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> 252d878178SMatthew Wilcox (Oracle) #include <linux/sched/mm.h> 260eeca283SRobert Love #include <linux/fsnotify.h> 271da177e4SLinus Torvalds #include <linux/personality.h> 281da177e4SLinus Torvalds #include <linux/security.h> 296146f0d5SMimi Zohar #include <linux/ima.h> 301da177e4SLinus Torvalds #include <linux/syscalls.h> 311da177e4SLinus Torvalds #include <linux/mount.h> 321da177e4SLinus Torvalds #include <linux/audit.h> 3316f7e0feSRandy Dunlap #include <linux/capability.h> 34834f2a4aSTrond Myklebust #include <linux/file.h> 355590ff0dSUlrich Drepper #include <linux/fcntl.h> 3608ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 375ad4e53bSAl Viro #include <linux/fs_struct.h> 38e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3999d263d4SLinus Torvalds #include <linux/hash.h> 402a18da7aSGeorge Spelvin #include <linux/bitops.h> 41aeaa4a79SEric W. Biederman #include <linux/init_task.h> 427c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 431da177e4SLinus Torvalds 44e81e3f4dSEric Paris #include "internal.h" 45c7105365SAl Viro #include "mount.h" 46e81e3f4dSEric Paris 471da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 481da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 491da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 501da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 511da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 521da177e4SLinus Torvalds * 531da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 541da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 551da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 561da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 571da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 581da177e4SLinus Torvalds * the special cases of the former code. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 611da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 621da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 631da177e4SLinus Torvalds * 641da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 651da177e4SLinus Torvalds * resolution to correspond with current state of the code. 661da177e4SLinus Torvalds * 671da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 681da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 691da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 701da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 711da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 721da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 731da177e4SLinus Torvalds */ 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 761da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 771da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 781da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 791da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 801da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8125985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 821da177e4SLinus Torvalds * 831da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 841da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 851da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 861da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 871da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 881da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 891da177e4SLinus Torvalds * and in the old Linux semantics. 901da177e4SLinus Torvalds */ 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 931da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 941da177e4SLinus Torvalds * 951da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 961da177e4SLinus Torvalds */ 971da177e4SLinus Torvalds 981da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 991da177e4SLinus Torvalds * inside the path - always follow. 1001da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1011da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1021da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1031da177e4SLinus Torvalds * otherwise - don't follow. 1041da177e4SLinus Torvalds * (applied in that order). 1051da177e4SLinus Torvalds * 1061da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1071da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1081da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1091da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1101da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1111da177e4SLinus Torvalds */ 1121da177e4SLinus Torvalds /* 1131da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 114a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1151da177e4SLinus Torvalds * any extra contention... 1161da177e4SLinus Torvalds */ 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1191da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1201da177e4SLinus Torvalds * kernel data space before using them.. 1211da177e4SLinus Torvalds * 1221da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1231da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1241da177e4SLinus Torvalds */ 1257950e385SJeff Layton 126fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12791a27b2aSJeff Layton 12851f39a1fSDavid Drysdale struct filename * 12991a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13091a27b2aSJeff Layton { 13194b5d262SAl Viro struct filename *result; 1327950e385SJeff Layton char *kname; 13394b5d262SAl Viro int len; 1341da177e4SLinus Torvalds 1357ac86265SJeff Layton result = audit_reusename(filename); 1367ac86265SJeff Layton if (result) 1377ac86265SJeff Layton return result; 1387ac86265SJeff Layton 1397950e385SJeff Layton result = __getname(); 1403f9f0aa6SLinus Torvalds if (unlikely(!result)) 1414043cde8SEric Paris return ERR_PTR(-ENOMEM); 1421da177e4SLinus Torvalds 1437950e385SJeff Layton /* 1447950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1457950e385SJeff Layton * allocation 1467950e385SJeff Layton */ 147fd2f7cb5SAl Viro kname = (char *)result->iname; 14891a27b2aSJeff Layton result->name = kname; 1497950e385SJeff Layton 15094b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15191a27b2aSJeff Layton if (unlikely(len < 0)) { 15294b5d262SAl Viro __putname(result); 15394b5d262SAl Viro return ERR_PTR(len); 15491a27b2aSJeff Layton } 1553f9f0aa6SLinus Torvalds 1567950e385SJeff Layton /* 1577950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1587950e385SJeff Layton * separate struct filename so we can dedicate the entire 1597950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1607950e385SJeff Layton * userland. 1617950e385SJeff Layton */ 16294b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 163fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1647950e385SJeff Layton kname = (char *)result; 1657950e385SJeff Layton 166fd2f7cb5SAl Viro /* 167fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 168fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 169fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 170fd2f7cb5SAl Viro */ 171fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17294b5d262SAl Viro if (unlikely(!result)) { 17394b5d262SAl Viro __putname(kname); 17494b5d262SAl Viro return ERR_PTR(-ENOMEM); 1757950e385SJeff Layton } 1767950e385SJeff Layton result->name = kname; 17794b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17894b5d262SAl Viro if (unlikely(len < 0)) { 17994b5d262SAl Viro __putname(kname); 18094b5d262SAl Viro kfree(result); 18194b5d262SAl Viro return ERR_PTR(len); 18294b5d262SAl Viro } 18394b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18494b5d262SAl Viro __putname(kname); 18594b5d262SAl Viro kfree(result); 18694b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18794b5d262SAl Viro } 1887950e385SJeff Layton } 1897950e385SJeff Layton 19094b5d262SAl Viro result->refcnt = 1; 1913f9f0aa6SLinus Torvalds /* The empty path is special. */ 1923f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1933f9f0aa6SLinus Torvalds if (empty) 1941fa1e7f6SAndy Whitcroft *empty = 1; 19594b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19694b5d262SAl Viro putname(result); 19794b5d262SAl Viro return ERR_PTR(-ENOENT); 1981da177e4SLinus Torvalds } 19994b5d262SAl Viro } 2007950e385SJeff Layton 2017950e385SJeff Layton result->uptr = filename; 202c4ad8f98SLinus Torvalds result->aname = NULL; 2031da177e4SLinus Torvalds audit_getname(result); 2041da177e4SLinus Torvalds return result; 2053f9f0aa6SLinus Torvalds } 2063f9f0aa6SLinus Torvalds 20791a27b2aSJeff Layton struct filename * 2088228e2c3SDmitry Kadashev getname_uflags(const char __user *filename, int uflags) 2098228e2c3SDmitry Kadashev { 2108228e2c3SDmitry Kadashev int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 2118228e2c3SDmitry Kadashev 2128228e2c3SDmitry Kadashev return getname_flags(filename, flags, NULL); 2138228e2c3SDmitry Kadashev } 2148228e2c3SDmitry Kadashev 2158228e2c3SDmitry Kadashev struct filename * 21691a27b2aSJeff Layton getname(const char __user * filename) 217f52e0c11SAl Viro { 218f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 219f52e0c11SAl Viro } 220f52e0c11SAl Viro 221c4ad8f98SLinus Torvalds struct filename * 222c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 223c4ad8f98SLinus Torvalds { 224c4ad8f98SLinus Torvalds struct filename *result; 22508518549SPaul Moore int len = strlen(filename) + 1; 226c4ad8f98SLinus Torvalds 227c4ad8f98SLinus Torvalds result = __getname(); 228c4ad8f98SLinus Torvalds if (unlikely(!result)) 229c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 230c4ad8f98SLinus Torvalds 23108518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 232fd2f7cb5SAl Viro result->name = (char *)result->iname; 23308518549SPaul Moore } else if (len <= PATH_MAX) { 23430ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 23508518549SPaul Moore struct filename *tmp; 23608518549SPaul Moore 23730ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 23808518549SPaul Moore if (unlikely(!tmp)) { 23908518549SPaul Moore __putname(result); 24008518549SPaul Moore return ERR_PTR(-ENOMEM); 24108518549SPaul Moore } 24208518549SPaul Moore tmp->name = (char *)result; 24308518549SPaul Moore result = tmp; 24408518549SPaul Moore } else { 24508518549SPaul Moore __putname(result); 24608518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 24708518549SPaul Moore } 24808518549SPaul Moore memcpy((char *)result->name, filename, len); 249c4ad8f98SLinus Torvalds result->uptr = NULL; 250c4ad8f98SLinus Torvalds result->aname = NULL; 25155422d0bSPaul Moore result->refcnt = 1; 252fd3522fdSPaul Moore audit_getname(result); 253c4ad8f98SLinus Torvalds 254c4ad8f98SLinus Torvalds return result; 255c4ad8f98SLinus Torvalds } 256c4ad8f98SLinus Torvalds 25791a27b2aSJeff Layton void putname(struct filename *name) 2581da177e4SLinus Torvalds { 259ea47ab11SAl Viro if (IS_ERR(name)) 26091ef658fSDmitry Kadashev return; 26191ef658fSDmitry Kadashev 26255422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 26355422d0bSPaul Moore 26455422d0bSPaul Moore if (--name->refcnt > 0) 26555422d0bSPaul Moore return; 26655422d0bSPaul Moore 267fd2f7cb5SAl Viro if (name->name != name->iname) { 26855422d0bSPaul Moore __putname(name->name); 26955422d0bSPaul Moore kfree(name); 27055422d0bSPaul Moore } else 27155422d0bSPaul Moore __putname(name); 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 27447291baaSChristian Brauner /** 27547291baaSChristian Brauner * check_acl - perform ACL permission checking 27647291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 27747291baaSChristian Brauner * @inode: inode to check permissions on 27847291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 27947291baaSChristian Brauner * 28047291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 28147291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 28247291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 28347291baaSChristian Brauner * 28447291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 28547291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 28647291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 28747291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 28847291baaSChristian Brauner * raw inode simply passs init_user_ns. 28947291baaSChristian Brauner */ 29047291baaSChristian Brauner static int check_acl(struct user_namespace *mnt_userns, 29147291baaSChristian Brauner struct inode *inode, int mask) 292e77819e5SLinus Torvalds { 29384635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 294e77819e5SLinus Torvalds struct posix_acl *acl; 295e77819e5SLinus Torvalds 296e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2973567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2983567866bSAl Viro if (!acl) 299e77819e5SLinus Torvalds return -EAGAIN; 3003567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 301b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 302e77819e5SLinus Torvalds return -ECHILD; 30347291baaSChristian Brauner return posix_acl_permission(mnt_userns, inode, acl, mask); 304e77819e5SLinus Torvalds } 305e77819e5SLinus Torvalds 3062982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 3074e34e719SChristoph Hellwig if (IS_ERR(acl)) 3084e34e719SChristoph Hellwig return PTR_ERR(acl); 309e77819e5SLinus Torvalds if (acl) { 31047291baaSChristian Brauner int error = posix_acl_permission(mnt_userns, inode, acl, mask); 311e77819e5SLinus Torvalds posix_acl_release(acl); 312e77819e5SLinus Torvalds return error; 313e77819e5SLinus Torvalds } 31484635d68SLinus Torvalds #endif 315e77819e5SLinus Torvalds 316e77819e5SLinus Torvalds return -EAGAIN; 317e77819e5SLinus Torvalds } 318e77819e5SLinus Torvalds 31947291baaSChristian Brauner /** 32047291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 32147291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 32247291baaSChristian Brauner * @inode: inode to check permissions on 32347291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3245fc475b7SLinus Torvalds * 32547291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 32647291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 32747291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 32847291baaSChristian Brauner * 32947291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 33047291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 33147291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 33247291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 33347291baaSChristian Brauner * raw inode simply passs init_user_ns. 3345909ccaaSLinus Torvalds */ 33547291baaSChristian Brauner static int acl_permission_check(struct user_namespace *mnt_userns, 33647291baaSChristian Brauner struct inode *inode, int mask) 3375909ccaaSLinus Torvalds { 33826cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 33947291baaSChristian Brauner kuid_t i_uid; 3405909ccaaSLinus Torvalds 3415fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 34247291baaSChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 34347291baaSChristian Brauner if (likely(uid_eq(current_fsuid(), i_uid))) { 3445fc475b7SLinus Torvalds mask &= 7; 3455909ccaaSLinus Torvalds mode >>= 6; 3465fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3475fc475b7SLinus Torvalds } 3485fc475b7SLinus Torvalds 3495fc475b7SLinus Torvalds /* Do we have ACL's? */ 350e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 35147291baaSChristian Brauner int error = check_acl(mnt_userns, inode, mask); 3525909ccaaSLinus Torvalds if (error != -EAGAIN) 3535909ccaaSLinus Torvalds return error; 3545909ccaaSLinus Torvalds } 3555909ccaaSLinus Torvalds 3565fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3575fc475b7SLinus Torvalds mask &= 7; 3585fc475b7SLinus Torvalds 3595fc475b7SLinus Torvalds /* 3605fc475b7SLinus Torvalds * Are the group permissions different from 3615fc475b7SLinus Torvalds * the other permissions in the bits we care 3625fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3635fc475b7SLinus Torvalds */ 3645fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 36547291baaSChristian Brauner kgid_t kgid = i_gid_into_mnt(mnt_userns, inode); 36647291baaSChristian Brauner if (in_group_p(kgid)) 3675909ccaaSLinus Torvalds mode >>= 3; 3685909ccaaSLinus Torvalds } 3695909ccaaSLinus Torvalds 3705fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3715fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3725909ccaaSLinus Torvalds } 3731da177e4SLinus Torvalds 3741da177e4SLinus Torvalds /** 3751da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 37647291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 3771da177e4SLinus Torvalds * @inode: inode to check access rights for 3785fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3795fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3801da177e4SLinus Torvalds * 3811da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3821da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3831da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 384b74c79e9SNick Piggin * are used for other things. 385b74c79e9SNick Piggin * 386b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 387b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 388b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 38947291baaSChristian Brauner * 39047291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39147291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39247291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39347291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39447291baaSChristian Brauner * raw inode simply passs init_user_ns. 3951da177e4SLinus Torvalds */ 39647291baaSChristian Brauner int generic_permission(struct user_namespace *mnt_userns, struct inode *inode, 39747291baaSChristian Brauner int mask) 3981da177e4SLinus Torvalds { 3995909ccaaSLinus Torvalds int ret; 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds /* 402948409c7SAndreas Gruenbacher * Do the basic permission checks. 4031da177e4SLinus Torvalds */ 40447291baaSChristian Brauner ret = acl_permission_check(mnt_userns, inode, mask); 4055909ccaaSLinus Torvalds if (ret != -EACCES) 4065909ccaaSLinus Torvalds return ret; 4071da177e4SLinus Torvalds 408d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 409d594e7ecSAl Viro /* DACs are overridable for directories */ 410d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 41147291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 41223adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 413d594e7ecSAl Viro return 0; 41447291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4150558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4161da177e4SLinus Torvalds return 0; 4172a4c2242SStephen Smalley return -EACCES; 4182a4c2242SStephen Smalley } 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds /* 4211da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4221da177e4SLinus Torvalds */ 4237ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 424d594e7ecSAl Viro if (mask == MAY_READ) 42547291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4260558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4271da177e4SLinus Torvalds return 0; 4282a4c2242SStephen Smalley /* 4292a4c2242SStephen Smalley * Read/write DACs are always overridable. 4302a4c2242SStephen Smalley * Executable DACs are overridable when there is 4312a4c2242SStephen Smalley * at least one exec bit set. 4322a4c2242SStephen Smalley */ 4332a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 43447291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4350558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4362a4c2242SStephen Smalley return 0; 4371da177e4SLinus Torvalds 4381da177e4SLinus Torvalds return -EACCES; 4391da177e4SLinus Torvalds } 4404d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4411da177e4SLinus Torvalds 44247291baaSChristian Brauner /** 44347291baaSChristian Brauner * do_inode_permission - UNIX permission checking 44447291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 44547291baaSChristian Brauner * @inode: inode to check permissions on 44647291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 44747291baaSChristian Brauner * 4483ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4493ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4503ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4513ddcd056SLinus Torvalds * permission function, use the fast case". 4523ddcd056SLinus Torvalds */ 45347291baaSChristian Brauner static inline int do_inode_permission(struct user_namespace *mnt_userns, 45447291baaSChristian Brauner struct inode *inode, int mask) 4553ddcd056SLinus Torvalds { 4563ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4573ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 458549c7297SChristian Brauner return inode->i_op->permission(mnt_userns, inode, mask); 4593ddcd056SLinus Torvalds 4603ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4613ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4623ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4633ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4643ddcd056SLinus Torvalds } 46547291baaSChristian Brauner return generic_permission(mnt_userns, inode, mask); 4663ddcd056SLinus Torvalds } 4673ddcd056SLinus Torvalds 468cb23beb5SChristoph Hellwig /** 4690bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4700bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 47155852635SRandy Dunlap * @inode: Inode to check permission on 4720bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4730bdaea90SDavid Howells * 4740bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4750bdaea90SDavid Howells */ 4760bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4770bdaea90SDavid Howells { 4780bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4790bdaea90SDavid Howells umode_t mode = inode->i_mode; 4800bdaea90SDavid Howells 4810bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 482bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4830bdaea90SDavid Howells return -EROFS; 4840bdaea90SDavid Howells } 4850bdaea90SDavid Howells return 0; 4860bdaea90SDavid Howells } 4870bdaea90SDavid Howells 4880bdaea90SDavid Howells /** 4890bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 49047291baaSChristian Brauner * @mnt_userns: User namespace of the mount the inode was found from 4910bdaea90SDavid Howells * @inode: Inode to check permission on 4920bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4930bdaea90SDavid Howells * 4940bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4950bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4960bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4970bdaea90SDavid Howells * 4980bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4990bdaea90SDavid Howells */ 50047291baaSChristian Brauner int inode_permission(struct user_namespace *mnt_userns, 50147291baaSChristian Brauner struct inode *inode, int mask) 5020bdaea90SDavid Howells { 5030bdaea90SDavid Howells int retval; 5040bdaea90SDavid Howells 5050bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 5060bdaea90SDavid Howells if (retval) 5070bdaea90SDavid Howells return retval; 5084bfd054aSEric Biggers 5094bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 5104bfd054aSEric Biggers /* 5114bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5124bfd054aSEric Biggers */ 5134bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5144bfd054aSEric Biggers return -EPERM; 5154bfd054aSEric Biggers 5164bfd054aSEric Biggers /* 5174bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5184bfd054aSEric Biggers * written back improperly if their true value is unknown 5194bfd054aSEric Biggers * to the vfs. 5204bfd054aSEric Biggers */ 521ba73d987SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 5224bfd054aSEric Biggers return -EACCES; 5234bfd054aSEric Biggers } 5244bfd054aSEric Biggers 52547291baaSChristian Brauner retval = do_inode_permission(mnt_userns, inode, mask); 5264bfd054aSEric Biggers if (retval) 5274bfd054aSEric Biggers return retval; 5284bfd054aSEric Biggers 5294bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5304bfd054aSEric Biggers if (retval) 5314bfd054aSEric Biggers return retval; 5324bfd054aSEric Biggers 5334bfd054aSEric Biggers return security_inode_permission(inode, mask); 5340bdaea90SDavid Howells } 5354d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5360bdaea90SDavid Howells 5370bdaea90SDavid Howells /** 5385dd784d0SJan Blunck * path_get - get a reference to a path 5395dd784d0SJan Blunck * @path: path to get the reference to 5405dd784d0SJan Blunck * 5415dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5425dd784d0SJan Blunck */ 543dcf787f3SAl Viro void path_get(const struct path *path) 5445dd784d0SJan Blunck { 5455dd784d0SJan Blunck mntget(path->mnt); 5465dd784d0SJan Blunck dget(path->dentry); 5475dd784d0SJan Blunck } 5485dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5495dd784d0SJan Blunck 5505dd784d0SJan Blunck /** 5511d957f9bSJan Blunck * path_put - put a reference to a path 5521d957f9bSJan Blunck * @path: path to put the reference to 5531d957f9bSJan Blunck * 5541d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5551d957f9bSJan Blunck */ 556dcf787f3SAl Viro void path_put(const struct path *path) 5571da177e4SLinus Torvalds { 5581d957f9bSJan Blunck dput(path->dentry); 5591d957f9bSJan Blunck mntput(path->mnt); 5601da177e4SLinus Torvalds } 5611d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5621da177e4SLinus Torvalds 563894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5641f55a6ecSAl Viro struct nameidata { 5651f55a6ecSAl Viro struct path path; 5661f55a6ecSAl Viro struct qstr last; 5671f55a6ecSAl Viro struct path root; 5681f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 569bcba1e7dSAl Viro unsigned int flags, state; 570ab87f9a5SAleksa Sarai unsigned seq, m_seq, r_seq; 5711f55a6ecSAl Viro int last_type; 5721f55a6ecSAl Viro unsigned depth; 573756daf26SNeilBrown int total_link_count; 574697fc6caSAl Viro struct saved { 575697fc6caSAl Viro struct path link; 576fceef393SAl Viro struct delayed_call done; 577697fc6caSAl Viro const char *name; 5780450b2d1SAl Viro unsigned seq; 579894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5809883d185SAl Viro struct filename *name; 5819883d185SAl Viro struct nameidata *saved; 5829883d185SAl Viro unsigned root_seq; 5839883d185SAl Viro int dfd; 5840f705953SAl Viro kuid_t dir_uid; 5850f705953SAl Viro umode_t dir_mode; 5863859a271SKees Cook } __randomize_layout; 5871f55a6ecSAl Viro 588bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 589bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 590bcba1e7dSAl Viro #define ND_JUMPED 4 591bcba1e7dSAl Viro 59206422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 593894bc8c4SAl Viro { 594756daf26SNeilBrown struct nameidata *old = current->nameidata; 595756daf26SNeilBrown p->stack = p->internal; 5967962c7d1SAl Viro p->depth = 0; 597c8a53ee5SAl Viro p->dfd = dfd; 598c8a53ee5SAl Viro p->name = name; 5997d01ef75SAl Viro p->path.mnt = NULL; 6007d01ef75SAl Viro p->path.dentry = NULL; 601756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 6029883d185SAl Viro p->saved = old; 603756daf26SNeilBrown current->nameidata = p; 604894bc8c4SAl Viro } 605894bc8c4SAl Viro 60606422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 60706422964SAl Viro const struct path *root) 60806422964SAl Viro { 60906422964SAl Viro __set_nameidata(p, dfd, name); 61006422964SAl Viro p->state = 0; 61106422964SAl Viro if (unlikely(root)) { 61206422964SAl Viro p->state = ND_ROOT_PRESET; 61306422964SAl Viro p->root = *root; 61406422964SAl Viro } 61506422964SAl Viro } 61606422964SAl Viro 6179883d185SAl Viro static void restore_nameidata(void) 618894bc8c4SAl Viro { 6199883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 620756daf26SNeilBrown 621756daf26SNeilBrown current->nameidata = old; 622756daf26SNeilBrown if (old) 623756daf26SNeilBrown old->total_link_count = now->total_link_count; 624e1a63bbcSAl Viro if (now->stack != now->internal) 625756daf26SNeilBrown kfree(now->stack); 626894bc8c4SAl Viro } 627894bc8c4SAl Viro 62860ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 629894bc8c4SAl Viro { 630bc40aee0SAl Viro struct saved *p; 631bc40aee0SAl Viro 6326da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 63360ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 634bc40aee0SAl Viro if (unlikely(!p)) 63560ef60c7SAl Viro return false; 636894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 637894bc8c4SAl Viro nd->stack = p; 63860ef60c7SAl Viro return true; 639894bc8c4SAl Viro } 640894bc8c4SAl Viro 641397d425dSEric W. Biederman /** 6426b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 643397d425dSEric W. Biederman * 644397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 645397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 646397d425dSEric W. Biederman */ 6476b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 648397d425dSEric W. Biederman { 64995dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 650397d425dSEric W. Biederman 651402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 652402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 653397d425dSEric W. Biederman return true; 654397d425dSEric W. Biederman 6556b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 656397d425dSEric W. Biederman } 657397d425dSEric W. Biederman 6587973387aSAl Viro static void drop_links(struct nameidata *nd) 6597973387aSAl Viro { 6607973387aSAl Viro int i = nd->depth; 6617973387aSAl Viro while (i--) { 6627973387aSAl Viro struct saved *last = nd->stack + i; 663fceef393SAl Viro do_delayed_call(&last->done); 664fceef393SAl Viro clear_delayed_call(&last->done); 6657973387aSAl Viro } 6667973387aSAl Viro } 6677973387aSAl Viro 6687973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6697973387aSAl Viro { 6707973387aSAl Viro drop_links(nd); 6717973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6727973387aSAl Viro int i; 6737973387aSAl Viro path_put(&nd->path); 6747973387aSAl Viro for (i = 0; i < nd->depth; i++) 6757973387aSAl Viro path_put(&nd->stack[i].link); 676bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 677102b8af2SAl Viro path_put(&nd->root); 678bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 679102b8af2SAl Viro } 6807973387aSAl Viro } else { 6817973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6827973387aSAl Viro rcu_read_unlock(); 6837973387aSAl Viro } 6847973387aSAl Viro nd->depth = 0; 6857d01ef75SAl Viro nd->path.mnt = NULL; 6867d01ef75SAl Viro nd->path.dentry = NULL; 6877973387aSAl Viro } 6887973387aSAl Viro 6897973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6902aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6917973387aSAl Viro { 6922aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6937973387aSAl Viro if (unlikely(res)) { 6947973387aSAl Viro if (res > 0) 6957973387aSAl Viro path->mnt = NULL; 6967973387aSAl Viro path->dentry = NULL; 6977973387aSAl Viro return false; 6987973387aSAl Viro } 6997973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 7007973387aSAl Viro path->dentry = NULL; 7017973387aSAl Viro return false; 7027973387aSAl Viro } 7037973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7047973387aSAl Viro } 7057973387aSAl Viro 7062aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7072aa38470SAl Viro struct path *path, unsigned seq) 7082aa38470SAl Viro { 7095bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7102aa38470SAl Viro } 7112aa38470SAl Viro 7127973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7137973387aSAl Viro { 7147973387aSAl Viro int i; 715eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 716eacd9aa8SAl Viro drop_links(nd); 717eacd9aa8SAl Viro nd->depth = 0; 718eacd9aa8SAl Viro return false; 719eacd9aa8SAl Viro } 7207973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7217973387aSAl Viro struct saved *last = nd->stack + i; 7227973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7237973387aSAl Viro drop_links(nd); 7247973387aSAl Viro nd->depth = i + 1; 7257973387aSAl Viro return false; 7267973387aSAl Viro } 7277973387aSAl Viro } 7287973387aSAl Viro return true; 7297973387aSAl Viro } 7307973387aSAl Viro 731ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 732ee594bffSAl Viro { 733adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 734bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 735ee594bffSAl Viro return true; 736bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 737ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 738ee594bffSAl Viro } 739ee594bffSAl Viro 74019660af7SAl Viro /* 74131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 74219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 74319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 74457e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 74519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 74619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 74719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 74819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 74931e6b01fSNick Piggin */ 75031e6b01fSNick Piggin 75131e6b01fSNick Piggin /** 752e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 75319660af7SAl Viro * @nd: nameidata pathwalk data 754e36cffedSJens Axboe * Returns: true on success, false on failure 75531e6b01fSNick Piggin * 756e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7574675ac39SAl Viro * for ref-walk mode. 7584675ac39SAl Viro * Must be called from rcu-walk context. 759e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7607973387aSAl Viro * terminate_walk(). 76131e6b01fSNick Piggin */ 762e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 76331e6b01fSNick Piggin { 76431e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 76531e6b01fSNick Piggin 76631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 767e5c832d5SLinus Torvalds 768e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 7697973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7704675ac39SAl Viro goto out1; 77184a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 77284a2bd39SAl Viro goto out; 773ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7744675ac39SAl Viro goto out; 7754675ac39SAl Viro rcu_read_unlock(); 7764675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 777e36cffedSJens Axboe return true; 7784675ac39SAl Viro 77984a2bd39SAl Viro out1: 7804675ac39SAl Viro nd->path.mnt = NULL; 7814675ac39SAl Viro nd->path.dentry = NULL; 7824675ac39SAl Viro out: 7834675ac39SAl Viro rcu_read_unlock(); 784e36cffedSJens Axboe return false; 7854675ac39SAl Viro } 7864675ac39SAl Viro 7874675ac39SAl Viro /** 788ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7894675ac39SAl Viro * @nd: nameidata pathwalk data 790ae66db45SAl Viro * @dentry: next dentry to step into 791ae66db45SAl Viro * @seq: seq number to check @dentry against 792ae66db45SAl Viro * Returns: true on success, false on failure 7934675ac39SAl Viro * 79430476f7eSTom Rix * Similar to try_to_unlazy(), but here we have the next dentry already 795ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 796ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 797ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 7984675ac39SAl Viro * terminate_walk(). 7994675ac39SAl Viro */ 800ae66db45SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq) 8014675ac39SAl Viro { 8024675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8034675ac39SAl Viro 8044675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 8054675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8064675ac39SAl Viro goto out2; 8077973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 8087973387aSAl Viro goto out2; 8094675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8107973387aSAl Viro goto out1; 81148a066e7SAl Viro 81215570086SLinus Torvalds /* 8134675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8144675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8154675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8164675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8174675ac39SAl Viro * know the parent must still be valid if the child sequence number is 81815570086SLinus Torvalds */ 8194675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 820e5c832d5SLinus Torvalds goto out; 82184a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 82284a2bd39SAl Viro goto out_dput; 823e5c832d5SLinus Torvalds /* 824e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 825e5c832d5SLinus Torvalds * still valid and get it if required. 826e5c832d5SLinus Torvalds */ 82784a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 82884a2bd39SAl Viro goto out_dput; 8298b61e74fSAl Viro rcu_read_unlock(); 830ae66db45SAl Viro return true; 83119660af7SAl Viro 8327973387aSAl Viro out2: 8337973387aSAl Viro nd->path.mnt = NULL; 8347973387aSAl Viro out1: 8357973387aSAl Viro nd->path.dentry = NULL; 836e5c832d5SLinus Torvalds out: 8378b61e74fSAl Viro rcu_read_unlock(); 838ae66db45SAl Viro return false; 83984a2bd39SAl Viro out_dput: 84084a2bd39SAl Viro rcu_read_unlock(); 84184a2bd39SAl Viro dput(dentry); 842ae66db45SAl Viro return false; 84331e6b01fSNick Piggin } 84431e6b01fSNick Piggin 8454ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 84634286d66SNick Piggin { 847a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8484ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 849a89f8337SAl Viro else 850a89f8337SAl Viro return 1; 85134286d66SNick Piggin } 85234286d66SNick Piggin 8539f1fafeeSAl Viro /** 8549f1fafeeSAl Viro * complete_walk - successful completion of path walk 8559f1fafeeSAl Viro * @nd: pointer nameidata 85639159de2SJeff Layton * 8579f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8589f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8599f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8609f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8619f1fafeeSAl Viro * need to drop nd->path. 86239159de2SJeff Layton */ 8639f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 86439159de2SJeff Layton { 86516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 86639159de2SJeff Layton int status; 86739159de2SJeff Layton 8689f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 869adb21d2bSAleksa Sarai /* 870adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 871adb21d2bSAleksa Sarai * externally-managed nd->root. 872adb21d2bSAleksa Sarai */ 873bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 874bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8759f1fafeeSAl Viro nd->root.mnt = NULL; 8766c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 877e36cffedSJens Axboe if (!try_to_unlazy(nd)) 87848a066e7SAl Viro return -ECHILD; 87948a066e7SAl Viro } 8809f1fafeeSAl Viro 881adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 882adb21d2bSAleksa Sarai /* 883adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 884adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 885adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 886adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 887adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 888adb21d2bSAleksa Sarai * 889adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 890adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 891adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 892adb21d2bSAleksa Sarai * requested root to userspace. 893adb21d2bSAleksa Sarai * 894adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 895adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 896adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 897adb21d2bSAleksa Sarai */ 898adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 899adb21d2bSAleksa Sarai return -EXDEV; 900adb21d2bSAleksa Sarai } 901adb21d2bSAleksa Sarai 902bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 90339159de2SJeff Layton return 0; 90439159de2SJeff Layton 905ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 90616c2cd71SAl Viro return 0; 90716c2cd71SAl Viro 908ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 90939159de2SJeff Layton if (status > 0) 91039159de2SJeff Layton return 0; 91139159de2SJeff Layton 91216c2cd71SAl Viro if (!status) 91339159de2SJeff Layton status = -ESTALE; 91416c2cd71SAl Viro 91539159de2SJeff Layton return status; 91639159de2SJeff Layton } 91739159de2SJeff Layton 918740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9192a737871SAl Viro { 92031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9219e6697e2SAl Viro 922adb21d2bSAleksa Sarai /* 923adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 924adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 925adb21d2bSAleksa Sarai * from the dirfd. 926adb21d2bSAleksa Sarai */ 927adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 928adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 929adb21d2bSAleksa Sarai 9309e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9318f47a016SAl Viro unsigned seq; 932c28cc364SNick Piggin 933c28cc364SNick Piggin do { 934c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 93531e6b01fSNick Piggin nd->root = fs->root; 9368f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 937c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9389e6697e2SAl Viro } else { 9399e6697e2SAl Viro get_fs_root(fs, &nd->root); 940bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9419e6697e2SAl Viro } 942740a1678SAleksa Sarai return 0; 94331e6b01fSNick Piggin } 94431e6b01fSNick Piggin 945248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 946248fb5b9SAl Viro { 947adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 948adb21d2bSAleksa Sarai return -EXDEV; 94972ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 95072ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 95172ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 95272ba2929SAleksa Sarai return -EXDEV; 95372ba2929SAleksa Sarai } 954740a1678SAleksa Sarai if (!nd->root.mnt) { 955740a1678SAleksa Sarai int error = set_root(nd); 956740a1678SAleksa Sarai if (error) 957740a1678SAleksa Sarai return error; 958740a1678SAleksa Sarai } 959248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 960248fb5b9SAl Viro struct dentry *d; 961248fb5b9SAl Viro nd->path = nd->root; 962248fb5b9SAl Viro d = nd->path.dentry; 963248fb5b9SAl Viro nd->inode = d->d_inode; 964248fb5b9SAl Viro nd->seq = nd->root_seq; 965248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 966248fb5b9SAl Viro return -ECHILD; 967248fb5b9SAl Viro } else { 968248fb5b9SAl Viro path_put(&nd->path); 969248fb5b9SAl Viro nd->path = nd->root; 970248fb5b9SAl Viro path_get(&nd->path); 971248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 972248fb5b9SAl Viro } 973bcba1e7dSAl Viro nd->state |= ND_JUMPED; 974248fb5b9SAl Viro return 0; 975248fb5b9SAl Viro } 976248fb5b9SAl Viro 977b5fb63c1SChristoph Hellwig /* 9786b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 979b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 980b5fb63c1SChristoph Hellwig */ 9811bc82070SAleksa Sarai int nd_jump_link(struct path *path) 982b5fb63c1SChristoph Hellwig { 9834b99d499SAleksa Sarai int error = -ELOOP; 9846e77137bSAl Viro struct nameidata *nd = current->nameidata; 985b5fb63c1SChristoph Hellwig 9864b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9874b99d499SAleksa Sarai goto err; 9884b99d499SAleksa Sarai 98972ba2929SAleksa Sarai error = -EXDEV; 99072ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 99172ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 99272ba2929SAleksa Sarai goto err; 99372ba2929SAleksa Sarai } 994adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 995adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 996adb21d2bSAleksa Sarai goto err; 99772ba2929SAleksa Sarai 9984b99d499SAleksa Sarai path_put(&nd->path); 999b5fb63c1SChristoph Hellwig nd->path = *path; 1000b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1001bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10021bc82070SAleksa Sarai return 0; 10034b99d499SAleksa Sarai 10044b99d499SAleksa Sarai err: 10054b99d499SAleksa Sarai path_put(path); 10064b99d499SAleksa Sarai return error; 1007b5fb63c1SChristoph Hellwig } 1008b5fb63c1SChristoph Hellwig 1009b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1010574197e0SAl Viro { 101121c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1012fceef393SAl Viro do_delayed_call(&last->done); 10136548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1014b9ff4429SAl Viro path_put(&last->link); 1015574197e0SAl Viro } 1016574197e0SAl Viro 10179c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10189c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10199c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10209c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10219c011be1SLuis Chamberlain 10229c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10239c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10249c011be1SLuis Chamberlain { 10259c011be1SLuis Chamberlain .procname = "protected_symlinks", 10269c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10279c011be1SLuis Chamberlain .maxlen = sizeof(int), 1028c7031c14SJulius Hemanth Pitti .mode = 0644, 10299c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10309c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10319c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10329c011be1SLuis Chamberlain }, 10339c011be1SLuis Chamberlain { 10349c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10359c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10369c011be1SLuis Chamberlain .maxlen = sizeof(int), 1037c7031c14SJulius Hemanth Pitti .mode = 0644, 10389c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10399c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10409c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10419c011be1SLuis Chamberlain }, 10429c011be1SLuis Chamberlain { 10439c011be1SLuis Chamberlain .procname = "protected_fifos", 10449c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10459c011be1SLuis Chamberlain .maxlen = sizeof(int), 1046c7031c14SJulius Hemanth Pitti .mode = 0644, 10479c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10489c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10499c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10509c011be1SLuis Chamberlain }, 10519c011be1SLuis Chamberlain { 10529c011be1SLuis Chamberlain .procname = "protected_regular", 10539c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10549c011be1SLuis Chamberlain .maxlen = sizeof(int), 1055c7031c14SJulius Hemanth Pitti .mode = 0644, 10569c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10579c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10589c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10599c011be1SLuis Chamberlain }, 10609c011be1SLuis Chamberlain { } 10619c011be1SLuis Chamberlain }; 10629c011be1SLuis Chamberlain 10639c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10649c011be1SLuis Chamberlain { 10659c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10669c011be1SLuis Chamberlain return 0; 10679c011be1SLuis Chamberlain } 10689c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10699c011be1SLuis Chamberlain 10709c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1071800179c9SKees Cook 1072800179c9SKees Cook /** 1073800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 107455852635SRandy Dunlap * @nd: nameidata pathwalk data 1075800179c9SKees Cook * 1076800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1077800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1078800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1079800179c9SKees Cook * processes from failing races against path names that may change out 1080800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1081800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1082800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1083800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1084800179c9SKees Cook * 1085800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1086800179c9SKees Cook */ 1087ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1088800179c9SKees Cook { 1089ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1090ba73d987SChristian Brauner kuid_t i_uid; 1091ba73d987SChristian Brauner 1092800179c9SKees Cook if (!sysctl_protected_symlinks) 1093800179c9SKees Cook return 0; 1094800179c9SKees Cook 1095ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1096ba73d987SChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 1097800179c9SKees Cook /* Allowed if owner and follower match. */ 1098ba73d987SChristian Brauner if (uid_eq(current_cred()->fsuid, i_uid)) 1099800179c9SKees Cook return 0; 1100800179c9SKees Cook 1101800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11020f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1103800179c9SKees Cook return 0; 1104800179c9SKees Cook 1105800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1106ba73d987SChristian Brauner if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) 1107800179c9SKees Cook return 0; 1108800179c9SKees Cook 110931956502SAl Viro if (nd->flags & LOOKUP_RCU) 111031956502SAl Viro return -ECHILD; 111131956502SAl Viro 1112ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1113245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1114800179c9SKees Cook return -EACCES; 1115800179c9SKees Cook } 1116800179c9SKees Cook 1117800179c9SKees Cook /** 1118800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1119ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1120800179c9SKees Cook * @inode: the source inode to hardlink from 1121800179c9SKees Cook * 1122800179c9SKees Cook * Return false if at least one of the following conditions: 1123800179c9SKees Cook * - inode is not a regular file 1124800179c9SKees Cook * - inode is setuid 1125800179c9SKees Cook * - inode is setgid and group-exec 1126800179c9SKees Cook * - access failure for read and write 1127800179c9SKees Cook * 1128800179c9SKees Cook * Otherwise returns true. 1129800179c9SKees Cook */ 1130ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1131ba73d987SChristian Brauner struct inode *inode) 1132800179c9SKees Cook { 1133800179c9SKees Cook umode_t mode = inode->i_mode; 1134800179c9SKees Cook 1135800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1136800179c9SKees Cook if (!S_ISREG(mode)) 1137800179c9SKees Cook return false; 1138800179c9SKees Cook 1139800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1140800179c9SKees Cook if (mode & S_ISUID) 1141800179c9SKees Cook return false; 1142800179c9SKees Cook 1143800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1144800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1145800179c9SKees Cook return false; 1146800179c9SKees Cook 1147800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1148ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1149800179c9SKees Cook return false; 1150800179c9SKees Cook 1151800179c9SKees Cook return true; 1152800179c9SKees Cook } 1153800179c9SKees Cook 1154800179c9SKees Cook /** 1155800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1156ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1157800179c9SKees Cook * @link: the source to hardlink from 1158800179c9SKees Cook * 1159800179c9SKees Cook * Block hardlink when all of: 1160800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1161800179c9SKees Cook * - fsuid does not match inode 1162800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1163f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1164800179c9SKees Cook * 1165ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1166ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1167ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1168ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1169ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1170ba73d987SChristian Brauner * 1171800179c9SKees Cook * Returns 0 if successful, -ve on error. 1172800179c9SKees Cook */ 1173ba73d987SChristian Brauner int may_linkat(struct user_namespace *mnt_userns, struct path *link) 1174800179c9SKees Cook { 1175593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1176593d1ce8SEric W. Biederman 1177593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1178ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 1179ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 1180593d1ce8SEric W. Biederman return -EOVERFLOW; 1181800179c9SKees Cook 1182800179c9SKees Cook if (!sysctl_protected_hardlinks) 1183800179c9SKees Cook return 0; 1184800179c9SKees Cook 1185800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1186800179c9SKees Cook * otherwise, it must be a safe source. 1187800179c9SKees Cook */ 1188ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1189ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1190800179c9SKees Cook return 0; 1191800179c9SKees Cook 1192245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1193800179c9SKees Cook return -EPERM; 1194800179c9SKees Cook } 1195800179c9SKees Cook 119630aba665SSalvatore Mesoraca /** 119730aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 119830aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 119930aba665SSalvatore Mesoraca * exist. 1200ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 12012111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 120230aba665SSalvatore Mesoraca * @inode: the inode of the file to open 120330aba665SSalvatore Mesoraca * 120430aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 120530aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 120630aba665SSalvatore Mesoraca * - the file already exists 120730aba665SSalvatore Mesoraca * - we are in a sticky directory 120830aba665SSalvatore Mesoraca * - we don't own the file 120930aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 121030aba665SSalvatore Mesoraca * - the directory is world writable 121130aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 121230aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 121330aba665SSalvatore Mesoraca * be enough. 121430aba665SSalvatore Mesoraca * 1215ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1216ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1217ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1218ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1219ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1220ba73d987SChristian Brauner * 122130aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 122230aba665SSalvatore Mesoraca */ 1223ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1224ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 122530aba665SSalvatore Mesoraca { 1226ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1227ba73d987SChristian Brauner kuid_t dir_uid = nd->dir_uid; 1228ba73d987SChristian Brauner 122930aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 123030aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1231d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1232ba73d987SChristian Brauner uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || 1233ba73d987SChristian Brauner uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) 123430aba665SSalvatore Mesoraca return 0; 123530aba665SSalvatore Mesoraca 1236d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1237d0cb5018SAl Viro (dir_mode & 0020 && 123830aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 123930aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1240245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1241245d7369SKees Cook "sticky_create_fifo" : 1242245d7369SKees Cook "sticky_create_regular"; 1243245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 124430aba665SSalvatore Mesoraca return -EACCES; 124530aba665SSalvatore Mesoraca } 124630aba665SSalvatore Mesoraca return 0; 124730aba665SSalvatore Mesoraca } 124830aba665SSalvatore Mesoraca 1249f015f126SDavid Howells /* 1250f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1251f015f126SDavid Howells * 1252f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1253f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1254f015f126SDavid Howells * Up is towards /. 1255f015f126SDavid Howells * 1256f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1257f015f126SDavid Howells * root. 1258f015f126SDavid Howells */ 1259bab77ebfSAl Viro int follow_up(struct path *path) 12601da177e4SLinus Torvalds { 12610714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12620714a533SAl Viro struct mount *parent; 12631da177e4SLinus Torvalds struct dentry *mountpoint; 126499b7db7bSNick Piggin 126548a066e7SAl Viro read_seqlock_excl(&mount_lock); 12660714a533SAl Viro parent = mnt->mnt_parent; 12673c0a6163SAl Viro if (parent == mnt) { 126848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12691da177e4SLinus Torvalds return 0; 12701da177e4SLinus Torvalds } 12710714a533SAl Viro mntget(&parent->mnt); 1272a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 127348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1274bab77ebfSAl Viro dput(path->dentry); 1275bab77ebfSAl Viro path->dentry = mountpoint; 1276bab77ebfSAl Viro mntput(path->mnt); 12770714a533SAl Viro path->mnt = &parent->mnt; 12781da177e4SLinus Torvalds return 1; 12791da177e4SLinus Torvalds } 12804d359507SAl Viro EXPORT_SYMBOL(follow_up); 12811da177e4SLinus Torvalds 12827ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12837ef482faSAl Viro struct path *path, unsigned *seqp) 12847ef482faSAl Viro { 12857ef482faSAl Viro while (mnt_has_parent(m)) { 12867ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12877ef482faSAl Viro 12887ef482faSAl Viro m = m->mnt_parent; 12897ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12907ef482faSAl Viro root->mnt == &m->mnt)) 12917ef482faSAl Viro break; 12927ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 12937ef482faSAl Viro path->mnt = &m->mnt; 12947ef482faSAl Viro path->dentry = mountpoint; 12957ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 12967ef482faSAl Viro return true; 12977ef482faSAl Viro } 12987ef482faSAl Viro } 12997ef482faSAl Viro return false; 13007ef482faSAl Viro } 13017ef482faSAl Viro 13022aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 13032aa38470SAl Viro struct path *path) 13042aa38470SAl Viro { 13052aa38470SAl Viro bool found; 13062aa38470SAl Viro 13072aa38470SAl Viro rcu_read_lock(); 13082aa38470SAl Viro while (1) { 13092aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 13102aa38470SAl Viro 13112aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 13122aa38470SAl Viro if (unlikely(!found)) { 13132aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 13142aa38470SAl Viro break; 13152aa38470SAl Viro } else { 13162aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 13172aa38470SAl Viro break; 13182aa38470SAl Viro rcu_read_unlock(); 13192aa38470SAl Viro path_put(path); 13202aa38470SAl Viro rcu_read_lock(); 13212aa38470SAl Viro } 13222aa38470SAl Viro } 13232aa38470SAl Viro rcu_read_unlock(); 13242aa38470SAl Viro return found; 13252aa38470SAl Viro } 13262aa38470SAl Viro 1327b5c84bf6SNick Piggin /* 13289875cf80SDavid Howells * Perform an automount 13299875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 13309875cf80SDavid Howells * were called with. 13311da177e4SLinus Torvalds */ 13321c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 133331e6b01fSNick Piggin { 133425e195aaSAl Viro struct dentry *dentry = path->dentry; 13359875cf80SDavid Howells 13360ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 13370ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 13380ec26fd0SMiklos Szeredi * the name. 13390ec26fd0SMiklos Szeredi * 13400ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 13415a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 13420ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 13430ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13440ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13450ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13465a30d8a2SDavid Howells */ 13471c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13485d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 134925e195aaSAl Viro dentry->d_inode) 13509875cf80SDavid Howells return -EISDIR; 13510ec26fd0SMiklos Szeredi 13521c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13539875cf80SDavid Howells return -ELOOP; 13549875cf80SDavid Howells 135525e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1356ea5b778aSDavid Howells } 13579875cf80SDavid Howells 13589875cf80SDavid Howells /* 13599deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13609deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13619deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13629deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13639875cf80SDavid Howells */ 13649deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13659deed3ebSAl Viro int *count, unsigned lookup_flags) 13669875cf80SDavid Howells { 13679deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13689875cf80SDavid Howells bool need_mntput = false; 13698aef1884SAl Viro int ret = 0; 13709875cf80SDavid Howells 13719deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1372cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1373cc53ce53SDavid Howells * being held. */ 1374d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1375fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1376508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1377cc53ce53SDavid Howells if (ret < 0) 13788aef1884SAl Viro break; 1379cc53ce53SDavid Howells } 1380cc53ce53SDavid Howells 13819deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13829875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13839deed3ebSAl Viro if (mounted) { // ... in our namespace 13849875cf80SDavid Howells dput(path->dentry); 13859875cf80SDavid Howells if (need_mntput) 1386463ffb2eSAl Viro mntput(path->mnt); 1387463ffb2eSAl Viro path->mnt = mounted; 1388463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13899deed3ebSAl Viro // here we know it's positive 13909deed3ebSAl Viro flags = path->dentry->d_flags; 13919875cf80SDavid Howells need_mntput = true; 13929875cf80SDavid Howells continue; 1393463ffb2eSAl Viro } 13941da177e4SLinus Torvalds } 13959875cf80SDavid Howells 13969deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 13979deed3ebSAl Viro break; 13989deed3ebSAl Viro 13999deed3ebSAl Viro // uncovered automount point 14009deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 14019deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 14029875cf80SDavid Howells if (ret < 0) 14038aef1884SAl Viro break; 14049875cf80SDavid Howells } 14059875cf80SDavid Howells 14069deed3ebSAl Viro if (ret == -EISDIR) 14079deed3ebSAl Viro ret = 0; 14089deed3ebSAl Viro // possible if you race with several mount --move 14099deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 14108aef1884SAl Viro mntput(path->mnt); 14119deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1412d41efb52SAl Viro ret = -ENOENT; 14139deed3ebSAl Viro *jumped = need_mntput; 14148402752eSAl Viro return ret; 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds 14179deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 14189deed3ebSAl Viro int *count, unsigned lookup_flags) 14199deed3ebSAl Viro { 14209deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 14219deed3ebSAl Viro 14229deed3ebSAl Viro /* fastpath */ 14239deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 14249deed3ebSAl Viro *jumped = false; 14259deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 14269deed3ebSAl Viro return -ENOENT; 14279deed3ebSAl Viro return 0; 14289deed3ebSAl Viro } 14299deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 14309deed3ebSAl Viro } 14319deed3ebSAl Viro 1432cc53ce53SDavid Howells int follow_down_one(struct path *path) 14331da177e4SLinus Torvalds { 14341da177e4SLinus Torvalds struct vfsmount *mounted; 14351da177e4SLinus Torvalds 14361c755af4SAl Viro mounted = lookup_mnt(path); 14371da177e4SLinus Torvalds if (mounted) { 14389393bd07SAl Viro dput(path->dentry); 14399393bd07SAl Viro mntput(path->mnt); 14409393bd07SAl Viro path->mnt = mounted; 14419393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 14421da177e4SLinus Torvalds return 1; 14431da177e4SLinus Torvalds } 14441da177e4SLinus Torvalds return 0; 14451da177e4SLinus Torvalds } 14464d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14471da177e4SLinus Torvalds 14489875cf80SDavid Howells /* 14499deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14509deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14519deed3ebSAl Viro * caller is permitted to proceed or not. 14529deed3ebSAl Viro */ 14539deed3ebSAl Viro int follow_down(struct path *path) 14549deed3ebSAl Viro { 14559deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14569deed3ebSAl Viro bool jumped; 14579deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 14589deed3ebSAl Viro 14599deed3ebSAl Viro if (path->mnt != mnt) 14609deed3ebSAl Viro mntput(mnt); 14619deed3ebSAl Viro return ret; 14629deed3ebSAl Viro } 14639deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14649deed3ebSAl Viro 14659deed3ebSAl Viro /* 1466287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1467287548e4SAl Viro * we meet a managed dentry that would need blocking. 14689875cf80SDavid Howells */ 14699875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1470254cf582SAl Viro struct inode **inode, unsigned *seqp) 14719875cf80SDavid Howells { 1472ea936aebSAl Viro struct dentry *dentry = path->dentry; 1473ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1474ea936aebSAl Viro 1475ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1476ea936aebSAl Viro return true; 1477ea936aebSAl Viro 1478ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1479ea936aebSAl Viro return false; 1480ea936aebSAl Viro 148162a7375eSIan Kent for (;;) { 148262a7375eSIan Kent /* 148362a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 148462a7375eSIan Kent * that wants to block transit. 148562a7375eSIan Kent */ 1486ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1487ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1488ea936aebSAl Viro if (res) 1489ea936aebSAl Viro return res == -EISDIR; 1490ea936aebSAl Viro flags = dentry->d_flags; 1491b8faf035SNeilBrown } 149262a7375eSIan Kent 1493ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1494ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1495ea936aebSAl Viro if (mounted) { 1496c7105365SAl Viro path->mnt = &mounted->mnt; 1497ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1498bcba1e7dSAl Viro nd->state |= ND_JUMPED; 1499ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1500ea936aebSAl Viro *inode = dentry->d_inode; 150159430262SLinus Torvalds /* 1502ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1503ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1504ea936aebSAl Viro * between mount hash removal and ->mnt_root 1505ea936aebSAl Viro * becoming unpinned. 150659430262SLinus Torvalds */ 1507ea936aebSAl Viro flags = dentry->d_flags; 1508ea936aebSAl Viro continue; 15099875cf80SDavid Howells } 1510ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1511ea936aebSAl Viro return false; 1512ea936aebSAl Viro } 1513ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1514ea936aebSAl Viro } 1515287548e4SAl Viro } 1516287548e4SAl Viro 1517db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1518db3c9adeSAl Viro struct path *path, struct inode **inode, 1519db3c9adeSAl Viro unsigned int *seqp) 1520bd7c4b50SAl Viro { 15219deed3ebSAl Viro bool jumped; 1522db3c9adeSAl Viro int ret; 1523bd7c4b50SAl Viro 1524db3c9adeSAl Viro path->mnt = nd->path.mnt; 1525db3c9adeSAl Viro path->dentry = dentry; 1526c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1527c153007bSAl Viro unsigned int seq = *seqp; 1528c153007bSAl Viro if (unlikely(!*inode)) 1529c153007bSAl Viro return -ENOENT; 1530c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15319deed3ebSAl Viro return 0; 1532ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 1533c153007bSAl Viro return -ECHILD; 1534c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1535c153007bSAl Viro path->mnt = nd->path.mnt; 1536c153007bSAl Viro path->dentry = dentry; 1537c153007bSAl Viro } 15389deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 15399deed3ebSAl Viro if (jumped) { 15409deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 15419deed3ebSAl Viro ret = -EXDEV; 15429deed3ebSAl Viro else 1543bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15449deed3ebSAl Viro } 15459deed3ebSAl Viro if (unlikely(ret)) { 15469deed3ebSAl Viro dput(path->dentry); 15479deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15489deed3ebSAl Viro mntput(path->mnt); 15499deed3ebSAl Viro } else { 1550bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1551bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1552bd7c4b50SAl Viro } 1553bd7c4b50SAl Viro return ret; 1554bd7c4b50SAl Viro } 1555bd7c4b50SAl Viro 15569875cf80SDavid Howells /* 1557f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1558f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1559baa03890SNick Piggin */ 1560e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1561e3c13928SAl Viro struct dentry *dir, 15626c51e513SAl Viro unsigned int flags) 1563baa03890SNick Piggin { 1564a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1565bad61189SMiklos Szeredi if (dentry) { 1566a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1567bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 156874ff0ffcSAl Viro if (!error) 15695542aa2fSEric W. Biederman d_invalidate(dentry); 1570bad61189SMiklos Szeredi dput(dentry); 157174ff0ffcSAl Viro return ERR_PTR(error); 1572bad61189SMiklos Szeredi } 1573bad61189SMiklos Szeredi } 1574baa03890SNick Piggin return dentry; 1575baa03890SNick Piggin } 1576baa03890SNick Piggin 1577baa03890SNick Piggin /* 1578a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1579a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1580a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1581a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1582a03ece5fSAl Viro * at all. 158344396f4bSJosef Bacik */ 1584a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1585a03ece5fSAl Viro struct dentry *base, unsigned int flags) 158644396f4bSJosef Bacik { 1587a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 158844396f4bSJosef Bacik struct dentry *old; 1589a03ece5fSAl Viro struct inode *dir = base->d_inode; 1590a03ece5fSAl Viro 1591a03ece5fSAl Viro if (dentry) 1592a03ece5fSAl Viro return dentry; 159344396f4bSJosef Bacik 159444396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1595a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 159644396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1597a03ece5fSAl Viro 1598a03ece5fSAl Viro dentry = d_alloc(base, name); 1599a03ece5fSAl Viro if (unlikely(!dentry)) 1600a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 160144396f4bSJosef Bacik 160272bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 160344396f4bSJosef Bacik if (unlikely(old)) { 160444396f4bSJosef Bacik dput(dentry); 160544396f4bSJosef Bacik dentry = old; 160644396f4bSJosef Bacik } 160744396f4bSJosef Bacik return dentry; 160844396f4bSJosef Bacik } 160944396f4bSJosef Bacik 161020e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 161120e34357SAl Viro struct inode **inode, 1612254cf582SAl Viro unsigned *seqp) 16131da177e4SLinus Torvalds { 161431e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16155a18fff2SAl Viro int status = 1; 16169875cf80SDavid Howells 16173cac260aSAl Viro /* 1618b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16195d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16205d0f49c1SAl Viro * going to fall back to non-racy lookup. 1621b04f784eSNick Piggin */ 162231e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 162331e6b01fSNick Piggin unsigned seq; 1624da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 16255d0f49c1SAl Viro if (unlikely(!dentry)) { 1626e36cffedSJens Axboe if (!try_to_unlazy(nd)) 162720e34357SAl Viro return ERR_PTR(-ECHILD); 162820e34357SAl Viro return NULL; 16295d0f49c1SAl Viro } 16305a18fff2SAl Viro 163112f8ad4bSLinus Torvalds /* 163212f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 163312f8ad4bSLinus Torvalds * the dentry name information from lookup. 163412f8ad4bSLinus Torvalds */ 163563afdfc7SDavid Howells *inode = d_backing_inode(dentry); 16365d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 163720e34357SAl Viro return ERR_PTR(-ECHILD); 163812f8ad4bSLinus Torvalds 163912f8ad4bSLinus Torvalds /* 164012f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 164112f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 164212f8ad4bSLinus Torvalds * 164312f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 164412f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 164512f8ad4bSLinus Torvalds */ 16465d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 164720e34357SAl Viro return ERR_PTR(-ECHILD); 16485a18fff2SAl Viro 1649254cf582SAl Viro *seqp = seq; 16504ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1651c153007bSAl Viro if (likely(status > 0)) 165220e34357SAl Viro return dentry; 1653ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 165420e34357SAl Viro return ERR_PTR(-ECHILD); 165526ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1656209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1657209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16585a18fff2SAl Viro } else { 1659e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 166081e6f520SAl Viro if (unlikely(!dentry)) 166120e34357SAl Viro return NULL; 16624ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16635d0f49c1SAl Viro } 16645a18fff2SAl Viro if (unlikely(status <= 0)) { 1665e9742b53SAl Viro if (!status) 16665d0f49c1SAl Viro d_invalidate(dentry); 16675a18fff2SAl Viro dput(dentry); 166820e34357SAl Viro return ERR_PTR(status); 16695a18fff2SAl Viro } 167020e34357SAl Viro return dentry; 1671697f514dSMiklos Szeredi } 1672697f514dSMiklos Szeredi 1673697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 167488d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1675e3c13928SAl Viro struct dentry *dir, 1676e3c13928SAl Viro unsigned int flags) 1677697f514dSMiklos Szeredi { 167888d8331aSAl Viro struct dentry *dentry, *old; 16791936386eSAl Viro struct inode *inode = dir->d_inode; 1680d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16811936386eSAl Viro 16821936386eSAl Viro /* Don't go there if it's already dead */ 168394bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 168488d8331aSAl Viro return ERR_PTR(-ENOENT); 168594bdd655SAl Viro again: 1686d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 168794bdd655SAl Viro if (IS_ERR(dentry)) 168888d8331aSAl Viro return dentry; 168994bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1690949a852eSAl Viro int error = d_revalidate(dentry, flags); 1691949a852eSAl Viro if (unlikely(error <= 0)) { 169294bdd655SAl Viro if (!error) { 1693949a852eSAl Viro d_invalidate(dentry); 1694949a852eSAl Viro dput(dentry); 169594bdd655SAl Viro goto again; 169694bdd655SAl Viro } 169794bdd655SAl Viro dput(dentry); 1698949a852eSAl Viro dentry = ERR_PTR(error); 1699949a852eSAl Viro } 170094bdd655SAl Viro } else { 17011936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 170285c7f810SAl Viro d_lookup_done(dentry); 17031936386eSAl Viro if (unlikely(old)) { 17041936386eSAl Viro dput(dentry); 17051936386eSAl Viro dentry = old; 1706949a852eSAl Viro } 1707949a852eSAl Viro } 1708e3c13928SAl Viro return dentry; 17091da177e4SLinus Torvalds } 17101da177e4SLinus Torvalds 171188d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 171288d8331aSAl Viro struct dentry *dir, 171388d8331aSAl Viro unsigned int flags) 171488d8331aSAl Viro { 171588d8331aSAl Viro struct inode *inode = dir->d_inode; 171688d8331aSAl Viro struct dentry *res; 171788d8331aSAl Viro inode_lock_shared(inode); 171888d8331aSAl Viro res = __lookup_slow(name, dir, flags); 171988d8331aSAl Viro inode_unlock_shared(inode); 172088d8331aSAl Viro return res; 172188d8331aSAl Viro } 172288d8331aSAl Viro 1723ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1724ba73d987SChristian Brauner struct nameidata *nd) 172552094c8aSAl Viro { 172652094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17277d6beb71SLinus Torvalds int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1728e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 172952094c8aSAl Viro return err; 173052094c8aSAl Viro } 1731ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 173252094c8aSAl Viro } 173352094c8aSAl Viro 173449055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1735d63ff28fSAl Viro { 173649055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 173749055906SAl Viro return -ELOOP; 17384542576bSAl Viro 17394542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17404542576bSAl Viro return 0; 17414542576bSAl Viro if (likely(nd->stack != nd->internal)) 17424542576bSAl Viro return 0; 174360ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 174449055906SAl Viro return 0; 174560ef60c7SAl Viro 174660ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 174760ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 174860ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1749aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 175060ef60c7SAl Viro 1751e5ca024eSAl Viro if (!try_to_unlazy(nd) || !grabbed_link) 175260ef60c7SAl Viro return -ECHILD; 175360ef60c7SAl Viro 175460ef60c7SAl Viro if (nd_alloc_stack(nd)) 175560ef60c7SAl Viro return 0; 1756bc40aee0SAl Viro } 175760ef60c7SAl Viro return -ENOMEM; 175849055906SAl Viro } 175949055906SAl Viro 176049055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 176149055906SAl Viro 176249055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 176349055906SAl Viro struct inode *inode, unsigned seq, int flags) 176449055906SAl Viro { 176549055906SAl Viro struct saved *last; 176649055906SAl Viro const char *res; 176749055906SAl Viro int error = reserve_stack(nd, link, seq); 176849055906SAl Viro 176949055906SAl Viro if (unlikely(error)) { 177049055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1771cd179f44SAl Viro path_put(link); 177206708adbSAl Viro return ERR_PTR(error); 1773626de996SAl Viro } 1774ab104923SAl Viro last = nd->stack + nd->depth++; 17751cf2665bSAl Viro last->link = *link; 1776fceef393SAl Viro clear_delayed_call(&last->done); 17770450b2d1SAl Viro last->seq = seq; 1778ad6cc4c3SAl Viro 1779b1a81972SAl Viro if (flags & WALK_TRAILING) { 1780ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1781ad6cc4c3SAl Viro if (unlikely(error)) 1782ad6cc4c3SAl Viro return ERR_PTR(error); 1783ad6cc4c3SAl Viro } 1784ad6cc4c3SAl Viro 1785dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1786dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1787ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1788ad6cc4c3SAl Viro 1789ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1790ad6cc4c3SAl Viro touch_atime(&last->link); 1791ad6cc4c3SAl Viro cond_resched(); 1792ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1793e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1794ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1795ad6cc4c3SAl Viro touch_atime(&last->link); 1796ad6cc4c3SAl Viro } 1797ad6cc4c3SAl Viro 1798ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1799ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1800ad6cc4c3SAl Viro if (unlikely(error)) 1801ad6cc4c3SAl Viro return ERR_PTR(error); 1802ad6cc4c3SAl Viro 1803ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1804ad6cc4c3SAl Viro if (!res) { 1805ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1806ad6cc4c3SAl Viro struct delayed_call *); 1807ad6cc4c3SAl Viro get = inode->i_op->get_link; 1808ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1809ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1810e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1811ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1812ad6cc4c3SAl Viro } else { 1813ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1814ad6cc4c3SAl Viro } 1815ad6cc4c3SAl Viro if (!res) 1816ad6cc4c3SAl Viro goto all_done; 1817ad6cc4c3SAl Viro if (IS_ERR(res)) 1818ad6cc4c3SAl Viro return res; 1819ad6cc4c3SAl Viro } 1820ad6cc4c3SAl Viro if (*res == '/') { 1821ad6cc4c3SAl Viro error = nd_jump_root(nd); 1822ad6cc4c3SAl Viro if (unlikely(error)) 1823ad6cc4c3SAl Viro return ERR_PTR(error); 1824ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1825ad6cc4c3SAl Viro ; 1826ad6cc4c3SAl Viro } 1827ad6cc4c3SAl Viro if (*res) 1828ad6cc4c3SAl Viro return res; 1829ad6cc4c3SAl Viro all_done: // pure jump 1830ad6cc4c3SAl Viro put_link(nd); 1831ad6cc4c3SAl Viro return NULL; 1832d63ff28fSAl Viro } 1833d63ff28fSAl Viro 18343ddcd056SLinus Torvalds /* 18353ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18363ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18373ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18383ddcd056SLinus Torvalds * for the common case. 18393ddcd056SLinus Torvalds */ 1840b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1841cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 18423ddcd056SLinus Torvalds { 1843cbae4d12SAl Viro struct path path; 1844cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1845cbae4d12SAl Viro 1846cbae4d12SAl Viro if (err < 0) 1847b0417d2cSAl Viro return ERR_PTR(err); 1848cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18498c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1850aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18518f64fb1cSAl Viro /* not a symlink or should not follow */ 1852c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1853c99687a0SAl Viro dput(nd->path.dentry); 1854c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1855c99687a0SAl Viro mntput(nd->path.mnt); 1856c99687a0SAl Viro } 1857c99687a0SAl Viro nd->path = path; 18588f64fb1cSAl Viro nd->inode = inode; 18598f64fb1cSAl Viro nd->seq = seq; 1860b0417d2cSAl Viro return NULL; 18618f64fb1cSAl Viro } 1862a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 186384f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1864cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1865b0417d2cSAl Viro return ERR_PTR(-ECHILD); 186684f0cd9eSAl Viro } else { 186784f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 186884f0cd9eSAl Viro mntget(path.mnt); 1869a7f77542SAl Viro } 1870b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18713ddcd056SLinus Torvalds } 18723ddcd056SLinus Torvalds 1873c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1874c2df1968SAl Viro struct inode **inodep, 1875c2df1968SAl Viro unsigned *seqp) 1876957dd41dSAl Viro { 187712487f30SAl Viro struct dentry *parent, *old; 1878957dd41dSAl Viro 187912487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 188012487f30SAl Viro goto in_root; 188112487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18827ef482faSAl Viro struct path path; 1883efe772d6SAl Viro unsigned seq; 18847ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18857ef482faSAl Viro &nd->root, &path, &seq)) 188612487f30SAl Viro goto in_root; 1887efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1888efe772d6SAl Viro return ERR_PTR(-ECHILD); 1889efe772d6SAl Viro nd->path = path; 1890efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1891efe772d6SAl Viro nd->seq = seq; 1892efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1893efe772d6SAl Viro return ERR_PTR(-ECHILD); 1894efe772d6SAl Viro /* we know that mountpoint was pinned */ 1895957dd41dSAl Viro } 189612487f30SAl Viro old = nd->path.dentry; 189712487f30SAl Viro parent = old->d_parent; 189812487f30SAl Viro *inodep = parent->d_inode; 189912487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 190012487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 190112487f30SAl Viro return ERR_PTR(-ECHILD); 190212487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 190312487f30SAl Viro return ERR_PTR(-ECHILD); 190412487f30SAl Viro return parent; 190512487f30SAl Viro in_root: 1906efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1907efe772d6SAl Viro return ERR_PTR(-ECHILD); 1908957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19097521f22bSAl Viro return ERR_PTR(-ECHILD); 1910c2df1968SAl Viro return NULL; 1911957dd41dSAl Viro } 1912957dd41dSAl Viro 1913c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1914c2df1968SAl Viro struct inode **inodep, 1915c2df1968SAl Viro unsigned *seqp) 1916957dd41dSAl Viro { 191712487f30SAl Viro struct dentry *parent; 191812487f30SAl Viro 1919957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 192012487f30SAl Viro goto in_root; 192112487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19222aa38470SAl Viro struct path path; 19232aa38470SAl Viro 19242aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19252aa38470SAl Viro &nd->root, &path)) 192612487f30SAl Viro goto in_root; 1927165200d6SAl Viro path_put(&nd->path); 1928165200d6SAl Viro nd->path = path; 19292aa38470SAl Viro nd->inode = path.dentry->d_inode; 1930165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1931165200d6SAl Viro return ERR_PTR(-EXDEV); 193212487f30SAl Viro } 1933957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 193412487f30SAl Viro parent = dget_parent(nd->path.dentry); 1935957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1936957dd41dSAl Viro dput(parent); 19377521f22bSAl Viro return ERR_PTR(-ENOENT); 1938957dd41dSAl Viro } 1939c2df1968SAl Viro *seqp = 0; 1940c2df1968SAl Viro *inodep = parent->d_inode; 1941c2df1968SAl Viro return parent; 194212487f30SAl Viro 194312487f30SAl Viro in_root: 1944957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19457521f22bSAl Viro return ERR_PTR(-EXDEV); 1946c2df1968SAl Viro dget(nd->path.dentry); 1947c2df1968SAl Viro return NULL; 1948957dd41dSAl Viro } 1949957dd41dSAl Viro 19507521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1951957dd41dSAl Viro { 1952957dd41dSAl Viro if (type == LAST_DOTDOT) { 19537521f22bSAl Viro const char *error = NULL; 1954c2df1968SAl Viro struct dentry *parent; 1955c2df1968SAl Viro struct inode *inode; 1956c2df1968SAl Viro unsigned seq; 1957957dd41dSAl Viro 1958957dd41dSAl Viro if (!nd->root.mnt) { 19597521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1960957dd41dSAl Viro if (error) 1961957dd41dSAl Viro return error; 1962957dd41dSAl Viro } 1963957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1964c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1965957dd41dSAl Viro else 1966c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1967c2df1968SAl Viro if (IS_ERR(parent)) 1968c2df1968SAl Viro return ERR_CAST(parent); 1969c2df1968SAl Viro if (unlikely(!parent)) 1970c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1971c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1972c2df1968SAl Viro else 1973c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1974c2df1968SAl Viro parent, inode, seq); 1975c2df1968SAl Viro if (unlikely(error)) 1976957dd41dSAl Viro return error; 1977957dd41dSAl Viro 1978957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1979957dd41dSAl Viro /* 1980957dd41dSAl Viro * If there was a racing rename or mount along our 1981957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1982957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1983957dd41dSAl Viro * some fallback). 1984957dd41dSAl Viro */ 1985957dd41dSAl Viro smp_rmb(); 1986957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 19877521f22bSAl Viro return ERR_PTR(-EAGAIN); 1988957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 19897521f22bSAl Viro return ERR_PTR(-EAGAIN); 1990957dd41dSAl Viro } 1991957dd41dSAl Viro } 19927521f22bSAl Viro return NULL; 1993957dd41dSAl Viro } 1994957dd41dSAl Viro 199592d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1996ce57dfc1SAl Viro { 1997db3c9adeSAl Viro struct dentry *dentry; 1998ce57dfc1SAl Viro struct inode *inode; 1999254cf582SAl Viro unsigned seq; 2000ce57dfc1SAl Viro /* 2001ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 2002ce57dfc1SAl Viro * to be able to know about the current root directory and 2003ce57dfc1SAl Viro * parent relationships. 2004ce57dfc1SAl Viro */ 20054693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 20061c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 20074693a547SAl Viro put_link(nd); 20087521f22bSAl Viro return handle_dots(nd, nd->last_type); 20094693a547SAl Viro } 201020e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 201120e34357SAl Viro if (IS_ERR(dentry)) 201292d27016SAl Viro return ERR_CAST(dentry); 201320e34357SAl Viro if (unlikely(!dentry)) { 2014db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 2015db3c9adeSAl Viro if (IS_ERR(dentry)) 201692d27016SAl Viro return ERR_CAST(dentry); 201720e34357SAl Viro } 201856676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 201956676ec3SAl Viro put_link(nd); 2020b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 2021ce57dfc1SAl Viro } 2022ce57dfc1SAl Viro 20231da177e4SLinus Torvalds /* 2024bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2025bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2026bfcfaa77SLinus Torvalds * 2027bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2028bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2029bfcfaa77SLinus Torvalds * fast. 2030bfcfaa77SLinus Torvalds * 2031bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2032bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2033bfcfaa77SLinus Torvalds * crossing operation. 2034bfcfaa77SLinus Torvalds * 2035bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2036bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2037bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2038bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2039bfcfaa77SLinus Torvalds */ 2040bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2041bfcfaa77SLinus Torvalds 2042f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2043bfcfaa77SLinus Torvalds 2044468a9428SGeorge Spelvin #ifdef HASH_MIX 2045bfcfaa77SLinus Torvalds 2046468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2047468a9428SGeorge Spelvin 2048468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20492a18da7aSGeorge Spelvin /* 20502a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20512a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20522a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20532a18da7aSGeorge Spelvin * and no temporaries. 20542a18da7aSGeorge Spelvin * 20552a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20562a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20572a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20582a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20592a18da7aSGeorge Spelvin * 20602a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20612a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20622a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20632a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20642a18da7aSGeorge Spelvin * 20652a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20662a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20672a18da7aSGeorge Spelvin * toggling any given output bit. 20682a18da7aSGeorge Spelvin * 20692a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20702a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20712a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20722a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20732a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20742a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20752a18da7aSGeorge Spelvin * Perfect: 8192 258048 20762a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20772a18da7aSGeorge Spelvin */ 20782a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20792a18da7aSGeorge Spelvin ( x ^= (a), \ 20802a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20812a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20822a18da7aSGeorge Spelvin y *= 9 ) 2083bfcfaa77SLinus Torvalds 20840fed3ac8SGeorge Spelvin /* 20852a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20862a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20872a18da7aSGeorge Spelvin * work done before the hash value is used. 20880fed3ac8SGeorge Spelvin */ 20892a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20900fed3ac8SGeorge Spelvin { 20912a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20922a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20932a18da7aSGeorge Spelvin return y >> 32; 20940fed3ac8SGeorge Spelvin } 20950fed3ac8SGeorge Spelvin 2096bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2097bfcfaa77SLinus Torvalds 20982a18da7aSGeorge Spelvin /* 20992a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 21002a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 21012a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 21022a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 21032a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 21042a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 21052a18da7aSGeorge Spelvin * Perfect: 2048 31744 21062a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 21072a18da7aSGeorge Spelvin */ 21082a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 21092a18da7aSGeorge Spelvin ( x ^= (a), \ 21102a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 21112a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 21122a18da7aSGeorge Spelvin y *= 9 ) 2113bfcfaa77SLinus Torvalds 21142a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 21150fed3ac8SGeorge Spelvin { 21162a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 21172a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 21180fed3ac8SGeorge Spelvin } 21190fed3ac8SGeorge Spelvin 2120bfcfaa77SLinus Torvalds #endif 2121bfcfaa77SLinus Torvalds 21222a18da7aSGeorge Spelvin /* 21232a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21242a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21252a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21262a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21272a18da7aSGeorge Spelvin * finds the delimiter after the name. 21282a18da7aSGeorge Spelvin */ 21298387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2130bfcfaa77SLinus Torvalds { 21318387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2132bfcfaa77SLinus Torvalds 2133bfcfaa77SLinus Torvalds for (;;) { 2134fcfd2fbfSGeorge Spelvin if (!len) 2135fcfd2fbfSGeorge Spelvin goto done; 2136e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2137bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2138bfcfaa77SLinus Torvalds break; 21392a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2140bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2141bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2142bfcfaa77SLinus Torvalds } 21432a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2144bfcfaa77SLinus Torvalds done: 21452a18da7aSGeorge Spelvin return fold_hash(x, y); 2146bfcfaa77SLinus Torvalds } 2147bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2148bfcfaa77SLinus Torvalds 2149fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21508387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2151fcfd2fbfSGeorge Spelvin { 21528387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21538387ff25SLinus Torvalds unsigned long adata, mask, len; 2154fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2155fcfd2fbfSGeorge Spelvin 21568387ff25SLinus Torvalds len = 0; 21578387ff25SLinus Torvalds goto inside; 21588387ff25SLinus Torvalds 2159fcfd2fbfSGeorge Spelvin do { 21602a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2161fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21628387ff25SLinus Torvalds inside: 2163fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2164fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2165fcfd2fbfSGeorge Spelvin 2166fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2167fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21682a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2169fcfd2fbfSGeorge Spelvin 21702a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2171fcfd2fbfSGeorge Spelvin } 2172fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2173fcfd2fbfSGeorge Spelvin 2174bfcfaa77SLinus Torvalds /* 2175bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2176d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2177bfcfaa77SLinus Torvalds */ 21788387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2179bfcfaa77SLinus Torvalds { 21808387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21818387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 218236126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2183bfcfaa77SLinus Torvalds 21848387ff25SLinus Torvalds len = 0; 21858387ff25SLinus Torvalds goto inside; 21868387ff25SLinus Torvalds 2187bfcfaa77SLinus Torvalds do { 21882a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2189bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21908387ff25SLinus Torvalds inside: 2191e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 219236126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 219336126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2194bfcfaa77SLinus Torvalds 219536126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 219636126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 219736126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21982a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 219936126f8fSLinus Torvalds 22002a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2201bfcfaa77SLinus Torvalds } 2202bfcfaa77SLinus Torvalds 22032a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2204bfcfaa77SLinus Torvalds 2205fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 22068387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 22070145acc2SLinus Torvalds { 22088387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 22090145acc2SLinus Torvalds while (len--) 2210fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 22110145acc2SLinus Torvalds return end_name_hash(hash); 22120145acc2SLinus Torvalds } 2213ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 22140145acc2SLinus Torvalds 2215fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 22168387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2217fcfd2fbfSGeorge Spelvin { 22188387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2219fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2220fcfd2fbfSGeorge Spelvin 2221fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2222e0ab7af9SGeorge Spelvin while (c) { 2223fcfd2fbfSGeorge Spelvin len++; 2224fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2225fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2226e0ab7af9SGeorge Spelvin } 2227fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2228fcfd2fbfSGeorge Spelvin } 2229f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2230fcfd2fbfSGeorge Spelvin 22313ddcd056SLinus Torvalds /* 2232200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2233200e9ef7SLinus Torvalds * one character. 2234200e9ef7SLinus Torvalds */ 22358387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2236200e9ef7SLinus Torvalds { 22378387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2238200e9ef7SLinus Torvalds unsigned long len = 0, c; 2239200e9ef7SLinus Torvalds 2240200e9ef7SLinus Torvalds c = (unsigned char)*name; 2241200e9ef7SLinus Torvalds do { 2242200e9ef7SLinus Torvalds len++; 2243200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2244200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2245200e9ef7SLinus Torvalds } while (c && c != '/'); 2246d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2247200e9ef7SLinus Torvalds } 2248200e9ef7SLinus Torvalds 2249bfcfaa77SLinus Torvalds #endif 2250bfcfaa77SLinus Torvalds 2251200e9ef7SLinus Torvalds /* 22521da177e4SLinus Torvalds * Name resolution. 2253ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2254ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22551da177e4SLinus Torvalds * 2256ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2257ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22581da177e4SLinus Torvalds */ 22596de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22601da177e4SLinus Torvalds { 2261d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22621da177e4SLinus Torvalds int err; 22631da177e4SLinus Torvalds 2264b4c03536SAl Viro nd->last_type = LAST_ROOT; 2265c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22669b5858e9SAl Viro if (IS_ERR(name)) 22679b5858e9SAl Viro return PTR_ERR(name); 22681da177e4SLinus Torvalds while (*name=='/') 22691da177e4SLinus Torvalds name++; 22701a97d899SAl Viro if (!*name) { 22711a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22729e18f10aSAl Viro return 0; 22731a97d899SAl Viro } 22741da177e4SLinus Torvalds 22751da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22761da177e4SLinus Torvalds for(;;) { 2277549c7297SChristian Brauner struct user_namespace *mnt_userns; 227892d27016SAl Viro const char *link; 2279d6bb3e90SLinus Torvalds u64 hash_len; 2280fe479a58SAl Viro int type; 22811da177e4SLinus Torvalds 2282549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2283549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22841da177e4SLinus Torvalds if (err) 22853595e234SAl Viro return err; 22861da177e4SLinus Torvalds 22878387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22881da177e4SLinus Torvalds 2289fe479a58SAl Viro type = LAST_NORM; 2290d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2291fe479a58SAl Viro case 2: 2292200e9ef7SLinus Torvalds if (name[1] == '.') { 2293fe479a58SAl Viro type = LAST_DOTDOT; 2294bcba1e7dSAl Viro nd->state |= ND_JUMPED; 229516c2cd71SAl Viro } 2296fe479a58SAl Viro break; 2297fe479a58SAl Viro case 1: 2298fe479a58SAl Viro type = LAST_DOT; 2299fe479a58SAl Viro } 23005a202bcdSAl Viro if (likely(type == LAST_NORM)) { 23015a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2302bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 23035a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2304a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2305da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 23065a202bcdSAl Viro if (err < 0) 23073595e234SAl Viro return err; 2308d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2309d6bb3e90SLinus Torvalds name = this.name; 23105a202bcdSAl Viro } 23115a202bcdSAl Viro } 2312fe479a58SAl Viro 2313d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2314d6bb3e90SLinus Torvalds nd->last.name = name; 23155f4a6a69SAl Viro nd->last_type = type; 23165f4a6a69SAl Viro 2317d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2318d6bb3e90SLinus Torvalds if (!*name) 2319bdf6cbf1SAl Viro goto OK; 2320200e9ef7SLinus Torvalds /* 2321200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2322200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2323200e9ef7SLinus Torvalds */ 2324200e9ef7SLinus Torvalds do { 2325d6bb3e90SLinus Torvalds name++; 2326d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 23278620c238SAl Viro if (unlikely(!*name)) { 23288620c238SAl Viro OK: 2329d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2330c108837eSAl Viro if (!depth) { 2331549c7297SChristian Brauner nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); 23320f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2333c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 23348620c238SAl Viro return 0; 2335c108837eSAl Viro } 23368620c238SAl Viro /* last component of nested symlink */ 2337d8d4611aSAl Viro name = nd->stack[--depth].name; 23388c4efe22SAl Viro link = walk_component(nd, 0); 23391c4ff1a8SAl Viro } else { 23401c4ff1a8SAl Viro /* not the last component */ 23418c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 23428620c238SAl Viro } 234392d27016SAl Viro if (unlikely(link)) { 234492d27016SAl Viro if (IS_ERR(link)) 234592d27016SAl Viro return PTR_ERR(link); 234692d27016SAl Viro /* a symlink to follow */ 2347d8d4611aSAl Viro nd->stack[depth++].name = name; 234892d27016SAl Viro name = link; 23499e18f10aSAl Viro continue; 235048c8b0c5SAl Viro } 235197242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 235297242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2353e36cffedSJens Axboe if (!try_to_unlazy(nd)) 235497242f99SAl Viro return -ECHILD; 235597242f99SAl Viro } 23563595e234SAl Viro return -ENOTDIR; 23575f4a6a69SAl Viro } 2358ce57dfc1SAl Viro } 235997242f99SAl Viro } 23601da177e4SLinus Torvalds 2361edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2362c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 236331e6b01fSNick Piggin { 2364740a1678SAleksa Sarai int error; 2365c8a53ee5SAl Viro const char *s = nd->name->name; 236631e6b01fSNick Piggin 23676c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23686c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23696c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23706c6ec2b0SJens Axboe 2371c0eb027eSLinus Torvalds if (!*s) 2372c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2373edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2374edc2b1daSAl Viro rcu_read_lock(); 2375c0eb027eSLinus Torvalds 2376bcba1e7dSAl Viro nd->flags = flags; 2377bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2378ab87f9a5SAleksa Sarai 2379ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2380ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2381ab87f9a5SAleksa Sarai smp_rmb(); 2382ab87f9a5SAleksa Sarai 2383bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2384b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2385b18825a7SDavid Howells struct inode *inode = root->d_inode; 238693893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2387368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23885b6ca027SAl Viro nd->path = nd->root; 23895b6ca027SAl Viro nd->inode = inode; 23905b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2391ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23928f47a016SAl Viro nd->root_seq = nd->seq; 23935b6ca027SAl Viro } else { 23945b6ca027SAl Viro path_get(&nd->path); 23955b6ca027SAl Viro } 2396368ee9baSAl Viro return s; 23975b6ca027SAl Viro } 23985b6ca027SAl Viro 239931e6b01fSNick Piggin nd->root.mnt = NULL; 240031e6b01fSNick Piggin 24018db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 24028db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2403740a1678SAleksa Sarai error = nd_jump_root(nd); 2404740a1678SAleksa Sarai if (unlikely(error)) 2405740a1678SAleksa Sarai return ERR_PTR(error); 2406ef55d917SAl Viro return s; 24078db52c7eSAleksa Sarai } 24088db52c7eSAleksa Sarai 24098db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 24108db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2411e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 241231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2413c28cc364SNick Piggin unsigned seq; 241431e6b01fSNick Piggin 2415c28cc364SNick Piggin do { 2416c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 241731e6b01fSNick Piggin nd->path = fs->pwd; 2418ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2419c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2420c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2421e41f7d4eSAl Viro } else { 2422e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2423ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2424e41f7d4eSAl Viro } 242531e6b01fSNick Piggin } else { 2426582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2427c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 242831e6b01fSNick Piggin struct dentry *dentry; 242931e6b01fSNick Piggin 24302903ff01SAl Viro if (!f.file) 2431368ee9baSAl Viro return ERR_PTR(-EBADF); 243231e6b01fSNick Piggin 24332903ff01SAl Viro dentry = f.file->f_path.dentry; 243431e6b01fSNick Piggin 2435edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 24362903ff01SAl Viro fdput(f); 2437368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2438f52e0c11SAl Viro } 24392903ff01SAl Viro 24402903ff01SAl Viro nd->path = f.file->f_path; 2441e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 244234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 244334a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24445590ff0dSUlrich Drepper } else { 24452903ff01SAl Viro path_get(&nd->path); 244634a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24471da177e4SLinus Torvalds } 244834a26b99SAl Viro fdput(f); 2449e41f7d4eSAl Viro } 24508db52c7eSAleksa Sarai 2451adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2452adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2453adb21d2bSAleksa Sarai nd->root = nd->path; 2454adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2455adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2456adb21d2bSAleksa Sarai } else { 2457adb21d2bSAleksa Sarai path_get(&nd->root); 2458bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2459adb21d2bSAleksa Sarai } 2460adb21d2bSAleksa Sarai } 2461adb21d2bSAleksa Sarai return s; 24629b4a9b14SAl Viro } 24639b4a9b14SAl Viro 24641ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 246595fa25d9SAl Viro { 2466bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2467bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2468bd92d7feSAl Viro 2469c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2470bd92d7feSAl Viro } 2471bd92d7feSAl Viro 24724f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24734f757f3cSAl Viro { 2474c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2475db3c9adeSAl Viro dget(nd->path.dentry); 2476b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2477b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 24784f757f3cSAl Viro } 24794f757f3cSAl Viro 24809b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2481c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24829b4a9b14SAl Viro { 2483c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2484bd92d7feSAl Viro int err; 248531e6b01fSNick Piggin 24869b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24874f757f3cSAl Viro err = handle_lookup_down(nd); 24885f336e72SAl Viro if (unlikely(err < 0)) 24895f336e72SAl Viro s = ERR_PTR(err); 24904f757f3cSAl Viro } 24914f757f3cSAl Viro 24921ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24931ccac622SAl Viro (s = lookup_last(nd)) != NULL) 24941ccac622SAl Viro ; 24954f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 24964f0ed93fSAl Viro err = handle_lookup_down(nd); 2497bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 24984f0ed93fSAl Viro } 24999f1fafeeSAl Viro if (!err) 25009f1fafeeSAl Viro err = complete_walk(nd); 2501bd92d7feSAl Viro 2502deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2503deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2504bd23a539SAl Viro err = -ENOTDIR; 2505625b6d10SAl Viro if (!err) { 2506625b6d10SAl Viro *path = nd->path; 2507625b6d10SAl Viro nd->path.mnt = NULL; 2508625b6d10SAl Viro nd->path.dentry = NULL; 2509625b6d10SAl Viro } 2510deb106c6SAl Viro terminate_walk(nd); 2511bd92d7feSAl Viro return err; 251231e6b01fSNick Piggin } 251331e6b01fSNick Piggin 2514794ebceaSStephen Brennan int filename_lookup(int dfd, struct filename *name, unsigned flags, 25159ad1aaa6SAl Viro struct path *path, struct path *root) 2516873f1eedSJeff Layton { 2517894bc8c4SAl Viro int retval; 25189883d185SAl Viro struct nameidata nd; 2519abc9f5beSAl Viro if (IS_ERR(name)) 2520abc9f5beSAl Viro return PTR_ERR(name); 252106422964SAl Viro set_nameidata(&nd, dfd, name, root); 2522c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2523873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2524c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2525873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2526c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2527873f1eedSJeff Layton 2528873f1eedSJeff Layton if (likely(!retval)) 2529161aff1dSAl Viro audit_inode(name, path->dentry, 2530161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 25319883d185SAl Viro restore_nameidata(); 2532020250f3SDmitry Kadashev return retval; 2533020250f3SDmitry Kadashev } 2534020250f3SDmitry Kadashev 25358bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2536c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2537391172c4SAl Viro struct path *parent) 25388bcb77faSAl Viro { 2539c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25409b5858e9SAl Viro int err = link_path_walk(s, nd); 25418bcb77faSAl Viro if (!err) 25428bcb77faSAl Viro err = complete_walk(nd); 2543391172c4SAl Viro if (!err) { 2544391172c4SAl Viro *parent = nd->path; 2545391172c4SAl Viro nd->path.mnt = NULL; 2546391172c4SAl Viro nd->path.dentry = NULL; 2547391172c4SAl Viro } 2548deb106c6SAl Viro terminate_walk(nd); 25498bcb77faSAl Viro return err; 25508bcb77faSAl Viro } 25518bcb77faSAl Viro 25520766ec82SStephen Brennan /* Note: this does not consume "name" */ 2553c5f563f9SAl Viro static int filename_parentat(int dfd, struct filename *name, 2554391172c4SAl Viro unsigned int flags, struct path *parent, 2555391172c4SAl Viro struct qstr *last, int *type) 25568bcb77faSAl Viro { 25578bcb77faSAl Viro int retval; 25589883d185SAl Viro struct nameidata nd; 25598bcb77faSAl Viro 25605c31b6ceSAl Viro if (IS_ERR(name)) 25610ee50b47SDmitry Kadashev return PTR_ERR(name); 256206422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2563c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25648bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2565c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25668bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2567c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2568391172c4SAl Viro if (likely(!retval)) { 2569391172c4SAl Viro *last = nd.last; 2570391172c4SAl Viro *type = nd.last_type; 2571c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2572391172c4SAl Viro } 25739883d185SAl Viro restore_nameidata(); 25740ee50b47SDmitry Kadashev return retval; 25750ee50b47SDmitry Kadashev } 25760ee50b47SDmitry Kadashev 257779714f72SAl Viro /* does lookup, returns the object with parent locked */ 25780766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25795590ff0dSUlrich Drepper { 25805c31b6ceSAl Viro struct dentry *d; 2581391172c4SAl Viro struct qstr last; 25820ee50b47SDmitry Kadashev int type, error; 258351689104SPaul Moore 2584c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25850ee50b47SDmitry Kadashev if (error) 25860ee50b47SDmitry Kadashev return ERR_PTR(error); 25875c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2588391172c4SAl Viro path_put(path); 25895c31b6ceSAl Viro return ERR_PTR(-EINVAL); 259079714f72SAl Viro } 25915955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2592391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 259379714f72SAl Viro if (IS_ERR(d)) { 25945955102cSAl Viro inode_unlock(path->dentry->d_inode); 2595391172c4SAl Viro path_put(path); 259679714f72SAl Viro } 259779714f72SAl Viro return d; 25985590ff0dSUlrich Drepper } 25995590ff0dSUlrich Drepper 26000766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 26010766ec82SStephen Brennan { 26020766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 26030766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 26040766ec82SStephen Brennan 26050766ec82SStephen Brennan putname(filename); 26060766ec82SStephen Brennan return res; 26070766ec82SStephen Brennan } 26080766ec82SStephen Brennan 2609d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2610d1811465SAl Viro { 2611794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2612794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2613794ebceaSStephen Brennan 2614794ebceaSStephen Brennan putname(filename); 2615794ebceaSStephen Brennan return ret; 2616794ebceaSStephen Brennan 2617d1811465SAl Viro } 26184d359507SAl Viro EXPORT_SYMBOL(kern_path); 2619d1811465SAl Viro 262016f18200SJosef 'Jeff' Sipek /** 262116f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 262216f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 262316f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 262416f18200SJosef 'Jeff' Sipek * @name: pointer to file name 262516f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2626e0a01249SAl Viro * @path: pointer to struct path to fill 262716f18200SJosef 'Jeff' Sipek */ 262816f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 262916f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2630e0a01249SAl Viro struct path *path) 263116f18200SJosef 'Jeff' Sipek { 2632794ebceaSStephen Brennan struct filename *filename; 26339ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2634794ebceaSStephen Brennan int ret; 2635794ebceaSStephen Brennan 2636794ebceaSStephen Brennan filename = getname_kernel(name); 26379ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2638794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2639794ebceaSStephen Brennan putname(filename); 2640794ebceaSStephen Brennan return ret; 264116f18200SJosef 'Jeff' Sipek } 26424d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 264316f18200SJosef 'Jeff' Sipek 2644c2fd68b6SChristian Brauner static int lookup_one_common(struct user_namespace *mnt_userns, 2645c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2646c2fd68b6SChristian Brauner struct qstr *this) 26473c95f0dcSAl Viro { 26483c95f0dcSAl Viro this->name = name; 26493c95f0dcSAl Viro this->len = len; 26503c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26513c95f0dcSAl Viro if (!len) 26523c95f0dcSAl Viro return -EACCES; 26533c95f0dcSAl Viro 26543c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26553c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26563c95f0dcSAl Viro return -EACCES; 26573c95f0dcSAl Viro } 26583c95f0dcSAl Viro 26593c95f0dcSAl Viro while (len--) { 26603c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26613c95f0dcSAl Viro if (c == '/' || c == '\0') 26623c95f0dcSAl Viro return -EACCES; 26633c95f0dcSAl Viro } 26643c95f0dcSAl Viro /* 26653c95f0dcSAl Viro * See if the low-level filesystem might want 26663c95f0dcSAl Viro * to use its own hash.. 26673c95f0dcSAl Viro */ 26683c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26693c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26703c95f0dcSAl Viro if (err < 0) 26713c95f0dcSAl Viro return err; 26723c95f0dcSAl Viro } 26733c95f0dcSAl Viro 2674c2fd68b6SChristian Brauner return inode_permission(mnt_userns, base->d_inode, MAY_EXEC); 26753c95f0dcSAl Viro } 26763c95f0dcSAl Viro 2677eead1911SChristoph Hellwig /** 26780da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26790da0b7fdSDavid Howells * @name: pathname component to lookup 26800da0b7fdSDavid Howells * @base: base directory to lookup from 26810da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26820da0b7fdSDavid Howells * 26830da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26840da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26850da0b7fdSDavid Howells * 26860da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26870da0b7fdSDavid Howells * not be called by generic code. 26880da0b7fdSDavid Howells * 26890da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26900da0b7fdSDavid Howells */ 26910da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26920da0b7fdSDavid Howells { 26930da0b7fdSDavid Howells struct qstr this; 26940da0b7fdSDavid Howells int err; 26950da0b7fdSDavid Howells 26960da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26970da0b7fdSDavid Howells 2698c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 26990da0b7fdSDavid Howells if (err) 27000da0b7fdSDavid Howells return ERR_PTR(err); 27010da0b7fdSDavid Howells 27020da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 27030da0b7fdSDavid Howells } 27040da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 27050da0b7fdSDavid Howells 27060da0b7fdSDavid Howells /** 2707a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2708eead1911SChristoph Hellwig * @name: pathname component to lookup 2709eead1911SChristoph Hellwig * @base: base directory to lookup from 2710eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2711eead1911SChristoph Hellwig * 2712a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 27139e7543e9SAl Viro * not be called by generic code. 2714bbddca8eSNeilBrown * 2715bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2716eead1911SChristoph Hellwig */ 2717057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2718057f6c01SJames Morris { 27198613a209SAl Viro struct dentry *dentry; 2720057f6c01SJames Morris struct qstr this; 2721cda309deSMiklos Szeredi int err; 2722057f6c01SJames Morris 27235955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27242f9092e1SDavid Woodhouse 2725c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2726cda309deSMiklos Szeredi if (err) 2727cda309deSMiklos Szeredi return ERR_PTR(err); 2728cda309deSMiklos Szeredi 27298613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27308613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2731057f6c01SJames Morris } 27324d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2733057f6c01SJames Morris 2734bbddca8eSNeilBrown /** 2735c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 2736c2fd68b6SChristian Brauner * @mnt_userns: user namespace of the mount the lookup is performed from 2737c2fd68b6SChristian Brauner * @name: pathname component to lookup 2738c2fd68b6SChristian Brauner * @base: base directory to lookup from 2739c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2740c2fd68b6SChristian Brauner * 2741c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2742c2fd68b6SChristian Brauner * not be called by generic code. 2743c2fd68b6SChristian Brauner * 2744c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2745c2fd68b6SChristian Brauner */ 2746c2fd68b6SChristian Brauner struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name, 2747c2fd68b6SChristian Brauner struct dentry *base, int len) 2748c2fd68b6SChristian Brauner { 2749c2fd68b6SChristian Brauner struct dentry *dentry; 2750c2fd68b6SChristian Brauner struct qstr this; 2751c2fd68b6SChristian Brauner int err; 2752c2fd68b6SChristian Brauner 2753c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2754c2fd68b6SChristian Brauner 2755c2fd68b6SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 2756c2fd68b6SChristian Brauner if (err) 2757c2fd68b6SChristian Brauner return ERR_PTR(err); 2758c2fd68b6SChristian Brauner 2759c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2760c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2761c2fd68b6SChristian Brauner } 2762c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2763c2fd68b6SChristian Brauner 2764c2fd68b6SChristian Brauner /** 276500675017SChristian Brauner * lookup_one_unlocked - filesystem helper to lookup single pathname component 276600675017SChristian Brauner * @mnt_userns: idmapping of the mount the lookup is performed from 276700675017SChristian Brauner * @name: pathname component to lookup 276800675017SChristian Brauner * @base: base directory to lookup from 276900675017SChristian Brauner * @len: maximum length @len should be interpreted to 277000675017SChristian Brauner * 277100675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 277200675017SChristian Brauner * not be called by generic code. 277300675017SChristian Brauner * 277400675017SChristian Brauner * Unlike lookup_one_len, it should be called without the parent 277500675017SChristian Brauner * i_mutex held, and will take the i_mutex itself if necessary. 277600675017SChristian Brauner */ 277700675017SChristian Brauner struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns, 277800675017SChristian Brauner const char *name, struct dentry *base, 277900675017SChristian Brauner int len) 278000675017SChristian Brauner { 278100675017SChristian Brauner struct qstr this; 278200675017SChristian Brauner int err; 278300675017SChristian Brauner struct dentry *ret; 278400675017SChristian Brauner 278500675017SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 278600675017SChristian Brauner if (err) 278700675017SChristian Brauner return ERR_PTR(err); 278800675017SChristian Brauner 278900675017SChristian Brauner ret = lookup_dcache(&this, base, 0); 279000675017SChristian Brauner if (!ret) 279100675017SChristian Brauner ret = lookup_slow(&this, base, 0); 279200675017SChristian Brauner return ret; 279300675017SChristian Brauner } 279400675017SChristian Brauner EXPORT_SYMBOL(lookup_one_unlocked); 279500675017SChristian Brauner 279600675017SChristian Brauner /** 279700675017SChristian Brauner * lookup_one_positive_unlocked - filesystem helper to lookup single 279800675017SChristian Brauner * pathname component 279900675017SChristian Brauner * @mnt_userns: idmapping of the mount the lookup is performed from 280000675017SChristian Brauner * @name: pathname component to lookup 280100675017SChristian Brauner * @base: base directory to lookup from 280200675017SChristian Brauner * @len: maximum length @len should be interpreted to 280300675017SChristian Brauner * 280400675017SChristian Brauner * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns 280500675017SChristian Brauner * known positive or ERR_PTR(). This is what most of the users want. 280600675017SChristian Brauner * 280700675017SChristian Brauner * Note that pinned negative with unlocked parent _can_ become positive at any 280800675017SChristian Brauner * time, so callers of lookup_one_unlocked() need to be very careful; pinned 280900675017SChristian Brauner * positives have >d_inode stable, so this one avoids such problems. 281000675017SChristian Brauner * 281100675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 281200675017SChristian Brauner * not be called by generic code. 281300675017SChristian Brauner * 281400675017SChristian Brauner * The helper should be called without i_mutex held. 281500675017SChristian Brauner */ 281600675017SChristian Brauner struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns, 281700675017SChristian Brauner const char *name, 281800675017SChristian Brauner struct dentry *base, int len) 281900675017SChristian Brauner { 282000675017SChristian Brauner struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len); 282100675017SChristian Brauner 282200675017SChristian Brauner if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 282300675017SChristian Brauner dput(ret); 282400675017SChristian Brauner ret = ERR_PTR(-ENOENT); 282500675017SChristian Brauner } 282600675017SChristian Brauner return ret; 282700675017SChristian Brauner } 282800675017SChristian Brauner EXPORT_SYMBOL(lookup_one_positive_unlocked); 282900675017SChristian Brauner 283000675017SChristian Brauner /** 2831bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2832bbddca8eSNeilBrown * @name: pathname component to lookup 2833bbddca8eSNeilBrown * @base: base directory to lookup from 2834bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2835bbddca8eSNeilBrown * 2836bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2837bbddca8eSNeilBrown * not be called by generic code. 2838bbddca8eSNeilBrown * 2839bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2840bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2841bbddca8eSNeilBrown */ 2842bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2843bbddca8eSNeilBrown struct dentry *base, int len) 2844bbddca8eSNeilBrown { 284500675017SChristian Brauner return lookup_one_unlocked(&init_user_ns, name, base, len); 2846bbddca8eSNeilBrown } 2847bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2848bbddca8eSNeilBrown 28496c2d4798SAl Viro /* 28506c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 28516c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 28526c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 28536c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 28546c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 28556c2d4798SAl Viro * this one avoids such problems. 28566c2d4798SAl Viro */ 28576c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28586c2d4798SAl Viro struct dentry *base, int len) 28596c2d4798SAl Viro { 286000675017SChristian Brauner return lookup_one_positive_unlocked(&init_user_ns, name, base, len); 28616c2d4798SAl Viro } 28626c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28636c2d4798SAl Viro 2864eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2865eedf265aSEric W. Biederman int path_pts(struct path *path) 2866eedf265aSEric W. Biederman { 2867eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2868eedf265aSEric W. Biederman * the input path. 2869eedf265aSEric W. Biederman */ 2870a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2871a6a7eb76SAl Viro struct dentry *child; 287219f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2873eedf265aSEric W. Biederman 2874a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2875a6a7eb76SAl Viro dput(parent); 287663b27720SAl Viro return -ENOENT; 2877a6a7eb76SAl Viro } 287863b27720SAl Viro dput(path->dentry); 287963b27720SAl Viro path->dentry = parent; 2880eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2881eedf265aSEric W. Biederman if (!child) 2882eedf265aSEric W. Biederman return -ENOENT; 2883eedf265aSEric W. Biederman 2884eedf265aSEric W. Biederman path->dentry = child; 2885eedf265aSEric W. Biederman dput(parent); 288619f6028aSAl Viro follow_down(path); 2887eedf265aSEric W. Biederman return 0; 2888eedf265aSEric W. Biederman } 2889eedf265aSEric W. Biederman #endif 2890eedf265aSEric W. Biederman 28911fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 28921fa1e7f6SAndy Whitcroft struct path *path, int *empty) 28931da177e4SLinus Torvalds { 2894794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2895794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2896794ebceaSStephen Brennan 2897794ebceaSStephen Brennan putname(filename); 2898794ebceaSStephen Brennan return ret; 28991da177e4SLinus Torvalds } 2900b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 29011fa1e7f6SAndy Whitcroft 2902ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2903ba73d987SChristian Brauner struct inode *inode) 29041da177e4SLinus Torvalds { 29058e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2906da9592edSDavid Howells 2907ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) 29081da177e4SLinus Torvalds return 0; 2909ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) 29101da177e4SLinus Torvalds return 0; 2911ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 29121da177e4SLinus Torvalds } 2913cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 29141da177e4SLinus Torvalds 29151da177e4SLinus Torvalds /* 29161da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 29171da177e4SLinus Torvalds * whether the type of victim is right. 29181da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 29191da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 29201da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 29211da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 29221da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 29231da177e4SLinus Torvalds * a. be owner of dir, or 29241da177e4SLinus Torvalds * b. be owner of victim, or 29251da177e4SLinus Torvalds * c. have CAP_FOWNER capability 29261da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 29271da177e4SLinus Torvalds * links pointing to it. 29280bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 29290bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 29300bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 29310bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 29320bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 29331da177e4SLinus Torvalds * nfs_async_unlink(). 29341da177e4SLinus Torvalds */ 2935ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2936ba73d987SChristian Brauner struct dentry *victim, bool isdir) 29371da177e4SLinus Torvalds { 293863afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 29391da177e4SLinus Torvalds int error; 29401da177e4SLinus Torvalds 2941b18825a7SDavid Howells if (d_is_negative(victim)) 29421da177e4SLinus Torvalds return -ENOENT; 2943b18825a7SDavid Howells BUG_ON(!inode); 29441da177e4SLinus Torvalds 29451da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2946593d1ce8SEric W. Biederman 2947593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2948ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 2949ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 2950593d1ce8SEric W. Biederman return -EOVERFLOW; 2951593d1ce8SEric W. Biederman 29524fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29531da177e4SLinus Torvalds 2954ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29551da177e4SLinus Torvalds if (error) 29561da177e4SLinus Torvalds return error; 29571da177e4SLinus Torvalds if (IS_APPEND(dir)) 29581da177e4SLinus Torvalds return -EPERM; 2959b18825a7SDavid Howells 2960ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2961ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2962ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 29631da177e4SLinus Torvalds return -EPERM; 29641da177e4SLinus Torvalds if (isdir) { 296544b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29661da177e4SLinus Torvalds return -ENOTDIR; 29671da177e4SLinus Torvalds if (IS_ROOT(victim)) 29681da177e4SLinus Torvalds return -EBUSY; 296944b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29701da177e4SLinus Torvalds return -EISDIR; 29711da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29721da177e4SLinus Torvalds return -ENOENT; 29731da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29741da177e4SLinus Torvalds return -EBUSY; 29751da177e4SLinus Torvalds return 0; 29761da177e4SLinus Torvalds } 29771da177e4SLinus Torvalds 29781da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29791da177e4SLinus Torvalds * dir. 29801da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29811da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29821da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2983036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2984036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2985036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 29861da177e4SLinus Torvalds */ 2987ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2988ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 29891da177e4SLinus Torvalds { 299014e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 29911da177e4SLinus Torvalds if (child->d_inode) 29921da177e4SLinus Torvalds return -EEXIST; 29931da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29941da177e4SLinus Torvalds return -ENOENT; 29958e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 2996036d5236SEric W. Biederman return -EOVERFLOW; 29978e538913SChristian Brauner 2998ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29991da177e4SLinus Torvalds } 30001da177e4SLinus Torvalds 30011da177e4SLinus Torvalds /* 30021da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 30031da177e4SLinus Torvalds */ 30041da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 30051da177e4SLinus Torvalds { 30061da177e4SLinus Torvalds struct dentry *p; 30071da177e4SLinus Torvalds 30081da177e4SLinus Torvalds if (p1 == p2) { 30095955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30101da177e4SLinus Torvalds return NULL; 30111da177e4SLinus Torvalds } 30121da177e4SLinus Torvalds 3013fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 30141da177e4SLinus Torvalds 3015e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 3016e2761a11SOGAWA Hirofumi if (p) { 30175955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30185955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 30191da177e4SLinus Torvalds return p; 30201da177e4SLinus Torvalds } 30211da177e4SLinus Torvalds 3022e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 3023e2761a11SOGAWA Hirofumi if (p) { 30245955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30255955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 30261da177e4SLinus Torvalds return p; 30271da177e4SLinus Torvalds } 30281da177e4SLinus Torvalds 30295955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30305955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 30311da177e4SLinus Torvalds return NULL; 30321da177e4SLinus Torvalds } 30334d359507SAl Viro EXPORT_SYMBOL(lock_rename); 30341da177e4SLinus Torvalds 30351da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 30361da177e4SLinus Torvalds { 30375955102cSAl Viro inode_unlock(p1->d_inode); 30381da177e4SLinus Torvalds if (p1 != p2) { 30395955102cSAl Viro inode_unlock(p2->d_inode); 3040fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 30411da177e4SLinus Torvalds } 30421da177e4SLinus Torvalds } 30434d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 30441da177e4SLinus Torvalds 30456521f891SChristian Brauner /** 3046*1639a49cSYang Xu * mode_strip_umask - handle vfs umask stripping 3047*1639a49cSYang Xu * @dir: parent directory of the new inode 3048*1639a49cSYang Xu * @mode: mode of the new inode to be created in @dir 3049*1639a49cSYang Xu * 3050*1639a49cSYang Xu * Umask stripping depends on whether or not the filesystem supports POSIX 3051*1639a49cSYang Xu * ACLs. If the filesystem doesn't support it umask stripping is done directly 3052*1639a49cSYang Xu * in here. If the filesystem does support POSIX ACLs umask stripping is 3053*1639a49cSYang Xu * deferred until the filesystem calls posix_acl_create(). 3054*1639a49cSYang Xu * 3055*1639a49cSYang Xu * Returns: mode 3056*1639a49cSYang Xu */ 3057*1639a49cSYang Xu static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode) 3058*1639a49cSYang Xu { 3059*1639a49cSYang Xu if (!IS_POSIXACL(dir)) 3060*1639a49cSYang Xu mode &= ~current_umask(); 3061*1639a49cSYang Xu return mode; 3062*1639a49cSYang Xu } 3063*1639a49cSYang Xu 3064*1639a49cSYang Xu /** 3065*1639a49cSYang Xu * vfs_prepare_mode - prepare the mode to be used for a new inode 3066*1639a49cSYang Xu * @mnt_userns: user namespace of the mount the inode was found from 3067*1639a49cSYang Xu * @dir: parent directory of the new inode 3068*1639a49cSYang Xu * @mode: mode of the new inode 3069*1639a49cSYang Xu * @mask_perms: allowed permission by the vfs 3070*1639a49cSYang Xu * @type: type of file to be created 3071*1639a49cSYang Xu * 3072*1639a49cSYang Xu * This helper consolidates and enforces vfs restrictions on the @mode of a new 3073*1639a49cSYang Xu * object to be created. 3074*1639a49cSYang Xu * 3075*1639a49cSYang Xu * Umask stripping depends on whether the filesystem supports POSIX ACLs (see 3076*1639a49cSYang Xu * the kernel documentation for mode_strip_umask()). Moving umask stripping 3077*1639a49cSYang Xu * after setgid stripping allows the same ordering for both non-POSIX ACL and 3078*1639a49cSYang Xu * POSIX ACL supporting filesystems. 3079*1639a49cSYang Xu * 3080*1639a49cSYang Xu * Note that it's currently valid for @type to be 0 if a directory is created. 3081*1639a49cSYang Xu * Filesystems raise that flag individually and we need to check whether each 3082*1639a49cSYang Xu * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a 3083*1639a49cSYang Xu * non-zero type. 3084*1639a49cSYang Xu * 3085*1639a49cSYang Xu * Returns: mode to be passed to the filesystem 3086*1639a49cSYang Xu */ 3087*1639a49cSYang Xu static inline umode_t vfs_prepare_mode(struct user_namespace *mnt_userns, 3088*1639a49cSYang Xu const struct inode *dir, umode_t mode, 3089*1639a49cSYang Xu umode_t mask_perms, umode_t type) 3090*1639a49cSYang Xu { 3091*1639a49cSYang Xu mode = mode_strip_sgid(mnt_userns, dir, mode); 3092*1639a49cSYang Xu mode = mode_strip_umask(dir, mode); 3093*1639a49cSYang Xu 3094*1639a49cSYang Xu /* 3095*1639a49cSYang Xu * Apply the vfs mandated allowed permission mask and set the type of 3096*1639a49cSYang Xu * file to be created before we call into the filesystem. 3097*1639a49cSYang Xu */ 3098*1639a49cSYang Xu mode &= (mask_perms & ~S_IFMT); 3099*1639a49cSYang Xu mode |= (type & S_IFMT); 3100*1639a49cSYang Xu 3101*1639a49cSYang Xu return mode; 3102*1639a49cSYang Xu } 3103*1639a49cSYang Xu 3104*1639a49cSYang Xu /** 31056521f891SChristian Brauner * vfs_create - create new file 31066521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 31076521f891SChristian Brauner * @dir: inode of @dentry 31086521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 31096521f891SChristian Brauner * @mode: mode of the new file 31106521f891SChristian Brauner * @want_excl: whether the file must not yet exist 31116521f891SChristian Brauner * 31126521f891SChristian Brauner * Create a new file. 31136521f891SChristian Brauner * 31146521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 31156521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 31166521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 31176521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 31186521f891SChristian Brauner * raw inode simply passs init_user_ns. 31196521f891SChristian Brauner */ 31206521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 31216521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 31221da177e4SLinus Torvalds { 31236521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 31241da177e4SLinus Torvalds if (error) 31251da177e4SLinus Torvalds return error; 31261da177e4SLinus Torvalds 3127acfa4380SAl Viro if (!dir->i_op->create) 31281da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 3129*1639a49cSYang Xu 3130*1639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IALLUGO, S_IFREG); 31311da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 31321da177e4SLinus Torvalds if (error) 31331da177e4SLinus Torvalds return error; 3134549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 3135a74574aaSStephen Smalley if (!error) 3136f38aa942SAmy Griffis fsnotify_create(dir, dentry); 31371da177e4SLinus Torvalds return error; 31381da177e4SLinus Torvalds } 31394d359507SAl Viro EXPORT_SYMBOL(vfs_create); 31401da177e4SLinus Torvalds 31418e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 31428e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 31438e6c848eSAl Viro void *arg) 31448e6c848eSAl Viro { 31458e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 3146ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 31478e6c848eSAl Viro if (error) 31488e6c848eSAl Viro return error; 31498e6c848eSAl Viro 31508e6c848eSAl Viro mode &= S_IALLUGO; 31518e6c848eSAl Viro mode |= S_IFREG; 31528e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 31538e6c848eSAl Viro if (error) 31548e6c848eSAl Viro return error; 31558e6c848eSAl Viro error = f(dentry, mode, arg); 31568e6c848eSAl Viro if (!error) 31578e6c848eSAl Viro fsnotify_create(dir, dentry); 31588e6c848eSAl Viro return error; 31598e6c848eSAl Viro } 31608e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 31618e6c848eSAl Viro 3162a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3163a2982cc9SEric W. Biederman { 3164a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3165a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3166a2982cc9SEric W. Biederman } 3167a2982cc9SEric W. Biederman 3168ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 3169ba73d987SChristian Brauner int acc_mode, int flag) 31701da177e4SLinus Torvalds { 31713fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 31721da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 31731da177e4SLinus Torvalds int error; 31741da177e4SLinus Torvalds 31751da177e4SLinus Torvalds if (!inode) 31761da177e4SLinus Torvalds return -ENOENT; 31771da177e4SLinus Torvalds 3178c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3179c8fe8f30SChristoph Hellwig case S_IFLNK: 31801da177e4SLinus Torvalds return -ELOOP; 3181c8fe8f30SChristoph Hellwig case S_IFDIR: 3182fc4177beSKees Cook if (acc_mode & MAY_WRITE) 31831da177e4SLinus Torvalds return -EISDIR; 3184fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3185fc4177beSKees Cook return -EACCES; 3186c8fe8f30SChristoph Hellwig break; 3187c8fe8f30SChristoph Hellwig case S_IFBLK: 3188c8fe8f30SChristoph Hellwig case S_IFCHR: 3189a2982cc9SEric W. Biederman if (!may_open_dev(path)) 31901da177e4SLinus Torvalds return -EACCES; 3191633fb6acSKees Cook fallthrough; 3192c8fe8f30SChristoph Hellwig case S_IFIFO: 3193c8fe8f30SChristoph Hellwig case S_IFSOCK: 3194633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3195633fb6acSKees Cook return -EACCES; 31961da177e4SLinus Torvalds flag &= ~O_TRUNC; 3197c8fe8f30SChristoph Hellwig break; 31980fd338b2SKees Cook case S_IFREG: 31990fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 32000fd338b2SKees Cook return -EACCES; 32010fd338b2SKees Cook break; 32024a3fd211SDave Hansen } 3203b41572e9SDave Hansen 3204ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 3205b41572e9SDave Hansen if (error) 3206b41572e9SDave Hansen return error; 32076146f0d5SMimi Zohar 32081da177e4SLinus Torvalds /* 32091da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 32101da177e4SLinus Torvalds */ 32111da177e4SLinus Torvalds if (IS_APPEND(inode)) { 32128737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 32137715b521SAl Viro return -EPERM; 32141da177e4SLinus Torvalds if (flag & O_TRUNC) 32157715b521SAl Viro return -EPERM; 32161da177e4SLinus Torvalds } 32171da177e4SLinus Torvalds 32181da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3219ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 32207715b521SAl Viro return -EPERM; 32211da177e4SLinus Torvalds 3222f3c7691eSJ. Bruce Fields return 0; 32237715b521SAl Viro } 32247715b521SAl Viro 3225549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 32267715b521SAl Viro { 3227f0bb5aafSAl Viro const struct path *path = &filp->f_path; 32287715b521SAl Viro struct inode *inode = path->dentry->d_inode; 32297715b521SAl Viro int error = get_write_access(inode); 32301da177e4SLinus Torvalds if (error) 32317715b521SAl Viro return error; 3232482e0007SJeff Layton 3233ea0d3ab2STetsuo Handa error = security_path_truncate(path); 32341da177e4SLinus Torvalds if (!error) { 3235549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3236d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3237e1181ee6SJeff Layton filp); 32381da177e4SLinus Torvalds } 32391da177e4SLinus Torvalds put_write_access(inode); 3240acd0c935SMimi Zohar return error; 32411da177e4SLinus Torvalds } 32421da177e4SLinus Torvalds 3243d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3244d57999e1SDave Hansen { 32458a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 32468a5e929dSAl Viro flag--; 3247d57999e1SDave Hansen return flag; 3248d57999e1SDave Hansen } 3249d57999e1SDave Hansen 3250ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3251ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3252ba73d987SChristian Brauner umode_t mode) 3253d18e9008SMiklos Szeredi { 3254d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3255d18e9008SMiklos Szeredi if (error) 3256d18e9008SMiklos Szeredi return error; 3257d18e9008SMiklos Szeredi 32588e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) 32591328c727SSeth Forshee return -EOVERFLOW; 32601328c727SSeth Forshee 3261ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 326247291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3263d18e9008SMiklos Szeredi if (error) 3264d18e9008SMiklos Szeredi return error; 3265d18e9008SMiklos Szeredi 3266d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3267d18e9008SMiklos Szeredi } 3268d18e9008SMiklos Szeredi 32691acf0af9SDavid Howells /* 32701acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 32711acf0af9SDavid Howells * dentry. 32721acf0af9SDavid Howells * 32731acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 32741acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 32751acf0af9SDavid Howells * 327600a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 327700a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 327800a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 32791acf0af9SDavid Howells * 32801acf0af9SDavid Howells * Returns an error code otherwise. 32811acf0af9SDavid Howells */ 3282239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3283239eb983SAl Viro struct file *file, 32843ec2eef1SAl Viro int open_flag, umode_t mode) 3285d18e9008SMiklos Szeredi { 3286d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3287d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3288d18e9008SMiklos Szeredi int error; 3289d18e9008SMiklos Szeredi 3290d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3291d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3292d18e9008SMiklos Szeredi 329330d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 329430d90494SAl Viro file->f_path.mnt = nd->path.mnt; 32950fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 329644907d79SAl Viro open_to_namei_flags(open_flag), mode); 32976fbd0714SAl Viro d_lookup_done(dentry); 3298384f26e2SAl Viro if (!error) { 329964e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 33006fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 33016fb968cdSAl Viro dput(dentry); 33026fb968cdSAl Viro dentry = dget(file->f_path.dentry); 33036fb968cdSAl Viro } 330464e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 33052675a4ebSAl Viro error = -EIO; 3306384f26e2SAl Viro } else { 330730d90494SAl Viro if (file->f_path.dentry) { 3308d18e9008SMiklos Szeredi dput(dentry); 3309d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 331010c64ceaSAl Viro } 3311239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3312a01e718fSAl Viro error = -ENOENT; 3313d18e9008SMiklos Szeredi } 3314d18e9008SMiklos Szeredi } 3315239eb983SAl Viro if (error) { 3316d18e9008SMiklos Szeredi dput(dentry); 3317239eb983SAl Viro dentry = ERR_PTR(error); 3318239eb983SAl Viro } 3319239eb983SAl Viro return dentry; 3320d18e9008SMiklos Szeredi } 3321d18e9008SMiklos Szeredi 332231e6b01fSNick Piggin /* 33231acf0af9SDavid Howells * Look up and maybe create and open the last component. 3324d58ffd35SMiklos Szeredi * 332500a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3326d58ffd35SMiklos Szeredi * 332700a07c15SAl Viro * Returns 0 on success, that is, if 332800a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 332900a07c15SAl Viro * the file was not completely opened at this time, though lookups and 333000a07c15SAl Viro * creations were performed. 333100a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 333200a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 333300a07c15SAl Viro * hadn't been specified. 33341acf0af9SDavid Howells * 333500a07c15SAl Viro * An error code is returned on failure. 3336d58ffd35SMiklos Szeredi */ 3337da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3338d58ffd35SMiklos Szeredi const struct open_flags *op, 33393ec2eef1SAl Viro bool got_write) 3340d58ffd35SMiklos Szeredi { 3341549c7297SChristian Brauner struct user_namespace *mnt_userns; 3342d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 334354ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 33441643b43fSAl Viro int open_flag = op->open_flag; 3345d58ffd35SMiklos Szeredi struct dentry *dentry; 33461643b43fSAl Viro int error, create_error = 0; 33471643b43fSAl Viro umode_t mode = op->mode; 33486fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3349d58ffd35SMiklos Szeredi 3350ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3351da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3352d58ffd35SMiklos Szeredi 335373a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 33546fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 33556fbd0714SAl Viro for (;;) { 33566fbd0714SAl Viro if (!dentry) { 33576fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3358d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3359da5ebf5aSAl Viro return dentry; 33606fbd0714SAl Viro } 33616fbd0714SAl Viro if (d_in_lookup(dentry)) 33626fbd0714SAl Viro break; 3363d58ffd35SMiklos Szeredi 33646fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 33656fbd0714SAl Viro if (likely(error > 0)) 33666fbd0714SAl Viro break; 33676fbd0714SAl Viro if (error) 33686fbd0714SAl Viro goto out_dput; 33696fbd0714SAl Viro d_invalidate(dentry); 33706fbd0714SAl Viro dput(dentry); 33716fbd0714SAl Viro dentry = NULL; 33726fbd0714SAl Viro } 33736fbd0714SAl Viro if (dentry->d_inode) { 3374d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3375da5ebf5aSAl Viro return dentry; 33766c51e513SAl Viro } 3377d18e9008SMiklos Szeredi 33781643b43fSAl Viro /* 33791643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 33801643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 33811643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 33821643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 33831643b43fSAl Viro * 33841643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 33851643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 33861643b43fSAl Viro */ 338799a4a90cSAl Viro if (unlikely(!got_write)) 338899a4a90cSAl Viro open_flag &= ~O_TRUNC; 3389549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 33901643b43fSAl Viro if (open_flag & O_CREAT) { 339199a4a90cSAl Viro if (open_flag & O_EXCL) 339299a4a90cSAl Viro open_flag &= ~O_TRUNC; 3393*1639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir->d_inode, mode, mode, mode); 339499a4a90cSAl Viro if (likely(got_write)) 3395549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3396ba73d987SChristian Brauner dentry, mode); 339799a4a90cSAl Viro else 339899a4a90cSAl Viro create_error = -EROFS; 339999a4a90cSAl Viro } 340099a4a90cSAl Viro if (create_error) 34011643b43fSAl Viro open_flag &= ~O_CREAT; 34021643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3403d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3404da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3405da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3406da5ebf5aSAl Viro return dentry; 3407239eb983SAl Viro } 340854ef4872SMiklos Szeredi 34096fbd0714SAl Viro if (d_in_lookup(dentry)) { 341012fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 341112fa5e24SAl Viro nd->flags); 34126fbd0714SAl Viro d_lookup_done(dentry); 341312fa5e24SAl Viro if (unlikely(res)) { 341412fa5e24SAl Viro if (IS_ERR(res)) { 341512fa5e24SAl Viro error = PTR_ERR(res); 341612fa5e24SAl Viro goto out_dput; 341712fa5e24SAl Viro } 341812fa5e24SAl Viro dput(dentry); 341912fa5e24SAl Viro dentry = res; 342012fa5e24SAl Viro } 342154ef4872SMiklos Szeredi } 342254ef4872SMiklos Szeredi 3423d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 34241643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 342573a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3426ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3427ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3428ce8644fcSAl Viro error = -EACCES; 3429d58ffd35SMiklos Szeredi goto out_dput; 343064894cf8SAl Viro } 3431549c7297SChristian Brauner 3432549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3433549c7297SChristian Brauner mode, open_flag & O_EXCL); 3434d58ffd35SMiklos Szeredi if (error) 3435d58ffd35SMiklos Szeredi goto out_dput; 3436d58ffd35SMiklos Szeredi } 34371643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 34381643b43fSAl Viro error = create_error; 3439d58ffd35SMiklos Szeredi goto out_dput; 3440d58ffd35SMiklos Szeredi } 3441da5ebf5aSAl Viro return dentry; 3442d58ffd35SMiklos Szeredi 3443d58ffd35SMiklos Szeredi out_dput: 3444d58ffd35SMiklos Szeredi dput(dentry); 3445da5ebf5aSAl Viro return ERR_PTR(error); 3446d58ffd35SMiklos Szeredi } 3447d58ffd35SMiklos Szeredi 3448c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 34493ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3450fb1cc555SAl Viro { 3451a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3452ca344a89SAl Viro int open_flag = op->open_flag; 345364894cf8SAl Viro bool got_write = false; 3454254cf582SAl Viro unsigned seq; 3455a1eb3315SMiklos Szeredi struct inode *inode; 3456da5ebf5aSAl Viro struct dentry *dentry; 3457b0417d2cSAl Viro const char *res; 3458fb1cc555SAl Viro 3459c3e380b0SAl Viro nd->flags |= op->intent; 3460c3e380b0SAl Viro 3461bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 346256676ec3SAl Viro if (nd->depth) 346356676ec3SAl Viro put_link(nd); 3464ff326a32SAl Viro return handle_dots(nd, nd->last_type); 34651f36f774SAl Viro } 3466a2c36b45SAl Viro 3467ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3468fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3469fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3470fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 347120e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 347220e34357SAl Viro if (IS_ERR(dentry)) 34731ccac622SAl Viro return ERR_CAST(dentry); 347420e34357SAl Viro if (likely(dentry)) 347571574865SMiklos Szeredi goto finish_lookup; 347671574865SMiklos Szeredi 34776583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3478b6183df7SMiklos Szeredi } else { 3479fe2d35ffSAl Viro /* create side of things */ 348072287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3481e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3482e36cffedSJens Axboe return ERR_PTR(-ECHILD); 348372287417SAl Viro } 3484c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 34851f36f774SAl Viro /* trailing slashes? */ 3486deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 34871ccac622SAl Viro return ERR_PTR(-EISDIR); 3488b6183df7SMiklos Szeredi } 34891f36f774SAl Viro 34909cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3491e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 349264894cf8SAl Viro /* 349364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 349464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 349564894cf8SAl Viro * dropping this one anyway. 349664894cf8SAl Viro */ 349764894cf8SAl Viro } 34989cf843e3SAl Viro if (open_flag & O_CREAT) 34995955102cSAl Viro inode_lock(dir->d_inode); 35009cf843e3SAl Viro else 35019cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3502da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3503f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3504f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 35059cf843e3SAl Viro if (open_flag & O_CREAT) 35065955102cSAl Viro inode_unlock(dir->d_inode); 35079cf843e3SAl Viro else 35089cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3509fb1cc555SAl Viro 3510c981a482SAl Viro if (got_write) 351159e96e65SAl Viro mnt_drop_write(nd->path.mnt); 35126c0d46c4SAl Viro 351359e96e65SAl Viro if (IS_ERR(dentry)) 351459e96e65SAl Viro return ERR_CAST(dentry); 351559e96e65SAl Viro 3516973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3517e73cabffSAl Viro dput(nd->path.dentry); 3518e73cabffSAl Viro nd->path.dentry = dentry; 3519c981a482SAl Viro return NULL; 3520fb1cc555SAl Viro } 3521fb1cc555SAl Viro 352220e34357SAl Viro finish_lookup: 352356676ec3SAl Viro if (nd->depth) 352456676ec3SAl Viro put_link(nd); 35258c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3526ff326a32SAl Viro if (unlikely(res)) 35271ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3528b0417d2cSAl Viro return res; 35291ccac622SAl Viro } 353031d1726dSAl Viro 3531c981a482SAl Viro /* 3532c981a482SAl Viro * Handle the last step of open() 3533c981a482SAl Viro */ 3534c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3535c981a482SAl Viro struct file *file, const struct open_flags *op) 3536c981a482SAl Viro { 3537549c7297SChristian Brauner struct user_namespace *mnt_userns; 3538c981a482SAl Viro int open_flag = op->open_flag; 3539c981a482SAl Viro bool do_truncate; 3540c981a482SAl Viro int acc_mode; 3541c981a482SAl Viro int error; 3542c981a482SAl Viro 3543ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3544ff326a32SAl Viro error = complete_walk(nd); 3545ff326a32SAl Viro if (error) 3546ff326a32SAl Viro return error; 3547ff326a32SAl Viro } 3548973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 354976ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3550549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 355130aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3552b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3553b94e0b32SAl Viro return -EEXIST; 355430aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3555c5971b8cSAl Viro return -EISDIR; 3556549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 355730aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 355830aba665SSalvatore Mesoraca if (unlikely(error)) 3559c5971b8cSAl Viro return error; 356030aba665SSalvatore Mesoraca } 356144b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3562c5971b8cSAl Viro return -ENOTDIR; 35636c0d46c4SAl Viro 35648795e7d4SAl Viro do_truncate = false; 35658795e7d4SAl Viro acc_mode = op->acc_mode; 35665a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 35675a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 35685a2d3eddSAl Viro open_flag &= ~O_TRUNC; 35695a2d3eddSAl Viro acc_mode = 0; 35708795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 35710f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 35720f9d1a10SAl Viro if (error) 3573c5971b8cSAl Viro return error; 35748795e7d4SAl Viro do_truncate = true; 35750f9d1a10SAl Viro } 3576549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 35778795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3578ae2bb293SAl Viro error = vfs_open(&nd->path, file); 35798795e7d4SAl Viro if (!error) 35806035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 35818795e7d4SAl Viro if (!error && do_truncate) 3582549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3583c80567c8SAl Viro if (unlikely(error > 0)) { 3584c80567c8SAl Viro WARN_ON(1); 3585c80567c8SAl Viro error = -EINVAL; 3586c80567c8SAl Viro } 35878795e7d4SAl Viro if (do_truncate) 35880f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3589c5971b8cSAl Viro return error; 3590fb1cc555SAl Viro } 3591fb1cc555SAl Viro 35926521f891SChristian Brauner /** 35936521f891SChristian Brauner * vfs_tmpfile - create tmpfile 35946521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 35956521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 35966521f891SChristian Brauner * @mode: mode of the new tmpfile 35972111c3c0SRandy Dunlap * @open_flag: flags 35986521f891SChristian Brauner * 35996521f891SChristian Brauner * Create a temporary file. 36006521f891SChristian Brauner * 36016521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 36026521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 36036521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 36046521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 36056521f891SChristian Brauner * raw inode simply passs init_user_ns. 36066521f891SChristian Brauner */ 36076521f891SChristian Brauner struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, 36086521f891SChristian Brauner struct dentry *dentry, umode_t mode, int open_flag) 3609af7bd4dcSAmir Goldstein { 3610af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3611af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3612af7bd4dcSAmir Goldstein struct inode *inode; 3613af7bd4dcSAmir Goldstein int error; 3614af7bd4dcSAmir Goldstein 3615af7bd4dcSAmir Goldstein /* we want directory to be writable */ 36166521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3617af7bd4dcSAmir Goldstein if (error) 3618af7bd4dcSAmir Goldstein goto out_err; 3619af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3620af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3621af7bd4dcSAmir Goldstein goto out_err; 3622af7bd4dcSAmir Goldstein error = -ENOMEM; 3623cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3624af7bd4dcSAmir Goldstein if (unlikely(!child)) 3625af7bd4dcSAmir Goldstein goto out_err; 3626*1639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode); 3627549c7297SChristian Brauner error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); 3628af7bd4dcSAmir Goldstein if (error) 3629af7bd4dcSAmir Goldstein goto out_err; 3630af7bd4dcSAmir Goldstein error = -ENOENT; 3631af7bd4dcSAmir Goldstein inode = child->d_inode; 3632af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3633af7bd4dcSAmir Goldstein goto out_err; 3634af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3635af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3636af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3637af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3638af7bd4dcSAmir Goldstein } 3639a2d2329eSChristian Brauner ima_post_create_tmpfile(mnt_userns, inode); 3640af7bd4dcSAmir Goldstein return child; 3641af7bd4dcSAmir Goldstein 3642af7bd4dcSAmir Goldstein out_err: 3643af7bd4dcSAmir Goldstein dput(child); 3644af7bd4dcSAmir Goldstein return ERR_PTR(error); 3645af7bd4dcSAmir Goldstein } 3646af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3647af7bd4dcSAmir Goldstein 3648c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 364960545d0dSAl Viro const struct open_flags *op, 36503ec2eef1SAl Viro struct file *file) 365160545d0dSAl Viro { 36526521f891SChristian Brauner struct user_namespace *mnt_userns; 3653625b6d10SAl Viro struct dentry *child; 3654625b6d10SAl Viro struct path path; 3655c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 365660545d0dSAl Viro if (unlikely(error)) 365760545d0dSAl Viro return error; 3658625b6d10SAl Viro error = mnt_want_write(path.mnt); 365960545d0dSAl Viro if (unlikely(error)) 366060545d0dSAl Viro goto out; 36616521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 36626521f891SChristian Brauner child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); 3663af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3664684e73beSHirofumi Nakagawa if (IS_ERR(child)) 366560545d0dSAl Viro goto out2; 3666625b6d10SAl Viro dput(path.dentry); 3667625b6d10SAl Viro path.dentry = child; 3668c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 366969a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3670549c7297SChristian Brauner error = may_open(mnt_userns, &path, 0, op->open_flag); 36711e8f44f1SAl Viro if (!error) 36721e8f44f1SAl Viro error = vfs_open(&path, file); 367360545d0dSAl Viro out2: 3674625b6d10SAl Viro mnt_drop_write(path.mnt); 367560545d0dSAl Viro out: 3676625b6d10SAl Viro path_put(&path); 367760545d0dSAl Viro return error; 367860545d0dSAl Viro } 367960545d0dSAl Viro 36806ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 36816ac08709SAl Viro { 36826ac08709SAl Viro struct path path; 36836ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 36846ac08709SAl Viro if (!error) { 36856ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3686ae2bb293SAl Viro error = vfs_open(&path, file); 36876ac08709SAl Viro path_put(&path); 36886ac08709SAl Viro } 36896ac08709SAl Viro return error; 36906ac08709SAl Viro } 36916ac08709SAl Viro 3692c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3693c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 36941da177e4SLinus Torvalds { 369530d90494SAl Viro struct file *file; 369613aab428SAl Viro int error; 369731e6b01fSNick Piggin 3698ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 36991afc99beSAl Viro if (IS_ERR(file)) 37001afc99beSAl Viro return file; 370131e6b01fSNick Piggin 3702bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 37033ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 37045f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 37056ac08709SAl Viro error = do_o_path(nd, flags, file); 37065f336e72SAl Viro } else { 37075f336e72SAl Viro const char *s = path_init(nd, flags); 37083bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3709c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 37101ccac622SAl Viro ; 3711c5971b8cSAl Viro if (!error) 3712c5971b8cSAl Viro error = do_open(nd, file, op); 3713deb106c6SAl Viro terminate_walk(nd); 37145f336e72SAl Viro } 37157c1c01ecSAl Viro if (likely(!error)) { 3716aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 37177c1c01ecSAl Viro return file; 37187c1c01ecSAl Viro WARN_ON(1); 37197c1c01ecSAl Viro error = -EINVAL; 3720015c3bbcSMiklos Szeredi } 37217c1c01ecSAl Viro fput(file); 37222675a4ebSAl Viro if (error == -EOPENSTALE) { 37232675a4ebSAl Viro if (flags & LOOKUP_RCU) 37242675a4ebSAl Viro error = -ECHILD; 37252675a4ebSAl Viro else 37262675a4ebSAl Viro error = -ESTALE; 37272675a4ebSAl Viro } 37287c1c01ecSAl Viro return ERR_PTR(error); 3729de459215SKirill Korotaev } 37301da177e4SLinus Torvalds 3731669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3732f9652e10SAl Viro const struct open_flags *op) 373313aab428SAl Viro { 37349883d185SAl Viro struct nameidata nd; 3735f9652e10SAl Viro int flags = op->lookup_flags; 373613aab428SAl Viro struct file *filp; 373713aab428SAl Viro 373806422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3739c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 374013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3741c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 374213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3743c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 37449883d185SAl Viro restore_nameidata(); 374513aab428SAl Viro return filp; 374613aab428SAl Viro } 374713aab428SAl Viro 3748ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3749f9652e10SAl Viro const char *name, const struct open_flags *op) 375073d049a4SAl Viro { 37519883d185SAl Viro struct nameidata nd; 375273d049a4SAl Viro struct file *file; 375351689104SPaul Moore struct filename *filename; 3754bcba1e7dSAl Viro int flags = op->lookup_flags; 375573d049a4SAl Viro 3756ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 375773d049a4SAl Viro return ERR_PTR(-ELOOP); 375873d049a4SAl Viro 375951689104SPaul Moore filename = getname_kernel(name); 3760a1c83681SViresh Kumar if (IS_ERR(filename)) 376151689104SPaul Moore return ERR_CAST(filename); 376251689104SPaul Moore 376306422964SAl Viro set_nameidata(&nd, -1, filename, root); 3764c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 376573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3766c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 376773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3768c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 37699883d185SAl Viro restore_nameidata(); 377051689104SPaul Moore putname(filename); 377173d049a4SAl Viro return file; 377273d049a4SAl Viro } 377373d049a4SAl Viro 3774b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 37751ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 37761da177e4SLinus Torvalds { 3777c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3778391172c4SAl Viro struct qstr last; 3779b3d4650dSNeilBrown bool want_dir = lookup_flags & LOOKUP_DIRECTORY; 3780b3d4650dSNeilBrown unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; 3781b3d4650dSNeilBrown unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; 3782391172c4SAl Viro int type; 3783c30dabfeSJan Kara int err2; 37841ac12b4bSJeff Layton int error; 37851ac12b4bSJeff Layton 3786b3d4650dSNeilBrown error = filename_parentat(dfd, name, reval_flag, path, &last, &type); 37870ee50b47SDmitry Kadashev if (error) 37880ee50b47SDmitry Kadashev return ERR_PTR(error); 37891da177e4SLinus Torvalds 3790c663e5d8SChristoph Hellwig /* 3791c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3792c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3793c663e5d8SChristoph Hellwig */ 37945c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3795ed75e95dSAl Viro goto out; 3796c663e5d8SChristoph Hellwig 3797c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3798391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3799c663e5d8SChristoph Hellwig /* 3800b3d4650dSNeilBrown * Do the final lookup. Suppress 'create' if there is a trailing 3801b3d4650dSNeilBrown * '/', and a directory wasn't requested. 3802c663e5d8SChristoph Hellwig */ 3803b3d4650dSNeilBrown if (last.name[last.len] && !want_dir) 3804b3d4650dSNeilBrown create_flags = 0; 38055955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3806b3d4650dSNeilBrown dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags); 38071da177e4SLinus Torvalds if (IS_ERR(dentry)) 3808a8104a9fSAl Viro goto unlock; 3809c663e5d8SChristoph Hellwig 3810a8104a9fSAl Viro error = -EEXIST; 3811b18825a7SDavid Howells if (d_is_positive(dentry)) 3812a8104a9fSAl Viro goto fail; 3813b18825a7SDavid Howells 3814c663e5d8SChristoph Hellwig /* 3815c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3816c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3817c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3818c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3819c663e5d8SChristoph Hellwig */ 3820b3d4650dSNeilBrown if (unlikely(!create_flags)) { 3821a8104a9fSAl Viro error = -ENOENT; 3822ed75e95dSAl Viro goto fail; 3823e9baf6e5SAl Viro } 3824c30dabfeSJan Kara if (unlikely(err2)) { 3825c30dabfeSJan Kara error = err2; 3826a8104a9fSAl Viro goto fail; 3827c30dabfeSJan Kara } 3828e9baf6e5SAl Viro return dentry; 38291da177e4SLinus Torvalds fail: 3830a8104a9fSAl Viro dput(dentry); 3831a8104a9fSAl Viro dentry = ERR_PTR(error); 3832a8104a9fSAl Viro unlock: 38335955102cSAl Viro inode_unlock(path->dentry->d_inode); 3834c30dabfeSJan Kara if (!err2) 3835391172c4SAl Viro mnt_drop_write(path->mnt); 3836ed75e95dSAl Viro out: 3837391172c4SAl Viro path_put(path); 3838ed75e95dSAl Viro return dentry; 3839dae6ad8fSAl Viro } 3840fa14a0b8SAl Viro 3841fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3842fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3843fa14a0b8SAl Viro { 3844b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3845b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3846b4a4f213SStephen Brennan 3847b4a4f213SStephen Brennan putname(filename); 3848b4a4f213SStephen Brennan return res; 3849fa14a0b8SAl Viro } 3850dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3851dae6ad8fSAl Viro 3852921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3853921a1650SAl Viro { 3854921a1650SAl Viro dput(dentry); 38555955102cSAl Viro inode_unlock(path->dentry->d_inode); 3856a8104a9fSAl Viro mnt_drop_write(path->mnt); 3857921a1650SAl Viro path_put(path); 3858921a1650SAl Viro } 3859921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3860921a1650SAl Viro 3861520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 38621ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3863dae6ad8fSAl Viro { 3864b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3865b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3866b4a4f213SStephen Brennan 3867b4a4f213SStephen Brennan putname(filename); 3868b4a4f213SStephen Brennan return res; 3869dae6ad8fSAl Viro } 3870dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3871dae6ad8fSAl Viro 38726521f891SChristian Brauner /** 38736521f891SChristian Brauner * vfs_mknod - create device node or file 38746521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 38756521f891SChristian Brauner * @dir: inode of @dentry 38766521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 38776521f891SChristian Brauner * @mode: mode of the new device node or file 38786521f891SChristian Brauner * @dev: device number of device to create 38796521f891SChristian Brauner * 38806521f891SChristian Brauner * Create a device node or file. 38816521f891SChristian Brauner * 38826521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 38836521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 38846521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 38856521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 38866521f891SChristian Brauner * raw inode simply passs init_user_ns. 38876521f891SChristian Brauner */ 38886521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 38896521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 38901da177e4SLinus Torvalds { 3891a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 38926521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 38931da177e4SLinus Torvalds 38941da177e4SLinus Torvalds if (error) 38951da177e4SLinus Torvalds return error; 38961da177e4SLinus Torvalds 3897a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3898a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 38991da177e4SLinus Torvalds return -EPERM; 39001da177e4SLinus Torvalds 3901acfa4380SAl Viro if (!dir->i_op->mknod) 39021da177e4SLinus Torvalds return -EPERM; 39031da177e4SLinus Torvalds 3904*1639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode); 390508ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 390608ce5f16SSerge E. Hallyn if (error) 390708ce5f16SSerge E. Hallyn return error; 390808ce5f16SSerge E. Hallyn 39091da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 39101da177e4SLinus Torvalds if (error) 39111da177e4SLinus Torvalds return error; 39121da177e4SLinus Torvalds 3913549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3914a74574aaSStephen Smalley if (!error) 3915f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39161da177e4SLinus Torvalds return error; 39171da177e4SLinus Torvalds } 39184d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 39191da177e4SLinus Torvalds 3920f69aac00SAl Viro static int may_mknod(umode_t mode) 3921463c3197SDave Hansen { 3922463c3197SDave Hansen switch (mode & S_IFMT) { 3923463c3197SDave Hansen case S_IFREG: 3924463c3197SDave Hansen case S_IFCHR: 3925463c3197SDave Hansen case S_IFBLK: 3926463c3197SDave Hansen case S_IFIFO: 3927463c3197SDave Hansen case S_IFSOCK: 3928463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3929463c3197SDave Hansen return 0; 3930463c3197SDave Hansen case S_IFDIR: 3931463c3197SDave Hansen return -EPERM; 3932463c3197SDave Hansen default: 3933463c3197SDave Hansen return -EINVAL; 3934463c3197SDave Hansen } 3935463c3197SDave Hansen } 3936463c3197SDave Hansen 393745f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 393887c4e192SDominik Brodowski unsigned int dev) 39391da177e4SLinus Torvalds { 39406521f891SChristian Brauner struct user_namespace *mnt_userns; 39411da177e4SLinus Torvalds struct dentry *dentry; 3942dae6ad8fSAl Viro struct path path; 3943dae6ad8fSAl Viro int error; 3944972567f1SJeff Layton unsigned int lookup_flags = 0; 39451da177e4SLinus Torvalds 39468e4bfca1SAl Viro error = may_mknod(mode); 39478e4bfca1SAl Viro if (error) 39487797251bSDmitry Kadashev goto out1; 3949972567f1SJeff Layton retry: 3950b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 39517797251bSDmitry Kadashev error = PTR_ERR(dentry); 3952dae6ad8fSAl Viro if (IS_ERR(dentry)) 39537797251bSDmitry Kadashev goto out1; 39542ad94ae6SAl Viro 3955*1639a49cSYang Xu error = security_path_mknod(&path, dentry, 3956*1639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode), dev); 3957be6d3e56SKentaro Takeda if (error) 39587797251bSDmitry Kadashev goto out2; 39596521f891SChristian Brauner 39606521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 39611da177e4SLinus Torvalds switch (mode & S_IFMT) { 39621da177e4SLinus Torvalds case 0: case S_IFREG: 39636521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 39646521f891SChristian Brauner dentry, mode, true); 396505d1a717SMimi Zohar if (!error) 3966a2d2329eSChristian Brauner ima_post_path_mknod(mnt_userns, dentry); 39671da177e4SLinus Torvalds break; 39681da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 39696521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 39706521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 39711da177e4SLinus Torvalds break; 39721da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 39736521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 39746521f891SChristian Brauner dentry, mode, 0); 39751da177e4SLinus Torvalds break; 39761da177e4SLinus Torvalds } 39777797251bSDmitry Kadashev out2: 3978921a1650SAl Viro done_path_create(&path, dentry); 3979972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3980972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3981972567f1SJeff Layton goto retry; 3982972567f1SJeff Layton } 39837797251bSDmitry Kadashev out1: 39847797251bSDmitry Kadashev putname(name); 39851da177e4SLinus Torvalds return error; 39861da177e4SLinus Torvalds } 39871da177e4SLinus Torvalds 398887c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 398987c4e192SDominik Brodowski unsigned int, dev) 399087c4e192SDominik Brodowski { 39917797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 399287c4e192SDominik Brodowski } 399387c4e192SDominik Brodowski 39948208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 39955590ff0dSUlrich Drepper { 39967797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 39975590ff0dSUlrich Drepper } 39985590ff0dSUlrich Drepper 39996521f891SChristian Brauner /** 40006521f891SChristian Brauner * vfs_mkdir - create directory 40016521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 40026521f891SChristian Brauner * @dir: inode of @dentry 40036521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 40046521f891SChristian Brauner * @mode: mode of the new directory 40056521f891SChristian Brauner * 40066521f891SChristian Brauner * Create a directory. 40076521f891SChristian Brauner * 40086521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 40096521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 40106521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 40116521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 40126521f891SChristian Brauner * raw inode simply passs init_user_ns. 40136521f891SChristian Brauner */ 40146521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 40156521f891SChristian Brauner struct dentry *dentry, umode_t mode) 40161da177e4SLinus Torvalds { 40176521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 40188de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40191da177e4SLinus Torvalds 40201da177e4SLinus Torvalds if (error) 40211da177e4SLinus Torvalds return error; 40221da177e4SLinus Torvalds 4023acfa4380SAl Viro if (!dir->i_op->mkdir) 40241da177e4SLinus Torvalds return -EPERM; 40251da177e4SLinus Torvalds 4026*1639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IRWXUGO | S_ISVTX, 0); 40271da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 40281da177e4SLinus Torvalds if (error) 40291da177e4SLinus Torvalds return error; 40301da177e4SLinus Torvalds 40318de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 40328de52778SAl Viro return -EMLINK; 40338de52778SAl Viro 4034549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 4035a74574aaSStephen Smalley if (!error) 4036f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 40371da177e4SLinus Torvalds return error; 40381da177e4SLinus Torvalds } 40394d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 40401da177e4SLinus Torvalds 404145f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 40421da177e4SLinus Torvalds { 40436902d925SDave Hansen struct dentry *dentry; 4044dae6ad8fSAl Viro struct path path; 4045dae6ad8fSAl Viro int error; 4046b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 40471da177e4SLinus Torvalds 4048b76d8b82SJeff Layton retry: 4049b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 4050584d3226SDmitry Kadashev error = PTR_ERR(dentry); 40516902d925SDave Hansen if (IS_ERR(dentry)) 4052584d3226SDmitry Kadashev goto out_putname; 40536902d925SDave Hansen 4054*1639a49cSYang Xu error = security_path_mkdir(&path, dentry, 4055*1639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode)); 40566521f891SChristian Brauner if (!error) { 40576521f891SChristian Brauner struct user_namespace *mnt_userns; 40586521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4059549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 4060549c7297SChristian Brauner mode); 40616521f891SChristian Brauner } 4062921a1650SAl Viro done_path_create(&path, dentry); 4063b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 4064b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 4065b76d8b82SJeff Layton goto retry; 4066b76d8b82SJeff Layton } 4067584d3226SDmitry Kadashev out_putname: 4068584d3226SDmitry Kadashev putname(name); 40691da177e4SLinus Torvalds return error; 40701da177e4SLinus Torvalds } 40711da177e4SLinus Torvalds 40720101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 40730101db7aSDominik Brodowski { 4074584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 40750101db7aSDominik Brodowski } 40760101db7aSDominik Brodowski 4077a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 40785590ff0dSUlrich Drepper { 4079584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 40805590ff0dSUlrich Drepper } 40815590ff0dSUlrich Drepper 40826521f891SChristian Brauner /** 40836521f891SChristian Brauner * vfs_rmdir - remove directory 40846521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 40856521f891SChristian Brauner * @dir: inode of @dentry 40866521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 40876521f891SChristian Brauner * 40886521f891SChristian Brauner * Remove a directory. 40896521f891SChristian Brauner * 40906521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 40916521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 40926521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 40936521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 40946521f891SChristian Brauner * raw inode simply passs init_user_ns. 40956521f891SChristian Brauner */ 40966521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 40976521f891SChristian Brauner struct dentry *dentry) 40981da177e4SLinus Torvalds { 40996521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 41001da177e4SLinus Torvalds 41011da177e4SLinus Torvalds if (error) 41021da177e4SLinus Torvalds return error; 41031da177e4SLinus Torvalds 4104acfa4380SAl Viro if (!dir->i_op->rmdir) 41051da177e4SLinus Torvalds return -EPERM; 41061da177e4SLinus Torvalds 41071d2ef590SAl Viro dget(dentry); 41085955102cSAl Viro inode_lock(dentry->d_inode); 4109912dbc15SSage Weil 41101da177e4SLinus Torvalds error = -EBUSY; 41111bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 41121bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4113912dbc15SSage Weil goto out; 4114912dbc15SSage Weil 41151da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4116912dbc15SSage Weil if (error) 4117912dbc15SSage Weil goto out; 4118912dbc15SSage Weil 41191da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4120912dbc15SSage Weil if (error) 4121912dbc15SSage Weil goto out; 4122912dbc15SSage Weil 41238767712fSAl Viro shrink_dcache_parent(dentry); 41241da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4125d83c49f3SAl Viro dont_mount(dentry); 41268ed936b5SEric W. Biederman detach_mounts(dentry); 41271da177e4SLinus Torvalds 4128912dbc15SSage Weil out: 41295955102cSAl Viro inode_unlock(dentry->d_inode); 41301d2ef590SAl Viro dput(dentry); 4131912dbc15SSage Weil if (!error) 4132a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 41331da177e4SLinus Torvalds return error; 41341da177e4SLinus Torvalds } 41354d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 41361da177e4SLinus Torvalds 413745f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 41381da177e4SLinus Torvalds { 41396521f891SChristian Brauner struct user_namespace *mnt_userns; 41400ee50b47SDmitry Kadashev int error; 41411da177e4SLinus Torvalds struct dentry *dentry; 4142f5beed75SAl Viro struct path path; 4143f5beed75SAl Viro struct qstr last; 4144f5beed75SAl Viro int type; 4145c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4146c6ee9206SJeff Layton retry: 4147c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 41480ee50b47SDmitry Kadashev if (error) 41490ee50b47SDmitry Kadashev goto exit1; 41501da177e4SLinus Torvalds 4151f5beed75SAl Viro switch (type) { 41521da177e4SLinus Torvalds case LAST_DOTDOT: 41531da177e4SLinus Torvalds error = -ENOTEMPTY; 41540ee50b47SDmitry Kadashev goto exit2; 41551da177e4SLinus Torvalds case LAST_DOT: 41561da177e4SLinus Torvalds error = -EINVAL; 41570ee50b47SDmitry Kadashev goto exit2; 41581da177e4SLinus Torvalds case LAST_ROOT: 41591da177e4SLinus Torvalds error = -EBUSY; 41600ee50b47SDmitry Kadashev goto exit2; 41611da177e4SLinus Torvalds } 41620612d9fbSOGAWA Hirofumi 4163f5beed75SAl Viro error = mnt_want_write(path.mnt); 4164c30dabfeSJan Kara if (error) 41650ee50b47SDmitry Kadashev goto exit2; 41660612d9fbSOGAWA Hirofumi 41675955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4168f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 41691da177e4SLinus Torvalds error = PTR_ERR(dentry); 41706902d925SDave Hansen if (IS_ERR(dentry)) 41710ee50b47SDmitry Kadashev goto exit3; 4172e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4173e6bc45d6STheodore Ts'o error = -ENOENT; 41740ee50b47SDmitry Kadashev goto exit4; 4175e6bc45d6STheodore Ts'o } 4176f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4177be6d3e56SKentaro Takeda if (error) 41780ee50b47SDmitry Kadashev goto exit4; 41796521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 41806521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 41810ee50b47SDmitry Kadashev exit4: 41821da177e4SLinus Torvalds dput(dentry); 41830ee50b47SDmitry Kadashev exit3: 41845955102cSAl Viro inode_unlock(path.dentry->d_inode); 4185f5beed75SAl Viro mnt_drop_write(path.mnt); 41860ee50b47SDmitry Kadashev exit2: 4187f5beed75SAl Viro path_put(&path); 4188c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4189c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4190c6ee9206SJeff Layton goto retry; 4191c6ee9206SJeff Layton } 41920ee50b47SDmitry Kadashev exit1: 419324fb33d4SAl Viro putname(name); 41941da177e4SLinus Torvalds return error; 41951da177e4SLinus Torvalds } 41961da177e4SLinus Torvalds 41973cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 41985590ff0dSUlrich Drepper { 4199e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 42005590ff0dSUlrich Drepper } 42015590ff0dSUlrich Drepper 4202b21996e3SJ. Bruce Fields /** 4203b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 42046521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 4205b21996e3SJ. Bruce Fields * @dir: parent directory 4206b21996e3SJ. Bruce Fields * @dentry: victim 4207b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4208b21996e3SJ. Bruce Fields * 4209b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4210b21996e3SJ. Bruce Fields * 4211b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4212b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4213b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4214b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4215b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4216b21996e3SJ. Bruce Fields * 4217b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4218b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4219b21996e3SJ. Bruce Fields * to be NFS exported. 42206521f891SChristian Brauner * 42216521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 42226521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42236521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42246521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42256521f891SChristian Brauner * raw inode simply passs init_user_ns. 4226b21996e3SJ. Bruce Fields */ 42276521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 42286521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 42291da177e4SLinus Torvalds { 42309accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 42316521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 42321da177e4SLinus Torvalds 42331da177e4SLinus Torvalds if (error) 42341da177e4SLinus Torvalds return error; 42351da177e4SLinus Torvalds 4236acfa4380SAl Viro if (!dir->i_op->unlink) 42371da177e4SLinus Torvalds return -EPERM; 42381da177e4SLinus Torvalds 42395955102cSAl Viro inode_lock(target); 424051cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 424151cc3a66SHugh Dickins error = -EPERM; 424251cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 42431da177e4SLinus Torvalds error = -EBUSY; 42441da177e4SLinus Torvalds else { 42451da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4246bec1052eSAl Viro if (!error) { 42475a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 42485a14696cSJ. Bruce Fields if (error) 4249b21996e3SJ. Bruce Fields goto out; 42501da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 42518ed936b5SEric W. Biederman if (!error) { 4252d83c49f3SAl Viro dont_mount(dentry); 42538ed936b5SEric W. Biederman detach_mounts(dentry); 42548ed936b5SEric W. Biederman } 4255bec1052eSAl Viro } 42561da177e4SLinus Torvalds } 4257b21996e3SJ. Bruce Fields out: 42585955102cSAl Viro inode_unlock(target); 42591da177e4SLinus Torvalds 42601da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4261a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4262a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4263a37d9a17SAmir Goldstein } else if (!error) { 42649accbb97SJ. Bruce Fields fsnotify_link_count(target); 4265a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 42661da177e4SLinus Torvalds } 42670eeca283SRobert Love 42681da177e4SLinus Torvalds return error; 42691da177e4SLinus Torvalds } 42704d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 42711da177e4SLinus Torvalds 42721da177e4SLinus Torvalds /* 42731da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 42741b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 42751da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 42761da177e4SLinus Torvalds * while waiting on the I/O. 42771da177e4SLinus Torvalds */ 427845f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 42791da177e4SLinus Torvalds { 42802ad94ae6SAl Viro int error; 42811da177e4SLinus Torvalds struct dentry *dentry; 4282f5beed75SAl Viro struct path path; 4283f5beed75SAl Viro struct qstr last; 4284f5beed75SAl Viro int type; 42851da177e4SLinus Torvalds struct inode *inode = NULL; 4286b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 42875d18f813SJeff Layton unsigned int lookup_flags = 0; 42885d18f813SJeff Layton retry: 4289c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 42900ee50b47SDmitry Kadashev if (error) 42910ee50b47SDmitry Kadashev goto exit1; 42922ad94ae6SAl Viro 42931da177e4SLinus Torvalds error = -EISDIR; 4294f5beed75SAl Viro if (type != LAST_NORM) 42950ee50b47SDmitry Kadashev goto exit2; 42960612d9fbSOGAWA Hirofumi 4297f5beed75SAl Viro error = mnt_want_write(path.mnt); 4298c30dabfeSJan Kara if (error) 42990ee50b47SDmitry Kadashev goto exit2; 4300b21996e3SJ. Bruce Fields retry_deleg: 43015955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4302f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 43031da177e4SLinus Torvalds error = PTR_ERR(dentry); 43041da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 43056521f891SChristian Brauner struct user_namespace *mnt_userns; 43066521f891SChristian Brauner 43071da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4308f5beed75SAl Viro if (last.name[last.len]) 430950338b88STörök Edwin goto slashes; 43101da177e4SLinus Torvalds inode = dentry->d_inode; 4311b18825a7SDavid Howells if (d_is_negative(dentry)) 4312e6bc45d6STheodore Ts'o goto slashes; 43137de9c6eeSAl Viro ihold(inode); 4314f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4315be6d3e56SKentaro Takeda if (error) 43160ee50b47SDmitry Kadashev goto exit3; 43176521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4318549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4319549c7297SChristian Brauner &delegated_inode); 43200ee50b47SDmitry Kadashev exit3: 43211da177e4SLinus Torvalds dput(dentry); 43221da177e4SLinus Torvalds } 43235955102cSAl Viro inode_unlock(path.dentry->d_inode); 43241da177e4SLinus Torvalds if (inode) 43251da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4326b21996e3SJ. Bruce Fields inode = NULL; 4327b21996e3SJ. Bruce Fields if (delegated_inode) { 43285a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4329b21996e3SJ. Bruce Fields if (!error) 4330b21996e3SJ. Bruce Fields goto retry_deleg; 4331b21996e3SJ. Bruce Fields } 4332f5beed75SAl Viro mnt_drop_write(path.mnt); 43330ee50b47SDmitry Kadashev exit2: 4334f5beed75SAl Viro path_put(&path); 43355d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 43365d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 43375d18f813SJeff Layton inode = NULL; 43385d18f813SJeff Layton goto retry; 43395d18f813SJeff Layton } 43400ee50b47SDmitry Kadashev exit1: 4341da2f1362SChristoph Hellwig putname(name); 43421da177e4SLinus Torvalds return error; 43431da177e4SLinus Torvalds 43441da177e4SLinus Torvalds slashes: 4345b18825a7SDavid Howells if (d_is_negative(dentry)) 4346b18825a7SDavid Howells error = -ENOENT; 434744b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4348b18825a7SDavid Howells error = -EISDIR; 4349b18825a7SDavid Howells else 4350b18825a7SDavid Howells error = -ENOTDIR; 43510ee50b47SDmitry Kadashev goto exit3; 43521da177e4SLinus Torvalds } 43531da177e4SLinus Torvalds 43542e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 43555590ff0dSUlrich Drepper { 43565590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 43575590ff0dSUlrich Drepper return -EINVAL; 43585590ff0dSUlrich Drepper 43595590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4360e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4361da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 43625590ff0dSUlrich Drepper } 43635590ff0dSUlrich Drepper 43643480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 43655590ff0dSUlrich Drepper { 4366da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 43675590ff0dSUlrich Drepper } 43685590ff0dSUlrich Drepper 43696521f891SChristian Brauner /** 43706521f891SChristian Brauner * vfs_symlink - create symlink 43716521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 43726521f891SChristian Brauner * @dir: inode of @dentry 43736521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 43746521f891SChristian Brauner * @oldname: name of the file to link to 43756521f891SChristian Brauner * 43766521f891SChristian Brauner * Create a symlink. 43776521f891SChristian Brauner * 43786521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 43796521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 43806521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 43816521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 43826521f891SChristian Brauner * raw inode simply passs init_user_ns. 43836521f891SChristian Brauner */ 43846521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 43856521f891SChristian Brauner struct dentry *dentry, const char *oldname) 43861da177e4SLinus Torvalds { 43876521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 43881da177e4SLinus Torvalds 43891da177e4SLinus Torvalds if (error) 43901da177e4SLinus Torvalds return error; 43911da177e4SLinus Torvalds 4392acfa4380SAl Viro if (!dir->i_op->symlink) 43931da177e4SLinus Torvalds return -EPERM; 43941da177e4SLinus Torvalds 43951da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 43961da177e4SLinus Torvalds if (error) 43971da177e4SLinus Torvalds return error; 43981da177e4SLinus Torvalds 4399549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4400a74574aaSStephen Smalley if (!error) 4401f38aa942SAmy Griffis fsnotify_create(dir, dentry); 44021da177e4SLinus Torvalds return error; 44031da177e4SLinus Torvalds } 44044d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 44051da177e4SLinus Torvalds 44067a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 44071da177e4SLinus Torvalds { 44082ad94ae6SAl Viro int error; 44096902d925SDave Hansen struct dentry *dentry; 4410dae6ad8fSAl Viro struct path path; 4411f46d3567SJeff Layton unsigned int lookup_flags = 0; 44121da177e4SLinus Torvalds 4413da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4414da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4415da2d0cedSDmitry Kadashev goto out_putnames; 4416da2d0cedSDmitry Kadashev } 4417f46d3567SJeff Layton retry: 4418b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 44191da177e4SLinus Torvalds error = PTR_ERR(dentry); 44206902d925SDave Hansen if (IS_ERR(dentry)) 4421da2d0cedSDmitry Kadashev goto out_putnames; 44226902d925SDave Hansen 442391a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 44246521f891SChristian Brauner if (!error) { 44256521f891SChristian Brauner struct user_namespace *mnt_userns; 44266521f891SChristian Brauner 44276521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 44286521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 44296521f891SChristian Brauner from->name); 44306521f891SChristian Brauner } 4431921a1650SAl Viro done_path_create(&path, dentry); 4432f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4433f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4434f46d3567SJeff Layton goto retry; 4435f46d3567SJeff Layton } 4436da2d0cedSDmitry Kadashev out_putnames: 4437da2d0cedSDmitry Kadashev putname(to); 44381da177e4SLinus Torvalds putname(from); 44391da177e4SLinus Torvalds return error; 44401da177e4SLinus Torvalds } 44411da177e4SLinus Torvalds 4442b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4443b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4444b724e846SDominik Brodowski { 4445da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4446b724e846SDominik Brodowski } 4447b724e846SDominik Brodowski 44483480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 44495590ff0dSUlrich Drepper { 4450da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 44515590ff0dSUlrich Drepper } 44525590ff0dSUlrich Drepper 4453146a8595SJ. Bruce Fields /** 4454146a8595SJ. Bruce Fields * vfs_link - create a new link 4455146a8595SJ. Bruce Fields * @old_dentry: object to be linked 44566521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4457146a8595SJ. Bruce Fields * @dir: new parent 4458146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4459146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4460146a8595SJ. Bruce Fields * 4461146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4462146a8595SJ. Bruce Fields * 4463146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4464146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4465146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4466146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4467146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4468146a8595SJ. Bruce Fields * 4469146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4470146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4471146a8595SJ. Bruce Fields * to be NFS exported. 44726521f891SChristian Brauner * 44736521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 44746521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 44756521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 44766521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 44776521f891SChristian Brauner * raw inode simply passs init_user_ns. 4478146a8595SJ. Bruce Fields */ 44796521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 44806521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 44816521f891SChristian Brauner struct inode **delegated_inode) 44821da177e4SLinus Torvalds { 44831da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 44848de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 44851da177e4SLinus Torvalds int error; 44861da177e4SLinus Torvalds 44871da177e4SLinus Torvalds if (!inode) 44881da177e4SLinus Torvalds return -ENOENT; 44891da177e4SLinus Torvalds 44906521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 44911da177e4SLinus Torvalds if (error) 44921da177e4SLinus Torvalds return error; 44931da177e4SLinus Torvalds 44941da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 44951da177e4SLinus Torvalds return -EXDEV; 44961da177e4SLinus Torvalds 44971da177e4SLinus Torvalds /* 44981da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 44991da177e4SLinus Torvalds */ 45001da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 45011da177e4SLinus Torvalds return -EPERM; 45020bd23d09SEric W. Biederman /* 45030bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 45040bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 45050bd23d09SEric W. Biederman * the vfs. 45060bd23d09SEric W. Biederman */ 45076521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 45080bd23d09SEric W. Biederman return -EPERM; 4509acfa4380SAl Viro if (!dir->i_op->link) 45101da177e4SLinus Torvalds return -EPERM; 45117e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 45121da177e4SLinus Torvalds return -EPERM; 45131da177e4SLinus Torvalds 45141da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 45151da177e4SLinus Torvalds if (error) 45161da177e4SLinus Torvalds return error; 45171da177e4SLinus Torvalds 45185955102cSAl Viro inode_lock(inode); 4519aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4520f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4521aae8a97dSAneesh Kumar K.V error = -ENOENT; 45228de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 45238de52778SAl Viro error = -EMLINK; 4524146a8595SJ. Bruce Fields else { 4525146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4526146a8595SJ. Bruce Fields if (!error) 45271da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4528146a8595SJ. Bruce Fields } 4529f4e0c30cSAl Viro 4530f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4531f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4532f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4533f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4534f4e0c30cSAl Viro } 45355955102cSAl Viro inode_unlock(inode); 4536e31e14ecSStephen Smalley if (!error) 45377e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 45381da177e4SLinus Torvalds return error; 45391da177e4SLinus Torvalds } 45404d359507SAl Viro EXPORT_SYMBOL(vfs_link); 45411da177e4SLinus Torvalds 45421da177e4SLinus Torvalds /* 45431da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 45441da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 45451da177e4SLinus Torvalds * newname. --KAB 45461da177e4SLinus Torvalds * 45471da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 45481da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 45491da177e4SLinus Torvalds * and other special files. --ADM 45501da177e4SLinus Torvalds */ 4551cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4552020250f3SDmitry Kadashev struct filename *new, int flags) 45531da177e4SLinus Torvalds { 45546521f891SChristian Brauner struct user_namespace *mnt_userns; 45551da177e4SLinus Torvalds struct dentry *new_dentry; 4556dae6ad8fSAl Viro struct path old_path, new_path; 4557146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 455811a7b371SAneesh Kumar K.V int how = 0; 45591da177e4SLinus Torvalds int error; 45601da177e4SLinus Torvalds 4561020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4562020250f3SDmitry Kadashev error = -EINVAL; 4563020250f3SDmitry Kadashev goto out_putnames; 4564020250f3SDmitry Kadashev } 456511a7b371SAneesh Kumar K.V /* 4566f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4567f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4568f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 456911a7b371SAneesh Kumar K.V */ 4570020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4571020250f3SDmitry Kadashev error = -ENOENT; 4572020250f3SDmitry Kadashev goto out_putnames; 4573f0cc6ffbSLinus Torvalds } 4574c04030e1SUlrich Drepper 457511a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 457611a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4577442e31caSJeff Layton retry: 4578794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 45791da177e4SLinus Torvalds if (error) 4580020250f3SDmitry Kadashev goto out_putnames; 45812ad94ae6SAl Viro 4582b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4583442e31caSJeff Layton (how & LOOKUP_REVAL)); 45841da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45856902d925SDave Hansen if (IS_ERR(new_dentry)) 4586020250f3SDmitry Kadashev goto out_putpath; 4587dae6ad8fSAl Viro 4588dae6ad8fSAl Viro error = -EXDEV; 4589dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4590dae6ad8fSAl Viro goto out_dput; 4591549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4592549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4593800179c9SKees Cook if (unlikely(error)) 4594800179c9SKees Cook goto out_dput; 4595dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4596be6d3e56SKentaro Takeda if (error) 4597a8104a9fSAl Viro goto out_dput; 45986521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 45996521f891SChristian Brauner new_dentry, &delegated_inode); 460075c3f29dSDave Hansen out_dput: 4601921a1650SAl Viro done_path_create(&new_path, new_dentry); 4602146a8595SJ. Bruce Fields if (delegated_inode) { 4603146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4604d22e6338SOleg Drokin if (!error) { 4605d22e6338SOleg Drokin path_put(&old_path); 4606146a8595SJ. Bruce Fields goto retry; 4607146a8595SJ. Bruce Fields } 4608d22e6338SOleg Drokin } 4609442e31caSJeff Layton if (retry_estale(error, how)) { 4610d22e6338SOleg Drokin path_put(&old_path); 4611442e31caSJeff Layton how |= LOOKUP_REVAL; 4612442e31caSJeff Layton goto retry; 4613442e31caSJeff Layton } 4614020250f3SDmitry Kadashev out_putpath: 46152d8f3038SAl Viro path_put(&old_path); 4616020250f3SDmitry Kadashev out_putnames: 4617020250f3SDmitry Kadashev putname(old); 4618020250f3SDmitry Kadashev putname(new); 46191da177e4SLinus Torvalds 46201da177e4SLinus Torvalds return error; 46211da177e4SLinus Torvalds } 46221da177e4SLinus Torvalds 462346ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 462446ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 462546ea89ebSDominik Brodowski { 4626020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4627020250f3SDmitry Kadashev newdfd, getname(newname), flags); 462846ea89ebSDominik Brodowski } 462946ea89ebSDominik Brodowski 46303480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 46315590ff0dSUlrich Drepper { 4632020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 46335590ff0dSUlrich Drepper } 46345590ff0dSUlrich Drepper 4635bc27027aSMiklos Szeredi /** 4636bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 46372111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4638bc27027aSMiklos Szeredi * 4639bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4640bc27027aSMiklos Szeredi * 4641bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4642bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4643bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4644bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4645bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4646bc27027aSMiklos Szeredi * so. 4647bc27027aSMiklos Szeredi * 4648bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4649bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4650bc27027aSMiklos Szeredi * to be NFS exported. 4651bc27027aSMiklos Szeredi * 46521da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 46531da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 46541da177e4SLinus Torvalds * Problems: 46550117d427SMauro Carvalho Chehab * 4656d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 46571da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 46581da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4659a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 46601da177e4SLinus Torvalds * story. 46616cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 46626cedba89SJ. Bruce Fields * and source (if it is not a directory). 46631b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 46641da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 46651da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4666a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 46671da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 46681da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 46691da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4670a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 46711da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 46721da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 46731da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4674e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 46751b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 46761da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4677c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 46781da177e4SLinus Torvalds * locking]. 46791da177e4SLinus Torvalds */ 46809fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 46811da177e4SLinus Torvalds { 46821da177e4SLinus Torvalds int error; 46839fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 46849fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 46859fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 46869fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 46879fe61450SChristian Brauner unsigned int flags = rd->flags; 4688bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4689bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4690bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4691da1ce067SMiklos Szeredi bool new_is_dir = false; 4692da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 469349d31c2fSAl Viro struct name_snapshot old_name; 46941da177e4SLinus Torvalds 46958d3e2936SMiklos Szeredi if (source == target) 46961da177e4SLinus Torvalds return 0; 46971da177e4SLinus Torvalds 46986521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 46991da177e4SLinus Torvalds if (error) 47001da177e4SLinus Torvalds return error; 47011da177e4SLinus Torvalds 4702da1ce067SMiklos Szeredi if (!target) { 47036521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4704da1ce067SMiklos Szeredi } else { 4705da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4706da1ce067SMiklos Szeredi 4707da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 47086521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 47096521f891SChristian Brauner new_dentry, is_dir); 4710da1ce067SMiklos Szeredi else 47116521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 47126521f891SChristian Brauner new_dentry, new_is_dir); 4713da1ce067SMiklos Szeredi } 47141da177e4SLinus Torvalds if (error) 47151da177e4SLinus Torvalds return error; 47161da177e4SLinus Torvalds 47172773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 47181da177e4SLinus Torvalds return -EPERM; 47191da177e4SLinus Torvalds 4720bc27027aSMiklos Szeredi /* 4721bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4722bc27027aSMiklos Szeredi * we'll need to flip '..'. 4723bc27027aSMiklos Szeredi */ 4724da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4725da1ce067SMiklos Szeredi if (is_dir) { 47266521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 472747291baaSChristian Brauner MAY_WRITE); 4728bc27027aSMiklos Szeredi if (error) 4729bc27027aSMiklos Szeredi return error; 4730bc27027aSMiklos Szeredi } 4731da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 47326521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 473347291baaSChristian Brauner MAY_WRITE); 4734da1ce067SMiklos Szeredi if (error) 4735da1ce067SMiklos Szeredi return error; 4736da1ce067SMiklos Szeredi } 4737da1ce067SMiklos Szeredi } 47380eeca283SRobert Love 47390b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 47400b3974ebSMiklos Szeredi flags); 4741bc27027aSMiklos Szeredi if (error) 4742bc27027aSMiklos Szeredi return error; 4743bc27027aSMiklos Szeredi 474449d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4745bc27027aSMiklos Szeredi dget(new_dentry); 4746da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4747bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4748bc27027aSMiklos Szeredi else if (target) 47495955102cSAl Viro inode_lock(target); 4750bc27027aSMiklos Szeredi 475151cc3a66SHugh Dickins error = -EPERM; 475251cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 475351cc3a66SHugh Dickins goto out; 475451cc3a66SHugh Dickins 4755bc27027aSMiklos Szeredi error = -EBUSY; 47567af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4757bc27027aSMiklos Szeredi goto out; 4758bc27027aSMiklos Szeredi 4759da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4760bc27027aSMiklos Szeredi error = -EMLINK; 4761da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4762bc27027aSMiklos Szeredi goto out; 4763da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4764da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4765da1ce067SMiklos Szeredi goto out; 4766da1ce067SMiklos Szeredi } 4767da1ce067SMiklos Szeredi if (!is_dir) { 4768bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4769bc27027aSMiklos Szeredi if (error) 4770bc27027aSMiklos Szeredi goto out; 4771da1ce067SMiklos Szeredi } 4772da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4773bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4774bc27027aSMiklos Szeredi if (error) 4775bc27027aSMiklos Szeredi goto out; 4776bc27027aSMiklos Szeredi } 4777549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4778520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4779bc27027aSMiklos Szeredi if (error) 4780bc27027aSMiklos Szeredi goto out; 4781bc27027aSMiklos Szeredi 4782da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 47838767712fSAl Viro if (is_dir) { 47848767712fSAl Viro shrink_dcache_parent(new_dentry); 4785bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 47868767712fSAl Viro } 4787bc27027aSMiklos Szeredi dont_mount(new_dentry); 47888ed936b5SEric W. Biederman detach_mounts(new_dentry); 4789bc27027aSMiklos Szeredi } 4790da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4791da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4792bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4793da1ce067SMiklos Szeredi else 4794da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4795da1ce067SMiklos Szeredi } 4796bc27027aSMiklos Szeredi out: 4797da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4798bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4799bc27027aSMiklos Szeredi else if (target) 48005955102cSAl Viro inode_unlock(target); 4801bc27027aSMiklos Szeredi dput(new_dentry); 4802da1ce067SMiklos Szeredi if (!error) { 4803f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4804da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4805da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4806f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4807da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4808da1ce067SMiklos Szeredi } 4809da1ce067SMiklos Szeredi } 481049d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 48110eeca283SRobert Love 48121da177e4SLinus Torvalds return error; 48131da177e4SLinus Torvalds } 48144d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 48151da177e4SLinus Torvalds 4816e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4817e886663cSJens Axboe struct filename *to, unsigned int flags) 48181da177e4SLinus Torvalds { 48199fe61450SChristian Brauner struct renamedata rd; 48201da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 48211da177e4SLinus Torvalds struct dentry *trap; 4822f5beed75SAl Viro struct path old_path, new_path; 4823f5beed75SAl Viro struct qstr old_last, new_last; 4824f5beed75SAl Viro int old_type, new_type; 48258e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4826f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4827c6a94284SJeff Layton bool should_retry = false; 4828e886663cSJens Axboe int error = -EINVAL; 4829520c8b16SMiklos Szeredi 48300d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 48310ee50b47SDmitry Kadashev goto put_names; 4832da1ce067SMiklos Szeredi 48330d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 48340d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 48350ee50b47SDmitry Kadashev goto put_names; 4836520c8b16SMiklos Szeredi 4837f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4838f5beed75SAl Viro target_flags = 0; 4839f5beed75SAl Viro 4840c6a94284SJeff Layton retry: 4841c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4842e886663cSJens Axboe &old_last, &old_type); 48430ee50b47SDmitry Kadashev if (error) 48440ee50b47SDmitry Kadashev goto put_names; 48451da177e4SLinus Torvalds 4846c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4847e886663cSJens Axboe &new_type); 48480ee50b47SDmitry Kadashev if (error) 48491da177e4SLinus Torvalds goto exit1; 48501da177e4SLinus Torvalds 48511da177e4SLinus Torvalds error = -EXDEV; 4852f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 48531da177e4SLinus Torvalds goto exit2; 48541da177e4SLinus Torvalds 48551da177e4SLinus Torvalds error = -EBUSY; 4856f5beed75SAl Viro if (old_type != LAST_NORM) 48571da177e4SLinus Torvalds goto exit2; 48581da177e4SLinus Torvalds 48590a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 48600a7c3937SMiklos Szeredi error = -EEXIST; 4861f5beed75SAl Viro if (new_type != LAST_NORM) 48621da177e4SLinus Torvalds goto exit2; 48631da177e4SLinus Torvalds 4864f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4865c30dabfeSJan Kara if (error) 4866c30dabfeSJan Kara goto exit2; 4867c30dabfeSJan Kara 48688e6d782cSJ. Bruce Fields retry_deleg: 4869f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 48701da177e4SLinus Torvalds 4871f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 48721da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 48731da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 48741da177e4SLinus Torvalds goto exit3; 48751da177e4SLinus Torvalds /* source must exist */ 48761da177e4SLinus Torvalds error = -ENOENT; 4877b18825a7SDavid Howells if (d_is_negative(old_dentry)) 48781da177e4SLinus Torvalds goto exit4; 4879f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 48801da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 48811da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 48821da177e4SLinus Torvalds goto exit4; 48830a7c3937SMiklos Szeredi error = -EEXIST; 48840a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 48850a7c3937SMiklos Szeredi goto exit5; 4886da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4887da1ce067SMiklos Szeredi error = -ENOENT; 4888da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4889da1ce067SMiklos Szeredi goto exit5; 4890da1ce067SMiklos Szeredi 4891da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4892da1ce067SMiklos Szeredi error = -ENOTDIR; 4893f5beed75SAl Viro if (new_last.name[new_last.len]) 4894da1ce067SMiklos Szeredi goto exit5; 4895da1ce067SMiklos Szeredi } 4896da1ce067SMiklos Szeredi } 48970a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 48980a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 48990a7c3937SMiklos Szeredi error = -ENOTDIR; 4900f5beed75SAl Viro if (old_last.name[old_last.len]) 49010a7c3937SMiklos Szeredi goto exit5; 4902f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 49030a7c3937SMiklos Szeredi goto exit5; 49040a7c3937SMiklos Szeredi } 49050a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 49060a7c3937SMiklos Szeredi error = -EINVAL; 49070a7c3937SMiklos Szeredi if (old_dentry == trap) 49080a7c3937SMiklos Szeredi goto exit5; 49091da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4910da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 49111da177e4SLinus Torvalds error = -ENOTEMPTY; 49121da177e4SLinus Torvalds if (new_dentry == trap) 49131da177e4SLinus Torvalds goto exit5; 49141da177e4SLinus Torvalds 4915f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4916f5beed75SAl Viro &new_path, new_dentry, flags); 4917be6d3e56SKentaro Takeda if (error) 4918c30dabfeSJan Kara goto exit5; 49199fe61450SChristian Brauner 49209fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 49219fe61450SChristian Brauner rd.old_dentry = old_dentry; 49226521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 49239fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 49249fe61450SChristian Brauner rd.new_dentry = new_dentry; 49256521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 49269fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 49279fe61450SChristian Brauner rd.flags = flags; 49289fe61450SChristian Brauner error = vfs_rename(&rd); 49291da177e4SLinus Torvalds exit5: 49301da177e4SLinus Torvalds dput(new_dentry); 49311da177e4SLinus Torvalds exit4: 49321da177e4SLinus Torvalds dput(old_dentry); 49331da177e4SLinus Torvalds exit3: 4934f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 49358e6d782cSJ. Bruce Fields if (delegated_inode) { 49368e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 49378e6d782cSJ. Bruce Fields if (!error) 49388e6d782cSJ. Bruce Fields goto retry_deleg; 49398e6d782cSJ. Bruce Fields } 4940f5beed75SAl Viro mnt_drop_write(old_path.mnt); 49411da177e4SLinus Torvalds exit2: 4942c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4943c6a94284SJeff Layton should_retry = true; 4944f5beed75SAl Viro path_put(&new_path); 49451da177e4SLinus Torvalds exit1: 4946f5beed75SAl Viro path_put(&old_path); 4947c6a94284SJeff Layton if (should_retry) { 4948c6a94284SJeff Layton should_retry = false; 4949c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4950c6a94284SJeff Layton goto retry; 4951c6a94284SJeff Layton } 49520ee50b47SDmitry Kadashev put_names: 4953e886663cSJens Axboe putname(from); 4954e886663cSJens Axboe putname(to); 49551da177e4SLinus Torvalds return error; 49561da177e4SLinus Torvalds } 49571da177e4SLinus Torvalds 4958ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4959ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4960ee81feb6SDominik Brodowski { 4961e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4962e886663cSJens Axboe flags); 4963ee81feb6SDominik Brodowski } 4964ee81feb6SDominik Brodowski 4965520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4966520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4967520c8b16SMiklos Szeredi { 4968e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4969e886663cSJens Axboe 0); 4970520c8b16SMiklos Szeredi } 4971520c8b16SMiklos Szeredi 4972a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 49735590ff0dSUlrich Drepper { 4974e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4975e886663cSJens Axboe getname(newname), 0); 49765590ff0dSUlrich Drepper } 49775590ff0dSUlrich Drepper 49785d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 49791da177e4SLinus Torvalds { 49805d826c84SAl Viro int len = PTR_ERR(link); 49811da177e4SLinus Torvalds if (IS_ERR(link)) 49821da177e4SLinus Torvalds goto out; 49831da177e4SLinus Torvalds 49841da177e4SLinus Torvalds len = strlen(link); 49851da177e4SLinus Torvalds if (len > (unsigned) buflen) 49861da177e4SLinus Torvalds len = buflen; 49871da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 49881da177e4SLinus Torvalds len = -EFAULT; 49891da177e4SLinus Torvalds out: 49901da177e4SLinus Torvalds return len; 49911da177e4SLinus Torvalds } 49921da177e4SLinus Torvalds 4993d60874cdSMiklos Szeredi /** 4994fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4995fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4996fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4997fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4998fd4a0edfSMiklos Szeredi * 4999fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 5000fd4a0edfSMiklos Szeredi * 5001fd4a0edfSMiklos Szeredi * Does not call security hook. 5002fd4a0edfSMiklos Szeredi */ 5003fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5004fd4a0edfSMiklos Szeredi { 5005fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 5006f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 5007f2df5da6SAl Viro const char *link; 5008f2df5da6SAl Viro int res; 5009fd4a0edfSMiklos Szeredi 501076fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 501176fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 501276fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 501376fca90eSMiklos Szeredi 501476fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 5015fd4a0edfSMiklos Szeredi return -EINVAL; 5016fd4a0edfSMiklos Szeredi 501776fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 501876fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 501976fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 502076fca90eSMiklos Szeredi } 502176fca90eSMiklos Szeredi 50224c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 5023f2df5da6SAl Viro if (!link) { 5024f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 5025f2df5da6SAl Viro if (IS_ERR(link)) 5026f2df5da6SAl Viro return PTR_ERR(link); 5027f2df5da6SAl Viro } 5028f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 5029f2df5da6SAl Viro do_delayed_call(&done); 5030f2df5da6SAl Viro return res; 5031fd4a0edfSMiklos Szeredi } 5032fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 50331da177e4SLinus Torvalds 5034d60874cdSMiklos Szeredi /** 5035d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 5036d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 5037d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 5038d60874cdSMiklos Szeredi * 5039d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 5040d60874cdSMiklos Szeredi * 5041d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 5042d60874cdSMiklos Szeredi * 5043d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 5044d60874cdSMiklos Szeredi */ 5045d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 5046d60874cdSMiklos Szeredi { 5047d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 5048d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 5049d60874cdSMiklos Szeredi 5050d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 5051d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 5052d60874cdSMiklos Szeredi if (!res) 5053d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 5054d60874cdSMiklos Szeredi } 5055d60874cdSMiklos Szeredi return res; 5056d60874cdSMiklos Szeredi } 5057d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 5058d60874cdSMiklos Szeredi 50591da177e4SLinus Torvalds /* get the link contents into pagecache */ 50606b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 5061fceef393SAl Viro struct delayed_call *callback) 50621da177e4SLinus Torvalds { 5063ebd09abbSDuane Griffin char *kaddr; 50641da177e4SLinus Torvalds struct page *page; 50656b255391SAl Viro struct address_space *mapping = inode->i_mapping; 50666b255391SAl Viro 5067d3883d4fSAl Viro if (!dentry) { 5068d3883d4fSAl Viro page = find_get_page(mapping, 0); 5069d3883d4fSAl Viro if (!page) 50706b255391SAl Viro return ERR_PTR(-ECHILD); 5071d3883d4fSAl Viro if (!PageUptodate(page)) { 5072d3883d4fSAl Viro put_page(page); 5073d3883d4fSAl Viro return ERR_PTR(-ECHILD); 5074d3883d4fSAl Viro } 5075d3883d4fSAl Viro } else { 5076090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 50771da177e4SLinus Torvalds if (IS_ERR(page)) 50786fe6900eSNick Piggin return (char*)page; 5079d3883d4fSAl Viro } 5080fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 508121fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 508221fc61c7SAl Viro kaddr = page_address(page); 50836b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 5084ebd09abbSDuane Griffin return kaddr; 50851da177e4SLinus Torvalds } 50861da177e4SLinus Torvalds 50876b255391SAl Viro EXPORT_SYMBOL(page_get_link); 50881da177e4SLinus Torvalds 5089fceef393SAl Viro void page_put_link(void *arg) 50901da177e4SLinus Torvalds { 5091fceef393SAl Viro put_page(arg); 50921da177e4SLinus Torvalds } 50934d359507SAl Viro EXPORT_SYMBOL(page_put_link); 50941da177e4SLinus Torvalds 5095aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5096aa80deabSAl Viro { 5097fceef393SAl Viro DEFINE_DELAYED_CALL(done); 50986b255391SAl Viro int res = readlink_copy(buffer, buflen, 50996b255391SAl Viro page_get_link(dentry, d_inode(dentry), 5100fceef393SAl Viro &done)); 5101fceef393SAl Viro do_delayed_call(&done); 5102aa80deabSAl Viro return res; 5103aa80deabSAl Viro } 5104aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 5105aa80deabSAl Viro 510656f5746cSMatthew Wilcox (Oracle) int page_symlink(struct inode *inode, const char *symname, int len) 51071da177e4SLinus Torvalds { 51081da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 510927a77913SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 511056f5746cSMatthew Wilcox (Oracle) bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS); 51110adb25d2SKirill Korotaev struct page *page; 5112afddba49SNick Piggin void *fsdata; 5113beb497abSDmitriy Monakhov int err; 51142d878178SMatthew Wilcox (Oracle) unsigned int flags; 51151da177e4SLinus Torvalds 51167e53cac4SNeilBrown retry: 51172d878178SMatthew Wilcox (Oracle) if (nofs) 51182d878178SMatthew Wilcox (Oracle) flags = memalloc_nofs_save(); 511927a77913SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata); 51202d878178SMatthew Wilcox (Oracle) if (nofs) 51212d878178SMatthew Wilcox (Oracle) memalloc_nofs_restore(flags); 51221da177e4SLinus Torvalds if (err) 5123afddba49SNick Piggin goto fail; 5124afddba49SNick Piggin 512521fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5126afddba49SNick Piggin 512727a77913SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, 0, len-1, len-1, 5128afddba49SNick Piggin page, fsdata); 51291da177e4SLinus Torvalds if (err < 0) 51301da177e4SLinus Torvalds goto fail; 5131afddba49SNick Piggin if (err < len-1) 5132afddba49SNick Piggin goto retry; 5133afddba49SNick Piggin 51341da177e4SLinus Torvalds mark_inode_dirty(inode); 51351da177e4SLinus Torvalds return 0; 51361da177e4SLinus Torvalds fail: 51371da177e4SLinus Torvalds return err; 51381da177e4SLinus Torvalds } 51394d359507SAl Viro EXPORT_SYMBOL(page_symlink); 51400adb25d2SKirill Korotaev 514192e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 51426b255391SAl Viro .get_link = page_get_link, 51431da177e4SLinus Torvalds }; 51441da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5145