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; 5069883d185SAl Viro unsigned root_seq; 5079883d185SAl Viro int dfd; 5083859a271SKees Cook } __randomize_layout; 5091f55a6ecSAl Viro 5109883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 511894bc8c4SAl Viro { 512756daf26SNeilBrown struct nameidata *old = current->nameidata; 513756daf26SNeilBrown p->stack = p->internal; 514c8a53ee5SAl Viro p->dfd = dfd; 515c8a53ee5SAl Viro p->name = name; 516756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5179883d185SAl Viro p->saved = old; 518756daf26SNeilBrown current->nameidata = p; 519894bc8c4SAl Viro } 520894bc8c4SAl Viro 5219883d185SAl Viro static void restore_nameidata(void) 522894bc8c4SAl Viro { 5239883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 524756daf26SNeilBrown 525756daf26SNeilBrown current->nameidata = old; 526756daf26SNeilBrown if (old) 527756daf26SNeilBrown old->total_link_count = now->total_link_count; 528e1a63bbcSAl Viro if (now->stack != now->internal) 529756daf26SNeilBrown kfree(now->stack); 530894bc8c4SAl Viro } 531894bc8c4SAl Viro 532894bc8c4SAl Viro static int __nd_alloc_stack(struct nameidata *nd) 533894bc8c4SAl Viro { 534bc40aee0SAl Viro struct saved *p; 535bc40aee0SAl Viro 536bc40aee0SAl Viro if (nd->flags & LOOKUP_RCU) { 5376da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 538bc40aee0SAl Viro GFP_ATOMIC); 539bc40aee0SAl Viro if (unlikely(!p)) 540bc40aee0SAl Viro return -ECHILD; 541bc40aee0SAl Viro } else { 5426da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 543894bc8c4SAl Viro GFP_KERNEL); 544894bc8c4SAl Viro if (unlikely(!p)) 545894bc8c4SAl Viro return -ENOMEM; 546bc40aee0SAl Viro } 547894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 548894bc8c4SAl Viro nd->stack = p; 549894bc8c4SAl Viro return 0; 550894bc8c4SAl Viro } 551894bc8c4SAl Viro 552397d425dSEric W. Biederman /** 553397d425dSEric W. Biederman * path_connected - Verify that a path->dentry is below path->mnt.mnt_root 554397d425dSEric W. Biederman * @path: nameidate to verify 555397d425dSEric W. Biederman * 556397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 557397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 558397d425dSEric W. Biederman */ 559397d425dSEric W. Biederman static bool path_connected(const struct path *path) 560397d425dSEric W. Biederman { 561397d425dSEric W. Biederman struct vfsmount *mnt = path->mnt; 56295dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 563397d425dSEric W. Biederman 56495dd7758SEric W. Biederman /* Bind mounts and multi-root filesystems can have disconnected paths */ 56595dd7758SEric W. Biederman if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root)) 566397d425dSEric W. Biederman return true; 567397d425dSEric W. Biederman 568397d425dSEric W. Biederman return is_subdir(path->dentry, mnt->mnt_root); 569397d425dSEric W. Biederman } 570397d425dSEric W. Biederman 571894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 572894bc8c4SAl Viro { 573da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 574894bc8c4SAl Viro return 0; 575894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 576894bc8c4SAl Viro return 0; 577894bc8c4SAl Viro return __nd_alloc_stack(nd); 578894bc8c4SAl Viro } 579894bc8c4SAl Viro 5807973387aSAl Viro static void drop_links(struct nameidata *nd) 5817973387aSAl Viro { 5827973387aSAl Viro int i = nd->depth; 5837973387aSAl Viro while (i--) { 5847973387aSAl Viro struct saved *last = nd->stack + i; 585fceef393SAl Viro do_delayed_call(&last->done); 586fceef393SAl Viro clear_delayed_call(&last->done); 5877973387aSAl Viro } 5887973387aSAl Viro } 5897973387aSAl Viro 5907973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5917973387aSAl Viro { 5927973387aSAl Viro drop_links(nd); 5937973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5947973387aSAl Viro int i; 5957973387aSAl Viro path_put(&nd->path); 5967973387aSAl Viro for (i = 0; i < nd->depth; i++) 5977973387aSAl Viro path_put(&nd->stack[i].link); 59884a2bd39SAl Viro if (nd->flags & LOOKUP_ROOT_GRABBED) { 599102b8af2SAl Viro path_put(&nd->root); 60084a2bd39SAl Viro nd->flags &= ~LOOKUP_ROOT_GRABBED; 601102b8af2SAl Viro } 6027973387aSAl Viro } else { 6037973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6047973387aSAl Viro rcu_read_unlock(); 6057973387aSAl Viro } 6067973387aSAl Viro nd->depth = 0; 6077973387aSAl Viro } 6087973387aSAl Viro 6097973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6107973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6117973387aSAl Viro struct path *path, unsigned seq) 6127973387aSAl Viro { 6137973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6147973387aSAl Viro if (unlikely(res)) { 6157973387aSAl Viro if (res > 0) 6167973387aSAl Viro path->mnt = NULL; 6177973387aSAl Viro path->dentry = NULL; 6187973387aSAl Viro return false; 6197973387aSAl Viro } 6207973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6217973387aSAl Viro path->dentry = NULL; 6227973387aSAl Viro return false; 6237973387aSAl Viro } 6247973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6257973387aSAl Viro } 6267973387aSAl Viro 6277973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6287973387aSAl Viro { 6297973387aSAl Viro int i; 6307973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6317973387aSAl Viro struct saved *last = nd->stack + i; 6327973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6337973387aSAl Viro drop_links(nd); 6347973387aSAl Viro nd->depth = i + 1; 6357973387aSAl Viro return false; 6367973387aSAl Viro } 6377973387aSAl Viro } 6387973387aSAl Viro return true; 6397973387aSAl Viro } 6407973387aSAl Viro 641ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 642ee594bffSAl Viro { 643adb21d2bSAleksa Sarai /* 644adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 645adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 646adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 647adb21d2bSAleksa Sarai */ 648adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 649adb21d2bSAleksa Sarai return false; 650adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 651ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 652ee594bffSAl Viro return true; 65384a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 654ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 655ee594bffSAl Viro } 656ee594bffSAl Viro 65719660af7SAl Viro /* 65831e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65919660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 66019660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 66157e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66219660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66319660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66419660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66519660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66631e6b01fSNick Piggin */ 66731e6b01fSNick Piggin 66831e6b01fSNick Piggin /** 66919660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 67019660af7SAl Viro * @nd: nameidata pathwalk data 67139191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67231e6b01fSNick Piggin * 6734675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6744675ac39SAl Viro * for ref-walk mode. 6754675ac39SAl Viro * Must be called from rcu-walk context. 6767973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6777973387aSAl Viro * terminate_walk(). 67831e6b01fSNick Piggin */ 6794675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 68031e6b01fSNick Piggin { 68131e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68231e6b01fSNick Piggin 68331e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 684e5c832d5SLinus Torvalds 685e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6867973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6874675ac39SAl Viro goto out1; 68884a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 68984a2bd39SAl Viro goto out; 690ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 6914675ac39SAl Viro goto out; 6924675ac39SAl Viro rcu_read_unlock(); 6934675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6944675ac39SAl Viro return 0; 6954675ac39SAl Viro 69684a2bd39SAl Viro out1: 6974675ac39SAl Viro nd->path.mnt = NULL; 6984675ac39SAl Viro nd->path.dentry = NULL; 6994675ac39SAl Viro out: 7004675ac39SAl Viro rcu_read_unlock(); 7014675ac39SAl Viro return -ECHILD; 7024675ac39SAl Viro } 7034675ac39SAl Viro 7044675ac39SAl Viro /** 7054675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7064675ac39SAl Viro * @nd: nameidata pathwalk data 7074675ac39SAl Viro * @dentry: child of nd->path.dentry 7084675ac39SAl Viro * @seq: seq number to check dentry against 7094675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7104675ac39SAl Viro * 7114675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7124675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7134675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7144675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7154675ac39SAl Viro * terminate_walk(). 7164675ac39SAl Viro */ 7174675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7184675ac39SAl Viro { 7194675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7204675ac39SAl Viro 7214675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7224675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7234675ac39SAl Viro goto out2; 7247973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7257973387aSAl Viro goto out2; 7264675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7277973387aSAl Viro goto out1; 72848a066e7SAl Viro 72915570086SLinus Torvalds /* 7304675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7314675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7324675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7334675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7344675ac39SAl Viro * know the parent must still be valid if the child sequence number is 73515570086SLinus Torvalds */ 7364675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 737e5c832d5SLinus Torvalds goto out; 73884a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 73984a2bd39SAl Viro goto out_dput; 740e5c832d5SLinus Torvalds /* 741e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 742e5c832d5SLinus Torvalds * still valid and get it if required. 743e5c832d5SLinus Torvalds */ 74484a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 74584a2bd39SAl Viro goto out_dput; 7468b61e74fSAl Viro rcu_read_unlock(); 74731e6b01fSNick Piggin return 0; 74819660af7SAl Viro 7497973387aSAl Viro out2: 7507973387aSAl Viro nd->path.mnt = NULL; 7517973387aSAl Viro out1: 7527973387aSAl Viro nd->path.dentry = NULL; 753e5c832d5SLinus Torvalds out: 7548b61e74fSAl Viro rcu_read_unlock(); 75584a2bd39SAl Viro return -ECHILD; 75684a2bd39SAl Viro out_dput: 75784a2bd39SAl Viro rcu_read_unlock(); 75884a2bd39SAl Viro dput(dentry); 75931e6b01fSNick Piggin return -ECHILD; 76031e6b01fSNick Piggin } 76131e6b01fSNick Piggin 7624ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76334286d66SNick Piggin { 764a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7654ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 766a89f8337SAl Viro else 767a89f8337SAl Viro return 1; 76834286d66SNick Piggin } 76934286d66SNick Piggin 7709f1fafeeSAl Viro /** 7719f1fafeeSAl Viro * complete_walk - successful completion of path walk 7729f1fafeeSAl Viro * @nd: pointer nameidata 77339159de2SJeff Layton * 7749f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7759f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7769f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7779f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7789f1fafeeSAl Viro * need to drop nd->path. 77939159de2SJeff Layton */ 7809f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 78139159de2SJeff Layton { 78216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78339159de2SJeff Layton int status; 78439159de2SJeff Layton 7859f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 786adb21d2bSAleksa Sarai /* 787adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 788adb21d2bSAleksa Sarai * externally-managed nd->root. 789adb21d2bSAleksa Sarai */ 790adb21d2bSAleksa Sarai if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) 7919f1fafeeSAl Viro nd->root.mnt = NULL; 7924675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 79348a066e7SAl Viro return -ECHILD; 79448a066e7SAl Viro } 7959f1fafeeSAl Viro 796adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 797adb21d2bSAleksa Sarai /* 798adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 799adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 800adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 801adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 802adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 803adb21d2bSAleksa Sarai * 804adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 805adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 806adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 807adb21d2bSAleksa Sarai * requested root to userspace. 808adb21d2bSAleksa Sarai * 809adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 810adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 811adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 812adb21d2bSAleksa Sarai */ 813adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 814adb21d2bSAleksa Sarai return -EXDEV; 815adb21d2bSAleksa Sarai } 816adb21d2bSAleksa Sarai 81716c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 81839159de2SJeff Layton return 0; 81939159de2SJeff Layton 820ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 82116c2cd71SAl Viro return 0; 82216c2cd71SAl Viro 823ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 82439159de2SJeff Layton if (status > 0) 82539159de2SJeff Layton return 0; 82639159de2SJeff Layton 82716c2cd71SAl Viro if (!status) 82839159de2SJeff Layton status = -ESTALE; 82916c2cd71SAl Viro 83039159de2SJeff Layton return status; 83139159de2SJeff Layton } 83239159de2SJeff Layton 833740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 8342a737871SAl Viro { 83531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8369e6697e2SAl Viro 837adb21d2bSAleksa Sarai /* 838adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 839adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 840adb21d2bSAleksa Sarai * from the dirfd. 841adb21d2bSAleksa Sarai */ 842adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 843adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 844adb21d2bSAleksa Sarai 8459e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8468f47a016SAl Viro unsigned seq; 847c28cc364SNick Piggin 848c28cc364SNick Piggin do { 849c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 85031e6b01fSNick Piggin nd->root = fs->root; 8518f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 852c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8539e6697e2SAl Viro } else { 8549e6697e2SAl Viro get_fs_root(fs, &nd->root); 85584a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 8569e6697e2SAl Viro } 857740a1678SAleksa Sarai return 0; 85831e6b01fSNick Piggin } 85931e6b01fSNick Piggin 8607b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8617b9337aaSNick Piggin struct nameidata *nd) 862051d3812SIan Kent { 86331e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8644ac91378SJan Blunck dput(nd->path.dentry); 86531e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8664ac91378SJan Blunck mntput(nd->path.mnt); 8679a229683SHuang Shijie } 86831e6b01fSNick Piggin nd->path.mnt = path->mnt; 8694ac91378SJan Blunck nd->path.dentry = path->dentry; 870051d3812SIan Kent } 871051d3812SIan Kent 872248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 873248fb5b9SAl Viro { 874adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 875adb21d2bSAleksa Sarai return -EXDEV; 87672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 87772ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 87872ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 87972ba2929SAleksa Sarai return -EXDEV; 88072ba2929SAleksa Sarai } 881740a1678SAleksa Sarai if (!nd->root.mnt) { 882740a1678SAleksa Sarai int error = set_root(nd); 883740a1678SAleksa Sarai if (error) 884740a1678SAleksa Sarai return error; 885740a1678SAleksa Sarai } 886248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 887248fb5b9SAl Viro struct dentry *d; 888248fb5b9SAl Viro nd->path = nd->root; 889248fb5b9SAl Viro d = nd->path.dentry; 890248fb5b9SAl Viro nd->inode = d->d_inode; 891248fb5b9SAl Viro nd->seq = nd->root_seq; 892248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 893248fb5b9SAl Viro return -ECHILD; 894248fb5b9SAl Viro } else { 895248fb5b9SAl Viro path_put(&nd->path); 896248fb5b9SAl Viro nd->path = nd->root; 897248fb5b9SAl Viro path_get(&nd->path); 898248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 899248fb5b9SAl Viro } 900248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 901248fb5b9SAl Viro return 0; 902248fb5b9SAl Viro } 903248fb5b9SAl Viro 904b5fb63c1SChristoph Hellwig /* 9056b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 906b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 907b5fb63c1SChristoph Hellwig */ 9081bc82070SAleksa Sarai int nd_jump_link(struct path *path) 909b5fb63c1SChristoph Hellwig { 9104b99d499SAleksa Sarai int error = -ELOOP; 9116e77137bSAl Viro struct nameidata *nd = current->nameidata; 912b5fb63c1SChristoph Hellwig 9134b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9144b99d499SAleksa Sarai goto err; 9154b99d499SAleksa Sarai 91672ba2929SAleksa Sarai error = -EXDEV; 91772ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 91872ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 91972ba2929SAleksa Sarai goto err; 92072ba2929SAleksa Sarai } 921adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 922adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 923adb21d2bSAleksa Sarai goto err; 92472ba2929SAleksa Sarai 9254b99d499SAleksa Sarai path_put(&nd->path); 926b5fb63c1SChristoph Hellwig nd->path = *path; 927b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 928b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 9291bc82070SAleksa Sarai return 0; 9304b99d499SAleksa Sarai 9314b99d499SAleksa Sarai err: 9324b99d499SAleksa Sarai path_put(path); 9334b99d499SAleksa Sarai return error; 934b5fb63c1SChristoph Hellwig } 935b5fb63c1SChristoph Hellwig 936b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 937574197e0SAl Viro { 93821c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 939fceef393SAl Viro do_delayed_call(&last->done); 9406548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 941b9ff4429SAl Viro path_put(&last->link); 942574197e0SAl Viro } 943574197e0SAl Viro 944561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 945561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 94630aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 94730aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 948800179c9SKees Cook 949800179c9SKees Cook /** 950800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 95155852635SRandy Dunlap * @nd: nameidata pathwalk data 952800179c9SKees Cook * 953800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 954800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 955800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 956800179c9SKees Cook * processes from failing races against path names that may change out 957800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 958800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 959800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 960800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 961800179c9SKees Cook * 962800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 963800179c9SKees Cook */ 964ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 965800179c9SKees Cook { 966800179c9SKees Cook const struct inode *parent; 9672d7f9e2aSSeth Forshee kuid_t puid; 968800179c9SKees Cook 969800179c9SKees Cook if (!sysctl_protected_symlinks) 970800179c9SKees Cook return 0; 971800179c9SKees Cook 972800179c9SKees Cook /* Allowed if owner and follower match. */ 97381abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 974800179c9SKees Cook return 0; 975800179c9SKees Cook 976800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 977aa65fa35SAl Viro parent = nd->inode; 978800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 979800179c9SKees Cook return 0; 980800179c9SKees Cook 981800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9822d7f9e2aSSeth Forshee puid = parent->i_uid; 9832d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 984800179c9SKees Cook return 0; 985800179c9SKees Cook 98631956502SAl Viro if (nd->flags & LOOKUP_RCU) 98731956502SAl Viro return -ECHILD; 98831956502SAl Viro 989ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 990245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 991800179c9SKees Cook return -EACCES; 992800179c9SKees Cook } 993800179c9SKees Cook 994800179c9SKees Cook /** 995800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 996800179c9SKees Cook * @inode: the source inode to hardlink from 997800179c9SKees Cook * 998800179c9SKees Cook * Return false if at least one of the following conditions: 999800179c9SKees Cook * - inode is not a regular file 1000800179c9SKees Cook * - inode is setuid 1001800179c9SKees Cook * - inode is setgid and group-exec 1002800179c9SKees Cook * - access failure for read and write 1003800179c9SKees Cook * 1004800179c9SKees Cook * Otherwise returns true. 1005800179c9SKees Cook */ 1006800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 1007800179c9SKees Cook { 1008800179c9SKees Cook umode_t mode = inode->i_mode; 1009800179c9SKees Cook 1010800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1011800179c9SKees Cook if (!S_ISREG(mode)) 1012800179c9SKees Cook return false; 1013800179c9SKees Cook 1014800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1015800179c9SKees Cook if (mode & S_ISUID) 1016800179c9SKees Cook return false; 1017800179c9SKees Cook 1018800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1019800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1020800179c9SKees Cook return false; 1021800179c9SKees Cook 1022800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1023800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 1024800179c9SKees Cook return false; 1025800179c9SKees Cook 1026800179c9SKees Cook return true; 1027800179c9SKees Cook } 1028800179c9SKees Cook 1029800179c9SKees Cook /** 1030800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1031800179c9SKees Cook * @link: the source to hardlink from 1032800179c9SKees Cook * 1033800179c9SKees Cook * Block hardlink when all of: 1034800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1035800179c9SKees Cook * - fsuid does not match inode 1036800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1037f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1038800179c9SKees Cook * 1039800179c9SKees Cook * Returns 0 if successful, -ve on error. 1040800179c9SKees Cook */ 1041800179c9SKees Cook static int may_linkat(struct path *link) 1042800179c9SKees Cook { 1043593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1044593d1ce8SEric W. Biederman 1045593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1046593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 1047593d1ce8SEric W. Biederman return -EOVERFLOW; 1048800179c9SKees Cook 1049800179c9SKees Cook if (!sysctl_protected_hardlinks) 1050800179c9SKees Cook return 0; 1051800179c9SKees Cook 1052800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1053800179c9SKees Cook * otherwise, it must be a safe source. 1054800179c9SKees Cook */ 1055cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1056800179c9SKees Cook return 0; 1057800179c9SKees Cook 1058245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1059800179c9SKees Cook return -EPERM; 1060800179c9SKees Cook } 1061800179c9SKees Cook 106230aba665SSalvatore Mesoraca /** 106330aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 106430aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 106530aba665SSalvatore Mesoraca * exist. 1066d0cb5018SAl Viro * @dir_mode: mode bits of directory 1067d0cb5018SAl Viro * @dir_uid: owner of directory 106830aba665SSalvatore Mesoraca * @inode: the inode of the file to open 106930aba665SSalvatore Mesoraca * 107030aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 107130aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 107230aba665SSalvatore Mesoraca * - the file already exists 107330aba665SSalvatore Mesoraca * - we are in a sticky directory 107430aba665SSalvatore Mesoraca * - we don't own the file 107530aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 107630aba665SSalvatore Mesoraca * - the directory is world writable 107730aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 107830aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 107930aba665SSalvatore Mesoraca * be enough. 108030aba665SSalvatore Mesoraca * 108130aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 108230aba665SSalvatore Mesoraca */ 1083d0cb5018SAl Viro static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid, 108430aba665SSalvatore Mesoraca struct inode * const inode) 108530aba665SSalvatore Mesoraca { 108630aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 108730aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1088d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1089d0cb5018SAl Viro uid_eq(inode->i_uid, dir_uid) || 109030aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 109130aba665SSalvatore Mesoraca return 0; 109230aba665SSalvatore Mesoraca 1093d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1094d0cb5018SAl Viro (dir_mode & 0020 && 109530aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 109630aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1097245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1098245d7369SKees Cook "sticky_create_fifo" : 1099245d7369SKees Cook "sticky_create_regular"; 1100245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 110130aba665SSalvatore Mesoraca return -EACCES; 110230aba665SSalvatore Mesoraca } 110330aba665SSalvatore Mesoraca return 0; 110430aba665SSalvatore Mesoraca } 110530aba665SSalvatore Mesoraca 1106f015f126SDavid Howells /* 1107f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1108f015f126SDavid Howells * 1109f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1110f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1111f015f126SDavid Howells * Up is towards /. 1112f015f126SDavid Howells * 1113f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1114f015f126SDavid Howells * root. 1115f015f126SDavid Howells */ 1116bab77ebfSAl Viro int follow_up(struct path *path) 11171da177e4SLinus Torvalds { 11180714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11190714a533SAl Viro struct mount *parent; 11201da177e4SLinus Torvalds struct dentry *mountpoint; 112199b7db7bSNick Piggin 112248a066e7SAl Viro read_seqlock_excl(&mount_lock); 11230714a533SAl Viro parent = mnt->mnt_parent; 11243c0a6163SAl Viro if (parent == mnt) { 112548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11261da177e4SLinus Torvalds return 0; 11271da177e4SLinus Torvalds } 11280714a533SAl Viro mntget(&parent->mnt); 1129a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 113048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1131bab77ebfSAl Viro dput(path->dentry); 1132bab77ebfSAl Viro path->dentry = mountpoint; 1133bab77ebfSAl Viro mntput(path->mnt); 11340714a533SAl Viro path->mnt = &parent->mnt; 11351da177e4SLinus Torvalds return 1; 11361da177e4SLinus Torvalds } 11374d359507SAl Viro EXPORT_SYMBOL(follow_up); 11381da177e4SLinus Torvalds 1139b5c84bf6SNick Piggin /* 11409875cf80SDavid Howells * Perform an automount 11419875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11429875cf80SDavid Howells * were called with. 11431da177e4SLinus Torvalds */ 11441c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 114531e6b01fSNick Piggin { 114625e195aaSAl Viro struct dentry *dentry = path->dentry; 11479875cf80SDavid Howells 11480ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11490ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11500ec26fd0SMiklos Szeredi * the name. 11510ec26fd0SMiklos Szeredi * 11520ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11535a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11540ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11550ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11560ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11570ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11585a30d8a2SDavid Howells */ 11591c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11605d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 116125e195aaSAl Viro dentry->d_inode) 11629875cf80SDavid Howells return -EISDIR; 11630ec26fd0SMiklos Szeredi 11641c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 11659875cf80SDavid Howells return -ELOOP; 11669875cf80SDavid Howells 116725e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1168ea5b778aSDavid Howells } 11699875cf80SDavid Howells 11709875cf80SDavid Howells /* 11719875cf80SDavid Howells * Handle a dentry that is managed in some way. 1172cc53ce53SDavid Howells * - Flagged for transit management (autofs) 11739875cf80SDavid Howells * - Flagged as mountpoint 11749875cf80SDavid Howells * - Flagged as automount point 11759875cf80SDavid Howells * 11769875cf80SDavid Howells * This may only be called in refwalk mode. 1177d41efb52SAl Viro * On success path->dentry is known positive. 11789875cf80SDavid Howells * 11799875cf80SDavid Howells * Serialization is taken care of in namespace.c 11809875cf80SDavid Howells */ 1181756daf26SNeilBrown static int follow_managed(struct path *path, struct nameidata *nd) 11829875cf80SDavid Howells { 11838aef1884SAl Viro struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */ 1184d41efb52SAl Viro unsigned flags; 11859875cf80SDavid Howells bool need_mntput = false; 11868aef1884SAl Viro int ret = 0; 11879875cf80SDavid Howells 11889875cf80SDavid Howells /* Given that we're not holding a lock here, we retain the value in a 11899875cf80SDavid Howells * local variable for each dentry as we look at it so that we don't see 11909875cf80SDavid Howells * the components of that value change under us */ 11912fa6b1e0SAl Viro while (flags = smp_load_acquire(&path->dentry->d_flags), 1192d41efb52SAl Viro unlikely(flags & DCACHE_MANAGED_DENTRY)) { 1193cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1194cc53ce53SDavid Howells * being held. */ 1195d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1196cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1197cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1198fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1199508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1200cc53ce53SDavid Howells if (ret < 0) 12018aef1884SAl Viro break; 1202cc53ce53SDavid Howells } 1203cc53ce53SDavid Howells 12049875cf80SDavid Howells /* Transit to a mounted filesystem. */ 1205d41efb52SAl Viro if (flags & DCACHE_MOUNTED) { 12069875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12079875cf80SDavid Howells if (mounted) { 12089875cf80SDavid Howells dput(path->dentry); 12099875cf80SDavid Howells if (need_mntput) 1210463ffb2eSAl Viro mntput(path->mnt); 1211463ffb2eSAl Viro path->mnt = mounted; 1212463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12139875cf80SDavid Howells need_mntput = true; 12149875cf80SDavid Howells continue; 1215463ffb2eSAl Viro } 1216463ffb2eSAl Viro 12179875cf80SDavid Howells /* Something is mounted on this dentry in another 12189875cf80SDavid Howells * namespace and/or whatever was mounted there in this 121948a066e7SAl Viro * namespace got unmounted before lookup_mnt() could 122048a066e7SAl Viro * get it */ 12211da177e4SLinus Torvalds } 12229875cf80SDavid Howells 12239875cf80SDavid Howells /* Handle an automount point */ 1224d41efb52SAl Viro if (flags & DCACHE_NEED_AUTOMOUNT) { 12251c9f5e06SAl Viro ret = follow_automount(path, &nd->total_link_count, 12261c9f5e06SAl Viro nd->flags); 12279875cf80SDavid Howells if (ret < 0) 12288aef1884SAl Viro break; 12299875cf80SDavid Howells continue; 12309875cf80SDavid Howells } 12319875cf80SDavid Howells 12329875cf80SDavid Howells /* We didn't change the current path point */ 12339875cf80SDavid Howells break; 12349875cf80SDavid Howells } 12358aef1884SAl Viro 123672ba2929SAleksa Sarai if (need_mntput) { 123772ba2929SAleksa Sarai if (path->mnt == mnt) 12388aef1884SAl Viro mntput(path->mnt); 123972ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 124072ba2929SAleksa Sarai ret = -EXDEV; 124172ba2929SAleksa Sarai else 12428402752eSAl Viro nd->flags |= LOOKUP_JUMPED; 124372ba2929SAleksa Sarai } 1244d41efb52SAl Viro if (ret == -EISDIR || !ret) 1245d41efb52SAl Viro ret = 1; 1246d41efb52SAl Viro if (ret > 0 && unlikely(d_flags_negative(flags))) 1247d41efb52SAl Viro ret = -ENOENT; 1248a9dc1494SAl Viro if (unlikely(ret < 0)) { 1249a9dc1494SAl Viro dput(path->dentry); 1250a9dc1494SAl Viro if (path->mnt != nd->path.mnt) 1251a9dc1494SAl Viro mntput(path->mnt); 1252a9dc1494SAl Viro } 12538402752eSAl Viro return ret; 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds 1256cc53ce53SDavid Howells int follow_down_one(struct path *path) 12571da177e4SLinus Torvalds { 12581da177e4SLinus Torvalds struct vfsmount *mounted; 12591da177e4SLinus Torvalds 12601c755af4SAl Viro mounted = lookup_mnt(path); 12611da177e4SLinus Torvalds if (mounted) { 12629393bd07SAl Viro dput(path->dentry); 12639393bd07SAl Viro mntput(path->mnt); 12649393bd07SAl Viro path->mnt = mounted; 12659393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12661da177e4SLinus Torvalds return 1; 12671da177e4SLinus Torvalds } 12681da177e4SLinus Torvalds return 0; 12691da177e4SLinus Torvalds } 12704d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12711da177e4SLinus Torvalds 1272fb5f51c7SIan Kent static inline int managed_dentry_rcu(const struct path *path) 127362a7375eSIan Kent { 1274fb5f51c7SIan Kent return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ? 1275fb5f51c7SIan Kent path->dentry->d_op->d_manage(path, true) : 0; 127662a7375eSIan Kent } 127762a7375eSIan Kent 12789875cf80SDavid Howells /* 1279287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1280287548e4SAl Viro * we meet a managed dentry that would need blocking. 12819875cf80SDavid Howells */ 12829875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1283254cf582SAl Viro struct inode **inode, unsigned *seqp) 12849875cf80SDavid Howells { 128562a7375eSIan Kent for (;;) { 1286c7105365SAl Viro struct mount *mounted; 128762a7375eSIan Kent /* 128862a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 128962a7375eSIan Kent * that wants to block transit. 129062a7375eSIan Kent */ 1291fb5f51c7SIan Kent switch (managed_dentry_rcu(path)) { 1292b8faf035SNeilBrown case -ECHILD: 1293b8faf035SNeilBrown default: 1294ab90911fSDavid Howells return false; 1295b8faf035SNeilBrown case -EISDIR: 1296b8faf035SNeilBrown return true; 1297b8faf035SNeilBrown case 0: 1298b8faf035SNeilBrown break; 1299b8faf035SNeilBrown } 130062a7375eSIan Kent 130162a7375eSIan Kent if (!d_mountpoint(path->dentry)) 1302b8faf035SNeilBrown return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 130362a7375eSIan Kent 1304474279dcSAl Viro mounted = __lookup_mnt(path->mnt, path->dentry); 13059875cf80SDavid Howells if (!mounted) 13069875cf80SDavid Howells break; 130772ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 130872ba2929SAleksa Sarai return false; 1309c7105365SAl Viro path->mnt = &mounted->mnt; 1310c7105365SAl Viro path->dentry = mounted->mnt.mnt_root; 1311a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1312254cf582SAl Viro *seqp = read_seqcount_begin(&path->dentry->d_seq); 131359430262SLinus Torvalds /* 131459430262SLinus Torvalds * Update the inode too. We don't need to re-check the 131559430262SLinus Torvalds * dentry sequence number here after this d_inode read, 131659430262SLinus Torvalds * because a mount-point is always pinned. 131759430262SLinus Torvalds */ 131859430262SLinus Torvalds *inode = path->dentry->d_inode; 13199875cf80SDavid Howells } 1320f5be3e29SAl Viro return !read_seqretry(&mount_lock, nd->m_seq) && 1321b8faf035SNeilBrown !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT); 1322287548e4SAl Viro } 1323287548e4SAl Viro 1324db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1325db3c9adeSAl Viro struct path *path, struct inode **inode, 1326db3c9adeSAl Viro unsigned int *seqp) 1327bd7c4b50SAl Viro { 1328db3c9adeSAl Viro int ret; 1329bd7c4b50SAl Viro 1330db3c9adeSAl Viro path->mnt = nd->path.mnt; 1331db3c9adeSAl Viro path->dentry = dentry; 1332c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1333c153007bSAl Viro unsigned int seq = *seqp; 1334c153007bSAl Viro if (unlikely(!*inode)) 1335c153007bSAl Viro return -ENOENT; 1336c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 1337c153007bSAl Viro return 1; 1338c153007bSAl Viro if (unlazy_child(nd, dentry, seq)) 1339c153007bSAl Viro return -ECHILD; 1340c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1341c153007bSAl Viro path->mnt = nd->path.mnt; 1342c153007bSAl Viro path->dentry = dentry; 1343c153007bSAl Viro } 1344db3c9adeSAl Viro ret = follow_managed(path, nd); 1345bd7c4b50SAl Viro if (likely(ret >= 0)) { 1346bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1347bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1348bd7c4b50SAl Viro } 1349bd7c4b50SAl Viro return ret; 1350bd7c4b50SAl Viro } 1351bd7c4b50SAl Viro 135231e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 135331e6b01fSNick Piggin { 13544023bfc9SAl Viro struct inode *inode = nd->inode; 135531e6b01fSNick Piggin 135631e6b01fSNick Piggin while (1) { 1357adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1358adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1359adb21d2bSAleksa Sarai return -ECHILD; 136031e6b01fSNick Piggin break; 1361adb21d2bSAleksa Sarai } 136231e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 136331e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 136431e6b01fSNick Piggin struct dentry *parent = old->d_parent; 136531e6b01fSNick Piggin unsigned seq; 136631e6b01fSNick Piggin 13674023bfc9SAl Viro inode = parent->d_inode; 136831e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1369aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1370aed434adSAl Viro return -ECHILD; 137131e6b01fSNick Piggin nd->path.dentry = parent; 137231e6b01fSNick Piggin nd->seq = seq; 1373397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 13742b98149cSAleksa Sarai return -ECHILD; 137531e6b01fSNick Piggin break; 1376aed434adSAl Viro } else { 1377aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1378aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1379aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1380aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1381aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1382aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1383aed434adSAl Viro return -ECHILD; 1384aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 138531e6b01fSNick Piggin break; 138672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 138772ba2929SAleksa Sarai return -ECHILD; 1388aed434adSAl Viro /* we know that mountpoint was pinned */ 1389aed434adSAl Viro nd->path.dentry = mountpoint; 1390aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1391aed434adSAl Viro inode = inode2; 1392aed434adSAl Viro nd->seq = seq; 139331e6b01fSNick Piggin } 1394aed434adSAl Viro } 1395aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1396b37199e6SAl Viro struct mount *mounted; 1397b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1398aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1399aed434adSAl Viro return -ECHILD; 1400b37199e6SAl Viro if (!mounted) 1401b37199e6SAl Viro break; 140272ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 140372ba2929SAleksa Sarai return -ECHILD; 1404b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1405b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 14064023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1407b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1408b37199e6SAl Viro } 14094023bfc9SAl Viro nd->inode = inode; 141031e6b01fSNick Piggin return 0; 141131e6b01fSNick Piggin } 141231e6b01fSNick Piggin 14139875cf80SDavid Howells /* 1414cc53ce53SDavid Howells * Follow down to the covering mount currently visible to userspace. At each 1415cc53ce53SDavid Howells * point, the filesystem owning that dentry may be queried as to whether the 1416cc53ce53SDavid Howells * caller is permitted to proceed or not. 1417cc53ce53SDavid Howells */ 14187cc90cc3SAl Viro int follow_down(struct path *path) 1419cc53ce53SDavid Howells { 1420cc53ce53SDavid Howells unsigned managed; 1421cc53ce53SDavid Howells int ret; 1422cc53ce53SDavid Howells 14236aa7de05SMark Rutland while (managed = READ_ONCE(path->dentry->d_flags), 1424cc53ce53SDavid Howells unlikely(managed & DCACHE_MANAGED_DENTRY)) { 1425cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1426cc53ce53SDavid Howells * being held. 1427cc53ce53SDavid Howells * 1428cc53ce53SDavid Howells * We indicate to the filesystem if someone is trying to mount 1429cc53ce53SDavid Howells * something here. This gives autofs the chance to deny anyone 1430cc53ce53SDavid Howells * other than its daemon the right to mount on its 1431cc53ce53SDavid Howells * superstructure. 1432cc53ce53SDavid Howells * 1433cc53ce53SDavid Howells * The filesystem may sleep at this point. 1434cc53ce53SDavid Howells */ 1435cc53ce53SDavid Howells if (managed & DCACHE_MANAGE_TRANSIT) { 1436cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op); 1437cc53ce53SDavid Howells BUG_ON(!path->dentry->d_op->d_manage); 1438fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1439cc53ce53SDavid Howells if (ret < 0) 1440cc53ce53SDavid Howells return ret == -EISDIR ? 0 : ret; 1441cc53ce53SDavid Howells } 1442cc53ce53SDavid Howells 1443cc53ce53SDavid Howells /* Transit to a mounted filesystem. */ 1444cc53ce53SDavid Howells if (managed & DCACHE_MOUNTED) { 1445cc53ce53SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 1446cc53ce53SDavid Howells if (!mounted) 1447cc53ce53SDavid Howells break; 1448cc53ce53SDavid Howells dput(path->dentry); 1449cc53ce53SDavid Howells mntput(path->mnt); 1450cc53ce53SDavid Howells path->mnt = mounted; 1451cc53ce53SDavid Howells path->dentry = dget(mounted->mnt_root); 1452cc53ce53SDavid Howells continue; 1453cc53ce53SDavid Howells } 1454cc53ce53SDavid Howells 1455cc53ce53SDavid Howells /* Don't handle automount points here */ 1456cc53ce53SDavid Howells break; 1457cc53ce53SDavid Howells } 1458cc53ce53SDavid Howells return 0; 1459cc53ce53SDavid Howells } 14604d359507SAl Viro EXPORT_SYMBOL(follow_down); 1461cc53ce53SDavid Howells 1462cc53ce53SDavid Howells /* 14639875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14649875cf80SDavid Howells */ 14659875cf80SDavid Howells static void follow_mount(struct path *path) 14669875cf80SDavid Howells { 14679875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14689875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14699875cf80SDavid Howells if (!mounted) 14709875cf80SDavid Howells break; 14719875cf80SDavid Howells dput(path->dentry); 14729875cf80SDavid Howells mntput(path->mnt); 14739875cf80SDavid Howells path->mnt = mounted; 14749875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14759875cf80SDavid Howells } 14769875cf80SDavid Howells } 14779875cf80SDavid Howells 1478eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1479eedf265aSEric W. Biederman { 1480eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1481eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1482eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1483eedf265aSEric W. Biederman dput(old); 1484eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1485eedf265aSEric W. Biederman return -ENOENT; 1486eedf265aSEric W. Biederman return 0; 1487eedf265aSEric W. Biederman } 1488eedf265aSEric W. Biederman 1489397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14901da177e4SLinus Torvalds { 14911da177e4SLinus Torvalds while (1) { 1492adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1493adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1494adb21d2bSAleksa Sarai return -EXDEV; 14951da177e4SLinus Torvalds break; 1496adb21d2bSAleksa Sarai } 14974ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1498eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1499eedf265aSEric W. Biederman if (ret) 1500eedf265aSEric W. Biederman return ret; 15011da177e4SLinus Torvalds break; 15021da177e4SLinus Torvalds } 15033088dd70SAl Viro if (!follow_up(&nd->path)) 15041da177e4SLinus Torvalds break; 150572ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 150672ba2929SAleksa Sarai return -EXDEV; 15071da177e4SLinus Torvalds } 150879ed0226SAl Viro follow_mount(&nd->path); 150931e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1510397d425dSEric W. Biederman return 0; 15111da177e4SLinus Torvalds } 15121da177e4SLinus Torvalds 15131da177e4SLinus Torvalds /* 1514f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1515f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1516baa03890SNick Piggin */ 1517e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1518e3c13928SAl Viro struct dentry *dir, 15196c51e513SAl Viro unsigned int flags) 1520baa03890SNick Piggin { 1521a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1522bad61189SMiklos Szeredi if (dentry) { 1523a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1524bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 152574ff0ffcSAl Viro if (!error) 15265542aa2fSEric W. Biederman d_invalidate(dentry); 1527bad61189SMiklos Szeredi dput(dentry); 152874ff0ffcSAl Viro return ERR_PTR(error); 1529bad61189SMiklos Szeredi } 1530bad61189SMiklos Szeredi } 1531baa03890SNick Piggin return dentry; 1532baa03890SNick Piggin } 1533baa03890SNick Piggin 1534baa03890SNick Piggin /* 1535a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1536a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1537a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1538a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1539a03ece5fSAl Viro * at all. 154044396f4bSJosef Bacik */ 1541a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1542a03ece5fSAl Viro struct dentry *base, unsigned int flags) 154344396f4bSJosef Bacik { 1544a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 154544396f4bSJosef Bacik struct dentry *old; 1546a03ece5fSAl Viro struct inode *dir = base->d_inode; 1547a03ece5fSAl Viro 1548a03ece5fSAl Viro if (dentry) 1549a03ece5fSAl Viro return dentry; 155044396f4bSJosef Bacik 155144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1552a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 155344396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1554a03ece5fSAl Viro 1555a03ece5fSAl Viro dentry = d_alloc(base, name); 1556a03ece5fSAl Viro if (unlikely(!dentry)) 1557a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 155844396f4bSJosef Bacik 155972bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 156044396f4bSJosef Bacik if (unlikely(old)) { 156144396f4bSJosef Bacik dput(dentry); 156244396f4bSJosef Bacik dentry = old; 156344396f4bSJosef Bacik } 156444396f4bSJosef Bacik return dentry; 156544396f4bSJosef Bacik } 156644396f4bSJosef Bacik 156720e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 156820e34357SAl Viro struct inode **inode, 1569254cf582SAl Viro unsigned *seqp) 15701da177e4SLinus Torvalds { 157131e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15725a18fff2SAl Viro int status = 1; 15739875cf80SDavid Howells 15743cac260aSAl Viro /* 1575b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15765d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15775d0f49c1SAl Viro * going to fall back to non-racy lookup. 1578b04f784eSNick Piggin */ 157931e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 158031e6b01fSNick Piggin unsigned seq; 1581da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15825d0f49c1SAl Viro if (unlikely(!dentry)) { 15834675ac39SAl Viro if (unlazy_walk(nd)) 158420e34357SAl Viro return ERR_PTR(-ECHILD); 158520e34357SAl Viro return NULL; 15865d0f49c1SAl Viro } 15875a18fff2SAl Viro 158812f8ad4bSLinus Torvalds /* 158912f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 159012f8ad4bSLinus Torvalds * the dentry name information from lookup. 159112f8ad4bSLinus Torvalds */ 159263afdfc7SDavid Howells *inode = d_backing_inode(dentry); 15935d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 159420e34357SAl Viro return ERR_PTR(-ECHILD); 159512f8ad4bSLinus Torvalds 159612f8ad4bSLinus Torvalds /* 159712f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 159812f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 159912f8ad4bSLinus Torvalds * 160012f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 160112f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 160212f8ad4bSLinus Torvalds */ 16035d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 160420e34357SAl Viro return ERR_PTR(-ECHILD); 16055a18fff2SAl Viro 1606254cf582SAl Viro *seqp = seq; 16074ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1608c153007bSAl Viro if (likely(status > 0)) 160920e34357SAl Viro return dentry; 16104675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 161120e34357SAl Viro return ERR_PTR(-ECHILD); 1612209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1613209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1614209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16155a18fff2SAl Viro } else { 1616e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 161781e6f520SAl Viro if (unlikely(!dentry)) 161820e34357SAl Viro return NULL; 16194ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16205d0f49c1SAl Viro } 16215a18fff2SAl Viro if (unlikely(status <= 0)) { 1622e9742b53SAl Viro if (!status) 16235d0f49c1SAl Viro d_invalidate(dentry); 16245a18fff2SAl Viro dput(dentry); 162520e34357SAl Viro return ERR_PTR(status); 16265a18fff2SAl Viro } 162720e34357SAl Viro return dentry; 1628697f514dSMiklos Szeredi } 1629697f514dSMiklos Szeredi 1630697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 163188d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1632e3c13928SAl Viro struct dentry *dir, 1633e3c13928SAl Viro unsigned int flags) 1634697f514dSMiklos Szeredi { 163588d8331aSAl Viro struct dentry *dentry, *old; 16361936386eSAl Viro struct inode *inode = dir->d_inode; 1637d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16381936386eSAl Viro 16391936386eSAl Viro /* Don't go there if it's already dead */ 164094bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 164188d8331aSAl Viro return ERR_PTR(-ENOENT); 164294bdd655SAl Viro again: 1643d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 164494bdd655SAl Viro if (IS_ERR(dentry)) 164588d8331aSAl Viro return dentry; 164694bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1647949a852eSAl Viro int error = d_revalidate(dentry, flags); 1648949a852eSAl Viro if (unlikely(error <= 0)) { 164994bdd655SAl Viro if (!error) { 1650949a852eSAl Viro d_invalidate(dentry); 1651949a852eSAl Viro dput(dentry); 165294bdd655SAl Viro goto again; 165394bdd655SAl Viro } 165494bdd655SAl Viro dput(dentry); 1655949a852eSAl Viro dentry = ERR_PTR(error); 1656949a852eSAl Viro } 165794bdd655SAl Viro } else { 16581936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 165985c7f810SAl Viro d_lookup_done(dentry); 16601936386eSAl Viro if (unlikely(old)) { 16611936386eSAl Viro dput(dentry); 16621936386eSAl Viro dentry = old; 1663949a852eSAl Viro } 1664949a852eSAl Viro } 1665e3c13928SAl Viro return dentry; 16661da177e4SLinus Torvalds } 16671da177e4SLinus Torvalds 166888d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 166988d8331aSAl Viro struct dentry *dir, 167088d8331aSAl Viro unsigned int flags) 167188d8331aSAl Viro { 167288d8331aSAl Viro struct inode *inode = dir->d_inode; 167388d8331aSAl Viro struct dentry *res; 167488d8331aSAl Viro inode_lock_shared(inode); 167588d8331aSAl Viro res = __lookup_slow(name, dir, flags); 167688d8331aSAl Viro inode_unlock_shared(inode); 167788d8331aSAl Viro return res; 167888d8331aSAl Viro } 167988d8331aSAl Viro 168052094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 168152094c8aSAl Viro { 168252094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16834ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 168452094c8aSAl Viro if (err != -ECHILD) 168552094c8aSAl Viro return err; 16864675ac39SAl Viro if (unlazy_walk(nd)) 168752094c8aSAl Viro return -ECHILD; 168852094c8aSAl Viro } 16894ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 169052094c8aSAl Viro } 169152094c8aSAl Viro 16929856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16939856fa1bSAl Viro { 16949856fa1bSAl Viro if (type == LAST_DOTDOT) { 1695740a1678SAleksa Sarai int error = 0; 1696740a1678SAleksa Sarai 1697740a1678SAleksa Sarai if (!nd->root.mnt) { 1698740a1678SAleksa Sarai error = set_root(nd); 1699740a1678SAleksa Sarai if (error) 1700740a1678SAleksa Sarai return error; 1701740a1678SAleksa Sarai } 1702ab87f9a5SAleksa Sarai if (nd->flags & LOOKUP_RCU) 1703ab87f9a5SAleksa Sarai error = follow_dotdot_rcu(nd); 1704ab87f9a5SAleksa Sarai else 1705ab87f9a5SAleksa Sarai error = follow_dotdot(nd); 1706ab87f9a5SAleksa Sarai if (error) 1707ab87f9a5SAleksa Sarai return error; 1708ab87f9a5SAleksa Sarai 1709ab87f9a5SAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1710ab87f9a5SAleksa Sarai /* 1711ab87f9a5SAleksa Sarai * If there was a racing rename or mount along our 1712ab87f9a5SAleksa Sarai * path, then we can't be sure that ".." hasn't jumped 1713ab87f9a5SAleksa Sarai * above nd->root (and so userspace should retry or use 1714ab87f9a5SAleksa Sarai * some fallback). 1715ab87f9a5SAleksa Sarai */ 1716ab87f9a5SAleksa Sarai smp_rmb(); 1717ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 1718ab87f9a5SAleksa Sarai return -EAGAIN; 1719ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 1720ab87f9a5SAleksa Sarai return -EAGAIN; 1721ab87f9a5SAleksa Sarai } 17229856fa1bSAl Viro } 17239856fa1bSAl Viro return 0; 17249856fa1bSAl Viro } 17259856fa1bSAl Viro 1726b1a81972SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 1727b1a81972SAl Viro 172806708adbSAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 1729b1a81972SAl Viro struct inode *inode, unsigned seq, int flags) 1730d63ff28fSAl Viro { 17311cf2665bSAl Viro struct saved *last; 1732ad6cc4c3SAl Viro const char *res; 1733ad6cc4c3SAl Viro int error; 1734ad6cc4c3SAl Viro 1735756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1736626de996SAl Viro path_to_nameidata(link, nd); 173706708adbSAl Viro return ERR_PTR(-ELOOP); 1738626de996SAl Viro } 1739bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1740cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1741cd179f44SAl Viro mntget(link->mnt); 17427973387aSAl Viro } 1743626de996SAl Viro error = nd_alloc_stack(nd); 1744626de996SAl Viro if (unlikely(error)) { 1745bc40aee0SAl Viro if (error == -ECHILD) { 1746ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1747ad1633a1SAl Viro drop_links(nd); 1748ad1633a1SAl Viro nd->depth = 0; 1749ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1750ad1633a1SAl Viro nd->path.mnt = NULL; 1751ad1633a1SAl Viro nd->path.dentry = NULL; 1752ad1633a1SAl Viro rcu_read_unlock(); 17534675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1754bc40aee0SAl Viro error = nd_alloc_stack(nd); 1755bc40aee0SAl Viro } 1756bc40aee0SAl Viro if (error) { 1757cd179f44SAl Viro path_put(link); 175806708adbSAl Viro return ERR_PTR(error); 1759626de996SAl Viro } 1760bc40aee0SAl Viro } 1761626de996SAl Viro 1762ab104923SAl Viro last = nd->stack + nd->depth++; 17631cf2665bSAl Viro last->link = *link; 1764fceef393SAl Viro clear_delayed_call(&last->done); 17650450b2d1SAl Viro last->seq = seq; 1766ad6cc4c3SAl Viro 1767b1a81972SAl Viro if (flags & WALK_TRAILING) { 1768ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1769ad6cc4c3SAl Viro if (unlikely(error)) 1770ad6cc4c3SAl Viro return ERR_PTR(error); 1771ad6cc4c3SAl Viro } 1772ad6cc4c3SAl Viro 1773ad6cc4c3SAl Viro if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) 1774ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1775ad6cc4c3SAl Viro 1776ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1777ad6cc4c3SAl Viro touch_atime(&last->link); 1778ad6cc4c3SAl Viro cond_resched(); 1779ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1780ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1781ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1782ad6cc4c3SAl Viro touch_atime(&last->link); 1783ad6cc4c3SAl Viro } 1784ad6cc4c3SAl Viro 1785ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1786ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1787ad6cc4c3SAl Viro if (unlikely(error)) 1788ad6cc4c3SAl Viro return ERR_PTR(error); 1789ad6cc4c3SAl Viro 1790ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1791ad6cc4c3SAl Viro if (!res) { 1792ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1793ad6cc4c3SAl Viro struct delayed_call *); 1794ad6cc4c3SAl Viro get = inode->i_op->get_link; 1795ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1796ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1797ad6cc4c3SAl Viro if (res == ERR_PTR(-ECHILD)) { 1798ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1799ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1800ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1801ad6cc4c3SAl Viro } 1802ad6cc4c3SAl Viro } else { 1803ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1804ad6cc4c3SAl Viro } 1805ad6cc4c3SAl Viro if (!res) 1806ad6cc4c3SAl Viro goto all_done; 1807ad6cc4c3SAl Viro if (IS_ERR(res)) 1808ad6cc4c3SAl Viro return res; 1809ad6cc4c3SAl Viro } 1810ad6cc4c3SAl Viro if (*res == '/') { 1811ad6cc4c3SAl Viro error = nd_jump_root(nd); 1812ad6cc4c3SAl Viro if (unlikely(error)) 1813ad6cc4c3SAl Viro return ERR_PTR(error); 1814ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1815ad6cc4c3SAl Viro ; 1816ad6cc4c3SAl Viro } 1817ad6cc4c3SAl Viro if (*res) 1818ad6cc4c3SAl Viro return res; 1819ad6cc4c3SAl Viro all_done: // pure jump 1820ad6cc4c3SAl Viro put_link(nd); 1821ad6cc4c3SAl Viro return NULL; 1822d63ff28fSAl Viro } 1823d63ff28fSAl Viro 18243ddcd056SLinus Torvalds /* 18253ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18263ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18273ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18283ddcd056SLinus Torvalds * for the common case. 18293ddcd056SLinus Torvalds */ 1830b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1831cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 18323ddcd056SLinus Torvalds { 1833cbae4d12SAl Viro struct path path; 1834cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1835cbae4d12SAl Viro 1836cbae4d12SAl Viro if (err < 0) 1837b0417d2cSAl Viro return ERR_PTR(err); 1838cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18398c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1840aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18418f64fb1cSAl Viro /* not a symlink or should not follow */ 1842cbae4d12SAl Viro path_to_nameidata(&path, nd); 18438f64fb1cSAl Viro nd->inode = inode; 18448f64fb1cSAl Viro nd->seq = seq; 1845b0417d2cSAl Viro return NULL; 18468f64fb1cSAl Viro } 1847a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1848a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 1849cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1850b0417d2cSAl Viro return ERR_PTR(-ECHILD); 1851a7f77542SAl Viro } 1852b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18533ddcd056SLinus Torvalds } 18543ddcd056SLinus Torvalds 185592d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1856ce57dfc1SAl Viro { 1857db3c9adeSAl Viro struct dentry *dentry; 1858ce57dfc1SAl Viro struct inode *inode; 1859254cf582SAl Viro unsigned seq; 1860ce57dfc1SAl Viro int err; 1861ce57dfc1SAl Viro /* 1862ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1863ce57dfc1SAl Viro * to be able to know about the current root directory and 1864ce57dfc1SAl Viro * parent relationships. 1865ce57dfc1SAl Viro */ 18664693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18671c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18684693a547SAl Viro put_link(nd); 186956676ec3SAl Viro err = handle_dots(nd, nd->last_type); 187092d27016SAl Viro return ERR_PTR(err); 18714693a547SAl Viro } 187220e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 187320e34357SAl Viro if (IS_ERR(dentry)) 187492d27016SAl Viro return ERR_CAST(dentry); 187520e34357SAl Viro if (unlikely(!dentry)) { 1876db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1877db3c9adeSAl Viro if (IS_ERR(dentry)) 187892d27016SAl Viro return ERR_CAST(dentry); 187920e34357SAl Viro } 188056676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 188156676ec3SAl Viro put_link(nd); 1882b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1883ce57dfc1SAl Viro } 1884ce57dfc1SAl Viro 18851da177e4SLinus Torvalds /* 1886bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1887bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1888bfcfaa77SLinus Torvalds * 1889bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1890bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1891bfcfaa77SLinus Torvalds * fast. 1892bfcfaa77SLinus Torvalds * 1893bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1894bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1895bfcfaa77SLinus Torvalds * crossing operation. 1896bfcfaa77SLinus Torvalds * 1897bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1898bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1899bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1900bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1901bfcfaa77SLinus Torvalds */ 1902bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1903bfcfaa77SLinus Torvalds 1904f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1905bfcfaa77SLinus Torvalds 1906468a9428SGeorge Spelvin #ifdef HASH_MIX 1907bfcfaa77SLinus Torvalds 1908468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1909468a9428SGeorge Spelvin 1910468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 19112a18da7aSGeorge Spelvin /* 19122a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 19132a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 19142a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 19152a18da7aSGeorge Spelvin * and no temporaries. 19162a18da7aSGeorge Spelvin * 19172a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 19182a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 19192a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 19202a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 19212a18da7aSGeorge Spelvin * 19222a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 19232a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 19242a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 19252a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 19262a18da7aSGeorge Spelvin * 19272a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 19282a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 19292a18da7aSGeorge Spelvin * toggling any given output bit. 19302a18da7aSGeorge Spelvin * 19312a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 19322a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19332a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 19342a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 19352a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 19362a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 19372a18da7aSGeorge Spelvin * Perfect: 8192 258048 19382a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 19392a18da7aSGeorge Spelvin */ 19402a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19412a18da7aSGeorge Spelvin ( x ^= (a), \ 19422a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 19432a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 19442a18da7aSGeorge Spelvin y *= 9 ) 1945bfcfaa77SLinus Torvalds 19460fed3ac8SGeorge Spelvin /* 19472a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 19482a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 19492a18da7aSGeorge Spelvin * work done before the hash value is used. 19500fed3ac8SGeorge Spelvin */ 19512a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19520fed3ac8SGeorge Spelvin { 19532a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 19542a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19552a18da7aSGeorge Spelvin return y >> 32; 19560fed3ac8SGeorge Spelvin } 19570fed3ac8SGeorge Spelvin 1958bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1959bfcfaa77SLinus Torvalds 19602a18da7aSGeorge Spelvin /* 19612a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19622a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19632a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19642a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19652a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19662a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19672a18da7aSGeorge Spelvin * Perfect: 2048 31744 19682a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19692a18da7aSGeorge Spelvin */ 19702a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19712a18da7aSGeorge Spelvin ( x ^= (a), \ 19722a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19732a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19742a18da7aSGeorge Spelvin y *= 9 ) 1975bfcfaa77SLinus Torvalds 19762a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19770fed3ac8SGeorge Spelvin { 19782a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19792a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19800fed3ac8SGeorge Spelvin } 19810fed3ac8SGeorge Spelvin 1982bfcfaa77SLinus Torvalds #endif 1983bfcfaa77SLinus Torvalds 19842a18da7aSGeorge Spelvin /* 19852a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19862a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19872a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19882a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19892a18da7aSGeorge Spelvin * finds the delimiter after the name. 19902a18da7aSGeorge Spelvin */ 19918387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1992bfcfaa77SLinus Torvalds { 19938387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1994bfcfaa77SLinus Torvalds 1995bfcfaa77SLinus Torvalds for (;;) { 1996fcfd2fbfSGeorge Spelvin if (!len) 1997fcfd2fbfSGeorge Spelvin goto done; 1998e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1999bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2000bfcfaa77SLinus Torvalds break; 20012a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2002bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2003bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2004bfcfaa77SLinus Torvalds } 20052a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2006bfcfaa77SLinus Torvalds done: 20072a18da7aSGeorge Spelvin return fold_hash(x, y); 2008bfcfaa77SLinus Torvalds } 2009bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2010bfcfaa77SLinus Torvalds 2011fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20128387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2013fcfd2fbfSGeorge Spelvin { 20148387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 20158387ff25SLinus Torvalds unsigned long adata, mask, len; 2016fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2017fcfd2fbfSGeorge Spelvin 20188387ff25SLinus Torvalds len = 0; 20198387ff25SLinus Torvalds goto inside; 20208387ff25SLinus Torvalds 2021fcfd2fbfSGeorge Spelvin do { 20222a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2023fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 20248387ff25SLinus Torvalds inside: 2025fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2026fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2027fcfd2fbfSGeorge Spelvin 2028fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2029fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 20302a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2031fcfd2fbfSGeorge Spelvin 20322a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2033fcfd2fbfSGeorge Spelvin } 2034fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2035fcfd2fbfSGeorge Spelvin 2036bfcfaa77SLinus Torvalds /* 2037bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2038d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2039bfcfaa77SLinus Torvalds */ 20408387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2041bfcfaa77SLinus Torvalds { 20428387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 20438387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 204436126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2045bfcfaa77SLinus Torvalds 20468387ff25SLinus Torvalds len = 0; 20478387ff25SLinus Torvalds goto inside; 20488387ff25SLinus Torvalds 2049bfcfaa77SLinus Torvalds do { 20502a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2051bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 20528387ff25SLinus Torvalds inside: 2053e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 205436126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 205536126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2056bfcfaa77SLinus Torvalds 205736126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 205836126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 205936126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20602a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 206136126f8fSLinus Torvalds 20622a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2063bfcfaa77SLinus Torvalds } 2064bfcfaa77SLinus Torvalds 20652a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2066bfcfaa77SLinus Torvalds 2067fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20688387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20690145acc2SLinus Torvalds { 20708387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20710145acc2SLinus Torvalds while (len--) 2072fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20730145acc2SLinus Torvalds return end_name_hash(hash); 20740145acc2SLinus Torvalds } 2075ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20760145acc2SLinus Torvalds 2077fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20788387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2079fcfd2fbfSGeorge Spelvin { 20808387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2081fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2082fcfd2fbfSGeorge Spelvin 2083fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2084e0ab7af9SGeorge Spelvin while (c) { 2085fcfd2fbfSGeorge Spelvin len++; 2086fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2087fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2088e0ab7af9SGeorge Spelvin } 2089fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2090fcfd2fbfSGeorge Spelvin } 2091f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2092fcfd2fbfSGeorge Spelvin 20933ddcd056SLinus Torvalds /* 2094200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2095200e9ef7SLinus Torvalds * one character. 2096200e9ef7SLinus Torvalds */ 20978387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2098200e9ef7SLinus Torvalds { 20998387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2100200e9ef7SLinus Torvalds unsigned long len = 0, c; 2101200e9ef7SLinus Torvalds 2102200e9ef7SLinus Torvalds c = (unsigned char)*name; 2103200e9ef7SLinus Torvalds do { 2104200e9ef7SLinus Torvalds len++; 2105200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2106200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2107200e9ef7SLinus Torvalds } while (c && c != '/'); 2108d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2109200e9ef7SLinus Torvalds } 2110200e9ef7SLinus Torvalds 2111bfcfaa77SLinus Torvalds #endif 2112bfcfaa77SLinus Torvalds 2113200e9ef7SLinus Torvalds /* 21141da177e4SLinus Torvalds * Name resolution. 2115ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2116ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 21171da177e4SLinus Torvalds * 2118ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2119ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 21201da177e4SLinus Torvalds */ 21216de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 21221da177e4SLinus Torvalds { 2123d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 21241da177e4SLinus Torvalds int err; 21251da177e4SLinus Torvalds 2126b4c03536SAl Viro nd->last_type = LAST_ROOT; 2127c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 21289b5858e9SAl Viro if (IS_ERR(name)) 21299b5858e9SAl Viro return PTR_ERR(name); 21301da177e4SLinus Torvalds while (*name=='/') 21311da177e4SLinus Torvalds name++; 21321da177e4SLinus Torvalds if (!*name) 21339e18f10aSAl Viro return 0; 21341da177e4SLinus Torvalds 21351da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 21361da177e4SLinus Torvalds for(;;) { 213792d27016SAl Viro const char *link; 2138d6bb3e90SLinus Torvalds u64 hash_len; 2139fe479a58SAl Viro int type; 21401da177e4SLinus Torvalds 214152094c8aSAl Viro err = may_lookup(nd); 21421da177e4SLinus Torvalds if (err) 21433595e234SAl Viro return err; 21441da177e4SLinus Torvalds 21458387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 21461da177e4SLinus Torvalds 2147fe479a58SAl Viro type = LAST_NORM; 2148d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2149fe479a58SAl Viro case 2: 2150200e9ef7SLinus Torvalds if (name[1] == '.') { 2151fe479a58SAl Viro type = LAST_DOTDOT; 215216c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 215316c2cd71SAl Viro } 2154fe479a58SAl Viro break; 2155fe479a58SAl Viro case 1: 2156fe479a58SAl Viro type = LAST_DOT; 2157fe479a58SAl Viro } 21585a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21595a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 216016c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21615a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2162a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2163da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21645a202bcdSAl Viro if (err < 0) 21653595e234SAl Viro return err; 2166d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2167d6bb3e90SLinus Torvalds name = this.name; 21685a202bcdSAl Viro } 21695a202bcdSAl Viro } 2170fe479a58SAl Viro 2171d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2172d6bb3e90SLinus Torvalds nd->last.name = name; 21735f4a6a69SAl Viro nd->last_type = type; 21745f4a6a69SAl Viro 2175d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2176d6bb3e90SLinus Torvalds if (!*name) 2177bdf6cbf1SAl Viro goto OK; 2178200e9ef7SLinus Torvalds /* 2179200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2180200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2181200e9ef7SLinus Torvalds */ 2182200e9ef7SLinus Torvalds do { 2183d6bb3e90SLinus Torvalds name++; 2184d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21858620c238SAl Viro if (unlikely(!*name)) { 21868620c238SAl Viro OK: 2187d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2188c108837eSAl Viro if (!depth) { 2189c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 21908620c238SAl Viro return 0; 2191c108837eSAl Viro } 21928620c238SAl Viro /* last component of nested symlink */ 2193d8d4611aSAl Viro name = nd->stack[--depth].name; 21948c4efe22SAl Viro link = walk_component(nd, 0); 21951c4ff1a8SAl Viro } else { 21961c4ff1a8SAl Viro /* not the last component */ 21978c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 21988620c238SAl Viro } 219992d27016SAl Viro if (unlikely(link)) { 220092d27016SAl Viro if (IS_ERR(link)) 220192d27016SAl Viro return PTR_ERR(link); 220292d27016SAl Viro /* a symlink to follow */ 2203d8d4611aSAl Viro nd->stack[depth++].name = name; 220492d27016SAl Viro name = link; 22059e18f10aSAl Viro continue; 220648c8b0c5SAl Viro } 220797242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 220897242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 22094675ac39SAl Viro if (unlazy_walk(nd)) 221097242f99SAl Viro return -ECHILD; 221197242f99SAl Viro } 22123595e234SAl Viro return -ENOTDIR; 22135f4a6a69SAl Viro } 2214ce57dfc1SAl Viro } 221597242f99SAl Viro } 22161da177e4SLinus Torvalds 2217edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2218c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 221931e6b01fSNick Piggin { 2220740a1678SAleksa Sarai int error; 2221c8a53ee5SAl Viro const char *s = nd->name->name; 222231e6b01fSNick Piggin 2223c0eb027eSLinus Torvalds if (!*s) 2224c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2225edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2226edc2b1daSAl Viro rcu_read_lock(); 2227c0eb027eSLinus Torvalds 2228c108837eSAl Viro nd->flags = flags | LOOKUP_JUMPED; 222931e6b01fSNick Piggin nd->depth = 0; 2230ab87f9a5SAleksa Sarai 2231ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2232ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2233ab87f9a5SAleksa Sarai smp_rmb(); 2234ab87f9a5SAleksa Sarai 22355b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2236b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2237b18825a7SDavid Howells struct inode *inode = root->d_inode; 223893893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2239368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 22405b6ca027SAl Viro nd->path = nd->root; 22415b6ca027SAl Viro nd->inode = inode; 22425b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2243ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22448f47a016SAl Viro nd->root_seq = nd->seq; 22455b6ca027SAl Viro } else { 22465b6ca027SAl Viro path_get(&nd->path); 22475b6ca027SAl Viro } 2248368ee9baSAl Viro return s; 22495b6ca027SAl Viro } 22505b6ca027SAl Viro 225131e6b01fSNick Piggin nd->root.mnt = NULL; 2252248fb5b9SAl Viro nd->path.mnt = NULL; 2253248fb5b9SAl Viro nd->path.dentry = NULL; 225431e6b01fSNick Piggin 22558db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 22568db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2257740a1678SAleksa Sarai error = nd_jump_root(nd); 2258740a1678SAleksa Sarai if (unlikely(error)) 2259740a1678SAleksa Sarai return ERR_PTR(error); 2260ef55d917SAl Viro return s; 22618db52c7eSAleksa Sarai } 22628db52c7eSAleksa Sarai 22638db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 22648db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2265e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 226631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2267c28cc364SNick Piggin unsigned seq; 226831e6b01fSNick Piggin 2269c28cc364SNick Piggin do { 2270c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 227131e6b01fSNick Piggin nd->path = fs->pwd; 2272ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2273c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2274c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2275e41f7d4eSAl Viro } else { 2276e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2277ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2278e41f7d4eSAl Viro } 227931e6b01fSNick Piggin } else { 2280582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2281c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 228231e6b01fSNick Piggin struct dentry *dentry; 228331e6b01fSNick Piggin 22842903ff01SAl Viro if (!f.file) 2285368ee9baSAl Viro return ERR_PTR(-EBADF); 228631e6b01fSNick Piggin 22872903ff01SAl Viro dentry = f.file->f_path.dentry; 228831e6b01fSNick Piggin 2289edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22902903ff01SAl Viro fdput(f); 2291368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2292f52e0c11SAl Viro } 22932903ff01SAl Viro 22942903ff01SAl Viro nd->path = f.file->f_path; 2295e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 229634a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 229734a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22985590ff0dSUlrich Drepper } else { 22992903ff01SAl Viro path_get(&nd->path); 230034a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 23011da177e4SLinus Torvalds } 230234a26b99SAl Viro fdput(f); 2303e41f7d4eSAl Viro } 23048db52c7eSAleksa Sarai 2305adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2306adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2307adb21d2bSAleksa Sarai nd->root = nd->path; 2308adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2309adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2310adb21d2bSAleksa Sarai } else { 2311adb21d2bSAleksa Sarai path_get(&nd->root); 2312adb21d2bSAleksa Sarai nd->flags |= LOOKUP_ROOT_GRABBED; 2313adb21d2bSAleksa Sarai } 2314adb21d2bSAleksa Sarai } 2315adb21d2bSAleksa Sarai return s; 23169b4a9b14SAl Viro } 23179b4a9b14SAl Viro 23181ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 231995fa25d9SAl Viro { 2320bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2321bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2322bd92d7feSAl Viro 2323c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2324bd92d7feSAl Viro } 2325bd92d7feSAl Viro 23264f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 23274f757f3cSAl Viro { 2328c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2329db3c9adeSAl Viro dget(nd->path.dentry); 2330b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2331b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 23324f757f3cSAl Viro } 23334f757f3cSAl Viro 23349b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2335c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23369b4a9b14SAl Viro { 2337c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2338bd92d7feSAl Viro int err; 233931e6b01fSNick Piggin 23409b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23414f757f3cSAl Viro err = handle_lookup_down(nd); 23425f336e72SAl Viro if (unlikely(err < 0)) 23435f336e72SAl Viro s = ERR_PTR(err); 23444f757f3cSAl Viro } 23454f757f3cSAl Viro 23461ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 23471ccac622SAl Viro (s = lookup_last(nd)) != NULL) 23481ccac622SAl Viro ; 23499f1fafeeSAl Viro if (!err) 23509f1fafeeSAl Viro err = complete_walk(nd); 2351bd92d7feSAl Viro 2352deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2353deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2354bd23a539SAl Viro err = -ENOTDIR; 2355161aff1dSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 2356161aff1dSAl Viro err = handle_lookup_down(nd); 2357161aff1dSAl Viro nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... 2358161aff1dSAl Viro } 2359625b6d10SAl Viro if (!err) { 2360625b6d10SAl Viro *path = nd->path; 2361625b6d10SAl Viro nd->path.mnt = NULL; 2362625b6d10SAl Viro nd->path.dentry = NULL; 2363625b6d10SAl Viro } 2364deb106c6SAl Viro terminate_walk(nd); 2365bd92d7feSAl Viro return err; 236631e6b01fSNick Piggin } 236731e6b01fSNick Piggin 236831d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23699ad1aaa6SAl Viro struct path *path, struct path *root) 2370873f1eedSJeff Layton { 2371894bc8c4SAl Viro int retval; 23729883d185SAl Viro struct nameidata nd; 2373abc9f5beSAl Viro if (IS_ERR(name)) 2374abc9f5beSAl Viro return PTR_ERR(name); 23759ad1aaa6SAl Viro if (unlikely(root)) { 23769ad1aaa6SAl Viro nd.root = *root; 23779ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23789ad1aaa6SAl Viro } 23799883d185SAl Viro set_nameidata(&nd, dfd, name); 2380c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2381873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2382c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2383873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2384c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2385873f1eedSJeff Layton 2386873f1eedSJeff Layton if (likely(!retval)) 2387161aff1dSAl Viro audit_inode(name, path->dentry, 2388161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 23899883d185SAl Viro restore_nameidata(); 2390e4bd1c1aSAl Viro putname(name); 2391873f1eedSJeff Layton return retval; 2392873f1eedSJeff Layton } 2393873f1eedSJeff Layton 23948bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2395c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2396391172c4SAl Viro struct path *parent) 23978bcb77faSAl Viro { 2398c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23999b5858e9SAl Viro int err = link_path_walk(s, nd); 24008bcb77faSAl Viro if (!err) 24018bcb77faSAl Viro err = complete_walk(nd); 2402391172c4SAl Viro if (!err) { 2403391172c4SAl Viro *parent = nd->path; 2404391172c4SAl Viro nd->path.mnt = NULL; 2405391172c4SAl Viro nd->path.dentry = NULL; 2406391172c4SAl Viro } 2407deb106c6SAl Viro terminate_walk(nd); 24088bcb77faSAl Viro return err; 24098bcb77faSAl Viro } 24108bcb77faSAl Viro 24115c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2412391172c4SAl Viro unsigned int flags, struct path *parent, 2413391172c4SAl Viro struct qstr *last, int *type) 24148bcb77faSAl Viro { 24158bcb77faSAl Viro int retval; 24169883d185SAl Viro struct nameidata nd; 24178bcb77faSAl Viro 24185c31b6ceSAl Viro if (IS_ERR(name)) 24195c31b6ceSAl Viro return name; 24209883d185SAl Viro set_nameidata(&nd, dfd, name); 2421c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 24228bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2423c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 24248bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2425c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2426391172c4SAl Viro if (likely(!retval)) { 2427391172c4SAl Viro *last = nd.last; 2428391172c4SAl Viro *type = nd.last_type; 2429c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 24305c31b6ceSAl Viro } else { 24315c31b6ceSAl Viro putname(name); 24325c31b6ceSAl Viro name = ERR_PTR(retval); 2433391172c4SAl Viro } 24349883d185SAl Viro restore_nameidata(); 24355c31b6ceSAl Viro return name; 24368bcb77faSAl Viro } 24378bcb77faSAl Viro 243879714f72SAl Viro /* does lookup, returns the object with parent locked */ 243979714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24405590ff0dSUlrich Drepper { 24415c31b6ceSAl Viro struct filename *filename; 24425c31b6ceSAl Viro struct dentry *d; 2443391172c4SAl Viro struct qstr last; 2444391172c4SAl Viro int type; 244551689104SPaul Moore 24465c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24475c31b6ceSAl Viro &last, &type); 244851689104SPaul Moore if (IS_ERR(filename)) 244951689104SPaul Moore return ERR_CAST(filename); 24505c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2451391172c4SAl Viro path_put(path); 24525c31b6ceSAl Viro putname(filename); 24535c31b6ceSAl Viro return ERR_PTR(-EINVAL); 245479714f72SAl Viro } 24555955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2456391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 245779714f72SAl Viro if (IS_ERR(d)) { 24585955102cSAl Viro inode_unlock(path->dentry->d_inode); 2459391172c4SAl Viro path_put(path); 246079714f72SAl Viro } 246151689104SPaul Moore putname(filename); 246279714f72SAl Viro return d; 24635590ff0dSUlrich Drepper } 24645590ff0dSUlrich Drepper 2465d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2466d1811465SAl Viro { 2467abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2468abc9f5beSAl Viro flags, path, NULL); 2469d1811465SAl Viro } 24704d359507SAl Viro EXPORT_SYMBOL(kern_path); 2471d1811465SAl Viro 247216f18200SJosef 'Jeff' Sipek /** 247316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 247416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 247516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 247616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 247716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2478e0a01249SAl Viro * @path: pointer to struct path to fill 247916f18200SJosef 'Jeff' Sipek */ 248016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 248116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2482e0a01249SAl Viro struct path *path) 248316f18200SJosef 'Jeff' Sipek { 24849ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24859ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2486abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2487abc9f5beSAl Viro flags , path, &root); 248816f18200SJosef 'Jeff' Sipek } 24894d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 249016f18200SJosef 'Jeff' Sipek 24913c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24923c95f0dcSAl Viro int len, struct qstr *this) 24933c95f0dcSAl Viro { 24943c95f0dcSAl Viro this->name = name; 24953c95f0dcSAl Viro this->len = len; 24963c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24973c95f0dcSAl Viro if (!len) 24983c95f0dcSAl Viro return -EACCES; 24993c95f0dcSAl Viro 25003c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 25013c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 25023c95f0dcSAl Viro return -EACCES; 25033c95f0dcSAl Viro } 25043c95f0dcSAl Viro 25053c95f0dcSAl Viro while (len--) { 25063c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 25073c95f0dcSAl Viro if (c == '/' || c == '\0') 25083c95f0dcSAl Viro return -EACCES; 25093c95f0dcSAl Viro } 25103c95f0dcSAl Viro /* 25113c95f0dcSAl Viro * See if the low-level filesystem might want 25123c95f0dcSAl Viro * to use its own hash.. 25133c95f0dcSAl Viro */ 25143c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 25153c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 25163c95f0dcSAl Viro if (err < 0) 25173c95f0dcSAl Viro return err; 25183c95f0dcSAl Viro } 25193c95f0dcSAl Viro 25203c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 25213c95f0dcSAl Viro } 25223c95f0dcSAl Viro 2523eead1911SChristoph Hellwig /** 25240da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 25250da0b7fdSDavid Howells * @name: pathname component to lookup 25260da0b7fdSDavid Howells * @base: base directory to lookup from 25270da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 25280da0b7fdSDavid Howells * 25290da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 25300da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 25310da0b7fdSDavid Howells * 25320da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25330da0b7fdSDavid Howells * not be called by generic code. 25340da0b7fdSDavid Howells * 25350da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25360da0b7fdSDavid Howells */ 25370da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25380da0b7fdSDavid Howells { 25390da0b7fdSDavid Howells struct qstr this; 25400da0b7fdSDavid Howells int err; 25410da0b7fdSDavid Howells 25420da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25430da0b7fdSDavid Howells 25440da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25450da0b7fdSDavid Howells if (err) 25460da0b7fdSDavid Howells return ERR_PTR(err); 25470da0b7fdSDavid Howells 25480da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25490da0b7fdSDavid Howells } 25500da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25510da0b7fdSDavid Howells 25520da0b7fdSDavid Howells /** 2553a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2554eead1911SChristoph Hellwig * @name: pathname component to lookup 2555eead1911SChristoph Hellwig * @base: base directory to lookup from 2556eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2557eead1911SChristoph Hellwig * 2558a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25599e7543e9SAl Viro * not be called by generic code. 2560bbddca8eSNeilBrown * 2561bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2562eead1911SChristoph Hellwig */ 2563057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2564057f6c01SJames Morris { 25658613a209SAl Viro struct dentry *dentry; 2566057f6c01SJames Morris struct qstr this; 2567cda309deSMiklos Szeredi int err; 2568057f6c01SJames Morris 25695955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25702f9092e1SDavid Woodhouse 25713c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2572cda309deSMiklos Szeredi if (err) 2573cda309deSMiklos Szeredi return ERR_PTR(err); 2574cda309deSMiklos Szeredi 25758613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25768613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2577057f6c01SJames Morris } 25784d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2579057f6c01SJames Morris 2580bbddca8eSNeilBrown /** 2581bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2582bbddca8eSNeilBrown * @name: pathname component to lookup 2583bbddca8eSNeilBrown * @base: base directory to lookup from 2584bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2585bbddca8eSNeilBrown * 2586bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2587bbddca8eSNeilBrown * not be called by generic code. 2588bbddca8eSNeilBrown * 2589bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2590bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2591bbddca8eSNeilBrown */ 2592bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2593bbddca8eSNeilBrown struct dentry *base, int len) 2594bbddca8eSNeilBrown { 2595bbddca8eSNeilBrown struct qstr this; 2596bbddca8eSNeilBrown int err; 259720d00ee8SLinus Torvalds struct dentry *ret; 2598bbddca8eSNeilBrown 25993c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2600bbddca8eSNeilBrown if (err) 2601bbddca8eSNeilBrown return ERR_PTR(err); 2602bbddca8eSNeilBrown 260320d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 260420d00ee8SLinus Torvalds if (!ret) 260520d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 260620d00ee8SLinus Torvalds return ret; 2607bbddca8eSNeilBrown } 2608bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2609bbddca8eSNeilBrown 26106c2d4798SAl Viro /* 26116c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 26126c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 26136c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 26146c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 26156c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 26166c2d4798SAl Viro * this one avoids such problems. 26176c2d4798SAl Viro */ 26186c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 26196c2d4798SAl Viro struct dentry *base, int len) 26206c2d4798SAl Viro { 26216c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 26222fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 26236c2d4798SAl Viro dput(ret); 26246c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 26256c2d4798SAl Viro } 26266c2d4798SAl Viro return ret; 26276c2d4798SAl Viro } 26286c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 26296c2d4798SAl Viro 2630eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2631eedf265aSEric W. Biederman int path_pts(struct path *path) 2632eedf265aSEric W. Biederman { 2633eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2634eedf265aSEric W. Biederman * the input path. 2635eedf265aSEric W. Biederman */ 2636eedf265aSEric W. Biederman struct dentry *child, *parent; 2637eedf265aSEric W. Biederman struct qstr this; 2638eedf265aSEric W. Biederman int ret; 2639eedf265aSEric W. Biederman 2640eedf265aSEric W. Biederman ret = path_parent_directory(path); 2641eedf265aSEric W. Biederman if (ret) 2642eedf265aSEric W. Biederman return ret; 2643eedf265aSEric W. Biederman 2644eedf265aSEric W. Biederman parent = path->dentry; 2645eedf265aSEric W. Biederman this.name = "pts"; 2646eedf265aSEric W. Biederman this.len = 3; 2647eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2648eedf265aSEric W. Biederman if (!child) 2649eedf265aSEric W. Biederman return -ENOENT; 2650eedf265aSEric W. Biederman 2651eedf265aSEric W. Biederman path->dentry = child; 2652eedf265aSEric W. Biederman dput(parent); 2653eedf265aSEric W. Biederman follow_mount(path); 2654eedf265aSEric W. Biederman return 0; 2655eedf265aSEric W. Biederman } 2656eedf265aSEric W. Biederman #endif 2657eedf265aSEric W. Biederman 26581fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26591fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26601da177e4SLinus Torvalds { 2661abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2662abc9f5beSAl Viro flags, path, NULL); 26631da177e4SLinus Torvalds } 2664b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26651fa1e7f6SAndy Whitcroft 2666cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 26671da177e4SLinus Torvalds { 26688e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2669da9592edSDavid Howells 26708e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 26711da177e4SLinus Torvalds return 0; 26728e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 26731da177e4SLinus Torvalds return 0; 267423adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 26751da177e4SLinus Torvalds } 2676cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 26771da177e4SLinus Torvalds 26781da177e4SLinus Torvalds /* 26791da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 26801da177e4SLinus Torvalds * whether the type of victim is right. 26811da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 26821da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 26831da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 26841da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 26851da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 26861da177e4SLinus Torvalds * a. be owner of dir, or 26871da177e4SLinus Torvalds * b. be owner of victim, or 26881da177e4SLinus Torvalds * c. have CAP_FOWNER capability 26891da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 26901da177e4SLinus Torvalds * links pointing to it. 26910bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 26920bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 26930bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 26940bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 26950bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 26961da177e4SLinus Torvalds * nfs_async_unlink(). 26971da177e4SLinus Torvalds */ 2698b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 26991da177e4SLinus Torvalds { 270063afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27011da177e4SLinus Torvalds int error; 27021da177e4SLinus Torvalds 2703b18825a7SDavid Howells if (d_is_negative(victim)) 27041da177e4SLinus Torvalds return -ENOENT; 2705b18825a7SDavid Howells BUG_ON(!inode); 27061da177e4SLinus Torvalds 27071da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2708593d1ce8SEric W. Biederman 2709593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2710593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2711593d1ce8SEric W. Biederman return -EOVERFLOW; 2712593d1ce8SEric W. Biederman 27134fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27141da177e4SLinus Torvalds 2715f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 27161da177e4SLinus Torvalds if (error) 27171da177e4SLinus Torvalds return error; 27181da177e4SLinus Torvalds if (IS_APPEND(dir)) 27191da177e4SLinus Torvalds return -EPERM; 2720b18825a7SDavid Howells 2721b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27220bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27231da177e4SLinus Torvalds return -EPERM; 27241da177e4SLinus Torvalds if (isdir) { 272544b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27261da177e4SLinus Torvalds return -ENOTDIR; 27271da177e4SLinus Torvalds if (IS_ROOT(victim)) 27281da177e4SLinus Torvalds return -EBUSY; 272944b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 27301da177e4SLinus Torvalds return -EISDIR; 27311da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27321da177e4SLinus Torvalds return -ENOENT; 27331da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27341da177e4SLinus Torvalds return -EBUSY; 27351da177e4SLinus Torvalds return 0; 27361da177e4SLinus Torvalds } 27371da177e4SLinus Torvalds 27381da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27391da177e4SLinus Torvalds * dir. 27401da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27411da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27421da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2743036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2744036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2745036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27461da177e4SLinus Torvalds */ 2747a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 27481da177e4SLinus Torvalds { 2749036d5236SEric W. Biederman struct user_namespace *s_user_ns; 275014e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 27511da177e4SLinus Torvalds if (child->d_inode) 27521da177e4SLinus Torvalds return -EEXIST; 27531da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27541da177e4SLinus Torvalds return -ENOENT; 2755036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2756036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2757036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2758036d5236SEric W. Biederman return -EOVERFLOW; 2759f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 27601da177e4SLinus Torvalds } 27611da177e4SLinus Torvalds 27621da177e4SLinus Torvalds /* 27631da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 27641da177e4SLinus Torvalds */ 27651da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 27661da177e4SLinus Torvalds { 27671da177e4SLinus Torvalds struct dentry *p; 27681da177e4SLinus Torvalds 27691da177e4SLinus Torvalds if (p1 == p2) { 27705955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27711da177e4SLinus Torvalds return NULL; 27721da177e4SLinus Torvalds } 27731da177e4SLinus Torvalds 2774fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 27751da177e4SLinus Torvalds 2776e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2777e2761a11SOGAWA Hirofumi if (p) { 27785955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 27795955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 27801da177e4SLinus Torvalds return p; 27811da177e4SLinus Torvalds } 27821da177e4SLinus Torvalds 2783e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2784e2761a11SOGAWA Hirofumi if (p) { 27855955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27865955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 27871da177e4SLinus Torvalds return p; 27881da177e4SLinus Torvalds } 27891da177e4SLinus Torvalds 27905955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27915955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 27921da177e4SLinus Torvalds return NULL; 27931da177e4SLinus Torvalds } 27944d359507SAl Viro EXPORT_SYMBOL(lock_rename); 27951da177e4SLinus Torvalds 27961da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 27971da177e4SLinus Torvalds { 27985955102cSAl Viro inode_unlock(p1->d_inode); 27991da177e4SLinus Torvalds if (p1 != p2) { 28005955102cSAl Viro inode_unlock(p2->d_inode); 2801fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28021da177e4SLinus Torvalds } 28031da177e4SLinus Torvalds } 28044d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28051da177e4SLinus Torvalds 28064acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2807312b63fbSAl Viro bool want_excl) 28081da177e4SLinus Torvalds { 2809a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 28101da177e4SLinus Torvalds if (error) 28111da177e4SLinus Torvalds return error; 28121da177e4SLinus Torvalds 2813acfa4380SAl Viro if (!dir->i_op->create) 28141da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 28151da177e4SLinus Torvalds mode &= S_IALLUGO; 28161da177e4SLinus Torvalds mode |= S_IFREG; 28171da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 28181da177e4SLinus Torvalds if (error) 28191da177e4SLinus Torvalds return error; 2820312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2821a74574aaSStephen Smalley if (!error) 2822f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28231da177e4SLinus Torvalds return error; 28241da177e4SLinus Torvalds } 28254d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28261da177e4SLinus Torvalds 28278e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 28288e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 28298e6c848eSAl Viro void *arg) 28308e6c848eSAl Viro { 28318e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 28328e6c848eSAl Viro int error = may_create(dir, dentry); 28338e6c848eSAl Viro if (error) 28348e6c848eSAl Viro return error; 28358e6c848eSAl Viro 28368e6c848eSAl Viro mode &= S_IALLUGO; 28378e6c848eSAl Viro mode |= S_IFREG; 28388e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 28398e6c848eSAl Viro if (error) 28408e6c848eSAl Viro return error; 28418e6c848eSAl Viro error = f(dentry, mode, arg); 28428e6c848eSAl Viro if (!error) 28438e6c848eSAl Viro fsnotify_create(dir, dentry); 28448e6c848eSAl Viro return error; 28458e6c848eSAl Viro } 28468e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 28478e6c848eSAl Viro 2848a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2849a2982cc9SEric W. Biederman { 2850a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2851a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2852a2982cc9SEric W. Biederman } 2853a2982cc9SEric W. Biederman 2854f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 28551da177e4SLinus Torvalds { 28563fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 28571da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 28581da177e4SLinus Torvalds int error; 28591da177e4SLinus Torvalds 28601da177e4SLinus Torvalds if (!inode) 28611da177e4SLinus Torvalds return -ENOENT; 28621da177e4SLinus Torvalds 2863c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2864c8fe8f30SChristoph Hellwig case S_IFLNK: 28651da177e4SLinus Torvalds return -ELOOP; 2866c8fe8f30SChristoph Hellwig case S_IFDIR: 2867c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 28681da177e4SLinus Torvalds return -EISDIR; 2869c8fe8f30SChristoph Hellwig break; 2870c8fe8f30SChristoph Hellwig case S_IFBLK: 2871c8fe8f30SChristoph Hellwig case S_IFCHR: 2872a2982cc9SEric W. Biederman if (!may_open_dev(path)) 28731da177e4SLinus Torvalds return -EACCES; 2874c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2875c8fe8f30SChristoph Hellwig case S_IFIFO: 2876c8fe8f30SChristoph Hellwig case S_IFSOCK: 28771da177e4SLinus Torvalds flag &= ~O_TRUNC; 2878c8fe8f30SChristoph Hellwig break; 28794a3fd211SDave Hansen } 2880b41572e9SDave Hansen 288162fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2882b41572e9SDave Hansen if (error) 2883b41572e9SDave Hansen return error; 28846146f0d5SMimi Zohar 28851da177e4SLinus Torvalds /* 28861da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 28871da177e4SLinus Torvalds */ 28881da177e4SLinus Torvalds if (IS_APPEND(inode)) { 28898737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 28907715b521SAl Viro return -EPERM; 28911da177e4SLinus Torvalds if (flag & O_TRUNC) 28927715b521SAl Viro return -EPERM; 28931da177e4SLinus Torvalds } 28941da177e4SLinus Torvalds 28951da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 28962e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 28977715b521SAl Viro return -EPERM; 28981da177e4SLinus Torvalds 2899f3c7691eSJ. Bruce Fields return 0; 29007715b521SAl Viro } 29017715b521SAl Viro 2902e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 29037715b521SAl Viro { 2904f0bb5aafSAl Viro const struct path *path = &filp->f_path; 29057715b521SAl Viro struct inode *inode = path->dentry->d_inode; 29067715b521SAl Viro int error = get_write_access(inode); 29071da177e4SLinus Torvalds if (error) 29087715b521SAl Viro return error; 29091da177e4SLinus Torvalds /* 29101da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 29111da177e4SLinus Torvalds */ 2912d7a06983SJeff Layton error = locks_verify_locked(filp); 2913be6d3e56SKentaro Takeda if (!error) 2914ea0d3ab2STetsuo Handa error = security_path_truncate(path); 29151da177e4SLinus Torvalds if (!error) { 29167715b521SAl Viro error = do_truncate(path->dentry, 0, 2917d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2918e1181ee6SJeff Layton filp); 29191da177e4SLinus Torvalds } 29201da177e4SLinus Torvalds put_write_access(inode); 2921acd0c935SMimi Zohar return error; 29221da177e4SLinus Torvalds } 29231da177e4SLinus Torvalds 2924d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2925d57999e1SDave Hansen { 29268a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29278a5e929dSAl Viro flag--; 2928d57999e1SDave Hansen return flag; 2929d57999e1SDave Hansen } 2930d57999e1SDave Hansen 2931d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2932d18e9008SMiklos Szeredi { 29331328c727SSeth Forshee struct user_namespace *s_user_ns; 2934d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2935d18e9008SMiklos Szeredi if (error) 2936d18e9008SMiklos Szeredi return error; 2937d18e9008SMiklos Szeredi 29381328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29391328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29401328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29411328c727SSeth Forshee return -EOVERFLOW; 29421328c727SSeth Forshee 2943d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2944d18e9008SMiklos Szeredi if (error) 2945d18e9008SMiklos Szeredi return error; 2946d18e9008SMiklos Szeredi 2947d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2948d18e9008SMiklos Szeredi } 2949d18e9008SMiklos Szeredi 29501acf0af9SDavid Howells /* 29511acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29521acf0af9SDavid Howells * dentry. 29531acf0af9SDavid Howells * 29541acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29551acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29561acf0af9SDavid Howells * 295700a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 295800a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 295900a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 29601acf0af9SDavid Howells * 29611acf0af9SDavid Howells * Returns an error code otherwise. 29621acf0af9SDavid Howells */ 2963239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 2964239eb983SAl Viro struct file *file, 2965015c3bbcSMiklos Szeredi const struct open_flags *op, 29663ec2eef1SAl Viro int open_flag, umode_t mode) 2967d18e9008SMiklos Szeredi { 2968d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2969d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2970d18e9008SMiklos Szeredi int error; 2971d18e9008SMiklos Szeredi 2972384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 2973d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2974d18e9008SMiklos Szeredi 2975d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2976d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2977d18e9008SMiklos Szeredi 297830d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 297930d90494SAl Viro file->f_path.mnt = nd->path.mnt; 29800fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 298144907d79SAl Viro open_to_namei_flags(open_flag), mode); 29826fbd0714SAl Viro d_lookup_done(dentry); 2983384f26e2SAl Viro if (!error) { 298464e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 2985d18e9008SMiklos Szeredi /* 2986384f26e2SAl Viro * We didn't have the inode before the open, so check open 2987384f26e2SAl Viro * permission here. 2988d18e9008SMiklos Szeredi */ 2989384f26e2SAl Viro int acc_mode = op->acc_mode; 299073a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 299103da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 299203da633aSAl Viro fsnotify_create(dir, dentry); 299362fb4a15SAl Viro acc_mode = 0; 299403da633aSAl Viro } 29952675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 2996384f26e2SAl Viro if (WARN_ON(error > 0)) 2997384f26e2SAl Viro error = -EINVAL; 299864e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 2999d18e9008SMiklos Szeredi error = -EIO; 3000384f26e2SAl Viro } else { 3001d18e9008SMiklos Szeredi if (file->f_path.dentry) { 3002d18e9008SMiklos Szeredi dput(dentry); 3003d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 300410c64ceaSAl Viro } 300573a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 3006d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 3007239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3008a01e718fSAl Viro error = -ENOENT; 3009d18e9008SMiklos Szeredi } 3010d18e9008SMiklos Szeredi } 3011239eb983SAl Viro if (error) { 3012d18e9008SMiklos Szeredi dput(dentry); 3013239eb983SAl Viro dentry = ERR_PTR(error); 3014239eb983SAl Viro } 3015239eb983SAl Viro return dentry; 3016d18e9008SMiklos Szeredi } 3017d18e9008SMiklos Szeredi 301831e6b01fSNick Piggin /* 30191acf0af9SDavid Howells * Look up and maybe create and open the last component. 3020d58ffd35SMiklos Szeredi * 302100a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3022d58ffd35SMiklos Szeredi * 302300a07c15SAl Viro * Returns 0 on success, that is, if 302400a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 302500a07c15SAl Viro * the file was not completely opened at this time, though lookups and 302600a07c15SAl Viro * creations were performed. 302700a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 302800a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 302900a07c15SAl Viro * hadn't been specified. 30301acf0af9SDavid Howells * 303100a07c15SAl Viro * An error code is returned on failure. 3032d58ffd35SMiklos Szeredi */ 3033da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3034d58ffd35SMiklos Szeredi const struct open_flags *op, 30353ec2eef1SAl Viro bool got_write) 3036d58ffd35SMiklos Szeredi { 3037d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 303854ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30391643b43fSAl Viro int open_flag = op->open_flag; 3040d58ffd35SMiklos Szeredi struct dentry *dentry; 30411643b43fSAl Viro int error, create_error = 0; 30421643b43fSAl Viro umode_t mode = op->mode; 30436fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3044d58ffd35SMiklos Szeredi 3045ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3046da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3047d58ffd35SMiklos Szeredi 304873a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 30496fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30506fbd0714SAl Viro for (;;) { 30516fbd0714SAl Viro if (!dentry) { 30526fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3053d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3054da5ebf5aSAl Viro return dentry; 30556fbd0714SAl Viro } 30566fbd0714SAl Viro if (d_in_lookup(dentry)) 30576fbd0714SAl Viro break; 3058d58ffd35SMiklos Szeredi 30596fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 30606fbd0714SAl Viro if (likely(error > 0)) 30616fbd0714SAl Viro break; 30626fbd0714SAl Viro if (error) 30636fbd0714SAl Viro goto out_dput; 30646fbd0714SAl Viro d_invalidate(dentry); 30656fbd0714SAl Viro dput(dentry); 30666fbd0714SAl Viro dentry = NULL; 30676fbd0714SAl Viro } 30686fbd0714SAl Viro if (dentry->d_inode) { 3069d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3070da5ebf5aSAl Viro return dentry; 30716c51e513SAl Viro } 3072d18e9008SMiklos Szeredi 30731643b43fSAl Viro /* 30741643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 30751643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 30761643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 30771643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 30781643b43fSAl Viro * 30791643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 30801643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 30811643b43fSAl Viro */ 30821643b43fSAl Viro if (open_flag & O_CREAT) { 30831643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 30841643b43fSAl Viro mode &= ~current_umask(); 30851643b43fSAl Viro if (unlikely(!got_write)) { 30861643b43fSAl Viro create_error = -EROFS; 30871643b43fSAl Viro open_flag &= ~O_CREAT; 30881643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 30891643b43fSAl Viro goto no_open; 30901643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 30911643b43fSAl Viro } else { 30921643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 30931643b43fSAl Viro if (create_error) { 30941643b43fSAl Viro open_flag &= ~O_CREAT; 30951643b43fSAl Viro if (open_flag & O_EXCL) 30961643b43fSAl Viro goto no_open; 30971643b43fSAl Viro } 30981643b43fSAl Viro } 30991643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 31001643b43fSAl Viro unlikely(!got_write)) { 31011643b43fSAl Viro /* 31021643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 31031643b43fSAl Viro * back to lookup + open 31041643b43fSAl Viro */ 31051643b43fSAl Viro goto no_open; 3106d18e9008SMiklos Szeredi } 3107d18e9008SMiklos Szeredi 31081643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3109239eb983SAl Viro dentry = atomic_open(nd, dentry, file, op, open_flag, mode); 3110da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3111da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3112da5ebf5aSAl Viro return dentry; 3113239eb983SAl Viro } 311454ef4872SMiklos Szeredi 31151643b43fSAl Viro no_open: 31166fbd0714SAl Viro if (d_in_lookup(dentry)) { 311712fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 311812fa5e24SAl Viro nd->flags); 31196fbd0714SAl Viro d_lookup_done(dentry); 312012fa5e24SAl Viro if (unlikely(res)) { 312112fa5e24SAl Viro if (IS_ERR(res)) { 312212fa5e24SAl Viro error = PTR_ERR(res); 312312fa5e24SAl Viro goto out_dput; 312412fa5e24SAl Viro } 312512fa5e24SAl Viro dput(dentry); 312612fa5e24SAl Viro dentry = res; 312712fa5e24SAl Viro } 312854ef4872SMiklos Szeredi } 312954ef4872SMiklos Szeredi 3130d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 31311643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 313273a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3133ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3134ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3135ce8644fcSAl Viro error = -EACCES; 3136d58ffd35SMiklos Szeredi goto out_dput; 313764894cf8SAl Viro } 3138ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31391643b43fSAl Viro open_flag & O_EXCL); 3140d58ffd35SMiklos Szeredi if (error) 3141d58ffd35SMiklos Szeredi goto out_dput; 3142ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3143d58ffd35SMiklos Szeredi } 31441643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31451643b43fSAl Viro error = create_error; 3146d58ffd35SMiklos Szeredi goto out_dput; 3147d58ffd35SMiklos Szeredi } 3148da5ebf5aSAl Viro return dentry; 3149d58ffd35SMiklos Szeredi 3150d58ffd35SMiklos Szeredi out_dput: 3151d58ffd35SMiklos Szeredi dput(dentry); 3152da5ebf5aSAl Viro return ERR_PTR(error); 3153d58ffd35SMiklos Szeredi } 3154d58ffd35SMiklos Szeredi 3155d58ffd35SMiklos Szeredi /* 3156fe2d35ffSAl Viro * Handle the last step of open() 315731e6b01fSNick Piggin */ 31581ccac622SAl Viro static const char *do_last(struct nameidata *nd, 31593ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3160fb1cc555SAl Viro { 3161a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 31626404674aSAl Viro kuid_t dir_uid = nd->inode->i_uid; 31636404674aSAl Viro umode_t dir_mode = nd->inode->i_mode; 3164ca344a89SAl Viro int open_flag = op->open_flag; 316577d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 316664894cf8SAl Viro bool got_write = false; 3167bcda7652SAl Viro int acc_mode = op->acc_mode; 3168254cf582SAl Viro unsigned seq; 3169a1eb3315SMiklos Szeredi struct inode *inode; 3170da5ebf5aSAl Viro struct dentry *dentry; 3171b0417d2cSAl Viro const char *res; 317216c2cd71SAl Viro int error; 3173fb1cc555SAl Viro 3174c3e380b0SAl Viro nd->flags |= op->intent; 3175c3e380b0SAl Viro 3176bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 317756676ec3SAl Viro if (nd->depth) 317856676ec3SAl Viro put_link(nd); 3179fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3180deb106c6SAl Viro if (unlikely(error)) 31811ccac622SAl Viro return ERR_PTR(error); 3182e83db167SMiklos Szeredi goto finish_open; 31831f36f774SAl Viro } 3184a2c36b45SAl Viro 3185ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3186fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3187fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3188fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 318920e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 319020e34357SAl Viro if (IS_ERR(dentry)) 31911ccac622SAl Viro return ERR_CAST(dentry); 319220e34357SAl Viro if (likely(dentry)) 319371574865SMiklos Szeredi goto finish_lookup; 319471574865SMiklos Szeredi 319537d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 31966583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3197b6183df7SMiklos Szeredi } else { 3198fe2d35ffSAl Viro /* create side of things */ 3199a3fbbde7SAl Viro /* 3200b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3201b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3202b6183df7SMiklos Szeredi * about to look up 3203a3fbbde7SAl Viro */ 32049f1fafeeSAl Viro error = complete_walk(nd); 3205e8bb73dfSAl Viro if (error) 32061ccac622SAl Viro return ERR_PTR(error); 3207fe2d35ffSAl Viro 3208c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 32091f36f774SAl Viro /* trailing slashes? */ 3210deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 32111ccac622SAl Viro return ERR_PTR(-EISDIR); 3212b6183df7SMiklos Szeredi } 32131f36f774SAl Viro 32149cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 321564894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 321664894cf8SAl Viro if (!error) 321764894cf8SAl Viro got_write = true; 321864894cf8SAl Viro /* 321964894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 322064894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 322164894cf8SAl Viro * dropping this one anyway. 322264894cf8SAl Viro */ 322364894cf8SAl Viro } 32249cf843e3SAl Viro if (open_flag & O_CREAT) 32255955102cSAl Viro inode_lock(dir->d_inode); 32269cf843e3SAl Viro else 32279cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3228da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 32299cf843e3SAl Viro if (open_flag & O_CREAT) 32305955102cSAl Viro inode_unlock(dir->d_inode); 32319cf843e3SAl Viro else 32329cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3233fb1cc555SAl Viro 3234da5ebf5aSAl Viro if (IS_ERR(dentry)) { 3235da5ebf5aSAl Viro error = PTR_ERR(dentry); 3236d58ffd35SMiklos Szeredi goto out; 3237da5ebf5aSAl Viro } 32386c0d46c4SAl Viro 323900a07c15SAl Viro if (file->f_mode & FMODE_OPENED) { 324073a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3241496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 324277d660a8SMiklos Szeredi will_truncate = false; 3243d18e9008SMiklos Szeredi 324476ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3245da5ebf5aSAl Viro dput(dentry); 3246d18e9008SMiklos Szeredi goto opened; 3247d18e9008SMiklos Szeredi } 3248d18e9008SMiklos Szeredi 324973a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 32509b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3251ca344a89SAl Viro open_flag &= ~O_TRUNC; 325277d660a8SMiklos Szeredi will_truncate = false; 325362fb4a15SAl Viro acc_mode = 0; 3254e73cabffSAl Viro dput(nd->path.dentry); 3255e73cabffSAl Viro nd->path.dentry = dentry; 3256e83db167SMiklos Szeredi goto finish_open_created; 3257fb1cc555SAl Viro } 3258fb1cc555SAl Viro 3259fb1cc555SAl Viro /* 3260d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3261d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3262d18e9008SMiklos Szeredi * necessary...) 3263d18e9008SMiklos Szeredi */ 326464894cf8SAl Viro if (got_write) { 3265d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 326664894cf8SAl Viro got_write = false; 3267d18e9008SMiklos Szeredi } 3268d18e9008SMiklos Szeredi 326920e34357SAl Viro finish_lookup: 327056676ec3SAl Viro if (nd->depth) 327156676ec3SAl Viro put_link(nd); 32728c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3273b0417d2cSAl Viro if (unlikely(res)) { 32741ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3275b0417d2cSAl Viro return res; 32761ccac622SAl Viro } 327731d1726dSAl Viro 327831d1726dSAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 327931d1726dSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 32801ccac622SAl Viro return ERR_PTR(-EEXIST); 328131d1726dSAl Viro } 3282bc77daa7SAl Viro finish_open: 32838f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3284a3fbbde7SAl Viro error = complete_walk(nd); 3285fac7d191SAl Viro if (error) 32861ccac622SAl Viro return ERR_PTR(error); 328776ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 328830aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3289fb1cc555SAl Viro error = -EISDIR; 329030aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 32912675a4ebSAl Viro goto out; 3292d0cb5018SAl Viro error = may_create_in_sticky(dir_mode, dir_uid, 329330aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 329430aba665SSalvatore Mesoraca if (unlikely(error)) 329530aba665SSalvatore Mesoraca goto out; 329630aba665SSalvatore Mesoraca } 3297af2f5542SMiklos Szeredi error = -ENOTDIR; 329844b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 32992675a4ebSAl Viro goto out; 33004bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 330177d660a8SMiklos Szeredi will_truncate = false; 33026c0d46c4SAl Viro 33030f9d1a10SAl Viro if (will_truncate) { 33040f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33050f9d1a10SAl Viro if (error) 33062675a4ebSAl Viro goto out; 330764894cf8SAl Viro got_write = true; 33080f9d1a10SAl Viro } 3309e83db167SMiklos Szeredi finish_open_created: 3310bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3311ca344a89SAl Viro if (error) 33122675a4ebSAl Viro goto out; 3313aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3314ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3315fac7d191SAl Viro if (error) 3316015c3bbcSMiklos Szeredi goto out; 3317a8277b9bSMiklos Szeredi opened: 33186035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3319fe9ec829SAl Viro if (!error && will_truncate) 33202675a4ebSAl Viro error = handle_truncate(file); 3321ca344a89SAl Viro out: 3322c80567c8SAl Viro if (unlikely(error > 0)) { 3323c80567c8SAl Viro WARN_ON(1); 3324c80567c8SAl Viro error = -EINVAL; 3325c80567c8SAl Viro } 332664894cf8SAl Viro if (got_write) 33270f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 33281ccac622SAl Viro return ERR_PTR(error); 3329fb1cc555SAl Viro } 3330fb1cc555SAl Viro 3331af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3332af7bd4dcSAmir Goldstein { 3333af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3334af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3335af7bd4dcSAmir Goldstein struct inode *inode; 3336af7bd4dcSAmir Goldstein int error; 3337af7bd4dcSAmir Goldstein 3338af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3339af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3340af7bd4dcSAmir Goldstein if (error) 3341af7bd4dcSAmir Goldstein goto out_err; 3342af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3343af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3344af7bd4dcSAmir Goldstein goto out_err; 3345af7bd4dcSAmir Goldstein error = -ENOMEM; 3346cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3347af7bd4dcSAmir Goldstein if (unlikely(!child)) 3348af7bd4dcSAmir Goldstein goto out_err; 3349af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3350af7bd4dcSAmir Goldstein if (error) 3351af7bd4dcSAmir Goldstein goto out_err; 3352af7bd4dcSAmir Goldstein error = -ENOENT; 3353af7bd4dcSAmir Goldstein inode = child->d_inode; 3354af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3355af7bd4dcSAmir Goldstein goto out_err; 3356af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3357af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3358af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3359af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3360af7bd4dcSAmir Goldstein } 3361fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3362af7bd4dcSAmir Goldstein return child; 3363af7bd4dcSAmir Goldstein 3364af7bd4dcSAmir Goldstein out_err: 3365af7bd4dcSAmir Goldstein dput(child); 3366af7bd4dcSAmir Goldstein return ERR_PTR(error); 3367af7bd4dcSAmir Goldstein } 3368af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3369af7bd4dcSAmir Goldstein 3370c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 337160545d0dSAl Viro const struct open_flags *op, 33723ec2eef1SAl Viro struct file *file) 337360545d0dSAl Viro { 3374625b6d10SAl Viro struct dentry *child; 3375625b6d10SAl Viro struct path path; 3376c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 337760545d0dSAl Viro if (unlikely(error)) 337860545d0dSAl Viro return error; 3379625b6d10SAl Viro error = mnt_want_write(path.mnt); 338060545d0dSAl Viro if (unlikely(error)) 338160545d0dSAl Viro goto out; 3382af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3383af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3384684e73beSHirofumi Nakagawa if (IS_ERR(child)) 338560545d0dSAl Viro goto out2; 3386625b6d10SAl Viro dput(path.dentry); 3387625b6d10SAl Viro path.dentry = child; 3388c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 338969a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 339062fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 339160545d0dSAl Viro if (error) 339260545d0dSAl Viro goto out2; 3393625b6d10SAl Viro file->f_path.mnt = path.mnt; 3394be12af3eSAl Viro error = finish_open(file, child, NULL); 339560545d0dSAl Viro out2: 3396625b6d10SAl Viro mnt_drop_write(path.mnt); 339760545d0dSAl Viro out: 3398625b6d10SAl Viro path_put(&path); 339960545d0dSAl Viro return error; 340060545d0dSAl Viro } 340160545d0dSAl Viro 34026ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34036ac08709SAl Viro { 34046ac08709SAl Viro struct path path; 34056ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34066ac08709SAl Viro if (!error) { 34076ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3408ae2bb293SAl Viro error = vfs_open(&path, file); 34096ac08709SAl Viro path_put(&path); 34106ac08709SAl Viro } 34116ac08709SAl Viro return error; 34126ac08709SAl Viro } 34136ac08709SAl Viro 3414c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3415c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34161da177e4SLinus Torvalds { 341730d90494SAl Viro struct file *file; 341813aab428SAl Viro int error; 341931e6b01fSNick Piggin 3420ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 34211afc99beSAl Viro if (IS_ERR(file)) 34221afc99beSAl Viro return file; 342331e6b01fSNick Piggin 3424bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 34253ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 34265f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 34276ac08709SAl Viro error = do_o_path(nd, flags, file); 34285f336e72SAl Viro } else { 34295f336e72SAl Viro const char *s = path_init(nd, flags); 34303bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 34311ccac622SAl Viro (s = do_last(nd, file, op)) != NULL) 34321ccac622SAl Viro ; 3433deb106c6SAl Viro terminate_walk(nd); 34345f336e72SAl Viro } 34357c1c01ecSAl Viro if (likely(!error)) { 3436aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 34377c1c01ecSAl Viro return file; 34387c1c01ecSAl Viro WARN_ON(1); 34397c1c01ecSAl Viro error = -EINVAL; 3440015c3bbcSMiklos Szeredi } 34417c1c01ecSAl Viro fput(file); 34422675a4ebSAl Viro if (error == -EOPENSTALE) { 34432675a4ebSAl Viro if (flags & LOOKUP_RCU) 34442675a4ebSAl Viro error = -ECHILD; 34452675a4ebSAl Viro else 34462675a4ebSAl Viro error = -ESTALE; 34472675a4ebSAl Viro } 34487c1c01ecSAl Viro return ERR_PTR(error); 3449de459215SKirill Korotaev } 34501da177e4SLinus Torvalds 3451669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3452f9652e10SAl Viro const struct open_flags *op) 345313aab428SAl Viro { 34549883d185SAl Viro struct nameidata nd; 3455f9652e10SAl Viro int flags = op->lookup_flags; 345613aab428SAl Viro struct file *filp; 345713aab428SAl Viro 34589883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3459c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 346013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3461c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 346213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3463c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34649883d185SAl Viro restore_nameidata(); 346513aab428SAl Viro return filp; 346613aab428SAl Viro } 346713aab428SAl Viro 346873d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3469f9652e10SAl Viro const char *name, const struct open_flags *op) 347073d049a4SAl Viro { 34719883d185SAl Viro struct nameidata nd; 347273d049a4SAl Viro struct file *file; 347351689104SPaul Moore struct filename *filename; 3474f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 347573d049a4SAl Viro 347673d049a4SAl Viro nd.root.mnt = mnt; 347773d049a4SAl Viro nd.root.dentry = dentry; 347873d049a4SAl Viro 3479b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 348073d049a4SAl Viro return ERR_PTR(-ELOOP); 348173d049a4SAl Viro 348251689104SPaul Moore filename = getname_kernel(name); 3483a1c83681SViresh Kumar if (IS_ERR(filename)) 348451689104SPaul Moore return ERR_CAST(filename); 348551689104SPaul Moore 34869883d185SAl Viro set_nameidata(&nd, -1, filename); 3487c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 348873d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3489c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 349073d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3491c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 34929883d185SAl Viro restore_nameidata(); 349351689104SPaul Moore putname(filename); 349473d049a4SAl Viro return file; 349573d049a4SAl Viro } 349673d049a4SAl Viro 3497fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34981ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34991da177e4SLinus Torvalds { 3500c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3501391172c4SAl Viro struct qstr last; 3502391172c4SAl Viro int type; 3503c30dabfeSJan Kara int err2; 35041ac12b4bSJeff Layton int error; 35051ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 35061ac12b4bSJeff Layton 35071ac12b4bSJeff Layton /* 35081ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 35091ac12b4bSJeff Layton * other flags passed in are ignored! 35101ac12b4bSJeff Layton */ 35111ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 35121ac12b4bSJeff Layton 35135c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 35145c31b6ceSAl Viro if (IS_ERR(name)) 35155c31b6ceSAl Viro return ERR_CAST(name); 35161da177e4SLinus Torvalds 3517c663e5d8SChristoph Hellwig /* 3518c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3519c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3520c663e5d8SChristoph Hellwig */ 35215c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3522ed75e95dSAl Viro goto out; 3523c663e5d8SChristoph Hellwig 3524c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3525391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3526c663e5d8SChristoph Hellwig /* 3527c663e5d8SChristoph Hellwig * Do the final lookup. 3528c663e5d8SChristoph Hellwig */ 3529391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 35305955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3531391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 35321da177e4SLinus Torvalds if (IS_ERR(dentry)) 3533a8104a9fSAl Viro goto unlock; 3534c663e5d8SChristoph Hellwig 3535a8104a9fSAl Viro error = -EEXIST; 3536b18825a7SDavid Howells if (d_is_positive(dentry)) 3537a8104a9fSAl Viro goto fail; 3538b18825a7SDavid Howells 3539c663e5d8SChristoph Hellwig /* 3540c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3541c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3542c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3543c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3544c663e5d8SChristoph Hellwig */ 3545391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3546a8104a9fSAl Viro error = -ENOENT; 3547ed75e95dSAl Viro goto fail; 3548e9baf6e5SAl Viro } 3549c30dabfeSJan Kara if (unlikely(err2)) { 3550c30dabfeSJan Kara error = err2; 3551a8104a9fSAl Viro goto fail; 3552c30dabfeSJan Kara } 3553181c37b6SAl Viro putname(name); 3554e9baf6e5SAl Viro return dentry; 35551da177e4SLinus Torvalds fail: 3556a8104a9fSAl Viro dput(dentry); 3557a8104a9fSAl Viro dentry = ERR_PTR(error); 3558a8104a9fSAl Viro unlock: 35595955102cSAl Viro inode_unlock(path->dentry->d_inode); 3560c30dabfeSJan Kara if (!err2) 3561391172c4SAl Viro mnt_drop_write(path->mnt); 3562ed75e95dSAl Viro out: 3563391172c4SAl Viro path_put(path); 3564181c37b6SAl Viro putname(name); 3565ed75e95dSAl Viro return dentry; 3566dae6ad8fSAl Viro } 3567fa14a0b8SAl Viro 3568fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3569fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3570fa14a0b8SAl Viro { 3571181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3572181c37b6SAl Viro path, lookup_flags); 3573fa14a0b8SAl Viro } 3574dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3575dae6ad8fSAl Viro 3576921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3577921a1650SAl Viro { 3578921a1650SAl Viro dput(dentry); 35795955102cSAl Viro inode_unlock(path->dentry->d_inode); 3580a8104a9fSAl Viro mnt_drop_write(path->mnt); 3581921a1650SAl Viro path_put(path); 3582921a1650SAl Viro } 3583921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3584921a1650SAl Viro 3585520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 35861ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3587dae6ad8fSAl Viro { 3588181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3589dae6ad8fSAl Viro } 3590dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3591dae6ad8fSAl Viro 35921a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35931da177e4SLinus Torvalds { 3594a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35951da177e4SLinus Torvalds 35961da177e4SLinus Torvalds if (error) 35971da177e4SLinus Torvalds return error; 35981da177e4SLinus Torvalds 359994f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 36001da177e4SLinus Torvalds return -EPERM; 36011da177e4SLinus Torvalds 3602acfa4380SAl Viro if (!dir->i_op->mknod) 36031da177e4SLinus Torvalds return -EPERM; 36041da177e4SLinus Torvalds 360508ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 360608ce5f16SSerge E. Hallyn if (error) 360708ce5f16SSerge E. Hallyn return error; 360808ce5f16SSerge E. Hallyn 36091da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 36101da177e4SLinus Torvalds if (error) 36111da177e4SLinus Torvalds return error; 36121da177e4SLinus Torvalds 36131da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3614a74574aaSStephen Smalley if (!error) 3615f38aa942SAmy Griffis fsnotify_create(dir, dentry); 36161da177e4SLinus Torvalds return error; 36171da177e4SLinus Torvalds } 36184d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 36191da177e4SLinus Torvalds 3620f69aac00SAl Viro static int may_mknod(umode_t mode) 3621463c3197SDave Hansen { 3622463c3197SDave Hansen switch (mode & S_IFMT) { 3623463c3197SDave Hansen case S_IFREG: 3624463c3197SDave Hansen case S_IFCHR: 3625463c3197SDave Hansen case S_IFBLK: 3626463c3197SDave Hansen case S_IFIFO: 3627463c3197SDave Hansen case S_IFSOCK: 3628463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3629463c3197SDave Hansen return 0; 3630463c3197SDave Hansen case S_IFDIR: 3631463c3197SDave Hansen return -EPERM; 3632463c3197SDave Hansen default: 3633463c3197SDave Hansen return -EINVAL; 3634463c3197SDave Hansen } 3635463c3197SDave Hansen } 3636463c3197SDave Hansen 363787c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 363887c4e192SDominik Brodowski unsigned int dev) 36391da177e4SLinus Torvalds { 36401da177e4SLinus Torvalds struct dentry *dentry; 3641dae6ad8fSAl Viro struct path path; 3642dae6ad8fSAl Viro int error; 3643972567f1SJeff Layton unsigned int lookup_flags = 0; 36441da177e4SLinus Torvalds 36458e4bfca1SAl Viro error = may_mknod(mode); 36468e4bfca1SAl Viro if (error) 36478e4bfca1SAl Viro return error; 3648972567f1SJeff Layton retry: 3649972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3650dae6ad8fSAl Viro if (IS_ERR(dentry)) 3651dae6ad8fSAl Viro return PTR_ERR(dentry); 36522ad94ae6SAl Viro 3653dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3654ce3b0f8dSAl Viro mode &= ~current_umask(); 3655dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3656be6d3e56SKentaro Takeda if (error) 3657a8104a9fSAl Viro goto out; 36581da177e4SLinus Torvalds switch (mode & S_IFMT) { 36591da177e4SLinus Torvalds case 0: case S_IFREG: 3660312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 366105d1a717SMimi Zohar if (!error) 366205d1a717SMimi Zohar ima_post_path_mknod(dentry); 36631da177e4SLinus Torvalds break; 36641da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3665dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36661da177e4SLinus Torvalds new_decode_dev(dev)); 36671da177e4SLinus Torvalds break; 36681da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3669dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 36701da177e4SLinus Torvalds break; 36711da177e4SLinus Torvalds } 3672a8104a9fSAl Viro out: 3673921a1650SAl Viro done_path_create(&path, dentry); 3674972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3675972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3676972567f1SJeff Layton goto retry; 3677972567f1SJeff Layton } 36781da177e4SLinus Torvalds return error; 36791da177e4SLinus Torvalds } 36801da177e4SLinus Torvalds 368187c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 368287c4e192SDominik Brodowski unsigned int, dev) 368387c4e192SDominik Brodowski { 368487c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 368587c4e192SDominik Brodowski } 368687c4e192SDominik Brodowski 36878208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 36885590ff0dSUlrich Drepper { 368987c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 36905590ff0dSUlrich Drepper } 36915590ff0dSUlrich Drepper 369218bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 36931da177e4SLinus Torvalds { 3694a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36958de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 36961da177e4SLinus Torvalds 36971da177e4SLinus Torvalds if (error) 36981da177e4SLinus Torvalds return error; 36991da177e4SLinus Torvalds 3700acfa4380SAl Viro if (!dir->i_op->mkdir) 37011da177e4SLinus Torvalds return -EPERM; 37021da177e4SLinus Torvalds 37031da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 37041da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 37051da177e4SLinus Torvalds if (error) 37061da177e4SLinus Torvalds return error; 37071da177e4SLinus Torvalds 37088de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 37098de52778SAl Viro return -EMLINK; 37108de52778SAl Viro 37111da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3712a74574aaSStephen Smalley if (!error) 3713f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 37141da177e4SLinus Torvalds return error; 37151da177e4SLinus Torvalds } 37164d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 37171da177e4SLinus Torvalds 37180101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 37191da177e4SLinus Torvalds { 37206902d925SDave Hansen struct dentry *dentry; 3721dae6ad8fSAl Viro struct path path; 3722dae6ad8fSAl Viro int error; 3723b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 37241da177e4SLinus Torvalds 3725b76d8b82SJeff Layton retry: 3726b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 37276902d925SDave Hansen if (IS_ERR(dentry)) 3728dae6ad8fSAl Viro return PTR_ERR(dentry); 37296902d925SDave Hansen 3730dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3731ce3b0f8dSAl Viro mode &= ~current_umask(); 3732dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3733a8104a9fSAl Viro if (!error) 3734dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3735921a1650SAl Viro done_path_create(&path, dentry); 3736b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3737b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3738b76d8b82SJeff Layton goto retry; 3739b76d8b82SJeff Layton } 37401da177e4SLinus Torvalds return error; 37411da177e4SLinus Torvalds } 37421da177e4SLinus Torvalds 37430101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 37440101db7aSDominik Brodowski { 37450101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 37460101db7aSDominik Brodowski } 37470101db7aSDominik Brodowski 3748a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 37495590ff0dSUlrich Drepper { 37500101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 37515590ff0dSUlrich Drepper } 37525590ff0dSUlrich Drepper 37531da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 37541da177e4SLinus Torvalds { 37551da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 37561da177e4SLinus Torvalds 37571da177e4SLinus Torvalds if (error) 37581da177e4SLinus Torvalds return error; 37591da177e4SLinus Torvalds 3760acfa4380SAl Viro if (!dir->i_op->rmdir) 37611da177e4SLinus Torvalds return -EPERM; 37621da177e4SLinus Torvalds 37631d2ef590SAl Viro dget(dentry); 37645955102cSAl Viro inode_lock(dentry->d_inode); 3765912dbc15SSage Weil 37661da177e4SLinus Torvalds error = -EBUSY; 37677af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3768912dbc15SSage Weil goto out; 3769912dbc15SSage Weil 37701da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3771912dbc15SSage Weil if (error) 3772912dbc15SSage Weil goto out; 3773912dbc15SSage Weil 37741da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3775912dbc15SSage Weil if (error) 3776912dbc15SSage Weil goto out; 3777912dbc15SSage Weil 37788767712fSAl Viro shrink_dcache_parent(dentry); 37791da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3780d83c49f3SAl Viro dont_mount(dentry); 37818ed936b5SEric W. Biederman detach_mounts(dentry); 3782116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 37831da177e4SLinus Torvalds 3784912dbc15SSage Weil out: 37855955102cSAl Viro inode_unlock(dentry->d_inode); 37861d2ef590SAl Viro dput(dentry); 3787912dbc15SSage Weil if (!error) 3788912dbc15SSage Weil d_delete(dentry); 37891da177e4SLinus Torvalds return error; 37901da177e4SLinus Torvalds } 37914d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37921da177e4SLinus Torvalds 3793f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 37941da177e4SLinus Torvalds { 37951da177e4SLinus Torvalds int error = 0; 379691a27b2aSJeff Layton struct filename *name; 37971da177e4SLinus Torvalds struct dentry *dentry; 3798f5beed75SAl Viro struct path path; 3799f5beed75SAl Viro struct qstr last; 3800f5beed75SAl Viro int type; 3801c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3802c6ee9206SJeff Layton retry: 3803c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3804c1d4dd27SAl Viro &path, &last, &type); 380591a27b2aSJeff Layton if (IS_ERR(name)) 380691a27b2aSJeff Layton return PTR_ERR(name); 38071da177e4SLinus Torvalds 3808f5beed75SAl Viro switch (type) { 38091da177e4SLinus Torvalds case LAST_DOTDOT: 38101da177e4SLinus Torvalds error = -ENOTEMPTY; 38111da177e4SLinus Torvalds goto exit1; 38121da177e4SLinus Torvalds case LAST_DOT: 38131da177e4SLinus Torvalds error = -EINVAL; 38141da177e4SLinus Torvalds goto exit1; 38151da177e4SLinus Torvalds case LAST_ROOT: 38161da177e4SLinus Torvalds error = -EBUSY; 38171da177e4SLinus Torvalds goto exit1; 38181da177e4SLinus Torvalds } 38190612d9fbSOGAWA Hirofumi 3820f5beed75SAl Viro error = mnt_want_write(path.mnt); 3821c30dabfeSJan Kara if (error) 3822c30dabfeSJan Kara goto exit1; 38230612d9fbSOGAWA Hirofumi 38245955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3825f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38261da177e4SLinus Torvalds error = PTR_ERR(dentry); 38276902d925SDave Hansen if (IS_ERR(dentry)) 38286902d925SDave Hansen goto exit2; 3829e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3830e6bc45d6STheodore Ts'o error = -ENOENT; 3831e6bc45d6STheodore Ts'o goto exit3; 3832e6bc45d6STheodore Ts'o } 3833f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3834be6d3e56SKentaro Takeda if (error) 3835c30dabfeSJan Kara goto exit3; 3836f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 38370622753bSDave Hansen exit3: 38381da177e4SLinus Torvalds dput(dentry); 38396902d925SDave Hansen exit2: 38405955102cSAl Viro inode_unlock(path.dentry->d_inode); 3841f5beed75SAl Viro mnt_drop_write(path.mnt); 38421da177e4SLinus Torvalds exit1: 3843f5beed75SAl Viro path_put(&path); 38441da177e4SLinus Torvalds putname(name); 3845c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3846c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3847c6ee9206SJeff Layton goto retry; 3848c6ee9206SJeff Layton } 38491da177e4SLinus Torvalds return error; 38501da177e4SLinus Torvalds } 38511da177e4SLinus Torvalds 38523cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 38535590ff0dSUlrich Drepper { 38545590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 38555590ff0dSUlrich Drepper } 38565590ff0dSUlrich Drepper 3857b21996e3SJ. Bruce Fields /** 3858b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3859b21996e3SJ. Bruce Fields * @dir: parent directory 3860b21996e3SJ. Bruce Fields * @dentry: victim 3861b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3862b21996e3SJ. Bruce Fields * 3863b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3864b21996e3SJ. Bruce Fields * 3865b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3866b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3867b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3868b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3869b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3870b21996e3SJ. Bruce Fields * 3871b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3872b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3873b21996e3SJ. Bruce Fields * to be NFS exported. 3874b21996e3SJ. Bruce Fields */ 3875b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38761da177e4SLinus Torvalds { 38779accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38781da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 38791da177e4SLinus Torvalds 38801da177e4SLinus Torvalds if (error) 38811da177e4SLinus Torvalds return error; 38821da177e4SLinus Torvalds 3883acfa4380SAl Viro if (!dir->i_op->unlink) 38841da177e4SLinus Torvalds return -EPERM; 38851da177e4SLinus Torvalds 38865955102cSAl Viro inode_lock(target); 38878ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38881da177e4SLinus Torvalds error = -EBUSY; 38891da177e4SLinus Torvalds else { 38901da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3891bec1052eSAl Viro if (!error) { 38925a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38935a14696cSJ. Bruce Fields if (error) 3894b21996e3SJ. Bruce Fields goto out; 38951da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38968ed936b5SEric W. Biederman if (!error) { 3897d83c49f3SAl Viro dont_mount(dentry); 38988ed936b5SEric W. Biederman detach_mounts(dentry); 3899116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 39008ed936b5SEric W. Biederman } 3901bec1052eSAl Viro } 39021da177e4SLinus Torvalds } 3903b21996e3SJ. Bruce Fields out: 39045955102cSAl Viro inode_unlock(target); 39051da177e4SLinus Torvalds 39061da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 39071da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 39089accbb97SJ. Bruce Fields fsnotify_link_count(target); 39091da177e4SLinus Torvalds d_delete(dentry); 39101da177e4SLinus Torvalds } 39110eeca283SRobert Love 39121da177e4SLinus Torvalds return error; 39131da177e4SLinus Torvalds } 39144d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 39151da177e4SLinus Torvalds 39161da177e4SLinus Torvalds /* 39171da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 39181b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 39191da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 39201da177e4SLinus Torvalds * while waiting on the I/O. 39211da177e4SLinus Torvalds */ 3922da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 39231da177e4SLinus Torvalds { 39242ad94ae6SAl Viro int error; 39251da177e4SLinus Torvalds struct dentry *dentry; 3926f5beed75SAl Viro struct path path; 3927f5beed75SAl Viro struct qstr last; 3928f5beed75SAl Viro int type; 39291da177e4SLinus Torvalds struct inode *inode = NULL; 3930b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 39315d18f813SJeff Layton unsigned int lookup_flags = 0; 39325d18f813SJeff Layton retry: 3933da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 393491a27b2aSJeff Layton if (IS_ERR(name)) 393591a27b2aSJeff Layton return PTR_ERR(name); 39362ad94ae6SAl Viro 39371da177e4SLinus Torvalds error = -EISDIR; 3938f5beed75SAl Viro if (type != LAST_NORM) 39391da177e4SLinus Torvalds goto exit1; 39400612d9fbSOGAWA Hirofumi 3941f5beed75SAl Viro error = mnt_want_write(path.mnt); 3942c30dabfeSJan Kara if (error) 3943c30dabfeSJan Kara goto exit1; 3944b21996e3SJ. Bruce Fields retry_deleg: 39455955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3946f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39471da177e4SLinus Torvalds error = PTR_ERR(dentry); 39481da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 39491da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3950f5beed75SAl Viro if (last.name[last.len]) 395150338b88STörök Edwin goto slashes; 39521da177e4SLinus Torvalds inode = dentry->d_inode; 3953b18825a7SDavid Howells if (d_is_negative(dentry)) 3954e6bc45d6STheodore Ts'o goto slashes; 39557de9c6eeSAl Viro ihold(inode); 3956f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3957be6d3e56SKentaro Takeda if (error) 3958c30dabfeSJan Kara goto exit2; 3959f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 39601da177e4SLinus Torvalds exit2: 39611da177e4SLinus Torvalds dput(dentry); 39621da177e4SLinus Torvalds } 39635955102cSAl Viro inode_unlock(path.dentry->d_inode); 39641da177e4SLinus Torvalds if (inode) 39651da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3966b21996e3SJ. Bruce Fields inode = NULL; 3967b21996e3SJ. Bruce Fields if (delegated_inode) { 39685a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3969b21996e3SJ. Bruce Fields if (!error) 3970b21996e3SJ. Bruce Fields goto retry_deleg; 3971b21996e3SJ. Bruce Fields } 3972f5beed75SAl Viro mnt_drop_write(path.mnt); 39731da177e4SLinus Torvalds exit1: 3974f5beed75SAl Viro path_put(&path); 39755d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39765d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 39775d18f813SJeff Layton inode = NULL; 39785d18f813SJeff Layton goto retry; 39795d18f813SJeff Layton } 3980da2f1362SChristoph Hellwig putname(name); 39811da177e4SLinus Torvalds return error; 39821da177e4SLinus Torvalds 39831da177e4SLinus Torvalds slashes: 3984b18825a7SDavid Howells if (d_is_negative(dentry)) 3985b18825a7SDavid Howells error = -ENOENT; 398644b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3987b18825a7SDavid Howells error = -EISDIR; 3988b18825a7SDavid Howells else 3989b18825a7SDavid Howells error = -ENOTDIR; 39901da177e4SLinus Torvalds goto exit2; 39911da177e4SLinus Torvalds } 39921da177e4SLinus Torvalds 39932e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39945590ff0dSUlrich Drepper { 39955590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39965590ff0dSUlrich Drepper return -EINVAL; 39975590ff0dSUlrich Drepper 39985590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39995590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 40005590ff0dSUlrich Drepper 4001da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 40025590ff0dSUlrich Drepper } 40035590ff0dSUlrich Drepper 40043480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 40055590ff0dSUlrich Drepper { 4006da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 40075590ff0dSUlrich Drepper } 40085590ff0dSUlrich Drepper 4009db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 40101da177e4SLinus Torvalds { 4011a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 40121da177e4SLinus Torvalds 40131da177e4SLinus Torvalds if (error) 40141da177e4SLinus Torvalds return error; 40151da177e4SLinus Torvalds 4016acfa4380SAl Viro if (!dir->i_op->symlink) 40171da177e4SLinus Torvalds return -EPERM; 40181da177e4SLinus Torvalds 40191da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 40201da177e4SLinus Torvalds if (error) 40211da177e4SLinus Torvalds return error; 40221da177e4SLinus Torvalds 40231da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 4024a74574aaSStephen Smalley if (!error) 4025f38aa942SAmy Griffis fsnotify_create(dir, dentry); 40261da177e4SLinus Torvalds return error; 40271da177e4SLinus Torvalds } 40284d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 40291da177e4SLinus Torvalds 4030b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 4031b724e846SDominik Brodowski const char __user *newname) 40321da177e4SLinus Torvalds { 40332ad94ae6SAl Viro int error; 403491a27b2aSJeff Layton struct filename *from; 40356902d925SDave Hansen struct dentry *dentry; 4036dae6ad8fSAl Viro struct path path; 4037f46d3567SJeff Layton unsigned int lookup_flags = 0; 40381da177e4SLinus Torvalds 40391da177e4SLinus Torvalds from = getname(oldname); 40401da177e4SLinus Torvalds if (IS_ERR(from)) 40411da177e4SLinus Torvalds return PTR_ERR(from); 4042f46d3567SJeff Layton retry: 4043f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 40441da177e4SLinus Torvalds error = PTR_ERR(dentry); 40456902d925SDave Hansen if (IS_ERR(dentry)) 4046dae6ad8fSAl Viro goto out_putname; 40476902d925SDave Hansen 404891a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4049a8104a9fSAl Viro if (!error) 405091a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4051921a1650SAl Viro done_path_create(&path, dentry); 4052f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4053f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4054f46d3567SJeff Layton goto retry; 4055f46d3567SJeff Layton } 40566902d925SDave Hansen out_putname: 40571da177e4SLinus Torvalds putname(from); 40581da177e4SLinus Torvalds return error; 40591da177e4SLinus Torvalds } 40601da177e4SLinus Torvalds 4061b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4062b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4063b724e846SDominik Brodowski { 4064b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4065b724e846SDominik Brodowski } 4066b724e846SDominik Brodowski 40673480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40685590ff0dSUlrich Drepper { 4069b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 40705590ff0dSUlrich Drepper } 40715590ff0dSUlrich Drepper 4072146a8595SJ. Bruce Fields /** 4073146a8595SJ. Bruce Fields * vfs_link - create a new link 4074146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4075146a8595SJ. Bruce Fields * @dir: new parent 4076146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4077146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4078146a8595SJ. Bruce Fields * 4079146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4080146a8595SJ. Bruce Fields * 4081146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4082146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4083146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4084146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4085146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4086146a8595SJ. Bruce Fields * 4087146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4088146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4089146a8595SJ. Bruce Fields * to be NFS exported. 4090146a8595SJ. Bruce Fields */ 4091146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40921da177e4SLinus Torvalds { 40931da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40948de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40951da177e4SLinus Torvalds int error; 40961da177e4SLinus Torvalds 40971da177e4SLinus Torvalds if (!inode) 40981da177e4SLinus Torvalds return -ENOENT; 40991da177e4SLinus Torvalds 4100a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 41011da177e4SLinus Torvalds if (error) 41021da177e4SLinus Torvalds return error; 41031da177e4SLinus Torvalds 41041da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 41051da177e4SLinus Torvalds return -EXDEV; 41061da177e4SLinus Torvalds 41071da177e4SLinus Torvalds /* 41081da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 41091da177e4SLinus Torvalds */ 41101da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 41111da177e4SLinus Torvalds return -EPERM; 41120bd23d09SEric W. Biederman /* 41130bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 41140bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 41150bd23d09SEric W. Biederman * the vfs. 41160bd23d09SEric W. Biederman */ 41170bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 41180bd23d09SEric W. Biederman return -EPERM; 4119acfa4380SAl Viro if (!dir->i_op->link) 41201da177e4SLinus Torvalds return -EPERM; 41217e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 41221da177e4SLinus Torvalds return -EPERM; 41231da177e4SLinus Torvalds 41241da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 41251da177e4SLinus Torvalds if (error) 41261da177e4SLinus Torvalds return error; 41271da177e4SLinus Torvalds 41285955102cSAl Viro inode_lock(inode); 4129aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4130f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4131aae8a97dSAneesh Kumar K.V error = -ENOENT; 41328de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 41338de52778SAl Viro error = -EMLINK; 4134146a8595SJ. Bruce Fields else { 4135146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4136146a8595SJ. Bruce Fields if (!error) 41371da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4138146a8595SJ. Bruce Fields } 4139f4e0c30cSAl Viro 4140f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4141f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4142f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4143f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4144f4e0c30cSAl Viro } 41455955102cSAl Viro inode_unlock(inode); 4146e31e14ecSStephen Smalley if (!error) 41477e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 41481da177e4SLinus Torvalds return error; 41491da177e4SLinus Torvalds } 41504d359507SAl Viro EXPORT_SYMBOL(vfs_link); 41511da177e4SLinus Torvalds 41521da177e4SLinus Torvalds /* 41531da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41541da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41551da177e4SLinus Torvalds * newname. --KAB 41561da177e4SLinus Torvalds * 41571da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41581da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 41591da177e4SLinus Torvalds * and other special files. --ADM 41601da177e4SLinus Torvalds */ 416146ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 416246ea89ebSDominik Brodowski const char __user *newname, int flags) 41631da177e4SLinus Torvalds { 41641da177e4SLinus Torvalds struct dentry *new_dentry; 4165dae6ad8fSAl Viro struct path old_path, new_path; 4166146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 416711a7b371SAneesh Kumar K.V int how = 0; 41681da177e4SLinus Torvalds int error; 41691da177e4SLinus Torvalds 417011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4171c04030e1SUlrich Drepper return -EINVAL; 417211a7b371SAneesh Kumar K.V /* 4173f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4174f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4175f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 417611a7b371SAneesh Kumar K.V */ 4177f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4178f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4179f0cc6ffbSLinus Torvalds return -ENOENT; 418011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4181f0cc6ffbSLinus Torvalds } 4182c04030e1SUlrich Drepper 418311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 418411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4185442e31caSJeff Layton retry: 418611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 41871da177e4SLinus Torvalds if (error) 41882ad94ae6SAl Viro return error; 41892ad94ae6SAl Viro 4190442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4191442e31caSJeff Layton (how & LOOKUP_REVAL)); 41921da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41936902d925SDave Hansen if (IS_ERR(new_dentry)) 4194dae6ad8fSAl Viro goto out; 4195dae6ad8fSAl Viro 4196dae6ad8fSAl Viro error = -EXDEV; 4197dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4198dae6ad8fSAl Viro goto out_dput; 4199800179c9SKees Cook error = may_linkat(&old_path); 4200800179c9SKees Cook if (unlikely(error)) 4201800179c9SKees Cook goto out_dput; 4202dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4203be6d3e56SKentaro Takeda if (error) 4204a8104a9fSAl Viro goto out_dput; 4205146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 420675c3f29dSDave Hansen out_dput: 4207921a1650SAl Viro done_path_create(&new_path, new_dentry); 4208146a8595SJ. Bruce Fields if (delegated_inode) { 4209146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4210d22e6338SOleg Drokin if (!error) { 4211d22e6338SOleg Drokin path_put(&old_path); 4212146a8595SJ. Bruce Fields goto retry; 4213146a8595SJ. Bruce Fields } 4214d22e6338SOleg Drokin } 4215442e31caSJeff Layton if (retry_estale(error, how)) { 4216d22e6338SOleg Drokin path_put(&old_path); 4217442e31caSJeff Layton how |= LOOKUP_REVAL; 4218442e31caSJeff Layton goto retry; 4219442e31caSJeff Layton } 42201da177e4SLinus Torvalds out: 42212d8f3038SAl Viro path_put(&old_path); 42221da177e4SLinus Torvalds 42231da177e4SLinus Torvalds return error; 42241da177e4SLinus Torvalds } 42251da177e4SLinus Torvalds 422646ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 422746ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 422846ea89ebSDominik Brodowski { 422946ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 423046ea89ebSDominik Brodowski } 423146ea89ebSDominik Brodowski 42323480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 42335590ff0dSUlrich Drepper { 423446ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 42355590ff0dSUlrich Drepper } 42365590ff0dSUlrich Drepper 4237bc27027aSMiklos Szeredi /** 4238bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4239bc27027aSMiklos Szeredi * @old_dir: parent of source 4240bc27027aSMiklos Szeredi * @old_dentry: source 4241bc27027aSMiklos Szeredi * @new_dir: parent of destination 4242bc27027aSMiklos Szeredi * @new_dentry: destination 4243bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4244520c8b16SMiklos Szeredi * @flags: rename flags 4245bc27027aSMiklos Szeredi * 4246bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4247bc27027aSMiklos Szeredi * 4248bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4249bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4250bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4251bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4252bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4253bc27027aSMiklos Szeredi * so. 4254bc27027aSMiklos Szeredi * 4255bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4256bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4257bc27027aSMiklos Szeredi * to be NFS exported. 4258bc27027aSMiklos Szeredi * 42591da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42601da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42611da177e4SLinus Torvalds * Problems: 42620117d427SMauro Carvalho Chehab * 4263d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42641da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42651da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4266a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42671da177e4SLinus Torvalds * story. 42686cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42696cedba89SJ. Bruce Fields * and source (if it is not a directory). 42701b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42711da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42721da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4273a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 42741da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 42751da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 42761da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4277a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 42781da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 42791da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 42801da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4281e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 42821b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 42831da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4284c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 42851da177e4SLinus Torvalds * locking]. 42861da177e4SLinus Torvalds */ 42871da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 42888e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4289520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 42901da177e4SLinus Torvalds { 42911da177e4SLinus Torvalds int error; 4292bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4293bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4294bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4295da1ce067SMiklos Szeredi bool new_is_dir = false; 4296da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 429749d31c2fSAl Viro struct name_snapshot old_name; 42981da177e4SLinus Torvalds 42998d3e2936SMiklos Szeredi if (source == target) 43001da177e4SLinus Torvalds return 0; 43011da177e4SLinus Torvalds 43021da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 43031da177e4SLinus Torvalds if (error) 43041da177e4SLinus Torvalds return error; 43051da177e4SLinus Torvalds 4306da1ce067SMiklos Szeredi if (!target) { 4307a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4308da1ce067SMiklos Szeredi } else { 4309da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4310da1ce067SMiklos Szeredi 4311da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 43121da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4313da1ce067SMiklos Szeredi else 4314da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4315da1ce067SMiklos Szeredi } 43161da177e4SLinus Torvalds if (error) 43171da177e4SLinus Torvalds return error; 43181da177e4SLinus Torvalds 43192773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 43201da177e4SLinus Torvalds return -EPERM; 43211da177e4SLinus Torvalds 4322bc27027aSMiklos Szeredi /* 4323bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4324bc27027aSMiklos Szeredi * we'll need to flip '..'. 4325bc27027aSMiklos Szeredi */ 4326da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4327da1ce067SMiklos Szeredi if (is_dir) { 4328bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4329bc27027aSMiklos Szeredi if (error) 4330bc27027aSMiklos Szeredi return error; 4331bc27027aSMiklos Szeredi } 4332da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4333da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4334da1ce067SMiklos Szeredi if (error) 4335da1ce067SMiklos Szeredi return error; 4336da1ce067SMiklos Szeredi } 4337da1ce067SMiklos Szeredi } 43380eeca283SRobert Love 43390b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 43400b3974ebSMiklos Szeredi flags); 4341bc27027aSMiklos Szeredi if (error) 4342bc27027aSMiklos Szeredi return error; 4343bc27027aSMiklos Szeredi 434449d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4345bc27027aSMiklos Szeredi dget(new_dentry); 4346da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4347bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4348bc27027aSMiklos Szeredi else if (target) 43495955102cSAl Viro inode_lock(target); 4350bc27027aSMiklos Szeredi 4351bc27027aSMiklos Szeredi error = -EBUSY; 43527af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4353bc27027aSMiklos Szeredi goto out; 4354bc27027aSMiklos Szeredi 4355da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4356bc27027aSMiklos Szeredi error = -EMLINK; 4357da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4358bc27027aSMiklos Szeredi goto out; 4359da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4360da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4361da1ce067SMiklos Szeredi goto out; 4362da1ce067SMiklos Szeredi } 4363da1ce067SMiklos Szeredi if (!is_dir) { 4364bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4365bc27027aSMiklos Szeredi if (error) 4366bc27027aSMiklos Szeredi goto out; 4367da1ce067SMiklos Szeredi } 4368da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4369bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4370bc27027aSMiklos Szeredi if (error) 4371bc27027aSMiklos Szeredi goto out; 4372bc27027aSMiklos Szeredi } 4373520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4374520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4375bc27027aSMiklos Szeredi if (error) 4376bc27027aSMiklos Szeredi goto out; 4377bc27027aSMiklos Szeredi 4378da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 43798767712fSAl Viro if (is_dir) { 43808767712fSAl Viro shrink_dcache_parent(new_dentry); 4381bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 43828767712fSAl Viro } 4383bc27027aSMiklos Szeredi dont_mount(new_dentry); 43848ed936b5SEric W. Biederman detach_mounts(new_dentry); 4385bc27027aSMiklos Szeredi } 4386da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4387da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4388bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4389da1ce067SMiklos Szeredi else 4390da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4391da1ce067SMiklos Szeredi } 4392bc27027aSMiklos Szeredi out: 4393da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4394bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4395bc27027aSMiklos Szeredi else if (target) 43965955102cSAl Viro inode_unlock(target); 4397bc27027aSMiklos Szeredi dput(new_dentry); 4398da1ce067SMiklos Szeredi if (!error) { 4399f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4400da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4401da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4402f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4403da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4404da1ce067SMiklos Szeredi } 4405da1ce067SMiklos Szeredi } 440649d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 44070eeca283SRobert Love 44081da177e4SLinus Torvalds return error; 44091da177e4SLinus Torvalds } 44104d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 44111da177e4SLinus Torvalds 4412ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4413ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 44141da177e4SLinus Torvalds { 44151da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 44161da177e4SLinus Torvalds struct dentry *trap; 4417f5beed75SAl Viro struct path old_path, new_path; 4418f5beed75SAl Viro struct qstr old_last, new_last; 4419f5beed75SAl Viro int old_type, new_type; 44208e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 442191a27b2aSJeff Layton struct filename *from; 442291a27b2aSJeff Layton struct filename *to; 4423f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4424c6a94284SJeff Layton bool should_retry = false; 44252ad94ae6SAl Viro int error; 4426520c8b16SMiklos Szeredi 44270d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4428da1ce067SMiklos Szeredi return -EINVAL; 4429da1ce067SMiklos Szeredi 44300d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 44310d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4432520c8b16SMiklos Szeredi return -EINVAL; 4433520c8b16SMiklos Szeredi 44340d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 44350d7a8555SMiklos Szeredi return -EPERM; 44360d7a8555SMiklos Szeredi 4437f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4438f5beed75SAl Viro target_flags = 0; 4439f5beed75SAl Viro 4440c6a94284SJeff Layton retry: 4441c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4442c1d4dd27SAl Viro &old_path, &old_last, &old_type); 444391a27b2aSJeff Layton if (IS_ERR(from)) { 444491a27b2aSJeff Layton error = PTR_ERR(from); 44451da177e4SLinus Torvalds goto exit; 444691a27b2aSJeff Layton } 44471da177e4SLinus Torvalds 4448c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4449c1d4dd27SAl Viro &new_path, &new_last, &new_type); 445091a27b2aSJeff Layton if (IS_ERR(to)) { 445191a27b2aSJeff Layton error = PTR_ERR(to); 44521da177e4SLinus Torvalds goto exit1; 445391a27b2aSJeff Layton } 44541da177e4SLinus Torvalds 44551da177e4SLinus Torvalds error = -EXDEV; 4456f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44571da177e4SLinus Torvalds goto exit2; 44581da177e4SLinus Torvalds 44591da177e4SLinus Torvalds error = -EBUSY; 4460f5beed75SAl Viro if (old_type != LAST_NORM) 44611da177e4SLinus Torvalds goto exit2; 44621da177e4SLinus Torvalds 44630a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44640a7c3937SMiklos Szeredi error = -EEXIST; 4465f5beed75SAl Viro if (new_type != LAST_NORM) 44661da177e4SLinus Torvalds goto exit2; 44671da177e4SLinus Torvalds 4468f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4469c30dabfeSJan Kara if (error) 4470c30dabfeSJan Kara goto exit2; 4471c30dabfeSJan Kara 44728e6d782cSJ. Bruce Fields retry_deleg: 4473f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 44741da177e4SLinus Torvalds 4475f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 44761da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 44771da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 44781da177e4SLinus Torvalds goto exit3; 44791da177e4SLinus Torvalds /* source must exist */ 44801da177e4SLinus Torvalds error = -ENOENT; 4481b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44821da177e4SLinus Torvalds goto exit4; 4483f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44841da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44851da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44861da177e4SLinus Torvalds goto exit4; 44870a7c3937SMiklos Szeredi error = -EEXIST; 44880a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44890a7c3937SMiklos Szeredi goto exit5; 4490da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4491da1ce067SMiklos Szeredi error = -ENOENT; 4492da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4493da1ce067SMiklos Szeredi goto exit5; 4494da1ce067SMiklos Szeredi 4495da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4496da1ce067SMiklos Szeredi error = -ENOTDIR; 4497f5beed75SAl Viro if (new_last.name[new_last.len]) 4498da1ce067SMiklos Szeredi goto exit5; 4499da1ce067SMiklos Szeredi } 4500da1ce067SMiklos Szeredi } 45010a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 45020a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 45030a7c3937SMiklos Szeredi error = -ENOTDIR; 4504f5beed75SAl Viro if (old_last.name[old_last.len]) 45050a7c3937SMiklos Szeredi goto exit5; 4506f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 45070a7c3937SMiklos Szeredi goto exit5; 45080a7c3937SMiklos Szeredi } 45090a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 45100a7c3937SMiklos Szeredi error = -EINVAL; 45110a7c3937SMiklos Szeredi if (old_dentry == trap) 45120a7c3937SMiklos Szeredi goto exit5; 45131da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4514da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 45151da177e4SLinus Torvalds error = -ENOTEMPTY; 45161da177e4SLinus Torvalds if (new_dentry == trap) 45171da177e4SLinus Torvalds goto exit5; 45181da177e4SLinus Torvalds 4519f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4520f5beed75SAl Viro &new_path, new_dentry, flags); 4521be6d3e56SKentaro Takeda if (error) 4522c30dabfeSJan Kara goto exit5; 4523f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4524f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4525520c8b16SMiklos Szeredi &delegated_inode, flags); 45261da177e4SLinus Torvalds exit5: 45271da177e4SLinus Torvalds dput(new_dentry); 45281da177e4SLinus Torvalds exit4: 45291da177e4SLinus Torvalds dput(old_dentry); 45301da177e4SLinus Torvalds exit3: 4531f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 45328e6d782cSJ. Bruce Fields if (delegated_inode) { 45338e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 45348e6d782cSJ. Bruce Fields if (!error) 45358e6d782cSJ. Bruce Fields goto retry_deleg; 45368e6d782cSJ. Bruce Fields } 4537f5beed75SAl Viro mnt_drop_write(old_path.mnt); 45381da177e4SLinus Torvalds exit2: 4539c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4540c6a94284SJeff Layton should_retry = true; 4541f5beed75SAl Viro path_put(&new_path); 45422ad94ae6SAl Viro putname(to); 45431da177e4SLinus Torvalds exit1: 4544f5beed75SAl Viro path_put(&old_path); 45451da177e4SLinus Torvalds putname(from); 4546c6a94284SJeff Layton if (should_retry) { 4547c6a94284SJeff Layton should_retry = false; 4548c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4549c6a94284SJeff Layton goto retry; 4550c6a94284SJeff Layton } 45512ad94ae6SAl Viro exit: 45521da177e4SLinus Torvalds return error; 45531da177e4SLinus Torvalds } 45541da177e4SLinus Torvalds 4555ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4556ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4557ee81feb6SDominik Brodowski { 4558ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4559ee81feb6SDominik Brodowski } 4560ee81feb6SDominik Brodowski 4561520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4562520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4563520c8b16SMiklos Szeredi { 4564ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4565520c8b16SMiklos Szeredi } 4566520c8b16SMiklos Szeredi 4567a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45685590ff0dSUlrich Drepper { 4569ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45705590ff0dSUlrich Drepper } 45715590ff0dSUlrich Drepper 4572787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4573787fb6bcSMiklos Szeredi { 4574787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4575787fb6bcSMiklos Szeredi if (error) 4576787fb6bcSMiklos Szeredi return error; 4577787fb6bcSMiklos Szeredi 4578787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4579787fb6bcSMiklos Szeredi return -EPERM; 4580787fb6bcSMiklos Szeredi 4581787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4582787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4583787fb6bcSMiklos Szeredi } 4584787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4585787fb6bcSMiklos Szeredi 45865d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 45871da177e4SLinus Torvalds { 45885d826c84SAl Viro int len = PTR_ERR(link); 45891da177e4SLinus Torvalds if (IS_ERR(link)) 45901da177e4SLinus Torvalds goto out; 45911da177e4SLinus Torvalds 45921da177e4SLinus Torvalds len = strlen(link); 45931da177e4SLinus Torvalds if (len > (unsigned) buflen) 45941da177e4SLinus Torvalds len = buflen; 45951da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45961da177e4SLinus Torvalds len = -EFAULT; 45971da177e4SLinus Torvalds out: 45981da177e4SLinus Torvalds return len; 45991da177e4SLinus Torvalds } 46001da177e4SLinus Torvalds 4601d60874cdSMiklos Szeredi /** 4602fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4603fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4604fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4605fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4606fd4a0edfSMiklos Szeredi * 4607fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4608fd4a0edfSMiklos Szeredi * 4609fd4a0edfSMiklos Szeredi * Does not call security hook. 4610fd4a0edfSMiklos Szeredi */ 4611fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4612fd4a0edfSMiklos Szeredi { 4613fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4614f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4615f2df5da6SAl Viro const char *link; 4616f2df5da6SAl Viro int res; 4617fd4a0edfSMiklos Szeredi 461876fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 461976fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 462076fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 462176fca90eSMiklos Szeredi 462276fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4623fd4a0edfSMiklos Szeredi return -EINVAL; 4624fd4a0edfSMiklos Szeredi 462576fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 462676fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 462776fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 462876fca90eSMiklos Szeredi } 462976fca90eSMiklos Szeredi 46304c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4631f2df5da6SAl Viro if (!link) { 4632f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4633f2df5da6SAl Viro if (IS_ERR(link)) 4634f2df5da6SAl Viro return PTR_ERR(link); 4635f2df5da6SAl Viro } 4636f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4637f2df5da6SAl Viro do_delayed_call(&done); 4638f2df5da6SAl Viro return res; 4639fd4a0edfSMiklos Szeredi } 4640fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 46411da177e4SLinus Torvalds 4642d60874cdSMiklos Szeredi /** 4643d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4644d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4645d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4646d60874cdSMiklos Szeredi * 4647d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4648d60874cdSMiklos Szeredi * 4649d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4650d60874cdSMiklos Szeredi * 4651d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4652d60874cdSMiklos Szeredi */ 4653d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4654d60874cdSMiklos Szeredi { 4655d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4656d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4657d60874cdSMiklos Szeredi 4658d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4659d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4660d60874cdSMiklos Szeredi if (!res) 4661d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4662d60874cdSMiklos Szeredi } 4663d60874cdSMiklos Szeredi return res; 4664d60874cdSMiklos Szeredi } 4665d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4666d60874cdSMiklos Szeredi 46671da177e4SLinus Torvalds /* get the link contents into pagecache */ 46686b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4669fceef393SAl Viro struct delayed_call *callback) 46701da177e4SLinus Torvalds { 4671ebd09abbSDuane Griffin char *kaddr; 46721da177e4SLinus Torvalds struct page *page; 46736b255391SAl Viro struct address_space *mapping = inode->i_mapping; 46746b255391SAl Viro 4675d3883d4fSAl Viro if (!dentry) { 4676d3883d4fSAl Viro page = find_get_page(mapping, 0); 4677d3883d4fSAl Viro if (!page) 46786b255391SAl Viro return ERR_PTR(-ECHILD); 4679d3883d4fSAl Viro if (!PageUptodate(page)) { 4680d3883d4fSAl Viro put_page(page); 4681d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4682d3883d4fSAl Viro } 4683d3883d4fSAl Viro } else { 4684090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 46851da177e4SLinus Torvalds if (IS_ERR(page)) 46866fe6900eSNick Piggin return (char*)page; 4687d3883d4fSAl Viro } 4688fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 468921fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 469021fc61c7SAl Viro kaddr = page_address(page); 46916b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4692ebd09abbSDuane Griffin return kaddr; 46931da177e4SLinus Torvalds } 46941da177e4SLinus Torvalds 46956b255391SAl Viro EXPORT_SYMBOL(page_get_link); 46961da177e4SLinus Torvalds 4697fceef393SAl Viro void page_put_link(void *arg) 46981da177e4SLinus Torvalds { 4699fceef393SAl Viro put_page(arg); 47001da177e4SLinus Torvalds } 47014d359507SAl Viro EXPORT_SYMBOL(page_put_link); 47021da177e4SLinus Torvalds 4703aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4704aa80deabSAl Viro { 4705fceef393SAl Viro DEFINE_DELAYED_CALL(done); 47066b255391SAl Viro int res = readlink_copy(buffer, buflen, 47076b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4708fceef393SAl Viro &done)); 4709fceef393SAl Viro do_delayed_call(&done); 4710aa80deabSAl Viro return res; 4711aa80deabSAl Viro } 4712aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4713aa80deabSAl Viro 471454566b2cSNick Piggin /* 471554566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 471654566b2cSNick Piggin */ 471754566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 47181da177e4SLinus Torvalds { 47191da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 47200adb25d2SKirill Korotaev struct page *page; 4721afddba49SNick Piggin void *fsdata; 4722beb497abSDmitriy Monakhov int err; 4723c718a975STetsuo Handa unsigned int flags = 0; 472454566b2cSNick Piggin if (nofs) 472554566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 47261da177e4SLinus Torvalds 47277e53cac4SNeilBrown retry: 4728afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 472954566b2cSNick Piggin flags, &page, &fsdata); 47301da177e4SLinus Torvalds if (err) 4731afddba49SNick Piggin goto fail; 4732afddba49SNick Piggin 473321fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4734afddba49SNick Piggin 4735afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4736afddba49SNick Piggin page, fsdata); 47371da177e4SLinus Torvalds if (err < 0) 47381da177e4SLinus Torvalds goto fail; 4739afddba49SNick Piggin if (err < len-1) 4740afddba49SNick Piggin goto retry; 4741afddba49SNick Piggin 47421da177e4SLinus Torvalds mark_inode_dirty(inode); 47431da177e4SLinus Torvalds return 0; 47441da177e4SLinus Torvalds fail: 47451da177e4SLinus Torvalds return err; 47461da177e4SLinus Torvalds } 47474d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 47481da177e4SLinus Torvalds 47490adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 47500adb25d2SKirill Korotaev { 47510adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4752c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 47530adb25d2SKirill Korotaev } 47544d359507SAl Viro EXPORT_SYMBOL(page_symlink); 47550adb25d2SKirill Korotaev 475692e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 47576b255391SAl Viro .get_link = page_get_link, 47581da177e4SLinus Torvalds }; 47591da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4760