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 5697973387aSAl Viro static void drop_links(struct nameidata *nd) 5707973387aSAl Viro { 5717973387aSAl Viro int i = nd->depth; 5727973387aSAl Viro while (i--) { 5737973387aSAl Viro struct saved *last = nd->stack + i; 574fceef393SAl Viro do_delayed_call(&last->done); 575fceef393SAl Viro clear_delayed_call(&last->done); 5767973387aSAl Viro } 5777973387aSAl Viro } 5787973387aSAl Viro 5797973387aSAl Viro static void terminate_walk(struct nameidata *nd) 5807973387aSAl Viro { 5817973387aSAl Viro drop_links(nd); 5827973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 5837973387aSAl Viro int i; 5847973387aSAl Viro path_put(&nd->path); 5857973387aSAl Viro for (i = 0; i < nd->depth; i++) 5867973387aSAl Viro path_put(&nd->stack[i].link); 58784a2bd39SAl Viro if (nd->flags & LOOKUP_ROOT_GRABBED) { 588102b8af2SAl Viro path_put(&nd->root); 58984a2bd39SAl Viro nd->flags &= ~LOOKUP_ROOT_GRABBED; 590102b8af2SAl Viro } 5917973387aSAl Viro } else { 5927973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 5937973387aSAl Viro rcu_read_unlock(); 5947973387aSAl Viro } 5957973387aSAl Viro nd->depth = 0; 5967973387aSAl Viro } 5977973387aSAl Viro 5987973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 5992aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6007973387aSAl Viro { 6012aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6027973387aSAl Viro if (unlikely(res)) { 6037973387aSAl Viro if (res > 0) 6047973387aSAl Viro path->mnt = NULL; 6057973387aSAl Viro path->dentry = NULL; 6067973387aSAl Viro return false; 6077973387aSAl Viro } 6087973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6097973387aSAl Viro path->dentry = NULL; 6107973387aSAl Viro return false; 6117973387aSAl Viro } 6127973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6137973387aSAl Viro } 6147973387aSAl Viro 6152aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 6162aa38470SAl Viro struct path *path, unsigned seq) 6172aa38470SAl Viro { 6182aa38470SAl Viro return __legitimize_path(path, nd->m_seq, seq); 6192aa38470SAl Viro } 6202aa38470SAl Viro 6217973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6227973387aSAl Viro { 6237973387aSAl Viro int i; 6247973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6257973387aSAl Viro struct saved *last = nd->stack + i; 6267973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6277973387aSAl Viro drop_links(nd); 6287973387aSAl Viro nd->depth = i + 1; 6297973387aSAl Viro return false; 6307973387aSAl Viro } 6317973387aSAl Viro } 6327973387aSAl Viro return true; 6337973387aSAl Viro } 6347973387aSAl Viro 635ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 636ee594bffSAl Viro { 637adb21d2bSAleksa Sarai /* 638adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 639adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 640adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 641adb21d2bSAleksa Sarai */ 642adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 643adb21d2bSAleksa Sarai return false; 644adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 645ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 646ee594bffSAl Viro return true; 64784a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 648ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 649ee594bffSAl Viro } 650ee594bffSAl Viro 65119660af7SAl Viro /* 65231e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 65319660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 65419660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 65557e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 65619660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 65719660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 65819660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 65919660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 66031e6b01fSNick Piggin */ 66131e6b01fSNick Piggin 66231e6b01fSNick Piggin /** 66319660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 66419660af7SAl Viro * @nd: nameidata pathwalk data 66539191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 66631e6b01fSNick Piggin * 6674675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 6684675ac39SAl Viro * for ref-walk mode. 6694675ac39SAl Viro * Must be called from rcu-walk context. 6707973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 6717973387aSAl Viro * terminate_walk(). 67231e6b01fSNick Piggin */ 6734675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 67431e6b01fSNick Piggin { 67531e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 67631e6b01fSNick Piggin 67731e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 678e5c832d5SLinus Torvalds 679e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 6807973387aSAl Viro if (unlikely(!legitimize_links(nd))) 6814675ac39SAl Viro goto out1; 68284a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 68384a2bd39SAl Viro goto out; 684ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 6854675ac39SAl Viro goto out; 6864675ac39SAl Viro rcu_read_unlock(); 6874675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 6884675ac39SAl Viro return 0; 6894675ac39SAl Viro 69084a2bd39SAl Viro out1: 6914675ac39SAl Viro nd->path.mnt = NULL; 6924675ac39SAl Viro nd->path.dentry = NULL; 6934675ac39SAl Viro out: 6944675ac39SAl Viro rcu_read_unlock(); 6954675ac39SAl Viro return -ECHILD; 6964675ac39SAl Viro } 6974675ac39SAl Viro 6984675ac39SAl Viro /** 6994675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7004675ac39SAl Viro * @nd: nameidata pathwalk data 7014675ac39SAl Viro * @dentry: child of nd->path.dentry 7024675ac39SAl Viro * @seq: seq number to check dentry against 7034675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7044675ac39SAl Viro * 7054675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7064675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7074675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7084675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7094675ac39SAl Viro * terminate_walk(). 7104675ac39SAl Viro */ 7114675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7124675ac39SAl Viro { 7134675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7144675ac39SAl Viro 7154675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7164675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7174675ac39SAl Viro goto out2; 7187973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7197973387aSAl Viro goto out2; 7204675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7217973387aSAl Viro goto out1; 72248a066e7SAl Viro 72315570086SLinus Torvalds /* 7244675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7254675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7264675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7274675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7284675ac39SAl Viro * know the parent must still be valid if the child sequence number is 72915570086SLinus Torvalds */ 7304675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 731e5c832d5SLinus Torvalds goto out; 73284a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 73384a2bd39SAl Viro goto out_dput; 734e5c832d5SLinus Torvalds /* 735e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 736e5c832d5SLinus Torvalds * still valid and get it if required. 737e5c832d5SLinus Torvalds */ 73884a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 73984a2bd39SAl Viro goto out_dput; 7408b61e74fSAl Viro rcu_read_unlock(); 74131e6b01fSNick Piggin return 0; 74219660af7SAl Viro 7437973387aSAl Viro out2: 7447973387aSAl Viro nd->path.mnt = NULL; 7457973387aSAl Viro out1: 7467973387aSAl Viro nd->path.dentry = NULL; 747e5c832d5SLinus Torvalds out: 7488b61e74fSAl Viro rcu_read_unlock(); 74984a2bd39SAl Viro return -ECHILD; 75084a2bd39SAl Viro out_dput: 75184a2bd39SAl Viro rcu_read_unlock(); 75284a2bd39SAl Viro dput(dentry); 75331e6b01fSNick Piggin return -ECHILD; 75431e6b01fSNick Piggin } 75531e6b01fSNick Piggin 7564ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 75734286d66SNick Piggin { 758a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 7594ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 760a89f8337SAl Viro else 761a89f8337SAl Viro return 1; 76234286d66SNick Piggin } 76334286d66SNick Piggin 7649f1fafeeSAl Viro /** 7659f1fafeeSAl Viro * complete_walk - successful completion of path walk 7669f1fafeeSAl Viro * @nd: pointer nameidata 76739159de2SJeff Layton * 7689f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 7699f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 7709f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 7719f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 7729f1fafeeSAl Viro * need to drop nd->path. 77339159de2SJeff Layton */ 7749f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 77539159de2SJeff Layton { 77616c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 77739159de2SJeff Layton int status; 77839159de2SJeff Layton 7799f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 780adb21d2bSAleksa Sarai /* 781adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 782adb21d2bSAleksa Sarai * externally-managed nd->root. 783adb21d2bSAleksa Sarai */ 784adb21d2bSAleksa Sarai if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) 7859f1fafeeSAl Viro nd->root.mnt = NULL; 7864675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 78748a066e7SAl Viro return -ECHILD; 78848a066e7SAl Viro } 7899f1fafeeSAl Viro 790adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 791adb21d2bSAleksa Sarai /* 792adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 793adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 794adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 795adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 796adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 797adb21d2bSAleksa Sarai * 798adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 799adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 800adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 801adb21d2bSAleksa Sarai * requested root to userspace. 802adb21d2bSAleksa Sarai * 803adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 804adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 805adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 806adb21d2bSAleksa Sarai */ 807adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 808adb21d2bSAleksa Sarai return -EXDEV; 809adb21d2bSAleksa Sarai } 810adb21d2bSAleksa Sarai 81116c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 81239159de2SJeff Layton return 0; 81339159de2SJeff Layton 814ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 81516c2cd71SAl Viro return 0; 81616c2cd71SAl Viro 817ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 81839159de2SJeff Layton if (status > 0) 81939159de2SJeff Layton return 0; 82039159de2SJeff Layton 82116c2cd71SAl Viro if (!status) 82239159de2SJeff Layton status = -ESTALE; 82316c2cd71SAl Viro 82439159de2SJeff Layton return status; 82539159de2SJeff Layton } 82639159de2SJeff Layton 827740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 8282a737871SAl Viro { 82931e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8309e6697e2SAl Viro 831adb21d2bSAleksa Sarai /* 832adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 833adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 834adb21d2bSAleksa Sarai * from the dirfd. 835adb21d2bSAleksa Sarai */ 836adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 837adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 838adb21d2bSAleksa Sarai 8399e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8408f47a016SAl Viro unsigned seq; 841c28cc364SNick Piggin 842c28cc364SNick Piggin do { 843c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 84431e6b01fSNick Piggin nd->root = fs->root; 8458f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 846c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 8479e6697e2SAl Viro } else { 8489e6697e2SAl Viro get_fs_root(fs, &nd->root); 84984a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 8509e6697e2SAl Viro } 851740a1678SAleksa Sarai return 0; 85231e6b01fSNick Piggin } 85331e6b01fSNick Piggin 854248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 855248fb5b9SAl Viro { 856adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 857adb21d2bSAleksa Sarai return -EXDEV; 85872ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 85972ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 86072ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 86172ba2929SAleksa Sarai return -EXDEV; 86272ba2929SAleksa Sarai } 863740a1678SAleksa Sarai if (!nd->root.mnt) { 864740a1678SAleksa Sarai int error = set_root(nd); 865740a1678SAleksa Sarai if (error) 866740a1678SAleksa Sarai return error; 867740a1678SAleksa Sarai } 868248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 869248fb5b9SAl Viro struct dentry *d; 870248fb5b9SAl Viro nd->path = nd->root; 871248fb5b9SAl Viro d = nd->path.dentry; 872248fb5b9SAl Viro nd->inode = d->d_inode; 873248fb5b9SAl Viro nd->seq = nd->root_seq; 874248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 875248fb5b9SAl Viro return -ECHILD; 876248fb5b9SAl Viro } else { 877248fb5b9SAl Viro path_put(&nd->path); 878248fb5b9SAl Viro nd->path = nd->root; 879248fb5b9SAl Viro path_get(&nd->path); 880248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 881248fb5b9SAl Viro } 882248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 883248fb5b9SAl Viro return 0; 884248fb5b9SAl Viro } 885248fb5b9SAl Viro 886b5fb63c1SChristoph Hellwig /* 8876b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 888b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 889b5fb63c1SChristoph Hellwig */ 8901bc82070SAleksa Sarai int nd_jump_link(struct path *path) 891b5fb63c1SChristoph Hellwig { 8924b99d499SAleksa Sarai int error = -ELOOP; 8936e77137bSAl Viro struct nameidata *nd = current->nameidata; 894b5fb63c1SChristoph Hellwig 8954b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 8964b99d499SAleksa Sarai goto err; 8974b99d499SAleksa Sarai 89872ba2929SAleksa Sarai error = -EXDEV; 89972ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 90072ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 90172ba2929SAleksa Sarai goto err; 90272ba2929SAleksa Sarai } 903adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 904adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 905adb21d2bSAleksa Sarai goto err; 90672ba2929SAleksa Sarai 9074b99d499SAleksa Sarai path_put(&nd->path); 908b5fb63c1SChristoph Hellwig nd->path = *path; 909b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 910b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 9111bc82070SAleksa Sarai return 0; 9124b99d499SAleksa Sarai 9134b99d499SAleksa Sarai err: 9144b99d499SAleksa Sarai path_put(path); 9154b99d499SAleksa Sarai return error; 916b5fb63c1SChristoph Hellwig } 917b5fb63c1SChristoph Hellwig 918b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 919574197e0SAl Viro { 92021c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 921fceef393SAl Viro do_delayed_call(&last->done); 9226548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 923b9ff4429SAl Viro path_put(&last->link); 924574197e0SAl Viro } 925574197e0SAl Viro 926561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 927561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 92830aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 92930aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 930800179c9SKees Cook 931800179c9SKees Cook /** 932800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 93355852635SRandy Dunlap * @nd: nameidata pathwalk data 934800179c9SKees Cook * 935800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 936800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 937800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 938800179c9SKees Cook * processes from failing races against path names that may change out 939800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 940800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 941800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 942800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 943800179c9SKees Cook * 944800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 945800179c9SKees Cook */ 946ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 947800179c9SKees Cook { 948800179c9SKees Cook const struct inode *parent; 9492d7f9e2aSSeth Forshee kuid_t puid; 950800179c9SKees Cook 951800179c9SKees Cook if (!sysctl_protected_symlinks) 952800179c9SKees Cook return 0; 953800179c9SKees Cook 954800179c9SKees Cook /* Allowed if owner and follower match. */ 95581abe27bSEric W. Biederman if (uid_eq(current_cred()->fsuid, inode->i_uid)) 956800179c9SKees Cook return 0; 957800179c9SKees Cook 958800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 959aa65fa35SAl Viro parent = nd->inode; 960800179c9SKees Cook if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 961800179c9SKees Cook return 0; 962800179c9SKees Cook 963800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 9642d7f9e2aSSeth Forshee puid = parent->i_uid; 9652d7f9e2aSSeth Forshee if (uid_valid(puid) && uid_eq(puid, inode->i_uid)) 966800179c9SKees Cook return 0; 967800179c9SKees Cook 96831956502SAl Viro if (nd->flags & LOOKUP_RCU) 96931956502SAl Viro return -ECHILD; 97031956502SAl Viro 971ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 972245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 973800179c9SKees Cook return -EACCES; 974800179c9SKees Cook } 975800179c9SKees Cook 976800179c9SKees Cook /** 977800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 978800179c9SKees Cook * @inode: the source inode to hardlink from 979800179c9SKees Cook * 980800179c9SKees Cook * Return false if at least one of the following conditions: 981800179c9SKees Cook * - inode is not a regular file 982800179c9SKees Cook * - inode is setuid 983800179c9SKees Cook * - inode is setgid and group-exec 984800179c9SKees Cook * - access failure for read and write 985800179c9SKees Cook * 986800179c9SKees Cook * Otherwise returns true. 987800179c9SKees Cook */ 988800179c9SKees Cook static bool safe_hardlink_source(struct inode *inode) 989800179c9SKees Cook { 990800179c9SKees Cook umode_t mode = inode->i_mode; 991800179c9SKees Cook 992800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 993800179c9SKees Cook if (!S_ISREG(mode)) 994800179c9SKees Cook return false; 995800179c9SKees Cook 996800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 997800179c9SKees Cook if (mode & S_ISUID) 998800179c9SKees Cook return false; 999800179c9SKees Cook 1000800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1001800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1002800179c9SKees Cook return false; 1003800179c9SKees Cook 1004800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1005800179c9SKees Cook if (inode_permission(inode, MAY_READ | MAY_WRITE)) 1006800179c9SKees Cook return false; 1007800179c9SKees Cook 1008800179c9SKees Cook return true; 1009800179c9SKees Cook } 1010800179c9SKees Cook 1011800179c9SKees Cook /** 1012800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1013800179c9SKees Cook * @link: the source to hardlink from 1014800179c9SKees Cook * 1015800179c9SKees Cook * Block hardlink when all of: 1016800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1017800179c9SKees Cook * - fsuid does not match inode 1018800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1019f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1020800179c9SKees Cook * 1021800179c9SKees Cook * Returns 0 if successful, -ve on error. 1022800179c9SKees Cook */ 1023800179c9SKees Cook static int may_linkat(struct path *link) 1024800179c9SKees Cook { 1025593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1026593d1ce8SEric W. Biederman 1027593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1028593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 1029593d1ce8SEric W. Biederman return -EOVERFLOW; 1030800179c9SKees Cook 1031800179c9SKees Cook if (!sysctl_protected_hardlinks) 1032800179c9SKees Cook return 0; 1033800179c9SKees Cook 1034800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1035800179c9SKees Cook * otherwise, it must be a safe source. 1036800179c9SKees Cook */ 1037cc658db4SKees Cook if (safe_hardlink_source(inode) || inode_owner_or_capable(inode)) 1038800179c9SKees Cook return 0; 1039800179c9SKees Cook 1040245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1041800179c9SKees Cook return -EPERM; 1042800179c9SKees Cook } 1043800179c9SKees Cook 104430aba665SSalvatore Mesoraca /** 104530aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 104630aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 104730aba665SSalvatore Mesoraca * exist. 1048d0cb5018SAl Viro * @dir_mode: mode bits of directory 1049d0cb5018SAl Viro * @dir_uid: owner of directory 105030aba665SSalvatore Mesoraca * @inode: the inode of the file to open 105130aba665SSalvatore Mesoraca * 105230aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 105330aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 105430aba665SSalvatore Mesoraca * - the file already exists 105530aba665SSalvatore Mesoraca * - we are in a sticky directory 105630aba665SSalvatore Mesoraca * - we don't own the file 105730aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 105830aba665SSalvatore Mesoraca * - the directory is world writable 105930aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 106030aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 106130aba665SSalvatore Mesoraca * be enough. 106230aba665SSalvatore Mesoraca * 106330aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 106430aba665SSalvatore Mesoraca */ 1065d0cb5018SAl Viro static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid, 106630aba665SSalvatore Mesoraca struct inode * const inode) 106730aba665SSalvatore Mesoraca { 106830aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 106930aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1070d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1071d0cb5018SAl Viro uid_eq(inode->i_uid, dir_uid) || 107230aba665SSalvatore Mesoraca uid_eq(current_fsuid(), inode->i_uid)) 107330aba665SSalvatore Mesoraca return 0; 107430aba665SSalvatore Mesoraca 1075d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1076d0cb5018SAl Viro (dir_mode & 0020 && 107730aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 107830aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1079245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1080245d7369SKees Cook "sticky_create_fifo" : 1081245d7369SKees Cook "sticky_create_regular"; 1082245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 108330aba665SSalvatore Mesoraca return -EACCES; 108430aba665SSalvatore Mesoraca } 108530aba665SSalvatore Mesoraca return 0; 108630aba665SSalvatore Mesoraca } 108730aba665SSalvatore Mesoraca 1088f015f126SDavid Howells /* 1089f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1090f015f126SDavid Howells * 1091f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1092f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1093f015f126SDavid Howells * Up is towards /. 1094f015f126SDavid Howells * 1095f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1096f015f126SDavid Howells * root. 1097f015f126SDavid Howells */ 1098bab77ebfSAl Viro int follow_up(struct path *path) 10991da177e4SLinus Torvalds { 11000714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11010714a533SAl Viro struct mount *parent; 11021da177e4SLinus Torvalds struct dentry *mountpoint; 110399b7db7bSNick Piggin 110448a066e7SAl Viro read_seqlock_excl(&mount_lock); 11050714a533SAl Viro parent = mnt->mnt_parent; 11063c0a6163SAl Viro if (parent == mnt) { 110748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11081da177e4SLinus Torvalds return 0; 11091da177e4SLinus Torvalds } 11100714a533SAl Viro mntget(&parent->mnt); 1111a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 111248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1113bab77ebfSAl Viro dput(path->dentry); 1114bab77ebfSAl Viro path->dentry = mountpoint; 1115bab77ebfSAl Viro mntput(path->mnt); 11160714a533SAl Viro path->mnt = &parent->mnt; 11171da177e4SLinus Torvalds return 1; 11181da177e4SLinus Torvalds } 11194d359507SAl Viro EXPORT_SYMBOL(follow_up); 11201da177e4SLinus Torvalds 11217ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 11227ef482faSAl Viro struct path *path, unsigned *seqp) 11237ef482faSAl Viro { 11247ef482faSAl Viro while (mnt_has_parent(m)) { 11257ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 11267ef482faSAl Viro 11277ef482faSAl Viro m = m->mnt_parent; 11287ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 11297ef482faSAl Viro root->mnt == &m->mnt)) 11307ef482faSAl Viro break; 11317ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 11327ef482faSAl Viro path->mnt = &m->mnt; 11337ef482faSAl Viro path->dentry = mountpoint; 11347ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 11357ef482faSAl Viro return true; 11367ef482faSAl Viro } 11377ef482faSAl Viro } 11387ef482faSAl Viro return false; 11397ef482faSAl Viro } 11407ef482faSAl Viro 11412aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 11422aa38470SAl Viro struct path *path) 11432aa38470SAl Viro { 11442aa38470SAl Viro bool found; 11452aa38470SAl Viro 11462aa38470SAl Viro rcu_read_lock(); 11472aa38470SAl Viro while (1) { 11482aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 11492aa38470SAl Viro 11502aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 11512aa38470SAl Viro if (unlikely(!found)) { 11522aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 11532aa38470SAl Viro break; 11542aa38470SAl Viro } else { 11552aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 11562aa38470SAl Viro break; 11572aa38470SAl Viro rcu_read_unlock(); 11582aa38470SAl Viro path_put(path); 11592aa38470SAl Viro rcu_read_lock(); 11602aa38470SAl Viro } 11612aa38470SAl Viro } 11622aa38470SAl Viro rcu_read_unlock(); 11632aa38470SAl Viro return found; 11642aa38470SAl Viro } 11652aa38470SAl Viro 1166b5c84bf6SNick Piggin /* 11679875cf80SDavid Howells * Perform an automount 11689875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 11699875cf80SDavid Howells * were called with. 11701da177e4SLinus Torvalds */ 11711c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 117231e6b01fSNick Piggin { 117325e195aaSAl Viro struct dentry *dentry = path->dentry; 11749875cf80SDavid Howells 11750ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 11760ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 11770ec26fd0SMiklos Szeredi * the name. 11780ec26fd0SMiklos Szeredi * 11790ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 11805a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 11810ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 11820ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 11830ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 11840ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 11855a30d8a2SDavid Howells */ 11861c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 11875d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 118825e195aaSAl Viro dentry->d_inode) 11899875cf80SDavid Howells return -EISDIR; 11900ec26fd0SMiklos Szeredi 11911c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 11929875cf80SDavid Howells return -ELOOP; 11939875cf80SDavid Howells 119425e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1195ea5b778aSDavid Howells } 11969875cf80SDavid Howells 11979875cf80SDavid Howells /* 11989deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 11999deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 12009deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 12019deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 12029875cf80SDavid Howells */ 12039deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 12049deed3ebSAl Viro int *count, unsigned lookup_flags) 12059875cf80SDavid Howells { 12069deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 12079875cf80SDavid Howells bool need_mntput = false; 12088aef1884SAl Viro int ret = 0; 12099875cf80SDavid Howells 12109deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1211cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1212cc53ce53SDavid Howells * being held. */ 1213d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1214fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1215508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1216cc53ce53SDavid Howells if (ret < 0) 12178aef1884SAl Viro break; 1218cc53ce53SDavid Howells } 1219cc53ce53SDavid Howells 12209deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 12219875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 12229deed3ebSAl Viro if (mounted) { // ... in our namespace 12239875cf80SDavid Howells dput(path->dentry); 12249875cf80SDavid Howells if (need_mntput) 1225463ffb2eSAl Viro mntput(path->mnt); 1226463ffb2eSAl Viro path->mnt = mounted; 1227463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 12289deed3ebSAl Viro // here we know it's positive 12299deed3ebSAl Viro flags = path->dentry->d_flags; 12309875cf80SDavid Howells need_mntput = true; 12319875cf80SDavid Howells continue; 1232463ffb2eSAl Viro } 12331da177e4SLinus Torvalds } 12349875cf80SDavid Howells 12359deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 12369deed3ebSAl Viro break; 12379deed3ebSAl Viro 12389deed3ebSAl Viro // uncovered automount point 12399deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 12409deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 12419875cf80SDavid Howells if (ret < 0) 12428aef1884SAl Viro break; 12439875cf80SDavid Howells } 12449875cf80SDavid Howells 12459deed3ebSAl Viro if (ret == -EISDIR) 12469deed3ebSAl Viro ret = 0; 12479deed3ebSAl Viro // possible if you race with several mount --move 12489deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 12498aef1884SAl Viro mntput(path->mnt); 12509deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1251d41efb52SAl Viro ret = -ENOENT; 12529deed3ebSAl Viro *jumped = need_mntput; 12538402752eSAl Viro return ret; 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds 12569deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 12579deed3ebSAl Viro int *count, unsigned lookup_flags) 12589deed3ebSAl Viro { 12599deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 12609deed3ebSAl Viro 12619deed3ebSAl Viro /* fastpath */ 12629deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 12639deed3ebSAl Viro *jumped = false; 12649deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 12659deed3ebSAl Viro return -ENOENT; 12669deed3ebSAl Viro return 0; 12679deed3ebSAl Viro } 12689deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 12699deed3ebSAl Viro } 12709deed3ebSAl Viro 1271cc53ce53SDavid Howells int follow_down_one(struct path *path) 12721da177e4SLinus Torvalds { 12731da177e4SLinus Torvalds struct vfsmount *mounted; 12741da177e4SLinus Torvalds 12751c755af4SAl Viro mounted = lookup_mnt(path); 12761da177e4SLinus Torvalds if (mounted) { 12779393bd07SAl Viro dput(path->dentry); 12789393bd07SAl Viro mntput(path->mnt); 12799393bd07SAl Viro path->mnt = mounted; 12809393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 12811da177e4SLinus Torvalds return 1; 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds return 0; 12841da177e4SLinus Torvalds } 12854d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 12861da177e4SLinus Torvalds 12879875cf80SDavid Howells /* 12889deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 12899deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 12909deed3ebSAl Viro * caller is permitted to proceed or not. 12919deed3ebSAl Viro */ 12929deed3ebSAl Viro int follow_down(struct path *path) 12939deed3ebSAl Viro { 12949deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 12959deed3ebSAl Viro bool jumped; 12969deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 12979deed3ebSAl Viro 12989deed3ebSAl Viro if (path->mnt != mnt) 12999deed3ebSAl Viro mntput(mnt); 13009deed3ebSAl Viro return ret; 13019deed3ebSAl Viro } 13029deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 13039deed3ebSAl Viro 13049deed3ebSAl Viro /* 1305287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1306287548e4SAl Viro * we meet a managed dentry that would need blocking. 13079875cf80SDavid Howells */ 13089875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1309254cf582SAl Viro struct inode **inode, unsigned *seqp) 13109875cf80SDavid Howells { 1311ea936aebSAl Viro struct dentry *dentry = path->dentry; 1312ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1313ea936aebSAl Viro 1314ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1315ea936aebSAl Viro return true; 1316ea936aebSAl Viro 1317ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1318ea936aebSAl Viro return false; 1319ea936aebSAl Viro 132062a7375eSIan Kent for (;;) { 132162a7375eSIan Kent /* 132262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 132362a7375eSIan Kent * that wants to block transit. 132462a7375eSIan Kent */ 1325ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1326ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1327ea936aebSAl Viro if (res) 1328ea936aebSAl Viro return res == -EISDIR; 1329ea936aebSAl Viro flags = dentry->d_flags; 1330b8faf035SNeilBrown } 133162a7375eSIan Kent 1332ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1333ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1334ea936aebSAl Viro if (mounted) { 1335c7105365SAl Viro path->mnt = &mounted->mnt; 1336ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1337a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1338ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1339ea936aebSAl Viro *inode = dentry->d_inode; 134059430262SLinus Torvalds /* 1341ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1342ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1343ea936aebSAl Viro * between mount hash removal and ->mnt_root 1344ea936aebSAl Viro * becoming unpinned. 134559430262SLinus Torvalds */ 1346ea936aebSAl Viro flags = dentry->d_flags; 1347ea936aebSAl Viro continue; 13489875cf80SDavid Howells } 1349ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1350ea936aebSAl Viro return false; 1351ea936aebSAl Viro } 1352ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1353ea936aebSAl Viro } 1354287548e4SAl Viro } 1355287548e4SAl Viro 1356db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1357db3c9adeSAl Viro struct path *path, struct inode **inode, 1358db3c9adeSAl Viro unsigned int *seqp) 1359bd7c4b50SAl Viro { 13609deed3ebSAl Viro bool jumped; 1361db3c9adeSAl Viro int ret; 1362bd7c4b50SAl Viro 1363db3c9adeSAl Viro path->mnt = nd->path.mnt; 1364db3c9adeSAl Viro path->dentry = dentry; 1365c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1366c153007bSAl Viro unsigned int seq = *seqp; 1367c153007bSAl Viro if (unlikely(!*inode)) 1368c153007bSAl Viro return -ENOENT; 1369c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 13709deed3ebSAl Viro return 0; 1371c153007bSAl Viro if (unlazy_child(nd, dentry, seq)) 1372c153007bSAl Viro return -ECHILD; 1373c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1374c153007bSAl Viro path->mnt = nd->path.mnt; 1375c153007bSAl Viro path->dentry = dentry; 1376c153007bSAl Viro } 13779deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 13789deed3ebSAl Viro if (jumped) { 13799deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 13809deed3ebSAl Viro ret = -EXDEV; 13819deed3ebSAl Viro else 13829deed3ebSAl Viro nd->flags |= LOOKUP_JUMPED; 13839deed3ebSAl Viro } 13849deed3ebSAl Viro if (unlikely(ret)) { 13859deed3ebSAl Viro dput(path->dentry); 13869deed3ebSAl Viro if (path->mnt != nd->path.mnt) 13879deed3ebSAl Viro mntput(path->mnt); 13889deed3ebSAl Viro } else { 1389bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1390bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1391bd7c4b50SAl Viro } 1392bd7c4b50SAl Viro return ret; 1393bd7c4b50SAl Viro } 1394bd7c4b50SAl Viro 13959875cf80SDavid Howells /* 1396f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1397f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1398baa03890SNick Piggin */ 1399e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1400e3c13928SAl Viro struct dentry *dir, 14016c51e513SAl Viro unsigned int flags) 1402baa03890SNick Piggin { 1403a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1404bad61189SMiklos Szeredi if (dentry) { 1405a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1406bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 140774ff0ffcSAl Viro if (!error) 14085542aa2fSEric W. Biederman d_invalidate(dentry); 1409bad61189SMiklos Szeredi dput(dentry); 141074ff0ffcSAl Viro return ERR_PTR(error); 1411bad61189SMiklos Szeredi } 1412bad61189SMiklos Szeredi } 1413baa03890SNick Piggin return dentry; 1414baa03890SNick Piggin } 1415baa03890SNick Piggin 1416baa03890SNick Piggin /* 1417a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1418a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1419a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1420a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1421a03ece5fSAl Viro * at all. 142244396f4bSJosef Bacik */ 1423a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1424a03ece5fSAl Viro struct dentry *base, unsigned int flags) 142544396f4bSJosef Bacik { 1426a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 142744396f4bSJosef Bacik struct dentry *old; 1428a03ece5fSAl Viro struct inode *dir = base->d_inode; 1429a03ece5fSAl Viro 1430a03ece5fSAl Viro if (dentry) 1431a03ece5fSAl Viro return dentry; 143244396f4bSJosef Bacik 143344396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1434a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 143544396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1436a03ece5fSAl Viro 1437a03ece5fSAl Viro dentry = d_alloc(base, name); 1438a03ece5fSAl Viro if (unlikely(!dentry)) 1439a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 144044396f4bSJosef Bacik 144172bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 144244396f4bSJosef Bacik if (unlikely(old)) { 144344396f4bSJosef Bacik dput(dentry); 144444396f4bSJosef Bacik dentry = old; 144544396f4bSJosef Bacik } 144644396f4bSJosef Bacik return dentry; 144744396f4bSJosef Bacik } 144844396f4bSJosef Bacik 144920e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 145020e34357SAl Viro struct inode **inode, 1451254cf582SAl Viro unsigned *seqp) 14521da177e4SLinus Torvalds { 145331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 14545a18fff2SAl Viro int status = 1; 14559875cf80SDavid Howells 14563cac260aSAl Viro /* 1457b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 14585d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 14595d0f49c1SAl Viro * going to fall back to non-racy lookup. 1460b04f784eSNick Piggin */ 146131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 146231e6b01fSNick Piggin unsigned seq; 1463da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 14645d0f49c1SAl Viro if (unlikely(!dentry)) { 14654675ac39SAl Viro if (unlazy_walk(nd)) 146620e34357SAl Viro return ERR_PTR(-ECHILD); 146720e34357SAl Viro return NULL; 14685d0f49c1SAl Viro } 14695a18fff2SAl Viro 147012f8ad4bSLinus Torvalds /* 147112f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 147212f8ad4bSLinus Torvalds * the dentry name information from lookup. 147312f8ad4bSLinus Torvalds */ 147463afdfc7SDavid Howells *inode = d_backing_inode(dentry); 14755d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 147620e34357SAl Viro return ERR_PTR(-ECHILD); 147712f8ad4bSLinus Torvalds 147812f8ad4bSLinus Torvalds /* 147912f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 148012f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 148112f8ad4bSLinus Torvalds * 148212f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 148312f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 148412f8ad4bSLinus Torvalds */ 14855d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 148620e34357SAl Viro return ERR_PTR(-ECHILD); 14875a18fff2SAl Viro 1488254cf582SAl Viro *seqp = seq; 14894ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1490c153007bSAl Viro if (likely(status > 0)) 149120e34357SAl Viro return dentry; 14924675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 149320e34357SAl Viro return ERR_PTR(-ECHILD); 1494209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1495209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1496209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 14975a18fff2SAl Viro } else { 1498e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 149981e6f520SAl Viro if (unlikely(!dentry)) 150020e34357SAl Viro return NULL; 15014ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15025d0f49c1SAl Viro } 15035a18fff2SAl Viro if (unlikely(status <= 0)) { 1504e9742b53SAl Viro if (!status) 15055d0f49c1SAl Viro d_invalidate(dentry); 15065a18fff2SAl Viro dput(dentry); 150720e34357SAl Viro return ERR_PTR(status); 15085a18fff2SAl Viro } 150920e34357SAl Viro return dentry; 1510697f514dSMiklos Szeredi } 1511697f514dSMiklos Szeredi 1512697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 151388d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1514e3c13928SAl Viro struct dentry *dir, 1515e3c13928SAl Viro unsigned int flags) 1516697f514dSMiklos Szeredi { 151788d8331aSAl Viro struct dentry *dentry, *old; 15181936386eSAl Viro struct inode *inode = dir->d_inode; 1519d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 15201936386eSAl Viro 15211936386eSAl Viro /* Don't go there if it's already dead */ 152294bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 152388d8331aSAl Viro return ERR_PTR(-ENOENT); 152494bdd655SAl Viro again: 1525d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 152694bdd655SAl Viro if (IS_ERR(dentry)) 152788d8331aSAl Viro return dentry; 152894bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1529949a852eSAl Viro int error = d_revalidate(dentry, flags); 1530949a852eSAl Viro if (unlikely(error <= 0)) { 153194bdd655SAl Viro if (!error) { 1532949a852eSAl Viro d_invalidate(dentry); 1533949a852eSAl Viro dput(dentry); 153494bdd655SAl Viro goto again; 153594bdd655SAl Viro } 153694bdd655SAl Viro dput(dentry); 1537949a852eSAl Viro dentry = ERR_PTR(error); 1538949a852eSAl Viro } 153994bdd655SAl Viro } else { 15401936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 154185c7f810SAl Viro d_lookup_done(dentry); 15421936386eSAl Viro if (unlikely(old)) { 15431936386eSAl Viro dput(dentry); 15441936386eSAl Viro dentry = old; 1545949a852eSAl Viro } 1546949a852eSAl Viro } 1547e3c13928SAl Viro return dentry; 15481da177e4SLinus Torvalds } 15491da177e4SLinus Torvalds 155088d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 155188d8331aSAl Viro struct dentry *dir, 155288d8331aSAl Viro unsigned int flags) 155388d8331aSAl Viro { 155488d8331aSAl Viro struct inode *inode = dir->d_inode; 155588d8331aSAl Viro struct dentry *res; 155688d8331aSAl Viro inode_lock_shared(inode); 155788d8331aSAl Viro res = __lookup_slow(name, dir, flags); 155888d8331aSAl Viro inode_unlock_shared(inode); 155988d8331aSAl Viro return res; 156088d8331aSAl Viro } 156188d8331aSAl Viro 156252094c8aSAl Viro static inline int may_lookup(struct nameidata *nd) 156352094c8aSAl Viro { 156452094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 15654ad5abb3SAl Viro int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 156652094c8aSAl Viro if (err != -ECHILD) 156752094c8aSAl Viro return err; 15684675ac39SAl Viro if (unlazy_walk(nd)) 156952094c8aSAl Viro return -ECHILD; 157052094c8aSAl Viro } 15714ad5abb3SAl Viro return inode_permission(nd->inode, MAY_EXEC); 157252094c8aSAl Viro } 157352094c8aSAl Viro 157449055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1575d63ff28fSAl Viro { 1576ad6cc4c3SAl Viro int error; 1577ad6cc4c3SAl Viro 157849055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 157949055906SAl Viro return -ELOOP; 15804542576bSAl Viro 15814542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 15824542576bSAl Viro return 0; 15834542576bSAl Viro if (likely(nd->stack != nd->internal)) 15844542576bSAl Viro return 0; 15854542576bSAl Viro 15864542576bSAl Viro error = __nd_alloc_stack(nd); 158749055906SAl Viro if (likely(!error)) 158849055906SAl Viro return 0; 1589bc40aee0SAl Viro if (error == -ECHILD) { 1590aef9404dSAl Viro // we must grab link first 1591aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 1592aef9404dSAl Viro // ... and we must unlazy to be able to clean up 1593aef9404dSAl Viro error = unlazy_walk(nd); 1594aef9404dSAl Viro if (unlikely(!grabbed_link)) 1595aef9404dSAl Viro error = -ECHILD; 1596aef9404dSAl Viro if (!error) 15974542576bSAl Viro error = __nd_alloc_stack(nd); 1598bc40aee0SAl Viro } 159949055906SAl Viro return error; 160049055906SAl Viro } 160149055906SAl Viro 160249055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 160349055906SAl Viro 160449055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 160549055906SAl Viro struct inode *inode, unsigned seq, int flags) 160649055906SAl Viro { 160749055906SAl Viro struct saved *last; 160849055906SAl Viro const char *res; 160949055906SAl Viro int error = reserve_stack(nd, link, seq); 161049055906SAl Viro 161149055906SAl Viro if (unlikely(error)) { 161249055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1613cd179f44SAl Viro path_put(link); 161406708adbSAl Viro return ERR_PTR(error); 1615626de996SAl Viro } 1616ab104923SAl Viro last = nd->stack + nd->depth++; 16171cf2665bSAl Viro last->link = *link; 1618fceef393SAl Viro clear_delayed_call(&last->done); 16190450b2d1SAl Viro last->seq = seq; 1620ad6cc4c3SAl Viro 1621b1a81972SAl Viro if (flags & WALK_TRAILING) { 1622ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1623ad6cc4c3SAl Viro if (unlikely(error)) 1624ad6cc4c3SAl Viro return ERR_PTR(error); 1625ad6cc4c3SAl Viro } 1626ad6cc4c3SAl Viro 1627ad6cc4c3SAl Viro if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS)) 1628ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1629ad6cc4c3SAl Viro 1630ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1631ad6cc4c3SAl Viro touch_atime(&last->link); 1632ad6cc4c3SAl Viro cond_resched(); 1633ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1634ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1635ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1636ad6cc4c3SAl Viro touch_atime(&last->link); 1637ad6cc4c3SAl Viro } 1638ad6cc4c3SAl Viro 1639ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1640ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1641ad6cc4c3SAl Viro if (unlikely(error)) 1642ad6cc4c3SAl Viro return ERR_PTR(error); 1643ad6cc4c3SAl Viro 1644ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1645ad6cc4c3SAl Viro if (!res) { 1646ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1647ad6cc4c3SAl Viro struct delayed_call *); 1648ad6cc4c3SAl Viro get = inode->i_op->get_link; 1649ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1650ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1651ad6cc4c3SAl Viro if (res == ERR_PTR(-ECHILD)) { 1652ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1653ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1654ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1655ad6cc4c3SAl Viro } 1656ad6cc4c3SAl Viro } else { 1657ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1658ad6cc4c3SAl Viro } 1659ad6cc4c3SAl Viro if (!res) 1660ad6cc4c3SAl Viro goto all_done; 1661ad6cc4c3SAl Viro if (IS_ERR(res)) 1662ad6cc4c3SAl Viro return res; 1663ad6cc4c3SAl Viro } 1664ad6cc4c3SAl Viro if (*res == '/') { 1665ad6cc4c3SAl Viro error = nd_jump_root(nd); 1666ad6cc4c3SAl Viro if (unlikely(error)) 1667ad6cc4c3SAl Viro return ERR_PTR(error); 1668ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1669ad6cc4c3SAl Viro ; 1670ad6cc4c3SAl Viro } 1671ad6cc4c3SAl Viro if (*res) 1672ad6cc4c3SAl Viro return res; 1673ad6cc4c3SAl Viro all_done: // pure jump 1674ad6cc4c3SAl Viro put_link(nd); 1675ad6cc4c3SAl Viro return NULL; 1676d63ff28fSAl Viro } 1677d63ff28fSAl Viro 16783ddcd056SLinus Torvalds /* 16793ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 16803ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 16813ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 16823ddcd056SLinus Torvalds * for the common case. 16833ddcd056SLinus Torvalds */ 1684b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1685cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 16863ddcd056SLinus Torvalds { 1687cbae4d12SAl Viro struct path path; 1688cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1689cbae4d12SAl Viro 1690cbae4d12SAl Viro if (err < 0) 1691b0417d2cSAl Viro return ERR_PTR(err); 1692cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 16938c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1694aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 16958f64fb1cSAl Viro /* not a symlink or should not follow */ 1696c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1697c99687a0SAl Viro dput(nd->path.dentry); 1698c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1699c99687a0SAl Viro mntput(nd->path.mnt); 1700c99687a0SAl Viro } 1701c99687a0SAl Viro nd->path = path; 17028f64fb1cSAl Viro nd->inode = inode; 17038f64fb1cSAl Viro nd->seq = seq; 1704b0417d2cSAl Viro return NULL; 17058f64fb1cSAl Viro } 1706a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 170784f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1708cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1709b0417d2cSAl Viro return ERR_PTR(-ECHILD); 171084f0cd9eSAl Viro } else { 171184f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 171284f0cd9eSAl Viro mntget(path.mnt); 1713a7f77542SAl Viro } 1714b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 17153ddcd056SLinus Torvalds } 17163ddcd056SLinus Torvalds 1717c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1718c2df1968SAl Viro struct inode **inodep, 1719c2df1968SAl Viro unsigned *seqp) 1720957dd41dSAl Viro { 172112487f30SAl Viro struct dentry *parent, *old; 1722957dd41dSAl Viro 172312487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 172412487f30SAl Viro goto in_root; 172512487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 17267ef482faSAl Viro struct path path; 1727efe772d6SAl Viro unsigned seq; 17287ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 17297ef482faSAl Viro &nd->root, &path, &seq)) 173012487f30SAl Viro goto in_root; 1731efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1732efe772d6SAl Viro return ERR_PTR(-ECHILD); 1733efe772d6SAl Viro nd->path = path; 1734efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1735efe772d6SAl Viro nd->seq = seq; 1736efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1737efe772d6SAl Viro return ERR_PTR(-ECHILD); 1738efe772d6SAl Viro /* we know that mountpoint was pinned */ 1739957dd41dSAl Viro } 174012487f30SAl Viro old = nd->path.dentry; 174112487f30SAl Viro parent = old->d_parent; 174212487f30SAl Viro *inodep = parent->d_inode; 174312487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 174412487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 174512487f30SAl Viro return ERR_PTR(-ECHILD); 174612487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 174712487f30SAl Viro return ERR_PTR(-ECHILD); 174812487f30SAl Viro return parent; 174912487f30SAl Viro in_root: 1750efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1751efe772d6SAl Viro return ERR_PTR(-ECHILD); 1752957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 17537521f22bSAl Viro return ERR_PTR(-ECHILD); 1754c2df1968SAl Viro return NULL; 1755957dd41dSAl Viro } 1756957dd41dSAl Viro 1757c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1758c2df1968SAl Viro struct inode **inodep, 1759c2df1968SAl Viro unsigned *seqp) 1760957dd41dSAl Viro { 176112487f30SAl Viro struct dentry *parent; 176212487f30SAl Viro 1763957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 176412487f30SAl Viro goto in_root; 176512487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 17662aa38470SAl Viro struct path path; 17672aa38470SAl Viro 17682aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 17692aa38470SAl Viro &nd->root, &path)) 177012487f30SAl Viro goto in_root; 1771165200d6SAl Viro path_put(&nd->path); 1772165200d6SAl Viro nd->path = path; 17732aa38470SAl Viro nd->inode = path.dentry->d_inode; 1774165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1775165200d6SAl Viro return ERR_PTR(-EXDEV); 177612487f30SAl Viro } 1777957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 177812487f30SAl Viro parent = dget_parent(nd->path.dentry); 1779957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1780957dd41dSAl Viro dput(parent); 17817521f22bSAl Viro return ERR_PTR(-ENOENT); 1782957dd41dSAl Viro } 1783c2df1968SAl Viro *seqp = 0; 1784c2df1968SAl Viro *inodep = parent->d_inode; 1785c2df1968SAl Viro return parent; 178612487f30SAl Viro 178712487f30SAl Viro in_root: 1788957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 17897521f22bSAl Viro return ERR_PTR(-EXDEV); 1790c2df1968SAl Viro dget(nd->path.dentry); 1791c2df1968SAl Viro return NULL; 1792957dd41dSAl Viro } 1793957dd41dSAl Viro 17947521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1795957dd41dSAl Viro { 1796957dd41dSAl Viro if (type == LAST_DOTDOT) { 17977521f22bSAl Viro const char *error = NULL; 1798c2df1968SAl Viro struct dentry *parent; 1799c2df1968SAl Viro struct inode *inode; 1800c2df1968SAl Viro unsigned seq; 1801957dd41dSAl Viro 1802957dd41dSAl Viro if (!nd->root.mnt) { 18037521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1804957dd41dSAl Viro if (error) 1805957dd41dSAl Viro return error; 1806957dd41dSAl Viro } 1807957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1808c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1809957dd41dSAl Viro else 1810c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1811c2df1968SAl Viro if (IS_ERR(parent)) 1812c2df1968SAl Viro return ERR_CAST(parent); 1813c2df1968SAl Viro if (unlikely(!parent)) 1814c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1815c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1816c2df1968SAl Viro else 1817c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1818c2df1968SAl Viro parent, inode, seq); 1819c2df1968SAl Viro if (unlikely(error)) 1820957dd41dSAl Viro return error; 1821957dd41dSAl Viro 1822957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1823957dd41dSAl Viro /* 1824957dd41dSAl Viro * If there was a racing rename or mount along our 1825957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1826957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1827957dd41dSAl Viro * some fallback). 1828957dd41dSAl Viro */ 1829957dd41dSAl Viro smp_rmb(); 1830957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 18317521f22bSAl Viro return ERR_PTR(-EAGAIN); 1832957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 18337521f22bSAl Viro return ERR_PTR(-EAGAIN); 1834957dd41dSAl Viro } 1835957dd41dSAl Viro } 18367521f22bSAl Viro return NULL; 1837957dd41dSAl Viro } 1838957dd41dSAl Viro 183992d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1840ce57dfc1SAl Viro { 1841db3c9adeSAl Viro struct dentry *dentry; 1842ce57dfc1SAl Viro struct inode *inode; 1843254cf582SAl Viro unsigned seq; 1844ce57dfc1SAl Viro /* 1845ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1846ce57dfc1SAl Viro * to be able to know about the current root directory and 1847ce57dfc1SAl Viro * parent relationships. 1848ce57dfc1SAl Viro */ 18494693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 18501c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 18514693a547SAl Viro put_link(nd); 18527521f22bSAl Viro return handle_dots(nd, nd->last_type); 18534693a547SAl Viro } 185420e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 185520e34357SAl Viro if (IS_ERR(dentry)) 185692d27016SAl Viro return ERR_CAST(dentry); 185720e34357SAl Viro if (unlikely(!dentry)) { 1858db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1859db3c9adeSAl Viro if (IS_ERR(dentry)) 186092d27016SAl Viro return ERR_CAST(dentry); 186120e34357SAl Viro } 186256676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 186356676ec3SAl Viro put_link(nd); 1864b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1865ce57dfc1SAl Viro } 1866ce57dfc1SAl Viro 18671da177e4SLinus Torvalds /* 1868bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1869bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1870bfcfaa77SLinus Torvalds * 1871bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1872bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1873bfcfaa77SLinus Torvalds * fast. 1874bfcfaa77SLinus Torvalds * 1875bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1876bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1877bfcfaa77SLinus Torvalds * crossing operation. 1878bfcfaa77SLinus Torvalds * 1879bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1880bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1881bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1882bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1883bfcfaa77SLinus Torvalds */ 1884bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1885bfcfaa77SLinus Torvalds 1886f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1887bfcfaa77SLinus Torvalds 1888468a9428SGeorge Spelvin #ifdef HASH_MIX 1889bfcfaa77SLinus Torvalds 1890468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1891468a9428SGeorge Spelvin 1892468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 18932a18da7aSGeorge Spelvin /* 18942a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 18952a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 18962a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 18972a18da7aSGeorge Spelvin * and no temporaries. 18982a18da7aSGeorge Spelvin * 18992a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 19002a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 19012a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 19022a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 19032a18da7aSGeorge Spelvin * 19042a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 19052a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 19062a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 19072a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 19082a18da7aSGeorge Spelvin * 19092a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 19102a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 19112a18da7aSGeorge Spelvin * toggling any given output bit. 19122a18da7aSGeorge Spelvin * 19132a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 19142a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19152a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 19162a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 19172a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 19182a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 19192a18da7aSGeorge Spelvin * Perfect: 8192 258048 19202a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 19212a18da7aSGeorge Spelvin */ 19222a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19232a18da7aSGeorge Spelvin ( x ^= (a), \ 19242a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 19252a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 19262a18da7aSGeorge Spelvin y *= 9 ) 1927bfcfaa77SLinus Torvalds 19280fed3ac8SGeorge Spelvin /* 19292a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 19302a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 19312a18da7aSGeorge Spelvin * work done before the hash value is used. 19320fed3ac8SGeorge Spelvin */ 19332a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19340fed3ac8SGeorge Spelvin { 19352a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 19362a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 19372a18da7aSGeorge Spelvin return y >> 32; 19380fed3ac8SGeorge Spelvin } 19390fed3ac8SGeorge Spelvin 1940bfcfaa77SLinus Torvalds #else /* 32-bit case */ 1941bfcfaa77SLinus Torvalds 19422a18da7aSGeorge Spelvin /* 19432a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 19442a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19452a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 19462a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 19472a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 19482a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 19492a18da7aSGeorge Spelvin * Perfect: 2048 31744 19502a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 19512a18da7aSGeorge Spelvin */ 19522a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 19532a18da7aSGeorge Spelvin ( x ^= (a), \ 19542a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 19552a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 19562a18da7aSGeorge Spelvin y *= 9 ) 1957bfcfaa77SLinus Torvalds 19582a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 19590fed3ac8SGeorge Spelvin { 19602a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 19612a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 19620fed3ac8SGeorge Spelvin } 19630fed3ac8SGeorge Spelvin 1964bfcfaa77SLinus Torvalds #endif 1965bfcfaa77SLinus Torvalds 19662a18da7aSGeorge Spelvin /* 19672a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 19682a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 19692a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 19702a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 19712a18da7aSGeorge Spelvin * finds the delimiter after the name. 19722a18da7aSGeorge Spelvin */ 19738387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 1974bfcfaa77SLinus Torvalds { 19758387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 1976bfcfaa77SLinus Torvalds 1977bfcfaa77SLinus Torvalds for (;;) { 1978fcfd2fbfSGeorge Spelvin if (!len) 1979fcfd2fbfSGeorge Spelvin goto done; 1980e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 1981bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 1982bfcfaa77SLinus Torvalds break; 19832a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 1984bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 1985bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 1986bfcfaa77SLinus Torvalds } 19872a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 1988bfcfaa77SLinus Torvalds done: 19892a18da7aSGeorge Spelvin return fold_hash(x, y); 1990bfcfaa77SLinus Torvalds } 1991bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 1992bfcfaa77SLinus Torvalds 1993fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 19948387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 1995fcfd2fbfSGeorge Spelvin { 19968387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 19978387ff25SLinus Torvalds unsigned long adata, mask, len; 1998fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 1999fcfd2fbfSGeorge Spelvin 20008387ff25SLinus Torvalds len = 0; 20018387ff25SLinus Torvalds goto inside; 20028387ff25SLinus Torvalds 2003fcfd2fbfSGeorge Spelvin do { 20042a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2005fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 20068387ff25SLinus Torvalds inside: 2007fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2008fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2009fcfd2fbfSGeorge Spelvin 2010fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2011fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 20122a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2013fcfd2fbfSGeorge Spelvin 20142a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2015fcfd2fbfSGeorge Spelvin } 2016fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2017fcfd2fbfSGeorge Spelvin 2018bfcfaa77SLinus Torvalds /* 2019bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2020d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2021bfcfaa77SLinus Torvalds */ 20228387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2023bfcfaa77SLinus Torvalds { 20248387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 20258387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 202636126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2027bfcfaa77SLinus Torvalds 20288387ff25SLinus Torvalds len = 0; 20298387ff25SLinus Torvalds goto inside; 20308387ff25SLinus Torvalds 2031bfcfaa77SLinus Torvalds do { 20322a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2033bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 20348387ff25SLinus Torvalds inside: 2035e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 203636126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 203736126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2038bfcfaa77SLinus Torvalds 203936126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 204036126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 204136126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 20422a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 204336126f8fSLinus Torvalds 20442a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2045bfcfaa77SLinus Torvalds } 2046bfcfaa77SLinus Torvalds 20472a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2048bfcfaa77SLinus Torvalds 2049fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 20508387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 20510145acc2SLinus Torvalds { 20528387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 20530145acc2SLinus Torvalds while (len--) 2054fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 20550145acc2SLinus Torvalds return end_name_hash(hash); 20560145acc2SLinus Torvalds } 2057ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 20580145acc2SLinus Torvalds 2059fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20608387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2061fcfd2fbfSGeorge Spelvin { 20628387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2063fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2064fcfd2fbfSGeorge Spelvin 2065fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2066e0ab7af9SGeorge Spelvin while (c) { 2067fcfd2fbfSGeorge Spelvin len++; 2068fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2069fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2070e0ab7af9SGeorge Spelvin } 2071fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2072fcfd2fbfSGeorge Spelvin } 2073f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2074fcfd2fbfSGeorge Spelvin 20753ddcd056SLinus Torvalds /* 2076200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2077200e9ef7SLinus Torvalds * one character. 2078200e9ef7SLinus Torvalds */ 20798387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2080200e9ef7SLinus Torvalds { 20818387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2082200e9ef7SLinus Torvalds unsigned long len = 0, c; 2083200e9ef7SLinus Torvalds 2084200e9ef7SLinus Torvalds c = (unsigned char)*name; 2085200e9ef7SLinus Torvalds do { 2086200e9ef7SLinus Torvalds len++; 2087200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2088200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2089200e9ef7SLinus Torvalds } while (c && c != '/'); 2090d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2091200e9ef7SLinus Torvalds } 2092200e9ef7SLinus Torvalds 2093bfcfaa77SLinus Torvalds #endif 2094bfcfaa77SLinus Torvalds 2095200e9ef7SLinus Torvalds /* 20961da177e4SLinus Torvalds * Name resolution. 2097ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2098ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 20991da177e4SLinus Torvalds * 2100ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2101ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 21021da177e4SLinus Torvalds */ 21036de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 21041da177e4SLinus Torvalds { 2105d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 21061da177e4SLinus Torvalds int err; 21071da177e4SLinus Torvalds 2108b4c03536SAl Viro nd->last_type = LAST_ROOT; 2109c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 21109b5858e9SAl Viro if (IS_ERR(name)) 21119b5858e9SAl Viro return PTR_ERR(name); 21121da177e4SLinus Torvalds while (*name=='/') 21131da177e4SLinus Torvalds name++; 21141da177e4SLinus Torvalds if (!*name) 21159e18f10aSAl Viro return 0; 21161da177e4SLinus Torvalds 21171da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 21181da177e4SLinus Torvalds for(;;) { 211992d27016SAl Viro const char *link; 2120d6bb3e90SLinus Torvalds u64 hash_len; 2121fe479a58SAl Viro int type; 21221da177e4SLinus Torvalds 212352094c8aSAl Viro err = may_lookup(nd); 21241da177e4SLinus Torvalds if (err) 21253595e234SAl Viro return err; 21261da177e4SLinus Torvalds 21278387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 21281da177e4SLinus Torvalds 2129fe479a58SAl Viro type = LAST_NORM; 2130d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2131fe479a58SAl Viro case 2: 2132200e9ef7SLinus Torvalds if (name[1] == '.') { 2133fe479a58SAl Viro type = LAST_DOTDOT; 213416c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 213516c2cd71SAl Viro } 2136fe479a58SAl Viro break; 2137fe479a58SAl Viro case 1: 2138fe479a58SAl Viro type = LAST_DOT; 2139fe479a58SAl Viro } 21405a202bcdSAl Viro if (likely(type == LAST_NORM)) { 21415a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 214216c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 21435a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2144a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2145da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 21465a202bcdSAl Viro if (err < 0) 21473595e234SAl Viro return err; 2148d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2149d6bb3e90SLinus Torvalds name = this.name; 21505a202bcdSAl Viro } 21515a202bcdSAl Viro } 2152fe479a58SAl Viro 2153d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2154d6bb3e90SLinus Torvalds nd->last.name = name; 21555f4a6a69SAl Viro nd->last_type = type; 21565f4a6a69SAl Viro 2157d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2158d6bb3e90SLinus Torvalds if (!*name) 2159bdf6cbf1SAl Viro goto OK; 2160200e9ef7SLinus Torvalds /* 2161200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2162200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2163200e9ef7SLinus Torvalds */ 2164200e9ef7SLinus Torvalds do { 2165d6bb3e90SLinus Torvalds name++; 2166d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 21678620c238SAl Viro if (unlikely(!*name)) { 21688620c238SAl Viro OK: 2169d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2170c108837eSAl Viro if (!depth) { 2171c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 21728620c238SAl Viro return 0; 2173c108837eSAl Viro } 21748620c238SAl Viro /* last component of nested symlink */ 2175d8d4611aSAl Viro name = nd->stack[--depth].name; 21768c4efe22SAl Viro link = walk_component(nd, 0); 21771c4ff1a8SAl Viro } else { 21781c4ff1a8SAl Viro /* not the last component */ 21798c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 21808620c238SAl Viro } 218192d27016SAl Viro if (unlikely(link)) { 218292d27016SAl Viro if (IS_ERR(link)) 218392d27016SAl Viro return PTR_ERR(link); 218492d27016SAl Viro /* a symlink to follow */ 2185d8d4611aSAl Viro nd->stack[depth++].name = name; 218692d27016SAl Viro name = link; 21879e18f10aSAl Viro continue; 218848c8b0c5SAl Viro } 218997242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 219097242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 21914675ac39SAl Viro if (unlazy_walk(nd)) 219297242f99SAl Viro return -ECHILD; 219397242f99SAl Viro } 21943595e234SAl Viro return -ENOTDIR; 21955f4a6a69SAl Viro } 2196ce57dfc1SAl Viro } 219797242f99SAl Viro } 21981da177e4SLinus Torvalds 2199edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2200c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 220131e6b01fSNick Piggin { 2202740a1678SAleksa Sarai int error; 2203c8a53ee5SAl Viro const char *s = nd->name->name; 220431e6b01fSNick Piggin 2205c0eb027eSLinus Torvalds if (!*s) 2206c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2207edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2208edc2b1daSAl Viro rcu_read_lock(); 2209c0eb027eSLinus Torvalds 2210c108837eSAl Viro nd->flags = flags | LOOKUP_JUMPED; 221131e6b01fSNick Piggin nd->depth = 0; 2212ab87f9a5SAleksa Sarai 2213ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2214ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2215ab87f9a5SAleksa Sarai smp_rmb(); 2216ab87f9a5SAleksa Sarai 22175b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2218b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2219b18825a7SDavid Howells struct inode *inode = root->d_inode; 222093893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2221368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 22225b6ca027SAl Viro nd->path = nd->root; 22235b6ca027SAl Viro nd->inode = inode; 22245b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2225ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22268f47a016SAl Viro nd->root_seq = nd->seq; 22275b6ca027SAl Viro } else { 22285b6ca027SAl Viro path_get(&nd->path); 22295b6ca027SAl Viro } 2230368ee9baSAl Viro return s; 22315b6ca027SAl Viro } 22325b6ca027SAl Viro 223331e6b01fSNick Piggin nd->root.mnt = NULL; 2234248fb5b9SAl Viro nd->path.mnt = NULL; 2235248fb5b9SAl Viro nd->path.dentry = NULL; 223631e6b01fSNick Piggin 22378db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 22388db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2239740a1678SAleksa Sarai error = nd_jump_root(nd); 2240740a1678SAleksa Sarai if (unlikely(error)) 2241740a1678SAleksa Sarai return ERR_PTR(error); 2242ef55d917SAl Viro return s; 22438db52c7eSAleksa Sarai } 22448db52c7eSAleksa Sarai 22458db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 22468db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2247e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 224831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2249c28cc364SNick Piggin unsigned seq; 225031e6b01fSNick Piggin 2251c28cc364SNick Piggin do { 2252c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 225331e6b01fSNick Piggin nd->path = fs->pwd; 2254ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2255c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2256c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2257e41f7d4eSAl Viro } else { 2258e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2259ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2260e41f7d4eSAl Viro } 226131e6b01fSNick Piggin } else { 2262582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2263c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 226431e6b01fSNick Piggin struct dentry *dentry; 226531e6b01fSNick Piggin 22662903ff01SAl Viro if (!f.file) 2267368ee9baSAl Viro return ERR_PTR(-EBADF); 226831e6b01fSNick Piggin 22692903ff01SAl Viro dentry = f.file->f_path.dentry; 227031e6b01fSNick Piggin 2271edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 22722903ff01SAl Viro fdput(f); 2273368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2274f52e0c11SAl Viro } 22752903ff01SAl Viro 22762903ff01SAl Viro nd->path = f.file->f_path; 2277e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 227834a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 227934a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 22805590ff0dSUlrich Drepper } else { 22812903ff01SAl Viro path_get(&nd->path); 228234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 22831da177e4SLinus Torvalds } 228434a26b99SAl Viro fdput(f); 2285e41f7d4eSAl Viro } 22868db52c7eSAleksa Sarai 2287adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2288adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2289adb21d2bSAleksa Sarai nd->root = nd->path; 2290adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2291adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2292adb21d2bSAleksa Sarai } else { 2293adb21d2bSAleksa Sarai path_get(&nd->root); 2294adb21d2bSAleksa Sarai nd->flags |= LOOKUP_ROOT_GRABBED; 2295adb21d2bSAleksa Sarai } 2296adb21d2bSAleksa Sarai } 2297adb21d2bSAleksa Sarai return s; 22989b4a9b14SAl Viro } 22999b4a9b14SAl Viro 23001ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 230195fa25d9SAl Viro { 2302bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2303bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2304bd92d7feSAl Viro 2305c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2306bd92d7feSAl Viro } 2307bd92d7feSAl Viro 23084f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 23094f757f3cSAl Viro { 2310c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2311db3c9adeSAl Viro dget(nd->path.dentry); 2312b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2313b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 23144f757f3cSAl Viro } 23154f757f3cSAl Viro 23169b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2317c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 23189b4a9b14SAl Viro { 2319c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2320bd92d7feSAl Viro int err; 232131e6b01fSNick Piggin 23229b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 23234f757f3cSAl Viro err = handle_lookup_down(nd); 23245f336e72SAl Viro if (unlikely(err < 0)) 23255f336e72SAl Viro s = ERR_PTR(err); 23264f757f3cSAl Viro } 23274f757f3cSAl Viro 23281ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 23291ccac622SAl Viro (s = lookup_last(nd)) != NULL) 23301ccac622SAl Viro ; 23319f1fafeeSAl Viro if (!err) 23329f1fafeeSAl Viro err = complete_walk(nd); 2333bd92d7feSAl Viro 2334deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2335deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2336bd23a539SAl Viro err = -ENOTDIR; 2337161aff1dSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 2338161aff1dSAl Viro err = handle_lookup_down(nd); 2339161aff1dSAl Viro nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... 2340161aff1dSAl Viro } 2341625b6d10SAl Viro if (!err) { 2342625b6d10SAl Viro *path = nd->path; 2343625b6d10SAl Viro nd->path.mnt = NULL; 2344625b6d10SAl Viro nd->path.dentry = NULL; 2345625b6d10SAl Viro } 2346deb106c6SAl Viro terminate_walk(nd); 2347bd92d7feSAl Viro return err; 234831e6b01fSNick Piggin } 234931e6b01fSNick Piggin 235031d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 23519ad1aaa6SAl Viro struct path *path, struct path *root) 2352873f1eedSJeff Layton { 2353894bc8c4SAl Viro int retval; 23549883d185SAl Viro struct nameidata nd; 2355abc9f5beSAl Viro if (IS_ERR(name)) 2356abc9f5beSAl Viro return PTR_ERR(name); 23579ad1aaa6SAl Viro if (unlikely(root)) { 23589ad1aaa6SAl Viro nd.root = *root; 23599ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 23609ad1aaa6SAl Viro } 23619883d185SAl Viro set_nameidata(&nd, dfd, name); 2362c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2363873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2364c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2365873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2366c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2367873f1eedSJeff Layton 2368873f1eedSJeff Layton if (likely(!retval)) 2369161aff1dSAl Viro audit_inode(name, path->dentry, 2370161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 23719883d185SAl Viro restore_nameidata(); 2372e4bd1c1aSAl Viro putname(name); 2373873f1eedSJeff Layton return retval; 2374873f1eedSJeff Layton } 2375873f1eedSJeff Layton 23768bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2377c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2378391172c4SAl Viro struct path *parent) 23798bcb77faSAl Viro { 2380c8a53ee5SAl Viro const char *s = path_init(nd, flags); 23819b5858e9SAl Viro int err = link_path_walk(s, nd); 23828bcb77faSAl Viro if (!err) 23838bcb77faSAl Viro err = complete_walk(nd); 2384391172c4SAl Viro if (!err) { 2385391172c4SAl Viro *parent = nd->path; 2386391172c4SAl Viro nd->path.mnt = NULL; 2387391172c4SAl Viro nd->path.dentry = NULL; 2388391172c4SAl Viro } 2389deb106c6SAl Viro terminate_walk(nd); 23908bcb77faSAl Viro return err; 23918bcb77faSAl Viro } 23928bcb77faSAl Viro 23935c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2394391172c4SAl Viro unsigned int flags, struct path *parent, 2395391172c4SAl Viro struct qstr *last, int *type) 23968bcb77faSAl Viro { 23978bcb77faSAl Viro int retval; 23989883d185SAl Viro struct nameidata nd; 23998bcb77faSAl Viro 24005c31b6ceSAl Viro if (IS_ERR(name)) 24015c31b6ceSAl Viro return name; 24029883d185SAl Viro set_nameidata(&nd, dfd, name); 2403c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 24048bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2405c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 24068bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2407c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2408391172c4SAl Viro if (likely(!retval)) { 2409391172c4SAl Viro *last = nd.last; 2410391172c4SAl Viro *type = nd.last_type; 2411c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 24125c31b6ceSAl Viro } else { 24135c31b6ceSAl Viro putname(name); 24145c31b6ceSAl Viro name = ERR_PTR(retval); 2415391172c4SAl Viro } 24169883d185SAl Viro restore_nameidata(); 24175c31b6ceSAl Viro return name; 24188bcb77faSAl Viro } 24198bcb77faSAl Viro 242079714f72SAl Viro /* does lookup, returns the object with parent locked */ 242179714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 24225590ff0dSUlrich Drepper { 24235c31b6ceSAl Viro struct filename *filename; 24245c31b6ceSAl Viro struct dentry *d; 2425391172c4SAl Viro struct qstr last; 2426391172c4SAl Viro int type; 242751689104SPaul Moore 24285c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 24295c31b6ceSAl Viro &last, &type); 243051689104SPaul Moore if (IS_ERR(filename)) 243151689104SPaul Moore return ERR_CAST(filename); 24325c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2433391172c4SAl Viro path_put(path); 24345c31b6ceSAl Viro putname(filename); 24355c31b6ceSAl Viro return ERR_PTR(-EINVAL); 243679714f72SAl Viro } 24375955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2438391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 243979714f72SAl Viro if (IS_ERR(d)) { 24405955102cSAl Viro inode_unlock(path->dentry->d_inode); 2441391172c4SAl Viro path_put(path); 244279714f72SAl Viro } 244351689104SPaul Moore putname(filename); 244479714f72SAl Viro return d; 24455590ff0dSUlrich Drepper } 24465590ff0dSUlrich Drepper 2447d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2448d1811465SAl Viro { 2449abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2450abc9f5beSAl Viro flags, path, NULL); 2451d1811465SAl Viro } 24524d359507SAl Viro EXPORT_SYMBOL(kern_path); 2453d1811465SAl Viro 245416f18200SJosef 'Jeff' Sipek /** 245516f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 245616f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 245716f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 245816f18200SJosef 'Jeff' Sipek * @name: pointer to file name 245916f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2460e0a01249SAl Viro * @path: pointer to struct path to fill 246116f18200SJosef 'Jeff' Sipek */ 246216f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 246316f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2464e0a01249SAl Viro struct path *path) 246516f18200SJosef 'Jeff' Sipek { 24669ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 24679ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2468abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2469abc9f5beSAl Viro flags , path, &root); 247016f18200SJosef 'Jeff' Sipek } 24714d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 247216f18200SJosef 'Jeff' Sipek 24733c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 24743c95f0dcSAl Viro int len, struct qstr *this) 24753c95f0dcSAl Viro { 24763c95f0dcSAl Viro this->name = name; 24773c95f0dcSAl Viro this->len = len; 24783c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 24793c95f0dcSAl Viro if (!len) 24803c95f0dcSAl Viro return -EACCES; 24813c95f0dcSAl Viro 24823c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 24833c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 24843c95f0dcSAl Viro return -EACCES; 24853c95f0dcSAl Viro } 24863c95f0dcSAl Viro 24873c95f0dcSAl Viro while (len--) { 24883c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 24893c95f0dcSAl Viro if (c == '/' || c == '\0') 24903c95f0dcSAl Viro return -EACCES; 24913c95f0dcSAl Viro } 24923c95f0dcSAl Viro /* 24933c95f0dcSAl Viro * See if the low-level filesystem might want 24943c95f0dcSAl Viro * to use its own hash.. 24953c95f0dcSAl Viro */ 24963c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 24973c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 24983c95f0dcSAl Viro if (err < 0) 24993c95f0dcSAl Viro return err; 25003c95f0dcSAl Viro } 25013c95f0dcSAl Viro 25023c95f0dcSAl Viro return inode_permission(base->d_inode, MAY_EXEC); 25033c95f0dcSAl Viro } 25043c95f0dcSAl Viro 2505eead1911SChristoph Hellwig /** 25060da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 25070da0b7fdSDavid Howells * @name: pathname component to lookup 25080da0b7fdSDavid Howells * @base: base directory to lookup from 25090da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 25100da0b7fdSDavid Howells * 25110da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 25120da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 25130da0b7fdSDavid Howells * 25140da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 25150da0b7fdSDavid Howells * not be called by generic code. 25160da0b7fdSDavid Howells * 25170da0b7fdSDavid Howells * The caller must hold base->i_mutex. 25180da0b7fdSDavid Howells */ 25190da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 25200da0b7fdSDavid Howells { 25210da0b7fdSDavid Howells struct qstr this; 25220da0b7fdSDavid Howells int err; 25230da0b7fdSDavid Howells 25240da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25250da0b7fdSDavid Howells 25260da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 25270da0b7fdSDavid Howells if (err) 25280da0b7fdSDavid Howells return ERR_PTR(err); 25290da0b7fdSDavid Howells 25300da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 25310da0b7fdSDavid Howells } 25320da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 25330da0b7fdSDavid Howells 25340da0b7fdSDavid Howells /** 2535a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2536eead1911SChristoph Hellwig * @name: pathname component to lookup 2537eead1911SChristoph Hellwig * @base: base directory to lookup from 2538eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2539eead1911SChristoph Hellwig * 2540a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 25419e7543e9SAl Viro * not be called by generic code. 2542bbddca8eSNeilBrown * 2543bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2544eead1911SChristoph Hellwig */ 2545057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2546057f6c01SJames Morris { 25478613a209SAl Viro struct dentry *dentry; 2548057f6c01SJames Morris struct qstr this; 2549cda309deSMiklos Szeredi int err; 2550057f6c01SJames Morris 25515955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 25522f9092e1SDavid Woodhouse 25533c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2554cda309deSMiklos Szeredi if (err) 2555cda309deSMiklos Szeredi return ERR_PTR(err); 2556cda309deSMiklos Szeredi 25578613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 25588613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2559057f6c01SJames Morris } 25604d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2561057f6c01SJames Morris 2562bbddca8eSNeilBrown /** 2563bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2564bbddca8eSNeilBrown * @name: pathname component to lookup 2565bbddca8eSNeilBrown * @base: base directory to lookup from 2566bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2567bbddca8eSNeilBrown * 2568bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2569bbddca8eSNeilBrown * not be called by generic code. 2570bbddca8eSNeilBrown * 2571bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2572bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2573bbddca8eSNeilBrown */ 2574bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2575bbddca8eSNeilBrown struct dentry *base, int len) 2576bbddca8eSNeilBrown { 2577bbddca8eSNeilBrown struct qstr this; 2578bbddca8eSNeilBrown int err; 257920d00ee8SLinus Torvalds struct dentry *ret; 2580bbddca8eSNeilBrown 25813c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2582bbddca8eSNeilBrown if (err) 2583bbddca8eSNeilBrown return ERR_PTR(err); 2584bbddca8eSNeilBrown 258520d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 258620d00ee8SLinus Torvalds if (!ret) 258720d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 258820d00ee8SLinus Torvalds return ret; 2589bbddca8eSNeilBrown } 2590bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2591bbddca8eSNeilBrown 25926c2d4798SAl Viro /* 25936c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 25946c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 25956c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 25966c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 25976c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 25986c2d4798SAl Viro * this one avoids such problems. 25996c2d4798SAl Viro */ 26006c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 26016c2d4798SAl Viro struct dentry *base, int len) 26026c2d4798SAl Viro { 26036c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 26042fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 26056c2d4798SAl Viro dput(ret); 26066c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 26076c2d4798SAl Viro } 26086c2d4798SAl Viro return ret; 26096c2d4798SAl Viro } 26106c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 26116c2d4798SAl Viro 2612eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2613eedf265aSEric W. Biederman int path_pts(struct path *path) 2614eedf265aSEric W. Biederman { 2615eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2616eedf265aSEric W. Biederman * the input path. 2617eedf265aSEric W. Biederman */ 2618a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2619a6a7eb76SAl Viro struct dentry *child; 262019f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2621eedf265aSEric W. Biederman 2622a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2623a6a7eb76SAl Viro dput(parent); 262463b27720SAl Viro return -ENOENT; 2625a6a7eb76SAl Viro } 262663b27720SAl Viro dput(path->dentry); 262763b27720SAl Viro path->dentry = parent; 2628eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2629eedf265aSEric W. Biederman if (!child) 2630eedf265aSEric W. Biederman return -ENOENT; 2631eedf265aSEric W. Biederman 2632eedf265aSEric W. Biederman path->dentry = child; 2633eedf265aSEric W. Biederman dput(parent); 263419f6028aSAl Viro follow_down(path); 2635eedf265aSEric W. Biederman return 0; 2636eedf265aSEric W. Biederman } 2637eedf265aSEric W. Biederman #endif 2638eedf265aSEric W. Biederman 26391fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 26401fa1e7f6SAndy Whitcroft struct path *path, int *empty) 26411da177e4SLinus Torvalds { 2642abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2643abc9f5beSAl Viro flags, path, NULL); 26441da177e4SLinus Torvalds } 2645b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 26461fa1e7f6SAndy Whitcroft 2647cbdf35bcSMiklos Szeredi int __check_sticky(struct inode *dir, struct inode *inode) 26481da177e4SLinus Torvalds { 26498e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2650da9592edSDavid Howells 26518e96e3b7SEric W. Biederman if (uid_eq(inode->i_uid, fsuid)) 26521da177e4SLinus Torvalds return 0; 26538e96e3b7SEric W. Biederman if (uid_eq(dir->i_uid, fsuid)) 26541da177e4SLinus Torvalds return 0; 265523adbe12SAndy Lutomirski return !capable_wrt_inode_uidgid(inode, CAP_FOWNER); 26561da177e4SLinus Torvalds } 2657cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 26581da177e4SLinus Torvalds 26591da177e4SLinus Torvalds /* 26601da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 26611da177e4SLinus Torvalds * whether the type of victim is right. 26621da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 26631da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 26641da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 26651da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 26661da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 26671da177e4SLinus Torvalds * a. be owner of dir, or 26681da177e4SLinus Torvalds * b. be owner of victim, or 26691da177e4SLinus Torvalds * c. have CAP_FOWNER capability 26701da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 26711da177e4SLinus Torvalds * links pointing to it. 26720bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 26730bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 26740bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 26750bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 26760bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 26771da177e4SLinus Torvalds * nfs_async_unlink(). 26781da177e4SLinus Torvalds */ 2679b18825a7SDavid Howells static int may_delete(struct inode *dir, struct dentry *victim, bool isdir) 26801da177e4SLinus Torvalds { 268163afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 26821da177e4SLinus Torvalds int error; 26831da177e4SLinus Torvalds 2684b18825a7SDavid Howells if (d_is_negative(victim)) 26851da177e4SLinus Torvalds return -ENOENT; 2686b18825a7SDavid Howells BUG_ON(!inode); 26871da177e4SLinus Torvalds 26881da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2689593d1ce8SEric W. Biederman 2690593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2691593d1ce8SEric W. Biederman if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid)) 2692593d1ce8SEric W. Biederman return -EOVERFLOW; 2693593d1ce8SEric W. Biederman 26944fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 26951da177e4SLinus Torvalds 2696f419a2e3SAl Viro error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 26971da177e4SLinus Torvalds if (error) 26981da177e4SLinus Torvalds return error; 26991da177e4SLinus Torvalds if (IS_APPEND(dir)) 27001da177e4SLinus Torvalds return -EPERM; 2701b18825a7SDavid Howells 2702b18825a7SDavid Howells if (check_sticky(dir, inode) || IS_APPEND(inode) || 27030bd23d09SEric W. Biederman IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode)) 27041da177e4SLinus Torvalds return -EPERM; 27051da177e4SLinus Torvalds if (isdir) { 270644b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27071da177e4SLinus Torvalds return -ENOTDIR; 27081da177e4SLinus Torvalds if (IS_ROOT(victim)) 27091da177e4SLinus Torvalds return -EBUSY; 271044b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 27111da177e4SLinus Torvalds return -EISDIR; 27121da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27131da177e4SLinus Torvalds return -ENOENT; 27141da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 27151da177e4SLinus Torvalds return -EBUSY; 27161da177e4SLinus Torvalds return 0; 27171da177e4SLinus Torvalds } 27181da177e4SLinus Torvalds 27191da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 27201da177e4SLinus Torvalds * dir. 27211da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 27221da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 27231da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2724036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2725036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2726036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 27271da177e4SLinus Torvalds */ 2728a95164d9SMiklos Szeredi static inline int may_create(struct inode *dir, struct dentry *child) 27291da177e4SLinus Torvalds { 2730036d5236SEric W. Biederman struct user_namespace *s_user_ns; 273114e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 27321da177e4SLinus Torvalds if (child->d_inode) 27331da177e4SLinus Torvalds return -EEXIST; 27341da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 27351da177e4SLinus Torvalds return -ENOENT; 2736036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2737036d5236SEric W. Biederman if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 2738036d5236SEric W. Biederman !kgid_has_mapping(s_user_ns, current_fsgid())) 2739036d5236SEric W. Biederman return -EOVERFLOW; 2740f419a2e3SAl Viro return inode_permission(dir, MAY_WRITE | MAY_EXEC); 27411da177e4SLinus Torvalds } 27421da177e4SLinus Torvalds 27431da177e4SLinus Torvalds /* 27441da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 27451da177e4SLinus Torvalds */ 27461da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 27471da177e4SLinus Torvalds { 27481da177e4SLinus Torvalds struct dentry *p; 27491da177e4SLinus Torvalds 27501da177e4SLinus Torvalds if (p1 == p2) { 27515955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27521da177e4SLinus Torvalds return NULL; 27531da177e4SLinus Torvalds } 27541da177e4SLinus Torvalds 2755fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 27561da177e4SLinus Torvalds 2757e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2758e2761a11SOGAWA Hirofumi if (p) { 27595955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 27605955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 27611da177e4SLinus Torvalds return p; 27621da177e4SLinus Torvalds } 27631da177e4SLinus Torvalds 2764e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2765e2761a11SOGAWA Hirofumi if (p) { 27665955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27675955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 27681da177e4SLinus Torvalds return p; 27691da177e4SLinus Torvalds } 27701da177e4SLinus Torvalds 27715955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 27725955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 27731da177e4SLinus Torvalds return NULL; 27741da177e4SLinus Torvalds } 27754d359507SAl Viro EXPORT_SYMBOL(lock_rename); 27761da177e4SLinus Torvalds 27771da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 27781da177e4SLinus Torvalds { 27795955102cSAl Viro inode_unlock(p1->d_inode); 27801da177e4SLinus Torvalds if (p1 != p2) { 27815955102cSAl Viro inode_unlock(p2->d_inode); 2782fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 27831da177e4SLinus Torvalds } 27841da177e4SLinus Torvalds } 27854d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 27861da177e4SLinus Torvalds 27874acdaf27SAl Viro int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, 2788312b63fbSAl Viro bool want_excl) 27891da177e4SLinus Torvalds { 2790a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 27911da177e4SLinus Torvalds if (error) 27921da177e4SLinus Torvalds return error; 27931da177e4SLinus Torvalds 2794acfa4380SAl Viro if (!dir->i_op->create) 27951da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 27961da177e4SLinus Torvalds mode &= S_IALLUGO; 27971da177e4SLinus Torvalds mode |= S_IFREG; 27981da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 27991da177e4SLinus Torvalds if (error) 28001da177e4SLinus Torvalds return error; 2801312b63fbSAl Viro error = dir->i_op->create(dir, dentry, mode, want_excl); 2802a74574aaSStephen Smalley if (!error) 2803f38aa942SAmy Griffis fsnotify_create(dir, dentry); 28041da177e4SLinus Torvalds return error; 28051da177e4SLinus Torvalds } 28064d359507SAl Viro EXPORT_SYMBOL(vfs_create); 28071da177e4SLinus Torvalds 28088e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 28098e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 28108e6c848eSAl Viro void *arg) 28118e6c848eSAl Viro { 28128e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 28138e6c848eSAl Viro int error = may_create(dir, dentry); 28148e6c848eSAl Viro if (error) 28158e6c848eSAl Viro return error; 28168e6c848eSAl Viro 28178e6c848eSAl Viro mode &= S_IALLUGO; 28188e6c848eSAl Viro mode |= S_IFREG; 28198e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 28208e6c848eSAl Viro if (error) 28218e6c848eSAl Viro return error; 28228e6c848eSAl Viro error = f(dentry, mode, arg); 28238e6c848eSAl Viro if (!error) 28248e6c848eSAl Viro fsnotify_create(dir, dentry); 28258e6c848eSAl Viro return error; 28268e6c848eSAl Viro } 28278e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 28288e6c848eSAl Viro 2829a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2830a2982cc9SEric W. Biederman { 2831a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2832a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2833a2982cc9SEric W. Biederman } 2834a2982cc9SEric W. Biederman 2835f0bb5aafSAl Viro static int may_open(const struct path *path, int acc_mode, int flag) 28361da177e4SLinus Torvalds { 28373fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 28381da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 28391da177e4SLinus Torvalds int error; 28401da177e4SLinus Torvalds 28411da177e4SLinus Torvalds if (!inode) 28421da177e4SLinus Torvalds return -ENOENT; 28431da177e4SLinus Torvalds 2844c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2845c8fe8f30SChristoph Hellwig case S_IFLNK: 28461da177e4SLinus Torvalds return -ELOOP; 2847c8fe8f30SChristoph Hellwig case S_IFDIR: 2848c8fe8f30SChristoph Hellwig if (acc_mode & MAY_WRITE) 28491da177e4SLinus Torvalds return -EISDIR; 2850c8fe8f30SChristoph Hellwig break; 2851c8fe8f30SChristoph Hellwig case S_IFBLK: 2852c8fe8f30SChristoph Hellwig case S_IFCHR: 2853a2982cc9SEric W. Biederman if (!may_open_dev(path)) 28541da177e4SLinus Torvalds return -EACCES; 2855c8fe8f30SChristoph Hellwig /*FALLTHRU*/ 2856c8fe8f30SChristoph Hellwig case S_IFIFO: 2857c8fe8f30SChristoph Hellwig case S_IFSOCK: 28581da177e4SLinus Torvalds flag &= ~O_TRUNC; 2859c8fe8f30SChristoph Hellwig break; 28604a3fd211SDave Hansen } 2861b41572e9SDave Hansen 286262fb4a15SAl Viro error = inode_permission(inode, MAY_OPEN | acc_mode); 2863b41572e9SDave Hansen if (error) 2864b41572e9SDave Hansen return error; 28656146f0d5SMimi Zohar 28661da177e4SLinus Torvalds /* 28671da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 28681da177e4SLinus Torvalds */ 28691da177e4SLinus Torvalds if (IS_APPEND(inode)) { 28708737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 28717715b521SAl Viro return -EPERM; 28721da177e4SLinus Torvalds if (flag & O_TRUNC) 28737715b521SAl Viro return -EPERM; 28741da177e4SLinus Torvalds } 28751da177e4SLinus Torvalds 28761da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 28772e149670SSerge E. Hallyn if (flag & O_NOATIME && !inode_owner_or_capable(inode)) 28787715b521SAl Viro return -EPERM; 28791da177e4SLinus Torvalds 2880f3c7691eSJ. Bruce Fields return 0; 28817715b521SAl Viro } 28827715b521SAl Viro 2883e1181ee6SJeff Layton static int handle_truncate(struct file *filp) 28847715b521SAl Viro { 2885f0bb5aafSAl Viro const struct path *path = &filp->f_path; 28867715b521SAl Viro struct inode *inode = path->dentry->d_inode; 28877715b521SAl Viro int error = get_write_access(inode); 28881da177e4SLinus Torvalds if (error) 28897715b521SAl Viro return error; 28901da177e4SLinus Torvalds /* 28911da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 28921da177e4SLinus Torvalds */ 2893d7a06983SJeff Layton error = locks_verify_locked(filp); 2894be6d3e56SKentaro Takeda if (!error) 2895ea0d3ab2STetsuo Handa error = security_path_truncate(path); 28961da177e4SLinus Torvalds if (!error) { 28977715b521SAl Viro error = do_truncate(path->dentry, 0, 2898d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 2899e1181ee6SJeff Layton filp); 29001da177e4SLinus Torvalds } 29011da177e4SLinus Torvalds put_write_access(inode); 2902acd0c935SMimi Zohar return error; 29031da177e4SLinus Torvalds } 29041da177e4SLinus Torvalds 2905d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 2906d57999e1SDave Hansen { 29078a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 29088a5e929dSAl Viro flag--; 2909d57999e1SDave Hansen return flag; 2910d57999e1SDave Hansen } 2911d57999e1SDave Hansen 2912d3607752SAl Viro static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode) 2913d18e9008SMiklos Szeredi { 29141328c727SSeth Forshee struct user_namespace *s_user_ns; 2915d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 2916d18e9008SMiklos Szeredi if (error) 2917d18e9008SMiklos Szeredi return error; 2918d18e9008SMiklos Szeredi 29191328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 29201328c727SSeth Forshee if (!kuid_has_mapping(s_user_ns, current_fsuid()) || 29211328c727SSeth Forshee !kgid_has_mapping(s_user_ns, current_fsgid())) 29221328c727SSeth Forshee return -EOVERFLOW; 29231328c727SSeth Forshee 2924d18e9008SMiklos Szeredi error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC); 2925d18e9008SMiklos Szeredi if (error) 2926d18e9008SMiklos Szeredi return error; 2927d18e9008SMiklos Szeredi 2928d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 2929d18e9008SMiklos Szeredi } 2930d18e9008SMiklos Szeredi 29311acf0af9SDavid Howells /* 29321acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 29331acf0af9SDavid Howells * dentry. 29341acf0af9SDavid Howells * 29351acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 29361acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 29371acf0af9SDavid Howells * 293800a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 293900a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 294000a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 29411acf0af9SDavid Howells * 29421acf0af9SDavid Howells * Returns an error code otherwise. 29431acf0af9SDavid Howells */ 2944239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 2945239eb983SAl Viro struct file *file, 2946015c3bbcSMiklos Szeredi const struct open_flags *op, 29473ec2eef1SAl Viro int open_flag, umode_t mode) 2948d18e9008SMiklos Szeredi { 2949d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 2950d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 2951d18e9008SMiklos Szeredi int error; 2952d18e9008SMiklos Szeredi 2953384f26e2SAl Viro if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */ 2954d18e9008SMiklos Szeredi open_flag &= ~O_TRUNC; 2955d18e9008SMiklos Szeredi 2956d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 2957d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 2958d18e9008SMiklos Szeredi 295930d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 296030d90494SAl Viro file->f_path.mnt = nd->path.mnt; 29610fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 296244907d79SAl Viro open_to_namei_flags(open_flag), mode); 29636fbd0714SAl Viro d_lookup_done(dentry); 2964384f26e2SAl Viro if (!error) { 296564e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 29666fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 29676fb968cdSAl Viro dput(dentry); 29686fb968cdSAl Viro dentry = dget(file->f_path.dentry); 29696fb968cdSAl Viro } 29707be219b4SAl Viro if (file->f_mode & FMODE_CREATED) 297103da633aSAl Viro fsnotify_create(dir, dentry); 297264e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 2973116cc022SMiklos Szeredi error = -EIO; 2974384f26e2SAl Viro } else { 2975d18e9008SMiklos Szeredi if (file->f_path.dentry) { 2976d18e9008SMiklos Szeredi dput(dentry); 2977d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 297810c64ceaSAl Viro } 297973a09dd9SAl Viro if (file->f_mode & FMODE_CREATED) 2980d18e9008SMiklos Szeredi fsnotify_create(dir, dentry); 2981239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 2982a01e718fSAl Viro error = -ENOENT; 2983d18e9008SMiklos Szeredi } 2984d18e9008SMiklos Szeredi } 2985239eb983SAl Viro if (error) { 2986d18e9008SMiklos Szeredi dput(dentry); 2987239eb983SAl Viro dentry = ERR_PTR(error); 2988239eb983SAl Viro } 2989239eb983SAl Viro return dentry; 2990d18e9008SMiklos Szeredi } 2991d18e9008SMiklos Szeredi 299231e6b01fSNick Piggin /* 29931acf0af9SDavid Howells * Look up and maybe create and open the last component. 2994d58ffd35SMiklos Szeredi * 299500a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 2996d58ffd35SMiklos Szeredi * 299700a07c15SAl Viro * Returns 0 on success, that is, if 299800a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 299900a07c15SAl Viro * the file was not completely opened at this time, though lookups and 300000a07c15SAl Viro * creations were performed. 300100a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 300200a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 300300a07c15SAl Viro * hadn't been specified. 30041acf0af9SDavid Howells * 300500a07c15SAl Viro * An error code is returned on failure. 3006d58ffd35SMiklos Szeredi */ 3007da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3008d58ffd35SMiklos Szeredi const struct open_flags *op, 30093ec2eef1SAl Viro bool got_write) 3010d58ffd35SMiklos Szeredi { 3011d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 301254ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 30131643b43fSAl Viro int open_flag = op->open_flag; 3014d58ffd35SMiklos Szeredi struct dentry *dentry; 30151643b43fSAl Viro int error, create_error = 0; 30161643b43fSAl Viro umode_t mode = op->mode; 30176fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3018d58ffd35SMiklos Szeredi 3019ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3020da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3021d58ffd35SMiklos Szeredi 302273a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 30236fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 30246fbd0714SAl Viro for (;;) { 30256fbd0714SAl Viro if (!dentry) { 30266fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3027d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3028da5ebf5aSAl Viro return dentry; 30296fbd0714SAl Viro } 30306fbd0714SAl Viro if (d_in_lookup(dentry)) 30316fbd0714SAl Viro break; 3032d58ffd35SMiklos Szeredi 30336fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 30346fbd0714SAl Viro if (likely(error > 0)) 30356fbd0714SAl Viro break; 30366fbd0714SAl Viro if (error) 30376fbd0714SAl Viro goto out_dput; 30386fbd0714SAl Viro d_invalidate(dentry); 30396fbd0714SAl Viro dput(dentry); 30406fbd0714SAl Viro dentry = NULL; 30416fbd0714SAl Viro } 30426fbd0714SAl Viro if (dentry->d_inode) { 3043d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3044da5ebf5aSAl Viro return dentry; 30456c51e513SAl Viro } 3046d18e9008SMiklos Szeredi 30471643b43fSAl Viro /* 30481643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 30491643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 30501643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 30511643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 30521643b43fSAl Viro * 30531643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 30541643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 30551643b43fSAl Viro */ 30561643b43fSAl Viro if (open_flag & O_CREAT) { 30571643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 30581643b43fSAl Viro mode &= ~current_umask(); 30591643b43fSAl Viro if (unlikely(!got_write)) { 30601643b43fSAl Viro create_error = -EROFS; 30611643b43fSAl Viro open_flag &= ~O_CREAT; 30621643b43fSAl Viro if (open_flag & (O_EXCL | O_TRUNC)) 30631643b43fSAl Viro goto no_open; 30641643b43fSAl Viro /* No side effects, safe to clear O_CREAT */ 30651643b43fSAl Viro } else { 30661643b43fSAl Viro create_error = may_o_create(&nd->path, dentry, mode); 30671643b43fSAl Viro if (create_error) { 30681643b43fSAl Viro open_flag &= ~O_CREAT; 30691643b43fSAl Viro if (open_flag & O_EXCL) 30701643b43fSAl Viro goto no_open; 30711643b43fSAl Viro } 30721643b43fSAl Viro } 30731643b43fSAl Viro } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) && 30741643b43fSAl Viro unlikely(!got_write)) { 30751643b43fSAl Viro /* 30761643b43fSAl Viro * No O_CREATE -> atomicity not a requirement -> fall 30771643b43fSAl Viro * back to lookup + open 30781643b43fSAl Viro */ 30791643b43fSAl Viro goto no_open; 3080d18e9008SMiklos Szeredi } 3081d18e9008SMiklos Szeredi 30821643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3083239eb983SAl Viro dentry = atomic_open(nd, dentry, file, op, open_flag, mode); 3084da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3085da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3086da5ebf5aSAl Viro return dentry; 3087239eb983SAl Viro } 308854ef4872SMiklos Szeredi 30891643b43fSAl Viro no_open: 30906fbd0714SAl Viro if (d_in_lookup(dentry)) { 309112fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 309212fa5e24SAl Viro nd->flags); 30936fbd0714SAl Viro d_lookup_done(dentry); 309412fa5e24SAl Viro if (unlikely(res)) { 309512fa5e24SAl Viro if (IS_ERR(res)) { 309612fa5e24SAl Viro error = PTR_ERR(res); 309712fa5e24SAl Viro goto out_dput; 309812fa5e24SAl Viro } 309912fa5e24SAl Viro dput(dentry); 310012fa5e24SAl Viro dentry = res; 310112fa5e24SAl Viro } 310254ef4872SMiklos Szeredi } 310354ef4872SMiklos Szeredi 3104d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 31051643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 310673a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3107ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3108ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3109ce8644fcSAl Viro error = -EACCES; 3110d58ffd35SMiklos Szeredi goto out_dput; 311164894cf8SAl Viro } 3112ce8644fcSAl Viro error = dir_inode->i_op->create(dir_inode, dentry, mode, 31131643b43fSAl Viro open_flag & O_EXCL); 3114d58ffd35SMiklos Szeredi if (error) 3115d58ffd35SMiklos Szeredi goto out_dput; 3116ce8644fcSAl Viro fsnotify_create(dir_inode, dentry); 3117d58ffd35SMiklos Szeredi } 31181643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 31191643b43fSAl Viro error = create_error; 3120d58ffd35SMiklos Szeredi goto out_dput; 3121d58ffd35SMiklos Szeredi } 3122da5ebf5aSAl Viro return dentry; 3123d58ffd35SMiklos Szeredi 3124d58ffd35SMiklos Szeredi out_dput: 3125d58ffd35SMiklos Szeredi dput(dentry); 3126da5ebf5aSAl Viro return ERR_PTR(error); 3127d58ffd35SMiklos Szeredi } 3128d58ffd35SMiklos Szeredi 3129c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 31303ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3131fb1cc555SAl Viro { 3132a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3133ca344a89SAl Viro int open_flag = op->open_flag; 313464894cf8SAl Viro bool got_write = false; 3135254cf582SAl Viro unsigned seq; 3136a1eb3315SMiklos Szeredi struct inode *inode; 3137da5ebf5aSAl Viro struct dentry *dentry; 3138b0417d2cSAl Viro const char *res; 313916c2cd71SAl Viro int error; 3140fb1cc555SAl Viro 3141c3e380b0SAl Viro nd->flags |= op->intent; 3142c3e380b0SAl Viro 3143bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 314456676ec3SAl Viro if (nd->depth) 314556676ec3SAl Viro put_link(nd); 31467521f22bSAl Viro res = handle_dots(nd, nd->last_type); 31477521f22bSAl Viro if (likely(!res)) 31487521f22bSAl Viro res = ERR_PTR(complete_walk(nd)); 31497521f22bSAl Viro return res; 31501f36f774SAl Viro } 3151a2c36b45SAl Viro 3152ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3153fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3154fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3155fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 315620e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 315720e34357SAl Viro if (IS_ERR(dentry)) 31581ccac622SAl Viro return ERR_CAST(dentry); 315920e34357SAl Viro if (likely(dentry)) 316071574865SMiklos Szeredi goto finish_lookup; 316171574865SMiklos Szeredi 31626583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3163b6183df7SMiklos Szeredi } else { 3164fe2d35ffSAl Viro /* create side of things */ 3165a3fbbde7SAl Viro /* 3166b6183df7SMiklos Szeredi * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED 3167b6183df7SMiklos Szeredi * has been cleared when we got to the last component we are 3168b6183df7SMiklos Szeredi * about to look up 3169a3fbbde7SAl Viro */ 31709f1fafeeSAl Viro error = complete_walk(nd); 3171c981a482SAl Viro if (unlikely(error)) 31721ccac622SAl Viro return ERR_PTR(error); 3173fe2d35ffSAl Viro 3174c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 31751f36f774SAl Viro /* trailing slashes? */ 3176deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 31771ccac622SAl Viro return ERR_PTR(-EISDIR); 3178b6183df7SMiklos Szeredi } 31791f36f774SAl Viro 31809cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 318164894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 318264894cf8SAl Viro if (!error) 318364894cf8SAl Viro got_write = true; 318464894cf8SAl Viro /* 318564894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 318664894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 318764894cf8SAl Viro * dropping this one anyway. 318864894cf8SAl Viro */ 318964894cf8SAl Viro } 31909cf843e3SAl Viro if (open_flag & O_CREAT) 31915955102cSAl Viro inode_lock(dir->d_inode); 31929cf843e3SAl Viro else 31939cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3194da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 31959cf843e3SAl Viro if (open_flag & O_CREAT) 31965955102cSAl Viro inode_unlock(dir->d_inode); 31979cf843e3SAl Viro else 31989cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3199fb1cc555SAl Viro 3200c981a482SAl Viro if (got_write) 320159e96e65SAl Viro mnt_drop_write(nd->path.mnt); 32026c0d46c4SAl Viro 320359e96e65SAl Viro if (IS_ERR(dentry)) 320459e96e65SAl Viro return ERR_CAST(dentry); 320559e96e65SAl Viro 3206973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3207e73cabffSAl Viro dput(nd->path.dentry); 3208e73cabffSAl Viro nd->path.dentry = dentry; 3209c981a482SAl Viro return NULL; 3210fb1cc555SAl Viro } 3211fb1cc555SAl Viro 321220e34357SAl Viro finish_lookup: 321356676ec3SAl Viro if (nd->depth) 321456676ec3SAl Viro put_link(nd); 32158c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3216b0417d2cSAl Viro if (unlikely(res)) { 32171ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3218b0417d2cSAl Viro return res; 32191ccac622SAl Viro } 322031d1726dSAl Viro 322131d1726dSAl Viro if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) { 322231d1726dSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 32231ccac622SAl Viro return ERR_PTR(-EEXIST); 322431d1726dSAl Viro } 3225c981a482SAl Viro 32268f64fb1cSAl Viro /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */ 3227c981a482SAl Viro return ERR_PTR(complete_walk(nd)); 3228c981a482SAl Viro } 3229c981a482SAl Viro 3230c981a482SAl Viro /* 3231c981a482SAl Viro * Handle the last step of open() 3232c981a482SAl Viro */ 3233c981a482SAl Viro static const char *do_last(struct nameidata *nd, 3234c981a482SAl Viro struct file *file, const struct open_flags *op) 3235c981a482SAl Viro { 3236c981a482SAl Viro kuid_t dir_uid = nd->inode->i_uid; 3237c981a482SAl Viro umode_t dir_mode = nd->inode->i_mode; 3238c981a482SAl Viro int open_flag = op->open_flag; 3239c981a482SAl Viro bool do_truncate; 3240c981a482SAl Viro int acc_mode; 3241c981a482SAl Viro const char *link; 3242c981a482SAl Viro int error; 3243c981a482SAl Viro 3244c981a482SAl Viro link = open_last_lookups(nd, file, op); 3245c981a482SAl Viro if (unlikely(link)) 3246c981a482SAl Viro return link; 3247c981a482SAl Viro 3248973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 324976ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 325030aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 325130aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 325259e96e65SAl Viro return ERR_PTR(-EISDIR); 3253d0cb5018SAl Viro error = may_create_in_sticky(dir_mode, dir_uid, 325430aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 325530aba665SSalvatore Mesoraca if (unlikely(error)) 325659e96e65SAl Viro return ERR_PTR(error); 325730aba665SSalvatore Mesoraca } 325844b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 325959e96e65SAl Viro return ERR_PTR(-ENOTDIR); 32606c0d46c4SAl Viro 32618795e7d4SAl Viro do_truncate = false; 32628795e7d4SAl Viro acc_mode = op->acc_mode; 32635a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 32645a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 32655a2d3eddSAl Viro open_flag &= ~O_TRUNC; 32665a2d3eddSAl Viro acc_mode = 0; 32678795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 32680f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 32690f9d1a10SAl Viro if (error) 327059e96e65SAl Viro return ERR_PTR(error); 32718795e7d4SAl Viro do_truncate = true; 32720f9d1a10SAl Viro } 3273bcda7652SAl Viro error = may_open(&nd->path, acc_mode, open_flag); 32748795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3275ae2bb293SAl Viro error = vfs_open(&nd->path, file); 32768795e7d4SAl Viro if (!error) 32776035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 32788795e7d4SAl Viro if (!error && do_truncate) 32792675a4ebSAl Viro error = handle_truncate(file); 3280c80567c8SAl Viro if (unlikely(error > 0)) { 3281c80567c8SAl Viro WARN_ON(1); 3282c80567c8SAl Viro error = -EINVAL; 3283c80567c8SAl Viro } 32848795e7d4SAl Viro if (do_truncate) 32850f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 32861ccac622SAl Viro return ERR_PTR(error); 3287fb1cc555SAl Viro } 3288fb1cc555SAl Viro 3289af7bd4dcSAmir Goldstein struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag) 3290af7bd4dcSAmir Goldstein { 3291af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3292af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3293af7bd4dcSAmir Goldstein struct inode *inode; 3294af7bd4dcSAmir Goldstein int error; 3295af7bd4dcSAmir Goldstein 3296af7bd4dcSAmir Goldstein /* we want directory to be writable */ 3297af7bd4dcSAmir Goldstein error = inode_permission(dir, MAY_WRITE | MAY_EXEC); 3298af7bd4dcSAmir Goldstein if (error) 3299af7bd4dcSAmir Goldstein goto out_err; 3300af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3301af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3302af7bd4dcSAmir Goldstein goto out_err; 3303af7bd4dcSAmir Goldstein error = -ENOMEM; 3304cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3305af7bd4dcSAmir Goldstein if (unlikely(!child)) 3306af7bd4dcSAmir Goldstein goto out_err; 3307af7bd4dcSAmir Goldstein error = dir->i_op->tmpfile(dir, child, mode); 3308af7bd4dcSAmir Goldstein if (error) 3309af7bd4dcSAmir Goldstein goto out_err; 3310af7bd4dcSAmir Goldstein error = -ENOENT; 3311af7bd4dcSAmir Goldstein inode = child->d_inode; 3312af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3313af7bd4dcSAmir Goldstein goto out_err; 3314af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3315af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3316af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3317af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3318af7bd4dcSAmir Goldstein } 3319fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3320af7bd4dcSAmir Goldstein return child; 3321af7bd4dcSAmir Goldstein 3322af7bd4dcSAmir Goldstein out_err: 3323af7bd4dcSAmir Goldstein dput(child); 3324af7bd4dcSAmir Goldstein return ERR_PTR(error); 3325af7bd4dcSAmir Goldstein } 3326af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3327af7bd4dcSAmir Goldstein 3328c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 332960545d0dSAl Viro const struct open_flags *op, 33303ec2eef1SAl Viro struct file *file) 333160545d0dSAl Viro { 3332625b6d10SAl Viro struct dentry *child; 3333625b6d10SAl Viro struct path path; 3334c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 333560545d0dSAl Viro if (unlikely(error)) 333660545d0dSAl Viro return error; 3337625b6d10SAl Viro error = mnt_want_write(path.mnt); 333860545d0dSAl Viro if (unlikely(error)) 333960545d0dSAl Viro goto out; 3340af7bd4dcSAmir Goldstein child = vfs_tmpfile(path.dentry, op->mode, op->open_flag); 3341af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3342684e73beSHirofumi Nakagawa if (IS_ERR(child)) 334360545d0dSAl Viro goto out2; 3344625b6d10SAl Viro dput(path.dentry); 3345625b6d10SAl Viro path.dentry = child; 3346c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 334769a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 334862fb4a15SAl Viro error = may_open(&path, 0, op->open_flag); 334960545d0dSAl Viro if (error) 335060545d0dSAl Viro goto out2; 3351625b6d10SAl Viro file->f_path.mnt = path.mnt; 3352be12af3eSAl Viro error = finish_open(file, child, NULL); 335360545d0dSAl Viro out2: 3354625b6d10SAl Viro mnt_drop_write(path.mnt); 335560545d0dSAl Viro out: 3356625b6d10SAl Viro path_put(&path); 335760545d0dSAl Viro return error; 335860545d0dSAl Viro } 335960545d0dSAl Viro 33606ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 33616ac08709SAl Viro { 33626ac08709SAl Viro struct path path; 33636ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 33646ac08709SAl Viro if (!error) { 33656ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3366ae2bb293SAl Viro error = vfs_open(&path, file); 33676ac08709SAl Viro path_put(&path); 33686ac08709SAl Viro } 33696ac08709SAl Viro return error; 33706ac08709SAl Viro } 33716ac08709SAl Viro 3372c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3373c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 33741da177e4SLinus Torvalds { 337530d90494SAl Viro struct file *file; 337613aab428SAl Viro int error; 337731e6b01fSNick Piggin 3378ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 33791afc99beSAl Viro if (IS_ERR(file)) 33801afc99beSAl Viro return file; 338131e6b01fSNick Piggin 3382bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 33833ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 33845f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 33856ac08709SAl Viro error = do_o_path(nd, flags, file); 33865f336e72SAl Viro } else { 33875f336e72SAl Viro const char *s = path_init(nd, flags); 33883bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 33891ccac622SAl Viro (s = do_last(nd, file, op)) != NULL) 33901ccac622SAl Viro ; 3391deb106c6SAl Viro terminate_walk(nd); 33925f336e72SAl Viro } 33937c1c01ecSAl Viro if (likely(!error)) { 3394aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 33957c1c01ecSAl Viro return file; 33967c1c01ecSAl Viro WARN_ON(1); 33977c1c01ecSAl Viro error = -EINVAL; 3398015c3bbcSMiklos Szeredi } 33997c1c01ecSAl Viro fput(file); 34002675a4ebSAl Viro if (error == -EOPENSTALE) { 34012675a4ebSAl Viro if (flags & LOOKUP_RCU) 34022675a4ebSAl Viro error = -ECHILD; 34032675a4ebSAl Viro else 34042675a4ebSAl Viro error = -ESTALE; 34052675a4ebSAl Viro } 34067c1c01ecSAl Viro return ERR_PTR(error); 3407de459215SKirill Korotaev } 34081da177e4SLinus Torvalds 3409669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3410f9652e10SAl Viro const struct open_flags *op) 341113aab428SAl Viro { 34129883d185SAl Viro struct nameidata nd; 3413f9652e10SAl Viro int flags = op->lookup_flags; 341413aab428SAl Viro struct file *filp; 341513aab428SAl Viro 34169883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3417c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 341813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3419c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 342013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3421c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 34229883d185SAl Viro restore_nameidata(); 342313aab428SAl Viro return filp; 342413aab428SAl Viro } 342513aab428SAl Viro 342673d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3427f9652e10SAl Viro const char *name, const struct open_flags *op) 342873d049a4SAl Viro { 34299883d185SAl Viro struct nameidata nd; 343073d049a4SAl Viro struct file *file; 343151689104SPaul Moore struct filename *filename; 3432f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 343373d049a4SAl Viro 343473d049a4SAl Viro nd.root.mnt = mnt; 343573d049a4SAl Viro nd.root.dentry = dentry; 343673d049a4SAl Viro 3437b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 343873d049a4SAl Viro return ERR_PTR(-ELOOP); 343973d049a4SAl Viro 344051689104SPaul Moore filename = getname_kernel(name); 3441a1c83681SViresh Kumar if (IS_ERR(filename)) 344251689104SPaul Moore return ERR_CAST(filename); 344351689104SPaul Moore 34449883d185SAl Viro set_nameidata(&nd, -1, filename); 3445c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 344673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3447c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 344873d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3449c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 34509883d185SAl Viro restore_nameidata(); 345151689104SPaul Moore putname(filename); 345273d049a4SAl Viro return file; 345373d049a4SAl Viro } 345473d049a4SAl Viro 3455fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 34561ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 34571da177e4SLinus Torvalds { 3458c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3459391172c4SAl Viro struct qstr last; 3460391172c4SAl Viro int type; 3461c30dabfeSJan Kara int err2; 34621ac12b4bSJeff Layton int error; 34631ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 34641ac12b4bSJeff Layton 34651ac12b4bSJeff Layton /* 34661ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 34671ac12b4bSJeff Layton * other flags passed in are ignored! 34681ac12b4bSJeff Layton */ 34691ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 34701ac12b4bSJeff Layton 34715c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 34725c31b6ceSAl Viro if (IS_ERR(name)) 34735c31b6ceSAl Viro return ERR_CAST(name); 34741da177e4SLinus Torvalds 3475c663e5d8SChristoph Hellwig /* 3476c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3477c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3478c663e5d8SChristoph Hellwig */ 34795c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3480ed75e95dSAl Viro goto out; 3481c663e5d8SChristoph Hellwig 3482c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3483391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3484c663e5d8SChristoph Hellwig /* 3485c663e5d8SChristoph Hellwig * Do the final lookup. 3486c663e5d8SChristoph Hellwig */ 3487391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 34885955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3489391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 34901da177e4SLinus Torvalds if (IS_ERR(dentry)) 3491a8104a9fSAl Viro goto unlock; 3492c663e5d8SChristoph Hellwig 3493a8104a9fSAl Viro error = -EEXIST; 3494b18825a7SDavid Howells if (d_is_positive(dentry)) 3495a8104a9fSAl Viro goto fail; 3496b18825a7SDavid Howells 3497c663e5d8SChristoph Hellwig /* 3498c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3499c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3500c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3501c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3502c663e5d8SChristoph Hellwig */ 3503391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3504a8104a9fSAl Viro error = -ENOENT; 3505ed75e95dSAl Viro goto fail; 3506e9baf6e5SAl Viro } 3507c30dabfeSJan Kara if (unlikely(err2)) { 3508c30dabfeSJan Kara error = err2; 3509a8104a9fSAl Viro goto fail; 3510c30dabfeSJan Kara } 3511181c37b6SAl Viro putname(name); 3512e9baf6e5SAl Viro return dentry; 35131da177e4SLinus Torvalds fail: 3514a8104a9fSAl Viro dput(dentry); 3515a8104a9fSAl Viro dentry = ERR_PTR(error); 3516a8104a9fSAl Viro unlock: 35175955102cSAl Viro inode_unlock(path->dentry->d_inode); 3518c30dabfeSJan Kara if (!err2) 3519391172c4SAl Viro mnt_drop_write(path->mnt); 3520ed75e95dSAl Viro out: 3521391172c4SAl Viro path_put(path); 3522181c37b6SAl Viro putname(name); 3523ed75e95dSAl Viro return dentry; 3524dae6ad8fSAl Viro } 3525fa14a0b8SAl Viro 3526fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3527fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3528fa14a0b8SAl Viro { 3529181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3530181c37b6SAl Viro path, lookup_flags); 3531fa14a0b8SAl Viro } 3532dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3533dae6ad8fSAl Viro 3534921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3535921a1650SAl Viro { 3536921a1650SAl Viro dput(dentry); 35375955102cSAl Viro inode_unlock(path->dentry->d_inode); 3538a8104a9fSAl Viro mnt_drop_write(path->mnt); 3539921a1650SAl Viro path_put(path); 3540921a1650SAl Viro } 3541921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3542921a1650SAl Viro 3543520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 35441ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3545dae6ad8fSAl Viro { 3546181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3547dae6ad8fSAl Viro } 3548dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3549dae6ad8fSAl Viro 35501a67aafbSAl Viro int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev) 35511da177e4SLinus Torvalds { 3552a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 35531da177e4SLinus Torvalds 35541da177e4SLinus Torvalds if (error) 35551da177e4SLinus Torvalds return error; 35561da177e4SLinus Torvalds 355794f82008SChristian Brauner if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD)) 35581da177e4SLinus Torvalds return -EPERM; 35591da177e4SLinus Torvalds 3560acfa4380SAl Viro if (!dir->i_op->mknod) 35611da177e4SLinus Torvalds return -EPERM; 35621da177e4SLinus Torvalds 356308ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 356408ce5f16SSerge E. Hallyn if (error) 356508ce5f16SSerge E. Hallyn return error; 356608ce5f16SSerge E. Hallyn 35671da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 35681da177e4SLinus Torvalds if (error) 35691da177e4SLinus Torvalds return error; 35701da177e4SLinus Torvalds 35711da177e4SLinus Torvalds error = dir->i_op->mknod(dir, dentry, mode, dev); 3572a74574aaSStephen Smalley if (!error) 3573f38aa942SAmy Griffis fsnotify_create(dir, dentry); 35741da177e4SLinus Torvalds return error; 35751da177e4SLinus Torvalds } 35764d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 35771da177e4SLinus Torvalds 3578f69aac00SAl Viro static int may_mknod(umode_t mode) 3579463c3197SDave Hansen { 3580463c3197SDave Hansen switch (mode & S_IFMT) { 3581463c3197SDave Hansen case S_IFREG: 3582463c3197SDave Hansen case S_IFCHR: 3583463c3197SDave Hansen case S_IFBLK: 3584463c3197SDave Hansen case S_IFIFO: 3585463c3197SDave Hansen case S_IFSOCK: 3586463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3587463c3197SDave Hansen return 0; 3588463c3197SDave Hansen case S_IFDIR: 3589463c3197SDave Hansen return -EPERM; 3590463c3197SDave Hansen default: 3591463c3197SDave Hansen return -EINVAL; 3592463c3197SDave Hansen } 3593463c3197SDave Hansen } 3594463c3197SDave Hansen 359587c4e192SDominik Brodowski long do_mknodat(int dfd, const char __user *filename, umode_t mode, 359687c4e192SDominik Brodowski unsigned int dev) 35971da177e4SLinus Torvalds { 35981da177e4SLinus Torvalds struct dentry *dentry; 3599dae6ad8fSAl Viro struct path path; 3600dae6ad8fSAl Viro int error; 3601972567f1SJeff Layton unsigned int lookup_flags = 0; 36021da177e4SLinus Torvalds 36038e4bfca1SAl Viro error = may_mknod(mode); 36048e4bfca1SAl Viro if (error) 36058e4bfca1SAl Viro return error; 3606972567f1SJeff Layton retry: 3607972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3608dae6ad8fSAl Viro if (IS_ERR(dentry)) 3609dae6ad8fSAl Viro return PTR_ERR(dentry); 36102ad94ae6SAl Viro 3611dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3612ce3b0f8dSAl Viro mode &= ~current_umask(); 3613dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3614be6d3e56SKentaro Takeda if (error) 3615a8104a9fSAl Viro goto out; 36161da177e4SLinus Torvalds switch (mode & S_IFMT) { 36171da177e4SLinus Torvalds case 0: case S_IFREG: 3618312b63fbSAl Viro error = vfs_create(path.dentry->d_inode,dentry,mode,true); 361905d1a717SMimi Zohar if (!error) 362005d1a717SMimi Zohar ima_post_path_mknod(dentry); 36211da177e4SLinus Torvalds break; 36221da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3623dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode, 36241da177e4SLinus Torvalds new_decode_dev(dev)); 36251da177e4SLinus Torvalds break; 36261da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3627dae6ad8fSAl Viro error = vfs_mknod(path.dentry->d_inode,dentry,mode,0); 36281da177e4SLinus Torvalds break; 36291da177e4SLinus Torvalds } 3630a8104a9fSAl Viro out: 3631921a1650SAl Viro done_path_create(&path, dentry); 3632972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3633972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3634972567f1SJeff Layton goto retry; 3635972567f1SJeff Layton } 36361da177e4SLinus Torvalds return error; 36371da177e4SLinus Torvalds } 36381da177e4SLinus Torvalds 363987c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 364087c4e192SDominik Brodowski unsigned int, dev) 364187c4e192SDominik Brodowski { 364287c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 364387c4e192SDominik Brodowski } 364487c4e192SDominik Brodowski 36458208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 36465590ff0dSUlrich Drepper { 364787c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 36485590ff0dSUlrich Drepper } 36495590ff0dSUlrich Drepper 365018bb1db3SAl Viro int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) 36511da177e4SLinus Torvalds { 3652a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 36538de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 36541da177e4SLinus Torvalds 36551da177e4SLinus Torvalds if (error) 36561da177e4SLinus Torvalds return error; 36571da177e4SLinus Torvalds 3658acfa4380SAl Viro if (!dir->i_op->mkdir) 36591da177e4SLinus Torvalds return -EPERM; 36601da177e4SLinus Torvalds 36611da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 36621da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 36631da177e4SLinus Torvalds if (error) 36641da177e4SLinus Torvalds return error; 36651da177e4SLinus Torvalds 36668de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 36678de52778SAl Viro return -EMLINK; 36688de52778SAl Viro 36691da177e4SLinus Torvalds error = dir->i_op->mkdir(dir, dentry, mode); 3670a74574aaSStephen Smalley if (!error) 3671f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 36721da177e4SLinus Torvalds return error; 36731da177e4SLinus Torvalds } 36744d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 36751da177e4SLinus Torvalds 36760101db7aSDominik Brodowski long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 36771da177e4SLinus Torvalds { 36786902d925SDave Hansen struct dentry *dentry; 3679dae6ad8fSAl Viro struct path path; 3680dae6ad8fSAl Viro int error; 3681b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 36821da177e4SLinus Torvalds 3683b76d8b82SJeff Layton retry: 3684b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 36856902d925SDave Hansen if (IS_ERR(dentry)) 3686dae6ad8fSAl Viro return PTR_ERR(dentry); 36876902d925SDave Hansen 3688dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3689ce3b0f8dSAl Viro mode &= ~current_umask(); 3690dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 3691a8104a9fSAl Viro if (!error) 3692dae6ad8fSAl Viro error = vfs_mkdir(path.dentry->d_inode, dentry, mode); 3693921a1650SAl Viro done_path_create(&path, dentry); 3694b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3695b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3696b76d8b82SJeff Layton goto retry; 3697b76d8b82SJeff Layton } 36981da177e4SLinus Torvalds return error; 36991da177e4SLinus Torvalds } 37001da177e4SLinus Torvalds 37010101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 37020101db7aSDominik Brodowski { 37030101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 37040101db7aSDominik Brodowski } 37050101db7aSDominik Brodowski 3706a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 37075590ff0dSUlrich Drepper { 37080101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 37095590ff0dSUlrich Drepper } 37105590ff0dSUlrich Drepper 37111da177e4SLinus Torvalds int vfs_rmdir(struct inode *dir, struct dentry *dentry) 37121da177e4SLinus Torvalds { 37131da177e4SLinus Torvalds int error = may_delete(dir, dentry, 1); 37141da177e4SLinus Torvalds 37151da177e4SLinus Torvalds if (error) 37161da177e4SLinus Torvalds return error; 37171da177e4SLinus Torvalds 3718acfa4380SAl Viro if (!dir->i_op->rmdir) 37191da177e4SLinus Torvalds return -EPERM; 37201da177e4SLinus Torvalds 37211d2ef590SAl Viro dget(dentry); 37225955102cSAl Viro inode_lock(dentry->d_inode); 3723912dbc15SSage Weil 37241da177e4SLinus Torvalds error = -EBUSY; 37257af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3726912dbc15SSage Weil goto out; 3727912dbc15SSage Weil 37281da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3729912dbc15SSage Weil if (error) 3730912dbc15SSage Weil goto out; 3731912dbc15SSage Weil 37321da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3733912dbc15SSage Weil if (error) 3734912dbc15SSage Weil goto out; 3735912dbc15SSage Weil 37368767712fSAl Viro shrink_dcache_parent(dentry); 37371da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3738d83c49f3SAl Viro dont_mount(dentry); 37398ed936b5SEric W. Biederman detach_mounts(dentry); 3740116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 37411da177e4SLinus Torvalds 3742912dbc15SSage Weil out: 37435955102cSAl Viro inode_unlock(dentry->d_inode); 37441d2ef590SAl Viro dput(dentry); 3745912dbc15SSage Weil if (!error) 3746912dbc15SSage Weil d_delete(dentry); 37471da177e4SLinus Torvalds return error; 37481da177e4SLinus Torvalds } 37494d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 37501da177e4SLinus Torvalds 3751f459dffaSDominik Brodowski long do_rmdir(int dfd, const char __user *pathname) 37521da177e4SLinus Torvalds { 37531da177e4SLinus Torvalds int error = 0; 375491a27b2aSJeff Layton struct filename *name; 37551da177e4SLinus Torvalds struct dentry *dentry; 3756f5beed75SAl Viro struct path path; 3757f5beed75SAl Viro struct qstr last; 3758f5beed75SAl Viro int type; 3759c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3760c6ee9206SJeff Layton retry: 3761c1d4dd27SAl Viro name = filename_parentat(dfd, getname(pathname), lookup_flags, 3762c1d4dd27SAl Viro &path, &last, &type); 376391a27b2aSJeff Layton if (IS_ERR(name)) 376491a27b2aSJeff Layton return PTR_ERR(name); 37651da177e4SLinus Torvalds 3766f5beed75SAl Viro switch (type) { 37671da177e4SLinus Torvalds case LAST_DOTDOT: 37681da177e4SLinus Torvalds error = -ENOTEMPTY; 37691da177e4SLinus Torvalds goto exit1; 37701da177e4SLinus Torvalds case LAST_DOT: 37711da177e4SLinus Torvalds error = -EINVAL; 37721da177e4SLinus Torvalds goto exit1; 37731da177e4SLinus Torvalds case LAST_ROOT: 37741da177e4SLinus Torvalds error = -EBUSY; 37751da177e4SLinus Torvalds goto exit1; 37761da177e4SLinus Torvalds } 37770612d9fbSOGAWA Hirofumi 3778f5beed75SAl Viro error = mnt_want_write(path.mnt); 3779c30dabfeSJan Kara if (error) 3780c30dabfeSJan Kara goto exit1; 37810612d9fbSOGAWA Hirofumi 37825955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3783f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 37841da177e4SLinus Torvalds error = PTR_ERR(dentry); 37856902d925SDave Hansen if (IS_ERR(dentry)) 37866902d925SDave Hansen goto exit2; 3787e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3788e6bc45d6STheodore Ts'o error = -ENOENT; 3789e6bc45d6STheodore Ts'o goto exit3; 3790e6bc45d6STheodore Ts'o } 3791f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3792be6d3e56SKentaro Takeda if (error) 3793c30dabfeSJan Kara goto exit3; 3794f5beed75SAl Viro error = vfs_rmdir(path.dentry->d_inode, dentry); 37950622753bSDave Hansen exit3: 37961da177e4SLinus Torvalds dput(dentry); 37976902d925SDave Hansen exit2: 37985955102cSAl Viro inode_unlock(path.dentry->d_inode); 3799f5beed75SAl Viro mnt_drop_write(path.mnt); 38001da177e4SLinus Torvalds exit1: 3801f5beed75SAl Viro path_put(&path); 38021da177e4SLinus Torvalds putname(name); 3803c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3804c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3805c6ee9206SJeff Layton goto retry; 3806c6ee9206SJeff Layton } 38071da177e4SLinus Torvalds return error; 38081da177e4SLinus Torvalds } 38091da177e4SLinus Torvalds 38103cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 38115590ff0dSUlrich Drepper { 38125590ff0dSUlrich Drepper return do_rmdir(AT_FDCWD, pathname); 38135590ff0dSUlrich Drepper } 38145590ff0dSUlrich Drepper 3815b21996e3SJ. Bruce Fields /** 3816b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 3817b21996e3SJ. Bruce Fields * @dir: parent directory 3818b21996e3SJ. Bruce Fields * @dentry: victim 3819b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3820b21996e3SJ. Bruce Fields * 3821b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3822b21996e3SJ. Bruce Fields * 3823b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3824b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3825b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3826b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3827b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3828b21996e3SJ. Bruce Fields * 3829b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 3830b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 3831b21996e3SJ. Bruce Fields * to be NFS exported. 3832b21996e3SJ. Bruce Fields */ 3833b21996e3SJ. Bruce Fields int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode) 38341da177e4SLinus Torvalds { 38359accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 38361da177e4SLinus Torvalds int error = may_delete(dir, dentry, 0); 38371da177e4SLinus Torvalds 38381da177e4SLinus Torvalds if (error) 38391da177e4SLinus Torvalds return error; 38401da177e4SLinus Torvalds 3841acfa4380SAl Viro if (!dir->i_op->unlink) 38421da177e4SLinus Torvalds return -EPERM; 38431da177e4SLinus Torvalds 38445955102cSAl Viro inode_lock(target); 38458ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 38461da177e4SLinus Torvalds error = -EBUSY; 38471da177e4SLinus Torvalds else { 38481da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 3849bec1052eSAl Viro if (!error) { 38505a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 38515a14696cSJ. Bruce Fields if (error) 3852b21996e3SJ. Bruce Fields goto out; 38531da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 38548ed936b5SEric W. Biederman if (!error) { 3855d83c49f3SAl Viro dont_mount(dentry); 38568ed936b5SEric W. Biederman detach_mounts(dentry); 3857116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 38588ed936b5SEric W. Biederman } 3859bec1052eSAl Viro } 38601da177e4SLinus Torvalds } 3861b21996e3SJ. Bruce Fields out: 38625955102cSAl Viro inode_unlock(target); 38631da177e4SLinus Torvalds 38641da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 38651da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 38669accbb97SJ. Bruce Fields fsnotify_link_count(target); 38671da177e4SLinus Torvalds d_delete(dentry); 38681da177e4SLinus Torvalds } 38690eeca283SRobert Love 38701da177e4SLinus Torvalds return error; 38711da177e4SLinus Torvalds } 38724d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 38731da177e4SLinus Torvalds 38741da177e4SLinus Torvalds /* 38751da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 38761b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 38771da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 38781da177e4SLinus Torvalds * while waiting on the I/O. 38791da177e4SLinus Torvalds */ 3880da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 38811da177e4SLinus Torvalds { 38822ad94ae6SAl Viro int error; 38831da177e4SLinus Torvalds struct dentry *dentry; 3884f5beed75SAl Viro struct path path; 3885f5beed75SAl Viro struct qstr last; 3886f5beed75SAl Viro int type; 38871da177e4SLinus Torvalds struct inode *inode = NULL; 3888b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 38895d18f813SJeff Layton unsigned int lookup_flags = 0; 38905d18f813SJeff Layton retry: 3891da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 389291a27b2aSJeff Layton if (IS_ERR(name)) 389391a27b2aSJeff Layton return PTR_ERR(name); 38942ad94ae6SAl Viro 38951da177e4SLinus Torvalds error = -EISDIR; 3896f5beed75SAl Viro if (type != LAST_NORM) 38971da177e4SLinus Torvalds goto exit1; 38980612d9fbSOGAWA Hirofumi 3899f5beed75SAl Viro error = mnt_want_write(path.mnt); 3900c30dabfeSJan Kara if (error) 3901c30dabfeSJan Kara goto exit1; 3902b21996e3SJ. Bruce Fields retry_deleg: 39035955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3904f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39051da177e4SLinus Torvalds error = PTR_ERR(dentry); 39061da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 39071da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 3908f5beed75SAl Viro if (last.name[last.len]) 390950338b88STörök Edwin goto slashes; 39101da177e4SLinus Torvalds inode = dentry->d_inode; 3911b18825a7SDavid Howells if (d_is_negative(dentry)) 3912e6bc45d6STheodore Ts'o goto slashes; 39137de9c6eeSAl Viro ihold(inode); 3914f5beed75SAl Viro error = security_path_unlink(&path, dentry); 3915be6d3e56SKentaro Takeda if (error) 3916c30dabfeSJan Kara goto exit2; 3917f5beed75SAl Viro error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode); 39181da177e4SLinus Torvalds exit2: 39191da177e4SLinus Torvalds dput(dentry); 39201da177e4SLinus Torvalds } 39215955102cSAl Viro inode_unlock(path.dentry->d_inode); 39221da177e4SLinus Torvalds if (inode) 39231da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 3924b21996e3SJ. Bruce Fields inode = NULL; 3925b21996e3SJ. Bruce Fields if (delegated_inode) { 39265a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 3927b21996e3SJ. Bruce Fields if (!error) 3928b21996e3SJ. Bruce Fields goto retry_deleg; 3929b21996e3SJ. Bruce Fields } 3930f5beed75SAl Viro mnt_drop_write(path.mnt); 39311da177e4SLinus Torvalds exit1: 3932f5beed75SAl Viro path_put(&path); 39335d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 39345d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 39355d18f813SJeff Layton inode = NULL; 39365d18f813SJeff Layton goto retry; 39375d18f813SJeff Layton } 3938da2f1362SChristoph Hellwig putname(name); 39391da177e4SLinus Torvalds return error; 39401da177e4SLinus Torvalds 39411da177e4SLinus Torvalds slashes: 3942b18825a7SDavid Howells if (d_is_negative(dentry)) 3943b18825a7SDavid Howells error = -ENOENT; 394444b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 3945b18825a7SDavid Howells error = -EISDIR; 3946b18825a7SDavid Howells else 3947b18825a7SDavid Howells error = -ENOTDIR; 39481da177e4SLinus Torvalds goto exit2; 39491da177e4SLinus Torvalds } 39501da177e4SLinus Torvalds 39512e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 39525590ff0dSUlrich Drepper { 39535590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 39545590ff0dSUlrich Drepper return -EINVAL; 39555590ff0dSUlrich Drepper 39565590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 39575590ff0dSUlrich Drepper return do_rmdir(dfd, pathname); 39585590ff0dSUlrich Drepper 3959da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 39605590ff0dSUlrich Drepper } 39615590ff0dSUlrich Drepper 39623480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 39635590ff0dSUlrich Drepper { 3964da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 39655590ff0dSUlrich Drepper } 39665590ff0dSUlrich Drepper 3967db2e747bSMiklos Szeredi int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname) 39681da177e4SLinus Torvalds { 3969a95164d9SMiklos Szeredi int error = may_create(dir, dentry); 39701da177e4SLinus Torvalds 39711da177e4SLinus Torvalds if (error) 39721da177e4SLinus Torvalds return error; 39731da177e4SLinus Torvalds 3974acfa4380SAl Viro if (!dir->i_op->symlink) 39751da177e4SLinus Torvalds return -EPERM; 39761da177e4SLinus Torvalds 39771da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 39781da177e4SLinus Torvalds if (error) 39791da177e4SLinus Torvalds return error; 39801da177e4SLinus Torvalds 39811da177e4SLinus Torvalds error = dir->i_op->symlink(dir, dentry, oldname); 3982a74574aaSStephen Smalley if (!error) 3983f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39841da177e4SLinus Torvalds return error; 39851da177e4SLinus Torvalds } 39864d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 39871da177e4SLinus Torvalds 3988b724e846SDominik Brodowski long do_symlinkat(const char __user *oldname, int newdfd, 3989b724e846SDominik Brodowski const char __user *newname) 39901da177e4SLinus Torvalds { 39912ad94ae6SAl Viro int error; 399291a27b2aSJeff Layton struct filename *from; 39936902d925SDave Hansen struct dentry *dentry; 3994dae6ad8fSAl Viro struct path path; 3995f46d3567SJeff Layton unsigned int lookup_flags = 0; 39961da177e4SLinus Torvalds 39971da177e4SLinus Torvalds from = getname(oldname); 39981da177e4SLinus Torvalds if (IS_ERR(from)) 39991da177e4SLinus Torvalds return PTR_ERR(from); 4000f46d3567SJeff Layton retry: 4001f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 40021da177e4SLinus Torvalds error = PTR_ERR(dentry); 40036902d925SDave Hansen if (IS_ERR(dentry)) 4004dae6ad8fSAl Viro goto out_putname; 40056902d925SDave Hansen 400691a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4007a8104a9fSAl Viro if (!error) 400891a27b2aSJeff Layton error = vfs_symlink(path.dentry->d_inode, dentry, from->name); 4009921a1650SAl Viro done_path_create(&path, dentry); 4010f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4011f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4012f46d3567SJeff Layton goto retry; 4013f46d3567SJeff Layton } 40146902d925SDave Hansen out_putname: 40151da177e4SLinus Torvalds putname(from); 40161da177e4SLinus Torvalds return error; 40171da177e4SLinus Torvalds } 40181da177e4SLinus Torvalds 4019b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4020b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4021b724e846SDominik Brodowski { 4022b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4023b724e846SDominik Brodowski } 4024b724e846SDominik Brodowski 40253480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 40265590ff0dSUlrich Drepper { 4027b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 40285590ff0dSUlrich Drepper } 40295590ff0dSUlrich Drepper 4030146a8595SJ. Bruce Fields /** 4031146a8595SJ. Bruce Fields * vfs_link - create a new link 4032146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4033146a8595SJ. Bruce Fields * @dir: new parent 4034146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4035146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4036146a8595SJ. Bruce Fields * 4037146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4038146a8595SJ. Bruce Fields * 4039146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4040146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4041146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4042146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4043146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4044146a8595SJ. Bruce Fields * 4045146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4046146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4047146a8595SJ. Bruce Fields * to be NFS exported. 4048146a8595SJ. Bruce Fields */ 4049146a8595SJ. Bruce Fields int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode) 40501da177e4SLinus Torvalds { 40511da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 40528de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40531da177e4SLinus Torvalds int error; 40541da177e4SLinus Torvalds 40551da177e4SLinus Torvalds if (!inode) 40561da177e4SLinus Torvalds return -ENOENT; 40571da177e4SLinus Torvalds 4058a95164d9SMiklos Szeredi error = may_create(dir, new_dentry); 40591da177e4SLinus Torvalds if (error) 40601da177e4SLinus Torvalds return error; 40611da177e4SLinus Torvalds 40621da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 40631da177e4SLinus Torvalds return -EXDEV; 40641da177e4SLinus Torvalds 40651da177e4SLinus Torvalds /* 40661da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 40671da177e4SLinus Torvalds */ 40681da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 40691da177e4SLinus Torvalds return -EPERM; 40700bd23d09SEric W. Biederman /* 40710bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 40720bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 40730bd23d09SEric W. Biederman * the vfs. 40740bd23d09SEric W. Biederman */ 40750bd23d09SEric W. Biederman if (HAS_UNMAPPED_ID(inode)) 40760bd23d09SEric W. Biederman return -EPERM; 4077acfa4380SAl Viro if (!dir->i_op->link) 40781da177e4SLinus Torvalds return -EPERM; 40797e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 40801da177e4SLinus Torvalds return -EPERM; 40811da177e4SLinus Torvalds 40821da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 40831da177e4SLinus Torvalds if (error) 40841da177e4SLinus Torvalds return error; 40851da177e4SLinus Torvalds 40865955102cSAl Viro inode_lock(inode); 4087aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4088f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4089aae8a97dSAneesh Kumar K.V error = -ENOENT; 40908de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 40918de52778SAl Viro error = -EMLINK; 4092146a8595SJ. Bruce Fields else { 4093146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4094146a8595SJ. Bruce Fields if (!error) 40951da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4096146a8595SJ. Bruce Fields } 4097f4e0c30cSAl Viro 4098f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4099f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4100f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4101f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4102f4e0c30cSAl Viro } 41035955102cSAl Viro inode_unlock(inode); 4104e31e14ecSStephen Smalley if (!error) 41057e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 41061da177e4SLinus Torvalds return error; 41071da177e4SLinus Torvalds } 41084d359507SAl Viro EXPORT_SYMBOL(vfs_link); 41091da177e4SLinus Torvalds 41101da177e4SLinus Torvalds /* 41111da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 41121da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 41131da177e4SLinus Torvalds * newname. --KAB 41141da177e4SLinus Torvalds * 41151da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 41161da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 41171da177e4SLinus Torvalds * and other special files. --ADM 41181da177e4SLinus Torvalds */ 411946ea89ebSDominik Brodowski int do_linkat(int olddfd, const char __user *oldname, int newdfd, 412046ea89ebSDominik Brodowski const char __user *newname, int flags) 41211da177e4SLinus Torvalds { 41221da177e4SLinus Torvalds struct dentry *new_dentry; 4123dae6ad8fSAl Viro struct path old_path, new_path; 4124146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 412511a7b371SAneesh Kumar K.V int how = 0; 41261da177e4SLinus Torvalds int error; 41271da177e4SLinus Torvalds 412811a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4129c04030e1SUlrich Drepper return -EINVAL; 413011a7b371SAneesh Kumar K.V /* 4131f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4132f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4133f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 413411a7b371SAneesh Kumar K.V */ 4135f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4136f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4137f0cc6ffbSLinus Torvalds return -ENOENT; 413811a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4139f0cc6ffbSLinus Torvalds } 4140c04030e1SUlrich Drepper 414111a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 414211a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4143442e31caSJeff Layton retry: 414411a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 41451da177e4SLinus Torvalds if (error) 41462ad94ae6SAl Viro return error; 41472ad94ae6SAl Viro 4148442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4149442e31caSJeff Layton (how & LOOKUP_REVAL)); 41501da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 41516902d925SDave Hansen if (IS_ERR(new_dentry)) 4152dae6ad8fSAl Viro goto out; 4153dae6ad8fSAl Viro 4154dae6ad8fSAl Viro error = -EXDEV; 4155dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4156dae6ad8fSAl Viro goto out_dput; 4157800179c9SKees Cook error = may_linkat(&old_path); 4158800179c9SKees Cook if (unlikely(error)) 4159800179c9SKees Cook goto out_dput; 4160dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4161be6d3e56SKentaro Takeda if (error) 4162a8104a9fSAl Viro goto out_dput; 4163146a8595SJ. Bruce Fields error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode); 416475c3f29dSDave Hansen out_dput: 4165921a1650SAl Viro done_path_create(&new_path, new_dentry); 4166146a8595SJ. Bruce Fields if (delegated_inode) { 4167146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4168d22e6338SOleg Drokin if (!error) { 4169d22e6338SOleg Drokin path_put(&old_path); 4170146a8595SJ. Bruce Fields goto retry; 4171146a8595SJ. Bruce Fields } 4172d22e6338SOleg Drokin } 4173442e31caSJeff Layton if (retry_estale(error, how)) { 4174d22e6338SOleg Drokin path_put(&old_path); 4175442e31caSJeff Layton how |= LOOKUP_REVAL; 4176442e31caSJeff Layton goto retry; 4177442e31caSJeff Layton } 41781da177e4SLinus Torvalds out: 41792d8f3038SAl Viro path_put(&old_path); 41801da177e4SLinus Torvalds 41811da177e4SLinus Torvalds return error; 41821da177e4SLinus Torvalds } 41831da177e4SLinus Torvalds 418446ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 418546ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 418646ea89ebSDominik Brodowski { 418746ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 418846ea89ebSDominik Brodowski } 418946ea89ebSDominik Brodowski 41903480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 41915590ff0dSUlrich Drepper { 419246ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 41935590ff0dSUlrich Drepper } 41945590ff0dSUlrich Drepper 4195bc27027aSMiklos Szeredi /** 4196bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 4197bc27027aSMiklos Szeredi * @old_dir: parent of source 4198bc27027aSMiklos Szeredi * @old_dentry: source 4199bc27027aSMiklos Szeredi * @new_dir: parent of destination 4200bc27027aSMiklos Szeredi * @new_dentry: destination 4201bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4202520c8b16SMiklos Szeredi * @flags: rename flags 4203bc27027aSMiklos Szeredi * 4204bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4205bc27027aSMiklos Szeredi * 4206bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4207bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4208bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4209bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4210bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4211bc27027aSMiklos Szeredi * so. 4212bc27027aSMiklos Szeredi * 4213bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4214bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4215bc27027aSMiklos Szeredi * to be NFS exported. 4216bc27027aSMiklos Szeredi * 42171da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 42181da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 42191da177e4SLinus Torvalds * Problems: 42200117d427SMauro Carvalho Chehab * 4221d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 42221da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 42231da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4224a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 42251da177e4SLinus Torvalds * story. 42266cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 42276cedba89SJ. Bruce Fields * and source (if it is not a directory). 42281b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 42291da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 42301da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4231a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 42321da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 42331da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 42341da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4235a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 42361da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 42371da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 42381da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4239e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 42401b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 42411da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4242c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 42431da177e4SLinus Torvalds * locking]. 42441da177e4SLinus Torvalds */ 42451da177e4SLinus Torvalds int vfs_rename(struct inode *old_dir, struct dentry *old_dentry, 42468e6d782cSJ. Bruce Fields struct inode *new_dir, struct dentry *new_dentry, 4247520c8b16SMiklos Szeredi struct inode **delegated_inode, unsigned int flags) 42481da177e4SLinus Torvalds { 42491da177e4SLinus Torvalds int error; 4250bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4251bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4252bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4253da1ce067SMiklos Szeredi bool new_is_dir = false; 4254da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 425549d31c2fSAl Viro struct name_snapshot old_name; 42561da177e4SLinus Torvalds 42578d3e2936SMiklos Szeredi if (source == target) 42581da177e4SLinus Torvalds return 0; 42591da177e4SLinus Torvalds 42601da177e4SLinus Torvalds error = may_delete(old_dir, old_dentry, is_dir); 42611da177e4SLinus Torvalds if (error) 42621da177e4SLinus Torvalds return error; 42631da177e4SLinus Torvalds 4264da1ce067SMiklos Szeredi if (!target) { 4265a95164d9SMiklos Szeredi error = may_create(new_dir, new_dentry); 4266da1ce067SMiklos Szeredi } else { 4267da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4268da1ce067SMiklos Szeredi 4269da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 42701da177e4SLinus Torvalds error = may_delete(new_dir, new_dentry, is_dir); 4271da1ce067SMiklos Szeredi else 4272da1ce067SMiklos Szeredi error = may_delete(new_dir, new_dentry, new_is_dir); 4273da1ce067SMiklos Szeredi } 42741da177e4SLinus Torvalds if (error) 42751da177e4SLinus Torvalds return error; 42761da177e4SLinus Torvalds 42772773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 42781da177e4SLinus Torvalds return -EPERM; 42791da177e4SLinus Torvalds 4280bc27027aSMiklos Szeredi /* 4281bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4282bc27027aSMiklos Szeredi * we'll need to flip '..'. 4283bc27027aSMiklos Szeredi */ 4284da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4285da1ce067SMiklos Szeredi if (is_dir) { 4286bc27027aSMiklos Szeredi error = inode_permission(source, MAY_WRITE); 4287bc27027aSMiklos Szeredi if (error) 4288bc27027aSMiklos Szeredi return error; 4289bc27027aSMiklos Szeredi } 4290da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 4291da1ce067SMiklos Szeredi error = inode_permission(target, MAY_WRITE); 4292da1ce067SMiklos Szeredi if (error) 4293da1ce067SMiklos Szeredi return error; 4294da1ce067SMiklos Szeredi } 4295da1ce067SMiklos Szeredi } 42960eeca283SRobert Love 42970b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 42980b3974ebSMiklos Szeredi flags); 4299bc27027aSMiklos Szeredi if (error) 4300bc27027aSMiklos Szeredi return error; 4301bc27027aSMiklos Szeredi 430249d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4303bc27027aSMiklos Szeredi dget(new_dentry); 4304da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4305bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4306bc27027aSMiklos Szeredi else if (target) 43075955102cSAl Viro inode_lock(target); 4308bc27027aSMiklos Szeredi 4309bc27027aSMiklos Szeredi error = -EBUSY; 43107af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4311bc27027aSMiklos Szeredi goto out; 4312bc27027aSMiklos Szeredi 4313da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4314bc27027aSMiklos Szeredi error = -EMLINK; 4315da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4316bc27027aSMiklos Szeredi goto out; 4317da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4318da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4319da1ce067SMiklos Szeredi goto out; 4320da1ce067SMiklos Szeredi } 4321da1ce067SMiklos Szeredi if (!is_dir) { 4322bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4323bc27027aSMiklos Szeredi if (error) 4324bc27027aSMiklos Szeredi goto out; 4325da1ce067SMiklos Szeredi } 4326da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4327bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4328bc27027aSMiklos Szeredi if (error) 4329bc27027aSMiklos Szeredi goto out; 4330bc27027aSMiklos Szeredi } 4331520c8b16SMiklos Szeredi error = old_dir->i_op->rename(old_dir, old_dentry, 4332520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4333bc27027aSMiklos Szeredi if (error) 4334bc27027aSMiklos Szeredi goto out; 4335bc27027aSMiklos Szeredi 4336da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 43378767712fSAl Viro if (is_dir) { 43388767712fSAl Viro shrink_dcache_parent(new_dentry); 4339bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 43408767712fSAl Viro } 4341bc27027aSMiklos Szeredi dont_mount(new_dentry); 43428ed936b5SEric W. Biederman detach_mounts(new_dentry); 4343bc27027aSMiklos Szeredi } 4344da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4345da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4346bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4347da1ce067SMiklos Szeredi else 4348da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4349da1ce067SMiklos Szeredi } 4350bc27027aSMiklos Szeredi out: 4351da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4352bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4353bc27027aSMiklos Szeredi else if (target) 43545955102cSAl Viro inode_unlock(target); 4355bc27027aSMiklos Szeredi dput(new_dentry); 4356da1ce067SMiklos Szeredi if (!error) { 4357f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4358da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4359da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4360f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4361da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4362da1ce067SMiklos Szeredi } 4363da1ce067SMiklos Szeredi } 436449d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 43650eeca283SRobert Love 43661da177e4SLinus Torvalds return error; 43671da177e4SLinus Torvalds } 43684d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 43691da177e4SLinus Torvalds 4370ee81feb6SDominik Brodowski static int do_renameat2(int olddfd, const char __user *oldname, int newdfd, 4371ee81feb6SDominik Brodowski const char __user *newname, unsigned int flags) 43721da177e4SLinus Torvalds { 43731da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 43741da177e4SLinus Torvalds struct dentry *trap; 4375f5beed75SAl Viro struct path old_path, new_path; 4376f5beed75SAl Viro struct qstr old_last, new_last; 4377f5beed75SAl Viro int old_type, new_type; 43788e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 437991a27b2aSJeff Layton struct filename *from; 438091a27b2aSJeff Layton struct filename *to; 4381f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4382c6a94284SJeff Layton bool should_retry = false; 43832ad94ae6SAl Viro int error; 4384520c8b16SMiklos Szeredi 43850d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4386da1ce067SMiklos Szeredi return -EINVAL; 4387da1ce067SMiklos Szeredi 43880d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 43890d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4390520c8b16SMiklos Szeredi return -EINVAL; 4391520c8b16SMiklos Szeredi 43920d7a8555SMiklos Szeredi if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD)) 43930d7a8555SMiklos Szeredi return -EPERM; 43940d7a8555SMiklos Szeredi 4395f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4396f5beed75SAl Viro target_flags = 0; 4397f5beed75SAl Viro 4398c6a94284SJeff Layton retry: 4399c1d4dd27SAl Viro from = filename_parentat(olddfd, getname(oldname), lookup_flags, 4400c1d4dd27SAl Viro &old_path, &old_last, &old_type); 440191a27b2aSJeff Layton if (IS_ERR(from)) { 440291a27b2aSJeff Layton error = PTR_ERR(from); 44031da177e4SLinus Torvalds goto exit; 440491a27b2aSJeff Layton } 44051da177e4SLinus Torvalds 4406c1d4dd27SAl Viro to = filename_parentat(newdfd, getname(newname), lookup_flags, 4407c1d4dd27SAl Viro &new_path, &new_last, &new_type); 440891a27b2aSJeff Layton if (IS_ERR(to)) { 440991a27b2aSJeff Layton error = PTR_ERR(to); 44101da177e4SLinus Torvalds goto exit1; 441191a27b2aSJeff Layton } 44121da177e4SLinus Torvalds 44131da177e4SLinus Torvalds error = -EXDEV; 4414f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 44151da177e4SLinus Torvalds goto exit2; 44161da177e4SLinus Torvalds 44171da177e4SLinus Torvalds error = -EBUSY; 4418f5beed75SAl Viro if (old_type != LAST_NORM) 44191da177e4SLinus Torvalds goto exit2; 44201da177e4SLinus Torvalds 44210a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 44220a7c3937SMiklos Szeredi error = -EEXIST; 4423f5beed75SAl Viro if (new_type != LAST_NORM) 44241da177e4SLinus Torvalds goto exit2; 44251da177e4SLinus Torvalds 4426f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4427c30dabfeSJan Kara if (error) 4428c30dabfeSJan Kara goto exit2; 4429c30dabfeSJan Kara 44308e6d782cSJ. Bruce Fields retry_deleg: 4431f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 44321da177e4SLinus Torvalds 4433f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 44341da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 44351da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 44361da177e4SLinus Torvalds goto exit3; 44371da177e4SLinus Torvalds /* source must exist */ 44381da177e4SLinus Torvalds error = -ENOENT; 4439b18825a7SDavid Howells if (d_is_negative(old_dentry)) 44401da177e4SLinus Torvalds goto exit4; 4441f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 44421da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44431da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 44441da177e4SLinus Torvalds goto exit4; 44450a7c3937SMiklos Szeredi error = -EEXIST; 44460a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 44470a7c3937SMiklos Szeredi goto exit5; 4448da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4449da1ce067SMiklos Szeredi error = -ENOENT; 4450da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4451da1ce067SMiklos Szeredi goto exit5; 4452da1ce067SMiklos Szeredi 4453da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4454da1ce067SMiklos Szeredi error = -ENOTDIR; 4455f5beed75SAl Viro if (new_last.name[new_last.len]) 4456da1ce067SMiklos Szeredi goto exit5; 4457da1ce067SMiklos Szeredi } 4458da1ce067SMiklos Szeredi } 44590a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 44600a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 44610a7c3937SMiklos Szeredi error = -ENOTDIR; 4462f5beed75SAl Viro if (old_last.name[old_last.len]) 44630a7c3937SMiklos Szeredi goto exit5; 4464f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 44650a7c3937SMiklos Szeredi goto exit5; 44660a7c3937SMiklos Szeredi } 44670a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 44680a7c3937SMiklos Szeredi error = -EINVAL; 44690a7c3937SMiklos Szeredi if (old_dentry == trap) 44700a7c3937SMiklos Szeredi goto exit5; 44711da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4472da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44731da177e4SLinus Torvalds error = -ENOTEMPTY; 44741da177e4SLinus Torvalds if (new_dentry == trap) 44751da177e4SLinus Torvalds goto exit5; 44761da177e4SLinus Torvalds 4477f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4478f5beed75SAl Viro &new_path, new_dentry, flags); 4479be6d3e56SKentaro Takeda if (error) 4480c30dabfeSJan Kara goto exit5; 4481f5beed75SAl Viro error = vfs_rename(old_path.dentry->d_inode, old_dentry, 4482f5beed75SAl Viro new_path.dentry->d_inode, new_dentry, 4483520c8b16SMiklos Szeredi &delegated_inode, flags); 44841da177e4SLinus Torvalds exit5: 44851da177e4SLinus Torvalds dput(new_dentry); 44861da177e4SLinus Torvalds exit4: 44871da177e4SLinus Torvalds dput(old_dentry); 44881da177e4SLinus Torvalds exit3: 4489f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 44908e6d782cSJ. Bruce Fields if (delegated_inode) { 44918e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 44928e6d782cSJ. Bruce Fields if (!error) 44938e6d782cSJ. Bruce Fields goto retry_deleg; 44948e6d782cSJ. Bruce Fields } 4495f5beed75SAl Viro mnt_drop_write(old_path.mnt); 44961da177e4SLinus Torvalds exit2: 4497c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4498c6a94284SJeff Layton should_retry = true; 4499f5beed75SAl Viro path_put(&new_path); 45002ad94ae6SAl Viro putname(to); 45011da177e4SLinus Torvalds exit1: 4502f5beed75SAl Viro path_put(&old_path); 45031da177e4SLinus Torvalds putname(from); 4504c6a94284SJeff Layton if (should_retry) { 4505c6a94284SJeff Layton should_retry = false; 4506c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4507c6a94284SJeff Layton goto retry; 4508c6a94284SJeff Layton } 45092ad94ae6SAl Viro exit: 45101da177e4SLinus Torvalds return error; 45111da177e4SLinus Torvalds } 45121da177e4SLinus Torvalds 4513ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4514ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4515ee81feb6SDominik Brodowski { 4516ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, flags); 4517ee81feb6SDominik Brodowski } 4518ee81feb6SDominik Brodowski 4519520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4520520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4521520c8b16SMiklos Szeredi { 4522ee81feb6SDominik Brodowski return do_renameat2(olddfd, oldname, newdfd, newname, 0); 4523520c8b16SMiklos Szeredi } 4524520c8b16SMiklos Szeredi 4525a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 45265590ff0dSUlrich Drepper { 4527ee81feb6SDominik Brodowski return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 45285590ff0dSUlrich Drepper } 45295590ff0dSUlrich Drepper 4530787fb6bcSMiklos Szeredi int vfs_whiteout(struct inode *dir, struct dentry *dentry) 4531787fb6bcSMiklos Szeredi { 4532787fb6bcSMiklos Szeredi int error = may_create(dir, dentry); 4533787fb6bcSMiklos Szeredi if (error) 4534787fb6bcSMiklos Szeredi return error; 4535787fb6bcSMiklos Szeredi 4536787fb6bcSMiklos Szeredi if (!dir->i_op->mknod) 4537787fb6bcSMiklos Szeredi return -EPERM; 4538787fb6bcSMiklos Szeredi 4539787fb6bcSMiklos Szeredi return dir->i_op->mknod(dir, dentry, 4540787fb6bcSMiklos Szeredi S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV); 4541787fb6bcSMiklos Szeredi } 4542787fb6bcSMiklos Szeredi EXPORT_SYMBOL(vfs_whiteout); 4543787fb6bcSMiklos Szeredi 45445d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 45451da177e4SLinus Torvalds { 45465d826c84SAl Viro int len = PTR_ERR(link); 45471da177e4SLinus Torvalds if (IS_ERR(link)) 45481da177e4SLinus Torvalds goto out; 45491da177e4SLinus Torvalds 45501da177e4SLinus Torvalds len = strlen(link); 45511da177e4SLinus Torvalds if (len > (unsigned) buflen) 45521da177e4SLinus Torvalds len = buflen; 45531da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 45541da177e4SLinus Torvalds len = -EFAULT; 45551da177e4SLinus Torvalds out: 45561da177e4SLinus Torvalds return len; 45571da177e4SLinus Torvalds } 45581da177e4SLinus Torvalds 4559d60874cdSMiklos Szeredi /** 4560fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4561fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4562fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4563fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4564fd4a0edfSMiklos Szeredi * 4565fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4566fd4a0edfSMiklos Szeredi * 4567fd4a0edfSMiklos Szeredi * Does not call security hook. 4568fd4a0edfSMiklos Szeredi */ 4569fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4570fd4a0edfSMiklos Szeredi { 4571fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4572f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4573f2df5da6SAl Viro const char *link; 4574f2df5da6SAl Viro int res; 4575fd4a0edfSMiklos Szeredi 457676fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 457776fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 457876fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 457976fca90eSMiklos Szeredi 458076fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4581fd4a0edfSMiklos Szeredi return -EINVAL; 4582fd4a0edfSMiklos Szeredi 458376fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 458476fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 458576fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 458676fca90eSMiklos Szeredi } 458776fca90eSMiklos Szeredi 45884c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4589f2df5da6SAl Viro if (!link) { 4590f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4591f2df5da6SAl Viro if (IS_ERR(link)) 4592f2df5da6SAl Viro return PTR_ERR(link); 4593f2df5da6SAl Viro } 4594f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4595f2df5da6SAl Viro do_delayed_call(&done); 4596f2df5da6SAl Viro return res; 4597fd4a0edfSMiklos Szeredi } 4598fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 45991da177e4SLinus Torvalds 4600d60874cdSMiklos Szeredi /** 4601d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4602d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4603d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4604d60874cdSMiklos Szeredi * 4605d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4606d60874cdSMiklos Szeredi * 4607d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4608d60874cdSMiklos Szeredi * 4609d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4610d60874cdSMiklos Szeredi */ 4611d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4612d60874cdSMiklos Szeredi { 4613d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4614d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4615d60874cdSMiklos Szeredi 4616d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4617d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4618d60874cdSMiklos Szeredi if (!res) 4619d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4620d60874cdSMiklos Szeredi } 4621d60874cdSMiklos Szeredi return res; 4622d60874cdSMiklos Szeredi } 4623d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4624d60874cdSMiklos Szeredi 46251da177e4SLinus Torvalds /* get the link contents into pagecache */ 46266b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4627fceef393SAl Viro struct delayed_call *callback) 46281da177e4SLinus Torvalds { 4629ebd09abbSDuane Griffin char *kaddr; 46301da177e4SLinus Torvalds struct page *page; 46316b255391SAl Viro struct address_space *mapping = inode->i_mapping; 46326b255391SAl Viro 4633d3883d4fSAl Viro if (!dentry) { 4634d3883d4fSAl Viro page = find_get_page(mapping, 0); 4635d3883d4fSAl Viro if (!page) 46366b255391SAl Viro return ERR_PTR(-ECHILD); 4637d3883d4fSAl Viro if (!PageUptodate(page)) { 4638d3883d4fSAl Viro put_page(page); 4639d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4640d3883d4fSAl Viro } 4641d3883d4fSAl Viro } else { 4642090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 46431da177e4SLinus Torvalds if (IS_ERR(page)) 46446fe6900eSNick Piggin return (char*)page; 4645d3883d4fSAl Viro } 4646fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 464721fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 464821fc61c7SAl Viro kaddr = page_address(page); 46496b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4650ebd09abbSDuane Griffin return kaddr; 46511da177e4SLinus Torvalds } 46521da177e4SLinus Torvalds 46536b255391SAl Viro EXPORT_SYMBOL(page_get_link); 46541da177e4SLinus Torvalds 4655fceef393SAl Viro void page_put_link(void *arg) 46561da177e4SLinus Torvalds { 4657fceef393SAl Viro put_page(arg); 46581da177e4SLinus Torvalds } 46594d359507SAl Viro EXPORT_SYMBOL(page_put_link); 46601da177e4SLinus Torvalds 4661aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4662aa80deabSAl Viro { 4663fceef393SAl Viro DEFINE_DELAYED_CALL(done); 46646b255391SAl Viro int res = readlink_copy(buffer, buflen, 46656b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4666fceef393SAl Viro &done)); 4667fceef393SAl Viro do_delayed_call(&done); 4668aa80deabSAl Viro return res; 4669aa80deabSAl Viro } 4670aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4671aa80deabSAl Viro 467254566b2cSNick Piggin /* 467354566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 467454566b2cSNick Piggin */ 467554566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 46761da177e4SLinus Torvalds { 46771da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 46780adb25d2SKirill Korotaev struct page *page; 4679afddba49SNick Piggin void *fsdata; 4680beb497abSDmitriy Monakhov int err; 4681c718a975STetsuo Handa unsigned int flags = 0; 468254566b2cSNick Piggin if (nofs) 468354566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 46841da177e4SLinus Torvalds 46857e53cac4SNeilBrown retry: 4686afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 468754566b2cSNick Piggin flags, &page, &fsdata); 46881da177e4SLinus Torvalds if (err) 4689afddba49SNick Piggin goto fail; 4690afddba49SNick Piggin 469121fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4692afddba49SNick Piggin 4693afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4694afddba49SNick Piggin page, fsdata); 46951da177e4SLinus Torvalds if (err < 0) 46961da177e4SLinus Torvalds goto fail; 4697afddba49SNick Piggin if (err < len-1) 4698afddba49SNick Piggin goto retry; 4699afddba49SNick Piggin 47001da177e4SLinus Torvalds mark_inode_dirty(inode); 47011da177e4SLinus Torvalds return 0; 47021da177e4SLinus Torvalds fail: 47031da177e4SLinus Torvalds return err; 47041da177e4SLinus Torvalds } 47054d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 47061da177e4SLinus Torvalds 47070adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 47080adb25d2SKirill Korotaev { 47090adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4710c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 47110adb25d2SKirill Korotaev } 47124d359507SAl Viro EXPORT_SYMBOL(page_symlink); 47130adb25d2SKirill Korotaev 471492e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 47156b255391SAl Viro .get_link = page_get_link, 47161da177e4SLinus Torvalds }; 47171da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4718