1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namei.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * Some corrections by tytso. 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 131da177e4SLinus Torvalds * lookup logic. 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 161da177e4SLinus Torvalds */ 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/init.h> 19630d9c47SPaul Gortmaker #include <linux/export.h> 2044696908SDavid S. Miller #include <linux/kernel.h> 211da177e4SLinus Torvalds #include <linux/slab.h> 221da177e4SLinus Torvalds #include <linux/fs.h> 231da177e4SLinus Torvalds #include <linux/namei.h> 241da177e4SLinus Torvalds #include <linux/pagemap.h> 250eeca283SRobert Love #include <linux/fsnotify.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 286146f0d5SMimi Zohar #include <linux/ima.h> 291da177e4SLinus Torvalds #include <linux/syscalls.h> 301da177e4SLinus Torvalds #include <linux/mount.h> 311da177e4SLinus Torvalds #include <linux/audit.h> 3216f7e0feSRandy Dunlap #include <linux/capability.h> 33834f2a4aSTrond Myklebust #include <linux/file.h> 345590ff0dSUlrich Drepper #include <linux/fcntl.h> 3508ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 365ad4e53bSAl Viro #include <linux/fs_struct.h> 37e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3899d263d4SLinus Torvalds #include <linux/hash.h> 392a18da7aSGeorge Spelvin #include <linux/bitops.h> 40aeaa4a79SEric W. Biederman #include <linux/init_task.h> 417c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 421da177e4SLinus Torvalds 43e81e3f4dSEric Paris #include "internal.h" 44c7105365SAl Viro #include "mount.h" 45e81e3f4dSEric Paris 461da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 471da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 481da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 491da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 501da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 511da177e4SLinus Torvalds * 521da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 531da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 541da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 551da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 561da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 571da177e4SLinus Torvalds * the special cases of the former code. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 601da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 611da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 641da177e4SLinus Torvalds * resolution to correspond with current state of the code. 651da177e4SLinus Torvalds * 661da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 671da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 681da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 691da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 701da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 711da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 721da177e4SLinus Torvalds */ 731da177e4SLinus Torvalds 741da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 751da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 761da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 771da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 781da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 791da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8025985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 811da177e4SLinus Torvalds * 821da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 831da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 841da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 851da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 861da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 871da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 881da177e4SLinus Torvalds * and in the old Linux semantics. 891da177e4SLinus Torvalds */ 901da177e4SLinus Torvalds 911da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 921da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 931da177e4SLinus Torvalds * 941da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 951da177e4SLinus Torvalds */ 961da177e4SLinus Torvalds 971da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 981da177e4SLinus Torvalds * inside the path - always follow. 991da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1001da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1011da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1021da177e4SLinus Torvalds * otherwise - don't follow. 1031da177e4SLinus Torvalds * (applied in that order). 1041da177e4SLinus Torvalds * 1051da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1061da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1071da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1081da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1091da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1101da177e4SLinus Torvalds */ 1111da177e4SLinus Torvalds /* 1121da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 113a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1141da177e4SLinus Torvalds * any extra contention... 1151da177e4SLinus Torvalds */ 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1181da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1191da177e4SLinus Torvalds * kernel data space before using them.. 1201da177e4SLinus Torvalds * 1211da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1221da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1231da177e4SLinus Torvalds */ 1247950e385SJeff Layton 125fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12691a27b2aSJeff Layton 12751f39a1fSDavid Drysdale struct filename * 12891a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12991a27b2aSJeff Layton { 13094b5d262SAl Viro struct filename *result; 1317950e385SJeff Layton char *kname; 13294b5d262SAl Viro int len; 1331da177e4SLinus Torvalds 1347ac86265SJeff Layton result = audit_reusename(filename); 1357ac86265SJeff Layton if (result) 1367ac86265SJeff Layton return result; 1377ac86265SJeff Layton 1387950e385SJeff Layton result = __getname(); 1393f9f0aa6SLinus Torvalds if (unlikely(!result)) 1404043cde8SEric Paris return ERR_PTR(-ENOMEM); 1411da177e4SLinus Torvalds 1427950e385SJeff Layton /* 1437950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1447950e385SJeff Layton * allocation 1457950e385SJeff Layton */ 146fd2f7cb5SAl Viro kname = (char *)result->iname; 14791a27b2aSJeff Layton result->name = kname; 1487950e385SJeff Layton 14994b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15091a27b2aSJeff Layton if (unlikely(len < 0)) { 15194b5d262SAl Viro __putname(result); 15294b5d262SAl Viro return ERR_PTR(len); 15391a27b2aSJeff Layton } 1543f9f0aa6SLinus Torvalds 1557950e385SJeff Layton /* 1567950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1577950e385SJeff Layton * separate struct filename so we can dedicate the entire 1587950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1597950e385SJeff Layton * userland. 1607950e385SJeff Layton */ 16194b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 162fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1637950e385SJeff Layton kname = (char *)result; 1647950e385SJeff Layton 165fd2f7cb5SAl Viro /* 166fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 167fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 168fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 169fd2f7cb5SAl Viro */ 170fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17194b5d262SAl Viro if (unlikely(!result)) { 17294b5d262SAl Viro __putname(kname); 17394b5d262SAl Viro return ERR_PTR(-ENOMEM); 1747950e385SJeff Layton } 1757950e385SJeff Layton result->name = kname; 17694b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17794b5d262SAl Viro if (unlikely(len < 0)) { 17894b5d262SAl Viro __putname(kname); 17994b5d262SAl Viro kfree(result); 18094b5d262SAl Viro return ERR_PTR(len); 18194b5d262SAl Viro } 18294b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18394b5d262SAl Viro __putname(kname); 18494b5d262SAl Viro kfree(result); 18594b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18694b5d262SAl Viro } 1877950e385SJeff Layton } 1887950e385SJeff Layton 18994b5d262SAl Viro result->refcnt = 1; 1903f9f0aa6SLinus Torvalds /* The empty path is special. */ 1913f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1923f9f0aa6SLinus Torvalds if (empty) 1931fa1e7f6SAndy Whitcroft *empty = 1; 19494b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19594b5d262SAl Viro putname(result); 19694b5d262SAl Viro return ERR_PTR(-ENOENT); 1971da177e4SLinus Torvalds } 19894b5d262SAl Viro } 1997950e385SJeff Layton 2007950e385SJeff Layton result->uptr = filename; 201c4ad8f98SLinus Torvalds result->aname = NULL; 2021da177e4SLinus Torvalds audit_getname(result); 2031da177e4SLinus Torvalds return result; 2043f9f0aa6SLinus Torvalds } 2053f9f0aa6SLinus Torvalds 20691a27b2aSJeff Layton struct filename * 20791a27b2aSJeff Layton getname(const char __user * filename) 208f52e0c11SAl Viro { 209f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 210f52e0c11SAl Viro } 211f52e0c11SAl Viro 212c4ad8f98SLinus Torvalds struct filename * 213c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 214c4ad8f98SLinus Torvalds { 215c4ad8f98SLinus Torvalds struct filename *result; 21608518549SPaul Moore int len = strlen(filename) + 1; 217c4ad8f98SLinus Torvalds 218c4ad8f98SLinus Torvalds result = __getname(); 219c4ad8f98SLinus Torvalds if (unlikely(!result)) 220c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 221c4ad8f98SLinus Torvalds 22208518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 223fd2f7cb5SAl Viro result->name = (char *)result->iname; 22408518549SPaul Moore } else if (len <= PATH_MAX) { 22530ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 22608518549SPaul Moore struct filename *tmp; 22708518549SPaul Moore 22830ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 22908518549SPaul Moore if (unlikely(!tmp)) { 23008518549SPaul Moore __putname(result); 23108518549SPaul Moore return ERR_PTR(-ENOMEM); 23208518549SPaul Moore } 23308518549SPaul Moore tmp->name = (char *)result; 23408518549SPaul Moore result = tmp; 23508518549SPaul Moore } else { 23608518549SPaul Moore __putname(result); 23708518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23808518549SPaul Moore } 23908518549SPaul Moore memcpy((char *)result->name, filename, len); 240c4ad8f98SLinus Torvalds result->uptr = NULL; 241c4ad8f98SLinus Torvalds result->aname = NULL; 24255422d0bSPaul Moore result->refcnt = 1; 243fd3522fdSPaul Moore audit_getname(result); 244c4ad8f98SLinus Torvalds 245c4ad8f98SLinus Torvalds return result; 246c4ad8f98SLinus Torvalds } 247c4ad8f98SLinus Torvalds 24891a27b2aSJeff Layton void putname(struct filename *name) 2491da177e4SLinus Torvalds { 25055422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 25155422d0bSPaul Moore 25255422d0bSPaul Moore if (--name->refcnt > 0) 25355422d0bSPaul Moore return; 25455422d0bSPaul Moore 255fd2f7cb5SAl Viro if (name->name != name->iname) { 25655422d0bSPaul Moore __putname(name->name); 25755422d0bSPaul Moore kfree(name); 25855422d0bSPaul Moore } else 25955422d0bSPaul Moore __putname(name); 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds 262e77819e5SLinus Torvalds static int check_acl(struct inode *inode, int mask) 263e77819e5SLinus Torvalds { 26484635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 265e77819e5SLinus Torvalds struct posix_acl *acl; 266e77819e5SLinus Torvalds 267e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2683567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2693567866bSAl Viro if (!acl) 270e77819e5SLinus Torvalds return -EAGAIN; 2713567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 272b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 273e77819e5SLinus Torvalds return -ECHILD; 274206b1d09SAri Savolainen return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK); 275e77819e5SLinus Torvalds } 276e77819e5SLinus Torvalds 2772982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2784e34e719SChristoph Hellwig if (IS_ERR(acl)) 2794e34e719SChristoph Hellwig return PTR_ERR(acl); 280e77819e5SLinus Torvalds if (acl) { 281e77819e5SLinus Torvalds int error = posix_acl_permission(inode, acl, mask); 282e77819e5SLinus Torvalds posix_acl_release(acl); 283e77819e5SLinus Torvalds return error; 284e77819e5SLinus Torvalds } 28584635d68SLinus Torvalds #endif 286e77819e5SLinus Torvalds 287e77819e5SLinus Torvalds return -EAGAIN; 288e77819e5SLinus Torvalds } 289e77819e5SLinus Torvalds 2905909ccaaSLinus Torvalds /* 291948409c7SAndreas Gruenbacher * This does the basic permission checking 2925909ccaaSLinus Torvalds */ 2937e40145eSAl Viro static int acl_permission_check(struct inode *inode, int mask) 2945909ccaaSLinus Torvalds { 29526cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 2965909ccaaSLinus Torvalds 2978e96e3b7SEric W. Biederman if (likely(uid_eq(current_fsuid(), inode->i_uid))) 2985909ccaaSLinus Torvalds mode >>= 6; 2995909ccaaSLinus Torvalds else { 300e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 3017e40145eSAl Viro int error = check_acl(inode, mask); 3025909ccaaSLinus Torvalds if (error != -EAGAIN) 3035909ccaaSLinus Torvalds return error; 3045909ccaaSLinus Torvalds } 3055909ccaaSLinus Torvalds 3065909ccaaSLinus Torvalds if (in_group_p(inode->i_gid)) 3075909ccaaSLinus Torvalds mode >>= 3; 3085909ccaaSLinus Torvalds } 3095909ccaaSLinus Torvalds 3105909ccaaSLinus Torvalds /* 3115909ccaaSLinus Torvalds * If the DACs are ok we don't need any capability check. 3125909ccaaSLinus Torvalds */ 3139c2c7039SAl Viro if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) 3145909ccaaSLinus Torvalds return 0; 3155909ccaaSLinus Torvalds return -EACCES; 3165909ccaaSLinus Torvalds } 3171da177e4SLinus Torvalds 3181da177e4SLinus Torvalds /** 3191da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3201da177e4SLinus Torvalds * @inode: inode to check access rights for 3218fd90c8dSAndreas Gruenbacher * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...) 3221da177e4SLinus Torvalds * 3231da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3241da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3251da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 326b74c79e9SNick Piggin * are used for other things. 327b74c79e9SNick Piggin * 328b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 329b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 330b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 3311da177e4SLinus Torvalds */ 3322830ba7fSAl Viro int generic_permission(struct inode *inode, int mask) 3331da177e4SLinus Torvalds { 3345909ccaaSLinus Torvalds int ret; 3351da177e4SLinus Torvalds 3361da177e4SLinus Torvalds /* 337948409c7SAndreas Gruenbacher * Do the basic permission checks. 3381da177e4SLinus Torvalds */ 3397e40145eSAl Viro ret = acl_permission_check(inode, mask); 3405909ccaaSLinus Torvalds if (ret != -EACCES) 3415909ccaaSLinus Torvalds return ret; 3421da177e4SLinus Torvalds 343d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 344d594e7ecSAl Viro /* DACs are overridable for directories */ 345d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 34623adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, 34723adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 348d594e7ecSAl Viro return 0; 34923adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3501da177e4SLinus Torvalds return 0; 3512a4c2242SStephen Smalley return -EACCES; 3522a4c2242SStephen Smalley } 3531da177e4SLinus Torvalds 3541da177e4SLinus Torvalds /* 3551da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 3561da177e4SLinus Torvalds */ 3577ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 358d594e7ecSAl Viro if (mask == MAY_READ) 35923adbe12SAndy Lutomirski if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH)) 3601da177e4SLinus Torvalds return 0; 3612a4c2242SStephen Smalley /* 3622a4c2242SStephen Smalley * Read/write DACs are always overridable. 3632a4c2242SStephen Smalley * Executable DACs are overridable when there is 3642a4c2242SStephen Smalley * at least one exec bit set. 3652a4c2242SStephen Smalley */ 3662a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 3672a4c2242SStephen Smalley if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE)) 3682a4c2242SStephen Smalley return 0; 3691da177e4SLinus Torvalds 3701da177e4SLinus Torvalds return -EACCES; 3711da177e4SLinus Torvalds } 3724d359507SAl Viro EXPORT_SYMBOL(generic_permission); 3731da177e4SLinus Torvalds 3743ddcd056SLinus Torvalds /* 3753ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 3763ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 3773ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 3783ddcd056SLinus Torvalds * permission function, use the fast case". 3793ddcd056SLinus Torvalds */ 3803ddcd056SLinus Torvalds static inline int do_inode_permission(struct inode *inode, int mask) 3813ddcd056SLinus Torvalds { 3823ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 3833ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 3843ddcd056SLinus Torvalds return inode->i_op->permission(inode, mask); 3853ddcd056SLinus Torvalds 3863ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 3873ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 3883ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 3893ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 3903ddcd056SLinus Torvalds } 3913ddcd056SLinus Torvalds return generic_permission(inode, mask); 3923ddcd056SLinus Torvalds } 3933ddcd056SLinus Torvalds 394cb23beb5SChristoph Hellwig /** 3950bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 3960bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 39755852635SRandy Dunlap * @inode: Inode to check permission on 3980bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 3990bdaea90SDavid Howells * 4000bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4010bdaea90SDavid Howells */ 4020bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4030bdaea90SDavid Howells { 4040bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4050bdaea90SDavid Howells umode_t mode = inode->i_mode; 4060bdaea90SDavid Howells 4070bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 408bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4090bdaea90SDavid Howells return -EROFS; 4100bdaea90SDavid Howells } 4110bdaea90SDavid Howells return 0; 4120bdaea90SDavid Howells } 4130bdaea90SDavid Howells 4140bdaea90SDavid Howells /** 4150bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4160bdaea90SDavid Howells * @inode: Inode to check permission on 4170bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4180bdaea90SDavid Howells * 4190bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4200bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4210bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4220bdaea90SDavid Howells * 4230bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4240bdaea90SDavid Howells */ 4250bdaea90SDavid Howells int inode_permission(struct inode *inode, int mask) 4260bdaea90SDavid Howells { 4270bdaea90SDavid Howells int retval; 4280bdaea90SDavid Howells 4290bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4300bdaea90SDavid Howells if (retval) 4310bdaea90SDavid Howells return retval; 4324bfd054aSEric Biggers 4334bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 4344bfd054aSEric Biggers /* 4354bfd054aSEric Biggers * Nobody gets write access to an immutable file. 4364bfd054aSEric Biggers */ 4374bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 4384bfd054aSEric Biggers return -EPERM; 4394bfd054aSEric Biggers 4404bfd054aSEric Biggers /* 4414bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 4424bfd054aSEric Biggers * written back improperly if their true value is unknown 4434bfd054aSEric Biggers * to the vfs. 4444bfd054aSEric Biggers */ 4454bfd054aSEric Biggers if (HAS_UNMAPPED_ID(inode)) 4464bfd054aSEric Biggers return -EACCES; 4474bfd054aSEric Biggers } 4484bfd054aSEric Biggers 4494bfd054aSEric Biggers retval = do_inode_permission(inode, mask); 4504bfd054aSEric Biggers if (retval) 4514bfd054aSEric Biggers return retval; 4524bfd054aSEric Biggers 4534bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 4544bfd054aSEric Biggers if (retval) 4554bfd054aSEric Biggers return retval; 4564bfd054aSEric Biggers 4574bfd054aSEric Biggers return security_inode_permission(inode, mask); 4580bdaea90SDavid Howells } 4594d359507SAl Viro EXPORT_SYMBOL(inode_permission); 4600bdaea90SDavid Howells 4610bdaea90SDavid Howells /** 4625dd784d0SJan Blunck * path_get - get a reference to a path 4635dd784d0SJan Blunck * @path: path to get the reference to 4645dd784d0SJan Blunck * 4655dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 4665dd784d0SJan Blunck */ 467dcf787f3SAl Viro void path_get(const struct path *path) 4685dd784d0SJan Blunck { 4695dd784d0SJan Blunck mntget(path->mnt); 4705dd784d0SJan Blunck dget(path->dentry); 4715dd784d0SJan Blunck } 4725dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 4735dd784d0SJan Blunck 4745dd784d0SJan Blunck /** 4751d957f9bSJan Blunck * path_put - put a reference to a path 4761d957f9bSJan Blunck * @path: path to put the reference to 4771d957f9bSJan Blunck * 4781d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 4791d957f9bSJan Blunck */ 480dcf787f3SAl Viro void path_put(const struct path *path) 4811da177e4SLinus Torvalds { 4821d957f9bSJan Blunck dput(path->dentry); 4831d957f9bSJan Blunck mntput(path->mnt); 4841da177e4SLinus Torvalds } 4851d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 4861da177e4SLinus Torvalds 487894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 4881f55a6ecSAl Viro struct nameidata { 4891f55a6ecSAl Viro struct path path; 4901f55a6ecSAl Viro struct qstr last; 4911f55a6ecSAl Viro struct path root; 4921f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 4931f55a6ecSAl Viro unsigned int flags; 494ab87f9a5SAleksa Sarai unsigned seq, m_seq, r_seq; 4951f55a6ecSAl Viro int last_type; 4961f55a6ecSAl Viro unsigned depth; 497756daf26SNeilBrown int total_link_count; 498697fc6caSAl Viro struct saved { 499697fc6caSAl Viro struct path link; 500fceef393SAl Viro struct delayed_call done; 501697fc6caSAl Viro const char *name; 5020450b2d1SAl Viro unsigned seq; 503894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5049883d185SAl Viro struct filename *name; 5059883d185SAl Viro struct nameidata *saved; 506fceef393SAl Viro struct inode *link_inode; 5079883d185SAl Viro unsigned root_seq; 5089883d185SAl Viro int dfd; 5093859a271SKees Cook } __randomize_layout; 5101f55a6ecSAl Viro 5119883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 512894bc8c4SAl Viro { 513756daf26SNeilBrown struct nameidata *old = current->nameidata; 514756daf26SNeilBrown p->stack = p->internal; 515c8a53ee5SAl Viro p->dfd = dfd; 516c8a53ee5SAl Viro p->name = name; 517756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5189883d185SAl Viro p->saved = old; 519756daf26SNeilBrown current->nameidata = p; 520894bc8c4SAl Viro } 521894bc8c4SAl Viro 5229883d185SAl Viro static void restore_nameidata(void) 523894bc8c4SAl Viro { 5249883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 525756daf26SNeilBrown 526756daf26SNeilBrown current->nameidata = old; 527756daf26SNeilBrown if (old) 528756daf26SNeilBrown old->total_link_count = now->total_link_count; 529e1a63bbcSAl Viro if (now->stack != now->internal) 530756daf26SNeilBrown kfree(now->stack); 531894bc8c4SAl Viro } 532894bc8c4SAl Viro 533894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 534894bc8c4SAl Viro { 535bc40aee0SAl Viro struct saved *p; 536bc40aee0SAl Viro 537bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 5386da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 539bc40aee0SAl Viro GFP_ATOMIC); 540bc40aee0SAl Viro if (unlikely(!p)) 541bc40aee0SAl Viro return -ECHILD; 542bc40aee0SAl Viro } else { 5436da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 544894bc8c4SAl Viro GFP_KERNEL); 545894bc8c4SAl Viro if (unlikely(!p)) 546894bc8c4SAl Viro return -ENOMEM; 547bc40aee0SAl Viro } 548894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 549894bc8c4SAl Viro nd->stack = p; 550894bc8c4SAl Viro return 0; 551894bc8c4SAl Viro } 552894bc8c4SAl Viro 553397d425dSEric W. Biederman /** 554397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 555397d425dSEric W. Biederman * @path: nameidate to verify 556397d425dSEric W. Biederman * 557397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 558397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 559397d425dSEric W. Biederman */ 560397d425dSEric W. Biederman static bool path_connected(const struct path *path) 561397d425dSEric W. Biederman { 562397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 56395dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 564397d425dSEric W. Biederman 56595dd7758SEric W. Biederman /* Bind mounts and multi-root filesystems can have disconnected paths */ 56695dd7758SEric W. Biederman if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root)) 567397d425dSEric W. Biederman return true; 568397d425dSEric W. Biederman 569397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 570397d425dSEric W. Biederman } 571397d425dSEric W. Biederman 572894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 573894bc8c4SAl Viro { 574da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 575894bc8c4SAl Viro return 0; 576894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 577894bc8c4SAl Viro return 0; 578894bc8c4SAl Viro return __nd_alloc_stack(nd); 579894bc8c4SAl Viro } 580894bc8c4SAl Viro 5817973387aSAl Viro static void drop_links(struct nameidata *nd) 5827973387aSAl Viro { 5837973387aSAl Viro int i = nd->depth; 5847973387aSAl Viro while (i--) { 5857973387aSAl Viro struct saved *last = nd->stack + i; 586fceef393SAl Viro do_delayed_call(&last->done); 587fceef393SAl Viro clear_delayed_call(&last->done); 5887973387aSAl Viro } 5897973387aSAl Viro } 5907973387aSAl Viro 5917973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5927973387aSAl Viro { 5937973387aSAl Viro drop_links(nd); 5947973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5957973387aSAl Viro int i; 5967973387aSAl Viro path_put(&nd->path); 5977973387aSAl Viro for (i = 0; i < nd->depth; i++) 5987973387aSAl Viro path_put(&nd->stack[i].link); 59984a2bd39SAl Viro if (nd->flags & LOOKUP_ROOT_GRABBED) { 600102b8af2SAl Viro path_put(&nd->root); 60184a2bd39SAl Viro nd->flags &= ~LOOKUP_ROOT_GRABBED; 602102b8af2SAl Viro } 6037973387aSAl Viro } else { 6047973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6057973387aSAl Viro rcu_read_unlock(); 6067973387aSAl Viro } 6077973387aSAl Viro nd->depth = 0; 6087973387aSAl Viro } 6097973387aSAl Viro 6107973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6117973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6127973387aSAl Viro struct path *path, unsigned seq) 6137973387aSAl Viro { 6147973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6157973387aSAl Viro if (unlikely(res)) { 6167973387aSAl Viro if (res > 0) 6177973387aSAl Viro path->mnt = NULL; 6187973387aSAl Viro path->dentry = NULL; 6197973387aSAl Viro return false; 6207973387aSAl Viro } 6217973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6227973387aSAl Viro path->dentry = NULL; 6237973387aSAl Viro return false; 6247973387aSAl Viro } 6257973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6267973387aSAl Viro } 6277973387aSAl Viro 6287973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6297973387aSAl Viro { 6307973387aSAl Viro int i; 6317973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6327973387aSAl Viro struct saved *last = nd->stack + i; 6337973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6347973387aSAl Viro drop_links(nd); 6357973387aSAl Viro nd->depth = i + 1; 6367973387aSAl Viro return false; 6377973387aSAl Viro } 6387973387aSAl Viro } 6397973387aSAl Viro return true; 6407973387aSAl Viro } 6417973387aSAl Viro 642ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 643ee594bffSAl Viro { 644adb21d2bSAleksa Sarai /* 645adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 646adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 647adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 648adb21d2bSAleksa Sarai */ 649adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 650adb21d2bSAleksa Sarai return false; 651adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 652ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 653ee594bffSAl Viro return true; 65484a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 655ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 656ee594bffSAl Viro } 657ee594bffSAl Viro 65819660af7SAl Viro /* 65931e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 66019660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 66119660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 66257e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66319660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66419660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66519660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66619660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66731e6b01fSNick Piggin */ 66831e6b01fSNick Piggin 66931e6b01fSNick Piggin /** 67019660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 67119660af7SAl Viro * @nd: nameidata pathwalk data 67239191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67331e6b01fSNick Piggin * 6744675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6754675ac39SAl Viro * for ref-walk mode. 6764675ac39SAl Viro * Must be called from rcu-walk context. 6777973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6787973387aSAl Viro * terminate_walk(). 67931e6b01fSNick Piggin */ 6804675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 68131e6b01fSNick Piggin { 68231e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68331e6b01fSNick Piggin 68431e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 685e5c832d5SLinus Torvalds 686e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6877973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6884675ac39SAl Viro goto out1; 68984a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 69084a2bd39SAl Viro goto out; 691ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 6924675ac39SAl Viro goto out; 6934675ac39SAl Viro rcu_read_unlock(); 6944675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6954675ac39SAl Viro return 0; 6964675ac39SAl Viro 69784a2bd39SAl Viro out1: 6984675ac39SAl Viro nd->path.mnt = NULL; 6994675ac39SAl Viro nd->path.dentry = NULL; 7004675ac39SAl Viro out: 7014675ac39SAl Viro rcu_read_unlock(); 7024675ac39SAl Viro return -ECHILD; 7034675ac39SAl Viro } 7044675ac39SAl Viro 7054675ac39SAl Viro /** 7064675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7074675ac39SAl Viro * @nd: nameidata pathwalk data 7084675ac39SAl Viro * @dentry: child of nd->path.dentry 7094675ac39SAl Viro * @seq: seq number to check dentry against 7104675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7114675ac39SAl Viro * 7124675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7134675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7144675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7154675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7164675ac39SAl Viro * terminate_walk(). 7174675ac39SAl Viro */ 7184675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7194675ac39SAl Viro { 7204675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7214675ac39SAl Viro 7224675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7234675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7244675ac39SAl Viro goto out2; 7257973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7267973387aSAl Viro goto out2; 7274675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7287973387aSAl Viro goto out1; 72948a066e7SAl Viro 73015570086SLinus Torvalds /* 7314675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7324675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7334675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7344675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7354675ac39SAl Viro * know the parent must still be valid if the child sequence number is 73615570086SLinus Torvalds */ 7374675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 738e5c832d5SLinus Torvalds goto out; 73984a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 74084a2bd39SAl Viro goto out_dput; 741e5c832d5SLinus Torvalds /* 742e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 743e5c832d5SLinus Torvalds * still valid and get it if required. 744e5c832d5SLinus Torvalds */ 74584a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 74684a2bd39SAl Viro goto out_dput; 7478b61e74fSAl Viro rcu_read_unlock(); 74831e6b01fSNick Piggin return 0; 74919660af7SAl Viro 7507973387aSAl Viro out2: 7517973387aSAl Viro nd->path.mnt = NULL; 7527973387aSAl Viro out1: 7537973387aSAl Viro nd->path.dentry = NULL; 754e5c832d5SLinus Torvalds out: 7558b61e74fSAl Viro rcu_read_unlock(); 75684a2bd39SAl Viro return -ECHILD; 75784a2bd39SAl Viro out_dput: 75884a2bd39SAl Viro rcu_read_unlock(); 75984a2bd39SAl Viro dput(dentry); 76031e6b01fSNick Piggin return -ECHILD; 76131e6b01fSNick Piggin } 76231e6b01fSNick Piggin 7634ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76434286d66SNick Piggin { 765a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7664ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 767a89f8337SAl Viro else 768a89f8337SAl Viro return 1; 76934286d66SNick Piggin } 77034286d66SNick Piggin 7719f1fafeeSAl Viro /** 7729f1fafeeSAl Viro * complete_walk - successful completion of path walk 7739f1fafeeSAl Viro * @nd: pointer nameidata 77439159de2SJeff Layton * 7759f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7769f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7779f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7789f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7799f1fafeeSAl Viro * need to drop nd->path. 78039159de2SJeff Layton */ 7819f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78239159de2SJeff Layton { 78316c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78439159de2SJeff Layton int status; 78539159de2SJeff Layton 7869f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 787adb21d2bSAleksa Sarai /* 788adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 789adb21d2bSAleksa Sarai * externally-managed nd->root. 790adb21d2bSAleksa Sarai */ 791adb21d2bSAleksa Sarai if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) 7929f1fafeeSAl Viro nd->root.mnt = NULL; 7934675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 79448a066e7SAl Viro return -ECHILD; 79548a066e7SAl Viro } 7969f1fafeeSAl Viro 797adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 798adb21d2bSAleksa Sarai /* 799adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 800adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 801adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 802adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 803adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 804adb21d2bSAleksa Sarai * 805adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 806adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 807adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 808adb21d2bSAleksa Sarai * requested root to userspace. 809adb21d2bSAleksa Sarai * 810adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 811adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 812adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 813adb21d2bSAleksa Sarai */ 814adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 815adb21d2bSAleksa Sarai return -EXDEV; 816adb21d2bSAleksa Sarai } 817adb21d2bSAleksa Sarai 81816c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 81939159de2SJeff Layton return 0; 82039159de2SJeff Layton 821ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 82216c2cd71SAl Viro return 0; 82316c2cd71SAl Viro 824ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 82539159de2SJeff Layton if (status > 0) 82639159de2SJeff Layton return 0; 82739159de2SJeff Layton 82816c2cd71SAl Viro if (!status) 82939159de2SJeff Layton status = -ESTALE; 83016c2cd71SAl Viro 83139159de2SJeff Layton return status; 83239159de2SJeff Layton } 83339159de2SJeff Layton 834740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 8352a737871SAl Viro { 83631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8379e6697e2SAl Viro 838adb21d2bSAleksa Sarai /* 839adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 840adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 841adb21d2bSAleksa Sarai * from the dirfd. 842adb21d2bSAleksa Sarai */ 843adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 844adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 845adb21d2bSAleksa Sarai 8469e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8478f47a016SAl Viro unsigned seq; 848c28cc364SNick Piggin 849c28cc364SNick Piggin do { 850c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 85131e6b01fSNick Piggin nd->root = fs->root; 8528f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 853c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8549e6697e2SAl Viro } else { 8559e6697e2SAl Viro get_fs_root(fs, &nd->root); 85684a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 8579e6697e2SAl Viro } 858740a1678SAleksa Sarai return 0; 85931e6b01fSNick Piggin } 86031e6b01fSNick Piggin 8617b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8627b9337aaSNick Piggin struct nameidata *nd) 863051d3812SIan Kent { 86431e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8654ac91378SJan Blunck dput(nd->path.dentry); 86631e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8674ac91378SJan Blunck mntput(nd->path.mnt); 8689a229683SHuang Shijie } 86931e6b01fSNick Piggin nd->path.mnt = path->mnt; 8704ac91378SJan Blunck nd->path.dentry = path->dentry; 871051d3812SIan Kent } 872051d3812SIan Kent 873248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 874248fb5b9SAl Viro { 875adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 876adb21d2bSAleksa Sarai return -EXDEV; 87772ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 87872ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 87972ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 88072ba2929SAleksa Sarai return -EXDEV; 88172ba2929SAleksa Sarai } 882740a1678SAleksa Sarai if (!nd->root.mnt) { 883740a1678SAleksa Sarai int error = set_root(nd); 884740a1678SAleksa Sarai if (error) 885740a1678SAleksa Sarai return error; 886740a1678SAleksa Sarai } 887248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 888248fb5b9SAl Viro struct dentry *d; 889248fb5b9SAl Viro nd->path = nd->root; 890248fb5b9SAl Viro d = nd->path.dentry; 891248fb5b9SAl Viro nd->inode = d->d_inode; 892248fb5b9SAl Viro nd->seq = nd->root_seq; 893248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 894248fb5b9SAl Viro return -ECHILD; 895248fb5b9SAl Viro } else { 896248fb5b9SAl Viro path_put(&nd->path); 897248fb5b9SAl Viro nd->path = nd->root; 898248fb5b9SAl Viro path_get(&nd->path); 899248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 900248fb5b9SAl Viro } 901248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 902248fb5b9SAl Viro return 0; 903248fb5b9SAl Viro } 904248fb5b9SAl Viro 905b5fb63c1SChristoph Hellwig /* 9066b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 907b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 908b5fb63c1SChristoph Hellwig */ 9091bc82070SAleksa Sarai int nd_jump_link(struct path *path) 910b5fb63c1SChristoph Hellwig { 9114b99d499SAleksa Sarai int error = -ELOOP; 9126e77137bSAl Viro struct nameidata *nd = current->nameidata; 913b5fb63c1SChristoph Hellwig 9144b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9154b99d499SAleksa Sarai goto err; 9164b99d499SAleksa Sarai 91772ba2929SAleksa Sarai error = -EXDEV; 91872ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 91972ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 92072ba2929SAleksa Sarai goto err; 92172ba2929SAleksa Sarai } 922adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 923adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 924adb21d2bSAleksa Sarai goto err; 92572ba2929SAleksa Sarai 9264b99d499SAleksa Sarai path_put(&nd->path); 927b5fb63c1SChristoph Hellwig nd->path = *path; 928b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 929b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 9301bc82070SAleksa Sarai return 0; 9314b99d499SAleksa Sarai 9324b99d499SAleksa Sarai err: 9334b99d499SAleksa Sarai path_put(path); 9344b99d499SAleksa Sarai return error; 935b5fb63c1SChristoph Hellwig } 936b5fb63c1SChristoph Hellwig 937b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 938574197e0SAl Viro { 93921c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 940fceef393SAl Viro do_delayed_call(&last->done); 9416548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 942b9ff4429SAl Viro path_put(&last->link); 943574197e0SAl Viro } 944574197e0SAl Viro 945561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 946561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 94730aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 94830aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 949800179c9SKees Cook 950800179c9SKees Cook /** 951800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 95255852635SRandy Dunlap * @nd: nameidata pathwalk data 953800179c9SKees Cook * 954800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 955800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 956800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 957800179c9SKees Cook * processes from failing races against path names that may change out 958800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 959800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 960800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 961800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 962800179c9SKees Cook * 963800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 964800179c9SKees Cook */ 965fec2fa24SAl Viro static inline int may_follow_link(struct nameidata *nd) 966800179c9SKees Cook { 967800179c9SKees Cook const struct inode *inode; 968800179c9SKees Cook const struct inode *parent; 9692d7f9e2aSSeth Forshee kuid_t puid; 970800179c9SKees Cook 971800179c9SKees Cook if (!sysctl_protected_symlinks) 972800179c9SKees Cook return 0; 973800179c9SKees Cook 974800179c9SKees Cook /* Allowed if owner and follower match. */ 975fceef393SAl Viro inode = nd->link_inode; 97681abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 977800179c9SKees Cook return 0; 978800179c9SKees Cook 979800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 980aa65fa35SAl Viro parent = nd->inode; 981800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 982800179c9SKees Cook return 0; 983800179c9SKees Cook 984800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9852d7f9e2aSSeth Forshee puid = parent->i_uid; 9862d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 987800179c9SKees Cook return 0; 988800179c9SKees Cook 98931956502SAl Viro if (nd->flags & LOOKUP_RCU) 99031956502SAl Viro return -ECHILD; 99131956502SAl Viro 992ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 993245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 994800179c9SKees Cook return -EACCES; 995800179c9SKees Cook } 996800179c9SKees Cook 997800179c9SKees Cook /** 998800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 999800179c9SKees Cook * @inode: the source inode to hardlink from 1000800179c9SKees Cook * 1001800179c9SKees Cook * Return false if at least one of the following conditions: 1002800179c9SKees Cook * - inode is not a regular file 1003800179c9SKees Cook * - inode is setuid 1004800179c9SKees Cook * - inode is setgid and group-exec 1005800179c9SKees Cook * - access failure for read and write 1006800179c9SKees Cook * 1007800179c9SKees Cook * Otherwise returns true. 1008800179c9SKees Cook */ 1009800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 1010800179c9SKees Cook { 1011800179c9SKees Cook umode_t mode = inode->i_mode; 1012800179c9SKees Cook 1013800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1014800179c9SKees Cook if (!S_ISREG(mode)) 1015800179c9SKees Cook return false; 1016800179c9SKees Cook 1017800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1018800179c9SKees Cook if (mode & S_ISUID) 1019800179c9SKees Cook return false; 1020800179c9SKees Cook 1021800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1022800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1023800179c9SKees Cook return false; 1024800179c9SKees Cook 1025800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1026800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 1027800179c9SKees Cook return false; 1028800179c9SKees Cook 1029800179c9SKees Cook return true; 1030800179c9SKees Cook } 1031800179c9SKees Cook 1032800179c9SKees Cook /** 1033800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1034800179c9SKees Cook * @link: the source to hardlink from 1035800179c9SKees Cook * 1036800179c9SKees Cook * Block hardlink when all of: 1037800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1038800179c9SKees Cook * - fsuid does not match inode 1039800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1040f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1041800179c9SKees Cook * 1042800179c9SKees Cook * Returns 0 if successful, -ve on error. 1043800179c9SKees Cook */ 1044800179c9SKees Cook static int may_linkat(struct path *link) 1045800179c9SKees Cook { 1046593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1047593d1ce8SEric W. Biederman 1048593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1049593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 1050593d1ce8SEric W. Biederman return -EOVERFLOW; 1051800179c9SKees Cook 1052800179c9SKees Cook if (!sysctl_protected_hardlinks) 1053800179c9SKees Cook return 0; 1054800179c9SKees Cook 1055800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1056800179c9SKees Cook * otherwise, it must be a safe source. 1057800179c9SKees Cook */ 1058cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1059800179c9SKees Cook return 0; 1060800179c9SKees Cook 1061245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1062800179c9SKees Cook return -EPERM; 1063800179c9SKees Cook } 1064800179c9SKees Cook 106530aba665SSalvatore Mesoraca /** 106630aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 106730aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 106830aba665SSalvatore Mesoraca * exist. 1069d0cb5018SAl Viro * @dir_mode: mode bits of directory 1070d0cb5018SAl Viro * @dir_uid: owner of directory 107130aba665SSalvatore Mesoraca * @inode: the inode of the file to open 107230aba665SSalvatore Mesoraca * 107330aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 107430aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 107530aba665SSalvatore Mesoraca * - the file already exists 107630aba665SSalvatore Mesoraca * - we are in a sticky directory 107730aba665SSalvatore Mesoraca * - we don't own the file 107830aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 107930aba665SSalvatore Mesoraca * - the directory is world writable 108030aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 108130aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 108230aba665SSalvatore Mesoraca * be enough. 108330aba665SSalvatore Mesoraca * 108430aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 108530aba665SSalvatore Mesoraca */ 1086d0cb5018SAl Viro static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid, 108730aba665SSalvatore Mesoraca struct inode * const inode) 108830aba665SSalvatore Mesoraca { 108930aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 109030aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1091d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1092d0cb5018SAl Viro uid_eq(inode->i_uid, dir_uid) || 109330aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 109430aba665SSalvatore Mesoraca return 0; 109530aba665SSalvatore Mesoraca 1096d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1097d0cb5018SAl Viro (dir_mode & 0020 && 109830aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 109930aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1100245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1101245d7369SKees Cook "sticky_create_fifo" : 1102245d7369SKees Cook "sticky_create_regular"; 1103245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 110430aba665SSalvatore Mesoraca return -EACCES; 110530aba665SSalvatore Mesoraca } 110630aba665SSalvatore Mesoraca return 0; 110730aba665SSalvatore Mesoraca } 110830aba665SSalvatore Mesoraca 11093b2e7f75SAl Viro static __always_inline 11103b2e7f75SAl Viro const char *get_link(struct nameidata *nd) 11111da177e4SLinus Torvalds { 1112ab104923SAl Viro struct saved *last = nd->stack + nd->depth - 1; 11131cf2665bSAl Viro struct dentry *dentry = last->link.dentry; 1114fceef393SAl Viro struct inode *inode = nd->link_inode; 11156d7b5aaeSAl Viro int error; 11160a959df5SAl Viro const char *res; 11171da177e4SLinus Torvalds 111843679723SAl Viro if (!(nd->flags & LOOKUP_PARENT)) { 111943679723SAl Viro error = may_follow_link(nd); 112043679723SAl Viro if (unlikely(error)) 112143679723SAl Viro return ERR_PTR(error); 112243679723SAl Viro } 112343679723SAl Viro 112427812141SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) 112527812141SAleksa Sarai return ERR_PTR(-ELOOP); 112627812141SAleksa Sarai 11278fa9dd24SNeilBrown if (!(nd->flags & LOOKUP_RCU)) { 11288fa9dd24SNeilBrown touch_atime(&last->link); 11298fa9dd24SNeilBrown cond_resched(); 1130c6718543SMiklos Szeredi } else if (atime_needs_update(&last->link, inode)) { 11314675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 11328fa9dd24SNeilBrown return ERR_PTR(-ECHILD); 11338fa9dd24SNeilBrown touch_atime(&last->link); 11348fa9dd24SNeilBrown } 11358fa9dd24SNeilBrown 1136bda0be7aSNeilBrown error = security_inode_follow_link(dentry, inode, 1137bda0be7aSNeilBrown nd->flags & LOOKUP_RCU); 1138bda0be7aSNeilBrown if (unlikely(error)) 11396920a440SAl Viro return ERR_PTR(error); 114036f3b4f6SAl Viro 114186acdca1SAl Viro nd->last_type = LAST_BIND; 11424c4f7c19SEric Biggers res = READ_ONCE(inode->i_link); 1143d4dee48bSAl Viro if (!res) { 1144fceef393SAl Viro const char * (*get)(struct dentry *, struct inode *, 1145fceef393SAl Viro struct delayed_call *); 1146fceef393SAl Viro get = inode->i_op->get_link; 11478c1b4566SAl Viro if (nd->flags & LOOKUP_RCU) { 1148fceef393SAl Viro res = get(NULL, inode, &last->done); 11496b255391SAl Viro if (res == ERR_PTR(-ECHILD)) { 11504675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 11518c1b4566SAl Viro return ERR_PTR(-ECHILD); 1152fceef393SAl Viro res = get(dentry, inode, &last->done); 11538c1b4566SAl Viro } 11546b255391SAl Viro } else { 1155fceef393SAl Viro res = get(dentry, inode, &last->done); 11566b255391SAl Viro } 115740fcf5a9SAl Viro if (!res) 115840fcf5a9SAl Viro goto all_done; 115940fcf5a9SAl Viro if (IS_ERR(res)) 1160fab51e8aSAl Viro return res; 11610a959df5SAl Viro } 1162fab51e8aSAl Viro if (*res == '/') { 1163740a1678SAleksa Sarai error = nd_jump_root(nd); 1164740a1678SAleksa Sarai if (unlikely(error)) 1165740a1678SAleksa Sarai return ERR_PTR(error); 1166fab51e8aSAl Viro while (unlikely(*++res == '/')) 1167fab51e8aSAl Viro ; 1168fab51e8aSAl Viro } 116940fcf5a9SAl Viro if (*res) 11700a959df5SAl Viro return res; 117140fcf5a9SAl Viro all_done: // pure jump 117240fcf5a9SAl Viro put_link(nd); 117340fcf5a9SAl Viro return NULL; 11740a959df5SAl Viro } 11756d7b5aaeSAl Viro 1176f015f126SDavid Howells /* 1177f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1178f015f126SDavid Howells * 1179f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1180f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1181f015f126SDavid Howells * Up is towards /. 1182f015f126SDavid Howells * 1183f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1184f015f126SDavid Howells * root. 1185f015f126SDavid Howells */ 1186bab77ebfSAl Viro int follow_up(struct path *path) 11871da177e4SLinus Torvalds { 11880714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11890714a533SAl Viro struct mount *parent; 11901da177e4SLinus Torvalds struct dentry *mountpoint; 119199b7db7bSNick Piggin 119248a066e7SAl Viro read_seqlock_excl(&mount_lock); 11930714a533SAl Viro parent = mnt->mnt_parent; 11943c0a6163SAl Viro if (parent == mnt) { 119548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11961da177e4SLinus Torvalds return 0; 11971da177e4SLinus Torvalds } 11980714a533SAl Viro mntget(&parent->mnt); 1199a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 120048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1201bab77ebfSAl Viro dput(path->dentry); 1202bab77ebfSAl Viro path->dentry = mountpoint; 1203bab77ebfSAl Viro mntput(path->mnt); 12040714a533SAl Viro path->mnt = &parent->mnt; 12051da177e4SLinus Torvalds return 1; 12061da177e4SLinus Torvalds } 12074d359507SAl Viro EXPORT_SYMBOL(follow_up); 12081da177e4SLinus Torvalds 1209b5c84bf6SNick Piggin /* 12109875cf80SDavid Howells * Perform an automount 12119875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 12129875cf80SDavid Howells * were called with. 12131da177e4SLinus Torvalds */ 12141c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 121531e6b01fSNick Piggin { 121625e195aaSAl Viro struct dentry *dentry = path->dentry; 12179875cf80SDavid Howells 12180ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 12190ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 12200ec26fd0SMiklos Szeredi * the name. 12210ec26fd0SMiklos Szeredi * 12220ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 12235a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 12240ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 12250ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 12260ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 12270ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 12285a30d8a2SDavid Howells */ 12291c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 12305d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 123125e195aaSAl Viro dentry->d_inode) 12329875cf80SDavid Howells return -EISDIR; 12330ec26fd0SMiklos Szeredi 12341c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 12359875cf80SDavid Howells return -ELOOP; 12369875cf80SDavid Howells 123725e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1238ea5b778aSDavid Howells } 12399875cf80SDavid Howells 12409875cf80SDavid Howells /* 12419875cf80SDavid Howells * Handle a dentry that is managed in some way. 1242cc53ce53SDavid Howells * - Flagged for transit management (autofs) 12439875cf80SDavid Howells * - Flagged as mountpoint 12449875cf80SDavid Howells * - Flagged as automount point 12459875cf80SDavid Howells * 12469875cf80SDavid Howells * This may only be called in refwalk mode. 1247d41efb52SAl Viro * On success path->dentry is known positive. 12489875cf80SDavid Howells * 12499875cf80SDavid Howells * Serialization is taken care of in namespace.c 12509875cf80SDavid Howells */ 1251756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 12529875cf80SDavid Howells { 12538aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 1254d41efb52SAl Viro unsigned flags; 12559875cf80SDavid Howells bool need_mntput = false; 12568aef1884SAl Viro int ret = 0; 12579875cf80SDavid Howells 12589875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 12599875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 12609875cf80SDavid Howells * the components of that value change under us */ 12612fa6b1e0SAl Viro while (flags = smp_load_acquire(&path->dentry->d_flags), 1262d41efb52SAl Viro unlikely(flags & DCACHE_MANAGED_DENTRY)) { 1263cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1264cc53ce53SDavid Howells * being held. */ 1265d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1266cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1267cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1268fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1269508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1270cc53ce53SDavid Howells if (ret < 0) 12718aef1884SAl Viro break; 1272cc53ce53SDavid Howells } 1273cc53ce53SDavid Howells 12749875cf80SDavid Howells /* Transit to a mounted filesystem. */ 1275d41efb52SAl Viro if (flags & DCACHE_MOUNTED) { 12769875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12779875cf80SDavid Howells if (mounted) { 12789875cf80SDavid Howells dput(path->dentry); 12799875cf80SDavid Howells if (need_mntput) 1280463ffb2eSAl Viro mntput(path->mnt); 1281463ffb2eSAl Viro path->mnt = mounted; 1282463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12839875cf80SDavid Howells need_mntput = true; 12849875cf80SDavid Howells continue; 1285463ffb2eSAl Viro } 1286463ffb2eSAl Viro 12879875cf80SDavid Howells /* Something is mounted on this dentry in another 12889875cf80SDavid Howells * namespace and/or whatever was mounted there in this 128948a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 129048a066e7SAl Viro * get it */ 12911da177e4SLinus Torvalds } 12929875cf80SDavid Howells 12939875cf80SDavid Howells /* Handle an automount point */ 1294d41efb52SAl Viro if (flags & DCACHE_NEED_AUTOMOUNT) { 12951c9f5e06SAl Viro ret = follow_automount(path, &nd->total_link_count, 12961c9f5e06SAl Viro nd->flags); 12979875cf80SDavid Howells if (ret < 0) 12988aef1884SAl Viro break; 12999875cf80SDavid Howells continue; 13009875cf80SDavid Howells } 13019875cf80SDavid Howells 13029875cf80SDavid Howells /* We didn't change the current path point */ 13039875cf80SDavid Howells break; 13049875cf80SDavid Howells } 13058aef1884SAl Viro 130672ba2929SAleksa Sarai if (need_mntput) { 130772ba2929SAleksa Sarai if (path->mnt == mnt) 13088aef1884SAl Viro mntput(path->mnt); 130972ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 131072ba2929SAleksa Sarai ret = -EXDEV; 131172ba2929SAleksa Sarai else 13128402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 131372ba2929SAleksa Sarai } 1314d41efb52SAl Viro if (ret == -EISDIR || !ret) 1315d41efb52SAl Viro ret = 1; 1316d41efb52SAl Viro if (ret > 0 && unlikely(d_flags_negative(flags))) 1317d41efb52SAl Viro ret = -ENOENT; 1318a9dc1494SAl Viro if (unlikely(ret < 0)) { 1319a9dc1494SAl Viro dput(path->dentry); 1320a9dc1494SAl Viro if (path->mnt != nd->path.mnt) 1321a9dc1494SAl Viro mntput(path->mnt); 1322a9dc1494SAl Viro } 13238402752eSAl Viro return ret; 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds 1326cc53ce53SDavid Howells int follow_down_one(struct path *path) 13271da177e4SLinus Torvalds { 13281da177e4SLinus Torvalds struct vfsmount *mounted; 13291da177e4SLinus Torvalds 13301c755af4SAl Viro mounted = lookup_mnt(path); 13311da177e4SLinus Torvalds if (mounted) { 13329393bd07SAl Viro dput(path->dentry); 13339393bd07SAl Viro mntput(path->mnt); 13349393bd07SAl Viro path->mnt = mounted; 13359393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 13361da177e4SLinus Torvalds return 1; 13371da177e4SLinus Torvalds } 13381da177e4SLinus Torvalds return 0; 13391da177e4SLinus Torvalds } 13404d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 13411da177e4SLinus Torvalds 1342fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 134362a7375eSIan Kent { 1344fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1345fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 134662a7375eSIan Kent } 134762a7375eSIan Kent 13489875cf80SDavid Howells /* 1349287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1350287548e4SAl Viro * we meet a managed dentry that would need blocking. 13519875cf80SDavid Howells */ 13529875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1353254cf582SAl Viro struct inode **inode, unsigned *seqp) 13549875cf80SDavid Howells { 135562a7375eSIan Kent for (;;) { 1356c7105365SAl Viro struct mount *mounted; 135762a7375eSIan Kent /* 135862a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 135962a7375eSIan Kent * that wants to block transit. 136062a7375eSIan Kent */ 1361fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1362b8faf035SNeilBrown case -ECHILD: 1363b8faf035SNeilBrown default: 1364ab90911fSDavid Howells return false; 1365b8faf035SNeilBrown case -EISDIR: 1366b8faf035SNeilBrown return true; 1367b8faf035SNeilBrown case 0: 1368b8faf035SNeilBrown break; 1369b8faf035SNeilBrown } 137062a7375eSIan Kent 137162a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1372b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 137362a7375eSIan Kent 1374474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 13759875cf80SDavid Howells if (!mounted) 13769875cf80SDavid Howells break; 137772ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 137872ba2929SAleksa Sarai return false; 1379c7105365SAl Viro path->mnt = &mounted->mnt; 1380c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1381a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1382254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 138359430262SLinus Torvalds /* 138459430262SLinus Torvalds * Update the inode too. We don't need to re-check the 138559430262SLinus Torvalds * dentry sequence number here after this d_inode read, 138659430262SLinus Torvalds * because a mount-point is always pinned. 138759430262SLinus Torvalds */ 138859430262SLinus Torvalds *inode = path->dentry->d_inode; 13899875cf80SDavid Howells } 1390f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1391b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1392287548e4SAl Viro } 1393287548e4SAl Viro 1394db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1395db3c9adeSAl Viro struct path *path, struct inode **inode, 1396db3c9adeSAl Viro unsigned int *seqp) 1397bd7c4b50SAl Viro { 1398db3c9adeSAl Viro int ret; 1399bd7c4b50SAl Viro 1400db3c9adeSAl Viro path->mnt = nd->path.mnt; 1401db3c9adeSAl Viro path->dentry = dentry; 1402c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1403c153007bSAl Viro unsigned int seq = *seqp; 1404c153007bSAl Viro if (unlikely(!*inode)) 1405c153007bSAl Viro return -ENOENT; 1406c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1407c153007bSAl Viro return 1; 1408c153007bSAl Viro if (unlazy_child(nd, dentry, seq)) 1409c153007bSAl Viro return -ECHILD; 1410c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1411c153007bSAl Viro path->mnt = nd->path.mnt; 1412c153007bSAl Viro path->dentry = dentry; 1413c153007bSAl Viro } 1414db3c9adeSAl Viro ret = follow_managed(path, nd); 1415bd7c4b50SAl Viro if (likely(ret >= 0)) { 1416bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1417bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1418bd7c4b50SAl Viro } 1419bd7c4b50SAl Viro return ret; 1420bd7c4b50SAl Viro } 1421bd7c4b50SAl Viro 142231e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 142331e6b01fSNick Piggin { 14244023bfc9SAl Viro struct inode *inode = nd->inode; 142531e6b01fSNick Piggin 142631e6b01fSNick Piggin while (1) { 1427adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1428adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1429adb21d2bSAleksa Sarai return -ECHILD; 143031e6b01fSNick Piggin break; 1431adb21d2bSAleksa Sarai } 143231e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 143331e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 143431e6b01fSNick Piggin struct dentry *parent = old->d_parent; 143531e6b01fSNick Piggin unsigned seq; 143631e6b01fSNick Piggin 14374023bfc9SAl Viro inode = parent->d_inode; 143831e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1439aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1440aed434adSAl Viro return -ECHILD; 144131e6b01fSNick Piggin nd->path.dentry = parent; 144231e6b01fSNick Piggin nd->seq = seq; 1443397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 14442b98149cSAleksa Sarai return -ECHILD; 144531e6b01fSNick Piggin break; 1446aed434adSAl Viro } else { 1447aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1448aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1449aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1450aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1451aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1452aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1453aed434adSAl Viro return -ECHILD; 1454aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 145531e6b01fSNick Piggin break; 145672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 145772ba2929SAleksa Sarai return -ECHILD; 1458aed434adSAl Viro /* we know that mountpoint was pinned */ 1459aed434adSAl Viro nd->path.dentry = mountpoint; 1460aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1461aed434adSAl Viro inode = inode2; 1462aed434adSAl Viro nd->seq = seq; 146331e6b01fSNick Piggin } 1464aed434adSAl Viro } 1465aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1466b37199e6SAl Viro struct mount *mounted; 1467b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1468aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1469aed434adSAl Viro return -ECHILD; 1470b37199e6SAl Viro if (!mounted) 1471b37199e6SAl Viro break; 147272ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 147372ba2929SAleksa Sarai return -ECHILD; 1474b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1475b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 14764023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1477b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1478b37199e6SAl Viro } 14794023bfc9SAl Viro nd->inode = inode; 148031e6b01fSNick Piggin return 0; 148131e6b01fSNick Piggin } 148231e6b01fSNick Piggin 14839875cf80SDavid Howells /* 1484cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1485cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1486cc53ce53SDavid Howells * caller is permitted to proceed or not. 1487cc53ce53SDavid Howells */ 14887cc90cc3SAl Viro int follow_down(struct path *path) 1489cc53ce53SDavid Howells { 1490cc53ce53SDavid Howells unsigned managed; 1491cc53ce53SDavid Howells int ret; 1492cc53ce53SDavid Howells 14936aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1494cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1495cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1496cc53ce53SDavid Howells * being held. 1497cc53ce53SDavid Howells * 1498cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1499cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1500cc53ce53SDavid Howells * other than its daemon the right to mount on its 1501cc53ce53SDavid Howells * superstructure. 1502cc53ce53SDavid Howells * 1503cc53ce53SDavid Howells * The filesystem may sleep at this point. 1504cc53ce53SDavid Howells */ 1505cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1506cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1507cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1508fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1509cc53ce53SDavid Howells if (ret < 0) 1510cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1511cc53ce53SDavid Howells } 1512cc53ce53SDavid Howells 1513cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1514cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1515cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1516cc53ce53SDavid Howells if (!mounted) 1517cc53ce53SDavid Howells break; 1518cc53ce53SDavid Howells dput(path->dentry); 1519cc53ce53SDavid Howells mntput(path->mnt); 1520cc53ce53SDavid Howells path->mnt = mounted; 1521cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1522cc53ce53SDavid Howells continue; 1523cc53ce53SDavid Howells } 1524cc53ce53SDavid Howells 1525cc53ce53SDavid Howells /* Don't handle automount points here */ 1526cc53ce53SDavid Howells break; 1527cc53ce53SDavid Howells } 1528cc53ce53SDavid Howells return 0; 1529cc53ce53SDavid Howells } 15304d359507SAl Viro EXPORT_SYMBOL(follow_down); 1531cc53ce53SDavid Howells 1532cc53ce53SDavid Howells /* 15339875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 15349875cf80SDavid Howells */ 15359875cf80SDavid Howells static void follow_mount(struct path *path) 15369875cf80SDavid Howells { 15379875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 15389875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 15399875cf80SDavid Howells if (!mounted) 15409875cf80SDavid Howells break; 15419875cf80SDavid Howells dput(path->dentry); 15429875cf80SDavid Howells mntput(path->mnt); 15439875cf80SDavid Howells path->mnt = mounted; 15449875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 15459875cf80SDavid Howells } 15469875cf80SDavid Howells } 15479875cf80SDavid Howells 1548eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1549eedf265aSEric W. Biederman { 1550eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1551eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1552eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1553eedf265aSEric W. Biederman dput(old); 1554eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1555eedf265aSEric W. Biederman return -ENOENT; 1556eedf265aSEric W. Biederman return 0; 1557eedf265aSEric W. Biederman } 1558eedf265aSEric W. Biederman 1559397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 15601da177e4SLinus Torvalds { 15611da177e4SLinus Torvalds while (1) { 1562adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1563adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1564adb21d2bSAleksa Sarai return -EXDEV; 15651da177e4SLinus Torvalds break; 1566adb21d2bSAleksa Sarai } 15674ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1568eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1569eedf265aSEric W. Biederman if (ret) 1570eedf265aSEric W. Biederman return ret; 15711da177e4SLinus Torvalds break; 15721da177e4SLinus Torvalds } 15733088dd70SAl Viro if (!follow_up(&nd->path)) 15741da177e4SLinus Torvalds break; 157572ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 157672ba2929SAleksa Sarai return -EXDEV; 15771da177e4SLinus Torvalds } 157879ed0226SAl Viro follow_mount(&nd->path); 157931e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1580397d425dSEric W. Biederman return 0; 15811da177e4SLinus Torvalds } 15821da177e4SLinus Torvalds 15831da177e4SLinus Torvalds /* 1584f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1585f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1586baa03890SNick Piggin */ 1587e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1588e3c13928SAl Viro struct dentry *dir, 15896c51e513SAl Viro unsigned int flags) 1590baa03890SNick Piggin { 1591a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1592bad61189SMiklos Szeredi if (dentry) { 1593a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1594bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 159574ff0ffcSAl Viro if (!error) 15965542aa2fSEric W. Biederman d_invalidate(dentry); 1597bad61189SMiklos Szeredi dput(dentry); 159874ff0ffcSAl Viro return ERR_PTR(error); 1599bad61189SMiklos Szeredi } 1600bad61189SMiklos Szeredi } 1601baa03890SNick Piggin return dentry; 1602baa03890SNick Piggin } 1603baa03890SNick Piggin 1604baa03890SNick Piggin /* 1605a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1606a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1607a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1608a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1609a03ece5fSAl Viro * at all. 161044396f4bSJosef Bacik */ 1611a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1612a03ece5fSAl Viro struct dentry *base, unsigned int flags) 161344396f4bSJosef Bacik { 1614a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 161544396f4bSJosef Bacik struct dentry *old; 1616a03ece5fSAl Viro struct inode *dir = base->d_inode; 1617a03ece5fSAl Viro 1618a03ece5fSAl Viro if (dentry) 1619a03ece5fSAl Viro return dentry; 162044396f4bSJosef Bacik 162144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1622a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 162344396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1624a03ece5fSAl Viro 1625a03ece5fSAl Viro dentry = d_alloc(base, name); 1626a03ece5fSAl Viro if (unlikely(!dentry)) 1627a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 162844396f4bSJosef Bacik 162972bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 163044396f4bSJosef Bacik if (unlikely(old)) { 163144396f4bSJosef Bacik dput(dentry); 163244396f4bSJosef Bacik dentry = old; 163344396f4bSJosef Bacik } 163444396f4bSJosef Bacik return dentry; 163544396f4bSJosef Bacik } 163644396f4bSJosef Bacik 163720e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 163820e34357SAl Viro struct inode **inode, 1639254cf582SAl Viro unsigned *seqp) 16401da177e4SLinus Torvalds { 164131e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16425a18fff2SAl Viro int status = 1; 16439875cf80SDavid Howells 16443cac260aSAl Viro /* 1645b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16465d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16475d0f49c1SAl Viro * going to fall back to non-racy lookup. 1648b04f784eSNick Piggin */ 164931e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 165031e6b01fSNick Piggin unsigned seq; 1651da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 16525d0f49c1SAl Viro if (unlikely(!dentry)) { 16534675ac39SAl Viro if (unlazy_walk(nd)) 165420e34357SAl Viro return ERR_PTR(-ECHILD); 165520e34357SAl Viro return NULL; 16565d0f49c1SAl Viro } 16575a18fff2SAl Viro 165812f8ad4bSLinus Torvalds /* 165912f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 166012f8ad4bSLinus Torvalds * the dentry name information from lookup. 166112f8ad4bSLinus Torvalds */ 166263afdfc7SDavid Howells *inode = d_backing_inode(dentry); 16635d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 166420e34357SAl Viro return ERR_PTR(-ECHILD); 166512f8ad4bSLinus Torvalds 166612f8ad4bSLinus Torvalds /* 166712f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 166812f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 166912f8ad4bSLinus Torvalds * 167012f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 167112f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 167212f8ad4bSLinus Torvalds */ 16735d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 167420e34357SAl Viro return ERR_PTR(-ECHILD); 16755a18fff2SAl Viro 1676254cf582SAl Viro *seqp = seq; 16774ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1678c153007bSAl Viro if (likely(status > 0)) 167920e34357SAl Viro return dentry; 16804675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 168120e34357SAl Viro return ERR_PTR(-ECHILD); 1682209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1683209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1684209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16855a18fff2SAl Viro } else { 1686e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 168781e6f520SAl Viro if (unlikely(!dentry)) 168820e34357SAl Viro return NULL; 16894ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16905d0f49c1SAl Viro } 16915a18fff2SAl Viro if (unlikely(status <= 0)) { 1692e9742b53SAl Viro if (!status) 16935d0f49c1SAl Viro d_invalidate(dentry); 16945a18fff2SAl Viro dput(dentry); 169520e34357SAl Viro return ERR_PTR(status); 16965a18fff2SAl Viro } 169720e34357SAl Viro return dentry; 1698697f514dSMiklos Szeredi } 1699697f514dSMiklos Szeredi 1700697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 170188d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1702e3c13928SAl Viro struct dentry *dir, 1703e3c13928SAl Viro unsigned int flags) 1704697f514dSMiklos Szeredi { 170588d8331aSAl Viro struct dentry *dentry, *old; 17061936386eSAl Viro struct inode *inode = dir->d_inode; 1707d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 17081936386eSAl Viro 17091936386eSAl Viro /* Don't go there if it's already dead */ 171094bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 171188d8331aSAl Viro return ERR_PTR(-ENOENT); 171294bdd655SAl Viro again: 1713d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 171494bdd655SAl Viro if (IS_ERR(dentry)) 171588d8331aSAl Viro return dentry; 171694bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1717949a852eSAl Viro int error = d_revalidate(dentry, flags); 1718949a852eSAl Viro if (unlikely(error <= 0)) { 171994bdd655SAl Viro if (!error) { 1720949a852eSAl Viro d_invalidate(dentry); 1721949a852eSAl Viro dput(dentry); 172294bdd655SAl Viro goto again; 172394bdd655SAl Viro } 172494bdd655SAl Viro dput(dentry); 1725949a852eSAl Viro dentry = ERR_PTR(error); 1726949a852eSAl Viro } 172794bdd655SAl Viro } else { 17281936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 172985c7f810SAl Viro d_lookup_done(dentry); 17301936386eSAl Viro if (unlikely(old)) { 17311936386eSAl Viro dput(dentry); 17321936386eSAl Viro dentry = old; 1733949a852eSAl Viro } 1734949a852eSAl Viro } 1735e3c13928SAl Viro return dentry; 17361da177e4SLinus Torvalds } 17371da177e4SLinus Torvalds 173888d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 173988d8331aSAl Viro struct dentry *dir, 174088d8331aSAl Viro unsigned int flags) 174188d8331aSAl Viro { 174288d8331aSAl Viro struct inode *inode = dir->d_inode; 174388d8331aSAl Viro struct dentry *res; 174488d8331aSAl Viro inode_lock_shared(inode); 174588d8331aSAl Viro res = __lookup_slow(name, dir, flags); 174688d8331aSAl Viro inode_unlock_shared(inode); 174788d8331aSAl Viro return res; 174888d8331aSAl Viro } 174988d8331aSAl Viro 175052094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 175152094c8aSAl Viro { 175252094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17534ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 175452094c8aSAl Viro if (err != -ECHILD) 175552094c8aSAl Viro return err; 17564675ac39SAl Viro if (unlazy_walk(nd)) 175752094c8aSAl Viro return -ECHILD; 175852094c8aSAl Viro } 17594ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 176052094c8aSAl Viro } 176152094c8aSAl Viro 17629856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 17639856fa1bSAl Viro { 17649856fa1bSAl Viro if (type == LAST_DOTDOT) { 1765740a1678SAleksa Sarai int error = 0; 1766740a1678SAleksa Sarai 1767740a1678SAleksa Sarai if (!nd->root.mnt) { 1768740a1678SAleksa Sarai error = set_root(nd); 1769740a1678SAleksa Sarai if (error) 1770740a1678SAleksa Sarai return error; 1771740a1678SAleksa Sarai } 1772ab87f9a5SAleksa Sarai if (nd->flags & LOOKUP_RCU) 1773ab87f9a5SAleksa Sarai error = follow_dotdot_rcu(nd); 1774ab87f9a5SAleksa Sarai else 1775ab87f9a5SAleksa Sarai error = follow_dotdot(nd); 1776ab87f9a5SAleksa Sarai if (error) 1777ab87f9a5SAleksa Sarai return error; 1778ab87f9a5SAleksa Sarai 1779ab87f9a5SAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1780ab87f9a5SAleksa Sarai /* 1781ab87f9a5SAleksa Sarai * If there was a racing rename or mount along our 1782ab87f9a5SAleksa Sarai * path, then we can't be sure that ".." hasn't jumped 1783ab87f9a5SAleksa Sarai * above nd->root (and so userspace should retry or use 1784ab87f9a5SAleksa Sarai * some fallback). 1785ab87f9a5SAleksa Sarai */ 1786ab87f9a5SAleksa Sarai smp_rmb(); 1787ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 1788ab87f9a5SAleksa Sarai return -EAGAIN; 1789ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 1790ab87f9a5SAleksa Sarai return -EAGAIN; 1791ab87f9a5SAleksa Sarai } 17929856fa1bSAl Viro } 17939856fa1bSAl Viro return 0; 17949856fa1bSAl Viro } 17959856fa1bSAl Viro 1796181548c0SAl Viro static int pick_link(struct nameidata *nd, struct path *link, 1797181548c0SAl Viro struct inode *inode, unsigned seq) 1798d63ff28fSAl Viro { 1799626de996SAl Viro int error; 18001cf2665bSAl Viro struct saved *last; 1801756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1802626de996SAl Viro path_to_nameidata(link, nd); 1803626de996SAl Viro return -ELOOP; 1804626de996SAl Viro } 1805bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1806cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1807cd179f44SAl Viro mntget(link->mnt); 18087973387aSAl Viro } 1809626de996SAl Viro error = nd_alloc_stack(nd); 1810626de996SAl Viro if (unlikely(error)) { 1811bc40aee0SAl Viro if (error == -ECHILD) { 1812ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1813ad1633a1SAl Viro drop_links(nd); 1814ad1633a1SAl Viro nd->depth = 0; 1815ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1816ad1633a1SAl Viro nd->path.mnt = NULL; 1817ad1633a1SAl Viro nd->path.dentry = NULL; 1818ad1633a1SAl Viro rcu_read_unlock(); 18194675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1820bc40aee0SAl Viro error = nd_alloc_stack(nd); 1821bc40aee0SAl Viro } 1822bc40aee0SAl Viro if (error) { 1823cd179f44SAl Viro path_put(link); 1824626de996SAl Viro return error; 1825626de996SAl Viro } 1826bc40aee0SAl Viro } 1827626de996SAl Viro 1828ab104923SAl Viro last = nd->stack + nd->depth++; 18291cf2665bSAl Viro last->link = *link; 1830fceef393SAl Viro clear_delayed_call(&last->done); 1831fceef393SAl Viro nd->link_inode = inode; 18320450b2d1SAl Viro last->seq = seq; 1833d63ff28fSAl Viro return 1; 1834d63ff28fSAl Viro } 1835d63ff28fSAl Viro 1836aca2903eSAl Viro enum {WALK_FOLLOW = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 183731d66bcdSAl Viro 18383ddcd056SLinus Torvalds /* 18393ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18403ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18413ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18423ddcd056SLinus Torvalds * for the common case. 18433ddcd056SLinus Torvalds */ 1844b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1845cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 18463ddcd056SLinus Torvalds { 1847cbae4d12SAl Viro struct path path; 1848cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1849cbae4d12SAl Viro 1850cbae4d12SAl Viro if (err < 0) 1851b0417d2cSAl Viro return ERR_PTR(err); 1852cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 1853aca2903eSAl Viro !((flags & WALK_FOLLOW) || (nd->flags & LOOKUP_FOLLOW)) || 1854aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18558f64fb1cSAl Viro /* not a symlink or should not follow */ 1856cbae4d12SAl Viro path_to_nameidata(&path, nd); 18578f64fb1cSAl Viro nd->inode = inode; 18588f64fb1cSAl Viro nd->seq = seq; 1859b0417d2cSAl Viro return NULL; 18608f64fb1cSAl Viro } 1861a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1862a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 1863cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1864b0417d2cSAl Viro return ERR_PTR(-ECHILD); 1865a7f77542SAl Viro } 1866b0417d2cSAl Viro err = pick_link(nd, &path, inode, seq); 1867b0417d2cSAl Viro if (err > 0) 1868b0417d2cSAl Viro return get_link(nd); 1869b0417d2cSAl Viro return ERR_PTR(err); 18703ddcd056SLinus Torvalds } 18713ddcd056SLinus Torvalds 187292d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1873ce57dfc1SAl Viro { 1874db3c9adeSAl Viro struct dentry *dentry; 1875ce57dfc1SAl Viro struct inode *inode; 1876254cf582SAl Viro unsigned seq; 1877ce57dfc1SAl Viro int err; 1878ce57dfc1SAl Viro /* 1879ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1880ce57dfc1SAl Viro * to be able to know about the current root directory and 1881ce57dfc1SAl Viro * parent relationships. 1882ce57dfc1SAl Viro */ 18834693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18841c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18854693a547SAl Viro put_link(nd); 188656676ec3SAl Viro err = handle_dots(nd, nd->last_type); 188792d27016SAl Viro return ERR_PTR(err); 18884693a547SAl Viro } 188920e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 189020e34357SAl Viro if (IS_ERR(dentry)) 189192d27016SAl Viro return ERR_CAST(dentry); 189220e34357SAl Viro if (unlikely(!dentry)) { 1893db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1894db3c9adeSAl Viro if (IS_ERR(dentry)) 189592d27016SAl Viro return ERR_CAST(dentry); 189620e34357SAl Viro } 189756676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 189856676ec3SAl Viro put_link(nd); 1899b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1900ce57dfc1SAl Viro } 1901ce57dfc1SAl Viro 19021da177e4SLinus Torvalds /* 1903bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1904bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1905bfcfaa77SLinus Torvalds * 1906bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1907bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1908bfcfaa77SLinus Torvalds * fast. 1909bfcfaa77SLinus Torvalds * 1910bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1911bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1912bfcfaa77SLinus Torvalds * crossing operation. 1913bfcfaa77SLinus Torvalds * 1914bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1915bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1916bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1917bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1918bfcfaa77SLinus Torvalds */ 1919bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1920bfcfaa77SLinus Torvalds 1921f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1922bfcfaa77SLinus Torvalds 1923468a9428SGeorge Spelvin #ifdef HASH_MIX 1924bfcfaa77SLinus Torvalds 1925468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1926468a9428SGeorge Spelvin 1927468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 19282a18da7aSGeorge Spelvin /* 19292a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 19302a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 19312a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 19322a18da7aSGeorge Spelvin * and no temporaries. 19332a18da7aSGeorge Spelvin * 19342a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 19352a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 19362a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 19372a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 19382a18da7aSGeorge Spelvin * 19392a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 19402a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 19412a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 19422a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 19432a18da7aSGeorge Spelvin * 19442a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 19452a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 19462a18da7aSGeorge Spelvin * toggling any given output bit. 19472a18da7aSGeorge Spelvin * 19482a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 19492a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19502a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 19512a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 19522a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 19532a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 19542a18da7aSGeorge Spelvin * Perfect: 8192 258048 19552a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 19562a18da7aSGeorge Spelvin */ 19572a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19582a18da7aSGeorge Spelvin ( x ^= (a), \ 19592a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 19602a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 19612a18da7aSGeorge Spelvin y *= 9 ) 1962bfcfaa77SLinus Torvalds 19630fed3ac8SGeorge Spelvin /* 19642a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 19652a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 19662a18da7aSGeorge Spelvin * work done before the hash value is used. 19670fed3ac8SGeorge Spelvin */ 19682a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19690fed3ac8SGeorge Spelvin { 19702a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 19712a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19722a18da7aSGeorge Spelvin return y >> 32; 19730fed3ac8SGeorge Spelvin } 19740fed3ac8SGeorge Spelvin 1975bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1976bfcfaa77SLinus Torvalds 19772a18da7aSGeorge Spelvin /* 19782a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19792a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19802a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19812a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19822a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19832a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19842a18da7aSGeorge Spelvin * Perfect: 2048 31744 19852a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19862a18da7aSGeorge Spelvin */ 19872a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19882a18da7aSGeorge Spelvin ( x ^= (a), \ 19892a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19902a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19912a18da7aSGeorge Spelvin y *= 9 ) 1992bfcfaa77SLinus Torvalds 19932a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19940fed3ac8SGeorge Spelvin { 19952a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19962a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19970fed3ac8SGeorge Spelvin } 19980fed3ac8SGeorge Spelvin 1999bfcfaa77SLinus Torvalds #endif 2000bfcfaa77SLinus Torvalds 20012a18da7aSGeorge Spelvin /* 20022a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 20032a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 20042a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 20052a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 20062a18da7aSGeorge Spelvin * finds the delimiter after the name. 20072a18da7aSGeorge Spelvin */ 20088387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2009bfcfaa77SLinus Torvalds { 20108387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2011bfcfaa77SLinus Torvalds 2012bfcfaa77SLinus Torvalds for (;;) { 2013fcfd2fbfSGeorge Spelvin if (!len) 2014fcfd2fbfSGeorge Spelvin goto done; 2015e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2016bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2017bfcfaa77SLinus Torvalds break; 20182a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2019bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2020bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2021bfcfaa77SLinus Torvalds } 20222a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2023bfcfaa77SLinus Torvalds done: 20242a18da7aSGeorge Spelvin return fold_hash(x, y); 2025bfcfaa77SLinus Torvalds } 2026bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2027bfcfaa77SLinus Torvalds 2028fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20298387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2030fcfd2fbfSGeorge Spelvin { 20318387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 20328387ff25SLinus Torvalds unsigned long adata, mask, len; 2033fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2034fcfd2fbfSGeorge Spelvin 20358387ff25SLinus Torvalds len = 0; 20368387ff25SLinus Torvalds goto inside; 20378387ff25SLinus Torvalds 2038fcfd2fbfSGeorge Spelvin do { 20392a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2040fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 20418387ff25SLinus Torvalds inside: 2042fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2043fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2044fcfd2fbfSGeorge Spelvin 2045fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2046fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 20472a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2048fcfd2fbfSGeorge Spelvin 20492a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2050fcfd2fbfSGeorge Spelvin } 2051fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2052fcfd2fbfSGeorge Spelvin 2053bfcfaa77SLinus Torvalds /* 2054bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2055d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2056bfcfaa77SLinus Torvalds */ 20578387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2058bfcfaa77SLinus Torvalds { 20598387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 20608387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 206136126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2062bfcfaa77SLinus Torvalds 20638387ff25SLinus Torvalds len = 0; 20648387ff25SLinus Torvalds goto inside; 20658387ff25SLinus Torvalds 2066bfcfaa77SLinus Torvalds do { 20672a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2068bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 20698387ff25SLinus Torvalds inside: 2070e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 207136126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 207236126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2073bfcfaa77SLinus Torvalds 207436126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 207536126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 207636126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20772a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 207836126f8fSLinus Torvalds 20792a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2080bfcfaa77SLinus Torvalds } 2081bfcfaa77SLinus Torvalds 20822a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2083bfcfaa77SLinus Torvalds 2084fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20858387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20860145acc2SLinus Torvalds { 20878387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20880145acc2SLinus Torvalds while (len--) 2089fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20900145acc2SLinus Torvalds return end_name_hash(hash); 20910145acc2SLinus Torvalds } 2092ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20930145acc2SLinus Torvalds 2094fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20958387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2096fcfd2fbfSGeorge Spelvin { 20978387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2098fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2099fcfd2fbfSGeorge Spelvin 2100fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2101e0ab7af9SGeorge Spelvin while (c) { 2102fcfd2fbfSGeorge Spelvin len++; 2103fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2104fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2105e0ab7af9SGeorge Spelvin } 2106fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2107fcfd2fbfSGeorge Spelvin } 2108f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2109fcfd2fbfSGeorge Spelvin 21103ddcd056SLinus Torvalds /* 2111200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2112200e9ef7SLinus Torvalds * one character. 2113200e9ef7SLinus Torvalds */ 21148387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2115200e9ef7SLinus Torvalds { 21168387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2117200e9ef7SLinus Torvalds unsigned long len = 0, c; 2118200e9ef7SLinus Torvalds 2119200e9ef7SLinus Torvalds c = (unsigned char)*name; 2120200e9ef7SLinus Torvalds do { 2121200e9ef7SLinus Torvalds len++; 2122200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2123200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2124200e9ef7SLinus Torvalds } while (c && c != '/'); 2125d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2126200e9ef7SLinus Torvalds } 2127200e9ef7SLinus Torvalds 2128bfcfaa77SLinus Torvalds #endif 2129bfcfaa77SLinus Torvalds 2130200e9ef7SLinus Torvalds /* 21311da177e4SLinus Torvalds * Name resolution. 2132ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2133ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 21341da177e4SLinus Torvalds * 2135ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2136ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 21371da177e4SLinus Torvalds */ 21386de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 21391da177e4SLinus Torvalds { 21401da177e4SLinus Torvalds int err; 21411da177e4SLinus Torvalds 21429b5858e9SAl Viro if (IS_ERR(name)) 21439b5858e9SAl Viro return PTR_ERR(name); 21441da177e4SLinus Torvalds while (*name=='/') 21451da177e4SLinus Torvalds name++; 21461da177e4SLinus Torvalds if (!*name) 21479e18f10aSAl Viro return 0; 21481da177e4SLinus Torvalds 21491da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 21501da177e4SLinus Torvalds for(;;) { 215192d27016SAl Viro const char *link; 2152d6bb3e90SLinus Torvalds u64 hash_len; 2153fe479a58SAl Viro int type; 21541da177e4SLinus Torvalds 215552094c8aSAl Viro err = may_lookup(nd); 21561da177e4SLinus Torvalds if (err) 21573595e234SAl Viro return err; 21581da177e4SLinus Torvalds 21598387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 21601da177e4SLinus Torvalds 2161fe479a58SAl Viro type = LAST_NORM; 2162d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2163fe479a58SAl Viro case 2: 2164200e9ef7SLinus Torvalds if (name[1] == '.') { 2165fe479a58SAl Viro type = LAST_DOTDOT; 216616c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 216716c2cd71SAl Viro } 2168fe479a58SAl Viro break; 2169fe479a58SAl Viro case 1: 2170fe479a58SAl Viro type = LAST_DOT; 2171fe479a58SAl Viro } 21725a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21735a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 217416c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21755a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2176a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2177da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21785a202bcdSAl Viro if (err < 0) 21793595e234SAl Viro return err; 2180d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2181d6bb3e90SLinus Torvalds name = this.name; 21825a202bcdSAl Viro } 21835a202bcdSAl Viro } 2184fe479a58SAl Viro 2185d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2186d6bb3e90SLinus Torvalds nd->last.name = name; 21875f4a6a69SAl Viro nd->last_type = type; 21885f4a6a69SAl Viro 2189d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2190d6bb3e90SLinus Torvalds if (!*name) 2191bdf6cbf1SAl Viro goto OK; 2192200e9ef7SLinus Torvalds /* 2193200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2194200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2195200e9ef7SLinus Torvalds */ 2196200e9ef7SLinus Torvalds do { 2197d6bb3e90SLinus Torvalds name++; 2198d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21998620c238SAl Viro if (unlikely(!*name)) { 22008620c238SAl Viro OK: 2201368ee9baSAl Viro /* pathname body, done */ 22028620c238SAl Viro if (!nd->depth) 22038620c238SAl Viro return 0; 22048620c238SAl Viro name = nd->stack[nd->depth - 1].name; 2205368ee9baSAl Viro /* trailing symlink, done */ 22068620c238SAl Viro if (!name) 22078620c238SAl Viro return 0; 22088620c238SAl Viro /* last component of nested symlink */ 220992d27016SAl Viro link = walk_component(nd, WALK_FOLLOW); 22101c4ff1a8SAl Viro } else { 22111c4ff1a8SAl Viro /* not the last component */ 221292d27016SAl Viro link = walk_component(nd, WALK_FOLLOW | WALK_MORE); 22138620c238SAl Viro } 221492d27016SAl Viro if (unlikely(link)) { 221592d27016SAl Viro if (IS_ERR(link)) 221692d27016SAl Viro return PTR_ERR(link); 221792d27016SAl Viro /* a symlink to follow */ 2218071bf501SAl Viro nd->stack[nd->depth - 1].name = name; 221992d27016SAl Viro name = link; 22209e18f10aSAl Viro continue; 222148c8b0c5SAl Viro } 222297242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 222397242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 22244675ac39SAl Viro if (unlazy_walk(nd)) 222597242f99SAl Viro return -ECHILD; 222697242f99SAl Viro } 22273595e234SAl Viro return -ENOTDIR; 22285f4a6a69SAl Viro } 2229ce57dfc1SAl Viro } 223097242f99SAl Viro } 22311da177e4SLinus Torvalds 2232edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2233c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 223431e6b01fSNick Piggin { 2235740a1678SAleksa Sarai int error; 2236c8a53ee5SAl Viro const char *s = nd->name->name; 223731e6b01fSNick Piggin 2238c0eb027eSLinus Torvalds if (!*s) 2239c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2240edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2241edc2b1daSAl Viro rcu_read_lock(); 2242c0eb027eSLinus Torvalds 224331e6b01fSNick Piggin nd->last_type = LAST_ROOT; /* if there are only slashes... */ 2244980f3ea2SAl Viro nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT; 224531e6b01fSNick Piggin nd->depth = 0; 2246ab87f9a5SAleksa Sarai 2247ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2248ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2249ab87f9a5SAleksa Sarai smp_rmb(); 2250ab87f9a5SAleksa Sarai 22515b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2252b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2253b18825a7SDavid Howells struct inode *inode = root->d_inode; 225493893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2255368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 22565b6ca027SAl Viro nd->path = nd->root; 22575b6ca027SAl Viro nd->inode = inode; 22585b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2259ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22608f47a016SAl Viro nd->root_seq = nd->seq; 22615b6ca027SAl Viro } else { 22625b6ca027SAl Viro path_get(&nd->path); 22635b6ca027SAl Viro } 2264368ee9baSAl Viro return s; 22655b6ca027SAl Viro } 22665b6ca027SAl Viro 226731e6b01fSNick Piggin nd->root.mnt = NULL; 2268248fb5b9SAl Viro nd->path.mnt = NULL; 2269248fb5b9SAl Viro nd->path.dentry = NULL; 227031e6b01fSNick Piggin 22718db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 22728db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2273740a1678SAleksa Sarai error = nd_jump_root(nd); 2274740a1678SAleksa Sarai if (unlikely(error)) 2275740a1678SAleksa Sarai return ERR_PTR(error); 2276ef55d917SAl Viro return s; 22778db52c7eSAleksa Sarai } 22788db52c7eSAleksa Sarai 22798db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 22808db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2281e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 228231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2283c28cc364SNick Piggin unsigned seq; 228431e6b01fSNick Piggin 2285c28cc364SNick Piggin do { 2286c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 228731e6b01fSNick Piggin nd->path = fs->pwd; 2288ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2289c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2290c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2291e41f7d4eSAl Viro } else { 2292e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2293ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2294e41f7d4eSAl Viro } 229531e6b01fSNick Piggin } else { 2296582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2297c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 229831e6b01fSNick Piggin struct dentry *dentry; 229931e6b01fSNick Piggin 23002903ff01SAl Viro if (!f.file) 2301368ee9baSAl Viro return ERR_PTR(-EBADF); 230231e6b01fSNick Piggin 23032903ff01SAl Viro dentry = f.file->f_path.dentry; 230431e6b01fSNick Piggin 2305edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 23062903ff01SAl Viro fdput(f); 2307368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2308f52e0c11SAl Viro } 23092903ff01SAl Viro 23102903ff01SAl Viro nd->path = f.file->f_path; 2311e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 231234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 231334a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23145590ff0dSUlrich Drepper } else { 23152903ff01SAl Viro path_get(&nd->path); 231634a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 23171da177e4SLinus Torvalds } 231834a26b99SAl Viro fdput(f); 2319e41f7d4eSAl Viro } 23208db52c7eSAleksa Sarai 2321adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2322adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2323adb21d2bSAleksa Sarai nd->root = nd->path; 2324adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2325adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2326adb21d2bSAleksa Sarai } else { 2327adb21d2bSAleksa Sarai path_get(&nd->root); 2328adb21d2bSAleksa Sarai nd->flags |= LOOKUP_ROOT_GRABBED; 2329adb21d2bSAleksa Sarai } 2330adb21d2bSAleksa Sarai } 2331adb21d2bSAleksa Sarai return s; 23329b4a9b14SAl Viro } 23339b4a9b14SAl Viro 23341ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 233595fa25d9SAl Viro { 233692d27016SAl Viro const char *link; 2337bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2338bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2339bd92d7feSAl Viro 2340bd92d7feSAl Viro nd->flags &= ~LOOKUP_PARENT; 234192d27016SAl Viro link = walk_component(nd, 0); 234292d27016SAl Viro if (link) { 23431ccac622SAl Viro nd->flags |= LOOKUP_PARENT; 23441ccac622SAl Viro nd->stack[0].name = NULL; 23451ccac622SAl Viro } 234692d27016SAl Viro return link; 2347bd92d7feSAl Viro } 2348bd92d7feSAl Viro 23494f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 23504f757f3cSAl Viro { 2351c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2352db3c9adeSAl Viro dget(nd->path.dentry); 2353b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2354b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 23554f757f3cSAl Viro } 23564f757f3cSAl Viro 23579b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2358c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23599b4a9b14SAl Viro { 2360c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2361bd92d7feSAl Viro int err; 236231e6b01fSNick Piggin 23639b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23644f757f3cSAl Viro err = handle_lookup_down(nd); 23655f336e72SAl Viro if (unlikely(err < 0)) 23665f336e72SAl Viro s = ERR_PTR(err); 23674f757f3cSAl Viro } 23684f757f3cSAl Viro 23691ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 23701ccac622SAl Viro (s = lookup_last(nd)) != NULL) 23711ccac622SAl Viro ; 23729f1fafeeSAl Viro if (!err) 23739f1fafeeSAl Viro err = complete_walk(nd); 2374bd92d7feSAl Viro 2375deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2376deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2377bd23a539SAl Viro err = -ENOTDIR; 2378161aff1dSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 2379161aff1dSAl Viro err = handle_lookup_down(nd); 2380161aff1dSAl Viro nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... 2381161aff1dSAl Viro } 2382625b6d10SAl Viro if (!err) { 2383625b6d10SAl Viro *path = nd->path; 2384625b6d10SAl Viro nd->path.mnt = NULL; 2385625b6d10SAl Viro nd->path.dentry = NULL; 2386625b6d10SAl Viro } 2387deb106c6SAl Viro terminate_walk(nd); 2388bd92d7feSAl Viro return err; 238931e6b01fSNick Piggin } 239031e6b01fSNick Piggin 239131d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23929ad1aaa6SAl Viro struct path *path, struct path *root) 2393873f1eedSJeff Layton { 2394894bc8c4SAl Viro int retval; 23959883d185SAl Viro struct nameidata nd; 2396abc9f5beSAl Viro if (IS_ERR(name)) 2397abc9f5beSAl Viro return PTR_ERR(name); 23989ad1aaa6SAl Viro if (unlikely(root)) { 23999ad1aaa6SAl Viro nd.root = *root; 24009ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 24019ad1aaa6SAl Viro } 24029883d185SAl Viro set_nameidata(&nd, dfd, name); 2403c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2404873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2405c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2406873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2407c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2408873f1eedSJeff Layton 2409873f1eedSJeff Layton if (likely(!retval)) 2410161aff1dSAl Viro audit_inode(name, path->dentry, 2411161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 24129883d185SAl Viro restore_nameidata(); 2413e4bd1c1aSAl Viro putname(name); 2414873f1eedSJeff Layton return retval; 2415873f1eedSJeff Layton } 2416873f1eedSJeff Layton 24178bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2418c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2419391172c4SAl Viro struct path *parent) 24208bcb77faSAl Viro { 2421c8a53ee5SAl Viro const char *s = path_init(nd, flags); 24229b5858e9SAl Viro int err = link_path_walk(s, nd); 24238bcb77faSAl Viro if (!err) 24248bcb77faSAl Viro err = complete_walk(nd); 2425391172c4SAl Viro if (!err) { 2426391172c4SAl Viro *parent = nd->path; 2427391172c4SAl Viro nd->path.mnt = NULL; 2428391172c4SAl Viro nd->path.dentry = NULL; 2429391172c4SAl Viro } 2430deb106c6SAl Viro terminate_walk(nd); 24318bcb77faSAl Viro return err; 24328bcb77faSAl Viro } 24338bcb77faSAl Viro 24345c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2435391172c4SAl Viro unsigned int flags, struct path *parent, 2436391172c4SAl Viro struct qstr *last, int *type) 24378bcb77faSAl Viro { 24388bcb77faSAl Viro int retval; 24399883d185SAl Viro struct nameidata nd; 24408bcb77faSAl Viro 24415c31b6ceSAl Viro if (IS_ERR(name)) 24425c31b6ceSAl Viro return name; 24439883d185SAl Viro set_nameidata(&nd, dfd, name); 2444c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 24458bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2446c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 24478bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2448c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2449391172c4SAl Viro if (likely(!retval)) { 2450391172c4SAl Viro *last = nd.last; 2451391172c4SAl Viro *type = nd.last_type; 2452c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 24535c31b6ceSAl Viro } else { 24545c31b6ceSAl Viro putname(name); 24555c31b6ceSAl Viro name = ERR_PTR(retval); 2456391172c4SAl Viro } 24579883d185SAl Viro restore_nameidata(); 24585c31b6ceSAl Viro return name; 24598bcb77faSAl Viro } 24608bcb77faSAl Viro 246179714f72SAl Viro /* does lookup, returns the object with parent locked */ 246279714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24635590ff0dSUlrich Drepper { 24645c31b6ceSAl Viro struct filename *filename; 24655c31b6ceSAl Viro struct dentry *d; 2466391172c4SAl Viro struct qstr last; 2467391172c4SAl Viro int type; 246851689104SPaul Moore 24695c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24705c31b6ceSAl Viro &last, &type); 247151689104SPaul Moore if (IS_ERR(filename)) 247251689104SPaul Moore return ERR_CAST(filename); 24735c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2474391172c4SAl Viro path_put(path); 24755c31b6ceSAl Viro putname(filename); 24765c31b6ceSAl Viro return ERR_PTR(-EINVAL); 247779714f72SAl Viro } 24785955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2479391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 248079714f72SAl Viro if (IS_ERR(d)) { 24815955102cSAl Viro inode_unlock(path->dentry->d_inode); 2482391172c4SAl Viro path_put(path); 248379714f72SAl Viro } 248451689104SPaul Moore putname(filename); 248579714f72SAl Viro return d; 24865590ff0dSUlrich Drepper } 24875590ff0dSUlrich Drepper 2488d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2489d1811465SAl Viro { 2490abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2491abc9f5beSAl Viro flags, path, NULL); 2492d1811465SAl Viro } 24934d359507SAl Viro EXPORT_SYMBOL(kern_path); 2494d1811465SAl Viro 249516f18200SJosef 'Jeff' Sipek /** 249616f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 249716f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 249816f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 249916f18200SJosef 'Jeff' Sipek * @name: pointer to file name 250016f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2501e0a01249SAl Viro * @path: pointer to struct path to fill 250216f18200SJosef 'Jeff' Sipek */ 250316f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 250416f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2505e0a01249SAl Viro struct path *path) 250616f18200SJosef 'Jeff' Sipek { 25079ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 25089ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2509abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2510abc9f5beSAl Viro flags , path, &root); 251116f18200SJosef 'Jeff' Sipek } 25124d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 251316f18200SJosef 'Jeff' Sipek 25143c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 25153c95f0dcSAl Viro int len, struct qstr *this) 25163c95f0dcSAl Viro { 25173c95f0dcSAl Viro this->name = name; 25183c95f0dcSAl Viro this->len = len; 25193c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 25203c95f0dcSAl Viro if (!len) 25213c95f0dcSAl Viro return -EACCES; 25223c95f0dcSAl Viro 25233c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 25243c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 25253c95f0dcSAl Viro return -EACCES; 25263c95f0dcSAl Viro } 25273c95f0dcSAl Viro 25283c95f0dcSAl Viro while (len--) { 25293c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 25303c95f0dcSAl Viro if (c == '/' || c == '\0') 25313c95f0dcSAl Viro return -EACCES; 25323c95f0dcSAl Viro } 25333c95f0dcSAl Viro /* 25343c95f0dcSAl Viro * See if the low-level filesystem might want 25353c95f0dcSAl Viro * to use its own hash.. 25363c95f0dcSAl Viro */ 25373c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 25383c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 25393c95f0dcSAl Viro if (err < 0) 25403c95f0dcSAl Viro return err; 25413c95f0dcSAl Viro } 25423c95f0dcSAl Viro 25433c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 25443c95f0dcSAl Viro } 25453c95f0dcSAl Viro 2546eead1911SChristoph Hellwig /** 25470da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 25480da0b7fdSDavid Howells * @name: pathname component to lookup 25490da0b7fdSDavid Howells * @base: base directory to lookup from 25500da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 25510da0b7fdSDavid Howells * 25520da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 25530da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 25540da0b7fdSDavid Howells * 25550da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25560da0b7fdSDavid Howells * not be called by generic code. 25570da0b7fdSDavid Howells * 25580da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25590da0b7fdSDavid Howells */ 25600da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25610da0b7fdSDavid Howells { 25620da0b7fdSDavid Howells struct qstr this; 25630da0b7fdSDavid Howells int err; 25640da0b7fdSDavid Howells 25650da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25660da0b7fdSDavid Howells 25670da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25680da0b7fdSDavid Howells if (err) 25690da0b7fdSDavid Howells return ERR_PTR(err); 25700da0b7fdSDavid Howells 25710da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25720da0b7fdSDavid Howells } 25730da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25740da0b7fdSDavid Howells 25750da0b7fdSDavid Howells /** 2576a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2577eead1911SChristoph Hellwig * @name: pathname component to lookup 2578eead1911SChristoph Hellwig * @base: base directory to lookup from 2579eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2580eead1911SChristoph Hellwig * 2581a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25829e7543e9SAl Viro * not be called by generic code. 2583bbddca8eSNeilBrown * 2584bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2585eead1911SChristoph Hellwig */ 2586057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2587057f6c01SJames Morris { 25888613a209SAl Viro struct dentry *dentry; 2589057f6c01SJames Morris struct qstr this; 2590cda309deSMiklos Szeredi int err; 2591057f6c01SJames Morris 25925955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25932f9092e1SDavid Woodhouse 25943c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2595cda309deSMiklos Szeredi if (err) 2596cda309deSMiklos Szeredi return ERR_PTR(err); 2597cda309deSMiklos Szeredi 25988613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25998613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2600057f6c01SJames Morris } 26014d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2602057f6c01SJames Morris 2603bbddca8eSNeilBrown /** 2604bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2605bbddca8eSNeilBrown * @name: pathname component to lookup 2606bbddca8eSNeilBrown * @base: base directory to lookup from 2607bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2608bbddca8eSNeilBrown * 2609bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2610bbddca8eSNeilBrown * not be called by generic code. 2611bbddca8eSNeilBrown * 2612bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2613bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2614bbddca8eSNeilBrown */ 2615bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2616bbddca8eSNeilBrown struct dentry *base, int len) 2617bbddca8eSNeilBrown { 2618bbddca8eSNeilBrown struct qstr this; 2619bbddca8eSNeilBrown int err; 262020d00ee8SLinus Torvalds struct dentry *ret; 2621bbddca8eSNeilBrown 26223c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2623bbddca8eSNeilBrown if (err) 2624bbddca8eSNeilBrown return ERR_PTR(err); 2625bbddca8eSNeilBrown 262620d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 262720d00ee8SLinus Torvalds if (!ret) 262820d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 262920d00ee8SLinus Torvalds return ret; 2630bbddca8eSNeilBrown } 2631bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2632bbddca8eSNeilBrown 26336c2d4798SAl Viro /* 26346c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 26356c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 26366c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 26376c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 26386c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 26396c2d4798SAl Viro * this one avoids such problems. 26406c2d4798SAl Viro */ 26416c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 26426c2d4798SAl Viro struct dentry *base, int len) 26436c2d4798SAl Viro { 26446c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 26452fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 26466c2d4798SAl Viro dput(ret); 26476c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 26486c2d4798SAl Viro } 26496c2d4798SAl Viro return ret; 26506c2d4798SAl Viro } 26516c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 26526c2d4798SAl Viro 2653eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2654eedf265aSEric W. Biederman int path_pts(struct path *path) 2655eedf265aSEric W. Biederman { 2656eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2657eedf265aSEric W. Biederman * the input path. 2658eedf265aSEric W. Biederman */ 2659eedf265aSEric W. Biederman struct dentry *child, *parent; 2660eedf265aSEric W. Biederman struct qstr this; 2661eedf265aSEric W. Biederman int ret; 2662eedf265aSEric W. Biederman 2663eedf265aSEric W. Biederman ret = path_parent_directory(path); 2664eedf265aSEric W. Biederman if (ret) 2665eedf265aSEric W. Biederman return ret; 2666eedf265aSEric W. Biederman 2667eedf265aSEric W. Biederman parent = path->dentry; 2668eedf265aSEric W. Biederman this.name = "pts"; 2669eedf265aSEric W. Biederman this.len = 3; 2670eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2671eedf265aSEric W. Biederman if (!child) 2672eedf265aSEric W. Biederman return -ENOENT; 2673eedf265aSEric W. Biederman 2674eedf265aSEric W. Biederman path->dentry = child; 2675eedf265aSEric W. Biederman dput(parent); 2676eedf265aSEric W. Biederman follow_mount(path); 2677eedf265aSEric W. Biederman return 0; 2678eedf265aSEric W. Biederman } 2679eedf265aSEric W. Biederman #endif 2680eedf265aSEric W. Biederman 26811fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26821fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26831da177e4SLinus Torvalds { 2684abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2685abc9f5beSAl Viro flags, path, NULL); 26861da177e4SLinus Torvalds } 2687b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26881fa1e7f6SAndy Whitcroft 2689cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 26901da177e4SLinus Torvalds { 26918e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2692da9592edSDavid Howells 26938e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 26941da177e4SLinus Torvalds return 0; 26958e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 26961da177e4SLinus Torvalds return 0; 269723adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 26981da177e4SLinus Torvalds } 2699cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27001da177e4SLinus Torvalds 27011da177e4SLinus Torvalds /* 27021da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27031da177e4SLinus Torvalds * whether the type of victim is right. 27041da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27051da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27061da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27071da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27081da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27091da177e4SLinus Torvalds * a. be owner of dir, or 27101da177e4SLinus Torvalds * b. be owner of victim, or 27111da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27121da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27131da177e4SLinus Torvalds * links pointing to it. 27140bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27150bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27160bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27170bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27180bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27191da177e4SLinus Torvalds * nfs_async_unlink(). 27201da177e4SLinus Torvalds */ 2721b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 27221da177e4SLinus Torvalds { 272363afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27241da177e4SLinus Torvalds int error; 27251da177e4SLinus Torvalds 2726b18825a7SDavid Howells if (d_is_negative(victim)) 27271da177e4SLinus Torvalds return -ENOENT; 2728b18825a7SDavid Howells BUG_ON(!inode); 27291da177e4SLinus Torvalds 27301da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2731593d1ce8SEric W. Biederman 2732593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2733593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2734593d1ce8SEric W. Biederman return -EOVERFLOW; 2735593d1ce8SEric W. Biederman 27364fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27371da177e4SLinus Torvalds 2738f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27391da177e4SLinus Torvalds if (error) 27401da177e4SLinus Torvalds return error; 27411da177e4SLinus Torvalds if (IS_APPEND(dir)) 27421da177e4SLinus Torvalds return -EPERM; 2743b18825a7SDavid Howells 2744b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27450bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27461da177e4SLinus Torvalds return -EPERM; 27471da177e4SLinus Torvalds if (isdir) { 274844b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27491da177e4SLinus Torvalds return -ENOTDIR; 27501da177e4SLinus Torvalds if (IS_ROOT(victim)) 27511da177e4SLinus Torvalds return -EBUSY; 275244b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 27531da177e4SLinus Torvalds return -EISDIR; 27541da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27551da177e4SLinus Torvalds return -ENOENT; 27561da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27571da177e4SLinus Torvalds return -EBUSY; 27581da177e4SLinus Torvalds return 0; 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds 27611da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27621da177e4SLinus Torvalds * dir. 27631da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27641da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27651da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2766036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2767036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2768036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27691da177e4SLinus Torvalds */ 2770a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 27711da177e4SLinus Torvalds { 2772036d5236SEric W. Biederman struct user_namespace *s_user_ns; 277314e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 27741da177e4SLinus Torvalds if (child->d_inode) 27751da177e4SLinus Torvalds return -EEXIST; 27761da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27771da177e4SLinus Torvalds return -ENOENT; 2778036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2779036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2780036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2781036d5236SEric W. Biederman return -EOVERFLOW; 2782f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 27831da177e4SLinus Torvalds } 27841da177e4SLinus Torvalds 27851da177e4SLinus Torvalds /* 27861da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 27871da177e4SLinus Torvalds */ 27881da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 27891da177e4SLinus Torvalds { 27901da177e4SLinus Torvalds struct dentry *p; 27911da177e4SLinus Torvalds 27921da177e4SLinus Torvalds if (p1 == p2) { 27935955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27941da177e4SLinus Torvalds return NULL; 27951da177e4SLinus Torvalds } 27961da177e4SLinus Torvalds 2797fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 27981da177e4SLinus Torvalds 2799e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2800e2761a11SOGAWA Hirofumi if (p) { 28015955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28025955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28031da177e4SLinus Torvalds return p; 28041da177e4SLinus Torvalds } 28051da177e4SLinus Torvalds 2806e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2807e2761a11SOGAWA Hirofumi if (p) { 28085955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28095955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28101da177e4SLinus Torvalds return p; 28111da177e4SLinus Torvalds } 28121da177e4SLinus Torvalds 28135955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28145955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28151da177e4SLinus Torvalds return NULL; 28161da177e4SLinus Torvalds } 28174d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28181da177e4SLinus Torvalds 28191da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28201da177e4SLinus Torvalds { 28215955102cSAl Viro inode_unlock(p1->d_inode); 28221da177e4SLinus Torvalds if (p1 != p2) { 28235955102cSAl Viro inode_unlock(p2->d_inode); 2824fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28251da177e4SLinus Torvalds } 28261da177e4SLinus Torvalds } 28274d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28281da177e4SLinus Torvalds 28294acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2830312b63fbSAl Viro bool want_excl) 28311da177e4SLinus Torvalds { 2832a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28331da177e4SLinus Torvalds if (error) 28341da177e4SLinus Torvalds return error; 28351da177e4SLinus Torvalds 2836acfa4380SAl Viro if (!dir->i_op->create) 28371da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28381da177e4SLinus Torvalds mode &= S_IALLUGO; 28391da177e4SLinus Torvalds mode |= S_IFREG; 28401da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28411da177e4SLinus Torvalds if (error) 28421da177e4SLinus Torvalds return error; 2843312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2844a74574aaSStephen Smalley if (!error) 2845f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28461da177e4SLinus Torvalds return error; 28471da177e4SLinus Torvalds } 28484d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28491da177e4SLinus Torvalds 28508e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 28518e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 28528e6c848eSAl Viro void *arg) 28538e6c848eSAl Viro { 28548e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 28558e6c848eSAl Viro int error = may_create(dir, dentry); 28568e6c848eSAl Viro if (error) 28578e6c848eSAl Viro return error; 28588e6c848eSAl Viro 28598e6c848eSAl Viro mode &= S_IALLUGO; 28608e6c848eSAl Viro mode |= S_IFREG; 28618e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 28628e6c848eSAl Viro if (error) 28638e6c848eSAl Viro return error; 28648e6c848eSAl Viro error = f(dentry, mode, arg); 28658e6c848eSAl Viro if (!error) 28668e6c848eSAl Viro fsnotify_create(dir, dentry); 28678e6c848eSAl Viro return error; 28688e6c848eSAl Viro } 28698e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 28708e6c848eSAl Viro 2871a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2872a2982cc9SEric W. Biederman { 2873a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2874a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2875a2982cc9SEric W. Biederman } 2876a2982cc9SEric W. Biederman 2877f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 28781da177e4SLinus Torvalds { 28793fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 28801da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 28811da177e4SLinus Torvalds int error; 28821da177e4SLinus Torvalds 28831da177e4SLinus Torvalds if (!inode) 28841da177e4SLinus Torvalds return -ENOENT; 28851da177e4SLinus Torvalds 2886c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2887c8fe8f30SChristoph Hellwig case S_IFLNK: 28881da177e4SLinus Torvalds return -ELOOP; 2889c8fe8f30SChristoph Hellwig case S_IFDIR: 2890c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 28911da177e4SLinus Torvalds return -EISDIR; 2892c8fe8f30SChristoph Hellwig break; 2893c8fe8f30SChristoph Hellwig case S_IFBLK: 2894c8fe8f30SChristoph Hellwig case S_IFCHR: 2895a2982cc9SEric W. Biederman if (!may_open_dev(path)) 28961da177e4SLinus Torvalds return -EACCES; 2897c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2898c8fe8f30SChristoph Hellwig case S_IFIFO: 2899c8fe8f30SChristoph Hellwig case S_IFSOCK: 29001da177e4SLinus Torvalds flag &= ~O_TRUNC; 2901c8fe8f30SChristoph Hellwig break; 29024a3fd211SDave Hansen } 2903b41572e9SDave Hansen 290462fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2905b41572e9SDave Hansen if (error) 2906b41572e9SDave Hansen return error; 29076146f0d5SMimi Zohar 29081da177e4SLinus Torvalds /* 29091da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29101da177e4SLinus Torvalds */ 29111da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29128737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29137715b521SAl Viro return -EPERM; 29141da177e4SLinus Torvalds if (flag & O_TRUNC) 29157715b521SAl Viro return -EPERM; 29161da177e4SLinus Torvalds } 29171da177e4SLinus Torvalds 29181da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 29192e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 29207715b521SAl Viro return -EPERM; 29211da177e4SLinus Torvalds 2922f3c7691eSJ. Bruce Fields return 0; 29237715b521SAl Viro } 29247715b521SAl Viro 2925e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29267715b521SAl Viro { 2927f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29287715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29297715b521SAl Viro int error = get_write_access(inode); 29301da177e4SLinus Torvalds if (error) 29317715b521SAl Viro return error; 29321da177e4SLinus Torvalds /* 29331da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29341da177e4SLinus Torvalds */ 2935d7a06983SJeff Layton error = locks_verify_locked(filp); 2936be6d3e56SKentaro Takeda if (!error) 2937ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29381da177e4SLinus Torvalds if (!error) { 29397715b521SAl Viro error = do_truncate(path->dentry, 0, 2940d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2941e1181ee6SJeff Layton filp); 29421da177e4SLinus Torvalds } 29431da177e4SLinus Torvalds put_write_access(inode); 2944acd0c935SMimi Zohar return error; 29451da177e4SLinus Torvalds } 29461da177e4SLinus Torvalds 2947d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2948d57999e1SDave Hansen { 29498a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29508a5e929dSAl Viro flag--; 2951d57999e1SDave Hansen return flag; 2952d57999e1SDave Hansen } 2953d57999e1SDave Hansen 2954d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2955d18e9008SMiklos Szeredi { 29561328c727SSeth Forshee struct user_namespace *s_user_ns; 2957d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2958d18e9008SMiklos Szeredi if (error) 2959d18e9008SMiklos Szeredi return error; 2960d18e9008SMiklos Szeredi 29611328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29621328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29631328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29641328c727SSeth Forshee return -EOVERFLOW; 29651328c727SSeth Forshee 2966d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2967d18e9008SMiklos Szeredi if (error) 2968d18e9008SMiklos Szeredi return error; 2969d18e9008SMiklos Szeredi 2970d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2971d18e9008SMiklos Szeredi } 2972d18e9008SMiklos Szeredi 29731acf0af9SDavid Howells /* 29741acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29751acf0af9SDavid Howells * dentry. 29761acf0af9SDavid Howells * 29771acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29781acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29791acf0af9SDavid Howells * 298000a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 298100a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 298200a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 29831acf0af9SDavid Howells * 29841acf0af9SDavid Howells * Returns an error code otherwise. 29851acf0af9SDavid Howells */ 2986239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 2987239eb983SAl Viro struct file *file, 2988015c3bbcSMiklos Szeredi const struct open_flags *op, 29893ec2eef1SAl Viro int open_flag, umode_t mode) 2990d18e9008SMiklos Szeredi { 2991d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2992d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2993d18e9008SMiklos Szeredi int error; 2994d18e9008SMiklos Szeredi 2995384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 2996d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2997d18e9008SMiklos Szeredi 2998d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2999d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3000d18e9008SMiklos Szeredi 300130d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 300230d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30030fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 300444907d79SAl Viro open_to_namei_flags(open_flag), mode); 30056fbd0714SAl Viro d_lookup_done(dentry); 3006384f26e2SAl Viro if (!error) { 300764e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 3008d18e9008SMiklos Szeredi /* 3009384f26e2SAl Viro * We didn't have the inode before the open, so check open 3010384f26e2SAl Viro * permission here. 3011d18e9008SMiklos Szeredi */ 3012384f26e2SAl Viro int acc_mode = op->acc_mode; 301373a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 301403da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 301503da633aSAl Viro fsnotify_create(dir, dentry); 301662fb4a15SAl Viro acc_mode = 0; 301703da633aSAl Viro } 30182675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 3019384f26e2SAl Viro if (WARN_ON(error > 0)) 3020384f26e2SAl Viro error = -EINVAL; 302164e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 3022d18e9008SMiklos Szeredi error = -EIO; 3023384f26e2SAl Viro } else { 3024d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3025d18e9008SMiklos Szeredi dput(dentry); 3026d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 302710c64ceaSAl Viro } 302873a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 3029d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3030239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3031a01e718fSAl Viro error = -ENOENT; 3032d18e9008SMiklos Szeredi } 3033d18e9008SMiklos Szeredi } 3034239eb983SAl Viro if (error) { 3035d18e9008SMiklos Szeredi dput(dentry); 3036239eb983SAl Viro dentry = ERR_PTR(error); 3037239eb983SAl Viro } 3038239eb983SAl Viro return dentry; 3039d18e9008SMiklos Szeredi } 3040d18e9008SMiklos Szeredi 304131e6b01fSNick Piggin /* 30421acf0af9SDavid Howells * Look up and maybe create and open the last component. 3043d58ffd35SMiklos Szeredi * 304400a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3045d58ffd35SMiklos Szeredi * 304600a07c15SAl Viro * Returns 0 on success, that is, if 304700a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 304800a07c15SAl Viro * the file was not completely opened at this time, though lookups and 304900a07c15SAl Viro * creations were performed. 305000a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 305100a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 305200a07c15SAl Viro * hadn't been specified. 30531acf0af9SDavid Howells * 305400a07c15SAl Viro * An error code is returned on failure. 3055d58ffd35SMiklos Szeredi */ 3056da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3057d58ffd35SMiklos Szeredi const struct open_flags *op, 30583ec2eef1SAl Viro bool got_write) 3059d58ffd35SMiklos Szeredi { 3060d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 306154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30621643b43fSAl Viro int open_flag = op->open_flag; 3063d58ffd35SMiklos Szeredi struct dentry *dentry; 30641643b43fSAl Viro int error, create_error = 0; 30651643b43fSAl Viro umode_t mode = op->mode; 30666fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3067d58ffd35SMiklos Szeredi 3068ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3069da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3070d58ffd35SMiklos Szeredi 307173a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 30726fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30736fbd0714SAl Viro for (;;) { 30746fbd0714SAl Viro if (!dentry) { 30756fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3076d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3077da5ebf5aSAl Viro return dentry; 30786fbd0714SAl Viro } 30796fbd0714SAl Viro if (d_in_lookup(dentry)) 30806fbd0714SAl Viro break; 3081d58ffd35SMiklos Szeredi 30826fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 30836fbd0714SAl Viro if (likely(error > 0)) 30846fbd0714SAl Viro break; 30856fbd0714SAl Viro if (error) 30866fbd0714SAl Viro goto out_dput; 30876fbd0714SAl Viro d_invalidate(dentry); 30886fbd0714SAl Viro dput(dentry); 30896fbd0714SAl Viro dentry = NULL; 30906fbd0714SAl Viro } 30916fbd0714SAl Viro if (dentry->d_inode) { 3092d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3093da5ebf5aSAl Viro return dentry; 30946c51e513SAl Viro } 3095d18e9008SMiklos Szeredi 30961643b43fSAl Viro /* 30971643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 30981643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 30991643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31001643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31011643b43fSAl Viro * 31021643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31031643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31041643b43fSAl Viro */ 31051643b43fSAl Viro if (open_flag & O_CREAT) { 31061643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31071643b43fSAl Viro mode &= ~current_umask(); 31081643b43fSAl Viro if (unlikely(!got_write)) { 31091643b43fSAl Viro create_error = -EROFS; 31101643b43fSAl Viro open_flag &= ~O_CREAT; 31111643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 31121643b43fSAl Viro goto no_open; 31131643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 31141643b43fSAl Viro } else { 31151643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 31161643b43fSAl Viro if (create_error) { 31171643b43fSAl Viro open_flag &= ~O_CREAT; 31181643b43fSAl Viro if (open_flag & O_EXCL) 31191643b43fSAl Viro goto no_open; 31201643b43fSAl Viro } 31211643b43fSAl Viro } 31221643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31231643b43fSAl Viro unlikely(!got_write)) { 31241643b43fSAl Viro /* 31251643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31261643b43fSAl Viro * back to lookup + open 31271643b43fSAl Viro */ 31281643b43fSAl Viro goto no_open; 3129d18e9008SMiklos Szeredi } 3130d18e9008SMiklos Szeredi 31311643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3132239eb983SAl Viro dentry = atomic_open(nd, dentry, file, op, open_flag, mode); 3133da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3134da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3135da5ebf5aSAl Viro return dentry; 3136239eb983SAl Viro } 313754ef4872SMiklos Szeredi 31381643b43fSAl Viro no_open: 31396fbd0714SAl Viro if (d_in_lookup(dentry)) { 314012fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 314112fa5e24SAl Viro nd->flags); 31426fbd0714SAl Viro d_lookup_done(dentry); 314312fa5e24SAl Viro if (unlikely(res)) { 314412fa5e24SAl Viro if (IS_ERR(res)) { 314512fa5e24SAl Viro error = PTR_ERR(res); 314612fa5e24SAl Viro goto out_dput; 314712fa5e24SAl Viro } 314812fa5e24SAl Viro dput(dentry); 314912fa5e24SAl Viro dentry = res; 315012fa5e24SAl Viro } 315154ef4872SMiklos Szeredi } 315254ef4872SMiklos Szeredi 3153d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 31541643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 315573a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3156ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3157ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3158ce8644fcSAl Viro error = -EACCES; 3159d58ffd35SMiklos Szeredi goto out_dput; 316064894cf8SAl Viro } 3161ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31621643b43fSAl Viro open_flag & O_EXCL); 3163d58ffd35SMiklos Szeredi if (error) 3164d58ffd35SMiklos Szeredi goto out_dput; 3165ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3166d58ffd35SMiklos Szeredi } 31671643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31681643b43fSAl Viro error = create_error; 3169d58ffd35SMiklos Szeredi goto out_dput; 3170d58ffd35SMiklos Szeredi } 3171da5ebf5aSAl Viro return dentry; 3172d58ffd35SMiklos Szeredi 3173d58ffd35SMiklos Szeredi out_dput: 3174d58ffd35SMiklos Szeredi dput(dentry); 3175da5ebf5aSAl Viro return ERR_PTR(error); 3176d58ffd35SMiklos Szeredi } 3177d58ffd35SMiklos Szeredi 3178d58ffd35SMiklos Szeredi /* 3179fe2d35ffSAl Viro * Handle the last step of open() 318031e6b01fSNick Piggin */ 31811ccac622SAl Viro static const char *do_last(struct nameidata *nd, 31823ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3183fb1cc555SAl Viro { 3184a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 31856404674aSAl Viro kuid_t dir_uid = nd->inode->i_uid; 31866404674aSAl Viro umode_t dir_mode = nd->inode->i_mode; 3187ca344a89SAl Viro int open_flag = op->open_flag; 318877d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 318964894cf8SAl Viro bool got_write = false; 3190bcda7652SAl Viro int acc_mode = op->acc_mode; 3191254cf582SAl Viro unsigned seq; 3192a1eb3315SMiklos Szeredi struct inode *inode; 3193da5ebf5aSAl Viro struct dentry *dentry; 3194b0417d2cSAl Viro const char *res; 319516c2cd71SAl Viro int error; 3196fb1cc555SAl Viro 3197c3e380b0SAl Viro nd->flags &= ~LOOKUP_PARENT; 3198c3e380b0SAl Viro nd->flags |= op->intent; 3199c3e380b0SAl Viro 3200bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 320156676ec3SAl Viro if (nd->depth) 320256676ec3SAl Viro put_link(nd); 3203fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3204deb106c6SAl Viro if (unlikely(error)) 32051ccac622SAl Viro return ERR_PTR(error); 3206e83db167SMiklos Szeredi goto finish_open; 32071f36f774SAl Viro } 3208a2c36b45SAl Viro 3209ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3210fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3211fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3212fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 321320e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 321420e34357SAl Viro if (IS_ERR(dentry)) 32151ccac622SAl Viro return ERR_CAST(dentry); 321620e34357SAl Viro if (likely(dentry)) 321771574865SMiklos Szeredi goto finish_lookup; 321871574865SMiklos Szeredi 321937d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 32206583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3221b6183df7SMiklos Szeredi } else { 3222fe2d35ffSAl Viro /* create side of things */ 3223a3fbbde7SAl Viro /* 3224b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3225b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3226b6183df7SMiklos Szeredi * about to look up 3227a3fbbde7SAl Viro */ 32289f1fafeeSAl Viro error = complete_walk(nd); 3229e8bb73dfSAl Viro if (error) 32301ccac622SAl Viro return ERR_PTR(error); 3231fe2d35ffSAl Viro 3232c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 32331f36f774SAl Viro /* trailing slashes? */ 3234deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 32351ccac622SAl Viro return ERR_PTR(-EISDIR); 3236b6183df7SMiklos Szeredi } 32371f36f774SAl Viro 32389cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 323964894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 324064894cf8SAl Viro if (!error) 324164894cf8SAl Viro got_write = true; 324264894cf8SAl Viro /* 324364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 324464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 324564894cf8SAl Viro * dropping this one anyway. 324664894cf8SAl Viro */ 324764894cf8SAl Viro } 32489cf843e3SAl Viro if (open_flag & O_CREAT) 32495955102cSAl Viro inode_lock(dir->d_inode); 32509cf843e3SAl Viro else 32519cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3252da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 32539cf843e3SAl Viro if (open_flag & O_CREAT) 32545955102cSAl Viro inode_unlock(dir->d_inode); 32559cf843e3SAl Viro else 32569cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3257fb1cc555SAl Viro 3258da5ebf5aSAl Viro if (IS_ERR(dentry)) { 3259da5ebf5aSAl Viro error = PTR_ERR(dentry); 3260d58ffd35SMiklos Szeredi goto out; 3261da5ebf5aSAl Viro } 32626c0d46c4SAl Viro 326300a07c15SAl Viro if (file->f_mode & FMODE_OPENED) { 326473a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3265496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 326677d660a8SMiklos Szeredi will_truncate = false; 3267d18e9008SMiklos Szeredi 326876ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3269da5ebf5aSAl Viro dput(dentry); 3270d18e9008SMiklos Szeredi goto opened; 3271d18e9008SMiklos Szeredi } 3272d18e9008SMiklos Szeredi 327373a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 32749b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3275ca344a89SAl Viro open_flag &= ~O_TRUNC; 327677d660a8SMiklos Szeredi will_truncate = false; 327762fb4a15SAl Viro acc_mode = 0; 3278e73cabffSAl Viro dput(nd->path.dentry); 3279e73cabffSAl Viro nd->path.dentry = dentry; 3280e83db167SMiklos Szeredi goto finish_open_created; 3281fb1cc555SAl Viro } 3282fb1cc555SAl Viro 3283fb1cc555SAl Viro /* 3284d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3285d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3286d18e9008SMiklos Szeredi * necessary...) 3287d18e9008SMiklos Szeredi */ 328864894cf8SAl Viro if (got_write) { 3289d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 329064894cf8SAl Viro got_write = false; 3291d18e9008SMiklos Szeredi } 3292d18e9008SMiklos Szeredi 329320e34357SAl Viro finish_lookup: 329456676ec3SAl Viro if (nd->depth) 329556676ec3SAl Viro put_link(nd); 3296b0417d2cSAl Viro res = step_into(nd, 0, dentry, inode, seq); 3297b0417d2cSAl Viro if (unlikely(res)) { 32981ccac622SAl Viro nd->flags |= LOOKUP_PARENT; 32991ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 33001ccac622SAl Viro nd->stack[0].name = NULL; 3301b0417d2cSAl Viro return res; 33021ccac622SAl Viro } 330331d1726dSAl Viro 330431d1726dSAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 330531d1726dSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 33061ccac622SAl Viro return ERR_PTR(-EEXIST); 330731d1726dSAl Viro } 3308bc77daa7SAl Viro finish_open: 33098f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3310a3fbbde7SAl Viro error = complete_walk(nd); 3311fac7d191SAl Viro if (error) 33121ccac622SAl Viro return ERR_PTR(error); 331376ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 331430aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3315fb1cc555SAl Viro error = -EISDIR; 331630aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 33172675a4ebSAl Viro goto out; 3318d0cb5018SAl Viro error = may_create_in_sticky(dir_mode, dir_uid, 331930aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 332030aba665SSalvatore Mesoraca if (unlikely(error)) 332130aba665SSalvatore Mesoraca goto out; 332230aba665SSalvatore Mesoraca } 3323af2f5542SMiklos Szeredi error = -ENOTDIR; 332444b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 33252675a4ebSAl Viro goto out; 33264bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 332777d660a8SMiklos Szeredi will_truncate = false; 33286c0d46c4SAl Viro 33290f9d1a10SAl Viro if (will_truncate) { 33300f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33310f9d1a10SAl Viro if (error) 33322675a4ebSAl Viro goto out; 333364894cf8SAl Viro got_write = true; 33340f9d1a10SAl Viro } 3335e83db167SMiklos Szeredi finish_open_created: 3336bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3337ca344a89SAl Viro if (error) 33382675a4ebSAl Viro goto out; 3339aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3340ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3341fac7d191SAl Viro if (error) 3342015c3bbcSMiklos Szeredi goto out; 3343a8277b9bSMiklos Szeredi opened: 33446035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3345fe9ec829SAl Viro if (!error && will_truncate) 33462675a4ebSAl Viro error = handle_truncate(file); 3347ca344a89SAl Viro out: 3348c80567c8SAl Viro if (unlikely(error > 0)) { 3349c80567c8SAl Viro WARN_ON(1); 3350c80567c8SAl Viro error = -EINVAL; 3351c80567c8SAl Viro } 335264894cf8SAl Viro if (got_write) 33530f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 33541ccac622SAl Viro return ERR_PTR(error); 3355fb1cc555SAl Viro } 3356fb1cc555SAl Viro 3357af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3358af7bd4dcSAmir Goldstein { 3359af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3360af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3361af7bd4dcSAmir Goldstein struct inode *inode; 3362af7bd4dcSAmir Goldstein int error; 3363af7bd4dcSAmir Goldstein 3364af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3365af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3366af7bd4dcSAmir Goldstein if (error) 3367af7bd4dcSAmir Goldstein goto out_err; 3368af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3369af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3370af7bd4dcSAmir Goldstein goto out_err; 3371af7bd4dcSAmir Goldstein error = -ENOMEM; 3372cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3373af7bd4dcSAmir Goldstein if (unlikely(!child)) 3374af7bd4dcSAmir Goldstein goto out_err; 3375af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3376af7bd4dcSAmir Goldstein if (error) 3377af7bd4dcSAmir Goldstein goto out_err; 3378af7bd4dcSAmir Goldstein error = -ENOENT; 3379af7bd4dcSAmir Goldstein inode = child->d_inode; 3380af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3381af7bd4dcSAmir Goldstein goto out_err; 3382af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3383af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3384af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3385af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3386af7bd4dcSAmir Goldstein } 3387fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3388af7bd4dcSAmir Goldstein return child; 3389af7bd4dcSAmir Goldstein 3390af7bd4dcSAmir Goldstein out_err: 3391af7bd4dcSAmir Goldstein dput(child); 3392af7bd4dcSAmir Goldstein return ERR_PTR(error); 3393af7bd4dcSAmir Goldstein } 3394af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3395af7bd4dcSAmir Goldstein 3396c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 339760545d0dSAl Viro const struct open_flags *op, 33983ec2eef1SAl Viro struct file *file) 339960545d0dSAl Viro { 3400625b6d10SAl Viro struct dentry *child; 3401625b6d10SAl Viro struct path path; 3402c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 340360545d0dSAl Viro if (unlikely(error)) 340460545d0dSAl Viro return error; 3405625b6d10SAl Viro error = mnt_want_write(path.mnt); 340660545d0dSAl Viro if (unlikely(error)) 340760545d0dSAl Viro goto out; 3408af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3409af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3410684e73beSHirofumi Nakagawa if (IS_ERR(child)) 341160545d0dSAl Viro goto out2; 3412625b6d10SAl Viro dput(path.dentry); 3413625b6d10SAl Viro path.dentry = child; 3414c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 341569a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 341662fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 341760545d0dSAl Viro if (error) 341860545d0dSAl Viro goto out2; 3419625b6d10SAl Viro file->f_path.mnt = path.mnt; 3420be12af3eSAl Viro error = finish_open(file, child, NULL); 342160545d0dSAl Viro out2: 3422625b6d10SAl Viro mnt_drop_write(path.mnt); 342360545d0dSAl Viro out: 3424625b6d10SAl Viro path_put(&path); 342560545d0dSAl Viro return error; 342660545d0dSAl Viro } 342760545d0dSAl Viro 34286ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34296ac08709SAl Viro { 34306ac08709SAl Viro struct path path; 34316ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34326ac08709SAl Viro if (!error) { 34336ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3434ae2bb293SAl Viro error = vfs_open(&path, file); 34356ac08709SAl Viro path_put(&path); 34366ac08709SAl Viro } 34376ac08709SAl Viro return error; 34386ac08709SAl Viro } 34396ac08709SAl Viro 3440c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3441c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34421da177e4SLinus Torvalds { 344330d90494SAl Viro struct file *file; 344413aab428SAl Viro int error; 344531e6b01fSNick Piggin 3446ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 34471afc99beSAl Viro if (IS_ERR(file)) 34481afc99beSAl Viro return file; 344931e6b01fSNick Piggin 3450bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 34513ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 34525f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 34536ac08709SAl Viro error = do_o_path(nd, flags, file); 34545f336e72SAl Viro } else { 34555f336e72SAl Viro const char *s = path_init(nd, flags); 34563bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 34571ccac622SAl Viro (s = do_last(nd, file, op)) != NULL) 34581ccac622SAl Viro ; 3459deb106c6SAl Viro terminate_walk(nd); 34605f336e72SAl Viro } 34617c1c01ecSAl Viro if (likely(!error)) { 3462aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 34637c1c01ecSAl Viro return file; 34647c1c01ecSAl Viro WARN_ON(1); 34657c1c01ecSAl Viro error = -EINVAL; 3466015c3bbcSMiklos Szeredi } 34677c1c01ecSAl Viro fput(file); 34682675a4ebSAl Viro if (error == -EOPENSTALE) { 34692675a4ebSAl Viro if (flags & LOOKUP_RCU) 34702675a4ebSAl Viro error = -ECHILD; 34712675a4ebSAl Viro else 34722675a4ebSAl Viro error = -ESTALE; 34732675a4ebSAl Viro } 34747c1c01ecSAl Viro return ERR_PTR(error); 3475de459215SKirill Korotaev } 34761da177e4SLinus Torvalds 3477669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3478f9652e10SAl Viro const struct open_flags *op) 347913aab428SAl Viro { 34809883d185SAl Viro struct nameidata nd; 3481f9652e10SAl Viro int flags = op->lookup_flags; 348213aab428SAl Viro struct file *filp; 348313aab428SAl Viro 34849883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3485c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 348613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3487c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 348813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3489c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34909883d185SAl Viro restore_nameidata(); 349113aab428SAl Viro return filp; 349213aab428SAl Viro } 349313aab428SAl Viro 349473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3495f9652e10SAl Viro const char *name, const struct open_flags *op) 349673d049a4SAl Viro { 34979883d185SAl Viro struct nameidata nd; 349873d049a4SAl Viro struct file *file; 349951689104SPaul Moore struct filename *filename; 3500f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 350173d049a4SAl Viro 350273d049a4SAl Viro nd.root.mnt = mnt; 350373d049a4SAl Viro nd.root.dentry = dentry; 350473d049a4SAl Viro 3505b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 350673d049a4SAl Viro return ERR_PTR(-ELOOP); 350773d049a4SAl Viro 350851689104SPaul Moore filename = getname_kernel(name); 3509a1c83681SViresh Kumar if (IS_ERR(filename)) 351051689104SPaul Moore return ERR_CAST(filename); 351151689104SPaul Moore 35129883d185SAl Viro set_nameidata(&nd, -1, filename); 3513c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 351473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3515c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 351673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3517c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35189883d185SAl Viro restore_nameidata(); 351951689104SPaul Moore putname(filename); 352073d049a4SAl Viro return file; 352173d049a4SAl Viro } 352273d049a4SAl Viro 3523fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35241ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35251da177e4SLinus Torvalds { 3526c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3527391172c4SAl Viro struct qstr last; 3528391172c4SAl Viro int type; 3529c30dabfeSJan Kara int err2; 35301ac12b4bSJeff Layton int error; 35311ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 35321ac12b4bSJeff Layton 35331ac12b4bSJeff Layton /* 35341ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 35351ac12b4bSJeff Layton * other flags passed in are ignored! 35361ac12b4bSJeff Layton */ 35371ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 35381ac12b4bSJeff Layton 35395c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 35405c31b6ceSAl Viro if (IS_ERR(name)) 35415c31b6ceSAl Viro return ERR_CAST(name); 35421da177e4SLinus Torvalds 3543c663e5d8SChristoph Hellwig /* 3544c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3545c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3546c663e5d8SChristoph Hellwig */ 35475c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3548ed75e95dSAl Viro goto out; 3549c663e5d8SChristoph Hellwig 3550c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3551391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3552c663e5d8SChristoph Hellwig /* 3553c663e5d8SChristoph Hellwig * Do the final lookup. 3554c663e5d8SChristoph Hellwig */ 3555391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 35565955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3557391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 35581da177e4SLinus Torvalds if (IS_ERR(dentry)) 3559a8104a9fSAl Viro goto unlock; 3560c663e5d8SChristoph Hellwig 3561a8104a9fSAl Viro error = -EEXIST; 3562b18825a7SDavid Howells if (d_is_positive(dentry)) 3563a8104a9fSAl Viro goto fail; 3564b18825a7SDavid Howells 3565c663e5d8SChristoph Hellwig /* 3566c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3567c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3568c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3569c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3570c663e5d8SChristoph Hellwig */ 3571391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3572a8104a9fSAl Viro error = -ENOENT; 3573ed75e95dSAl Viro goto fail; 3574e9baf6e5SAl Viro } 3575c30dabfeSJan Kara if (unlikely(err2)) { 3576c30dabfeSJan Kara error = err2; 3577a8104a9fSAl Viro goto fail; 3578c30dabfeSJan Kara } 3579181c37b6SAl Viro putname(name); 3580e9baf6e5SAl Viro return dentry; 35811da177e4SLinus Torvalds fail: 3582a8104a9fSAl Viro dput(dentry); 3583a8104a9fSAl Viro dentry = ERR_PTR(error); 3584a8104a9fSAl Viro unlock: 35855955102cSAl Viro inode_unlock(path->dentry->d_inode); 3586c30dabfeSJan Kara if (!err2) 3587391172c4SAl Viro mnt_drop_write(path->mnt); 3588ed75e95dSAl Viro out: 3589391172c4SAl Viro path_put(path); 3590181c37b6SAl Viro putname(name); 3591ed75e95dSAl Viro return dentry; 3592dae6ad8fSAl Viro } 3593fa14a0b8SAl Viro 3594fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3595fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3596fa14a0b8SAl Viro { 3597181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3598181c37b6SAl Viro path, lookup_flags); 3599fa14a0b8SAl Viro } 3600dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3601dae6ad8fSAl Viro 3602921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3603921a1650SAl Viro { 3604921a1650SAl Viro dput(dentry); 36055955102cSAl Viro inode_unlock(path->dentry->d_inode); 3606a8104a9fSAl Viro mnt_drop_write(path->mnt); 3607921a1650SAl Viro path_put(path); 3608921a1650SAl Viro } 3609921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3610921a1650SAl Viro 3611520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36121ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3613dae6ad8fSAl Viro { 3614181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3615dae6ad8fSAl Viro } 3616dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3617dae6ad8fSAl Viro 36181a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 36191da177e4SLinus Torvalds { 3620a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36211da177e4SLinus Torvalds 36221da177e4SLinus Torvalds if (error) 36231da177e4SLinus Torvalds return error; 36241da177e4SLinus Torvalds 362594f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 36261da177e4SLinus Torvalds return -EPERM; 36271da177e4SLinus Torvalds 3628acfa4380SAl Viro if (!dir->i_op->mknod) 36291da177e4SLinus Torvalds return -EPERM; 36301da177e4SLinus Torvalds 363108ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 363208ce5f16SSerge E. Hallyn if (error) 363308ce5f16SSerge E. Hallyn return error; 363408ce5f16SSerge E. Hallyn 36351da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 36361da177e4SLinus Torvalds if (error) 36371da177e4SLinus Torvalds return error; 36381da177e4SLinus Torvalds 36391da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3640a74574aaSStephen Smalley if (!error) 3641f38aa942SAmy Griffis fsnotify_create(dir, dentry); 36421da177e4SLinus Torvalds return error; 36431da177e4SLinus Torvalds } 36444d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 36451da177e4SLinus Torvalds 3646f69aac00SAl Viro static int may_mknod(umode_t mode) 3647463c3197SDave Hansen { 3648463c3197SDave Hansen switch (mode & S_IFMT) { 3649463c3197SDave Hansen case S_IFREG: 3650463c3197SDave Hansen case S_IFCHR: 3651463c3197SDave Hansen case S_IFBLK: 3652463c3197SDave Hansen case S_IFIFO: 3653463c3197SDave Hansen case S_IFSOCK: 3654463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3655463c3197SDave Hansen return 0; 3656463c3197SDave Hansen case S_IFDIR: 3657463c3197SDave Hansen return -EPERM; 3658463c3197SDave Hansen default: 3659463c3197SDave Hansen return -EINVAL; 3660463c3197SDave Hansen } 3661463c3197SDave Hansen } 3662463c3197SDave Hansen 366387c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 366487c4e192SDominik Brodowski unsigned int dev) 36651da177e4SLinus Torvalds { 36661da177e4SLinus Torvalds struct dentry *dentry; 3667dae6ad8fSAl Viro struct path path; 3668dae6ad8fSAl Viro int error; 3669972567f1SJeff Layton unsigned int lookup_flags = 0; 36701da177e4SLinus Torvalds 36718e4bfca1SAl Viro error = may_mknod(mode); 36728e4bfca1SAl Viro if (error) 36738e4bfca1SAl Viro return error; 3674972567f1SJeff Layton retry: 3675972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3676dae6ad8fSAl Viro if (IS_ERR(dentry)) 3677dae6ad8fSAl Viro return PTR_ERR(dentry); 36782ad94ae6SAl Viro 3679dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3680ce3b0f8dSAl Viro mode &= ~current_umask(); 3681dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3682be6d3e56SKentaro Takeda if (error) 3683a8104a9fSAl Viro goto out; 36841da177e4SLinus Torvalds switch (mode & S_IFMT) { 36851da177e4SLinus Torvalds case 0: case S_IFREG: 3686312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 368705d1a717SMimi Zohar if (!error) 368805d1a717SMimi Zohar ima_post_path_mknod(dentry); 36891da177e4SLinus Torvalds break; 36901da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3691dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36921da177e4SLinus Torvalds new_decode_dev(dev)); 36931da177e4SLinus Torvalds break; 36941da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3695dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 36961da177e4SLinus Torvalds break; 36971da177e4SLinus Torvalds } 3698a8104a9fSAl Viro out: 3699921a1650SAl Viro done_path_create(&path, dentry); 3700972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3701972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3702972567f1SJeff Layton goto retry; 3703972567f1SJeff Layton } 37041da177e4SLinus Torvalds return error; 37051da177e4SLinus Torvalds } 37061da177e4SLinus Torvalds 370787c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 370887c4e192SDominik Brodowski unsigned int, dev) 370987c4e192SDominik Brodowski { 371087c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 371187c4e192SDominik Brodowski } 371287c4e192SDominik Brodowski 37138208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37145590ff0dSUlrich Drepper { 371587c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 37165590ff0dSUlrich Drepper } 37175590ff0dSUlrich Drepper 371818bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 37191da177e4SLinus Torvalds { 3720a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 37218de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 37221da177e4SLinus Torvalds 37231da177e4SLinus Torvalds if (error) 37241da177e4SLinus Torvalds return error; 37251da177e4SLinus Torvalds 3726acfa4380SAl Viro if (!dir->i_op->mkdir) 37271da177e4SLinus Torvalds return -EPERM; 37281da177e4SLinus Torvalds 37291da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 37301da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 37311da177e4SLinus Torvalds if (error) 37321da177e4SLinus Torvalds return error; 37331da177e4SLinus Torvalds 37348de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 37358de52778SAl Viro return -EMLINK; 37368de52778SAl Viro 37371da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3738a74574aaSStephen Smalley if (!error) 3739f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 37401da177e4SLinus Torvalds return error; 37411da177e4SLinus Torvalds } 37424d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 37431da177e4SLinus Torvalds 37440101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 37451da177e4SLinus Torvalds { 37466902d925SDave Hansen struct dentry *dentry; 3747dae6ad8fSAl Viro struct path path; 3748dae6ad8fSAl Viro int error; 3749b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 37501da177e4SLinus Torvalds 3751b76d8b82SJeff Layton retry: 3752b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 37536902d925SDave Hansen if (IS_ERR(dentry)) 3754dae6ad8fSAl Viro return PTR_ERR(dentry); 37556902d925SDave Hansen 3756dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3757ce3b0f8dSAl Viro mode &= ~current_umask(); 3758dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3759a8104a9fSAl Viro if (!error) 3760dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3761921a1650SAl Viro done_path_create(&path, dentry); 3762b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3763b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3764b76d8b82SJeff Layton goto retry; 3765b76d8b82SJeff Layton } 37661da177e4SLinus Torvalds return error; 37671da177e4SLinus Torvalds } 37681da177e4SLinus Torvalds 37690101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 37700101db7aSDominik Brodowski { 37710101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 37720101db7aSDominik Brodowski } 37730101db7aSDominik Brodowski 3774a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 37755590ff0dSUlrich Drepper { 37760101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 37775590ff0dSUlrich Drepper } 37785590ff0dSUlrich Drepper 37791da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 37801da177e4SLinus Torvalds { 37811da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 37821da177e4SLinus Torvalds 37831da177e4SLinus Torvalds if (error) 37841da177e4SLinus Torvalds return error; 37851da177e4SLinus Torvalds 3786acfa4380SAl Viro if (!dir->i_op->rmdir) 37871da177e4SLinus Torvalds return -EPERM; 37881da177e4SLinus Torvalds 37891d2ef590SAl Viro dget(dentry); 37905955102cSAl Viro inode_lock(dentry->d_inode); 3791912dbc15SSage Weil 37921da177e4SLinus Torvalds error = -EBUSY; 37937af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3794912dbc15SSage Weil goto out; 3795912dbc15SSage Weil 37961da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3797912dbc15SSage Weil if (error) 3798912dbc15SSage Weil goto out; 3799912dbc15SSage Weil 38001da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3801912dbc15SSage Weil if (error) 3802912dbc15SSage Weil goto out; 3803912dbc15SSage Weil 38048767712fSAl Viro shrink_dcache_parent(dentry); 38051da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3806d83c49f3SAl Viro dont_mount(dentry); 38078ed936b5SEric W. Biederman detach_mounts(dentry); 3808116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 38091da177e4SLinus Torvalds 3810912dbc15SSage Weil out: 38115955102cSAl Viro inode_unlock(dentry->d_inode); 38121d2ef590SAl Viro dput(dentry); 3813912dbc15SSage Weil if (!error) 3814912dbc15SSage Weil d_delete(dentry); 38151da177e4SLinus Torvalds return error; 38161da177e4SLinus Torvalds } 38174d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 38181da177e4SLinus Torvalds 3819f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 38201da177e4SLinus Torvalds { 38211da177e4SLinus Torvalds int error = 0; 382291a27b2aSJeff Layton struct filename *name; 38231da177e4SLinus Torvalds struct dentry *dentry; 3824f5beed75SAl Viro struct path path; 3825f5beed75SAl Viro struct qstr last; 3826f5beed75SAl Viro int type; 3827c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3828c6ee9206SJeff Layton retry: 3829c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3830c1d4dd27SAl Viro &path, &last, &type); 383191a27b2aSJeff Layton if (IS_ERR(name)) 383291a27b2aSJeff Layton return PTR_ERR(name); 38331da177e4SLinus Torvalds 3834f5beed75SAl Viro switch (type) { 38351da177e4SLinus Torvalds case LAST_DOTDOT: 38361da177e4SLinus Torvalds error = -ENOTEMPTY; 38371da177e4SLinus Torvalds goto exit1; 38381da177e4SLinus Torvalds case LAST_DOT: 38391da177e4SLinus Torvalds error = -EINVAL; 38401da177e4SLinus Torvalds goto exit1; 38411da177e4SLinus Torvalds case LAST_ROOT: 38421da177e4SLinus Torvalds error = -EBUSY; 38431da177e4SLinus Torvalds goto exit1; 38441da177e4SLinus Torvalds } 38450612d9fbSOGAWA Hirofumi 3846f5beed75SAl Viro error = mnt_want_write(path.mnt); 3847c30dabfeSJan Kara if (error) 3848c30dabfeSJan Kara goto exit1; 38490612d9fbSOGAWA Hirofumi 38505955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3851f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38521da177e4SLinus Torvalds error = PTR_ERR(dentry); 38536902d925SDave Hansen if (IS_ERR(dentry)) 38546902d925SDave Hansen goto exit2; 3855e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3856e6bc45d6STheodore Ts'o error = -ENOENT; 3857e6bc45d6STheodore Ts'o goto exit3; 3858e6bc45d6STheodore Ts'o } 3859f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3860be6d3e56SKentaro Takeda if (error) 3861c30dabfeSJan Kara goto exit3; 3862f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 38630622753bSDave Hansen exit3: 38641da177e4SLinus Torvalds dput(dentry); 38656902d925SDave Hansen exit2: 38665955102cSAl Viro inode_unlock(path.dentry->d_inode); 3867f5beed75SAl Viro mnt_drop_write(path.mnt); 38681da177e4SLinus Torvalds exit1: 3869f5beed75SAl Viro path_put(&path); 38701da177e4SLinus Torvalds putname(name); 3871c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3872c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3873c6ee9206SJeff Layton goto retry; 3874c6ee9206SJeff Layton } 38751da177e4SLinus Torvalds return error; 38761da177e4SLinus Torvalds } 38771da177e4SLinus Torvalds 38783cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 38795590ff0dSUlrich Drepper { 38805590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 38815590ff0dSUlrich Drepper } 38825590ff0dSUlrich Drepper 3883b21996e3SJ. Bruce Fields /** 3884b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3885b21996e3SJ. Bruce Fields * @dir: parent directory 3886b21996e3SJ. Bruce Fields * @dentry: victim 3887b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3888b21996e3SJ. Bruce Fields * 3889b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3890b21996e3SJ. Bruce Fields * 3891b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3892b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3893b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3894b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3895b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3896b21996e3SJ. Bruce Fields * 3897b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3898b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3899b21996e3SJ. Bruce Fields * to be NFS exported. 3900b21996e3SJ. Bruce Fields */ 3901b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 39021da177e4SLinus Torvalds { 39039accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 39041da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 39051da177e4SLinus Torvalds 39061da177e4SLinus Torvalds if (error) 39071da177e4SLinus Torvalds return error; 39081da177e4SLinus Torvalds 3909acfa4380SAl Viro if (!dir->i_op->unlink) 39101da177e4SLinus Torvalds return -EPERM; 39111da177e4SLinus Torvalds 39125955102cSAl Viro inode_lock(target); 39138ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 39141da177e4SLinus Torvalds error = -EBUSY; 39151da177e4SLinus Torvalds else { 39161da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3917bec1052eSAl Viro if (!error) { 39185a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 39195a14696cSJ. Bruce Fields if (error) 3920b21996e3SJ. Bruce Fields goto out; 39211da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 39228ed936b5SEric W. Biederman if (!error) { 3923d83c49f3SAl Viro dont_mount(dentry); 39248ed936b5SEric W. Biederman detach_mounts(dentry); 3925116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 39268ed936b5SEric W. Biederman } 3927bec1052eSAl Viro } 39281da177e4SLinus Torvalds } 3929b21996e3SJ. Bruce Fields out: 39305955102cSAl Viro inode_unlock(target); 39311da177e4SLinus Torvalds 39321da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 39331da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 39349accbb97SJ. Bruce Fields fsnotify_link_count(target); 39351da177e4SLinus Torvalds d_delete(dentry); 39361da177e4SLinus Torvalds } 39370eeca283SRobert Love 39381da177e4SLinus Torvalds return error; 39391da177e4SLinus Torvalds } 39404d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 39411da177e4SLinus Torvalds 39421da177e4SLinus Torvalds /* 39431da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 39441b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 39451da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 39461da177e4SLinus Torvalds * while waiting on the I/O. 39471da177e4SLinus Torvalds */ 3948da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 39491da177e4SLinus Torvalds { 39502ad94ae6SAl Viro int error; 39511da177e4SLinus Torvalds struct dentry *dentry; 3952f5beed75SAl Viro struct path path; 3953f5beed75SAl Viro struct qstr last; 3954f5beed75SAl Viro int type; 39551da177e4SLinus Torvalds struct inode *inode = NULL; 3956b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 39575d18f813SJeff Layton unsigned int lookup_flags = 0; 39585d18f813SJeff Layton retry: 3959da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 396091a27b2aSJeff Layton if (IS_ERR(name)) 396191a27b2aSJeff Layton return PTR_ERR(name); 39622ad94ae6SAl Viro 39631da177e4SLinus Torvalds error = -EISDIR; 3964f5beed75SAl Viro if (type != LAST_NORM) 39651da177e4SLinus Torvalds goto exit1; 39660612d9fbSOGAWA Hirofumi 3967f5beed75SAl Viro error = mnt_want_write(path.mnt); 3968c30dabfeSJan Kara if (error) 3969c30dabfeSJan Kara goto exit1; 3970b21996e3SJ. Bruce Fields retry_deleg: 39715955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3972f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39731da177e4SLinus Torvalds error = PTR_ERR(dentry); 39741da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 39751da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3976f5beed75SAl Viro if (last.name[last.len]) 397750338b88STörök Edwin goto slashes; 39781da177e4SLinus Torvalds inode = dentry->d_inode; 3979b18825a7SDavid Howells if (d_is_negative(dentry)) 3980e6bc45d6STheodore Ts'o goto slashes; 39817de9c6eeSAl Viro ihold(inode); 3982f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3983be6d3e56SKentaro Takeda if (error) 3984c30dabfeSJan Kara goto exit2; 3985f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 39861da177e4SLinus Torvalds exit2: 39871da177e4SLinus Torvalds dput(dentry); 39881da177e4SLinus Torvalds } 39895955102cSAl Viro inode_unlock(path.dentry->d_inode); 39901da177e4SLinus Torvalds if (inode) 39911da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3992b21996e3SJ. Bruce Fields inode = NULL; 3993b21996e3SJ. Bruce Fields if (delegated_inode) { 39945a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3995b21996e3SJ. Bruce Fields if (!error) 3996b21996e3SJ. Bruce Fields goto retry_deleg; 3997b21996e3SJ. Bruce Fields } 3998f5beed75SAl Viro mnt_drop_write(path.mnt); 39991da177e4SLinus Torvalds exit1: 4000f5beed75SAl Viro path_put(&path); 40015d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 40025d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 40035d18f813SJeff Layton inode = NULL; 40045d18f813SJeff Layton goto retry; 40055d18f813SJeff Layton } 4006da2f1362SChristoph Hellwig putname(name); 40071da177e4SLinus Torvalds return error; 40081da177e4SLinus Torvalds 40091da177e4SLinus Torvalds slashes: 4010b18825a7SDavid Howells if (d_is_negative(dentry)) 4011b18825a7SDavid Howells error = -ENOENT; 401244b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4013b18825a7SDavid Howells error = -EISDIR; 4014b18825a7SDavid Howells else 4015b18825a7SDavid Howells error = -ENOTDIR; 40161da177e4SLinus Torvalds goto exit2; 40171da177e4SLinus Torvalds } 40181da177e4SLinus Torvalds 40192e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 40205590ff0dSUlrich Drepper { 40215590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 40225590ff0dSUlrich Drepper return -EINVAL; 40235590ff0dSUlrich Drepper 40245590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 40255590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 40265590ff0dSUlrich Drepper 4027da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 40285590ff0dSUlrich Drepper } 40295590ff0dSUlrich Drepper 40303480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 40315590ff0dSUlrich Drepper { 4032da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 40335590ff0dSUlrich Drepper } 40345590ff0dSUlrich Drepper 4035db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 40361da177e4SLinus Torvalds { 4037a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 40381da177e4SLinus Torvalds 40391da177e4SLinus Torvalds if (error) 40401da177e4SLinus Torvalds return error; 40411da177e4SLinus Torvalds 4042acfa4380SAl Viro if (!dir->i_op->symlink) 40431da177e4SLinus Torvalds return -EPERM; 40441da177e4SLinus Torvalds 40451da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 40461da177e4SLinus Torvalds if (error) 40471da177e4SLinus Torvalds return error; 40481da177e4SLinus Torvalds 40491da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4050a74574aaSStephen Smalley if (!error) 4051f38aa942SAmy Griffis fsnotify_create(dir, dentry); 40521da177e4SLinus Torvalds return error; 40531da177e4SLinus Torvalds } 40544d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 40551da177e4SLinus Torvalds 4056b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 4057b724e846SDominik Brodowski const char __user *newname) 40581da177e4SLinus Torvalds { 40592ad94ae6SAl Viro int error; 406091a27b2aSJeff Layton struct filename *from; 40616902d925SDave Hansen struct dentry *dentry; 4062dae6ad8fSAl Viro struct path path; 4063f46d3567SJeff Layton unsigned int lookup_flags = 0; 40641da177e4SLinus Torvalds 40651da177e4SLinus Torvalds from = getname(oldname); 40661da177e4SLinus Torvalds if (IS_ERR(from)) 40671da177e4SLinus Torvalds return PTR_ERR(from); 4068f46d3567SJeff Layton retry: 4069f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 40701da177e4SLinus Torvalds error = PTR_ERR(dentry); 40716902d925SDave Hansen if (IS_ERR(dentry)) 4072dae6ad8fSAl Viro goto out_putname; 40736902d925SDave Hansen 407491a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4075a8104a9fSAl Viro if (!error) 407691a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4077921a1650SAl Viro done_path_create(&path, dentry); 4078f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4079f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4080f46d3567SJeff Layton goto retry; 4081f46d3567SJeff Layton } 40826902d925SDave Hansen out_putname: 40831da177e4SLinus Torvalds putname(from); 40841da177e4SLinus Torvalds return error; 40851da177e4SLinus Torvalds } 40861da177e4SLinus Torvalds 4087b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4088b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4089b724e846SDominik Brodowski { 4090b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4091b724e846SDominik Brodowski } 4092b724e846SDominik Brodowski 40933480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40945590ff0dSUlrich Drepper { 4095b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 40965590ff0dSUlrich Drepper } 40975590ff0dSUlrich Drepper 4098146a8595SJ. Bruce Fields /** 4099146a8595SJ. Bruce Fields * vfs_link - create a new link 4100146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4101146a8595SJ. Bruce Fields * @dir: new parent 4102146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4103146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4104146a8595SJ. Bruce Fields * 4105146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4106146a8595SJ. Bruce Fields * 4107146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4108146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4109146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4110146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4111146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4112146a8595SJ. Bruce Fields * 4113146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4114146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4115146a8595SJ. Bruce Fields * to be NFS exported. 4116146a8595SJ. Bruce Fields */ 4117146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 41181da177e4SLinus Torvalds { 41191da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 41208de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41211da177e4SLinus Torvalds int error; 41221da177e4SLinus Torvalds 41231da177e4SLinus Torvalds if (!inode) 41241da177e4SLinus Torvalds return -ENOENT; 41251da177e4SLinus Torvalds 4126a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 41271da177e4SLinus Torvalds if (error) 41281da177e4SLinus Torvalds return error; 41291da177e4SLinus Torvalds 41301da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 41311da177e4SLinus Torvalds return -EXDEV; 41321da177e4SLinus Torvalds 41331da177e4SLinus Torvalds /* 41341da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 41351da177e4SLinus Torvalds */ 41361da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 41371da177e4SLinus Torvalds return -EPERM; 41380bd23d09SEric W. Biederman /* 41390bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 41400bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 41410bd23d09SEric W. Biederman * the vfs. 41420bd23d09SEric W. Biederman */ 41430bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 41440bd23d09SEric W. Biederman return -EPERM; 4145acfa4380SAl Viro if (!dir->i_op->link) 41461da177e4SLinus Torvalds return -EPERM; 41477e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 41481da177e4SLinus Torvalds return -EPERM; 41491da177e4SLinus Torvalds 41501da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 41511da177e4SLinus Torvalds if (error) 41521da177e4SLinus Torvalds return error; 41531da177e4SLinus Torvalds 41545955102cSAl Viro inode_lock(inode); 4155aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4156f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4157aae8a97dSAneesh Kumar K.V error = -ENOENT; 41588de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 41598de52778SAl Viro error = -EMLINK; 4160146a8595SJ. Bruce Fields else { 4161146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4162146a8595SJ. Bruce Fields if (!error) 41631da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4164146a8595SJ. Bruce Fields } 4165f4e0c30cSAl Viro 4166f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4167f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4168f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4169f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4170f4e0c30cSAl Viro } 41715955102cSAl Viro inode_unlock(inode); 4172e31e14ecSStephen Smalley if (!error) 41737e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 41741da177e4SLinus Torvalds return error; 41751da177e4SLinus Torvalds } 41764d359507SAl Viro EXPORT_SYMBOL(vfs_link); 41771da177e4SLinus Torvalds 41781da177e4SLinus Torvalds /* 41791da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41801da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41811da177e4SLinus Torvalds * newname. --KAB 41821da177e4SLinus Torvalds * 41831da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41841da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 41851da177e4SLinus Torvalds * and other special files. --ADM 41861da177e4SLinus Torvalds */ 418746ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 418846ea89ebSDominik Brodowski const char __user *newname, int flags) 41891da177e4SLinus Torvalds { 41901da177e4SLinus Torvalds struct dentry *new_dentry; 4191dae6ad8fSAl Viro struct path old_path, new_path; 4192146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 419311a7b371SAneesh Kumar K.V int how = 0; 41941da177e4SLinus Torvalds int error; 41951da177e4SLinus Torvalds 419611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4197c04030e1SUlrich Drepper return -EINVAL; 419811a7b371SAneesh Kumar K.V /* 4199f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4200f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4201f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 420211a7b371SAneesh Kumar K.V */ 4203f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4204f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4205f0cc6ffbSLinus Torvalds return -ENOENT; 420611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4207f0cc6ffbSLinus Torvalds } 4208c04030e1SUlrich Drepper 420911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 421011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4211442e31caSJeff Layton retry: 421211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 42131da177e4SLinus Torvalds if (error) 42142ad94ae6SAl Viro return error; 42152ad94ae6SAl Viro 4216442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4217442e31caSJeff Layton (how & LOOKUP_REVAL)); 42181da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 42196902d925SDave Hansen if (IS_ERR(new_dentry)) 4220dae6ad8fSAl Viro goto out; 4221dae6ad8fSAl Viro 4222dae6ad8fSAl Viro error = -EXDEV; 4223dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4224dae6ad8fSAl Viro goto out_dput; 4225800179c9SKees Cook error = may_linkat(&old_path); 4226800179c9SKees Cook if (unlikely(error)) 4227800179c9SKees Cook goto out_dput; 4228dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4229be6d3e56SKentaro Takeda if (error) 4230a8104a9fSAl Viro goto out_dput; 4231146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 423275c3f29dSDave Hansen out_dput: 4233921a1650SAl Viro done_path_create(&new_path, new_dentry); 4234146a8595SJ. Bruce Fields if (delegated_inode) { 4235146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4236d22e6338SOleg Drokin if (!error) { 4237d22e6338SOleg Drokin path_put(&old_path); 4238146a8595SJ. Bruce Fields goto retry; 4239146a8595SJ. Bruce Fields } 4240d22e6338SOleg Drokin } 4241442e31caSJeff Layton if (retry_estale(error, how)) { 4242d22e6338SOleg Drokin path_put(&old_path); 4243442e31caSJeff Layton how |= LOOKUP_REVAL; 4244442e31caSJeff Layton goto retry; 4245442e31caSJeff Layton } 42461da177e4SLinus Torvalds out: 42472d8f3038SAl Viro path_put(&old_path); 42481da177e4SLinus Torvalds 42491da177e4SLinus Torvalds return error; 42501da177e4SLinus Torvalds } 42511da177e4SLinus Torvalds 425246ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 425346ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 425446ea89ebSDominik Brodowski { 425546ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 425646ea89ebSDominik Brodowski } 425746ea89ebSDominik Brodowski 42583480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 42595590ff0dSUlrich Drepper { 426046ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 42615590ff0dSUlrich Drepper } 42625590ff0dSUlrich Drepper 4263bc27027aSMiklos Szeredi /** 4264bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4265bc27027aSMiklos Szeredi * @old_dir: parent of source 4266bc27027aSMiklos Szeredi * @old_dentry: source 4267bc27027aSMiklos Szeredi * @new_dir: parent of destination 4268bc27027aSMiklos Szeredi * @new_dentry: destination 4269bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4270520c8b16SMiklos Szeredi * @flags: rename flags 4271bc27027aSMiklos Szeredi * 4272bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4273bc27027aSMiklos Szeredi * 4274bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4275bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4276bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4277bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4278bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4279bc27027aSMiklos Szeredi * so. 4280bc27027aSMiklos Szeredi * 4281bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4282bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4283bc27027aSMiklos Szeredi * to be NFS exported. 4284bc27027aSMiklos Szeredi * 42851da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42861da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42871da177e4SLinus Torvalds * Problems: 42880117d427SMauro Carvalho Chehab * 4289d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42901da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42911da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4292a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42931da177e4SLinus Torvalds * story. 42946cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42956cedba89SJ. Bruce Fields * and source (if it is not a directory). 42961b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42971da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42981da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4299a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 43001da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 43011da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 43021da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4303a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 43041da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 43051da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 43061da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4307e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 43081b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 43091da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4310c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 43111da177e4SLinus Torvalds * locking]. 43121da177e4SLinus Torvalds */ 43131da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 43148e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4315520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 43161da177e4SLinus Torvalds { 43171da177e4SLinus Torvalds int error; 4318bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4319bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4320bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4321da1ce067SMiklos Szeredi bool new_is_dir = false; 4322da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 432349d31c2fSAl Viro struct name_snapshot old_name; 43241da177e4SLinus Torvalds 43258d3e2936SMiklos Szeredi if (source == target) 43261da177e4SLinus Torvalds return 0; 43271da177e4SLinus Torvalds 43281da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 43291da177e4SLinus Torvalds if (error) 43301da177e4SLinus Torvalds return error; 43311da177e4SLinus Torvalds 4332da1ce067SMiklos Szeredi if (!target) { 4333a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4334da1ce067SMiklos Szeredi } else { 4335da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4336da1ce067SMiklos Szeredi 4337da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43381da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4339da1ce067SMiklos Szeredi else 4340da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4341da1ce067SMiklos Szeredi } 43421da177e4SLinus Torvalds if (error) 43431da177e4SLinus Torvalds return error; 43441da177e4SLinus Torvalds 43452773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 43461da177e4SLinus Torvalds return -EPERM; 43471da177e4SLinus Torvalds 4348bc27027aSMiklos Szeredi /* 4349bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4350bc27027aSMiklos Szeredi * we'll need to flip '..'. 4351bc27027aSMiklos Szeredi */ 4352da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4353da1ce067SMiklos Szeredi if (is_dir) { 4354bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4355bc27027aSMiklos Szeredi if (error) 4356bc27027aSMiklos Szeredi return error; 4357bc27027aSMiklos Szeredi } 4358da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4359da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4360da1ce067SMiklos Szeredi if (error) 4361da1ce067SMiklos Szeredi return error; 4362da1ce067SMiklos Szeredi } 4363da1ce067SMiklos Szeredi } 43640eeca283SRobert Love 43650b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 43660b3974ebSMiklos Szeredi flags); 4367bc27027aSMiklos Szeredi if (error) 4368bc27027aSMiklos Szeredi return error; 4369bc27027aSMiklos Szeredi 437049d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4371bc27027aSMiklos Szeredi dget(new_dentry); 4372da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4373bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4374bc27027aSMiklos Szeredi else if (target) 43755955102cSAl Viro inode_lock(target); 4376bc27027aSMiklos Szeredi 4377bc27027aSMiklos Szeredi error = -EBUSY; 43787af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4379bc27027aSMiklos Szeredi goto out; 4380bc27027aSMiklos Szeredi 4381da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4382bc27027aSMiklos Szeredi error = -EMLINK; 4383da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4384bc27027aSMiklos Szeredi goto out; 4385da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4386da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4387da1ce067SMiklos Szeredi goto out; 4388da1ce067SMiklos Szeredi } 4389da1ce067SMiklos Szeredi if (!is_dir) { 4390bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4391bc27027aSMiklos Szeredi if (error) 4392bc27027aSMiklos Szeredi goto out; 4393da1ce067SMiklos Szeredi } 4394da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4395bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4396bc27027aSMiklos Szeredi if (error) 4397bc27027aSMiklos Szeredi goto out; 4398bc27027aSMiklos Szeredi } 4399520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4400520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4401bc27027aSMiklos Szeredi if (error) 4402bc27027aSMiklos Szeredi goto out; 4403bc27027aSMiklos Szeredi 4404da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 44058767712fSAl Viro if (is_dir) { 44068767712fSAl Viro shrink_dcache_parent(new_dentry); 4407bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 44088767712fSAl Viro } 4409bc27027aSMiklos Szeredi dont_mount(new_dentry); 44108ed936b5SEric W. Biederman detach_mounts(new_dentry); 4411bc27027aSMiklos Szeredi } 4412da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4413da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4414bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4415da1ce067SMiklos Szeredi else 4416da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4417da1ce067SMiklos Szeredi } 4418bc27027aSMiklos Szeredi out: 4419da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4420bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4421bc27027aSMiklos Szeredi else if (target) 44225955102cSAl Viro inode_unlock(target); 4423bc27027aSMiklos Szeredi dput(new_dentry); 4424da1ce067SMiklos Szeredi if (!error) { 4425f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4426da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4427da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4428f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4429da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4430da1ce067SMiklos Szeredi } 4431da1ce067SMiklos Szeredi } 443249d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 44330eeca283SRobert Love 44341da177e4SLinus Torvalds return error; 44351da177e4SLinus Torvalds } 44364d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 44371da177e4SLinus Torvalds 4438ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4439ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 44401da177e4SLinus Torvalds { 44411da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 44421da177e4SLinus Torvalds struct dentry *trap; 4443f5beed75SAl Viro struct path old_path, new_path; 4444f5beed75SAl Viro struct qstr old_last, new_last; 4445f5beed75SAl Viro int old_type, new_type; 44468e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 444791a27b2aSJeff Layton struct filename *from; 444891a27b2aSJeff Layton struct filename *to; 4449f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4450c6a94284SJeff Layton bool should_retry = false; 44512ad94ae6SAl Viro int error; 4452520c8b16SMiklos Szeredi 44530d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4454da1ce067SMiklos Szeredi return -EINVAL; 4455da1ce067SMiklos Szeredi 44560d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 44570d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4458520c8b16SMiklos Szeredi return -EINVAL; 4459520c8b16SMiklos Szeredi 44600d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 44610d7a8555SMiklos Szeredi return -EPERM; 44620d7a8555SMiklos Szeredi 4463f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4464f5beed75SAl Viro target_flags = 0; 4465f5beed75SAl Viro 4466c6a94284SJeff Layton retry: 4467c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4468c1d4dd27SAl Viro &old_path, &old_last, &old_type); 446991a27b2aSJeff Layton if (IS_ERR(from)) { 447091a27b2aSJeff Layton error = PTR_ERR(from); 44711da177e4SLinus Torvalds goto exit; 447291a27b2aSJeff Layton } 44731da177e4SLinus Torvalds 4474c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4475c1d4dd27SAl Viro &new_path, &new_last, &new_type); 447691a27b2aSJeff Layton if (IS_ERR(to)) { 447791a27b2aSJeff Layton error = PTR_ERR(to); 44781da177e4SLinus Torvalds goto exit1; 447991a27b2aSJeff Layton } 44801da177e4SLinus Torvalds 44811da177e4SLinus Torvalds error = -EXDEV; 4482f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44831da177e4SLinus Torvalds goto exit2; 44841da177e4SLinus Torvalds 44851da177e4SLinus Torvalds error = -EBUSY; 4486f5beed75SAl Viro if (old_type != LAST_NORM) 44871da177e4SLinus Torvalds goto exit2; 44881da177e4SLinus Torvalds 44890a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44900a7c3937SMiklos Szeredi error = -EEXIST; 4491f5beed75SAl Viro if (new_type != LAST_NORM) 44921da177e4SLinus Torvalds goto exit2; 44931da177e4SLinus Torvalds 4494f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4495c30dabfeSJan Kara if (error) 4496c30dabfeSJan Kara goto exit2; 4497c30dabfeSJan Kara 44988e6d782cSJ. Bruce Fields retry_deleg: 4499f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 45001da177e4SLinus Torvalds 4501f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 45021da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 45031da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 45041da177e4SLinus Torvalds goto exit3; 45051da177e4SLinus Torvalds /* source must exist */ 45061da177e4SLinus Torvalds error = -ENOENT; 4507b18825a7SDavid Howells if (d_is_negative(old_dentry)) 45081da177e4SLinus Torvalds goto exit4; 4509f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 45101da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45111da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 45121da177e4SLinus Torvalds goto exit4; 45130a7c3937SMiklos Szeredi error = -EEXIST; 45140a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 45150a7c3937SMiklos Szeredi goto exit5; 4516da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4517da1ce067SMiklos Szeredi error = -ENOENT; 4518da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4519da1ce067SMiklos Szeredi goto exit5; 4520da1ce067SMiklos Szeredi 4521da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4522da1ce067SMiklos Szeredi error = -ENOTDIR; 4523f5beed75SAl Viro if (new_last.name[new_last.len]) 4524da1ce067SMiklos Szeredi goto exit5; 4525da1ce067SMiklos Szeredi } 4526da1ce067SMiklos Szeredi } 45270a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 45280a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 45290a7c3937SMiklos Szeredi error = -ENOTDIR; 4530f5beed75SAl Viro if (old_last.name[old_last.len]) 45310a7c3937SMiklos Szeredi goto exit5; 4532f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 45330a7c3937SMiklos Szeredi goto exit5; 45340a7c3937SMiklos Szeredi } 45350a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 45360a7c3937SMiklos Szeredi error = -EINVAL; 45370a7c3937SMiklos Szeredi if (old_dentry == trap) 45380a7c3937SMiklos Szeredi goto exit5; 45391da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4540da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 45411da177e4SLinus Torvalds error = -ENOTEMPTY; 45421da177e4SLinus Torvalds if (new_dentry == trap) 45431da177e4SLinus Torvalds goto exit5; 45441da177e4SLinus Torvalds 4545f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4546f5beed75SAl Viro &new_path, new_dentry, flags); 4547be6d3e56SKentaro Takeda if (error) 4548c30dabfeSJan Kara goto exit5; 4549f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4550f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4551520c8b16SMiklos Szeredi &delegated_inode, flags); 45521da177e4SLinus Torvalds exit5: 45531da177e4SLinus Torvalds dput(new_dentry); 45541da177e4SLinus Torvalds exit4: 45551da177e4SLinus Torvalds dput(old_dentry); 45561da177e4SLinus Torvalds exit3: 4557f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 45588e6d782cSJ. Bruce Fields if (delegated_inode) { 45598e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 45608e6d782cSJ. Bruce Fields if (!error) 45618e6d782cSJ. Bruce Fields goto retry_deleg; 45628e6d782cSJ. Bruce Fields } 4563f5beed75SAl Viro mnt_drop_write(old_path.mnt); 45641da177e4SLinus Torvalds exit2: 4565c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4566c6a94284SJeff Layton should_retry = true; 4567f5beed75SAl Viro path_put(&new_path); 45682ad94ae6SAl Viro putname(to); 45691da177e4SLinus Torvalds exit1: 4570f5beed75SAl Viro path_put(&old_path); 45711da177e4SLinus Torvalds putname(from); 4572c6a94284SJeff Layton if (should_retry) { 4573c6a94284SJeff Layton should_retry = false; 4574c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4575c6a94284SJeff Layton goto retry; 4576c6a94284SJeff Layton } 45772ad94ae6SAl Viro exit: 45781da177e4SLinus Torvalds return error; 45791da177e4SLinus Torvalds } 45801da177e4SLinus Torvalds 4581ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4582ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4583ee81feb6SDominik Brodowski { 4584ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4585ee81feb6SDominik Brodowski } 4586ee81feb6SDominik Brodowski 4587520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4588520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4589520c8b16SMiklos Szeredi { 4590ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4591520c8b16SMiklos Szeredi } 4592520c8b16SMiklos Szeredi 4593a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45945590ff0dSUlrich Drepper { 4595ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45965590ff0dSUlrich Drepper } 45975590ff0dSUlrich Drepper 4598787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4599787fb6bcSMiklos Szeredi { 4600787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4601787fb6bcSMiklos Szeredi if (error) 4602787fb6bcSMiklos Szeredi return error; 4603787fb6bcSMiklos Szeredi 4604787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4605787fb6bcSMiklos Szeredi return -EPERM; 4606787fb6bcSMiklos Szeredi 4607787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4608787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4609787fb6bcSMiklos Szeredi } 4610787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4611787fb6bcSMiklos Szeredi 46125d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 46131da177e4SLinus Torvalds { 46145d826c84SAl Viro int len = PTR_ERR(link); 46151da177e4SLinus Torvalds if (IS_ERR(link)) 46161da177e4SLinus Torvalds goto out; 46171da177e4SLinus Torvalds 46181da177e4SLinus Torvalds len = strlen(link); 46191da177e4SLinus Torvalds if (len > (unsigned) buflen) 46201da177e4SLinus Torvalds len = buflen; 46211da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 46221da177e4SLinus Torvalds len = -EFAULT; 46231da177e4SLinus Torvalds out: 46241da177e4SLinus Torvalds return len; 46251da177e4SLinus Torvalds } 46261da177e4SLinus Torvalds 4627d60874cdSMiklos Szeredi /** 4628fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4629fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4630fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4631fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4632fd4a0edfSMiklos Szeredi * 4633fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4634fd4a0edfSMiklos Szeredi * 4635fd4a0edfSMiklos Szeredi * Does not call security hook. 4636fd4a0edfSMiklos Szeredi */ 4637fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4638fd4a0edfSMiklos Szeredi { 4639fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4640f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4641f2df5da6SAl Viro const char *link; 4642f2df5da6SAl Viro int res; 4643fd4a0edfSMiklos Szeredi 464476fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 464576fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 464676fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 464776fca90eSMiklos Szeredi 464876fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4649fd4a0edfSMiklos Szeredi return -EINVAL; 4650fd4a0edfSMiklos Szeredi 465176fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 465276fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 465376fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 465476fca90eSMiklos Szeredi } 465576fca90eSMiklos Szeredi 46564c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4657f2df5da6SAl Viro if (!link) { 4658f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4659f2df5da6SAl Viro if (IS_ERR(link)) 4660f2df5da6SAl Viro return PTR_ERR(link); 4661f2df5da6SAl Viro } 4662f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4663f2df5da6SAl Viro do_delayed_call(&done); 4664f2df5da6SAl Viro return res; 4665fd4a0edfSMiklos Szeredi } 4666fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 46671da177e4SLinus Torvalds 4668d60874cdSMiklos Szeredi /** 4669d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4670d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4671d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4672d60874cdSMiklos Szeredi * 4673d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4674d60874cdSMiklos Szeredi * 4675d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4676d60874cdSMiklos Szeredi * 4677d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4678d60874cdSMiklos Szeredi */ 4679d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4680d60874cdSMiklos Szeredi { 4681d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4682d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4683d60874cdSMiklos Szeredi 4684d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4685d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4686d60874cdSMiklos Szeredi if (!res) 4687d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4688d60874cdSMiklos Szeredi } 4689d60874cdSMiklos Szeredi return res; 4690d60874cdSMiklos Szeredi } 4691d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4692d60874cdSMiklos Szeredi 46931da177e4SLinus Torvalds /* get the link contents into pagecache */ 46946b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4695fceef393SAl Viro struct delayed_call *callback) 46961da177e4SLinus Torvalds { 4697ebd09abbSDuane Griffin char *kaddr; 46981da177e4SLinus Torvalds struct page *page; 46996b255391SAl Viro struct address_space *mapping = inode->i_mapping; 47006b255391SAl Viro 4701d3883d4fSAl Viro if (!dentry) { 4702d3883d4fSAl Viro page = find_get_page(mapping, 0); 4703d3883d4fSAl Viro if (!page) 47046b255391SAl Viro return ERR_PTR(-ECHILD); 4705d3883d4fSAl Viro if (!PageUptodate(page)) { 4706d3883d4fSAl Viro put_page(page); 4707d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4708d3883d4fSAl Viro } 4709d3883d4fSAl Viro } else { 4710090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 47111da177e4SLinus Torvalds if (IS_ERR(page)) 47126fe6900eSNick Piggin return (char*)page; 4713d3883d4fSAl Viro } 4714fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 471521fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 471621fc61c7SAl Viro kaddr = page_address(page); 47176b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4718ebd09abbSDuane Griffin return kaddr; 47191da177e4SLinus Torvalds } 47201da177e4SLinus Torvalds 47216b255391SAl Viro EXPORT_SYMBOL(page_get_link); 47221da177e4SLinus Torvalds 4723fceef393SAl Viro void page_put_link(void *arg) 47241da177e4SLinus Torvalds { 4725fceef393SAl Viro put_page(arg); 47261da177e4SLinus Torvalds } 47274d359507SAl Viro EXPORT_SYMBOL(page_put_link); 47281da177e4SLinus Torvalds 4729aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4730aa80deabSAl Viro { 4731fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47326b255391SAl Viro int res = readlink_copy(buffer, buflen, 47336b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4734fceef393SAl Viro &done)); 4735fceef393SAl Viro do_delayed_call(&done); 4736aa80deabSAl Viro return res; 4737aa80deabSAl Viro } 4738aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4739aa80deabSAl Viro 474054566b2cSNick Piggin /* 474154566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 474254566b2cSNick Piggin */ 474354566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 47441da177e4SLinus Torvalds { 47451da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 47460adb25d2SKirill Korotaev struct page *page; 4747afddba49SNick Piggin void *fsdata; 4748beb497abSDmitriy Monakhov int err; 4749c718a975STetsuo Handa unsigned int flags = 0; 475054566b2cSNick Piggin if (nofs) 475154566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 47521da177e4SLinus Torvalds 47537e53cac4SNeilBrown retry: 4754afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 475554566b2cSNick Piggin flags, &page, &fsdata); 47561da177e4SLinus Torvalds if (err) 4757afddba49SNick Piggin goto fail; 4758afddba49SNick Piggin 475921fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4760afddba49SNick Piggin 4761afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4762afddba49SNick Piggin page, fsdata); 47631da177e4SLinus Torvalds if (err < 0) 47641da177e4SLinus Torvalds goto fail; 4765afddba49SNick Piggin if (err < len-1) 4766afddba49SNick Piggin goto retry; 4767afddba49SNick Piggin 47681da177e4SLinus Torvalds mark_inode_dirty(inode); 47691da177e4SLinus Torvalds return 0; 47701da177e4SLinus Torvalds fail: 47711da177e4SLinus Torvalds return err; 47721da177e4SLinus Torvalds } 47734d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 47741da177e4SLinus Torvalds 47750adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 47760adb25d2SKirill Korotaev { 47770adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4778c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 47790adb25d2SKirill Korotaev } 47804d359507SAl Viro EXPORT_SYMBOL(page_symlink); 47810adb25d2SKirill Korotaev 478292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 47836b255391SAl Viro .get_link = page_get_link, 47841da177e4SLinus Torvalds }; 47851da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4786