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 668*6e180327SAl Viro static void leave_rcu(struct nameidata *nd) 669*6e180327SAl Viro { 670*6e180327SAl Viro nd->flags &= ~LOOKUP_RCU; 671*6e180327SAl Viro rcu_read_unlock(); 672*6e180327SAl Viro } 673*6e180327SAl Viro 6747973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6757973387aSAl Viro { 6767973387aSAl Viro drop_links(nd); 6777973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6787973387aSAl Viro int i; 6797973387aSAl Viro path_put(&nd->path); 6807973387aSAl Viro for (i = 0; i < nd->depth; i++) 6817973387aSAl Viro path_put(&nd->stack[i].link); 682bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 683102b8af2SAl Viro path_put(&nd->root); 684bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 685102b8af2SAl Viro } 6867973387aSAl Viro } else { 687*6e180327SAl Viro leave_rcu(nd); 6887973387aSAl Viro } 6897973387aSAl Viro nd->depth = 0; 6907d01ef75SAl Viro nd->path.mnt = NULL; 6917d01ef75SAl Viro nd->path.dentry = NULL; 6927973387aSAl Viro } 6937973387aSAl Viro 6947973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6952aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6967973387aSAl Viro { 6972aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6987973387aSAl Viro if (unlikely(res)) { 6997973387aSAl Viro if (res > 0) 7007973387aSAl Viro path->mnt = NULL; 7017973387aSAl Viro path->dentry = NULL; 7027973387aSAl Viro return false; 7037973387aSAl Viro } 7047973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 7057973387aSAl Viro path->dentry = NULL; 7067973387aSAl Viro return false; 7077973387aSAl Viro } 7087973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7097973387aSAl Viro } 7107973387aSAl Viro 7112aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7122aa38470SAl Viro struct path *path, unsigned seq) 7132aa38470SAl Viro { 7145bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7152aa38470SAl Viro } 7162aa38470SAl Viro 7177973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7187973387aSAl Viro { 7197973387aSAl Viro int i; 720eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 721eacd9aa8SAl Viro drop_links(nd); 722eacd9aa8SAl Viro nd->depth = 0; 723eacd9aa8SAl Viro return false; 724eacd9aa8SAl Viro } 7257973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7267973387aSAl Viro struct saved *last = nd->stack + i; 7277973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7287973387aSAl Viro drop_links(nd); 7297973387aSAl Viro nd->depth = i + 1; 7307973387aSAl Viro return false; 7317973387aSAl Viro } 7327973387aSAl Viro } 7337973387aSAl Viro return true; 7347973387aSAl Viro } 7357973387aSAl Viro 736ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 737ee594bffSAl Viro { 738adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 739bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 740ee594bffSAl Viro return true; 741bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 742ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 743ee594bffSAl Viro } 744ee594bffSAl Viro 74519660af7SAl Viro /* 74631e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 74719660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 74819660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 74957e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 75019660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 75119660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 75219660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 75319660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 75431e6b01fSNick Piggin */ 75531e6b01fSNick Piggin 75631e6b01fSNick Piggin /** 757e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 75819660af7SAl Viro * @nd: nameidata pathwalk data 759e36cffedSJens Axboe * Returns: true on success, false on failure 76031e6b01fSNick Piggin * 761e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7624675ac39SAl Viro * for ref-walk mode. 7634675ac39SAl Viro * Must be called from rcu-walk context. 764e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7657973387aSAl Viro * terminate_walk(). 76631e6b01fSNick Piggin */ 767e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 76831e6b01fSNick Piggin { 76931e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 77031e6b01fSNick Piggin 77131e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 772e5c832d5SLinus Torvalds 7737973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7744675ac39SAl Viro goto out1; 77584a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 77684a2bd39SAl Viro goto out; 777ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7784675ac39SAl Viro goto out; 779*6e180327SAl Viro leave_rcu(nd); 7804675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 781e36cffedSJens Axboe return true; 7824675ac39SAl Viro 78384a2bd39SAl Viro out1: 7844675ac39SAl Viro nd->path.mnt = NULL; 7854675ac39SAl Viro nd->path.dentry = NULL; 7864675ac39SAl Viro out: 787*6e180327SAl Viro leave_rcu(nd); 788e36cffedSJens Axboe return false; 7894675ac39SAl Viro } 7904675ac39SAl Viro 7914675ac39SAl Viro /** 792ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7934675ac39SAl Viro * @nd: nameidata pathwalk data 794ae66db45SAl Viro * @dentry: next dentry to step into 795ae66db45SAl Viro * @seq: seq number to check @dentry against 796ae66db45SAl Viro * Returns: true on success, false on failure 7974675ac39SAl Viro * 79830476f7eSTom Rix * Similar to try_to_unlazy(), but here we have the next dentry already 799ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 800ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 801ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 8024675ac39SAl Viro * terminate_walk(). 8034675ac39SAl Viro */ 804ae66db45SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq) 8054675ac39SAl Viro { 8067e4745a0SAl Viro int res; 8074675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8084675ac39SAl Viro 8094675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8104675ac39SAl Viro goto out2; 8117e4745a0SAl Viro res = __legitimize_mnt(nd->path.mnt, nd->m_seq); 8127e4745a0SAl Viro if (unlikely(res)) { 8137e4745a0SAl Viro if (res > 0) 8147973387aSAl Viro goto out2; 8157e4745a0SAl Viro goto out1; 8167e4745a0SAl Viro } 8174675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8187973387aSAl Viro goto out1; 81948a066e7SAl Viro 82015570086SLinus Torvalds /* 8214675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8224675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8234675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8244675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8254675ac39SAl Viro * know the parent must still be valid if the child sequence number is 82615570086SLinus Torvalds */ 8274675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 828e5c832d5SLinus Torvalds goto out; 82982ef0698SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 83084a2bd39SAl Viro goto out_dput; 831e5c832d5SLinus Torvalds /* 832e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 833e5c832d5SLinus Torvalds * still valid and get it if required. 834e5c832d5SLinus Torvalds */ 83584a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 83684a2bd39SAl Viro goto out_dput; 837*6e180327SAl Viro leave_rcu(nd); 838ae66db45SAl Viro return true; 83919660af7SAl Viro 8407973387aSAl Viro out2: 8417973387aSAl Viro nd->path.mnt = NULL; 8427973387aSAl Viro out1: 8437973387aSAl Viro nd->path.dentry = NULL; 844e5c832d5SLinus Torvalds out: 845*6e180327SAl Viro leave_rcu(nd); 846ae66db45SAl Viro return false; 84784a2bd39SAl Viro out_dput: 848*6e180327SAl Viro leave_rcu(nd); 84984a2bd39SAl Viro dput(dentry); 850ae66db45SAl Viro return false; 85131e6b01fSNick Piggin } 85231e6b01fSNick Piggin 8534ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 85434286d66SNick Piggin { 855a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8564ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 857a89f8337SAl Viro else 858a89f8337SAl Viro return 1; 85934286d66SNick Piggin } 86034286d66SNick Piggin 8619f1fafeeSAl Viro /** 8629f1fafeeSAl Viro * complete_walk - successful completion of path walk 8639f1fafeeSAl Viro * @nd: pointer nameidata 86439159de2SJeff Layton * 8659f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8669f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8679f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8689f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8699f1fafeeSAl Viro * need to drop nd->path. 87039159de2SJeff Layton */ 8719f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 87239159de2SJeff Layton { 87316c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 87439159de2SJeff Layton int status; 87539159de2SJeff Layton 8769f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 877adb21d2bSAleksa Sarai /* 878adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 879adb21d2bSAleksa Sarai * externally-managed nd->root. 880adb21d2bSAleksa Sarai */ 881bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 882bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8839f1fafeeSAl Viro nd->root.mnt = NULL; 8846c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 885e36cffedSJens Axboe if (!try_to_unlazy(nd)) 88648a066e7SAl Viro return -ECHILD; 88748a066e7SAl Viro } 8889f1fafeeSAl Viro 889adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 890adb21d2bSAleksa Sarai /* 891adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 892adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 893adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 894adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 895adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 896adb21d2bSAleksa Sarai * 897adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 898adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 899adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 900adb21d2bSAleksa Sarai * requested root to userspace. 901adb21d2bSAleksa Sarai * 902adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 903adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 904adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 905adb21d2bSAleksa Sarai */ 906adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 907adb21d2bSAleksa Sarai return -EXDEV; 908adb21d2bSAleksa Sarai } 909adb21d2bSAleksa Sarai 910bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 91139159de2SJeff Layton return 0; 91239159de2SJeff Layton 913ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 91416c2cd71SAl Viro return 0; 91516c2cd71SAl Viro 916ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 91739159de2SJeff Layton if (status > 0) 91839159de2SJeff Layton return 0; 91939159de2SJeff Layton 92016c2cd71SAl Viro if (!status) 92139159de2SJeff Layton status = -ESTALE; 92216c2cd71SAl Viro 92339159de2SJeff Layton return status; 92439159de2SJeff Layton } 92539159de2SJeff Layton 926740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9272a737871SAl Viro { 92831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9299e6697e2SAl Viro 930adb21d2bSAleksa Sarai /* 931adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 932adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 933adb21d2bSAleksa Sarai * from the dirfd. 934adb21d2bSAleksa Sarai */ 935adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 936adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 937adb21d2bSAleksa Sarai 9389e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9398f47a016SAl Viro unsigned seq; 940c28cc364SNick Piggin 941c28cc364SNick Piggin do { 942c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 94331e6b01fSNick Piggin nd->root = fs->root; 9448f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 945c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9469e6697e2SAl Viro } else { 9479e6697e2SAl Viro get_fs_root(fs, &nd->root); 948bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9499e6697e2SAl Viro } 950740a1678SAleksa Sarai return 0; 95131e6b01fSNick Piggin } 95231e6b01fSNick Piggin 953248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 954248fb5b9SAl Viro { 955adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 956adb21d2bSAleksa Sarai return -EXDEV; 95772ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 95872ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 95972ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 96072ba2929SAleksa Sarai return -EXDEV; 96172ba2929SAleksa Sarai } 962740a1678SAleksa Sarai if (!nd->root.mnt) { 963740a1678SAleksa Sarai int error = set_root(nd); 964740a1678SAleksa Sarai if (error) 965740a1678SAleksa Sarai return error; 966740a1678SAleksa Sarai } 967248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 968248fb5b9SAl Viro struct dentry *d; 969248fb5b9SAl Viro nd->path = nd->root; 970248fb5b9SAl Viro d = nd->path.dentry; 971248fb5b9SAl Viro nd->inode = d->d_inode; 972248fb5b9SAl Viro nd->seq = nd->root_seq; 97382ef0698SAl Viro if (read_seqcount_retry(&d->d_seq, nd->seq)) 974248fb5b9SAl Viro return -ECHILD; 975248fb5b9SAl Viro } else { 976248fb5b9SAl Viro path_put(&nd->path); 977248fb5b9SAl Viro nd->path = nd->root; 978248fb5b9SAl Viro path_get(&nd->path); 979248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 980248fb5b9SAl Viro } 981bcba1e7dSAl Viro nd->state |= ND_JUMPED; 982248fb5b9SAl Viro return 0; 983248fb5b9SAl Viro } 984248fb5b9SAl Viro 985b5fb63c1SChristoph Hellwig /* 9866b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 987b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 988b5fb63c1SChristoph Hellwig */ 9891bc82070SAleksa Sarai int nd_jump_link(struct path *path) 990b5fb63c1SChristoph Hellwig { 9914b99d499SAleksa Sarai int error = -ELOOP; 9926e77137bSAl Viro struct nameidata *nd = current->nameidata; 993b5fb63c1SChristoph Hellwig 9944b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9954b99d499SAleksa Sarai goto err; 9964b99d499SAleksa Sarai 99772ba2929SAleksa Sarai error = -EXDEV; 99872ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 99972ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 100072ba2929SAleksa Sarai goto err; 100172ba2929SAleksa Sarai } 1002adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 1003adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 1004adb21d2bSAleksa Sarai goto err; 100572ba2929SAleksa Sarai 10064b99d499SAleksa Sarai path_put(&nd->path); 1007b5fb63c1SChristoph Hellwig nd->path = *path; 1008b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1009bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10101bc82070SAleksa Sarai return 0; 10114b99d499SAleksa Sarai 10124b99d499SAleksa Sarai err: 10134b99d499SAleksa Sarai path_put(path); 10144b99d499SAleksa Sarai return error; 1015b5fb63c1SChristoph Hellwig } 1016b5fb63c1SChristoph Hellwig 1017b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1018574197e0SAl Viro { 101921c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1020fceef393SAl Viro do_delayed_call(&last->done); 10216548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1022b9ff4429SAl Viro path_put(&last->link); 1023574197e0SAl Viro } 1024574197e0SAl Viro 10259c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10269c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10279c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10289c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10299c011be1SLuis Chamberlain 10309c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10319c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10329c011be1SLuis Chamberlain { 10339c011be1SLuis Chamberlain .procname = "protected_symlinks", 10349c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10359c011be1SLuis Chamberlain .maxlen = sizeof(int), 1036c7031c14SJulius Hemanth Pitti .mode = 0644, 10379c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10389c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10399c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10409c011be1SLuis Chamberlain }, 10419c011be1SLuis Chamberlain { 10429c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10439c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10449c011be1SLuis Chamberlain .maxlen = sizeof(int), 1045c7031c14SJulius Hemanth Pitti .mode = 0644, 10469c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10479c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10489c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10499c011be1SLuis Chamberlain }, 10509c011be1SLuis Chamberlain { 10519c011be1SLuis Chamberlain .procname = "protected_fifos", 10529c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10539c011be1SLuis Chamberlain .maxlen = sizeof(int), 1054c7031c14SJulius Hemanth Pitti .mode = 0644, 10559c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10569c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10579c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10589c011be1SLuis Chamberlain }, 10599c011be1SLuis Chamberlain { 10609c011be1SLuis Chamberlain .procname = "protected_regular", 10619c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10629c011be1SLuis Chamberlain .maxlen = sizeof(int), 1063c7031c14SJulius Hemanth Pitti .mode = 0644, 10649c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10659c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10669c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10679c011be1SLuis Chamberlain }, 10689c011be1SLuis Chamberlain { } 10699c011be1SLuis Chamberlain }; 10709c011be1SLuis Chamberlain 10719c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10729c011be1SLuis Chamberlain { 10739c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10749c011be1SLuis Chamberlain return 0; 10759c011be1SLuis Chamberlain } 10769c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10779c011be1SLuis Chamberlain 10789c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1079800179c9SKees Cook 1080800179c9SKees Cook /** 1081800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 108255852635SRandy Dunlap * @nd: nameidata pathwalk data 1083800179c9SKees Cook * 1084800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1085800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1086800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1087800179c9SKees Cook * processes from failing races against path names that may change out 1088800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1089800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1090800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1091800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1092800179c9SKees Cook * 1093800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1094800179c9SKees Cook */ 1095ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1096800179c9SKees Cook { 1097ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1098ba73d987SChristian Brauner kuid_t i_uid; 1099ba73d987SChristian Brauner 1100800179c9SKees Cook if (!sysctl_protected_symlinks) 1101800179c9SKees Cook return 0; 1102800179c9SKees Cook 1103ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1104ba73d987SChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 1105800179c9SKees Cook /* Allowed if owner and follower match. */ 1106ba73d987SChristian Brauner if (uid_eq(current_cred()->fsuid, i_uid)) 1107800179c9SKees Cook return 0; 1108800179c9SKees Cook 1109800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11100f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1111800179c9SKees Cook return 0; 1112800179c9SKees Cook 1113800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1114ba73d987SChristian Brauner if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) 1115800179c9SKees Cook return 0; 1116800179c9SKees Cook 111731956502SAl Viro if (nd->flags & LOOKUP_RCU) 111831956502SAl Viro return -ECHILD; 111931956502SAl Viro 1120ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1121245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1122800179c9SKees Cook return -EACCES; 1123800179c9SKees Cook } 1124800179c9SKees Cook 1125800179c9SKees Cook /** 1126800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1127ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1128800179c9SKees Cook * @inode: the source inode to hardlink from 1129800179c9SKees Cook * 1130800179c9SKees Cook * Return false if at least one of the following conditions: 1131800179c9SKees Cook * - inode is not a regular file 1132800179c9SKees Cook * - inode is setuid 1133800179c9SKees Cook * - inode is setgid and group-exec 1134800179c9SKees Cook * - access failure for read and write 1135800179c9SKees Cook * 1136800179c9SKees Cook * Otherwise returns true. 1137800179c9SKees Cook */ 1138ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1139ba73d987SChristian Brauner struct inode *inode) 1140800179c9SKees Cook { 1141800179c9SKees Cook umode_t mode = inode->i_mode; 1142800179c9SKees Cook 1143800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1144800179c9SKees Cook if (!S_ISREG(mode)) 1145800179c9SKees Cook return false; 1146800179c9SKees Cook 1147800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1148800179c9SKees Cook if (mode & S_ISUID) 1149800179c9SKees Cook return false; 1150800179c9SKees Cook 1151800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1152800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1153800179c9SKees Cook return false; 1154800179c9SKees Cook 1155800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1156ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1157800179c9SKees Cook return false; 1158800179c9SKees Cook 1159800179c9SKees Cook return true; 1160800179c9SKees Cook } 1161800179c9SKees Cook 1162800179c9SKees Cook /** 1163800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1164ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1165800179c9SKees Cook * @link: the source to hardlink from 1166800179c9SKees Cook * 1167800179c9SKees Cook * Block hardlink when all of: 1168800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1169800179c9SKees Cook * - fsuid does not match inode 1170800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1171f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1172800179c9SKees Cook * 1173ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1174ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1175ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1176ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1177ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1178ba73d987SChristian Brauner * 1179800179c9SKees Cook * Returns 0 if successful, -ve on error. 1180800179c9SKees Cook */ 1181ba73d987SChristian Brauner int may_linkat(struct user_namespace *mnt_userns, struct path *link) 1182800179c9SKees Cook { 1183593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1184593d1ce8SEric W. Biederman 1185593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1186ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 1187ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 1188593d1ce8SEric W. Biederman return -EOVERFLOW; 1189800179c9SKees Cook 1190800179c9SKees Cook if (!sysctl_protected_hardlinks) 1191800179c9SKees Cook return 0; 1192800179c9SKees Cook 1193800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1194800179c9SKees Cook * otherwise, it must be a safe source. 1195800179c9SKees Cook */ 1196ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1197ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1198800179c9SKees Cook return 0; 1199800179c9SKees Cook 1200245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1201800179c9SKees Cook return -EPERM; 1202800179c9SKees Cook } 1203800179c9SKees Cook 120430aba665SSalvatore Mesoraca /** 120530aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 120630aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 120730aba665SSalvatore Mesoraca * exist. 1208ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 12092111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 121030aba665SSalvatore Mesoraca * @inode: the inode of the file to open 121130aba665SSalvatore Mesoraca * 121230aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 121330aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 121430aba665SSalvatore Mesoraca * - the file already exists 121530aba665SSalvatore Mesoraca * - we are in a sticky directory 121630aba665SSalvatore Mesoraca * - we don't own the file 121730aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 121830aba665SSalvatore Mesoraca * - the directory is world writable 121930aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 122030aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 122130aba665SSalvatore Mesoraca * be enough. 122230aba665SSalvatore Mesoraca * 1223ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1224ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1225ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1226ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1227ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1228ba73d987SChristian Brauner * 122930aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 123030aba665SSalvatore Mesoraca */ 1231ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1232ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 123330aba665SSalvatore Mesoraca { 1234ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1235ba73d987SChristian Brauner kuid_t dir_uid = nd->dir_uid; 1236ba73d987SChristian Brauner 123730aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 123830aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1239d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1240ba73d987SChristian Brauner uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || 1241ba73d987SChristian Brauner uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) 124230aba665SSalvatore Mesoraca return 0; 124330aba665SSalvatore Mesoraca 1244d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1245d0cb5018SAl Viro (dir_mode & 0020 && 124630aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 124730aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1248245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1249245d7369SKees Cook "sticky_create_fifo" : 1250245d7369SKees Cook "sticky_create_regular"; 1251245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 125230aba665SSalvatore Mesoraca return -EACCES; 125330aba665SSalvatore Mesoraca } 125430aba665SSalvatore Mesoraca return 0; 125530aba665SSalvatore Mesoraca } 125630aba665SSalvatore Mesoraca 1257f015f126SDavid Howells /* 1258f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1259f015f126SDavid Howells * 1260f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1261f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1262f015f126SDavid Howells * Up is towards /. 1263f015f126SDavid Howells * 1264f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1265f015f126SDavid Howells * root. 1266f015f126SDavid Howells */ 1267bab77ebfSAl Viro int follow_up(struct path *path) 12681da177e4SLinus Torvalds { 12690714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12700714a533SAl Viro struct mount *parent; 12711da177e4SLinus Torvalds struct dentry *mountpoint; 127299b7db7bSNick Piggin 127348a066e7SAl Viro read_seqlock_excl(&mount_lock); 12740714a533SAl Viro parent = mnt->mnt_parent; 12753c0a6163SAl Viro if (parent == mnt) { 127648a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12771da177e4SLinus Torvalds return 0; 12781da177e4SLinus Torvalds } 12790714a533SAl Viro mntget(&parent->mnt); 1280a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 128148a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1282bab77ebfSAl Viro dput(path->dentry); 1283bab77ebfSAl Viro path->dentry = mountpoint; 1284bab77ebfSAl Viro mntput(path->mnt); 12850714a533SAl Viro path->mnt = &parent->mnt; 12861da177e4SLinus Torvalds return 1; 12871da177e4SLinus Torvalds } 12884d359507SAl Viro EXPORT_SYMBOL(follow_up); 12891da177e4SLinus Torvalds 12907ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12917ef482faSAl Viro struct path *path, unsigned *seqp) 12927ef482faSAl Viro { 12937ef482faSAl Viro while (mnt_has_parent(m)) { 12947ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12957ef482faSAl Viro 12967ef482faSAl Viro m = m->mnt_parent; 12977ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12987ef482faSAl Viro root->mnt == &m->mnt)) 12997ef482faSAl Viro break; 13007ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 13017ef482faSAl Viro path->mnt = &m->mnt; 13027ef482faSAl Viro path->dentry = mountpoint; 13037ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 13047ef482faSAl Viro return true; 13057ef482faSAl Viro } 13067ef482faSAl Viro } 13077ef482faSAl Viro return false; 13087ef482faSAl Viro } 13097ef482faSAl Viro 13102aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 13112aa38470SAl Viro struct path *path) 13122aa38470SAl Viro { 13132aa38470SAl Viro bool found; 13142aa38470SAl Viro 13152aa38470SAl Viro rcu_read_lock(); 13162aa38470SAl Viro while (1) { 13172aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 13182aa38470SAl Viro 13192aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 13202aa38470SAl Viro if (unlikely(!found)) { 13212aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 13222aa38470SAl Viro break; 13232aa38470SAl Viro } else { 13242aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 13252aa38470SAl Viro break; 13262aa38470SAl Viro rcu_read_unlock(); 13272aa38470SAl Viro path_put(path); 13282aa38470SAl Viro rcu_read_lock(); 13292aa38470SAl Viro } 13302aa38470SAl Viro } 13312aa38470SAl Viro rcu_read_unlock(); 13322aa38470SAl Viro return found; 13332aa38470SAl Viro } 13342aa38470SAl Viro 1335b5c84bf6SNick Piggin /* 13369875cf80SDavid Howells * Perform an automount 13379875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 13389875cf80SDavid Howells * were called with. 13391da177e4SLinus Torvalds */ 13401c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 134131e6b01fSNick Piggin { 134225e195aaSAl Viro struct dentry *dentry = path->dentry; 13439875cf80SDavid Howells 13440ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 13450ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 13460ec26fd0SMiklos Szeredi * the name. 13470ec26fd0SMiklos Szeredi * 13480ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 13495a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 13500ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 13510ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13520ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13530ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13545a30d8a2SDavid Howells */ 13551c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13565d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 135725e195aaSAl Viro dentry->d_inode) 13589875cf80SDavid Howells return -EISDIR; 13590ec26fd0SMiklos Szeredi 13601c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13619875cf80SDavid Howells return -ELOOP; 13629875cf80SDavid Howells 136325e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1364ea5b778aSDavid Howells } 13659875cf80SDavid Howells 13669875cf80SDavid Howells /* 13679deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13689deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13699deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13709deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13719875cf80SDavid Howells */ 13729deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13739deed3ebSAl Viro int *count, unsigned lookup_flags) 13749875cf80SDavid Howells { 13759deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13769875cf80SDavid Howells bool need_mntput = false; 13778aef1884SAl Viro int ret = 0; 13789875cf80SDavid Howells 13799deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1380cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1381cc53ce53SDavid Howells * being held. */ 1382d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1383fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1384508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1385cc53ce53SDavid Howells if (ret < 0) 13868aef1884SAl Viro break; 1387cc53ce53SDavid Howells } 1388cc53ce53SDavid Howells 13899deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13909875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13919deed3ebSAl Viro if (mounted) { // ... in our namespace 13929875cf80SDavid Howells dput(path->dentry); 13939875cf80SDavid Howells if (need_mntput) 1394463ffb2eSAl Viro mntput(path->mnt); 1395463ffb2eSAl Viro path->mnt = mounted; 1396463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13979deed3ebSAl Viro // here we know it's positive 13989deed3ebSAl Viro flags = path->dentry->d_flags; 13999875cf80SDavid Howells need_mntput = true; 14009875cf80SDavid Howells continue; 1401463ffb2eSAl Viro } 14021da177e4SLinus Torvalds } 14039875cf80SDavid Howells 14049deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 14059deed3ebSAl Viro break; 14069deed3ebSAl Viro 14079deed3ebSAl Viro // uncovered automount point 14089deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 14099deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 14109875cf80SDavid Howells if (ret < 0) 14118aef1884SAl Viro break; 14129875cf80SDavid Howells } 14139875cf80SDavid Howells 14149deed3ebSAl Viro if (ret == -EISDIR) 14159deed3ebSAl Viro ret = 0; 14169deed3ebSAl Viro // possible if you race with several mount --move 14179deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 14188aef1884SAl Viro mntput(path->mnt); 14199deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1420d41efb52SAl Viro ret = -ENOENT; 14219deed3ebSAl Viro *jumped = need_mntput; 14228402752eSAl Viro return ret; 14231da177e4SLinus Torvalds } 14241da177e4SLinus Torvalds 14259deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 14269deed3ebSAl Viro int *count, unsigned lookup_flags) 14279deed3ebSAl Viro { 14289deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 14299deed3ebSAl Viro 14309deed3ebSAl Viro /* fastpath */ 14319deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 14329deed3ebSAl Viro *jumped = false; 14339deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 14349deed3ebSAl Viro return -ENOENT; 14359deed3ebSAl Viro return 0; 14369deed3ebSAl Viro } 14379deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 14389deed3ebSAl Viro } 14399deed3ebSAl Viro 1440cc53ce53SDavid Howells int follow_down_one(struct path *path) 14411da177e4SLinus Torvalds { 14421da177e4SLinus Torvalds struct vfsmount *mounted; 14431da177e4SLinus Torvalds 14441c755af4SAl Viro mounted = lookup_mnt(path); 14451da177e4SLinus Torvalds if (mounted) { 14469393bd07SAl Viro dput(path->dentry); 14479393bd07SAl Viro mntput(path->mnt); 14489393bd07SAl Viro path->mnt = mounted; 14499393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 14501da177e4SLinus Torvalds return 1; 14511da177e4SLinus Torvalds } 14521da177e4SLinus Torvalds return 0; 14531da177e4SLinus Torvalds } 14544d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14551da177e4SLinus Torvalds 14569875cf80SDavid Howells /* 14579deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14589deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14599deed3ebSAl Viro * caller is permitted to proceed or not. 14609deed3ebSAl Viro */ 14619deed3ebSAl Viro int follow_down(struct path *path) 14629deed3ebSAl Viro { 14639deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14649deed3ebSAl Viro bool jumped; 14659deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 14669deed3ebSAl Viro 14679deed3ebSAl Viro if (path->mnt != mnt) 14689deed3ebSAl Viro mntput(mnt); 14699deed3ebSAl Viro return ret; 14709deed3ebSAl Viro } 14719deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14729deed3ebSAl Viro 14739deed3ebSAl Viro /* 1474287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1475287548e4SAl Viro * we meet a managed dentry that would need blocking. 14769875cf80SDavid Howells */ 14779875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1478254cf582SAl Viro struct inode **inode, unsigned *seqp) 14799875cf80SDavid Howells { 1480ea936aebSAl Viro struct dentry *dentry = path->dentry; 1481ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1482ea936aebSAl Viro 1483ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1484ea936aebSAl Viro return true; 1485ea936aebSAl Viro 1486ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1487ea936aebSAl Viro return false; 1488ea936aebSAl Viro 148962a7375eSIan Kent for (;;) { 149062a7375eSIan Kent /* 149162a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 149262a7375eSIan Kent * that wants to block transit. 149362a7375eSIan Kent */ 1494ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1495ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1496ea936aebSAl Viro if (res) 1497ea936aebSAl Viro return res == -EISDIR; 1498ea936aebSAl Viro flags = dentry->d_flags; 1499b8faf035SNeilBrown } 150062a7375eSIan Kent 1501ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1502ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1503ea936aebSAl Viro if (mounted) { 1504c7105365SAl Viro path->mnt = &mounted->mnt; 1505ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1506bcba1e7dSAl Viro nd->state |= ND_JUMPED; 1507ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1508ea936aebSAl Viro *inode = dentry->d_inode; 150959430262SLinus Torvalds /* 1510ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1511ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1512ea936aebSAl Viro * between mount hash removal and ->mnt_root 1513ea936aebSAl Viro * becoming unpinned. 151459430262SLinus Torvalds */ 1515ea936aebSAl Viro flags = dentry->d_flags; 151620aac6c6SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 151720aac6c6SAl Viro return false; 1518ea936aebSAl Viro continue; 15199875cf80SDavid Howells } 1520ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1521ea936aebSAl Viro return false; 1522ea936aebSAl Viro } 1523ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1524ea936aebSAl Viro } 1525287548e4SAl Viro } 1526287548e4SAl Viro 1527db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1528db3c9adeSAl Viro struct path *path, struct inode **inode, 1529db3c9adeSAl Viro unsigned int *seqp) 1530bd7c4b50SAl Viro { 15319deed3ebSAl Viro bool jumped; 1532db3c9adeSAl Viro int ret; 1533bd7c4b50SAl Viro 1534db3c9adeSAl Viro path->mnt = nd->path.mnt; 1535db3c9adeSAl Viro path->dentry = dentry; 1536c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1537c153007bSAl Viro unsigned int seq = *seqp; 1538c153007bSAl Viro if (unlikely(!*inode)) 1539c153007bSAl Viro return -ENOENT; 1540c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15419deed3ebSAl Viro return 0; 1542ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 1543c153007bSAl Viro return -ECHILD; 1544c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1545c153007bSAl Viro path->mnt = nd->path.mnt; 1546c153007bSAl Viro path->dentry = dentry; 1547c153007bSAl Viro } 15489deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 15499deed3ebSAl Viro if (jumped) { 15509deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 15519deed3ebSAl Viro ret = -EXDEV; 15529deed3ebSAl Viro else 1553bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15549deed3ebSAl Viro } 15559deed3ebSAl Viro if (unlikely(ret)) { 15569deed3ebSAl Viro dput(path->dentry); 15579deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15589deed3ebSAl Viro mntput(path->mnt); 15599deed3ebSAl Viro } else { 1560bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1561bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1562bd7c4b50SAl Viro } 1563bd7c4b50SAl Viro return ret; 1564bd7c4b50SAl Viro } 1565bd7c4b50SAl Viro 15669875cf80SDavid Howells /* 1567f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1568f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1569baa03890SNick Piggin */ 1570e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1571e3c13928SAl Viro struct dentry *dir, 15726c51e513SAl Viro unsigned int flags) 1573baa03890SNick Piggin { 1574a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1575bad61189SMiklos Szeredi if (dentry) { 1576a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1577bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 157874ff0ffcSAl Viro if (!error) 15795542aa2fSEric W. Biederman d_invalidate(dentry); 1580bad61189SMiklos Szeredi dput(dentry); 158174ff0ffcSAl Viro return ERR_PTR(error); 1582bad61189SMiklos Szeredi } 1583bad61189SMiklos Szeredi } 1584baa03890SNick Piggin return dentry; 1585baa03890SNick Piggin } 1586baa03890SNick Piggin 1587baa03890SNick Piggin /* 1588a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1589a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1590a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1591a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1592a03ece5fSAl Viro * at all. 159344396f4bSJosef Bacik */ 1594a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1595a03ece5fSAl Viro struct dentry *base, unsigned int flags) 159644396f4bSJosef Bacik { 1597a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 159844396f4bSJosef Bacik struct dentry *old; 1599a03ece5fSAl Viro struct inode *dir = base->d_inode; 1600a03ece5fSAl Viro 1601a03ece5fSAl Viro if (dentry) 1602a03ece5fSAl Viro return dentry; 160344396f4bSJosef Bacik 160444396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1605a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 160644396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1607a03ece5fSAl Viro 1608a03ece5fSAl Viro dentry = d_alloc(base, name); 1609a03ece5fSAl Viro if (unlikely(!dentry)) 1610a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 161144396f4bSJosef Bacik 161272bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 161344396f4bSJosef Bacik if (unlikely(old)) { 161444396f4bSJosef Bacik dput(dentry); 161544396f4bSJosef Bacik dentry = old; 161644396f4bSJosef Bacik } 161744396f4bSJosef Bacik return dentry; 161844396f4bSJosef Bacik } 161944396f4bSJosef Bacik 162020e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 162120e34357SAl Viro struct inode **inode, 1622254cf582SAl Viro unsigned *seqp) 16231da177e4SLinus Torvalds { 162431e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16255a18fff2SAl Viro int status = 1; 16269875cf80SDavid Howells 16273cac260aSAl Viro /* 1628b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16295d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16305d0f49c1SAl Viro * going to fall back to non-racy lookup. 1631b04f784eSNick Piggin */ 163231e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 163331e6b01fSNick Piggin unsigned seq; 1634da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 16355d0f49c1SAl Viro if (unlikely(!dentry)) { 1636e36cffedSJens Axboe if (!try_to_unlazy(nd)) 163720e34357SAl Viro return ERR_PTR(-ECHILD); 163820e34357SAl Viro return NULL; 16395d0f49c1SAl Viro } 16405a18fff2SAl Viro 164112f8ad4bSLinus Torvalds /* 164212f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 164312f8ad4bSLinus Torvalds * the dentry name information from lookup. 164412f8ad4bSLinus Torvalds */ 164563afdfc7SDavid Howells *inode = d_backing_inode(dentry); 164682ef0698SAl Viro if (read_seqcount_retry(&dentry->d_seq, seq)) 164720e34357SAl Viro return ERR_PTR(-ECHILD); 164812f8ad4bSLinus Torvalds 164912f8ad4bSLinus Torvalds /* 165012f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 165112f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 165212f8ad4bSLinus Torvalds * 165312f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 165412f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 165512f8ad4bSLinus Torvalds */ 165682ef0698SAl Viro if (__read_seqcount_retry(&parent->d_seq, nd->seq)) 165720e34357SAl Viro return ERR_PTR(-ECHILD); 16585a18fff2SAl Viro 1659254cf582SAl Viro *seqp = seq; 16604ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1661c153007bSAl Viro if (likely(status > 0)) 166220e34357SAl Viro return dentry; 1663ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 166420e34357SAl Viro return ERR_PTR(-ECHILD); 166526ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1666209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1667209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16685a18fff2SAl Viro } else { 1669e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 167081e6f520SAl Viro if (unlikely(!dentry)) 167120e34357SAl Viro return NULL; 16724ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16735d0f49c1SAl Viro } 16745a18fff2SAl Viro if (unlikely(status <= 0)) { 1675e9742b53SAl Viro if (!status) 16765d0f49c1SAl Viro d_invalidate(dentry); 16775a18fff2SAl Viro dput(dentry); 167820e34357SAl Viro return ERR_PTR(status); 16795a18fff2SAl Viro } 168020e34357SAl Viro return dentry; 1681697f514dSMiklos Szeredi } 1682697f514dSMiklos Szeredi 1683697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 168488d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1685e3c13928SAl Viro struct dentry *dir, 1686e3c13928SAl Viro unsigned int flags) 1687697f514dSMiklos Szeredi { 168888d8331aSAl Viro struct dentry *dentry, *old; 16891936386eSAl Viro struct inode *inode = dir->d_inode; 1690d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16911936386eSAl Viro 16921936386eSAl Viro /* Don't go there if it's already dead */ 169394bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 169488d8331aSAl Viro return ERR_PTR(-ENOENT); 169594bdd655SAl Viro again: 1696d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 169794bdd655SAl Viro if (IS_ERR(dentry)) 169888d8331aSAl Viro return dentry; 169994bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1700949a852eSAl Viro int error = d_revalidate(dentry, flags); 1701949a852eSAl Viro if (unlikely(error <= 0)) { 170294bdd655SAl Viro if (!error) { 1703949a852eSAl Viro d_invalidate(dentry); 1704949a852eSAl Viro dput(dentry); 170594bdd655SAl Viro goto again; 170694bdd655SAl Viro } 170794bdd655SAl Viro dput(dentry); 1708949a852eSAl Viro dentry = ERR_PTR(error); 1709949a852eSAl Viro } 171094bdd655SAl Viro } else { 17111936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 171285c7f810SAl Viro d_lookup_done(dentry); 17131936386eSAl Viro if (unlikely(old)) { 17141936386eSAl Viro dput(dentry); 17151936386eSAl Viro dentry = old; 1716949a852eSAl Viro } 1717949a852eSAl Viro } 1718e3c13928SAl Viro return dentry; 17191da177e4SLinus Torvalds } 17201da177e4SLinus Torvalds 172188d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 172288d8331aSAl Viro struct dentry *dir, 172388d8331aSAl Viro unsigned int flags) 172488d8331aSAl Viro { 172588d8331aSAl Viro struct inode *inode = dir->d_inode; 172688d8331aSAl Viro struct dentry *res; 172788d8331aSAl Viro inode_lock_shared(inode); 172888d8331aSAl Viro res = __lookup_slow(name, dir, flags); 172988d8331aSAl Viro inode_unlock_shared(inode); 173088d8331aSAl Viro return res; 173188d8331aSAl Viro } 173288d8331aSAl Viro 1733ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1734ba73d987SChristian Brauner struct nameidata *nd) 173552094c8aSAl Viro { 173652094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17377d6beb71SLinus Torvalds int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1738e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 173952094c8aSAl Viro return err; 174052094c8aSAl Viro } 1741ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 174252094c8aSAl Viro } 174352094c8aSAl Viro 174449055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1745d63ff28fSAl Viro { 174649055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 174749055906SAl Viro return -ELOOP; 17484542576bSAl Viro 17494542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17504542576bSAl Viro return 0; 17514542576bSAl Viro if (likely(nd->stack != nd->internal)) 17524542576bSAl Viro return 0; 175360ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 175449055906SAl Viro return 0; 175560ef60c7SAl Viro 175660ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 175760ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 175860ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1759aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 176060ef60c7SAl Viro 1761e5ca024eSAl Viro if (!try_to_unlazy(nd) || !grabbed_link) 176260ef60c7SAl Viro return -ECHILD; 176360ef60c7SAl Viro 176460ef60c7SAl Viro if (nd_alloc_stack(nd)) 176560ef60c7SAl Viro return 0; 1766bc40aee0SAl Viro } 176760ef60c7SAl Viro return -ENOMEM; 176849055906SAl Viro } 176949055906SAl Viro 177049055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 177149055906SAl Viro 177249055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 177349055906SAl Viro struct inode *inode, unsigned seq, int flags) 177449055906SAl Viro { 177549055906SAl Viro struct saved *last; 177649055906SAl Viro const char *res; 177749055906SAl Viro int error = reserve_stack(nd, link, seq); 177849055906SAl Viro 177949055906SAl Viro if (unlikely(error)) { 178049055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1781cd179f44SAl Viro path_put(link); 178206708adbSAl Viro return ERR_PTR(error); 1783626de996SAl Viro } 1784ab104923SAl Viro last = nd->stack + nd->depth++; 17851cf2665bSAl Viro last->link = *link; 1786fceef393SAl Viro clear_delayed_call(&last->done); 17870450b2d1SAl Viro last->seq = seq; 1788ad6cc4c3SAl Viro 1789b1a81972SAl Viro if (flags & WALK_TRAILING) { 1790ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1791ad6cc4c3SAl Viro if (unlikely(error)) 1792ad6cc4c3SAl Viro return ERR_PTR(error); 1793ad6cc4c3SAl Viro } 1794ad6cc4c3SAl Viro 1795dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1796dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1797ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1798ad6cc4c3SAl Viro 1799ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1800ad6cc4c3SAl Viro touch_atime(&last->link); 1801ad6cc4c3SAl Viro cond_resched(); 1802ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1803e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1804ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1805ad6cc4c3SAl Viro touch_atime(&last->link); 1806ad6cc4c3SAl Viro } 1807ad6cc4c3SAl Viro 1808ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1809ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1810ad6cc4c3SAl Viro if (unlikely(error)) 1811ad6cc4c3SAl Viro return ERR_PTR(error); 1812ad6cc4c3SAl Viro 1813ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1814ad6cc4c3SAl Viro if (!res) { 1815ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1816ad6cc4c3SAl Viro struct delayed_call *); 1817ad6cc4c3SAl Viro get = inode->i_op->get_link; 1818ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1819ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1820e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1821ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1822ad6cc4c3SAl Viro } else { 1823ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1824ad6cc4c3SAl Viro } 1825ad6cc4c3SAl Viro if (!res) 1826ad6cc4c3SAl Viro goto all_done; 1827ad6cc4c3SAl Viro if (IS_ERR(res)) 1828ad6cc4c3SAl Viro return res; 1829ad6cc4c3SAl Viro } 1830ad6cc4c3SAl Viro if (*res == '/') { 1831ad6cc4c3SAl Viro error = nd_jump_root(nd); 1832ad6cc4c3SAl Viro if (unlikely(error)) 1833ad6cc4c3SAl Viro return ERR_PTR(error); 1834ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1835ad6cc4c3SAl Viro ; 1836ad6cc4c3SAl Viro } 1837ad6cc4c3SAl Viro if (*res) 1838ad6cc4c3SAl Viro return res; 1839ad6cc4c3SAl Viro all_done: // pure jump 1840ad6cc4c3SAl Viro put_link(nd); 1841ad6cc4c3SAl Viro return NULL; 1842d63ff28fSAl Viro } 1843d63ff28fSAl Viro 18443ddcd056SLinus Torvalds /* 18453ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18463ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18473ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18483ddcd056SLinus Torvalds * for the common case. 18493ddcd056SLinus Torvalds */ 1850b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1851cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 18523ddcd056SLinus Torvalds { 1853cbae4d12SAl Viro struct path path; 1854cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1855cbae4d12SAl Viro 1856cbae4d12SAl Viro if (err < 0) 1857b0417d2cSAl Viro return ERR_PTR(err); 1858cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18598c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1860aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18618f64fb1cSAl Viro /* not a symlink or should not follow */ 1862c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1863c99687a0SAl Viro dput(nd->path.dentry); 1864c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1865c99687a0SAl Viro mntput(nd->path.mnt); 1866c99687a0SAl Viro } 1867c99687a0SAl Viro nd->path = path; 18688f64fb1cSAl Viro nd->inode = inode; 18698f64fb1cSAl Viro nd->seq = seq; 1870b0417d2cSAl Viro return NULL; 18718f64fb1cSAl Viro } 1872a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 187384f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1874cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1875b0417d2cSAl Viro return ERR_PTR(-ECHILD); 187684f0cd9eSAl Viro } else { 187784f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 187884f0cd9eSAl Viro mntget(path.mnt); 1879a7f77542SAl Viro } 1880b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18813ddcd056SLinus Torvalds } 18823ddcd056SLinus Torvalds 1883c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1884c2df1968SAl Viro struct inode **inodep, 1885c2df1968SAl Viro unsigned *seqp) 1886957dd41dSAl Viro { 188712487f30SAl Viro struct dentry *parent, *old; 1888957dd41dSAl Viro 188912487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 189012487f30SAl Viro goto in_root; 189112487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18927ef482faSAl Viro struct path path; 1893efe772d6SAl Viro unsigned seq; 18947ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18957ef482faSAl Viro &nd->root, &path, &seq)) 189612487f30SAl Viro goto in_root; 1897efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1898efe772d6SAl Viro return ERR_PTR(-ECHILD); 1899efe772d6SAl Viro nd->path = path; 1900efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1901efe772d6SAl Viro nd->seq = seq; 190282ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1903efe772d6SAl Viro return ERR_PTR(-ECHILD); 1904efe772d6SAl Viro /* we know that mountpoint was pinned */ 1905957dd41dSAl Viro } 190612487f30SAl Viro old = nd->path.dentry; 190712487f30SAl Viro parent = old->d_parent; 190812487f30SAl Viro *inodep = parent->d_inode; 190912487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 191082ef0698SAl Viro if (read_seqcount_retry(&old->d_seq, nd->seq)) 191112487f30SAl Viro return ERR_PTR(-ECHILD); 191212487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 191312487f30SAl Viro return ERR_PTR(-ECHILD); 191412487f30SAl Viro return parent; 191512487f30SAl Viro in_root: 191682ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1917efe772d6SAl Viro return ERR_PTR(-ECHILD); 1918957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19197521f22bSAl Viro return ERR_PTR(-ECHILD); 192051c6546cSAl Viro *seqp = nd->seq; 192151c6546cSAl Viro *inodep = nd->path.dentry->d_inode; 192251c6546cSAl Viro return nd->path.dentry; 1923957dd41dSAl Viro } 1924957dd41dSAl Viro 1925c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1926c2df1968SAl Viro struct inode **inodep, 1927c2df1968SAl Viro unsigned *seqp) 1928957dd41dSAl Viro { 192912487f30SAl Viro struct dentry *parent; 193012487f30SAl Viro 1931957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 193212487f30SAl Viro goto in_root; 193312487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19342aa38470SAl Viro struct path path; 19352aa38470SAl Viro 19362aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19372aa38470SAl Viro &nd->root, &path)) 193812487f30SAl Viro goto in_root; 1939165200d6SAl Viro path_put(&nd->path); 1940165200d6SAl Viro nd->path = path; 19412aa38470SAl Viro nd->inode = path.dentry->d_inode; 1942165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1943165200d6SAl Viro return ERR_PTR(-EXDEV); 194412487f30SAl Viro } 1945957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 194612487f30SAl Viro parent = dget_parent(nd->path.dentry); 1947957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1948957dd41dSAl Viro dput(parent); 19497521f22bSAl Viro return ERR_PTR(-ENOENT); 1950957dd41dSAl Viro } 1951c2df1968SAl Viro *seqp = 0; 1952c2df1968SAl Viro *inodep = parent->d_inode; 1953c2df1968SAl Viro return parent; 195412487f30SAl Viro 195512487f30SAl Viro in_root: 1956957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19577521f22bSAl Viro return ERR_PTR(-EXDEV); 195851c6546cSAl Viro *seqp = 0; 195951c6546cSAl Viro *inodep = nd->path.dentry->d_inode; 196051c6546cSAl Viro return dget(nd->path.dentry); 1961957dd41dSAl Viro } 1962957dd41dSAl Viro 19637521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1964957dd41dSAl Viro { 1965957dd41dSAl Viro if (type == LAST_DOTDOT) { 19667521f22bSAl Viro const char *error = NULL; 1967c2df1968SAl Viro struct dentry *parent; 1968c2df1968SAl Viro struct inode *inode; 1969c2df1968SAl Viro unsigned seq; 1970957dd41dSAl Viro 1971957dd41dSAl Viro if (!nd->root.mnt) { 19727521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1973957dd41dSAl Viro if (error) 1974957dd41dSAl Viro return error; 1975957dd41dSAl Viro } 1976957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1977c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1978957dd41dSAl Viro else 1979c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1980c2df1968SAl Viro if (IS_ERR(parent)) 1981c2df1968SAl Viro return ERR_CAST(parent); 198251c6546cSAl Viro error = step_into(nd, WALK_NOFOLLOW, parent, inode, seq); 1983c2df1968SAl Viro if (unlikely(error)) 1984957dd41dSAl Viro return error; 1985957dd41dSAl Viro 1986957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1987957dd41dSAl Viro /* 1988957dd41dSAl Viro * If there was a racing rename or mount along our 1989957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1990957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1991957dd41dSAl Viro * some fallback). 1992957dd41dSAl Viro */ 1993957dd41dSAl Viro smp_rmb(); 199482ef0698SAl Viro if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)) 19957521f22bSAl Viro return ERR_PTR(-EAGAIN); 199682ef0698SAl Viro if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)) 19977521f22bSAl Viro return ERR_PTR(-EAGAIN); 1998957dd41dSAl Viro } 1999957dd41dSAl Viro } 20007521f22bSAl Viro return NULL; 2001957dd41dSAl Viro } 2002957dd41dSAl Viro 200392d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 2004ce57dfc1SAl Viro { 2005db3c9adeSAl Viro struct dentry *dentry; 2006ce57dfc1SAl Viro struct inode *inode; 2007254cf582SAl Viro unsigned seq; 2008ce57dfc1SAl Viro /* 2009ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 2010ce57dfc1SAl Viro * to be able to know about the current root directory and 2011ce57dfc1SAl Viro * parent relationships. 2012ce57dfc1SAl Viro */ 20134693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 20141c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 20154693a547SAl Viro put_link(nd); 20167521f22bSAl Viro return handle_dots(nd, nd->last_type); 20174693a547SAl Viro } 201820e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 201920e34357SAl Viro if (IS_ERR(dentry)) 202092d27016SAl Viro return ERR_CAST(dentry); 202120e34357SAl Viro if (unlikely(!dentry)) { 2022db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 2023db3c9adeSAl Viro if (IS_ERR(dentry)) 202492d27016SAl Viro return ERR_CAST(dentry); 202520e34357SAl Viro } 202656676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 202756676ec3SAl Viro put_link(nd); 2028b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 2029ce57dfc1SAl Viro } 2030ce57dfc1SAl Viro 20311da177e4SLinus Torvalds /* 2032bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2033bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2034bfcfaa77SLinus Torvalds * 2035bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2036bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2037bfcfaa77SLinus Torvalds * fast. 2038bfcfaa77SLinus Torvalds * 2039bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2040bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2041bfcfaa77SLinus Torvalds * crossing operation. 2042bfcfaa77SLinus Torvalds * 2043bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2044bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2045bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2046bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2047bfcfaa77SLinus Torvalds */ 2048bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2049bfcfaa77SLinus Torvalds 2050f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2051bfcfaa77SLinus Torvalds 2052468a9428SGeorge Spelvin #ifdef HASH_MIX 2053bfcfaa77SLinus Torvalds 2054468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2055468a9428SGeorge Spelvin 2056468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20572a18da7aSGeorge Spelvin /* 20582a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20592a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20602a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20612a18da7aSGeorge Spelvin * and no temporaries. 20622a18da7aSGeorge Spelvin * 20632a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20642a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20652a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20662a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20672a18da7aSGeorge Spelvin * 20682a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20692a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20702a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20712a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20722a18da7aSGeorge Spelvin * 20732a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20742a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20752a18da7aSGeorge Spelvin * toggling any given output bit. 20762a18da7aSGeorge Spelvin * 20772a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20782a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20792a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20802a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20812a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20822a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20832a18da7aSGeorge Spelvin * Perfect: 8192 258048 20842a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20852a18da7aSGeorge Spelvin */ 20862a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20872a18da7aSGeorge Spelvin ( x ^= (a), \ 20882a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20892a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20902a18da7aSGeorge Spelvin y *= 9 ) 2091bfcfaa77SLinus Torvalds 20920fed3ac8SGeorge Spelvin /* 20932a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20942a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20952a18da7aSGeorge Spelvin * work done before the hash value is used. 20960fed3ac8SGeorge Spelvin */ 20972a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20980fed3ac8SGeorge Spelvin { 20992a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 21002a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 21012a18da7aSGeorge Spelvin return y >> 32; 21020fed3ac8SGeorge Spelvin } 21030fed3ac8SGeorge Spelvin 2104bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2105bfcfaa77SLinus Torvalds 21062a18da7aSGeorge Spelvin /* 21072a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 21082a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 21092a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 21102a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 21112a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 21122a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 21132a18da7aSGeorge Spelvin * Perfect: 2048 31744 21142a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 21152a18da7aSGeorge Spelvin */ 21162a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 21172a18da7aSGeorge Spelvin ( x ^= (a), \ 21182a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 21192a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 21202a18da7aSGeorge Spelvin y *= 9 ) 2121bfcfaa77SLinus Torvalds 21222a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 21230fed3ac8SGeorge Spelvin { 21242a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 21252a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 21260fed3ac8SGeorge Spelvin } 21270fed3ac8SGeorge Spelvin 2128bfcfaa77SLinus Torvalds #endif 2129bfcfaa77SLinus Torvalds 21302a18da7aSGeorge Spelvin /* 21312a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21322a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21332a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21342a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21352a18da7aSGeorge Spelvin * finds the delimiter after the name. 21362a18da7aSGeorge Spelvin */ 21378387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2138bfcfaa77SLinus Torvalds { 21398387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2140bfcfaa77SLinus Torvalds 2141bfcfaa77SLinus Torvalds for (;;) { 2142fcfd2fbfSGeorge Spelvin if (!len) 2143fcfd2fbfSGeorge Spelvin goto done; 2144e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2145bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2146bfcfaa77SLinus Torvalds break; 21472a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2148bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2149bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2150bfcfaa77SLinus Torvalds } 21512a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2152bfcfaa77SLinus Torvalds done: 21532a18da7aSGeorge Spelvin return fold_hash(x, y); 2154bfcfaa77SLinus Torvalds } 2155bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2156bfcfaa77SLinus Torvalds 2157fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21588387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2159fcfd2fbfSGeorge Spelvin { 21608387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21618387ff25SLinus Torvalds unsigned long adata, mask, len; 2162fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2163fcfd2fbfSGeorge Spelvin 21648387ff25SLinus Torvalds len = 0; 21658387ff25SLinus Torvalds goto inside; 21668387ff25SLinus Torvalds 2167fcfd2fbfSGeorge Spelvin do { 21682a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2169fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21708387ff25SLinus Torvalds inside: 2171fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2172fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2173fcfd2fbfSGeorge Spelvin 2174fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2175fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21762a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2177fcfd2fbfSGeorge Spelvin 21782a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2179fcfd2fbfSGeorge Spelvin } 2180fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2181fcfd2fbfSGeorge Spelvin 2182bfcfaa77SLinus Torvalds /* 2183bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2184d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2185bfcfaa77SLinus Torvalds */ 21868387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2187bfcfaa77SLinus Torvalds { 21888387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21898387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 219036126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2191bfcfaa77SLinus Torvalds 21928387ff25SLinus Torvalds len = 0; 21938387ff25SLinus Torvalds goto inside; 21948387ff25SLinus Torvalds 2195bfcfaa77SLinus Torvalds do { 21962a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2197bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21988387ff25SLinus Torvalds inside: 2199e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 220036126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 220136126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2202bfcfaa77SLinus Torvalds 220336126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 220436126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 220536126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 22062a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 220736126f8fSLinus Torvalds 22082a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2209bfcfaa77SLinus Torvalds } 2210bfcfaa77SLinus Torvalds 22112a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2212bfcfaa77SLinus Torvalds 2213fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 22148387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 22150145acc2SLinus Torvalds { 22168387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 22170145acc2SLinus Torvalds while (len--) 2218fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 22190145acc2SLinus Torvalds return end_name_hash(hash); 22200145acc2SLinus Torvalds } 2221ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 22220145acc2SLinus Torvalds 2223fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 22248387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2225fcfd2fbfSGeorge Spelvin { 22268387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2227fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2228fcfd2fbfSGeorge Spelvin 2229fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2230e0ab7af9SGeorge Spelvin while (c) { 2231fcfd2fbfSGeorge Spelvin len++; 2232fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2233fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2234e0ab7af9SGeorge Spelvin } 2235fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2236fcfd2fbfSGeorge Spelvin } 2237f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2238fcfd2fbfSGeorge Spelvin 22393ddcd056SLinus Torvalds /* 2240200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2241200e9ef7SLinus Torvalds * one character. 2242200e9ef7SLinus Torvalds */ 22438387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2244200e9ef7SLinus Torvalds { 22458387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2246200e9ef7SLinus Torvalds unsigned long len = 0, c; 2247200e9ef7SLinus Torvalds 2248200e9ef7SLinus Torvalds c = (unsigned char)*name; 2249200e9ef7SLinus Torvalds do { 2250200e9ef7SLinus Torvalds len++; 2251200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2252200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2253200e9ef7SLinus Torvalds } while (c && c != '/'); 2254d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2255200e9ef7SLinus Torvalds } 2256200e9ef7SLinus Torvalds 2257bfcfaa77SLinus Torvalds #endif 2258bfcfaa77SLinus Torvalds 2259200e9ef7SLinus Torvalds /* 22601da177e4SLinus Torvalds * Name resolution. 2261ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2262ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22631da177e4SLinus Torvalds * 2264ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2265ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22661da177e4SLinus Torvalds */ 22676de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22681da177e4SLinus Torvalds { 2269d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22701da177e4SLinus Torvalds int err; 22711da177e4SLinus Torvalds 2272b4c03536SAl Viro nd->last_type = LAST_ROOT; 2273c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22749b5858e9SAl Viro if (IS_ERR(name)) 22759b5858e9SAl Viro return PTR_ERR(name); 22761da177e4SLinus Torvalds while (*name=='/') 22771da177e4SLinus Torvalds name++; 22781a97d899SAl Viro if (!*name) { 22791a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22809e18f10aSAl Viro return 0; 22811a97d899SAl Viro } 22821da177e4SLinus Torvalds 22831da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22841da177e4SLinus Torvalds for(;;) { 2285549c7297SChristian Brauner struct user_namespace *mnt_userns; 228692d27016SAl Viro const char *link; 2287d6bb3e90SLinus Torvalds u64 hash_len; 2288fe479a58SAl Viro int type; 22891da177e4SLinus Torvalds 2290549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2291549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22921da177e4SLinus Torvalds if (err) 22933595e234SAl Viro return err; 22941da177e4SLinus Torvalds 22958387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22961da177e4SLinus Torvalds 2297fe479a58SAl Viro type = LAST_NORM; 2298d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2299fe479a58SAl Viro case 2: 2300200e9ef7SLinus Torvalds if (name[1] == '.') { 2301fe479a58SAl Viro type = LAST_DOTDOT; 2302bcba1e7dSAl Viro nd->state |= ND_JUMPED; 230316c2cd71SAl Viro } 2304fe479a58SAl Viro break; 2305fe479a58SAl Viro case 1: 2306fe479a58SAl Viro type = LAST_DOT; 2307fe479a58SAl Viro } 23085a202bcdSAl Viro if (likely(type == LAST_NORM)) { 23095a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2310bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 23115a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2312a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2313da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 23145a202bcdSAl Viro if (err < 0) 23153595e234SAl Viro return err; 2316d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2317d6bb3e90SLinus Torvalds name = this.name; 23185a202bcdSAl Viro } 23195a202bcdSAl Viro } 2320fe479a58SAl Viro 2321d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2322d6bb3e90SLinus Torvalds nd->last.name = name; 23235f4a6a69SAl Viro nd->last_type = type; 23245f4a6a69SAl Viro 2325d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2326d6bb3e90SLinus Torvalds if (!*name) 2327bdf6cbf1SAl Viro goto OK; 2328200e9ef7SLinus Torvalds /* 2329200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2330200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2331200e9ef7SLinus Torvalds */ 2332200e9ef7SLinus Torvalds do { 2333d6bb3e90SLinus Torvalds name++; 2334d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 23358620c238SAl Viro if (unlikely(!*name)) { 23368620c238SAl Viro OK: 2337d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2338c108837eSAl Viro if (!depth) { 2339549c7297SChristian Brauner nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); 23400f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2341c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 23428620c238SAl Viro return 0; 2343c108837eSAl Viro } 23448620c238SAl Viro /* last component of nested symlink */ 2345d8d4611aSAl Viro name = nd->stack[--depth].name; 23468c4efe22SAl Viro link = walk_component(nd, 0); 23471c4ff1a8SAl Viro } else { 23481c4ff1a8SAl Viro /* not the last component */ 23498c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 23508620c238SAl Viro } 235192d27016SAl Viro if (unlikely(link)) { 235292d27016SAl Viro if (IS_ERR(link)) 235392d27016SAl Viro return PTR_ERR(link); 235492d27016SAl Viro /* a symlink to follow */ 2355d8d4611aSAl Viro nd->stack[depth++].name = name; 235692d27016SAl Viro name = link; 23579e18f10aSAl Viro continue; 235848c8b0c5SAl Viro } 235997242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 236097242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2361e36cffedSJens Axboe if (!try_to_unlazy(nd)) 236297242f99SAl Viro return -ECHILD; 236397242f99SAl Viro } 23643595e234SAl Viro return -ENOTDIR; 23655f4a6a69SAl Viro } 2366ce57dfc1SAl Viro } 236797242f99SAl Viro } 23681da177e4SLinus Torvalds 2369edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2370c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 237131e6b01fSNick Piggin { 2372740a1678SAleksa Sarai int error; 2373c8a53ee5SAl Viro const char *s = nd->name->name; 237431e6b01fSNick Piggin 23756c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23766c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23776c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23786c6ec2b0SJens Axboe 2379c0eb027eSLinus Torvalds if (!*s) 2380c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2381edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2382edc2b1daSAl Viro rcu_read_lock(); 2383c0eb027eSLinus Torvalds 2384bcba1e7dSAl Viro nd->flags = flags; 2385bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2386ab87f9a5SAleksa Sarai 2387ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2388ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2389ab87f9a5SAleksa Sarai smp_rmb(); 2390ab87f9a5SAleksa Sarai 2391bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2392b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2393b18825a7SDavid Howells struct inode *inode = root->d_inode; 239493893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2395368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23965b6ca027SAl Viro nd->path = nd->root; 23975b6ca027SAl Viro nd->inode = inode; 23985b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2399ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24008f47a016SAl Viro nd->root_seq = nd->seq; 24015b6ca027SAl Viro } else { 24025b6ca027SAl Viro path_get(&nd->path); 24035b6ca027SAl Viro } 2404368ee9baSAl Viro return s; 24055b6ca027SAl Viro } 24065b6ca027SAl Viro 240731e6b01fSNick Piggin nd->root.mnt = NULL; 240831e6b01fSNick Piggin 24098db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 24108db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2411740a1678SAleksa Sarai error = nd_jump_root(nd); 2412740a1678SAleksa Sarai if (unlikely(error)) 2413740a1678SAleksa Sarai return ERR_PTR(error); 2414ef55d917SAl Viro return s; 24158db52c7eSAleksa Sarai } 24168db52c7eSAleksa Sarai 24178db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 24188db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2419e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 242031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2421c28cc364SNick Piggin unsigned seq; 242231e6b01fSNick Piggin 2423c28cc364SNick Piggin do { 2424c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 242531e6b01fSNick Piggin nd->path = fs->pwd; 2426ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2427c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2428c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2429e41f7d4eSAl Viro } else { 2430e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2431ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2432e41f7d4eSAl Viro } 243331e6b01fSNick Piggin } else { 2434582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2435c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 243631e6b01fSNick Piggin struct dentry *dentry; 243731e6b01fSNick Piggin 24382903ff01SAl Viro if (!f.file) 2439368ee9baSAl Viro return ERR_PTR(-EBADF); 244031e6b01fSNick Piggin 24412903ff01SAl Viro dentry = f.file->f_path.dentry; 244231e6b01fSNick Piggin 2443edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 24442903ff01SAl Viro fdput(f); 2445368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2446f52e0c11SAl Viro } 24472903ff01SAl Viro 24482903ff01SAl Viro nd->path = f.file->f_path; 2449e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 245034a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 245134a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24525590ff0dSUlrich Drepper } else { 24532903ff01SAl Viro path_get(&nd->path); 245434a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24551da177e4SLinus Torvalds } 245634a26b99SAl Viro fdput(f); 2457e41f7d4eSAl Viro } 24588db52c7eSAleksa Sarai 2459adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2460adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2461adb21d2bSAleksa Sarai nd->root = nd->path; 2462adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2463adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2464adb21d2bSAleksa Sarai } else { 2465adb21d2bSAleksa Sarai path_get(&nd->root); 2466bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2467adb21d2bSAleksa Sarai } 2468adb21d2bSAleksa Sarai } 2469adb21d2bSAleksa Sarai return s; 24709b4a9b14SAl Viro } 24719b4a9b14SAl Viro 24721ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 247395fa25d9SAl Viro { 2474bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2475bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2476bd92d7feSAl Viro 2477c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2478bd92d7feSAl Viro } 2479bd92d7feSAl Viro 24804f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24814f757f3cSAl Viro { 2482c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2483db3c9adeSAl Viro dget(nd->path.dentry); 2484b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2485b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 24864f757f3cSAl Viro } 24874f757f3cSAl Viro 24889b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2489c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24909b4a9b14SAl Viro { 2491c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2492bd92d7feSAl Viro int err; 249331e6b01fSNick Piggin 24949b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24954f757f3cSAl Viro err = handle_lookup_down(nd); 24965f336e72SAl Viro if (unlikely(err < 0)) 24975f336e72SAl Viro s = ERR_PTR(err); 24984f757f3cSAl Viro } 24994f757f3cSAl Viro 25001ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 25011ccac622SAl Viro (s = lookup_last(nd)) != NULL) 25021ccac622SAl Viro ; 25034f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 25044f0ed93fSAl Viro err = handle_lookup_down(nd); 2505bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 25064f0ed93fSAl Viro } 25079f1fafeeSAl Viro if (!err) 25089f1fafeeSAl Viro err = complete_walk(nd); 2509bd92d7feSAl Viro 2510deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2511deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2512bd23a539SAl Viro err = -ENOTDIR; 2513625b6d10SAl Viro if (!err) { 2514625b6d10SAl Viro *path = nd->path; 2515625b6d10SAl Viro nd->path.mnt = NULL; 2516625b6d10SAl Viro nd->path.dentry = NULL; 2517625b6d10SAl Viro } 2518deb106c6SAl Viro terminate_walk(nd); 2519bd92d7feSAl Viro return err; 252031e6b01fSNick Piggin } 252131e6b01fSNick Piggin 2522794ebceaSStephen Brennan int filename_lookup(int dfd, struct filename *name, unsigned flags, 25239ad1aaa6SAl Viro struct path *path, struct path *root) 2524873f1eedSJeff Layton { 2525894bc8c4SAl Viro int retval; 25269883d185SAl Viro struct nameidata nd; 2527abc9f5beSAl Viro if (IS_ERR(name)) 2528abc9f5beSAl Viro return PTR_ERR(name); 252906422964SAl Viro set_nameidata(&nd, dfd, name, root); 2530c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2531873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2532c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2533873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2534c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2535873f1eedSJeff Layton 2536873f1eedSJeff Layton if (likely(!retval)) 2537161aff1dSAl Viro audit_inode(name, path->dentry, 2538161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 25399883d185SAl Viro restore_nameidata(); 2540020250f3SDmitry Kadashev return retval; 2541020250f3SDmitry Kadashev } 2542020250f3SDmitry Kadashev 25438bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2544c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2545391172c4SAl Viro struct path *parent) 25468bcb77faSAl Viro { 2547c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25489b5858e9SAl Viro int err = link_path_walk(s, nd); 25498bcb77faSAl Viro if (!err) 25508bcb77faSAl Viro err = complete_walk(nd); 2551391172c4SAl Viro if (!err) { 2552391172c4SAl Viro *parent = nd->path; 2553391172c4SAl Viro nd->path.mnt = NULL; 2554391172c4SAl Viro nd->path.dentry = NULL; 2555391172c4SAl Viro } 2556deb106c6SAl Viro terminate_walk(nd); 25578bcb77faSAl Viro return err; 25588bcb77faSAl Viro } 25598bcb77faSAl Viro 25600766ec82SStephen Brennan /* Note: this does not consume "name" */ 2561c5f563f9SAl Viro static int filename_parentat(int dfd, struct filename *name, 2562391172c4SAl Viro unsigned int flags, struct path *parent, 2563391172c4SAl Viro struct qstr *last, int *type) 25648bcb77faSAl Viro { 25658bcb77faSAl Viro int retval; 25669883d185SAl Viro struct nameidata nd; 25678bcb77faSAl Viro 25685c31b6ceSAl Viro if (IS_ERR(name)) 25690ee50b47SDmitry Kadashev return PTR_ERR(name); 257006422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2571c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25728bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2573c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25748bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2575c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2576391172c4SAl Viro if (likely(!retval)) { 2577391172c4SAl Viro *last = nd.last; 2578391172c4SAl Viro *type = nd.last_type; 2579c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2580391172c4SAl Viro } 25819883d185SAl Viro restore_nameidata(); 25820ee50b47SDmitry Kadashev return retval; 25830ee50b47SDmitry Kadashev } 25840ee50b47SDmitry Kadashev 258579714f72SAl Viro /* does lookup, returns the object with parent locked */ 25860766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25875590ff0dSUlrich Drepper { 25885c31b6ceSAl Viro struct dentry *d; 2589391172c4SAl Viro struct qstr last; 25900ee50b47SDmitry Kadashev int type, error; 259151689104SPaul Moore 2592c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25930ee50b47SDmitry Kadashev if (error) 25940ee50b47SDmitry Kadashev return ERR_PTR(error); 25955c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2596391172c4SAl Viro path_put(path); 25975c31b6ceSAl Viro return ERR_PTR(-EINVAL); 259879714f72SAl Viro } 25995955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2600391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 260179714f72SAl Viro if (IS_ERR(d)) { 26025955102cSAl Viro inode_unlock(path->dentry->d_inode); 2603391172c4SAl Viro path_put(path); 260479714f72SAl Viro } 260579714f72SAl Viro return d; 26065590ff0dSUlrich Drepper } 26075590ff0dSUlrich Drepper 26080766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 26090766ec82SStephen Brennan { 26100766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 26110766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 26120766ec82SStephen Brennan 26130766ec82SStephen Brennan putname(filename); 26140766ec82SStephen Brennan return res; 26150766ec82SStephen Brennan } 26160766ec82SStephen Brennan 2617d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2618d1811465SAl Viro { 2619794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2620794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2621794ebceaSStephen Brennan 2622794ebceaSStephen Brennan putname(filename); 2623794ebceaSStephen Brennan return ret; 2624794ebceaSStephen Brennan 2625d1811465SAl Viro } 26264d359507SAl Viro EXPORT_SYMBOL(kern_path); 2627d1811465SAl Viro 262816f18200SJosef 'Jeff' Sipek /** 262916f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 263016f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 263116f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 263216f18200SJosef 'Jeff' Sipek * @name: pointer to file name 263316f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2634e0a01249SAl Viro * @path: pointer to struct path to fill 263516f18200SJosef 'Jeff' Sipek */ 263616f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 263716f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2638e0a01249SAl Viro struct path *path) 263916f18200SJosef 'Jeff' Sipek { 2640794ebceaSStephen Brennan struct filename *filename; 26419ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2642794ebceaSStephen Brennan int ret; 2643794ebceaSStephen Brennan 2644794ebceaSStephen Brennan filename = getname_kernel(name); 26459ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2646794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2647794ebceaSStephen Brennan putname(filename); 2648794ebceaSStephen Brennan return ret; 264916f18200SJosef 'Jeff' Sipek } 26504d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 265116f18200SJosef 'Jeff' Sipek 2652c2fd68b6SChristian Brauner static int lookup_one_common(struct user_namespace *mnt_userns, 2653c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2654c2fd68b6SChristian Brauner struct qstr *this) 26553c95f0dcSAl Viro { 26563c95f0dcSAl Viro this->name = name; 26573c95f0dcSAl Viro this->len = len; 26583c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26593c95f0dcSAl Viro if (!len) 26603c95f0dcSAl Viro return -EACCES; 26613c95f0dcSAl Viro 26623c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26633c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26643c95f0dcSAl Viro return -EACCES; 26653c95f0dcSAl Viro } 26663c95f0dcSAl Viro 26673c95f0dcSAl Viro while (len--) { 26683c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26693c95f0dcSAl Viro if (c == '/' || c == '\0') 26703c95f0dcSAl Viro return -EACCES; 26713c95f0dcSAl Viro } 26723c95f0dcSAl Viro /* 26733c95f0dcSAl Viro * See if the low-level filesystem might want 26743c95f0dcSAl Viro * to use its own hash.. 26753c95f0dcSAl Viro */ 26763c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26773c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26783c95f0dcSAl Viro if (err < 0) 26793c95f0dcSAl Viro return err; 26803c95f0dcSAl Viro } 26813c95f0dcSAl Viro 2682c2fd68b6SChristian Brauner return inode_permission(mnt_userns, base->d_inode, MAY_EXEC); 26833c95f0dcSAl Viro } 26843c95f0dcSAl Viro 2685eead1911SChristoph Hellwig /** 26860da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26870da0b7fdSDavid Howells * @name: pathname component to lookup 26880da0b7fdSDavid Howells * @base: base directory to lookup from 26890da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26900da0b7fdSDavid Howells * 26910da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26920da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26930da0b7fdSDavid Howells * 26940da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26950da0b7fdSDavid Howells * not be called by generic code. 26960da0b7fdSDavid Howells * 26970da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26980da0b7fdSDavid Howells */ 26990da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 27000da0b7fdSDavid Howells { 27010da0b7fdSDavid Howells struct qstr this; 27020da0b7fdSDavid Howells int err; 27030da0b7fdSDavid Howells 27040da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27050da0b7fdSDavid Howells 2706c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 27070da0b7fdSDavid Howells if (err) 27080da0b7fdSDavid Howells return ERR_PTR(err); 27090da0b7fdSDavid Howells 27100da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 27110da0b7fdSDavid Howells } 27120da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 27130da0b7fdSDavid Howells 27140da0b7fdSDavid Howells /** 2715a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2716eead1911SChristoph Hellwig * @name: pathname component to lookup 2717eead1911SChristoph Hellwig * @base: base directory to lookup from 2718eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2719eead1911SChristoph Hellwig * 2720a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 27219e7543e9SAl Viro * not be called by generic code. 2722bbddca8eSNeilBrown * 2723bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2724eead1911SChristoph Hellwig */ 2725057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2726057f6c01SJames Morris { 27278613a209SAl Viro struct dentry *dentry; 2728057f6c01SJames Morris struct qstr this; 2729cda309deSMiklos Szeredi int err; 2730057f6c01SJames Morris 27315955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27322f9092e1SDavid Woodhouse 2733c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2734cda309deSMiklos Szeredi if (err) 2735cda309deSMiklos Szeredi return ERR_PTR(err); 2736cda309deSMiklos Szeredi 27378613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27388613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2739057f6c01SJames Morris } 27404d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2741057f6c01SJames Morris 2742bbddca8eSNeilBrown /** 2743c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 2744c2fd68b6SChristian Brauner * @mnt_userns: user namespace of the mount the lookup is performed from 2745c2fd68b6SChristian Brauner * @name: pathname component to lookup 2746c2fd68b6SChristian Brauner * @base: base directory to lookup from 2747c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2748c2fd68b6SChristian Brauner * 2749c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2750c2fd68b6SChristian Brauner * not be called by generic code. 2751c2fd68b6SChristian Brauner * 2752c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2753c2fd68b6SChristian Brauner */ 2754c2fd68b6SChristian Brauner struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name, 2755c2fd68b6SChristian Brauner struct dentry *base, int len) 2756c2fd68b6SChristian Brauner { 2757c2fd68b6SChristian Brauner struct dentry *dentry; 2758c2fd68b6SChristian Brauner struct qstr this; 2759c2fd68b6SChristian Brauner int err; 2760c2fd68b6SChristian Brauner 2761c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2762c2fd68b6SChristian Brauner 2763c2fd68b6SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 2764c2fd68b6SChristian Brauner if (err) 2765c2fd68b6SChristian Brauner return ERR_PTR(err); 2766c2fd68b6SChristian Brauner 2767c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2768c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2769c2fd68b6SChristian Brauner } 2770c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2771c2fd68b6SChristian Brauner 2772c2fd68b6SChristian Brauner /** 277300675017SChristian Brauner * lookup_one_unlocked - filesystem helper to lookup single pathname component 277400675017SChristian Brauner * @mnt_userns: idmapping of the mount the lookup is performed from 277500675017SChristian Brauner * @name: pathname component to lookup 277600675017SChristian Brauner * @base: base directory to lookup from 277700675017SChristian Brauner * @len: maximum length @len should be interpreted to 277800675017SChristian Brauner * 277900675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 278000675017SChristian Brauner * not be called by generic code. 278100675017SChristian Brauner * 278200675017SChristian Brauner * Unlike lookup_one_len, it should be called without the parent 278300675017SChristian Brauner * i_mutex held, and will take the i_mutex itself if necessary. 278400675017SChristian Brauner */ 278500675017SChristian Brauner struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns, 278600675017SChristian Brauner const char *name, struct dentry *base, 278700675017SChristian Brauner int len) 278800675017SChristian Brauner { 278900675017SChristian Brauner struct qstr this; 279000675017SChristian Brauner int err; 279100675017SChristian Brauner struct dentry *ret; 279200675017SChristian Brauner 279300675017SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 279400675017SChristian Brauner if (err) 279500675017SChristian Brauner return ERR_PTR(err); 279600675017SChristian Brauner 279700675017SChristian Brauner ret = lookup_dcache(&this, base, 0); 279800675017SChristian Brauner if (!ret) 279900675017SChristian Brauner ret = lookup_slow(&this, base, 0); 280000675017SChristian Brauner return ret; 280100675017SChristian Brauner } 280200675017SChristian Brauner EXPORT_SYMBOL(lookup_one_unlocked); 280300675017SChristian Brauner 280400675017SChristian Brauner /** 280500675017SChristian Brauner * lookup_one_positive_unlocked - filesystem helper to lookup single 280600675017SChristian Brauner * pathname component 280700675017SChristian Brauner * @mnt_userns: idmapping of the mount the lookup is performed from 280800675017SChristian Brauner * @name: pathname component to lookup 280900675017SChristian Brauner * @base: base directory to lookup from 281000675017SChristian Brauner * @len: maximum length @len should be interpreted to 281100675017SChristian Brauner * 281200675017SChristian Brauner * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns 281300675017SChristian Brauner * known positive or ERR_PTR(). This is what most of the users want. 281400675017SChristian Brauner * 281500675017SChristian Brauner * Note that pinned negative with unlocked parent _can_ become positive at any 281600675017SChristian Brauner * time, so callers of lookup_one_unlocked() need to be very careful; pinned 281700675017SChristian Brauner * positives have >d_inode stable, so this one avoids such problems. 281800675017SChristian Brauner * 281900675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 282000675017SChristian Brauner * not be called by generic code. 282100675017SChristian Brauner * 282200675017SChristian Brauner * The helper should be called without i_mutex held. 282300675017SChristian Brauner */ 282400675017SChristian Brauner struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns, 282500675017SChristian Brauner const char *name, 282600675017SChristian Brauner struct dentry *base, int len) 282700675017SChristian Brauner { 282800675017SChristian Brauner struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len); 282900675017SChristian Brauner 283000675017SChristian Brauner if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 283100675017SChristian Brauner dput(ret); 283200675017SChristian Brauner ret = ERR_PTR(-ENOENT); 283300675017SChristian Brauner } 283400675017SChristian Brauner return ret; 283500675017SChristian Brauner } 283600675017SChristian Brauner EXPORT_SYMBOL(lookup_one_positive_unlocked); 283700675017SChristian Brauner 283800675017SChristian Brauner /** 2839bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2840bbddca8eSNeilBrown * @name: pathname component to lookup 2841bbddca8eSNeilBrown * @base: base directory to lookup from 2842bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2843bbddca8eSNeilBrown * 2844bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2845bbddca8eSNeilBrown * not be called by generic code. 2846bbddca8eSNeilBrown * 2847bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2848bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2849bbddca8eSNeilBrown */ 2850bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2851bbddca8eSNeilBrown struct dentry *base, int len) 2852bbddca8eSNeilBrown { 285300675017SChristian Brauner return lookup_one_unlocked(&init_user_ns, name, base, len); 2854bbddca8eSNeilBrown } 2855bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2856bbddca8eSNeilBrown 28576c2d4798SAl Viro /* 28586c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 28596c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 28606c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 28616c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 28626c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 28636c2d4798SAl Viro * this one avoids such problems. 28646c2d4798SAl Viro */ 28656c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28666c2d4798SAl Viro struct dentry *base, int len) 28676c2d4798SAl Viro { 286800675017SChristian Brauner return lookup_one_positive_unlocked(&init_user_ns, name, base, len); 28696c2d4798SAl Viro } 28706c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28716c2d4798SAl Viro 2872eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2873eedf265aSEric W. Biederman int path_pts(struct path *path) 2874eedf265aSEric W. Biederman { 2875eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2876eedf265aSEric W. Biederman * the input path. 2877eedf265aSEric W. Biederman */ 2878a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2879a6a7eb76SAl Viro struct dentry *child; 288019f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2881eedf265aSEric W. Biederman 2882a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2883a6a7eb76SAl Viro dput(parent); 288463b27720SAl Viro return -ENOENT; 2885a6a7eb76SAl Viro } 288663b27720SAl Viro dput(path->dentry); 288763b27720SAl Viro path->dentry = parent; 2888eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2889eedf265aSEric W. Biederman if (!child) 2890eedf265aSEric W. Biederman return -ENOENT; 2891eedf265aSEric W. Biederman 2892eedf265aSEric W. Biederman path->dentry = child; 2893eedf265aSEric W. Biederman dput(parent); 289419f6028aSAl Viro follow_down(path); 2895eedf265aSEric W. Biederman return 0; 2896eedf265aSEric W. Biederman } 2897eedf265aSEric W. Biederman #endif 2898eedf265aSEric W. Biederman 28991fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 29001fa1e7f6SAndy Whitcroft struct path *path, int *empty) 29011da177e4SLinus Torvalds { 2902794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2903794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2904794ebceaSStephen Brennan 2905794ebceaSStephen Brennan putname(filename); 2906794ebceaSStephen Brennan return ret; 29071da177e4SLinus Torvalds } 2908b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 29091fa1e7f6SAndy Whitcroft 2910ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2911ba73d987SChristian Brauner struct inode *inode) 29121da177e4SLinus Torvalds { 29138e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2914da9592edSDavid Howells 2915ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) 29161da177e4SLinus Torvalds return 0; 2917ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) 29181da177e4SLinus Torvalds return 0; 2919ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 29201da177e4SLinus Torvalds } 2921cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 29221da177e4SLinus Torvalds 29231da177e4SLinus Torvalds /* 29241da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 29251da177e4SLinus Torvalds * whether the type of victim is right. 29261da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 29271da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 29281da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 29291da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 29301da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 29311da177e4SLinus Torvalds * a. be owner of dir, or 29321da177e4SLinus Torvalds * b. be owner of victim, or 29331da177e4SLinus Torvalds * c. have CAP_FOWNER capability 29341da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 29351da177e4SLinus Torvalds * links pointing to it. 29360bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 29370bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 29380bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 29390bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 29400bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 29411da177e4SLinus Torvalds * nfs_async_unlink(). 29421da177e4SLinus Torvalds */ 2943ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2944ba73d987SChristian Brauner struct dentry *victim, bool isdir) 29451da177e4SLinus Torvalds { 294663afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 29471da177e4SLinus Torvalds int error; 29481da177e4SLinus Torvalds 2949b18825a7SDavid Howells if (d_is_negative(victim)) 29501da177e4SLinus Torvalds return -ENOENT; 2951b18825a7SDavid Howells BUG_ON(!inode); 29521da177e4SLinus Torvalds 29531da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2954593d1ce8SEric W. Biederman 2955593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2956ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 2957ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 2958593d1ce8SEric W. Biederman return -EOVERFLOW; 2959593d1ce8SEric W. Biederman 29604fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29611da177e4SLinus Torvalds 2962ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29631da177e4SLinus Torvalds if (error) 29641da177e4SLinus Torvalds return error; 29651da177e4SLinus Torvalds if (IS_APPEND(dir)) 29661da177e4SLinus Torvalds return -EPERM; 2967b18825a7SDavid Howells 2968ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2969ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2970ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 29711da177e4SLinus Torvalds return -EPERM; 29721da177e4SLinus Torvalds if (isdir) { 297344b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29741da177e4SLinus Torvalds return -ENOTDIR; 29751da177e4SLinus Torvalds if (IS_ROOT(victim)) 29761da177e4SLinus Torvalds return -EBUSY; 297744b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29781da177e4SLinus Torvalds return -EISDIR; 29791da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29801da177e4SLinus Torvalds return -ENOENT; 29811da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29821da177e4SLinus Torvalds return -EBUSY; 29831da177e4SLinus Torvalds return 0; 29841da177e4SLinus Torvalds } 29851da177e4SLinus Torvalds 29861da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29871da177e4SLinus Torvalds * dir. 29881da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29891da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29901da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2991036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2992036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2993036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 29941da177e4SLinus Torvalds */ 2995ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2996ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 29971da177e4SLinus Torvalds { 299814e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 29991da177e4SLinus Torvalds if (child->d_inode) 30001da177e4SLinus Torvalds return -EEXIST; 30011da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 30021da177e4SLinus Torvalds return -ENOENT; 30038e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 3004036d5236SEric W. Biederman return -EOVERFLOW; 30058e538913SChristian Brauner 3006ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 30071da177e4SLinus Torvalds } 30081da177e4SLinus Torvalds 30091da177e4SLinus Torvalds /* 30101da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 30111da177e4SLinus Torvalds */ 30121da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 30131da177e4SLinus Torvalds { 30141da177e4SLinus Torvalds struct dentry *p; 30151da177e4SLinus Torvalds 30161da177e4SLinus Torvalds if (p1 == p2) { 30175955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30181da177e4SLinus Torvalds return NULL; 30191da177e4SLinus Torvalds } 30201da177e4SLinus Torvalds 3021fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 30221da177e4SLinus Torvalds 3023e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 3024e2761a11SOGAWA Hirofumi if (p) { 30255955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30265955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 30271da177e4SLinus Torvalds return p; 30281da177e4SLinus Torvalds } 30291da177e4SLinus Torvalds 3030e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 3031e2761a11SOGAWA Hirofumi if (p) { 30325955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30335955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 30341da177e4SLinus Torvalds return p; 30351da177e4SLinus Torvalds } 30361da177e4SLinus Torvalds 30375955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30385955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 30391da177e4SLinus Torvalds return NULL; 30401da177e4SLinus Torvalds } 30414d359507SAl Viro EXPORT_SYMBOL(lock_rename); 30421da177e4SLinus Torvalds 30431da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 30441da177e4SLinus Torvalds { 30455955102cSAl Viro inode_unlock(p1->d_inode); 30461da177e4SLinus Torvalds if (p1 != p2) { 30475955102cSAl Viro inode_unlock(p2->d_inode); 3048fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 30491da177e4SLinus Torvalds } 30501da177e4SLinus Torvalds } 30514d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 30521da177e4SLinus Torvalds 30536521f891SChristian Brauner /** 30546521f891SChristian Brauner * vfs_create - create new file 30556521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 30566521f891SChristian Brauner * @dir: inode of @dentry 30576521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 30586521f891SChristian Brauner * @mode: mode of the new file 30596521f891SChristian Brauner * @want_excl: whether the file must not yet exist 30606521f891SChristian Brauner * 30616521f891SChristian Brauner * Create a new file. 30626521f891SChristian Brauner * 30636521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 30646521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 30656521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 30666521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 30676521f891SChristian Brauner * raw inode simply passs init_user_ns. 30686521f891SChristian Brauner */ 30696521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 30706521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 30711da177e4SLinus Torvalds { 30726521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 30731da177e4SLinus Torvalds if (error) 30741da177e4SLinus Torvalds return error; 30751da177e4SLinus Torvalds 3076acfa4380SAl Viro if (!dir->i_op->create) 30771da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 30781da177e4SLinus Torvalds mode &= S_IALLUGO; 30791da177e4SLinus Torvalds mode |= S_IFREG; 30801da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 30811da177e4SLinus Torvalds if (error) 30821da177e4SLinus Torvalds return error; 3083549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 3084a74574aaSStephen Smalley if (!error) 3085f38aa942SAmy Griffis fsnotify_create(dir, dentry); 30861da177e4SLinus Torvalds return error; 30871da177e4SLinus Torvalds } 30884d359507SAl Viro EXPORT_SYMBOL(vfs_create); 30891da177e4SLinus Torvalds 30908e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 30918e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 30928e6c848eSAl Viro void *arg) 30938e6c848eSAl Viro { 30948e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 3095ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 30968e6c848eSAl Viro if (error) 30978e6c848eSAl Viro return error; 30988e6c848eSAl Viro 30998e6c848eSAl Viro mode &= S_IALLUGO; 31008e6c848eSAl Viro mode |= S_IFREG; 31018e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 31028e6c848eSAl Viro if (error) 31038e6c848eSAl Viro return error; 31048e6c848eSAl Viro error = f(dentry, mode, arg); 31058e6c848eSAl Viro if (!error) 31068e6c848eSAl Viro fsnotify_create(dir, dentry); 31078e6c848eSAl Viro return error; 31088e6c848eSAl Viro } 31098e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 31108e6c848eSAl Viro 3111a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3112a2982cc9SEric W. Biederman { 3113a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3114a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3115a2982cc9SEric W. Biederman } 3116a2982cc9SEric W. Biederman 3117ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 3118ba73d987SChristian Brauner int acc_mode, int flag) 31191da177e4SLinus Torvalds { 31203fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 31211da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 31221da177e4SLinus Torvalds int error; 31231da177e4SLinus Torvalds 31241da177e4SLinus Torvalds if (!inode) 31251da177e4SLinus Torvalds return -ENOENT; 31261da177e4SLinus Torvalds 3127c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3128c8fe8f30SChristoph Hellwig case S_IFLNK: 31291da177e4SLinus Torvalds return -ELOOP; 3130c8fe8f30SChristoph Hellwig case S_IFDIR: 3131fc4177beSKees Cook if (acc_mode & MAY_WRITE) 31321da177e4SLinus Torvalds return -EISDIR; 3133fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3134fc4177beSKees Cook return -EACCES; 3135c8fe8f30SChristoph Hellwig break; 3136c8fe8f30SChristoph Hellwig case S_IFBLK: 3137c8fe8f30SChristoph Hellwig case S_IFCHR: 3138a2982cc9SEric W. Biederman if (!may_open_dev(path)) 31391da177e4SLinus Torvalds return -EACCES; 3140633fb6acSKees Cook fallthrough; 3141c8fe8f30SChristoph Hellwig case S_IFIFO: 3142c8fe8f30SChristoph Hellwig case S_IFSOCK: 3143633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3144633fb6acSKees Cook return -EACCES; 31451da177e4SLinus Torvalds flag &= ~O_TRUNC; 3146c8fe8f30SChristoph Hellwig break; 31470fd338b2SKees Cook case S_IFREG: 31480fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 31490fd338b2SKees Cook return -EACCES; 31500fd338b2SKees Cook break; 31514a3fd211SDave Hansen } 3152b41572e9SDave Hansen 3153ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 3154b41572e9SDave Hansen if (error) 3155b41572e9SDave Hansen return error; 31566146f0d5SMimi Zohar 31571da177e4SLinus Torvalds /* 31581da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 31591da177e4SLinus Torvalds */ 31601da177e4SLinus Torvalds if (IS_APPEND(inode)) { 31618737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 31627715b521SAl Viro return -EPERM; 31631da177e4SLinus Torvalds if (flag & O_TRUNC) 31647715b521SAl Viro return -EPERM; 31651da177e4SLinus Torvalds } 31661da177e4SLinus Torvalds 31671da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3168ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 31697715b521SAl Viro return -EPERM; 31701da177e4SLinus Torvalds 3171f3c7691eSJ. Bruce Fields return 0; 31727715b521SAl Viro } 31737715b521SAl Viro 3174549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 31757715b521SAl Viro { 3176f0bb5aafSAl Viro const struct path *path = &filp->f_path; 31777715b521SAl Viro struct inode *inode = path->dentry->d_inode; 31787715b521SAl Viro int error = get_write_access(inode); 31791da177e4SLinus Torvalds if (error) 31807715b521SAl Viro return error; 3181482e0007SJeff Layton 3182ea0d3ab2STetsuo Handa error = security_path_truncate(path); 31831da177e4SLinus Torvalds if (!error) { 3184549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3185d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3186e1181ee6SJeff Layton filp); 31871da177e4SLinus Torvalds } 31881da177e4SLinus Torvalds put_write_access(inode); 3189acd0c935SMimi Zohar return error; 31901da177e4SLinus Torvalds } 31911da177e4SLinus Torvalds 3192d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3193d57999e1SDave Hansen { 31948a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 31958a5e929dSAl Viro flag--; 3196d57999e1SDave Hansen return flag; 3197d57999e1SDave Hansen } 3198d57999e1SDave Hansen 3199ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3200ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3201ba73d987SChristian Brauner umode_t mode) 3202d18e9008SMiklos Szeredi { 3203d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3204d18e9008SMiklos Szeredi if (error) 3205d18e9008SMiklos Szeredi return error; 3206d18e9008SMiklos Szeredi 32078e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) 32081328c727SSeth Forshee return -EOVERFLOW; 32091328c727SSeth Forshee 3210ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 321147291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3212d18e9008SMiklos Szeredi if (error) 3213d18e9008SMiklos Szeredi return error; 3214d18e9008SMiklos Szeredi 3215d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3216d18e9008SMiklos Szeredi } 3217d18e9008SMiklos Szeredi 32181acf0af9SDavid Howells /* 32191acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 32201acf0af9SDavid Howells * dentry. 32211acf0af9SDavid Howells * 32221acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 32231acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 32241acf0af9SDavid Howells * 322500a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 322600a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 322700a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 32281acf0af9SDavid Howells * 32291acf0af9SDavid Howells * Returns an error code otherwise. 32301acf0af9SDavid Howells */ 3231239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3232239eb983SAl Viro struct file *file, 32333ec2eef1SAl Viro int open_flag, umode_t mode) 3234d18e9008SMiklos Szeredi { 3235d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3236d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3237d18e9008SMiklos Szeredi int error; 3238d18e9008SMiklos Szeredi 3239d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3240d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3241d18e9008SMiklos Szeredi 324230d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 324330d90494SAl Viro file->f_path.mnt = nd->path.mnt; 32440fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 324544907d79SAl Viro open_to_namei_flags(open_flag), mode); 32466fbd0714SAl Viro d_lookup_done(dentry); 3247384f26e2SAl Viro if (!error) { 324864e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 32496fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 32506fb968cdSAl Viro dput(dentry); 32516fb968cdSAl Viro dentry = dget(file->f_path.dentry); 32526fb968cdSAl Viro } 325364e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 32542675a4ebSAl Viro error = -EIO; 3255384f26e2SAl Viro } else { 325630d90494SAl Viro if (file->f_path.dentry) { 3257d18e9008SMiklos Szeredi dput(dentry); 3258d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 325910c64ceaSAl Viro } 3260239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3261a01e718fSAl Viro error = -ENOENT; 3262d18e9008SMiklos Szeredi } 3263d18e9008SMiklos Szeredi } 3264239eb983SAl Viro if (error) { 3265d18e9008SMiklos Szeredi dput(dentry); 3266239eb983SAl Viro dentry = ERR_PTR(error); 3267239eb983SAl Viro } 3268239eb983SAl Viro return dentry; 3269d18e9008SMiklos Szeredi } 3270d18e9008SMiklos Szeredi 327131e6b01fSNick Piggin /* 32721acf0af9SDavid Howells * Look up and maybe create and open the last component. 3273d58ffd35SMiklos Szeredi * 327400a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3275d58ffd35SMiklos Szeredi * 327600a07c15SAl Viro * Returns 0 on success, that is, if 327700a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 327800a07c15SAl Viro * the file was not completely opened at this time, though lookups and 327900a07c15SAl Viro * creations were performed. 328000a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 328100a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 328200a07c15SAl Viro * hadn't been specified. 32831acf0af9SDavid Howells * 328400a07c15SAl Viro * An error code is returned on failure. 3285d58ffd35SMiklos Szeredi */ 3286da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3287d58ffd35SMiklos Szeredi const struct open_flags *op, 32883ec2eef1SAl Viro bool got_write) 3289d58ffd35SMiklos Szeredi { 3290549c7297SChristian Brauner struct user_namespace *mnt_userns; 3291d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 329254ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 32931643b43fSAl Viro int open_flag = op->open_flag; 3294d58ffd35SMiklos Szeredi struct dentry *dentry; 32951643b43fSAl Viro int error, create_error = 0; 32961643b43fSAl Viro umode_t mode = op->mode; 32976fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3298d58ffd35SMiklos Szeredi 3299ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3300da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3301d58ffd35SMiklos Szeredi 330273a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 33036fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 33046fbd0714SAl Viro for (;;) { 33056fbd0714SAl Viro if (!dentry) { 33066fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3307d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3308da5ebf5aSAl Viro return dentry; 33096fbd0714SAl Viro } 33106fbd0714SAl Viro if (d_in_lookup(dentry)) 33116fbd0714SAl Viro break; 3312d58ffd35SMiklos Szeredi 33136fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 33146fbd0714SAl Viro if (likely(error > 0)) 33156fbd0714SAl Viro break; 33166fbd0714SAl Viro if (error) 33176fbd0714SAl Viro goto out_dput; 33186fbd0714SAl Viro d_invalidate(dentry); 33196fbd0714SAl Viro dput(dentry); 33206fbd0714SAl Viro dentry = NULL; 33216fbd0714SAl Viro } 33226fbd0714SAl Viro if (dentry->d_inode) { 3323d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3324da5ebf5aSAl Viro return dentry; 33256c51e513SAl Viro } 3326d18e9008SMiklos Szeredi 33271643b43fSAl Viro /* 33281643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 33291643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 33301643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 33311643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 33321643b43fSAl Viro * 33331643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 33341643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 33351643b43fSAl Viro */ 333699a4a90cSAl Viro if (unlikely(!got_write)) 333799a4a90cSAl Viro open_flag &= ~O_TRUNC; 3338549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 33391643b43fSAl Viro if (open_flag & O_CREAT) { 334099a4a90cSAl Viro if (open_flag & O_EXCL) 334199a4a90cSAl Viro open_flag &= ~O_TRUNC; 33421643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 33431643b43fSAl Viro mode &= ~current_umask(); 334499a4a90cSAl Viro if (likely(got_write)) 3345549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3346ba73d987SChristian Brauner dentry, mode); 334799a4a90cSAl Viro else 334899a4a90cSAl Viro create_error = -EROFS; 334999a4a90cSAl Viro } 335099a4a90cSAl Viro if (create_error) 33511643b43fSAl Viro open_flag &= ~O_CREAT; 33521643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3353d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3354da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3355da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3356da5ebf5aSAl Viro return dentry; 3357239eb983SAl Viro } 335854ef4872SMiklos Szeredi 33596fbd0714SAl Viro if (d_in_lookup(dentry)) { 336012fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 336112fa5e24SAl Viro nd->flags); 33626fbd0714SAl Viro d_lookup_done(dentry); 336312fa5e24SAl Viro if (unlikely(res)) { 336412fa5e24SAl Viro if (IS_ERR(res)) { 336512fa5e24SAl Viro error = PTR_ERR(res); 336612fa5e24SAl Viro goto out_dput; 336712fa5e24SAl Viro } 336812fa5e24SAl Viro dput(dentry); 336912fa5e24SAl Viro dentry = res; 337012fa5e24SAl Viro } 337154ef4872SMiklos Szeredi } 337254ef4872SMiklos Szeredi 3373d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 33741643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 337573a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3376ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3377ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3378ce8644fcSAl Viro error = -EACCES; 3379d58ffd35SMiklos Szeredi goto out_dput; 338064894cf8SAl Viro } 3381549c7297SChristian Brauner 3382549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3383549c7297SChristian Brauner mode, open_flag & O_EXCL); 3384d58ffd35SMiklos Szeredi if (error) 3385d58ffd35SMiklos Szeredi goto out_dput; 3386d58ffd35SMiklos Szeredi } 33871643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 33881643b43fSAl Viro error = create_error; 3389d58ffd35SMiklos Szeredi goto out_dput; 3390d58ffd35SMiklos Szeredi } 3391da5ebf5aSAl Viro return dentry; 3392d58ffd35SMiklos Szeredi 3393d58ffd35SMiklos Szeredi out_dput: 3394d58ffd35SMiklos Szeredi dput(dentry); 3395da5ebf5aSAl Viro return ERR_PTR(error); 3396d58ffd35SMiklos Szeredi } 3397d58ffd35SMiklos Szeredi 3398c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 33993ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3400fb1cc555SAl Viro { 3401a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3402ca344a89SAl Viro int open_flag = op->open_flag; 340364894cf8SAl Viro bool got_write = false; 3404254cf582SAl Viro unsigned seq; 3405a1eb3315SMiklos Szeredi struct inode *inode; 3406da5ebf5aSAl Viro struct dentry *dentry; 3407b0417d2cSAl Viro const char *res; 3408fb1cc555SAl Viro 3409c3e380b0SAl Viro nd->flags |= op->intent; 3410c3e380b0SAl Viro 3411bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 341256676ec3SAl Viro if (nd->depth) 341356676ec3SAl Viro put_link(nd); 3414ff326a32SAl Viro return handle_dots(nd, nd->last_type); 34151f36f774SAl Viro } 3416a2c36b45SAl Viro 3417ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3418fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3419fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3420fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 342120e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 342220e34357SAl Viro if (IS_ERR(dentry)) 34231ccac622SAl Viro return ERR_CAST(dentry); 342420e34357SAl Viro if (likely(dentry)) 342571574865SMiklos Szeredi goto finish_lookup; 342671574865SMiklos Szeredi 34276583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3428b6183df7SMiklos Szeredi } else { 3429fe2d35ffSAl Viro /* create side of things */ 343072287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3431e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3432e36cffedSJens Axboe return ERR_PTR(-ECHILD); 343372287417SAl Viro } 3434c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 34351f36f774SAl Viro /* trailing slashes? */ 3436deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 34371ccac622SAl Viro return ERR_PTR(-EISDIR); 3438b6183df7SMiklos Szeredi } 34391f36f774SAl Viro 34409cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3441e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 344264894cf8SAl Viro /* 344364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 344464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 344564894cf8SAl Viro * dropping this one anyway. 344664894cf8SAl Viro */ 344764894cf8SAl Viro } 34489cf843e3SAl Viro if (open_flag & O_CREAT) 34495955102cSAl Viro inode_lock(dir->d_inode); 34509cf843e3SAl Viro else 34519cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3452da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3453f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3454f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 34559cf843e3SAl Viro if (open_flag & O_CREAT) 34565955102cSAl Viro inode_unlock(dir->d_inode); 34579cf843e3SAl Viro else 34589cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3459fb1cc555SAl Viro 3460c981a482SAl Viro if (got_write) 346159e96e65SAl Viro mnt_drop_write(nd->path.mnt); 34626c0d46c4SAl Viro 346359e96e65SAl Viro if (IS_ERR(dentry)) 346459e96e65SAl Viro return ERR_CAST(dentry); 346559e96e65SAl Viro 3466973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3467e73cabffSAl Viro dput(nd->path.dentry); 3468e73cabffSAl Viro nd->path.dentry = dentry; 3469c981a482SAl Viro return NULL; 3470fb1cc555SAl Viro } 3471fb1cc555SAl Viro 347220e34357SAl Viro finish_lookup: 347356676ec3SAl Viro if (nd->depth) 347456676ec3SAl Viro put_link(nd); 34758c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3476ff326a32SAl Viro if (unlikely(res)) 34771ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3478b0417d2cSAl Viro return res; 34791ccac622SAl Viro } 348031d1726dSAl Viro 3481c981a482SAl Viro /* 3482c981a482SAl Viro * Handle the last step of open() 3483c981a482SAl Viro */ 3484c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3485c981a482SAl Viro struct file *file, const struct open_flags *op) 3486c981a482SAl Viro { 3487549c7297SChristian Brauner struct user_namespace *mnt_userns; 3488c981a482SAl Viro int open_flag = op->open_flag; 3489c981a482SAl Viro bool do_truncate; 3490c981a482SAl Viro int acc_mode; 3491c981a482SAl Viro int error; 3492c981a482SAl Viro 3493ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3494ff326a32SAl Viro error = complete_walk(nd); 3495ff326a32SAl Viro if (error) 3496ff326a32SAl Viro return error; 3497ff326a32SAl Viro } 3498973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 349976ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3500549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 350130aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3502b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3503b94e0b32SAl Viro return -EEXIST; 350430aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3505c5971b8cSAl Viro return -EISDIR; 3506549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 350730aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 350830aba665SSalvatore Mesoraca if (unlikely(error)) 3509c5971b8cSAl Viro return error; 351030aba665SSalvatore Mesoraca } 351144b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3512c5971b8cSAl Viro return -ENOTDIR; 35136c0d46c4SAl Viro 35148795e7d4SAl Viro do_truncate = false; 35158795e7d4SAl Viro acc_mode = op->acc_mode; 35165a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 35175a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 35185a2d3eddSAl Viro open_flag &= ~O_TRUNC; 35195a2d3eddSAl Viro acc_mode = 0; 35208795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 35210f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 35220f9d1a10SAl Viro if (error) 3523c5971b8cSAl Viro return error; 35248795e7d4SAl Viro do_truncate = true; 35250f9d1a10SAl Viro } 3526549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 35278795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3528ae2bb293SAl Viro error = vfs_open(&nd->path, file); 35298795e7d4SAl Viro if (!error) 35306035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 35318795e7d4SAl Viro if (!error && do_truncate) 3532549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3533c80567c8SAl Viro if (unlikely(error > 0)) { 3534c80567c8SAl Viro WARN_ON(1); 3535c80567c8SAl Viro error = -EINVAL; 3536c80567c8SAl Viro } 35378795e7d4SAl Viro if (do_truncate) 35380f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3539c5971b8cSAl Viro return error; 3540fb1cc555SAl Viro } 3541fb1cc555SAl Viro 35426521f891SChristian Brauner /** 35436521f891SChristian Brauner * vfs_tmpfile - create tmpfile 35446521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 35456521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 35466521f891SChristian Brauner * @mode: mode of the new tmpfile 35472111c3c0SRandy Dunlap * @open_flag: flags 35486521f891SChristian Brauner * 35496521f891SChristian Brauner * Create a temporary file. 35506521f891SChristian Brauner * 35516521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 35526521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 35536521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 35546521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 35556521f891SChristian Brauner * raw inode simply passs init_user_ns. 35566521f891SChristian Brauner */ 35576521f891SChristian Brauner struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, 35586521f891SChristian Brauner struct dentry *dentry, umode_t mode, int open_flag) 3559af7bd4dcSAmir Goldstein { 3560af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3561af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3562af7bd4dcSAmir Goldstein struct inode *inode; 3563af7bd4dcSAmir Goldstein int error; 3564af7bd4dcSAmir Goldstein 3565af7bd4dcSAmir Goldstein /* we want directory to be writable */ 35666521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3567af7bd4dcSAmir Goldstein if (error) 3568af7bd4dcSAmir Goldstein goto out_err; 3569af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3570af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3571af7bd4dcSAmir Goldstein goto out_err; 3572af7bd4dcSAmir Goldstein error = -ENOMEM; 3573cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3574af7bd4dcSAmir Goldstein if (unlikely(!child)) 3575af7bd4dcSAmir Goldstein goto out_err; 3576549c7297SChristian Brauner error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); 3577af7bd4dcSAmir Goldstein if (error) 3578af7bd4dcSAmir Goldstein goto out_err; 3579af7bd4dcSAmir Goldstein error = -ENOENT; 3580af7bd4dcSAmir Goldstein inode = child->d_inode; 3581af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3582af7bd4dcSAmir Goldstein goto out_err; 3583af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3584af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3585af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3586af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3587af7bd4dcSAmir Goldstein } 3588a2d2329eSChristian Brauner ima_post_create_tmpfile(mnt_userns, inode); 3589af7bd4dcSAmir Goldstein return child; 3590af7bd4dcSAmir Goldstein 3591af7bd4dcSAmir Goldstein out_err: 3592af7bd4dcSAmir Goldstein dput(child); 3593af7bd4dcSAmir Goldstein return ERR_PTR(error); 3594af7bd4dcSAmir Goldstein } 3595af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3596af7bd4dcSAmir Goldstein 3597c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 359860545d0dSAl Viro const struct open_flags *op, 35993ec2eef1SAl Viro struct file *file) 360060545d0dSAl Viro { 36016521f891SChristian Brauner struct user_namespace *mnt_userns; 3602625b6d10SAl Viro struct dentry *child; 3603625b6d10SAl Viro struct path path; 3604c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 360560545d0dSAl Viro if (unlikely(error)) 360660545d0dSAl Viro return error; 3607625b6d10SAl Viro error = mnt_want_write(path.mnt); 360860545d0dSAl Viro if (unlikely(error)) 360960545d0dSAl Viro goto out; 36106521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 36116521f891SChristian Brauner child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); 3612af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3613684e73beSHirofumi Nakagawa if (IS_ERR(child)) 361460545d0dSAl Viro goto out2; 3615625b6d10SAl Viro dput(path.dentry); 3616625b6d10SAl Viro path.dentry = child; 3617c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 361869a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3619549c7297SChristian Brauner error = may_open(mnt_userns, &path, 0, op->open_flag); 36201e8f44f1SAl Viro if (!error) 36211e8f44f1SAl Viro error = vfs_open(&path, file); 362260545d0dSAl Viro out2: 3623625b6d10SAl Viro mnt_drop_write(path.mnt); 362460545d0dSAl Viro out: 3625625b6d10SAl Viro path_put(&path); 362660545d0dSAl Viro return error; 362760545d0dSAl Viro } 362860545d0dSAl Viro 36296ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 36306ac08709SAl Viro { 36316ac08709SAl Viro struct path path; 36326ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 36336ac08709SAl Viro if (!error) { 36346ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3635ae2bb293SAl Viro error = vfs_open(&path, file); 36366ac08709SAl Viro path_put(&path); 36376ac08709SAl Viro } 36386ac08709SAl Viro return error; 36396ac08709SAl Viro } 36406ac08709SAl Viro 3641c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3642c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 36431da177e4SLinus Torvalds { 364430d90494SAl Viro struct file *file; 364513aab428SAl Viro int error; 364631e6b01fSNick Piggin 3647ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 36481afc99beSAl Viro if (IS_ERR(file)) 36491afc99beSAl Viro return file; 365031e6b01fSNick Piggin 3651bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 36523ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 36535f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 36546ac08709SAl Viro error = do_o_path(nd, flags, file); 36555f336e72SAl Viro } else { 36565f336e72SAl Viro const char *s = path_init(nd, flags); 36573bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3658c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 36591ccac622SAl Viro ; 3660c5971b8cSAl Viro if (!error) 3661c5971b8cSAl Viro error = do_open(nd, file, op); 3662deb106c6SAl Viro terminate_walk(nd); 36635f336e72SAl Viro } 36647c1c01ecSAl Viro if (likely(!error)) { 3665aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 36667c1c01ecSAl Viro return file; 36677c1c01ecSAl Viro WARN_ON(1); 36687c1c01ecSAl Viro error = -EINVAL; 3669015c3bbcSMiklos Szeredi } 36707c1c01ecSAl Viro fput(file); 36712675a4ebSAl Viro if (error == -EOPENSTALE) { 36722675a4ebSAl Viro if (flags & LOOKUP_RCU) 36732675a4ebSAl Viro error = -ECHILD; 36742675a4ebSAl Viro else 36752675a4ebSAl Viro error = -ESTALE; 36762675a4ebSAl Viro } 36777c1c01ecSAl Viro return ERR_PTR(error); 3678de459215SKirill Korotaev } 36791da177e4SLinus Torvalds 3680669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3681f9652e10SAl Viro const struct open_flags *op) 368213aab428SAl Viro { 36839883d185SAl Viro struct nameidata nd; 3684f9652e10SAl Viro int flags = op->lookup_flags; 368513aab428SAl Viro struct file *filp; 368613aab428SAl Viro 368706422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3688c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 368913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3690c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 369113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3692c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 36939883d185SAl Viro restore_nameidata(); 369413aab428SAl Viro return filp; 369513aab428SAl Viro } 369613aab428SAl Viro 3697ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3698f9652e10SAl Viro const char *name, const struct open_flags *op) 369973d049a4SAl Viro { 37009883d185SAl Viro struct nameidata nd; 370173d049a4SAl Viro struct file *file; 370251689104SPaul Moore struct filename *filename; 3703bcba1e7dSAl Viro int flags = op->lookup_flags; 370473d049a4SAl Viro 3705ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 370673d049a4SAl Viro return ERR_PTR(-ELOOP); 370773d049a4SAl Viro 370851689104SPaul Moore filename = getname_kernel(name); 3709a1c83681SViresh Kumar if (IS_ERR(filename)) 371051689104SPaul Moore return ERR_CAST(filename); 371151689104SPaul Moore 371206422964SAl Viro set_nameidata(&nd, -1, filename, root); 3713c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 371473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3715c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 371673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3717c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 37189883d185SAl Viro restore_nameidata(); 371951689104SPaul Moore putname(filename); 372073d049a4SAl Viro return file; 372173d049a4SAl Viro } 372273d049a4SAl Viro 3723b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 37241ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 37251da177e4SLinus Torvalds { 3726c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3727391172c4SAl Viro struct qstr last; 3728b3d4650dSNeilBrown bool want_dir = lookup_flags & LOOKUP_DIRECTORY; 3729b3d4650dSNeilBrown unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; 3730b3d4650dSNeilBrown unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; 3731391172c4SAl Viro int type; 3732c30dabfeSJan Kara int err2; 37331ac12b4bSJeff Layton int error; 37341ac12b4bSJeff Layton 3735b3d4650dSNeilBrown error = filename_parentat(dfd, name, reval_flag, path, &last, &type); 37360ee50b47SDmitry Kadashev if (error) 37370ee50b47SDmitry Kadashev return ERR_PTR(error); 37381da177e4SLinus Torvalds 3739c663e5d8SChristoph Hellwig /* 3740c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3741c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3742c663e5d8SChristoph Hellwig */ 37435c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3744ed75e95dSAl Viro goto out; 3745c663e5d8SChristoph Hellwig 3746c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3747391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3748c663e5d8SChristoph Hellwig /* 3749b3d4650dSNeilBrown * Do the final lookup. Suppress 'create' if there is a trailing 3750b3d4650dSNeilBrown * '/', and a directory wasn't requested. 3751c663e5d8SChristoph Hellwig */ 3752b3d4650dSNeilBrown if (last.name[last.len] && !want_dir) 3753b3d4650dSNeilBrown create_flags = 0; 37545955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3755b3d4650dSNeilBrown dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags); 37561da177e4SLinus Torvalds if (IS_ERR(dentry)) 3757a8104a9fSAl Viro goto unlock; 3758c663e5d8SChristoph Hellwig 3759a8104a9fSAl Viro error = -EEXIST; 3760b18825a7SDavid Howells if (d_is_positive(dentry)) 3761a8104a9fSAl Viro goto fail; 3762b18825a7SDavid Howells 3763c663e5d8SChristoph Hellwig /* 3764c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3765c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3766c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3767c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3768c663e5d8SChristoph Hellwig */ 3769b3d4650dSNeilBrown if (unlikely(!create_flags)) { 3770a8104a9fSAl Viro error = -ENOENT; 3771ed75e95dSAl Viro goto fail; 3772e9baf6e5SAl Viro } 3773c30dabfeSJan Kara if (unlikely(err2)) { 3774c30dabfeSJan Kara error = err2; 3775a8104a9fSAl Viro goto fail; 3776c30dabfeSJan Kara } 3777e9baf6e5SAl Viro return dentry; 37781da177e4SLinus Torvalds fail: 3779a8104a9fSAl Viro dput(dentry); 3780a8104a9fSAl Viro dentry = ERR_PTR(error); 3781a8104a9fSAl Viro unlock: 37825955102cSAl Viro inode_unlock(path->dentry->d_inode); 3783c30dabfeSJan Kara if (!err2) 3784391172c4SAl Viro mnt_drop_write(path->mnt); 3785ed75e95dSAl Viro out: 3786391172c4SAl Viro path_put(path); 3787ed75e95dSAl Viro return dentry; 3788dae6ad8fSAl Viro } 3789fa14a0b8SAl Viro 3790fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3791fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3792fa14a0b8SAl Viro { 3793b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3794b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3795b4a4f213SStephen Brennan 3796b4a4f213SStephen Brennan putname(filename); 3797b4a4f213SStephen Brennan return res; 3798fa14a0b8SAl Viro } 3799dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3800dae6ad8fSAl Viro 3801921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3802921a1650SAl Viro { 3803921a1650SAl Viro dput(dentry); 38045955102cSAl Viro inode_unlock(path->dentry->d_inode); 3805a8104a9fSAl Viro mnt_drop_write(path->mnt); 3806921a1650SAl Viro path_put(path); 3807921a1650SAl Viro } 3808921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3809921a1650SAl Viro 3810520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 38111ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3812dae6ad8fSAl Viro { 3813b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3814b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3815b4a4f213SStephen Brennan 3816b4a4f213SStephen Brennan putname(filename); 3817b4a4f213SStephen Brennan return res; 3818dae6ad8fSAl Viro } 3819dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3820dae6ad8fSAl Viro 38216521f891SChristian Brauner /** 38226521f891SChristian Brauner * vfs_mknod - create device node or file 38236521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 38246521f891SChristian Brauner * @dir: inode of @dentry 38256521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 38266521f891SChristian Brauner * @mode: mode of the new device node or file 38276521f891SChristian Brauner * @dev: device number of device to create 38286521f891SChristian Brauner * 38296521f891SChristian Brauner * Create a device node or file. 38306521f891SChristian Brauner * 38316521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 38326521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 38336521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 38346521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 38356521f891SChristian Brauner * raw inode simply passs init_user_ns. 38366521f891SChristian Brauner */ 38376521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 38386521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 38391da177e4SLinus Torvalds { 3840a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 38416521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 38421da177e4SLinus Torvalds 38431da177e4SLinus Torvalds if (error) 38441da177e4SLinus Torvalds return error; 38451da177e4SLinus Torvalds 3846a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3847a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 38481da177e4SLinus Torvalds return -EPERM; 38491da177e4SLinus Torvalds 3850acfa4380SAl Viro if (!dir->i_op->mknod) 38511da177e4SLinus Torvalds return -EPERM; 38521da177e4SLinus Torvalds 385308ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 385408ce5f16SSerge E. Hallyn if (error) 385508ce5f16SSerge E. Hallyn return error; 385608ce5f16SSerge E. Hallyn 38571da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 38581da177e4SLinus Torvalds if (error) 38591da177e4SLinus Torvalds return error; 38601da177e4SLinus Torvalds 3861549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3862a74574aaSStephen Smalley if (!error) 3863f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38641da177e4SLinus Torvalds return error; 38651da177e4SLinus Torvalds } 38664d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 38671da177e4SLinus Torvalds 3868f69aac00SAl Viro static int may_mknod(umode_t mode) 3869463c3197SDave Hansen { 3870463c3197SDave Hansen switch (mode & S_IFMT) { 3871463c3197SDave Hansen case S_IFREG: 3872463c3197SDave Hansen case S_IFCHR: 3873463c3197SDave Hansen case S_IFBLK: 3874463c3197SDave Hansen case S_IFIFO: 3875463c3197SDave Hansen case S_IFSOCK: 3876463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3877463c3197SDave Hansen return 0; 3878463c3197SDave Hansen case S_IFDIR: 3879463c3197SDave Hansen return -EPERM; 3880463c3197SDave Hansen default: 3881463c3197SDave Hansen return -EINVAL; 3882463c3197SDave Hansen } 3883463c3197SDave Hansen } 3884463c3197SDave Hansen 388545f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 388687c4e192SDominik Brodowski unsigned int dev) 38871da177e4SLinus Torvalds { 38886521f891SChristian Brauner struct user_namespace *mnt_userns; 38891da177e4SLinus Torvalds struct dentry *dentry; 3890dae6ad8fSAl Viro struct path path; 3891dae6ad8fSAl Viro int error; 3892972567f1SJeff Layton unsigned int lookup_flags = 0; 38931da177e4SLinus Torvalds 38948e4bfca1SAl Viro error = may_mknod(mode); 38958e4bfca1SAl Viro if (error) 38967797251bSDmitry Kadashev goto out1; 3897972567f1SJeff Layton retry: 3898b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 38997797251bSDmitry Kadashev error = PTR_ERR(dentry); 3900dae6ad8fSAl Viro if (IS_ERR(dentry)) 39017797251bSDmitry Kadashev goto out1; 39022ad94ae6SAl Viro 3903dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3904ce3b0f8dSAl Viro mode &= ~current_umask(); 3905dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3906be6d3e56SKentaro Takeda if (error) 39077797251bSDmitry Kadashev goto out2; 39086521f891SChristian Brauner 39096521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 39101da177e4SLinus Torvalds switch (mode & S_IFMT) { 39111da177e4SLinus Torvalds case 0: case S_IFREG: 39126521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 39136521f891SChristian Brauner dentry, mode, true); 391405d1a717SMimi Zohar if (!error) 3915a2d2329eSChristian Brauner ima_post_path_mknod(mnt_userns, dentry); 39161da177e4SLinus Torvalds break; 39171da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 39186521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 39196521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 39201da177e4SLinus Torvalds break; 39211da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 39226521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 39236521f891SChristian Brauner dentry, mode, 0); 39241da177e4SLinus Torvalds break; 39251da177e4SLinus Torvalds } 39267797251bSDmitry Kadashev out2: 3927921a1650SAl Viro done_path_create(&path, dentry); 3928972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3929972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3930972567f1SJeff Layton goto retry; 3931972567f1SJeff Layton } 39327797251bSDmitry Kadashev out1: 39337797251bSDmitry Kadashev putname(name); 39341da177e4SLinus Torvalds return error; 39351da177e4SLinus Torvalds } 39361da177e4SLinus Torvalds 393787c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 393887c4e192SDominik Brodowski unsigned int, dev) 393987c4e192SDominik Brodowski { 39407797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 394187c4e192SDominik Brodowski } 394287c4e192SDominik Brodowski 39438208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 39445590ff0dSUlrich Drepper { 39457797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 39465590ff0dSUlrich Drepper } 39475590ff0dSUlrich Drepper 39486521f891SChristian Brauner /** 39496521f891SChristian Brauner * vfs_mkdir - create directory 39506521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 39516521f891SChristian Brauner * @dir: inode of @dentry 39526521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39536521f891SChristian Brauner * @mode: mode of the new directory 39546521f891SChristian Brauner * 39556521f891SChristian Brauner * Create a directory. 39566521f891SChristian Brauner * 39576521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39586521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39596521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39606521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39616521f891SChristian Brauner * raw inode simply passs init_user_ns. 39626521f891SChristian Brauner */ 39636521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 39646521f891SChristian Brauner struct dentry *dentry, umode_t mode) 39651da177e4SLinus Torvalds { 39666521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 39678de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39681da177e4SLinus Torvalds 39691da177e4SLinus Torvalds if (error) 39701da177e4SLinus Torvalds return error; 39711da177e4SLinus Torvalds 3972acfa4380SAl Viro if (!dir->i_op->mkdir) 39731da177e4SLinus Torvalds return -EPERM; 39741da177e4SLinus Torvalds 39751da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 39761da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 39771da177e4SLinus Torvalds if (error) 39781da177e4SLinus Torvalds return error; 39791da177e4SLinus Torvalds 39808de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 39818de52778SAl Viro return -EMLINK; 39828de52778SAl Viro 3983549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 3984a74574aaSStephen Smalley if (!error) 3985f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 39861da177e4SLinus Torvalds return error; 39871da177e4SLinus Torvalds } 39884d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 39891da177e4SLinus Torvalds 399045f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 39911da177e4SLinus Torvalds { 39926902d925SDave Hansen struct dentry *dentry; 3993dae6ad8fSAl Viro struct path path; 3994dae6ad8fSAl Viro int error; 3995b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 39961da177e4SLinus Torvalds 3997b76d8b82SJeff Layton retry: 3998b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 3999584d3226SDmitry Kadashev error = PTR_ERR(dentry); 40006902d925SDave Hansen if (IS_ERR(dentry)) 4001584d3226SDmitry Kadashev goto out_putname; 40026902d925SDave Hansen 4003dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 4004ce3b0f8dSAl Viro mode &= ~current_umask(); 4005dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 40066521f891SChristian Brauner if (!error) { 40076521f891SChristian Brauner struct user_namespace *mnt_userns; 40086521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4009549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 4010549c7297SChristian Brauner mode); 40116521f891SChristian Brauner } 4012921a1650SAl Viro done_path_create(&path, dentry); 4013b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 4014b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 4015b76d8b82SJeff Layton goto retry; 4016b76d8b82SJeff Layton } 4017584d3226SDmitry Kadashev out_putname: 4018584d3226SDmitry Kadashev putname(name); 40191da177e4SLinus Torvalds return error; 40201da177e4SLinus Torvalds } 40211da177e4SLinus Torvalds 40220101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 40230101db7aSDominik Brodowski { 4024584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 40250101db7aSDominik Brodowski } 40260101db7aSDominik Brodowski 4027a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 40285590ff0dSUlrich Drepper { 4029584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 40305590ff0dSUlrich Drepper } 40315590ff0dSUlrich Drepper 40326521f891SChristian Brauner /** 40336521f891SChristian Brauner * vfs_rmdir - remove directory 40346521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 40356521f891SChristian Brauner * @dir: inode of @dentry 40366521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 40376521f891SChristian Brauner * 40386521f891SChristian Brauner * Remove a directory. 40396521f891SChristian Brauner * 40406521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 40416521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 40426521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 40436521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 40446521f891SChristian Brauner * raw inode simply passs init_user_ns. 40456521f891SChristian Brauner */ 40466521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 40476521f891SChristian Brauner struct dentry *dentry) 40481da177e4SLinus Torvalds { 40496521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 40501da177e4SLinus Torvalds 40511da177e4SLinus Torvalds if (error) 40521da177e4SLinus Torvalds return error; 40531da177e4SLinus Torvalds 4054acfa4380SAl Viro if (!dir->i_op->rmdir) 40551da177e4SLinus Torvalds return -EPERM; 40561da177e4SLinus Torvalds 40571d2ef590SAl Viro dget(dentry); 40585955102cSAl Viro inode_lock(dentry->d_inode); 4059912dbc15SSage Weil 40601da177e4SLinus Torvalds error = -EBUSY; 40611bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 40621bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4063912dbc15SSage Weil goto out; 4064912dbc15SSage Weil 40651da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4066912dbc15SSage Weil if (error) 4067912dbc15SSage Weil goto out; 4068912dbc15SSage Weil 40691da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4070912dbc15SSage Weil if (error) 4071912dbc15SSage Weil goto out; 4072912dbc15SSage Weil 40738767712fSAl Viro shrink_dcache_parent(dentry); 40741da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4075d83c49f3SAl Viro dont_mount(dentry); 40768ed936b5SEric W. Biederman detach_mounts(dentry); 40771da177e4SLinus Torvalds 4078912dbc15SSage Weil out: 40795955102cSAl Viro inode_unlock(dentry->d_inode); 40801d2ef590SAl Viro dput(dentry); 4081912dbc15SSage Weil if (!error) 4082a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 40831da177e4SLinus Torvalds return error; 40841da177e4SLinus Torvalds } 40854d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 40861da177e4SLinus Torvalds 408745f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 40881da177e4SLinus Torvalds { 40896521f891SChristian Brauner struct user_namespace *mnt_userns; 40900ee50b47SDmitry Kadashev int error; 40911da177e4SLinus Torvalds struct dentry *dentry; 4092f5beed75SAl Viro struct path path; 4093f5beed75SAl Viro struct qstr last; 4094f5beed75SAl Viro int type; 4095c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4096c6ee9206SJeff Layton retry: 4097c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 40980ee50b47SDmitry Kadashev if (error) 40990ee50b47SDmitry Kadashev goto exit1; 41001da177e4SLinus Torvalds 4101f5beed75SAl Viro switch (type) { 41021da177e4SLinus Torvalds case LAST_DOTDOT: 41031da177e4SLinus Torvalds error = -ENOTEMPTY; 41040ee50b47SDmitry Kadashev goto exit2; 41051da177e4SLinus Torvalds case LAST_DOT: 41061da177e4SLinus Torvalds error = -EINVAL; 41070ee50b47SDmitry Kadashev goto exit2; 41081da177e4SLinus Torvalds case LAST_ROOT: 41091da177e4SLinus Torvalds error = -EBUSY; 41100ee50b47SDmitry Kadashev goto exit2; 41111da177e4SLinus Torvalds } 41120612d9fbSOGAWA Hirofumi 4113f5beed75SAl Viro error = mnt_want_write(path.mnt); 4114c30dabfeSJan Kara if (error) 41150ee50b47SDmitry Kadashev goto exit2; 41160612d9fbSOGAWA Hirofumi 41175955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4118f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 41191da177e4SLinus Torvalds error = PTR_ERR(dentry); 41206902d925SDave Hansen if (IS_ERR(dentry)) 41210ee50b47SDmitry Kadashev goto exit3; 4122e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4123e6bc45d6STheodore Ts'o error = -ENOENT; 41240ee50b47SDmitry Kadashev goto exit4; 4125e6bc45d6STheodore Ts'o } 4126f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4127be6d3e56SKentaro Takeda if (error) 41280ee50b47SDmitry Kadashev goto exit4; 41296521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 41306521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 41310ee50b47SDmitry Kadashev exit4: 41321da177e4SLinus Torvalds dput(dentry); 41330ee50b47SDmitry Kadashev exit3: 41345955102cSAl Viro inode_unlock(path.dentry->d_inode); 4135f5beed75SAl Viro mnt_drop_write(path.mnt); 41360ee50b47SDmitry Kadashev exit2: 4137f5beed75SAl Viro path_put(&path); 4138c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4139c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4140c6ee9206SJeff Layton goto retry; 4141c6ee9206SJeff Layton } 41420ee50b47SDmitry Kadashev exit1: 414324fb33d4SAl Viro putname(name); 41441da177e4SLinus Torvalds return error; 41451da177e4SLinus Torvalds } 41461da177e4SLinus Torvalds 41473cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 41485590ff0dSUlrich Drepper { 4149e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 41505590ff0dSUlrich Drepper } 41515590ff0dSUlrich Drepper 4152b21996e3SJ. Bruce Fields /** 4153b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 41546521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 4155b21996e3SJ. Bruce Fields * @dir: parent directory 4156b21996e3SJ. Bruce Fields * @dentry: victim 4157b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4158b21996e3SJ. Bruce Fields * 4159b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4160b21996e3SJ. Bruce Fields * 4161b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4162b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4163b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4164b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4165b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4166b21996e3SJ. Bruce Fields * 4167b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4168b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4169b21996e3SJ. Bruce Fields * to be NFS exported. 41706521f891SChristian Brauner * 41716521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 41726521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 41736521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 41746521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 41756521f891SChristian Brauner * raw inode simply passs init_user_ns. 4176b21996e3SJ. Bruce Fields */ 41776521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 41786521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 41791da177e4SLinus Torvalds { 41809accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 41816521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 41821da177e4SLinus Torvalds 41831da177e4SLinus Torvalds if (error) 41841da177e4SLinus Torvalds return error; 41851da177e4SLinus Torvalds 4186acfa4380SAl Viro if (!dir->i_op->unlink) 41871da177e4SLinus Torvalds return -EPERM; 41881da177e4SLinus Torvalds 41895955102cSAl Viro inode_lock(target); 419051cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 419151cc3a66SHugh Dickins error = -EPERM; 419251cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 41931da177e4SLinus Torvalds error = -EBUSY; 41941da177e4SLinus Torvalds else { 41951da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4196bec1052eSAl Viro if (!error) { 41975a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 41985a14696cSJ. Bruce Fields if (error) 4199b21996e3SJ. Bruce Fields goto out; 42001da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 42018ed936b5SEric W. Biederman if (!error) { 4202d83c49f3SAl Viro dont_mount(dentry); 42038ed936b5SEric W. Biederman detach_mounts(dentry); 42048ed936b5SEric W. Biederman } 4205bec1052eSAl Viro } 42061da177e4SLinus Torvalds } 4207b21996e3SJ. Bruce Fields out: 42085955102cSAl Viro inode_unlock(target); 42091da177e4SLinus Torvalds 42101da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4211a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4212a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4213a37d9a17SAmir Goldstein } else if (!error) { 42149accbb97SJ. Bruce Fields fsnotify_link_count(target); 4215a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 42161da177e4SLinus Torvalds } 42170eeca283SRobert Love 42181da177e4SLinus Torvalds return error; 42191da177e4SLinus Torvalds } 42204d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 42211da177e4SLinus Torvalds 42221da177e4SLinus Torvalds /* 42231da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 42241b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 42251da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 42261da177e4SLinus Torvalds * while waiting on the I/O. 42271da177e4SLinus Torvalds */ 422845f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 42291da177e4SLinus Torvalds { 42302ad94ae6SAl Viro int error; 42311da177e4SLinus Torvalds struct dentry *dentry; 4232f5beed75SAl Viro struct path path; 4233f5beed75SAl Viro struct qstr last; 4234f5beed75SAl Viro int type; 42351da177e4SLinus Torvalds struct inode *inode = NULL; 4236b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 42375d18f813SJeff Layton unsigned int lookup_flags = 0; 42385d18f813SJeff Layton retry: 4239c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 42400ee50b47SDmitry Kadashev if (error) 42410ee50b47SDmitry Kadashev goto exit1; 42422ad94ae6SAl Viro 42431da177e4SLinus Torvalds error = -EISDIR; 4244f5beed75SAl Viro if (type != LAST_NORM) 42450ee50b47SDmitry Kadashev goto exit2; 42460612d9fbSOGAWA Hirofumi 4247f5beed75SAl Viro error = mnt_want_write(path.mnt); 4248c30dabfeSJan Kara if (error) 42490ee50b47SDmitry Kadashev goto exit2; 4250b21996e3SJ. Bruce Fields retry_deleg: 42515955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4252f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 42531da177e4SLinus Torvalds error = PTR_ERR(dentry); 42541da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 42556521f891SChristian Brauner struct user_namespace *mnt_userns; 42566521f891SChristian Brauner 42571da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4258f5beed75SAl Viro if (last.name[last.len]) 425950338b88STörök Edwin goto slashes; 42601da177e4SLinus Torvalds inode = dentry->d_inode; 4261b18825a7SDavid Howells if (d_is_negative(dentry)) 4262e6bc45d6STheodore Ts'o goto slashes; 42637de9c6eeSAl Viro ihold(inode); 4264f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4265be6d3e56SKentaro Takeda if (error) 42660ee50b47SDmitry Kadashev goto exit3; 42676521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4268549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4269549c7297SChristian Brauner &delegated_inode); 42700ee50b47SDmitry Kadashev exit3: 42711da177e4SLinus Torvalds dput(dentry); 42721da177e4SLinus Torvalds } 42735955102cSAl Viro inode_unlock(path.dentry->d_inode); 42741da177e4SLinus Torvalds if (inode) 42751da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4276b21996e3SJ. Bruce Fields inode = NULL; 4277b21996e3SJ. Bruce Fields if (delegated_inode) { 42785a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4279b21996e3SJ. Bruce Fields if (!error) 4280b21996e3SJ. Bruce Fields goto retry_deleg; 4281b21996e3SJ. Bruce Fields } 4282f5beed75SAl Viro mnt_drop_write(path.mnt); 42830ee50b47SDmitry Kadashev exit2: 4284f5beed75SAl Viro path_put(&path); 42855d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 42865d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 42875d18f813SJeff Layton inode = NULL; 42885d18f813SJeff Layton goto retry; 42895d18f813SJeff Layton } 42900ee50b47SDmitry Kadashev exit1: 4291da2f1362SChristoph Hellwig putname(name); 42921da177e4SLinus Torvalds return error; 42931da177e4SLinus Torvalds 42941da177e4SLinus Torvalds slashes: 4295b18825a7SDavid Howells if (d_is_negative(dentry)) 4296b18825a7SDavid Howells error = -ENOENT; 429744b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4298b18825a7SDavid Howells error = -EISDIR; 4299b18825a7SDavid Howells else 4300b18825a7SDavid Howells error = -ENOTDIR; 43010ee50b47SDmitry Kadashev goto exit3; 43021da177e4SLinus Torvalds } 43031da177e4SLinus Torvalds 43042e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 43055590ff0dSUlrich Drepper { 43065590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 43075590ff0dSUlrich Drepper return -EINVAL; 43085590ff0dSUlrich Drepper 43095590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4310e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4311da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 43125590ff0dSUlrich Drepper } 43135590ff0dSUlrich Drepper 43143480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 43155590ff0dSUlrich Drepper { 4316da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 43175590ff0dSUlrich Drepper } 43185590ff0dSUlrich Drepper 43196521f891SChristian Brauner /** 43206521f891SChristian Brauner * vfs_symlink - create symlink 43216521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 43226521f891SChristian Brauner * @dir: inode of @dentry 43236521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 43246521f891SChristian Brauner * @oldname: name of the file to link to 43256521f891SChristian Brauner * 43266521f891SChristian Brauner * Create a symlink. 43276521f891SChristian Brauner * 43286521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 43296521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 43306521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 43316521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 43326521f891SChristian Brauner * raw inode simply passs init_user_ns. 43336521f891SChristian Brauner */ 43346521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 43356521f891SChristian Brauner struct dentry *dentry, const char *oldname) 43361da177e4SLinus Torvalds { 43376521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 43381da177e4SLinus Torvalds 43391da177e4SLinus Torvalds if (error) 43401da177e4SLinus Torvalds return error; 43411da177e4SLinus Torvalds 4342acfa4380SAl Viro if (!dir->i_op->symlink) 43431da177e4SLinus Torvalds return -EPERM; 43441da177e4SLinus Torvalds 43451da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 43461da177e4SLinus Torvalds if (error) 43471da177e4SLinus Torvalds return error; 43481da177e4SLinus Torvalds 4349549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4350a74574aaSStephen Smalley if (!error) 4351f38aa942SAmy Griffis fsnotify_create(dir, dentry); 43521da177e4SLinus Torvalds return error; 43531da177e4SLinus Torvalds } 43544d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 43551da177e4SLinus Torvalds 43567a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 43571da177e4SLinus Torvalds { 43582ad94ae6SAl Viro int error; 43596902d925SDave Hansen struct dentry *dentry; 4360dae6ad8fSAl Viro struct path path; 4361f46d3567SJeff Layton unsigned int lookup_flags = 0; 43621da177e4SLinus Torvalds 4363da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4364da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4365da2d0cedSDmitry Kadashev goto out_putnames; 4366da2d0cedSDmitry Kadashev } 4367f46d3567SJeff Layton retry: 4368b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 43691da177e4SLinus Torvalds error = PTR_ERR(dentry); 43706902d925SDave Hansen if (IS_ERR(dentry)) 4371da2d0cedSDmitry Kadashev goto out_putnames; 43726902d925SDave Hansen 437391a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 43746521f891SChristian Brauner if (!error) { 43756521f891SChristian Brauner struct user_namespace *mnt_userns; 43766521f891SChristian Brauner 43776521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 43786521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 43796521f891SChristian Brauner from->name); 43806521f891SChristian Brauner } 4381921a1650SAl Viro done_path_create(&path, dentry); 4382f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4383f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4384f46d3567SJeff Layton goto retry; 4385f46d3567SJeff Layton } 4386da2d0cedSDmitry Kadashev out_putnames: 4387da2d0cedSDmitry Kadashev putname(to); 43881da177e4SLinus Torvalds putname(from); 43891da177e4SLinus Torvalds return error; 43901da177e4SLinus Torvalds } 43911da177e4SLinus Torvalds 4392b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4393b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4394b724e846SDominik Brodowski { 4395da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4396b724e846SDominik Brodowski } 4397b724e846SDominik Brodowski 43983480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 43995590ff0dSUlrich Drepper { 4400da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 44015590ff0dSUlrich Drepper } 44025590ff0dSUlrich Drepper 4403146a8595SJ. Bruce Fields /** 4404146a8595SJ. Bruce Fields * vfs_link - create a new link 4405146a8595SJ. Bruce Fields * @old_dentry: object to be linked 44066521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4407146a8595SJ. Bruce Fields * @dir: new parent 4408146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4409146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4410146a8595SJ. Bruce Fields * 4411146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4412146a8595SJ. Bruce Fields * 4413146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4414146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4415146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4416146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4417146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4418146a8595SJ. Bruce Fields * 4419146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4420146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4421146a8595SJ. Bruce Fields * to be NFS exported. 44226521f891SChristian Brauner * 44236521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 44246521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 44256521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 44266521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 44276521f891SChristian Brauner * raw inode simply passs init_user_ns. 4428146a8595SJ. Bruce Fields */ 44296521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 44306521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 44316521f891SChristian Brauner struct inode **delegated_inode) 44321da177e4SLinus Torvalds { 44331da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 44348de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 44351da177e4SLinus Torvalds int error; 44361da177e4SLinus Torvalds 44371da177e4SLinus Torvalds if (!inode) 44381da177e4SLinus Torvalds return -ENOENT; 44391da177e4SLinus Torvalds 44406521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 44411da177e4SLinus Torvalds if (error) 44421da177e4SLinus Torvalds return error; 44431da177e4SLinus Torvalds 44441da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 44451da177e4SLinus Torvalds return -EXDEV; 44461da177e4SLinus Torvalds 44471da177e4SLinus Torvalds /* 44481da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 44491da177e4SLinus Torvalds */ 44501da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 44511da177e4SLinus Torvalds return -EPERM; 44520bd23d09SEric W. Biederman /* 44530bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 44540bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 44550bd23d09SEric W. Biederman * the vfs. 44560bd23d09SEric W. Biederman */ 44576521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 44580bd23d09SEric W. Biederman return -EPERM; 4459acfa4380SAl Viro if (!dir->i_op->link) 44601da177e4SLinus Torvalds return -EPERM; 44617e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 44621da177e4SLinus Torvalds return -EPERM; 44631da177e4SLinus Torvalds 44641da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 44651da177e4SLinus Torvalds if (error) 44661da177e4SLinus Torvalds return error; 44671da177e4SLinus Torvalds 44685955102cSAl Viro inode_lock(inode); 4469aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4470f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4471aae8a97dSAneesh Kumar K.V error = -ENOENT; 44728de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 44738de52778SAl Viro error = -EMLINK; 4474146a8595SJ. Bruce Fields else { 4475146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4476146a8595SJ. Bruce Fields if (!error) 44771da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4478146a8595SJ. Bruce Fields } 4479f4e0c30cSAl Viro 4480f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4481f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4482f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4483f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4484f4e0c30cSAl Viro } 44855955102cSAl Viro inode_unlock(inode); 4486e31e14ecSStephen Smalley if (!error) 44877e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 44881da177e4SLinus Torvalds return error; 44891da177e4SLinus Torvalds } 44904d359507SAl Viro EXPORT_SYMBOL(vfs_link); 44911da177e4SLinus Torvalds 44921da177e4SLinus Torvalds /* 44931da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 44941da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 44951da177e4SLinus Torvalds * newname. --KAB 44961da177e4SLinus Torvalds * 44971da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 44981da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 44991da177e4SLinus Torvalds * and other special files. --ADM 45001da177e4SLinus Torvalds */ 4501cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4502020250f3SDmitry Kadashev struct filename *new, int flags) 45031da177e4SLinus Torvalds { 45046521f891SChristian Brauner struct user_namespace *mnt_userns; 45051da177e4SLinus Torvalds struct dentry *new_dentry; 4506dae6ad8fSAl Viro struct path old_path, new_path; 4507146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 450811a7b371SAneesh Kumar K.V int how = 0; 45091da177e4SLinus Torvalds int error; 45101da177e4SLinus Torvalds 4511020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4512020250f3SDmitry Kadashev error = -EINVAL; 4513020250f3SDmitry Kadashev goto out_putnames; 4514020250f3SDmitry Kadashev } 451511a7b371SAneesh Kumar K.V /* 4516f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4517f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4518f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 451911a7b371SAneesh Kumar K.V */ 4520020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4521020250f3SDmitry Kadashev error = -ENOENT; 4522020250f3SDmitry Kadashev goto out_putnames; 4523f0cc6ffbSLinus Torvalds } 4524c04030e1SUlrich Drepper 452511a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 452611a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4527442e31caSJeff Layton retry: 4528794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 45291da177e4SLinus Torvalds if (error) 4530020250f3SDmitry Kadashev goto out_putnames; 45312ad94ae6SAl Viro 4532b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4533442e31caSJeff Layton (how & LOOKUP_REVAL)); 45341da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45356902d925SDave Hansen if (IS_ERR(new_dentry)) 4536020250f3SDmitry Kadashev goto out_putpath; 4537dae6ad8fSAl Viro 4538dae6ad8fSAl Viro error = -EXDEV; 4539dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4540dae6ad8fSAl Viro goto out_dput; 4541549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4542549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4543800179c9SKees Cook if (unlikely(error)) 4544800179c9SKees Cook goto out_dput; 4545dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4546be6d3e56SKentaro Takeda if (error) 4547a8104a9fSAl Viro goto out_dput; 45486521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 45496521f891SChristian Brauner new_dentry, &delegated_inode); 455075c3f29dSDave Hansen out_dput: 4551921a1650SAl Viro done_path_create(&new_path, new_dentry); 4552146a8595SJ. Bruce Fields if (delegated_inode) { 4553146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4554d22e6338SOleg Drokin if (!error) { 4555d22e6338SOleg Drokin path_put(&old_path); 4556146a8595SJ. Bruce Fields goto retry; 4557146a8595SJ. Bruce Fields } 4558d22e6338SOleg Drokin } 4559442e31caSJeff Layton if (retry_estale(error, how)) { 4560d22e6338SOleg Drokin path_put(&old_path); 4561442e31caSJeff Layton how |= LOOKUP_REVAL; 4562442e31caSJeff Layton goto retry; 4563442e31caSJeff Layton } 4564020250f3SDmitry Kadashev out_putpath: 45652d8f3038SAl Viro path_put(&old_path); 4566020250f3SDmitry Kadashev out_putnames: 4567020250f3SDmitry Kadashev putname(old); 4568020250f3SDmitry Kadashev putname(new); 45691da177e4SLinus Torvalds 45701da177e4SLinus Torvalds return error; 45711da177e4SLinus Torvalds } 45721da177e4SLinus Torvalds 457346ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 457446ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 457546ea89ebSDominik Brodowski { 4576020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4577020250f3SDmitry Kadashev newdfd, getname(newname), flags); 457846ea89ebSDominik Brodowski } 457946ea89ebSDominik Brodowski 45803480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 45815590ff0dSUlrich Drepper { 4582020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 45835590ff0dSUlrich Drepper } 45845590ff0dSUlrich Drepper 4585bc27027aSMiklos Szeredi /** 4586bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 45872111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4588bc27027aSMiklos Szeredi * 4589bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4590bc27027aSMiklos Szeredi * 4591bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4592bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4593bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4594bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4595bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4596bc27027aSMiklos Szeredi * so. 4597bc27027aSMiklos Szeredi * 4598bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4599bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4600bc27027aSMiklos Szeredi * to be NFS exported. 4601bc27027aSMiklos Szeredi * 46021da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 46031da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 46041da177e4SLinus Torvalds * Problems: 46050117d427SMauro Carvalho Chehab * 4606d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 46071da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 46081da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4609a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 46101da177e4SLinus Torvalds * story. 46116cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 46126cedba89SJ. Bruce Fields * and source (if it is not a directory). 46131b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 46141da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 46151da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4616a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 46171da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 46181da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 46191da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4620a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 46211da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 46221da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 46231da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4624e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 46251b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 46261da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4627c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 46281da177e4SLinus Torvalds * locking]. 46291da177e4SLinus Torvalds */ 46309fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 46311da177e4SLinus Torvalds { 46321da177e4SLinus Torvalds int error; 46339fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 46349fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 46359fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 46369fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 46379fe61450SChristian Brauner unsigned int flags = rd->flags; 4638bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4639bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4640bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4641da1ce067SMiklos Szeredi bool new_is_dir = false; 4642da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 464349d31c2fSAl Viro struct name_snapshot old_name; 46441da177e4SLinus Torvalds 46458d3e2936SMiklos Szeredi if (source == target) 46461da177e4SLinus Torvalds return 0; 46471da177e4SLinus Torvalds 46486521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 46491da177e4SLinus Torvalds if (error) 46501da177e4SLinus Torvalds return error; 46511da177e4SLinus Torvalds 4652da1ce067SMiklos Szeredi if (!target) { 46536521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4654da1ce067SMiklos Szeredi } else { 4655da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4656da1ce067SMiklos Szeredi 4657da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46586521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 46596521f891SChristian Brauner new_dentry, is_dir); 4660da1ce067SMiklos Szeredi else 46616521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 46626521f891SChristian Brauner new_dentry, new_is_dir); 4663da1ce067SMiklos Szeredi } 46641da177e4SLinus Torvalds if (error) 46651da177e4SLinus Torvalds return error; 46661da177e4SLinus Torvalds 46672773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 46681da177e4SLinus Torvalds return -EPERM; 46691da177e4SLinus Torvalds 4670bc27027aSMiklos Szeredi /* 4671bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4672bc27027aSMiklos Szeredi * we'll need to flip '..'. 4673bc27027aSMiklos Szeredi */ 4674da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4675da1ce067SMiklos Szeredi if (is_dir) { 46766521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 467747291baaSChristian Brauner MAY_WRITE); 4678bc27027aSMiklos Szeredi if (error) 4679bc27027aSMiklos Szeredi return error; 4680bc27027aSMiklos Szeredi } 4681da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 46826521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 468347291baaSChristian Brauner MAY_WRITE); 4684da1ce067SMiklos Szeredi if (error) 4685da1ce067SMiklos Szeredi return error; 4686da1ce067SMiklos Szeredi } 4687da1ce067SMiklos Szeredi } 46880eeca283SRobert Love 46890b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 46900b3974ebSMiklos Szeredi flags); 4691bc27027aSMiklos Szeredi if (error) 4692bc27027aSMiklos Szeredi return error; 4693bc27027aSMiklos Szeredi 469449d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4695bc27027aSMiklos Szeredi dget(new_dentry); 4696da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4697bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4698bc27027aSMiklos Szeredi else if (target) 46995955102cSAl Viro inode_lock(target); 4700bc27027aSMiklos Szeredi 470151cc3a66SHugh Dickins error = -EPERM; 470251cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 470351cc3a66SHugh Dickins goto out; 470451cc3a66SHugh Dickins 4705bc27027aSMiklos Szeredi error = -EBUSY; 47067af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4707bc27027aSMiklos Szeredi goto out; 4708bc27027aSMiklos Szeredi 4709da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4710bc27027aSMiklos Szeredi error = -EMLINK; 4711da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4712bc27027aSMiklos Szeredi goto out; 4713da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4714da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4715da1ce067SMiklos Szeredi goto out; 4716da1ce067SMiklos Szeredi } 4717da1ce067SMiklos Szeredi if (!is_dir) { 4718bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4719bc27027aSMiklos Szeredi if (error) 4720bc27027aSMiklos Szeredi goto out; 4721da1ce067SMiklos Szeredi } 4722da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4723bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4724bc27027aSMiklos Szeredi if (error) 4725bc27027aSMiklos Szeredi goto out; 4726bc27027aSMiklos Szeredi } 4727549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4728520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4729bc27027aSMiklos Szeredi if (error) 4730bc27027aSMiklos Szeredi goto out; 4731bc27027aSMiklos Szeredi 4732da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 47338767712fSAl Viro if (is_dir) { 47348767712fSAl Viro shrink_dcache_parent(new_dentry); 4735bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 47368767712fSAl Viro } 4737bc27027aSMiklos Szeredi dont_mount(new_dentry); 47388ed936b5SEric W. Biederman detach_mounts(new_dentry); 4739bc27027aSMiklos Szeredi } 4740da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4741da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4742bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4743da1ce067SMiklos Szeredi else 4744da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4745da1ce067SMiklos Szeredi } 4746bc27027aSMiklos Szeredi out: 4747da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4748bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4749bc27027aSMiklos Szeredi else if (target) 47505955102cSAl Viro inode_unlock(target); 4751bc27027aSMiklos Szeredi dput(new_dentry); 4752da1ce067SMiklos Szeredi if (!error) { 4753f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4754da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4755da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4756f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4757da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4758da1ce067SMiklos Szeredi } 4759da1ce067SMiklos Szeredi } 476049d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 47610eeca283SRobert Love 47621da177e4SLinus Torvalds return error; 47631da177e4SLinus Torvalds } 47644d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 47651da177e4SLinus Torvalds 4766e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4767e886663cSJens Axboe struct filename *to, unsigned int flags) 47681da177e4SLinus Torvalds { 47699fe61450SChristian Brauner struct renamedata rd; 47701da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 47711da177e4SLinus Torvalds struct dentry *trap; 4772f5beed75SAl Viro struct path old_path, new_path; 4773f5beed75SAl Viro struct qstr old_last, new_last; 4774f5beed75SAl Viro int old_type, new_type; 47758e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4776f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4777c6a94284SJeff Layton bool should_retry = false; 4778e886663cSJens Axboe int error = -EINVAL; 4779520c8b16SMiklos Szeredi 47800d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 47810ee50b47SDmitry Kadashev goto put_names; 4782da1ce067SMiklos Szeredi 47830d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 47840d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 47850ee50b47SDmitry Kadashev goto put_names; 4786520c8b16SMiklos Szeredi 4787f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4788f5beed75SAl Viro target_flags = 0; 4789f5beed75SAl Viro 4790c6a94284SJeff Layton retry: 4791c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4792e886663cSJens Axboe &old_last, &old_type); 47930ee50b47SDmitry Kadashev if (error) 47940ee50b47SDmitry Kadashev goto put_names; 47951da177e4SLinus Torvalds 4796c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4797e886663cSJens Axboe &new_type); 47980ee50b47SDmitry Kadashev if (error) 47991da177e4SLinus Torvalds goto exit1; 48001da177e4SLinus Torvalds 48011da177e4SLinus Torvalds error = -EXDEV; 4802f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 48031da177e4SLinus Torvalds goto exit2; 48041da177e4SLinus Torvalds 48051da177e4SLinus Torvalds error = -EBUSY; 4806f5beed75SAl Viro if (old_type != LAST_NORM) 48071da177e4SLinus Torvalds goto exit2; 48081da177e4SLinus Torvalds 48090a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 48100a7c3937SMiklos Szeredi error = -EEXIST; 4811f5beed75SAl Viro if (new_type != LAST_NORM) 48121da177e4SLinus Torvalds goto exit2; 48131da177e4SLinus Torvalds 4814f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4815c30dabfeSJan Kara if (error) 4816c30dabfeSJan Kara goto exit2; 4817c30dabfeSJan Kara 48188e6d782cSJ. Bruce Fields retry_deleg: 4819f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 48201da177e4SLinus Torvalds 4821f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 48221da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 48231da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 48241da177e4SLinus Torvalds goto exit3; 48251da177e4SLinus Torvalds /* source must exist */ 48261da177e4SLinus Torvalds error = -ENOENT; 4827b18825a7SDavid Howells if (d_is_negative(old_dentry)) 48281da177e4SLinus Torvalds goto exit4; 4829f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 48301da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 48311da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 48321da177e4SLinus Torvalds goto exit4; 48330a7c3937SMiklos Szeredi error = -EEXIST; 48340a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 48350a7c3937SMiklos Szeredi goto exit5; 4836da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4837da1ce067SMiklos Szeredi error = -ENOENT; 4838da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4839da1ce067SMiklos Szeredi goto exit5; 4840da1ce067SMiklos Szeredi 4841da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4842da1ce067SMiklos Szeredi error = -ENOTDIR; 4843f5beed75SAl Viro if (new_last.name[new_last.len]) 4844da1ce067SMiklos Szeredi goto exit5; 4845da1ce067SMiklos Szeredi } 4846da1ce067SMiklos Szeredi } 48470a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 48480a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 48490a7c3937SMiklos Szeredi error = -ENOTDIR; 4850f5beed75SAl Viro if (old_last.name[old_last.len]) 48510a7c3937SMiklos Szeredi goto exit5; 4852f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 48530a7c3937SMiklos Szeredi goto exit5; 48540a7c3937SMiklos Szeredi } 48550a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 48560a7c3937SMiklos Szeredi error = -EINVAL; 48570a7c3937SMiklos Szeredi if (old_dentry == trap) 48580a7c3937SMiklos Szeredi goto exit5; 48591da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4860da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 48611da177e4SLinus Torvalds error = -ENOTEMPTY; 48621da177e4SLinus Torvalds if (new_dentry == trap) 48631da177e4SLinus Torvalds goto exit5; 48641da177e4SLinus Torvalds 4865f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4866f5beed75SAl Viro &new_path, new_dentry, flags); 4867be6d3e56SKentaro Takeda if (error) 4868c30dabfeSJan Kara goto exit5; 48699fe61450SChristian Brauner 48709fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 48719fe61450SChristian Brauner rd.old_dentry = old_dentry; 48726521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 48739fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 48749fe61450SChristian Brauner rd.new_dentry = new_dentry; 48756521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 48769fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 48779fe61450SChristian Brauner rd.flags = flags; 48789fe61450SChristian Brauner error = vfs_rename(&rd); 48791da177e4SLinus Torvalds exit5: 48801da177e4SLinus Torvalds dput(new_dentry); 48811da177e4SLinus Torvalds exit4: 48821da177e4SLinus Torvalds dput(old_dentry); 48831da177e4SLinus Torvalds exit3: 4884f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 48858e6d782cSJ. Bruce Fields if (delegated_inode) { 48868e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 48878e6d782cSJ. Bruce Fields if (!error) 48888e6d782cSJ. Bruce Fields goto retry_deleg; 48898e6d782cSJ. Bruce Fields } 4890f5beed75SAl Viro mnt_drop_write(old_path.mnt); 48911da177e4SLinus Torvalds exit2: 4892c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4893c6a94284SJeff Layton should_retry = true; 4894f5beed75SAl Viro path_put(&new_path); 48951da177e4SLinus Torvalds exit1: 4896f5beed75SAl Viro path_put(&old_path); 4897c6a94284SJeff Layton if (should_retry) { 4898c6a94284SJeff Layton should_retry = false; 4899c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4900c6a94284SJeff Layton goto retry; 4901c6a94284SJeff Layton } 49020ee50b47SDmitry Kadashev put_names: 4903e886663cSJens Axboe putname(from); 4904e886663cSJens Axboe putname(to); 49051da177e4SLinus Torvalds return error; 49061da177e4SLinus Torvalds } 49071da177e4SLinus Torvalds 4908ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4909ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4910ee81feb6SDominik Brodowski { 4911e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4912e886663cSJens Axboe flags); 4913ee81feb6SDominik Brodowski } 4914ee81feb6SDominik Brodowski 4915520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4916520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4917520c8b16SMiklos Szeredi { 4918e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4919e886663cSJens Axboe 0); 4920520c8b16SMiklos Szeredi } 4921520c8b16SMiklos Szeredi 4922a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 49235590ff0dSUlrich Drepper { 4924e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4925e886663cSJens Axboe getname(newname), 0); 49265590ff0dSUlrich Drepper } 49275590ff0dSUlrich Drepper 49285d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 49291da177e4SLinus Torvalds { 49305d826c84SAl Viro int len = PTR_ERR(link); 49311da177e4SLinus Torvalds if (IS_ERR(link)) 49321da177e4SLinus Torvalds goto out; 49331da177e4SLinus Torvalds 49341da177e4SLinus Torvalds len = strlen(link); 49351da177e4SLinus Torvalds if (len > (unsigned) buflen) 49361da177e4SLinus Torvalds len = buflen; 49371da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 49381da177e4SLinus Torvalds len = -EFAULT; 49391da177e4SLinus Torvalds out: 49401da177e4SLinus Torvalds return len; 49411da177e4SLinus Torvalds } 49421da177e4SLinus Torvalds 4943d60874cdSMiklos Szeredi /** 4944fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4945fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4946fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4947fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4948fd4a0edfSMiklos Szeredi * 4949fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4950fd4a0edfSMiklos Szeredi * 4951fd4a0edfSMiklos Szeredi * Does not call security hook. 4952fd4a0edfSMiklos Szeredi */ 4953fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4954fd4a0edfSMiklos Szeredi { 4955fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4956f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4957f2df5da6SAl Viro const char *link; 4958f2df5da6SAl Viro int res; 4959fd4a0edfSMiklos Szeredi 496076fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 496176fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 496276fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 496376fca90eSMiklos Szeredi 496476fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4965fd4a0edfSMiklos Szeredi return -EINVAL; 4966fd4a0edfSMiklos Szeredi 496776fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 496876fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 496976fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 497076fca90eSMiklos Szeredi } 497176fca90eSMiklos Szeredi 49724c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4973f2df5da6SAl Viro if (!link) { 4974f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4975f2df5da6SAl Viro if (IS_ERR(link)) 4976f2df5da6SAl Viro return PTR_ERR(link); 4977f2df5da6SAl Viro } 4978f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4979f2df5da6SAl Viro do_delayed_call(&done); 4980f2df5da6SAl Viro return res; 4981fd4a0edfSMiklos Szeredi } 4982fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 49831da177e4SLinus Torvalds 4984d60874cdSMiklos Szeredi /** 4985d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4986d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4987d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4988d60874cdSMiklos Szeredi * 4989d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4990d60874cdSMiklos Szeredi * 4991d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4992d60874cdSMiklos Szeredi * 4993d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4994d60874cdSMiklos Szeredi */ 4995d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4996d60874cdSMiklos Szeredi { 4997d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4998d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4999d60874cdSMiklos Szeredi 5000d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 5001d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 5002d60874cdSMiklos Szeredi if (!res) 5003d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 5004d60874cdSMiklos Szeredi } 5005d60874cdSMiklos Szeredi return res; 5006d60874cdSMiklos Szeredi } 5007d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 5008d60874cdSMiklos Szeredi 50091da177e4SLinus Torvalds /* get the link contents into pagecache */ 50106b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 5011fceef393SAl Viro struct delayed_call *callback) 50121da177e4SLinus Torvalds { 5013ebd09abbSDuane Griffin char *kaddr; 50141da177e4SLinus Torvalds struct page *page; 50156b255391SAl Viro struct address_space *mapping = inode->i_mapping; 50166b255391SAl Viro 5017d3883d4fSAl Viro if (!dentry) { 5018d3883d4fSAl Viro page = find_get_page(mapping, 0); 5019d3883d4fSAl Viro if (!page) 50206b255391SAl Viro return ERR_PTR(-ECHILD); 5021d3883d4fSAl Viro if (!PageUptodate(page)) { 5022d3883d4fSAl Viro put_page(page); 5023d3883d4fSAl Viro return ERR_PTR(-ECHILD); 5024d3883d4fSAl Viro } 5025d3883d4fSAl Viro } else { 5026090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 50271da177e4SLinus Torvalds if (IS_ERR(page)) 50286fe6900eSNick Piggin return (char*)page; 5029d3883d4fSAl Viro } 5030fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 503121fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 503221fc61c7SAl Viro kaddr = page_address(page); 50336b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 5034ebd09abbSDuane Griffin return kaddr; 50351da177e4SLinus Torvalds } 50361da177e4SLinus Torvalds 50376b255391SAl Viro EXPORT_SYMBOL(page_get_link); 50381da177e4SLinus Torvalds 5039fceef393SAl Viro void page_put_link(void *arg) 50401da177e4SLinus Torvalds { 5041fceef393SAl Viro put_page(arg); 50421da177e4SLinus Torvalds } 50434d359507SAl Viro EXPORT_SYMBOL(page_put_link); 50441da177e4SLinus Torvalds 5045aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5046aa80deabSAl Viro { 5047fceef393SAl Viro DEFINE_DELAYED_CALL(done); 50486b255391SAl Viro int res = readlink_copy(buffer, buflen, 50496b255391SAl Viro page_get_link(dentry, d_inode(dentry), 5050fceef393SAl Viro &done)); 5051fceef393SAl Viro do_delayed_call(&done); 5052aa80deabSAl Viro return res; 5053aa80deabSAl Viro } 5054aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 5055aa80deabSAl Viro 505656f5746cSMatthew Wilcox (Oracle) int page_symlink(struct inode *inode, const char *symname, int len) 50571da177e4SLinus Torvalds { 50581da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 505927a77913SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 506056f5746cSMatthew Wilcox (Oracle) bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS); 50610adb25d2SKirill Korotaev struct page *page; 5062afddba49SNick Piggin void *fsdata; 5063beb497abSDmitriy Monakhov int err; 50642d878178SMatthew Wilcox (Oracle) unsigned int flags; 50651da177e4SLinus Torvalds 50667e53cac4SNeilBrown retry: 50672d878178SMatthew Wilcox (Oracle) if (nofs) 50682d878178SMatthew Wilcox (Oracle) flags = memalloc_nofs_save(); 506927a77913SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata); 50702d878178SMatthew Wilcox (Oracle) if (nofs) 50712d878178SMatthew Wilcox (Oracle) memalloc_nofs_restore(flags); 50721da177e4SLinus Torvalds if (err) 5073afddba49SNick Piggin goto fail; 5074afddba49SNick Piggin 507521fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5076afddba49SNick Piggin 507727a77913SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, 0, len-1, len-1, 5078afddba49SNick Piggin page, fsdata); 50791da177e4SLinus Torvalds if (err < 0) 50801da177e4SLinus Torvalds goto fail; 5081afddba49SNick Piggin if (err < len-1) 5082afddba49SNick Piggin goto retry; 5083afddba49SNick Piggin 50841da177e4SLinus Torvalds mark_inode_dirty(inode); 50851da177e4SLinus Torvalds return 0; 50861da177e4SLinus Torvalds fail: 50871da177e4SLinus Torvalds return err; 50881da177e4SLinus Torvalds } 50894d359507SAl Viro EXPORT_SYMBOL(page_symlink); 50900adb25d2SKirill Korotaev 509192e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 50926b255391SAl Viro .get_link = page_get_link, 50931da177e4SLinus Torvalds }; 50941da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5095