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 * 2078228e2c3SDmitry Kadashev getname_uflags(const char __user *filename, int uflags) 2088228e2c3SDmitry Kadashev { 2098228e2c3SDmitry Kadashev int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 2108228e2c3SDmitry Kadashev 2118228e2c3SDmitry Kadashev return getname_flags(filename, flags, NULL); 2128228e2c3SDmitry Kadashev } 2138228e2c3SDmitry Kadashev 2148228e2c3SDmitry Kadashev struct filename * 21591a27b2aSJeff Layton getname(const char __user * filename) 216f52e0c11SAl Viro { 217f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 218f52e0c11SAl Viro } 219f52e0c11SAl Viro 220c4ad8f98SLinus Torvalds struct filename * 221c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 222c4ad8f98SLinus Torvalds { 223c4ad8f98SLinus Torvalds struct filename *result; 22408518549SPaul Moore int len = strlen(filename) + 1; 225c4ad8f98SLinus Torvalds 226c4ad8f98SLinus Torvalds result = __getname(); 227c4ad8f98SLinus Torvalds if (unlikely(!result)) 228c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 229c4ad8f98SLinus Torvalds 23008518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 231fd2f7cb5SAl Viro result->name = (char *)result->iname; 23208518549SPaul Moore } else if (len <= PATH_MAX) { 23330ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 23408518549SPaul Moore struct filename *tmp; 23508518549SPaul Moore 23630ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 23708518549SPaul Moore if (unlikely(!tmp)) { 23808518549SPaul Moore __putname(result); 23908518549SPaul Moore return ERR_PTR(-ENOMEM); 24008518549SPaul Moore } 24108518549SPaul Moore tmp->name = (char *)result; 24208518549SPaul Moore result = tmp; 24308518549SPaul Moore } else { 24408518549SPaul Moore __putname(result); 24508518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 24608518549SPaul Moore } 24708518549SPaul Moore memcpy((char *)result->name, filename, len); 248c4ad8f98SLinus Torvalds result->uptr = NULL; 249c4ad8f98SLinus Torvalds result->aname = NULL; 25055422d0bSPaul Moore result->refcnt = 1; 251fd3522fdSPaul Moore audit_getname(result); 252c4ad8f98SLinus Torvalds 253c4ad8f98SLinus Torvalds return result; 254c4ad8f98SLinus Torvalds } 255c4ad8f98SLinus Torvalds 25691a27b2aSJeff Layton void putname(struct filename *name) 2571da177e4SLinus Torvalds { 25891ef658fSDmitry Kadashev if (IS_ERR_OR_NULL(name)) 25991ef658fSDmitry Kadashev return; 26091ef658fSDmitry Kadashev 26155422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 26255422d0bSPaul Moore 26355422d0bSPaul Moore if (--name->refcnt > 0) 26455422d0bSPaul Moore return; 26555422d0bSPaul Moore 266fd2f7cb5SAl Viro if (name->name != name->iname) { 26755422d0bSPaul Moore __putname(name->name); 26855422d0bSPaul Moore kfree(name); 26955422d0bSPaul Moore } else 27055422d0bSPaul Moore __putname(name); 2711da177e4SLinus Torvalds } 2721da177e4SLinus Torvalds 27347291baaSChristian Brauner /** 27447291baaSChristian Brauner * check_acl - perform ACL permission checking 27547291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 27647291baaSChristian Brauner * @inode: inode to check permissions on 27747291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 27847291baaSChristian Brauner * 27947291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 28047291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 28147291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 28247291baaSChristian Brauner * 28347291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 28447291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 28547291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 28647291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 28747291baaSChristian Brauner * raw inode simply passs init_user_ns. 28847291baaSChristian Brauner */ 28947291baaSChristian Brauner static int check_acl(struct user_namespace *mnt_userns, 29047291baaSChristian Brauner struct inode *inode, int mask) 291e77819e5SLinus Torvalds { 29284635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 293e77819e5SLinus Torvalds struct posix_acl *acl; 294e77819e5SLinus Torvalds 295e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2963567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2973567866bSAl Viro if (!acl) 298e77819e5SLinus Torvalds return -EAGAIN; 2993567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 300b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 301e77819e5SLinus Torvalds return -ECHILD; 30247291baaSChristian Brauner return posix_acl_permission(mnt_userns, inode, acl, mask); 303e77819e5SLinus Torvalds } 304e77819e5SLinus Torvalds 3052982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 3064e34e719SChristoph Hellwig if (IS_ERR(acl)) 3074e34e719SChristoph Hellwig return PTR_ERR(acl); 308e77819e5SLinus Torvalds if (acl) { 30947291baaSChristian Brauner int error = posix_acl_permission(mnt_userns, inode, acl, mask); 310e77819e5SLinus Torvalds posix_acl_release(acl); 311e77819e5SLinus Torvalds return error; 312e77819e5SLinus Torvalds } 31384635d68SLinus Torvalds #endif 314e77819e5SLinus Torvalds 315e77819e5SLinus Torvalds return -EAGAIN; 316e77819e5SLinus Torvalds } 317e77819e5SLinus Torvalds 31847291baaSChristian Brauner /** 31947291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 32047291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 32147291baaSChristian Brauner * @inode: inode to check permissions on 32247291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3235fc475b7SLinus Torvalds * 32447291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 32547291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 32647291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 32747291baaSChristian Brauner * 32847291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 32947291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 33047291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 33147291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 33247291baaSChristian Brauner * raw inode simply passs init_user_ns. 3335909ccaaSLinus Torvalds */ 33447291baaSChristian Brauner static int acl_permission_check(struct user_namespace *mnt_userns, 33547291baaSChristian Brauner struct inode *inode, int mask) 3365909ccaaSLinus Torvalds { 33726cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 33847291baaSChristian Brauner kuid_t i_uid; 3395909ccaaSLinus Torvalds 3405fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 34147291baaSChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 34247291baaSChristian Brauner if (likely(uid_eq(current_fsuid(), i_uid))) { 3435fc475b7SLinus Torvalds mask &= 7; 3445909ccaaSLinus Torvalds mode >>= 6; 3455fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3465fc475b7SLinus Torvalds } 3475fc475b7SLinus Torvalds 3485fc475b7SLinus Torvalds /* Do we have ACL's? */ 349e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 35047291baaSChristian Brauner int error = check_acl(mnt_userns, inode, mask); 3515909ccaaSLinus Torvalds if (error != -EAGAIN) 3525909ccaaSLinus Torvalds return error; 3535909ccaaSLinus Torvalds } 3545909ccaaSLinus Torvalds 3555fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3565fc475b7SLinus Torvalds mask &= 7; 3575fc475b7SLinus Torvalds 3585fc475b7SLinus Torvalds /* 3595fc475b7SLinus Torvalds * Are the group permissions different from 3605fc475b7SLinus Torvalds * the other permissions in the bits we care 3615fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3625fc475b7SLinus Torvalds */ 3635fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 36447291baaSChristian Brauner kgid_t kgid = i_gid_into_mnt(mnt_userns, inode); 36547291baaSChristian Brauner if (in_group_p(kgid)) 3665909ccaaSLinus Torvalds mode >>= 3; 3675909ccaaSLinus Torvalds } 3685909ccaaSLinus Torvalds 3695fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3705fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3715909ccaaSLinus Torvalds } 3721da177e4SLinus Torvalds 3731da177e4SLinus Torvalds /** 3741da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 37547291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 3761da177e4SLinus Torvalds * @inode: inode to check access rights for 3775fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3785fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3791da177e4SLinus Torvalds * 3801da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3811da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3821da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 383b74c79e9SNick Piggin * are used for other things. 384b74c79e9SNick Piggin * 385b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 386b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 387b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 38847291baaSChristian Brauner * 38947291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39047291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39147291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39247291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39347291baaSChristian Brauner * raw inode simply passs init_user_ns. 3941da177e4SLinus Torvalds */ 39547291baaSChristian Brauner int generic_permission(struct user_namespace *mnt_userns, struct inode *inode, 39647291baaSChristian Brauner int mask) 3971da177e4SLinus Torvalds { 3985909ccaaSLinus Torvalds int ret; 3991da177e4SLinus Torvalds 4001da177e4SLinus Torvalds /* 401948409c7SAndreas Gruenbacher * Do the basic permission checks. 4021da177e4SLinus Torvalds */ 40347291baaSChristian Brauner ret = acl_permission_check(mnt_userns, inode, mask); 4045909ccaaSLinus Torvalds if (ret != -EACCES) 4055909ccaaSLinus Torvalds return ret; 4061da177e4SLinus Torvalds 407d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 408d594e7ecSAl Viro /* DACs are overridable for directories */ 409d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 41047291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 41123adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 412d594e7ecSAl Viro return 0; 41347291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4140558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4151da177e4SLinus Torvalds return 0; 4162a4c2242SStephen Smalley return -EACCES; 4172a4c2242SStephen Smalley } 4181da177e4SLinus Torvalds 4191da177e4SLinus Torvalds /* 4201da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4211da177e4SLinus Torvalds */ 4227ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 423d594e7ecSAl Viro if (mask == MAY_READ) 42447291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4250558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4261da177e4SLinus Torvalds return 0; 4272a4c2242SStephen Smalley /* 4282a4c2242SStephen Smalley * Read/write DACs are always overridable. 4292a4c2242SStephen Smalley * Executable DACs are overridable when there is 4302a4c2242SStephen Smalley * at least one exec bit set. 4312a4c2242SStephen Smalley */ 4322a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 43347291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4340558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4352a4c2242SStephen Smalley return 0; 4361da177e4SLinus Torvalds 4371da177e4SLinus Torvalds return -EACCES; 4381da177e4SLinus Torvalds } 4394d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4401da177e4SLinus Torvalds 44147291baaSChristian Brauner /** 44247291baaSChristian Brauner * do_inode_permission - UNIX permission checking 44347291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 44447291baaSChristian Brauner * @inode: inode to check permissions on 44547291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 44647291baaSChristian Brauner * 4473ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4483ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4493ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4503ddcd056SLinus Torvalds * permission function, use the fast case". 4513ddcd056SLinus Torvalds */ 45247291baaSChristian Brauner static inline int do_inode_permission(struct user_namespace *mnt_userns, 45347291baaSChristian Brauner struct inode *inode, int mask) 4543ddcd056SLinus Torvalds { 4553ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4563ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 457549c7297SChristian Brauner return inode->i_op->permission(mnt_userns, inode, mask); 4583ddcd056SLinus Torvalds 4593ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4603ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4613ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4623ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4633ddcd056SLinus Torvalds } 46447291baaSChristian Brauner return generic_permission(mnt_userns, inode, mask); 4653ddcd056SLinus Torvalds } 4663ddcd056SLinus Torvalds 467cb23beb5SChristoph Hellwig /** 4680bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4690bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 47055852635SRandy Dunlap * @inode: Inode to check permission on 4710bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4720bdaea90SDavid Howells * 4730bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4740bdaea90SDavid Howells */ 4750bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4760bdaea90SDavid Howells { 4770bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4780bdaea90SDavid Howells umode_t mode = inode->i_mode; 4790bdaea90SDavid Howells 4800bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 481bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4820bdaea90SDavid Howells return -EROFS; 4830bdaea90SDavid Howells } 4840bdaea90SDavid Howells return 0; 4850bdaea90SDavid Howells } 4860bdaea90SDavid Howells 4870bdaea90SDavid Howells /** 4880bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 48947291baaSChristian Brauner * @mnt_userns: User namespace of the mount the inode was found from 4900bdaea90SDavid Howells * @inode: Inode to check permission on 4910bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4920bdaea90SDavid Howells * 4930bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4940bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4950bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4960bdaea90SDavid Howells * 4970bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4980bdaea90SDavid Howells */ 49947291baaSChristian Brauner int inode_permission(struct user_namespace *mnt_userns, 50047291baaSChristian Brauner struct inode *inode, int mask) 5010bdaea90SDavid Howells { 5020bdaea90SDavid Howells int retval; 5030bdaea90SDavid Howells 5040bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 5050bdaea90SDavid Howells if (retval) 5060bdaea90SDavid Howells return retval; 5074bfd054aSEric Biggers 5084bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 5094bfd054aSEric Biggers /* 5104bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5114bfd054aSEric Biggers */ 5124bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5134bfd054aSEric Biggers return -EPERM; 5144bfd054aSEric Biggers 5154bfd054aSEric Biggers /* 5164bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5174bfd054aSEric Biggers * written back improperly if their true value is unknown 5184bfd054aSEric Biggers * to the vfs. 5194bfd054aSEric Biggers */ 520ba73d987SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 5214bfd054aSEric Biggers return -EACCES; 5224bfd054aSEric Biggers } 5234bfd054aSEric Biggers 52447291baaSChristian Brauner retval = do_inode_permission(mnt_userns, inode, mask); 5254bfd054aSEric Biggers if (retval) 5264bfd054aSEric Biggers return retval; 5274bfd054aSEric Biggers 5284bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5294bfd054aSEric Biggers if (retval) 5304bfd054aSEric Biggers return retval; 5314bfd054aSEric Biggers 5324bfd054aSEric Biggers return security_inode_permission(inode, mask); 5330bdaea90SDavid Howells } 5344d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5350bdaea90SDavid Howells 5360bdaea90SDavid Howells /** 5375dd784d0SJan Blunck * path_get - get a reference to a path 5385dd784d0SJan Blunck * @path: path to get the reference to 5395dd784d0SJan Blunck * 5405dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5415dd784d0SJan Blunck */ 542dcf787f3SAl Viro void path_get(const struct path *path) 5435dd784d0SJan Blunck { 5445dd784d0SJan Blunck mntget(path->mnt); 5455dd784d0SJan Blunck dget(path->dentry); 5465dd784d0SJan Blunck } 5475dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5485dd784d0SJan Blunck 5495dd784d0SJan Blunck /** 5501d957f9bSJan Blunck * path_put - put a reference to a path 5511d957f9bSJan Blunck * @path: path to put the reference to 5521d957f9bSJan Blunck * 5531d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5541d957f9bSJan Blunck */ 555dcf787f3SAl Viro void path_put(const struct path *path) 5561da177e4SLinus Torvalds { 5571d957f9bSJan Blunck dput(path->dentry); 5581d957f9bSJan Blunck mntput(path->mnt); 5591da177e4SLinus Torvalds } 5601d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5611da177e4SLinus Torvalds 562894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5631f55a6ecSAl Viro struct nameidata { 5641f55a6ecSAl Viro struct path path; 5651f55a6ecSAl Viro struct qstr last; 5661f55a6ecSAl Viro struct path root; 5671f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 568bcba1e7dSAl Viro unsigned int flags, state; 569ab87f9a5SAleksa Sarai unsigned seq, m_seq, r_seq; 5701f55a6ecSAl Viro int last_type; 5711f55a6ecSAl Viro unsigned depth; 572756daf26SNeilBrown int total_link_count; 573697fc6caSAl Viro struct saved { 574697fc6caSAl Viro struct path link; 575fceef393SAl Viro struct delayed_call done; 576697fc6caSAl Viro const char *name; 5770450b2d1SAl Viro unsigned seq; 578894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5799883d185SAl Viro struct filename *name; 5809883d185SAl Viro struct nameidata *saved; 5819883d185SAl Viro unsigned root_seq; 5829883d185SAl Viro int dfd; 5830f705953SAl Viro kuid_t dir_uid; 5840f705953SAl Viro umode_t dir_mode; 5853859a271SKees Cook } __randomize_layout; 5861f55a6ecSAl Viro 587bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 588bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 589bcba1e7dSAl Viro #define ND_JUMPED 4 590bcba1e7dSAl Viro 59106422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 592894bc8c4SAl Viro { 593756daf26SNeilBrown struct nameidata *old = current->nameidata; 594756daf26SNeilBrown p->stack = p->internal; 5957962c7d1SAl Viro p->depth = 0; 596c8a53ee5SAl Viro p->dfd = dfd; 597c8a53ee5SAl Viro p->name = name; 5987d01ef75SAl Viro p->path.mnt = NULL; 5997d01ef75SAl Viro p->path.dentry = NULL; 600756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 6019883d185SAl Viro p->saved = old; 602756daf26SNeilBrown current->nameidata = p; 603894bc8c4SAl Viro } 604894bc8c4SAl Viro 60506422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 60606422964SAl Viro const struct path *root) 60706422964SAl Viro { 60806422964SAl Viro __set_nameidata(p, dfd, name); 60906422964SAl Viro p->state = 0; 61006422964SAl Viro if (unlikely(root)) { 61106422964SAl Viro p->state = ND_ROOT_PRESET; 61206422964SAl Viro p->root = *root; 61306422964SAl Viro } 61406422964SAl Viro } 61506422964SAl Viro 6169883d185SAl Viro static void restore_nameidata(void) 617894bc8c4SAl Viro { 6189883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 619756daf26SNeilBrown 620756daf26SNeilBrown current->nameidata = old; 621756daf26SNeilBrown if (old) 622756daf26SNeilBrown old->total_link_count = now->total_link_count; 623e1a63bbcSAl Viro if (now->stack != now->internal) 624756daf26SNeilBrown kfree(now->stack); 625894bc8c4SAl Viro } 626894bc8c4SAl Viro 62760ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 628894bc8c4SAl Viro { 629bc40aee0SAl Viro struct saved *p; 630bc40aee0SAl Viro 6316da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 63260ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 633bc40aee0SAl Viro if (unlikely(!p)) 63460ef60c7SAl Viro return false; 635894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 636894bc8c4SAl Viro nd->stack = p; 63760ef60c7SAl Viro return true; 638894bc8c4SAl Viro } 639894bc8c4SAl Viro 640397d425dSEric W. Biederman /** 6416b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 642397d425dSEric W. Biederman * 643397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 644397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 645397d425dSEric W. Biederman */ 6466b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 647397d425dSEric W. Biederman { 64895dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 649397d425dSEric W. Biederman 650402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 651402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 652397d425dSEric W. Biederman return true; 653397d425dSEric W. Biederman 6546b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 655397d425dSEric W. Biederman } 656397d425dSEric W. Biederman 6577973387aSAl Viro static void drop_links(struct nameidata *nd) 6587973387aSAl Viro { 6597973387aSAl Viro int i = nd->depth; 6607973387aSAl Viro while (i--) { 6617973387aSAl Viro struct saved *last = nd->stack + i; 662fceef393SAl Viro do_delayed_call(&last->done); 663fceef393SAl Viro clear_delayed_call(&last->done); 6647973387aSAl Viro } 6657973387aSAl Viro } 6667973387aSAl Viro 6677973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6687973387aSAl Viro { 6697973387aSAl Viro drop_links(nd); 6707973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6717973387aSAl Viro int i; 6727973387aSAl Viro path_put(&nd->path); 6737973387aSAl Viro for (i = 0; i < nd->depth; i++) 6747973387aSAl Viro path_put(&nd->stack[i].link); 675bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 676102b8af2SAl Viro path_put(&nd->root); 677bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 678102b8af2SAl Viro } 6797973387aSAl Viro } else { 6807973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6817973387aSAl Viro rcu_read_unlock(); 6827973387aSAl Viro } 6837973387aSAl Viro nd->depth = 0; 6847d01ef75SAl Viro nd->path.mnt = NULL; 6857d01ef75SAl Viro nd->path.dentry = NULL; 6867973387aSAl Viro } 6877973387aSAl Viro 6887973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6892aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6907973387aSAl Viro { 6912aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6927973387aSAl Viro if (unlikely(res)) { 6937973387aSAl Viro if (res > 0) 6947973387aSAl Viro path->mnt = NULL; 6957973387aSAl Viro path->dentry = NULL; 6967973387aSAl Viro return false; 6977973387aSAl Viro } 6987973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6997973387aSAl Viro path->dentry = NULL; 7007973387aSAl Viro return false; 7017973387aSAl Viro } 7027973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7037973387aSAl Viro } 7047973387aSAl Viro 7052aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7062aa38470SAl Viro struct path *path, unsigned seq) 7072aa38470SAl Viro { 7085bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7092aa38470SAl Viro } 7102aa38470SAl Viro 7117973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7127973387aSAl Viro { 7137973387aSAl Viro int i; 714eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 715eacd9aa8SAl Viro drop_links(nd); 716eacd9aa8SAl Viro nd->depth = 0; 717eacd9aa8SAl Viro return false; 718eacd9aa8SAl Viro } 7197973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7207973387aSAl Viro struct saved *last = nd->stack + i; 7217973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7227973387aSAl Viro drop_links(nd); 7237973387aSAl Viro nd->depth = i + 1; 7247973387aSAl Viro return false; 7257973387aSAl Viro } 7267973387aSAl Viro } 7277973387aSAl Viro return true; 7287973387aSAl Viro } 7297973387aSAl Viro 730ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 731ee594bffSAl Viro { 732adb21d2bSAleksa Sarai /* 733adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 734adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 735adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 736adb21d2bSAleksa Sarai */ 737adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 738adb21d2bSAleksa Sarai return false; 739adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 740bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 741ee594bffSAl Viro return true; 742bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 743ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 744ee594bffSAl Viro } 745ee594bffSAl Viro 74619660af7SAl Viro /* 74731e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 74819660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 74919660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 75057e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 75119660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 75219660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 75319660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 75419660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 75531e6b01fSNick Piggin */ 75631e6b01fSNick Piggin 75731e6b01fSNick Piggin /** 758e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 75919660af7SAl Viro * @nd: nameidata pathwalk data 760e36cffedSJens Axboe * Returns: true on success, false on failure 76131e6b01fSNick Piggin * 762e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7634675ac39SAl Viro * for ref-walk mode. 7644675ac39SAl Viro * Must be called from rcu-walk context. 765e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7667973387aSAl Viro * terminate_walk(). 76731e6b01fSNick Piggin */ 768e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 76931e6b01fSNick Piggin { 77031e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 77131e6b01fSNick Piggin 77231e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 773e5c832d5SLinus Torvalds 774e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 7757973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7764675ac39SAl Viro goto out1; 77784a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 77884a2bd39SAl Viro goto out; 779ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7804675ac39SAl Viro goto out; 7814675ac39SAl Viro rcu_read_unlock(); 7824675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 783e36cffedSJens Axboe return true; 7844675ac39SAl Viro 78584a2bd39SAl Viro out1: 7864675ac39SAl Viro nd->path.mnt = NULL; 7874675ac39SAl Viro nd->path.dentry = NULL; 7884675ac39SAl Viro out: 7894675ac39SAl Viro rcu_read_unlock(); 790e36cffedSJens Axboe return false; 7914675ac39SAl Viro } 7924675ac39SAl Viro 7934675ac39SAl Viro /** 794ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7954675ac39SAl Viro * @nd: nameidata pathwalk data 796ae66db45SAl Viro * @dentry: next dentry to step into 797ae66db45SAl Viro * @seq: seq number to check @dentry against 798ae66db45SAl Viro * Returns: true on success, false on failure 7994675ac39SAl Viro * 800ae66db45SAl Viro * Similar to to try_to_unlazy(), but here we have the next dentry already 801ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 802ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 803ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 8044675ac39SAl Viro * terminate_walk(). 8054675ac39SAl Viro */ 806ae66db45SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq) 8074675ac39SAl Viro { 8084675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8094675ac39SAl Viro 8104675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 8114675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8124675ac39SAl Viro goto out2; 8137973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 8147973387aSAl Viro goto out2; 8154675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8167973387aSAl Viro goto out1; 81748a066e7SAl Viro 81815570086SLinus Torvalds /* 8194675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8204675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8214675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8224675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8234675ac39SAl Viro * know the parent must still be valid if the child sequence number is 82415570086SLinus Torvalds */ 8254675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 826e5c832d5SLinus Torvalds goto out; 82784a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 82884a2bd39SAl Viro goto out_dput; 829e5c832d5SLinus Torvalds /* 830e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 831e5c832d5SLinus Torvalds * still valid and get it if required. 832e5c832d5SLinus Torvalds */ 83384a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 83484a2bd39SAl Viro goto out_dput; 8358b61e74fSAl Viro rcu_read_unlock(); 836ae66db45SAl Viro return true; 83719660af7SAl Viro 8387973387aSAl Viro out2: 8397973387aSAl Viro nd->path.mnt = NULL; 8407973387aSAl Viro out1: 8417973387aSAl Viro nd->path.dentry = NULL; 842e5c832d5SLinus Torvalds out: 8438b61e74fSAl Viro rcu_read_unlock(); 844ae66db45SAl Viro return false; 84584a2bd39SAl Viro out_dput: 84684a2bd39SAl Viro rcu_read_unlock(); 84784a2bd39SAl Viro dput(dentry); 848ae66db45SAl Viro return false; 84931e6b01fSNick Piggin } 85031e6b01fSNick Piggin 8514ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 85234286d66SNick Piggin { 853a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8544ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 855a89f8337SAl Viro else 856a89f8337SAl Viro return 1; 85734286d66SNick Piggin } 85834286d66SNick Piggin 8599f1fafeeSAl Viro /** 8609f1fafeeSAl Viro * complete_walk - successful completion of path walk 8619f1fafeeSAl Viro * @nd: pointer nameidata 86239159de2SJeff Layton * 8639f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8649f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8659f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8669f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8679f1fafeeSAl Viro * need to drop nd->path. 86839159de2SJeff Layton */ 8699f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 87039159de2SJeff Layton { 87116c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 87239159de2SJeff Layton int status; 87339159de2SJeff Layton 8749f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 875adb21d2bSAleksa Sarai /* 876adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 877adb21d2bSAleksa Sarai * externally-managed nd->root. 878adb21d2bSAleksa Sarai */ 879bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 880bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8819f1fafeeSAl Viro nd->root.mnt = NULL; 8826c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 883e36cffedSJens Axboe if (!try_to_unlazy(nd)) 88448a066e7SAl Viro return -ECHILD; 88548a066e7SAl Viro } 8869f1fafeeSAl Viro 887adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 888adb21d2bSAleksa Sarai /* 889adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 890adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 891adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 892adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 893adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 894adb21d2bSAleksa Sarai * 895adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 896adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 897adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 898adb21d2bSAleksa Sarai * requested root to userspace. 899adb21d2bSAleksa Sarai * 900adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 901adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 902adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 903adb21d2bSAleksa Sarai */ 904adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 905adb21d2bSAleksa Sarai return -EXDEV; 906adb21d2bSAleksa Sarai } 907adb21d2bSAleksa Sarai 908bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 90939159de2SJeff Layton return 0; 91039159de2SJeff Layton 911ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 91216c2cd71SAl Viro return 0; 91316c2cd71SAl Viro 914ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 91539159de2SJeff Layton if (status > 0) 91639159de2SJeff Layton return 0; 91739159de2SJeff Layton 91816c2cd71SAl Viro if (!status) 91939159de2SJeff Layton status = -ESTALE; 92016c2cd71SAl Viro 92139159de2SJeff Layton return status; 92239159de2SJeff Layton } 92339159de2SJeff Layton 924740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9252a737871SAl Viro { 92631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9279e6697e2SAl Viro 928adb21d2bSAleksa Sarai /* 929adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 930adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 931adb21d2bSAleksa Sarai * from the dirfd. 932adb21d2bSAleksa Sarai */ 933adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 934adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 935adb21d2bSAleksa Sarai 9369e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9378f47a016SAl Viro unsigned seq; 938c28cc364SNick Piggin 939c28cc364SNick Piggin do { 940c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 94131e6b01fSNick Piggin nd->root = fs->root; 9428f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 943c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9449e6697e2SAl Viro } else { 9459e6697e2SAl Viro get_fs_root(fs, &nd->root); 946bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9479e6697e2SAl Viro } 948740a1678SAleksa Sarai return 0; 94931e6b01fSNick Piggin } 95031e6b01fSNick Piggin 951248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 952248fb5b9SAl Viro { 953adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 954adb21d2bSAleksa Sarai return -EXDEV; 95572ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 95672ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 95772ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 95872ba2929SAleksa Sarai return -EXDEV; 95972ba2929SAleksa Sarai } 960740a1678SAleksa Sarai if (!nd->root.mnt) { 961740a1678SAleksa Sarai int error = set_root(nd); 962740a1678SAleksa Sarai if (error) 963740a1678SAleksa Sarai return error; 964740a1678SAleksa Sarai } 965248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 966248fb5b9SAl Viro struct dentry *d; 967248fb5b9SAl Viro nd->path = nd->root; 968248fb5b9SAl Viro d = nd->path.dentry; 969248fb5b9SAl Viro nd->inode = d->d_inode; 970248fb5b9SAl Viro nd->seq = nd->root_seq; 971248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 972248fb5b9SAl Viro return -ECHILD; 973248fb5b9SAl Viro } else { 974248fb5b9SAl Viro path_put(&nd->path); 975248fb5b9SAl Viro nd->path = nd->root; 976248fb5b9SAl Viro path_get(&nd->path); 977248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 978248fb5b9SAl Viro } 979bcba1e7dSAl Viro nd->state |= ND_JUMPED; 980248fb5b9SAl Viro return 0; 981248fb5b9SAl Viro } 982248fb5b9SAl Viro 983b5fb63c1SChristoph Hellwig /* 9846b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 985b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 986b5fb63c1SChristoph Hellwig */ 9871bc82070SAleksa Sarai int nd_jump_link(struct path *path) 988b5fb63c1SChristoph Hellwig { 9894b99d499SAleksa Sarai int error = -ELOOP; 9906e77137bSAl Viro struct nameidata *nd = current->nameidata; 991b5fb63c1SChristoph Hellwig 9924b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9934b99d499SAleksa Sarai goto err; 9944b99d499SAleksa Sarai 99572ba2929SAleksa Sarai error = -EXDEV; 99672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 99772ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 99872ba2929SAleksa Sarai goto err; 99972ba2929SAleksa Sarai } 1000adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 1001adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 1002adb21d2bSAleksa Sarai goto err; 100372ba2929SAleksa Sarai 10044b99d499SAleksa Sarai path_put(&nd->path); 1005b5fb63c1SChristoph Hellwig nd->path = *path; 1006b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1007bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10081bc82070SAleksa Sarai return 0; 10094b99d499SAleksa Sarai 10104b99d499SAleksa Sarai err: 10114b99d499SAleksa Sarai path_put(path); 10124b99d499SAleksa Sarai return error; 1013b5fb63c1SChristoph Hellwig } 1014b5fb63c1SChristoph Hellwig 1015b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1016574197e0SAl Viro { 101721c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1018fceef393SAl Viro do_delayed_call(&last->done); 10196548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1020b9ff4429SAl Viro path_put(&last->link); 1021574197e0SAl Viro } 1022574197e0SAl Viro 1023561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 1024561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 102530aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 102630aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 1027800179c9SKees Cook 1028800179c9SKees Cook /** 1029800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 103055852635SRandy Dunlap * @nd: nameidata pathwalk data 1031800179c9SKees Cook * 1032800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1033800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1034800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1035800179c9SKees Cook * processes from failing races against path names that may change out 1036800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1037800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1038800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1039800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1040800179c9SKees Cook * 1041800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1042800179c9SKees Cook */ 1043ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1044800179c9SKees Cook { 1045ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1046ba73d987SChristian Brauner kuid_t i_uid; 1047ba73d987SChristian Brauner 1048800179c9SKees Cook if (!sysctl_protected_symlinks) 1049800179c9SKees Cook return 0; 1050800179c9SKees Cook 1051ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1052ba73d987SChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 1053800179c9SKees Cook /* Allowed if owner and follower match. */ 1054ba73d987SChristian Brauner if (uid_eq(current_cred()->fsuid, i_uid)) 1055800179c9SKees Cook return 0; 1056800179c9SKees Cook 1057800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 10580f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1059800179c9SKees Cook return 0; 1060800179c9SKees Cook 1061800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1062ba73d987SChristian Brauner if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) 1063800179c9SKees Cook return 0; 1064800179c9SKees Cook 106531956502SAl Viro if (nd->flags & LOOKUP_RCU) 106631956502SAl Viro return -ECHILD; 106731956502SAl Viro 1068ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1069245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1070800179c9SKees Cook return -EACCES; 1071800179c9SKees Cook } 1072800179c9SKees Cook 1073800179c9SKees Cook /** 1074800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1075ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1076800179c9SKees Cook * @inode: the source inode to hardlink from 1077800179c9SKees Cook * 1078800179c9SKees Cook * Return false if at least one of the following conditions: 1079800179c9SKees Cook * - inode is not a regular file 1080800179c9SKees Cook * - inode is setuid 1081800179c9SKees Cook * - inode is setgid and group-exec 1082800179c9SKees Cook * - access failure for read and write 1083800179c9SKees Cook * 1084800179c9SKees Cook * Otherwise returns true. 1085800179c9SKees Cook */ 1086ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1087ba73d987SChristian Brauner struct inode *inode) 1088800179c9SKees Cook { 1089800179c9SKees Cook umode_t mode = inode->i_mode; 1090800179c9SKees Cook 1091800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1092800179c9SKees Cook if (!S_ISREG(mode)) 1093800179c9SKees Cook return false; 1094800179c9SKees Cook 1095800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1096800179c9SKees Cook if (mode & S_ISUID) 1097800179c9SKees Cook return false; 1098800179c9SKees Cook 1099800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1100800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1101800179c9SKees Cook return false; 1102800179c9SKees Cook 1103800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1104ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1105800179c9SKees Cook return false; 1106800179c9SKees Cook 1107800179c9SKees Cook return true; 1108800179c9SKees Cook } 1109800179c9SKees Cook 1110800179c9SKees Cook /** 1111800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1112ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1113800179c9SKees Cook * @link: the source to hardlink from 1114800179c9SKees Cook * 1115800179c9SKees Cook * Block hardlink when all of: 1116800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1117800179c9SKees Cook * - fsuid does not match inode 1118800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1119f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1120800179c9SKees Cook * 1121ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1122ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1123ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1124ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1125ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1126ba73d987SChristian Brauner * 1127800179c9SKees Cook * Returns 0 if successful, -ve on error. 1128800179c9SKees Cook */ 1129ba73d987SChristian Brauner int may_linkat(struct user_namespace *mnt_userns, struct path *link) 1130800179c9SKees Cook { 1131593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1132593d1ce8SEric W. Biederman 1133593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1134ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 1135ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 1136593d1ce8SEric W. Biederman return -EOVERFLOW; 1137800179c9SKees Cook 1138800179c9SKees Cook if (!sysctl_protected_hardlinks) 1139800179c9SKees Cook return 0; 1140800179c9SKees Cook 1141800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1142800179c9SKees Cook * otherwise, it must be a safe source. 1143800179c9SKees Cook */ 1144ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1145ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1146800179c9SKees Cook return 0; 1147800179c9SKees Cook 1148245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1149800179c9SKees Cook return -EPERM; 1150800179c9SKees Cook } 1151800179c9SKees Cook 115230aba665SSalvatore Mesoraca /** 115330aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 115430aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 115530aba665SSalvatore Mesoraca * exist. 1156ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 11572111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 115830aba665SSalvatore Mesoraca * @inode: the inode of the file to open 115930aba665SSalvatore Mesoraca * 116030aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 116130aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 116230aba665SSalvatore Mesoraca * - the file already exists 116330aba665SSalvatore Mesoraca * - we are in a sticky directory 116430aba665SSalvatore Mesoraca * - we don't own the file 116530aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 116630aba665SSalvatore Mesoraca * - the directory is world writable 116730aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 116830aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 116930aba665SSalvatore Mesoraca * be enough. 117030aba665SSalvatore Mesoraca * 1171ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1172ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1173ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1174ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1175ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1176ba73d987SChristian Brauner * 117730aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 117830aba665SSalvatore Mesoraca */ 1179ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1180ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 118130aba665SSalvatore Mesoraca { 1182ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1183ba73d987SChristian Brauner kuid_t dir_uid = nd->dir_uid; 1184ba73d987SChristian Brauner 118530aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 118630aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1187d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1188ba73d987SChristian Brauner uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || 1189ba73d987SChristian Brauner uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) 119030aba665SSalvatore Mesoraca return 0; 119130aba665SSalvatore Mesoraca 1192d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1193d0cb5018SAl Viro (dir_mode & 0020 && 119430aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 119530aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1196245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1197245d7369SKees Cook "sticky_create_fifo" : 1198245d7369SKees Cook "sticky_create_regular"; 1199245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 120030aba665SSalvatore Mesoraca return -EACCES; 120130aba665SSalvatore Mesoraca } 120230aba665SSalvatore Mesoraca return 0; 120330aba665SSalvatore Mesoraca } 120430aba665SSalvatore Mesoraca 1205f015f126SDavid Howells /* 1206f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1207f015f126SDavid Howells * 1208f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1209f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1210f015f126SDavid Howells * Up is towards /. 1211f015f126SDavid Howells * 1212f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1213f015f126SDavid Howells * root. 1214f015f126SDavid Howells */ 1215bab77ebfSAl Viro int follow_up(struct path *path) 12161da177e4SLinus Torvalds { 12170714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12180714a533SAl Viro struct mount *parent; 12191da177e4SLinus Torvalds struct dentry *mountpoint; 122099b7db7bSNick Piggin 122148a066e7SAl Viro read_seqlock_excl(&mount_lock); 12220714a533SAl Viro parent = mnt->mnt_parent; 12233c0a6163SAl Viro if (parent == mnt) { 122448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12251da177e4SLinus Torvalds return 0; 12261da177e4SLinus Torvalds } 12270714a533SAl Viro mntget(&parent->mnt); 1228a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 122948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1230bab77ebfSAl Viro dput(path->dentry); 1231bab77ebfSAl Viro path->dentry = mountpoint; 1232bab77ebfSAl Viro mntput(path->mnt); 12330714a533SAl Viro path->mnt = &parent->mnt; 12341da177e4SLinus Torvalds return 1; 12351da177e4SLinus Torvalds } 12364d359507SAl Viro EXPORT_SYMBOL(follow_up); 12371da177e4SLinus Torvalds 12387ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12397ef482faSAl Viro struct path *path, unsigned *seqp) 12407ef482faSAl Viro { 12417ef482faSAl Viro while (mnt_has_parent(m)) { 12427ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12437ef482faSAl Viro 12447ef482faSAl Viro m = m->mnt_parent; 12457ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12467ef482faSAl Viro root->mnt == &m->mnt)) 12477ef482faSAl Viro break; 12487ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 12497ef482faSAl Viro path->mnt = &m->mnt; 12507ef482faSAl Viro path->dentry = mountpoint; 12517ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 12527ef482faSAl Viro return true; 12537ef482faSAl Viro } 12547ef482faSAl Viro } 12557ef482faSAl Viro return false; 12567ef482faSAl Viro } 12577ef482faSAl Viro 12582aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 12592aa38470SAl Viro struct path *path) 12602aa38470SAl Viro { 12612aa38470SAl Viro bool found; 12622aa38470SAl Viro 12632aa38470SAl Viro rcu_read_lock(); 12642aa38470SAl Viro while (1) { 12652aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 12662aa38470SAl Viro 12672aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 12682aa38470SAl Viro if (unlikely(!found)) { 12692aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 12702aa38470SAl Viro break; 12712aa38470SAl Viro } else { 12722aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 12732aa38470SAl Viro break; 12742aa38470SAl Viro rcu_read_unlock(); 12752aa38470SAl Viro path_put(path); 12762aa38470SAl Viro rcu_read_lock(); 12772aa38470SAl Viro } 12782aa38470SAl Viro } 12792aa38470SAl Viro rcu_read_unlock(); 12802aa38470SAl Viro return found; 12812aa38470SAl Viro } 12822aa38470SAl Viro 1283b5c84bf6SNick Piggin /* 12849875cf80SDavid Howells * Perform an automount 12859875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 12869875cf80SDavid Howells * were called with. 12871da177e4SLinus Torvalds */ 12881c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 128931e6b01fSNick Piggin { 129025e195aaSAl Viro struct dentry *dentry = path->dentry; 12919875cf80SDavid Howells 12920ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 12930ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 12940ec26fd0SMiklos Szeredi * the name. 12950ec26fd0SMiklos Szeredi * 12960ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 12975a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 12980ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 12990ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13000ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13010ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13025a30d8a2SDavid Howells */ 13031c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13045d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 130525e195aaSAl Viro dentry->d_inode) 13069875cf80SDavid Howells return -EISDIR; 13070ec26fd0SMiklos Szeredi 13081c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13099875cf80SDavid Howells return -ELOOP; 13109875cf80SDavid Howells 131125e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1312ea5b778aSDavid Howells } 13139875cf80SDavid Howells 13149875cf80SDavid Howells /* 13159deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13169deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13179deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13189deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13199875cf80SDavid Howells */ 13209deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13219deed3ebSAl Viro int *count, unsigned lookup_flags) 13229875cf80SDavid Howells { 13239deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13249875cf80SDavid Howells bool need_mntput = false; 13258aef1884SAl Viro int ret = 0; 13269875cf80SDavid Howells 13279deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1328cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1329cc53ce53SDavid Howells * being held. */ 1330d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1331fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1332508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1333cc53ce53SDavid Howells if (ret < 0) 13348aef1884SAl Viro break; 1335cc53ce53SDavid Howells } 1336cc53ce53SDavid Howells 13379deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13389875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13399deed3ebSAl Viro if (mounted) { // ... in our namespace 13409875cf80SDavid Howells dput(path->dentry); 13419875cf80SDavid Howells if (need_mntput) 1342463ffb2eSAl Viro mntput(path->mnt); 1343463ffb2eSAl Viro path->mnt = mounted; 1344463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13459deed3ebSAl Viro // here we know it's positive 13469deed3ebSAl Viro flags = path->dentry->d_flags; 13479875cf80SDavid Howells need_mntput = true; 13489875cf80SDavid Howells continue; 1349463ffb2eSAl Viro } 13501da177e4SLinus Torvalds } 13519875cf80SDavid Howells 13529deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 13539deed3ebSAl Viro break; 13549deed3ebSAl Viro 13559deed3ebSAl Viro // uncovered automount point 13569deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 13579deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 13589875cf80SDavid Howells if (ret < 0) 13598aef1884SAl Viro break; 13609875cf80SDavid Howells } 13619875cf80SDavid Howells 13629deed3ebSAl Viro if (ret == -EISDIR) 13639deed3ebSAl Viro ret = 0; 13649deed3ebSAl Viro // possible if you race with several mount --move 13659deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 13668aef1884SAl Viro mntput(path->mnt); 13679deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1368d41efb52SAl Viro ret = -ENOENT; 13699deed3ebSAl Viro *jumped = need_mntput; 13708402752eSAl Viro return ret; 13711da177e4SLinus Torvalds } 13721da177e4SLinus Torvalds 13739deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 13749deed3ebSAl Viro int *count, unsigned lookup_flags) 13759deed3ebSAl Viro { 13769deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 13779deed3ebSAl Viro 13789deed3ebSAl Viro /* fastpath */ 13799deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 13809deed3ebSAl Viro *jumped = false; 13819deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 13829deed3ebSAl Viro return -ENOENT; 13839deed3ebSAl Viro return 0; 13849deed3ebSAl Viro } 13859deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 13869deed3ebSAl Viro } 13879deed3ebSAl Viro 1388cc53ce53SDavid Howells int follow_down_one(struct path *path) 13891da177e4SLinus Torvalds { 13901da177e4SLinus Torvalds struct vfsmount *mounted; 13911da177e4SLinus Torvalds 13921c755af4SAl Viro mounted = lookup_mnt(path); 13931da177e4SLinus Torvalds if (mounted) { 13949393bd07SAl Viro dput(path->dentry); 13959393bd07SAl Viro mntput(path->mnt); 13969393bd07SAl Viro path->mnt = mounted; 13979393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 13981da177e4SLinus Torvalds return 1; 13991da177e4SLinus Torvalds } 14001da177e4SLinus Torvalds return 0; 14011da177e4SLinus Torvalds } 14024d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14031da177e4SLinus Torvalds 14049875cf80SDavid Howells /* 14059deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14069deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14079deed3ebSAl Viro * caller is permitted to proceed or not. 14089deed3ebSAl Viro */ 14099deed3ebSAl Viro int follow_down(struct path *path) 14109deed3ebSAl Viro { 14119deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14129deed3ebSAl Viro bool jumped; 14139deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 14149deed3ebSAl Viro 14159deed3ebSAl Viro if (path->mnt != mnt) 14169deed3ebSAl Viro mntput(mnt); 14179deed3ebSAl Viro return ret; 14189deed3ebSAl Viro } 14199deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14209deed3ebSAl Viro 14219deed3ebSAl Viro /* 1422287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1423287548e4SAl Viro * we meet a managed dentry that would need blocking. 14249875cf80SDavid Howells */ 14259875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1426254cf582SAl Viro struct inode **inode, unsigned *seqp) 14279875cf80SDavid Howells { 1428ea936aebSAl Viro struct dentry *dentry = path->dentry; 1429ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1430ea936aebSAl Viro 1431ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1432ea936aebSAl Viro return true; 1433ea936aebSAl Viro 1434ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1435ea936aebSAl Viro return false; 1436ea936aebSAl Viro 143762a7375eSIan Kent for (;;) { 143862a7375eSIan Kent /* 143962a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 144062a7375eSIan Kent * that wants to block transit. 144162a7375eSIan Kent */ 1442ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1443ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1444ea936aebSAl Viro if (res) 1445ea936aebSAl Viro return res == -EISDIR; 1446ea936aebSAl Viro flags = dentry->d_flags; 1447b8faf035SNeilBrown } 144862a7375eSIan Kent 1449ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1450ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1451ea936aebSAl Viro if (mounted) { 1452c7105365SAl Viro path->mnt = &mounted->mnt; 1453ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1454bcba1e7dSAl Viro nd->state |= ND_JUMPED; 1455ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1456ea936aebSAl Viro *inode = dentry->d_inode; 145759430262SLinus Torvalds /* 1458ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1459ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1460ea936aebSAl Viro * between mount hash removal and ->mnt_root 1461ea936aebSAl Viro * becoming unpinned. 146259430262SLinus Torvalds */ 1463ea936aebSAl Viro flags = dentry->d_flags; 1464ea936aebSAl Viro continue; 14659875cf80SDavid Howells } 1466ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1467ea936aebSAl Viro return false; 1468ea936aebSAl Viro } 1469ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1470ea936aebSAl Viro } 1471287548e4SAl Viro } 1472287548e4SAl Viro 1473db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1474db3c9adeSAl Viro struct path *path, struct inode **inode, 1475db3c9adeSAl Viro unsigned int *seqp) 1476bd7c4b50SAl Viro { 14779deed3ebSAl Viro bool jumped; 1478db3c9adeSAl Viro int ret; 1479bd7c4b50SAl Viro 1480db3c9adeSAl Viro path->mnt = nd->path.mnt; 1481db3c9adeSAl Viro path->dentry = dentry; 1482c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1483c153007bSAl Viro unsigned int seq = *seqp; 1484c153007bSAl Viro if (unlikely(!*inode)) 1485c153007bSAl Viro return -ENOENT; 1486c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 14879deed3ebSAl Viro return 0; 1488ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 1489c153007bSAl Viro return -ECHILD; 1490c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1491c153007bSAl Viro path->mnt = nd->path.mnt; 1492c153007bSAl Viro path->dentry = dentry; 1493c153007bSAl Viro } 14949deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 14959deed3ebSAl Viro if (jumped) { 14969deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 14979deed3ebSAl Viro ret = -EXDEV; 14989deed3ebSAl Viro else 1499bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15009deed3ebSAl Viro } 15019deed3ebSAl Viro if (unlikely(ret)) { 15029deed3ebSAl Viro dput(path->dentry); 15039deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15049deed3ebSAl Viro mntput(path->mnt); 15059deed3ebSAl Viro } else { 1506bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1507bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1508bd7c4b50SAl Viro } 1509bd7c4b50SAl Viro return ret; 1510bd7c4b50SAl Viro } 1511bd7c4b50SAl Viro 15129875cf80SDavid Howells /* 1513f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1514f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1515baa03890SNick Piggin */ 1516e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1517e3c13928SAl Viro struct dentry *dir, 15186c51e513SAl Viro unsigned int flags) 1519baa03890SNick Piggin { 1520a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1521bad61189SMiklos Szeredi if (dentry) { 1522a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1523bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 152474ff0ffcSAl Viro if (!error) 15255542aa2fSEric W. Biederman d_invalidate(dentry); 1526bad61189SMiklos Szeredi dput(dentry); 152774ff0ffcSAl Viro return ERR_PTR(error); 1528bad61189SMiklos Szeredi } 1529bad61189SMiklos Szeredi } 1530baa03890SNick Piggin return dentry; 1531baa03890SNick Piggin } 1532baa03890SNick Piggin 1533baa03890SNick Piggin /* 1534a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1535a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1536a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1537a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1538a03ece5fSAl Viro * at all. 153944396f4bSJosef Bacik */ 1540a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1541a03ece5fSAl Viro struct dentry *base, unsigned int flags) 154244396f4bSJosef Bacik { 1543a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 154444396f4bSJosef Bacik struct dentry *old; 1545a03ece5fSAl Viro struct inode *dir = base->d_inode; 1546a03ece5fSAl Viro 1547a03ece5fSAl Viro if (dentry) 1548a03ece5fSAl Viro return dentry; 154944396f4bSJosef Bacik 155044396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1551a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 155244396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1553a03ece5fSAl Viro 1554a03ece5fSAl Viro dentry = d_alloc(base, name); 1555a03ece5fSAl Viro if (unlikely(!dentry)) 1556a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 155744396f4bSJosef Bacik 155872bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 155944396f4bSJosef Bacik if (unlikely(old)) { 156044396f4bSJosef Bacik dput(dentry); 156144396f4bSJosef Bacik dentry = old; 156244396f4bSJosef Bacik } 156344396f4bSJosef Bacik return dentry; 156444396f4bSJosef Bacik } 156544396f4bSJosef Bacik 156620e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 156720e34357SAl Viro struct inode **inode, 1568254cf582SAl Viro unsigned *seqp) 15691da177e4SLinus Torvalds { 157031e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15715a18fff2SAl Viro int status = 1; 15729875cf80SDavid Howells 15733cac260aSAl Viro /* 1574b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15755d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15765d0f49c1SAl Viro * going to fall back to non-racy lookup. 1577b04f784eSNick Piggin */ 157831e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 157931e6b01fSNick Piggin unsigned seq; 1580da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15815d0f49c1SAl Viro if (unlikely(!dentry)) { 1582e36cffedSJens Axboe if (!try_to_unlazy(nd)) 158320e34357SAl Viro return ERR_PTR(-ECHILD); 158420e34357SAl Viro return NULL; 15855d0f49c1SAl Viro } 15865a18fff2SAl Viro 158712f8ad4bSLinus Torvalds /* 158812f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 158912f8ad4bSLinus Torvalds * the dentry name information from lookup. 159012f8ad4bSLinus Torvalds */ 159163afdfc7SDavid Howells *inode = d_backing_inode(dentry); 15925d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 159320e34357SAl Viro return ERR_PTR(-ECHILD); 159412f8ad4bSLinus Torvalds 159512f8ad4bSLinus Torvalds /* 159612f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 159712f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 159812f8ad4bSLinus Torvalds * 159912f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 160012f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 160112f8ad4bSLinus Torvalds */ 16025d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 160320e34357SAl Viro return ERR_PTR(-ECHILD); 16045a18fff2SAl Viro 1605254cf582SAl Viro *seqp = seq; 16064ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1607c153007bSAl Viro if (likely(status > 0)) 160820e34357SAl Viro return dentry; 1609ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 161020e34357SAl Viro return ERR_PTR(-ECHILD); 161126ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1612209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1613209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16145a18fff2SAl Viro } else { 1615e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 161681e6f520SAl Viro if (unlikely(!dentry)) 161720e34357SAl Viro return NULL; 16184ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16195d0f49c1SAl Viro } 16205a18fff2SAl Viro if (unlikely(status <= 0)) { 1621e9742b53SAl Viro if (!status) 16225d0f49c1SAl Viro d_invalidate(dentry); 16235a18fff2SAl Viro dput(dentry); 162420e34357SAl Viro return ERR_PTR(status); 16255a18fff2SAl Viro } 162620e34357SAl Viro return dentry; 1627697f514dSMiklos Szeredi } 1628697f514dSMiklos Szeredi 1629697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 163088d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1631e3c13928SAl Viro struct dentry *dir, 1632e3c13928SAl Viro unsigned int flags) 1633697f514dSMiklos Szeredi { 163488d8331aSAl Viro struct dentry *dentry, *old; 16351936386eSAl Viro struct inode *inode = dir->d_inode; 1636d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16371936386eSAl Viro 16381936386eSAl Viro /* Don't go there if it's already dead */ 163994bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 164088d8331aSAl Viro return ERR_PTR(-ENOENT); 164194bdd655SAl Viro again: 1642d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 164394bdd655SAl Viro if (IS_ERR(dentry)) 164488d8331aSAl Viro return dentry; 164594bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1646949a852eSAl Viro int error = d_revalidate(dentry, flags); 1647949a852eSAl Viro if (unlikely(error <= 0)) { 164894bdd655SAl Viro if (!error) { 1649949a852eSAl Viro d_invalidate(dentry); 1650949a852eSAl Viro dput(dentry); 165194bdd655SAl Viro goto again; 165294bdd655SAl Viro } 165394bdd655SAl Viro dput(dentry); 1654949a852eSAl Viro dentry = ERR_PTR(error); 1655949a852eSAl Viro } 165694bdd655SAl Viro } else { 16571936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 165885c7f810SAl Viro d_lookup_done(dentry); 16591936386eSAl Viro if (unlikely(old)) { 16601936386eSAl Viro dput(dentry); 16611936386eSAl Viro dentry = old; 1662949a852eSAl Viro } 1663949a852eSAl Viro } 1664e3c13928SAl Viro return dentry; 16651da177e4SLinus Torvalds } 16661da177e4SLinus Torvalds 166788d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 166888d8331aSAl Viro struct dentry *dir, 166988d8331aSAl Viro unsigned int flags) 167088d8331aSAl Viro { 167188d8331aSAl Viro struct inode *inode = dir->d_inode; 167288d8331aSAl Viro struct dentry *res; 167388d8331aSAl Viro inode_lock_shared(inode); 167488d8331aSAl Viro res = __lookup_slow(name, dir, flags); 167588d8331aSAl Viro inode_unlock_shared(inode); 167688d8331aSAl Viro return res; 167788d8331aSAl Viro } 167888d8331aSAl Viro 1679ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1680ba73d987SChristian Brauner struct nameidata *nd) 168152094c8aSAl Viro { 168252094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16837d6beb71SLinus Torvalds int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1684e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 168552094c8aSAl Viro return err; 168652094c8aSAl Viro } 1687ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 168852094c8aSAl Viro } 168952094c8aSAl Viro 169049055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1691d63ff28fSAl Viro { 169249055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 169349055906SAl Viro return -ELOOP; 16944542576bSAl Viro 16954542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 16964542576bSAl Viro return 0; 16974542576bSAl Viro if (likely(nd->stack != nd->internal)) 16984542576bSAl Viro return 0; 169960ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 170049055906SAl Viro return 0; 170160ef60c7SAl Viro 170260ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 170360ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 170460ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1705aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 170660ef60c7SAl Viro 1707e36cffedSJens Axboe if (!try_to_unlazy(nd) != 0 || !grabbed_link) 170860ef60c7SAl Viro return -ECHILD; 170960ef60c7SAl Viro 171060ef60c7SAl Viro if (nd_alloc_stack(nd)) 171160ef60c7SAl Viro return 0; 1712bc40aee0SAl Viro } 171360ef60c7SAl Viro return -ENOMEM; 171449055906SAl Viro } 171549055906SAl Viro 171649055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 171749055906SAl Viro 171849055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 171949055906SAl Viro struct inode *inode, unsigned seq, int flags) 172049055906SAl Viro { 172149055906SAl Viro struct saved *last; 172249055906SAl Viro const char *res; 172349055906SAl Viro int error = reserve_stack(nd, link, seq); 172449055906SAl Viro 172549055906SAl Viro if (unlikely(error)) { 172649055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1727cd179f44SAl Viro path_put(link); 172806708adbSAl Viro return ERR_PTR(error); 1729626de996SAl Viro } 1730ab104923SAl Viro last = nd->stack + nd->depth++; 17311cf2665bSAl Viro last->link = *link; 1732fceef393SAl Viro clear_delayed_call(&last->done); 17330450b2d1SAl Viro last->seq = seq; 1734ad6cc4c3SAl Viro 1735b1a81972SAl Viro if (flags & WALK_TRAILING) { 1736ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1737ad6cc4c3SAl Viro if (unlikely(error)) 1738ad6cc4c3SAl Viro return ERR_PTR(error); 1739ad6cc4c3SAl Viro } 1740ad6cc4c3SAl Viro 1741dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1742dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1743ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1744ad6cc4c3SAl Viro 1745ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1746ad6cc4c3SAl Viro touch_atime(&last->link); 1747ad6cc4c3SAl Viro cond_resched(); 1748ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1749e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1750ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1751ad6cc4c3SAl Viro touch_atime(&last->link); 1752ad6cc4c3SAl Viro } 1753ad6cc4c3SAl Viro 1754ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1755ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1756ad6cc4c3SAl Viro if (unlikely(error)) 1757ad6cc4c3SAl Viro return ERR_PTR(error); 1758ad6cc4c3SAl Viro 1759ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1760ad6cc4c3SAl Viro if (!res) { 1761ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1762ad6cc4c3SAl Viro struct delayed_call *); 1763ad6cc4c3SAl Viro get = inode->i_op->get_link; 1764ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1765ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1766e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1767ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1768ad6cc4c3SAl Viro } else { 1769ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1770ad6cc4c3SAl Viro } 1771ad6cc4c3SAl Viro if (!res) 1772ad6cc4c3SAl Viro goto all_done; 1773ad6cc4c3SAl Viro if (IS_ERR(res)) 1774ad6cc4c3SAl Viro return res; 1775ad6cc4c3SAl Viro } 1776ad6cc4c3SAl Viro if (*res == '/') { 1777ad6cc4c3SAl Viro error = nd_jump_root(nd); 1778ad6cc4c3SAl Viro if (unlikely(error)) 1779ad6cc4c3SAl Viro return ERR_PTR(error); 1780ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1781ad6cc4c3SAl Viro ; 1782ad6cc4c3SAl Viro } 1783ad6cc4c3SAl Viro if (*res) 1784ad6cc4c3SAl Viro return res; 1785ad6cc4c3SAl Viro all_done: // pure jump 1786ad6cc4c3SAl Viro put_link(nd); 1787ad6cc4c3SAl Viro return NULL; 1788d63ff28fSAl Viro } 1789d63ff28fSAl Viro 17903ddcd056SLinus Torvalds /* 17913ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17923ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17933ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17943ddcd056SLinus Torvalds * for the common case. 17953ddcd056SLinus Torvalds */ 1796b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1797cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 17983ddcd056SLinus Torvalds { 1799cbae4d12SAl Viro struct path path; 1800cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1801cbae4d12SAl Viro 1802cbae4d12SAl Viro if (err < 0) 1803b0417d2cSAl Viro return ERR_PTR(err); 1804cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18058c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1806aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18078f64fb1cSAl Viro /* not a symlink or should not follow */ 1808c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1809c99687a0SAl Viro dput(nd->path.dentry); 1810c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1811c99687a0SAl Viro mntput(nd->path.mnt); 1812c99687a0SAl Viro } 1813c99687a0SAl Viro nd->path = path; 18148f64fb1cSAl Viro nd->inode = inode; 18158f64fb1cSAl Viro nd->seq = seq; 1816b0417d2cSAl Viro return NULL; 18178f64fb1cSAl Viro } 1818a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 181984f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1820cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1821b0417d2cSAl Viro return ERR_PTR(-ECHILD); 182284f0cd9eSAl Viro } else { 182384f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 182484f0cd9eSAl Viro mntget(path.mnt); 1825a7f77542SAl Viro } 1826b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18273ddcd056SLinus Torvalds } 18283ddcd056SLinus Torvalds 1829c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1830c2df1968SAl Viro struct inode **inodep, 1831c2df1968SAl Viro unsigned *seqp) 1832957dd41dSAl Viro { 183312487f30SAl Viro struct dentry *parent, *old; 1834957dd41dSAl Viro 183512487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 183612487f30SAl Viro goto in_root; 183712487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18387ef482faSAl Viro struct path path; 1839efe772d6SAl Viro unsigned seq; 18407ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18417ef482faSAl Viro &nd->root, &path, &seq)) 184212487f30SAl Viro goto in_root; 1843efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1844efe772d6SAl Viro return ERR_PTR(-ECHILD); 1845efe772d6SAl Viro nd->path = path; 1846efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1847efe772d6SAl Viro nd->seq = seq; 1848efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1849efe772d6SAl Viro return ERR_PTR(-ECHILD); 1850efe772d6SAl Viro /* we know that mountpoint was pinned */ 1851957dd41dSAl Viro } 185212487f30SAl Viro old = nd->path.dentry; 185312487f30SAl Viro parent = old->d_parent; 185412487f30SAl Viro *inodep = parent->d_inode; 185512487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 185612487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 185712487f30SAl Viro return ERR_PTR(-ECHILD); 185812487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 185912487f30SAl Viro return ERR_PTR(-ECHILD); 186012487f30SAl Viro return parent; 186112487f30SAl Viro in_root: 1862efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1863efe772d6SAl Viro return ERR_PTR(-ECHILD); 1864957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 18657521f22bSAl Viro return ERR_PTR(-ECHILD); 1866c2df1968SAl Viro return NULL; 1867957dd41dSAl Viro } 1868957dd41dSAl Viro 1869c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1870c2df1968SAl Viro struct inode **inodep, 1871c2df1968SAl Viro unsigned *seqp) 1872957dd41dSAl Viro { 187312487f30SAl Viro struct dentry *parent; 187412487f30SAl Viro 1875957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 187612487f30SAl Viro goto in_root; 187712487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18782aa38470SAl Viro struct path path; 18792aa38470SAl Viro 18802aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 18812aa38470SAl Viro &nd->root, &path)) 188212487f30SAl Viro goto in_root; 1883165200d6SAl Viro path_put(&nd->path); 1884165200d6SAl Viro nd->path = path; 18852aa38470SAl Viro nd->inode = path.dentry->d_inode; 1886165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1887165200d6SAl Viro return ERR_PTR(-EXDEV); 188812487f30SAl Viro } 1889957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 189012487f30SAl Viro parent = dget_parent(nd->path.dentry); 1891957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1892957dd41dSAl Viro dput(parent); 18937521f22bSAl Viro return ERR_PTR(-ENOENT); 1894957dd41dSAl Viro } 1895c2df1968SAl Viro *seqp = 0; 1896c2df1968SAl Viro *inodep = parent->d_inode; 1897c2df1968SAl Viro return parent; 189812487f30SAl Viro 189912487f30SAl Viro in_root: 1900957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19017521f22bSAl Viro return ERR_PTR(-EXDEV); 1902c2df1968SAl Viro dget(nd->path.dentry); 1903c2df1968SAl Viro return NULL; 1904957dd41dSAl Viro } 1905957dd41dSAl Viro 19067521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1907957dd41dSAl Viro { 1908957dd41dSAl Viro if (type == LAST_DOTDOT) { 19097521f22bSAl Viro const char *error = NULL; 1910c2df1968SAl Viro struct dentry *parent; 1911c2df1968SAl Viro struct inode *inode; 1912c2df1968SAl Viro unsigned seq; 1913957dd41dSAl Viro 1914957dd41dSAl Viro if (!nd->root.mnt) { 19157521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1916957dd41dSAl Viro if (error) 1917957dd41dSAl Viro return error; 1918957dd41dSAl Viro } 1919957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1920c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1921957dd41dSAl Viro else 1922c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1923c2df1968SAl Viro if (IS_ERR(parent)) 1924c2df1968SAl Viro return ERR_CAST(parent); 1925c2df1968SAl Viro if (unlikely(!parent)) 1926c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1927c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1928c2df1968SAl Viro else 1929c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1930c2df1968SAl Viro parent, inode, seq); 1931c2df1968SAl Viro if (unlikely(error)) 1932957dd41dSAl Viro return error; 1933957dd41dSAl Viro 1934957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1935957dd41dSAl Viro /* 1936957dd41dSAl Viro * If there was a racing rename or mount along our 1937957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1938957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1939957dd41dSAl Viro * some fallback). 1940957dd41dSAl Viro */ 1941957dd41dSAl Viro smp_rmb(); 1942957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 19437521f22bSAl Viro return ERR_PTR(-EAGAIN); 1944957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 19457521f22bSAl Viro return ERR_PTR(-EAGAIN); 1946957dd41dSAl Viro } 1947957dd41dSAl Viro } 19487521f22bSAl Viro return NULL; 1949957dd41dSAl Viro } 1950957dd41dSAl Viro 195192d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1952ce57dfc1SAl Viro { 1953db3c9adeSAl Viro struct dentry *dentry; 1954ce57dfc1SAl Viro struct inode *inode; 1955254cf582SAl Viro unsigned seq; 1956ce57dfc1SAl Viro /* 1957ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1958ce57dfc1SAl Viro * to be able to know about the current root directory and 1959ce57dfc1SAl Viro * parent relationships. 1960ce57dfc1SAl Viro */ 19614693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 19621c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 19634693a547SAl Viro put_link(nd); 19647521f22bSAl Viro return handle_dots(nd, nd->last_type); 19654693a547SAl Viro } 196620e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 196720e34357SAl Viro if (IS_ERR(dentry)) 196892d27016SAl Viro return ERR_CAST(dentry); 196920e34357SAl Viro if (unlikely(!dentry)) { 1970db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1971db3c9adeSAl Viro if (IS_ERR(dentry)) 197292d27016SAl Viro return ERR_CAST(dentry); 197320e34357SAl Viro } 197456676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 197556676ec3SAl Viro put_link(nd); 1976b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1977ce57dfc1SAl Viro } 1978ce57dfc1SAl Viro 19791da177e4SLinus Torvalds /* 1980bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1981bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1982bfcfaa77SLinus Torvalds * 1983bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1984bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1985bfcfaa77SLinus Torvalds * fast. 1986bfcfaa77SLinus Torvalds * 1987bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1988bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1989bfcfaa77SLinus Torvalds * crossing operation. 1990bfcfaa77SLinus Torvalds * 1991bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1992bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1993bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1994bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1995bfcfaa77SLinus Torvalds */ 1996bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1997bfcfaa77SLinus Torvalds 1998f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1999bfcfaa77SLinus Torvalds 2000468a9428SGeorge Spelvin #ifdef HASH_MIX 2001bfcfaa77SLinus Torvalds 2002468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2003468a9428SGeorge Spelvin 2004468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20052a18da7aSGeorge Spelvin /* 20062a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20072a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20082a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20092a18da7aSGeorge Spelvin * and no temporaries. 20102a18da7aSGeorge Spelvin * 20112a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20122a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20132a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20142a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20152a18da7aSGeorge Spelvin * 20162a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20172a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20182a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20192a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20202a18da7aSGeorge Spelvin * 20212a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20222a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20232a18da7aSGeorge Spelvin * toggling any given output bit. 20242a18da7aSGeorge Spelvin * 20252a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20262a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20272a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20282a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20292a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20302a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20312a18da7aSGeorge Spelvin * Perfect: 8192 258048 20322a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20332a18da7aSGeorge Spelvin */ 20342a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20352a18da7aSGeorge Spelvin ( x ^= (a), \ 20362a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20372a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20382a18da7aSGeorge Spelvin y *= 9 ) 2039bfcfaa77SLinus Torvalds 20400fed3ac8SGeorge Spelvin /* 20412a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20422a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20432a18da7aSGeorge Spelvin * work done before the hash value is used. 20440fed3ac8SGeorge Spelvin */ 20452a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20460fed3ac8SGeorge Spelvin { 20472a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20482a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20492a18da7aSGeorge Spelvin return y >> 32; 20500fed3ac8SGeorge Spelvin } 20510fed3ac8SGeorge Spelvin 2052bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2053bfcfaa77SLinus Torvalds 20542a18da7aSGeorge Spelvin /* 20552a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 20562a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20572a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 20582a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 20592a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 20602a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 20612a18da7aSGeorge Spelvin * Perfect: 2048 31744 20622a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 20632a18da7aSGeorge Spelvin */ 20642a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20652a18da7aSGeorge Spelvin ( x ^= (a), \ 20662a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 20672a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 20682a18da7aSGeorge Spelvin y *= 9 ) 2069bfcfaa77SLinus Torvalds 20702a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20710fed3ac8SGeorge Spelvin { 20722a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 20732a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 20740fed3ac8SGeorge Spelvin } 20750fed3ac8SGeorge Spelvin 2076bfcfaa77SLinus Torvalds #endif 2077bfcfaa77SLinus Torvalds 20782a18da7aSGeorge Spelvin /* 20792a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 20802a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 20812a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 20822a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 20832a18da7aSGeorge Spelvin * finds the delimiter after the name. 20842a18da7aSGeorge Spelvin */ 20858387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2086bfcfaa77SLinus Torvalds { 20878387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2088bfcfaa77SLinus Torvalds 2089bfcfaa77SLinus Torvalds for (;;) { 2090fcfd2fbfSGeorge Spelvin if (!len) 2091fcfd2fbfSGeorge Spelvin goto done; 2092e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2093bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2094bfcfaa77SLinus Torvalds break; 20952a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2096bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2097bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2098bfcfaa77SLinus Torvalds } 20992a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2100bfcfaa77SLinus Torvalds done: 21012a18da7aSGeorge Spelvin return fold_hash(x, y); 2102bfcfaa77SLinus Torvalds } 2103bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2104bfcfaa77SLinus Torvalds 2105fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21068387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2107fcfd2fbfSGeorge Spelvin { 21088387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21098387ff25SLinus Torvalds unsigned long adata, mask, len; 2110fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2111fcfd2fbfSGeorge Spelvin 21128387ff25SLinus Torvalds len = 0; 21138387ff25SLinus Torvalds goto inside; 21148387ff25SLinus Torvalds 2115fcfd2fbfSGeorge Spelvin do { 21162a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2117fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21188387ff25SLinus Torvalds inside: 2119fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2120fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2121fcfd2fbfSGeorge Spelvin 2122fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2123fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21242a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2125fcfd2fbfSGeorge Spelvin 21262a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2127fcfd2fbfSGeorge Spelvin } 2128fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2129fcfd2fbfSGeorge Spelvin 2130bfcfaa77SLinus Torvalds /* 2131bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2132d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2133bfcfaa77SLinus Torvalds */ 21348387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2135bfcfaa77SLinus Torvalds { 21368387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21378387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 213836126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2139bfcfaa77SLinus Torvalds 21408387ff25SLinus Torvalds len = 0; 21418387ff25SLinus Torvalds goto inside; 21428387ff25SLinus Torvalds 2143bfcfaa77SLinus Torvalds do { 21442a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2145bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21468387ff25SLinus Torvalds inside: 2147e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 214836126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 214936126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2150bfcfaa77SLinus Torvalds 215136126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 215236126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 215336126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21542a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 215536126f8fSLinus Torvalds 21562a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2157bfcfaa77SLinus Torvalds } 2158bfcfaa77SLinus Torvalds 21592a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2160bfcfaa77SLinus Torvalds 2161fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 21628387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 21630145acc2SLinus Torvalds { 21648387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 21650145acc2SLinus Torvalds while (len--) 2166fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 21670145acc2SLinus Torvalds return end_name_hash(hash); 21680145acc2SLinus Torvalds } 2169ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 21700145acc2SLinus Torvalds 2171fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21728387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2173fcfd2fbfSGeorge Spelvin { 21748387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2175fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2176fcfd2fbfSGeorge Spelvin 2177fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2178e0ab7af9SGeorge Spelvin while (c) { 2179fcfd2fbfSGeorge Spelvin len++; 2180fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2181fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2182e0ab7af9SGeorge Spelvin } 2183fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2184fcfd2fbfSGeorge Spelvin } 2185f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2186fcfd2fbfSGeorge Spelvin 21873ddcd056SLinus Torvalds /* 2188200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2189200e9ef7SLinus Torvalds * one character. 2190200e9ef7SLinus Torvalds */ 21918387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2192200e9ef7SLinus Torvalds { 21938387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2194200e9ef7SLinus Torvalds unsigned long len = 0, c; 2195200e9ef7SLinus Torvalds 2196200e9ef7SLinus Torvalds c = (unsigned char)*name; 2197200e9ef7SLinus Torvalds do { 2198200e9ef7SLinus Torvalds len++; 2199200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2200200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2201200e9ef7SLinus Torvalds } while (c && c != '/'); 2202d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2203200e9ef7SLinus Torvalds } 2204200e9ef7SLinus Torvalds 2205bfcfaa77SLinus Torvalds #endif 2206bfcfaa77SLinus Torvalds 2207200e9ef7SLinus Torvalds /* 22081da177e4SLinus Torvalds * Name resolution. 2209ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2210ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22111da177e4SLinus Torvalds * 2212ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2213ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22141da177e4SLinus Torvalds */ 22156de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22161da177e4SLinus Torvalds { 2217d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22181da177e4SLinus Torvalds int err; 22191da177e4SLinus Torvalds 2220b4c03536SAl Viro nd->last_type = LAST_ROOT; 2221c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22229b5858e9SAl Viro if (IS_ERR(name)) 22239b5858e9SAl Viro return PTR_ERR(name); 22241da177e4SLinus Torvalds while (*name=='/') 22251da177e4SLinus Torvalds name++; 22261a97d899SAl Viro if (!*name) { 22271a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22289e18f10aSAl Viro return 0; 22291a97d899SAl Viro } 22301da177e4SLinus Torvalds 22311da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22321da177e4SLinus Torvalds for(;;) { 2233549c7297SChristian Brauner struct user_namespace *mnt_userns; 223492d27016SAl Viro const char *link; 2235d6bb3e90SLinus Torvalds u64 hash_len; 2236fe479a58SAl Viro int type; 22371da177e4SLinus Torvalds 2238549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2239549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22401da177e4SLinus Torvalds if (err) 22413595e234SAl Viro return err; 22421da177e4SLinus Torvalds 22438387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22441da177e4SLinus Torvalds 2245fe479a58SAl Viro type = LAST_NORM; 2246d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2247fe479a58SAl Viro case 2: 2248200e9ef7SLinus Torvalds if (name[1] == '.') { 2249fe479a58SAl Viro type = LAST_DOTDOT; 2250bcba1e7dSAl Viro nd->state |= ND_JUMPED; 225116c2cd71SAl Viro } 2252fe479a58SAl Viro break; 2253fe479a58SAl Viro case 1: 2254fe479a58SAl Viro type = LAST_DOT; 2255fe479a58SAl Viro } 22565a202bcdSAl Viro if (likely(type == LAST_NORM)) { 22575a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2258bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 22595a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2260a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2261da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 22625a202bcdSAl Viro if (err < 0) 22633595e234SAl Viro return err; 2264d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2265d6bb3e90SLinus Torvalds name = this.name; 22665a202bcdSAl Viro } 22675a202bcdSAl Viro } 2268fe479a58SAl Viro 2269d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2270d6bb3e90SLinus Torvalds nd->last.name = name; 22715f4a6a69SAl Viro nd->last_type = type; 22725f4a6a69SAl Viro 2273d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2274d6bb3e90SLinus Torvalds if (!*name) 2275bdf6cbf1SAl Viro goto OK; 2276200e9ef7SLinus Torvalds /* 2277200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2278200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2279200e9ef7SLinus Torvalds */ 2280200e9ef7SLinus Torvalds do { 2281d6bb3e90SLinus Torvalds name++; 2282d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 22838620c238SAl Viro if (unlikely(!*name)) { 22848620c238SAl Viro OK: 2285d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2286c108837eSAl Viro if (!depth) { 2287549c7297SChristian Brauner nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); 22880f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2289c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 22908620c238SAl Viro return 0; 2291c108837eSAl Viro } 22928620c238SAl Viro /* last component of nested symlink */ 2293d8d4611aSAl Viro name = nd->stack[--depth].name; 22948c4efe22SAl Viro link = walk_component(nd, 0); 22951c4ff1a8SAl Viro } else { 22961c4ff1a8SAl Viro /* not the last component */ 22978c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 22988620c238SAl Viro } 229992d27016SAl Viro if (unlikely(link)) { 230092d27016SAl Viro if (IS_ERR(link)) 230192d27016SAl Viro return PTR_ERR(link); 230292d27016SAl Viro /* a symlink to follow */ 2303d8d4611aSAl Viro nd->stack[depth++].name = name; 230492d27016SAl Viro name = link; 23059e18f10aSAl Viro continue; 230648c8b0c5SAl Viro } 230797242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 230897242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2309e36cffedSJens Axboe if (!try_to_unlazy(nd)) 231097242f99SAl Viro return -ECHILD; 231197242f99SAl Viro } 23123595e234SAl Viro return -ENOTDIR; 23135f4a6a69SAl Viro } 2314ce57dfc1SAl Viro } 231597242f99SAl Viro } 23161da177e4SLinus Torvalds 2317edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2318c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 231931e6b01fSNick Piggin { 2320740a1678SAleksa Sarai int error; 2321c8a53ee5SAl Viro const char *s = nd->name->name; 232231e6b01fSNick Piggin 23236c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23246c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23256c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23266c6ec2b0SJens Axboe 2327c0eb027eSLinus Torvalds if (!*s) 2328c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2329edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2330edc2b1daSAl Viro rcu_read_lock(); 2331c0eb027eSLinus Torvalds 2332bcba1e7dSAl Viro nd->flags = flags; 2333bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2334ab87f9a5SAleksa Sarai 2335ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2336ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2337ab87f9a5SAleksa Sarai smp_rmb(); 2338ab87f9a5SAleksa Sarai 2339bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2340b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2341b18825a7SDavid Howells struct inode *inode = root->d_inode; 234293893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2343368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23445b6ca027SAl Viro nd->path = nd->root; 23455b6ca027SAl Viro nd->inode = inode; 23465b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2347ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23488f47a016SAl Viro nd->root_seq = nd->seq; 23495b6ca027SAl Viro } else { 23505b6ca027SAl Viro path_get(&nd->path); 23515b6ca027SAl Viro } 2352368ee9baSAl Viro return s; 23535b6ca027SAl Viro } 23545b6ca027SAl Viro 235531e6b01fSNick Piggin nd->root.mnt = NULL; 235631e6b01fSNick Piggin 23578db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 23588db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2359740a1678SAleksa Sarai error = nd_jump_root(nd); 2360740a1678SAleksa Sarai if (unlikely(error)) 2361740a1678SAleksa Sarai return ERR_PTR(error); 2362ef55d917SAl Viro return s; 23638db52c7eSAleksa Sarai } 23648db52c7eSAleksa Sarai 23658db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 23668db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2367e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 236831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2369c28cc364SNick Piggin unsigned seq; 237031e6b01fSNick Piggin 2371c28cc364SNick Piggin do { 2372c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 237331e6b01fSNick Piggin nd->path = fs->pwd; 2374ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2375c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2376c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2377e41f7d4eSAl Viro } else { 2378e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2379ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2380e41f7d4eSAl Viro } 238131e6b01fSNick Piggin } else { 2382582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2383c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 238431e6b01fSNick Piggin struct dentry *dentry; 238531e6b01fSNick Piggin 23862903ff01SAl Viro if (!f.file) 2387368ee9baSAl Viro return ERR_PTR(-EBADF); 238831e6b01fSNick Piggin 23892903ff01SAl Viro dentry = f.file->f_path.dentry; 239031e6b01fSNick Piggin 2391edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 23922903ff01SAl Viro fdput(f); 2393368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2394f52e0c11SAl Viro } 23952903ff01SAl Viro 23962903ff01SAl Viro nd->path = f.file->f_path; 2397e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 239834a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 239934a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24005590ff0dSUlrich Drepper } else { 24012903ff01SAl Viro path_get(&nd->path); 240234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24031da177e4SLinus Torvalds } 240434a26b99SAl Viro fdput(f); 2405e41f7d4eSAl Viro } 24068db52c7eSAleksa Sarai 2407adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2408adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2409adb21d2bSAleksa Sarai nd->root = nd->path; 2410adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2411adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2412adb21d2bSAleksa Sarai } else { 2413adb21d2bSAleksa Sarai path_get(&nd->root); 2414bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2415adb21d2bSAleksa Sarai } 2416adb21d2bSAleksa Sarai } 2417adb21d2bSAleksa Sarai return s; 24189b4a9b14SAl Viro } 24199b4a9b14SAl Viro 24201ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 242195fa25d9SAl Viro { 2422bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2423bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2424bd92d7feSAl Viro 2425c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2426bd92d7feSAl Viro } 2427bd92d7feSAl Viro 24284f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24294f757f3cSAl Viro { 2430c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2431db3c9adeSAl Viro dget(nd->path.dentry); 2432b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2433b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 24344f757f3cSAl Viro } 24354f757f3cSAl Viro 24369b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2437c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24389b4a9b14SAl Viro { 2439c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2440bd92d7feSAl Viro int err; 244131e6b01fSNick Piggin 24429b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24434f757f3cSAl Viro err = handle_lookup_down(nd); 24445f336e72SAl Viro if (unlikely(err < 0)) 24455f336e72SAl Viro s = ERR_PTR(err); 24464f757f3cSAl Viro } 24474f757f3cSAl Viro 24481ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24491ccac622SAl Viro (s = lookup_last(nd)) != NULL) 24501ccac622SAl Viro ; 24514f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 24524f0ed93fSAl Viro err = handle_lookup_down(nd); 2453bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 24544f0ed93fSAl Viro } 24559f1fafeeSAl Viro if (!err) 24569f1fafeeSAl Viro err = complete_walk(nd); 2457bd92d7feSAl Viro 2458deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2459deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2460bd23a539SAl Viro err = -ENOTDIR; 2461625b6d10SAl Viro if (!err) { 2462625b6d10SAl Viro *path = nd->path; 2463625b6d10SAl Viro nd->path.mnt = NULL; 2464625b6d10SAl Viro nd->path.dentry = NULL; 2465625b6d10SAl Viro } 2466deb106c6SAl Viro terminate_walk(nd); 2467bd92d7feSAl Viro return err; 246831e6b01fSNick Piggin } 246931e6b01fSNick Piggin 2470020250f3SDmitry Kadashev static int __filename_lookup(int dfd, struct filename *name, unsigned flags, 24719ad1aaa6SAl Viro struct path *path, struct path *root) 2472873f1eedSJeff Layton { 2473894bc8c4SAl Viro int retval; 24749883d185SAl Viro struct nameidata nd; 2475abc9f5beSAl Viro if (IS_ERR(name)) 2476abc9f5beSAl Viro return PTR_ERR(name); 247706422964SAl Viro set_nameidata(&nd, dfd, name, root); 2478c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2479873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2480c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2481873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2482c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2483873f1eedSJeff Layton 2484873f1eedSJeff Layton if (likely(!retval)) 2485161aff1dSAl Viro audit_inode(name, path->dentry, 2486161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 24879883d185SAl Viro restore_nameidata(); 2488020250f3SDmitry Kadashev return retval; 2489020250f3SDmitry Kadashev } 2490020250f3SDmitry Kadashev 2491020250f3SDmitry Kadashev int filename_lookup(int dfd, struct filename *name, unsigned flags, 2492020250f3SDmitry Kadashev struct path *path, struct path *root) 2493020250f3SDmitry Kadashev { 2494020250f3SDmitry Kadashev int retval = __filename_lookup(dfd, name, flags, path, root); 2495020250f3SDmitry Kadashev 2496e4bd1c1aSAl Viro putname(name); 2497873f1eedSJeff Layton return retval; 2498873f1eedSJeff Layton } 2499873f1eedSJeff Layton 25008bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2501c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2502391172c4SAl Viro struct path *parent) 25038bcb77faSAl Viro { 2504c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25059b5858e9SAl Viro int err = link_path_walk(s, nd); 25068bcb77faSAl Viro if (!err) 25078bcb77faSAl Viro err = complete_walk(nd); 2508391172c4SAl Viro if (!err) { 2509391172c4SAl Viro *parent = nd->path; 2510391172c4SAl Viro nd->path.mnt = NULL; 2511391172c4SAl Viro nd->path.dentry = NULL; 2512391172c4SAl Viro } 2513deb106c6SAl Viro terminate_walk(nd); 25148bcb77faSAl Viro return err; 25158bcb77faSAl Viro } 25168bcb77faSAl Viro 25170ee50b47SDmitry Kadashev static int __filename_parentat(int dfd, struct filename *name, 2518391172c4SAl Viro unsigned int flags, struct path *parent, 2519391172c4SAl Viro struct qstr *last, int *type) 25208bcb77faSAl Viro { 25218bcb77faSAl Viro int retval; 25229883d185SAl Viro struct nameidata nd; 25238bcb77faSAl Viro 25245c31b6ceSAl Viro if (IS_ERR(name)) 25250ee50b47SDmitry Kadashev return PTR_ERR(name); 252606422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2527c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25288bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2529c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25308bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2531c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2532391172c4SAl Viro if (likely(!retval)) { 2533391172c4SAl Viro *last = nd.last; 2534391172c4SAl Viro *type = nd.last_type; 2535c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2536391172c4SAl Viro } 25379883d185SAl Viro restore_nameidata(); 25380ee50b47SDmitry Kadashev return retval; 25390ee50b47SDmitry Kadashev } 25400ee50b47SDmitry Kadashev 25410ee50b47SDmitry Kadashev static int filename_parentat(int dfd, struct filename *name, 25420ee50b47SDmitry Kadashev unsigned int flags, struct path *parent, 25430ee50b47SDmitry Kadashev struct qstr *last, int *type) 25440ee50b47SDmitry Kadashev { 25450ee50b47SDmitry Kadashev int retval = __filename_parentat(dfd, name, flags, parent, last, type); 25460ee50b47SDmitry Kadashev 25470ee50b47SDmitry Kadashev putname(name); 25480ee50b47SDmitry Kadashev return retval; 25498bcb77faSAl Viro } 25508bcb77faSAl Viro 255179714f72SAl Viro /* does lookup, returns the object with parent locked */ 255279714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 25535590ff0dSUlrich Drepper { 25545c31b6ceSAl Viro struct dentry *d; 2555391172c4SAl Viro struct qstr last; 25560ee50b47SDmitry Kadashev int type, error; 255751689104SPaul Moore 25580ee50b47SDmitry Kadashev error = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 25595c31b6ceSAl Viro &last, &type); 25600ee50b47SDmitry Kadashev if (error) 25610ee50b47SDmitry Kadashev return ERR_PTR(error); 25625c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2563391172c4SAl Viro path_put(path); 25645c31b6ceSAl Viro return ERR_PTR(-EINVAL); 256579714f72SAl Viro } 25665955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2567391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 256879714f72SAl Viro if (IS_ERR(d)) { 25695955102cSAl Viro inode_unlock(path->dentry->d_inode); 2570391172c4SAl Viro path_put(path); 257179714f72SAl Viro } 257279714f72SAl Viro return d; 25735590ff0dSUlrich Drepper } 25745590ff0dSUlrich Drepper 2575d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2576d1811465SAl Viro { 2577abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2578abc9f5beSAl Viro flags, path, NULL); 2579d1811465SAl Viro } 25804d359507SAl Viro EXPORT_SYMBOL(kern_path); 2581d1811465SAl Viro 258216f18200SJosef 'Jeff' Sipek /** 258316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 258416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 258516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 258616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 258716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2588e0a01249SAl Viro * @path: pointer to struct path to fill 258916f18200SJosef 'Jeff' Sipek */ 259016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 259116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2592e0a01249SAl Viro struct path *path) 259316f18200SJosef 'Jeff' Sipek { 25949ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 25959ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2596abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2597abc9f5beSAl Viro flags , path, &root); 259816f18200SJosef 'Jeff' Sipek } 25994d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 260016f18200SJosef 'Jeff' Sipek 26013c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 26023c95f0dcSAl Viro int len, struct qstr *this) 26033c95f0dcSAl Viro { 26043c95f0dcSAl Viro this->name = name; 26053c95f0dcSAl Viro this->len = len; 26063c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26073c95f0dcSAl Viro if (!len) 26083c95f0dcSAl Viro return -EACCES; 26093c95f0dcSAl Viro 26103c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26113c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26123c95f0dcSAl Viro return -EACCES; 26133c95f0dcSAl Viro } 26143c95f0dcSAl Viro 26153c95f0dcSAl Viro while (len--) { 26163c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26173c95f0dcSAl Viro if (c == '/' || c == '\0') 26183c95f0dcSAl Viro return -EACCES; 26193c95f0dcSAl Viro } 26203c95f0dcSAl Viro /* 26213c95f0dcSAl Viro * See if the low-level filesystem might want 26223c95f0dcSAl Viro * to use its own hash.. 26233c95f0dcSAl Viro */ 26243c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26253c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26263c95f0dcSAl Viro if (err < 0) 26273c95f0dcSAl Viro return err; 26283c95f0dcSAl Viro } 26293c95f0dcSAl Viro 263047291baaSChristian Brauner return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC); 26313c95f0dcSAl Viro } 26323c95f0dcSAl Viro 2633eead1911SChristoph Hellwig /** 26340da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26350da0b7fdSDavid Howells * @name: pathname component to lookup 26360da0b7fdSDavid Howells * @base: base directory to lookup from 26370da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26380da0b7fdSDavid Howells * 26390da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26400da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26410da0b7fdSDavid Howells * 26420da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26430da0b7fdSDavid Howells * not be called by generic code. 26440da0b7fdSDavid Howells * 26450da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26460da0b7fdSDavid Howells */ 26470da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26480da0b7fdSDavid Howells { 26490da0b7fdSDavid Howells struct qstr this; 26500da0b7fdSDavid Howells int err; 26510da0b7fdSDavid Howells 26520da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26530da0b7fdSDavid Howells 26540da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 26550da0b7fdSDavid Howells if (err) 26560da0b7fdSDavid Howells return ERR_PTR(err); 26570da0b7fdSDavid Howells 26580da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 26590da0b7fdSDavid Howells } 26600da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 26610da0b7fdSDavid Howells 26620da0b7fdSDavid Howells /** 2663a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2664eead1911SChristoph Hellwig * @name: pathname component to lookup 2665eead1911SChristoph Hellwig * @base: base directory to lookup from 2666eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2667eead1911SChristoph Hellwig * 2668a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 26699e7543e9SAl Viro * not be called by generic code. 2670bbddca8eSNeilBrown * 2671bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2672eead1911SChristoph Hellwig */ 2673057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2674057f6c01SJames Morris { 26758613a209SAl Viro struct dentry *dentry; 2676057f6c01SJames Morris struct qstr this; 2677cda309deSMiklos Szeredi int err; 2678057f6c01SJames Morris 26795955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26802f9092e1SDavid Woodhouse 26813c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2682cda309deSMiklos Szeredi if (err) 2683cda309deSMiklos Szeredi return ERR_PTR(err); 2684cda309deSMiklos Szeredi 26858613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 26868613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2687057f6c01SJames Morris } 26884d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2689057f6c01SJames Morris 2690bbddca8eSNeilBrown /** 2691bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2692bbddca8eSNeilBrown * @name: pathname component to lookup 2693bbddca8eSNeilBrown * @base: base directory to lookup from 2694bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2695bbddca8eSNeilBrown * 2696bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2697bbddca8eSNeilBrown * not be called by generic code. 2698bbddca8eSNeilBrown * 2699bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2700bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2701bbddca8eSNeilBrown */ 2702bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2703bbddca8eSNeilBrown struct dentry *base, int len) 2704bbddca8eSNeilBrown { 2705bbddca8eSNeilBrown struct qstr this; 2706bbddca8eSNeilBrown int err; 270720d00ee8SLinus Torvalds struct dentry *ret; 2708bbddca8eSNeilBrown 27093c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2710bbddca8eSNeilBrown if (err) 2711bbddca8eSNeilBrown return ERR_PTR(err); 2712bbddca8eSNeilBrown 271320d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 271420d00ee8SLinus Torvalds if (!ret) 271520d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 271620d00ee8SLinus Torvalds return ret; 2717bbddca8eSNeilBrown } 2718bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2719bbddca8eSNeilBrown 27206c2d4798SAl Viro /* 27216c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 27226c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 27236c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 27246c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 27256c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 27266c2d4798SAl Viro * this one avoids such problems. 27276c2d4798SAl Viro */ 27286c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 27296c2d4798SAl Viro struct dentry *base, int len) 27306c2d4798SAl Viro { 27316c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 27322fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 27336c2d4798SAl Viro dput(ret); 27346c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 27356c2d4798SAl Viro } 27366c2d4798SAl Viro return ret; 27376c2d4798SAl Viro } 27386c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 27396c2d4798SAl Viro 2740eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2741eedf265aSEric W. Biederman int path_pts(struct path *path) 2742eedf265aSEric W. Biederman { 2743eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2744eedf265aSEric W. Biederman * the input path. 2745eedf265aSEric W. Biederman */ 2746a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2747a6a7eb76SAl Viro struct dentry *child; 274819f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2749eedf265aSEric W. Biederman 2750a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2751a6a7eb76SAl Viro dput(parent); 275263b27720SAl Viro return -ENOENT; 2753a6a7eb76SAl Viro } 275463b27720SAl Viro dput(path->dentry); 275563b27720SAl Viro path->dentry = parent; 2756eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2757eedf265aSEric W. Biederman if (!child) 2758eedf265aSEric W. Biederman return -ENOENT; 2759eedf265aSEric W. Biederman 2760eedf265aSEric W. Biederman path->dentry = child; 2761eedf265aSEric W. Biederman dput(parent); 276219f6028aSAl Viro follow_down(path); 2763eedf265aSEric W. Biederman return 0; 2764eedf265aSEric W. Biederman } 2765eedf265aSEric W. Biederman #endif 2766eedf265aSEric W. Biederman 27671fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 27681fa1e7f6SAndy Whitcroft struct path *path, int *empty) 27691da177e4SLinus Torvalds { 2770abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2771abc9f5beSAl Viro flags, path, NULL); 27721da177e4SLinus Torvalds } 2773b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 27741fa1e7f6SAndy Whitcroft 2775ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2776ba73d987SChristian Brauner struct inode *inode) 27771da177e4SLinus Torvalds { 27788e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2779da9592edSDavid Howells 2780ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) 27811da177e4SLinus Torvalds return 0; 2782ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) 27831da177e4SLinus Torvalds return 0; 2784ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 27851da177e4SLinus Torvalds } 2786cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27871da177e4SLinus Torvalds 27881da177e4SLinus Torvalds /* 27891da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27901da177e4SLinus Torvalds * whether the type of victim is right. 27911da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27921da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27931da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27941da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27951da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27961da177e4SLinus Torvalds * a. be owner of dir, or 27971da177e4SLinus Torvalds * b. be owner of victim, or 27981da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27991da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 28001da177e4SLinus Torvalds * links pointing to it. 28010bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 28020bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 28030bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 28040bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 28050bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 28061da177e4SLinus Torvalds * nfs_async_unlink(). 28071da177e4SLinus Torvalds */ 2808ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2809ba73d987SChristian Brauner struct dentry *victim, bool isdir) 28101da177e4SLinus Torvalds { 281163afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 28121da177e4SLinus Torvalds int error; 28131da177e4SLinus Torvalds 2814b18825a7SDavid Howells if (d_is_negative(victim)) 28151da177e4SLinus Torvalds return -ENOENT; 2816b18825a7SDavid Howells BUG_ON(!inode); 28171da177e4SLinus Torvalds 28181da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2819593d1ce8SEric W. Biederman 2820593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2821ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 2822ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 2823593d1ce8SEric W. Biederman return -EOVERFLOW; 2824593d1ce8SEric W. Biederman 28254fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 28261da177e4SLinus Torvalds 2827ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 28281da177e4SLinus Torvalds if (error) 28291da177e4SLinus Torvalds return error; 28301da177e4SLinus Torvalds if (IS_APPEND(dir)) 28311da177e4SLinus Torvalds return -EPERM; 2832b18825a7SDavid Howells 2833ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2834ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2835ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 28361da177e4SLinus Torvalds return -EPERM; 28371da177e4SLinus Torvalds if (isdir) { 283844b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28391da177e4SLinus Torvalds return -ENOTDIR; 28401da177e4SLinus Torvalds if (IS_ROOT(victim)) 28411da177e4SLinus Torvalds return -EBUSY; 284244b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28431da177e4SLinus Torvalds return -EISDIR; 28441da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28451da177e4SLinus Torvalds return -ENOENT; 28461da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28471da177e4SLinus Torvalds return -EBUSY; 28481da177e4SLinus Torvalds return 0; 28491da177e4SLinus Torvalds } 28501da177e4SLinus Torvalds 28511da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28521da177e4SLinus Torvalds * dir. 28531da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28541da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28551da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2856036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2857036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2858036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28591da177e4SLinus Torvalds */ 2860ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2861ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 28621da177e4SLinus Torvalds { 286314e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28641da177e4SLinus Torvalds if (child->d_inode) 28651da177e4SLinus Torvalds return -EEXIST; 28661da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28671da177e4SLinus Torvalds return -ENOENT; 28688e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 2869036d5236SEric W. Biederman return -EOVERFLOW; 28708e538913SChristian Brauner 2871ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 28721da177e4SLinus Torvalds } 28731da177e4SLinus Torvalds 28741da177e4SLinus Torvalds /* 28751da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28761da177e4SLinus Torvalds */ 28771da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28781da177e4SLinus Torvalds { 28791da177e4SLinus Torvalds struct dentry *p; 28801da177e4SLinus Torvalds 28811da177e4SLinus Torvalds if (p1 == p2) { 28825955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28831da177e4SLinus Torvalds return NULL; 28841da177e4SLinus Torvalds } 28851da177e4SLinus Torvalds 2886fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28871da177e4SLinus Torvalds 2888e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2889e2761a11SOGAWA Hirofumi if (p) { 28905955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28915955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28921da177e4SLinus Torvalds return p; 28931da177e4SLinus Torvalds } 28941da177e4SLinus Torvalds 2895e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2896e2761a11SOGAWA Hirofumi if (p) { 28975955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28985955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28991da177e4SLinus Torvalds return p; 29001da177e4SLinus Torvalds } 29011da177e4SLinus Torvalds 29025955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29035955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 29041da177e4SLinus Torvalds return NULL; 29051da177e4SLinus Torvalds } 29064d359507SAl Viro EXPORT_SYMBOL(lock_rename); 29071da177e4SLinus Torvalds 29081da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 29091da177e4SLinus Torvalds { 29105955102cSAl Viro inode_unlock(p1->d_inode); 29111da177e4SLinus Torvalds if (p1 != p2) { 29125955102cSAl Viro inode_unlock(p2->d_inode); 2913fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 29141da177e4SLinus Torvalds } 29151da177e4SLinus Torvalds } 29164d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 29171da177e4SLinus Torvalds 29186521f891SChristian Brauner /** 29196521f891SChristian Brauner * vfs_create - create new file 29206521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 29216521f891SChristian Brauner * @dir: inode of @dentry 29226521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 29236521f891SChristian Brauner * @mode: mode of the new file 29246521f891SChristian Brauner * @want_excl: whether the file must not yet exist 29256521f891SChristian Brauner * 29266521f891SChristian Brauner * Create a new file. 29276521f891SChristian Brauner * 29286521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 29296521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 29306521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 29316521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 29326521f891SChristian Brauner * raw inode simply passs init_user_ns. 29336521f891SChristian Brauner */ 29346521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 29356521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 29361da177e4SLinus Torvalds { 29376521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 29381da177e4SLinus Torvalds if (error) 29391da177e4SLinus Torvalds return error; 29401da177e4SLinus Torvalds 2941acfa4380SAl Viro if (!dir->i_op->create) 29421da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 29431da177e4SLinus Torvalds mode &= S_IALLUGO; 29441da177e4SLinus Torvalds mode |= S_IFREG; 29451da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 29461da177e4SLinus Torvalds if (error) 29471da177e4SLinus Torvalds return error; 2948549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 2949a74574aaSStephen Smalley if (!error) 2950f38aa942SAmy Griffis fsnotify_create(dir, dentry); 29511da177e4SLinus Torvalds return error; 29521da177e4SLinus Torvalds } 29534d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29541da177e4SLinus Torvalds 29558e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29568e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29578e6c848eSAl Viro void *arg) 29588e6c848eSAl Viro { 29598e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 2960ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 29618e6c848eSAl Viro if (error) 29628e6c848eSAl Viro return error; 29638e6c848eSAl Viro 29648e6c848eSAl Viro mode &= S_IALLUGO; 29658e6c848eSAl Viro mode |= S_IFREG; 29668e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29678e6c848eSAl Viro if (error) 29688e6c848eSAl Viro return error; 29698e6c848eSAl Viro error = f(dentry, mode, arg); 29708e6c848eSAl Viro if (!error) 29718e6c848eSAl Viro fsnotify_create(dir, dentry); 29728e6c848eSAl Viro return error; 29738e6c848eSAl Viro } 29748e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29758e6c848eSAl Viro 2976a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2977a2982cc9SEric W. Biederman { 2978a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2979a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2980a2982cc9SEric W. Biederman } 2981a2982cc9SEric W. Biederman 2982ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 2983ba73d987SChristian Brauner int acc_mode, int flag) 29841da177e4SLinus Torvalds { 29853fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29861da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29871da177e4SLinus Torvalds int error; 29881da177e4SLinus Torvalds 29891da177e4SLinus Torvalds if (!inode) 29901da177e4SLinus Torvalds return -ENOENT; 29911da177e4SLinus Torvalds 2992c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2993c8fe8f30SChristoph Hellwig case S_IFLNK: 29941da177e4SLinus Torvalds return -ELOOP; 2995c8fe8f30SChristoph Hellwig case S_IFDIR: 2996fc4177beSKees Cook if (acc_mode & MAY_WRITE) 29971da177e4SLinus Torvalds return -EISDIR; 2998fc4177beSKees Cook if (acc_mode & MAY_EXEC) 2999fc4177beSKees Cook return -EACCES; 3000c8fe8f30SChristoph Hellwig break; 3001c8fe8f30SChristoph Hellwig case S_IFBLK: 3002c8fe8f30SChristoph Hellwig case S_IFCHR: 3003a2982cc9SEric W. Biederman if (!may_open_dev(path)) 30041da177e4SLinus Torvalds return -EACCES; 3005633fb6acSKees Cook fallthrough; 3006c8fe8f30SChristoph Hellwig case S_IFIFO: 3007c8fe8f30SChristoph Hellwig case S_IFSOCK: 3008633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3009633fb6acSKees Cook return -EACCES; 30101da177e4SLinus Torvalds flag &= ~O_TRUNC; 3011c8fe8f30SChristoph Hellwig break; 30120fd338b2SKees Cook case S_IFREG: 30130fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 30140fd338b2SKees Cook return -EACCES; 30150fd338b2SKees Cook break; 30164a3fd211SDave Hansen } 3017b41572e9SDave Hansen 3018ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 3019b41572e9SDave Hansen if (error) 3020b41572e9SDave Hansen return error; 30216146f0d5SMimi Zohar 30221da177e4SLinus Torvalds /* 30231da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 30241da177e4SLinus Torvalds */ 30251da177e4SLinus Torvalds if (IS_APPEND(inode)) { 30268737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 30277715b521SAl Viro return -EPERM; 30281da177e4SLinus Torvalds if (flag & O_TRUNC) 30297715b521SAl Viro return -EPERM; 30301da177e4SLinus Torvalds } 30311da177e4SLinus Torvalds 30321da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3033ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 30347715b521SAl Viro return -EPERM; 30351da177e4SLinus Torvalds 3036f3c7691eSJ. Bruce Fields return 0; 30377715b521SAl Viro } 30387715b521SAl Viro 3039549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 30407715b521SAl Viro { 3041f0bb5aafSAl Viro const struct path *path = &filp->f_path; 30427715b521SAl Viro struct inode *inode = path->dentry->d_inode; 30437715b521SAl Viro int error = get_write_access(inode); 30441da177e4SLinus Torvalds if (error) 30457715b521SAl Viro return error; 30461da177e4SLinus Torvalds /* 30471da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 30481da177e4SLinus Torvalds */ 3049d7a06983SJeff Layton error = locks_verify_locked(filp); 3050be6d3e56SKentaro Takeda if (!error) 3051ea0d3ab2STetsuo Handa error = security_path_truncate(path); 30521da177e4SLinus Torvalds if (!error) { 3053549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3054d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3055e1181ee6SJeff Layton filp); 30561da177e4SLinus Torvalds } 30571da177e4SLinus Torvalds put_write_access(inode); 3058acd0c935SMimi Zohar return error; 30591da177e4SLinus Torvalds } 30601da177e4SLinus Torvalds 3061d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3062d57999e1SDave Hansen { 30638a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30648a5e929dSAl Viro flag--; 3065d57999e1SDave Hansen return flag; 3066d57999e1SDave Hansen } 3067d57999e1SDave Hansen 3068ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3069ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3070ba73d987SChristian Brauner umode_t mode) 3071d18e9008SMiklos Szeredi { 3072d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3073d18e9008SMiklos Szeredi if (error) 3074d18e9008SMiklos Szeredi return error; 3075d18e9008SMiklos Szeredi 30768e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) 30771328c727SSeth Forshee return -EOVERFLOW; 30781328c727SSeth Forshee 3079ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 308047291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3081d18e9008SMiklos Szeredi if (error) 3082d18e9008SMiklos Szeredi return error; 3083d18e9008SMiklos Szeredi 3084d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3085d18e9008SMiklos Szeredi } 3086d18e9008SMiklos Szeredi 30871acf0af9SDavid Howells /* 30881acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30891acf0af9SDavid Howells * dentry. 30901acf0af9SDavid Howells * 30911acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30921acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30931acf0af9SDavid Howells * 309400a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 309500a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 309600a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 30971acf0af9SDavid Howells * 30981acf0af9SDavid Howells * Returns an error code otherwise. 30991acf0af9SDavid Howells */ 3100239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3101239eb983SAl Viro struct file *file, 31023ec2eef1SAl Viro int open_flag, umode_t mode) 3103d18e9008SMiklos Szeredi { 3104d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3105d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3106d18e9008SMiklos Szeredi int error; 3107d18e9008SMiklos Szeredi 3108d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3109d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3110d18e9008SMiklos Szeredi 311130d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 311230d90494SAl Viro file->f_path.mnt = nd->path.mnt; 31130fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 311444907d79SAl Viro open_to_namei_flags(open_flag), mode); 31156fbd0714SAl Viro d_lookup_done(dentry); 3116384f26e2SAl Viro if (!error) { 311764e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 31186fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 31196fb968cdSAl Viro dput(dentry); 31206fb968cdSAl Viro dentry = dget(file->f_path.dentry); 31216fb968cdSAl Viro } 312264e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 31232675a4ebSAl Viro error = -EIO; 3124384f26e2SAl Viro } else { 312530d90494SAl Viro if (file->f_path.dentry) { 3126d18e9008SMiklos Szeredi dput(dentry); 3127d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 312810c64ceaSAl Viro } 3129239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3130a01e718fSAl Viro error = -ENOENT; 3131d18e9008SMiklos Szeredi } 3132d18e9008SMiklos Szeredi } 3133239eb983SAl Viro if (error) { 3134d18e9008SMiklos Szeredi dput(dentry); 3135239eb983SAl Viro dentry = ERR_PTR(error); 3136239eb983SAl Viro } 3137239eb983SAl Viro return dentry; 3138d18e9008SMiklos Szeredi } 3139d18e9008SMiklos Szeredi 314031e6b01fSNick Piggin /* 31411acf0af9SDavid Howells * Look up and maybe create and open the last component. 3142d58ffd35SMiklos Szeredi * 314300a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3144d58ffd35SMiklos Szeredi * 314500a07c15SAl Viro * Returns 0 on success, that is, if 314600a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 314700a07c15SAl Viro * the file was not completely opened at this time, though lookups and 314800a07c15SAl Viro * creations were performed. 314900a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 315000a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 315100a07c15SAl Viro * hadn't been specified. 31521acf0af9SDavid Howells * 315300a07c15SAl Viro * An error code is returned on failure. 3154d58ffd35SMiklos Szeredi */ 3155da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3156d58ffd35SMiklos Szeredi const struct open_flags *op, 31573ec2eef1SAl Viro bool got_write) 3158d58ffd35SMiklos Szeredi { 3159549c7297SChristian Brauner struct user_namespace *mnt_userns; 3160d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 316154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31621643b43fSAl Viro int open_flag = op->open_flag; 3163d58ffd35SMiklos Szeredi struct dentry *dentry; 31641643b43fSAl Viro int error, create_error = 0; 31651643b43fSAl Viro umode_t mode = op->mode; 31666fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3167d58ffd35SMiklos Szeredi 3168ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3169da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3170d58ffd35SMiklos Szeredi 317173a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 31726fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31736fbd0714SAl Viro for (;;) { 31746fbd0714SAl Viro if (!dentry) { 31756fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3176d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3177da5ebf5aSAl Viro return dentry; 31786fbd0714SAl Viro } 31796fbd0714SAl Viro if (d_in_lookup(dentry)) 31806fbd0714SAl Viro break; 3181d58ffd35SMiklos Szeredi 31826fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31836fbd0714SAl Viro if (likely(error > 0)) 31846fbd0714SAl Viro break; 31856fbd0714SAl Viro if (error) 31866fbd0714SAl Viro goto out_dput; 31876fbd0714SAl Viro d_invalidate(dentry); 31886fbd0714SAl Viro dput(dentry); 31896fbd0714SAl Viro dentry = NULL; 31906fbd0714SAl Viro } 31916fbd0714SAl Viro if (dentry->d_inode) { 3192d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3193da5ebf5aSAl Viro return dentry; 31946c51e513SAl Viro } 3195d18e9008SMiklos Szeredi 31961643b43fSAl Viro /* 31971643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31981643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31991643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 32001643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 32011643b43fSAl Viro * 32021643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 32031643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 32041643b43fSAl Viro */ 320599a4a90cSAl Viro if (unlikely(!got_write)) 320699a4a90cSAl Viro open_flag &= ~O_TRUNC; 3207549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 32081643b43fSAl Viro if (open_flag & O_CREAT) { 320999a4a90cSAl Viro if (open_flag & O_EXCL) 321099a4a90cSAl Viro open_flag &= ~O_TRUNC; 32111643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 32121643b43fSAl Viro mode &= ~current_umask(); 321399a4a90cSAl Viro if (likely(got_write)) 3214549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3215ba73d987SChristian Brauner dentry, mode); 321699a4a90cSAl Viro else 321799a4a90cSAl Viro create_error = -EROFS; 321899a4a90cSAl Viro } 321999a4a90cSAl Viro if (create_error) 32201643b43fSAl Viro open_flag &= ~O_CREAT; 32211643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3222d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3223da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3224da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3225da5ebf5aSAl Viro return dentry; 3226239eb983SAl Viro } 322754ef4872SMiklos Szeredi 32286fbd0714SAl Viro if (d_in_lookup(dentry)) { 322912fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 323012fa5e24SAl Viro nd->flags); 32316fbd0714SAl Viro d_lookup_done(dentry); 323212fa5e24SAl Viro if (unlikely(res)) { 323312fa5e24SAl Viro if (IS_ERR(res)) { 323412fa5e24SAl Viro error = PTR_ERR(res); 323512fa5e24SAl Viro goto out_dput; 323612fa5e24SAl Viro } 323712fa5e24SAl Viro dput(dentry); 323812fa5e24SAl Viro dentry = res; 323912fa5e24SAl Viro } 324054ef4872SMiklos Szeredi } 324154ef4872SMiklos Szeredi 3242d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32431643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 324473a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3245ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3246ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3247ce8644fcSAl Viro error = -EACCES; 3248d58ffd35SMiklos Szeredi goto out_dput; 324964894cf8SAl Viro } 3250549c7297SChristian Brauner 3251549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3252549c7297SChristian Brauner mode, open_flag & O_EXCL); 3253d58ffd35SMiklos Szeredi if (error) 3254d58ffd35SMiklos Szeredi goto out_dput; 3255d58ffd35SMiklos Szeredi } 32561643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32571643b43fSAl Viro error = create_error; 3258d58ffd35SMiklos Szeredi goto out_dput; 3259d58ffd35SMiklos Szeredi } 3260da5ebf5aSAl Viro return dentry; 3261d58ffd35SMiklos Szeredi 3262d58ffd35SMiklos Szeredi out_dput: 3263d58ffd35SMiklos Szeredi dput(dentry); 3264da5ebf5aSAl Viro return ERR_PTR(error); 3265d58ffd35SMiklos Szeredi } 3266d58ffd35SMiklos Szeredi 3267c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 32683ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3269fb1cc555SAl Viro { 3270a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3271ca344a89SAl Viro int open_flag = op->open_flag; 327264894cf8SAl Viro bool got_write = false; 3273254cf582SAl Viro unsigned seq; 3274a1eb3315SMiklos Szeredi struct inode *inode; 3275da5ebf5aSAl Viro struct dentry *dentry; 3276b0417d2cSAl Viro const char *res; 3277fb1cc555SAl Viro 3278c3e380b0SAl Viro nd->flags |= op->intent; 3279c3e380b0SAl Viro 3280bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 328156676ec3SAl Viro if (nd->depth) 328256676ec3SAl Viro put_link(nd); 3283ff326a32SAl Viro return handle_dots(nd, nd->last_type); 32841f36f774SAl Viro } 3285a2c36b45SAl Viro 3286ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3287fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3288fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3289fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 329020e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 329120e34357SAl Viro if (IS_ERR(dentry)) 32921ccac622SAl Viro return ERR_CAST(dentry); 329320e34357SAl Viro if (likely(dentry)) 329471574865SMiklos Szeredi goto finish_lookup; 329571574865SMiklos Szeredi 32966583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3297b6183df7SMiklos Szeredi } else { 3298fe2d35ffSAl Viro /* create side of things */ 329972287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3300e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3301e36cffedSJens Axboe return ERR_PTR(-ECHILD); 330272287417SAl Viro } 3303c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 33041f36f774SAl Viro /* trailing slashes? */ 3305deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 33061ccac622SAl Viro return ERR_PTR(-EISDIR); 3307b6183df7SMiklos Szeredi } 33081f36f774SAl Viro 33099cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3310e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 331164894cf8SAl Viro /* 331264894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 331364894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 331464894cf8SAl Viro * dropping this one anyway. 331564894cf8SAl Viro */ 331664894cf8SAl Viro } 33179cf843e3SAl Viro if (open_flag & O_CREAT) 33185955102cSAl Viro inode_lock(dir->d_inode); 33199cf843e3SAl Viro else 33209cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3321da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3322f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3323f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 33249cf843e3SAl Viro if (open_flag & O_CREAT) 33255955102cSAl Viro inode_unlock(dir->d_inode); 33269cf843e3SAl Viro else 33279cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3328fb1cc555SAl Viro 3329c981a482SAl Viro if (got_write) 333059e96e65SAl Viro mnt_drop_write(nd->path.mnt); 33316c0d46c4SAl Viro 333259e96e65SAl Viro if (IS_ERR(dentry)) 333359e96e65SAl Viro return ERR_CAST(dentry); 333459e96e65SAl Viro 3335973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3336e73cabffSAl Viro dput(nd->path.dentry); 3337e73cabffSAl Viro nd->path.dentry = dentry; 3338c981a482SAl Viro return NULL; 3339fb1cc555SAl Viro } 3340fb1cc555SAl Viro 334120e34357SAl Viro finish_lookup: 334256676ec3SAl Viro if (nd->depth) 334356676ec3SAl Viro put_link(nd); 33448c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3345ff326a32SAl Viro if (unlikely(res)) 33461ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3347b0417d2cSAl Viro return res; 33481ccac622SAl Viro } 334931d1726dSAl Viro 3350c981a482SAl Viro /* 3351c981a482SAl Viro * Handle the last step of open() 3352c981a482SAl Viro */ 3353c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3354c981a482SAl Viro struct file *file, const struct open_flags *op) 3355c981a482SAl Viro { 3356549c7297SChristian Brauner struct user_namespace *mnt_userns; 3357c981a482SAl Viro int open_flag = op->open_flag; 3358c981a482SAl Viro bool do_truncate; 3359c981a482SAl Viro int acc_mode; 3360c981a482SAl Viro int error; 3361c981a482SAl Viro 3362ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3363ff326a32SAl Viro error = complete_walk(nd); 3364ff326a32SAl Viro if (error) 3365ff326a32SAl Viro return error; 3366ff326a32SAl Viro } 3367973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 336876ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3369549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 337030aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3371b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3372b94e0b32SAl Viro return -EEXIST; 337330aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3374c5971b8cSAl Viro return -EISDIR; 3375549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 337630aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 337730aba665SSalvatore Mesoraca if (unlikely(error)) 3378c5971b8cSAl Viro return error; 337930aba665SSalvatore Mesoraca } 338044b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3381c5971b8cSAl Viro return -ENOTDIR; 33826c0d46c4SAl Viro 33838795e7d4SAl Viro do_truncate = false; 33848795e7d4SAl Viro acc_mode = op->acc_mode; 33855a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 33865a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 33875a2d3eddSAl Viro open_flag &= ~O_TRUNC; 33885a2d3eddSAl Viro acc_mode = 0; 33898795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 33900f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33910f9d1a10SAl Viro if (error) 3392c5971b8cSAl Viro return error; 33938795e7d4SAl Viro do_truncate = true; 33940f9d1a10SAl Viro } 3395549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 33968795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3397ae2bb293SAl Viro error = vfs_open(&nd->path, file); 33988795e7d4SAl Viro if (!error) 33996035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 34008795e7d4SAl Viro if (!error && do_truncate) 3401549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3402c80567c8SAl Viro if (unlikely(error > 0)) { 3403c80567c8SAl Viro WARN_ON(1); 3404c80567c8SAl Viro error = -EINVAL; 3405c80567c8SAl Viro } 34068795e7d4SAl Viro if (do_truncate) 34070f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3408c5971b8cSAl Viro return error; 3409fb1cc555SAl Viro } 3410fb1cc555SAl Viro 34116521f891SChristian Brauner /** 34126521f891SChristian Brauner * vfs_tmpfile - create tmpfile 34136521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 34146521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 34156521f891SChristian Brauner * @mode: mode of the new tmpfile 34162111c3c0SRandy Dunlap * @open_flag: flags 34176521f891SChristian Brauner * 34186521f891SChristian Brauner * Create a temporary file. 34196521f891SChristian Brauner * 34206521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 34216521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 34226521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 34236521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 34246521f891SChristian Brauner * raw inode simply passs init_user_ns. 34256521f891SChristian Brauner */ 34266521f891SChristian Brauner struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, 34276521f891SChristian Brauner struct dentry *dentry, umode_t mode, int open_flag) 3428af7bd4dcSAmir Goldstein { 3429af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3430af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3431af7bd4dcSAmir Goldstein struct inode *inode; 3432af7bd4dcSAmir Goldstein int error; 3433af7bd4dcSAmir Goldstein 3434af7bd4dcSAmir Goldstein /* we want directory to be writable */ 34356521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3436af7bd4dcSAmir Goldstein if (error) 3437af7bd4dcSAmir Goldstein goto out_err; 3438af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3439af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3440af7bd4dcSAmir Goldstein goto out_err; 3441af7bd4dcSAmir Goldstein error = -ENOMEM; 3442cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3443af7bd4dcSAmir Goldstein if (unlikely(!child)) 3444af7bd4dcSAmir Goldstein goto out_err; 3445549c7297SChristian Brauner error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); 3446af7bd4dcSAmir Goldstein if (error) 3447af7bd4dcSAmir Goldstein goto out_err; 3448af7bd4dcSAmir Goldstein error = -ENOENT; 3449af7bd4dcSAmir Goldstein inode = child->d_inode; 3450af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3451af7bd4dcSAmir Goldstein goto out_err; 3452af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3453af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3454af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3455af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3456af7bd4dcSAmir Goldstein } 3457a2d2329eSChristian Brauner ima_post_create_tmpfile(mnt_userns, inode); 3458af7bd4dcSAmir Goldstein return child; 3459af7bd4dcSAmir Goldstein 3460af7bd4dcSAmir Goldstein out_err: 3461af7bd4dcSAmir Goldstein dput(child); 3462af7bd4dcSAmir Goldstein return ERR_PTR(error); 3463af7bd4dcSAmir Goldstein } 3464af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3465af7bd4dcSAmir Goldstein 3466c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 346760545d0dSAl Viro const struct open_flags *op, 34683ec2eef1SAl Viro struct file *file) 346960545d0dSAl Viro { 34706521f891SChristian Brauner struct user_namespace *mnt_userns; 3471625b6d10SAl Viro struct dentry *child; 3472625b6d10SAl Viro struct path path; 3473c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 347460545d0dSAl Viro if (unlikely(error)) 347560545d0dSAl Viro return error; 3476625b6d10SAl Viro error = mnt_want_write(path.mnt); 347760545d0dSAl Viro if (unlikely(error)) 347860545d0dSAl Viro goto out; 34796521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 34806521f891SChristian Brauner child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); 3481af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3482684e73beSHirofumi Nakagawa if (IS_ERR(child)) 348360545d0dSAl Viro goto out2; 3484625b6d10SAl Viro dput(path.dentry); 3485625b6d10SAl Viro path.dentry = child; 3486c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 348769a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3488549c7297SChristian Brauner error = may_open(mnt_userns, &path, 0, op->open_flag); 34891e8f44f1SAl Viro if (!error) 34901e8f44f1SAl Viro error = vfs_open(&path, file); 349160545d0dSAl Viro out2: 3492625b6d10SAl Viro mnt_drop_write(path.mnt); 349360545d0dSAl Viro out: 3494625b6d10SAl Viro path_put(&path); 349560545d0dSAl Viro return error; 349660545d0dSAl Viro } 349760545d0dSAl Viro 34986ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34996ac08709SAl Viro { 35006ac08709SAl Viro struct path path; 35016ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 35026ac08709SAl Viro if (!error) { 35036ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3504ae2bb293SAl Viro error = vfs_open(&path, file); 35056ac08709SAl Viro path_put(&path); 35066ac08709SAl Viro } 35076ac08709SAl Viro return error; 35086ac08709SAl Viro } 35096ac08709SAl Viro 3510c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3511c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35121da177e4SLinus Torvalds { 351330d90494SAl Viro struct file *file; 351413aab428SAl Viro int error; 351531e6b01fSNick Piggin 3516ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35171afc99beSAl Viro if (IS_ERR(file)) 35181afc99beSAl Viro return file; 351931e6b01fSNick Piggin 3520bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 35213ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 35225f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 35236ac08709SAl Viro error = do_o_path(nd, flags, file); 35245f336e72SAl Viro } else { 35255f336e72SAl Viro const char *s = path_init(nd, flags); 35263bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3527c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 35281ccac622SAl Viro ; 3529c5971b8cSAl Viro if (!error) 3530c5971b8cSAl Viro error = do_open(nd, file, op); 3531deb106c6SAl Viro terminate_walk(nd); 35325f336e72SAl Viro } 35337c1c01ecSAl Viro if (likely(!error)) { 3534aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35357c1c01ecSAl Viro return file; 35367c1c01ecSAl Viro WARN_ON(1); 35377c1c01ecSAl Viro error = -EINVAL; 3538015c3bbcSMiklos Szeredi } 35397c1c01ecSAl Viro fput(file); 35402675a4ebSAl Viro if (error == -EOPENSTALE) { 35412675a4ebSAl Viro if (flags & LOOKUP_RCU) 35422675a4ebSAl Viro error = -ECHILD; 35432675a4ebSAl Viro else 35442675a4ebSAl Viro error = -ESTALE; 35452675a4ebSAl Viro } 35467c1c01ecSAl Viro return ERR_PTR(error); 3547de459215SKirill Korotaev } 35481da177e4SLinus Torvalds 3549669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3550f9652e10SAl Viro const struct open_flags *op) 355113aab428SAl Viro { 35529883d185SAl Viro struct nameidata nd; 3553f9652e10SAl Viro int flags = op->lookup_flags; 355413aab428SAl Viro struct file *filp; 355513aab428SAl Viro 355606422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3557c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 355813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3559c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 356013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3561c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35629883d185SAl Viro restore_nameidata(); 356313aab428SAl Viro return filp; 356413aab428SAl Viro } 356513aab428SAl Viro 3566ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3567f9652e10SAl Viro const char *name, const struct open_flags *op) 356873d049a4SAl Viro { 35699883d185SAl Viro struct nameidata nd; 357073d049a4SAl Viro struct file *file; 357151689104SPaul Moore struct filename *filename; 3572bcba1e7dSAl Viro int flags = op->lookup_flags; 357373d049a4SAl Viro 3574ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 357573d049a4SAl Viro return ERR_PTR(-ELOOP); 357673d049a4SAl Viro 357751689104SPaul Moore filename = getname_kernel(name); 3578a1c83681SViresh Kumar if (IS_ERR(filename)) 357951689104SPaul Moore return ERR_CAST(filename); 358051689104SPaul Moore 358106422964SAl Viro set_nameidata(&nd, -1, filename, root); 3582c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 358373d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3584c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 358573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3586c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35879883d185SAl Viro restore_nameidata(); 358851689104SPaul Moore putname(filename); 358973d049a4SAl Viro return file; 359073d049a4SAl Viro } 359173d049a4SAl Viro 3592584d3226SDmitry Kadashev static struct dentry *__filename_create(int dfd, struct filename *name, 35931ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35941da177e4SLinus Torvalds { 3595c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3596391172c4SAl Viro struct qstr last; 3597391172c4SAl Viro int type; 3598c30dabfeSJan Kara int err2; 35991ac12b4bSJeff Layton int error; 36001ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36011ac12b4bSJeff Layton 36021ac12b4bSJeff Layton /* 36031ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36041ac12b4bSJeff Layton * other flags passed in are ignored! 36051ac12b4bSJeff Layton */ 36061ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36071ac12b4bSJeff Layton 3608584d3226SDmitry Kadashev error = __filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36090ee50b47SDmitry Kadashev if (error) 36100ee50b47SDmitry Kadashev return ERR_PTR(error); 36111da177e4SLinus Torvalds 3612c663e5d8SChristoph Hellwig /* 3613c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3614c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3615c663e5d8SChristoph Hellwig */ 36165c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3617ed75e95dSAl Viro goto out; 3618c663e5d8SChristoph Hellwig 3619c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3620391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3621c663e5d8SChristoph Hellwig /* 3622c663e5d8SChristoph Hellwig * Do the final lookup. 3623c663e5d8SChristoph Hellwig */ 3624391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36255955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3626391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36271da177e4SLinus Torvalds if (IS_ERR(dentry)) 3628a8104a9fSAl Viro goto unlock; 3629c663e5d8SChristoph Hellwig 3630a8104a9fSAl Viro error = -EEXIST; 3631b18825a7SDavid Howells if (d_is_positive(dentry)) 3632a8104a9fSAl Viro goto fail; 3633b18825a7SDavid Howells 3634c663e5d8SChristoph Hellwig /* 3635c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3636c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3637c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3638c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3639c663e5d8SChristoph Hellwig */ 3640391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3641a8104a9fSAl Viro error = -ENOENT; 3642ed75e95dSAl Viro goto fail; 3643e9baf6e5SAl Viro } 3644c30dabfeSJan Kara if (unlikely(err2)) { 3645c30dabfeSJan Kara error = err2; 3646a8104a9fSAl Viro goto fail; 3647c30dabfeSJan Kara } 3648e9baf6e5SAl Viro return dentry; 36491da177e4SLinus Torvalds fail: 3650a8104a9fSAl Viro dput(dentry); 3651a8104a9fSAl Viro dentry = ERR_PTR(error); 3652a8104a9fSAl Viro unlock: 36535955102cSAl Viro inode_unlock(path->dentry->d_inode); 3654c30dabfeSJan Kara if (!err2) 3655391172c4SAl Viro mnt_drop_write(path->mnt); 3656ed75e95dSAl Viro out: 3657391172c4SAl Viro path_put(path); 3658ed75e95dSAl Viro return dentry; 3659dae6ad8fSAl Viro } 3660fa14a0b8SAl Viro 3661584d3226SDmitry Kadashev static inline struct dentry *filename_create(int dfd, struct filename *name, 3662584d3226SDmitry Kadashev struct path *path, unsigned int lookup_flags) 3663584d3226SDmitry Kadashev { 3664584d3226SDmitry Kadashev struct dentry *res = __filename_create(dfd, name, path, lookup_flags); 3665584d3226SDmitry Kadashev 3666584d3226SDmitry Kadashev putname(name); 3667584d3226SDmitry Kadashev return res; 3668584d3226SDmitry Kadashev } 3669584d3226SDmitry Kadashev 3670fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3671fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3672fa14a0b8SAl Viro { 3673181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3674181c37b6SAl Viro path, lookup_flags); 3675fa14a0b8SAl Viro } 3676dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3677dae6ad8fSAl Viro 3678921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3679921a1650SAl Viro { 3680921a1650SAl Viro dput(dentry); 36815955102cSAl Viro inode_unlock(path->dentry->d_inode); 3682a8104a9fSAl Viro mnt_drop_write(path->mnt); 3683921a1650SAl Viro path_put(path); 3684921a1650SAl Viro } 3685921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3686921a1650SAl Viro 3687520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36881ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3689dae6ad8fSAl Viro { 3690181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3691dae6ad8fSAl Viro } 3692dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3693dae6ad8fSAl Viro 36946521f891SChristian Brauner /** 36956521f891SChristian Brauner * vfs_mknod - create device node or file 36966521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 36976521f891SChristian Brauner * @dir: inode of @dentry 36986521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 36996521f891SChristian Brauner * @mode: mode of the new device node or file 37006521f891SChristian Brauner * @dev: device number of device to create 37016521f891SChristian Brauner * 37026521f891SChristian Brauner * Create a device node or file. 37036521f891SChristian Brauner * 37046521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 37056521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 37066521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 37076521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 37086521f891SChristian Brauner * raw inode simply passs init_user_ns. 37096521f891SChristian Brauner */ 37106521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 37116521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 37121da177e4SLinus Torvalds { 3713a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 37146521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 37151da177e4SLinus Torvalds 37161da177e4SLinus Torvalds if (error) 37171da177e4SLinus Torvalds return error; 37181da177e4SLinus Torvalds 3719a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3720a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 37211da177e4SLinus Torvalds return -EPERM; 37221da177e4SLinus Torvalds 3723acfa4380SAl Viro if (!dir->i_op->mknod) 37241da177e4SLinus Torvalds return -EPERM; 37251da177e4SLinus Torvalds 372608ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 372708ce5f16SSerge E. Hallyn if (error) 372808ce5f16SSerge E. Hallyn return error; 372908ce5f16SSerge E. Hallyn 37301da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37311da177e4SLinus Torvalds if (error) 37321da177e4SLinus Torvalds return error; 37331da177e4SLinus Torvalds 3734549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3735a74574aaSStephen Smalley if (!error) 3736f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37371da177e4SLinus Torvalds return error; 37381da177e4SLinus Torvalds } 37394d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37401da177e4SLinus Torvalds 3741f69aac00SAl Viro static int may_mknod(umode_t mode) 3742463c3197SDave Hansen { 3743463c3197SDave Hansen switch (mode & S_IFMT) { 3744463c3197SDave Hansen case S_IFREG: 3745463c3197SDave Hansen case S_IFCHR: 3746463c3197SDave Hansen case S_IFBLK: 3747463c3197SDave Hansen case S_IFIFO: 3748463c3197SDave Hansen case S_IFSOCK: 3749463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3750463c3197SDave Hansen return 0; 3751463c3197SDave Hansen case S_IFDIR: 3752463c3197SDave Hansen return -EPERM; 3753463c3197SDave Hansen default: 3754463c3197SDave Hansen return -EINVAL; 3755463c3197SDave Hansen } 3756463c3197SDave Hansen } 3757463c3197SDave Hansen 375845f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 375987c4e192SDominik Brodowski unsigned int dev) 37601da177e4SLinus Torvalds { 37616521f891SChristian Brauner struct user_namespace *mnt_userns; 37621da177e4SLinus Torvalds struct dentry *dentry; 3763dae6ad8fSAl Viro struct path path; 3764dae6ad8fSAl Viro int error; 3765972567f1SJeff Layton unsigned int lookup_flags = 0; 37661da177e4SLinus Torvalds 37678e4bfca1SAl Viro error = may_mknod(mode); 37688e4bfca1SAl Viro if (error) 37697797251bSDmitry Kadashev goto out1; 3770972567f1SJeff Layton retry: 37717797251bSDmitry Kadashev dentry = __filename_create(dfd, name, &path, lookup_flags); 37727797251bSDmitry Kadashev error = PTR_ERR(dentry); 3773dae6ad8fSAl Viro if (IS_ERR(dentry)) 37747797251bSDmitry Kadashev goto out1; 37752ad94ae6SAl Viro 3776dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3777ce3b0f8dSAl Viro mode &= ~current_umask(); 3778dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3779be6d3e56SKentaro Takeda if (error) 37807797251bSDmitry Kadashev goto out2; 37816521f891SChristian Brauner 37826521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 37831da177e4SLinus Torvalds switch (mode & S_IFMT) { 37841da177e4SLinus Torvalds case 0: case S_IFREG: 37856521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 37866521f891SChristian Brauner dentry, mode, true); 378705d1a717SMimi Zohar if (!error) 3788a2d2329eSChristian Brauner ima_post_path_mknod(mnt_userns, dentry); 37891da177e4SLinus Torvalds break; 37901da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 37916521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 37926521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 37931da177e4SLinus Torvalds break; 37941da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 37956521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 37966521f891SChristian Brauner dentry, mode, 0); 37971da177e4SLinus Torvalds break; 37981da177e4SLinus Torvalds } 37997797251bSDmitry Kadashev out2: 3800921a1650SAl Viro done_path_create(&path, dentry); 3801972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3802972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3803972567f1SJeff Layton goto retry; 3804972567f1SJeff Layton } 38057797251bSDmitry Kadashev out1: 38067797251bSDmitry Kadashev putname(name); 38071da177e4SLinus Torvalds return error; 38081da177e4SLinus Torvalds } 38091da177e4SLinus Torvalds 381087c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 381187c4e192SDominik Brodowski unsigned int, dev) 381287c4e192SDominik Brodowski { 38137797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 381487c4e192SDominik Brodowski } 381587c4e192SDominik Brodowski 38168208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 38175590ff0dSUlrich Drepper { 38187797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 38195590ff0dSUlrich Drepper } 38205590ff0dSUlrich Drepper 38216521f891SChristian Brauner /** 38226521f891SChristian Brauner * vfs_mkdir - create directory 38236521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 38246521f891SChristian Brauner * @dir: inode of @dentry 38256521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 38266521f891SChristian Brauner * @mode: mode of the new directory 38276521f891SChristian Brauner * 38286521f891SChristian Brauner * Create a directory. 38296521f891SChristian Brauner * 38306521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 38316521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 38326521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 38336521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 38346521f891SChristian Brauner * raw inode simply passs init_user_ns. 38356521f891SChristian Brauner */ 38366521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 38376521f891SChristian Brauner struct dentry *dentry, umode_t mode) 38381da177e4SLinus Torvalds { 38396521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 38408de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38411da177e4SLinus Torvalds 38421da177e4SLinus Torvalds if (error) 38431da177e4SLinus Torvalds return error; 38441da177e4SLinus Torvalds 3845acfa4380SAl Viro if (!dir->i_op->mkdir) 38461da177e4SLinus Torvalds return -EPERM; 38471da177e4SLinus Torvalds 38481da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38491da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38501da177e4SLinus Torvalds if (error) 38511da177e4SLinus Torvalds return error; 38521da177e4SLinus Torvalds 38538de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38548de52778SAl Viro return -EMLINK; 38558de52778SAl Viro 3856549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 3857a74574aaSStephen Smalley if (!error) 3858f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38591da177e4SLinus Torvalds return error; 38601da177e4SLinus Torvalds } 38614d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38621da177e4SLinus Torvalds 386345f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 38641da177e4SLinus Torvalds { 38656902d925SDave Hansen struct dentry *dentry; 3866dae6ad8fSAl Viro struct path path; 3867dae6ad8fSAl Viro int error; 3868b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38691da177e4SLinus Torvalds 3870b76d8b82SJeff Layton retry: 3871584d3226SDmitry Kadashev dentry = __filename_create(dfd, name, &path, lookup_flags); 3872584d3226SDmitry Kadashev error = PTR_ERR(dentry); 38736902d925SDave Hansen if (IS_ERR(dentry)) 3874584d3226SDmitry Kadashev goto out_putname; 38756902d925SDave Hansen 3876dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3877ce3b0f8dSAl Viro mode &= ~current_umask(); 3878dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 38796521f891SChristian Brauner if (!error) { 38806521f891SChristian Brauner struct user_namespace *mnt_userns; 38816521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 3882549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 3883549c7297SChristian Brauner mode); 38846521f891SChristian Brauner } 3885921a1650SAl Viro done_path_create(&path, dentry); 3886b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3887b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3888b76d8b82SJeff Layton goto retry; 3889b76d8b82SJeff Layton } 3890584d3226SDmitry Kadashev out_putname: 3891584d3226SDmitry Kadashev putname(name); 38921da177e4SLinus Torvalds return error; 38931da177e4SLinus Torvalds } 38941da177e4SLinus Torvalds 38950101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38960101db7aSDominik Brodowski { 3897584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 38980101db7aSDominik Brodowski } 38990101db7aSDominik Brodowski 3900a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 39015590ff0dSUlrich Drepper { 3902584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 39035590ff0dSUlrich Drepper } 39045590ff0dSUlrich Drepper 39056521f891SChristian Brauner /** 39066521f891SChristian Brauner * vfs_rmdir - remove directory 39076521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 39086521f891SChristian Brauner * @dir: inode of @dentry 39096521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39106521f891SChristian Brauner * 39116521f891SChristian Brauner * Remove a directory. 39126521f891SChristian Brauner * 39136521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39146521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39156521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39166521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39176521f891SChristian Brauner * raw inode simply passs init_user_ns. 39186521f891SChristian Brauner */ 39196521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 39206521f891SChristian Brauner struct dentry *dentry) 39211da177e4SLinus Torvalds { 39226521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 39231da177e4SLinus Torvalds 39241da177e4SLinus Torvalds if (error) 39251da177e4SLinus Torvalds return error; 39261da177e4SLinus Torvalds 3927acfa4380SAl Viro if (!dir->i_op->rmdir) 39281da177e4SLinus Torvalds return -EPERM; 39291da177e4SLinus Torvalds 39301d2ef590SAl Viro dget(dentry); 39315955102cSAl Viro inode_lock(dentry->d_inode); 3932912dbc15SSage Weil 39331da177e4SLinus Torvalds error = -EBUSY; 39347af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3935912dbc15SSage Weil goto out; 3936912dbc15SSage Weil 39371da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3938912dbc15SSage Weil if (error) 3939912dbc15SSage Weil goto out; 3940912dbc15SSage Weil 39411da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3942912dbc15SSage Weil if (error) 3943912dbc15SSage Weil goto out; 3944912dbc15SSage Weil 39458767712fSAl Viro shrink_dcache_parent(dentry); 39461da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3947d83c49f3SAl Viro dont_mount(dentry); 39488ed936b5SEric W. Biederman detach_mounts(dentry); 3949116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 39501da177e4SLinus Torvalds 3951912dbc15SSage Weil out: 39525955102cSAl Viro inode_unlock(dentry->d_inode); 39531d2ef590SAl Viro dput(dentry); 3954912dbc15SSage Weil if (!error) 3955912dbc15SSage Weil d_delete(dentry); 39561da177e4SLinus Torvalds return error; 39571da177e4SLinus Torvalds } 39584d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 39591da177e4SLinus Torvalds 396045f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 39611da177e4SLinus Torvalds { 39626521f891SChristian Brauner struct user_namespace *mnt_userns; 39630ee50b47SDmitry Kadashev int error; 39641da177e4SLinus Torvalds struct dentry *dentry; 3965f5beed75SAl Viro struct path path; 3966f5beed75SAl Viro struct qstr last; 3967f5beed75SAl Viro int type; 3968c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3969c6ee9206SJeff Layton retry: 39700ee50b47SDmitry Kadashev error = __filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 39710ee50b47SDmitry Kadashev if (error) 39720ee50b47SDmitry Kadashev goto exit1; 39731da177e4SLinus Torvalds 3974f5beed75SAl Viro switch (type) { 39751da177e4SLinus Torvalds case LAST_DOTDOT: 39761da177e4SLinus Torvalds error = -ENOTEMPTY; 39770ee50b47SDmitry Kadashev goto exit2; 39781da177e4SLinus Torvalds case LAST_DOT: 39791da177e4SLinus Torvalds error = -EINVAL; 39800ee50b47SDmitry Kadashev goto exit2; 39811da177e4SLinus Torvalds case LAST_ROOT: 39821da177e4SLinus Torvalds error = -EBUSY; 39830ee50b47SDmitry Kadashev goto exit2; 39841da177e4SLinus Torvalds } 39850612d9fbSOGAWA Hirofumi 3986f5beed75SAl Viro error = mnt_want_write(path.mnt); 3987c30dabfeSJan Kara if (error) 39880ee50b47SDmitry Kadashev goto exit2; 39890612d9fbSOGAWA Hirofumi 39905955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3991f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39921da177e4SLinus Torvalds error = PTR_ERR(dentry); 39936902d925SDave Hansen if (IS_ERR(dentry)) 39940ee50b47SDmitry Kadashev goto exit3; 3995e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3996e6bc45d6STheodore Ts'o error = -ENOENT; 39970ee50b47SDmitry Kadashev goto exit4; 3998e6bc45d6STheodore Ts'o } 3999f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4000be6d3e56SKentaro Takeda if (error) 40010ee50b47SDmitry Kadashev goto exit4; 40026521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 40036521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 40040ee50b47SDmitry Kadashev exit4: 40051da177e4SLinus Torvalds dput(dentry); 40060ee50b47SDmitry Kadashev exit3: 40075955102cSAl Viro inode_unlock(path.dentry->d_inode); 4008f5beed75SAl Viro mnt_drop_write(path.mnt); 40090ee50b47SDmitry Kadashev exit2: 4010f5beed75SAl Viro path_put(&path); 4011c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4012c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4013c6ee9206SJeff Layton goto retry; 4014c6ee9206SJeff Layton } 40150ee50b47SDmitry Kadashev exit1: 401624fb33d4SAl Viro putname(name); 40171da177e4SLinus Torvalds return error; 40181da177e4SLinus Torvalds } 40191da177e4SLinus Torvalds 40203cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 40215590ff0dSUlrich Drepper { 4022e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 40235590ff0dSUlrich Drepper } 40245590ff0dSUlrich Drepper 4025b21996e3SJ. Bruce Fields /** 4026b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 40276521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 4028b21996e3SJ. Bruce Fields * @dir: parent directory 4029b21996e3SJ. Bruce Fields * @dentry: victim 4030b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4031b21996e3SJ. Bruce Fields * 4032b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4033b21996e3SJ. Bruce Fields * 4034b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4035b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4036b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4037b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4038b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4039b21996e3SJ. Bruce Fields * 4040b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4041b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4042b21996e3SJ. Bruce Fields * to be NFS exported. 40436521f891SChristian Brauner * 40446521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 40456521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 40466521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 40476521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 40486521f891SChristian Brauner * raw inode simply passs init_user_ns. 4049b21996e3SJ. Bruce Fields */ 40506521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 40516521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 40521da177e4SLinus Torvalds { 40539accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 40546521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 40551da177e4SLinus Torvalds 40561da177e4SLinus Torvalds if (error) 40571da177e4SLinus Torvalds return error; 40581da177e4SLinus Torvalds 4059acfa4380SAl Viro if (!dir->i_op->unlink) 40601da177e4SLinus Torvalds return -EPERM; 40611da177e4SLinus Torvalds 40625955102cSAl Viro inode_lock(target); 40638ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 40641da177e4SLinus Torvalds error = -EBUSY; 40651da177e4SLinus Torvalds else { 40661da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4067bec1052eSAl Viro if (!error) { 40685a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 40695a14696cSJ. Bruce Fields if (error) 4070b21996e3SJ. Bruce Fields goto out; 40711da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 40728ed936b5SEric W. Biederman if (!error) { 4073d83c49f3SAl Viro dont_mount(dentry); 40748ed936b5SEric W. Biederman detach_mounts(dentry); 4075116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 40768ed936b5SEric W. Biederman } 4077bec1052eSAl Viro } 40781da177e4SLinus Torvalds } 4079b21996e3SJ. Bruce Fields out: 40805955102cSAl Viro inode_unlock(target); 40811da177e4SLinus Torvalds 40821da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40831da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40849accbb97SJ. Bruce Fields fsnotify_link_count(target); 40851da177e4SLinus Torvalds d_delete(dentry); 40861da177e4SLinus Torvalds } 40870eeca283SRobert Love 40881da177e4SLinus Torvalds return error; 40891da177e4SLinus Torvalds } 40904d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40911da177e4SLinus Torvalds 40921da177e4SLinus Torvalds /* 40931da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40941b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40951da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40961da177e4SLinus Torvalds * while waiting on the I/O. 40971da177e4SLinus Torvalds */ 409845f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 40991da177e4SLinus Torvalds { 41002ad94ae6SAl Viro int error; 41011da177e4SLinus Torvalds struct dentry *dentry; 4102f5beed75SAl Viro struct path path; 4103f5beed75SAl Viro struct qstr last; 4104f5beed75SAl Viro int type; 41051da177e4SLinus Torvalds struct inode *inode = NULL; 4106b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 41075d18f813SJeff Layton unsigned int lookup_flags = 0; 41085d18f813SJeff Layton retry: 41090ee50b47SDmitry Kadashev error = __filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 41100ee50b47SDmitry Kadashev if (error) 41110ee50b47SDmitry Kadashev goto exit1; 41122ad94ae6SAl Viro 41131da177e4SLinus Torvalds error = -EISDIR; 4114f5beed75SAl Viro if (type != LAST_NORM) 41150ee50b47SDmitry Kadashev goto exit2; 41160612d9fbSOGAWA Hirofumi 4117f5beed75SAl Viro error = mnt_want_write(path.mnt); 4118c30dabfeSJan Kara if (error) 41190ee50b47SDmitry Kadashev goto exit2; 4120b21996e3SJ. Bruce Fields retry_deleg: 41215955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4122f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 41231da177e4SLinus Torvalds error = PTR_ERR(dentry); 41241da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 41256521f891SChristian Brauner struct user_namespace *mnt_userns; 41266521f891SChristian Brauner 41271da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4128f5beed75SAl Viro if (last.name[last.len]) 412950338b88STörök Edwin goto slashes; 41301da177e4SLinus Torvalds inode = dentry->d_inode; 4131b18825a7SDavid Howells if (d_is_negative(dentry)) 4132e6bc45d6STheodore Ts'o goto slashes; 41337de9c6eeSAl Viro ihold(inode); 4134f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4135be6d3e56SKentaro Takeda if (error) 41360ee50b47SDmitry Kadashev goto exit3; 41376521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4138549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4139549c7297SChristian Brauner &delegated_inode); 41400ee50b47SDmitry Kadashev exit3: 41411da177e4SLinus Torvalds dput(dentry); 41421da177e4SLinus Torvalds } 41435955102cSAl Viro inode_unlock(path.dentry->d_inode); 41441da177e4SLinus Torvalds if (inode) 41451da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4146b21996e3SJ. Bruce Fields inode = NULL; 4147b21996e3SJ. Bruce Fields if (delegated_inode) { 41485a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4149b21996e3SJ. Bruce Fields if (!error) 4150b21996e3SJ. Bruce Fields goto retry_deleg; 4151b21996e3SJ. Bruce Fields } 4152f5beed75SAl Viro mnt_drop_write(path.mnt); 41530ee50b47SDmitry Kadashev exit2: 4154f5beed75SAl Viro path_put(&path); 41555d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 41565d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 41575d18f813SJeff Layton inode = NULL; 41585d18f813SJeff Layton goto retry; 41595d18f813SJeff Layton } 41600ee50b47SDmitry Kadashev exit1: 4161da2f1362SChristoph Hellwig putname(name); 41621da177e4SLinus Torvalds return error; 41631da177e4SLinus Torvalds 41641da177e4SLinus Torvalds slashes: 4165b18825a7SDavid Howells if (d_is_negative(dentry)) 4166b18825a7SDavid Howells error = -ENOENT; 416744b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4168b18825a7SDavid Howells error = -EISDIR; 4169b18825a7SDavid Howells else 4170b18825a7SDavid Howells error = -ENOTDIR; 41710ee50b47SDmitry Kadashev goto exit3; 41721da177e4SLinus Torvalds } 41731da177e4SLinus Torvalds 41742e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 41755590ff0dSUlrich Drepper { 41765590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 41775590ff0dSUlrich Drepper return -EINVAL; 41785590ff0dSUlrich Drepper 41795590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4180e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4181da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 41825590ff0dSUlrich Drepper } 41835590ff0dSUlrich Drepper 41843480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41855590ff0dSUlrich Drepper { 4186da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 41875590ff0dSUlrich Drepper } 41885590ff0dSUlrich Drepper 41896521f891SChristian Brauner /** 41906521f891SChristian Brauner * vfs_symlink - create symlink 41916521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 41926521f891SChristian Brauner * @dir: inode of @dentry 41936521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 41946521f891SChristian Brauner * @oldname: name of the file to link to 41956521f891SChristian Brauner * 41966521f891SChristian Brauner * Create a symlink. 41976521f891SChristian Brauner * 41986521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 41996521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42006521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42016521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42026521f891SChristian Brauner * raw inode simply passs init_user_ns. 42036521f891SChristian Brauner */ 42046521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 42056521f891SChristian Brauner struct dentry *dentry, const char *oldname) 42061da177e4SLinus Torvalds { 42076521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 42081da177e4SLinus Torvalds 42091da177e4SLinus Torvalds if (error) 42101da177e4SLinus Torvalds return error; 42111da177e4SLinus Torvalds 4212acfa4380SAl Viro if (!dir->i_op->symlink) 42131da177e4SLinus Torvalds return -EPERM; 42141da177e4SLinus Torvalds 42151da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 42161da177e4SLinus Torvalds if (error) 42171da177e4SLinus Torvalds return error; 42181da177e4SLinus Torvalds 4219549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4220a74574aaSStephen Smalley if (!error) 4221f38aa942SAmy Griffis fsnotify_create(dir, dentry); 42221da177e4SLinus Torvalds return error; 42231da177e4SLinus Torvalds } 42244d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 42251da177e4SLinus Torvalds 42267a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 42271da177e4SLinus Torvalds { 42282ad94ae6SAl Viro int error; 42296902d925SDave Hansen struct dentry *dentry; 4230dae6ad8fSAl Viro struct path path; 4231f46d3567SJeff Layton unsigned int lookup_flags = 0; 42321da177e4SLinus Torvalds 4233da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4234da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4235da2d0cedSDmitry Kadashev goto out_putnames; 4236da2d0cedSDmitry Kadashev } 4237f46d3567SJeff Layton retry: 4238da2d0cedSDmitry Kadashev dentry = __filename_create(newdfd, to, &path, lookup_flags); 42391da177e4SLinus Torvalds error = PTR_ERR(dentry); 42406902d925SDave Hansen if (IS_ERR(dentry)) 4241da2d0cedSDmitry Kadashev goto out_putnames; 42426902d925SDave Hansen 424391a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 42446521f891SChristian Brauner if (!error) { 42456521f891SChristian Brauner struct user_namespace *mnt_userns; 42466521f891SChristian Brauner 42476521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 42486521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 42496521f891SChristian Brauner from->name); 42506521f891SChristian Brauner } 4251921a1650SAl Viro done_path_create(&path, dentry); 4252f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4253f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4254f46d3567SJeff Layton goto retry; 4255f46d3567SJeff Layton } 4256da2d0cedSDmitry Kadashev out_putnames: 4257da2d0cedSDmitry Kadashev putname(to); 42581da177e4SLinus Torvalds putname(from); 42591da177e4SLinus Torvalds return error; 42601da177e4SLinus Torvalds } 42611da177e4SLinus Torvalds 4262b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4263b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4264b724e846SDominik Brodowski { 4265da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4266b724e846SDominik Brodowski } 4267b724e846SDominik Brodowski 42683480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 42695590ff0dSUlrich Drepper { 4270da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 42715590ff0dSUlrich Drepper } 42725590ff0dSUlrich Drepper 4273146a8595SJ. Bruce Fields /** 4274146a8595SJ. Bruce Fields * vfs_link - create a new link 4275146a8595SJ. Bruce Fields * @old_dentry: object to be linked 42766521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4277146a8595SJ. Bruce Fields * @dir: new parent 4278146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4279146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4280146a8595SJ. Bruce Fields * 4281146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4282146a8595SJ. Bruce Fields * 4283146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4284146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4285146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4286146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4287146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4288146a8595SJ. Bruce Fields * 4289146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4290146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4291146a8595SJ. Bruce Fields * to be NFS exported. 42926521f891SChristian Brauner * 42936521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 42946521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42956521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42966521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42976521f891SChristian Brauner * raw inode simply passs init_user_ns. 4298146a8595SJ. Bruce Fields */ 42996521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 43006521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 43016521f891SChristian Brauner struct inode **delegated_inode) 43021da177e4SLinus Torvalds { 43031da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 43048de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 43051da177e4SLinus Torvalds int error; 43061da177e4SLinus Torvalds 43071da177e4SLinus Torvalds if (!inode) 43081da177e4SLinus Torvalds return -ENOENT; 43091da177e4SLinus Torvalds 43106521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 43111da177e4SLinus Torvalds if (error) 43121da177e4SLinus Torvalds return error; 43131da177e4SLinus Torvalds 43141da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 43151da177e4SLinus Torvalds return -EXDEV; 43161da177e4SLinus Torvalds 43171da177e4SLinus Torvalds /* 43181da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 43191da177e4SLinus Torvalds */ 43201da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 43211da177e4SLinus Torvalds return -EPERM; 43220bd23d09SEric W. Biederman /* 43230bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 43240bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 43250bd23d09SEric W. Biederman * the vfs. 43260bd23d09SEric W. Biederman */ 43276521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 43280bd23d09SEric W. Biederman return -EPERM; 4329acfa4380SAl Viro if (!dir->i_op->link) 43301da177e4SLinus Torvalds return -EPERM; 43317e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 43321da177e4SLinus Torvalds return -EPERM; 43331da177e4SLinus Torvalds 43341da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 43351da177e4SLinus Torvalds if (error) 43361da177e4SLinus Torvalds return error; 43371da177e4SLinus Torvalds 43385955102cSAl Viro inode_lock(inode); 4339aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4340f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4341aae8a97dSAneesh Kumar K.V error = -ENOENT; 43428de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 43438de52778SAl Viro error = -EMLINK; 4344146a8595SJ. Bruce Fields else { 4345146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4346146a8595SJ. Bruce Fields if (!error) 43471da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4348146a8595SJ. Bruce Fields } 4349f4e0c30cSAl Viro 4350f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4351f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4352f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4353f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4354f4e0c30cSAl Viro } 43555955102cSAl Viro inode_unlock(inode); 4356e31e14ecSStephen Smalley if (!error) 43577e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 43581da177e4SLinus Torvalds return error; 43591da177e4SLinus Torvalds } 43604d359507SAl Viro EXPORT_SYMBOL(vfs_link); 43611da177e4SLinus Torvalds 43621da177e4SLinus Torvalds /* 43631da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 43641da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 43651da177e4SLinus Torvalds * newname. --KAB 43661da177e4SLinus Torvalds * 43671da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 43681da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 43691da177e4SLinus Torvalds * and other special files. --ADM 43701da177e4SLinus Torvalds */ 4371*cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4372020250f3SDmitry Kadashev struct filename *new, int flags) 43731da177e4SLinus Torvalds { 43746521f891SChristian Brauner struct user_namespace *mnt_userns; 43751da177e4SLinus Torvalds struct dentry *new_dentry; 4376dae6ad8fSAl Viro struct path old_path, new_path; 4377146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 437811a7b371SAneesh Kumar K.V int how = 0; 43791da177e4SLinus Torvalds int error; 43801da177e4SLinus Torvalds 4381020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4382020250f3SDmitry Kadashev error = -EINVAL; 4383020250f3SDmitry Kadashev goto out_putnames; 4384020250f3SDmitry Kadashev } 438511a7b371SAneesh Kumar K.V /* 4386f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4387f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4388f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 438911a7b371SAneesh Kumar K.V */ 4390020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4391020250f3SDmitry Kadashev error = -ENOENT; 4392020250f3SDmitry Kadashev goto out_putnames; 4393f0cc6ffbSLinus Torvalds } 4394c04030e1SUlrich Drepper 439511a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 439611a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4397442e31caSJeff Layton retry: 4398020250f3SDmitry Kadashev error = __filename_lookup(olddfd, old, how, &old_path, NULL); 43991da177e4SLinus Torvalds if (error) 4400020250f3SDmitry Kadashev goto out_putnames; 44012ad94ae6SAl Viro 4402020250f3SDmitry Kadashev new_dentry = __filename_create(newdfd, new, &new_path, 4403442e31caSJeff Layton (how & LOOKUP_REVAL)); 44041da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44056902d925SDave Hansen if (IS_ERR(new_dentry)) 4406020250f3SDmitry Kadashev goto out_putpath; 4407dae6ad8fSAl Viro 4408dae6ad8fSAl Viro error = -EXDEV; 4409dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4410dae6ad8fSAl Viro goto out_dput; 4411549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4412549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4413800179c9SKees Cook if (unlikely(error)) 4414800179c9SKees Cook goto out_dput; 4415dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4416be6d3e56SKentaro Takeda if (error) 4417a8104a9fSAl Viro goto out_dput; 44186521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 44196521f891SChristian Brauner new_dentry, &delegated_inode); 442075c3f29dSDave Hansen out_dput: 4421921a1650SAl Viro done_path_create(&new_path, new_dentry); 4422146a8595SJ. Bruce Fields if (delegated_inode) { 4423146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4424d22e6338SOleg Drokin if (!error) { 4425d22e6338SOleg Drokin path_put(&old_path); 4426146a8595SJ. Bruce Fields goto retry; 4427146a8595SJ. Bruce Fields } 4428d22e6338SOleg Drokin } 4429442e31caSJeff Layton if (retry_estale(error, how)) { 4430d22e6338SOleg Drokin path_put(&old_path); 4431442e31caSJeff Layton how |= LOOKUP_REVAL; 4432442e31caSJeff Layton goto retry; 4433442e31caSJeff Layton } 4434020250f3SDmitry Kadashev out_putpath: 44352d8f3038SAl Viro path_put(&old_path); 4436020250f3SDmitry Kadashev out_putnames: 4437020250f3SDmitry Kadashev putname(old); 4438020250f3SDmitry Kadashev putname(new); 44391da177e4SLinus Torvalds 44401da177e4SLinus Torvalds return error; 44411da177e4SLinus Torvalds } 44421da177e4SLinus Torvalds 444346ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 444446ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 444546ea89ebSDominik Brodowski { 4446020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4447020250f3SDmitry Kadashev newdfd, getname(newname), flags); 444846ea89ebSDominik Brodowski } 444946ea89ebSDominik Brodowski 44503480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 44515590ff0dSUlrich Drepper { 4452020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 44535590ff0dSUlrich Drepper } 44545590ff0dSUlrich Drepper 4455bc27027aSMiklos Szeredi /** 4456bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 44572111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4458bc27027aSMiklos Szeredi * 4459bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4460bc27027aSMiklos Szeredi * 4461bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4462bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4463bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4464bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4465bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4466bc27027aSMiklos Szeredi * so. 4467bc27027aSMiklos Szeredi * 4468bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4469bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4470bc27027aSMiklos Szeredi * to be NFS exported. 4471bc27027aSMiklos Szeredi * 44721da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 44731da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 44741da177e4SLinus Torvalds * Problems: 44750117d427SMauro Carvalho Chehab * 4476d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 44771da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 44781da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4479a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 44801da177e4SLinus Torvalds * story. 44816cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 44826cedba89SJ. Bruce Fields * and source (if it is not a directory). 44831b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 44841da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 44851da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4486a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 44871da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 44881da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 44891da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4490a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 44911da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 44921da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 44931da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4494e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 44951b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 44961da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4497c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 44981da177e4SLinus Torvalds * locking]. 44991da177e4SLinus Torvalds */ 45009fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 45011da177e4SLinus Torvalds { 45021da177e4SLinus Torvalds int error; 45039fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 45049fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 45059fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 45069fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 45079fe61450SChristian Brauner unsigned int flags = rd->flags; 4508bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4509bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4510bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4511da1ce067SMiklos Szeredi bool new_is_dir = false; 4512da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 451349d31c2fSAl Viro struct name_snapshot old_name; 45141da177e4SLinus Torvalds 45158d3e2936SMiklos Szeredi if (source == target) 45161da177e4SLinus Torvalds return 0; 45171da177e4SLinus Torvalds 45186521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 45191da177e4SLinus Torvalds if (error) 45201da177e4SLinus Torvalds return error; 45211da177e4SLinus Torvalds 4522da1ce067SMiklos Szeredi if (!target) { 45236521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4524da1ce067SMiklos Szeredi } else { 4525da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4526da1ce067SMiklos Szeredi 4527da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 45286521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 45296521f891SChristian Brauner new_dentry, is_dir); 4530da1ce067SMiklos Szeredi else 45316521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 45326521f891SChristian Brauner new_dentry, new_is_dir); 4533da1ce067SMiklos Szeredi } 45341da177e4SLinus Torvalds if (error) 45351da177e4SLinus Torvalds return error; 45361da177e4SLinus Torvalds 45372773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 45381da177e4SLinus Torvalds return -EPERM; 45391da177e4SLinus Torvalds 4540bc27027aSMiklos Szeredi /* 4541bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4542bc27027aSMiklos Szeredi * we'll need to flip '..'. 4543bc27027aSMiklos Szeredi */ 4544da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4545da1ce067SMiklos Szeredi if (is_dir) { 45466521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 454747291baaSChristian Brauner MAY_WRITE); 4548bc27027aSMiklos Szeredi if (error) 4549bc27027aSMiklos Szeredi return error; 4550bc27027aSMiklos Szeredi } 4551da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 45526521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 455347291baaSChristian Brauner MAY_WRITE); 4554da1ce067SMiklos Szeredi if (error) 4555da1ce067SMiklos Szeredi return error; 4556da1ce067SMiklos Szeredi } 4557da1ce067SMiklos Szeredi } 45580eeca283SRobert Love 45590b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 45600b3974ebSMiklos Szeredi flags); 4561bc27027aSMiklos Szeredi if (error) 4562bc27027aSMiklos Szeredi return error; 4563bc27027aSMiklos Szeredi 456449d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4565bc27027aSMiklos Szeredi dget(new_dentry); 4566da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4567bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4568bc27027aSMiklos Szeredi else if (target) 45695955102cSAl Viro inode_lock(target); 4570bc27027aSMiklos Szeredi 4571bc27027aSMiklos Szeredi error = -EBUSY; 45727af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4573bc27027aSMiklos Szeredi goto out; 4574bc27027aSMiklos Szeredi 4575da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4576bc27027aSMiklos Szeredi error = -EMLINK; 4577da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4578bc27027aSMiklos Szeredi goto out; 4579da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4580da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4581da1ce067SMiklos Szeredi goto out; 4582da1ce067SMiklos Szeredi } 4583da1ce067SMiklos Szeredi if (!is_dir) { 4584bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4585bc27027aSMiklos Szeredi if (error) 4586bc27027aSMiklos Szeredi goto out; 4587da1ce067SMiklos Szeredi } 4588da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4589bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4590bc27027aSMiklos Szeredi if (error) 4591bc27027aSMiklos Szeredi goto out; 4592bc27027aSMiklos Szeredi } 4593549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4594520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4595bc27027aSMiklos Szeredi if (error) 4596bc27027aSMiklos Szeredi goto out; 4597bc27027aSMiklos Szeredi 4598da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 45998767712fSAl Viro if (is_dir) { 46008767712fSAl Viro shrink_dcache_parent(new_dentry); 4601bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 46028767712fSAl Viro } 4603bc27027aSMiklos Szeredi dont_mount(new_dentry); 46048ed936b5SEric W. Biederman detach_mounts(new_dentry); 4605bc27027aSMiklos Szeredi } 4606da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4607da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4608bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4609da1ce067SMiklos Szeredi else 4610da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4611da1ce067SMiklos Szeredi } 4612bc27027aSMiklos Szeredi out: 4613da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4614bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4615bc27027aSMiklos Szeredi else if (target) 46165955102cSAl Viro inode_unlock(target); 4617bc27027aSMiklos Szeredi dput(new_dentry); 4618da1ce067SMiklos Szeredi if (!error) { 4619f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4620da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4621da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4622f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4623da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4624da1ce067SMiklos Szeredi } 4625da1ce067SMiklos Szeredi } 462649d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 46270eeca283SRobert Love 46281da177e4SLinus Torvalds return error; 46291da177e4SLinus Torvalds } 46304d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 46311da177e4SLinus Torvalds 4632e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4633e886663cSJens Axboe struct filename *to, unsigned int flags) 46341da177e4SLinus Torvalds { 46359fe61450SChristian Brauner struct renamedata rd; 46361da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 46371da177e4SLinus Torvalds struct dentry *trap; 4638f5beed75SAl Viro struct path old_path, new_path; 4639f5beed75SAl Viro struct qstr old_last, new_last; 4640f5beed75SAl Viro int old_type, new_type; 46418e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4642f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4643c6a94284SJeff Layton bool should_retry = false; 4644e886663cSJens Axboe int error = -EINVAL; 4645520c8b16SMiklos Szeredi 46460d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 46470ee50b47SDmitry Kadashev goto put_names; 4648da1ce067SMiklos Szeredi 46490d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 46500d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 46510ee50b47SDmitry Kadashev goto put_names; 4652520c8b16SMiklos Szeredi 4653f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4654f5beed75SAl Viro target_flags = 0; 4655f5beed75SAl Viro 4656c6a94284SJeff Layton retry: 46570ee50b47SDmitry Kadashev error = __filename_parentat(olddfd, from, lookup_flags, &old_path, 4658e886663cSJens Axboe &old_last, &old_type); 46590ee50b47SDmitry Kadashev if (error) 46600ee50b47SDmitry Kadashev goto put_names; 46611da177e4SLinus Torvalds 46620ee50b47SDmitry Kadashev error = __filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4663e886663cSJens Axboe &new_type); 46640ee50b47SDmitry Kadashev if (error) 46651da177e4SLinus Torvalds goto exit1; 46661da177e4SLinus Torvalds 46671da177e4SLinus Torvalds error = -EXDEV; 4668f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 46691da177e4SLinus Torvalds goto exit2; 46701da177e4SLinus Torvalds 46711da177e4SLinus Torvalds error = -EBUSY; 4672f5beed75SAl Viro if (old_type != LAST_NORM) 46731da177e4SLinus Torvalds goto exit2; 46741da177e4SLinus Torvalds 46750a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 46760a7c3937SMiklos Szeredi error = -EEXIST; 4677f5beed75SAl Viro if (new_type != LAST_NORM) 46781da177e4SLinus Torvalds goto exit2; 46791da177e4SLinus Torvalds 4680f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4681c30dabfeSJan Kara if (error) 4682c30dabfeSJan Kara goto exit2; 4683c30dabfeSJan Kara 46848e6d782cSJ. Bruce Fields retry_deleg: 4685f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 46861da177e4SLinus Torvalds 4687f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 46881da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 46891da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 46901da177e4SLinus Torvalds goto exit3; 46911da177e4SLinus Torvalds /* source must exist */ 46921da177e4SLinus Torvalds error = -ENOENT; 4693b18825a7SDavid Howells if (d_is_negative(old_dentry)) 46941da177e4SLinus Torvalds goto exit4; 4695f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 46961da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 46971da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 46981da177e4SLinus Torvalds goto exit4; 46990a7c3937SMiklos Szeredi error = -EEXIST; 47000a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 47010a7c3937SMiklos Szeredi goto exit5; 4702da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4703da1ce067SMiklos Szeredi error = -ENOENT; 4704da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4705da1ce067SMiklos Szeredi goto exit5; 4706da1ce067SMiklos Szeredi 4707da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4708da1ce067SMiklos Szeredi error = -ENOTDIR; 4709f5beed75SAl Viro if (new_last.name[new_last.len]) 4710da1ce067SMiklos Szeredi goto exit5; 4711da1ce067SMiklos Szeredi } 4712da1ce067SMiklos Szeredi } 47130a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 47140a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 47150a7c3937SMiklos Szeredi error = -ENOTDIR; 4716f5beed75SAl Viro if (old_last.name[old_last.len]) 47170a7c3937SMiklos Szeredi goto exit5; 4718f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 47190a7c3937SMiklos Szeredi goto exit5; 47200a7c3937SMiklos Szeredi } 47210a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 47220a7c3937SMiklos Szeredi error = -EINVAL; 47230a7c3937SMiklos Szeredi if (old_dentry == trap) 47240a7c3937SMiklos Szeredi goto exit5; 47251da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4726da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 47271da177e4SLinus Torvalds error = -ENOTEMPTY; 47281da177e4SLinus Torvalds if (new_dentry == trap) 47291da177e4SLinus Torvalds goto exit5; 47301da177e4SLinus Torvalds 4731f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4732f5beed75SAl Viro &new_path, new_dentry, flags); 4733be6d3e56SKentaro Takeda if (error) 4734c30dabfeSJan Kara goto exit5; 47359fe61450SChristian Brauner 47369fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 47379fe61450SChristian Brauner rd.old_dentry = old_dentry; 47386521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 47399fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 47409fe61450SChristian Brauner rd.new_dentry = new_dentry; 47416521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 47429fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 47439fe61450SChristian Brauner rd.flags = flags; 47449fe61450SChristian Brauner error = vfs_rename(&rd); 47451da177e4SLinus Torvalds exit5: 47461da177e4SLinus Torvalds dput(new_dentry); 47471da177e4SLinus Torvalds exit4: 47481da177e4SLinus Torvalds dput(old_dentry); 47491da177e4SLinus Torvalds exit3: 4750f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 47518e6d782cSJ. Bruce Fields if (delegated_inode) { 47528e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 47538e6d782cSJ. Bruce Fields if (!error) 47548e6d782cSJ. Bruce Fields goto retry_deleg; 47558e6d782cSJ. Bruce Fields } 4756f5beed75SAl Viro mnt_drop_write(old_path.mnt); 47571da177e4SLinus Torvalds exit2: 4758c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4759c6a94284SJeff Layton should_retry = true; 4760f5beed75SAl Viro path_put(&new_path); 47611da177e4SLinus Torvalds exit1: 4762f5beed75SAl Viro path_put(&old_path); 4763c6a94284SJeff Layton if (should_retry) { 4764c6a94284SJeff Layton should_retry = false; 4765c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4766c6a94284SJeff Layton goto retry; 4767c6a94284SJeff Layton } 47680ee50b47SDmitry Kadashev put_names: 4769e886663cSJens Axboe putname(from); 4770e886663cSJens Axboe putname(to); 47711da177e4SLinus Torvalds return error; 47721da177e4SLinus Torvalds } 47731da177e4SLinus Torvalds 4774ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4775ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4776ee81feb6SDominik Brodowski { 4777e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4778e886663cSJens Axboe flags); 4779ee81feb6SDominik Brodowski } 4780ee81feb6SDominik Brodowski 4781520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4782520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4783520c8b16SMiklos Szeredi { 4784e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4785e886663cSJens Axboe 0); 4786520c8b16SMiklos Szeredi } 4787520c8b16SMiklos Szeredi 4788a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 47895590ff0dSUlrich Drepper { 4790e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4791e886663cSJens Axboe getname(newname), 0); 47925590ff0dSUlrich Drepper } 47935590ff0dSUlrich Drepper 47945d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 47951da177e4SLinus Torvalds { 47965d826c84SAl Viro int len = PTR_ERR(link); 47971da177e4SLinus Torvalds if (IS_ERR(link)) 47981da177e4SLinus Torvalds goto out; 47991da177e4SLinus Torvalds 48001da177e4SLinus Torvalds len = strlen(link); 48011da177e4SLinus Torvalds if (len > (unsigned) buflen) 48021da177e4SLinus Torvalds len = buflen; 48031da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 48041da177e4SLinus Torvalds len = -EFAULT; 48051da177e4SLinus Torvalds out: 48061da177e4SLinus Torvalds return len; 48071da177e4SLinus Torvalds } 48081da177e4SLinus Torvalds 4809d60874cdSMiklos Szeredi /** 4810fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4811fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4812fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4813fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4814fd4a0edfSMiklos Szeredi * 4815fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4816fd4a0edfSMiklos Szeredi * 4817fd4a0edfSMiklos Szeredi * Does not call security hook. 4818fd4a0edfSMiklos Szeredi */ 4819fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4820fd4a0edfSMiklos Szeredi { 4821fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4822f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4823f2df5da6SAl Viro const char *link; 4824f2df5da6SAl Viro int res; 4825fd4a0edfSMiklos Szeredi 482676fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 482776fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 482876fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 482976fca90eSMiklos Szeredi 483076fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4831fd4a0edfSMiklos Szeredi return -EINVAL; 4832fd4a0edfSMiklos Szeredi 483376fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 483476fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 483576fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 483676fca90eSMiklos Szeredi } 483776fca90eSMiklos Szeredi 48384c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4839f2df5da6SAl Viro if (!link) { 4840f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4841f2df5da6SAl Viro if (IS_ERR(link)) 4842f2df5da6SAl Viro return PTR_ERR(link); 4843f2df5da6SAl Viro } 4844f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4845f2df5da6SAl Viro do_delayed_call(&done); 4846f2df5da6SAl Viro return res; 4847fd4a0edfSMiklos Szeredi } 4848fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 48491da177e4SLinus Torvalds 4850d60874cdSMiklos Szeredi /** 4851d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4852d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4853d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4854d60874cdSMiklos Szeredi * 4855d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4856d60874cdSMiklos Szeredi * 4857d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4858d60874cdSMiklos Szeredi * 4859d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4860d60874cdSMiklos Szeredi */ 4861d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4862d60874cdSMiklos Szeredi { 4863d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4864d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4865d60874cdSMiklos Szeredi 4866d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4867d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4868d60874cdSMiklos Szeredi if (!res) 4869d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4870d60874cdSMiklos Szeredi } 4871d60874cdSMiklos Szeredi return res; 4872d60874cdSMiklos Szeredi } 4873d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4874d60874cdSMiklos Szeredi 48751da177e4SLinus Torvalds /* get the link contents into pagecache */ 48766b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4877fceef393SAl Viro struct delayed_call *callback) 48781da177e4SLinus Torvalds { 4879ebd09abbSDuane Griffin char *kaddr; 48801da177e4SLinus Torvalds struct page *page; 48816b255391SAl Viro struct address_space *mapping = inode->i_mapping; 48826b255391SAl Viro 4883d3883d4fSAl Viro if (!dentry) { 4884d3883d4fSAl Viro page = find_get_page(mapping, 0); 4885d3883d4fSAl Viro if (!page) 48866b255391SAl Viro return ERR_PTR(-ECHILD); 4887d3883d4fSAl Viro if (!PageUptodate(page)) { 4888d3883d4fSAl Viro put_page(page); 4889d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4890d3883d4fSAl Viro } 4891d3883d4fSAl Viro } else { 4892090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 48931da177e4SLinus Torvalds if (IS_ERR(page)) 48946fe6900eSNick Piggin return (char*)page; 4895d3883d4fSAl Viro } 4896fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 489721fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 489821fc61c7SAl Viro kaddr = page_address(page); 48996b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4900ebd09abbSDuane Griffin return kaddr; 49011da177e4SLinus Torvalds } 49021da177e4SLinus Torvalds 49036b255391SAl Viro EXPORT_SYMBOL(page_get_link); 49041da177e4SLinus Torvalds 4905fceef393SAl Viro void page_put_link(void *arg) 49061da177e4SLinus Torvalds { 4907fceef393SAl Viro put_page(arg); 49081da177e4SLinus Torvalds } 49094d359507SAl Viro EXPORT_SYMBOL(page_put_link); 49101da177e4SLinus Torvalds 4911aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4912aa80deabSAl Viro { 4913fceef393SAl Viro DEFINE_DELAYED_CALL(done); 49146b255391SAl Viro int res = readlink_copy(buffer, buflen, 49156b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4916fceef393SAl Viro &done)); 4917fceef393SAl Viro do_delayed_call(&done); 4918aa80deabSAl Viro return res; 4919aa80deabSAl Viro } 4920aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4921aa80deabSAl Viro 492254566b2cSNick Piggin /* 492354566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 492454566b2cSNick Piggin */ 492554566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 49261da177e4SLinus Torvalds { 49271da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 49280adb25d2SKirill Korotaev struct page *page; 4929afddba49SNick Piggin void *fsdata; 4930beb497abSDmitriy Monakhov int err; 4931c718a975STetsuo Handa unsigned int flags = 0; 493254566b2cSNick Piggin if (nofs) 493354566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 49341da177e4SLinus Torvalds 49357e53cac4SNeilBrown retry: 4936afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 493754566b2cSNick Piggin flags, &page, &fsdata); 49381da177e4SLinus Torvalds if (err) 4939afddba49SNick Piggin goto fail; 4940afddba49SNick Piggin 494121fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4942afddba49SNick Piggin 4943afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4944afddba49SNick Piggin page, fsdata); 49451da177e4SLinus Torvalds if (err < 0) 49461da177e4SLinus Torvalds goto fail; 4947afddba49SNick Piggin if (err < len-1) 4948afddba49SNick Piggin goto retry; 4949afddba49SNick Piggin 49501da177e4SLinus Torvalds mark_inode_dirty(inode); 49511da177e4SLinus Torvalds return 0; 49521da177e4SLinus Torvalds fail: 49531da177e4SLinus Torvalds return err; 49541da177e4SLinus Torvalds } 49554d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 49561da177e4SLinus Torvalds 49570adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 49580adb25d2SKirill Korotaev { 49590adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4960c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 49610adb25d2SKirill Korotaev } 49624d359507SAl Viro EXPORT_SYMBOL(page_symlink); 49630adb25d2SKirill Korotaev 496492e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 49656b255391SAl Viro .get_link = page_get_link, 49661da177e4SLinus Torvalds }; 49671da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4968