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 /* 11719deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 11729deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 11739deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 11749deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 11759875cf80SDavid Howells */ 11769deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 11779deed3ebSAl Viro int *count, unsigned lookup_flags) 11789875cf80SDavid Howells { 11799deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 11809875cf80SDavid Howells bool need_mntput = false; 11818aef1884SAl Viro int ret = 0; 11829875cf80SDavid Howells 11839deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1184cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1185cc53ce53SDavid Howells * being held. */ 1186d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1187fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1188508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1189cc53ce53SDavid Howells if (ret < 0) 11908aef1884SAl Viro break; 1191cc53ce53SDavid Howells } 1192cc53ce53SDavid Howells 11939deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 11949875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11959deed3ebSAl Viro if (mounted) { // ... in our namespace 11969875cf80SDavid Howells dput(path->dentry); 11979875cf80SDavid Howells if (need_mntput) 1198463ffb2eSAl Viro mntput(path->mnt); 1199463ffb2eSAl Viro path->mnt = mounted; 1200463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12019deed3ebSAl Viro // here we know it's positive 12029deed3ebSAl Viro flags = path->dentry->d_flags; 12039875cf80SDavid Howells need_mntput = true; 12049875cf80SDavid Howells continue; 1205463ffb2eSAl Viro } 12061da177e4SLinus Torvalds } 12079875cf80SDavid Howells 12089deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 12099deed3ebSAl Viro break; 12109deed3ebSAl Viro 12119deed3ebSAl Viro // uncovered automount point 12129deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 12139deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 12149875cf80SDavid Howells if (ret < 0) 12158aef1884SAl Viro break; 12169875cf80SDavid Howells } 12179875cf80SDavid Howells 12189deed3ebSAl Viro if (ret == -EISDIR) 12199deed3ebSAl Viro ret = 0; 12209deed3ebSAl Viro // possible if you race with several mount --move 12219deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 12228aef1884SAl Viro mntput(path->mnt); 12239deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1224d41efb52SAl Viro ret = -ENOENT; 12259deed3ebSAl Viro *jumped = need_mntput; 12268402752eSAl Viro return ret; 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds 12299deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 12309deed3ebSAl Viro int *count, unsigned lookup_flags) 12319deed3ebSAl Viro { 12329deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 12339deed3ebSAl Viro 12349deed3ebSAl Viro /* fastpath */ 12359deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 12369deed3ebSAl Viro *jumped = false; 12379deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 12389deed3ebSAl Viro return -ENOENT; 12399deed3ebSAl Viro return 0; 12409deed3ebSAl Viro } 12419deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 12429deed3ebSAl Viro } 12439deed3ebSAl Viro 1244cc53ce53SDavid Howells int follow_down_one(struct path *path) 12451da177e4SLinus Torvalds { 12461da177e4SLinus Torvalds struct vfsmount *mounted; 12471da177e4SLinus Torvalds 12481c755af4SAl Viro mounted = lookup_mnt(path); 12491da177e4SLinus Torvalds if (mounted) { 12509393bd07SAl Viro dput(path->dentry); 12519393bd07SAl Viro mntput(path->mnt); 12529393bd07SAl Viro path->mnt = mounted; 12539393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12541da177e4SLinus Torvalds return 1; 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds return 0; 12571da177e4SLinus Torvalds } 12584d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12591da177e4SLinus Torvalds 12609875cf80SDavid Howells /* 12619deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 12629deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 12639deed3ebSAl Viro * caller is permitted to proceed or not. 12649deed3ebSAl Viro */ 12659deed3ebSAl Viro int follow_down(struct path *path) 12669deed3ebSAl Viro { 12679deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 12689deed3ebSAl Viro bool jumped; 12699deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 12709deed3ebSAl Viro 12719deed3ebSAl Viro if (path->mnt != mnt) 12729deed3ebSAl Viro mntput(mnt); 12739deed3ebSAl Viro return ret; 12749deed3ebSAl Viro } 12759deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 12769deed3ebSAl Viro 12779deed3ebSAl Viro /* 1278287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1279287548e4SAl Viro * we meet a managed dentry that would need blocking. 12809875cf80SDavid Howells */ 12819875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1282254cf582SAl Viro struct inode **inode, unsigned *seqp) 12839875cf80SDavid Howells { 1284ea936aebSAl Viro struct dentry *dentry = path->dentry; 1285ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1286ea936aebSAl Viro 1287ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1288ea936aebSAl Viro return true; 1289ea936aebSAl Viro 1290ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1291ea936aebSAl Viro return false; 1292ea936aebSAl Viro 129362a7375eSIan Kent for (;;) { 129462a7375eSIan Kent /* 129562a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 129662a7375eSIan Kent * that wants to block transit. 129762a7375eSIan Kent */ 1298ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1299ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1300ea936aebSAl Viro if (res) 1301ea936aebSAl Viro return res == -EISDIR; 1302ea936aebSAl Viro flags = dentry->d_flags; 1303b8faf035SNeilBrown } 130462a7375eSIan Kent 1305ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1306ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1307ea936aebSAl Viro if (mounted) { 1308c7105365SAl Viro path->mnt = &mounted->mnt; 1309ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1310a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1311ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1312ea936aebSAl Viro *inode = dentry->d_inode; 131359430262SLinus Torvalds /* 1314ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1315ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1316ea936aebSAl Viro * between mount hash removal and ->mnt_root 1317ea936aebSAl Viro * becoming unpinned. 131859430262SLinus Torvalds */ 1319ea936aebSAl Viro flags = dentry->d_flags; 1320ea936aebSAl Viro continue; 13219875cf80SDavid Howells } 1322ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1323ea936aebSAl Viro return false; 1324ea936aebSAl Viro } 1325ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1326ea936aebSAl Viro } 1327287548e4SAl Viro } 1328287548e4SAl Viro 1329db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1330db3c9adeSAl Viro struct path *path, struct inode **inode, 1331db3c9adeSAl Viro unsigned int *seqp) 1332bd7c4b50SAl Viro { 13339deed3ebSAl Viro bool jumped; 1334db3c9adeSAl Viro int ret; 1335bd7c4b50SAl Viro 1336db3c9adeSAl Viro path->mnt = nd->path.mnt; 1337db3c9adeSAl Viro path->dentry = dentry; 1338c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1339c153007bSAl Viro unsigned int seq = *seqp; 1340c153007bSAl Viro if (unlikely(!*inode)) 1341c153007bSAl Viro return -ENOENT; 1342c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 13439deed3ebSAl Viro return 0; 1344c153007bSAl Viro if (unlazy_child(nd, dentry, seq)) 1345c153007bSAl Viro return -ECHILD; 1346c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1347c153007bSAl Viro path->mnt = nd->path.mnt; 1348c153007bSAl Viro path->dentry = dentry; 1349c153007bSAl Viro } 13509deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 13519deed3ebSAl Viro if (jumped) { 13529deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 13539deed3ebSAl Viro ret = -EXDEV; 13549deed3ebSAl Viro else 13559deed3ebSAl Viro nd->flags |= LOOKUP_JUMPED; 13569deed3ebSAl Viro } 13579deed3ebSAl Viro if (unlikely(ret)) { 13589deed3ebSAl Viro dput(path->dentry); 13599deed3ebSAl Viro if (path->mnt != nd->path.mnt) 13609deed3ebSAl Viro mntput(path->mnt); 13619deed3ebSAl Viro } else { 1362bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1363bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1364bd7c4b50SAl Viro } 1365bd7c4b50SAl Viro return ret; 1366bd7c4b50SAl Viro } 1367bd7c4b50SAl Viro 136831e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 136931e6b01fSNick Piggin { 13704023bfc9SAl Viro struct inode *inode = nd->inode; 137131e6b01fSNick Piggin 137231e6b01fSNick Piggin while (1) { 1373adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1374adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1375adb21d2bSAleksa Sarai return -ECHILD; 137631e6b01fSNick Piggin break; 1377adb21d2bSAleksa Sarai } 137831e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 137931e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 138031e6b01fSNick Piggin struct dentry *parent = old->d_parent; 138131e6b01fSNick Piggin unsigned seq; 138231e6b01fSNick Piggin 13834023bfc9SAl Viro inode = parent->d_inode; 138431e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1385aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1386aed434adSAl Viro return -ECHILD; 138731e6b01fSNick Piggin nd->path.dentry = parent; 138831e6b01fSNick Piggin nd->seq = seq; 1389397d425dSEric W. Biederman if (unlikely(!path_connected(&nd->path))) 13902b98149cSAleksa Sarai return -ECHILD; 139131e6b01fSNick Piggin break; 1392aed434adSAl Viro } else { 1393aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1394aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1395aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1396aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1397aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1398aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1399aed434adSAl Viro return -ECHILD; 1400aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 140131e6b01fSNick Piggin break; 140272ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 140372ba2929SAleksa Sarai return -ECHILD; 1404aed434adSAl Viro /* we know that mountpoint was pinned */ 1405aed434adSAl Viro nd->path.dentry = mountpoint; 1406aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1407aed434adSAl Viro inode = inode2; 1408aed434adSAl Viro nd->seq = seq; 140931e6b01fSNick Piggin } 1410aed434adSAl Viro } 1411aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1412b37199e6SAl Viro struct mount *mounted; 1413b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1414aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1415aed434adSAl Viro return -ECHILD; 1416b37199e6SAl Viro if (!mounted) 1417b37199e6SAl Viro break; 141872ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 141972ba2929SAleksa Sarai return -ECHILD; 1420b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1421b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 14224023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1423b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1424b37199e6SAl Viro } 14254023bfc9SAl Viro nd->inode = inode; 142631e6b01fSNick Piggin return 0; 142731e6b01fSNick Piggin } 142831e6b01fSNick Piggin 14299875cf80SDavid Howells /* 14309875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14319875cf80SDavid Howells */ 14329875cf80SDavid Howells static void follow_mount(struct path *path) 14339875cf80SDavid Howells { 14349875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14359875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14369875cf80SDavid Howells if (!mounted) 14379875cf80SDavid Howells break; 14389875cf80SDavid Howells dput(path->dentry); 14399875cf80SDavid Howells mntput(path->mnt); 14409875cf80SDavid Howells path->mnt = mounted; 14419875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14429875cf80SDavid Howells } 14439875cf80SDavid Howells } 14449875cf80SDavid Howells 1445eedf265aSEric W. Biederman static int path_parent_directory(struct path *path) 1446eedf265aSEric W. Biederman { 1447eedf265aSEric W. Biederman struct dentry *old = path->dentry; 1448eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 1449eedf265aSEric W. Biederman path->dentry = dget_parent(path->dentry); 1450eedf265aSEric W. Biederman dput(old); 1451eedf265aSEric W. Biederman if (unlikely(!path_connected(path))) 1452eedf265aSEric W. Biederman return -ENOENT; 1453eedf265aSEric W. Biederman return 0; 1454eedf265aSEric W. Biederman } 1455eedf265aSEric W. Biederman 1456397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14571da177e4SLinus Torvalds { 14581da177e4SLinus Torvalds while (1) { 1459adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1460adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1461adb21d2bSAleksa Sarai return -EXDEV; 14621da177e4SLinus Torvalds break; 1463adb21d2bSAleksa Sarai } 14644ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) { 1465eedf265aSEric W. Biederman int ret = path_parent_directory(&nd->path); 1466eedf265aSEric W. Biederman if (ret) 1467eedf265aSEric W. Biederman return ret; 14681da177e4SLinus Torvalds break; 14691da177e4SLinus Torvalds } 14703088dd70SAl Viro if (!follow_up(&nd->path)) 14711da177e4SLinus Torvalds break; 147272ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 147372ba2929SAleksa Sarai return -EXDEV; 14741da177e4SLinus Torvalds } 147579ed0226SAl Viro follow_mount(&nd->path); 147631e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1477397d425dSEric W. Biederman return 0; 14781da177e4SLinus Torvalds } 14791da177e4SLinus Torvalds 14801da177e4SLinus Torvalds /* 1481f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1482f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1483baa03890SNick Piggin */ 1484e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1485e3c13928SAl Viro struct dentry *dir, 14866c51e513SAl Viro unsigned int flags) 1487baa03890SNick Piggin { 1488a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1489bad61189SMiklos Szeredi if (dentry) { 1490a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1491bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 149274ff0ffcSAl Viro if (!error) 14935542aa2fSEric W. Biederman d_invalidate(dentry); 1494bad61189SMiklos Szeredi dput(dentry); 149574ff0ffcSAl Viro return ERR_PTR(error); 1496bad61189SMiklos Szeredi } 1497bad61189SMiklos Szeredi } 1498baa03890SNick Piggin return dentry; 1499baa03890SNick Piggin } 1500baa03890SNick Piggin 1501baa03890SNick Piggin /* 1502a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1503a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1504a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1505a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1506a03ece5fSAl Viro * at all. 150744396f4bSJosef Bacik */ 1508a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1509a03ece5fSAl Viro struct dentry *base, unsigned int flags) 151044396f4bSJosef Bacik { 1511a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 151244396f4bSJosef Bacik struct dentry *old; 1513a03ece5fSAl Viro struct inode *dir = base->d_inode; 1514a03ece5fSAl Viro 1515a03ece5fSAl Viro if (dentry) 1516a03ece5fSAl Viro return dentry; 151744396f4bSJosef Bacik 151844396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1519a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 152044396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1521a03ece5fSAl Viro 1522a03ece5fSAl Viro dentry = d_alloc(base, name); 1523a03ece5fSAl Viro if (unlikely(!dentry)) 1524a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 152544396f4bSJosef Bacik 152672bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 152744396f4bSJosef Bacik if (unlikely(old)) { 152844396f4bSJosef Bacik dput(dentry); 152944396f4bSJosef Bacik dentry = old; 153044396f4bSJosef Bacik } 153144396f4bSJosef Bacik return dentry; 153244396f4bSJosef Bacik } 153344396f4bSJosef Bacik 153420e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 153520e34357SAl Viro struct inode **inode, 1536254cf582SAl Viro unsigned *seqp) 15371da177e4SLinus Torvalds { 153831e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15395a18fff2SAl Viro int status = 1; 15409875cf80SDavid Howells 15413cac260aSAl Viro /* 1542b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15435d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15445d0f49c1SAl Viro * going to fall back to non-racy lookup. 1545b04f784eSNick Piggin */ 154631e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 154731e6b01fSNick Piggin unsigned seq; 1548da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15495d0f49c1SAl Viro if (unlikely(!dentry)) { 15504675ac39SAl Viro if (unlazy_walk(nd)) 155120e34357SAl Viro return ERR_PTR(-ECHILD); 155220e34357SAl Viro return NULL; 15535d0f49c1SAl Viro } 15545a18fff2SAl Viro 155512f8ad4bSLinus Torvalds /* 155612f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 155712f8ad4bSLinus Torvalds * the dentry name information from lookup. 155812f8ad4bSLinus Torvalds */ 155963afdfc7SDavid Howells *inode = d_backing_inode(dentry); 15605d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 156120e34357SAl Viro return ERR_PTR(-ECHILD); 156212f8ad4bSLinus Torvalds 156312f8ad4bSLinus Torvalds /* 156412f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 156512f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 156612f8ad4bSLinus Torvalds * 156712f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 156812f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 156912f8ad4bSLinus Torvalds */ 15705d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 157120e34357SAl Viro return ERR_PTR(-ECHILD); 15725a18fff2SAl Viro 1573254cf582SAl Viro *seqp = seq; 15744ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1575c153007bSAl Viro if (likely(status > 0)) 157620e34357SAl Viro return dentry; 15774675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 157820e34357SAl Viro return ERR_PTR(-ECHILD); 1579209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1580209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1581209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 15825a18fff2SAl Viro } else { 1583e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 158481e6f520SAl Viro if (unlikely(!dentry)) 158520e34357SAl Viro return NULL; 15864ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15875d0f49c1SAl Viro } 15885a18fff2SAl Viro if (unlikely(status <= 0)) { 1589e9742b53SAl Viro if (!status) 15905d0f49c1SAl Viro d_invalidate(dentry); 15915a18fff2SAl Viro dput(dentry); 159220e34357SAl Viro return ERR_PTR(status); 15935a18fff2SAl Viro } 159420e34357SAl Viro return dentry; 1595697f514dSMiklos Szeredi } 1596697f514dSMiklos Szeredi 1597697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 159888d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1599e3c13928SAl Viro struct dentry *dir, 1600e3c13928SAl Viro unsigned int flags) 1601697f514dSMiklos Szeredi { 160288d8331aSAl Viro struct dentry *dentry, *old; 16031936386eSAl Viro struct inode *inode = dir->d_inode; 1604d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16051936386eSAl Viro 16061936386eSAl Viro /* Don't go there if it's already dead */ 160794bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 160888d8331aSAl Viro return ERR_PTR(-ENOENT); 160994bdd655SAl Viro again: 1610d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 161194bdd655SAl Viro if (IS_ERR(dentry)) 161288d8331aSAl Viro return dentry; 161394bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1614949a852eSAl Viro int error = d_revalidate(dentry, flags); 1615949a852eSAl Viro if (unlikely(error <= 0)) { 161694bdd655SAl Viro if (!error) { 1617949a852eSAl Viro d_invalidate(dentry); 1618949a852eSAl Viro dput(dentry); 161994bdd655SAl Viro goto again; 162094bdd655SAl Viro } 162194bdd655SAl Viro dput(dentry); 1622949a852eSAl Viro dentry = ERR_PTR(error); 1623949a852eSAl Viro } 162494bdd655SAl Viro } else { 16251936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 162685c7f810SAl Viro d_lookup_done(dentry); 16271936386eSAl Viro if (unlikely(old)) { 16281936386eSAl Viro dput(dentry); 16291936386eSAl Viro dentry = old; 1630949a852eSAl Viro } 1631949a852eSAl Viro } 1632e3c13928SAl Viro return dentry; 16331da177e4SLinus Torvalds } 16341da177e4SLinus Torvalds 163588d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 163688d8331aSAl Viro struct dentry *dir, 163788d8331aSAl Viro unsigned int flags) 163888d8331aSAl Viro { 163988d8331aSAl Viro struct inode *inode = dir->d_inode; 164088d8331aSAl Viro struct dentry *res; 164188d8331aSAl Viro inode_lock_shared(inode); 164288d8331aSAl Viro res = __lookup_slow(name, dir, flags); 164388d8331aSAl Viro inode_unlock_shared(inode); 164488d8331aSAl Viro return res; 164588d8331aSAl Viro } 164688d8331aSAl Viro 164752094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 164852094c8aSAl Viro { 164952094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16504ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 165152094c8aSAl Viro if (err != -ECHILD) 165252094c8aSAl Viro return err; 16534675ac39SAl Viro if (unlazy_walk(nd)) 165452094c8aSAl Viro return -ECHILD; 165552094c8aSAl Viro } 16564ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 165752094c8aSAl Viro } 165852094c8aSAl Viro 16599856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16609856fa1bSAl Viro { 16619856fa1bSAl Viro if (type == LAST_DOTDOT) { 1662740a1678SAleksa Sarai int error = 0; 1663740a1678SAleksa Sarai 1664740a1678SAleksa Sarai if (!nd->root.mnt) { 1665740a1678SAleksa Sarai error = set_root(nd); 1666740a1678SAleksa Sarai if (error) 1667740a1678SAleksa Sarai return error; 1668740a1678SAleksa Sarai } 1669ab87f9a5SAleksa Sarai if (nd->flags & LOOKUP_RCU) 1670ab87f9a5SAleksa Sarai error = follow_dotdot_rcu(nd); 1671ab87f9a5SAleksa Sarai else 1672ab87f9a5SAleksa Sarai error = follow_dotdot(nd); 1673ab87f9a5SAleksa Sarai if (error) 1674ab87f9a5SAleksa Sarai return error; 1675ab87f9a5SAleksa Sarai 1676ab87f9a5SAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1677ab87f9a5SAleksa Sarai /* 1678ab87f9a5SAleksa Sarai * If there was a racing rename or mount along our 1679ab87f9a5SAleksa Sarai * path, then we can't be sure that ".." hasn't jumped 1680ab87f9a5SAleksa Sarai * above nd->root (and so userspace should retry or use 1681ab87f9a5SAleksa Sarai * some fallback). 1682ab87f9a5SAleksa Sarai */ 1683ab87f9a5SAleksa Sarai smp_rmb(); 1684ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 1685ab87f9a5SAleksa Sarai return -EAGAIN; 1686ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 1687ab87f9a5SAleksa Sarai return -EAGAIN; 1688ab87f9a5SAleksa Sarai } 16899856fa1bSAl Viro } 16909856fa1bSAl Viro return 0; 16919856fa1bSAl Viro } 16929856fa1bSAl Viro 1693b1a81972SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 1694b1a81972SAl Viro 169506708adbSAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 1696b1a81972SAl Viro struct inode *inode, unsigned seq, int flags) 1697d63ff28fSAl Viro { 16981cf2665bSAl Viro struct saved *last; 1699ad6cc4c3SAl Viro const char *res; 1700ad6cc4c3SAl Viro int error; 1701ad6cc4c3SAl Viro 1702756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1703626de996SAl Viro path_to_nameidata(link, nd); 170406708adbSAl Viro return ERR_PTR(-ELOOP); 1705626de996SAl Viro } 1706bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1707cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1708cd179f44SAl Viro mntget(link->mnt); 17097973387aSAl Viro } 1710626de996SAl Viro error = nd_alloc_stack(nd); 1711626de996SAl Viro if (unlikely(error)) { 1712bc40aee0SAl Viro if (error == -ECHILD) { 1713ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1714ad1633a1SAl Viro drop_links(nd); 1715ad1633a1SAl Viro nd->depth = 0; 1716ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1717ad1633a1SAl Viro nd->path.mnt = NULL; 1718ad1633a1SAl Viro nd->path.dentry = NULL; 1719ad1633a1SAl Viro rcu_read_unlock(); 17204675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1721bc40aee0SAl Viro error = nd_alloc_stack(nd); 1722bc40aee0SAl Viro } 1723bc40aee0SAl Viro if (error) { 1724cd179f44SAl Viro path_put(link); 172506708adbSAl Viro return ERR_PTR(error); 1726626de996SAl Viro } 1727bc40aee0SAl Viro } 1728626de996SAl Viro 1729ab104923SAl Viro last = nd->stack + nd->depth++; 17301cf2665bSAl Viro last->link = *link; 1731fceef393SAl Viro clear_delayed_call(&last->done); 17320450b2d1SAl Viro last->seq = seq; 1733ad6cc4c3SAl Viro 1734b1a81972SAl Viro if (flags & WALK_TRAILING) { 1735ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1736ad6cc4c3SAl Viro if (unlikely(error)) 1737ad6cc4c3SAl Viro return ERR_PTR(error); 1738ad6cc4c3SAl Viro } 1739ad6cc4c3SAl Viro 1740ad6cc4c3SAl Viro if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) 1741ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1742ad6cc4c3SAl Viro 1743ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1744ad6cc4c3SAl Viro touch_atime(&last->link); 1745ad6cc4c3SAl Viro cond_resched(); 1746ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1747ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1748ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1749ad6cc4c3SAl Viro touch_atime(&last->link); 1750ad6cc4c3SAl Viro } 1751ad6cc4c3SAl Viro 1752ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1753ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1754ad6cc4c3SAl Viro if (unlikely(error)) 1755ad6cc4c3SAl Viro return ERR_PTR(error); 1756ad6cc4c3SAl Viro 1757ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1758ad6cc4c3SAl Viro if (!res) { 1759ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1760ad6cc4c3SAl Viro struct delayed_call *); 1761ad6cc4c3SAl Viro get = inode->i_op->get_link; 1762ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1763ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1764ad6cc4c3SAl Viro if (res == ERR_PTR(-ECHILD)) { 1765ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1766ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1767ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1768ad6cc4c3SAl Viro } 1769ad6cc4c3SAl Viro } else { 1770ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1771ad6cc4c3SAl Viro } 1772ad6cc4c3SAl Viro if (!res) 1773ad6cc4c3SAl Viro goto all_done; 1774ad6cc4c3SAl Viro if (IS_ERR(res)) 1775ad6cc4c3SAl Viro return res; 1776ad6cc4c3SAl Viro } 1777ad6cc4c3SAl Viro if (*res == '/') { 1778ad6cc4c3SAl Viro error = nd_jump_root(nd); 1779ad6cc4c3SAl Viro if (unlikely(error)) 1780ad6cc4c3SAl Viro return ERR_PTR(error); 1781ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1782ad6cc4c3SAl Viro ; 1783ad6cc4c3SAl Viro } 1784ad6cc4c3SAl Viro if (*res) 1785ad6cc4c3SAl Viro return res; 1786ad6cc4c3SAl Viro all_done: // pure jump 1787ad6cc4c3SAl Viro put_link(nd); 1788ad6cc4c3SAl Viro return NULL; 1789d63ff28fSAl Viro } 1790d63ff28fSAl Viro 17913ddcd056SLinus Torvalds /* 17923ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17933ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17943ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17953ddcd056SLinus Torvalds * for the common case. 17963ddcd056SLinus Torvalds */ 1797b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1798cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 17993ddcd056SLinus Torvalds { 1800cbae4d12SAl Viro struct path path; 1801cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1802cbae4d12SAl Viro 1803cbae4d12SAl Viro if (err < 0) 1804b0417d2cSAl Viro return ERR_PTR(err); 1805cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18068c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1807aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18088f64fb1cSAl Viro /* not a symlink or should not follow */ 1809cbae4d12SAl Viro path_to_nameidata(&path, nd); 18108f64fb1cSAl Viro nd->inode = inode; 18118f64fb1cSAl Viro nd->seq = seq; 1812b0417d2cSAl Viro return NULL; 18138f64fb1cSAl Viro } 1814a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1815a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 1816cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1817b0417d2cSAl Viro return ERR_PTR(-ECHILD); 1818a7f77542SAl Viro } 1819b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18203ddcd056SLinus Torvalds } 18213ddcd056SLinus Torvalds 182292d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1823ce57dfc1SAl Viro { 1824db3c9adeSAl Viro struct dentry *dentry; 1825ce57dfc1SAl Viro struct inode *inode; 1826254cf582SAl Viro unsigned seq; 1827ce57dfc1SAl Viro int err; 1828ce57dfc1SAl Viro /* 1829ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1830ce57dfc1SAl Viro * to be able to know about the current root directory and 1831ce57dfc1SAl Viro * parent relationships. 1832ce57dfc1SAl Viro */ 18334693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18341c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18354693a547SAl Viro put_link(nd); 183656676ec3SAl Viro err = handle_dots(nd, nd->last_type); 183792d27016SAl Viro return ERR_PTR(err); 18384693a547SAl Viro } 183920e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 184020e34357SAl Viro if (IS_ERR(dentry)) 184192d27016SAl Viro return ERR_CAST(dentry); 184220e34357SAl Viro if (unlikely(!dentry)) { 1843db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1844db3c9adeSAl Viro if (IS_ERR(dentry)) 184592d27016SAl Viro return ERR_CAST(dentry); 184620e34357SAl Viro } 184756676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 184856676ec3SAl Viro put_link(nd); 1849b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1850ce57dfc1SAl Viro } 1851ce57dfc1SAl Viro 18521da177e4SLinus Torvalds /* 1853bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1854bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1855bfcfaa77SLinus Torvalds * 1856bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1857bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1858bfcfaa77SLinus Torvalds * fast. 1859bfcfaa77SLinus Torvalds * 1860bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1861bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1862bfcfaa77SLinus Torvalds * crossing operation. 1863bfcfaa77SLinus Torvalds * 1864bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1865bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1866bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1867bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1868bfcfaa77SLinus Torvalds */ 1869bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1870bfcfaa77SLinus Torvalds 1871f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1872bfcfaa77SLinus Torvalds 1873468a9428SGeorge Spelvin #ifdef HASH_MIX 1874bfcfaa77SLinus Torvalds 1875468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1876468a9428SGeorge Spelvin 1877468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18782a18da7aSGeorge Spelvin /* 18792a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18802a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18812a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18822a18da7aSGeorge Spelvin * and no temporaries. 18832a18da7aSGeorge Spelvin * 18842a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18852a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18862a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18872a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18882a18da7aSGeorge Spelvin * 18892a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18902a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18912a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18922a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18932a18da7aSGeorge Spelvin * 18942a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18952a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18962a18da7aSGeorge Spelvin * toggling any given output bit. 18972a18da7aSGeorge Spelvin * 18982a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 18992a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19002a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 19012a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 19022a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 19032a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 19042a18da7aSGeorge Spelvin * Perfect: 8192 258048 19052a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 19062a18da7aSGeorge Spelvin */ 19072a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19082a18da7aSGeorge Spelvin ( x ^= (a), \ 19092a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 19102a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 19112a18da7aSGeorge Spelvin y *= 9 ) 1912bfcfaa77SLinus Torvalds 19130fed3ac8SGeorge Spelvin /* 19142a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 19152a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 19162a18da7aSGeorge Spelvin * work done before the hash value is used. 19170fed3ac8SGeorge Spelvin */ 19182a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19190fed3ac8SGeorge Spelvin { 19202a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 19212a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19222a18da7aSGeorge Spelvin return y >> 32; 19230fed3ac8SGeorge Spelvin } 19240fed3ac8SGeorge Spelvin 1925bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1926bfcfaa77SLinus Torvalds 19272a18da7aSGeorge Spelvin /* 19282a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19292a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19302a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19312a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19322a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19332a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19342a18da7aSGeorge Spelvin * Perfect: 2048 31744 19352a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19362a18da7aSGeorge Spelvin */ 19372a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19382a18da7aSGeorge Spelvin ( x ^= (a), \ 19392a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19402a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19412a18da7aSGeorge Spelvin y *= 9 ) 1942bfcfaa77SLinus Torvalds 19432a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19440fed3ac8SGeorge Spelvin { 19452a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19462a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19470fed3ac8SGeorge Spelvin } 19480fed3ac8SGeorge Spelvin 1949bfcfaa77SLinus Torvalds #endif 1950bfcfaa77SLinus Torvalds 19512a18da7aSGeorge Spelvin /* 19522a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19532a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19542a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19552a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19562a18da7aSGeorge Spelvin * finds the delimiter after the name. 19572a18da7aSGeorge Spelvin */ 19588387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1959bfcfaa77SLinus Torvalds { 19608387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1961bfcfaa77SLinus Torvalds 1962bfcfaa77SLinus Torvalds for (;;) { 1963fcfd2fbfSGeorge Spelvin if (!len) 1964fcfd2fbfSGeorge Spelvin goto done; 1965e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1966bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1967bfcfaa77SLinus Torvalds break; 19682a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1969bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1970bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1971bfcfaa77SLinus Torvalds } 19722a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1973bfcfaa77SLinus Torvalds done: 19742a18da7aSGeorge Spelvin return fold_hash(x, y); 1975bfcfaa77SLinus Torvalds } 1976bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1977bfcfaa77SLinus Torvalds 1978fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19798387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1980fcfd2fbfSGeorge Spelvin { 19818387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19828387ff25SLinus Torvalds unsigned long adata, mask, len; 1983fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1984fcfd2fbfSGeorge Spelvin 19858387ff25SLinus Torvalds len = 0; 19868387ff25SLinus Torvalds goto inside; 19878387ff25SLinus Torvalds 1988fcfd2fbfSGeorge Spelvin do { 19892a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1990fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19918387ff25SLinus Torvalds inside: 1992fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1993fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1994fcfd2fbfSGeorge Spelvin 1995fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1996fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19972a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 1998fcfd2fbfSGeorge Spelvin 19992a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2000fcfd2fbfSGeorge Spelvin } 2001fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2002fcfd2fbfSGeorge Spelvin 2003bfcfaa77SLinus Torvalds /* 2004bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2005d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2006bfcfaa77SLinus Torvalds */ 20078387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2008bfcfaa77SLinus Torvalds { 20098387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 20108387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 201136126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2012bfcfaa77SLinus Torvalds 20138387ff25SLinus Torvalds len = 0; 20148387ff25SLinus Torvalds goto inside; 20158387ff25SLinus Torvalds 2016bfcfaa77SLinus Torvalds do { 20172a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2018bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 20198387ff25SLinus Torvalds inside: 2020e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 202136126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 202236126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2023bfcfaa77SLinus Torvalds 202436126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 202536126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 202636126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20272a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 202836126f8fSLinus Torvalds 20292a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2030bfcfaa77SLinus Torvalds } 2031bfcfaa77SLinus Torvalds 20322a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2033bfcfaa77SLinus Torvalds 2034fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20358387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20360145acc2SLinus Torvalds { 20378387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20380145acc2SLinus Torvalds while (len--) 2039fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20400145acc2SLinus Torvalds return end_name_hash(hash); 20410145acc2SLinus Torvalds } 2042ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20430145acc2SLinus Torvalds 2044fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20458387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2046fcfd2fbfSGeorge Spelvin { 20478387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2048fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2049fcfd2fbfSGeorge Spelvin 2050fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2051e0ab7af9SGeorge Spelvin while (c) { 2052fcfd2fbfSGeorge Spelvin len++; 2053fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2054fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2055e0ab7af9SGeorge Spelvin } 2056fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2057fcfd2fbfSGeorge Spelvin } 2058f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2059fcfd2fbfSGeorge Spelvin 20603ddcd056SLinus Torvalds /* 2061200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2062200e9ef7SLinus Torvalds * one character. 2063200e9ef7SLinus Torvalds */ 20648387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2065200e9ef7SLinus Torvalds { 20668387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2067200e9ef7SLinus Torvalds unsigned long len = 0, c; 2068200e9ef7SLinus Torvalds 2069200e9ef7SLinus Torvalds c = (unsigned char)*name; 2070200e9ef7SLinus Torvalds do { 2071200e9ef7SLinus Torvalds len++; 2072200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2073200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2074200e9ef7SLinus Torvalds } while (c && c != '/'); 2075d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2076200e9ef7SLinus Torvalds } 2077200e9ef7SLinus Torvalds 2078bfcfaa77SLinus Torvalds #endif 2079bfcfaa77SLinus Torvalds 2080200e9ef7SLinus Torvalds /* 20811da177e4SLinus Torvalds * Name resolution. 2082ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2083ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20841da177e4SLinus Torvalds * 2085ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2086ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20871da177e4SLinus Torvalds */ 20886de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20891da177e4SLinus Torvalds { 2090d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 20911da177e4SLinus Torvalds int err; 20921da177e4SLinus Torvalds 2093b4c03536SAl Viro nd->last_type = LAST_ROOT; 2094c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 20959b5858e9SAl Viro if (IS_ERR(name)) 20969b5858e9SAl Viro return PTR_ERR(name); 20971da177e4SLinus Torvalds while (*name=='/') 20981da177e4SLinus Torvalds name++; 20991da177e4SLinus Torvalds if (!*name) 21009e18f10aSAl Viro return 0; 21011da177e4SLinus Torvalds 21021da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 21031da177e4SLinus Torvalds for(;;) { 210492d27016SAl Viro const char *link; 2105d6bb3e90SLinus Torvalds u64 hash_len; 2106fe479a58SAl Viro int type; 21071da177e4SLinus Torvalds 210852094c8aSAl Viro err = may_lookup(nd); 21091da177e4SLinus Torvalds if (err) 21103595e234SAl Viro return err; 21111da177e4SLinus Torvalds 21128387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 21131da177e4SLinus Torvalds 2114fe479a58SAl Viro type = LAST_NORM; 2115d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2116fe479a58SAl Viro case 2: 2117200e9ef7SLinus Torvalds if (name[1] == '.') { 2118fe479a58SAl Viro type = LAST_DOTDOT; 211916c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 212016c2cd71SAl Viro } 2121fe479a58SAl Viro break; 2122fe479a58SAl Viro case 1: 2123fe479a58SAl Viro type = LAST_DOT; 2124fe479a58SAl Viro } 21255a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21265a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 212716c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21285a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2129a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2130da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21315a202bcdSAl Viro if (err < 0) 21323595e234SAl Viro return err; 2133d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2134d6bb3e90SLinus Torvalds name = this.name; 21355a202bcdSAl Viro } 21365a202bcdSAl Viro } 2137fe479a58SAl Viro 2138d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2139d6bb3e90SLinus Torvalds nd->last.name = name; 21405f4a6a69SAl Viro nd->last_type = type; 21415f4a6a69SAl Viro 2142d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2143d6bb3e90SLinus Torvalds if (!*name) 2144bdf6cbf1SAl Viro goto OK; 2145200e9ef7SLinus Torvalds /* 2146200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2147200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2148200e9ef7SLinus Torvalds */ 2149200e9ef7SLinus Torvalds do { 2150d6bb3e90SLinus Torvalds name++; 2151d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21528620c238SAl Viro if (unlikely(!*name)) { 21538620c238SAl Viro OK: 2154d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2155c108837eSAl Viro if (!depth) { 2156c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 21578620c238SAl Viro return 0; 2158c108837eSAl Viro } 21598620c238SAl Viro /* last component of nested symlink */ 2160d8d4611aSAl Viro name = nd->stack[--depth].name; 21618c4efe22SAl Viro link = walk_component(nd, 0); 21621c4ff1a8SAl Viro } else { 21631c4ff1a8SAl Viro /* not the last component */ 21648c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 21658620c238SAl Viro } 216692d27016SAl Viro if (unlikely(link)) { 216792d27016SAl Viro if (IS_ERR(link)) 216892d27016SAl Viro return PTR_ERR(link); 216992d27016SAl Viro /* a symlink to follow */ 2170d8d4611aSAl Viro nd->stack[depth++].name = name; 217192d27016SAl Viro name = link; 21729e18f10aSAl Viro continue; 217348c8b0c5SAl Viro } 217497242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 217597242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21764675ac39SAl Viro if (unlazy_walk(nd)) 217797242f99SAl Viro return -ECHILD; 217897242f99SAl Viro } 21793595e234SAl Viro return -ENOTDIR; 21805f4a6a69SAl Viro } 2181ce57dfc1SAl Viro } 218297242f99SAl Viro } 21831da177e4SLinus Torvalds 2184edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2185c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 218631e6b01fSNick Piggin { 2187740a1678SAleksa Sarai int error; 2188c8a53ee5SAl Viro const char *s = nd->name->name; 218931e6b01fSNick Piggin 2190c0eb027eSLinus Torvalds if (!*s) 2191c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2192edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2193edc2b1daSAl Viro rcu_read_lock(); 2194c0eb027eSLinus Torvalds 2195c108837eSAl Viro nd->flags = flags | LOOKUP_JUMPED; 219631e6b01fSNick Piggin nd->depth = 0; 2197ab87f9a5SAleksa Sarai 2198ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2199ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2200ab87f9a5SAleksa Sarai smp_rmb(); 2201ab87f9a5SAleksa Sarai 22025b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2203b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2204b18825a7SDavid Howells struct inode *inode = root->d_inode; 220593893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2206368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 22075b6ca027SAl Viro nd->path = nd->root; 22085b6ca027SAl Viro nd->inode = inode; 22095b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2210ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22118f47a016SAl Viro nd->root_seq = nd->seq; 22125b6ca027SAl Viro } else { 22135b6ca027SAl Viro path_get(&nd->path); 22145b6ca027SAl Viro } 2215368ee9baSAl Viro return s; 22165b6ca027SAl Viro } 22175b6ca027SAl Viro 221831e6b01fSNick Piggin nd->root.mnt = NULL; 2219248fb5b9SAl Viro nd->path.mnt = NULL; 2220248fb5b9SAl Viro nd->path.dentry = NULL; 222131e6b01fSNick Piggin 22228db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 22238db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2224740a1678SAleksa Sarai error = nd_jump_root(nd); 2225740a1678SAleksa Sarai if (unlikely(error)) 2226740a1678SAleksa Sarai return ERR_PTR(error); 2227ef55d917SAl Viro return s; 22288db52c7eSAleksa Sarai } 22298db52c7eSAleksa Sarai 22308db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 22318db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2232e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 223331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2234c28cc364SNick Piggin unsigned seq; 223531e6b01fSNick Piggin 2236c28cc364SNick Piggin do { 2237c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 223831e6b01fSNick Piggin nd->path = fs->pwd; 2239ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2240c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2241c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2242e41f7d4eSAl Viro } else { 2243e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2244ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2245e41f7d4eSAl Viro } 224631e6b01fSNick Piggin } else { 2247582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2248c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 224931e6b01fSNick Piggin struct dentry *dentry; 225031e6b01fSNick Piggin 22512903ff01SAl Viro if (!f.file) 2252368ee9baSAl Viro return ERR_PTR(-EBADF); 225331e6b01fSNick Piggin 22542903ff01SAl Viro dentry = f.file->f_path.dentry; 225531e6b01fSNick Piggin 2256edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22572903ff01SAl Viro fdput(f); 2258368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2259f52e0c11SAl Viro } 22602903ff01SAl Viro 22612903ff01SAl Viro nd->path = f.file->f_path; 2262e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 226334a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 226434a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22655590ff0dSUlrich Drepper } else { 22662903ff01SAl Viro path_get(&nd->path); 226734a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22681da177e4SLinus Torvalds } 226934a26b99SAl Viro fdput(f); 2270e41f7d4eSAl Viro } 22718db52c7eSAleksa Sarai 2272adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2273adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2274adb21d2bSAleksa Sarai nd->root = nd->path; 2275adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2276adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2277adb21d2bSAleksa Sarai } else { 2278adb21d2bSAleksa Sarai path_get(&nd->root); 2279adb21d2bSAleksa Sarai nd->flags |= LOOKUP_ROOT_GRABBED; 2280adb21d2bSAleksa Sarai } 2281adb21d2bSAleksa Sarai } 2282adb21d2bSAleksa Sarai return s; 22839b4a9b14SAl Viro } 22849b4a9b14SAl Viro 22851ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 228695fa25d9SAl Viro { 2287bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2288bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2289bd92d7feSAl Viro 2290c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2291bd92d7feSAl Viro } 2292bd92d7feSAl Viro 22934f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22944f757f3cSAl Viro { 2295c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2296db3c9adeSAl Viro dget(nd->path.dentry); 2297b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2298b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 22994f757f3cSAl Viro } 23004f757f3cSAl Viro 23019b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2302c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23039b4a9b14SAl Viro { 2304c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2305bd92d7feSAl Viro int err; 230631e6b01fSNick Piggin 23079b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23084f757f3cSAl Viro err = handle_lookup_down(nd); 23095f336e72SAl Viro if (unlikely(err < 0)) 23105f336e72SAl Viro s = ERR_PTR(err); 23114f757f3cSAl Viro } 23124f757f3cSAl Viro 23131ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 23141ccac622SAl Viro (s = lookup_last(nd)) != NULL) 23151ccac622SAl Viro ; 23169f1fafeeSAl Viro if (!err) 23179f1fafeeSAl Viro err = complete_walk(nd); 2318bd92d7feSAl Viro 2319deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2320deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2321bd23a539SAl Viro err = -ENOTDIR; 2322161aff1dSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 2323161aff1dSAl Viro err = handle_lookup_down(nd); 2324161aff1dSAl Viro nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... 2325161aff1dSAl Viro } 2326625b6d10SAl Viro if (!err) { 2327625b6d10SAl Viro *path = nd->path; 2328625b6d10SAl Viro nd->path.mnt = NULL; 2329625b6d10SAl Viro nd->path.dentry = NULL; 2330625b6d10SAl Viro } 2331deb106c6SAl Viro terminate_walk(nd); 2332bd92d7feSAl Viro return err; 233331e6b01fSNick Piggin } 233431e6b01fSNick Piggin 233531d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23369ad1aaa6SAl Viro struct path *path, struct path *root) 2337873f1eedSJeff Layton { 2338894bc8c4SAl Viro int retval; 23399883d185SAl Viro struct nameidata nd; 2340abc9f5beSAl Viro if (IS_ERR(name)) 2341abc9f5beSAl Viro return PTR_ERR(name); 23429ad1aaa6SAl Viro if (unlikely(root)) { 23439ad1aaa6SAl Viro nd.root = *root; 23449ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23459ad1aaa6SAl Viro } 23469883d185SAl Viro set_nameidata(&nd, dfd, name); 2347c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2348873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2349c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2350873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2351c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2352873f1eedSJeff Layton 2353873f1eedSJeff Layton if (likely(!retval)) 2354161aff1dSAl Viro audit_inode(name, path->dentry, 2355161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 23569883d185SAl Viro restore_nameidata(); 2357e4bd1c1aSAl Viro putname(name); 2358873f1eedSJeff Layton return retval; 2359873f1eedSJeff Layton } 2360873f1eedSJeff Layton 23618bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2362c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2363391172c4SAl Viro struct path *parent) 23648bcb77faSAl Viro { 2365c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23669b5858e9SAl Viro int err = link_path_walk(s, nd); 23678bcb77faSAl Viro if (!err) 23688bcb77faSAl Viro err = complete_walk(nd); 2369391172c4SAl Viro if (!err) { 2370391172c4SAl Viro *parent = nd->path; 2371391172c4SAl Viro nd->path.mnt = NULL; 2372391172c4SAl Viro nd->path.dentry = NULL; 2373391172c4SAl Viro } 2374deb106c6SAl Viro terminate_walk(nd); 23758bcb77faSAl Viro return err; 23768bcb77faSAl Viro } 23778bcb77faSAl Viro 23785c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2379391172c4SAl Viro unsigned int flags, struct path *parent, 2380391172c4SAl Viro struct qstr *last, int *type) 23818bcb77faSAl Viro { 23828bcb77faSAl Viro int retval; 23839883d185SAl Viro struct nameidata nd; 23848bcb77faSAl Viro 23855c31b6ceSAl Viro if (IS_ERR(name)) 23865c31b6ceSAl Viro return name; 23879883d185SAl Viro set_nameidata(&nd, dfd, name); 2388c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23898bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2390c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23918bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2392c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2393391172c4SAl Viro if (likely(!retval)) { 2394391172c4SAl Viro *last = nd.last; 2395391172c4SAl Viro *type = nd.last_type; 2396c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 23975c31b6ceSAl Viro } else { 23985c31b6ceSAl Viro putname(name); 23995c31b6ceSAl Viro name = ERR_PTR(retval); 2400391172c4SAl Viro } 24019883d185SAl Viro restore_nameidata(); 24025c31b6ceSAl Viro return name; 24038bcb77faSAl Viro } 24048bcb77faSAl Viro 240579714f72SAl Viro /* does lookup, returns the object with parent locked */ 240679714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24075590ff0dSUlrich Drepper { 24085c31b6ceSAl Viro struct filename *filename; 24095c31b6ceSAl Viro struct dentry *d; 2410391172c4SAl Viro struct qstr last; 2411391172c4SAl Viro int type; 241251689104SPaul Moore 24135c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24145c31b6ceSAl Viro &last, &type); 241551689104SPaul Moore if (IS_ERR(filename)) 241651689104SPaul Moore return ERR_CAST(filename); 24175c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2418391172c4SAl Viro path_put(path); 24195c31b6ceSAl Viro putname(filename); 24205c31b6ceSAl Viro return ERR_PTR(-EINVAL); 242179714f72SAl Viro } 24225955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2423391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 242479714f72SAl Viro if (IS_ERR(d)) { 24255955102cSAl Viro inode_unlock(path->dentry->d_inode); 2426391172c4SAl Viro path_put(path); 242779714f72SAl Viro } 242851689104SPaul Moore putname(filename); 242979714f72SAl Viro return d; 24305590ff0dSUlrich Drepper } 24315590ff0dSUlrich Drepper 2432d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2433d1811465SAl Viro { 2434abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2435abc9f5beSAl Viro flags, path, NULL); 2436d1811465SAl Viro } 24374d359507SAl Viro EXPORT_SYMBOL(kern_path); 2438d1811465SAl Viro 243916f18200SJosef 'Jeff' Sipek /** 244016f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 244116f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 244216f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 244316f18200SJosef 'Jeff' Sipek * @name: pointer to file name 244416f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2445e0a01249SAl Viro * @path: pointer to struct path to fill 244616f18200SJosef 'Jeff' Sipek */ 244716f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 244816f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2449e0a01249SAl Viro struct path *path) 245016f18200SJosef 'Jeff' Sipek { 24519ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24529ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2453abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2454abc9f5beSAl Viro flags , path, &root); 245516f18200SJosef 'Jeff' Sipek } 24564d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 245716f18200SJosef 'Jeff' Sipek 24583c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24593c95f0dcSAl Viro int len, struct qstr *this) 24603c95f0dcSAl Viro { 24613c95f0dcSAl Viro this->name = name; 24623c95f0dcSAl Viro this->len = len; 24633c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24643c95f0dcSAl Viro if (!len) 24653c95f0dcSAl Viro return -EACCES; 24663c95f0dcSAl Viro 24673c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24683c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24693c95f0dcSAl Viro return -EACCES; 24703c95f0dcSAl Viro } 24713c95f0dcSAl Viro 24723c95f0dcSAl Viro while (len--) { 24733c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24743c95f0dcSAl Viro if (c == '/' || c == '\0') 24753c95f0dcSAl Viro return -EACCES; 24763c95f0dcSAl Viro } 24773c95f0dcSAl Viro /* 24783c95f0dcSAl Viro * See if the low-level filesystem might want 24793c95f0dcSAl Viro * to use its own hash.. 24803c95f0dcSAl Viro */ 24813c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 24823c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 24833c95f0dcSAl Viro if (err < 0) 24843c95f0dcSAl Viro return err; 24853c95f0dcSAl Viro } 24863c95f0dcSAl Viro 24873c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 24883c95f0dcSAl Viro } 24893c95f0dcSAl Viro 2490eead1911SChristoph Hellwig /** 24910da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 24920da0b7fdSDavid Howells * @name: pathname component to lookup 24930da0b7fdSDavid Howells * @base: base directory to lookup from 24940da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 24950da0b7fdSDavid Howells * 24960da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 24970da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 24980da0b7fdSDavid Howells * 24990da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25000da0b7fdSDavid Howells * not be called by generic code. 25010da0b7fdSDavid Howells * 25020da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25030da0b7fdSDavid Howells */ 25040da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25050da0b7fdSDavid Howells { 25060da0b7fdSDavid Howells struct qstr this; 25070da0b7fdSDavid Howells int err; 25080da0b7fdSDavid Howells 25090da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25100da0b7fdSDavid Howells 25110da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25120da0b7fdSDavid Howells if (err) 25130da0b7fdSDavid Howells return ERR_PTR(err); 25140da0b7fdSDavid Howells 25150da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25160da0b7fdSDavid Howells } 25170da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25180da0b7fdSDavid Howells 25190da0b7fdSDavid Howells /** 2520a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2521eead1911SChristoph Hellwig * @name: pathname component to lookup 2522eead1911SChristoph Hellwig * @base: base directory to lookup from 2523eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2524eead1911SChristoph Hellwig * 2525a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25269e7543e9SAl Viro * not be called by generic code. 2527bbddca8eSNeilBrown * 2528bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2529eead1911SChristoph Hellwig */ 2530057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2531057f6c01SJames Morris { 25328613a209SAl Viro struct dentry *dentry; 2533057f6c01SJames Morris struct qstr this; 2534cda309deSMiklos Szeredi int err; 2535057f6c01SJames Morris 25365955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25372f9092e1SDavid Woodhouse 25383c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2539cda309deSMiklos Szeredi if (err) 2540cda309deSMiklos Szeredi return ERR_PTR(err); 2541cda309deSMiklos Szeredi 25428613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25438613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2544057f6c01SJames Morris } 25454d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2546057f6c01SJames Morris 2547bbddca8eSNeilBrown /** 2548bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2549bbddca8eSNeilBrown * @name: pathname component to lookup 2550bbddca8eSNeilBrown * @base: base directory to lookup from 2551bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2552bbddca8eSNeilBrown * 2553bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2554bbddca8eSNeilBrown * not be called by generic code. 2555bbddca8eSNeilBrown * 2556bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2557bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2558bbddca8eSNeilBrown */ 2559bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2560bbddca8eSNeilBrown struct dentry *base, int len) 2561bbddca8eSNeilBrown { 2562bbddca8eSNeilBrown struct qstr this; 2563bbddca8eSNeilBrown int err; 256420d00ee8SLinus Torvalds struct dentry *ret; 2565bbddca8eSNeilBrown 25663c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2567bbddca8eSNeilBrown if (err) 2568bbddca8eSNeilBrown return ERR_PTR(err); 2569bbddca8eSNeilBrown 257020d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 257120d00ee8SLinus Torvalds if (!ret) 257220d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 257320d00ee8SLinus Torvalds return ret; 2574bbddca8eSNeilBrown } 2575bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2576bbddca8eSNeilBrown 25776c2d4798SAl Viro /* 25786c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 25796c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 25806c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 25816c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 25826c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 25836c2d4798SAl Viro * this one avoids such problems. 25846c2d4798SAl Viro */ 25856c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 25866c2d4798SAl Viro struct dentry *base, int len) 25876c2d4798SAl Viro { 25886c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 25892fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 25906c2d4798SAl Viro dput(ret); 25916c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 25926c2d4798SAl Viro } 25936c2d4798SAl Viro return ret; 25946c2d4798SAl Viro } 25956c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 25966c2d4798SAl Viro 2597eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2598eedf265aSEric W. Biederman int path_pts(struct path *path) 2599eedf265aSEric W. Biederman { 2600eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2601eedf265aSEric W. Biederman * the input path. 2602eedf265aSEric W. Biederman */ 2603eedf265aSEric W. Biederman struct dentry *child, *parent; 2604eedf265aSEric W. Biederman struct qstr this; 2605eedf265aSEric W. Biederman int ret; 2606eedf265aSEric W. Biederman 2607eedf265aSEric W. Biederman ret = path_parent_directory(path); 2608eedf265aSEric W. Biederman if (ret) 2609eedf265aSEric W. Biederman return ret; 2610eedf265aSEric W. Biederman 2611eedf265aSEric W. Biederman parent = path->dentry; 2612eedf265aSEric W. Biederman this.name = "pts"; 2613eedf265aSEric W. Biederman this.len = 3; 2614eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2615eedf265aSEric W. Biederman if (!child) 2616eedf265aSEric W. Biederman return -ENOENT; 2617eedf265aSEric W. Biederman 2618eedf265aSEric W. Biederman path->dentry = child; 2619eedf265aSEric W. Biederman dput(parent); 2620eedf265aSEric W. Biederman follow_mount(path); 2621eedf265aSEric W. Biederman return 0; 2622eedf265aSEric W. Biederman } 2623eedf265aSEric W. Biederman #endif 2624eedf265aSEric W. Biederman 26251fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26261fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26271da177e4SLinus Torvalds { 2628abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2629abc9f5beSAl Viro flags, path, NULL); 26301da177e4SLinus Torvalds } 2631b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26321fa1e7f6SAndy Whitcroft 2633cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 26341da177e4SLinus Torvalds { 26358e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2636da9592edSDavid Howells 26378e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 26381da177e4SLinus Torvalds return 0; 26398e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 26401da177e4SLinus Torvalds return 0; 264123adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 26421da177e4SLinus Torvalds } 2643cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 26441da177e4SLinus Torvalds 26451da177e4SLinus Torvalds /* 26461da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 26471da177e4SLinus Torvalds * whether the type of victim is right. 26481da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 26491da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 26501da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 26511da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 26521da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 26531da177e4SLinus Torvalds * a. be owner of dir, or 26541da177e4SLinus Torvalds * b. be owner of victim, or 26551da177e4SLinus Torvalds * c. have CAP_FOWNER capability 26561da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 26571da177e4SLinus Torvalds * links pointing to it. 26580bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 26590bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 26600bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 26610bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 26620bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 26631da177e4SLinus Torvalds * nfs_async_unlink(). 26641da177e4SLinus Torvalds */ 2665b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 26661da177e4SLinus Torvalds { 266763afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 26681da177e4SLinus Torvalds int error; 26691da177e4SLinus Torvalds 2670b18825a7SDavid Howells if (d_is_negative(victim)) 26711da177e4SLinus Torvalds return -ENOENT; 2672b18825a7SDavid Howells BUG_ON(!inode); 26731da177e4SLinus Torvalds 26741da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2675593d1ce8SEric W. Biederman 2676593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2677593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2678593d1ce8SEric W. Biederman return -EOVERFLOW; 2679593d1ce8SEric W. Biederman 26804fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 26811da177e4SLinus Torvalds 2682f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 26831da177e4SLinus Torvalds if (error) 26841da177e4SLinus Torvalds return error; 26851da177e4SLinus Torvalds if (IS_APPEND(dir)) 26861da177e4SLinus Torvalds return -EPERM; 2687b18825a7SDavid Howells 2688b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 26890bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 26901da177e4SLinus Torvalds return -EPERM; 26911da177e4SLinus Torvalds if (isdir) { 269244b1d530SMiklos Szeredi if (!d_is_dir(victim)) 26931da177e4SLinus Torvalds return -ENOTDIR; 26941da177e4SLinus Torvalds if (IS_ROOT(victim)) 26951da177e4SLinus Torvalds return -EBUSY; 269644b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 26971da177e4SLinus Torvalds return -EISDIR; 26981da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 26991da177e4SLinus Torvalds return -ENOENT; 27001da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27011da177e4SLinus Torvalds return -EBUSY; 27021da177e4SLinus Torvalds return 0; 27031da177e4SLinus Torvalds } 27041da177e4SLinus Torvalds 27051da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27061da177e4SLinus Torvalds * dir. 27071da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27081da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27091da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2710036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2711036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2712036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27131da177e4SLinus Torvalds */ 2714a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 27151da177e4SLinus Torvalds { 2716036d5236SEric W. Biederman struct user_namespace *s_user_ns; 271714e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 27181da177e4SLinus Torvalds if (child->d_inode) 27191da177e4SLinus Torvalds return -EEXIST; 27201da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27211da177e4SLinus Torvalds return -ENOENT; 2722036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2723036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2724036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2725036d5236SEric W. Biederman return -EOVERFLOW; 2726f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 27271da177e4SLinus Torvalds } 27281da177e4SLinus Torvalds 27291da177e4SLinus Torvalds /* 27301da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 27311da177e4SLinus Torvalds */ 27321da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 27331da177e4SLinus Torvalds { 27341da177e4SLinus Torvalds struct dentry *p; 27351da177e4SLinus Torvalds 27361da177e4SLinus Torvalds if (p1 == p2) { 27375955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27381da177e4SLinus Torvalds return NULL; 27391da177e4SLinus Torvalds } 27401da177e4SLinus Torvalds 2741fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 27421da177e4SLinus Torvalds 2743e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2744e2761a11SOGAWA Hirofumi if (p) { 27455955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 27465955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 27471da177e4SLinus Torvalds return p; 27481da177e4SLinus Torvalds } 27491da177e4SLinus Torvalds 2750e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2751e2761a11SOGAWA Hirofumi if (p) { 27525955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27535955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 27541da177e4SLinus Torvalds return p; 27551da177e4SLinus Torvalds } 27561da177e4SLinus Torvalds 27575955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27585955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 27591da177e4SLinus Torvalds return NULL; 27601da177e4SLinus Torvalds } 27614d359507SAl Viro EXPORT_SYMBOL(lock_rename); 27621da177e4SLinus Torvalds 27631da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 27641da177e4SLinus Torvalds { 27655955102cSAl Viro inode_unlock(p1->d_inode); 27661da177e4SLinus Torvalds if (p1 != p2) { 27675955102cSAl Viro inode_unlock(p2->d_inode); 2768fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 27691da177e4SLinus Torvalds } 27701da177e4SLinus Torvalds } 27714d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 27721da177e4SLinus Torvalds 27734acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2774312b63fbSAl Viro bool want_excl) 27751da177e4SLinus Torvalds { 2776a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 27771da177e4SLinus Torvalds if (error) 27781da177e4SLinus Torvalds return error; 27791da177e4SLinus Torvalds 2780acfa4380SAl Viro if (!dir->i_op->create) 27811da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 27821da177e4SLinus Torvalds mode &= S_IALLUGO; 27831da177e4SLinus Torvalds mode |= S_IFREG; 27841da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 27851da177e4SLinus Torvalds if (error) 27861da177e4SLinus Torvalds return error; 2787312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2788a74574aaSStephen Smalley if (!error) 2789f38aa942SAmy Griffis fsnotify_create(dir, dentry); 27901da177e4SLinus Torvalds return error; 27911da177e4SLinus Torvalds } 27924d359507SAl Viro EXPORT_SYMBOL(vfs_create); 27931da177e4SLinus Torvalds 27948e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 27958e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 27968e6c848eSAl Viro void *arg) 27978e6c848eSAl Viro { 27988e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 27998e6c848eSAl Viro int error = may_create(dir, dentry); 28008e6c848eSAl Viro if (error) 28018e6c848eSAl Viro return error; 28028e6c848eSAl Viro 28038e6c848eSAl Viro mode &= S_IALLUGO; 28048e6c848eSAl Viro mode |= S_IFREG; 28058e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 28068e6c848eSAl Viro if (error) 28078e6c848eSAl Viro return error; 28088e6c848eSAl Viro error = f(dentry, mode, arg); 28098e6c848eSAl Viro if (!error) 28108e6c848eSAl Viro fsnotify_create(dir, dentry); 28118e6c848eSAl Viro return error; 28128e6c848eSAl Viro } 28138e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 28148e6c848eSAl Viro 2815a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2816a2982cc9SEric W. Biederman { 2817a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2818a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2819a2982cc9SEric W. Biederman } 2820a2982cc9SEric W. Biederman 2821f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 28221da177e4SLinus Torvalds { 28233fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 28241da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 28251da177e4SLinus Torvalds int error; 28261da177e4SLinus Torvalds 28271da177e4SLinus Torvalds if (!inode) 28281da177e4SLinus Torvalds return -ENOENT; 28291da177e4SLinus Torvalds 2830c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2831c8fe8f30SChristoph Hellwig case S_IFLNK: 28321da177e4SLinus Torvalds return -ELOOP; 2833c8fe8f30SChristoph Hellwig case S_IFDIR: 2834c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 28351da177e4SLinus Torvalds return -EISDIR; 2836c8fe8f30SChristoph Hellwig break; 2837c8fe8f30SChristoph Hellwig case S_IFBLK: 2838c8fe8f30SChristoph Hellwig case S_IFCHR: 2839a2982cc9SEric W. Biederman if (!may_open_dev(path)) 28401da177e4SLinus Torvalds return -EACCES; 2841c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2842c8fe8f30SChristoph Hellwig case S_IFIFO: 2843c8fe8f30SChristoph Hellwig case S_IFSOCK: 28441da177e4SLinus Torvalds flag &= ~O_TRUNC; 2845c8fe8f30SChristoph Hellwig break; 28464a3fd211SDave Hansen } 2847b41572e9SDave Hansen 284862fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2849b41572e9SDave Hansen if (error) 2850b41572e9SDave Hansen return error; 28516146f0d5SMimi Zohar 28521da177e4SLinus Torvalds /* 28531da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 28541da177e4SLinus Torvalds */ 28551da177e4SLinus Torvalds if (IS_APPEND(inode)) { 28568737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 28577715b521SAl Viro return -EPERM; 28581da177e4SLinus Torvalds if (flag & O_TRUNC) 28597715b521SAl Viro return -EPERM; 28601da177e4SLinus Torvalds } 28611da177e4SLinus Torvalds 28621da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 28632e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 28647715b521SAl Viro return -EPERM; 28651da177e4SLinus Torvalds 2866f3c7691eSJ. Bruce Fields return 0; 28677715b521SAl Viro } 28687715b521SAl Viro 2869e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 28707715b521SAl Viro { 2871f0bb5aafSAl Viro const struct path *path = &filp->f_path; 28727715b521SAl Viro struct inode *inode = path->dentry->d_inode; 28737715b521SAl Viro int error = get_write_access(inode); 28741da177e4SLinus Torvalds if (error) 28757715b521SAl Viro return error; 28761da177e4SLinus Torvalds /* 28771da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 28781da177e4SLinus Torvalds */ 2879d7a06983SJeff Layton error = locks_verify_locked(filp); 2880be6d3e56SKentaro Takeda if (!error) 2881ea0d3ab2STetsuo Handa error = security_path_truncate(path); 28821da177e4SLinus Torvalds if (!error) { 28837715b521SAl Viro error = do_truncate(path->dentry, 0, 2884d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2885e1181ee6SJeff Layton filp); 28861da177e4SLinus Torvalds } 28871da177e4SLinus Torvalds put_write_access(inode); 2888acd0c935SMimi Zohar return error; 28891da177e4SLinus Torvalds } 28901da177e4SLinus Torvalds 2891d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2892d57999e1SDave Hansen { 28938a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 28948a5e929dSAl Viro flag--; 2895d57999e1SDave Hansen return flag; 2896d57999e1SDave Hansen } 2897d57999e1SDave Hansen 2898d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2899d18e9008SMiklos Szeredi { 29001328c727SSeth Forshee struct user_namespace *s_user_ns; 2901d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2902d18e9008SMiklos Szeredi if (error) 2903d18e9008SMiklos Szeredi return error; 2904d18e9008SMiklos Szeredi 29051328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29061328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29071328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29081328c727SSeth Forshee return -EOVERFLOW; 29091328c727SSeth Forshee 2910d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2911d18e9008SMiklos Szeredi if (error) 2912d18e9008SMiklos Szeredi return error; 2913d18e9008SMiklos Szeredi 2914d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2915d18e9008SMiklos Szeredi } 2916d18e9008SMiklos Szeredi 29171acf0af9SDavid Howells /* 29181acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29191acf0af9SDavid Howells * dentry. 29201acf0af9SDavid Howells * 29211acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29221acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29231acf0af9SDavid Howells * 292400a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 292500a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 292600a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 29271acf0af9SDavid Howells * 29281acf0af9SDavid Howells * Returns an error code otherwise. 29291acf0af9SDavid Howells */ 2930239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 2931239eb983SAl Viro struct file *file, 2932015c3bbcSMiklos Szeredi const struct open_flags *op, 29333ec2eef1SAl Viro int open_flag, umode_t mode) 2934d18e9008SMiklos Szeredi { 2935d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2936d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2937d18e9008SMiklos Szeredi int error; 2938d18e9008SMiklos Szeredi 2939384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 2940d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2941d18e9008SMiklos Szeredi 2942d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2943d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2944d18e9008SMiklos Szeredi 294530d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 294630d90494SAl Viro file->f_path.mnt = nd->path.mnt; 29470fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 294844907d79SAl Viro open_to_namei_flags(open_flag), mode); 29496fbd0714SAl Viro d_lookup_done(dentry); 2950384f26e2SAl Viro if (!error) { 295164e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 2952d18e9008SMiklos Szeredi /* 2953384f26e2SAl Viro * We didn't have the inode before the open, so check open 2954384f26e2SAl Viro * permission here. 2955d18e9008SMiklos Szeredi */ 2956384f26e2SAl Viro int acc_mode = op->acc_mode; 295773a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 295803da633aSAl Viro WARN_ON(!(open_flag & O_CREAT)); 295903da633aSAl Viro fsnotify_create(dir, dentry); 296062fb4a15SAl Viro acc_mode = 0; 296103da633aSAl Viro } 29622675a4ebSAl Viro error = may_open(&file->f_path, acc_mode, open_flag); 2963384f26e2SAl Viro if (WARN_ON(error > 0)) 2964384f26e2SAl Viro error = -EINVAL; 296564e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 2966d18e9008SMiklos Szeredi error = -EIO; 2967384f26e2SAl Viro } else { 2968d18e9008SMiklos Szeredi if (file->f_path.dentry) { 2969d18e9008SMiklos Szeredi dput(dentry); 2970d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 297110c64ceaSAl Viro } 297273a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 2973d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 2974239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 2975a01e718fSAl Viro error = -ENOENT; 2976d18e9008SMiklos Szeredi } 2977d18e9008SMiklos Szeredi } 2978239eb983SAl Viro if (error) { 2979d18e9008SMiklos Szeredi dput(dentry); 2980239eb983SAl Viro dentry = ERR_PTR(error); 2981239eb983SAl Viro } 2982239eb983SAl Viro return dentry; 2983d18e9008SMiklos Szeredi } 2984d18e9008SMiklos Szeredi 298531e6b01fSNick Piggin /* 29861acf0af9SDavid Howells * Look up and maybe create and open the last component. 2987d58ffd35SMiklos Szeredi * 298800a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 2989d58ffd35SMiklos Szeredi * 299000a07c15SAl Viro * Returns 0 on success, that is, if 299100a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 299200a07c15SAl Viro * the file was not completely opened at this time, though lookups and 299300a07c15SAl Viro * creations were performed. 299400a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 299500a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 299600a07c15SAl Viro * hadn't been specified. 29971acf0af9SDavid Howells * 299800a07c15SAl Viro * An error code is returned on failure. 2999d58ffd35SMiklos Szeredi */ 3000da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3001d58ffd35SMiklos Szeredi const struct open_flags *op, 30023ec2eef1SAl Viro bool got_write) 3003d58ffd35SMiklos Szeredi { 3004d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 300554ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30061643b43fSAl Viro int open_flag = op->open_flag; 3007d58ffd35SMiklos Szeredi struct dentry *dentry; 30081643b43fSAl Viro int error, create_error = 0; 30091643b43fSAl Viro umode_t mode = op->mode; 30106fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3011d58ffd35SMiklos Szeredi 3012ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3013da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3014d58ffd35SMiklos Szeredi 301573a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 30166fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30176fbd0714SAl Viro for (;;) { 30186fbd0714SAl Viro if (!dentry) { 30196fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3020d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3021da5ebf5aSAl Viro return dentry; 30226fbd0714SAl Viro } 30236fbd0714SAl Viro if (d_in_lookup(dentry)) 30246fbd0714SAl Viro break; 3025d58ffd35SMiklos Szeredi 30266fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 30276fbd0714SAl Viro if (likely(error > 0)) 30286fbd0714SAl Viro break; 30296fbd0714SAl Viro if (error) 30306fbd0714SAl Viro goto out_dput; 30316fbd0714SAl Viro d_invalidate(dentry); 30326fbd0714SAl Viro dput(dentry); 30336fbd0714SAl Viro dentry = NULL; 30346fbd0714SAl Viro } 30356fbd0714SAl Viro if (dentry->d_inode) { 3036d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3037da5ebf5aSAl Viro return dentry; 30386c51e513SAl Viro } 3039d18e9008SMiklos Szeredi 30401643b43fSAl Viro /* 30411643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 30421643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 30431643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 30441643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 30451643b43fSAl Viro * 30461643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 30471643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 30481643b43fSAl Viro */ 30491643b43fSAl Viro if (open_flag & O_CREAT) { 30501643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 30511643b43fSAl Viro mode &= ~current_umask(); 30521643b43fSAl Viro if (unlikely(!got_write)) { 30531643b43fSAl Viro create_error = -EROFS; 30541643b43fSAl Viro open_flag &= ~O_CREAT; 30551643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 30561643b43fSAl Viro goto no_open; 30571643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 30581643b43fSAl Viro } else { 30591643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 30601643b43fSAl Viro if (create_error) { 30611643b43fSAl Viro open_flag &= ~O_CREAT; 30621643b43fSAl Viro if (open_flag & O_EXCL) 30631643b43fSAl Viro goto no_open; 30641643b43fSAl Viro } 30651643b43fSAl Viro } 30661643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 30671643b43fSAl Viro unlikely(!got_write)) { 30681643b43fSAl Viro /* 30691643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 30701643b43fSAl Viro * back to lookup + open 30711643b43fSAl Viro */ 30721643b43fSAl Viro goto no_open; 3073d18e9008SMiklos Szeredi } 3074d18e9008SMiklos Szeredi 30751643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3076239eb983SAl Viro dentry = atomic_open(nd, dentry, file, op, open_flag, mode); 3077da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3078da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3079da5ebf5aSAl Viro return dentry; 3080239eb983SAl Viro } 308154ef4872SMiklos Szeredi 30821643b43fSAl Viro no_open: 30836fbd0714SAl Viro if (d_in_lookup(dentry)) { 308412fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 308512fa5e24SAl Viro nd->flags); 30866fbd0714SAl Viro d_lookup_done(dentry); 308712fa5e24SAl Viro if (unlikely(res)) { 308812fa5e24SAl Viro if (IS_ERR(res)) { 308912fa5e24SAl Viro error = PTR_ERR(res); 309012fa5e24SAl Viro goto out_dput; 309112fa5e24SAl Viro } 309212fa5e24SAl Viro dput(dentry); 309312fa5e24SAl Viro dentry = res; 309412fa5e24SAl Viro } 309554ef4872SMiklos Szeredi } 309654ef4872SMiklos Szeredi 3097d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 30981643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 309973a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3100ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3101ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3102ce8644fcSAl Viro error = -EACCES; 3103d58ffd35SMiklos Szeredi goto out_dput; 310464894cf8SAl Viro } 3105ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31061643b43fSAl Viro open_flag & O_EXCL); 3107d58ffd35SMiklos Szeredi if (error) 3108d58ffd35SMiklos Szeredi goto out_dput; 3109ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3110d58ffd35SMiklos Szeredi } 31111643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31121643b43fSAl Viro error = create_error; 3113d58ffd35SMiklos Szeredi goto out_dput; 3114d58ffd35SMiklos Szeredi } 3115da5ebf5aSAl Viro return dentry; 3116d58ffd35SMiklos Szeredi 3117d58ffd35SMiklos Szeredi out_dput: 3118d58ffd35SMiklos Szeredi dput(dentry); 3119da5ebf5aSAl Viro return ERR_PTR(error); 3120d58ffd35SMiklos Szeredi } 3121d58ffd35SMiklos Szeredi 3122d58ffd35SMiklos Szeredi /* 3123fe2d35ffSAl Viro * Handle the last step of open() 312431e6b01fSNick Piggin */ 31251ccac622SAl Viro static const char *do_last(struct nameidata *nd, 31263ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3127fb1cc555SAl Viro { 3128a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 31296404674aSAl Viro kuid_t dir_uid = nd->inode->i_uid; 31306404674aSAl Viro umode_t dir_mode = nd->inode->i_mode; 3131ca344a89SAl Viro int open_flag = op->open_flag; 313277d660a8SMiklos Szeredi bool will_truncate = (open_flag & O_TRUNC) != 0; 313364894cf8SAl Viro bool got_write = false; 3134bcda7652SAl Viro int acc_mode = op->acc_mode; 3135254cf582SAl Viro unsigned seq; 3136a1eb3315SMiklos Szeredi struct inode *inode; 3137da5ebf5aSAl Viro struct dentry *dentry; 3138b0417d2cSAl Viro const char *res; 313916c2cd71SAl Viro int error; 3140fb1cc555SAl Viro 3141c3e380b0SAl Viro nd->flags |= op->intent; 3142c3e380b0SAl Viro 3143bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 314456676ec3SAl Viro if (nd->depth) 314556676ec3SAl Viro put_link(nd); 3146fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3147deb106c6SAl Viro if (unlikely(error)) 31481ccac622SAl Viro return ERR_PTR(error); 3149e83db167SMiklos Szeredi goto finish_open; 31501f36f774SAl Viro } 3151a2c36b45SAl Viro 3152ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3153fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3154fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3155fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 315620e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 315720e34357SAl Viro if (IS_ERR(dentry)) 31581ccac622SAl Viro return ERR_CAST(dentry); 315920e34357SAl Viro if (likely(dentry)) 316071574865SMiklos Szeredi goto finish_lookup; 316171574865SMiklos Szeredi 316237d7fffcSMiklos Szeredi BUG_ON(nd->inode != dir->d_inode); 31636583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3164b6183df7SMiklos Szeredi } else { 3165fe2d35ffSAl Viro /* create side of things */ 3166a3fbbde7SAl Viro /* 3167b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3168b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3169b6183df7SMiklos Szeredi * about to look up 3170a3fbbde7SAl Viro */ 31719f1fafeeSAl Viro error = complete_walk(nd); 3172e8bb73dfSAl Viro if (error) 31731ccac622SAl Viro return ERR_PTR(error); 3174fe2d35ffSAl Viro 3175c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 31761f36f774SAl Viro /* trailing slashes? */ 3177deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 31781ccac622SAl Viro return ERR_PTR(-EISDIR); 3179b6183df7SMiklos Szeredi } 31801f36f774SAl Viro 31819cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 318264894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 318364894cf8SAl Viro if (!error) 318464894cf8SAl Viro got_write = true; 318564894cf8SAl Viro /* 318664894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 318764894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 318864894cf8SAl Viro * dropping this one anyway. 318964894cf8SAl Viro */ 319064894cf8SAl Viro } 31919cf843e3SAl Viro if (open_flag & O_CREAT) 31925955102cSAl Viro inode_lock(dir->d_inode); 31939cf843e3SAl Viro else 31949cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3195da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 31969cf843e3SAl Viro if (open_flag & O_CREAT) 31975955102cSAl Viro inode_unlock(dir->d_inode); 31989cf843e3SAl Viro else 31999cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3200fb1cc555SAl Viro 3201da5ebf5aSAl Viro if (IS_ERR(dentry)) { 3202da5ebf5aSAl Viro error = PTR_ERR(dentry); 3203d58ffd35SMiklos Szeredi goto out; 3204da5ebf5aSAl Viro } 32056c0d46c4SAl Viro 320600a07c15SAl Viro if (file->f_mode & FMODE_OPENED) { 320773a09dd9SAl Viro if ((file->f_mode & FMODE_CREATED) || 3208496ad9aaSAl Viro !S_ISREG(file_inode(file)->i_mode)) 320977d660a8SMiklos Szeredi will_truncate = false; 3210d18e9008SMiklos Szeredi 321176ae2a5aSAl Viro audit_inode(nd->name, file->f_path.dentry, 0); 3212da5ebf5aSAl Viro dput(dentry); 3213d18e9008SMiklos Szeredi goto opened; 3214d18e9008SMiklos Szeredi } 3215d18e9008SMiklos Szeredi 321673a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) { 32179b44f1b3SAl Viro /* Don't check for write permission, don't truncate */ 3218ca344a89SAl Viro open_flag &= ~O_TRUNC; 321977d660a8SMiklos Szeredi will_truncate = false; 322062fb4a15SAl Viro acc_mode = 0; 3221e73cabffSAl Viro dput(nd->path.dentry); 3222e73cabffSAl Viro nd->path.dentry = dentry; 3223e83db167SMiklos Szeredi goto finish_open_created; 3224fb1cc555SAl Viro } 3225fb1cc555SAl Viro 3226fb1cc555SAl Viro /* 3227d18e9008SMiklos Szeredi * If atomic_open() acquired write access it is dropped now due to 3228d18e9008SMiklos Szeredi * possible mount and symlink following (this might be optimized away if 3229d18e9008SMiklos Szeredi * necessary...) 3230d18e9008SMiklos Szeredi */ 323164894cf8SAl Viro if (got_write) { 3232d18e9008SMiklos Szeredi mnt_drop_write(nd->path.mnt); 323364894cf8SAl Viro got_write = false; 3234d18e9008SMiklos Szeredi } 3235d18e9008SMiklos Szeredi 323620e34357SAl Viro finish_lookup: 323756676ec3SAl Viro if (nd->depth) 323856676ec3SAl Viro put_link(nd); 32398c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3240b0417d2cSAl Viro if (unlikely(res)) { 32411ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3242b0417d2cSAl Viro return res; 32431ccac622SAl Viro } 324431d1726dSAl Viro 324531d1726dSAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 324631d1726dSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 32471ccac622SAl Viro return ERR_PTR(-EEXIST); 324831d1726dSAl Viro } 3249bc77daa7SAl Viro finish_open: 32508f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3251a3fbbde7SAl Viro error = complete_walk(nd); 3252fac7d191SAl Viro if (error) 32531ccac622SAl Viro return ERR_PTR(error); 325476ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 325530aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3256fb1cc555SAl Viro error = -EISDIR; 325730aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 32582675a4ebSAl Viro goto out; 3259d0cb5018SAl Viro error = may_create_in_sticky(dir_mode, dir_uid, 326030aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 326130aba665SSalvatore Mesoraca if (unlikely(error)) 326230aba665SSalvatore Mesoraca goto out; 326330aba665SSalvatore Mesoraca } 3264af2f5542SMiklos Szeredi error = -ENOTDIR; 326544b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 32662675a4ebSAl Viro goto out; 32674bbcbd3bSDavid Howells if (!d_is_reg(nd->path.dentry)) 326877d660a8SMiklos Szeredi will_truncate = false; 32696c0d46c4SAl Viro 32700f9d1a10SAl Viro if (will_truncate) { 32710f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 32720f9d1a10SAl Viro if (error) 32732675a4ebSAl Viro goto out; 327464894cf8SAl Viro got_write = true; 32750f9d1a10SAl Viro } 3276e83db167SMiklos Szeredi finish_open_created: 3277bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 3278ca344a89SAl Viro if (error) 32792675a4ebSAl Viro goto out; 3280aad888f8SAl Viro BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */ 3281ae2bb293SAl Viro error = vfs_open(&nd->path, file); 3282fac7d191SAl Viro if (error) 3283015c3bbcSMiklos Szeredi goto out; 3284a8277b9bSMiklos Szeredi opened: 32856035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 3286fe9ec829SAl Viro if (!error && will_truncate) 32872675a4ebSAl Viro error = handle_truncate(file); 3288ca344a89SAl Viro out: 3289c80567c8SAl Viro if (unlikely(error > 0)) { 3290c80567c8SAl Viro WARN_ON(1); 3291c80567c8SAl Viro error = -EINVAL; 3292c80567c8SAl Viro } 329364894cf8SAl Viro if (got_write) 32940f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 32951ccac622SAl Viro return ERR_PTR(error); 3296fb1cc555SAl Viro } 3297fb1cc555SAl Viro 3298af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3299af7bd4dcSAmir Goldstein { 3300af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3301af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3302af7bd4dcSAmir Goldstein struct inode *inode; 3303af7bd4dcSAmir Goldstein int error; 3304af7bd4dcSAmir Goldstein 3305af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3306af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3307af7bd4dcSAmir Goldstein if (error) 3308af7bd4dcSAmir Goldstein goto out_err; 3309af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3310af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3311af7bd4dcSAmir Goldstein goto out_err; 3312af7bd4dcSAmir Goldstein error = -ENOMEM; 3313cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3314af7bd4dcSAmir Goldstein if (unlikely(!child)) 3315af7bd4dcSAmir Goldstein goto out_err; 3316af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3317af7bd4dcSAmir Goldstein if (error) 3318af7bd4dcSAmir Goldstein goto out_err; 3319af7bd4dcSAmir Goldstein error = -ENOENT; 3320af7bd4dcSAmir Goldstein inode = child->d_inode; 3321af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3322af7bd4dcSAmir Goldstein goto out_err; 3323af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3324af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3325af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3326af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3327af7bd4dcSAmir Goldstein } 3328fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3329af7bd4dcSAmir Goldstein return child; 3330af7bd4dcSAmir Goldstein 3331af7bd4dcSAmir Goldstein out_err: 3332af7bd4dcSAmir Goldstein dput(child); 3333af7bd4dcSAmir Goldstein return ERR_PTR(error); 3334af7bd4dcSAmir Goldstein } 3335af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3336af7bd4dcSAmir Goldstein 3337c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 333860545d0dSAl Viro const struct open_flags *op, 33393ec2eef1SAl Viro struct file *file) 334060545d0dSAl Viro { 3341625b6d10SAl Viro struct dentry *child; 3342625b6d10SAl Viro struct path path; 3343c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 334460545d0dSAl Viro if (unlikely(error)) 334560545d0dSAl Viro return error; 3346625b6d10SAl Viro error = mnt_want_write(path.mnt); 334760545d0dSAl Viro if (unlikely(error)) 334860545d0dSAl Viro goto out; 3349af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3350af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3351684e73beSHirofumi Nakagawa if (IS_ERR(child)) 335260545d0dSAl Viro goto out2; 3353625b6d10SAl Viro dput(path.dentry); 3354625b6d10SAl Viro path.dentry = child; 3355c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 335669a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 335762fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 335860545d0dSAl Viro if (error) 335960545d0dSAl Viro goto out2; 3360625b6d10SAl Viro file->f_path.mnt = path.mnt; 3361be12af3eSAl Viro error = finish_open(file, child, NULL); 336260545d0dSAl Viro out2: 3363625b6d10SAl Viro mnt_drop_write(path.mnt); 336460545d0dSAl Viro out: 3365625b6d10SAl Viro path_put(&path); 336660545d0dSAl Viro return error; 336760545d0dSAl Viro } 336860545d0dSAl Viro 33696ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 33706ac08709SAl Viro { 33716ac08709SAl Viro struct path path; 33726ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 33736ac08709SAl Viro if (!error) { 33746ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3375ae2bb293SAl Viro error = vfs_open(&path, file); 33766ac08709SAl Viro path_put(&path); 33776ac08709SAl Viro } 33786ac08709SAl Viro return error; 33796ac08709SAl Viro } 33806ac08709SAl Viro 3381c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3382c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 33831da177e4SLinus Torvalds { 338430d90494SAl Viro struct file *file; 338513aab428SAl Viro int error; 338631e6b01fSNick Piggin 3387ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 33881afc99beSAl Viro if (IS_ERR(file)) 33891afc99beSAl Viro return file; 339031e6b01fSNick Piggin 3391bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 33923ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 33935f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 33946ac08709SAl Viro error = do_o_path(nd, flags, file); 33955f336e72SAl Viro } else { 33965f336e72SAl Viro const char *s = path_init(nd, flags); 33973bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 33981ccac622SAl Viro (s = do_last(nd, file, op)) != NULL) 33991ccac622SAl Viro ; 3400deb106c6SAl Viro terminate_walk(nd); 34015f336e72SAl Viro } 34027c1c01ecSAl Viro if (likely(!error)) { 3403aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 34047c1c01ecSAl Viro return file; 34057c1c01ecSAl Viro WARN_ON(1); 34067c1c01ecSAl Viro error = -EINVAL; 3407015c3bbcSMiklos Szeredi } 34087c1c01ecSAl Viro fput(file); 34092675a4ebSAl Viro if (error == -EOPENSTALE) { 34102675a4ebSAl Viro if (flags & LOOKUP_RCU) 34112675a4ebSAl Viro error = -ECHILD; 34122675a4ebSAl Viro else 34132675a4ebSAl Viro error = -ESTALE; 34142675a4ebSAl Viro } 34157c1c01ecSAl Viro return ERR_PTR(error); 3416de459215SKirill Korotaev } 34171da177e4SLinus Torvalds 3418669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3419f9652e10SAl Viro const struct open_flags *op) 342013aab428SAl Viro { 34219883d185SAl Viro struct nameidata nd; 3422f9652e10SAl Viro int flags = op->lookup_flags; 342313aab428SAl Viro struct file *filp; 342413aab428SAl Viro 34259883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3426c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 342713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3428c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 342913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3430c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34319883d185SAl Viro restore_nameidata(); 343213aab428SAl Viro return filp; 343313aab428SAl Viro } 343413aab428SAl Viro 343573d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3436f9652e10SAl Viro const char *name, const struct open_flags *op) 343773d049a4SAl Viro { 34389883d185SAl Viro struct nameidata nd; 343973d049a4SAl Viro struct file *file; 344051689104SPaul Moore struct filename *filename; 3441f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 344273d049a4SAl Viro 344373d049a4SAl Viro nd.root.mnt = mnt; 344473d049a4SAl Viro nd.root.dentry = dentry; 344573d049a4SAl Viro 3446b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 344773d049a4SAl Viro return ERR_PTR(-ELOOP); 344873d049a4SAl Viro 344951689104SPaul Moore filename = getname_kernel(name); 3450a1c83681SViresh Kumar if (IS_ERR(filename)) 345151689104SPaul Moore return ERR_CAST(filename); 345251689104SPaul Moore 34539883d185SAl Viro set_nameidata(&nd, -1, filename); 3454c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 345573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3456c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 345773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3458c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 34599883d185SAl Viro restore_nameidata(); 346051689104SPaul Moore putname(filename); 346173d049a4SAl Viro return file; 346273d049a4SAl Viro } 346373d049a4SAl Viro 3464fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34651ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34661da177e4SLinus Torvalds { 3467c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3468391172c4SAl Viro struct qstr last; 3469391172c4SAl Viro int type; 3470c30dabfeSJan Kara int err2; 34711ac12b4bSJeff Layton int error; 34721ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34731ac12b4bSJeff Layton 34741ac12b4bSJeff Layton /* 34751ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34761ac12b4bSJeff Layton * other flags passed in are ignored! 34771ac12b4bSJeff Layton */ 34781ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34791ac12b4bSJeff Layton 34805c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34815c31b6ceSAl Viro if (IS_ERR(name)) 34825c31b6ceSAl Viro return ERR_CAST(name); 34831da177e4SLinus Torvalds 3484c663e5d8SChristoph Hellwig /* 3485c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3486c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3487c663e5d8SChristoph Hellwig */ 34885c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3489ed75e95dSAl Viro goto out; 3490c663e5d8SChristoph Hellwig 3491c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3492391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3493c663e5d8SChristoph Hellwig /* 3494c663e5d8SChristoph Hellwig * Do the final lookup. 3495c663e5d8SChristoph Hellwig */ 3496391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 34975955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3498391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34991da177e4SLinus Torvalds if (IS_ERR(dentry)) 3500a8104a9fSAl Viro goto unlock; 3501c663e5d8SChristoph Hellwig 3502a8104a9fSAl Viro error = -EEXIST; 3503b18825a7SDavid Howells if (d_is_positive(dentry)) 3504a8104a9fSAl Viro goto fail; 3505b18825a7SDavid Howells 3506c663e5d8SChristoph Hellwig /* 3507c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3508c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3509c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3510c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3511c663e5d8SChristoph Hellwig */ 3512391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3513a8104a9fSAl Viro error = -ENOENT; 3514ed75e95dSAl Viro goto fail; 3515e9baf6e5SAl Viro } 3516c30dabfeSJan Kara if (unlikely(err2)) { 3517c30dabfeSJan Kara error = err2; 3518a8104a9fSAl Viro goto fail; 3519c30dabfeSJan Kara } 3520181c37b6SAl Viro putname(name); 3521e9baf6e5SAl Viro return dentry; 35221da177e4SLinus Torvalds fail: 3523a8104a9fSAl Viro dput(dentry); 3524a8104a9fSAl Viro dentry = ERR_PTR(error); 3525a8104a9fSAl Viro unlock: 35265955102cSAl Viro inode_unlock(path->dentry->d_inode); 3527c30dabfeSJan Kara if (!err2) 3528391172c4SAl Viro mnt_drop_write(path->mnt); 3529ed75e95dSAl Viro out: 3530391172c4SAl Viro path_put(path); 3531181c37b6SAl Viro putname(name); 3532ed75e95dSAl Viro return dentry; 3533dae6ad8fSAl Viro } 3534fa14a0b8SAl Viro 3535fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3536fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3537fa14a0b8SAl Viro { 3538181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3539181c37b6SAl Viro path, lookup_flags); 3540fa14a0b8SAl Viro } 3541dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3542dae6ad8fSAl Viro 3543921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3544921a1650SAl Viro { 3545921a1650SAl Viro dput(dentry); 35465955102cSAl Viro inode_unlock(path->dentry->d_inode); 3547a8104a9fSAl Viro mnt_drop_write(path->mnt); 3548921a1650SAl Viro path_put(path); 3549921a1650SAl Viro } 3550921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3551921a1650SAl Viro 3552520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 35531ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3554dae6ad8fSAl Viro { 3555181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3556dae6ad8fSAl Viro } 3557dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3558dae6ad8fSAl Viro 35591a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35601da177e4SLinus Torvalds { 3561a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35621da177e4SLinus Torvalds 35631da177e4SLinus Torvalds if (error) 35641da177e4SLinus Torvalds return error; 35651da177e4SLinus Torvalds 356694f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 35671da177e4SLinus Torvalds return -EPERM; 35681da177e4SLinus Torvalds 3569acfa4380SAl Viro if (!dir->i_op->mknod) 35701da177e4SLinus Torvalds return -EPERM; 35711da177e4SLinus Torvalds 357208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 357308ce5f16SSerge E. Hallyn if (error) 357408ce5f16SSerge E. Hallyn return error; 357508ce5f16SSerge E. Hallyn 35761da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35771da177e4SLinus Torvalds if (error) 35781da177e4SLinus Torvalds return error; 35791da177e4SLinus Torvalds 35801da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3581a74574aaSStephen Smalley if (!error) 3582f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35831da177e4SLinus Torvalds return error; 35841da177e4SLinus Torvalds } 35854d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35861da177e4SLinus Torvalds 3587f69aac00SAl Viro static int may_mknod(umode_t mode) 3588463c3197SDave Hansen { 3589463c3197SDave Hansen switch (mode & S_IFMT) { 3590463c3197SDave Hansen case S_IFREG: 3591463c3197SDave Hansen case S_IFCHR: 3592463c3197SDave Hansen case S_IFBLK: 3593463c3197SDave Hansen case S_IFIFO: 3594463c3197SDave Hansen case S_IFSOCK: 3595463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3596463c3197SDave Hansen return 0; 3597463c3197SDave Hansen case S_IFDIR: 3598463c3197SDave Hansen return -EPERM; 3599463c3197SDave Hansen default: 3600463c3197SDave Hansen return -EINVAL; 3601463c3197SDave Hansen } 3602463c3197SDave Hansen } 3603463c3197SDave Hansen 360487c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 360587c4e192SDominik Brodowski unsigned int dev) 36061da177e4SLinus Torvalds { 36071da177e4SLinus Torvalds struct dentry *dentry; 3608dae6ad8fSAl Viro struct path path; 3609dae6ad8fSAl Viro int error; 3610972567f1SJeff Layton unsigned int lookup_flags = 0; 36111da177e4SLinus Torvalds 36128e4bfca1SAl Viro error = may_mknod(mode); 36138e4bfca1SAl Viro if (error) 36148e4bfca1SAl Viro return error; 3615972567f1SJeff Layton retry: 3616972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3617dae6ad8fSAl Viro if (IS_ERR(dentry)) 3618dae6ad8fSAl Viro return PTR_ERR(dentry); 36192ad94ae6SAl Viro 3620dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3621ce3b0f8dSAl Viro mode &= ~current_umask(); 3622dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3623be6d3e56SKentaro Takeda if (error) 3624a8104a9fSAl Viro goto out; 36251da177e4SLinus Torvalds switch (mode & S_IFMT) { 36261da177e4SLinus Torvalds case 0: case S_IFREG: 3627312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 362805d1a717SMimi Zohar if (!error) 362905d1a717SMimi Zohar ima_post_path_mknod(dentry); 36301da177e4SLinus Torvalds break; 36311da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3632dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36331da177e4SLinus Torvalds new_decode_dev(dev)); 36341da177e4SLinus Torvalds break; 36351da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3636dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 36371da177e4SLinus Torvalds break; 36381da177e4SLinus Torvalds } 3639a8104a9fSAl Viro out: 3640921a1650SAl Viro done_path_create(&path, dentry); 3641972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3642972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3643972567f1SJeff Layton goto retry; 3644972567f1SJeff Layton } 36451da177e4SLinus Torvalds return error; 36461da177e4SLinus Torvalds } 36471da177e4SLinus Torvalds 364887c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 364987c4e192SDominik Brodowski unsigned int, dev) 365087c4e192SDominik Brodowski { 365187c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 365287c4e192SDominik Brodowski } 365387c4e192SDominik Brodowski 36548208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 36555590ff0dSUlrich Drepper { 365687c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 36575590ff0dSUlrich Drepper } 36585590ff0dSUlrich Drepper 365918bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 36601da177e4SLinus Torvalds { 3661a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36628de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 36631da177e4SLinus Torvalds 36641da177e4SLinus Torvalds if (error) 36651da177e4SLinus Torvalds return error; 36661da177e4SLinus Torvalds 3667acfa4380SAl Viro if (!dir->i_op->mkdir) 36681da177e4SLinus Torvalds return -EPERM; 36691da177e4SLinus Torvalds 36701da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 36711da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 36721da177e4SLinus Torvalds if (error) 36731da177e4SLinus Torvalds return error; 36741da177e4SLinus Torvalds 36758de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36768de52778SAl Viro return -EMLINK; 36778de52778SAl Viro 36781da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3679a74574aaSStephen Smalley if (!error) 3680f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36811da177e4SLinus Torvalds return error; 36821da177e4SLinus Torvalds } 36834d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36841da177e4SLinus Torvalds 36850101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 36861da177e4SLinus Torvalds { 36876902d925SDave Hansen struct dentry *dentry; 3688dae6ad8fSAl Viro struct path path; 3689dae6ad8fSAl Viro int error; 3690b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36911da177e4SLinus Torvalds 3692b76d8b82SJeff Layton retry: 3693b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36946902d925SDave Hansen if (IS_ERR(dentry)) 3695dae6ad8fSAl Viro return PTR_ERR(dentry); 36966902d925SDave Hansen 3697dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3698ce3b0f8dSAl Viro mode &= ~current_umask(); 3699dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3700a8104a9fSAl Viro if (!error) 3701dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3702921a1650SAl Viro done_path_create(&path, dentry); 3703b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3704b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3705b76d8b82SJeff Layton goto retry; 3706b76d8b82SJeff Layton } 37071da177e4SLinus Torvalds return error; 37081da177e4SLinus Torvalds } 37091da177e4SLinus Torvalds 37100101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 37110101db7aSDominik Brodowski { 37120101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 37130101db7aSDominik Brodowski } 37140101db7aSDominik Brodowski 3715a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 37165590ff0dSUlrich Drepper { 37170101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 37185590ff0dSUlrich Drepper } 37195590ff0dSUlrich Drepper 37201da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 37211da177e4SLinus Torvalds { 37221da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 37231da177e4SLinus Torvalds 37241da177e4SLinus Torvalds if (error) 37251da177e4SLinus Torvalds return error; 37261da177e4SLinus Torvalds 3727acfa4380SAl Viro if (!dir->i_op->rmdir) 37281da177e4SLinus Torvalds return -EPERM; 37291da177e4SLinus Torvalds 37301d2ef590SAl Viro dget(dentry); 37315955102cSAl Viro inode_lock(dentry->d_inode); 3732912dbc15SSage Weil 37331da177e4SLinus Torvalds error = -EBUSY; 37347af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3735912dbc15SSage Weil goto out; 3736912dbc15SSage Weil 37371da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3738912dbc15SSage Weil if (error) 3739912dbc15SSage Weil goto out; 3740912dbc15SSage Weil 37411da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3742912dbc15SSage Weil if (error) 3743912dbc15SSage Weil goto out; 3744912dbc15SSage Weil 37458767712fSAl Viro shrink_dcache_parent(dentry); 37461da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3747d83c49f3SAl Viro dont_mount(dentry); 37488ed936b5SEric W. Biederman detach_mounts(dentry); 3749116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 37501da177e4SLinus Torvalds 3751912dbc15SSage Weil out: 37525955102cSAl Viro inode_unlock(dentry->d_inode); 37531d2ef590SAl Viro dput(dentry); 3754912dbc15SSage Weil if (!error) 3755912dbc15SSage Weil d_delete(dentry); 37561da177e4SLinus Torvalds return error; 37571da177e4SLinus Torvalds } 37584d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37591da177e4SLinus Torvalds 3760f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 37611da177e4SLinus Torvalds { 37621da177e4SLinus Torvalds int error = 0; 376391a27b2aSJeff Layton struct filename *name; 37641da177e4SLinus Torvalds struct dentry *dentry; 3765f5beed75SAl Viro struct path path; 3766f5beed75SAl Viro struct qstr last; 3767f5beed75SAl Viro int type; 3768c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3769c6ee9206SJeff Layton retry: 3770c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3771c1d4dd27SAl Viro &path, &last, &type); 377291a27b2aSJeff Layton if (IS_ERR(name)) 377391a27b2aSJeff Layton return PTR_ERR(name); 37741da177e4SLinus Torvalds 3775f5beed75SAl Viro switch (type) { 37761da177e4SLinus Torvalds case LAST_DOTDOT: 37771da177e4SLinus Torvalds error = -ENOTEMPTY; 37781da177e4SLinus Torvalds goto exit1; 37791da177e4SLinus Torvalds case LAST_DOT: 37801da177e4SLinus Torvalds error = -EINVAL; 37811da177e4SLinus Torvalds goto exit1; 37821da177e4SLinus Torvalds case LAST_ROOT: 37831da177e4SLinus Torvalds error = -EBUSY; 37841da177e4SLinus Torvalds goto exit1; 37851da177e4SLinus Torvalds } 37860612d9fbSOGAWA Hirofumi 3787f5beed75SAl Viro error = mnt_want_write(path.mnt); 3788c30dabfeSJan Kara if (error) 3789c30dabfeSJan Kara goto exit1; 37900612d9fbSOGAWA Hirofumi 37915955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3792f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37931da177e4SLinus Torvalds error = PTR_ERR(dentry); 37946902d925SDave Hansen if (IS_ERR(dentry)) 37956902d925SDave Hansen goto exit2; 3796e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3797e6bc45d6STheodore Ts'o error = -ENOENT; 3798e6bc45d6STheodore Ts'o goto exit3; 3799e6bc45d6STheodore Ts'o } 3800f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3801be6d3e56SKentaro Takeda if (error) 3802c30dabfeSJan Kara goto exit3; 3803f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 38040622753bSDave Hansen exit3: 38051da177e4SLinus Torvalds dput(dentry); 38066902d925SDave Hansen exit2: 38075955102cSAl Viro inode_unlock(path.dentry->d_inode); 3808f5beed75SAl Viro mnt_drop_write(path.mnt); 38091da177e4SLinus Torvalds exit1: 3810f5beed75SAl Viro path_put(&path); 38111da177e4SLinus Torvalds putname(name); 3812c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3813c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3814c6ee9206SJeff Layton goto retry; 3815c6ee9206SJeff Layton } 38161da177e4SLinus Torvalds return error; 38171da177e4SLinus Torvalds } 38181da177e4SLinus Torvalds 38193cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 38205590ff0dSUlrich Drepper { 38215590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 38225590ff0dSUlrich Drepper } 38235590ff0dSUlrich Drepper 3824b21996e3SJ. Bruce Fields /** 3825b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3826b21996e3SJ. Bruce Fields * @dir: parent directory 3827b21996e3SJ. Bruce Fields * @dentry: victim 3828b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3829b21996e3SJ. Bruce Fields * 3830b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3831b21996e3SJ. Bruce Fields * 3832b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3833b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3834b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3835b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3836b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3837b21996e3SJ. Bruce Fields * 3838b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3839b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3840b21996e3SJ. Bruce Fields * to be NFS exported. 3841b21996e3SJ. Bruce Fields */ 3842b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38431da177e4SLinus Torvalds { 38449accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38451da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 38461da177e4SLinus Torvalds 38471da177e4SLinus Torvalds if (error) 38481da177e4SLinus Torvalds return error; 38491da177e4SLinus Torvalds 3850acfa4380SAl Viro if (!dir->i_op->unlink) 38511da177e4SLinus Torvalds return -EPERM; 38521da177e4SLinus Torvalds 38535955102cSAl Viro inode_lock(target); 38548ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38551da177e4SLinus Torvalds error = -EBUSY; 38561da177e4SLinus Torvalds else { 38571da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3858bec1052eSAl Viro if (!error) { 38595a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38605a14696cSJ. Bruce Fields if (error) 3861b21996e3SJ. Bruce Fields goto out; 38621da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38638ed936b5SEric W. Biederman if (!error) { 3864d83c49f3SAl Viro dont_mount(dentry); 38658ed936b5SEric W. Biederman detach_mounts(dentry); 3866116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 38678ed936b5SEric W. Biederman } 3868bec1052eSAl Viro } 38691da177e4SLinus Torvalds } 3870b21996e3SJ. Bruce Fields out: 38715955102cSAl Viro inode_unlock(target); 38721da177e4SLinus Torvalds 38731da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38741da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38759accbb97SJ. Bruce Fields fsnotify_link_count(target); 38761da177e4SLinus Torvalds d_delete(dentry); 38771da177e4SLinus Torvalds } 38780eeca283SRobert Love 38791da177e4SLinus Torvalds return error; 38801da177e4SLinus Torvalds } 38814d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38821da177e4SLinus Torvalds 38831da177e4SLinus Torvalds /* 38841da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38851b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38861da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38871da177e4SLinus Torvalds * while waiting on the I/O. 38881da177e4SLinus Torvalds */ 3889da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 38901da177e4SLinus Torvalds { 38912ad94ae6SAl Viro int error; 38921da177e4SLinus Torvalds struct dentry *dentry; 3893f5beed75SAl Viro struct path path; 3894f5beed75SAl Viro struct qstr last; 3895f5beed75SAl Viro int type; 38961da177e4SLinus Torvalds struct inode *inode = NULL; 3897b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38985d18f813SJeff Layton unsigned int lookup_flags = 0; 38995d18f813SJeff Layton retry: 3900da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 390191a27b2aSJeff Layton if (IS_ERR(name)) 390291a27b2aSJeff Layton return PTR_ERR(name); 39032ad94ae6SAl Viro 39041da177e4SLinus Torvalds error = -EISDIR; 3905f5beed75SAl Viro if (type != LAST_NORM) 39061da177e4SLinus Torvalds goto exit1; 39070612d9fbSOGAWA Hirofumi 3908f5beed75SAl Viro error = mnt_want_write(path.mnt); 3909c30dabfeSJan Kara if (error) 3910c30dabfeSJan Kara goto exit1; 3911b21996e3SJ. Bruce Fields retry_deleg: 39125955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3913f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39141da177e4SLinus Torvalds error = PTR_ERR(dentry); 39151da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 39161da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3917f5beed75SAl Viro if (last.name[last.len]) 391850338b88STörök Edwin goto slashes; 39191da177e4SLinus Torvalds inode = dentry->d_inode; 3920b18825a7SDavid Howells if (d_is_negative(dentry)) 3921e6bc45d6STheodore Ts'o goto slashes; 39227de9c6eeSAl Viro ihold(inode); 3923f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3924be6d3e56SKentaro Takeda if (error) 3925c30dabfeSJan Kara goto exit2; 3926f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 39271da177e4SLinus Torvalds exit2: 39281da177e4SLinus Torvalds dput(dentry); 39291da177e4SLinus Torvalds } 39305955102cSAl Viro inode_unlock(path.dentry->d_inode); 39311da177e4SLinus Torvalds if (inode) 39321da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3933b21996e3SJ. Bruce Fields inode = NULL; 3934b21996e3SJ. Bruce Fields if (delegated_inode) { 39355a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3936b21996e3SJ. Bruce Fields if (!error) 3937b21996e3SJ. Bruce Fields goto retry_deleg; 3938b21996e3SJ. Bruce Fields } 3939f5beed75SAl Viro mnt_drop_write(path.mnt); 39401da177e4SLinus Torvalds exit1: 3941f5beed75SAl Viro path_put(&path); 39425d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39435d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 39445d18f813SJeff Layton inode = NULL; 39455d18f813SJeff Layton goto retry; 39465d18f813SJeff Layton } 3947da2f1362SChristoph Hellwig putname(name); 39481da177e4SLinus Torvalds return error; 39491da177e4SLinus Torvalds 39501da177e4SLinus Torvalds slashes: 3951b18825a7SDavid Howells if (d_is_negative(dentry)) 3952b18825a7SDavid Howells error = -ENOENT; 395344b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3954b18825a7SDavid Howells error = -EISDIR; 3955b18825a7SDavid Howells else 3956b18825a7SDavid Howells error = -ENOTDIR; 39571da177e4SLinus Torvalds goto exit2; 39581da177e4SLinus Torvalds } 39591da177e4SLinus Torvalds 39602e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39615590ff0dSUlrich Drepper { 39625590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39635590ff0dSUlrich Drepper return -EINVAL; 39645590ff0dSUlrich Drepper 39655590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39665590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39675590ff0dSUlrich Drepper 3968da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 39695590ff0dSUlrich Drepper } 39705590ff0dSUlrich Drepper 39713480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39725590ff0dSUlrich Drepper { 3973da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 39745590ff0dSUlrich Drepper } 39755590ff0dSUlrich Drepper 3976db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39771da177e4SLinus Torvalds { 3978a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39791da177e4SLinus Torvalds 39801da177e4SLinus Torvalds if (error) 39811da177e4SLinus Torvalds return error; 39821da177e4SLinus Torvalds 3983acfa4380SAl Viro if (!dir->i_op->symlink) 39841da177e4SLinus Torvalds return -EPERM; 39851da177e4SLinus Torvalds 39861da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39871da177e4SLinus Torvalds if (error) 39881da177e4SLinus Torvalds return error; 39891da177e4SLinus Torvalds 39901da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3991a74574aaSStephen Smalley if (!error) 3992f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39931da177e4SLinus Torvalds return error; 39941da177e4SLinus Torvalds } 39954d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39961da177e4SLinus Torvalds 3997b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 3998b724e846SDominik Brodowski const char __user *newname) 39991da177e4SLinus Torvalds { 40002ad94ae6SAl Viro int error; 400191a27b2aSJeff Layton struct filename *from; 40026902d925SDave Hansen struct dentry *dentry; 4003dae6ad8fSAl Viro struct path path; 4004f46d3567SJeff Layton unsigned int lookup_flags = 0; 40051da177e4SLinus Torvalds 40061da177e4SLinus Torvalds from = getname(oldname); 40071da177e4SLinus Torvalds if (IS_ERR(from)) 40081da177e4SLinus Torvalds return PTR_ERR(from); 4009f46d3567SJeff Layton retry: 4010f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 40111da177e4SLinus Torvalds error = PTR_ERR(dentry); 40126902d925SDave Hansen if (IS_ERR(dentry)) 4013dae6ad8fSAl Viro goto out_putname; 40146902d925SDave Hansen 401591a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4016a8104a9fSAl Viro if (!error) 401791a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4018921a1650SAl Viro done_path_create(&path, dentry); 4019f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4020f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4021f46d3567SJeff Layton goto retry; 4022f46d3567SJeff Layton } 40236902d925SDave Hansen out_putname: 40241da177e4SLinus Torvalds putname(from); 40251da177e4SLinus Torvalds return error; 40261da177e4SLinus Torvalds } 40271da177e4SLinus Torvalds 4028b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4029b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4030b724e846SDominik Brodowski { 4031b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4032b724e846SDominik Brodowski } 4033b724e846SDominik Brodowski 40343480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40355590ff0dSUlrich Drepper { 4036b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 40375590ff0dSUlrich Drepper } 40385590ff0dSUlrich Drepper 4039146a8595SJ. Bruce Fields /** 4040146a8595SJ. Bruce Fields * vfs_link - create a new link 4041146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4042146a8595SJ. Bruce Fields * @dir: new parent 4043146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4044146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4045146a8595SJ. Bruce Fields * 4046146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4047146a8595SJ. Bruce Fields * 4048146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4049146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4050146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4051146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4052146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4053146a8595SJ. Bruce Fields * 4054146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4055146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4056146a8595SJ. Bruce Fields * to be NFS exported. 4057146a8595SJ. Bruce Fields */ 4058146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40591da177e4SLinus Torvalds { 40601da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40618de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40621da177e4SLinus Torvalds int error; 40631da177e4SLinus Torvalds 40641da177e4SLinus Torvalds if (!inode) 40651da177e4SLinus Torvalds return -ENOENT; 40661da177e4SLinus Torvalds 4067a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40681da177e4SLinus Torvalds if (error) 40691da177e4SLinus Torvalds return error; 40701da177e4SLinus Torvalds 40711da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40721da177e4SLinus Torvalds return -EXDEV; 40731da177e4SLinus Torvalds 40741da177e4SLinus Torvalds /* 40751da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40761da177e4SLinus Torvalds */ 40771da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40781da177e4SLinus Torvalds return -EPERM; 40790bd23d09SEric W. Biederman /* 40800bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 40810bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 40820bd23d09SEric W. Biederman * the vfs. 40830bd23d09SEric W. Biederman */ 40840bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 40850bd23d09SEric W. Biederman return -EPERM; 4086acfa4380SAl Viro if (!dir->i_op->link) 40871da177e4SLinus Torvalds return -EPERM; 40887e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40891da177e4SLinus Torvalds return -EPERM; 40901da177e4SLinus Torvalds 40911da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40921da177e4SLinus Torvalds if (error) 40931da177e4SLinus Torvalds return error; 40941da177e4SLinus Torvalds 40955955102cSAl Viro inode_lock(inode); 4096aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4097f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4098aae8a97dSAneesh Kumar K.V error = -ENOENT; 40998de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 41008de52778SAl Viro error = -EMLINK; 4101146a8595SJ. Bruce Fields else { 4102146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4103146a8595SJ. Bruce Fields if (!error) 41041da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4105146a8595SJ. Bruce Fields } 4106f4e0c30cSAl Viro 4107f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4108f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4109f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4110f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4111f4e0c30cSAl Viro } 41125955102cSAl Viro inode_unlock(inode); 4113e31e14ecSStephen Smalley if (!error) 41147e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 41151da177e4SLinus Torvalds return error; 41161da177e4SLinus Torvalds } 41174d359507SAl Viro EXPORT_SYMBOL(vfs_link); 41181da177e4SLinus Torvalds 41191da177e4SLinus Torvalds /* 41201da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41211da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41221da177e4SLinus Torvalds * newname. --KAB 41231da177e4SLinus Torvalds * 41241da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41251da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 41261da177e4SLinus Torvalds * and other special files. --ADM 41271da177e4SLinus Torvalds */ 412846ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 412946ea89ebSDominik Brodowski const char __user *newname, int flags) 41301da177e4SLinus Torvalds { 41311da177e4SLinus Torvalds struct dentry *new_dentry; 4132dae6ad8fSAl Viro struct path old_path, new_path; 4133146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 413411a7b371SAneesh Kumar K.V int how = 0; 41351da177e4SLinus Torvalds int error; 41361da177e4SLinus Torvalds 413711a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4138c04030e1SUlrich Drepper return -EINVAL; 413911a7b371SAneesh Kumar K.V /* 4140f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4141f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4142f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 414311a7b371SAneesh Kumar K.V */ 4144f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4145f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4146f0cc6ffbSLinus Torvalds return -ENOENT; 414711a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4148f0cc6ffbSLinus Torvalds } 4149c04030e1SUlrich Drepper 415011a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 415111a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4152442e31caSJeff Layton retry: 415311a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 41541da177e4SLinus Torvalds if (error) 41552ad94ae6SAl Viro return error; 41562ad94ae6SAl Viro 4157442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4158442e31caSJeff Layton (how & LOOKUP_REVAL)); 41591da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41606902d925SDave Hansen if (IS_ERR(new_dentry)) 4161dae6ad8fSAl Viro goto out; 4162dae6ad8fSAl Viro 4163dae6ad8fSAl Viro error = -EXDEV; 4164dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4165dae6ad8fSAl Viro goto out_dput; 4166800179c9SKees Cook error = may_linkat(&old_path); 4167800179c9SKees Cook if (unlikely(error)) 4168800179c9SKees Cook goto out_dput; 4169dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4170be6d3e56SKentaro Takeda if (error) 4171a8104a9fSAl Viro goto out_dput; 4172146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 417375c3f29dSDave Hansen out_dput: 4174921a1650SAl Viro done_path_create(&new_path, new_dentry); 4175146a8595SJ. Bruce Fields if (delegated_inode) { 4176146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4177d22e6338SOleg Drokin if (!error) { 4178d22e6338SOleg Drokin path_put(&old_path); 4179146a8595SJ. Bruce Fields goto retry; 4180146a8595SJ. Bruce Fields } 4181d22e6338SOleg Drokin } 4182442e31caSJeff Layton if (retry_estale(error, how)) { 4183d22e6338SOleg Drokin path_put(&old_path); 4184442e31caSJeff Layton how |= LOOKUP_REVAL; 4185442e31caSJeff Layton goto retry; 4186442e31caSJeff Layton } 41871da177e4SLinus Torvalds out: 41882d8f3038SAl Viro path_put(&old_path); 41891da177e4SLinus Torvalds 41901da177e4SLinus Torvalds return error; 41911da177e4SLinus Torvalds } 41921da177e4SLinus Torvalds 419346ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 419446ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 419546ea89ebSDominik Brodowski { 419646ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 419746ea89ebSDominik Brodowski } 419846ea89ebSDominik Brodowski 41993480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 42005590ff0dSUlrich Drepper { 420146ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 42025590ff0dSUlrich Drepper } 42035590ff0dSUlrich Drepper 4204bc27027aSMiklos Szeredi /** 4205bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4206bc27027aSMiklos Szeredi * @old_dir: parent of source 4207bc27027aSMiklos Szeredi * @old_dentry: source 4208bc27027aSMiklos Szeredi * @new_dir: parent of destination 4209bc27027aSMiklos Szeredi * @new_dentry: destination 4210bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4211520c8b16SMiklos Szeredi * @flags: rename flags 4212bc27027aSMiklos Szeredi * 4213bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4214bc27027aSMiklos Szeredi * 4215bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4216bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4217bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4218bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4219bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4220bc27027aSMiklos Szeredi * so. 4221bc27027aSMiklos Szeredi * 4222bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4223bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4224bc27027aSMiklos Szeredi * to be NFS exported. 4225bc27027aSMiklos Szeredi * 42261da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42271da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42281da177e4SLinus Torvalds * Problems: 42290117d427SMauro Carvalho Chehab * 4230d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42311da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42321da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4233a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42341da177e4SLinus Torvalds * story. 42356cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42366cedba89SJ. Bruce Fields * and source (if it is not a directory). 42371b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42381da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42391da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4240a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 42411da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 42421da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 42431da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4244a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 42451da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 42461da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 42471da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4248e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 42491b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 42501da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4251c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 42521da177e4SLinus Torvalds * locking]. 42531da177e4SLinus Torvalds */ 42541da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 42558e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4256520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 42571da177e4SLinus Torvalds { 42581da177e4SLinus Torvalds int error; 4259bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4260bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4261bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4262da1ce067SMiklos Szeredi bool new_is_dir = false; 4263da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 426449d31c2fSAl Viro struct name_snapshot old_name; 42651da177e4SLinus Torvalds 42668d3e2936SMiklos Szeredi if (source == target) 42671da177e4SLinus Torvalds return 0; 42681da177e4SLinus Torvalds 42691da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 42701da177e4SLinus Torvalds if (error) 42711da177e4SLinus Torvalds return error; 42721da177e4SLinus Torvalds 4273da1ce067SMiklos Szeredi if (!target) { 4274a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4275da1ce067SMiklos Szeredi } else { 4276da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4277da1ce067SMiklos Szeredi 4278da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42791da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4280da1ce067SMiklos Szeredi else 4281da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4282da1ce067SMiklos Szeredi } 42831da177e4SLinus Torvalds if (error) 42841da177e4SLinus Torvalds return error; 42851da177e4SLinus Torvalds 42862773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 42871da177e4SLinus Torvalds return -EPERM; 42881da177e4SLinus Torvalds 4289bc27027aSMiklos Szeredi /* 4290bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4291bc27027aSMiklos Szeredi * we'll need to flip '..'. 4292bc27027aSMiklos Szeredi */ 4293da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4294da1ce067SMiklos Szeredi if (is_dir) { 4295bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4296bc27027aSMiklos Szeredi if (error) 4297bc27027aSMiklos Szeredi return error; 4298bc27027aSMiklos Szeredi } 4299da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4300da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4301da1ce067SMiklos Szeredi if (error) 4302da1ce067SMiklos Szeredi return error; 4303da1ce067SMiklos Szeredi } 4304da1ce067SMiklos Szeredi } 43050eeca283SRobert Love 43060b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 43070b3974ebSMiklos Szeredi flags); 4308bc27027aSMiklos Szeredi if (error) 4309bc27027aSMiklos Szeredi return error; 4310bc27027aSMiklos Szeredi 431149d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4312bc27027aSMiklos Szeredi dget(new_dentry); 4313da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4314bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4315bc27027aSMiklos Szeredi else if (target) 43165955102cSAl Viro inode_lock(target); 4317bc27027aSMiklos Szeredi 4318bc27027aSMiklos Szeredi error = -EBUSY; 43197af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4320bc27027aSMiklos Szeredi goto out; 4321bc27027aSMiklos Szeredi 4322da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4323bc27027aSMiklos Szeredi error = -EMLINK; 4324da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4325bc27027aSMiklos Szeredi goto out; 4326da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4327da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4328da1ce067SMiklos Szeredi goto out; 4329da1ce067SMiklos Szeredi } 4330da1ce067SMiklos Szeredi if (!is_dir) { 4331bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4332bc27027aSMiklos Szeredi if (error) 4333bc27027aSMiklos Szeredi goto out; 4334da1ce067SMiklos Szeredi } 4335da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4336bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4337bc27027aSMiklos Szeredi if (error) 4338bc27027aSMiklos Szeredi goto out; 4339bc27027aSMiklos Szeredi } 4340520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4341520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4342bc27027aSMiklos Szeredi if (error) 4343bc27027aSMiklos Szeredi goto out; 4344bc27027aSMiklos Szeredi 4345da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 43468767712fSAl Viro if (is_dir) { 43478767712fSAl Viro shrink_dcache_parent(new_dentry); 4348bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 43498767712fSAl Viro } 4350bc27027aSMiklos Szeredi dont_mount(new_dentry); 43518ed936b5SEric W. Biederman detach_mounts(new_dentry); 4352bc27027aSMiklos Szeredi } 4353da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4354da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4355bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4356da1ce067SMiklos Szeredi else 4357da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4358da1ce067SMiklos Szeredi } 4359bc27027aSMiklos Szeredi out: 4360da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4361bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4362bc27027aSMiklos Szeredi else if (target) 43635955102cSAl Viro inode_unlock(target); 4364bc27027aSMiklos Szeredi dput(new_dentry); 4365da1ce067SMiklos Szeredi if (!error) { 4366f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4367da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4368da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4369f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4370da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4371da1ce067SMiklos Szeredi } 4372da1ce067SMiklos Szeredi } 437349d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 43740eeca283SRobert Love 43751da177e4SLinus Torvalds return error; 43761da177e4SLinus Torvalds } 43774d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43781da177e4SLinus Torvalds 4379ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4380ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 43811da177e4SLinus Torvalds { 43821da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43831da177e4SLinus Torvalds struct dentry *trap; 4384f5beed75SAl Viro struct path old_path, new_path; 4385f5beed75SAl Viro struct qstr old_last, new_last; 4386f5beed75SAl Viro int old_type, new_type; 43878e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 438891a27b2aSJeff Layton struct filename *from; 438991a27b2aSJeff Layton struct filename *to; 4390f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4391c6a94284SJeff Layton bool should_retry = false; 43922ad94ae6SAl Viro int error; 4393520c8b16SMiklos Szeredi 43940d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4395da1ce067SMiklos Szeredi return -EINVAL; 4396da1ce067SMiklos Szeredi 43970d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43980d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4399520c8b16SMiklos Szeredi return -EINVAL; 4400520c8b16SMiklos Szeredi 44010d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 44020d7a8555SMiklos Szeredi return -EPERM; 44030d7a8555SMiklos Szeredi 4404f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4405f5beed75SAl Viro target_flags = 0; 4406f5beed75SAl Viro 4407c6a94284SJeff Layton retry: 4408c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4409c1d4dd27SAl Viro &old_path, &old_last, &old_type); 441091a27b2aSJeff Layton if (IS_ERR(from)) { 441191a27b2aSJeff Layton error = PTR_ERR(from); 44121da177e4SLinus Torvalds goto exit; 441391a27b2aSJeff Layton } 44141da177e4SLinus Torvalds 4415c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4416c1d4dd27SAl Viro &new_path, &new_last, &new_type); 441791a27b2aSJeff Layton if (IS_ERR(to)) { 441891a27b2aSJeff Layton error = PTR_ERR(to); 44191da177e4SLinus Torvalds goto exit1; 442091a27b2aSJeff Layton } 44211da177e4SLinus Torvalds 44221da177e4SLinus Torvalds error = -EXDEV; 4423f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44241da177e4SLinus Torvalds goto exit2; 44251da177e4SLinus Torvalds 44261da177e4SLinus Torvalds error = -EBUSY; 4427f5beed75SAl Viro if (old_type != LAST_NORM) 44281da177e4SLinus Torvalds goto exit2; 44291da177e4SLinus Torvalds 44300a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44310a7c3937SMiklos Szeredi error = -EEXIST; 4432f5beed75SAl Viro if (new_type != LAST_NORM) 44331da177e4SLinus Torvalds goto exit2; 44341da177e4SLinus Torvalds 4435f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4436c30dabfeSJan Kara if (error) 4437c30dabfeSJan Kara goto exit2; 4438c30dabfeSJan Kara 44398e6d782cSJ. Bruce Fields retry_deleg: 4440f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 44411da177e4SLinus Torvalds 4442f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 44431da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 44441da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 44451da177e4SLinus Torvalds goto exit3; 44461da177e4SLinus Torvalds /* source must exist */ 44471da177e4SLinus Torvalds error = -ENOENT; 4448b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44491da177e4SLinus Torvalds goto exit4; 4450f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44511da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44521da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44531da177e4SLinus Torvalds goto exit4; 44540a7c3937SMiklos Szeredi error = -EEXIST; 44550a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44560a7c3937SMiklos Szeredi goto exit5; 4457da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4458da1ce067SMiklos Szeredi error = -ENOENT; 4459da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4460da1ce067SMiklos Szeredi goto exit5; 4461da1ce067SMiklos Szeredi 4462da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4463da1ce067SMiklos Szeredi error = -ENOTDIR; 4464f5beed75SAl Viro if (new_last.name[new_last.len]) 4465da1ce067SMiklos Szeredi goto exit5; 4466da1ce067SMiklos Szeredi } 4467da1ce067SMiklos Szeredi } 44680a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44690a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44700a7c3937SMiklos Szeredi error = -ENOTDIR; 4471f5beed75SAl Viro if (old_last.name[old_last.len]) 44720a7c3937SMiklos Szeredi goto exit5; 4473f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44740a7c3937SMiklos Szeredi goto exit5; 44750a7c3937SMiklos Szeredi } 44760a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44770a7c3937SMiklos Szeredi error = -EINVAL; 44780a7c3937SMiklos Szeredi if (old_dentry == trap) 44790a7c3937SMiklos Szeredi goto exit5; 44801da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4481da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44821da177e4SLinus Torvalds error = -ENOTEMPTY; 44831da177e4SLinus Torvalds if (new_dentry == trap) 44841da177e4SLinus Torvalds goto exit5; 44851da177e4SLinus Torvalds 4486f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4487f5beed75SAl Viro &new_path, new_dentry, flags); 4488be6d3e56SKentaro Takeda if (error) 4489c30dabfeSJan Kara goto exit5; 4490f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4491f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4492520c8b16SMiklos Szeredi &delegated_inode, flags); 44931da177e4SLinus Torvalds exit5: 44941da177e4SLinus Torvalds dput(new_dentry); 44951da177e4SLinus Torvalds exit4: 44961da177e4SLinus Torvalds dput(old_dentry); 44971da177e4SLinus Torvalds exit3: 4498f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44998e6d782cSJ. Bruce Fields if (delegated_inode) { 45008e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 45018e6d782cSJ. Bruce Fields if (!error) 45028e6d782cSJ. Bruce Fields goto retry_deleg; 45038e6d782cSJ. Bruce Fields } 4504f5beed75SAl Viro mnt_drop_write(old_path.mnt); 45051da177e4SLinus Torvalds exit2: 4506c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4507c6a94284SJeff Layton should_retry = true; 4508f5beed75SAl Viro path_put(&new_path); 45092ad94ae6SAl Viro putname(to); 45101da177e4SLinus Torvalds exit1: 4511f5beed75SAl Viro path_put(&old_path); 45121da177e4SLinus Torvalds putname(from); 4513c6a94284SJeff Layton if (should_retry) { 4514c6a94284SJeff Layton should_retry = false; 4515c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4516c6a94284SJeff Layton goto retry; 4517c6a94284SJeff Layton } 45182ad94ae6SAl Viro exit: 45191da177e4SLinus Torvalds return error; 45201da177e4SLinus Torvalds } 45211da177e4SLinus Torvalds 4522ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4523ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4524ee81feb6SDominik Brodowski { 4525ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4526ee81feb6SDominik Brodowski } 4527ee81feb6SDominik Brodowski 4528520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4529520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4530520c8b16SMiklos Szeredi { 4531ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4532520c8b16SMiklos Szeredi } 4533520c8b16SMiklos Szeredi 4534a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45355590ff0dSUlrich Drepper { 4536ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45375590ff0dSUlrich Drepper } 45385590ff0dSUlrich Drepper 4539787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4540787fb6bcSMiklos Szeredi { 4541787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4542787fb6bcSMiklos Szeredi if (error) 4543787fb6bcSMiklos Szeredi return error; 4544787fb6bcSMiklos Szeredi 4545787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4546787fb6bcSMiklos Szeredi return -EPERM; 4547787fb6bcSMiklos Szeredi 4548787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4549787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4550787fb6bcSMiklos Szeredi } 4551787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4552787fb6bcSMiklos Szeredi 45535d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 45541da177e4SLinus Torvalds { 45555d826c84SAl Viro int len = PTR_ERR(link); 45561da177e4SLinus Torvalds if (IS_ERR(link)) 45571da177e4SLinus Torvalds goto out; 45581da177e4SLinus Torvalds 45591da177e4SLinus Torvalds len = strlen(link); 45601da177e4SLinus Torvalds if (len > (unsigned) buflen) 45611da177e4SLinus Torvalds len = buflen; 45621da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45631da177e4SLinus Torvalds len = -EFAULT; 45641da177e4SLinus Torvalds out: 45651da177e4SLinus Torvalds return len; 45661da177e4SLinus Torvalds } 45671da177e4SLinus Torvalds 4568d60874cdSMiklos Szeredi /** 4569fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4570fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4571fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4572fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4573fd4a0edfSMiklos Szeredi * 4574fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4575fd4a0edfSMiklos Szeredi * 4576fd4a0edfSMiklos Szeredi * Does not call security hook. 4577fd4a0edfSMiklos Szeredi */ 4578fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4579fd4a0edfSMiklos Szeredi { 4580fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4581f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4582f2df5da6SAl Viro const char *link; 4583f2df5da6SAl Viro int res; 4584fd4a0edfSMiklos Szeredi 458576fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 458676fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 458776fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 458876fca90eSMiklos Szeredi 458976fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4590fd4a0edfSMiklos Szeredi return -EINVAL; 4591fd4a0edfSMiklos Szeredi 459276fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 459376fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 459476fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 459576fca90eSMiklos Szeredi } 459676fca90eSMiklos Szeredi 45974c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4598f2df5da6SAl Viro if (!link) { 4599f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4600f2df5da6SAl Viro if (IS_ERR(link)) 4601f2df5da6SAl Viro return PTR_ERR(link); 4602f2df5da6SAl Viro } 4603f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4604f2df5da6SAl Viro do_delayed_call(&done); 4605f2df5da6SAl Viro return res; 4606fd4a0edfSMiklos Szeredi } 4607fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 46081da177e4SLinus Torvalds 4609d60874cdSMiklos Szeredi /** 4610d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4611d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4612d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4613d60874cdSMiklos Szeredi * 4614d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4615d60874cdSMiklos Szeredi * 4616d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4617d60874cdSMiklos Szeredi * 4618d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4619d60874cdSMiklos Szeredi */ 4620d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4621d60874cdSMiklos Szeredi { 4622d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4623d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4624d60874cdSMiklos Szeredi 4625d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4626d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4627d60874cdSMiklos Szeredi if (!res) 4628d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4629d60874cdSMiklos Szeredi } 4630d60874cdSMiklos Szeredi return res; 4631d60874cdSMiklos Szeredi } 4632d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4633d60874cdSMiklos Szeredi 46341da177e4SLinus Torvalds /* get the link contents into pagecache */ 46356b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4636fceef393SAl Viro struct delayed_call *callback) 46371da177e4SLinus Torvalds { 4638ebd09abbSDuane Griffin char *kaddr; 46391da177e4SLinus Torvalds struct page *page; 46406b255391SAl Viro struct address_space *mapping = inode->i_mapping; 46416b255391SAl Viro 4642d3883d4fSAl Viro if (!dentry) { 4643d3883d4fSAl Viro page = find_get_page(mapping, 0); 4644d3883d4fSAl Viro if (!page) 46456b255391SAl Viro return ERR_PTR(-ECHILD); 4646d3883d4fSAl Viro if (!PageUptodate(page)) { 4647d3883d4fSAl Viro put_page(page); 4648d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4649d3883d4fSAl Viro } 4650d3883d4fSAl Viro } else { 4651090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 46521da177e4SLinus Torvalds if (IS_ERR(page)) 46536fe6900eSNick Piggin return (char*)page; 4654d3883d4fSAl Viro } 4655fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 465621fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 465721fc61c7SAl Viro kaddr = page_address(page); 46586b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4659ebd09abbSDuane Griffin return kaddr; 46601da177e4SLinus Torvalds } 46611da177e4SLinus Torvalds 46626b255391SAl Viro EXPORT_SYMBOL(page_get_link); 46631da177e4SLinus Torvalds 4664fceef393SAl Viro void page_put_link(void *arg) 46651da177e4SLinus Torvalds { 4666fceef393SAl Viro put_page(arg); 46671da177e4SLinus Torvalds } 46684d359507SAl Viro EXPORT_SYMBOL(page_put_link); 46691da177e4SLinus Torvalds 4670aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4671aa80deabSAl Viro { 4672fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46736b255391SAl Viro int res = readlink_copy(buffer, buflen, 46746b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4675fceef393SAl Viro &done)); 4676fceef393SAl Viro do_delayed_call(&done); 4677aa80deabSAl Viro return res; 4678aa80deabSAl Viro } 4679aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4680aa80deabSAl Viro 468154566b2cSNick Piggin /* 468254566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 468354566b2cSNick Piggin */ 468454566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 46851da177e4SLinus Torvalds { 46861da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 46870adb25d2SKirill Korotaev struct page *page; 4688afddba49SNick Piggin void *fsdata; 4689beb497abSDmitriy Monakhov int err; 4690c718a975STetsuo Handa unsigned int flags = 0; 469154566b2cSNick Piggin if (nofs) 469254566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 46931da177e4SLinus Torvalds 46947e53cac4SNeilBrown retry: 4695afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 469654566b2cSNick Piggin flags, &page, &fsdata); 46971da177e4SLinus Torvalds if (err) 4698afddba49SNick Piggin goto fail; 4699afddba49SNick Piggin 470021fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4701afddba49SNick Piggin 4702afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4703afddba49SNick Piggin page, fsdata); 47041da177e4SLinus Torvalds if (err < 0) 47051da177e4SLinus Torvalds goto fail; 4706afddba49SNick Piggin if (err < len-1) 4707afddba49SNick Piggin goto retry; 4708afddba49SNick Piggin 47091da177e4SLinus Torvalds mark_inode_dirty(inode); 47101da177e4SLinus Torvalds return 0; 47111da177e4SLinus Torvalds fail: 47121da177e4SLinus Torvalds return err; 47131da177e4SLinus Torvalds } 47144d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 47151da177e4SLinus Torvalds 47160adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 47170adb25d2SKirill Korotaev { 47180adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4719c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 47200adb25d2SKirill Korotaev } 47214d359507SAl Viro EXPORT_SYMBOL(page_symlink); 47220adb25d2SKirill Korotaev 472392e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 47246b255391SAl Viro .get_link = page_get_link, 47251da177e4SLinus Torvalds }; 47261da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4727