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 /** 5536b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 554397d425dSEric W. Biederman * 555397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 556397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 557397d425dSEric W. Biederman */ 5586b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 559397d425dSEric W. Biederman { 56095dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 561397d425dSEric W. Biederman 56295dd7758SEric W. Biederman /* Bind mounts and multi-root filesystems can have disconnected paths */ 56395dd7758SEric W. Biederman if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root)) 564397d425dSEric W. Biederman return true; 565397d425dSEric W. Biederman 5666b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 567397d425dSEric W. Biederman } 568397d425dSEric W. Biederman 569894bc8c4SAl Viro static inline int nd_alloc_stack(struct nameidata *nd) 570894bc8c4SAl Viro { 571da4e0be0SAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 572894bc8c4SAl Viro return 0; 573894bc8c4SAl Viro if (likely(nd->stack != nd->internal)) 574894bc8c4SAl Viro return 0; 575894bc8c4SAl Viro return __nd_alloc_stack(nd); 576894bc8c4SAl Viro } 577894bc8c4SAl Viro 5787973387aSAl Viro static void drop_links(struct nameidata *nd) 5797973387aSAl Viro { 5807973387aSAl Viro int i = nd->depth; 5817973387aSAl Viro while (i--) { 5827973387aSAl Viro struct saved *last = nd->stack + i; 583fceef393SAl Viro do_delayed_call(&last->done); 584fceef393SAl Viro clear_delayed_call(&last->done); 5857973387aSAl Viro } 5867973387aSAl Viro } 5877973387aSAl Viro 5887973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5897973387aSAl Viro { 5907973387aSAl Viro drop_links(nd); 5917973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5927973387aSAl Viro int i; 5937973387aSAl Viro path_put(&nd->path); 5947973387aSAl Viro for (i = 0; i < nd->depth; i++) 5957973387aSAl Viro path_put(&nd->stack[i].link); 59684a2bd39SAl Viro if (nd->flags & LOOKUP_ROOT_GRABBED) { 597102b8af2SAl Viro path_put(&nd->root); 59884a2bd39SAl Viro nd->flags &= ~LOOKUP_ROOT_GRABBED; 599102b8af2SAl Viro } 6007973387aSAl Viro } else { 6017973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6027973387aSAl Viro rcu_read_unlock(); 6037973387aSAl Viro } 6047973387aSAl Viro nd->depth = 0; 6057973387aSAl Viro } 6067973387aSAl Viro 6077973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6087973387aSAl Viro static bool legitimize_path(struct nameidata *nd, 6097973387aSAl Viro struct path *path, unsigned seq) 6107973387aSAl Viro { 6117973387aSAl Viro int res = __legitimize_mnt(path->mnt, nd->m_seq); 6127973387aSAl Viro if (unlikely(res)) { 6137973387aSAl Viro if (res > 0) 6147973387aSAl Viro path->mnt = NULL; 6157973387aSAl Viro path->dentry = NULL; 6167973387aSAl Viro return false; 6177973387aSAl Viro } 6187973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6197973387aSAl Viro path->dentry = NULL; 6207973387aSAl Viro return false; 6217973387aSAl Viro } 6227973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6237973387aSAl Viro } 6247973387aSAl Viro 6257973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6267973387aSAl Viro { 6277973387aSAl Viro int i; 6287973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6297973387aSAl Viro struct saved *last = nd->stack + i; 6307973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6317973387aSAl Viro drop_links(nd); 6327973387aSAl Viro nd->depth = i + 1; 6337973387aSAl Viro return false; 6347973387aSAl Viro } 6357973387aSAl Viro } 6367973387aSAl Viro return true; 6377973387aSAl Viro } 6387973387aSAl Viro 639ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 640ee594bffSAl Viro { 641adb21d2bSAleksa Sarai /* 642adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 643adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 644adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 645adb21d2bSAleksa Sarai */ 646adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 647adb21d2bSAleksa Sarai return false; 648adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 649ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 650ee594bffSAl Viro return true; 65184a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 652ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 653ee594bffSAl Viro } 654ee594bffSAl Viro 65519660af7SAl Viro /* 65631e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65719660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 65819660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 65957e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 66019660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 66119660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 66219660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 66319660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66431e6b01fSNick Piggin */ 66531e6b01fSNick Piggin 66631e6b01fSNick Piggin /** 66719660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 66819660af7SAl Viro * @nd: nameidata pathwalk data 66939191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 67031e6b01fSNick Piggin * 6714675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6724675ac39SAl Viro * for ref-walk mode. 6734675ac39SAl Viro * Must be called from rcu-walk context. 6747973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6757973387aSAl Viro * terminate_walk(). 67631e6b01fSNick Piggin */ 6774675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 67831e6b01fSNick Piggin { 67931e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 68031e6b01fSNick Piggin 68131e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 682e5c832d5SLinus Torvalds 683e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6847973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6854675ac39SAl Viro goto out1; 68684a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 68784a2bd39SAl Viro goto out; 688ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 6894675ac39SAl Viro goto out; 6904675ac39SAl Viro rcu_read_unlock(); 6914675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6924675ac39SAl Viro return 0; 6934675ac39SAl Viro 69484a2bd39SAl Viro out1: 6954675ac39SAl Viro nd->path.mnt = NULL; 6964675ac39SAl Viro nd->path.dentry = NULL; 6974675ac39SAl Viro out: 6984675ac39SAl Viro rcu_read_unlock(); 6994675ac39SAl Viro return -ECHILD; 7004675ac39SAl Viro } 7014675ac39SAl Viro 7024675ac39SAl Viro /** 7034675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7044675ac39SAl Viro * @nd: nameidata pathwalk data 7054675ac39SAl Viro * @dentry: child of nd->path.dentry 7064675ac39SAl Viro * @seq: seq number to check dentry against 7074675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7084675ac39SAl Viro * 7094675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7104675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7114675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7124675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7134675ac39SAl Viro * terminate_walk(). 7144675ac39SAl Viro */ 7154675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7164675ac39SAl Viro { 7174675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7184675ac39SAl Viro 7194675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7204675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7214675ac39SAl Viro goto out2; 7227973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7237973387aSAl Viro goto out2; 7244675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7257973387aSAl Viro goto out1; 72648a066e7SAl Viro 72715570086SLinus Torvalds /* 7284675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7294675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7304675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7314675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7324675ac39SAl Viro * know the parent must still be valid if the child sequence number is 73315570086SLinus Torvalds */ 7344675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 735e5c832d5SLinus Torvalds goto out; 73684a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 73784a2bd39SAl Viro goto out_dput; 738e5c832d5SLinus Torvalds /* 739e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 740e5c832d5SLinus Torvalds * still valid and get it if required. 741e5c832d5SLinus Torvalds */ 74284a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 74384a2bd39SAl Viro goto out_dput; 7448b61e74fSAl Viro rcu_read_unlock(); 74531e6b01fSNick Piggin return 0; 74619660af7SAl Viro 7477973387aSAl Viro out2: 7487973387aSAl Viro nd->path.mnt = NULL; 7497973387aSAl Viro out1: 7507973387aSAl Viro nd->path.dentry = NULL; 751e5c832d5SLinus Torvalds out: 7528b61e74fSAl Viro rcu_read_unlock(); 75384a2bd39SAl Viro return -ECHILD; 75484a2bd39SAl Viro out_dput: 75584a2bd39SAl Viro rcu_read_unlock(); 75684a2bd39SAl Viro dput(dentry); 75731e6b01fSNick Piggin return -ECHILD; 75831e6b01fSNick Piggin } 75931e6b01fSNick Piggin 7604ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 76134286d66SNick Piggin { 762a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7634ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 764a89f8337SAl Viro else 765a89f8337SAl Viro return 1; 76634286d66SNick Piggin } 76734286d66SNick Piggin 7689f1fafeeSAl Viro /** 7699f1fafeeSAl Viro * complete_walk - successful completion of path walk 7709f1fafeeSAl Viro * @nd: pointer nameidata 77139159de2SJeff Layton * 7729f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7739f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7749f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7759f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7769f1fafeeSAl Viro * need to drop nd->path. 77739159de2SJeff Layton */ 7789f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 77939159de2SJeff Layton { 78016c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 78139159de2SJeff Layton int status; 78239159de2SJeff Layton 7839f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 784adb21d2bSAleksa Sarai /* 785adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 786adb21d2bSAleksa Sarai * externally-managed nd->root. 787adb21d2bSAleksa Sarai */ 788adb21d2bSAleksa Sarai if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) 7899f1fafeeSAl Viro nd->root.mnt = NULL; 7904675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 79148a066e7SAl Viro return -ECHILD; 79248a066e7SAl Viro } 7939f1fafeeSAl Viro 794adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 795adb21d2bSAleksa Sarai /* 796adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 797adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 798adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 799adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 800adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 801adb21d2bSAleksa Sarai * 802adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 803adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 804adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 805adb21d2bSAleksa Sarai * requested root to userspace. 806adb21d2bSAleksa Sarai * 807adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 808adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 809adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 810adb21d2bSAleksa Sarai */ 811adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 812adb21d2bSAleksa Sarai return -EXDEV; 813adb21d2bSAleksa Sarai } 814adb21d2bSAleksa Sarai 81516c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 81639159de2SJeff Layton return 0; 81739159de2SJeff Layton 818ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 81916c2cd71SAl Viro return 0; 82016c2cd71SAl Viro 821ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 82239159de2SJeff Layton if (status > 0) 82339159de2SJeff Layton return 0; 82439159de2SJeff Layton 82516c2cd71SAl Viro if (!status) 82639159de2SJeff Layton status = -ESTALE; 82716c2cd71SAl Viro 82839159de2SJeff Layton return status; 82939159de2SJeff Layton } 83039159de2SJeff Layton 831740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 8322a737871SAl Viro { 83331e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8349e6697e2SAl Viro 835adb21d2bSAleksa Sarai /* 836adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 837adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 838adb21d2bSAleksa Sarai * from the dirfd. 839adb21d2bSAleksa Sarai */ 840adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 841adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 842adb21d2bSAleksa Sarai 8439e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8448f47a016SAl Viro unsigned seq; 845c28cc364SNick Piggin 846c28cc364SNick Piggin do { 847c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 84831e6b01fSNick Piggin nd->root = fs->root; 8498f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 850c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8519e6697e2SAl Viro } else { 8529e6697e2SAl Viro get_fs_root(fs, &nd->root); 85384a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 8549e6697e2SAl Viro } 855740a1678SAleksa Sarai return 0; 85631e6b01fSNick Piggin } 85731e6b01fSNick Piggin 8587b9337aaSNick Piggin static inline void path_to_nameidata(const struct path *path, 8597b9337aaSNick Piggin struct nameidata *nd) 860051d3812SIan Kent { 86131e6b01fSNick Piggin if (!(nd->flags & LOOKUP_RCU)) { 8624ac91378SJan Blunck dput(nd->path.dentry); 86331e6b01fSNick Piggin if (nd->path.mnt != path->mnt) 8644ac91378SJan Blunck mntput(nd->path.mnt); 8659a229683SHuang Shijie } 86631e6b01fSNick Piggin nd->path.mnt = path->mnt; 8674ac91378SJan Blunck nd->path.dentry = path->dentry; 868051d3812SIan Kent } 869051d3812SIan Kent 870248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 871248fb5b9SAl Viro { 872adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 873adb21d2bSAleksa Sarai return -EXDEV; 87472ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 87572ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 87672ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 87772ba2929SAleksa Sarai return -EXDEV; 87872ba2929SAleksa Sarai } 879740a1678SAleksa Sarai if (!nd->root.mnt) { 880740a1678SAleksa Sarai int error = set_root(nd); 881740a1678SAleksa Sarai if (error) 882740a1678SAleksa Sarai return error; 883740a1678SAleksa Sarai } 884248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 885248fb5b9SAl Viro struct dentry *d; 886248fb5b9SAl Viro nd->path = nd->root; 887248fb5b9SAl Viro d = nd->path.dentry; 888248fb5b9SAl Viro nd->inode = d->d_inode; 889248fb5b9SAl Viro nd->seq = nd->root_seq; 890248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 891248fb5b9SAl Viro return -ECHILD; 892248fb5b9SAl Viro } else { 893248fb5b9SAl Viro path_put(&nd->path); 894248fb5b9SAl Viro nd->path = nd->root; 895248fb5b9SAl Viro path_get(&nd->path); 896248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 897248fb5b9SAl Viro } 898248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 899248fb5b9SAl Viro return 0; 900248fb5b9SAl Viro } 901248fb5b9SAl Viro 902b5fb63c1SChristoph Hellwig /* 9036b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 904b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 905b5fb63c1SChristoph Hellwig */ 9061bc82070SAleksa Sarai int nd_jump_link(struct path *path) 907b5fb63c1SChristoph Hellwig { 9084b99d499SAleksa Sarai int error = -ELOOP; 9096e77137bSAl Viro struct nameidata *nd = current->nameidata; 910b5fb63c1SChristoph Hellwig 9114b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9124b99d499SAleksa Sarai goto err; 9134b99d499SAleksa Sarai 91472ba2929SAleksa Sarai error = -EXDEV; 91572ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 91672ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 91772ba2929SAleksa Sarai goto err; 91872ba2929SAleksa Sarai } 919adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 920adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 921adb21d2bSAleksa Sarai goto err; 92272ba2929SAleksa Sarai 9234b99d499SAleksa Sarai path_put(&nd->path); 924b5fb63c1SChristoph Hellwig nd->path = *path; 925b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 926b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 9271bc82070SAleksa Sarai return 0; 9284b99d499SAleksa Sarai 9294b99d499SAleksa Sarai err: 9304b99d499SAleksa Sarai path_put(path); 9314b99d499SAleksa Sarai return error; 932b5fb63c1SChristoph Hellwig } 933b5fb63c1SChristoph Hellwig 934b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 935574197e0SAl Viro { 93621c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 937fceef393SAl Viro do_delayed_call(&last->done); 9386548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 939b9ff4429SAl Viro path_put(&last->link); 940574197e0SAl Viro } 941574197e0SAl Viro 942561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 943561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 94430aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 94530aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 946800179c9SKees Cook 947800179c9SKees Cook /** 948800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 94955852635SRandy Dunlap * @nd: nameidata pathwalk data 950800179c9SKees Cook * 951800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 952800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 953800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 954800179c9SKees Cook * processes from failing races against path names that may change out 955800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 956800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 957800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 958800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 959800179c9SKees Cook * 960800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 961800179c9SKees Cook */ 962ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 963800179c9SKees Cook { 964800179c9SKees Cook const struct inode *parent; 9652d7f9e2aSSeth Forshee kuid_t puid; 966800179c9SKees Cook 967800179c9SKees Cook if (!sysctl_protected_symlinks) 968800179c9SKees Cook return 0; 969800179c9SKees Cook 970800179c9SKees Cook /* Allowed if owner and follower match. */ 97181abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 972800179c9SKees Cook return 0; 973800179c9SKees Cook 974800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 975aa65fa35SAl Viro parent = nd->inode; 976800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 977800179c9SKees Cook return 0; 978800179c9SKees Cook 979800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9802d7f9e2aSSeth Forshee puid = parent->i_uid; 9812d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 982800179c9SKees Cook return 0; 983800179c9SKees Cook 98431956502SAl Viro if (nd->flags & LOOKUP_RCU) 98531956502SAl Viro return -ECHILD; 98631956502SAl Viro 987ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 988245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 989800179c9SKees Cook return -EACCES; 990800179c9SKees Cook } 991800179c9SKees Cook 992800179c9SKees Cook /** 993800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 994800179c9SKees Cook * @inode: the source inode to hardlink from 995800179c9SKees Cook * 996800179c9SKees Cook * Return false if at least one of the following conditions: 997800179c9SKees Cook * - inode is not a regular file 998800179c9SKees Cook * - inode is setuid 999800179c9SKees Cook * - inode is setgid and group-exec 1000800179c9SKees Cook * - access failure for read and write 1001800179c9SKees Cook * 1002800179c9SKees Cook * Otherwise returns true. 1003800179c9SKees Cook */ 1004800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 1005800179c9SKees Cook { 1006800179c9SKees Cook umode_t mode = inode->i_mode; 1007800179c9SKees Cook 1008800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1009800179c9SKees Cook if (!S_ISREG(mode)) 1010800179c9SKees Cook return false; 1011800179c9SKees Cook 1012800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1013800179c9SKees Cook if (mode & S_ISUID) 1014800179c9SKees Cook return false; 1015800179c9SKees Cook 1016800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1017800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1018800179c9SKees Cook return false; 1019800179c9SKees Cook 1020800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1021800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 1022800179c9SKees Cook return false; 1023800179c9SKees Cook 1024800179c9SKees Cook return true; 1025800179c9SKees Cook } 1026800179c9SKees Cook 1027800179c9SKees Cook /** 1028800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1029800179c9SKees Cook * @link: the source to hardlink from 1030800179c9SKees Cook * 1031800179c9SKees Cook * Block hardlink when all of: 1032800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1033800179c9SKees Cook * - fsuid does not match inode 1034800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1035f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1036800179c9SKees Cook * 1037800179c9SKees Cook * Returns 0 if successful, -ve on error. 1038800179c9SKees Cook */ 1039800179c9SKees Cook static int may_linkat(struct path *link) 1040800179c9SKees Cook { 1041593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1042593d1ce8SEric W. Biederman 1043593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1044593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 1045593d1ce8SEric W. Biederman return -EOVERFLOW; 1046800179c9SKees Cook 1047800179c9SKees Cook if (!sysctl_protected_hardlinks) 1048800179c9SKees Cook return 0; 1049800179c9SKees Cook 1050800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1051800179c9SKees Cook * otherwise, it must be a safe source. 1052800179c9SKees Cook */ 1053cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1054800179c9SKees Cook return 0; 1055800179c9SKees Cook 1056245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1057800179c9SKees Cook return -EPERM; 1058800179c9SKees Cook } 1059800179c9SKees Cook 106030aba665SSalvatore Mesoraca /** 106130aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 106230aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 106330aba665SSalvatore Mesoraca * exist. 1064d0cb5018SAl Viro * @dir_mode: mode bits of directory 1065d0cb5018SAl Viro * @dir_uid: owner of directory 106630aba665SSalvatore Mesoraca * @inode: the inode of the file to open 106730aba665SSalvatore Mesoraca * 106830aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 106930aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 107030aba665SSalvatore Mesoraca * - the file already exists 107130aba665SSalvatore Mesoraca * - we are in a sticky directory 107230aba665SSalvatore Mesoraca * - we don't own the file 107330aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 107430aba665SSalvatore Mesoraca * - the directory is world writable 107530aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 107630aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 107730aba665SSalvatore Mesoraca * be enough. 107830aba665SSalvatore Mesoraca * 107930aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 108030aba665SSalvatore Mesoraca */ 1081d0cb5018SAl Viro static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid, 108230aba665SSalvatore Mesoraca struct inode * const inode) 108330aba665SSalvatore Mesoraca { 108430aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 108530aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1086d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1087d0cb5018SAl Viro uid_eq(inode->i_uid, dir_uid) || 108830aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 108930aba665SSalvatore Mesoraca return 0; 109030aba665SSalvatore Mesoraca 1091d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1092d0cb5018SAl Viro (dir_mode & 0020 && 109330aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 109430aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1095245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1096245d7369SKees Cook "sticky_create_fifo" : 1097245d7369SKees Cook "sticky_create_regular"; 1098245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 109930aba665SSalvatore Mesoraca return -EACCES; 110030aba665SSalvatore Mesoraca } 110130aba665SSalvatore Mesoraca return 0; 110230aba665SSalvatore Mesoraca } 110330aba665SSalvatore Mesoraca 1104f015f126SDavid Howells /* 1105f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1106f015f126SDavid Howells * 1107f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1108f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1109f015f126SDavid Howells * Up is towards /. 1110f015f126SDavid Howells * 1111f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1112f015f126SDavid Howells * root. 1113f015f126SDavid Howells */ 1114bab77ebfSAl Viro int follow_up(struct path *path) 11151da177e4SLinus Torvalds { 11160714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11170714a533SAl Viro struct mount *parent; 11181da177e4SLinus Torvalds struct dentry *mountpoint; 111999b7db7bSNick Piggin 112048a066e7SAl Viro read_seqlock_excl(&mount_lock); 11210714a533SAl Viro parent = mnt->mnt_parent; 11223c0a6163SAl Viro if (parent == mnt) { 112348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11241da177e4SLinus Torvalds return 0; 11251da177e4SLinus Torvalds } 11260714a533SAl Viro mntget(&parent->mnt); 1127a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 112848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1129bab77ebfSAl Viro dput(path->dentry); 1130bab77ebfSAl Viro path->dentry = mountpoint; 1131bab77ebfSAl Viro mntput(path->mnt); 11320714a533SAl Viro path->mnt = &parent->mnt; 11331da177e4SLinus Torvalds return 1; 11341da177e4SLinus Torvalds } 11354d359507SAl Viro EXPORT_SYMBOL(follow_up); 11361da177e4SLinus Torvalds 1137b5c84bf6SNick Piggin /* 11389875cf80SDavid Howells * Perform an automount 11399875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11409875cf80SDavid Howells * were called with. 11411da177e4SLinus Torvalds */ 11421c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 114331e6b01fSNick Piggin { 114425e195aaSAl Viro struct dentry *dentry = path->dentry; 11459875cf80SDavid Howells 11460ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11470ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11480ec26fd0SMiklos Szeredi * the name. 11490ec26fd0SMiklos Szeredi * 11500ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11515a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11520ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11530ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11540ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11550ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11565a30d8a2SDavid Howells */ 11571c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11585d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 115925e195aaSAl Viro dentry->d_inode) 11609875cf80SDavid Howells return -EISDIR; 11610ec26fd0SMiklos Szeredi 11621c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 11639875cf80SDavid Howells return -ELOOP; 11649875cf80SDavid Howells 116525e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1166ea5b778aSDavid Howells } 11679875cf80SDavid Howells 11689875cf80SDavid Howells /* 11699deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 11709deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 11719deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 11729deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 11739875cf80SDavid Howells */ 11749deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 11759deed3ebSAl Viro int *count, unsigned lookup_flags) 11769875cf80SDavid Howells { 11779deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 11789875cf80SDavid Howells bool need_mntput = false; 11798aef1884SAl Viro int ret = 0; 11809875cf80SDavid Howells 11819deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1182cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1183cc53ce53SDavid Howells * being held. */ 1184d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1185fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1186508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1187cc53ce53SDavid Howells if (ret < 0) 11888aef1884SAl Viro break; 1189cc53ce53SDavid Howells } 1190cc53ce53SDavid Howells 11919deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 11929875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 11939deed3ebSAl Viro if (mounted) { // ... in our namespace 11949875cf80SDavid Howells dput(path->dentry); 11959875cf80SDavid Howells if (need_mntput) 1196463ffb2eSAl Viro mntput(path->mnt); 1197463ffb2eSAl Viro path->mnt = mounted; 1198463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 11999deed3ebSAl Viro // here we know it's positive 12009deed3ebSAl Viro flags = path->dentry->d_flags; 12019875cf80SDavid Howells need_mntput = true; 12029875cf80SDavid Howells continue; 1203463ffb2eSAl Viro } 12041da177e4SLinus Torvalds } 12059875cf80SDavid Howells 12069deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 12079deed3ebSAl Viro break; 12089deed3ebSAl Viro 12099deed3ebSAl Viro // uncovered automount point 12109deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 12119deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 12129875cf80SDavid Howells if (ret < 0) 12138aef1884SAl Viro break; 12149875cf80SDavid Howells } 12159875cf80SDavid Howells 12169deed3ebSAl Viro if (ret == -EISDIR) 12179deed3ebSAl Viro ret = 0; 12189deed3ebSAl Viro // possible if you race with several mount --move 12199deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 12208aef1884SAl Viro mntput(path->mnt); 12219deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1222d41efb52SAl Viro ret = -ENOENT; 12239deed3ebSAl Viro *jumped = need_mntput; 12248402752eSAl Viro return ret; 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds 12279deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 12289deed3ebSAl Viro int *count, unsigned lookup_flags) 12299deed3ebSAl Viro { 12309deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 12319deed3ebSAl Viro 12329deed3ebSAl Viro /* fastpath */ 12339deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 12349deed3ebSAl Viro *jumped = false; 12359deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 12369deed3ebSAl Viro return -ENOENT; 12379deed3ebSAl Viro return 0; 12389deed3ebSAl Viro } 12399deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 12409deed3ebSAl Viro } 12419deed3ebSAl Viro 1242cc53ce53SDavid Howells int follow_down_one(struct path *path) 12431da177e4SLinus Torvalds { 12441da177e4SLinus Torvalds struct vfsmount *mounted; 12451da177e4SLinus Torvalds 12461c755af4SAl Viro mounted = lookup_mnt(path); 12471da177e4SLinus Torvalds if (mounted) { 12489393bd07SAl Viro dput(path->dentry); 12499393bd07SAl Viro mntput(path->mnt); 12509393bd07SAl Viro path->mnt = mounted; 12519393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12521da177e4SLinus Torvalds return 1; 12531da177e4SLinus Torvalds } 12541da177e4SLinus Torvalds return 0; 12551da177e4SLinus Torvalds } 12564d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12571da177e4SLinus Torvalds 12589875cf80SDavid Howells /* 12599deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 12609deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 12619deed3ebSAl Viro * caller is permitted to proceed or not. 12629deed3ebSAl Viro */ 12639deed3ebSAl Viro int follow_down(struct path *path) 12649deed3ebSAl Viro { 12659deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 12669deed3ebSAl Viro bool jumped; 12679deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 12689deed3ebSAl Viro 12699deed3ebSAl Viro if (path->mnt != mnt) 12709deed3ebSAl Viro mntput(mnt); 12719deed3ebSAl Viro return ret; 12729deed3ebSAl Viro } 12739deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 12749deed3ebSAl Viro 12759deed3ebSAl Viro /* 1276287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1277287548e4SAl Viro * we meet a managed dentry that would need blocking. 12789875cf80SDavid Howells */ 12799875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1280254cf582SAl Viro struct inode **inode, unsigned *seqp) 12819875cf80SDavid Howells { 1282ea936aebSAl Viro struct dentry *dentry = path->dentry; 1283ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1284ea936aebSAl Viro 1285ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1286ea936aebSAl Viro return true; 1287ea936aebSAl Viro 1288ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1289ea936aebSAl Viro return false; 1290ea936aebSAl Viro 129162a7375eSIan Kent for (;;) { 129262a7375eSIan Kent /* 129362a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 129462a7375eSIan Kent * that wants to block transit. 129562a7375eSIan Kent */ 1296ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1297ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1298ea936aebSAl Viro if (res) 1299ea936aebSAl Viro return res == -EISDIR; 1300ea936aebSAl Viro flags = dentry->d_flags; 1301b8faf035SNeilBrown } 130262a7375eSIan Kent 1303ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1304ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1305ea936aebSAl Viro if (mounted) { 1306c7105365SAl Viro path->mnt = &mounted->mnt; 1307ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1308a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1309ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1310ea936aebSAl Viro *inode = dentry->d_inode; 131159430262SLinus Torvalds /* 1312ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1313ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1314ea936aebSAl Viro * between mount hash removal and ->mnt_root 1315ea936aebSAl Viro * becoming unpinned. 131659430262SLinus Torvalds */ 1317ea936aebSAl Viro flags = dentry->d_flags; 1318ea936aebSAl Viro continue; 13199875cf80SDavid Howells } 1320ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1321ea936aebSAl Viro return false; 1322ea936aebSAl Viro } 1323ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1324ea936aebSAl Viro } 1325287548e4SAl Viro } 1326287548e4SAl Viro 1327db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1328db3c9adeSAl Viro struct path *path, struct inode **inode, 1329db3c9adeSAl Viro unsigned int *seqp) 1330bd7c4b50SAl Viro { 13319deed3ebSAl Viro bool jumped; 1332db3c9adeSAl Viro int ret; 1333bd7c4b50SAl Viro 1334db3c9adeSAl Viro path->mnt = nd->path.mnt; 1335db3c9adeSAl Viro path->dentry = dentry; 1336c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1337c153007bSAl Viro unsigned int seq = *seqp; 1338c153007bSAl Viro if (unlikely(!*inode)) 1339c153007bSAl Viro return -ENOENT; 1340c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 13419deed3ebSAl Viro return 0; 1342c153007bSAl Viro if (unlazy_child(nd, dentry, seq)) 1343c153007bSAl Viro return -ECHILD; 1344c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1345c153007bSAl Viro path->mnt = nd->path.mnt; 1346c153007bSAl Viro path->dentry = dentry; 1347c153007bSAl Viro } 13489deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 13499deed3ebSAl Viro if (jumped) { 13509deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 13519deed3ebSAl Viro ret = -EXDEV; 13529deed3ebSAl Viro else 13539deed3ebSAl Viro nd->flags |= LOOKUP_JUMPED; 13549deed3ebSAl Viro } 13559deed3ebSAl Viro if (unlikely(ret)) { 13569deed3ebSAl Viro dput(path->dentry); 13579deed3ebSAl Viro if (path->mnt != nd->path.mnt) 13589deed3ebSAl Viro mntput(path->mnt); 13599deed3ebSAl Viro } else { 1360bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1361bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1362bd7c4b50SAl Viro } 1363bd7c4b50SAl Viro return ret; 1364bd7c4b50SAl Viro } 1365bd7c4b50SAl Viro 136631e6b01fSNick Piggin static int follow_dotdot_rcu(struct nameidata *nd) 136731e6b01fSNick Piggin { 13684023bfc9SAl Viro struct inode *inode = nd->inode; 136931e6b01fSNick Piggin 137031e6b01fSNick Piggin while (1) { 1371adb21d2bSAleksa Sarai if (path_equal(&nd->path, &nd->root)) { 1372adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 1373adb21d2bSAleksa Sarai return -ECHILD; 137431e6b01fSNick Piggin break; 1375adb21d2bSAleksa Sarai } 137631e6b01fSNick Piggin if (nd->path.dentry != nd->path.mnt->mnt_root) { 137731e6b01fSNick Piggin struct dentry *old = nd->path.dentry; 137831e6b01fSNick Piggin struct dentry *parent = old->d_parent; 137931e6b01fSNick Piggin unsigned seq; 138031e6b01fSNick Piggin 13814023bfc9SAl Viro inode = parent->d_inode; 138231e6b01fSNick Piggin seq = read_seqcount_begin(&parent->d_seq); 1383aed434adSAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 1384aed434adSAl Viro return -ECHILD; 138531e6b01fSNick Piggin nd->path.dentry = parent; 138631e6b01fSNick Piggin nd->seq = seq; 13876b03f7edSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 13882b98149cSAleksa Sarai return -ECHILD; 138931e6b01fSNick Piggin break; 1390aed434adSAl Viro } else { 1391aed434adSAl Viro struct mount *mnt = real_mount(nd->path.mnt); 1392aed434adSAl Viro struct mount *mparent = mnt->mnt_parent; 1393aed434adSAl Viro struct dentry *mountpoint = mnt->mnt_mountpoint; 1394aed434adSAl Viro struct inode *inode2 = mountpoint->d_inode; 1395aed434adSAl Viro unsigned seq = read_seqcount_begin(&mountpoint->d_seq); 1396aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1397aed434adSAl Viro return -ECHILD; 1398aed434adSAl Viro if (&mparent->mnt == nd->path.mnt) 139931e6b01fSNick Piggin break; 140072ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 140172ba2929SAleksa Sarai return -ECHILD; 1402aed434adSAl Viro /* we know that mountpoint was pinned */ 1403aed434adSAl Viro nd->path.dentry = mountpoint; 1404aed434adSAl Viro nd->path.mnt = &mparent->mnt; 1405aed434adSAl Viro inode = inode2; 1406aed434adSAl Viro nd->seq = seq; 140731e6b01fSNick Piggin } 1408aed434adSAl Viro } 1409aed434adSAl Viro while (unlikely(d_mountpoint(nd->path.dentry))) { 1410b37199e6SAl Viro struct mount *mounted; 1411b37199e6SAl Viro mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry); 1412aed434adSAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1413aed434adSAl Viro return -ECHILD; 1414b37199e6SAl Viro if (!mounted) 1415b37199e6SAl Viro break; 141672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 141772ba2929SAleksa Sarai return -ECHILD; 1418b37199e6SAl Viro nd->path.mnt = &mounted->mnt; 1419b37199e6SAl Viro nd->path.dentry = mounted->mnt.mnt_root; 14204023bfc9SAl Viro inode = nd->path.dentry->d_inode; 1421b37199e6SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 1422b37199e6SAl Viro } 14234023bfc9SAl Viro nd->inode = inode; 142431e6b01fSNick Piggin return 0; 142531e6b01fSNick Piggin } 142631e6b01fSNick Piggin 14279875cf80SDavid Howells /* 14289875cf80SDavid Howells * Skip to top of mountpoint pile in refwalk mode for follow_dotdot() 14299875cf80SDavid Howells */ 14309875cf80SDavid Howells static void follow_mount(struct path *path) 14319875cf80SDavid Howells { 14329875cf80SDavid Howells while (d_mountpoint(path->dentry)) { 14339875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 14349875cf80SDavid Howells if (!mounted) 14359875cf80SDavid Howells break; 14369875cf80SDavid Howells dput(path->dentry); 14379875cf80SDavid Howells mntput(path->mnt); 14389875cf80SDavid Howells path->mnt = mounted; 14399875cf80SDavid Howells path->dentry = dget(mounted->mnt_root); 14409875cf80SDavid Howells } 14419875cf80SDavid Howells } 14429875cf80SDavid Howells 144363b27720SAl Viro static struct dentry *path_parent_directory(struct path *path) 1444eedf265aSEric W. Biederman { 1445eedf265aSEric W. Biederman /* rare case of legitimate dget_parent()... */ 144663b27720SAl Viro struct dentry *parent = dget_parent(path->dentry); 144763b27720SAl Viro 144863b27720SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 144963b27720SAl Viro dput(parent); 145063b27720SAl Viro parent = NULL; 145163b27720SAl Viro } 145263b27720SAl Viro return parent; 1453eedf265aSEric W. Biederman } 1454eedf265aSEric W. Biederman 1455397d425dSEric W. Biederman static int follow_dotdot(struct nameidata *nd) 14561da177e4SLinus Torvalds { 145763b27720SAl Viro struct dentry *parent; 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) { 146563b27720SAl Viro parent = path_parent_directory(&nd->path); 146663b27720SAl Viro if (!parent) 146763b27720SAl Viro return -ENOENT; 146863b27720SAl Viro dput(nd->path.dentry); 146963b27720SAl Viro nd->path.dentry = parent; 14701da177e4SLinus Torvalds break; 14711da177e4SLinus Torvalds } 14723088dd70SAl Viro if (!follow_up(&nd->path)) 14731da177e4SLinus Torvalds break; 147472ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 147572ba2929SAleksa Sarai return -EXDEV; 14761da177e4SLinus Torvalds } 147779ed0226SAl Viro follow_mount(&nd->path); 147831e6b01fSNick Piggin nd->inode = nd->path.dentry->d_inode; 1479397d425dSEric W. Biederman return 0; 14801da177e4SLinus Torvalds } 14811da177e4SLinus Torvalds 14821da177e4SLinus Torvalds /* 1483f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1484f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1485baa03890SNick Piggin */ 1486e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1487e3c13928SAl Viro struct dentry *dir, 14886c51e513SAl Viro unsigned int flags) 1489baa03890SNick Piggin { 1490a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1491bad61189SMiklos Szeredi if (dentry) { 1492a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1493bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 149474ff0ffcSAl Viro if (!error) 14955542aa2fSEric W. Biederman d_invalidate(dentry); 1496bad61189SMiklos Szeredi dput(dentry); 149774ff0ffcSAl Viro return ERR_PTR(error); 1498bad61189SMiklos Szeredi } 1499bad61189SMiklos Szeredi } 1500baa03890SNick Piggin return dentry; 1501baa03890SNick Piggin } 1502baa03890SNick Piggin 1503baa03890SNick Piggin /* 1504a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1505a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1506a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1507a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1508a03ece5fSAl Viro * at all. 150944396f4bSJosef Bacik */ 1510a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1511a03ece5fSAl Viro struct dentry *base, unsigned int flags) 151244396f4bSJosef Bacik { 1513a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 151444396f4bSJosef Bacik struct dentry *old; 1515a03ece5fSAl Viro struct inode *dir = base->d_inode; 1516a03ece5fSAl Viro 1517a03ece5fSAl Viro if (dentry) 1518a03ece5fSAl Viro return dentry; 151944396f4bSJosef Bacik 152044396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1521a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 152244396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1523a03ece5fSAl Viro 1524a03ece5fSAl Viro dentry = d_alloc(base, name); 1525a03ece5fSAl Viro if (unlikely(!dentry)) 1526a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 152744396f4bSJosef Bacik 152872bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 152944396f4bSJosef Bacik if (unlikely(old)) { 153044396f4bSJosef Bacik dput(dentry); 153144396f4bSJosef Bacik dentry = old; 153244396f4bSJosef Bacik } 153344396f4bSJosef Bacik return dentry; 153444396f4bSJosef Bacik } 153544396f4bSJosef Bacik 153620e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 153720e34357SAl Viro struct inode **inode, 1538254cf582SAl Viro unsigned *seqp) 15391da177e4SLinus Torvalds { 154031e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15415a18fff2SAl Viro int status = 1; 15429875cf80SDavid Howells 15433cac260aSAl Viro /* 1544b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15455d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15465d0f49c1SAl Viro * going to fall back to non-racy lookup. 1547b04f784eSNick Piggin */ 154831e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 154931e6b01fSNick Piggin unsigned seq; 1550da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15515d0f49c1SAl Viro if (unlikely(!dentry)) { 15524675ac39SAl Viro if (unlazy_walk(nd)) 155320e34357SAl Viro return ERR_PTR(-ECHILD); 155420e34357SAl Viro return NULL; 15555d0f49c1SAl Viro } 15565a18fff2SAl Viro 155712f8ad4bSLinus Torvalds /* 155812f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 155912f8ad4bSLinus Torvalds * the dentry name information from lookup. 156012f8ad4bSLinus Torvalds */ 156163afdfc7SDavid Howells *inode = d_backing_inode(dentry); 15625d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 156320e34357SAl Viro return ERR_PTR(-ECHILD); 156412f8ad4bSLinus Torvalds 156512f8ad4bSLinus Torvalds /* 156612f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 156712f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 156812f8ad4bSLinus Torvalds * 156912f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 157012f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 157112f8ad4bSLinus Torvalds */ 15725d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 157320e34357SAl Viro return ERR_PTR(-ECHILD); 15745a18fff2SAl Viro 1575254cf582SAl Viro *seqp = seq; 15764ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1577c153007bSAl Viro if (likely(status > 0)) 157820e34357SAl Viro return dentry; 15794675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 158020e34357SAl Viro return ERR_PTR(-ECHILD); 1581209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1582209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1583209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 15845a18fff2SAl Viro } else { 1585e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 158681e6f520SAl Viro if (unlikely(!dentry)) 158720e34357SAl Viro return NULL; 15884ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15895d0f49c1SAl Viro } 15905a18fff2SAl Viro if (unlikely(status <= 0)) { 1591e9742b53SAl Viro if (!status) 15925d0f49c1SAl Viro d_invalidate(dentry); 15935a18fff2SAl Viro dput(dentry); 159420e34357SAl Viro return ERR_PTR(status); 15955a18fff2SAl Viro } 159620e34357SAl Viro return dentry; 1597697f514dSMiklos Szeredi } 1598697f514dSMiklos Szeredi 1599697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 160088d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1601e3c13928SAl Viro struct dentry *dir, 1602e3c13928SAl Viro unsigned int flags) 1603697f514dSMiklos Szeredi { 160488d8331aSAl Viro struct dentry *dentry, *old; 16051936386eSAl Viro struct inode *inode = dir->d_inode; 1606d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16071936386eSAl Viro 16081936386eSAl Viro /* Don't go there if it's already dead */ 160994bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 161088d8331aSAl Viro return ERR_PTR(-ENOENT); 161194bdd655SAl Viro again: 1612d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 161394bdd655SAl Viro if (IS_ERR(dentry)) 161488d8331aSAl Viro return dentry; 161594bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1616949a852eSAl Viro int error = d_revalidate(dentry, flags); 1617949a852eSAl Viro if (unlikely(error <= 0)) { 161894bdd655SAl Viro if (!error) { 1619949a852eSAl Viro d_invalidate(dentry); 1620949a852eSAl Viro dput(dentry); 162194bdd655SAl Viro goto again; 162294bdd655SAl Viro } 162394bdd655SAl Viro dput(dentry); 1624949a852eSAl Viro dentry = ERR_PTR(error); 1625949a852eSAl Viro } 162694bdd655SAl Viro } else { 16271936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 162885c7f810SAl Viro d_lookup_done(dentry); 16291936386eSAl Viro if (unlikely(old)) { 16301936386eSAl Viro dput(dentry); 16311936386eSAl Viro dentry = old; 1632949a852eSAl Viro } 1633949a852eSAl Viro } 1634e3c13928SAl Viro return dentry; 16351da177e4SLinus Torvalds } 16361da177e4SLinus Torvalds 163788d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 163888d8331aSAl Viro struct dentry *dir, 163988d8331aSAl Viro unsigned int flags) 164088d8331aSAl Viro { 164188d8331aSAl Viro struct inode *inode = dir->d_inode; 164288d8331aSAl Viro struct dentry *res; 164388d8331aSAl Viro inode_lock_shared(inode); 164488d8331aSAl Viro res = __lookup_slow(name, dir, flags); 164588d8331aSAl Viro inode_unlock_shared(inode); 164688d8331aSAl Viro return res; 164788d8331aSAl Viro } 164888d8331aSAl Viro 164952094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 165052094c8aSAl Viro { 165152094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16524ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 165352094c8aSAl Viro if (err != -ECHILD) 165452094c8aSAl Viro return err; 16554675ac39SAl Viro if (unlazy_walk(nd)) 165652094c8aSAl Viro return -ECHILD; 165752094c8aSAl Viro } 16584ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 165952094c8aSAl Viro } 166052094c8aSAl Viro 16619856fa1bSAl Viro static inline int handle_dots(struct nameidata *nd, int type) 16629856fa1bSAl Viro { 16639856fa1bSAl Viro if (type == LAST_DOTDOT) { 1664740a1678SAleksa Sarai int error = 0; 1665740a1678SAleksa Sarai 1666740a1678SAleksa Sarai if (!nd->root.mnt) { 1667740a1678SAleksa Sarai error = set_root(nd); 1668740a1678SAleksa Sarai if (error) 1669740a1678SAleksa Sarai return error; 1670740a1678SAleksa Sarai } 1671ab87f9a5SAleksa Sarai if (nd->flags & LOOKUP_RCU) 1672ab87f9a5SAleksa Sarai error = follow_dotdot_rcu(nd); 1673ab87f9a5SAleksa Sarai else 1674ab87f9a5SAleksa Sarai error = follow_dotdot(nd); 1675ab87f9a5SAleksa Sarai if (error) 1676ab87f9a5SAleksa Sarai return error; 1677ab87f9a5SAleksa Sarai 1678ab87f9a5SAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1679ab87f9a5SAleksa Sarai /* 1680ab87f9a5SAleksa Sarai * If there was a racing rename or mount along our 1681ab87f9a5SAleksa Sarai * path, then we can't be sure that ".." hasn't jumped 1682ab87f9a5SAleksa Sarai * above nd->root (and so userspace should retry or use 1683ab87f9a5SAleksa Sarai * some fallback). 1684ab87f9a5SAleksa Sarai */ 1685ab87f9a5SAleksa Sarai smp_rmb(); 1686ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 1687ab87f9a5SAleksa Sarai return -EAGAIN; 1688ab87f9a5SAleksa Sarai if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 1689ab87f9a5SAleksa Sarai return -EAGAIN; 1690ab87f9a5SAleksa Sarai } 16919856fa1bSAl Viro } 16929856fa1bSAl Viro return 0; 16939856fa1bSAl Viro } 16949856fa1bSAl Viro 1695b1a81972SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 1696b1a81972SAl Viro 169706708adbSAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 1698b1a81972SAl Viro struct inode *inode, unsigned seq, int flags) 1699d63ff28fSAl Viro { 17001cf2665bSAl Viro struct saved *last; 1701ad6cc4c3SAl Viro const char *res; 1702ad6cc4c3SAl Viro int error; 1703ad6cc4c3SAl Viro 1704756daf26SNeilBrown if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) { 1705626de996SAl Viro path_to_nameidata(link, nd); 170606708adbSAl Viro return ERR_PTR(-ELOOP); 1707626de996SAl Viro } 1708bc40aee0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1709cd179f44SAl Viro if (link->mnt == nd->path.mnt) 1710cd179f44SAl Viro mntget(link->mnt); 17117973387aSAl Viro } 1712626de996SAl Viro error = nd_alloc_stack(nd); 1713626de996SAl Viro if (unlikely(error)) { 1714bc40aee0SAl Viro if (error == -ECHILD) { 1715ad1633a1SAl Viro if (unlikely(!legitimize_path(nd, link, seq))) { 1716ad1633a1SAl Viro drop_links(nd); 1717ad1633a1SAl Viro nd->depth = 0; 1718ad1633a1SAl Viro nd->flags &= ~LOOKUP_RCU; 1719ad1633a1SAl Viro nd->path.mnt = NULL; 1720ad1633a1SAl Viro nd->path.dentry = NULL; 1721ad1633a1SAl Viro rcu_read_unlock(); 17224675ac39SAl Viro } else if (likely(unlazy_walk(nd)) == 0) 1723bc40aee0SAl Viro error = nd_alloc_stack(nd); 1724bc40aee0SAl Viro } 1725bc40aee0SAl Viro if (error) { 1726cd179f44SAl Viro path_put(link); 172706708adbSAl Viro return ERR_PTR(error); 1728626de996SAl Viro } 1729bc40aee0SAl Viro } 1730626de996SAl Viro 1731ab104923SAl Viro last = nd->stack + nd->depth++; 17321cf2665bSAl Viro last->link = *link; 1733fceef393SAl Viro clear_delayed_call(&last->done); 17340450b2d1SAl Viro last->seq = seq; 1735ad6cc4c3SAl Viro 1736b1a81972SAl Viro if (flags & WALK_TRAILING) { 1737ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1738ad6cc4c3SAl Viro if (unlikely(error)) 1739ad6cc4c3SAl Viro return ERR_PTR(error); 1740ad6cc4c3SAl Viro } 1741ad6cc4c3SAl Viro 1742ad6cc4c3SAl Viro if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) 1743ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1744ad6cc4c3SAl Viro 1745ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1746ad6cc4c3SAl Viro touch_atime(&last->link); 1747ad6cc4c3SAl Viro cond_resched(); 1748ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1749ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1750ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1751ad6cc4c3SAl Viro touch_atime(&last->link); 1752ad6cc4c3SAl Viro } 1753ad6cc4c3SAl Viro 1754ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1755ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1756ad6cc4c3SAl Viro if (unlikely(error)) 1757ad6cc4c3SAl Viro return ERR_PTR(error); 1758ad6cc4c3SAl Viro 1759ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1760ad6cc4c3SAl Viro if (!res) { 1761ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1762ad6cc4c3SAl Viro struct delayed_call *); 1763ad6cc4c3SAl Viro get = inode->i_op->get_link; 1764ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1765ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1766ad6cc4c3SAl Viro if (res == ERR_PTR(-ECHILD)) { 1767ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1768ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1769ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1770ad6cc4c3SAl Viro } 1771ad6cc4c3SAl Viro } else { 1772ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1773ad6cc4c3SAl Viro } 1774ad6cc4c3SAl Viro if (!res) 1775ad6cc4c3SAl Viro goto all_done; 1776ad6cc4c3SAl Viro if (IS_ERR(res)) 1777ad6cc4c3SAl Viro return res; 1778ad6cc4c3SAl Viro } 1779ad6cc4c3SAl Viro if (*res == '/') { 1780ad6cc4c3SAl Viro error = nd_jump_root(nd); 1781ad6cc4c3SAl Viro if (unlikely(error)) 1782ad6cc4c3SAl Viro return ERR_PTR(error); 1783ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1784ad6cc4c3SAl Viro ; 1785ad6cc4c3SAl Viro } 1786ad6cc4c3SAl Viro if (*res) 1787ad6cc4c3SAl Viro return res; 1788ad6cc4c3SAl Viro all_done: // pure jump 1789ad6cc4c3SAl Viro put_link(nd); 1790ad6cc4c3SAl Viro return NULL; 1791d63ff28fSAl Viro } 1792d63ff28fSAl Viro 17933ddcd056SLinus Torvalds /* 17943ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17953ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17963ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17973ddcd056SLinus Torvalds * for the common case. 17983ddcd056SLinus Torvalds */ 1799b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1800cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 18013ddcd056SLinus Torvalds { 1802cbae4d12SAl Viro struct path path; 1803cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1804cbae4d12SAl Viro 1805cbae4d12SAl Viro if (err < 0) 1806b0417d2cSAl Viro return ERR_PTR(err); 1807cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18088c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1809aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18108f64fb1cSAl Viro /* not a symlink or should not follow */ 1811cbae4d12SAl Viro path_to_nameidata(&path, nd); 18128f64fb1cSAl Viro nd->inode = inode; 18138f64fb1cSAl Viro nd->seq = seq; 1814b0417d2cSAl Viro return NULL; 18158f64fb1cSAl Viro } 1816a7f77542SAl Viro /* make sure that d_is_symlink above matches inode */ 1817a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 1818cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1819b0417d2cSAl Viro return ERR_PTR(-ECHILD); 1820a7f77542SAl Viro } 1821b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18223ddcd056SLinus Torvalds } 18233ddcd056SLinus Torvalds 182492d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1825ce57dfc1SAl Viro { 1826db3c9adeSAl Viro struct dentry *dentry; 1827ce57dfc1SAl Viro struct inode *inode; 1828254cf582SAl Viro unsigned seq; 1829ce57dfc1SAl Viro int err; 1830ce57dfc1SAl Viro /* 1831ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1832ce57dfc1SAl Viro * to be able to know about the current root directory and 1833ce57dfc1SAl Viro * parent relationships. 1834ce57dfc1SAl Viro */ 18354693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18361c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18374693a547SAl Viro put_link(nd); 183856676ec3SAl Viro err = handle_dots(nd, nd->last_type); 183992d27016SAl Viro return ERR_PTR(err); 18404693a547SAl Viro } 184120e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 184220e34357SAl Viro if (IS_ERR(dentry)) 184392d27016SAl Viro return ERR_CAST(dentry); 184420e34357SAl Viro if (unlikely(!dentry)) { 1845db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1846db3c9adeSAl Viro if (IS_ERR(dentry)) 184792d27016SAl Viro return ERR_CAST(dentry); 184820e34357SAl Viro } 184956676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 185056676ec3SAl Viro put_link(nd); 1851b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1852ce57dfc1SAl Viro } 1853ce57dfc1SAl Viro 18541da177e4SLinus Torvalds /* 1855bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1856bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1857bfcfaa77SLinus Torvalds * 1858bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1859bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1860bfcfaa77SLinus Torvalds * fast. 1861bfcfaa77SLinus Torvalds * 1862bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1863bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1864bfcfaa77SLinus Torvalds * crossing operation. 1865bfcfaa77SLinus Torvalds * 1866bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1867bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1868bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1869bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1870bfcfaa77SLinus Torvalds */ 1871bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1872bfcfaa77SLinus Torvalds 1873f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1874bfcfaa77SLinus Torvalds 1875468a9428SGeorge Spelvin #ifdef HASH_MIX 1876bfcfaa77SLinus Torvalds 1877468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1878468a9428SGeorge Spelvin 1879468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18802a18da7aSGeorge Spelvin /* 18812a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18822a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18832a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18842a18da7aSGeorge Spelvin * and no temporaries. 18852a18da7aSGeorge Spelvin * 18862a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 18872a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 18882a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 18892a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 18902a18da7aSGeorge Spelvin * 18912a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 18922a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 18932a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 18942a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 18952a18da7aSGeorge Spelvin * 18962a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 18972a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 18982a18da7aSGeorge Spelvin * toggling any given output bit. 18992a18da7aSGeorge Spelvin * 19002a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 19012a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19022a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 19032a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 19042a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 19052a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 19062a18da7aSGeorge Spelvin * Perfect: 8192 258048 19072a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 19082a18da7aSGeorge Spelvin */ 19092a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19102a18da7aSGeorge Spelvin ( x ^= (a), \ 19112a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 19122a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 19132a18da7aSGeorge Spelvin y *= 9 ) 1914bfcfaa77SLinus Torvalds 19150fed3ac8SGeorge Spelvin /* 19162a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 19172a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 19182a18da7aSGeorge Spelvin * work done before the hash value is used. 19190fed3ac8SGeorge Spelvin */ 19202a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19210fed3ac8SGeorge Spelvin { 19222a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 19232a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19242a18da7aSGeorge Spelvin return y >> 32; 19250fed3ac8SGeorge Spelvin } 19260fed3ac8SGeorge Spelvin 1927bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1928bfcfaa77SLinus Torvalds 19292a18da7aSGeorge Spelvin /* 19302a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19312a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19322a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19332a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19342a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19352a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19362a18da7aSGeorge Spelvin * Perfect: 2048 31744 19372a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19382a18da7aSGeorge Spelvin */ 19392a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19402a18da7aSGeorge Spelvin ( x ^= (a), \ 19412a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19422a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19432a18da7aSGeorge Spelvin y *= 9 ) 1944bfcfaa77SLinus Torvalds 19452a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19460fed3ac8SGeorge Spelvin { 19472a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19482a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19490fed3ac8SGeorge Spelvin } 19500fed3ac8SGeorge Spelvin 1951bfcfaa77SLinus Torvalds #endif 1952bfcfaa77SLinus Torvalds 19532a18da7aSGeorge Spelvin /* 19542a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19552a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19562a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19572a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19582a18da7aSGeorge Spelvin * finds the delimiter after the name. 19592a18da7aSGeorge Spelvin */ 19608387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1961bfcfaa77SLinus Torvalds { 19628387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1963bfcfaa77SLinus Torvalds 1964bfcfaa77SLinus Torvalds for (;;) { 1965fcfd2fbfSGeorge Spelvin if (!len) 1966fcfd2fbfSGeorge Spelvin goto done; 1967e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1968bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1969bfcfaa77SLinus Torvalds break; 19702a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1971bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1972bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1973bfcfaa77SLinus Torvalds } 19742a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1975bfcfaa77SLinus Torvalds done: 19762a18da7aSGeorge Spelvin return fold_hash(x, y); 1977bfcfaa77SLinus Torvalds } 1978bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1979bfcfaa77SLinus Torvalds 1980fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19818387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1982fcfd2fbfSGeorge Spelvin { 19838387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19848387ff25SLinus Torvalds unsigned long adata, mask, len; 1985fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1986fcfd2fbfSGeorge Spelvin 19878387ff25SLinus Torvalds len = 0; 19888387ff25SLinus Torvalds goto inside; 19898387ff25SLinus Torvalds 1990fcfd2fbfSGeorge Spelvin do { 19912a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1992fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 19938387ff25SLinus Torvalds inside: 1994fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 1995fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 1996fcfd2fbfSGeorge Spelvin 1997fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 1998fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 19992a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2000fcfd2fbfSGeorge Spelvin 20012a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2002fcfd2fbfSGeorge Spelvin } 2003fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2004fcfd2fbfSGeorge Spelvin 2005bfcfaa77SLinus Torvalds /* 2006bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2007d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2008bfcfaa77SLinus Torvalds */ 20098387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2010bfcfaa77SLinus Torvalds { 20118387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 20128387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 201336126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2014bfcfaa77SLinus Torvalds 20158387ff25SLinus Torvalds len = 0; 20168387ff25SLinus Torvalds goto inside; 20178387ff25SLinus Torvalds 2018bfcfaa77SLinus Torvalds do { 20192a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2020bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 20218387ff25SLinus Torvalds inside: 2022e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 202336126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 202436126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2025bfcfaa77SLinus Torvalds 202636126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 202736126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 202836126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20292a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 203036126f8fSLinus Torvalds 20312a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2032bfcfaa77SLinus Torvalds } 2033bfcfaa77SLinus Torvalds 20342a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2035bfcfaa77SLinus Torvalds 2036fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20378387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20380145acc2SLinus Torvalds { 20398387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20400145acc2SLinus Torvalds while (len--) 2041fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20420145acc2SLinus Torvalds return end_name_hash(hash); 20430145acc2SLinus Torvalds } 2044ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20450145acc2SLinus Torvalds 2046fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20478387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2048fcfd2fbfSGeorge Spelvin { 20498387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2050fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2051fcfd2fbfSGeorge Spelvin 2052fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2053e0ab7af9SGeorge Spelvin while (c) { 2054fcfd2fbfSGeorge Spelvin len++; 2055fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2056fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2057e0ab7af9SGeorge Spelvin } 2058fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2059fcfd2fbfSGeorge Spelvin } 2060f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2061fcfd2fbfSGeorge Spelvin 20623ddcd056SLinus Torvalds /* 2063200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2064200e9ef7SLinus Torvalds * one character. 2065200e9ef7SLinus Torvalds */ 20668387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2067200e9ef7SLinus Torvalds { 20688387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2069200e9ef7SLinus Torvalds unsigned long len = 0, c; 2070200e9ef7SLinus Torvalds 2071200e9ef7SLinus Torvalds c = (unsigned char)*name; 2072200e9ef7SLinus Torvalds do { 2073200e9ef7SLinus Torvalds len++; 2074200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2075200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2076200e9ef7SLinus Torvalds } while (c && c != '/'); 2077d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2078200e9ef7SLinus Torvalds } 2079200e9ef7SLinus Torvalds 2080bfcfaa77SLinus Torvalds #endif 2081bfcfaa77SLinus Torvalds 2082200e9ef7SLinus Torvalds /* 20831da177e4SLinus Torvalds * Name resolution. 2084ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2085ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20861da177e4SLinus Torvalds * 2087ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2088ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 20891da177e4SLinus Torvalds */ 20906de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 20911da177e4SLinus Torvalds { 2092d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 20931da177e4SLinus Torvalds int err; 20941da177e4SLinus Torvalds 2095b4c03536SAl Viro nd->last_type = LAST_ROOT; 2096c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 20979b5858e9SAl Viro if (IS_ERR(name)) 20989b5858e9SAl Viro return PTR_ERR(name); 20991da177e4SLinus Torvalds while (*name=='/') 21001da177e4SLinus Torvalds name++; 21011da177e4SLinus Torvalds if (!*name) 21029e18f10aSAl Viro return 0; 21031da177e4SLinus Torvalds 21041da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 21051da177e4SLinus Torvalds for(;;) { 210692d27016SAl Viro const char *link; 2107d6bb3e90SLinus Torvalds u64 hash_len; 2108fe479a58SAl Viro int type; 21091da177e4SLinus Torvalds 211052094c8aSAl Viro err = may_lookup(nd); 21111da177e4SLinus Torvalds if (err) 21123595e234SAl Viro return err; 21131da177e4SLinus Torvalds 21148387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 21151da177e4SLinus Torvalds 2116fe479a58SAl Viro type = LAST_NORM; 2117d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2118fe479a58SAl Viro case 2: 2119200e9ef7SLinus Torvalds if (name[1] == '.') { 2120fe479a58SAl Viro type = LAST_DOTDOT; 212116c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 212216c2cd71SAl Viro } 2123fe479a58SAl Viro break; 2124fe479a58SAl Viro case 1: 2125fe479a58SAl Viro type = LAST_DOT; 2126fe479a58SAl Viro } 21275a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21285a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 212916c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21305a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2131a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2132da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21335a202bcdSAl Viro if (err < 0) 21343595e234SAl Viro return err; 2135d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2136d6bb3e90SLinus Torvalds name = this.name; 21375a202bcdSAl Viro } 21385a202bcdSAl Viro } 2139fe479a58SAl Viro 2140d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2141d6bb3e90SLinus Torvalds nd->last.name = name; 21425f4a6a69SAl Viro nd->last_type = type; 21435f4a6a69SAl Viro 2144d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2145d6bb3e90SLinus Torvalds if (!*name) 2146bdf6cbf1SAl Viro goto OK; 2147200e9ef7SLinus Torvalds /* 2148200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2149200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2150200e9ef7SLinus Torvalds */ 2151200e9ef7SLinus Torvalds do { 2152d6bb3e90SLinus Torvalds name++; 2153d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21548620c238SAl Viro if (unlikely(!*name)) { 21558620c238SAl Viro OK: 2156d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2157c108837eSAl Viro if (!depth) { 2158c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 21598620c238SAl Viro return 0; 2160c108837eSAl Viro } 21618620c238SAl Viro /* last component of nested symlink */ 2162d8d4611aSAl Viro name = nd->stack[--depth].name; 21638c4efe22SAl Viro link = walk_component(nd, 0); 21641c4ff1a8SAl Viro } else { 21651c4ff1a8SAl Viro /* not the last component */ 21668c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 21678620c238SAl Viro } 216892d27016SAl Viro if (unlikely(link)) { 216992d27016SAl Viro if (IS_ERR(link)) 217092d27016SAl Viro return PTR_ERR(link); 217192d27016SAl Viro /* a symlink to follow */ 2172d8d4611aSAl Viro nd->stack[depth++].name = name; 217392d27016SAl Viro name = link; 21749e18f10aSAl Viro continue; 217548c8b0c5SAl Viro } 217697242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 217797242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21784675ac39SAl Viro if (unlazy_walk(nd)) 217997242f99SAl Viro return -ECHILD; 218097242f99SAl Viro } 21813595e234SAl Viro return -ENOTDIR; 21825f4a6a69SAl Viro } 2183ce57dfc1SAl Viro } 218497242f99SAl Viro } 21851da177e4SLinus Torvalds 2186edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2187c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 218831e6b01fSNick Piggin { 2189740a1678SAleksa Sarai int error; 2190c8a53ee5SAl Viro const char *s = nd->name->name; 219131e6b01fSNick Piggin 2192c0eb027eSLinus Torvalds if (!*s) 2193c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2194edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2195edc2b1daSAl Viro rcu_read_lock(); 2196c0eb027eSLinus Torvalds 2197c108837eSAl Viro nd->flags = flags | LOOKUP_JUMPED; 219831e6b01fSNick Piggin nd->depth = 0; 2199ab87f9a5SAleksa Sarai 2200ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2201ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2202ab87f9a5SAleksa Sarai smp_rmb(); 2203ab87f9a5SAleksa Sarai 22045b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2205b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2206b18825a7SDavid Howells struct inode *inode = root->d_inode; 220793893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2208368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 22095b6ca027SAl Viro nd->path = nd->root; 22105b6ca027SAl Viro nd->inode = inode; 22115b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2212ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22138f47a016SAl Viro nd->root_seq = nd->seq; 22145b6ca027SAl Viro } else { 22155b6ca027SAl Viro path_get(&nd->path); 22165b6ca027SAl Viro } 2217368ee9baSAl Viro return s; 22185b6ca027SAl Viro } 22195b6ca027SAl Viro 222031e6b01fSNick Piggin nd->root.mnt = NULL; 2221248fb5b9SAl Viro nd->path.mnt = NULL; 2222248fb5b9SAl Viro nd->path.dentry = NULL; 222331e6b01fSNick Piggin 22248db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 22258db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2226740a1678SAleksa Sarai error = nd_jump_root(nd); 2227740a1678SAleksa Sarai if (unlikely(error)) 2228740a1678SAleksa Sarai return ERR_PTR(error); 2229ef55d917SAl Viro return s; 22308db52c7eSAleksa Sarai } 22318db52c7eSAleksa Sarai 22328db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 22338db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2234e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 223531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2236c28cc364SNick Piggin unsigned seq; 223731e6b01fSNick Piggin 2238c28cc364SNick Piggin do { 2239c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 224031e6b01fSNick Piggin nd->path = fs->pwd; 2241ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2242c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2243c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2244e41f7d4eSAl Viro } else { 2245e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2246ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2247e41f7d4eSAl Viro } 224831e6b01fSNick Piggin } else { 2249582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2250c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 225131e6b01fSNick Piggin struct dentry *dentry; 225231e6b01fSNick Piggin 22532903ff01SAl Viro if (!f.file) 2254368ee9baSAl Viro return ERR_PTR(-EBADF); 225531e6b01fSNick Piggin 22562903ff01SAl Viro dentry = f.file->f_path.dentry; 225731e6b01fSNick Piggin 2258edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22592903ff01SAl Viro fdput(f); 2260368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2261f52e0c11SAl Viro } 22622903ff01SAl Viro 22632903ff01SAl Viro nd->path = f.file->f_path; 2264e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 226534a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 226634a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22675590ff0dSUlrich Drepper } else { 22682903ff01SAl Viro path_get(&nd->path); 226934a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22701da177e4SLinus Torvalds } 227134a26b99SAl Viro fdput(f); 2272e41f7d4eSAl Viro } 22738db52c7eSAleksa Sarai 2274adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2275adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2276adb21d2bSAleksa Sarai nd->root = nd->path; 2277adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2278adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2279adb21d2bSAleksa Sarai } else { 2280adb21d2bSAleksa Sarai path_get(&nd->root); 2281adb21d2bSAleksa Sarai nd->flags |= LOOKUP_ROOT_GRABBED; 2282adb21d2bSAleksa Sarai } 2283adb21d2bSAleksa Sarai } 2284adb21d2bSAleksa Sarai return s; 22859b4a9b14SAl Viro } 22869b4a9b14SAl Viro 22871ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 228895fa25d9SAl Viro { 2289bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2290bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2291bd92d7feSAl Viro 2292c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2293bd92d7feSAl Viro } 2294bd92d7feSAl Viro 22954f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 22964f757f3cSAl Viro { 2297c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2298db3c9adeSAl Viro dget(nd->path.dentry); 2299b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2300b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 23014f757f3cSAl Viro } 23024f757f3cSAl Viro 23039b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2304c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23059b4a9b14SAl Viro { 2306c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2307bd92d7feSAl Viro int err; 230831e6b01fSNick Piggin 23099b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23104f757f3cSAl Viro err = handle_lookup_down(nd); 23115f336e72SAl Viro if (unlikely(err < 0)) 23125f336e72SAl Viro s = ERR_PTR(err); 23134f757f3cSAl Viro } 23144f757f3cSAl Viro 23151ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 23161ccac622SAl Viro (s = lookup_last(nd)) != NULL) 23171ccac622SAl Viro ; 23189f1fafeeSAl Viro if (!err) 23199f1fafeeSAl Viro err = complete_walk(nd); 2320bd92d7feSAl Viro 2321deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2322deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2323bd23a539SAl Viro err = -ENOTDIR; 2324161aff1dSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 2325161aff1dSAl Viro err = handle_lookup_down(nd); 2326161aff1dSAl Viro nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... 2327161aff1dSAl Viro } 2328625b6d10SAl Viro if (!err) { 2329625b6d10SAl Viro *path = nd->path; 2330625b6d10SAl Viro nd->path.mnt = NULL; 2331625b6d10SAl Viro nd->path.dentry = NULL; 2332625b6d10SAl Viro } 2333deb106c6SAl Viro terminate_walk(nd); 2334bd92d7feSAl Viro return err; 233531e6b01fSNick Piggin } 233631e6b01fSNick Piggin 233731d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23389ad1aaa6SAl Viro struct path *path, struct path *root) 2339873f1eedSJeff Layton { 2340894bc8c4SAl Viro int retval; 23419883d185SAl Viro struct nameidata nd; 2342abc9f5beSAl Viro if (IS_ERR(name)) 2343abc9f5beSAl Viro return PTR_ERR(name); 23449ad1aaa6SAl Viro if (unlikely(root)) { 23459ad1aaa6SAl Viro nd.root = *root; 23469ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23479ad1aaa6SAl Viro } 23489883d185SAl Viro set_nameidata(&nd, dfd, name); 2349c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2350873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2351c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2352873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2353c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2354873f1eedSJeff Layton 2355873f1eedSJeff Layton if (likely(!retval)) 2356161aff1dSAl Viro audit_inode(name, path->dentry, 2357161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 23589883d185SAl Viro restore_nameidata(); 2359e4bd1c1aSAl Viro putname(name); 2360873f1eedSJeff Layton return retval; 2361873f1eedSJeff Layton } 2362873f1eedSJeff Layton 23638bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2364c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2365391172c4SAl Viro struct path *parent) 23668bcb77faSAl Viro { 2367c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23689b5858e9SAl Viro int err = link_path_walk(s, nd); 23698bcb77faSAl Viro if (!err) 23708bcb77faSAl Viro err = complete_walk(nd); 2371391172c4SAl Viro if (!err) { 2372391172c4SAl Viro *parent = nd->path; 2373391172c4SAl Viro nd->path.mnt = NULL; 2374391172c4SAl Viro nd->path.dentry = NULL; 2375391172c4SAl Viro } 2376deb106c6SAl Viro terminate_walk(nd); 23778bcb77faSAl Viro return err; 23788bcb77faSAl Viro } 23798bcb77faSAl Viro 23805c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2381391172c4SAl Viro unsigned int flags, struct path *parent, 2382391172c4SAl Viro struct qstr *last, int *type) 23838bcb77faSAl Viro { 23848bcb77faSAl Viro int retval; 23859883d185SAl Viro struct nameidata nd; 23868bcb77faSAl Viro 23875c31b6ceSAl Viro if (IS_ERR(name)) 23885c31b6ceSAl Viro return name; 23899883d185SAl Viro set_nameidata(&nd, dfd, name); 2390c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 23918bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2392c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 23938bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2394c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2395391172c4SAl Viro if (likely(!retval)) { 2396391172c4SAl Viro *last = nd.last; 2397391172c4SAl Viro *type = nd.last_type; 2398c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 23995c31b6ceSAl Viro } else { 24005c31b6ceSAl Viro putname(name); 24015c31b6ceSAl Viro name = ERR_PTR(retval); 2402391172c4SAl Viro } 24039883d185SAl Viro restore_nameidata(); 24045c31b6ceSAl Viro return name; 24058bcb77faSAl Viro } 24068bcb77faSAl Viro 240779714f72SAl Viro /* does lookup, returns the object with parent locked */ 240879714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24095590ff0dSUlrich Drepper { 24105c31b6ceSAl Viro struct filename *filename; 24115c31b6ceSAl Viro struct dentry *d; 2412391172c4SAl Viro struct qstr last; 2413391172c4SAl Viro int type; 241451689104SPaul Moore 24155c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24165c31b6ceSAl Viro &last, &type); 241751689104SPaul Moore if (IS_ERR(filename)) 241851689104SPaul Moore return ERR_CAST(filename); 24195c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2420391172c4SAl Viro path_put(path); 24215c31b6ceSAl Viro putname(filename); 24225c31b6ceSAl Viro return ERR_PTR(-EINVAL); 242379714f72SAl Viro } 24245955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2425391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 242679714f72SAl Viro if (IS_ERR(d)) { 24275955102cSAl Viro inode_unlock(path->dentry->d_inode); 2428391172c4SAl Viro path_put(path); 242979714f72SAl Viro } 243051689104SPaul Moore putname(filename); 243179714f72SAl Viro return d; 24325590ff0dSUlrich Drepper } 24335590ff0dSUlrich Drepper 2434d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2435d1811465SAl Viro { 2436abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2437abc9f5beSAl Viro flags, path, NULL); 2438d1811465SAl Viro } 24394d359507SAl Viro EXPORT_SYMBOL(kern_path); 2440d1811465SAl Viro 244116f18200SJosef 'Jeff' Sipek /** 244216f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 244316f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 244416f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 244516f18200SJosef 'Jeff' Sipek * @name: pointer to file name 244616f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2447e0a01249SAl Viro * @path: pointer to struct path to fill 244816f18200SJosef 'Jeff' Sipek */ 244916f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 245016f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2451e0a01249SAl Viro struct path *path) 245216f18200SJosef 'Jeff' Sipek { 24539ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24549ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2455abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2456abc9f5beSAl Viro flags , path, &root); 245716f18200SJosef 'Jeff' Sipek } 24584d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 245916f18200SJosef 'Jeff' Sipek 24603c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24613c95f0dcSAl Viro int len, struct qstr *this) 24623c95f0dcSAl Viro { 24633c95f0dcSAl Viro this->name = name; 24643c95f0dcSAl Viro this->len = len; 24653c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24663c95f0dcSAl Viro if (!len) 24673c95f0dcSAl Viro return -EACCES; 24683c95f0dcSAl Viro 24693c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24703c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24713c95f0dcSAl Viro return -EACCES; 24723c95f0dcSAl Viro } 24733c95f0dcSAl Viro 24743c95f0dcSAl Viro while (len--) { 24753c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24763c95f0dcSAl Viro if (c == '/' || c == '\0') 24773c95f0dcSAl Viro return -EACCES; 24783c95f0dcSAl Viro } 24793c95f0dcSAl Viro /* 24803c95f0dcSAl Viro * See if the low-level filesystem might want 24813c95f0dcSAl Viro * to use its own hash.. 24823c95f0dcSAl Viro */ 24833c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 24843c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 24853c95f0dcSAl Viro if (err < 0) 24863c95f0dcSAl Viro return err; 24873c95f0dcSAl Viro } 24883c95f0dcSAl Viro 24893c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 24903c95f0dcSAl Viro } 24913c95f0dcSAl Viro 2492eead1911SChristoph Hellwig /** 24930da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 24940da0b7fdSDavid Howells * @name: pathname component to lookup 24950da0b7fdSDavid Howells * @base: base directory to lookup from 24960da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 24970da0b7fdSDavid Howells * 24980da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 24990da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 25000da0b7fdSDavid Howells * 25010da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25020da0b7fdSDavid Howells * not be called by generic code. 25030da0b7fdSDavid Howells * 25040da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25050da0b7fdSDavid Howells */ 25060da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25070da0b7fdSDavid Howells { 25080da0b7fdSDavid Howells struct qstr this; 25090da0b7fdSDavid Howells int err; 25100da0b7fdSDavid Howells 25110da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25120da0b7fdSDavid Howells 25130da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25140da0b7fdSDavid Howells if (err) 25150da0b7fdSDavid Howells return ERR_PTR(err); 25160da0b7fdSDavid Howells 25170da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25180da0b7fdSDavid Howells } 25190da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25200da0b7fdSDavid Howells 25210da0b7fdSDavid Howells /** 2522a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2523eead1911SChristoph Hellwig * @name: pathname component to lookup 2524eead1911SChristoph Hellwig * @base: base directory to lookup from 2525eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2526eead1911SChristoph Hellwig * 2527a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25289e7543e9SAl Viro * not be called by generic code. 2529bbddca8eSNeilBrown * 2530bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2531eead1911SChristoph Hellwig */ 2532057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2533057f6c01SJames Morris { 25348613a209SAl Viro struct dentry *dentry; 2535057f6c01SJames Morris struct qstr this; 2536cda309deSMiklos Szeredi int err; 2537057f6c01SJames Morris 25385955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25392f9092e1SDavid Woodhouse 25403c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2541cda309deSMiklos Szeredi if (err) 2542cda309deSMiklos Szeredi return ERR_PTR(err); 2543cda309deSMiklos Szeredi 25448613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25458613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2546057f6c01SJames Morris } 25474d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2548057f6c01SJames Morris 2549bbddca8eSNeilBrown /** 2550bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2551bbddca8eSNeilBrown * @name: pathname component to lookup 2552bbddca8eSNeilBrown * @base: base directory to lookup from 2553bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2554bbddca8eSNeilBrown * 2555bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2556bbddca8eSNeilBrown * not be called by generic code. 2557bbddca8eSNeilBrown * 2558bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2559bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2560bbddca8eSNeilBrown */ 2561bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2562bbddca8eSNeilBrown struct dentry *base, int len) 2563bbddca8eSNeilBrown { 2564bbddca8eSNeilBrown struct qstr this; 2565bbddca8eSNeilBrown int err; 256620d00ee8SLinus Torvalds struct dentry *ret; 2567bbddca8eSNeilBrown 25683c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2569bbddca8eSNeilBrown if (err) 2570bbddca8eSNeilBrown return ERR_PTR(err); 2571bbddca8eSNeilBrown 257220d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 257320d00ee8SLinus Torvalds if (!ret) 257420d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 257520d00ee8SLinus Torvalds return ret; 2576bbddca8eSNeilBrown } 2577bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2578bbddca8eSNeilBrown 25796c2d4798SAl Viro /* 25806c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 25816c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 25826c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 25836c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 25846c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 25856c2d4798SAl Viro * this one avoids such problems. 25866c2d4798SAl Viro */ 25876c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 25886c2d4798SAl Viro struct dentry *base, int len) 25896c2d4798SAl Viro { 25906c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 25912fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 25926c2d4798SAl Viro dput(ret); 25936c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 25946c2d4798SAl Viro } 25956c2d4798SAl Viro return ret; 25966c2d4798SAl Viro } 25976c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 25986c2d4798SAl Viro 2599eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2600eedf265aSEric W. Biederman int path_pts(struct path *path) 2601eedf265aSEric W. Biederman { 2602eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2603eedf265aSEric W. Biederman * the input path. 2604eedf265aSEric W. Biederman */ 2605eedf265aSEric W. Biederman struct dentry *child, *parent; 2606eedf265aSEric W. Biederman struct qstr this; 2607eedf265aSEric W. Biederman 260863b27720SAl Viro parent = path_parent_directory(path); 260963b27720SAl Viro if (!parent) 261063b27720SAl Viro return -ENOENT; 2611eedf265aSEric W. Biederman 261263b27720SAl Viro dput(path->dentry); 261363b27720SAl Viro path->dentry = parent; 2614eedf265aSEric W. Biederman this.name = "pts"; 2615eedf265aSEric W. Biederman this.len = 3; 2616eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2617eedf265aSEric W. Biederman if (!child) 2618eedf265aSEric W. Biederman return -ENOENT; 2619eedf265aSEric W. Biederman 2620eedf265aSEric W. Biederman path->dentry = child; 2621eedf265aSEric W. Biederman dput(parent); 2622eedf265aSEric W. Biederman follow_mount(path); 2623eedf265aSEric W. Biederman return 0; 2624eedf265aSEric W. Biederman } 2625eedf265aSEric W. Biederman #endif 2626eedf265aSEric W. Biederman 26271fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26281fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26291da177e4SLinus Torvalds { 2630abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2631abc9f5beSAl Viro flags, path, NULL); 26321da177e4SLinus Torvalds } 2633b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26341fa1e7f6SAndy Whitcroft 2635cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 26361da177e4SLinus Torvalds { 26378e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2638da9592edSDavid Howells 26398e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 26401da177e4SLinus Torvalds return 0; 26418e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 26421da177e4SLinus Torvalds return 0; 264323adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 26441da177e4SLinus Torvalds } 2645cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 26461da177e4SLinus Torvalds 26471da177e4SLinus Torvalds /* 26481da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 26491da177e4SLinus Torvalds * whether the type of victim is right. 26501da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 26511da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 26521da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 26531da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 26541da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 26551da177e4SLinus Torvalds * a. be owner of dir, or 26561da177e4SLinus Torvalds * b. be owner of victim, or 26571da177e4SLinus Torvalds * c. have CAP_FOWNER capability 26581da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 26591da177e4SLinus Torvalds * links pointing to it. 26600bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 26610bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 26620bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 26630bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 26640bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 26651da177e4SLinus Torvalds * nfs_async_unlink(). 26661da177e4SLinus Torvalds */ 2667b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 26681da177e4SLinus Torvalds { 266963afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 26701da177e4SLinus Torvalds int error; 26711da177e4SLinus Torvalds 2672b18825a7SDavid Howells if (d_is_negative(victim)) 26731da177e4SLinus Torvalds return -ENOENT; 2674b18825a7SDavid Howells BUG_ON(!inode); 26751da177e4SLinus Torvalds 26761da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2677593d1ce8SEric W. Biederman 2678593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2679593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2680593d1ce8SEric W. Biederman return -EOVERFLOW; 2681593d1ce8SEric W. Biederman 26824fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 26831da177e4SLinus Torvalds 2684f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 26851da177e4SLinus Torvalds if (error) 26861da177e4SLinus Torvalds return error; 26871da177e4SLinus Torvalds if (IS_APPEND(dir)) 26881da177e4SLinus Torvalds return -EPERM; 2689b18825a7SDavid Howells 2690b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 26910bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 26921da177e4SLinus Torvalds return -EPERM; 26931da177e4SLinus Torvalds if (isdir) { 269444b1d530SMiklos Szeredi if (!d_is_dir(victim)) 26951da177e4SLinus Torvalds return -ENOTDIR; 26961da177e4SLinus Torvalds if (IS_ROOT(victim)) 26971da177e4SLinus Torvalds return -EBUSY; 269844b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 26991da177e4SLinus Torvalds return -EISDIR; 27001da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27011da177e4SLinus Torvalds return -ENOENT; 27021da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27031da177e4SLinus Torvalds return -EBUSY; 27041da177e4SLinus Torvalds return 0; 27051da177e4SLinus Torvalds } 27061da177e4SLinus Torvalds 27071da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27081da177e4SLinus Torvalds * dir. 27091da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27101da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27111da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2712036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2713036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2714036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27151da177e4SLinus Torvalds */ 2716a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 27171da177e4SLinus Torvalds { 2718036d5236SEric W. Biederman struct user_namespace *s_user_ns; 271914e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 27201da177e4SLinus Torvalds if (child->d_inode) 27211da177e4SLinus Torvalds return -EEXIST; 27221da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27231da177e4SLinus Torvalds return -ENOENT; 2724036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2725036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2726036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2727036d5236SEric W. Biederman return -EOVERFLOW; 2728f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 27291da177e4SLinus Torvalds } 27301da177e4SLinus Torvalds 27311da177e4SLinus Torvalds /* 27321da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 27331da177e4SLinus Torvalds */ 27341da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 27351da177e4SLinus Torvalds { 27361da177e4SLinus Torvalds struct dentry *p; 27371da177e4SLinus Torvalds 27381da177e4SLinus Torvalds if (p1 == p2) { 27395955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27401da177e4SLinus Torvalds return NULL; 27411da177e4SLinus Torvalds } 27421da177e4SLinus Torvalds 2743fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 27441da177e4SLinus Torvalds 2745e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2746e2761a11SOGAWA Hirofumi if (p) { 27475955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 27485955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 27491da177e4SLinus Torvalds return p; 27501da177e4SLinus Torvalds } 27511da177e4SLinus Torvalds 2752e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2753e2761a11SOGAWA Hirofumi if (p) { 27545955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27555955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 27561da177e4SLinus Torvalds return p; 27571da177e4SLinus Torvalds } 27581da177e4SLinus Torvalds 27595955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27605955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 27611da177e4SLinus Torvalds return NULL; 27621da177e4SLinus Torvalds } 27634d359507SAl Viro EXPORT_SYMBOL(lock_rename); 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 27661da177e4SLinus Torvalds { 27675955102cSAl Viro inode_unlock(p1->d_inode); 27681da177e4SLinus Torvalds if (p1 != p2) { 27695955102cSAl Viro inode_unlock(p2->d_inode); 2770fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 27711da177e4SLinus Torvalds } 27721da177e4SLinus Torvalds } 27734d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 27741da177e4SLinus Torvalds 27754acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2776312b63fbSAl Viro bool want_excl) 27771da177e4SLinus Torvalds { 2778a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 27791da177e4SLinus Torvalds if (error) 27801da177e4SLinus Torvalds return error; 27811da177e4SLinus Torvalds 2782acfa4380SAl Viro if (!dir->i_op->create) 27831da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 27841da177e4SLinus Torvalds mode &= S_IALLUGO; 27851da177e4SLinus Torvalds mode |= S_IFREG; 27861da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 27871da177e4SLinus Torvalds if (error) 27881da177e4SLinus Torvalds return error; 2789312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2790a74574aaSStephen Smalley if (!error) 2791f38aa942SAmy Griffis fsnotify_create(dir, dentry); 27921da177e4SLinus Torvalds return error; 27931da177e4SLinus Torvalds } 27944d359507SAl Viro EXPORT_SYMBOL(vfs_create); 27951da177e4SLinus Torvalds 27968e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 27978e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 27988e6c848eSAl Viro void *arg) 27998e6c848eSAl Viro { 28008e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 28018e6c848eSAl Viro int error = may_create(dir, dentry); 28028e6c848eSAl Viro if (error) 28038e6c848eSAl Viro return error; 28048e6c848eSAl Viro 28058e6c848eSAl Viro mode &= S_IALLUGO; 28068e6c848eSAl Viro mode |= S_IFREG; 28078e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 28088e6c848eSAl Viro if (error) 28098e6c848eSAl Viro return error; 28108e6c848eSAl Viro error = f(dentry, mode, arg); 28118e6c848eSAl Viro if (!error) 28128e6c848eSAl Viro fsnotify_create(dir, dentry); 28138e6c848eSAl Viro return error; 28148e6c848eSAl Viro } 28158e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 28168e6c848eSAl Viro 2817a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2818a2982cc9SEric W. Biederman { 2819a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2820a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2821a2982cc9SEric W. Biederman } 2822a2982cc9SEric W. Biederman 2823f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 28241da177e4SLinus Torvalds { 28253fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 28261da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 28271da177e4SLinus Torvalds int error; 28281da177e4SLinus Torvalds 28291da177e4SLinus Torvalds if (!inode) 28301da177e4SLinus Torvalds return -ENOENT; 28311da177e4SLinus Torvalds 2832c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2833c8fe8f30SChristoph Hellwig case S_IFLNK: 28341da177e4SLinus Torvalds return -ELOOP; 2835c8fe8f30SChristoph Hellwig case S_IFDIR: 2836c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 28371da177e4SLinus Torvalds return -EISDIR; 2838c8fe8f30SChristoph Hellwig break; 2839c8fe8f30SChristoph Hellwig case S_IFBLK: 2840c8fe8f30SChristoph Hellwig case S_IFCHR: 2841a2982cc9SEric W. Biederman if (!may_open_dev(path)) 28421da177e4SLinus Torvalds return -EACCES; 2843c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2844c8fe8f30SChristoph Hellwig case S_IFIFO: 2845c8fe8f30SChristoph Hellwig case S_IFSOCK: 28461da177e4SLinus Torvalds flag &= ~O_TRUNC; 2847c8fe8f30SChristoph Hellwig break; 28484a3fd211SDave Hansen } 2849b41572e9SDave Hansen 285062fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2851b41572e9SDave Hansen if (error) 2852b41572e9SDave Hansen return error; 28536146f0d5SMimi Zohar 28541da177e4SLinus Torvalds /* 28551da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 28561da177e4SLinus Torvalds */ 28571da177e4SLinus Torvalds if (IS_APPEND(inode)) { 28588737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 28597715b521SAl Viro return -EPERM; 28601da177e4SLinus Torvalds if (flag & O_TRUNC) 28617715b521SAl Viro return -EPERM; 28621da177e4SLinus Torvalds } 28631da177e4SLinus Torvalds 28641da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 28652e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 28667715b521SAl Viro return -EPERM; 28671da177e4SLinus Torvalds 2868f3c7691eSJ. Bruce Fields return 0; 28697715b521SAl Viro } 28707715b521SAl Viro 2871e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 28727715b521SAl Viro { 2873f0bb5aafSAl Viro const struct path *path = &filp->f_path; 28747715b521SAl Viro struct inode *inode = path->dentry->d_inode; 28757715b521SAl Viro int error = get_write_access(inode); 28761da177e4SLinus Torvalds if (error) 28777715b521SAl Viro return error; 28781da177e4SLinus Torvalds /* 28791da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 28801da177e4SLinus Torvalds */ 2881d7a06983SJeff Layton error = locks_verify_locked(filp); 2882be6d3e56SKentaro Takeda if (!error) 2883ea0d3ab2STetsuo Handa error = security_path_truncate(path); 28841da177e4SLinus Torvalds if (!error) { 28857715b521SAl Viro error = do_truncate(path->dentry, 0, 2886d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2887e1181ee6SJeff Layton filp); 28881da177e4SLinus Torvalds } 28891da177e4SLinus Torvalds put_write_access(inode); 2890acd0c935SMimi Zohar return error; 28911da177e4SLinus Torvalds } 28921da177e4SLinus Torvalds 2893d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2894d57999e1SDave Hansen { 28958a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 28968a5e929dSAl Viro flag--; 2897d57999e1SDave Hansen return flag; 2898d57999e1SDave Hansen } 2899d57999e1SDave Hansen 2900d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2901d18e9008SMiklos Szeredi { 29021328c727SSeth Forshee struct user_namespace *s_user_ns; 2903d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2904d18e9008SMiklos Szeredi if (error) 2905d18e9008SMiklos Szeredi return error; 2906d18e9008SMiklos Szeredi 29071328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29081328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29091328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29101328c727SSeth Forshee return -EOVERFLOW; 29111328c727SSeth Forshee 2912d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2913d18e9008SMiklos Szeredi if (error) 2914d18e9008SMiklos Szeredi return error; 2915d18e9008SMiklos Szeredi 2916d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2917d18e9008SMiklos Szeredi } 2918d18e9008SMiklos Szeredi 29191acf0af9SDavid Howells /* 29201acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29211acf0af9SDavid Howells * dentry. 29221acf0af9SDavid Howells * 29231acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29241acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29251acf0af9SDavid Howells * 292600a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 292700a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 292800a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 29291acf0af9SDavid Howells * 29301acf0af9SDavid Howells * Returns an error code otherwise. 29311acf0af9SDavid Howells */ 2932239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 2933239eb983SAl Viro struct file *file, 2934015c3bbcSMiklos Szeredi const struct open_flags *op, 29353ec2eef1SAl Viro int open_flag, umode_t mode) 2936d18e9008SMiklos Szeredi { 2937d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2938d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2939d18e9008SMiklos Szeredi int error; 2940d18e9008SMiklos Szeredi 2941384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 2942d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2943d18e9008SMiklos Szeredi 2944d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2945d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2946d18e9008SMiklos Szeredi 294730d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 294830d90494SAl Viro file->f_path.mnt = nd->path.mnt; 29490fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 295044907d79SAl Viro open_to_namei_flags(open_flag), mode); 29516fbd0714SAl Viro d_lookup_done(dentry); 2952384f26e2SAl Viro if (!error) { 295364e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 29546fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 29556fb968cdSAl Viro dput(dentry); 29566fb968cdSAl Viro dentry = dget(file->f_path.dentry); 29576fb968cdSAl Viro } 29587be219b4SAl Viro if (file->f_mode & FMODE_CREATED) 295903da633aSAl Viro fsnotify_create(dir, dentry); 296064e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 2961116cc022SMiklos Szeredi error = -EIO; 2962384f26e2SAl Viro } else { 2963d18e9008SMiklos Szeredi if (file->f_path.dentry) { 2964d18e9008SMiklos Szeredi dput(dentry); 2965d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 296610c64ceaSAl Viro } 296773a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 2968d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 2969239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 2970a01e718fSAl Viro error = -ENOENT; 2971d18e9008SMiklos Szeredi } 2972d18e9008SMiklos Szeredi } 2973239eb983SAl Viro if (error) { 2974d18e9008SMiklos Szeredi dput(dentry); 2975239eb983SAl Viro dentry = ERR_PTR(error); 2976239eb983SAl Viro } 2977239eb983SAl Viro return dentry; 2978d18e9008SMiklos Szeredi } 2979d18e9008SMiklos Szeredi 298031e6b01fSNick Piggin /* 29811acf0af9SDavid Howells * Look up and maybe create and open the last component. 2982d58ffd35SMiklos Szeredi * 298300a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 2984d58ffd35SMiklos Szeredi * 298500a07c15SAl Viro * Returns 0 on success, that is, if 298600a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 298700a07c15SAl Viro * the file was not completely opened at this time, though lookups and 298800a07c15SAl Viro * creations were performed. 298900a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 299000a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 299100a07c15SAl Viro * hadn't been specified. 29921acf0af9SDavid Howells * 299300a07c15SAl Viro * An error code is returned on failure. 2994d58ffd35SMiklos Szeredi */ 2995da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 2996d58ffd35SMiklos Szeredi const struct open_flags *op, 29973ec2eef1SAl Viro bool got_write) 2998d58ffd35SMiklos Szeredi { 2999d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 300054ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30011643b43fSAl Viro int open_flag = op->open_flag; 3002d58ffd35SMiklos Szeredi struct dentry *dentry; 30031643b43fSAl Viro int error, create_error = 0; 30041643b43fSAl Viro umode_t mode = op->mode; 30056fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3006d58ffd35SMiklos Szeredi 3007ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3008da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3009d58ffd35SMiklos Szeredi 301073a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 30116fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30126fbd0714SAl Viro for (;;) { 30136fbd0714SAl Viro if (!dentry) { 30146fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3015d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3016da5ebf5aSAl Viro return dentry; 30176fbd0714SAl Viro } 30186fbd0714SAl Viro if (d_in_lookup(dentry)) 30196fbd0714SAl Viro break; 3020d58ffd35SMiklos Szeredi 30216fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 30226fbd0714SAl Viro if (likely(error > 0)) 30236fbd0714SAl Viro break; 30246fbd0714SAl Viro if (error) 30256fbd0714SAl Viro goto out_dput; 30266fbd0714SAl Viro d_invalidate(dentry); 30276fbd0714SAl Viro dput(dentry); 30286fbd0714SAl Viro dentry = NULL; 30296fbd0714SAl Viro } 30306fbd0714SAl Viro if (dentry->d_inode) { 3031d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3032da5ebf5aSAl Viro return dentry; 30336c51e513SAl Viro } 3034d18e9008SMiklos Szeredi 30351643b43fSAl Viro /* 30361643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 30371643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 30381643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 30391643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 30401643b43fSAl Viro * 30411643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 30421643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 30431643b43fSAl Viro */ 30441643b43fSAl Viro if (open_flag & O_CREAT) { 30451643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 30461643b43fSAl Viro mode &= ~current_umask(); 30471643b43fSAl Viro if (unlikely(!got_write)) { 30481643b43fSAl Viro create_error = -EROFS; 30491643b43fSAl Viro open_flag &= ~O_CREAT; 30501643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 30511643b43fSAl Viro goto no_open; 30521643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 30531643b43fSAl Viro } else { 30541643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 30551643b43fSAl Viro if (create_error) { 30561643b43fSAl Viro open_flag &= ~O_CREAT; 30571643b43fSAl Viro if (open_flag & O_EXCL) 30581643b43fSAl Viro goto no_open; 30591643b43fSAl Viro } 30601643b43fSAl Viro } 30611643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 30621643b43fSAl Viro unlikely(!got_write)) { 30631643b43fSAl Viro /* 30641643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 30651643b43fSAl Viro * back to lookup + open 30661643b43fSAl Viro */ 30671643b43fSAl Viro goto no_open; 3068d18e9008SMiklos Szeredi } 3069d18e9008SMiklos Szeredi 30701643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3071239eb983SAl Viro dentry = atomic_open(nd, dentry, file, op, open_flag, mode); 3072da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3073da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3074da5ebf5aSAl Viro return dentry; 3075239eb983SAl Viro } 307654ef4872SMiklos Szeredi 30771643b43fSAl Viro no_open: 30786fbd0714SAl Viro if (d_in_lookup(dentry)) { 307912fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 308012fa5e24SAl Viro nd->flags); 30816fbd0714SAl Viro d_lookup_done(dentry); 308212fa5e24SAl Viro if (unlikely(res)) { 308312fa5e24SAl Viro if (IS_ERR(res)) { 308412fa5e24SAl Viro error = PTR_ERR(res); 308512fa5e24SAl Viro goto out_dput; 308612fa5e24SAl Viro } 308712fa5e24SAl Viro dput(dentry); 308812fa5e24SAl Viro dentry = res; 308912fa5e24SAl Viro } 309054ef4872SMiklos Szeredi } 309154ef4872SMiklos Szeredi 3092d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 30931643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 309473a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3095ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3096ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3097ce8644fcSAl Viro error = -EACCES; 3098d58ffd35SMiklos Szeredi goto out_dput; 309964894cf8SAl Viro } 3100ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31011643b43fSAl Viro open_flag & O_EXCL); 3102d58ffd35SMiklos Szeredi if (error) 3103d58ffd35SMiklos Szeredi goto out_dput; 3104ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3105d58ffd35SMiklos Szeredi } 31061643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31071643b43fSAl Viro error = create_error; 3108d58ffd35SMiklos Szeredi goto out_dput; 3109d58ffd35SMiklos Szeredi } 3110da5ebf5aSAl Viro return dentry; 3111d58ffd35SMiklos Szeredi 3112d58ffd35SMiklos Szeredi out_dput: 3113d58ffd35SMiklos Szeredi dput(dentry); 3114da5ebf5aSAl Viro return ERR_PTR(error); 3115d58ffd35SMiklos Szeredi } 3116d58ffd35SMiklos Szeredi 3117c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 31183ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3119fb1cc555SAl Viro { 3120a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3121ca344a89SAl Viro int open_flag = op->open_flag; 312264894cf8SAl Viro bool got_write = false; 3123254cf582SAl Viro unsigned seq; 3124a1eb3315SMiklos Szeredi struct inode *inode; 3125da5ebf5aSAl Viro struct dentry *dentry; 3126b0417d2cSAl Viro const char *res; 312716c2cd71SAl Viro int error; 3128fb1cc555SAl Viro 3129c3e380b0SAl Viro nd->flags |= op->intent; 3130c3e380b0SAl Viro 3131bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 313256676ec3SAl Viro if (nd->depth) 313356676ec3SAl Viro put_link(nd); 3134fe2d35ffSAl Viro error = handle_dots(nd, nd->last_type); 3135c981a482SAl Viro if (likely(!error)) 3136c981a482SAl Viro error = complete_walk(nd); 31371ccac622SAl Viro return ERR_PTR(error); 31381f36f774SAl Viro } 3139a2c36b45SAl Viro 3140ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3141fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3142fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3143fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 314420e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 314520e34357SAl Viro if (IS_ERR(dentry)) 31461ccac622SAl Viro return ERR_CAST(dentry); 314720e34357SAl Viro if (likely(dentry)) 314871574865SMiklos Szeredi goto finish_lookup; 314971574865SMiklos Szeredi 31506583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3151b6183df7SMiklos Szeredi } else { 3152fe2d35ffSAl Viro /* create side of things */ 3153a3fbbde7SAl Viro /* 3154b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3155b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3156b6183df7SMiklos Szeredi * about to look up 3157a3fbbde7SAl Viro */ 31589f1fafeeSAl Viro error = complete_walk(nd); 3159c981a482SAl Viro if (unlikely(error)) 31601ccac622SAl Viro return ERR_PTR(error); 3161fe2d35ffSAl Viro 3162c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 31631f36f774SAl Viro /* trailing slashes? */ 3164deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 31651ccac622SAl Viro return ERR_PTR(-EISDIR); 3166b6183df7SMiklos Szeredi } 31671f36f774SAl Viro 31689cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 316964894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 317064894cf8SAl Viro if (!error) 317164894cf8SAl Viro got_write = true; 317264894cf8SAl Viro /* 317364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 317464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 317564894cf8SAl Viro * dropping this one anyway. 317664894cf8SAl Viro */ 317764894cf8SAl Viro } 31789cf843e3SAl Viro if (open_flag & O_CREAT) 31795955102cSAl Viro inode_lock(dir->d_inode); 31809cf843e3SAl Viro else 31819cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3182da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 31839cf843e3SAl Viro if (open_flag & O_CREAT) 31845955102cSAl Viro inode_unlock(dir->d_inode); 31859cf843e3SAl Viro else 31869cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3187fb1cc555SAl Viro 3188c981a482SAl Viro if (got_write) 318959e96e65SAl Viro mnt_drop_write(nd->path.mnt); 31906c0d46c4SAl Viro 319159e96e65SAl Viro if (IS_ERR(dentry)) 319259e96e65SAl Viro return ERR_CAST(dentry); 319359e96e65SAl Viro 3194973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3195e73cabffSAl Viro dput(nd->path.dentry); 3196e73cabffSAl Viro nd->path.dentry = dentry; 3197c981a482SAl Viro return NULL; 3198fb1cc555SAl Viro } 3199fb1cc555SAl Viro 320020e34357SAl Viro finish_lookup: 320156676ec3SAl Viro if (nd->depth) 320256676ec3SAl Viro put_link(nd); 32038c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3204b0417d2cSAl Viro if (unlikely(res)) { 32051ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3206b0417d2cSAl Viro return res; 32071ccac622SAl Viro } 320831d1726dSAl Viro 320931d1726dSAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 321031d1726dSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 32111ccac622SAl Viro return ERR_PTR(-EEXIST); 321231d1726dSAl Viro } 3213c981a482SAl Viro 32148f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3215c981a482SAl Viro return ERR_PTR(complete_walk(nd)); 3216c981a482SAl Viro } 3217c981a482SAl Viro 3218c981a482SAl Viro /* 3219c981a482SAl Viro * Handle the last step of open() 3220c981a482SAl Viro */ 3221c981a482SAl Viro static const char *do_last(struct nameidata *nd, 3222c981a482SAl Viro struct file *file, const struct open_flags *op) 3223c981a482SAl Viro { 3224c981a482SAl Viro kuid_t dir_uid = nd->inode->i_uid; 3225c981a482SAl Viro umode_t dir_mode = nd->inode->i_mode; 3226c981a482SAl Viro int open_flag = op->open_flag; 3227c981a482SAl Viro bool do_truncate; 3228c981a482SAl Viro int acc_mode; 3229c981a482SAl Viro const char *link; 3230c981a482SAl Viro int error; 3231c981a482SAl Viro 3232c981a482SAl Viro link = open_last_lookups(nd, file, op); 3233c981a482SAl Viro if (unlikely(link)) 3234c981a482SAl Viro return link; 3235c981a482SAl Viro 3236973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 323776ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 323830aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 323930aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 324059e96e65SAl Viro return ERR_PTR(-EISDIR); 3241d0cb5018SAl Viro error = may_create_in_sticky(dir_mode, dir_uid, 324230aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 324330aba665SSalvatore Mesoraca if (unlikely(error)) 324459e96e65SAl Viro return ERR_PTR(error); 324530aba665SSalvatore Mesoraca } 324644b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 324759e96e65SAl Viro return ERR_PTR(-ENOTDIR); 32486c0d46c4SAl Viro 32498795e7d4SAl Viro do_truncate = false; 32508795e7d4SAl Viro acc_mode = op->acc_mode; 32515a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 32525a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 32535a2d3eddSAl Viro open_flag &= ~O_TRUNC; 32545a2d3eddSAl Viro acc_mode = 0; 32558795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 32560f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 32570f9d1a10SAl Viro if (error) 325859e96e65SAl Viro return ERR_PTR(error); 32598795e7d4SAl Viro do_truncate = true; 32600f9d1a10SAl Viro } 3261bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 32628795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3263ae2bb293SAl Viro error = vfs_open(&nd->path, file); 32648795e7d4SAl Viro if (!error) 32656035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 32668795e7d4SAl Viro if (!error && do_truncate) 32672675a4ebSAl Viro error = handle_truncate(file); 3268c80567c8SAl Viro if (unlikely(error > 0)) { 3269c80567c8SAl Viro WARN_ON(1); 3270c80567c8SAl Viro error = -EINVAL; 3271c80567c8SAl Viro } 32728795e7d4SAl Viro if (do_truncate) 32730f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 32741ccac622SAl Viro return ERR_PTR(error); 3275fb1cc555SAl Viro } 3276fb1cc555SAl Viro 3277af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3278af7bd4dcSAmir Goldstein { 3279af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3280af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3281af7bd4dcSAmir Goldstein struct inode *inode; 3282af7bd4dcSAmir Goldstein int error; 3283af7bd4dcSAmir Goldstein 3284af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3285af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3286af7bd4dcSAmir Goldstein if (error) 3287af7bd4dcSAmir Goldstein goto out_err; 3288af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3289af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3290af7bd4dcSAmir Goldstein goto out_err; 3291af7bd4dcSAmir Goldstein error = -ENOMEM; 3292cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3293af7bd4dcSAmir Goldstein if (unlikely(!child)) 3294af7bd4dcSAmir Goldstein goto out_err; 3295af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3296af7bd4dcSAmir Goldstein if (error) 3297af7bd4dcSAmir Goldstein goto out_err; 3298af7bd4dcSAmir Goldstein error = -ENOENT; 3299af7bd4dcSAmir Goldstein inode = child->d_inode; 3300af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3301af7bd4dcSAmir Goldstein goto out_err; 3302af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3303af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3304af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3305af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3306af7bd4dcSAmir Goldstein } 3307fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3308af7bd4dcSAmir Goldstein return child; 3309af7bd4dcSAmir Goldstein 3310af7bd4dcSAmir Goldstein out_err: 3311af7bd4dcSAmir Goldstein dput(child); 3312af7bd4dcSAmir Goldstein return ERR_PTR(error); 3313af7bd4dcSAmir Goldstein } 3314af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3315af7bd4dcSAmir Goldstein 3316c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 331760545d0dSAl Viro const struct open_flags *op, 33183ec2eef1SAl Viro struct file *file) 331960545d0dSAl Viro { 3320625b6d10SAl Viro struct dentry *child; 3321625b6d10SAl Viro struct path path; 3322c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 332360545d0dSAl Viro if (unlikely(error)) 332460545d0dSAl Viro return error; 3325625b6d10SAl Viro error = mnt_want_write(path.mnt); 332660545d0dSAl Viro if (unlikely(error)) 332760545d0dSAl Viro goto out; 3328af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3329af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3330684e73beSHirofumi Nakagawa if (IS_ERR(child)) 333160545d0dSAl Viro goto out2; 3332625b6d10SAl Viro dput(path.dentry); 3333625b6d10SAl Viro path.dentry = child; 3334c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 333569a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 333662fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 333760545d0dSAl Viro if (error) 333860545d0dSAl Viro goto out2; 3339625b6d10SAl Viro file->f_path.mnt = path.mnt; 3340be12af3eSAl Viro error = finish_open(file, child, NULL); 334160545d0dSAl Viro out2: 3342625b6d10SAl Viro mnt_drop_write(path.mnt); 334360545d0dSAl Viro out: 3344625b6d10SAl Viro path_put(&path); 334560545d0dSAl Viro return error; 334660545d0dSAl Viro } 334760545d0dSAl Viro 33486ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 33496ac08709SAl Viro { 33506ac08709SAl Viro struct path path; 33516ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 33526ac08709SAl Viro if (!error) { 33536ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3354ae2bb293SAl Viro error = vfs_open(&path, file); 33556ac08709SAl Viro path_put(&path); 33566ac08709SAl Viro } 33576ac08709SAl Viro return error; 33586ac08709SAl Viro } 33596ac08709SAl Viro 3360c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3361c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 33621da177e4SLinus Torvalds { 336330d90494SAl Viro struct file *file; 336413aab428SAl Viro int error; 336531e6b01fSNick Piggin 3366ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 33671afc99beSAl Viro if (IS_ERR(file)) 33681afc99beSAl Viro return file; 336931e6b01fSNick Piggin 3370bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 33713ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 33725f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 33736ac08709SAl Viro error = do_o_path(nd, flags, file); 33745f336e72SAl Viro } else { 33755f336e72SAl Viro const char *s = path_init(nd, flags); 33763bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 33771ccac622SAl Viro (s = do_last(nd, file, op)) != NULL) 33781ccac622SAl Viro ; 3379deb106c6SAl Viro terminate_walk(nd); 33805f336e72SAl Viro } 33817c1c01ecSAl Viro if (likely(!error)) { 3382aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 33837c1c01ecSAl Viro return file; 33847c1c01ecSAl Viro WARN_ON(1); 33857c1c01ecSAl Viro error = -EINVAL; 3386015c3bbcSMiklos Szeredi } 33877c1c01ecSAl Viro fput(file); 33882675a4ebSAl Viro if (error == -EOPENSTALE) { 33892675a4ebSAl Viro if (flags & LOOKUP_RCU) 33902675a4ebSAl Viro error = -ECHILD; 33912675a4ebSAl Viro else 33922675a4ebSAl Viro error = -ESTALE; 33932675a4ebSAl Viro } 33947c1c01ecSAl Viro return ERR_PTR(error); 3395de459215SKirill Korotaev } 33961da177e4SLinus Torvalds 3397669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3398f9652e10SAl Viro const struct open_flags *op) 339913aab428SAl Viro { 34009883d185SAl Viro struct nameidata nd; 3401f9652e10SAl Viro int flags = op->lookup_flags; 340213aab428SAl Viro struct file *filp; 340313aab428SAl Viro 34049883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3405c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 340613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3407c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 340813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3409c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34109883d185SAl Viro restore_nameidata(); 341113aab428SAl Viro return filp; 341213aab428SAl Viro } 341313aab428SAl Viro 341473d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3415f9652e10SAl Viro const char *name, const struct open_flags *op) 341673d049a4SAl Viro { 34179883d185SAl Viro struct nameidata nd; 341873d049a4SAl Viro struct file *file; 341951689104SPaul Moore struct filename *filename; 3420f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 342173d049a4SAl Viro 342273d049a4SAl Viro nd.root.mnt = mnt; 342373d049a4SAl Viro nd.root.dentry = dentry; 342473d049a4SAl Viro 3425b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 342673d049a4SAl Viro return ERR_PTR(-ELOOP); 342773d049a4SAl Viro 342851689104SPaul Moore filename = getname_kernel(name); 3429a1c83681SViresh Kumar if (IS_ERR(filename)) 343051689104SPaul Moore return ERR_CAST(filename); 343151689104SPaul Moore 34329883d185SAl Viro set_nameidata(&nd, -1, filename); 3433c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 343473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3435c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 343673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3437c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 34389883d185SAl Viro restore_nameidata(); 343951689104SPaul Moore putname(filename); 344073d049a4SAl Viro return file; 344173d049a4SAl Viro } 344273d049a4SAl Viro 3443fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34441ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34451da177e4SLinus Torvalds { 3446c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3447391172c4SAl Viro struct qstr last; 3448391172c4SAl Viro int type; 3449c30dabfeSJan Kara int err2; 34501ac12b4bSJeff Layton int error; 34511ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34521ac12b4bSJeff Layton 34531ac12b4bSJeff Layton /* 34541ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34551ac12b4bSJeff Layton * other flags passed in are ignored! 34561ac12b4bSJeff Layton */ 34571ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34581ac12b4bSJeff Layton 34595c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34605c31b6ceSAl Viro if (IS_ERR(name)) 34615c31b6ceSAl Viro return ERR_CAST(name); 34621da177e4SLinus Torvalds 3463c663e5d8SChristoph Hellwig /* 3464c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3465c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3466c663e5d8SChristoph Hellwig */ 34675c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3468ed75e95dSAl Viro goto out; 3469c663e5d8SChristoph Hellwig 3470c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3471391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3472c663e5d8SChristoph Hellwig /* 3473c663e5d8SChristoph Hellwig * Do the final lookup. 3474c663e5d8SChristoph Hellwig */ 3475391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 34765955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3477391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34781da177e4SLinus Torvalds if (IS_ERR(dentry)) 3479a8104a9fSAl Viro goto unlock; 3480c663e5d8SChristoph Hellwig 3481a8104a9fSAl Viro error = -EEXIST; 3482b18825a7SDavid Howells if (d_is_positive(dentry)) 3483a8104a9fSAl Viro goto fail; 3484b18825a7SDavid Howells 3485c663e5d8SChristoph Hellwig /* 3486c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3487c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3488c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3489c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3490c663e5d8SChristoph Hellwig */ 3491391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3492a8104a9fSAl Viro error = -ENOENT; 3493ed75e95dSAl Viro goto fail; 3494e9baf6e5SAl Viro } 3495c30dabfeSJan Kara if (unlikely(err2)) { 3496c30dabfeSJan Kara error = err2; 3497a8104a9fSAl Viro goto fail; 3498c30dabfeSJan Kara } 3499181c37b6SAl Viro putname(name); 3500e9baf6e5SAl Viro return dentry; 35011da177e4SLinus Torvalds fail: 3502a8104a9fSAl Viro dput(dentry); 3503a8104a9fSAl Viro dentry = ERR_PTR(error); 3504a8104a9fSAl Viro unlock: 35055955102cSAl Viro inode_unlock(path->dentry->d_inode); 3506c30dabfeSJan Kara if (!err2) 3507391172c4SAl Viro mnt_drop_write(path->mnt); 3508ed75e95dSAl Viro out: 3509391172c4SAl Viro path_put(path); 3510181c37b6SAl Viro putname(name); 3511ed75e95dSAl Viro return dentry; 3512dae6ad8fSAl Viro } 3513fa14a0b8SAl Viro 3514fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3515fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3516fa14a0b8SAl Viro { 3517181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3518181c37b6SAl Viro path, lookup_flags); 3519fa14a0b8SAl Viro } 3520dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3521dae6ad8fSAl Viro 3522921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3523921a1650SAl Viro { 3524921a1650SAl Viro dput(dentry); 35255955102cSAl Viro inode_unlock(path->dentry->d_inode); 3526a8104a9fSAl Viro mnt_drop_write(path->mnt); 3527921a1650SAl Viro path_put(path); 3528921a1650SAl Viro } 3529921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3530921a1650SAl Viro 3531520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 35321ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3533dae6ad8fSAl Viro { 3534181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3535dae6ad8fSAl Viro } 3536dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3537dae6ad8fSAl Viro 35381a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35391da177e4SLinus Torvalds { 3540a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35411da177e4SLinus Torvalds 35421da177e4SLinus Torvalds if (error) 35431da177e4SLinus Torvalds return error; 35441da177e4SLinus Torvalds 354594f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 35461da177e4SLinus Torvalds return -EPERM; 35471da177e4SLinus Torvalds 3548acfa4380SAl Viro if (!dir->i_op->mknod) 35491da177e4SLinus Torvalds return -EPERM; 35501da177e4SLinus Torvalds 355108ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 355208ce5f16SSerge E. Hallyn if (error) 355308ce5f16SSerge E. Hallyn return error; 355408ce5f16SSerge E. Hallyn 35551da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35561da177e4SLinus Torvalds if (error) 35571da177e4SLinus Torvalds return error; 35581da177e4SLinus Torvalds 35591da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3560a74574aaSStephen Smalley if (!error) 3561f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35621da177e4SLinus Torvalds return error; 35631da177e4SLinus Torvalds } 35644d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35651da177e4SLinus Torvalds 3566f69aac00SAl Viro static int may_mknod(umode_t mode) 3567463c3197SDave Hansen { 3568463c3197SDave Hansen switch (mode & S_IFMT) { 3569463c3197SDave Hansen case S_IFREG: 3570463c3197SDave Hansen case S_IFCHR: 3571463c3197SDave Hansen case S_IFBLK: 3572463c3197SDave Hansen case S_IFIFO: 3573463c3197SDave Hansen case S_IFSOCK: 3574463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3575463c3197SDave Hansen return 0; 3576463c3197SDave Hansen case S_IFDIR: 3577463c3197SDave Hansen return -EPERM; 3578463c3197SDave Hansen default: 3579463c3197SDave Hansen return -EINVAL; 3580463c3197SDave Hansen } 3581463c3197SDave Hansen } 3582463c3197SDave Hansen 358387c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 358487c4e192SDominik Brodowski unsigned int dev) 35851da177e4SLinus Torvalds { 35861da177e4SLinus Torvalds struct dentry *dentry; 3587dae6ad8fSAl Viro struct path path; 3588dae6ad8fSAl Viro int error; 3589972567f1SJeff Layton unsigned int lookup_flags = 0; 35901da177e4SLinus Torvalds 35918e4bfca1SAl Viro error = may_mknod(mode); 35928e4bfca1SAl Viro if (error) 35938e4bfca1SAl Viro return error; 3594972567f1SJeff Layton retry: 3595972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3596dae6ad8fSAl Viro if (IS_ERR(dentry)) 3597dae6ad8fSAl Viro return PTR_ERR(dentry); 35982ad94ae6SAl Viro 3599dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3600ce3b0f8dSAl Viro mode &= ~current_umask(); 3601dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3602be6d3e56SKentaro Takeda if (error) 3603a8104a9fSAl Viro goto out; 36041da177e4SLinus Torvalds switch (mode & S_IFMT) { 36051da177e4SLinus Torvalds case 0: case S_IFREG: 3606312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 360705d1a717SMimi Zohar if (!error) 360805d1a717SMimi Zohar ima_post_path_mknod(dentry); 36091da177e4SLinus Torvalds break; 36101da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3611dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36121da177e4SLinus Torvalds new_decode_dev(dev)); 36131da177e4SLinus Torvalds break; 36141da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3615dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 36161da177e4SLinus Torvalds break; 36171da177e4SLinus Torvalds } 3618a8104a9fSAl Viro out: 3619921a1650SAl Viro done_path_create(&path, dentry); 3620972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3621972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3622972567f1SJeff Layton goto retry; 3623972567f1SJeff Layton } 36241da177e4SLinus Torvalds return error; 36251da177e4SLinus Torvalds } 36261da177e4SLinus Torvalds 362787c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 362887c4e192SDominik Brodowski unsigned int, dev) 362987c4e192SDominik Brodowski { 363087c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 363187c4e192SDominik Brodowski } 363287c4e192SDominik Brodowski 36338208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 36345590ff0dSUlrich Drepper { 363587c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 36365590ff0dSUlrich Drepper } 36375590ff0dSUlrich Drepper 363818bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 36391da177e4SLinus Torvalds { 3640a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36418de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 36421da177e4SLinus Torvalds 36431da177e4SLinus Torvalds if (error) 36441da177e4SLinus Torvalds return error; 36451da177e4SLinus Torvalds 3646acfa4380SAl Viro if (!dir->i_op->mkdir) 36471da177e4SLinus Torvalds return -EPERM; 36481da177e4SLinus Torvalds 36491da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 36501da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 36511da177e4SLinus Torvalds if (error) 36521da177e4SLinus Torvalds return error; 36531da177e4SLinus Torvalds 36548de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36558de52778SAl Viro return -EMLINK; 36568de52778SAl Viro 36571da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3658a74574aaSStephen Smalley if (!error) 3659f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36601da177e4SLinus Torvalds return error; 36611da177e4SLinus Torvalds } 36624d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36631da177e4SLinus Torvalds 36640101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 36651da177e4SLinus Torvalds { 36666902d925SDave Hansen struct dentry *dentry; 3667dae6ad8fSAl Viro struct path path; 3668dae6ad8fSAl Viro int error; 3669b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36701da177e4SLinus Torvalds 3671b76d8b82SJeff Layton retry: 3672b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36736902d925SDave Hansen if (IS_ERR(dentry)) 3674dae6ad8fSAl Viro return PTR_ERR(dentry); 36756902d925SDave Hansen 3676dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3677ce3b0f8dSAl Viro mode &= ~current_umask(); 3678dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3679a8104a9fSAl Viro if (!error) 3680dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3681921a1650SAl Viro done_path_create(&path, dentry); 3682b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3683b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3684b76d8b82SJeff Layton goto retry; 3685b76d8b82SJeff Layton } 36861da177e4SLinus Torvalds return error; 36871da177e4SLinus Torvalds } 36881da177e4SLinus Torvalds 36890101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 36900101db7aSDominik Brodowski { 36910101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 36920101db7aSDominik Brodowski } 36930101db7aSDominik Brodowski 3694a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 36955590ff0dSUlrich Drepper { 36960101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 36975590ff0dSUlrich Drepper } 36985590ff0dSUlrich Drepper 36991da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 37001da177e4SLinus Torvalds { 37011da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 37021da177e4SLinus Torvalds 37031da177e4SLinus Torvalds if (error) 37041da177e4SLinus Torvalds return error; 37051da177e4SLinus Torvalds 3706acfa4380SAl Viro if (!dir->i_op->rmdir) 37071da177e4SLinus Torvalds return -EPERM; 37081da177e4SLinus Torvalds 37091d2ef590SAl Viro dget(dentry); 37105955102cSAl Viro inode_lock(dentry->d_inode); 3711912dbc15SSage Weil 37121da177e4SLinus Torvalds error = -EBUSY; 37137af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3714912dbc15SSage Weil goto out; 3715912dbc15SSage Weil 37161da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3717912dbc15SSage Weil if (error) 3718912dbc15SSage Weil goto out; 3719912dbc15SSage Weil 37201da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3721912dbc15SSage Weil if (error) 3722912dbc15SSage Weil goto out; 3723912dbc15SSage Weil 37248767712fSAl Viro shrink_dcache_parent(dentry); 37251da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3726d83c49f3SAl Viro dont_mount(dentry); 37278ed936b5SEric W. Biederman detach_mounts(dentry); 3728116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 37291da177e4SLinus Torvalds 3730912dbc15SSage Weil out: 37315955102cSAl Viro inode_unlock(dentry->d_inode); 37321d2ef590SAl Viro dput(dentry); 3733912dbc15SSage Weil if (!error) 3734912dbc15SSage Weil d_delete(dentry); 37351da177e4SLinus Torvalds return error; 37361da177e4SLinus Torvalds } 37374d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37381da177e4SLinus Torvalds 3739f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 37401da177e4SLinus Torvalds { 37411da177e4SLinus Torvalds int error = 0; 374291a27b2aSJeff Layton struct filename *name; 37431da177e4SLinus Torvalds struct dentry *dentry; 3744f5beed75SAl Viro struct path path; 3745f5beed75SAl Viro struct qstr last; 3746f5beed75SAl Viro int type; 3747c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3748c6ee9206SJeff Layton retry: 3749c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3750c1d4dd27SAl Viro &path, &last, &type); 375191a27b2aSJeff Layton if (IS_ERR(name)) 375291a27b2aSJeff Layton return PTR_ERR(name); 37531da177e4SLinus Torvalds 3754f5beed75SAl Viro switch (type) { 37551da177e4SLinus Torvalds case LAST_DOTDOT: 37561da177e4SLinus Torvalds error = -ENOTEMPTY; 37571da177e4SLinus Torvalds goto exit1; 37581da177e4SLinus Torvalds case LAST_DOT: 37591da177e4SLinus Torvalds error = -EINVAL; 37601da177e4SLinus Torvalds goto exit1; 37611da177e4SLinus Torvalds case LAST_ROOT: 37621da177e4SLinus Torvalds error = -EBUSY; 37631da177e4SLinus Torvalds goto exit1; 37641da177e4SLinus Torvalds } 37650612d9fbSOGAWA Hirofumi 3766f5beed75SAl Viro error = mnt_want_write(path.mnt); 3767c30dabfeSJan Kara if (error) 3768c30dabfeSJan Kara goto exit1; 37690612d9fbSOGAWA Hirofumi 37705955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3771f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37721da177e4SLinus Torvalds error = PTR_ERR(dentry); 37736902d925SDave Hansen if (IS_ERR(dentry)) 37746902d925SDave Hansen goto exit2; 3775e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3776e6bc45d6STheodore Ts'o error = -ENOENT; 3777e6bc45d6STheodore Ts'o goto exit3; 3778e6bc45d6STheodore Ts'o } 3779f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3780be6d3e56SKentaro Takeda if (error) 3781c30dabfeSJan Kara goto exit3; 3782f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37830622753bSDave Hansen exit3: 37841da177e4SLinus Torvalds dput(dentry); 37856902d925SDave Hansen exit2: 37865955102cSAl Viro inode_unlock(path.dentry->d_inode); 3787f5beed75SAl Viro mnt_drop_write(path.mnt); 37881da177e4SLinus Torvalds exit1: 3789f5beed75SAl Viro path_put(&path); 37901da177e4SLinus Torvalds putname(name); 3791c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3792c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3793c6ee9206SJeff Layton goto retry; 3794c6ee9206SJeff Layton } 37951da177e4SLinus Torvalds return error; 37961da177e4SLinus Torvalds } 37971da177e4SLinus Torvalds 37983cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 37995590ff0dSUlrich Drepper { 38005590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 38015590ff0dSUlrich Drepper } 38025590ff0dSUlrich Drepper 3803b21996e3SJ. Bruce Fields /** 3804b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3805b21996e3SJ. Bruce Fields * @dir: parent directory 3806b21996e3SJ. Bruce Fields * @dentry: victim 3807b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3808b21996e3SJ. Bruce Fields * 3809b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3810b21996e3SJ. Bruce Fields * 3811b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3812b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3813b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3814b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3815b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3816b21996e3SJ. Bruce Fields * 3817b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3818b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3819b21996e3SJ. Bruce Fields * to be NFS exported. 3820b21996e3SJ. Bruce Fields */ 3821b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38221da177e4SLinus Torvalds { 38239accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38241da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 38251da177e4SLinus Torvalds 38261da177e4SLinus Torvalds if (error) 38271da177e4SLinus Torvalds return error; 38281da177e4SLinus Torvalds 3829acfa4380SAl Viro if (!dir->i_op->unlink) 38301da177e4SLinus Torvalds return -EPERM; 38311da177e4SLinus Torvalds 38325955102cSAl Viro inode_lock(target); 38338ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38341da177e4SLinus Torvalds error = -EBUSY; 38351da177e4SLinus Torvalds else { 38361da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3837bec1052eSAl Viro if (!error) { 38385a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38395a14696cSJ. Bruce Fields if (error) 3840b21996e3SJ. Bruce Fields goto out; 38411da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38428ed936b5SEric W. Biederman if (!error) { 3843d83c49f3SAl Viro dont_mount(dentry); 38448ed936b5SEric W. Biederman detach_mounts(dentry); 3845116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 38468ed936b5SEric W. Biederman } 3847bec1052eSAl Viro } 38481da177e4SLinus Torvalds } 3849b21996e3SJ. Bruce Fields out: 38505955102cSAl Viro inode_unlock(target); 38511da177e4SLinus Torvalds 38521da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38531da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38549accbb97SJ. Bruce Fields fsnotify_link_count(target); 38551da177e4SLinus Torvalds d_delete(dentry); 38561da177e4SLinus Torvalds } 38570eeca283SRobert Love 38581da177e4SLinus Torvalds return error; 38591da177e4SLinus Torvalds } 38604d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38611da177e4SLinus Torvalds 38621da177e4SLinus Torvalds /* 38631da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38641b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38651da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38661da177e4SLinus Torvalds * while waiting on the I/O. 38671da177e4SLinus Torvalds */ 3868da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 38691da177e4SLinus Torvalds { 38702ad94ae6SAl Viro int error; 38711da177e4SLinus Torvalds struct dentry *dentry; 3872f5beed75SAl Viro struct path path; 3873f5beed75SAl Viro struct qstr last; 3874f5beed75SAl Viro int type; 38751da177e4SLinus Torvalds struct inode *inode = NULL; 3876b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38775d18f813SJeff Layton unsigned int lookup_flags = 0; 38785d18f813SJeff Layton retry: 3879da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 388091a27b2aSJeff Layton if (IS_ERR(name)) 388191a27b2aSJeff Layton return PTR_ERR(name); 38822ad94ae6SAl Viro 38831da177e4SLinus Torvalds error = -EISDIR; 3884f5beed75SAl Viro if (type != LAST_NORM) 38851da177e4SLinus Torvalds goto exit1; 38860612d9fbSOGAWA Hirofumi 3887f5beed75SAl Viro error = mnt_want_write(path.mnt); 3888c30dabfeSJan Kara if (error) 3889c30dabfeSJan Kara goto exit1; 3890b21996e3SJ. Bruce Fields retry_deleg: 38915955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3892f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 38931da177e4SLinus Torvalds error = PTR_ERR(dentry); 38941da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 38951da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3896f5beed75SAl Viro if (last.name[last.len]) 389750338b88STörök Edwin goto slashes; 38981da177e4SLinus Torvalds inode = dentry->d_inode; 3899b18825a7SDavid Howells if (d_is_negative(dentry)) 3900e6bc45d6STheodore Ts'o goto slashes; 39017de9c6eeSAl Viro ihold(inode); 3902f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3903be6d3e56SKentaro Takeda if (error) 3904c30dabfeSJan Kara goto exit2; 3905f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 39061da177e4SLinus Torvalds exit2: 39071da177e4SLinus Torvalds dput(dentry); 39081da177e4SLinus Torvalds } 39095955102cSAl Viro inode_unlock(path.dentry->d_inode); 39101da177e4SLinus Torvalds if (inode) 39111da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3912b21996e3SJ. Bruce Fields inode = NULL; 3913b21996e3SJ. Bruce Fields if (delegated_inode) { 39145a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3915b21996e3SJ. Bruce Fields if (!error) 3916b21996e3SJ. Bruce Fields goto retry_deleg; 3917b21996e3SJ. Bruce Fields } 3918f5beed75SAl Viro mnt_drop_write(path.mnt); 39191da177e4SLinus Torvalds exit1: 3920f5beed75SAl Viro path_put(&path); 39215d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39225d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 39235d18f813SJeff Layton inode = NULL; 39245d18f813SJeff Layton goto retry; 39255d18f813SJeff Layton } 3926da2f1362SChristoph Hellwig putname(name); 39271da177e4SLinus Torvalds return error; 39281da177e4SLinus Torvalds 39291da177e4SLinus Torvalds slashes: 3930b18825a7SDavid Howells if (d_is_negative(dentry)) 3931b18825a7SDavid Howells error = -ENOENT; 393244b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3933b18825a7SDavid Howells error = -EISDIR; 3934b18825a7SDavid Howells else 3935b18825a7SDavid Howells error = -ENOTDIR; 39361da177e4SLinus Torvalds goto exit2; 39371da177e4SLinus Torvalds } 39381da177e4SLinus Torvalds 39392e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39405590ff0dSUlrich Drepper { 39415590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39425590ff0dSUlrich Drepper return -EINVAL; 39435590ff0dSUlrich Drepper 39445590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39455590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39465590ff0dSUlrich Drepper 3947da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 39485590ff0dSUlrich Drepper } 39495590ff0dSUlrich Drepper 39503480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39515590ff0dSUlrich Drepper { 3952da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 39535590ff0dSUlrich Drepper } 39545590ff0dSUlrich Drepper 3955db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39561da177e4SLinus Torvalds { 3957a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39581da177e4SLinus Torvalds 39591da177e4SLinus Torvalds if (error) 39601da177e4SLinus Torvalds return error; 39611da177e4SLinus Torvalds 3962acfa4380SAl Viro if (!dir->i_op->symlink) 39631da177e4SLinus Torvalds return -EPERM; 39641da177e4SLinus Torvalds 39651da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39661da177e4SLinus Torvalds if (error) 39671da177e4SLinus Torvalds return error; 39681da177e4SLinus Torvalds 39691da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3970a74574aaSStephen Smalley if (!error) 3971f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39721da177e4SLinus Torvalds return error; 39731da177e4SLinus Torvalds } 39744d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39751da177e4SLinus Torvalds 3976b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 3977b724e846SDominik Brodowski const char __user *newname) 39781da177e4SLinus Torvalds { 39792ad94ae6SAl Viro int error; 398091a27b2aSJeff Layton struct filename *from; 39816902d925SDave Hansen struct dentry *dentry; 3982dae6ad8fSAl Viro struct path path; 3983f46d3567SJeff Layton unsigned int lookup_flags = 0; 39841da177e4SLinus Torvalds 39851da177e4SLinus Torvalds from = getname(oldname); 39861da177e4SLinus Torvalds if (IS_ERR(from)) 39871da177e4SLinus Torvalds return PTR_ERR(from); 3988f46d3567SJeff Layton retry: 3989f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 39901da177e4SLinus Torvalds error = PTR_ERR(dentry); 39916902d925SDave Hansen if (IS_ERR(dentry)) 3992dae6ad8fSAl Viro goto out_putname; 39936902d925SDave Hansen 399491a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 3995a8104a9fSAl Viro if (!error) 399691a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 3997921a1650SAl Viro done_path_create(&path, dentry); 3998f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 3999f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4000f46d3567SJeff Layton goto retry; 4001f46d3567SJeff Layton } 40026902d925SDave Hansen out_putname: 40031da177e4SLinus Torvalds putname(from); 40041da177e4SLinus Torvalds return error; 40051da177e4SLinus Torvalds } 40061da177e4SLinus Torvalds 4007b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4008b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4009b724e846SDominik Brodowski { 4010b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4011b724e846SDominik Brodowski } 4012b724e846SDominik Brodowski 40133480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40145590ff0dSUlrich Drepper { 4015b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 40165590ff0dSUlrich Drepper } 40175590ff0dSUlrich Drepper 4018146a8595SJ. Bruce Fields /** 4019146a8595SJ. Bruce Fields * vfs_link - create a new link 4020146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4021146a8595SJ. Bruce Fields * @dir: new parent 4022146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4023146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4024146a8595SJ. Bruce Fields * 4025146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4026146a8595SJ. Bruce Fields * 4027146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4028146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4029146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4030146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4031146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4032146a8595SJ. Bruce Fields * 4033146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4034146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4035146a8595SJ. Bruce Fields * to be NFS exported. 4036146a8595SJ. Bruce Fields */ 4037146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40381da177e4SLinus Torvalds { 40391da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40408de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40411da177e4SLinus Torvalds int error; 40421da177e4SLinus Torvalds 40431da177e4SLinus Torvalds if (!inode) 40441da177e4SLinus Torvalds return -ENOENT; 40451da177e4SLinus Torvalds 4046a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40471da177e4SLinus Torvalds if (error) 40481da177e4SLinus Torvalds return error; 40491da177e4SLinus Torvalds 40501da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40511da177e4SLinus Torvalds return -EXDEV; 40521da177e4SLinus Torvalds 40531da177e4SLinus Torvalds /* 40541da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40551da177e4SLinus Torvalds */ 40561da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40571da177e4SLinus Torvalds return -EPERM; 40580bd23d09SEric W. Biederman /* 40590bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 40600bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 40610bd23d09SEric W. Biederman * the vfs. 40620bd23d09SEric W. Biederman */ 40630bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 40640bd23d09SEric W. Biederman return -EPERM; 4065acfa4380SAl Viro if (!dir->i_op->link) 40661da177e4SLinus Torvalds return -EPERM; 40677e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40681da177e4SLinus Torvalds return -EPERM; 40691da177e4SLinus Torvalds 40701da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40711da177e4SLinus Torvalds if (error) 40721da177e4SLinus Torvalds return error; 40731da177e4SLinus Torvalds 40745955102cSAl Viro inode_lock(inode); 4075aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4076f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4077aae8a97dSAneesh Kumar K.V error = -ENOENT; 40788de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40798de52778SAl Viro error = -EMLINK; 4080146a8595SJ. Bruce Fields else { 4081146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4082146a8595SJ. Bruce Fields if (!error) 40831da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4084146a8595SJ. Bruce Fields } 4085f4e0c30cSAl Viro 4086f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4087f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4088f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4089f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4090f4e0c30cSAl Viro } 40915955102cSAl Viro inode_unlock(inode); 4092e31e14ecSStephen Smalley if (!error) 40937e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 40941da177e4SLinus Torvalds return error; 40951da177e4SLinus Torvalds } 40964d359507SAl Viro EXPORT_SYMBOL(vfs_link); 40971da177e4SLinus Torvalds 40981da177e4SLinus Torvalds /* 40991da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41001da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41011da177e4SLinus Torvalds * newname. --KAB 41021da177e4SLinus Torvalds * 41031da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41041da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 41051da177e4SLinus Torvalds * and other special files. --ADM 41061da177e4SLinus Torvalds */ 410746ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 410846ea89ebSDominik Brodowski const char __user *newname, int flags) 41091da177e4SLinus Torvalds { 41101da177e4SLinus Torvalds struct dentry *new_dentry; 4111dae6ad8fSAl Viro struct path old_path, new_path; 4112146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 411311a7b371SAneesh Kumar K.V int how = 0; 41141da177e4SLinus Torvalds int error; 41151da177e4SLinus Torvalds 411611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4117c04030e1SUlrich Drepper return -EINVAL; 411811a7b371SAneesh Kumar K.V /* 4119f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4120f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4121f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 412211a7b371SAneesh Kumar K.V */ 4123f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4124f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4125f0cc6ffbSLinus Torvalds return -ENOENT; 412611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4127f0cc6ffbSLinus Torvalds } 4128c04030e1SUlrich Drepper 412911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 413011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4131442e31caSJeff Layton retry: 413211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 41331da177e4SLinus Torvalds if (error) 41342ad94ae6SAl Viro return error; 41352ad94ae6SAl Viro 4136442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4137442e31caSJeff Layton (how & LOOKUP_REVAL)); 41381da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41396902d925SDave Hansen if (IS_ERR(new_dentry)) 4140dae6ad8fSAl Viro goto out; 4141dae6ad8fSAl Viro 4142dae6ad8fSAl Viro error = -EXDEV; 4143dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4144dae6ad8fSAl Viro goto out_dput; 4145800179c9SKees Cook error = may_linkat(&old_path); 4146800179c9SKees Cook if (unlikely(error)) 4147800179c9SKees Cook goto out_dput; 4148dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4149be6d3e56SKentaro Takeda if (error) 4150a8104a9fSAl Viro goto out_dput; 4151146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 415275c3f29dSDave Hansen out_dput: 4153921a1650SAl Viro done_path_create(&new_path, new_dentry); 4154146a8595SJ. Bruce Fields if (delegated_inode) { 4155146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4156d22e6338SOleg Drokin if (!error) { 4157d22e6338SOleg Drokin path_put(&old_path); 4158146a8595SJ. Bruce Fields goto retry; 4159146a8595SJ. Bruce Fields } 4160d22e6338SOleg Drokin } 4161442e31caSJeff Layton if (retry_estale(error, how)) { 4162d22e6338SOleg Drokin path_put(&old_path); 4163442e31caSJeff Layton how |= LOOKUP_REVAL; 4164442e31caSJeff Layton goto retry; 4165442e31caSJeff Layton } 41661da177e4SLinus Torvalds out: 41672d8f3038SAl Viro path_put(&old_path); 41681da177e4SLinus Torvalds 41691da177e4SLinus Torvalds return error; 41701da177e4SLinus Torvalds } 41711da177e4SLinus Torvalds 417246ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 417346ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 417446ea89ebSDominik Brodowski { 417546ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 417646ea89ebSDominik Brodowski } 417746ea89ebSDominik Brodowski 41783480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41795590ff0dSUlrich Drepper { 418046ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41815590ff0dSUlrich Drepper } 41825590ff0dSUlrich Drepper 4183bc27027aSMiklos Szeredi /** 4184bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4185bc27027aSMiklos Szeredi * @old_dir: parent of source 4186bc27027aSMiklos Szeredi * @old_dentry: source 4187bc27027aSMiklos Szeredi * @new_dir: parent of destination 4188bc27027aSMiklos Szeredi * @new_dentry: destination 4189bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4190520c8b16SMiklos Szeredi * @flags: rename flags 4191bc27027aSMiklos Szeredi * 4192bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4193bc27027aSMiklos Szeredi * 4194bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4195bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4196bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4197bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4198bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4199bc27027aSMiklos Szeredi * so. 4200bc27027aSMiklos Szeredi * 4201bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4202bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4203bc27027aSMiklos Szeredi * to be NFS exported. 4204bc27027aSMiklos Szeredi * 42051da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42061da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42071da177e4SLinus Torvalds * Problems: 42080117d427SMauro Carvalho Chehab * 4209d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42101da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42111da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4212a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42131da177e4SLinus Torvalds * story. 42146cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42156cedba89SJ. Bruce Fields * and source (if it is not a directory). 42161b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42171da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42181da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4219a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 42201da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 42211da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 42221da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4223a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 42241da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 42251da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 42261da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4227e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 42281b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 42291da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4230c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 42311da177e4SLinus Torvalds * locking]. 42321da177e4SLinus Torvalds */ 42331da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 42348e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4235520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 42361da177e4SLinus Torvalds { 42371da177e4SLinus Torvalds int error; 4238bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4239bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4240bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4241da1ce067SMiklos Szeredi bool new_is_dir = false; 4242da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 424349d31c2fSAl Viro struct name_snapshot old_name; 42441da177e4SLinus Torvalds 42458d3e2936SMiklos Szeredi if (source == target) 42461da177e4SLinus Torvalds return 0; 42471da177e4SLinus Torvalds 42481da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 42491da177e4SLinus Torvalds if (error) 42501da177e4SLinus Torvalds return error; 42511da177e4SLinus Torvalds 4252da1ce067SMiklos Szeredi if (!target) { 4253a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4254da1ce067SMiklos Szeredi } else { 4255da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4256da1ce067SMiklos Szeredi 4257da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42581da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4259da1ce067SMiklos Szeredi else 4260da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4261da1ce067SMiklos Szeredi } 42621da177e4SLinus Torvalds if (error) 42631da177e4SLinus Torvalds return error; 42641da177e4SLinus Torvalds 42652773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 42661da177e4SLinus Torvalds return -EPERM; 42671da177e4SLinus Torvalds 4268bc27027aSMiklos Szeredi /* 4269bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4270bc27027aSMiklos Szeredi * we'll need to flip '..'. 4271bc27027aSMiklos Szeredi */ 4272da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4273da1ce067SMiklos Szeredi if (is_dir) { 4274bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4275bc27027aSMiklos Szeredi if (error) 4276bc27027aSMiklos Szeredi return error; 4277bc27027aSMiklos Szeredi } 4278da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4279da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4280da1ce067SMiklos Szeredi if (error) 4281da1ce067SMiklos Szeredi return error; 4282da1ce067SMiklos Szeredi } 4283da1ce067SMiklos Szeredi } 42840eeca283SRobert Love 42850b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42860b3974ebSMiklos Szeredi flags); 4287bc27027aSMiklos Szeredi if (error) 4288bc27027aSMiklos Szeredi return error; 4289bc27027aSMiklos Szeredi 429049d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4291bc27027aSMiklos Szeredi dget(new_dentry); 4292da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4293bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4294bc27027aSMiklos Szeredi else if (target) 42955955102cSAl Viro inode_lock(target); 4296bc27027aSMiklos Szeredi 4297bc27027aSMiklos Szeredi error = -EBUSY; 42987af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4299bc27027aSMiklos Szeredi goto out; 4300bc27027aSMiklos Szeredi 4301da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4302bc27027aSMiklos Szeredi error = -EMLINK; 4303da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4304bc27027aSMiklos Szeredi goto out; 4305da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4306da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4307da1ce067SMiklos Szeredi goto out; 4308da1ce067SMiklos Szeredi } 4309da1ce067SMiklos Szeredi if (!is_dir) { 4310bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4311bc27027aSMiklos Szeredi if (error) 4312bc27027aSMiklos Szeredi goto out; 4313da1ce067SMiklos Szeredi } 4314da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4315bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4316bc27027aSMiklos Szeredi if (error) 4317bc27027aSMiklos Szeredi goto out; 4318bc27027aSMiklos Szeredi } 4319520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4320520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4321bc27027aSMiklos Szeredi if (error) 4322bc27027aSMiklos Szeredi goto out; 4323bc27027aSMiklos Szeredi 4324da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 43258767712fSAl Viro if (is_dir) { 43268767712fSAl Viro shrink_dcache_parent(new_dentry); 4327bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 43288767712fSAl Viro } 4329bc27027aSMiklos Szeredi dont_mount(new_dentry); 43308ed936b5SEric W. Biederman detach_mounts(new_dentry); 4331bc27027aSMiklos Szeredi } 4332da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4333da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4334bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4335da1ce067SMiklos Szeredi else 4336da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4337da1ce067SMiklos Szeredi } 4338bc27027aSMiklos Szeredi out: 4339da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4340bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4341bc27027aSMiklos Szeredi else if (target) 43425955102cSAl Viro inode_unlock(target); 4343bc27027aSMiklos Szeredi dput(new_dentry); 4344da1ce067SMiklos Szeredi if (!error) { 4345f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4346da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4347da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4348f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4349da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4350da1ce067SMiklos Szeredi } 4351da1ce067SMiklos Szeredi } 435249d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 43530eeca283SRobert Love 43541da177e4SLinus Torvalds return error; 43551da177e4SLinus Torvalds } 43564d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43571da177e4SLinus Torvalds 4358ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4359ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 43601da177e4SLinus Torvalds { 43611da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43621da177e4SLinus Torvalds struct dentry *trap; 4363f5beed75SAl Viro struct path old_path, new_path; 4364f5beed75SAl Viro struct qstr old_last, new_last; 4365f5beed75SAl Viro int old_type, new_type; 43668e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 436791a27b2aSJeff Layton struct filename *from; 436891a27b2aSJeff Layton struct filename *to; 4369f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4370c6a94284SJeff Layton bool should_retry = false; 43712ad94ae6SAl Viro int error; 4372520c8b16SMiklos Szeredi 43730d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4374da1ce067SMiklos Szeredi return -EINVAL; 4375da1ce067SMiklos Szeredi 43760d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43770d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4378520c8b16SMiklos Szeredi return -EINVAL; 4379520c8b16SMiklos Szeredi 43800d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43810d7a8555SMiklos Szeredi return -EPERM; 43820d7a8555SMiklos Szeredi 4383f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4384f5beed75SAl Viro target_flags = 0; 4385f5beed75SAl Viro 4386c6a94284SJeff Layton retry: 4387c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4388c1d4dd27SAl Viro &old_path, &old_last, &old_type); 438991a27b2aSJeff Layton if (IS_ERR(from)) { 439091a27b2aSJeff Layton error = PTR_ERR(from); 43911da177e4SLinus Torvalds goto exit; 439291a27b2aSJeff Layton } 43931da177e4SLinus Torvalds 4394c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4395c1d4dd27SAl Viro &new_path, &new_last, &new_type); 439691a27b2aSJeff Layton if (IS_ERR(to)) { 439791a27b2aSJeff Layton error = PTR_ERR(to); 43981da177e4SLinus Torvalds goto exit1; 439991a27b2aSJeff Layton } 44001da177e4SLinus Torvalds 44011da177e4SLinus Torvalds error = -EXDEV; 4402f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44031da177e4SLinus Torvalds goto exit2; 44041da177e4SLinus Torvalds 44051da177e4SLinus Torvalds error = -EBUSY; 4406f5beed75SAl Viro if (old_type != LAST_NORM) 44071da177e4SLinus Torvalds goto exit2; 44081da177e4SLinus Torvalds 44090a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44100a7c3937SMiklos Szeredi error = -EEXIST; 4411f5beed75SAl Viro if (new_type != LAST_NORM) 44121da177e4SLinus Torvalds goto exit2; 44131da177e4SLinus Torvalds 4414f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4415c30dabfeSJan Kara if (error) 4416c30dabfeSJan Kara goto exit2; 4417c30dabfeSJan Kara 44188e6d782cSJ. Bruce Fields retry_deleg: 4419f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 44201da177e4SLinus Torvalds 4421f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 44221da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 44231da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 44241da177e4SLinus Torvalds goto exit3; 44251da177e4SLinus Torvalds /* source must exist */ 44261da177e4SLinus Torvalds error = -ENOENT; 4427b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44281da177e4SLinus Torvalds goto exit4; 4429f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44301da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44311da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44321da177e4SLinus Torvalds goto exit4; 44330a7c3937SMiklos Szeredi error = -EEXIST; 44340a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44350a7c3937SMiklos Szeredi goto exit5; 4436da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4437da1ce067SMiklos Szeredi error = -ENOENT; 4438da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4439da1ce067SMiklos Szeredi goto exit5; 4440da1ce067SMiklos Szeredi 4441da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4442da1ce067SMiklos Szeredi error = -ENOTDIR; 4443f5beed75SAl Viro if (new_last.name[new_last.len]) 4444da1ce067SMiklos Szeredi goto exit5; 4445da1ce067SMiklos Szeredi } 4446da1ce067SMiklos Szeredi } 44470a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44480a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44490a7c3937SMiklos Szeredi error = -ENOTDIR; 4450f5beed75SAl Viro if (old_last.name[old_last.len]) 44510a7c3937SMiklos Szeredi goto exit5; 4452f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44530a7c3937SMiklos Szeredi goto exit5; 44540a7c3937SMiklos Szeredi } 44550a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44560a7c3937SMiklos Szeredi error = -EINVAL; 44570a7c3937SMiklos Szeredi if (old_dentry == trap) 44580a7c3937SMiklos Szeredi goto exit5; 44591da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4460da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44611da177e4SLinus Torvalds error = -ENOTEMPTY; 44621da177e4SLinus Torvalds if (new_dentry == trap) 44631da177e4SLinus Torvalds goto exit5; 44641da177e4SLinus Torvalds 4465f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4466f5beed75SAl Viro &new_path, new_dentry, flags); 4467be6d3e56SKentaro Takeda if (error) 4468c30dabfeSJan Kara goto exit5; 4469f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4470f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4471520c8b16SMiklos Szeredi &delegated_inode, flags); 44721da177e4SLinus Torvalds exit5: 44731da177e4SLinus Torvalds dput(new_dentry); 44741da177e4SLinus Torvalds exit4: 44751da177e4SLinus Torvalds dput(old_dentry); 44761da177e4SLinus Torvalds exit3: 4477f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44788e6d782cSJ. Bruce Fields if (delegated_inode) { 44798e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44808e6d782cSJ. Bruce Fields if (!error) 44818e6d782cSJ. Bruce Fields goto retry_deleg; 44828e6d782cSJ. Bruce Fields } 4483f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44841da177e4SLinus Torvalds exit2: 4485c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4486c6a94284SJeff Layton should_retry = true; 4487f5beed75SAl Viro path_put(&new_path); 44882ad94ae6SAl Viro putname(to); 44891da177e4SLinus Torvalds exit1: 4490f5beed75SAl Viro path_put(&old_path); 44911da177e4SLinus Torvalds putname(from); 4492c6a94284SJeff Layton if (should_retry) { 4493c6a94284SJeff Layton should_retry = false; 4494c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4495c6a94284SJeff Layton goto retry; 4496c6a94284SJeff Layton } 44972ad94ae6SAl Viro exit: 44981da177e4SLinus Torvalds return error; 44991da177e4SLinus Torvalds } 45001da177e4SLinus Torvalds 4501ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4502ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4503ee81feb6SDominik Brodowski { 4504ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4505ee81feb6SDominik Brodowski } 4506ee81feb6SDominik Brodowski 4507520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4508520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4509520c8b16SMiklos Szeredi { 4510ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4511520c8b16SMiklos Szeredi } 4512520c8b16SMiklos Szeredi 4513a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45145590ff0dSUlrich Drepper { 4515ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45165590ff0dSUlrich Drepper } 45175590ff0dSUlrich Drepper 4518787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4519787fb6bcSMiklos Szeredi { 4520787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4521787fb6bcSMiklos Szeredi if (error) 4522787fb6bcSMiklos Szeredi return error; 4523787fb6bcSMiklos Szeredi 4524787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4525787fb6bcSMiklos Szeredi return -EPERM; 4526787fb6bcSMiklos Szeredi 4527787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4528787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4529787fb6bcSMiklos Szeredi } 4530787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4531787fb6bcSMiklos Szeredi 45325d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 45331da177e4SLinus Torvalds { 45345d826c84SAl Viro int len = PTR_ERR(link); 45351da177e4SLinus Torvalds if (IS_ERR(link)) 45361da177e4SLinus Torvalds goto out; 45371da177e4SLinus Torvalds 45381da177e4SLinus Torvalds len = strlen(link); 45391da177e4SLinus Torvalds if (len > (unsigned) buflen) 45401da177e4SLinus Torvalds len = buflen; 45411da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45421da177e4SLinus Torvalds len = -EFAULT; 45431da177e4SLinus Torvalds out: 45441da177e4SLinus Torvalds return len; 45451da177e4SLinus Torvalds } 45461da177e4SLinus Torvalds 4547d60874cdSMiklos Szeredi /** 4548fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4549fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4550fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4551fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4552fd4a0edfSMiklos Szeredi * 4553fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4554fd4a0edfSMiklos Szeredi * 4555fd4a0edfSMiklos Szeredi * Does not call security hook. 4556fd4a0edfSMiklos Szeredi */ 4557fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4558fd4a0edfSMiklos Szeredi { 4559fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4560f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4561f2df5da6SAl Viro const char *link; 4562f2df5da6SAl Viro int res; 4563fd4a0edfSMiklos Szeredi 456476fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 456576fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 456676fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 456776fca90eSMiklos Szeredi 456876fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4569fd4a0edfSMiklos Szeredi return -EINVAL; 4570fd4a0edfSMiklos Szeredi 457176fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 457276fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 457376fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 457476fca90eSMiklos Szeredi } 457576fca90eSMiklos Szeredi 45764c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4577f2df5da6SAl Viro if (!link) { 4578f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4579f2df5da6SAl Viro if (IS_ERR(link)) 4580f2df5da6SAl Viro return PTR_ERR(link); 4581f2df5da6SAl Viro } 4582f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4583f2df5da6SAl Viro do_delayed_call(&done); 4584f2df5da6SAl Viro return res; 4585fd4a0edfSMiklos Szeredi } 4586fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 45871da177e4SLinus Torvalds 4588d60874cdSMiklos Szeredi /** 4589d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4590d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4591d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4592d60874cdSMiklos Szeredi * 4593d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4594d60874cdSMiklos Szeredi * 4595d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4596d60874cdSMiklos Szeredi * 4597d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4598d60874cdSMiklos Szeredi */ 4599d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4600d60874cdSMiklos Szeredi { 4601d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4602d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4603d60874cdSMiklos Szeredi 4604d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4605d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4606d60874cdSMiklos Szeredi if (!res) 4607d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4608d60874cdSMiklos Szeredi } 4609d60874cdSMiklos Szeredi return res; 4610d60874cdSMiklos Szeredi } 4611d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4612d60874cdSMiklos Szeredi 46131da177e4SLinus Torvalds /* get the link contents into pagecache */ 46146b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4615fceef393SAl Viro struct delayed_call *callback) 46161da177e4SLinus Torvalds { 4617ebd09abbSDuane Griffin char *kaddr; 46181da177e4SLinus Torvalds struct page *page; 46196b255391SAl Viro struct address_space *mapping = inode->i_mapping; 46206b255391SAl Viro 4621d3883d4fSAl Viro if (!dentry) { 4622d3883d4fSAl Viro page = find_get_page(mapping, 0); 4623d3883d4fSAl Viro if (!page) 46246b255391SAl Viro return ERR_PTR(-ECHILD); 4625d3883d4fSAl Viro if (!PageUptodate(page)) { 4626d3883d4fSAl Viro put_page(page); 4627d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4628d3883d4fSAl Viro } 4629d3883d4fSAl Viro } else { 4630090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 46311da177e4SLinus Torvalds if (IS_ERR(page)) 46326fe6900eSNick Piggin return (char*)page; 4633d3883d4fSAl Viro } 4634fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 463521fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 463621fc61c7SAl Viro kaddr = page_address(page); 46376b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4638ebd09abbSDuane Griffin return kaddr; 46391da177e4SLinus Torvalds } 46401da177e4SLinus Torvalds 46416b255391SAl Viro EXPORT_SYMBOL(page_get_link); 46421da177e4SLinus Torvalds 4643fceef393SAl Viro void page_put_link(void *arg) 46441da177e4SLinus Torvalds { 4645fceef393SAl Viro put_page(arg); 46461da177e4SLinus Torvalds } 46474d359507SAl Viro EXPORT_SYMBOL(page_put_link); 46481da177e4SLinus Torvalds 4649aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4650aa80deabSAl Viro { 4651fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46526b255391SAl Viro int res = readlink_copy(buffer, buflen, 46536b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4654fceef393SAl Viro &done)); 4655fceef393SAl Viro do_delayed_call(&done); 4656aa80deabSAl Viro return res; 4657aa80deabSAl Viro } 4658aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4659aa80deabSAl Viro 466054566b2cSNick Piggin /* 466154566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 466254566b2cSNick Piggin */ 466354566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 46641da177e4SLinus Torvalds { 46651da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 46660adb25d2SKirill Korotaev struct page *page; 4667afddba49SNick Piggin void *fsdata; 4668beb497abSDmitriy Monakhov int err; 4669c718a975STetsuo Handa unsigned int flags = 0; 467054566b2cSNick Piggin if (nofs) 467154566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 46721da177e4SLinus Torvalds 46737e53cac4SNeilBrown retry: 4674afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 467554566b2cSNick Piggin flags, &page, &fsdata); 46761da177e4SLinus Torvalds if (err) 4677afddba49SNick Piggin goto fail; 4678afddba49SNick Piggin 467921fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4680afddba49SNick Piggin 4681afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4682afddba49SNick Piggin page, fsdata); 46831da177e4SLinus Torvalds if (err < 0) 46841da177e4SLinus Torvalds goto fail; 4685afddba49SNick Piggin if (err < len-1) 4686afddba49SNick Piggin goto retry; 4687afddba49SNick Piggin 46881da177e4SLinus Torvalds mark_inode_dirty(inode); 46891da177e4SLinus Torvalds return 0; 46901da177e4SLinus Torvalds fail: 46911da177e4SLinus Torvalds return err; 46921da177e4SLinus Torvalds } 46934d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 46941da177e4SLinus Torvalds 46950adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 46960adb25d2SKirill Korotaev { 46970adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4698c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 46990adb25d2SKirill Korotaev } 47004d359507SAl Viro EXPORT_SYMBOL(page_symlink); 47010adb25d2SKirill Korotaev 470292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 47036b255391SAl Viro .get_link = page_get_link, 47041da177e4SLinus Torvalds }; 47051da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4706