (likely(type == LAST_NORM)) { 20595a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 206016c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 20615a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2062a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2063da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 20645a202bcdSAl Viro if (err < 0) 20653595e234SAl Viro return err; 2066d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2067d6bb3e90SLinus Torvalds name = this.name; 20685a202bcdSAl Viro } 20695a202bcdSAl Viro } 2070fe479a58SAl Viro 2071d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2072d6bb3e90SLinus Torvalds nd->last.name = name; 20735f4a6a69SAl Viro nd->last_type = type; 20745f4a6a69SAl Viro 2075d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2076d6bb3e90SLinus Torvalds if (!*name) 2077bdf6cbf1SAl Viro goto OK; 2078200e9ef7SLinus Torvalds /* 2079200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2080200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2081200e9ef7SLinus Torvalds */ 2082200e9ef7SLinus Torvalds do { 2083d6bb3e90SLinus Torvalds name++; 2084d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 20858620c238SAl Viro if (unlikely(!*name)) { 20868620c238SAl Viro OK: 2087368ee9baSAl Viro /* pathname body, done */ 20888620c238SAl Viro if (!nd->depth) 20898620c238SAl Viro return 0; 20908620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2091368ee9baSAl Viro /* trailing symlink, done */ 20928620c238SAl Viro if (!name) 20938620c238SAl Viro return 0; 20948620c238SAl Viro /* last component of nested symlink */ 20958f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW); 20961c4ff1a8SAl Viro } else { 20971c4ff1a8SAl Viro /* not the last component */ 20988f64fb1cSAl Viro err = walk_component(nd, WALK_FOLLOW | WALK_MORE); 20998620c238SAl Viro } 2100ce57dfc1SAl Viro if (err < 0) 21013595e234SAl Viro return err; 2102fe479a58SAl Viro 2103ce57dfc1SAl Viro if (err) { 2104626de996SAl Viro const char *s = get_link(nd); 21055a460275SAl Viro 2106a1c83681SViresh Kumar if (IS_ERR(s)) 21073595e234SAl Viro return PTR_ERR(s); 2108172a39a0SAl Viro err = 0; 210912b09578SAl Viro if (unlikely(!s)) { 211012b09578SAl Viro /* jumped */ 2111b9ff4429SAl Viro put_link(nd); 211212b09578SAl Viro } else { 2113071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 211432cd7468SAl Viro name = s; 21159e18f10aSAl Viro continue; 211648c8b0c5SAl Viro } 211731e6b01fSNick Piggin } 211897242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 211997242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21204675ac39SAl Viro if (unlazy_walk(nd)) 212197242f99SAl Viro return -ECHILD; 212297242f99SAl Viro } 21233595e234SAl Viro return -ENOTDIR; 21245f4a6a69SAl Viro } 2125ce57dfc1SAl Viro } 212697242f99SAl Viro } 21271da177e4SLinus Torvalds 2128c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 212931e6b01fSNick Piggin { 2130c8a53ee5SAl Viro const char *s = nd->name->name; 213131e6b01fSNick Piggin 2132c0eb027eSLinus Torvalds if (!*s) 2133c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2134c0eb027eSLinus Torvalds 213531e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2136980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 213731e6b01fSNick Piggin nd->depth = 0; 21385b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2139b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2140b18825a7SDavid Howells struct inode *inode = root->d_inode; 214193893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2142368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 21435b6ca027SAl Viro nd->path = nd->root; 21445b6ca027SAl Viro nd->inode = inode; 21455b6ca027SAl Viro if (flags & LOOKUP_RCU) { 21468b61e74fSAl Viro rcu_read_lock(); 21475b6ca027SAl Viro nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 21488f47a016SAl Viro nd->root_seq = nd->seq; 214948a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 21505b6ca027SAl Viro } else { 21515b6ca027SAl Viro path_get(&nd->path); 21525b6ca027SAl Viro } 2153368ee9baSAl Viro return s; 21545b6ca027SAl Viro } 21555b6ca027SAl Viro 215631e6b01fSNick Piggin nd->root.mnt = NULL; 2157248fb5b9SAl Viro nd->path.mnt = NULL; 2158248fb5b9SAl Viro nd->path.dentry = NULL; 215931e6b01fSNick Piggin 216048a066e7SAl Viro nd->m_seq = read_seqbegin(&mount_lock); 2161fd2f7cb5SAl Viro if (*s == '/') { 21629e6697e2SAl Viro if (flags & LOOKUP_RCU) 21638b61e74fSAl Viro rcu_read_lock(); 2164e41f7d4eSAl Viro set_root(nd); 2165248fb5b9SAl Viro if (likely(!nd_jump_root(nd))) 2166ef55d917SAl Viro return s; 2167ef55d917SAl Viro nd->root.mnt = NULL; 2168ef55d917SAl Viro rcu_read_unlock(); 2169ef55d917SAl Viro return ERR_PTR(-ECHILD); 2170c8a53ee5SAl Viro } else if (nd->dfd == AT_FDCWD) { 2171e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 217231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2173c28cc364SNick Piggin unsigned seq; 217431e6b01fSNick Piggin 21758b61e74fSAl Viro rcu_read_lock(); 217631e6b01fSNick Piggin 2177c28cc364SNick Piggin do { 2178c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 217931e6b01fSNick Piggin nd->path = fs->pwd; 2180ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2181c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2182c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2183e41f7d4eSAl Viro } else { 2184e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2185ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2186e41f7d4eSAl Viro } 2187ef55d917SAl Viro return s; 218831e6b01fSNick Piggin } else { 2189582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2190c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 219131e6b01fSNick Piggin struct dentry *dentry; 219231e6b01fSNick Piggin 21932903ff01SAl Viro if (!f.file) 2194368ee9baSAl Viro return ERR_PTR(-EBADF); 219531e6b01fSNick Piggin 21962903ff01SAl Viro dentry = f.file->f_path.dentry; 219731e6b01fSNick Piggin 2198fd2f7cb5SAl Viro if (*s) { 219944b1d530SMiklos Szeredi if (!d_can_lookup(dentry)) { 22002903ff01SAl Viro fdput(f); 2201368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2202f52e0c11SAl Viro } 22032903ff01SAl Viro } 22042903ff01SAl Viro 22052903ff01SAl Viro nd->path = f.file->f_path; 2206e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 22078b61e74fSAl Viro rcu_read_lock(); 220834a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 220934a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22105590ff0dSUlrich Drepper } else { 22112903ff01SAl Viro path_get(&nd->path); 221234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22131da177e4SLinus Torvalds } 221434a26b99SAl Viro fdput(f); 2215368ee9baSAl Viro return s; 2216e41f7d4eSAl Viro } 22179b4a9b14SAl Viro } 22189b4a9b14SAl Viro 22193bdba28bSAl Viro static const char *trailing_symlink(struct nameidata *nd) 222095fa25d9SAl Viro { 222195fa25d9SAl Viro const char *s; 2222fec2fa24SAl Viro int error = may_follow_link(nd); 2223deb106c6SAl Viro if (unlikely(error)) 22243bdba28bSAl Viro return ERR_PTR(error); 222595fa25d9SAl Viro nd->flags |= LOOKUP_PARENT; 2226fab51e8aSAl Viro nd->stack[0].name = NULL; 22273b2e7f75SAl Viro s = get_link(nd); 2228deb106c6SAl Viro return s ? s : ""; 222995fa25d9SAl Viro } 223095fa25d9SAl Viro 2231caa85634SAl Viro static inline int lookup_last(struct nameidata *nd) 2232bd92d7feSAl Viro { 2233bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2234bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2235bd92d7feSAl Viro 2236bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 22371c4ff1a8SAl Viro return walk_component(nd, 0); 2238bd92d7feSAl Viro } 2239bd92d7feSAl Viro 22404f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22414f757f3cSAl Viro { 22424f757f3cSAl Viro struct path path = nd->path; 22434f757f3cSAl Viro struct inode *inode = nd->inode; 22444f757f3cSAl Viro unsigned seq = nd->seq; 22454f757f3cSAl Viro int err; 22464f757f3cSAl Viro 22474f757f3cSAl Viro if (nd->flags & LOOKUP_RCU) { 22484f757f3cSAl Viro /* 22494f757f3cSAl Viro * don't bother with unlazy_walk on failure - we are 22504f757f3cSAl Viro * at the very beginning of walk, so we lose nothing 22514f757f3cSAl Viro * if we simply redo everything in non-RCU mode 22524f757f3cSAl Viro */ 22534f757f3cSAl Viro if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq))) 22544f757f3cSAl Viro return -ECHILD; 22554f757f3cSAl Viro } else { 22564f757f3cSAl Viro dget(path.dentry); 22574f757f3cSAl Viro err = follow_managed(&path, nd); 22584f757f3cSAl Viro if (unlikely(err < 0)) 22594f757f3cSAl Viro return err; 22604f757f3cSAl Viro inode = d_backing_inode(path.dentry); 22614f757f3cSAl Viro seq = 0; 22624f757f3cSAl Viro } 22634f757f3cSAl Viro path_to_nameidata(&path, nd); 22644f757f3cSAl Viro nd->inode = inode; 22654f757f3cSAl Viro nd->seq = seq; 22664f757f3cSAl Viro return 0; 22674f757f3cSAl Viro } 22684f757f3cSAl Viro 22699b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2270c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 22719b4a9b14SAl Viro { 2272c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2273bd92d7feSAl Viro int err; 227431e6b01fSNick Piggin 2275368ee9baSAl Viro if (IS_ERR(s)) 2276368ee9baSAl Viro return PTR_ERR(s); 22774f757f3cSAl Viro 22784f757f3cSAl Viro if (unlikely(flags & LOOKUP_DOWN)) { 22794f757f3cSAl Viro err = handle_lookup_down(nd); 22804f757f3cSAl Viro if (unlikely(err < 0)) { 22814f757f3cSAl Viro terminate_walk(nd); 22824f757f3cSAl Viro return err; 22834f757f3cSAl Viro } 22844f757f3cSAl Viro } 22854f757f3cSAl Viro 22863bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) 22873bdba28bSAl Viro && ((err = lookup_last(nd)) > 0)) { 22883bdba28bSAl Viro s = trailing_symlink(nd); 22893bdba28bSAl Viro if (IS_ERR(s)) { 22903bdba28bSAl Viro err = PTR_ERR(s); 22916d7b5aaeSAl Viro break; 2292bd92d7feSAl Viro } 2293bd92d7feSAl Viro } 22949f1fafeeSAl Viro if (!err) 22959f1fafeeSAl Viro err = complete_walk(nd); 2296bd92d7feSAl Viro 2297deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2298deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2299bd23a539SAl Viro err = -ENOTDIR; 2300625b6d10SAl Viro if (!err) { 2301625b6d10SAl Viro *path = nd->path; 2302625b6d10SAl Viro nd->path.mnt = NULL; 2303625b6d10SAl Viro nd->path.dentry = NULL; 2304625b6d10SAl Viro } 2305deb106c6SAl Viro terminate_walk(nd); 2306bd92d7feSAl Viro return err; 230731e6b01fSNick Piggin } 230831e6b01fSNick Piggin 2309625b6d10SAl Viro static int filename_lookup(int dfd, struct filename *name, unsigned flags, 23109ad1aaa6SAl Viro struct path *path, struct path *root) 2311873f1eedSJeff Layton { 2312894bc8c4SAl Viro int retval; 23139883d185SAl Viro struct nameidata nd; 2314abc9f5beSAl Viro if (IS_ERR(name)) 2315abc9f5beSAl Viro return PTR_ERR(name); 23169ad1aaa6SAl Viro if (unlikely(root)) { 23179ad1aaa6SAl Viro nd.root = *root; 23189ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23199ad1aaa6SAl Viro } 23209883d185SAl Viro set_nameidata(&nd, dfd, name); 2321c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2322873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2323c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2324873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2325c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2326873f1eedSJeff Layton 2327873f1eedSJeff Layton if (likely(!retval)) 2328625b6d10SAl Viro audit_inode(name, path->dentry, flags & LOOKUP_PARENT); 23299883d185SAl Viro restore_nameidata(); 2330e4bd1c1aSAl Viro putname(name); 2331873f1eedSJeff Layton return retval; 2332873f1eedSJeff Layton } 2333873f1eedSJeff Layton 23348bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2335c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2336391172c4SAl Viro struct path *parent) 23378bcb77faSAl Viro { 2338c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2339368ee9baSAl Viro int err; 2340368ee9baSAl Viro if (IS_ERR(s)) 2341368ee9baSAl Viro return PTR_ERR(s); 2342368ee9baSAl Viro err = link_path_walk(s, nd); 23438bcb77faSAl Viro if (!err) 23448bcb77faSAl Viro err = complete_walk(nd); 2345391172c4SAl Viro if (!err) { 2346391172c4SAl Viro *parent = nd->path; 2347391172c4SAl Viro nd->path.mnt = NULL; 2348391172c4SAl Viro nd->path.dentry = NULL; 2349391172c4SAl Viro } 2350deb106c6SAl Viro terminate_walk(nd); 23518bcb77faSAl Viro return err; 23528bcb77faSAl Viro } 23538bcb77faSAl Viro 23545c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2355391172c4SAl Viro unsigned int flags, struct path *parent, 2356391172c4SAl Viro struct qstr *last, int *type) 23578bcb77faSAl Viro { 23588bcb77faSAl Viro int retval; 23599883d185SAl Viro struct nameidata nd; 23608bcb77faSAl Viro 23615c31b6ceSAl Viro if (IS_ERR(name)) 23625c31b6ceSAl Viro return name; 23639883d185SAl Viro set_nameidata(&nd, dfd, name); 2364c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23658bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2366c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23678bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2368c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2369391172c4SAl Viro if (likely(!retval)) { 2370391172c4SAl Viro *last = nd.last; 2371391172c4SAl Viro *type = nd.last_type; 2372391172c4SAl Viro audit_inode(name, parent->dentry, LOOKUP_PARENT); 23735c31b6ceSAl Viro } else { 23745c31b6ceSAl Viro putname(name); 23755c31b6ceSAl Viro name = ERR_PTR(retval); 2376391172c4SAl Viro } 23779883d185SAl Viro restore_nameidata(); 23785c31b6ceSAl Viro return name; 23798bcb77faSAl Viro } 23808bcb77faSAl Viro 238179714f72SAl Viro /* does lookup, returns the object with parent locked */ 238279714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 23835590ff0dSUlrich Drepper { 23845c31b6ceSAl Viro struct filename *filename; 23855c31b6ceSAl Viro struct dentry *d; 2386391172c4SAl Viro struct qstr last; 2387391172c4SAl Viro int type; 238851689104SPaul Moore 23895c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 23905c31b6ceSAl Viro &last, &type); 239151689104SPaul Moore if (IS_ERR(filename)) 239251689104SPaul Moore return ERR_CAST(filename); 23935c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2394391172c4SAl Viro path_put(path); 23955c31b6ceSAl Viro putname(filename); 23965c31b6ceSAl Viro return ERR_PTR(-EINVAL); 239779714f72SAl Viro } 23985955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2399391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 240079714f72SAl Viro if (IS_ERR(d)) { 24015955102cSAl Viro inode_unlock(path->dentry->d_inode); 2402391172c4SAl Viro path_put(path); 240379714f72SAl Viro } 240451689104SPaul Moore putname(filename); 240579714f72SAl Viro return d; 24065590ff0dSUlrich Drepper } 24075590ff0dSUlrich Drepper 2408d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2409d1811465SAl Viro { 2410abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2411abc9f5beSAl Viro flags, path, NULL); 2412d1811465SAl Viro } 24134d359507SAl Viro EXPORT_SYMBOL(kern_path); 2414d1811465SAl Viro 241516f18200SJosef 'Jeff' Sipek /** 241616f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 241716f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 241816f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 241916f18200SJosef 'Jeff' Sipek * @name: pointer to file name 242016f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2421e0a01249SAl Viro * @path: pointer to struct path to fill 242216f18200SJosef 'Jeff' Sipek */ 242316f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 242416f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2425e0a01249SAl Viro struct path *path) 242616f18200SJosef 'Jeff' Sipek { 24279ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24289ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2429abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2430abc9f5beSAl Viro flags , path, &root); 243116f18200SJosef 'Jeff' Sipek } 24324d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 243316f18200SJosef 'Jeff' Sipek 24343c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24353c95f0dcSAl Viro int len, struct qstr *this) 24363c95f0dcSAl Viro { 24373c95f0dcSAl Viro this->name = name; 24383c95f0dcSAl Viro this->len = len; 24393c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24403c95f0dcSAl Viro if (!len) 24413c95f0dcSAl Viro return -EACCES; 24423c95f0dcSAl Viro 24433c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24443c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24453c95f0dcSAl Viro return -EACCES; 24463c95f0dcSAl Viro } 24473c95f0dcSAl Viro 24483c95f0dcSAl Viro while (len--) { 24493c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24503c95f0dcSAl Viro if (c == '/' || c == '\0') 24513c95f0dcSAl Viro return -EACCES; 24523c95f0dcSAl Viro } 24533c95f0dcSAl Viro /* 24543c95f0dcSAl Viro * See if the low-level filesystem might want 24553c95f0dcSAl Viro * to use its own hash.. 24563c95f0dcSAl Viro */ 24573c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 24583c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 24593c95f0dcSAl Viro if (err < 0) 24603c95f0dcSAl Viro return err; 24613c95f0dcSAl Viro } 24623c95f0dcSAl Viro 24633c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 24643c95f0dcSAl Viro } 24653c95f0dcSAl Viro 2466eead1911SChristoph Hellwig /** 24670da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 24680da0b7fdSDavid Howells * @name: pathname component to lookup 24690da0b7fdSDavid Howells * @base: base directory to lookup from 24700da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 24710da0b7fdSDavid Howells * 24720da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 24730da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 24740da0b7fdSDavid Howells * 24750da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 24760da0b7fdSDavid Howells * not be called by generic code. 24770da0b7fdSDavid Howells * 24780da0b7fdSDavid Howells * The caller must hold base->i_mutex. 24790da0b7fdSDavid Howells */ 24800da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 24810da0b7fdSDavid Howells { 24820da0b7fdSDavid Howells struct qstr this; 24830da0b7fdSDavid Howells int err; 24840da0b7fdSDavid Howells 24850da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 24860da0b7fdSDavid Howells 24870da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 24880da0b7fdSDavid Howells if (err) 24890da0b7fdSDavid Howells return ERR_PTR(err); 24900da0b7fdSDavid Howells 24910da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 24920da0b7fdSDavid Howells } 24930da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 24940da0b7fdSDavid Howells 24950da0b7fdSDavid Howells /** 2496a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2497eead1911SChristoph Hellwig * @name: pathname component to lookup 2498eead1911SChristoph Hellwig * @base: base directory to lookup from 2499eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2500eead1911SChristoph Hellwig * 2501a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25029e7543e9SAl Viro * not be called by generic code. 2503bbddca8eSNeilBrown * 2504bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2505eead1911SChristoph Hellwig */ 2506057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2507057f6c01SJames Morris { 25088613a209SAl Viro struct dentry *dentry; 2509057f6c01SJames Morris struct qstr this; 2510cda309deSMiklos Szeredi int err; 2511057f6c01SJames Morris 25125955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25132f9092e1SDavid Woodhouse 25143c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2515cda309deSMiklos Szeredi if (err) 2516cda309deSMiklos Szeredi return ERR_PTR(err); 2517cda309deSMiklos Szeredi 25188613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25198613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2520057f6c01SJames Morris } 25214d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2522057f6c01SJames Morris 2523bbddca8eSNeilBrown /** 2524bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2525bbddca8eSNeilBrown * @name: pathname component to lookup 2526bbddca8eSNeilBrown * @base: base directory to lookup from 2527bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2528bbddca8eSNeilBrown * 2529bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2530bbddca8eSNeilBrown * not be called by generic code. 2531bbddca8eSNeilBrown * 2532bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2533bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2534bbddca8eSNeilBrown */ 2535bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2536bbddca8eSNeilBrown struct dentry *base, int len) 2537bbddca8eSNeilBrown { 2538bbddca8eSNeilBrown struct qstr this; 2539bbddca8eSNeilBrown int err; 254020d00ee8SLinus Torvalds struct dentry *ret; 2541bbddca8eSNeilBrown 25423c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2543bbddca8eSNeilBrown if (err) 2544bbddca8eSNeilBrown return ERR_PTR(err); 2545bbddca8eSNeilBrown 254620d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 254720d00ee8SLinus Torvalds if (!ret) 254820d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 254920d00ee8SLinus Torvalds return ret; 2550bbddca8eSNeilBrown } 2551bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2552bbddca8eSNeilBrown 2553eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2554eedf265aSEric W. Biederman int path_pts(struct path *path) 2555eedf265aSEric W. Biederman { 2556eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2557eedf265aSEric W. Biederman * the input path. 2558eedf265aSEric W. Biederman */ 2559eedf265aSEric W. Biederman struct dentry *child, *parent; 2560eedf265aSEric W. Biederman struct qstr this; 2561eedf265aSEric W. Biederman int ret; 2562eedf265aSEric W. Biederman 2563eedf265aSEric W. Biederman ret = path_parent_directory(path); 2564eedf265aSEric W. Biederman if (ret) 2565eedf265aSEric W. Biederman return ret; 2566eedf265aSEric W. Biederman 2567eedf265aSEric W. Biederman parent = path->dentry; 2568eedf265aSEric W. Biederman this.name = "pts"; 2569eedf265aSEric W. Biederman this.len = 3; 2570eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2571eedf265aSEric W. Biederman if (!child) 2572eedf265aSEric W. Biederman return -ENOENT; 2573eedf265aSEric W. Biederman 2574eedf265aSEric W. Biederman path->dentry = child; 2575eedf265aSEric W. Biederman dput(parent); 2576eedf265aSEric W. Biederman follow_mount(path); 2577eedf265aSEric W. Biederman return 0; 2578eedf265aSEric W. Biederman } 2579eedf265aSEric W. Biederman #endif 2580eedf265aSEric W. Biederman 25811fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 25821fa1e7f6SAndy Whitcroft struct path *path, int *empty) 25831da177e4SLinus Torvalds { 2584abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2585abc9f5beSAl Viro flags, path, NULL); 25861da177e4SLinus Torvalds } 2587b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 25881fa1e7f6SAndy Whitcroft 25898033426eSJeff Layton /** 2590197df04cSAl Viro * mountpoint_last - look up last component for umount 25918033426eSJeff Layton * @nd: pathwalk nameidata - currently pointing at parent directory of "last" 25928033426eSJeff Layton * 25938033426eSJeff Layton * This is a special lookup_last function just for umount. In this case, we 25948033426eSJeff Layton * need to resolve the path without doing any revalidation. 25958033426eSJeff Layton * 25968033426eSJeff Layton * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since 25978033426eSJeff Layton * mountpoints are always pinned in the dcache, their ancestors are too. Thus, 25988033426eSJeff Layton * in almost all cases, this lookup will be served out of the dcache. The only 25998033426eSJeff Layton * cases where it won't are if nd->last refers to a symlink or the path is 26008033426eSJeff Layton * bogus and it doesn't exist. 26018033426eSJeff Layton * 26028033426eSJeff Layton * Returns: 26038033426eSJeff Layton * -error: if there was an error during lookup. This includes -ENOENT if the 2604ba8f4613SAl Viro * lookup found a negative dentry. 26058033426eSJeff Layton * 2606ba8f4613SAl Viro * 0: if we successfully resolved nd->last and found it to not to be a 2607ba8f4613SAl Viro * symlink that needs to be followed. 26088033426eSJeff Layton * 26098033426eSJeff Layton * 1: if we successfully resolved nd->last and found it to be a symlink 2610ba8f4613SAl Viro * that needs to be followed. 26118033426eSJeff Layton */ 26128033426eSJeff Layton static int 2613ba8f4613SAl Viro mountpoint_last(struct nameidata *nd) 26148033426eSJeff Layton { 26158033426eSJeff Layton int error = 0; 26168033426eSJeff Layton struct dentry *dir = nd->path.dentry; 2617ba8f4613SAl Viro struct path path; 26188033426eSJeff Layton 261935759521SAl Viro /* If we're in rcuwalk, drop out of it to handle last component */ 262035759521SAl Viro if (nd->flags & LOOKUP_RCU) { 26214675ac39SAl Viro if (unlazy_walk(nd)) 2622deb106c6SAl Viro return -ECHILD; 26238033426eSJeff Layton } 26248033426eSJeff Layton 26258033426eSJeff Layton nd->flags &= ~LOOKUP_PARENT; 26268033426eSJeff Layton 26278033426eSJeff Layton if (unlikely(nd->last_type != LAST_NORM)) { 26288033426eSJeff Layton error = handle_dots(nd, nd->last_type); 262935759521SAl Viro if (error) 2630deb106c6SAl Viro return error; 2631ba8f4613SAl Viro path.dentry = dget(nd->path.dentry); 2632949a852eSAl Viro } else { 2633ba8f4613SAl Viro path.dentry = d_lookup(dir, &nd->last); 2634ba8f4613SAl Viro if (!path.dentry) { 26358033426eSJeff Layton /* 2636949a852eSAl Viro * No cached dentry. Mounted dentries are pinned in the 2637949a852eSAl Viro * cache, so that means that this dentry is probably 2638949a852eSAl Viro * a symlink or the path doesn't actually point 2639949a852eSAl Viro * to a mounted dentry. 26408033426eSJeff Layton */ 2641ba8f4613SAl Viro path.dentry = lookup_slow(&nd->last, dir, 2642949a852eSAl Viro nd->flags | LOOKUP_NO_REVAL); 2643ba8f4613SAl Viro if (IS_ERR(path.dentry)) 2644ba8f4613SAl Viro return PTR_ERR(path.dentry); 26458033426eSJeff Layton } 2646bcceeebaSDave Jones } 2647ba8f4613SAl Viro if (d_is_negative(path.dentry)) { 2648ba8f4613SAl Viro dput(path.dentry); 2649deb106c6SAl Viro return -ENOENT; 265035759521SAl Viro } 2651ba8f4613SAl Viro path.mnt = nd->path.mnt; 26528f64fb1cSAl Viro return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0); 26538033426eSJeff Layton } 26548033426eSJeff Layton 26558033426eSJeff Layton /** 2656197df04cSAl Viro * path_mountpoint - look up a path to be umounted 26572a78b857SMasanari Iida * @nd: lookup context 26588033426eSJeff Layton * @flags: lookup flags 2659c8a53ee5SAl Viro * @path: pointer to container for result 26608033426eSJeff Layton * 26618033426eSJeff Layton * Look up the given name, but don't attempt to revalidate the last component. 2662606d6fe3SRandy Dunlap * Returns 0 and "path" will be valid on success; Returns error otherwise. 26638033426eSJeff Layton */ 26648033426eSJeff Layton static int 2665c8a53ee5SAl Viro path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path) 26668033426eSJeff Layton { 2667c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2668368ee9baSAl Viro int err; 2669368ee9baSAl Viro if (IS_ERR(s)) 2670368ee9baSAl Viro return PTR_ERR(s); 26713bdba28bSAl Viro while (!(err = link_path_walk(s, nd)) && 2672ba8f4613SAl Viro (err = mountpoint_last(nd)) > 0) { 26733bdba28bSAl Viro s = trailing_symlink(nd); 26743bdba28bSAl Viro if (IS_ERR(s)) { 26753bdba28bSAl Viro err = PTR_ERR(s); 26768033426eSJeff Layton break; 26778033426eSJeff Layton } 26783bdba28bSAl Viro } 2679ba8f4613SAl Viro if (!err) { 2680ba8f4613SAl Viro *path = nd->path; 2681ba8f4613SAl Viro nd->path.mnt = NULL; 2682ba8f4613SAl Viro nd->path.dentry = NULL; 2683ba8f4613SAl Viro follow_mount(path); 2684ba8f4613SAl Viro } 2685deb106c6SAl Viro terminate_walk(nd); 26868033426eSJeff Layton return err; 26878033426eSJeff Layton } 26888033426eSJeff Layton 26892d864651SAl Viro static int 2690668696dcSAl Viro filename_mountpoint(int dfd, struct filename *name, struct path *path, 26912d864651SAl Viro unsigned int flags) 26922d864651SAl Viro { 26939883d185SAl Viro struct nameidata nd; 2694cbaab2dbSAl Viro int error; 2695668696dcSAl Viro if (IS_ERR(name)) 2696668696dcSAl Viro return PTR_ERR(name); 26979883d185SAl Viro set_nameidata(&nd, dfd, name); 2698c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_RCU, path); 26992d864651SAl Viro if (unlikely(error == -ECHILD)) 2700c8a53ee5SAl Viro error = path_mountpoint(&nd, flags, path); 27012d864651SAl Viro if (unlikely(error == -ESTALE)) 2702c8a53ee5SAl Viro error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path); 27032d864651SAl Viro if (likely(!error)) 2704668696dcSAl Viro audit_inode(name, path->dentry, 0); 27059883d185SAl Viro restore_nameidata(); 2706668696dcSAl Viro putname(name); 27072d864651SAl Viro return error; 27082d864651SAl Viro } 27092d864651SAl Viro 27108033426eSJeff Layton /** 2711197df04cSAl Viro * user_path_mountpoint_at - lookup a path from userland in order to umount it 27128033426eSJeff Layton * @dfd: directory file descriptor 27138033426eSJeff Layton * @name: pathname from userland 27148033426eSJeff Layton * @flags: lookup flags 27158033426eSJeff Layton * @path: pointer to container to hold result 27168033426eSJeff Layton * 27178033426eSJeff Layton * A umount is a special case for path walking. We're not actually interested 27188033426eSJeff Layton * in the inode in this situation, and ESTALE errors can be a problem. We 27198033426eSJeff Layton * simply want track down the dentry and vfsmount attached at the mountpoint 27208033426eSJeff Layton * and avoid revalidating the last component. 27218033426eSJeff Layton * 27228033426eSJeff Layton * Returns 0 and populates "path" on success. 27238033426eSJeff Layton */ 27248033426eSJeff Layton int 2725197df04cSAl Viro user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags, 27268033426eSJeff Layton struct path *path) 27278033426eSJeff Layton { 2728cbaab2dbSAl Viro return filename_mountpoint(dfd, getname(name), path, flags); 27298033426eSJeff Layton } 27308033426eSJeff Layton 27312d864651SAl Viro int 27322d864651SAl Viro kern_path_mountpoint(int dfd, const char *name, struct path *path, 27332d864651SAl Viro unsigned int flags) 27342d864651SAl Viro { 2735cbaab2dbSAl Viro return filename_mountpoint(dfd, getname_kernel(name), path, flags); 27362d864651SAl Viro } 27372d864651SAl Viro EXPORT_SYMBOL(kern_path_mountpoint); 27382d864651SAl Viro 2739cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 27401da177e4SLinus Torvalds { 27418e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2742da9592edSDavid Howells 27438e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 27441da177e4SLinus Torvalds return 0; 27458e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 27461da177e4SLinus Torvalds return 0; 274723adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 27481da177e4SLinus Torvalds } 2749cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27501da177e4SLinus Torvalds 27511da177e4SLinus Torvalds /* 27521da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27531da177e4SLinus Torvalds * whether the type of victim is right. 27541da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27551da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27561da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27571da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27581da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27591da177e4SLinus Torvalds * a. be owner of dir, or 27601da177e4SLinus Torvalds * b. be owner of victim, or 27611da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27621da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27631da177e4SLinus Torvalds * links pointing to it. 27640bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27650bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27660bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27670bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27680bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27691da177e4SLinus Torvalds * nfs_async_unlink(). 27701da177e4SLinus Torvalds */ 2771b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27721da177e4SLinus Torvalds { 277363afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27741da177e4SLinus Torvalds int error; 27751da177e4SLinus Torvalds 2776b18825a7SDavid Howells if (d_is_negative(victim)) 27771da177e4SLinus Torvalds return -ENOENT; 2778b18825a7SDavid Howells BUG_ON(!inode); 27791da177e4SLinus Torvalds 27801da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2781593d1ce8SEric W. Biederman 2782593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2783593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2784593d1ce8SEric W. Biederman return -EOVERFLOW; 2785593d1ce8SEric W. Biederman 27864fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27871da177e4SLinus Torvalds 2788f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27891da177e4SLinus Torvalds if (error) 27901da177e4SLinus Torvalds return error; 27911da177e4SLinus Torvalds if (IS_APPEND(dir)) 27921da177e4SLinus Torvalds return -EPERM; 2793b18825a7SDavid Howells 2794b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27950bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27961da177e4SLinus Torvalds return -EPERM; 27971da177e4SLinus Torvalds if (isdir) { 279844b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27991da177e4SLinus Torvalds return -ENOTDIR; 28001da177e4SLinus Torvalds if (IS_ROOT(victim)) 28011da177e4SLinus Torvalds return -EBUSY; 280244b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28031da177e4SLinus Torvalds return -EISDIR; 28041da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28051da177e4SLinus Torvalds return -ENOENT; 28061da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28071da177e4SLinus Torvalds return -EBUSY; 28081da177e4SLinus Torvalds return 0; 28091da177e4SLinus Torvalds } 28101da177e4SLinus Torvalds 28111da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28121da177e4SLinus Torvalds * dir. 28131da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28141da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28151da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2816036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2817036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2818036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28191da177e4SLinus Torvalds */ 2820a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 28211da177e4SLinus Torvalds { 2822036d5236SEric W. Biederman struct user_namespace *s_user_ns; 282314e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28241da177e4SLinus Torvalds if (child->d_inode) 28251da177e4SLinus Torvalds return -EEXIST; 28261da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28271da177e4SLinus Torvalds return -ENOENT; 2828036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2829036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2830036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2831036d5236SEric W. Biederman return -EOVERFLOW; 2832f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 28331da177e4SLinus Torvalds } 28341da177e4SLinus Torvalds 28351da177e4SLinus Torvalds /* 28361da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28371da177e4SLinus Torvalds */ 28381da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28391da177e4SLinus Torvalds { 28401da177e4SLinus Torvalds struct dentry *p; 28411da177e4SLinus Torvalds 28421da177e4SLinus Torvalds if (p1 == p2) { 28435955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28441da177e4SLinus Torvalds return NULL; 28451da177e4SLinus Torvalds } 28461da177e4SLinus Torvalds 2847fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28481da177e4SLinus Torvalds 2849e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2850e2761a11SOGAWA Hirofumi if (p) { 28515955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28525955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28531da177e4SLinus Torvalds return p; 28541da177e4SLinus Torvalds } 28551da177e4SLinus Torvalds 2856e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2857e2761a11SOGAWA Hirofumi if (p) { 28585955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28595955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28601da177e4SLinus Torvalds return p; 28611da177e4SLinus Torvalds } 28621da177e4SLinus Torvalds 28635955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28645955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28651da177e4SLinus Torvalds return NULL; 28661da177e4SLinus Torvalds } 28674d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28681da177e4SLinus Torvalds 28691da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28701da177e4SLinus Torvalds { 28715955102cSAl Viro inode_unlock(p1->d_inode); 28721da177e4SLinus Torvalds if (p1 != p2) { 28735955102cSAl Viro inode_unlock(p2->d_inode); 2874fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28751da177e4SLinus Torvalds } 28761da177e4SLinus Torvalds } 28774d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28781da177e4SLinus Torvalds 28794acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2880312b63fbSAl Viro bool want_excl) 28811da177e4SLinus Torvalds { 2882a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28831da177e4SLinus Torvalds if (error) 28841da177e4SLinus Torvalds return error; 28851da177e4SLinus Torvalds 2886acfa4380SAl Viro if (!dir->i_op->create) 28871da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28881da177e4SLinus Torvalds mode &= S_IALLUGO; 28891da177e4SLinus Torvalds mode |= S_IFREG; 28901da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28911da177e4SLinus Torvalds if (error) 28921da177e4SLinus Torvalds return error; 2893312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2894a74574aaSStephen Smalley if (!error) 2895f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28961da177e4SLinus Torvalds return error; 28971da177e4SLinus Torvalds } 28984d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28991da177e4SLinus Torvalds 29008e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29018e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29028e6c848eSAl Viro void *arg) 29038e6c848eSAl Viro { 29048e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 29058e6c848eSAl Viro int error = may_create(dir, dentry); 29068e6c848eSAl Viro if (error) 29078e6c848eSAl Viro return error; 29088e6c848eSAl Viro 29098e6c848eSAl Viro mode &= S_IALLUGO; 29108e6c848eSAl Viro mode |= S_IFREG; 29118e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29128e6c848eSAl Viro if (error) 29138e6c848eSAl Viro return error; 29148e6c848eSAl Viro error = f(dentry, mode, arg); 29158e6c848eSAl Viro if (!error) 29168e6c848eSAl Viro fsnotify_create(dir, dentry); 29178e6c848eSAl Viro return error; 29188e6c848eSAl Viro } 29198e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29208e6c848eSAl Viro 2921a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2922a2982cc9SEric W. Biederman { 2923a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2924a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2925a2982cc9SEric W. Biederman } 2926a2982cc9SEric W. Biederman 2927f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 29281da177e4SLinus Torvalds { 29293fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29301da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29311da177e4SLinus Torvalds int error; 29321da177e4SLinus Torvalds 29331da177e4SLinus Torvalds if (!inode) 29341da177e4SLinus Torvalds return -ENOENT; 29351da177e4SLinus Torvalds 2936c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2937c8fe8f30SChristoph Hellwig case S_IFLNK: 29381da177e4SLinus Torvalds return -ELOOP; 2939c8fe8f30SChristoph Hellwig case S_IFDIR: 2940c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 29411da177e4SLinus Torvalds return -EISDIR; 2942c8fe8f30SChristoph Hellwig break; 2943c8fe8f30SChristoph Hellwig case S_IFBLK: 2944c8fe8f30SChristoph Hellwig case S_IFCHR: 2945a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29461da177e4SLinus Torvalds return -EACCES; 2947c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2948c8fe8f30SChristoph Hellwig case S_IFIFO: 2949c8fe8f30SChristoph Hellwig case S_IFSOCK: 29501da177e4SLinus Torvalds flag &= ~O_TRUNC; 2951c8fe8f30SChristoph Hellwig break; 29524a3fd211SDave Hansen } 2953b41572e9SDave Hansen 295462fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2955b41572e9SDave Hansen if (error) 2956b41572e9SDave Hansen return error; 29576146f0d5SMimi Zohar 29581da177e4SLinus Torvalds /* 29591da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29601da177e4SLinus Torvalds */ 29611da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29628737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29637715b521SAl Viro return -EPERM; 29641da177e4SLinus Torvalds if (flag & O_TRUNC) 29657715b521SAl Viro return -EPERM; 29661da177e4SLinus Torvalds } 29671da177e4SLinus Torvalds 29681da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29692e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29707715b521SAl Viro return -EPERM; 29711da177e4SLinus Torvalds 2972f3c7691eSJ. Bruce Fields return 0; 29737715b521SAl Viro } 29747715b521SAl Viro 2975e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29767715b521SAl Viro { 2977f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29787715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29797715b521SAl Viro int error = get_write_access(inode); 29801da177e4SLinus Torvalds if (error) 29817715b521SAl Viro return error; 29821da177e4SLinus Torvalds /* 29831da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29841da177e4SLinus Torvalds */ 2985d7a06983SJeff Layton error = locks_verify_locked(filp); 2986be6d3e56SKentaro Takeda if (!error) 2987ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29881da177e4SLinus Torvalds if (!error) { 29897715b521SAl Viro error = do_truncate(path->dentry, 0, 2990d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2991e1181ee6SJeff Layton filp); 29921da177e4SLinus Torvalds } 29931da177e4SLinus Torvalds put_write_access(inode); 2994acd0c935SMimi Zohar return error; 29951da177e4SLinus Torvalds } 29961da177e4SLinus Torvalds 2997d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2998d57999e1SDave Hansen { 29998a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30008a5e929dSAl Viro flag--; 3001d57999e1SDave Hansen return flag; 3002d57999e1SDave Hansen } 3003d57999e1SDave Hansen 3004d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 3005d18e9008SMiklos Szeredi { 30061328c727SSeth Forshee struct user_namespace *s_user_ns; 3007d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3008d18e9008SMiklos Szeredi if (error) 3009d18e9008SMiklos Szeredi return error; 3010d18e9008SMiklos Szeredi 30111328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 30121328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 30131328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 30141328c727SSeth Forshee return -EOVERFLOW; 30151328c727SSeth Forshee 3016d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 3017d18e9008SMiklos Szeredi if (error) 3018d18e9008SMiklos Szeredi return error; 3019d18e9008SMiklos Szeredi 3020d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3021d18e9008SMiklos Szeredi } 3022d18e9008SMiklos Szeredi 30231acf0af9SDavid Howells /* 30241acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30251acf0af9SDavid Howells * dentry. 30261acf0af9SDavid Howells * 30271acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30281acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30291acf0af9SDavid Howells * 30301acf0af9SDavid Howells * Returns 1 if the file was looked up only or didn't need creating. The 30311acf0af9SDavid Howells * caller will need to perform the open themselves. @path will have been 30321acf0af9SDavid Howells * updated to point to the new dentry. This may be negative. 30331acf0af9SDavid Howells * 30341acf0af9SDavid Howells * Returns an error code otherwise. 30351acf0af9SDavid Howells */ 30362675a4ebSAl Viro static int atomic_open(struct nameidata *nd, struct dentry *dentry, 303730d90494SAl Viro struct path *path, struct file *file, 3038015c3bbcSMiklos Szeredi const struct open_flags *op, 30391643b43fSAl Viro int open_flag, umode_t mode, 304047237687SAl Viro int *opened) 3041d18e9008SMiklos Szeredi { 3042d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3043d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3044d18e9008SMiklos Szeredi int error; 3045d18e9008SMiklos Szeredi 3046384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 3047d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 3048d18e9008SMiklos Szeredi 3049d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3050d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3051d18e9008SMiklos Szeredi 305230d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 305330d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30540fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 30550fb1ea09SAl Viro open_to_namei_flags(open_flag), 30560fb1ea09SAl Viro mode, opened); 30576fbd0714SAl Viro d_lookup_done(dentry); 3058384f26e2SAl Viro if (!error) { 3059d18e9008SMiklos Szeredi /* 3060384f26e2SAl Viro * We didn't have the inode before the open, so check open 3061384f26e2SAl Viro * permission here. 3062d18e9008SMiklos Szeredi */ 3063384f26e2SAl Viro int acc_mode = op->acc_mode; 306473a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 306503da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 306603da633aSAl Viro fsnotify_create(dir, dentry); 306762fb4a15SAl Viro acc_mode = 0; 306803da633aSAl Viro } 30692675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3070384f26e2SAl Viro if (WARN_ON(error > 0)) 3071384f26e2SAl Viro error = -EINVAL; 3072384f26e2SAl Viro } else if (error > 0) { 3073d18e9008SMiklos Szeredi 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; 30872675a4ebSAl Viro return 1; 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 * 3098d58ffd35SMiklos Szeredi * Must be called with i_mutex held on parent. 3099d58ffd35SMiklos Szeredi * 31001acf0af9SDavid Howells * Returns 0 if the file was successfully atomically created (if necessary) and 31011acf0af9SDavid Howells * opened. In this case the file will be returned attached to @file. 31021acf0af9SDavid Howells * 31031acf0af9SDavid Howells * Returns 1 if the file was not completely opened at this time, though lookups 31041acf0af9SDavid Howells * and creations will have been performed and the dentry returned in @path will 31051acf0af9SDavid Howells * be positive upon return if O_CREAT was specified. If O_CREAT wasn't 31061acf0af9SDavid Howells * specified then a negative dentry may be returned. 31071acf0af9SDavid Howells * 31081acf0af9SDavid Howells * An error code is returned otherwise. 31091acf0af9SDavid Howells * 31101acf0af9SDavid Howells * FILE_CREATE will be set in @*opened if the dentry was created and will be 31111acf0af9SDavid Howells * cleared otherwise prior to returning. 3112d58ffd35SMiklos Szeredi */ 31132675a4ebSAl Viro static int lookup_open(struct nameidata *nd, struct path *path, 311430d90494SAl Viro struct file *file, 3115d58ffd35SMiklos Szeredi const struct open_flags *op, 311664894cf8SAl Viro bool got_write, int *opened) 3117d58ffd35SMiklos Szeredi { 3118d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 311954ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31201643b43fSAl Viro int open_flag = op->open_flag; 3121d58ffd35SMiklos Szeredi struct dentry *dentry; 31221643b43fSAl Viro int error, create_error = 0; 31231643b43fSAl Viro umode_t mode = op->mode; 31246fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3125d58ffd35SMiklos Szeredi 3126ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3127ce8644fcSAl Viro return -ENOENT; 3128d58ffd35SMiklos Szeredi 312973a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 31306fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31316fbd0714SAl Viro for (;;) { 31326fbd0714SAl Viro if (!dentry) { 31336fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3134d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 31352675a4ebSAl Viro return PTR_ERR(dentry); 31366fbd0714SAl Viro } 31376fbd0714SAl Viro if (d_in_lookup(dentry)) 31386fbd0714SAl Viro break; 3139d58ffd35SMiklos Szeredi 31406fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31416fbd0714SAl Viro if (likely(error > 0)) 31426fbd0714SAl Viro break; 31436fbd0714SAl Viro if (error) 31446fbd0714SAl Viro goto out_dput; 31456fbd0714SAl Viro d_invalidate(dentry); 31466fbd0714SAl Viro dput(dentry); 31476fbd0714SAl Viro dentry = NULL; 31486fbd0714SAl Viro } 31496fbd0714SAl Viro if (dentry->d_inode) { 3150d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3151d18e9008SMiklos Szeredi goto out_no_open; 31526c51e513SAl Viro } 3153d18e9008SMiklos Szeredi 31541643b43fSAl Viro /* 31551643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31561643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31571643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31581643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31591643b43fSAl Viro * 31601643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31611643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31621643b43fSAl Viro */ 31631643b43fSAl Viro if (open_flag & O_CREAT) { 31641643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31651643b43fSAl Viro mode &= ~current_umask(); 31661643b43fSAl Viro if (unlikely(!got_write)) { 31671643b43fSAl Viro create_error = -EROFS; 31681643b43fSAl Viro open_flag &= ~O_CREAT; 31691643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31701643b43fSAl Viro goto no_open; 31711643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31721643b43fSAl Viro } else { 31731643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31741643b43fSAl Viro if (create_error) { 31751643b43fSAl Viro open_flag &= ~O_CREAT; 31761643b43fSAl Viro if (open_flag & O_EXCL) 31771643b43fSAl Viro goto no_open; 31781643b43fSAl Viro } 31791643b43fSAl Viro } 31801643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31811643b43fSAl Viro unlikely(!got_write)) { 31821643b43fSAl Viro /* 31831643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31841643b43fSAl Viro * back to lookup + open 31851643b43fSAl Viro */ 31861643b43fSAl Viro goto no_open; 3187d18e9008SMiklos Szeredi } 3188d18e9008SMiklos Szeredi 31891643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 31901643b43fSAl Viro error = atomic_open(nd, dentry, path, file, op, open_flag, 31911643b43fSAl Viro mode, opened); 31921643b43fSAl Viro if (unlikely(error == -ENOENT) && create_error) 31931643b43fSAl Viro error = create_error; 31941643b43fSAl Viro return error; 31951643b43fSAl Viro } 319654ef4872SMiklos Szeredi 31971643b43fSAl Viro no_open: 31986fbd0714SAl Viro if (d_in_lookup(dentry)) { 319912fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 320012fa5e24SAl Viro nd->flags); 32016fbd0714SAl Viro d_lookup_done(dentry); 320212fa5e24SAl Viro if (unlikely(res)) { 320312fa5e24SAl Viro if (IS_ERR(res)) { 320412fa5e24SAl Viro error = PTR_ERR(res); 320512fa5e24SAl Viro goto out_dput; 320612fa5e24SAl Viro } 320712fa5e24SAl Viro dput(dentry); 320812fa5e24SAl Viro dentry = res; 320912fa5e24SAl Viro } 321054ef4872SMiklos Szeredi } 321154ef4872SMiklos Szeredi 3212d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32131643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 321473a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3215ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3216ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3217ce8644fcSAl Viro error = -EACCES; 3218d58ffd35SMiklos Szeredi goto out_dput; 321964894cf8SAl Viro } 3220ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 32211643b43fSAl Viro open_flag & O_EXCL); 3222d58ffd35SMiklos Szeredi if (error) 3223d58ffd35SMiklos Szeredi goto out_dput; 3224ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3225d58ffd35SMiklos Szeredi } 32261643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32271643b43fSAl Viro error = create_error; 3228d58ffd35SMiklos Szeredi goto out_dput; 3229d58ffd35SMiklos Szeredi } 3230d18e9008SMiklos Szeredi out_no_open: 3231d58ffd35SMiklos Szeredi path->dentry = dentry; 3232d58ffd35SMiklos Szeredi path->mnt = nd->path.mnt; 32332675a4ebSAl Viro return 1; 3234d58ffd35SMiklos Szeredi 3235d58ffd35SMiklos Szeredi out_dput: 3236d58ffd35SMiklos Szeredi dput(dentry); 32372675a4ebSAl Viro return error; 3238d58ffd35SMiklos Szeredi } 3239d58ffd35SMiklos Szeredi 3240d58ffd35SMiklos Szeredi /* 3241fe2d35ffSAl Viro * Handle the last step of open() 324231e6b01fSNick Piggin */ 3243896475d5SAl Viro static int do_last(struct nameidata *nd, 324430d90494SAl Viro struct file *file, const struct open_flags *op, 324576ae2a5aSAl Viro int *opened) 3246fb1cc555SAl Viro { 3247a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3248ca344a89SAl Viro int open_flag = op->open_flag; 324977d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 325064894cf8SAl Viro bool got_write = false; 3251bcda7652SAl Viro int acc_mode = op->acc_mode; 3252254cf582SAl Viro unsigned seq; 3253a1eb3315SMiklos Szeredi struct inode *inode; 3254896475d5SAl Viro struct path path; 325516c2cd71SAl Viro int error; 3256fb1cc555SAl Viro 3257c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3258c3e380b0SAl Viro nd->flags |= op->intent; 3259c3e380b0SAl Viro 3260bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 3261fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3262deb106c6SAl Viro if (unlikely(error)) 32632675a4ebSAl Viro return error; 3264e83db167SMiklos Szeredi goto finish_open; 32651f36f774SAl Viro } 3266a2c36b45SAl Viro 3267ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3268fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3269fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3270fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 3271254cf582SAl Viro error = lookup_fast(nd, &path, &inode, &seq); 3272e9742b53SAl Viro if (likely(error > 0)) 327371574865SMiklos Szeredi goto finish_lookup; 327471574865SMiklos Szeredi 3275ce57dfc1SAl Viro if (error < 0) 3276deb106c6SAl Viro return error; 3277a1eb3315SMiklos Szeredi 327837d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32796583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3280b6183df7SMiklos Szeredi } else { 3281fe2d35ffSAl Viro /* create side of things */ 3282a3fbbde7SAl Viro /* 3283b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3284b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3285b6183df7SMiklos Szeredi * about to look up 3286a3fbbde7SAl Viro */ 32879f1fafeeSAl Viro error = complete_walk(nd); 3288e8bb73dfSAl Viro if (error) 32892675a4ebSAl Viro return error; 3290fe2d35ffSAl Viro 329176ae2a5aSAl Viro audit_inode(nd->name, dir, LOOKUP_PARENT); 32921f36f774SAl Viro /* trailing slashes? */ 3293deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 3294deb106c6SAl Viro return -EISDIR; 3295b6183df7SMiklos Szeredi } 32961f36f774SAl Viro 32979cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 329864894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 329964894cf8SAl Viro if (!error) 330064894cf8SAl Viro got_write = true; 330164894cf8SAl Viro /* 330264894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 330364894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 330464894cf8SAl Viro * dropping this one anyway. 330564894cf8SAl Viro */ 330664894cf8SAl Viro } 33079cf843e3SAl Viro if (open_flag & O_CREAT) 33085955102cSAl Viro inode_lock(dir->d_inode); 33099cf843e3SAl Viro else 33109cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3311896475d5SAl Viro error = lookup_open(nd, &path, file, op, got_write, opened); 33129cf843e3SAl Viro if (open_flag & O_CREAT) 33135955102cSAl Viro inode_unlock(dir->d_inode); 33149cf843e3SAl Viro else 33159cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3316fb1cc555SAl Viro 33172675a4ebSAl Viro if (error <= 0) { 33182675a4ebSAl Viro if (error) 3319d58ffd35SMiklos Szeredi goto out; 33206c0d46c4SAl Viro 332173a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3322496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 332377d660a8SMiklos Szeredi will_truncate = false; 3324d18e9008SMiklos Szeredi 332576ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3326d18e9008SMiklos Szeredi goto opened; 3327d18e9008SMiklos Szeredi } 3328d18e9008SMiklos Szeredi 332973a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 33309b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3331ca344a89SAl Viro open_flag &= ~O_TRUNC; 333277d660a8SMiklos Szeredi will_truncate = false; 333362fb4a15SAl Viro acc_mode = 0; 3334896475d5SAl Viro path_to_nameidata(&path, nd); 3335e83db167SMiklos Szeredi goto finish_open_created; 3336fb1cc555SAl Viro } 3337fb1cc555SAl Viro 3338fb1cc555SAl Viro /* 3339d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3340d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3341d18e9008SMiklos Szeredi * necessary...) 3342d18e9008SMiklos Szeredi */ 334364894cf8SAl Viro if (got_write) { 3344d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 334564894cf8SAl Viro got_write = false; 3346d18e9008SMiklos Szeredi } 3347d18e9008SMiklos Szeredi 3348e6ec03a2SAl Viro error = follow_managed(&path, nd); 3349e6ec03a2SAl Viro if (unlikely(error < 0)) 3350e6ec03a2SAl Viro return error; 3351e6ec03a2SAl Viro 33526583fe22SAl Viro if (unlikely(d_is_negative(path.dentry))) { 33536583fe22SAl Viro path_to_nameidata(&path, nd); 33546583fe22SAl Viro return -ENOENT; 33556583fe22SAl Viro } 33566583fe22SAl Viro 33576583fe22SAl Viro /* 33586583fe22SAl Viro * create/update audit record if it already exists. 33596583fe22SAl Viro */ 33606583fe22SAl Viro audit_inode(nd->name, path.dentry, 0); 33616583fe22SAl Viro 3362deb106c6SAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 3363deb106c6SAl Viro path_to_nameidata(&path, nd); 3364deb106c6SAl Viro return -EEXIST; 3365deb106c6SAl Viro } 3366fb1cc555SAl Viro 3367254cf582SAl Viro seq = 0; /* out of RCU mode, so the value doesn't matter */ 3368d4565649SAl Viro inode = d_backing_inode(path.dentry); 3369766c4cbfSAl Viro finish_lookup: 33708f64fb1cSAl Viro error = step_into(nd, &path, 0, inode, seq); 3371deb106c6SAl Viro if (unlikely(error)) 3372d63ff28fSAl Viro return error; 3373bc77daa7SAl Viro finish_open: 33748f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3375a3fbbde7SAl Viro error = complete_walk(nd); 3376fac7d191SAl Viro if (error) 33772675a4ebSAl Viro return error; 337876ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3379fb1cc555SAl Viro error = -EISDIR; 338044b1d530SMiklos Szeredi if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry)) 33812675a4ebSAl Viro goto out; 3382af2f5542SMiklos Szeredi error = -ENOTDIR; 338344b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 33842675a4ebSAl Viro goto out; 33854bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 338677d660a8SMiklos Szeredi will_truncate = false; 33876c0d46c4SAl Viro 33880f9d1a10SAl Viro if (will_truncate) { 33890f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33900f9d1a10SAl Viro if (error) 33912675a4ebSAl Viro goto out; 339264894cf8SAl Viro got_write = true; 33930f9d1a10SAl Viro } 3394e83db167SMiklos Szeredi finish_open_created: 3395bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3396ca344a89SAl Viro if (error) 33972675a4ebSAl Viro goto out; 3398aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3399ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3400fac7d191SAl Viro if (error) 3401015c3bbcSMiklos Szeredi goto out; 3402a8277b9bSMiklos Szeredi opened: 34036035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3404fe9ec829SAl Viro if (!error && will_truncate) 34052675a4ebSAl Viro error = handle_truncate(file); 3406ca344a89SAl Viro out: 3407c80567c8SAl Viro if (unlikely(error > 0)) { 3408c80567c8SAl Viro WARN_ON(1); 3409c80567c8SAl Viro error = -EINVAL; 3410c80567c8SAl Viro } 341164894cf8SAl Viro if (got_write) 34120f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 34132675a4ebSAl Viro return error; 3414fb1cc555SAl Viro } 3415fb1cc555SAl Viro 3416af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3417af7bd4dcSAmir Goldstein { 3418af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3419af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3420af7bd4dcSAmir Goldstein struct inode *inode; 3421af7bd4dcSAmir Goldstein int error; 3422af7bd4dcSAmir Goldstein 3423af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3424af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3425af7bd4dcSAmir Goldstein if (error) 3426af7bd4dcSAmir Goldstein goto out_err; 3427af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3428af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3429af7bd4dcSAmir Goldstein goto out_err; 3430af7bd4dcSAmir Goldstein error = -ENOMEM; 3431cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3432af7bd4dcSAmir Goldstein if (unlikely(!child)) 3433af7bd4dcSAmir Goldstein goto out_err; 3434af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3435af7bd4dcSAmir Goldstein if (error) 3436af7bd4dcSAmir Goldstein goto out_err; 3437af7bd4dcSAmir Goldstein error = -ENOENT; 3438af7bd4dcSAmir Goldstein inode = child->d_inode; 3439af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3440af7bd4dcSAmir Goldstein goto out_err; 3441af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3442af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3443af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3444af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3445af7bd4dcSAmir Goldstein } 3446af7bd4dcSAmir Goldstein return child; 3447af7bd4dcSAmir Goldstein 3448af7bd4dcSAmir Goldstein out_err: 3449af7bd4dcSAmir Goldstein dput(child); 3450af7bd4dcSAmir Goldstein return ERR_PTR(error); 3451af7bd4dcSAmir Goldstein } 3452af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3453af7bd4dcSAmir Goldstein 3454c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 345560545d0dSAl Viro const struct open_flags *op, 345660545d0dSAl Viro struct file *file, int *opened) 345760545d0dSAl Viro { 3458625b6d10SAl Viro struct dentry *child; 3459625b6d10SAl Viro struct path path; 3460c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 346160545d0dSAl Viro if (unlikely(error)) 346260545d0dSAl Viro return error; 3463625b6d10SAl Viro error = mnt_want_write(path.mnt); 346460545d0dSAl Viro if (unlikely(error)) 346560545d0dSAl Viro goto out; 3466af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3467af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3468684e73beSHirofumi Nakagawa if (IS_ERR(child)) 346960545d0dSAl Viro goto out2; 3470625b6d10SAl Viro dput(path.dentry); 3471625b6d10SAl Viro path.dentry = child; 3472c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 347369a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 347462fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 347560545d0dSAl Viro if (error) 347660545d0dSAl Viro goto out2; 3477625b6d10SAl Viro file->f_path.mnt = path.mnt; 3478625b6d10SAl Viro error = finish_open(file, child, NULL, opened); 347960545d0dSAl Viro out2: 3480625b6d10SAl Viro mnt_drop_write(path.mnt); 348160545d0dSAl Viro out: 3482625b6d10SAl Viro path_put(&path); 348360545d0dSAl Viro return error; 348460545d0dSAl Viro } 348560545d0dSAl Viro 34866ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34876ac08709SAl Viro { 34886ac08709SAl Viro struct path path; 34896ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34906ac08709SAl Viro if (!error) { 34916ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3492ae2bb293SAl Viro error = vfs_open(&path, file); 34936ac08709SAl Viro path_put(&path); 34946ac08709SAl Viro } 34956ac08709SAl Viro return error; 34966ac08709SAl Viro } 34976ac08709SAl Viro 3498c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3499c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35001da177e4SLinus Torvalds { 3501368ee9baSAl Viro const char *s; 350230d90494SAl Viro struct file *file; 350347237687SAl Viro int opened = 0; 350413aab428SAl Viro int error; 350531e6b01fSNick Piggin 3506ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35071afc99beSAl Viro if (IS_ERR(file)) 35081afc99beSAl Viro return file; 350931e6b01fSNick Piggin 3510bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 3511c8a53ee5SAl Viro error = do_tmpfile(nd, flags, op, file, &opened); 3512f15133dfSAl Viro goto out2; 351360545d0dSAl Viro } 351460545d0dSAl Viro 35156ac08709SAl Viro if (unlikely(file->f_flags & O_PATH)) { 35166ac08709SAl Viro error = do_o_path(nd, flags, file); 35176ac08709SAl Viro goto out2; 35186ac08709SAl Viro } 35196ac08709SAl Viro 3520c8a53ee5SAl Viro s = path_init(nd, flags); 3521368ee9baSAl Viro if (IS_ERR(s)) { 35224d27f326SAl Viro fput(file); 3523368ee9baSAl Viro return ERR_CAST(s); 3524368ee9baSAl Viro } 35253bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 352676ae2a5aSAl Viro (error = do_last(nd, file, op, &opened)) > 0) { 352773d049a4SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 35283bdba28bSAl Viro s = trailing_symlink(nd); 35293bdba28bSAl Viro if (IS_ERR(s)) { 35303bdba28bSAl Viro error = PTR_ERR(s); 35312675a4ebSAl Viro break; 3532806b681cSAl Viro } 35333bdba28bSAl Viro } 3534deb106c6SAl Viro terminate_walk(nd); 3535f15133dfSAl Viro out2: 35367c1c01ecSAl Viro if (likely(!error)) { 3537aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35387c1c01ecSAl Viro return file; 35397c1c01ecSAl Viro WARN_ON(1); 35407c1c01ecSAl Viro error = -EINVAL; 3541015c3bbcSMiklos Szeredi } 35427c1c01ecSAl Viro fput(file); 35432675a4ebSAl Viro if (error == -EOPENSTALE) { 35442675a4ebSAl Viro if (flags & LOOKUP_RCU) 35452675a4ebSAl Viro error = -ECHILD; 35462675a4ebSAl Viro else 35472675a4ebSAl Viro error = -ESTALE; 35482675a4ebSAl Viro } 35497c1c01ecSAl Viro return ERR_PTR(error); 3550de459215SKirill Korotaev } 35511da177e4SLinus Torvalds 3552669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3553f9652e10SAl Viro const struct open_flags *op) 355413aab428SAl Viro { 35559883d185SAl Viro struct nameidata nd; 3556f9652e10SAl Viro int flags = op->lookup_flags; 355713aab428SAl Viro struct file *filp; 355813aab428SAl Viro 35599883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3560c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 356113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3562c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 356313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3564c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35659883d185SAl Viro restore_nameidata(); 356613aab428SAl Viro return filp; 356713aab428SAl Viro } 356813aab428SAl Viro 356973d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3570f9652e10SAl Viro const char *name, const struct open_flags *op) 357173d049a4SAl Viro { 35729883d185SAl Viro struct nameidata nd; 357373d049a4SAl Viro struct file *file; 357451689104SPaul Moore struct filename *filename; 3575f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 357673d049a4SAl Viro 357773d049a4SAl Viro nd.root.mnt = mnt; 357873d049a4SAl Viro nd.root.dentry = dentry; 357973d049a4SAl Viro 3580b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 358173d049a4SAl Viro return ERR_PTR(-ELOOP); 358273d049a4SAl Viro 358351689104SPaul Moore filename = getname_kernel(name); 3584a1c83681SViresh Kumar if (IS_ERR(filename)) 358551689104SPaul Moore return ERR_CAST(filename); 358651689104SPaul Moore 35879883d185SAl Viro set_nameidata(&nd, -1, filename); 3588c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 358973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3590c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 359173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3592c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35939883d185SAl Viro restore_nameidata(); 359451689104SPaul Moore putname(filename); 359573d049a4SAl Viro return file; 359673d049a4SAl Viro } 359773d049a4SAl Viro 3598fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35991ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36001da177e4SLinus Torvalds { 3601c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3602391172c4SAl Viro struct qstr last; 3603391172c4SAl Viro int type; 3604c30dabfeSJan Kara int err2; 36051ac12b4bSJeff Layton int error; 36061ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36071ac12b4bSJeff Layton 36081ac12b4bSJeff Layton /* 36091ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36101ac12b4bSJeff Layton * other flags passed in are ignored! 36111ac12b4bSJeff Layton */ 36121ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36131ac12b4bSJeff Layton 36145c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36155c31b6ceSAl Viro if (IS_ERR(name)) 36165c31b6ceSAl Viro return ERR_CAST(name); 36171da177e4SLinus Torvalds 3618c663e5d8SChristoph Hellwig /* 3619c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3620c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3621c663e5d8SChristoph Hellwig */ 36225c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3623ed75e95dSAl Viro goto out; 3624c663e5d8SChristoph Hellwig 3625c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3626391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3627c663e5d8SChristoph Hellwig /* 3628c663e5d8SChristoph Hellwig * Do the final lookup. 3629c663e5d8SChristoph Hellwig */ 3630391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36315955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3632391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36331da177e4SLinus Torvalds if (IS_ERR(dentry)) 3634a8104a9fSAl Viro goto unlock; 3635c663e5d8SChristoph Hellwig 3636a8104a9fSAl Viro error = -EEXIST; 3637b18825a7SDavid Howells if (d_is_positive(dentry)) 3638a8104a9fSAl Viro goto fail; 3639b18825a7SDavid Howells 3640c663e5d8SChristoph Hellwig /* 3641c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3642c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3643c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3644c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3645c663e5d8SChristoph Hellwig */ 3646391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3647a8104a9fSAl Viro error = -ENOENT; 3648ed75e95dSAl Viro goto fail; 3649e9baf6e5SAl Viro } 3650c30dabfeSJan Kara if (unlikely(err2)) { 3651c30dabfeSJan Kara error = err2; 3652a8104a9fSAl Viro goto fail; 3653c30dabfeSJan Kara } 3654181c37b6SAl Viro putname(name); 3655e9baf6e5SAl Viro return dentry; 36561da177e4SLinus Torvalds fail: 3657a8104a9fSAl Viro dput(dentry); 3658a8104a9fSAl Viro dentry = ERR_PTR(error); 3659a8104a9fSAl Viro unlock: 36605955102cSAl Viro inode_unlock(path->dentry->d_inode); 3661c30dabfeSJan Kara if (!err2) 3662391172c4SAl Viro mnt_drop_write(path->mnt); 3663ed75e95dSAl Viro out: 3664391172c4SAl Viro path_put(path); 3665181c37b6SAl Viro putname(name); 3666ed75e95dSAl Viro return dentry; 3667dae6ad8fSAl Viro } 3668fa14a0b8SAl Viro 3669fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3670fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3671fa14a0b8SAl Viro { 3672181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3673181c37b6SAl Viro path, lookup_flags); 3674fa14a0b8SAl Viro } 3675dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3676dae6ad8fSAl Viro 3677921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3678921a1650SAl Viro { 3679921a1650SAl Viro dput(dentry); 36805955102cSAl Viro inode_unlock(path->dentry->d_inode); 3681a8104a9fSAl Viro mnt_drop_write(path->mnt); 3682921a1650SAl Viro path_put(path); 3683921a1650SAl Viro } 3684921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3685921a1650SAl Viro 3686520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36871ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3688dae6ad8fSAl Viro { 3689181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3690dae6ad8fSAl Viro } 3691dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3692dae6ad8fSAl Viro 36931a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 36941da177e4SLinus Torvalds { 3695a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36961da177e4SLinus Torvalds 36971da177e4SLinus Torvalds if (error) 36981da177e4SLinus Torvalds return error; 36991da177e4SLinus Torvalds 370055956b59SEric W. Biederman if ((S_ISCHR(mode) || S_ISBLK(mode)) && 370155956b59SEric W. Biederman !ns_capable(dentry->d_sb->s_user_ns, CAP_MKNOD)) 37021da177e4SLinus Torvalds return -EPERM; 37031da177e4SLinus Torvalds 3704acfa4380SAl Viro if (!dir->i_op->mknod) 37051da177e4SLinus Torvalds return -EPERM; 37061da177e4SLinus Torvalds 370708ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 370808ce5f16SSerge E. Hallyn if (error) 370908ce5f16SSerge E. Hallyn return error; 371008ce5f16SSerge E. Hallyn 37111da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37121da177e4SLinus Torvalds if (error) 37131da177e4SLinus Torvalds return error; 37141da177e4SLinus Torvalds 37151da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3716a74574aaSStephen Smalley if (!error) 3717f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37181da177e4SLinus Torvalds return error; 37191da177e4SLinus Torvalds } 37204d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37211da177e4SLinus Torvalds 3722f69aac00SAl Viro static int may_mknod(umode_t mode) 3723463c3197SDave Hansen { 3724463c3197SDave Hansen switch (mode & S_IFMT) { 3725463c3197SDave Hansen case S_IFREG: 3726463c3197SDave Hansen case S_IFCHR: 3727463c3197SDave Hansen case S_IFBLK: 3728463c3197SDave Hansen case S_IFIFO: 3729463c3197SDave Hansen case S_IFSOCK: 3730463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3731463c3197SDave Hansen return 0; 3732463c3197SDave Hansen case S_IFDIR: 3733463c3197SDave Hansen return -EPERM; 3734463c3197SDave Hansen default: 3735463c3197SDave Hansen return -EINVAL; 3736463c3197SDave Hansen } 3737463c3197SDave Hansen } 3738463c3197SDave Hansen 373987c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 374087c4e192SDominik Brodowski unsigned int dev) 37411da177e4SLinus Torvalds { 37421da177e4SLinus Torvalds struct dentry *dentry; 3743dae6ad8fSAl Viro struct path path; 3744dae6ad8fSAl Viro int error; 3745972567f1SJeff Layton unsigned int lookup_flags = 0; 37461da177e4SLinus Torvalds 37478e4bfca1SAl Viro error = may_mknod(mode); 37488e4bfca1SAl Viro if (error) 37498e4bfca1SAl Viro return error; 3750972567f1SJeff Layton retry: 3751972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3752dae6ad8fSAl Viro if (IS_ERR(dentry)) 3753dae6ad8fSAl Viro return PTR_ERR(dentry); 37542ad94ae6SAl Viro 3755dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3756ce3b0f8dSAl Viro mode &= ~current_umask(); 3757dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3758be6d3e56SKentaro Takeda if (error) 3759a8104a9fSAl Viro goto out; 37601da177e4SLinus Torvalds switch (mode & S_IFMT) { 37611da177e4SLinus Torvalds case 0: case S_IFREG: 3762312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 376305d1a717SMimi Zohar if (!error) 376405d1a717SMimi Zohar ima_post_path_mknod(dentry); 37651da177e4SLinus Torvalds break; 37661da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3767dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 37681da177e4SLinus Torvalds new_decode_dev(dev)); 37691da177e4SLinus Torvalds break; 37701da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3771dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 37721da177e4SLinus Torvalds break; 37731da177e4SLinus Torvalds } 3774a8104a9fSAl Viro out: 3775921a1650SAl Viro done_path_create(&path, dentry); 3776972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3777972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3778972567f1SJeff Layton goto retry; 3779972567f1SJeff Layton } 37801da177e4SLinus Torvalds return error; 37811da177e4SLinus Torvalds } 37821da177e4SLinus Torvalds 378387c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 378487c4e192SDominik Brodowski unsigned int, dev) 378587c4e192SDominik Brodowski { 378687c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 378787c4e192SDominik Brodowski } 378887c4e192SDominik Brodowski 37898208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37905590ff0dSUlrich Drepper { 379187c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 37925590ff0dSUlrich Drepper } 37935590ff0dSUlrich Drepper 379418bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37951da177e4SLinus Torvalds { 3796a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37978de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37981da177e4SLinus Torvalds 37991da177e4SLinus Torvalds if (error) 38001da177e4SLinus Torvalds return error; 38011da177e4SLinus Torvalds 3802acfa4380SAl Viro if (!dir->i_op->mkdir) 38031da177e4SLinus Torvalds return -EPERM; 38041da177e4SLinus Torvalds 38051da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38061da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38071da177e4SLinus Torvalds if (error) 38081da177e4SLinus Torvalds return error; 38091da177e4SLinus Torvalds 38108de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38118de52778SAl Viro return -EMLINK; 38128de52778SAl Viro 38131da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3814a74574aaSStephen Smalley if (!error) 3815f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38161da177e4SLinus Torvalds return error; 38171da177e4SLinus Torvalds } 38184d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38191da177e4SLinus Torvalds 38200101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38211da177e4SLinus Torvalds { 38226902d925SDave Hansen struct dentry *dentry; 3823dae6ad8fSAl Viro struct path path; 3824dae6ad8fSAl Viro int error; 3825b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38261da177e4SLinus Torvalds 3827b76d8b82SJeff Layton retry: 3828b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38296902d925SDave Hansen if (IS_ERR(dentry)) 3830dae6ad8fSAl Viro return PTR_ERR(dentry); 38316902d925SDave Hansen 3832dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3833ce3b0f8dSAl Viro mode &= ~current_umask(); 3834dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3835a8104a9fSAl Viro if (!error) 3836dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3837921a1650SAl Viro done_path_create(&path, dentry); 3838b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3839b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3840b76d8b82SJeff Layton goto retry; 3841b76d8b82SJeff Layton } 38421da177e4SLinus Torvalds return error; 38431da177e4SLinus Torvalds } 38441da177e4SLinus Torvalds 38450101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38460101db7aSDominik Brodowski { 38470101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38480101db7aSDominik Brodowski } 38490101db7aSDominik Brodowski 3850a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38515590ff0dSUlrich Drepper { 38520101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38535590ff0dSUlrich Drepper } 38545590ff0dSUlrich Drepper 38551da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 38561da177e4SLinus Torvalds { 38571da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 38581da177e4SLinus Torvalds 38591da177e4SLinus Torvalds if (error) 38601da177e4SLinus Torvalds return error; 38611da177e4SLinus Torvalds 3862acfa4380SAl Viro if (!dir->i_op->rmdir) 38631da177e4SLinus Torvalds return -EPERM; 38641da177e4SLinus Torvalds 38651d2ef590SAl Viro dget(dentry); 38665955102cSAl Viro inode_lock(dentry->d_inode); 3867912dbc15SSage Weil 38681da177e4SLinus Torvalds error = -EBUSY; 38697af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3870912dbc15SSage Weil goto out; 3871912dbc15SSage Weil 38721da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3873912dbc15SSage Weil if (error) 3874912dbc15SSage Weil goto out; 3875912dbc15SSage Weil 38761da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3877912dbc15SSage Weil if (error) 3878912dbc15SSage Weil goto out; 3879912dbc15SSage Weil 38808767712fSAl Viro shrink_dcache_parent(dentry); 38811da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3882d83c49f3SAl Viro dont_mount(dentry); 38838ed936b5SEric W. Biederman detach_mounts(dentry); 38841da177e4SLinus Torvalds 3885912dbc15SSage Weil out: 38865955102cSAl Viro inode_unlock(dentry->d_inode); 38871d2ef590SAl Viro dput(dentry); 3888912dbc15SSage Weil if (!error) 3889912dbc15SSage Weil d_delete(dentry); 38901da177e4SLinus Torvalds return error; 38911da177e4SLinus Torvalds } 38924d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38931da177e4SLinus Torvalds 3894f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 38951da177e4SLinus Torvalds { 38961da177e4SLinus Torvalds int error = 0; 389791a27b2aSJeff Layton struct filename *name; 38981da177e4SLinus Torvalds struct dentry *dentry; 3899f5beed75SAl Viro struct path path; 3900f5beed75SAl Viro struct qstr last; 3901f5beed75SAl Viro int type; 3902c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3903c6ee9206SJeff Layton retry: 3904c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3905c1d4dd27SAl Viro &path, &last, &type); 390691a27b2aSJeff Layton if (IS_ERR(name)) 390791a27b2aSJeff Layton return PTR_ERR(name); 39081da177e4SLinus Torvalds 3909f5beed75SAl Viro switch (type) { 39101da177e4SLinus Torvalds case LAST_DOTDOT: 39111da177e4SLinus Torvalds error = -ENOTEMPTY; 39121da177e4SLinus Torvalds goto exit1; 39131da177e4SLinus Torvalds case LAST_DOT: 39141da177e4SLinus Torvalds error = -EINVAL; 39151da177e4SLinus Torvalds goto exit1; 39161da177e4SLinus Torvalds case LAST_ROOT: 39171da177e4SLinus Torvalds error = -EBUSY; 39181da177e4SLinus Torvalds goto exit1; 39191da177e4SLinus Torvalds } 39200612d9fbSOGAWA Hirofumi 3921f5beed75SAl Viro error = mnt_want_write(path.mnt); 3922c30dabfeSJan Kara if (error) 3923c30dabfeSJan Kara goto exit1; 39240612d9fbSOGAWA Hirofumi 39255955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3926f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39271da177e4SLinus Torvalds error = PTR_ERR(dentry); 39286902d925SDave Hansen if (IS_ERR(dentry)) 39296902d925SDave Hansen goto exit2; 3930e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3931e6bc45d6STheodore Ts'o error = -ENOENT; 3932e6bc45d6STheodore Ts'o goto exit3; 3933e6bc45d6STheodore Ts'o } 3934f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3935be6d3e56SKentaro Takeda if (error) 3936c30dabfeSJan Kara goto exit3; 3937f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 39380622753bSDave Hansen exit3: 39391da177e4SLinus Torvalds dput(dentry); 39406902d925SDave Hansen exit2: 39415955102cSAl Viro inode_unlock(path.dentry->d_inode); 3942f5beed75SAl Viro mnt_drop_write(path.mnt); 39431da177e4SLinus Torvalds exit1: 3944f5beed75SAl Viro path_put(&path); 39451da177e4SLinus Torvalds putname(name); 3946c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3947c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3948c6ee9206SJeff Layton goto retry; 3949c6ee9206SJeff Layton } 39501da177e4SLinus Torvalds return error; 39511da177e4SLinus Torvalds } 39521da177e4SLinus Torvalds 39533cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39545590ff0dSUlrich Drepper { 39555590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 39565590ff0dSUlrich Drepper } 39575590ff0dSUlrich Drepper 3958b21996e3SJ. Bruce Fields /** 3959b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3960b21996e3SJ. Bruce Fields * @dir: parent directory 3961b21996e3SJ. Bruce Fields * @dentry: victim 3962b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3963b21996e3SJ. Bruce Fields * 3964b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3965b21996e3SJ. Bruce Fields * 3966b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3967b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3968b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3969b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3970b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3971b21996e3SJ. Bruce Fields * 3972b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3973b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3974b21996e3SJ. Bruce Fields * to be NFS exported. 3975b21996e3SJ. Bruce Fields */ 3976b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39771da177e4SLinus Torvalds { 39789accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39791da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39801da177e4SLinus Torvalds 39811da177e4SLinus Torvalds if (error) 39821da177e4SLinus Torvalds return error; 39831da177e4SLinus Torvalds 3984acfa4380SAl Viro if (!dir->i_op->unlink) 39851da177e4SLinus Torvalds return -EPERM; 39861da177e4SLinus Torvalds 39875955102cSAl Viro inode_lock(target); 39888ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39891da177e4SLinus Torvalds error = -EBUSY; 39901da177e4SLinus Torvalds else { 39911da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3992bec1052eSAl Viro if (!error) { 39935a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39945a14696cSJ. Bruce Fields if (error) 3995b21996e3SJ. Bruce Fields goto out; 39961da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39978ed936b5SEric W. Biederman if (!error) { 3998d83c49f3SAl Viro dont_mount(dentry); 39998ed936b5SEric W. Biederman detach_mounts(dentry); 40008ed936b5SEric W. Biederman } 4001bec1052eSAl Viro } 40021da177e4SLinus Torvalds } 4003b21996e3SJ. Bruce Fields out: 40045955102cSAl Viro inode_unlock(target); 40051da177e4SLinus Torvalds 40061da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40071da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40089accbb97SJ. Bruce Fields fsnotify_link_count(target); 40091da177e4SLinus Torvalds d_delete(dentry); 40101da177e4SLinus Torvalds } 40110eeca283SRobert Love 40121da177e4SLinus Torvalds return error; 40131da177e4SLinus Torvalds } 40144d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40151da177e4SLinus Torvalds 40161da177e4SLinus Torvalds /* 40171da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40181b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40191da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40201da177e4SLinus Torvalds * while waiting on the I/O. 40211da177e4SLinus Torvalds */ 4022da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40231da177e4SLinus Torvalds { 40242ad94ae6SAl Viro int error; 40251da177e4SLinus Torvalds struct dentry *dentry; 4026f5beed75SAl Viro struct path path; 4027f5beed75SAl Viro struct qstr last; 4028f5beed75SAl Viro int type; 40291da177e4SLinus Torvalds struct inode *inode = NULL; 4030b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40315d18f813SJeff Layton unsigned int lookup_flags = 0; 40325d18f813SJeff Layton retry: 4033da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 403491a27b2aSJeff Layton if (IS_ERR(name)) 403591a27b2aSJeff Layton return PTR_ERR(name); 40362ad94ae6SAl Viro 40371da177e4SLinus Torvalds error = -EISDIR; 4038f5beed75SAl Viro if (type != LAST_NORM) 40391da177e4SLinus Torvalds goto exit1; 40400612d9fbSOGAWA Hirofumi 4041f5beed75SAl Viro error = mnt_want_write(path.mnt); 4042c30dabfeSJan Kara if (error) 4043c30dabfeSJan Kara goto exit1; 4044b21996e3SJ. Bruce Fields retry_deleg: 40455955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4046f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40471da177e4SLinus Torvalds error = PTR_ERR(dentry); 40481da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40491da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4050f5beed75SAl Viro if (last.name[last.len]) 405150338b88STörök Edwin goto slashes; 40521da177e4SLinus Torvalds inode = dentry->d_inode; 4053b18825a7SDavid Howells if (d_is_negative(dentry)) 4054e6bc45d6STheodore Ts'o goto slashes; 40557de9c6eeSAl Viro ihold(inode); 4056f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4057be6d3e56SKentaro Takeda if (error) 4058c30dabfeSJan Kara goto exit2; 4059f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 40601da177e4SLinus Torvalds exit2: 40611da177e4SLinus Torvalds dput(dentry); 40621da177e4SLinus Torvalds } 40635955102cSAl Viro inode_unlock(path.dentry->d_inode); 40641da177e4SLinus Torvalds if (inode) 40651da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4066b21996e3SJ. Bruce Fields inode = NULL; 4067b21996e3SJ. Bruce Fields if (delegated_inode) { 40685a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4069b21996e3SJ. Bruce Fields if (!error) 4070b21996e3SJ. Bruce Fields goto retry_deleg; 4071b21996e3SJ. Bruce Fields } 4072f5beed75SAl Viro mnt_drop_write(path.mnt); 40731da177e4SLinus Torvalds exit1: 4074f5beed75SAl Viro path_put(&path); 40755d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40765d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40775d18f813SJeff Layton inode = NULL; 40785d18f813SJeff Layton goto retry; 40795d18f813SJeff Layton } 4080da2f1362SChristoph Hellwig putname(name); 40811da177e4SLinus Torvalds return error; 40821da177e4SLinus Torvalds 40831da177e4SLinus Torvalds slashes: 4084b18825a7SDavid Howells if (d_is_negative(dentry)) 4085b18825a7SDavid Howells error = -ENOENT; 408644b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4087b18825a7SDavid Howells error = -EISDIR; 4088b18825a7SDavid Howells else 4089b18825a7SDavid Howells error = -ENOTDIR; 40901da177e4SLinus Torvalds goto exit2; 40911da177e4SLinus Torvalds } 40921da177e4SLinus Torvalds 40932e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40945590ff0dSUlrich Drepper { 40955590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40965590ff0dSUlrich Drepper return -EINVAL; 40975590ff0dSUlrich Drepper 40985590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 40995590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 41005590ff0dSUlrich Drepper 4101da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 41025590ff0dSUlrich Drepper } 41035590ff0dSUlrich Drepper 41043480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41055590ff0dSUlrich Drepper { 4106da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 41075590ff0dSUlrich Drepper } 41085590ff0dSUlrich Drepper 4109db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 41101da177e4SLinus Torvalds { 4111a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 41121da177e4SLinus Torvalds 41131da177e4SLinus Torvalds if (error) 41141da177e4SLinus Torvalds return error; 41151da177e4SLinus Torvalds 4116acfa4380SAl Viro if (!dir->i_op->symlink) 41171da177e4SLinus Torvalds return -EPERM; 41181da177e4SLinus Torvalds 41191da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41201da177e4SLinus Torvalds if (error) 41211da177e4SLinus Torvalds return error; 41221da177e4SLinus Torvalds 41231da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4124a74574aaSStephen Smalley if (!error) 4125f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41261da177e4SLinus Torvalds return error; 41271da177e4SLinus Torvalds } 41284d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41291da177e4SLinus Torvalds 4130b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 4131b724e846SDominik Brodowski const char __user *newname) 41321da177e4SLinus Torvalds { 41332ad94ae6SAl Viro int error; 413491a27b2aSJeff Layton struct filename *from; 41356902d925SDave Hansen struct dentry *dentry; 4136dae6ad8fSAl Viro struct path path; 4137f46d3567SJeff Layton unsigned int lookup_flags = 0; 41381da177e4SLinus Torvalds 41391da177e4SLinus Torvalds from = getname(oldname); 41401da177e4SLinus Torvalds if (IS_ERR(from)) 41411da177e4SLinus Torvalds return PTR_ERR(from); 4142f46d3567SJeff Layton retry: 4143f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41441da177e4SLinus Torvalds error = PTR_ERR(dentry); 41456902d925SDave Hansen if (IS_ERR(dentry)) 4146dae6ad8fSAl Viro goto out_putname; 41476902d925SDave Hansen 414891a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4149a8104a9fSAl Viro if (!error) 415091a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4151921a1650SAl Viro done_path_create(&path, dentry); 4152f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4153f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4154f46d3567SJeff Layton goto retry; 4155f46d3567SJeff Layton } 41566902d925SDave Hansen out_putname: 41571da177e4SLinus Torvalds putname(from); 41581da177e4SLinus Torvalds return error; 41591da177e4SLinus Torvalds } 41601da177e4SLinus Torvalds 4161b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4162b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4163b724e846SDominik Brodowski { 4164b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4165b724e846SDominik Brodowski } 4166b724e846SDominik Brodowski 41673480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 41685590ff0dSUlrich Drepper { 4169b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 41705590ff0dSUlrich Drepper } 41715590ff0dSUlrich Drepper 4172146a8595SJ. Bruce Fields /** 4173146a8595SJ. Bruce Fields * vfs_link - create a new link 4174146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4175146a8595SJ. Bruce Fields * @dir: new parent 4176146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4177146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4178146a8595SJ. Bruce Fields * 4179146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4180146a8595SJ. Bruce Fields * 4181146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4182146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4183146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4184146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4185146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4186146a8595SJ. Bruce Fields * 4187146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4188146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4189146a8595SJ. Bruce Fields * to be NFS exported. 4190146a8595SJ. Bruce Fields */ 4191146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41921da177e4SLinus Torvalds { 41931da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41948de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41951da177e4SLinus Torvalds int error; 41961da177e4SLinus Torvalds 41971da177e4SLinus Torvalds if (!inode) 41981da177e4SLinus Torvalds return -ENOENT; 41991da177e4SLinus Torvalds 4200a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 42011da177e4SLinus Torvalds if (error) 42021da177e4SLinus Torvalds return error; 42031da177e4SLinus Torvalds 42041da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 42051da177e4SLinus Torvalds return -EXDEV; 42061da177e4SLinus Torvalds 42071da177e4SLinus Torvalds /* 42081da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 42091da177e4SLinus Torvalds */ 42101da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 42111da177e4SLinus Torvalds return -EPERM; 42120bd23d09SEric W. Biederman /* 42130bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 42140bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 42150bd23d09SEric W. Biederman * the vfs. 42160bd23d09SEric W. Biederman */ 42170bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 42180bd23d09SEric W. Biederman return -EPERM; 4219acfa4380SAl Viro if (!dir->i_op->link) 42201da177e4SLinus Torvalds return -EPERM; 42217e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42221da177e4SLinus Torvalds return -EPERM; 42231da177e4SLinus Torvalds 42241da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42251da177e4SLinus Torvalds if (error) 42261da177e4SLinus Torvalds return error; 42271da177e4SLinus Torvalds 42285955102cSAl Viro inode_lock(inode); 4229aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4230f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4231aae8a97dSAneesh Kumar K.V error = -ENOENT; 42328de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 42338de52778SAl Viro error = -EMLINK; 4234146a8595SJ. Bruce Fields else { 4235146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4236146a8595SJ. Bruce Fields if (!error) 42371da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4238146a8595SJ. Bruce Fields } 4239f4e0c30cSAl Viro 4240f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4241f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4242f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4243f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4244f4e0c30cSAl Viro } 42455955102cSAl Viro inode_unlock(inode); 4246e31e14ecSStephen Smalley if (!error) 42477e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 42481da177e4SLinus Torvalds return error; 42491da177e4SLinus Torvalds } 42504d359507SAl Viro EXPORT_SYMBOL(vfs_link); 42511da177e4SLinus Torvalds 42521da177e4SLinus Torvalds /* 42531da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 42541da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 42551da177e4SLinus Torvalds * newname. --KAB 42561da177e4SLinus Torvalds * 42571da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 42581da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 42591da177e4SLinus Torvalds * and other special files. --ADM 42601da177e4SLinus Torvalds */ 426146ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 426246ea89ebSDominik Brodowski const char __user *newname, int flags) 42631da177e4SLinus Torvalds { 42641da177e4SLinus Torvalds struct dentry *new_dentry; 4265dae6ad8fSAl Viro struct path old_path, new_path; 4266146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 426711a7b371SAneesh Kumar K.V int how = 0; 42681da177e4SLinus Torvalds int error; 42691da177e4SLinus Torvalds 427011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4271c04030e1SUlrich Drepper return -EINVAL; 427211a7b371SAneesh Kumar K.V /* 4273f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4274f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4275f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 427611a7b371SAneesh Kumar K.V */ 4277f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4278f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4279f0cc6ffbSLinus Torvalds return -ENOENT; 428011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4281f0cc6ffbSLinus Torvalds } 4282c04030e1SUlrich Drepper 428311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 428411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4285442e31caSJeff Layton retry: 428611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42871da177e4SLinus Torvalds if (error) 42882ad94ae6SAl Viro return error; 42892ad94ae6SAl Viro 4290442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4291442e31caSJeff Layton (how & LOOKUP_REVAL)); 42921da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42936902d925SDave Hansen if (IS_ERR(new_dentry)) 4294dae6ad8fSAl Viro goto out; 4295dae6ad8fSAl Viro 4296dae6ad8fSAl Viro error = -EXDEV; 4297dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4298dae6ad8fSAl Viro goto out_dput; 4299800179c9SKees Cook error = may_linkat(&old_path); 4300800179c9SKees Cook if (unlikely(error)) 4301800179c9SKees Cook goto out_dput; 4302dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4303be6d3e56SKentaro Takeda if (error) 4304a8104a9fSAl Viro goto out_dput; 4305146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 430675c3f29dSDave Hansen out_dput: 4307921a1650SAl Viro done_path_create(&new_path, new_dentry); 4308146a8595SJ. Bruce Fields if (delegated_inode) { 4309146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4310d22e6338SOleg Drokin if (!error) { 4311d22e6338SOleg Drokin path_put(&old_path); 4312146a8595SJ. Bruce Fields goto retry; 4313146a8595SJ. Bruce Fields } 4314d22e6338SOleg Drokin } 4315442e31caSJeff Layton if (retry_estale(error, how)) { 4316d22e6338SOleg Drokin path_put(&old_path); 4317442e31caSJeff Layton how |= LOOKUP_REVAL; 4318442e31caSJeff Layton goto retry; 4319442e31caSJeff Layton } 43201da177e4SLinus Torvalds out: 43212d8f3038SAl Viro path_put(&old_path); 43221da177e4SLinus Torvalds 43231da177e4SLinus Torvalds return error; 43241da177e4SLinus Torvalds } 43251da177e4SLinus Torvalds 432646ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 432746ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 432846ea89ebSDominik Brodowski { 432946ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 433046ea89ebSDominik Brodowski } 433146ea89ebSDominik Brodowski 43323480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 43335590ff0dSUlrich Drepper { 433446ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 43355590ff0dSUlrich Drepper } 43365590ff0dSUlrich Drepper 4337bc27027aSMiklos Szeredi /** 4338bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4339bc27027aSMiklos Szeredi * @old_dir: parent of source 4340bc27027aSMiklos Szeredi * @old_dentry: source 4341bc27027aSMiklos Szeredi * @new_dir: parent of destination 4342bc27027aSMiklos Szeredi * @new_dentry: destination 4343bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4344520c8b16SMiklos Szeredi * @flags: rename flags 4345bc27027aSMiklos Szeredi * 4346bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4347bc27027aSMiklos Szeredi * 4348bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4349bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4350bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4351bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4352bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4353bc27027aSMiklos Szeredi * so. 4354bc27027aSMiklos Szeredi * 4355bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4356bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4357bc27027aSMiklos Szeredi * to be NFS exported. 4358bc27027aSMiklos Szeredi * 43591da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 43601da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 43611da177e4SLinus Torvalds * Problems: 43620117d427SMauro Carvalho Chehab * 4363d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 43641da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 43651da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4366a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 43671da177e4SLinus Torvalds * story. 43686cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 43696cedba89SJ. Bruce Fields * and source (if it is not a directory). 43701b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 43711da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 43721da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4373a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43741da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43751da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43761da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4377a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43781da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43791da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 43801da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4381e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 43821b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 43831da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4384c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 43851da177e4SLinus Torvalds * locking]. 43861da177e4SLinus Torvalds */ 43871da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 43888e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4389520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 43901da177e4SLinus Torvalds { 43911da177e4SLinus Torvalds int error; 4392bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4393bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4394bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4395da1ce067SMiklos Szeredi bool new_is_dir = false; 4396da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 439749d31c2fSAl Viro struct name_snapshot old_name; 43981da177e4SLinus Torvalds 43998d3e2936SMiklos Szeredi if (source == target) 44001da177e4SLinus Torvalds return 0; 44011da177e4SLinus Torvalds 44021da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 44031da177e4SLinus Torvalds if (error) 44041da177e4SLinus Torvalds return error; 44051da177e4SLinus Torvalds 4406da1ce067SMiklos Szeredi if (!target) { 4407a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4408da1ce067SMiklos Szeredi } else { 4409da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4410da1ce067SMiklos Szeredi 4411da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44121da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4413da1ce067SMiklos Szeredi else 4414da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4415da1ce067SMiklos Szeredi } 44161da177e4SLinus Torvalds if (error) 44171da177e4SLinus Torvalds return error; 44181da177e4SLinus Torvalds 44192773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 44201da177e4SLinus Torvalds return -EPERM; 44211da177e4SLinus Torvalds 4422bc27027aSMiklos Szeredi /* 4423bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4424bc27027aSMiklos Szeredi * we'll need to flip '..'. 4425bc27027aSMiklos Szeredi */ 4426da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4427da1ce067SMiklos Szeredi if (is_dir) { 4428bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4429bc27027aSMiklos Szeredi if (error) 4430bc27027aSMiklos Szeredi return error; 4431bc27027aSMiklos Szeredi } 4432da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4433da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4434da1ce067SMiklos Szeredi if (error) 4435da1ce067SMiklos Szeredi return error; 4436da1ce067SMiklos Szeredi } 4437da1ce067SMiklos Szeredi } 44380eeca283SRobert Love 44390b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 44400b3974ebSMiklos Szeredi flags); 4441bc27027aSMiklos Szeredi if (error) 4442bc27027aSMiklos Szeredi return error; 4443bc27027aSMiklos Szeredi 444449d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4445bc27027aSMiklos Szeredi dget(new_dentry); 4446da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4447bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4448bc27027aSMiklos Szeredi else if (target) 44495955102cSAl Viro inode_lock(target); 4450bc27027aSMiklos Szeredi 4451bc27027aSMiklos Szeredi error = -EBUSY; 44527af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4453bc27027aSMiklos Szeredi goto out; 4454bc27027aSMiklos Szeredi 4455da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4456bc27027aSMiklos Szeredi error = -EMLINK; 4457da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4458bc27027aSMiklos Szeredi goto out; 4459da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4460da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4461da1ce067SMiklos Szeredi goto out; 4462da1ce067SMiklos Szeredi } 4463da1ce067SMiklos Szeredi if (!is_dir) { 4464bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4465bc27027aSMiklos Szeredi if (error) 4466bc27027aSMiklos Szeredi goto out; 4467da1ce067SMiklos Szeredi } 4468da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4469bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4470bc27027aSMiklos Szeredi if (error) 4471bc27027aSMiklos Szeredi goto out; 4472bc27027aSMiklos Szeredi } 4473520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4474520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4475bc27027aSMiklos Szeredi if (error) 4476bc27027aSMiklos Szeredi goto out; 4477bc27027aSMiklos Szeredi 4478da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44798767712fSAl Viro if (is_dir) { 44808767712fSAl Viro shrink_dcache_parent(new_dentry); 4481bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 44828767712fSAl Viro } 4483bc27027aSMiklos Szeredi dont_mount(new_dentry); 44848ed936b5SEric W. Biederman detach_mounts(new_dentry); 4485bc27027aSMiklos Szeredi } 4486da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4487da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4488bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4489da1ce067SMiklos Szeredi else 4490da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4491da1ce067SMiklos Szeredi } 4492bc27027aSMiklos Szeredi out: 4493da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4494bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4495bc27027aSMiklos Szeredi else if (target) 44965955102cSAl Viro inode_unlock(target); 4497bc27027aSMiklos Szeredi dput(new_dentry); 4498da1ce067SMiklos Szeredi if (!error) { 449949d31c2fSAl Viro fsnotify_move(old_dir, new_dir, old_name.name, is_dir, 4500da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4501da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4502da1ce067SMiklos Szeredi fsnotify_move(new_dir, old_dir, old_dentry->d_name.name, 4503da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4504da1ce067SMiklos Szeredi } 4505da1ce067SMiklos Szeredi } 450649d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 45070eeca283SRobert Love 45081da177e4SLinus Torvalds return error; 45091da177e4SLinus Torvalds } 45104d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 45111da177e4SLinus Torvalds 4512ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4513ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 45141da177e4SLinus Torvalds { 45151da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 45161da177e4SLinus Torvalds struct dentry *trap; 4517f5beed75SAl Viro struct path old_path, new_path; 4518f5beed75SAl Viro struct qstr old_last, new_last; 4519f5beed75SAl Viro int old_type, new_type; 45208e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 452191a27b2aSJeff Layton struct filename *from; 452291a27b2aSJeff Layton struct filename *to; 4523f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4524c6a94284SJeff Layton bool should_retry = false; 45252ad94ae6SAl Viro int error; 4526520c8b16SMiklos Szeredi 45270d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4528da1ce067SMiklos Szeredi return -EINVAL; 4529da1ce067SMiklos Szeredi 45300d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 45310d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4532520c8b16SMiklos Szeredi return -EINVAL; 4533520c8b16SMiklos Szeredi 45340d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 45350d7a8555SMiklos Szeredi return -EPERM; 45360d7a8555SMiklos Szeredi 4537f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4538f5beed75SAl Viro target_flags = 0; 4539f5beed75SAl Viro 4540c6a94284SJeff Layton retry: 4541c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4542c1d4dd27SAl Viro &old_path, &old_last, &old_type); 454391a27b2aSJeff Layton if (IS_ERR(from)) { 454491a27b2aSJeff Layton error = PTR_ERR(from); 45451da177e4SLinus Torvalds goto exit; 454691a27b2aSJeff Layton } 45471da177e4SLinus Torvalds 4548c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4549c1d4dd27SAl Viro &new_path, &new_last, &new_type); 455091a27b2aSJeff Layton if (IS_ERR(to)) { 455191a27b2aSJeff Layton error = PTR_ERR(to); 45521da177e4SLinus Torvalds goto exit1; 455391a27b2aSJeff Layton } 45541da177e4SLinus Torvalds 45551da177e4SLinus Torvalds error = -EXDEV; 4556f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 45571da177e4SLinus Torvalds goto exit2; 45581da177e4SLinus Torvalds 45591da177e4SLinus Torvalds error = -EBUSY; 4560f5beed75SAl Viro if (old_type != LAST_NORM) 45611da177e4SLinus Torvalds goto exit2; 45621da177e4SLinus Torvalds 45630a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 45640a7c3937SMiklos Szeredi error = -EEXIST; 4565f5beed75SAl Viro if (new_type != LAST_NORM) 45661da177e4SLinus Torvalds goto exit2; 45671da177e4SLinus Torvalds 4568f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4569c30dabfeSJan Kara if (error) 4570c30dabfeSJan Kara goto exit2; 4571c30dabfeSJan Kara 45728e6d782cSJ. Bruce Fields retry_deleg: 4573f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45741da177e4SLinus Torvalds 4575f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45761da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45771da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45781da177e4SLinus Torvalds goto exit3; 45791da177e4SLinus Torvalds /* source must exist */ 45801da177e4SLinus Torvalds error = -ENOENT; 4581b18825a7SDavid Howells if (d_is_negative(old_dentry)) 45821da177e4SLinus Torvalds goto exit4; 4583f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 45841da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45851da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 45861da177e4SLinus Torvalds goto exit4; 45870a7c3937SMiklos Szeredi error = -EEXIST; 45880a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 45890a7c3937SMiklos Szeredi goto exit5; 4590da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4591da1ce067SMiklos Szeredi error = -ENOENT; 4592da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4593da1ce067SMiklos Szeredi goto exit5; 4594da1ce067SMiklos Szeredi 4595da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4596da1ce067SMiklos Szeredi error = -ENOTDIR; 4597f5beed75SAl Viro if (new_last.name[new_last.len]) 4598da1ce067SMiklos Szeredi goto exit5; 4599da1ce067SMiklos Szeredi } 4600da1ce067SMiklos Szeredi } 46010a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 46020a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 46030a7c3937SMiklos Szeredi error = -ENOTDIR; 4604f5beed75SAl Viro if (old_last.name[old_last.len]) 46050a7c3937SMiklos Szeredi goto exit5; 4606f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 46070a7c3937SMiklos Szeredi goto exit5; 46080a7c3937SMiklos Szeredi } 46090a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 46100a7c3937SMiklos Szeredi error = -EINVAL; 46110a7c3937SMiklos Szeredi if (old_dentry == trap) 46120a7c3937SMiklos Szeredi goto exit5; 46131da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4614da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46151da177e4SLinus Torvalds error = -ENOTEMPTY; 46161da177e4SLinus Torvalds if (new_dentry == trap) 46171da177e4SLinus Torvalds goto exit5; 46181da177e4SLinus Torvalds 4619f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4620f5beed75SAl Viro &new_path, new_dentry, flags); 4621be6d3e56SKentaro Takeda if (error) 4622c30dabfeSJan Kara goto exit5; 4623f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4624f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4625520c8b16SMiklos Szeredi &delegated_inode, flags); 46261da177e4SLinus Torvalds exit5: 46271da177e4SLinus Torvalds dput(new_dentry); 46281da177e4SLinus Torvalds exit4: 46291da177e4SLinus Torvalds dput(old_dentry); 46301da177e4SLinus Torvalds exit3: 4631f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 46328e6d782cSJ. Bruce Fields if (delegated_inode) { 46338e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 46348e6d782cSJ. Bruce Fields if (!error) 46358e6d782cSJ. Bruce Fields goto retry_deleg; 46368e6d782cSJ. Bruce Fields } 4637f5beed75SAl Viro mnt_drop_write(old_path.mnt); 46381da177e4SLinus Torvalds exit2: 4639c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4640c6a94284SJeff Layton should_retry = true; 4641f5beed75SAl Viro path_put(&new_path); 46422ad94ae6SAl Viro putname(to); 46431da177e4SLinus Torvalds exit1: 4644f5beed75SAl Viro path_put(&old_path); 46451da177e4SLinus Torvalds putname(from); 4646c6a94284SJeff Layton if (should_retry) { 4647c6a94284SJeff Layton should_retry = false; 4648c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4649c6a94284SJeff Layton goto retry; 4650c6a94284SJeff Layton } 46512ad94ae6SAl Viro exit: 46521da177e4SLinus Torvalds return error; 46531da177e4SLinus Torvalds } 46541da177e4SLinus Torvalds 4655ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4656ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4657ee81feb6SDominik Brodowski { 4658ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4659ee81feb6SDominik Brodowski } 4660ee81feb6SDominik Brodowski 4661520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4662520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4663520c8b16SMiklos Szeredi { 4664ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4665520c8b16SMiklos Szeredi } 4666520c8b16SMiklos Szeredi 4667a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 46685590ff0dSUlrich Drepper { 4669ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 46705590ff0dSUlrich Drepper } 46715590ff0dSUlrich Drepper 4672787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4673787fb6bcSMiklos Szeredi { 4674787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4675787fb6bcSMiklos Szeredi if (error) 4676787fb6bcSMiklos Szeredi return error; 4677787fb6bcSMiklos Szeredi 4678787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4679787fb6bcSMiklos Szeredi return -EPERM; 4680787fb6bcSMiklos Szeredi 4681787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4682787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4683787fb6bcSMiklos Szeredi } 4684787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4685787fb6bcSMiklos Szeredi 46865d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 46871da177e4SLinus Torvalds { 46885d826c84SAl Viro int len = PTR_ERR(link); 46891da177e4SLinus Torvalds if (IS_ERR(link)) 46901da177e4SLinus Torvalds goto out; 46911da177e4SLinus Torvalds 46921da177e4SLinus Torvalds len = strlen(link); 46931da177e4SLinus Torvalds if (len > (unsigned) buflen) 46941da177e4SLinus Torvalds len = buflen; 46951da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 46961da177e4SLinus Torvalds len = -EFAULT; 46971da177e4SLinus Torvalds out: 46981da177e4SLinus Torvalds return len; 46991da177e4SLinus Torvalds } 47001da177e4SLinus Torvalds 47011da177e4SLinus Torvalds /* 47021da177e4SLinus Torvalds * A helper for ->readlink(). This should be used *ONLY* for symlinks that 47036b255391SAl Viro * have ->get_link() not calling nd_jump_link(). Using (or not using) it 47046b255391SAl Viro * for any given inode is up to filesystem. 47051da177e4SLinus Torvalds */ 4706d16744ecSMiklos Szeredi static int generic_readlink(struct dentry *dentry, char __user *buffer, 4707d16744ecSMiklos Szeredi int buflen) 47081da177e4SLinus Torvalds { 4709fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47105f2c4179SAl Viro struct inode *inode = d_inode(dentry); 47115f2c4179SAl Viro const char *link = inode->i_link; 4712694a1764SMarcin Slusarz int res; 4713cc314eefSLinus Torvalds 4714d4dee48bSAl Viro if (!link) { 4715fceef393SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4716680baacbSAl Viro if (IS_ERR(link)) 4717680baacbSAl Viro return PTR_ERR(link); 4718d4dee48bSAl Viro } 4719680baacbSAl Viro res = readlink_copy(buffer, buflen, link); 4720fceef393SAl Viro do_delayed_call(&done); 4721694a1764SMarcin Slusarz return res; 47221da177e4SLinus Torvalds } 47231da177e4SLinus Torvalds 4724d60874cdSMiklos Szeredi /** 4725fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4726fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4727fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4728fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4729fd4a0edfSMiklos Szeredi * 4730fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4731fd4a0edfSMiklos Szeredi * 4732fd4a0edfSMiklos Szeredi * Does not call security hook. 4733fd4a0edfSMiklos Szeredi */ 4734fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4735fd4a0edfSMiklos Szeredi { 4736fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4737fd4a0edfSMiklos Szeredi 473876fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 473976fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 474076fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 474176fca90eSMiklos Szeredi 474276fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4743fd4a0edfSMiklos Szeredi return -EINVAL; 4744fd4a0edfSMiklos Szeredi 474576fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 474676fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 474776fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 474876fca90eSMiklos Szeredi } 474976fca90eSMiklos Szeredi 475076fca90eSMiklos Szeredi return generic_readlink(dentry, buffer, buflen); 4751fd4a0edfSMiklos Szeredi } 4752fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 47531da177e4SLinus Torvalds 4754d60874cdSMiklos Szeredi /** 4755d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4756d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4757d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4758d60874cdSMiklos Szeredi * 4759d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4760d60874cdSMiklos Szeredi * 4761d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4762d60874cdSMiklos Szeredi * 4763d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4764d60874cdSMiklos Szeredi */ 4765d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4766d60874cdSMiklos Szeredi { 4767d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4768d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4769d60874cdSMiklos Szeredi 4770d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4771d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4772d60874cdSMiklos Szeredi if (!res) 4773d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4774d60874cdSMiklos Szeredi } 4775d60874cdSMiklos Szeredi return res; 4776d60874cdSMiklos Szeredi } 4777d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4778d60874cdSMiklos Szeredi 47791da177e4SLinus Torvalds /* get the link contents into pagecache */ 47806b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4781fceef393SAl Viro struct delayed_call *callback) 47821da177e4SLinus Torvalds { 4783ebd09abbSDuane Griffin char *kaddr; 47841da177e4SLinus Torvalds struct page *page; 47856b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47866b255391SAl Viro 4787d3883d4fSAl Viro if (!dentry) { 4788d3883d4fSAl Viro page = find_get_page(mapping, 0); 4789d3883d4fSAl Viro if (!page) 47906b255391SAl Viro return ERR_PTR(-ECHILD); 4791d3883d4fSAl Viro if (!PageUptodate(page)) { 4792d3883d4fSAl Viro put_page(page); 4793d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4794d3883d4fSAl Viro } 4795d3883d4fSAl Viro } else { 4796090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47971da177e4SLinus Torvalds if (IS_ERR(page)) 47986fe6900eSNick Piggin return (char*)page; 4799d3883d4fSAl Viro } 4800fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 480121fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 480221fc61c7SAl Viro kaddr = page_address(page); 48036b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4804ebd09abbSDuane Griffin return kaddr; 48051da177e4SLinus Torvalds } 48061da177e4SLinus Torvalds 48076b255391SAl Viro EXPORT_SYMBOL(page_get_link); 48081da177e4SLinus Torvalds 4809fceef393SAl Viro void page_put_link(void *arg) 48101da177e4SLinus Torvalds { 4811fceef393SAl Viro put_page(arg); 48121da177e4SLinus Torvalds } 48134d359507SAl Viro EXPORT_SYMBOL(page_put_link); 48141da177e4SLinus Torvalds 4815aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4816aa80deabSAl Viro { 4817fceef393SAl Viro DEFINE_DELAYED_CALL(done); 48186b255391SAl Viro int res = readlink_copy(buffer, buflen, 48196b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4820fceef393SAl Viro &done)); 4821fceef393SAl Viro do_delayed_call(&done); 4822aa80deabSAl Viro return res; 4823aa80deabSAl Viro } 4824aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4825aa80deabSAl Viro 482654566b2cSNick Piggin /* 482754566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 482854566b2cSNick Piggin */ 482954566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48301da177e4SLinus Torvalds { 48311da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48320adb25d2SKirill Korotaev struct page *page; 4833afddba49SNick Piggin void *fsdata; 4834beb497abSDmitriy Monakhov int err; 4835c718a975STetsuo Handa unsigned int flags = 0; 483654566b2cSNick Piggin if (nofs) 483754566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 48381da177e4SLinus Torvalds 48397e53cac4SNeilBrown retry: 4840afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 484154566b2cSNick Piggin flags, &page, &fsdata); 48421da177e4SLinus Torvalds if (err) 4843afddba49SNick Piggin goto fail; 4844afddba49SNick Piggin 484521fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4846afddba49SNick Piggin 4847afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4848afddba49SNick Piggin page, fsdata); 48491da177e4SLinus Torvalds if (err < 0) 48501da177e4SLinus Torvalds goto fail; 4851afddba49SNick Piggin if (err < len-1) 4852afddba49SNick Piggin goto retry; 4853afddba49SNick Piggin 48541da177e4SLinus Torvalds mark_inode_dirty(inode); 48551da177e4SLinus Torvalds return 0; 48561da177e4SLinus Torvalds fail: 48571da177e4SLinus Torvalds return err; 48581da177e4SLinus Torvalds } 48594d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 48601da177e4SLinus Torvalds 48610adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 48620adb25d2SKirill Korotaev { 48630adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4864c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 48650adb25d2SKirill Korotaev } 48664d359507SAl Viro EXPORT_SYMBOL(page_symlink); 48670adb25d2SKirill Korotaev 486892e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 48696b255391SAl Viro .get_link = page_get_link, 48701da177e4SLinus Torvalds }; 48711da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4872