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> 25*2d878178SMatthew Wilcox (Oracle) #include <linux/sched/mm.h> 260eeca283SRobert Love #include <linux/fsnotify.h> 271da177e4SLinus Torvalds #include <linux/personality.h> 281da177e4SLinus Torvalds #include <linux/security.h> 296146f0d5SMimi Zohar #include <linux/ima.h> 301da177e4SLinus Torvalds #include <linux/syscalls.h> 311da177e4SLinus Torvalds #include <linux/mount.h> 321da177e4SLinus Torvalds #include <linux/audit.h> 3316f7e0feSRandy Dunlap #include <linux/capability.h> 34834f2a4aSTrond Myklebust #include <linux/file.h> 355590ff0dSUlrich Drepper #include <linux/fcntl.h> 3608ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 375ad4e53bSAl Viro #include <linux/fs_struct.h> 38e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3999d263d4SLinus Torvalds #include <linux/hash.h> 402a18da7aSGeorge Spelvin #include <linux/bitops.h> 41aeaa4a79SEric W. Biederman #include <linux/init_task.h> 427c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 431da177e4SLinus Torvalds 44e81e3f4dSEric Paris #include "internal.h" 45c7105365SAl Viro #include "mount.h" 46e81e3f4dSEric Paris 471da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 481da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 491da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 501da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 511da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 521da177e4SLinus Torvalds * 531da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 541da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 551da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 561da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 571da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 581da177e4SLinus Torvalds * the special cases of the former code. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 611da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 621da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 631da177e4SLinus Torvalds * 641da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 651da177e4SLinus Torvalds * resolution to correspond with current state of the code. 661da177e4SLinus Torvalds * 671da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 681da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 691da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 701da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 711da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 721da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 731da177e4SLinus Torvalds */ 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 761da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 771da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 781da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 791da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 801da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8125985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 821da177e4SLinus Torvalds * 831da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 841da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 851da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 861da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 871da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 881da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 891da177e4SLinus Torvalds * and in the old Linux semantics. 901da177e4SLinus Torvalds */ 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 931da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 941da177e4SLinus Torvalds * 951da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 961da177e4SLinus Torvalds */ 971da177e4SLinus Torvalds 981da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 991da177e4SLinus Torvalds * inside the path - always follow. 1001da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1011da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1021da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1031da177e4SLinus Torvalds * otherwise - don't follow. 1041da177e4SLinus Torvalds * (applied in that order). 1051da177e4SLinus Torvalds * 1061da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1071da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1081da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1091da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1101da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1111da177e4SLinus Torvalds */ 1121da177e4SLinus Torvalds /* 1131da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 114a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1151da177e4SLinus Torvalds * any extra contention... 1161da177e4SLinus Torvalds */ 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1191da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1201da177e4SLinus Torvalds * kernel data space before using them.. 1211da177e4SLinus Torvalds * 1221da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1231da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1241da177e4SLinus Torvalds */ 1257950e385SJeff Layton 126fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12791a27b2aSJeff Layton 12851f39a1fSDavid Drysdale struct filename * 12991a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13091a27b2aSJeff Layton { 13194b5d262SAl Viro struct filename *result; 1327950e385SJeff Layton char *kname; 13394b5d262SAl Viro int len; 1341da177e4SLinus Torvalds 1357ac86265SJeff Layton result = audit_reusename(filename); 1367ac86265SJeff Layton if (result) 1377ac86265SJeff Layton return result; 1387ac86265SJeff Layton 1397950e385SJeff Layton result = __getname(); 1403f9f0aa6SLinus Torvalds if (unlikely(!result)) 1414043cde8SEric Paris return ERR_PTR(-ENOMEM); 1421da177e4SLinus Torvalds 1437950e385SJeff Layton /* 1447950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1457950e385SJeff Layton * allocation 1467950e385SJeff Layton */ 147fd2f7cb5SAl Viro kname = (char *)result->iname; 14891a27b2aSJeff Layton result->name = kname; 1497950e385SJeff Layton 15094b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15191a27b2aSJeff Layton if (unlikely(len < 0)) { 15294b5d262SAl Viro __putname(result); 15394b5d262SAl Viro return ERR_PTR(len); 15491a27b2aSJeff Layton } 1553f9f0aa6SLinus Torvalds 1567950e385SJeff Layton /* 1577950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1587950e385SJeff Layton * separate struct filename so we can dedicate the entire 1597950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1607950e385SJeff Layton * userland. 1617950e385SJeff Layton */ 16294b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 163fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1647950e385SJeff Layton kname = (char *)result; 1657950e385SJeff Layton 166fd2f7cb5SAl Viro /* 167fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 168fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 169fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 170fd2f7cb5SAl Viro */ 171fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17294b5d262SAl Viro if (unlikely(!result)) { 17394b5d262SAl Viro __putname(kname); 17494b5d262SAl Viro return ERR_PTR(-ENOMEM); 1757950e385SJeff Layton } 1767950e385SJeff Layton result->name = kname; 17794b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17894b5d262SAl Viro if (unlikely(len < 0)) { 17994b5d262SAl Viro __putname(kname); 18094b5d262SAl Viro kfree(result); 18194b5d262SAl Viro return ERR_PTR(len); 18294b5d262SAl Viro } 18394b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18494b5d262SAl Viro __putname(kname); 18594b5d262SAl Viro kfree(result); 18694b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18794b5d262SAl Viro } 1887950e385SJeff Layton } 1897950e385SJeff Layton 19094b5d262SAl Viro result->refcnt = 1; 1913f9f0aa6SLinus Torvalds /* The empty path is special. */ 1923f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1933f9f0aa6SLinus Torvalds if (empty) 1941fa1e7f6SAndy Whitcroft *empty = 1; 19594b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19694b5d262SAl Viro putname(result); 19794b5d262SAl Viro return ERR_PTR(-ENOENT); 1981da177e4SLinus Torvalds } 19994b5d262SAl Viro } 2007950e385SJeff Layton 2017950e385SJeff Layton result->uptr = filename; 202c4ad8f98SLinus Torvalds result->aname = NULL; 2031da177e4SLinus Torvalds audit_getname(result); 2041da177e4SLinus Torvalds return result; 2053f9f0aa6SLinus Torvalds } 2063f9f0aa6SLinus Torvalds 20791a27b2aSJeff Layton struct filename * 2088228e2c3SDmitry Kadashev getname_uflags(const char __user *filename, int uflags) 2098228e2c3SDmitry Kadashev { 2108228e2c3SDmitry Kadashev int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 2118228e2c3SDmitry Kadashev 2128228e2c3SDmitry Kadashev return getname_flags(filename, flags, NULL); 2138228e2c3SDmitry Kadashev } 2148228e2c3SDmitry Kadashev 2158228e2c3SDmitry Kadashev struct filename * 21691a27b2aSJeff Layton getname(const char __user * filename) 217f52e0c11SAl Viro { 218f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 219f52e0c11SAl Viro } 220f52e0c11SAl Viro 221c4ad8f98SLinus Torvalds struct filename * 222c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 223c4ad8f98SLinus Torvalds { 224c4ad8f98SLinus Torvalds struct filename *result; 22508518549SPaul Moore int len = strlen(filename) + 1; 226c4ad8f98SLinus Torvalds 227c4ad8f98SLinus Torvalds result = __getname(); 228c4ad8f98SLinus Torvalds if (unlikely(!result)) 229c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 230c4ad8f98SLinus Torvalds 23108518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 232fd2f7cb5SAl Viro result->name = (char *)result->iname; 23308518549SPaul Moore } else if (len <= PATH_MAX) { 23430ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 23508518549SPaul Moore struct filename *tmp; 23608518549SPaul Moore 23730ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 23808518549SPaul Moore if (unlikely(!tmp)) { 23908518549SPaul Moore __putname(result); 24008518549SPaul Moore return ERR_PTR(-ENOMEM); 24108518549SPaul Moore } 24208518549SPaul Moore tmp->name = (char *)result; 24308518549SPaul Moore result = tmp; 24408518549SPaul Moore } else { 24508518549SPaul Moore __putname(result); 24608518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 24708518549SPaul Moore } 24808518549SPaul Moore memcpy((char *)result->name, filename, len); 249c4ad8f98SLinus Torvalds result->uptr = NULL; 250c4ad8f98SLinus Torvalds result->aname = NULL; 25155422d0bSPaul Moore result->refcnt = 1; 252fd3522fdSPaul Moore audit_getname(result); 253c4ad8f98SLinus Torvalds 254c4ad8f98SLinus Torvalds return result; 255c4ad8f98SLinus Torvalds } 256c4ad8f98SLinus Torvalds 25791a27b2aSJeff Layton void putname(struct filename *name) 2581da177e4SLinus Torvalds { 259ea47ab11SAl Viro if (IS_ERR(name)) 26091ef658fSDmitry Kadashev return; 26191ef658fSDmitry Kadashev 26255422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 26355422d0bSPaul Moore 26455422d0bSPaul Moore if (--name->refcnt > 0) 26555422d0bSPaul Moore return; 26655422d0bSPaul Moore 267fd2f7cb5SAl Viro if (name->name != name->iname) { 26855422d0bSPaul Moore __putname(name->name); 26955422d0bSPaul Moore kfree(name); 27055422d0bSPaul Moore } else 27155422d0bSPaul Moore __putname(name); 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 27447291baaSChristian Brauner /** 27547291baaSChristian Brauner * check_acl - perform ACL permission checking 27647291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 27747291baaSChristian Brauner * @inode: inode to check permissions on 27847291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 27947291baaSChristian Brauner * 28047291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 28147291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 28247291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 28347291baaSChristian Brauner * 28447291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 28547291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 28647291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 28747291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 28847291baaSChristian Brauner * raw inode simply passs init_user_ns. 28947291baaSChristian Brauner */ 29047291baaSChristian Brauner static int check_acl(struct user_namespace *mnt_userns, 29147291baaSChristian Brauner struct inode *inode, int mask) 292e77819e5SLinus Torvalds { 29384635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 294e77819e5SLinus Torvalds struct posix_acl *acl; 295e77819e5SLinus Torvalds 296e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2973567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2983567866bSAl Viro if (!acl) 299e77819e5SLinus Torvalds return -EAGAIN; 3003567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 301b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 302e77819e5SLinus Torvalds return -ECHILD; 30347291baaSChristian Brauner return posix_acl_permission(mnt_userns, inode, acl, mask); 304e77819e5SLinus Torvalds } 305e77819e5SLinus Torvalds 3062982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 3074e34e719SChristoph Hellwig if (IS_ERR(acl)) 3084e34e719SChristoph Hellwig return PTR_ERR(acl); 309e77819e5SLinus Torvalds if (acl) { 31047291baaSChristian Brauner int error = posix_acl_permission(mnt_userns, inode, acl, mask); 311e77819e5SLinus Torvalds posix_acl_release(acl); 312e77819e5SLinus Torvalds return error; 313e77819e5SLinus Torvalds } 31484635d68SLinus Torvalds #endif 315e77819e5SLinus Torvalds 316e77819e5SLinus Torvalds return -EAGAIN; 317e77819e5SLinus Torvalds } 318e77819e5SLinus Torvalds 31947291baaSChristian Brauner /** 32047291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 32147291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 32247291baaSChristian Brauner * @inode: inode to check permissions on 32347291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3245fc475b7SLinus Torvalds * 32547291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 32647291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 32747291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 32847291baaSChristian Brauner * 32947291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 33047291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 33147291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 33247291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 33347291baaSChristian Brauner * raw inode simply passs init_user_ns. 3345909ccaaSLinus Torvalds */ 33547291baaSChristian Brauner static int acl_permission_check(struct user_namespace *mnt_userns, 33647291baaSChristian Brauner struct inode *inode, int mask) 3375909ccaaSLinus Torvalds { 33826cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 33947291baaSChristian Brauner kuid_t i_uid; 3405909ccaaSLinus Torvalds 3415fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 34247291baaSChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 34347291baaSChristian Brauner if (likely(uid_eq(current_fsuid(), i_uid))) { 3445fc475b7SLinus Torvalds mask &= 7; 3455909ccaaSLinus Torvalds mode >>= 6; 3465fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3475fc475b7SLinus Torvalds } 3485fc475b7SLinus Torvalds 3495fc475b7SLinus Torvalds /* Do we have ACL's? */ 350e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 35147291baaSChristian Brauner int error = check_acl(mnt_userns, inode, mask); 3525909ccaaSLinus Torvalds if (error != -EAGAIN) 3535909ccaaSLinus Torvalds return error; 3545909ccaaSLinus Torvalds } 3555909ccaaSLinus Torvalds 3565fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3575fc475b7SLinus Torvalds mask &= 7; 3585fc475b7SLinus Torvalds 3595fc475b7SLinus Torvalds /* 3605fc475b7SLinus Torvalds * Are the group permissions different from 3615fc475b7SLinus Torvalds * the other permissions in the bits we care 3625fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3635fc475b7SLinus Torvalds */ 3645fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 36547291baaSChristian Brauner kgid_t kgid = i_gid_into_mnt(mnt_userns, inode); 36647291baaSChristian Brauner if (in_group_p(kgid)) 3675909ccaaSLinus Torvalds mode >>= 3; 3685909ccaaSLinus Torvalds } 3695909ccaaSLinus Torvalds 3705fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3715fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3725909ccaaSLinus Torvalds } 3731da177e4SLinus Torvalds 3741da177e4SLinus Torvalds /** 3751da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 37647291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 3771da177e4SLinus Torvalds * @inode: inode to check access rights for 3785fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3795fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3801da177e4SLinus Torvalds * 3811da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3821da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3831da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 384b74c79e9SNick Piggin * are used for other things. 385b74c79e9SNick Piggin * 386b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 387b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 388b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 38947291baaSChristian Brauner * 39047291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39147291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39247291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39347291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39447291baaSChristian Brauner * raw inode simply passs init_user_ns. 3951da177e4SLinus Torvalds */ 39647291baaSChristian Brauner int generic_permission(struct user_namespace *mnt_userns, struct inode *inode, 39747291baaSChristian Brauner int mask) 3981da177e4SLinus Torvalds { 3995909ccaaSLinus Torvalds int ret; 4001da177e4SLinus Torvalds 4011da177e4SLinus Torvalds /* 402948409c7SAndreas Gruenbacher * Do the basic permission checks. 4031da177e4SLinus Torvalds */ 40447291baaSChristian Brauner ret = acl_permission_check(mnt_userns, inode, mask); 4055909ccaaSLinus Torvalds if (ret != -EACCES) 4065909ccaaSLinus Torvalds return ret; 4071da177e4SLinus Torvalds 408d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 409d594e7ecSAl Viro /* DACs are overridable for directories */ 410d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 41147291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 41223adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 413d594e7ecSAl Viro return 0; 41447291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4150558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4161da177e4SLinus Torvalds return 0; 4172a4c2242SStephen Smalley return -EACCES; 4182a4c2242SStephen Smalley } 4191da177e4SLinus Torvalds 4201da177e4SLinus Torvalds /* 4211da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4221da177e4SLinus Torvalds */ 4237ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 424d594e7ecSAl Viro if (mask == MAY_READ) 42547291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4260558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4271da177e4SLinus Torvalds return 0; 4282a4c2242SStephen Smalley /* 4292a4c2242SStephen Smalley * Read/write DACs are always overridable. 4302a4c2242SStephen Smalley * Executable DACs are overridable when there is 4312a4c2242SStephen Smalley * at least one exec bit set. 4322a4c2242SStephen Smalley */ 4332a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 43447291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4350558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4362a4c2242SStephen Smalley return 0; 4371da177e4SLinus Torvalds 4381da177e4SLinus Torvalds return -EACCES; 4391da177e4SLinus Torvalds } 4404d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4411da177e4SLinus Torvalds 44247291baaSChristian Brauner /** 44347291baaSChristian Brauner * do_inode_permission - UNIX permission checking 44447291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 44547291baaSChristian Brauner * @inode: inode to check permissions on 44647291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 44747291baaSChristian Brauner * 4483ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4493ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4503ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4513ddcd056SLinus Torvalds * permission function, use the fast case". 4523ddcd056SLinus Torvalds */ 45347291baaSChristian Brauner static inline int do_inode_permission(struct user_namespace *mnt_userns, 45447291baaSChristian Brauner struct inode *inode, int mask) 4553ddcd056SLinus Torvalds { 4563ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4573ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 458549c7297SChristian Brauner return inode->i_op->permission(mnt_userns, inode, mask); 4593ddcd056SLinus Torvalds 4603ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4613ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4623ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4633ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4643ddcd056SLinus Torvalds } 46547291baaSChristian Brauner return generic_permission(mnt_userns, inode, mask); 4663ddcd056SLinus Torvalds } 4673ddcd056SLinus Torvalds 468cb23beb5SChristoph Hellwig /** 4690bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4700bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 47155852635SRandy Dunlap * @inode: Inode to check permission on 4720bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4730bdaea90SDavid Howells * 4740bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4750bdaea90SDavid Howells */ 4760bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4770bdaea90SDavid Howells { 4780bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4790bdaea90SDavid Howells umode_t mode = inode->i_mode; 4800bdaea90SDavid Howells 4810bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 482bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4830bdaea90SDavid Howells return -EROFS; 4840bdaea90SDavid Howells } 4850bdaea90SDavid Howells return 0; 4860bdaea90SDavid Howells } 4870bdaea90SDavid Howells 4880bdaea90SDavid Howells /** 4890bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 49047291baaSChristian Brauner * @mnt_userns: User namespace of the mount the inode was found from 4910bdaea90SDavid Howells * @inode: Inode to check permission on 4920bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4930bdaea90SDavid Howells * 4940bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4950bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4960bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4970bdaea90SDavid Howells * 4980bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4990bdaea90SDavid Howells */ 50047291baaSChristian Brauner int inode_permission(struct user_namespace *mnt_userns, 50147291baaSChristian Brauner struct inode *inode, int mask) 5020bdaea90SDavid Howells { 5030bdaea90SDavid Howells int retval; 5040bdaea90SDavid Howells 5050bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 5060bdaea90SDavid Howells if (retval) 5070bdaea90SDavid Howells return retval; 5084bfd054aSEric Biggers 5094bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 5104bfd054aSEric Biggers /* 5114bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5124bfd054aSEric Biggers */ 5134bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5144bfd054aSEric Biggers return -EPERM; 5154bfd054aSEric Biggers 5164bfd054aSEric Biggers /* 5174bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5184bfd054aSEric Biggers * written back improperly if their true value is unknown 5194bfd054aSEric Biggers * to the vfs. 5204bfd054aSEric Biggers */ 521ba73d987SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 5224bfd054aSEric Biggers return -EACCES; 5234bfd054aSEric Biggers } 5244bfd054aSEric Biggers 52547291baaSChristian Brauner retval = do_inode_permission(mnt_userns, inode, mask); 5264bfd054aSEric Biggers if (retval) 5274bfd054aSEric Biggers return retval; 5284bfd054aSEric Biggers 5294bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5304bfd054aSEric Biggers if (retval) 5314bfd054aSEric Biggers return retval; 5324bfd054aSEric Biggers 5334bfd054aSEric Biggers return security_inode_permission(inode, mask); 5340bdaea90SDavid Howells } 5354d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5360bdaea90SDavid Howells 5370bdaea90SDavid Howells /** 5385dd784d0SJan Blunck * path_get - get a reference to a path 5395dd784d0SJan Blunck * @path: path to get the reference to 5405dd784d0SJan Blunck * 5415dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5425dd784d0SJan Blunck */ 543dcf787f3SAl Viro void path_get(const struct path *path) 5445dd784d0SJan Blunck { 5455dd784d0SJan Blunck mntget(path->mnt); 5465dd784d0SJan Blunck dget(path->dentry); 5475dd784d0SJan Blunck } 5485dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5495dd784d0SJan Blunck 5505dd784d0SJan Blunck /** 5511d957f9bSJan Blunck * path_put - put a reference to a path 5521d957f9bSJan Blunck * @path: path to put the reference to 5531d957f9bSJan Blunck * 5541d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5551d957f9bSJan Blunck */ 556dcf787f3SAl Viro void path_put(const struct path *path) 5571da177e4SLinus Torvalds { 5581d957f9bSJan Blunck dput(path->dentry); 5591d957f9bSJan Blunck mntput(path->mnt); 5601da177e4SLinus Torvalds } 5611d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5621da177e4SLinus Torvalds 563894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5641f55a6ecSAl Viro struct nameidata { 5651f55a6ecSAl Viro struct path path; 5661f55a6ecSAl Viro struct qstr last; 5671f55a6ecSAl Viro struct path root; 5681f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 569bcba1e7dSAl Viro unsigned int flags, state; 570ab87f9a5SAleksa Sarai unsigned seq, m_seq, r_seq; 5711f55a6ecSAl Viro int last_type; 5721f55a6ecSAl Viro unsigned depth; 573756daf26SNeilBrown int total_link_count; 574697fc6caSAl Viro struct saved { 575697fc6caSAl Viro struct path link; 576fceef393SAl Viro struct delayed_call done; 577697fc6caSAl Viro const char *name; 5780450b2d1SAl Viro unsigned seq; 579894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5809883d185SAl Viro struct filename *name; 5819883d185SAl Viro struct nameidata *saved; 5829883d185SAl Viro unsigned root_seq; 5839883d185SAl Viro int dfd; 5840f705953SAl Viro kuid_t dir_uid; 5850f705953SAl Viro umode_t dir_mode; 5863859a271SKees Cook } __randomize_layout; 5871f55a6ecSAl Viro 588bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 589bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 590bcba1e7dSAl Viro #define ND_JUMPED 4 591bcba1e7dSAl Viro 59206422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 593894bc8c4SAl Viro { 594756daf26SNeilBrown struct nameidata *old = current->nameidata; 595756daf26SNeilBrown p->stack = p->internal; 5967962c7d1SAl Viro p->depth = 0; 597c8a53ee5SAl Viro p->dfd = dfd; 598c8a53ee5SAl Viro p->name = name; 5997d01ef75SAl Viro p->path.mnt = NULL; 6007d01ef75SAl Viro p->path.dentry = NULL; 601756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 6029883d185SAl Viro p->saved = old; 603756daf26SNeilBrown current->nameidata = p; 604894bc8c4SAl Viro } 605894bc8c4SAl Viro 60606422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 60706422964SAl Viro const struct path *root) 60806422964SAl Viro { 60906422964SAl Viro __set_nameidata(p, dfd, name); 61006422964SAl Viro p->state = 0; 61106422964SAl Viro if (unlikely(root)) { 61206422964SAl Viro p->state = ND_ROOT_PRESET; 61306422964SAl Viro p->root = *root; 61406422964SAl Viro } 61506422964SAl Viro } 61606422964SAl Viro 6179883d185SAl Viro static void restore_nameidata(void) 618894bc8c4SAl Viro { 6199883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 620756daf26SNeilBrown 621756daf26SNeilBrown current->nameidata = old; 622756daf26SNeilBrown if (old) 623756daf26SNeilBrown old->total_link_count = now->total_link_count; 624e1a63bbcSAl Viro if (now->stack != now->internal) 625756daf26SNeilBrown kfree(now->stack); 626894bc8c4SAl Viro } 627894bc8c4SAl Viro 62860ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 629894bc8c4SAl Viro { 630bc40aee0SAl Viro struct saved *p; 631bc40aee0SAl Viro 6326da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 63360ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 634bc40aee0SAl Viro if (unlikely(!p)) 63560ef60c7SAl Viro return false; 636894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 637894bc8c4SAl Viro nd->stack = p; 63860ef60c7SAl Viro return true; 639894bc8c4SAl Viro } 640894bc8c4SAl Viro 641397d425dSEric W. Biederman /** 6426b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 643397d425dSEric W. Biederman * 644397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 645397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 646397d425dSEric W. Biederman */ 6476b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 648397d425dSEric W. Biederman { 64995dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 650397d425dSEric W. Biederman 651402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 652402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 653397d425dSEric W. Biederman return true; 654397d425dSEric W. Biederman 6556b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 656397d425dSEric W. Biederman } 657397d425dSEric W. Biederman 6587973387aSAl Viro static void drop_links(struct nameidata *nd) 6597973387aSAl Viro { 6607973387aSAl Viro int i = nd->depth; 6617973387aSAl Viro while (i--) { 6627973387aSAl Viro struct saved *last = nd->stack + i; 663fceef393SAl Viro do_delayed_call(&last->done); 664fceef393SAl Viro clear_delayed_call(&last->done); 6657973387aSAl Viro } 6667973387aSAl Viro } 6677973387aSAl Viro 6687973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6697973387aSAl Viro { 6707973387aSAl Viro drop_links(nd); 6717973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6727973387aSAl Viro int i; 6737973387aSAl Viro path_put(&nd->path); 6747973387aSAl Viro for (i = 0; i < nd->depth; i++) 6757973387aSAl Viro path_put(&nd->stack[i].link); 676bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 677102b8af2SAl Viro path_put(&nd->root); 678bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 679102b8af2SAl Viro } 6807973387aSAl Viro } else { 6817973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6827973387aSAl Viro rcu_read_unlock(); 6837973387aSAl Viro } 6847973387aSAl Viro nd->depth = 0; 6857d01ef75SAl Viro nd->path.mnt = NULL; 6867d01ef75SAl Viro nd->path.dentry = NULL; 6877973387aSAl Viro } 6887973387aSAl Viro 6897973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6902aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6917973387aSAl Viro { 6922aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6937973387aSAl Viro if (unlikely(res)) { 6947973387aSAl Viro if (res > 0) 6957973387aSAl Viro path->mnt = NULL; 6967973387aSAl Viro path->dentry = NULL; 6977973387aSAl Viro return false; 6987973387aSAl Viro } 6997973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 7007973387aSAl Viro path->dentry = NULL; 7017973387aSAl Viro return false; 7027973387aSAl Viro } 7037973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7047973387aSAl Viro } 7057973387aSAl Viro 7062aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7072aa38470SAl Viro struct path *path, unsigned seq) 7082aa38470SAl Viro { 7095bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7102aa38470SAl Viro } 7112aa38470SAl Viro 7127973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7137973387aSAl Viro { 7147973387aSAl Viro int i; 715eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 716eacd9aa8SAl Viro drop_links(nd); 717eacd9aa8SAl Viro nd->depth = 0; 718eacd9aa8SAl Viro return false; 719eacd9aa8SAl Viro } 7207973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7217973387aSAl Viro struct saved *last = nd->stack + i; 7227973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7237973387aSAl Viro drop_links(nd); 7247973387aSAl Viro nd->depth = i + 1; 7257973387aSAl Viro return false; 7267973387aSAl Viro } 7277973387aSAl Viro } 7287973387aSAl Viro return true; 7297973387aSAl Viro } 7307973387aSAl Viro 731ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 732ee594bffSAl Viro { 733adb21d2bSAleksa Sarai /* 734adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 735adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 736adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 737adb21d2bSAleksa Sarai */ 738adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 739adb21d2bSAleksa Sarai return false; 740adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 741bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 742ee594bffSAl Viro return true; 743bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 744ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 745ee594bffSAl Viro } 746ee594bffSAl Viro 74719660af7SAl Viro /* 74831e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 74919660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 75019660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 75157e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 75219660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 75319660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 75419660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 75519660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 75631e6b01fSNick Piggin */ 75731e6b01fSNick Piggin 75831e6b01fSNick Piggin /** 759e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 76019660af7SAl Viro * @nd: nameidata pathwalk data 761e36cffedSJens Axboe * Returns: true on success, false on failure 76231e6b01fSNick Piggin * 763e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7644675ac39SAl Viro * for ref-walk mode. 7654675ac39SAl Viro * Must be called from rcu-walk context. 766e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7677973387aSAl Viro * terminate_walk(). 76831e6b01fSNick Piggin */ 769e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 77031e6b01fSNick Piggin { 77131e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 77231e6b01fSNick Piggin 77331e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 774e5c832d5SLinus Torvalds 775e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 7767973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7774675ac39SAl Viro goto out1; 77884a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 77984a2bd39SAl Viro goto out; 780ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7814675ac39SAl Viro goto out; 7824675ac39SAl Viro rcu_read_unlock(); 7834675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 784e36cffedSJens Axboe return true; 7854675ac39SAl Viro 78684a2bd39SAl Viro out1: 7874675ac39SAl Viro nd->path.mnt = NULL; 7884675ac39SAl Viro nd->path.dentry = NULL; 7894675ac39SAl Viro out: 7904675ac39SAl Viro rcu_read_unlock(); 791e36cffedSJens Axboe return false; 7924675ac39SAl Viro } 7934675ac39SAl Viro 7944675ac39SAl Viro /** 795ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7964675ac39SAl Viro * @nd: nameidata pathwalk data 797ae66db45SAl Viro * @dentry: next dentry to step into 798ae66db45SAl Viro * @seq: seq number to check @dentry against 799ae66db45SAl Viro * Returns: true on success, false on failure 8004675ac39SAl Viro * 801ae66db45SAl Viro * Similar to to try_to_unlazy(), but here we have the next dentry already 802ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 803ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 804ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 8054675ac39SAl Viro * terminate_walk(). 8064675ac39SAl Viro */ 807ae66db45SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq) 8084675ac39SAl Viro { 8094675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8104675ac39SAl Viro 8114675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 8124675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8134675ac39SAl Viro goto out2; 8147973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 8157973387aSAl Viro goto out2; 8164675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8177973387aSAl Viro goto out1; 81848a066e7SAl Viro 81915570086SLinus Torvalds /* 8204675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8214675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8224675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8234675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8244675ac39SAl Viro * know the parent must still be valid if the child sequence number is 82515570086SLinus Torvalds */ 8264675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 827e5c832d5SLinus Torvalds goto out; 82884a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 82984a2bd39SAl Viro goto out_dput; 830e5c832d5SLinus Torvalds /* 831e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 832e5c832d5SLinus Torvalds * still valid and get it if required. 833e5c832d5SLinus Torvalds */ 83484a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 83584a2bd39SAl Viro goto out_dput; 8368b61e74fSAl Viro rcu_read_unlock(); 837ae66db45SAl Viro return true; 83819660af7SAl Viro 8397973387aSAl Viro out2: 8407973387aSAl Viro nd->path.mnt = NULL; 8417973387aSAl Viro out1: 8427973387aSAl Viro nd->path.dentry = NULL; 843e5c832d5SLinus Torvalds out: 8448b61e74fSAl Viro rcu_read_unlock(); 845ae66db45SAl Viro return false; 84684a2bd39SAl Viro out_dput: 84784a2bd39SAl Viro rcu_read_unlock(); 84884a2bd39SAl Viro dput(dentry); 849ae66db45SAl Viro return false; 85031e6b01fSNick Piggin } 85131e6b01fSNick Piggin 8524ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 85334286d66SNick Piggin { 854a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8554ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 856a89f8337SAl Viro else 857a89f8337SAl Viro return 1; 85834286d66SNick Piggin } 85934286d66SNick Piggin 8609f1fafeeSAl Viro /** 8619f1fafeeSAl Viro * complete_walk - successful completion of path walk 8629f1fafeeSAl Viro * @nd: pointer nameidata 86339159de2SJeff Layton * 8649f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8659f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8669f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8679f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8689f1fafeeSAl Viro * need to drop nd->path. 86939159de2SJeff Layton */ 8709f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 87139159de2SJeff Layton { 87216c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 87339159de2SJeff Layton int status; 87439159de2SJeff Layton 8759f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 876adb21d2bSAleksa Sarai /* 877adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 878adb21d2bSAleksa Sarai * externally-managed nd->root. 879adb21d2bSAleksa Sarai */ 880bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 881bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8829f1fafeeSAl Viro nd->root.mnt = NULL; 8836c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 884e36cffedSJens Axboe if (!try_to_unlazy(nd)) 88548a066e7SAl Viro return -ECHILD; 88648a066e7SAl Viro } 8879f1fafeeSAl Viro 888adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 889adb21d2bSAleksa Sarai /* 890adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 891adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 892adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 893adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 894adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 895adb21d2bSAleksa Sarai * 896adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 897adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 898adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 899adb21d2bSAleksa Sarai * requested root to userspace. 900adb21d2bSAleksa Sarai * 901adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 902adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 903adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 904adb21d2bSAleksa Sarai */ 905adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 906adb21d2bSAleksa Sarai return -EXDEV; 907adb21d2bSAleksa Sarai } 908adb21d2bSAleksa Sarai 909bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 91039159de2SJeff Layton return 0; 91139159de2SJeff Layton 912ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 91316c2cd71SAl Viro return 0; 91416c2cd71SAl Viro 915ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 91639159de2SJeff Layton if (status > 0) 91739159de2SJeff Layton return 0; 91839159de2SJeff Layton 91916c2cd71SAl Viro if (!status) 92039159de2SJeff Layton status = -ESTALE; 92116c2cd71SAl Viro 92239159de2SJeff Layton return status; 92339159de2SJeff Layton } 92439159de2SJeff Layton 925740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9262a737871SAl Viro { 92731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9289e6697e2SAl Viro 929adb21d2bSAleksa Sarai /* 930adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 931adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 932adb21d2bSAleksa Sarai * from the dirfd. 933adb21d2bSAleksa Sarai */ 934adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 935adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 936adb21d2bSAleksa Sarai 9379e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9388f47a016SAl Viro unsigned seq; 939c28cc364SNick Piggin 940c28cc364SNick Piggin do { 941c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 94231e6b01fSNick Piggin nd->root = fs->root; 9438f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 944c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9459e6697e2SAl Viro } else { 9469e6697e2SAl Viro get_fs_root(fs, &nd->root); 947bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9489e6697e2SAl Viro } 949740a1678SAleksa Sarai return 0; 95031e6b01fSNick Piggin } 95131e6b01fSNick Piggin 952248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 953248fb5b9SAl Viro { 954adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 955adb21d2bSAleksa Sarai return -EXDEV; 95672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 95772ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 95872ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 95972ba2929SAleksa Sarai return -EXDEV; 96072ba2929SAleksa Sarai } 961740a1678SAleksa Sarai if (!nd->root.mnt) { 962740a1678SAleksa Sarai int error = set_root(nd); 963740a1678SAleksa Sarai if (error) 964740a1678SAleksa Sarai return error; 965740a1678SAleksa Sarai } 966248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 967248fb5b9SAl Viro struct dentry *d; 968248fb5b9SAl Viro nd->path = nd->root; 969248fb5b9SAl Viro d = nd->path.dentry; 970248fb5b9SAl Viro nd->inode = d->d_inode; 971248fb5b9SAl Viro nd->seq = nd->root_seq; 972248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 973248fb5b9SAl Viro return -ECHILD; 974248fb5b9SAl Viro } else { 975248fb5b9SAl Viro path_put(&nd->path); 976248fb5b9SAl Viro nd->path = nd->root; 977248fb5b9SAl Viro path_get(&nd->path); 978248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 979248fb5b9SAl Viro } 980bcba1e7dSAl Viro nd->state |= ND_JUMPED; 981248fb5b9SAl Viro return 0; 982248fb5b9SAl Viro } 983248fb5b9SAl Viro 984b5fb63c1SChristoph Hellwig /* 9856b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 986b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 987b5fb63c1SChristoph Hellwig */ 9881bc82070SAleksa Sarai int nd_jump_link(struct path *path) 989b5fb63c1SChristoph Hellwig { 9904b99d499SAleksa Sarai int error = -ELOOP; 9916e77137bSAl Viro struct nameidata *nd = current->nameidata; 992b5fb63c1SChristoph Hellwig 9934b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9944b99d499SAleksa Sarai goto err; 9954b99d499SAleksa Sarai 99672ba2929SAleksa Sarai error = -EXDEV; 99772ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 99872ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 99972ba2929SAleksa Sarai goto err; 100072ba2929SAleksa Sarai } 1001adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 1002adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 1003adb21d2bSAleksa Sarai goto err; 100472ba2929SAleksa Sarai 10054b99d499SAleksa Sarai path_put(&nd->path); 1006b5fb63c1SChristoph Hellwig nd->path = *path; 1007b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1008bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10091bc82070SAleksa Sarai return 0; 10104b99d499SAleksa Sarai 10114b99d499SAleksa Sarai err: 10124b99d499SAleksa Sarai path_put(path); 10134b99d499SAleksa Sarai return error; 1014b5fb63c1SChristoph Hellwig } 1015b5fb63c1SChristoph Hellwig 1016b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1017574197e0SAl Viro { 101821c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1019fceef393SAl Viro do_delayed_call(&last->done); 10206548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1021b9ff4429SAl Viro path_put(&last->link); 1022574197e0SAl Viro } 1023574197e0SAl Viro 10249c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10259c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10269c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10279c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10289c011be1SLuis Chamberlain 10299c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10309c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10319c011be1SLuis Chamberlain { 10329c011be1SLuis Chamberlain .procname = "protected_symlinks", 10339c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10349c011be1SLuis Chamberlain .maxlen = sizeof(int), 10359c011be1SLuis Chamberlain .mode = 0600, 10369c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10379c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10389c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10399c011be1SLuis Chamberlain }, 10409c011be1SLuis Chamberlain { 10419c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10429c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10439c011be1SLuis Chamberlain .maxlen = sizeof(int), 10449c011be1SLuis Chamberlain .mode = 0600, 10459c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10469c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10479c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10489c011be1SLuis Chamberlain }, 10499c011be1SLuis Chamberlain { 10509c011be1SLuis Chamberlain .procname = "protected_fifos", 10519c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10529c011be1SLuis Chamberlain .maxlen = sizeof(int), 10539c011be1SLuis Chamberlain .mode = 0600, 10549c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10559c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10569c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10579c011be1SLuis Chamberlain }, 10589c011be1SLuis Chamberlain { 10599c011be1SLuis Chamberlain .procname = "protected_regular", 10609c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10619c011be1SLuis Chamberlain .maxlen = sizeof(int), 10629c011be1SLuis Chamberlain .mode = 0600, 10639c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10649c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10659c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10669c011be1SLuis Chamberlain }, 10679c011be1SLuis Chamberlain { } 10689c011be1SLuis Chamberlain }; 10699c011be1SLuis Chamberlain 10709c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10719c011be1SLuis Chamberlain { 10729c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10739c011be1SLuis Chamberlain return 0; 10749c011be1SLuis Chamberlain } 10759c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10769c011be1SLuis Chamberlain 10779c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1078800179c9SKees Cook 1079800179c9SKees Cook /** 1080800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 108155852635SRandy Dunlap * @nd: nameidata pathwalk data 1082800179c9SKees Cook * 1083800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1084800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1085800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1086800179c9SKees Cook * processes from failing races against path names that may change out 1087800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1088800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1089800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1090800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1091800179c9SKees Cook * 1092800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1093800179c9SKees Cook */ 1094ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1095800179c9SKees Cook { 1096ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1097ba73d987SChristian Brauner kuid_t i_uid; 1098ba73d987SChristian Brauner 1099800179c9SKees Cook if (!sysctl_protected_symlinks) 1100800179c9SKees Cook return 0; 1101800179c9SKees Cook 1102ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1103ba73d987SChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 1104800179c9SKees Cook /* Allowed if owner and follower match. */ 1105ba73d987SChristian Brauner if (uid_eq(current_cred()->fsuid, i_uid)) 1106800179c9SKees Cook return 0; 1107800179c9SKees Cook 1108800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11090f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1110800179c9SKees Cook return 0; 1111800179c9SKees Cook 1112800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1113ba73d987SChristian Brauner if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) 1114800179c9SKees Cook return 0; 1115800179c9SKees Cook 111631956502SAl Viro if (nd->flags & LOOKUP_RCU) 111731956502SAl Viro return -ECHILD; 111831956502SAl Viro 1119ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1120245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1121800179c9SKees Cook return -EACCES; 1122800179c9SKees Cook } 1123800179c9SKees Cook 1124800179c9SKees Cook /** 1125800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1126ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1127800179c9SKees Cook * @inode: the source inode to hardlink from 1128800179c9SKees Cook * 1129800179c9SKees Cook * Return false if at least one of the following conditions: 1130800179c9SKees Cook * - inode is not a regular file 1131800179c9SKees Cook * - inode is setuid 1132800179c9SKees Cook * - inode is setgid and group-exec 1133800179c9SKees Cook * - access failure for read and write 1134800179c9SKees Cook * 1135800179c9SKees Cook * Otherwise returns true. 1136800179c9SKees Cook */ 1137ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1138ba73d987SChristian Brauner struct inode *inode) 1139800179c9SKees Cook { 1140800179c9SKees Cook umode_t mode = inode->i_mode; 1141800179c9SKees Cook 1142800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1143800179c9SKees Cook if (!S_ISREG(mode)) 1144800179c9SKees Cook return false; 1145800179c9SKees Cook 1146800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1147800179c9SKees Cook if (mode & S_ISUID) 1148800179c9SKees Cook return false; 1149800179c9SKees Cook 1150800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1151800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1152800179c9SKees Cook return false; 1153800179c9SKees Cook 1154800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1155ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1156800179c9SKees Cook return false; 1157800179c9SKees Cook 1158800179c9SKees Cook return true; 1159800179c9SKees Cook } 1160800179c9SKees Cook 1161800179c9SKees Cook /** 1162800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1163ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1164800179c9SKees Cook * @link: the source to hardlink from 1165800179c9SKees Cook * 1166800179c9SKees Cook * Block hardlink when all of: 1167800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1168800179c9SKees Cook * - fsuid does not match inode 1169800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1170f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1171800179c9SKees Cook * 1172ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1173ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1174ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1175ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1176ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1177ba73d987SChristian Brauner * 1178800179c9SKees Cook * Returns 0 if successful, -ve on error. 1179800179c9SKees Cook */ 1180ba73d987SChristian Brauner int may_linkat(struct user_namespace *mnt_userns, struct path *link) 1181800179c9SKees Cook { 1182593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1183593d1ce8SEric W. Biederman 1184593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1185ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 1186ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 1187593d1ce8SEric W. Biederman return -EOVERFLOW; 1188800179c9SKees Cook 1189800179c9SKees Cook if (!sysctl_protected_hardlinks) 1190800179c9SKees Cook return 0; 1191800179c9SKees Cook 1192800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1193800179c9SKees Cook * otherwise, it must be a safe source. 1194800179c9SKees Cook */ 1195ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1196ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1197800179c9SKees Cook return 0; 1198800179c9SKees Cook 1199245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1200800179c9SKees Cook return -EPERM; 1201800179c9SKees Cook } 1202800179c9SKees Cook 120330aba665SSalvatore Mesoraca /** 120430aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 120530aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 120630aba665SSalvatore Mesoraca * exist. 1207ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 12082111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 120930aba665SSalvatore Mesoraca * @inode: the inode of the file to open 121030aba665SSalvatore Mesoraca * 121130aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 121230aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 121330aba665SSalvatore Mesoraca * - the file already exists 121430aba665SSalvatore Mesoraca * - we are in a sticky directory 121530aba665SSalvatore Mesoraca * - we don't own the file 121630aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 121730aba665SSalvatore Mesoraca * - the directory is world writable 121830aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 121930aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 122030aba665SSalvatore Mesoraca * be enough. 122130aba665SSalvatore Mesoraca * 1222ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1223ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1224ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1225ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1226ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1227ba73d987SChristian Brauner * 122830aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 122930aba665SSalvatore Mesoraca */ 1230ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1231ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 123230aba665SSalvatore Mesoraca { 1233ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1234ba73d987SChristian Brauner kuid_t dir_uid = nd->dir_uid; 1235ba73d987SChristian Brauner 123630aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 123730aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1238d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1239ba73d987SChristian Brauner uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || 1240ba73d987SChristian Brauner uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) 124130aba665SSalvatore Mesoraca return 0; 124230aba665SSalvatore Mesoraca 1243d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1244d0cb5018SAl Viro (dir_mode & 0020 && 124530aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 124630aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1247245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1248245d7369SKees Cook "sticky_create_fifo" : 1249245d7369SKees Cook "sticky_create_regular"; 1250245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 125130aba665SSalvatore Mesoraca return -EACCES; 125230aba665SSalvatore Mesoraca } 125330aba665SSalvatore Mesoraca return 0; 125430aba665SSalvatore Mesoraca } 125530aba665SSalvatore Mesoraca 1256f015f126SDavid Howells /* 1257f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1258f015f126SDavid Howells * 1259f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1260f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1261f015f126SDavid Howells * Up is towards /. 1262f015f126SDavid Howells * 1263f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1264f015f126SDavid Howells * root. 1265f015f126SDavid Howells */ 1266bab77ebfSAl Viro int follow_up(struct path *path) 12671da177e4SLinus Torvalds { 12680714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12690714a533SAl Viro struct mount *parent; 12701da177e4SLinus Torvalds struct dentry *mountpoint; 127199b7db7bSNick Piggin 127248a066e7SAl Viro read_seqlock_excl(&mount_lock); 12730714a533SAl Viro parent = mnt->mnt_parent; 12743c0a6163SAl Viro if (parent == mnt) { 127548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12761da177e4SLinus Torvalds return 0; 12771da177e4SLinus Torvalds } 12780714a533SAl Viro mntget(&parent->mnt); 1279a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 128048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1281bab77ebfSAl Viro dput(path->dentry); 1282bab77ebfSAl Viro path->dentry = mountpoint; 1283bab77ebfSAl Viro mntput(path->mnt); 12840714a533SAl Viro path->mnt = &parent->mnt; 12851da177e4SLinus Torvalds return 1; 12861da177e4SLinus Torvalds } 12874d359507SAl Viro EXPORT_SYMBOL(follow_up); 12881da177e4SLinus Torvalds 12897ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12907ef482faSAl Viro struct path *path, unsigned *seqp) 12917ef482faSAl Viro { 12927ef482faSAl Viro while (mnt_has_parent(m)) { 12937ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12947ef482faSAl Viro 12957ef482faSAl Viro m = m->mnt_parent; 12967ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12977ef482faSAl Viro root->mnt == &m->mnt)) 12987ef482faSAl Viro break; 12997ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 13007ef482faSAl Viro path->mnt = &m->mnt; 13017ef482faSAl Viro path->dentry = mountpoint; 13027ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 13037ef482faSAl Viro return true; 13047ef482faSAl Viro } 13057ef482faSAl Viro } 13067ef482faSAl Viro return false; 13077ef482faSAl Viro } 13087ef482faSAl Viro 13092aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 13102aa38470SAl Viro struct path *path) 13112aa38470SAl Viro { 13122aa38470SAl Viro bool found; 13132aa38470SAl Viro 13142aa38470SAl Viro rcu_read_lock(); 13152aa38470SAl Viro while (1) { 13162aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 13172aa38470SAl Viro 13182aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 13192aa38470SAl Viro if (unlikely(!found)) { 13202aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 13212aa38470SAl Viro break; 13222aa38470SAl Viro } else { 13232aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 13242aa38470SAl Viro break; 13252aa38470SAl Viro rcu_read_unlock(); 13262aa38470SAl Viro path_put(path); 13272aa38470SAl Viro rcu_read_lock(); 13282aa38470SAl Viro } 13292aa38470SAl Viro } 13302aa38470SAl Viro rcu_read_unlock(); 13312aa38470SAl Viro return found; 13322aa38470SAl Viro } 13332aa38470SAl Viro 1334b5c84bf6SNick Piggin /* 13359875cf80SDavid Howells * Perform an automount 13369875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 13379875cf80SDavid Howells * were called with. 13381da177e4SLinus Torvalds */ 13391c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 134031e6b01fSNick Piggin { 134125e195aaSAl Viro struct dentry *dentry = path->dentry; 13429875cf80SDavid Howells 13430ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 13440ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 13450ec26fd0SMiklos Szeredi * the name. 13460ec26fd0SMiklos Szeredi * 13470ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 13485a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 13490ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 13500ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13510ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13520ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13535a30d8a2SDavid Howells */ 13541c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13555d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 135625e195aaSAl Viro dentry->d_inode) 13579875cf80SDavid Howells return -EISDIR; 13580ec26fd0SMiklos Szeredi 13591c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13609875cf80SDavid Howells return -ELOOP; 13619875cf80SDavid Howells 136225e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1363ea5b778aSDavid Howells } 13649875cf80SDavid Howells 13659875cf80SDavid Howells /* 13669deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13679deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13689deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13699deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13709875cf80SDavid Howells */ 13719deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13729deed3ebSAl Viro int *count, unsigned lookup_flags) 13739875cf80SDavid Howells { 13749deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13759875cf80SDavid Howells bool need_mntput = false; 13768aef1884SAl Viro int ret = 0; 13779875cf80SDavid Howells 13789deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1379cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1380cc53ce53SDavid Howells * being held. */ 1381d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1382fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1383508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1384cc53ce53SDavid Howells if (ret < 0) 13858aef1884SAl Viro break; 1386cc53ce53SDavid Howells } 1387cc53ce53SDavid Howells 13889deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13899875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13909deed3ebSAl Viro if (mounted) { // ... in our namespace 13919875cf80SDavid Howells dput(path->dentry); 13929875cf80SDavid Howells if (need_mntput) 1393463ffb2eSAl Viro mntput(path->mnt); 1394463ffb2eSAl Viro path->mnt = mounted; 1395463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13969deed3ebSAl Viro // here we know it's positive 13979deed3ebSAl Viro flags = path->dentry->d_flags; 13989875cf80SDavid Howells need_mntput = true; 13999875cf80SDavid Howells continue; 1400463ffb2eSAl Viro } 14011da177e4SLinus Torvalds } 14029875cf80SDavid Howells 14039deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 14049deed3ebSAl Viro break; 14059deed3ebSAl Viro 14069deed3ebSAl Viro // uncovered automount point 14079deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 14089deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 14099875cf80SDavid Howells if (ret < 0) 14108aef1884SAl Viro break; 14119875cf80SDavid Howells } 14129875cf80SDavid Howells 14139deed3ebSAl Viro if (ret == -EISDIR) 14149deed3ebSAl Viro ret = 0; 14159deed3ebSAl Viro // possible if you race with several mount --move 14169deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 14178aef1884SAl Viro mntput(path->mnt); 14189deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1419d41efb52SAl Viro ret = -ENOENT; 14209deed3ebSAl Viro *jumped = need_mntput; 14218402752eSAl Viro return ret; 14221da177e4SLinus Torvalds } 14231da177e4SLinus Torvalds 14249deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 14259deed3ebSAl Viro int *count, unsigned lookup_flags) 14269deed3ebSAl Viro { 14279deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 14289deed3ebSAl Viro 14299deed3ebSAl Viro /* fastpath */ 14309deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 14319deed3ebSAl Viro *jumped = false; 14329deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 14339deed3ebSAl Viro return -ENOENT; 14349deed3ebSAl Viro return 0; 14359deed3ebSAl Viro } 14369deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 14379deed3ebSAl Viro } 14389deed3ebSAl Viro 1439cc53ce53SDavid Howells int follow_down_one(struct path *path) 14401da177e4SLinus Torvalds { 14411da177e4SLinus Torvalds struct vfsmount *mounted; 14421da177e4SLinus Torvalds 14431c755af4SAl Viro mounted = lookup_mnt(path); 14441da177e4SLinus Torvalds if (mounted) { 14459393bd07SAl Viro dput(path->dentry); 14469393bd07SAl Viro mntput(path->mnt); 14479393bd07SAl Viro path->mnt = mounted; 14489393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 14491da177e4SLinus Torvalds return 1; 14501da177e4SLinus Torvalds } 14511da177e4SLinus Torvalds return 0; 14521da177e4SLinus Torvalds } 14534d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14541da177e4SLinus Torvalds 14559875cf80SDavid Howells /* 14569deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14579deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14589deed3ebSAl Viro * caller is permitted to proceed or not. 14599deed3ebSAl Viro */ 14609deed3ebSAl Viro int follow_down(struct path *path) 14619deed3ebSAl Viro { 14629deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14639deed3ebSAl Viro bool jumped; 14649deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 14659deed3ebSAl Viro 14669deed3ebSAl Viro if (path->mnt != mnt) 14679deed3ebSAl Viro mntput(mnt); 14689deed3ebSAl Viro return ret; 14699deed3ebSAl Viro } 14709deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14719deed3ebSAl Viro 14729deed3ebSAl Viro /* 1473287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1474287548e4SAl Viro * we meet a managed dentry that would need blocking. 14759875cf80SDavid Howells */ 14769875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1477254cf582SAl Viro struct inode **inode, unsigned *seqp) 14789875cf80SDavid Howells { 1479ea936aebSAl Viro struct dentry *dentry = path->dentry; 1480ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1481ea936aebSAl Viro 1482ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1483ea936aebSAl Viro return true; 1484ea936aebSAl Viro 1485ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1486ea936aebSAl Viro return false; 1487ea936aebSAl Viro 148862a7375eSIan Kent for (;;) { 148962a7375eSIan Kent /* 149062a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 149162a7375eSIan Kent * that wants to block transit. 149262a7375eSIan Kent */ 1493ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1494ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1495ea936aebSAl Viro if (res) 1496ea936aebSAl Viro return res == -EISDIR; 1497ea936aebSAl Viro flags = dentry->d_flags; 1498b8faf035SNeilBrown } 149962a7375eSIan Kent 1500ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1501ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1502ea936aebSAl Viro if (mounted) { 1503c7105365SAl Viro path->mnt = &mounted->mnt; 1504ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1505bcba1e7dSAl Viro nd->state |= ND_JUMPED; 1506ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1507ea936aebSAl Viro *inode = dentry->d_inode; 150859430262SLinus Torvalds /* 1509ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1510ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1511ea936aebSAl Viro * between mount hash removal and ->mnt_root 1512ea936aebSAl Viro * becoming unpinned. 151359430262SLinus Torvalds */ 1514ea936aebSAl Viro flags = dentry->d_flags; 1515ea936aebSAl Viro continue; 15169875cf80SDavid Howells } 1517ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1518ea936aebSAl Viro return false; 1519ea936aebSAl Viro } 1520ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1521ea936aebSAl Viro } 1522287548e4SAl Viro } 1523287548e4SAl Viro 1524db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1525db3c9adeSAl Viro struct path *path, struct inode **inode, 1526db3c9adeSAl Viro unsigned int *seqp) 1527bd7c4b50SAl Viro { 15289deed3ebSAl Viro bool jumped; 1529db3c9adeSAl Viro int ret; 1530bd7c4b50SAl Viro 1531db3c9adeSAl Viro path->mnt = nd->path.mnt; 1532db3c9adeSAl Viro path->dentry = dentry; 1533c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1534c153007bSAl Viro unsigned int seq = *seqp; 1535c153007bSAl Viro if (unlikely(!*inode)) 1536c153007bSAl Viro return -ENOENT; 1537c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15389deed3ebSAl Viro return 0; 1539ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 1540c153007bSAl Viro return -ECHILD; 1541c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1542c153007bSAl Viro path->mnt = nd->path.mnt; 1543c153007bSAl Viro path->dentry = dentry; 1544c153007bSAl Viro } 15459deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 15469deed3ebSAl Viro if (jumped) { 15479deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 15489deed3ebSAl Viro ret = -EXDEV; 15499deed3ebSAl Viro else 1550bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15519deed3ebSAl Viro } 15529deed3ebSAl Viro if (unlikely(ret)) { 15539deed3ebSAl Viro dput(path->dentry); 15549deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15559deed3ebSAl Viro mntput(path->mnt); 15569deed3ebSAl Viro } else { 1557bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1558bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1559bd7c4b50SAl Viro } 1560bd7c4b50SAl Viro return ret; 1561bd7c4b50SAl Viro } 1562bd7c4b50SAl Viro 15639875cf80SDavid Howells /* 1564f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1565f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1566baa03890SNick Piggin */ 1567e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1568e3c13928SAl Viro struct dentry *dir, 15696c51e513SAl Viro unsigned int flags) 1570baa03890SNick Piggin { 1571a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1572bad61189SMiklos Szeredi if (dentry) { 1573a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1574bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 157574ff0ffcSAl Viro if (!error) 15765542aa2fSEric W. Biederman d_invalidate(dentry); 1577bad61189SMiklos Szeredi dput(dentry); 157874ff0ffcSAl Viro return ERR_PTR(error); 1579bad61189SMiklos Szeredi } 1580bad61189SMiklos Szeredi } 1581baa03890SNick Piggin return dentry; 1582baa03890SNick Piggin } 1583baa03890SNick Piggin 1584baa03890SNick Piggin /* 1585a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1586a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1587a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1588a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1589a03ece5fSAl Viro * at all. 159044396f4bSJosef Bacik */ 1591a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1592a03ece5fSAl Viro struct dentry *base, unsigned int flags) 159344396f4bSJosef Bacik { 1594a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 159544396f4bSJosef Bacik struct dentry *old; 1596a03ece5fSAl Viro struct inode *dir = base->d_inode; 1597a03ece5fSAl Viro 1598a03ece5fSAl Viro if (dentry) 1599a03ece5fSAl Viro return dentry; 160044396f4bSJosef Bacik 160144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1602a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 160344396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1604a03ece5fSAl Viro 1605a03ece5fSAl Viro dentry = d_alloc(base, name); 1606a03ece5fSAl Viro if (unlikely(!dentry)) 1607a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 160844396f4bSJosef Bacik 160972bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 161044396f4bSJosef Bacik if (unlikely(old)) { 161144396f4bSJosef Bacik dput(dentry); 161244396f4bSJosef Bacik dentry = old; 161344396f4bSJosef Bacik } 161444396f4bSJosef Bacik return dentry; 161544396f4bSJosef Bacik } 161644396f4bSJosef Bacik 161720e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 161820e34357SAl Viro struct inode **inode, 1619254cf582SAl Viro unsigned *seqp) 16201da177e4SLinus Torvalds { 162131e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16225a18fff2SAl Viro int status = 1; 16239875cf80SDavid Howells 16243cac260aSAl Viro /* 1625b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16265d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16275d0f49c1SAl Viro * going to fall back to non-racy lookup. 1628b04f784eSNick Piggin */ 162931e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 163031e6b01fSNick Piggin unsigned seq; 1631da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 16325d0f49c1SAl Viro if (unlikely(!dentry)) { 1633e36cffedSJens Axboe if (!try_to_unlazy(nd)) 163420e34357SAl Viro return ERR_PTR(-ECHILD); 163520e34357SAl Viro return NULL; 16365d0f49c1SAl Viro } 16375a18fff2SAl Viro 163812f8ad4bSLinus Torvalds /* 163912f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 164012f8ad4bSLinus Torvalds * the dentry name information from lookup. 164112f8ad4bSLinus Torvalds */ 164263afdfc7SDavid Howells *inode = d_backing_inode(dentry); 16435d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 164420e34357SAl Viro return ERR_PTR(-ECHILD); 164512f8ad4bSLinus Torvalds 164612f8ad4bSLinus Torvalds /* 164712f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 164812f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 164912f8ad4bSLinus Torvalds * 165012f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 165112f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 165212f8ad4bSLinus Torvalds */ 16535d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 165420e34357SAl Viro return ERR_PTR(-ECHILD); 16555a18fff2SAl Viro 1656254cf582SAl Viro *seqp = seq; 16574ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1658c153007bSAl Viro if (likely(status > 0)) 165920e34357SAl Viro return dentry; 1660ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 166120e34357SAl Viro return ERR_PTR(-ECHILD); 166226ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1663209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1664209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16655a18fff2SAl Viro } else { 1666e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 166781e6f520SAl Viro if (unlikely(!dentry)) 166820e34357SAl Viro return NULL; 16694ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16705d0f49c1SAl Viro } 16715a18fff2SAl Viro if (unlikely(status <= 0)) { 1672e9742b53SAl Viro if (!status) 16735d0f49c1SAl Viro d_invalidate(dentry); 16745a18fff2SAl Viro dput(dentry); 167520e34357SAl Viro return ERR_PTR(status); 16765a18fff2SAl Viro } 167720e34357SAl Viro return dentry; 1678697f514dSMiklos Szeredi } 1679697f514dSMiklos Szeredi 1680697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 168188d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1682e3c13928SAl Viro struct dentry *dir, 1683e3c13928SAl Viro unsigned int flags) 1684697f514dSMiklos Szeredi { 168588d8331aSAl Viro struct dentry *dentry, *old; 16861936386eSAl Viro struct inode *inode = dir->d_inode; 1687d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16881936386eSAl Viro 16891936386eSAl Viro /* Don't go there if it's already dead */ 169094bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 169188d8331aSAl Viro return ERR_PTR(-ENOENT); 169294bdd655SAl Viro again: 1693d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 169494bdd655SAl Viro if (IS_ERR(dentry)) 169588d8331aSAl Viro return dentry; 169694bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1697949a852eSAl Viro int error = d_revalidate(dentry, flags); 1698949a852eSAl Viro if (unlikely(error <= 0)) { 169994bdd655SAl Viro if (!error) { 1700949a852eSAl Viro d_invalidate(dentry); 1701949a852eSAl Viro dput(dentry); 170294bdd655SAl Viro goto again; 170394bdd655SAl Viro } 170494bdd655SAl Viro dput(dentry); 1705949a852eSAl Viro dentry = ERR_PTR(error); 1706949a852eSAl Viro } 170794bdd655SAl Viro } else { 17081936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 170985c7f810SAl Viro d_lookup_done(dentry); 17101936386eSAl Viro if (unlikely(old)) { 17111936386eSAl Viro dput(dentry); 17121936386eSAl Viro dentry = old; 1713949a852eSAl Viro } 1714949a852eSAl Viro } 1715e3c13928SAl Viro return dentry; 17161da177e4SLinus Torvalds } 17171da177e4SLinus Torvalds 171888d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 171988d8331aSAl Viro struct dentry *dir, 172088d8331aSAl Viro unsigned int flags) 172188d8331aSAl Viro { 172288d8331aSAl Viro struct inode *inode = dir->d_inode; 172388d8331aSAl Viro struct dentry *res; 172488d8331aSAl Viro inode_lock_shared(inode); 172588d8331aSAl Viro res = __lookup_slow(name, dir, flags); 172688d8331aSAl Viro inode_unlock_shared(inode); 172788d8331aSAl Viro return res; 172888d8331aSAl Viro } 172988d8331aSAl Viro 1730ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1731ba73d987SChristian Brauner struct nameidata *nd) 173252094c8aSAl Viro { 173352094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17347d6beb71SLinus Torvalds int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1735e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 173652094c8aSAl Viro return err; 173752094c8aSAl Viro } 1738ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 173952094c8aSAl Viro } 174052094c8aSAl Viro 174149055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1742d63ff28fSAl Viro { 174349055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 174449055906SAl Viro return -ELOOP; 17454542576bSAl Viro 17464542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17474542576bSAl Viro return 0; 17484542576bSAl Viro if (likely(nd->stack != nd->internal)) 17494542576bSAl Viro return 0; 175060ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 175149055906SAl Viro return 0; 175260ef60c7SAl Viro 175360ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 175460ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 175560ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1756aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 175760ef60c7SAl Viro 1758e36cffedSJens Axboe if (!try_to_unlazy(nd) != 0 || !grabbed_link) 175960ef60c7SAl Viro return -ECHILD; 176060ef60c7SAl Viro 176160ef60c7SAl Viro if (nd_alloc_stack(nd)) 176260ef60c7SAl Viro return 0; 1763bc40aee0SAl Viro } 176460ef60c7SAl Viro return -ENOMEM; 176549055906SAl Viro } 176649055906SAl Viro 176749055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 176849055906SAl Viro 176949055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 177049055906SAl Viro struct inode *inode, unsigned seq, int flags) 177149055906SAl Viro { 177249055906SAl Viro struct saved *last; 177349055906SAl Viro const char *res; 177449055906SAl Viro int error = reserve_stack(nd, link, seq); 177549055906SAl Viro 177649055906SAl Viro if (unlikely(error)) { 177749055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1778cd179f44SAl Viro path_put(link); 177906708adbSAl Viro return ERR_PTR(error); 1780626de996SAl Viro } 1781ab104923SAl Viro last = nd->stack + nd->depth++; 17821cf2665bSAl Viro last->link = *link; 1783fceef393SAl Viro clear_delayed_call(&last->done); 17840450b2d1SAl Viro last->seq = seq; 1785ad6cc4c3SAl Viro 1786b1a81972SAl Viro if (flags & WALK_TRAILING) { 1787ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1788ad6cc4c3SAl Viro if (unlikely(error)) 1789ad6cc4c3SAl Viro return ERR_PTR(error); 1790ad6cc4c3SAl Viro } 1791ad6cc4c3SAl Viro 1792dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1793dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1794ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1795ad6cc4c3SAl Viro 1796ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1797ad6cc4c3SAl Viro touch_atime(&last->link); 1798ad6cc4c3SAl Viro cond_resched(); 1799ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1800e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1801ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1802ad6cc4c3SAl Viro touch_atime(&last->link); 1803ad6cc4c3SAl Viro } 1804ad6cc4c3SAl Viro 1805ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1806ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1807ad6cc4c3SAl Viro if (unlikely(error)) 1808ad6cc4c3SAl Viro return ERR_PTR(error); 1809ad6cc4c3SAl Viro 1810ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1811ad6cc4c3SAl Viro if (!res) { 1812ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1813ad6cc4c3SAl Viro struct delayed_call *); 1814ad6cc4c3SAl Viro get = inode->i_op->get_link; 1815ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1816ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1817e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1818ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1819ad6cc4c3SAl Viro } else { 1820ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1821ad6cc4c3SAl Viro } 1822ad6cc4c3SAl Viro if (!res) 1823ad6cc4c3SAl Viro goto all_done; 1824ad6cc4c3SAl Viro if (IS_ERR(res)) 1825ad6cc4c3SAl Viro return res; 1826ad6cc4c3SAl Viro } 1827ad6cc4c3SAl Viro if (*res == '/') { 1828ad6cc4c3SAl Viro error = nd_jump_root(nd); 1829ad6cc4c3SAl Viro if (unlikely(error)) 1830ad6cc4c3SAl Viro return ERR_PTR(error); 1831ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1832ad6cc4c3SAl Viro ; 1833ad6cc4c3SAl Viro } 1834ad6cc4c3SAl Viro if (*res) 1835ad6cc4c3SAl Viro return res; 1836ad6cc4c3SAl Viro all_done: // pure jump 1837ad6cc4c3SAl Viro put_link(nd); 1838ad6cc4c3SAl Viro return NULL; 1839d63ff28fSAl Viro } 1840d63ff28fSAl Viro 18413ddcd056SLinus Torvalds /* 18423ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18433ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18443ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18453ddcd056SLinus Torvalds * for the common case. 18463ddcd056SLinus Torvalds */ 1847b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1848cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 18493ddcd056SLinus Torvalds { 1850cbae4d12SAl Viro struct path path; 1851cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1852cbae4d12SAl Viro 1853cbae4d12SAl Viro if (err < 0) 1854b0417d2cSAl Viro return ERR_PTR(err); 1855cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18568c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1857aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18588f64fb1cSAl Viro /* not a symlink or should not follow */ 1859c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1860c99687a0SAl Viro dput(nd->path.dentry); 1861c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1862c99687a0SAl Viro mntput(nd->path.mnt); 1863c99687a0SAl Viro } 1864c99687a0SAl Viro nd->path = path; 18658f64fb1cSAl Viro nd->inode = inode; 18668f64fb1cSAl Viro nd->seq = seq; 1867b0417d2cSAl Viro return NULL; 18688f64fb1cSAl Viro } 1869a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 187084f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1871cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1872b0417d2cSAl Viro return ERR_PTR(-ECHILD); 187384f0cd9eSAl Viro } else { 187484f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 187584f0cd9eSAl Viro mntget(path.mnt); 1876a7f77542SAl Viro } 1877b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18783ddcd056SLinus Torvalds } 18793ddcd056SLinus Torvalds 1880c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1881c2df1968SAl Viro struct inode **inodep, 1882c2df1968SAl Viro unsigned *seqp) 1883957dd41dSAl Viro { 188412487f30SAl Viro struct dentry *parent, *old; 1885957dd41dSAl Viro 188612487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 188712487f30SAl Viro goto in_root; 188812487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18897ef482faSAl Viro struct path path; 1890efe772d6SAl Viro unsigned seq; 18917ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18927ef482faSAl Viro &nd->root, &path, &seq)) 189312487f30SAl Viro goto in_root; 1894efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1895efe772d6SAl Viro return ERR_PTR(-ECHILD); 1896efe772d6SAl Viro nd->path = path; 1897efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1898efe772d6SAl Viro nd->seq = seq; 1899efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1900efe772d6SAl Viro return ERR_PTR(-ECHILD); 1901efe772d6SAl Viro /* we know that mountpoint was pinned */ 1902957dd41dSAl Viro } 190312487f30SAl Viro old = nd->path.dentry; 190412487f30SAl Viro parent = old->d_parent; 190512487f30SAl Viro *inodep = parent->d_inode; 190612487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 190712487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 190812487f30SAl Viro return ERR_PTR(-ECHILD); 190912487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 191012487f30SAl Viro return ERR_PTR(-ECHILD); 191112487f30SAl Viro return parent; 191212487f30SAl Viro in_root: 1913efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1914efe772d6SAl Viro return ERR_PTR(-ECHILD); 1915957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19167521f22bSAl Viro return ERR_PTR(-ECHILD); 1917c2df1968SAl Viro return NULL; 1918957dd41dSAl Viro } 1919957dd41dSAl Viro 1920c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1921c2df1968SAl Viro struct inode **inodep, 1922c2df1968SAl Viro unsigned *seqp) 1923957dd41dSAl Viro { 192412487f30SAl Viro struct dentry *parent; 192512487f30SAl Viro 1926957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 192712487f30SAl Viro goto in_root; 192812487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19292aa38470SAl Viro struct path path; 19302aa38470SAl Viro 19312aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19322aa38470SAl Viro &nd->root, &path)) 193312487f30SAl Viro goto in_root; 1934165200d6SAl Viro path_put(&nd->path); 1935165200d6SAl Viro nd->path = path; 19362aa38470SAl Viro nd->inode = path.dentry->d_inode; 1937165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1938165200d6SAl Viro return ERR_PTR(-EXDEV); 193912487f30SAl Viro } 1940957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 194112487f30SAl Viro parent = dget_parent(nd->path.dentry); 1942957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1943957dd41dSAl Viro dput(parent); 19447521f22bSAl Viro return ERR_PTR(-ENOENT); 1945957dd41dSAl Viro } 1946c2df1968SAl Viro *seqp = 0; 1947c2df1968SAl Viro *inodep = parent->d_inode; 1948c2df1968SAl Viro return parent; 194912487f30SAl Viro 195012487f30SAl Viro in_root: 1951957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19527521f22bSAl Viro return ERR_PTR(-EXDEV); 1953c2df1968SAl Viro dget(nd->path.dentry); 1954c2df1968SAl Viro return NULL; 1955957dd41dSAl Viro } 1956957dd41dSAl Viro 19577521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1958957dd41dSAl Viro { 1959957dd41dSAl Viro if (type == LAST_DOTDOT) { 19607521f22bSAl Viro const char *error = NULL; 1961c2df1968SAl Viro struct dentry *parent; 1962c2df1968SAl Viro struct inode *inode; 1963c2df1968SAl Viro unsigned seq; 1964957dd41dSAl Viro 1965957dd41dSAl Viro if (!nd->root.mnt) { 19667521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1967957dd41dSAl Viro if (error) 1968957dd41dSAl Viro return error; 1969957dd41dSAl Viro } 1970957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1971c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1972957dd41dSAl Viro else 1973c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1974c2df1968SAl Viro if (IS_ERR(parent)) 1975c2df1968SAl Viro return ERR_CAST(parent); 1976c2df1968SAl Viro if (unlikely(!parent)) 1977c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1978c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1979c2df1968SAl Viro else 1980c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1981c2df1968SAl Viro parent, inode, seq); 1982c2df1968SAl Viro if (unlikely(error)) 1983957dd41dSAl Viro return error; 1984957dd41dSAl Viro 1985957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1986957dd41dSAl Viro /* 1987957dd41dSAl Viro * If there was a racing rename or mount along our 1988957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1989957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1990957dd41dSAl Viro * some fallback). 1991957dd41dSAl Viro */ 1992957dd41dSAl Viro smp_rmb(); 1993957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 19947521f22bSAl Viro return ERR_PTR(-EAGAIN); 1995957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 19967521f22bSAl Viro return ERR_PTR(-EAGAIN); 1997957dd41dSAl Viro } 1998957dd41dSAl Viro } 19997521f22bSAl Viro return NULL; 2000957dd41dSAl Viro } 2001957dd41dSAl Viro 200292d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 2003ce57dfc1SAl Viro { 2004db3c9adeSAl Viro struct dentry *dentry; 2005ce57dfc1SAl Viro struct inode *inode; 2006254cf582SAl Viro unsigned seq; 2007ce57dfc1SAl Viro /* 2008ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 2009ce57dfc1SAl Viro * to be able to know about the current root directory and 2010ce57dfc1SAl Viro * parent relationships. 2011ce57dfc1SAl Viro */ 20124693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 20131c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 20144693a547SAl Viro put_link(nd); 20157521f22bSAl Viro return handle_dots(nd, nd->last_type); 20164693a547SAl Viro } 201720e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 201820e34357SAl Viro if (IS_ERR(dentry)) 201992d27016SAl Viro return ERR_CAST(dentry); 202020e34357SAl Viro if (unlikely(!dentry)) { 2021db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 2022db3c9adeSAl Viro if (IS_ERR(dentry)) 202392d27016SAl Viro return ERR_CAST(dentry); 202420e34357SAl Viro } 202556676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 202656676ec3SAl Viro put_link(nd); 2027b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 2028ce57dfc1SAl Viro } 2029ce57dfc1SAl Viro 20301da177e4SLinus Torvalds /* 2031bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2032bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2033bfcfaa77SLinus Torvalds * 2034bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2035bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2036bfcfaa77SLinus Torvalds * fast. 2037bfcfaa77SLinus Torvalds * 2038bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2039bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2040bfcfaa77SLinus Torvalds * crossing operation. 2041bfcfaa77SLinus Torvalds * 2042bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2043bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2044bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2045bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2046bfcfaa77SLinus Torvalds */ 2047bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2048bfcfaa77SLinus Torvalds 2049f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2050bfcfaa77SLinus Torvalds 2051468a9428SGeorge Spelvin #ifdef HASH_MIX 2052bfcfaa77SLinus Torvalds 2053468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2054468a9428SGeorge Spelvin 2055468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20562a18da7aSGeorge Spelvin /* 20572a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20582a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20592a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20602a18da7aSGeorge Spelvin * and no temporaries. 20612a18da7aSGeorge Spelvin * 20622a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20632a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20642a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20652a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20662a18da7aSGeorge Spelvin * 20672a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20682a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20692a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20702a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20712a18da7aSGeorge Spelvin * 20722a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20732a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20742a18da7aSGeorge Spelvin * toggling any given output bit. 20752a18da7aSGeorge Spelvin * 20762a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20772a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20782a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20792a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20802a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20812a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20822a18da7aSGeorge Spelvin * Perfect: 8192 258048 20832a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20842a18da7aSGeorge Spelvin */ 20852a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20862a18da7aSGeorge Spelvin ( x ^= (a), \ 20872a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20882a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20892a18da7aSGeorge Spelvin y *= 9 ) 2090bfcfaa77SLinus Torvalds 20910fed3ac8SGeorge Spelvin /* 20922a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20932a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20942a18da7aSGeorge Spelvin * work done before the hash value is used. 20950fed3ac8SGeorge Spelvin */ 20962a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20970fed3ac8SGeorge Spelvin { 20982a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20992a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 21002a18da7aSGeorge Spelvin return y >> 32; 21010fed3ac8SGeorge Spelvin } 21020fed3ac8SGeorge Spelvin 2103bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2104bfcfaa77SLinus Torvalds 21052a18da7aSGeorge Spelvin /* 21062a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 21072a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 21082a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 21092a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 21102a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 21112a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 21122a18da7aSGeorge Spelvin * Perfect: 2048 31744 21132a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 21142a18da7aSGeorge Spelvin */ 21152a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 21162a18da7aSGeorge Spelvin ( x ^= (a), \ 21172a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 21182a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 21192a18da7aSGeorge Spelvin y *= 9 ) 2120bfcfaa77SLinus Torvalds 21212a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 21220fed3ac8SGeorge Spelvin { 21232a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 21242a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 21250fed3ac8SGeorge Spelvin } 21260fed3ac8SGeorge Spelvin 2127bfcfaa77SLinus Torvalds #endif 2128bfcfaa77SLinus Torvalds 21292a18da7aSGeorge Spelvin /* 21302a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21312a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21322a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21332a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21342a18da7aSGeorge Spelvin * finds the delimiter after the name. 21352a18da7aSGeorge Spelvin */ 21368387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2137bfcfaa77SLinus Torvalds { 21388387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2139bfcfaa77SLinus Torvalds 2140bfcfaa77SLinus Torvalds for (;;) { 2141fcfd2fbfSGeorge Spelvin if (!len) 2142fcfd2fbfSGeorge Spelvin goto done; 2143e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2144bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2145bfcfaa77SLinus Torvalds break; 21462a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2147bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2148bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2149bfcfaa77SLinus Torvalds } 21502a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2151bfcfaa77SLinus Torvalds done: 21522a18da7aSGeorge Spelvin return fold_hash(x, y); 2153bfcfaa77SLinus Torvalds } 2154bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2155bfcfaa77SLinus Torvalds 2156fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21578387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2158fcfd2fbfSGeorge Spelvin { 21598387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21608387ff25SLinus Torvalds unsigned long adata, mask, len; 2161fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2162fcfd2fbfSGeorge Spelvin 21638387ff25SLinus Torvalds len = 0; 21648387ff25SLinus Torvalds goto inside; 21658387ff25SLinus Torvalds 2166fcfd2fbfSGeorge Spelvin do { 21672a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2168fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21698387ff25SLinus Torvalds inside: 2170fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2171fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2172fcfd2fbfSGeorge Spelvin 2173fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2174fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21752a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2176fcfd2fbfSGeorge Spelvin 21772a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2178fcfd2fbfSGeorge Spelvin } 2179fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2180fcfd2fbfSGeorge Spelvin 2181bfcfaa77SLinus Torvalds /* 2182bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2183d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2184bfcfaa77SLinus Torvalds */ 21858387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2186bfcfaa77SLinus Torvalds { 21878387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21888387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 218936126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2190bfcfaa77SLinus Torvalds 21918387ff25SLinus Torvalds len = 0; 21928387ff25SLinus Torvalds goto inside; 21938387ff25SLinus Torvalds 2194bfcfaa77SLinus Torvalds do { 21952a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2196bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21978387ff25SLinus Torvalds inside: 2198e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 219936126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 220036126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2201bfcfaa77SLinus Torvalds 220236126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 220336126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 220436126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 22052a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 220636126f8fSLinus Torvalds 22072a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2208bfcfaa77SLinus Torvalds } 2209bfcfaa77SLinus Torvalds 22102a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2211bfcfaa77SLinus Torvalds 2212fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 22138387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 22140145acc2SLinus Torvalds { 22158387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 22160145acc2SLinus Torvalds while (len--) 2217fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 22180145acc2SLinus Torvalds return end_name_hash(hash); 22190145acc2SLinus Torvalds } 2220ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 22210145acc2SLinus Torvalds 2222fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 22238387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2224fcfd2fbfSGeorge Spelvin { 22258387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2226fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2227fcfd2fbfSGeorge Spelvin 2228fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2229e0ab7af9SGeorge Spelvin while (c) { 2230fcfd2fbfSGeorge Spelvin len++; 2231fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2232fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2233e0ab7af9SGeorge Spelvin } 2234fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2235fcfd2fbfSGeorge Spelvin } 2236f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2237fcfd2fbfSGeorge Spelvin 22383ddcd056SLinus Torvalds /* 2239200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2240200e9ef7SLinus Torvalds * one character. 2241200e9ef7SLinus Torvalds */ 22428387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2243200e9ef7SLinus Torvalds { 22448387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2245200e9ef7SLinus Torvalds unsigned long len = 0, c; 2246200e9ef7SLinus Torvalds 2247200e9ef7SLinus Torvalds c = (unsigned char)*name; 2248200e9ef7SLinus Torvalds do { 2249200e9ef7SLinus Torvalds len++; 2250200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2251200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2252200e9ef7SLinus Torvalds } while (c && c != '/'); 2253d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2254200e9ef7SLinus Torvalds } 2255200e9ef7SLinus Torvalds 2256bfcfaa77SLinus Torvalds #endif 2257bfcfaa77SLinus Torvalds 2258200e9ef7SLinus Torvalds /* 22591da177e4SLinus Torvalds * Name resolution. 2260ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2261ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22621da177e4SLinus Torvalds * 2263ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2264ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22651da177e4SLinus Torvalds */ 22666de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22671da177e4SLinus Torvalds { 2268d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22691da177e4SLinus Torvalds int err; 22701da177e4SLinus Torvalds 2271b4c03536SAl Viro nd->last_type = LAST_ROOT; 2272c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22739b5858e9SAl Viro if (IS_ERR(name)) 22749b5858e9SAl Viro return PTR_ERR(name); 22751da177e4SLinus Torvalds while (*name=='/') 22761da177e4SLinus Torvalds name++; 22771a97d899SAl Viro if (!*name) { 22781a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22799e18f10aSAl Viro return 0; 22801a97d899SAl Viro } 22811da177e4SLinus Torvalds 22821da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22831da177e4SLinus Torvalds for(;;) { 2284549c7297SChristian Brauner struct user_namespace *mnt_userns; 228592d27016SAl Viro const char *link; 2286d6bb3e90SLinus Torvalds u64 hash_len; 2287fe479a58SAl Viro int type; 22881da177e4SLinus Torvalds 2289549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2290549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22911da177e4SLinus Torvalds if (err) 22923595e234SAl Viro return err; 22931da177e4SLinus Torvalds 22948387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22951da177e4SLinus Torvalds 2296fe479a58SAl Viro type = LAST_NORM; 2297d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2298fe479a58SAl Viro case 2: 2299200e9ef7SLinus Torvalds if (name[1] == '.') { 2300fe479a58SAl Viro type = LAST_DOTDOT; 2301bcba1e7dSAl Viro nd->state |= ND_JUMPED; 230216c2cd71SAl Viro } 2303fe479a58SAl Viro break; 2304fe479a58SAl Viro case 1: 2305fe479a58SAl Viro type = LAST_DOT; 2306fe479a58SAl Viro } 23075a202bcdSAl Viro if (likely(type == LAST_NORM)) { 23085a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2309bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 23105a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2311a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2312da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 23135a202bcdSAl Viro if (err < 0) 23143595e234SAl Viro return err; 2315d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2316d6bb3e90SLinus Torvalds name = this.name; 23175a202bcdSAl Viro } 23185a202bcdSAl Viro } 2319fe479a58SAl Viro 2320d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2321d6bb3e90SLinus Torvalds nd->last.name = name; 23225f4a6a69SAl Viro nd->last_type = type; 23235f4a6a69SAl Viro 2324d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2325d6bb3e90SLinus Torvalds if (!*name) 2326bdf6cbf1SAl Viro goto OK; 2327200e9ef7SLinus Torvalds /* 2328200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2329200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2330200e9ef7SLinus Torvalds */ 2331200e9ef7SLinus Torvalds do { 2332d6bb3e90SLinus Torvalds name++; 2333d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 23348620c238SAl Viro if (unlikely(!*name)) { 23358620c238SAl Viro OK: 2336d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2337c108837eSAl Viro if (!depth) { 2338549c7297SChristian Brauner nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); 23390f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2340c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 23418620c238SAl Viro return 0; 2342c108837eSAl Viro } 23438620c238SAl Viro /* last component of nested symlink */ 2344d8d4611aSAl Viro name = nd->stack[--depth].name; 23458c4efe22SAl Viro link = walk_component(nd, 0); 23461c4ff1a8SAl Viro } else { 23471c4ff1a8SAl Viro /* not the last component */ 23488c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 23498620c238SAl Viro } 235092d27016SAl Viro if (unlikely(link)) { 235192d27016SAl Viro if (IS_ERR(link)) 235292d27016SAl Viro return PTR_ERR(link); 235392d27016SAl Viro /* a symlink to follow */ 2354d8d4611aSAl Viro nd->stack[depth++].name = name; 235592d27016SAl Viro name = link; 23569e18f10aSAl Viro continue; 235748c8b0c5SAl Viro } 235897242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 235997242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2360e36cffedSJens Axboe if (!try_to_unlazy(nd)) 236197242f99SAl Viro return -ECHILD; 236297242f99SAl Viro } 23633595e234SAl Viro return -ENOTDIR; 23645f4a6a69SAl Viro } 2365ce57dfc1SAl Viro } 236697242f99SAl Viro } 23671da177e4SLinus Torvalds 2368edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2369c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 237031e6b01fSNick Piggin { 2371740a1678SAleksa Sarai int error; 2372c8a53ee5SAl Viro const char *s = nd->name->name; 237331e6b01fSNick Piggin 23746c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23756c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23766c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23776c6ec2b0SJens Axboe 2378c0eb027eSLinus Torvalds if (!*s) 2379c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2380edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2381edc2b1daSAl Viro rcu_read_lock(); 2382c0eb027eSLinus Torvalds 2383bcba1e7dSAl Viro nd->flags = flags; 2384bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2385ab87f9a5SAleksa Sarai 2386ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2387ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2388ab87f9a5SAleksa Sarai smp_rmb(); 2389ab87f9a5SAleksa Sarai 2390bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2391b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2392b18825a7SDavid Howells struct inode *inode = root->d_inode; 239393893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2394368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23955b6ca027SAl Viro nd->path = nd->root; 23965b6ca027SAl Viro nd->inode = inode; 23975b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2398ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23998f47a016SAl Viro nd->root_seq = nd->seq; 24005b6ca027SAl Viro } else { 24015b6ca027SAl Viro path_get(&nd->path); 24025b6ca027SAl Viro } 2403368ee9baSAl Viro return s; 24045b6ca027SAl Viro } 24055b6ca027SAl Viro 240631e6b01fSNick Piggin nd->root.mnt = NULL; 240731e6b01fSNick Piggin 24088db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 24098db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2410740a1678SAleksa Sarai error = nd_jump_root(nd); 2411740a1678SAleksa Sarai if (unlikely(error)) 2412740a1678SAleksa Sarai return ERR_PTR(error); 2413ef55d917SAl Viro return s; 24148db52c7eSAleksa Sarai } 24158db52c7eSAleksa Sarai 24168db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 24178db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2418e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 241931e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2420c28cc364SNick Piggin unsigned seq; 242131e6b01fSNick Piggin 2422c28cc364SNick Piggin do { 2423c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 242431e6b01fSNick Piggin nd->path = fs->pwd; 2425ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2426c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2427c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2428e41f7d4eSAl Viro } else { 2429e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2430ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2431e41f7d4eSAl Viro } 243231e6b01fSNick Piggin } else { 2433582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2434c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 243531e6b01fSNick Piggin struct dentry *dentry; 243631e6b01fSNick Piggin 24372903ff01SAl Viro if (!f.file) 2438368ee9baSAl Viro return ERR_PTR(-EBADF); 243931e6b01fSNick Piggin 24402903ff01SAl Viro dentry = f.file->f_path.dentry; 244131e6b01fSNick Piggin 2442edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 24432903ff01SAl Viro fdput(f); 2444368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2445f52e0c11SAl Viro } 24462903ff01SAl Viro 24472903ff01SAl Viro nd->path = f.file->f_path; 2448e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 244934a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 245034a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24515590ff0dSUlrich Drepper } else { 24522903ff01SAl Viro path_get(&nd->path); 245334a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24541da177e4SLinus Torvalds } 245534a26b99SAl Viro fdput(f); 2456e41f7d4eSAl Viro } 24578db52c7eSAleksa Sarai 2458adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2459adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2460adb21d2bSAleksa Sarai nd->root = nd->path; 2461adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2462adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2463adb21d2bSAleksa Sarai } else { 2464adb21d2bSAleksa Sarai path_get(&nd->root); 2465bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2466adb21d2bSAleksa Sarai } 2467adb21d2bSAleksa Sarai } 2468adb21d2bSAleksa Sarai return s; 24699b4a9b14SAl Viro } 24709b4a9b14SAl Viro 24711ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 247295fa25d9SAl Viro { 2473bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2474bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2475bd92d7feSAl Viro 2476c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2477bd92d7feSAl Viro } 2478bd92d7feSAl Viro 24794f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24804f757f3cSAl Viro { 2481c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2482db3c9adeSAl Viro dget(nd->path.dentry); 2483b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2484b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 24854f757f3cSAl Viro } 24864f757f3cSAl Viro 24879b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2488c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24899b4a9b14SAl Viro { 2490c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2491bd92d7feSAl Viro int err; 249231e6b01fSNick Piggin 24939b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24944f757f3cSAl Viro err = handle_lookup_down(nd); 24955f336e72SAl Viro if (unlikely(err < 0)) 24965f336e72SAl Viro s = ERR_PTR(err); 24974f757f3cSAl Viro } 24984f757f3cSAl Viro 24991ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 25001ccac622SAl Viro (s = lookup_last(nd)) != NULL) 25011ccac622SAl Viro ; 25024f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 25034f0ed93fSAl Viro err = handle_lookup_down(nd); 2504bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 25054f0ed93fSAl Viro } 25069f1fafeeSAl Viro if (!err) 25079f1fafeeSAl Viro err = complete_walk(nd); 2508bd92d7feSAl Viro 2509deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2510deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2511bd23a539SAl Viro err = -ENOTDIR; 2512625b6d10SAl Viro if (!err) { 2513625b6d10SAl Viro *path = nd->path; 2514625b6d10SAl Viro nd->path.mnt = NULL; 2515625b6d10SAl Viro nd->path.dentry = NULL; 2516625b6d10SAl Viro } 2517deb106c6SAl Viro terminate_walk(nd); 2518bd92d7feSAl Viro return err; 251931e6b01fSNick Piggin } 252031e6b01fSNick Piggin 2521794ebceaSStephen Brennan int filename_lookup(int dfd, struct filename *name, unsigned flags, 25229ad1aaa6SAl Viro struct path *path, struct path *root) 2523873f1eedSJeff Layton { 2524894bc8c4SAl Viro int retval; 25259883d185SAl Viro struct nameidata nd; 2526abc9f5beSAl Viro if (IS_ERR(name)) 2527abc9f5beSAl Viro return PTR_ERR(name); 252806422964SAl Viro set_nameidata(&nd, dfd, name, root); 2529c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2530873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2531c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2532873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2533c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2534873f1eedSJeff Layton 2535873f1eedSJeff Layton if (likely(!retval)) 2536161aff1dSAl Viro audit_inode(name, path->dentry, 2537161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 25389883d185SAl Viro restore_nameidata(); 2539020250f3SDmitry Kadashev return retval; 2540020250f3SDmitry Kadashev } 2541020250f3SDmitry Kadashev 25428bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2543c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2544391172c4SAl Viro struct path *parent) 25458bcb77faSAl Viro { 2546c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25479b5858e9SAl Viro int err = link_path_walk(s, nd); 25488bcb77faSAl Viro if (!err) 25498bcb77faSAl Viro err = complete_walk(nd); 2550391172c4SAl Viro if (!err) { 2551391172c4SAl Viro *parent = nd->path; 2552391172c4SAl Viro nd->path.mnt = NULL; 2553391172c4SAl Viro nd->path.dentry = NULL; 2554391172c4SAl Viro } 2555deb106c6SAl Viro terminate_walk(nd); 25568bcb77faSAl Viro return err; 25578bcb77faSAl Viro } 25588bcb77faSAl Viro 25590766ec82SStephen Brennan /* Note: this does not consume "name" */ 2560c5f563f9SAl Viro static int filename_parentat(int dfd, struct filename *name, 2561391172c4SAl Viro unsigned int flags, struct path *parent, 2562391172c4SAl Viro struct qstr *last, int *type) 25638bcb77faSAl Viro { 25648bcb77faSAl Viro int retval; 25659883d185SAl Viro struct nameidata nd; 25668bcb77faSAl Viro 25675c31b6ceSAl Viro if (IS_ERR(name)) 25680ee50b47SDmitry Kadashev return PTR_ERR(name); 256906422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2570c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25718bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2572c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25738bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2574c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2575391172c4SAl Viro if (likely(!retval)) { 2576391172c4SAl Viro *last = nd.last; 2577391172c4SAl Viro *type = nd.last_type; 2578c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2579391172c4SAl Viro } 25809883d185SAl Viro restore_nameidata(); 25810ee50b47SDmitry Kadashev return retval; 25820ee50b47SDmitry Kadashev } 25830ee50b47SDmitry Kadashev 258479714f72SAl Viro /* does lookup, returns the object with parent locked */ 25850766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25865590ff0dSUlrich Drepper { 25875c31b6ceSAl Viro struct dentry *d; 2588391172c4SAl Viro struct qstr last; 25890ee50b47SDmitry Kadashev int type, error; 259051689104SPaul Moore 2591c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25920ee50b47SDmitry Kadashev if (error) 25930ee50b47SDmitry Kadashev return ERR_PTR(error); 25945c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2595391172c4SAl Viro path_put(path); 25965c31b6ceSAl Viro return ERR_PTR(-EINVAL); 259779714f72SAl Viro } 25985955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2599391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 260079714f72SAl Viro if (IS_ERR(d)) { 26015955102cSAl Viro inode_unlock(path->dentry->d_inode); 2602391172c4SAl Viro path_put(path); 260379714f72SAl Viro } 260479714f72SAl Viro return d; 26055590ff0dSUlrich Drepper } 26065590ff0dSUlrich Drepper 26070766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 26080766ec82SStephen Brennan { 26090766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 26100766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 26110766ec82SStephen Brennan 26120766ec82SStephen Brennan putname(filename); 26130766ec82SStephen Brennan return res; 26140766ec82SStephen Brennan } 26150766ec82SStephen Brennan 2616d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2617d1811465SAl Viro { 2618794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2619794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2620794ebceaSStephen Brennan 2621794ebceaSStephen Brennan putname(filename); 2622794ebceaSStephen Brennan return ret; 2623794ebceaSStephen Brennan 2624d1811465SAl Viro } 26254d359507SAl Viro EXPORT_SYMBOL(kern_path); 2626d1811465SAl Viro 262716f18200SJosef 'Jeff' Sipek /** 262816f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 262916f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 263016f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 263116f18200SJosef 'Jeff' Sipek * @name: pointer to file name 263216f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2633e0a01249SAl Viro * @path: pointer to struct path to fill 263416f18200SJosef 'Jeff' Sipek */ 263516f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 263616f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2637e0a01249SAl Viro struct path *path) 263816f18200SJosef 'Jeff' Sipek { 2639794ebceaSStephen Brennan struct filename *filename; 26409ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2641794ebceaSStephen Brennan int ret; 2642794ebceaSStephen Brennan 2643794ebceaSStephen Brennan filename = getname_kernel(name); 26449ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2645794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2646794ebceaSStephen Brennan putname(filename); 2647794ebceaSStephen Brennan return ret; 264816f18200SJosef 'Jeff' Sipek } 26494d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 265016f18200SJosef 'Jeff' Sipek 2651c2fd68b6SChristian Brauner static int lookup_one_common(struct user_namespace *mnt_userns, 2652c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2653c2fd68b6SChristian Brauner struct qstr *this) 26543c95f0dcSAl Viro { 26553c95f0dcSAl Viro this->name = name; 26563c95f0dcSAl Viro this->len = len; 26573c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26583c95f0dcSAl Viro if (!len) 26593c95f0dcSAl Viro return -EACCES; 26603c95f0dcSAl Viro 26613c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26623c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26633c95f0dcSAl Viro return -EACCES; 26643c95f0dcSAl Viro } 26653c95f0dcSAl Viro 26663c95f0dcSAl Viro while (len--) { 26673c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26683c95f0dcSAl Viro if (c == '/' || c == '\0') 26693c95f0dcSAl Viro return -EACCES; 26703c95f0dcSAl Viro } 26713c95f0dcSAl Viro /* 26723c95f0dcSAl Viro * See if the low-level filesystem might want 26733c95f0dcSAl Viro * to use its own hash.. 26743c95f0dcSAl Viro */ 26753c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26763c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26773c95f0dcSAl Viro if (err < 0) 26783c95f0dcSAl Viro return err; 26793c95f0dcSAl Viro } 26803c95f0dcSAl Viro 2681c2fd68b6SChristian Brauner return inode_permission(mnt_userns, base->d_inode, MAY_EXEC); 26823c95f0dcSAl Viro } 26833c95f0dcSAl Viro 2684eead1911SChristoph Hellwig /** 26850da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26860da0b7fdSDavid Howells * @name: pathname component to lookup 26870da0b7fdSDavid Howells * @base: base directory to lookup from 26880da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26890da0b7fdSDavid Howells * 26900da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26910da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26920da0b7fdSDavid Howells * 26930da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26940da0b7fdSDavid Howells * not be called by generic code. 26950da0b7fdSDavid Howells * 26960da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26970da0b7fdSDavid Howells */ 26980da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26990da0b7fdSDavid Howells { 27000da0b7fdSDavid Howells struct qstr this; 27010da0b7fdSDavid Howells int err; 27020da0b7fdSDavid Howells 27030da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27040da0b7fdSDavid Howells 2705c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 27060da0b7fdSDavid Howells if (err) 27070da0b7fdSDavid Howells return ERR_PTR(err); 27080da0b7fdSDavid Howells 27090da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 27100da0b7fdSDavid Howells } 27110da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 27120da0b7fdSDavid Howells 27130da0b7fdSDavid Howells /** 2714a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2715eead1911SChristoph Hellwig * @name: pathname component to lookup 2716eead1911SChristoph Hellwig * @base: base directory to lookup from 2717eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2718eead1911SChristoph Hellwig * 2719a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 27209e7543e9SAl Viro * not be called by generic code. 2721bbddca8eSNeilBrown * 2722bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2723eead1911SChristoph Hellwig */ 2724057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2725057f6c01SJames Morris { 27268613a209SAl Viro struct dentry *dentry; 2727057f6c01SJames Morris struct qstr this; 2728cda309deSMiklos Szeredi int err; 2729057f6c01SJames Morris 27305955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27312f9092e1SDavid Woodhouse 2732c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2733cda309deSMiklos Szeredi if (err) 2734cda309deSMiklos Szeredi return ERR_PTR(err); 2735cda309deSMiklos Szeredi 27368613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27378613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2738057f6c01SJames Morris } 27394d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2740057f6c01SJames Morris 2741bbddca8eSNeilBrown /** 2742c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 2743c2fd68b6SChristian Brauner * @mnt_userns: user namespace of the mount the lookup is performed from 2744c2fd68b6SChristian Brauner * @name: pathname component to lookup 2745c2fd68b6SChristian Brauner * @base: base directory to lookup from 2746c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2747c2fd68b6SChristian Brauner * 2748c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2749c2fd68b6SChristian Brauner * not be called by generic code. 2750c2fd68b6SChristian Brauner * 2751c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2752c2fd68b6SChristian Brauner */ 2753c2fd68b6SChristian Brauner struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name, 2754c2fd68b6SChristian Brauner struct dentry *base, int len) 2755c2fd68b6SChristian Brauner { 2756c2fd68b6SChristian Brauner struct dentry *dentry; 2757c2fd68b6SChristian Brauner struct qstr this; 2758c2fd68b6SChristian Brauner int err; 2759c2fd68b6SChristian Brauner 2760c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2761c2fd68b6SChristian Brauner 2762c2fd68b6SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 2763c2fd68b6SChristian Brauner if (err) 2764c2fd68b6SChristian Brauner return ERR_PTR(err); 2765c2fd68b6SChristian Brauner 2766c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2767c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2768c2fd68b6SChristian Brauner } 2769c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2770c2fd68b6SChristian Brauner 2771c2fd68b6SChristian Brauner /** 2772bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2773bbddca8eSNeilBrown * @name: pathname component to lookup 2774bbddca8eSNeilBrown * @base: base directory to lookup from 2775bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2776bbddca8eSNeilBrown * 2777bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2778bbddca8eSNeilBrown * not be called by generic code. 2779bbddca8eSNeilBrown * 2780bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2781bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2782bbddca8eSNeilBrown */ 2783bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2784bbddca8eSNeilBrown struct dentry *base, int len) 2785bbddca8eSNeilBrown { 2786bbddca8eSNeilBrown struct qstr this; 2787bbddca8eSNeilBrown int err; 278820d00ee8SLinus Torvalds struct dentry *ret; 2789bbddca8eSNeilBrown 2790c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2791bbddca8eSNeilBrown if (err) 2792bbddca8eSNeilBrown return ERR_PTR(err); 2793bbddca8eSNeilBrown 279420d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 279520d00ee8SLinus Torvalds if (!ret) 279620d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 279720d00ee8SLinus Torvalds return ret; 2798bbddca8eSNeilBrown } 2799bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2800bbddca8eSNeilBrown 28016c2d4798SAl Viro /* 28026c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 28036c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 28046c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 28056c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 28066c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 28076c2d4798SAl Viro * this one avoids such problems. 28086c2d4798SAl Viro */ 28096c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28106c2d4798SAl Viro struct dentry *base, int len) 28116c2d4798SAl Viro { 28126c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 28132fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 28146c2d4798SAl Viro dput(ret); 28156c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 28166c2d4798SAl Viro } 28176c2d4798SAl Viro return ret; 28186c2d4798SAl Viro } 28196c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28206c2d4798SAl Viro 2821eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2822eedf265aSEric W. Biederman int path_pts(struct path *path) 2823eedf265aSEric W. Biederman { 2824eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2825eedf265aSEric W. Biederman * the input path. 2826eedf265aSEric W. Biederman */ 2827a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2828a6a7eb76SAl Viro struct dentry *child; 282919f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2830eedf265aSEric W. Biederman 2831a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2832a6a7eb76SAl Viro dput(parent); 283363b27720SAl Viro return -ENOENT; 2834a6a7eb76SAl Viro } 283563b27720SAl Viro dput(path->dentry); 283663b27720SAl Viro path->dentry = parent; 2837eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2838eedf265aSEric W. Biederman if (!child) 2839eedf265aSEric W. Biederman return -ENOENT; 2840eedf265aSEric W. Biederman 2841eedf265aSEric W. Biederman path->dentry = child; 2842eedf265aSEric W. Biederman dput(parent); 284319f6028aSAl Viro follow_down(path); 2844eedf265aSEric W. Biederman return 0; 2845eedf265aSEric W. Biederman } 2846eedf265aSEric W. Biederman #endif 2847eedf265aSEric W. Biederman 28481fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 28491fa1e7f6SAndy Whitcroft struct path *path, int *empty) 28501da177e4SLinus Torvalds { 2851794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2852794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2853794ebceaSStephen Brennan 2854794ebceaSStephen Brennan putname(filename); 2855794ebceaSStephen Brennan return ret; 28561da177e4SLinus Torvalds } 2857b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 28581fa1e7f6SAndy Whitcroft 2859ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2860ba73d987SChristian Brauner struct inode *inode) 28611da177e4SLinus Torvalds { 28628e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2863da9592edSDavid Howells 2864ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) 28651da177e4SLinus Torvalds return 0; 2866ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) 28671da177e4SLinus Torvalds return 0; 2868ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 28691da177e4SLinus Torvalds } 2870cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 28711da177e4SLinus Torvalds 28721da177e4SLinus Torvalds /* 28731da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 28741da177e4SLinus Torvalds * whether the type of victim is right. 28751da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 28761da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 28771da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 28781da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 28791da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 28801da177e4SLinus Torvalds * a. be owner of dir, or 28811da177e4SLinus Torvalds * b. be owner of victim, or 28821da177e4SLinus Torvalds * c. have CAP_FOWNER capability 28831da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 28841da177e4SLinus Torvalds * links pointing to it. 28850bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 28860bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 28870bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 28880bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 28890bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 28901da177e4SLinus Torvalds * nfs_async_unlink(). 28911da177e4SLinus Torvalds */ 2892ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2893ba73d987SChristian Brauner struct dentry *victim, bool isdir) 28941da177e4SLinus Torvalds { 289563afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 28961da177e4SLinus Torvalds int error; 28971da177e4SLinus Torvalds 2898b18825a7SDavid Howells if (d_is_negative(victim)) 28991da177e4SLinus Torvalds return -ENOENT; 2900b18825a7SDavid Howells BUG_ON(!inode); 29011da177e4SLinus Torvalds 29021da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2903593d1ce8SEric W. Biederman 2904593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2905ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 2906ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 2907593d1ce8SEric W. Biederman return -EOVERFLOW; 2908593d1ce8SEric W. Biederman 29094fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29101da177e4SLinus Torvalds 2911ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29121da177e4SLinus Torvalds if (error) 29131da177e4SLinus Torvalds return error; 29141da177e4SLinus Torvalds if (IS_APPEND(dir)) 29151da177e4SLinus Torvalds return -EPERM; 2916b18825a7SDavid Howells 2917ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2918ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2919ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 29201da177e4SLinus Torvalds return -EPERM; 29211da177e4SLinus Torvalds if (isdir) { 292244b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29231da177e4SLinus Torvalds return -ENOTDIR; 29241da177e4SLinus Torvalds if (IS_ROOT(victim)) 29251da177e4SLinus Torvalds return -EBUSY; 292644b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29271da177e4SLinus Torvalds return -EISDIR; 29281da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29291da177e4SLinus Torvalds return -ENOENT; 29301da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29311da177e4SLinus Torvalds return -EBUSY; 29321da177e4SLinus Torvalds return 0; 29331da177e4SLinus Torvalds } 29341da177e4SLinus Torvalds 29351da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29361da177e4SLinus Torvalds * dir. 29371da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29381da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29391da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2940036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2941036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2942036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 29431da177e4SLinus Torvalds */ 2944ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2945ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 29461da177e4SLinus Torvalds { 294714e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 29481da177e4SLinus Torvalds if (child->d_inode) 29491da177e4SLinus Torvalds return -EEXIST; 29501da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29511da177e4SLinus Torvalds return -ENOENT; 29528e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 2953036d5236SEric W. Biederman return -EOVERFLOW; 29548e538913SChristian Brauner 2955ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29561da177e4SLinus Torvalds } 29571da177e4SLinus Torvalds 29581da177e4SLinus Torvalds /* 29591da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 29601da177e4SLinus Torvalds */ 29611da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 29621da177e4SLinus Torvalds { 29631da177e4SLinus Torvalds struct dentry *p; 29641da177e4SLinus Torvalds 29651da177e4SLinus Torvalds if (p1 == p2) { 29665955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29671da177e4SLinus Torvalds return NULL; 29681da177e4SLinus Torvalds } 29691da177e4SLinus Torvalds 2970fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 29711da177e4SLinus Torvalds 2972e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2973e2761a11SOGAWA Hirofumi if (p) { 29745955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 29755955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 29761da177e4SLinus Torvalds return p; 29771da177e4SLinus Torvalds } 29781da177e4SLinus Torvalds 2979e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2980e2761a11SOGAWA Hirofumi if (p) { 29815955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29825955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 29831da177e4SLinus Torvalds return p; 29841da177e4SLinus Torvalds } 29851da177e4SLinus Torvalds 29865955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29875955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 29881da177e4SLinus Torvalds return NULL; 29891da177e4SLinus Torvalds } 29904d359507SAl Viro EXPORT_SYMBOL(lock_rename); 29911da177e4SLinus Torvalds 29921da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 29931da177e4SLinus Torvalds { 29945955102cSAl Viro inode_unlock(p1->d_inode); 29951da177e4SLinus Torvalds if (p1 != p2) { 29965955102cSAl Viro inode_unlock(p2->d_inode); 2997fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 29981da177e4SLinus Torvalds } 29991da177e4SLinus Torvalds } 30004d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 30011da177e4SLinus Torvalds 30026521f891SChristian Brauner /** 30036521f891SChristian Brauner * vfs_create - create new file 30046521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 30056521f891SChristian Brauner * @dir: inode of @dentry 30066521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 30076521f891SChristian Brauner * @mode: mode of the new file 30086521f891SChristian Brauner * @want_excl: whether the file must not yet exist 30096521f891SChristian Brauner * 30106521f891SChristian Brauner * Create a new file. 30116521f891SChristian Brauner * 30126521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 30136521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 30146521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 30156521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 30166521f891SChristian Brauner * raw inode simply passs init_user_ns. 30176521f891SChristian Brauner */ 30186521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 30196521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 30201da177e4SLinus Torvalds { 30216521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 30221da177e4SLinus Torvalds if (error) 30231da177e4SLinus Torvalds return error; 30241da177e4SLinus Torvalds 3025acfa4380SAl Viro if (!dir->i_op->create) 30261da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 30271da177e4SLinus Torvalds mode &= S_IALLUGO; 30281da177e4SLinus Torvalds mode |= S_IFREG; 30291da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 30301da177e4SLinus Torvalds if (error) 30311da177e4SLinus Torvalds return error; 3032549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 3033a74574aaSStephen Smalley if (!error) 3034f38aa942SAmy Griffis fsnotify_create(dir, dentry); 30351da177e4SLinus Torvalds return error; 30361da177e4SLinus Torvalds } 30374d359507SAl Viro EXPORT_SYMBOL(vfs_create); 30381da177e4SLinus Torvalds 30398e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 30408e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 30418e6c848eSAl Viro void *arg) 30428e6c848eSAl Viro { 30438e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 3044ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 30458e6c848eSAl Viro if (error) 30468e6c848eSAl Viro return error; 30478e6c848eSAl Viro 30488e6c848eSAl Viro mode &= S_IALLUGO; 30498e6c848eSAl Viro mode |= S_IFREG; 30508e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 30518e6c848eSAl Viro if (error) 30528e6c848eSAl Viro return error; 30538e6c848eSAl Viro error = f(dentry, mode, arg); 30548e6c848eSAl Viro if (!error) 30558e6c848eSAl Viro fsnotify_create(dir, dentry); 30568e6c848eSAl Viro return error; 30578e6c848eSAl Viro } 30588e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 30598e6c848eSAl Viro 3060a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3061a2982cc9SEric W. Biederman { 3062a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3063a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3064a2982cc9SEric W. Biederman } 3065a2982cc9SEric W. Biederman 3066ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 3067ba73d987SChristian Brauner int acc_mode, int flag) 30681da177e4SLinus Torvalds { 30693fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 30701da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 30711da177e4SLinus Torvalds int error; 30721da177e4SLinus Torvalds 30731da177e4SLinus Torvalds if (!inode) 30741da177e4SLinus Torvalds return -ENOENT; 30751da177e4SLinus Torvalds 3076c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3077c8fe8f30SChristoph Hellwig case S_IFLNK: 30781da177e4SLinus Torvalds return -ELOOP; 3079c8fe8f30SChristoph Hellwig case S_IFDIR: 3080fc4177beSKees Cook if (acc_mode & MAY_WRITE) 30811da177e4SLinus Torvalds return -EISDIR; 3082fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3083fc4177beSKees Cook return -EACCES; 3084c8fe8f30SChristoph Hellwig break; 3085c8fe8f30SChristoph Hellwig case S_IFBLK: 3086c8fe8f30SChristoph Hellwig case S_IFCHR: 3087a2982cc9SEric W. Biederman if (!may_open_dev(path)) 30881da177e4SLinus Torvalds return -EACCES; 3089633fb6acSKees Cook fallthrough; 3090c8fe8f30SChristoph Hellwig case S_IFIFO: 3091c8fe8f30SChristoph Hellwig case S_IFSOCK: 3092633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3093633fb6acSKees Cook return -EACCES; 30941da177e4SLinus Torvalds flag &= ~O_TRUNC; 3095c8fe8f30SChristoph Hellwig break; 30960fd338b2SKees Cook case S_IFREG: 30970fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 30980fd338b2SKees Cook return -EACCES; 30990fd338b2SKees Cook break; 31004a3fd211SDave Hansen } 3101b41572e9SDave Hansen 3102ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 3103b41572e9SDave Hansen if (error) 3104b41572e9SDave Hansen return error; 31056146f0d5SMimi Zohar 31061da177e4SLinus Torvalds /* 31071da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 31081da177e4SLinus Torvalds */ 31091da177e4SLinus Torvalds if (IS_APPEND(inode)) { 31108737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 31117715b521SAl Viro return -EPERM; 31121da177e4SLinus Torvalds if (flag & O_TRUNC) 31137715b521SAl Viro return -EPERM; 31141da177e4SLinus Torvalds } 31151da177e4SLinus Torvalds 31161da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3117ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 31187715b521SAl Viro return -EPERM; 31191da177e4SLinus Torvalds 3120f3c7691eSJ. Bruce Fields return 0; 31217715b521SAl Viro } 31227715b521SAl Viro 3123549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 31247715b521SAl Viro { 3125f0bb5aafSAl Viro const struct path *path = &filp->f_path; 31267715b521SAl Viro struct inode *inode = path->dentry->d_inode; 31277715b521SAl Viro int error = get_write_access(inode); 31281da177e4SLinus Torvalds if (error) 31297715b521SAl Viro return error; 3130482e0007SJeff Layton 3131ea0d3ab2STetsuo Handa error = security_path_truncate(path); 31321da177e4SLinus Torvalds if (!error) { 3133549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3134d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3135e1181ee6SJeff Layton filp); 31361da177e4SLinus Torvalds } 31371da177e4SLinus Torvalds put_write_access(inode); 3138acd0c935SMimi Zohar return error; 31391da177e4SLinus Torvalds } 31401da177e4SLinus Torvalds 3141d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3142d57999e1SDave Hansen { 31438a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 31448a5e929dSAl Viro flag--; 3145d57999e1SDave Hansen return flag; 3146d57999e1SDave Hansen } 3147d57999e1SDave Hansen 3148ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3149ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3150ba73d987SChristian Brauner umode_t mode) 3151d18e9008SMiklos Szeredi { 3152d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3153d18e9008SMiklos Szeredi if (error) 3154d18e9008SMiklos Szeredi return error; 3155d18e9008SMiklos Szeredi 31568e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) 31571328c727SSeth Forshee return -EOVERFLOW; 31581328c727SSeth Forshee 3159ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 316047291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3161d18e9008SMiklos Szeredi if (error) 3162d18e9008SMiklos Szeredi return error; 3163d18e9008SMiklos Szeredi 3164d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3165d18e9008SMiklos Szeredi } 3166d18e9008SMiklos Szeredi 31671acf0af9SDavid Howells /* 31681acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 31691acf0af9SDavid Howells * dentry. 31701acf0af9SDavid Howells * 31711acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 31721acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 31731acf0af9SDavid Howells * 317400a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 317500a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 317600a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 31771acf0af9SDavid Howells * 31781acf0af9SDavid Howells * Returns an error code otherwise. 31791acf0af9SDavid Howells */ 3180239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3181239eb983SAl Viro struct file *file, 31823ec2eef1SAl Viro int open_flag, umode_t mode) 3183d18e9008SMiklos Szeredi { 3184d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3185d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3186d18e9008SMiklos Szeredi int error; 3187d18e9008SMiklos Szeredi 3188d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3189d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3190d18e9008SMiklos Szeredi 319130d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 319230d90494SAl Viro file->f_path.mnt = nd->path.mnt; 31930fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 319444907d79SAl Viro open_to_namei_flags(open_flag), mode); 31956fbd0714SAl Viro d_lookup_done(dentry); 3196384f26e2SAl Viro if (!error) { 319764e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 31986fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 31996fb968cdSAl Viro dput(dentry); 32006fb968cdSAl Viro dentry = dget(file->f_path.dentry); 32016fb968cdSAl Viro } 320264e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 32032675a4ebSAl Viro error = -EIO; 3204384f26e2SAl Viro } else { 320530d90494SAl Viro if (file->f_path.dentry) { 3206d18e9008SMiklos Szeredi dput(dentry); 3207d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 320810c64ceaSAl Viro } 3209239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3210a01e718fSAl Viro error = -ENOENT; 3211d18e9008SMiklos Szeredi } 3212d18e9008SMiklos Szeredi } 3213239eb983SAl Viro if (error) { 3214d18e9008SMiklos Szeredi dput(dentry); 3215239eb983SAl Viro dentry = ERR_PTR(error); 3216239eb983SAl Viro } 3217239eb983SAl Viro return dentry; 3218d18e9008SMiklos Szeredi } 3219d18e9008SMiklos Szeredi 322031e6b01fSNick Piggin /* 32211acf0af9SDavid Howells * Look up and maybe create and open the last component. 3222d58ffd35SMiklos Szeredi * 322300a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3224d58ffd35SMiklos Szeredi * 322500a07c15SAl Viro * Returns 0 on success, that is, if 322600a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 322700a07c15SAl Viro * the file was not completely opened at this time, though lookups and 322800a07c15SAl Viro * creations were performed. 322900a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 323000a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 323100a07c15SAl Viro * hadn't been specified. 32321acf0af9SDavid Howells * 323300a07c15SAl Viro * An error code is returned on failure. 3234d58ffd35SMiklos Szeredi */ 3235da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3236d58ffd35SMiklos Szeredi const struct open_flags *op, 32373ec2eef1SAl Viro bool got_write) 3238d58ffd35SMiklos Szeredi { 3239549c7297SChristian Brauner struct user_namespace *mnt_userns; 3240d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 324154ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 32421643b43fSAl Viro int open_flag = op->open_flag; 3243d58ffd35SMiklos Szeredi struct dentry *dentry; 32441643b43fSAl Viro int error, create_error = 0; 32451643b43fSAl Viro umode_t mode = op->mode; 32466fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3247d58ffd35SMiklos Szeredi 3248ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3249da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3250d58ffd35SMiklos Szeredi 325173a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 32526fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 32536fbd0714SAl Viro for (;;) { 32546fbd0714SAl Viro if (!dentry) { 32556fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3256d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3257da5ebf5aSAl Viro return dentry; 32586fbd0714SAl Viro } 32596fbd0714SAl Viro if (d_in_lookup(dentry)) 32606fbd0714SAl Viro break; 3261d58ffd35SMiklos Szeredi 32626fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 32636fbd0714SAl Viro if (likely(error > 0)) 32646fbd0714SAl Viro break; 32656fbd0714SAl Viro if (error) 32666fbd0714SAl Viro goto out_dput; 32676fbd0714SAl Viro d_invalidate(dentry); 32686fbd0714SAl Viro dput(dentry); 32696fbd0714SAl Viro dentry = NULL; 32706fbd0714SAl Viro } 32716fbd0714SAl Viro if (dentry->d_inode) { 3272d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3273da5ebf5aSAl Viro return dentry; 32746c51e513SAl Viro } 3275d18e9008SMiklos Szeredi 32761643b43fSAl Viro /* 32771643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 32781643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 32791643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 32801643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 32811643b43fSAl Viro * 32821643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 32831643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 32841643b43fSAl Viro */ 328599a4a90cSAl Viro if (unlikely(!got_write)) 328699a4a90cSAl Viro open_flag &= ~O_TRUNC; 3287549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 32881643b43fSAl Viro if (open_flag & O_CREAT) { 328999a4a90cSAl Viro if (open_flag & O_EXCL) 329099a4a90cSAl Viro open_flag &= ~O_TRUNC; 32911643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 32921643b43fSAl Viro mode &= ~current_umask(); 329399a4a90cSAl Viro if (likely(got_write)) 3294549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3295ba73d987SChristian Brauner dentry, mode); 329699a4a90cSAl Viro else 329799a4a90cSAl Viro create_error = -EROFS; 329899a4a90cSAl Viro } 329999a4a90cSAl Viro if (create_error) 33001643b43fSAl Viro open_flag &= ~O_CREAT; 33011643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3302d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3303da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3304da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3305da5ebf5aSAl Viro return dentry; 3306239eb983SAl Viro } 330754ef4872SMiklos Szeredi 33086fbd0714SAl Viro if (d_in_lookup(dentry)) { 330912fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 331012fa5e24SAl Viro nd->flags); 33116fbd0714SAl Viro d_lookup_done(dentry); 331212fa5e24SAl Viro if (unlikely(res)) { 331312fa5e24SAl Viro if (IS_ERR(res)) { 331412fa5e24SAl Viro error = PTR_ERR(res); 331512fa5e24SAl Viro goto out_dput; 331612fa5e24SAl Viro } 331712fa5e24SAl Viro dput(dentry); 331812fa5e24SAl Viro dentry = res; 331912fa5e24SAl Viro } 332054ef4872SMiklos Szeredi } 332154ef4872SMiklos Szeredi 3322d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 33231643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 332473a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3325ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3326ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3327ce8644fcSAl Viro error = -EACCES; 3328d58ffd35SMiklos Szeredi goto out_dput; 332964894cf8SAl Viro } 3330549c7297SChristian Brauner 3331549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3332549c7297SChristian Brauner mode, open_flag & O_EXCL); 3333d58ffd35SMiklos Szeredi if (error) 3334d58ffd35SMiklos Szeredi goto out_dput; 3335d58ffd35SMiklos Szeredi } 33361643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 33371643b43fSAl Viro error = create_error; 3338d58ffd35SMiklos Szeredi goto out_dput; 3339d58ffd35SMiklos Szeredi } 3340da5ebf5aSAl Viro return dentry; 3341d58ffd35SMiklos Szeredi 3342d58ffd35SMiklos Szeredi out_dput: 3343d58ffd35SMiklos Szeredi dput(dentry); 3344da5ebf5aSAl Viro return ERR_PTR(error); 3345d58ffd35SMiklos Szeredi } 3346d58ffd35SMiklos Szeredi 3347c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 33483ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3349fb1cc555SAl Viro { 3350a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3351ca344a89SAl Viro int open_flag = op->open_flag; 335264894cf8SAl Viro bool got_write = false; 3353254cf582SAl Viro unsigned seq; 3354a1eb3315SMiklos Szeredi struct inode *inode; 3355da5ebf5aSAl Viro struct dentry *dentry; 3356b0417d2cSAl Viro const char *res; 3357fb1cc555SAl Viro 3358c3e380b0SAl Viro nd->flags |= op->intent; 3359c3e380b0SAl Viro 3360bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 336156676ec3SAl Viro if (nd->depth) 336256676ec3SAl Viro put_link(nd); 3363ff326a32SAl Viro return handle_dots(nd, nd->last_type); 33641f36f774SAl Viro } 3365a2c36b45SAl Viro 3366ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3367fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3368fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3369fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 337020e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 337120e34357SAl Viro if (IS_ERR(dentry)) 33721ccac622SAl Viro return ERR_CAST(dentry); 337320e34357SAl Viro if (likely(dentry)) 337471574865SMiklos Szeredi goto finish_lookup; 337571574865SMiklos Szeredi 33766583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3377b6183df7SMiklos Szeredi } else { 3378fe2d35ffSAl Viro /* create side of things */ 337972287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3380e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3381e36cffedSJens Axboe return ERR_PTR(-ECHILD); 338272287417SAl Viro } 3383c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 33841f36f774SAl Viro /* trailing slashes? */ 3385deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 33861ccac622SAl Viro return ERR_PTR(-EISDIR); 3387b6183df7SMiklos Szeredi } 33881f36f774SAl Viro 33899cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3390e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 339164894cf8SAl Viro /* 339264894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 339364894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 339464894cf8SAl Viro * dropping this one anyway. 339564894cf8SAl Viro */ 339664894cf8SAl Viro } 33979cf843e3SAl Viro if (open_flag & O_CREAT) 33985955102cSAl Viro inode_lock(dir->d_inode); 33999cf843e3SAl Viro else 34009cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3401da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3402f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3403f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 34049cf843e3SAl Viro if (open_flag & O_CREAT) 34055955102cSAl Viro inode_unlock(dir->d_inode); 34069cf843e3SAl Viro else 34079cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3408fb1cc555SAl Viro 3409c981a482SAl Viro if (got_write) 341059e96e65SAl Viro mnt_drop_write(nd->path.mnt); 34116c0d46c4SAl Viro 341259e96e65SAl Viro if (IS_ERR(dentry)) 341359e96e65SAl Viro return ERR_CAST(dentry); 341459e96e65SAl Viro 3415973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3416e73cabffSAl Viro dput(nd->path.dentry); 3417e73cabffSAl Viro nd->path.dentry = dentry; 3418c981a482SAl Viro return NULL; 3419fb1cc555SAl Viro } 3420fb1cc555SAl Viro 342120e34357SAl Viro finish_lookup: 342256676ec3SAl Viro if (nd->depth) 342356676ec3SAl Viro put_link(nd); 34248c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3425ff326a32SAl Viro if (unlikely(res)) 34261ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3427b0417d2cSAl Viro return res; 34281ccac622SAl Viro } 342931d1726dSAl Viro 3430c981a482SAl Viro /* 3431c981a482SAl Viro * Handle the last step of open() 3432c981a482SAl Viro */ 3433c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3434c981a482SAl Viro struct file *file, const struct open_flags *op) 3435c981a482SAl Viro { 3436549c7297SChristian Brauner struct user_namespace *mnt_userns; 3437c981a482SAl Viro int open_flag = op->open_flag; 3438c981a482SAl Viro bool do_truncate; 3439c981a482SAl Viro int acc_mode; 3440c981a482SAl Viro int error; 3441c981a482SAl Viro 3442ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3443ff326a32SAl Viro error = complete_walk(nd); 3444ff326a32SAl Viro if (error) 3445ff326a32SAl Viro return error; 3446ff326a32SAl Viro } 3447973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 344876ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3449549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 345030aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3451b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3452b94e0b32SAl Viro return -EEXIST; 345330aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3454c5971b8cSAl Viro return -EISDIR; 3455549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 345630aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 345730aba665SSalvatore Mesoraca if (unlikely(error)) 3458c5971b8cSAl Viro return error; 345930aba665SSalvatore Mesoraca } 346044b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3461c5971b8cSAl Viro return -ENOTDIR; 34626c0d46c4SAl Viro 34638795e7d4SAl Viro do_truncate = false; 34648795e7d4SAl Viro acc_mode = op->acc_mode; 34655a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 34665a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 34675a2d3eddSAl Viro open_flag &= ~O_TRUNC; 34685a2d3eddSAl Viro acc_mode = 0; 34698795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 34700f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 34710f9d1a10SAl Viro if (error) 3472c5971b8cSAl Viro return error; 34738795e7d4SAl Viro do_truncate = true; 34740f9d1a10SAl Viro } 3475549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 34768795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3477ae2bb293SAl Viro error = vfs_open(&nd->path, file); 34788795e7d4SAl Viro if (!error) 34796035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 34808795e7d4SAl Viro if (!error && do_truncate) 3481549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3482c80567c8SAl Viro if (unlikely(error > 0)) { 3483c80567c8SAl Viro WARN_ON(1); 3484c80567c8SAl Viro error = -EINVAL; 3485c80567c8SAl Viro } 34868795e7d4SAl Viro if (do_truncate) 34870f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3488c5971b8cSAl Viro return error; 3489fb1cc555SAl Viro } 3490fb1cc555SAl Viro 34916521f891SChristian Brauner /** 34926521f891SChristian Brauner * vfs_tmpfile - create tmpfile 34936521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 34946521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 34956521f891SChristian Brauner * @mode: mode of the new tmpfile 34962111c3c0SRandy Dunlap * @open_flag: flags 34976521f891SChristian Brauner * 34986521f891SChristian Brauner * Create a temporary file. 34996521f891SChristian Brauner * 35006521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 35016521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 35026521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 35036521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 35046521f891SChristian Brauner * raw inode simply passs init_user_ns. 35056521f891SChristian Brauner */ 35066521f891SChristian Brauner struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, 35076521f891SChristian Brauner struct dentry *dentry, umode_t mode, int open_flag) 3508af7bd4dcSAmir Goldstein { 3509af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3510af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3511af7bd4dcSAmir Goldstein struct inode *inode; 3512af7bd4dcSAmir Goldstein int error; 3513af7bd4dcSAmir Goldstein 3514af7bd4dcSAmir Goldstein /* we want directory to be writable */ 35156521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3516af7bd4dcSAmir Goldstein if (error) 3517af7bd4dcSAmir Goldstein goto out_err; 3518af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3519af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3520af7bd4dcSAmir Goldstein goto out_err; 3521af7bd4dcSAmir Goldstein error = -ENOMEM; 3522cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3523af7bd4dcSAmir Goldstein if (unlikely(!child)) 3524af7bd4dcSAmir Goldstein goto out_err; 3525549c7297SChristian Brauner error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); 3526af7bd4dcSAmir Goldstein if (error) 3527af7bd4dcSAmir Goldstein goto out_err; 3528af7bd4dcSAmir Goldstein error = -ENOENT; 3529af7bd4dcSAmir Goldstein inode = child->d_inode; 3530af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3531af7bd4dcSAmir Goldstein goto out_err; 3532af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3533af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3534af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3535af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3536af7bd4dcSAmir Goldstein } 3537a2d2329eSChristian Brauner ima_post_create_tmpfile(mnt_userns, inode); 3538af7bd4dcSAmir Goldstein return child; 3539af7bd4dcSAmir Goldstein 3540af7bd4dcSAmir Goldstein out_err: 3541af7bd4dcSAmir Goldstein dput(child); 3542af7bd4dcSAmir Goldstein return ERR_PTR(error); 3543af7bd4dcSAmir Goldstein } 3544af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3545af7bd4dcSAmir Goldstein 3546c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 354760545d0dSAl Viro const struct open_flags *op, 35483ec2eef1SAl Viro struct file *file) 354960545d0dSAl Viro { 35506521f891SChristian Brauner struct user_namespace *mnt_userns; 3551625b6d10SAl Viro struct dentry *child; 3552625b6d10SAl Viro struct path path; 3553c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 355460545d0dSAl Viro if (unlikely(error)) 355560545d0dSAl Viro return error; 3556625b6d10SAl Viro error = mnt_want_write(path.mnt); 355760545d0dSAl Viro if (unlikely(error)) 355860545d0dSAl Viro goto out; 35596521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 35606521f891SChristian Brauner child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); 3561af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3562684e73beSHirofumi Nakagawa if (IS_ERR(child)) 356360545d0dSAl Viro goto out2; 3564625b6d10SAl Viro dput(path.dentry); 3565625b6d10SAl Viro path.dentry = child; 3566c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 356769a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3568549c7297SChristian Brauner error = may_open(mnt_userns, &path, 0, op->open_flag); 35691e8f44f1SAl Viro if (!error) 35701e8f44f1SAl Viro error = vfs_open(&path, file); 357160545d0dSAl Viro out2: 3572625b6d10SAl Viro mnt_drop_write(path.mnt); 357360545d0dSAl Viro out: 3574625b6d10SAl Viro path_put(&path); 357560545d0dSAl Viro return error; 357660545d0dSAl Viro } 357760545d0dSAl Viro 35786ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 35796ac08709SAl Viro { 35806ac08709SAl Viro struct path path; 35816ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 35826ac08709SAl Viro if (!error) { 35836ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3584ae2bb293SAl Viro error = vfs_open(&path, file); 35856ac08709SAl Viro path_put(&path); 35866ac08709SAl Viro } 35876ac08709SAl Viro return error; 35886ac08709SAl Viro } 35896ac08709SAl Viro 3590c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3591c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35921da177e4SLinus Torvalds { 359330d90494SAl Viro struct file *file; 359413aab428SAl Viro int error; 359531e6b01fSNick Piggin 3596ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35971afc99beSAl Viro if (IS_ERR(file)) 35981afc99beSAl Viro return file; 359931e6b01fSNick Piggin 3600bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 36013ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 36025f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 36036ac08709SAl Viro error = do_o_path(nd, flags, file); 36045f336e72SAl Viro } else { 36055f336e72SAl Viro const char *s = path_init(nd, flags); 36063bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3607c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 36081ccac622SAl Viro ; 3609c5971b8cSAl Viro if (!error) 3610c5971b8cSAl Viro error = do_open(nd, file, op); 3611deb106c6SAl Viro terminate_walk(nd); 36125f336e72SAl Viro } 36137c1c01ecSAl Viro if (likely(!error)) { 3614aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 36157c1c01ecSAl Viro return file; 36167c1c01ecSAl Viro WARN_ON(1); 36177c1c01ecSAl Viro error = -EINVAL; 3618015c3bbcSMiklos Szeredi } 36197c1c01ecSAl Viro fput(file); 36202675a4ebSAl Viro if (error == -EOPENSTALE) { 36212675a4ebSAl Viro if (flags & LOOKUP_RCU) 36222675a4ebSAl Viro error = -ECHILD; 36232675a4ebSAl Viro else 36242675a4ebSAl Viro error = -ESTALE; 36252675a4ebSAl Viro } 36267c1c01ecSAl Viro return ERR_PTR(error); 3627de459215SKirill Korotaev } 36281da177e4SLinus Torvalds 3629669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3630f9652e10SAl Viro const struct open_flags *op) 363113aab428SAl Viro { 36329883d185SAl Viro struct nameidata nd; 3633f9652e10SAl Viro int flags = op->lookup_flags; 363413aab428SAl Viro struct file *filp; 363513aab428SAl Viro 363606422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3637c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 363813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3639c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 364013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3641c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 36429883d185SAl Viro restore_nameidata(); 364313aab428SAl Viro return filp; 364413aab428SAl Viro } 364513aab428SAl Viro 3646ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3647f9652e10SAl Viro const char *name, const struct open_flags *op) 364873d049a4SAl Viro { 36499883d185SAl Viro struct nameidata nd; 365073d049a4SAl Viro struct file *file; 365151689104SPaul Moore struct filename *filename; 3652bcba1e7dSAl Viro int flags = op->lookup_flags; 365373d049a4SAl Viro 3654ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 365573d049a4SAl Viro return ERR_PTR(-ELOOP); 365673d049a4SAl Viro 365751689104SPaul Moore filename = getname_kernel(name); 3658a1c83681SViresh Kumar if (IS_ERR(filename)) 365951689104SPaul Moore return ERR_CAST(filename); 366051689104SPaul Moore 366106422964SAl Viro set_nameidata(&nd, -1, filename, root); 3662c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 366373d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3664c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 366573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3666c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 36679883d185SAl Viro restore_nameidata(); 366851689104SPaul Moore putname(filename); 366973d049a4SAl Viro return file; 367073d049a4SAl Viro } 367173d049a4SAl Viro 3672b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 36731ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36741da177e4SLinus Torvalds { 3675c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3676391172c4SAl Viro struct qstr last; 3677b3d4650dSNeilBrown bool want_dir = lookup_flags & LOOKUP_DIRECTORY; 3678b3d4650dSNeilBrown unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; 3679b3d4650dSNeilBrown unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; 3680391172c4SAl Viro int type; 3681c30dabfeSJan Kara int err2; 36821ac12b4bSJeff Layton int error; 36831ac12b4bSJeff Layton 3684b3d4650dSNeilBrown error = filename_parentat(dfd, name, reval_flag, path, &last, &type); 36850ee50b47SDmitry Kadashev if (error) 36860ee50b47SDmitry Kadashev return ERR_PTR(error); 36871da177e4SLinus Torvalds 3688c663e5d8SChristoph Hellwig /* 3689c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3690c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3691c663e5d8SChristoph Hellwig */ 36925c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3693ed75e95dSAl Viro goto out; 3694c663e5d8SChristoph Hellwig 3695c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3696391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3697c663e5d8SChristoph Hellwig /* 3698b3d4650dSNeilBrown * Do the final lookup. Suppress 'create' if there is a trailing 3699b3d4650dSNeilBrown * '/', and a directory wasn't requested. 3700c663e5d8SChristoph Hellwig */ 3701b3d4650dSNeilBrown if (last.name[last.len] && !want_dir) 3702b3d4650dSNeilBrown create_flags = 0; 37035955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3704b3d4650dSNeilBrown dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags); 37051da177e4SLinus Torvalds if (IS_ERR(dentry)) 3706a8104a9fSAl Viro goto unlock; 3707c663e5d8SChristoph Hellwig 3708a8104a9fSAl Viro error = -EEXIST; 3709b18825a7SDavid Howells if (d_is_positive(dentry)) 3710a8104a9fSAl Viro goto fail; 3711b18825a7SDavid Howells 3712c663e5d8SChristoph Hellwig /* 3713c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3714c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3715c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3716c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3717c663e5d8SChristoph Hellwig */ 3718b3d4650dSNeilBrown if (unlikely(!create_flags)) { 3719a8104a9fSAl Viro error = -ENOENT; 3720ed75e95dSAl Viro goto fail; 3721e9baf6e5SAl Viro } 3722c30dabfeSJan Kara if (unlikely(err2)) { 3723c30dabfeSJan Kara error = err2; 3724a8104a9fSAl Viro goto fail; 3725c30dabfeSJan Kara } 3726e9baf6e5SAl Viro return dentry; 37271da177e4SLinus Torvalds fail: 3728a8104a9fSAl Viro dput(dentry); 3729a8104a9fSAl Viro dentry = ERR_PTR(error); 3730a8104a9fSAl Viro unlock: 37315955102cSAl Viro inode_unlock(path->dentry->d_inode); 3732c30dabfeSJan Kara if (!err2) 3733391172c4SAl Viro mnt_drop_write(path->mnt); 3734ed75e95dSAl Viro out: 3735391172c4SAl Viro path_put(path); 3736ed75e95dSAl Viro return dentry; 3737dae6ad8fSAl Viro } 3738fa14a0b8SAl Viro 3739fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3740fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3741fa14a0b8SAl Viro { 3742b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3743b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3744b4a4f213SStephen Brennan 3745b4a4f213SStephen Brennan putname(filename); 3746b4a4f213SStephen Brennan return res; 3747fa14a0b8SAl Viro } 3748dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3749dae6ad8fSAl Viro 3750921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3751921a1650SAl Viro { 3752921a1650SAl Viro dput(dentry); 37535955102cSAl Viro inode_unlock(path->dentry->d_inode); 3754a8104a9fSAl Viro mnt_drop_write(path->mnt); 3755921a1650SAl Viro path_put(path); 3756921a1650SAl Viro } 3757921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3758921a1650SAl Viro 3759520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 37601ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3761dae6ad8fSAl Viro { 3762b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3763b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3764b4a4f213SStephen Brennan 3765b4a4f213SStephen Brennan putname(filename); 3766b4a4f213SStephen Brennan return res; 3767dae6ad8fSAl Viro } 3768dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3769dae6ad8fSAl Viro 37706521f891SChristian Brauner /** 37716521f891SChristian Brauner * vfs_mknod - create device node or file 37726521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 37736521f891SChristian Brauner * @dir: inode of @dentry 37746521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 37756521f891SChristian Brauner * @mode: mode of the new device node or file 37766521f891SChristian Brauner * @dev: device number of device to create 37776521f891SChristian Brauner * 37786521f891SChristian Brauner * Create a device node or file. 37796521f891SChristian Brauner * 37806521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 37816521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 37826521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 37836521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 37846521f891SChristian Brauner * raw inode simply passs init_user_ns. 37856521f891SChristian Brauner */ 37866521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 37876521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 37881da177e4SLinus Torvalds { 3789a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 37906521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 37911da177e4SLinus Torvalds 37921da177e4SLinus Torvalds if (error) 37931da177e4SLinus Torvalds return error; 37941da177e4SLinus Torvalds 3795a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3796a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 37971da177e4SLinus Torvalds return -EPERM; 37981da177e4SLinus Torvalds 3799acfa4380SAl Viro if (!dir->i_op->mknod) 38001da177e4SLinus Torvalds return -EPERM; 38011da177e4SLinus Torvalds 380208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 380308ce5f16SSerge E. Hallyn if (error) 380408ce5f16SSerge E. Hallyn return error; 380508ce5f16SSerge E. Hallyn 38061da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 38071da177e4SLinus Torvalds if (error) 38081da177e4SLinus Torvalds return error; 38091da177e4SLinus Torvalds 3810549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3811a74574aaSStephen Smalley if (!error) 3812f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38131da177e4SLinus Torvalds return error; 38141da177e4SLinus Torvalds } 38154d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 38161da177e4SLinus Torvalds 3817f69aac00SAl Viro static int may_mknod(umode_t mode) 3818463c3197SDave Hansen { 3819463c3197SDave Hansen switch (mode & S_IFMT) { 3820463c3197SDave Hansen case S_IFREG: 3821463c3197SDave Hansen case S_IFCHR: 3822463c3197SDave Hansen case S_IFBLK: 3823463c3197SDave Hansen case S_IFIFO: 3824463c3197SDave Hansen case S_IFSOCK: 3825463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3826463c3197SDave Hansen return 0; 3827463c3197SDave Hansen case S_IFDIR: 3828463c3197SDave Hansen return -EPERM; 3829463c3197SDave Hansen default: 3830463c3197SDave Hansen return -EINVAL; 3831463c3197SDave Hansen } 3832463c3197SDave Hansen } 3833463c3197SDave Hansen 383445f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 383587c4e192SDominik Brodowski unsigned int dev) 38361da177e4SLinus Torvalds { 38376521f891SChristian Brauner struct user_namespace *mnt_userns; 38381da177e4SLinus Torvalds struct dentry *dentry; 3839dae6ad8fSAl Viro struct path path; 3840dae6ad8fSAl Viro int error; 3841972567f1SJeff Layton unsigned int lookup_flags = 0; 38421da177e4SLinus Torvalds 38438e4bfca1SAl Viro error = may_mknod(mode); 38448e4bfca1SAl Viro if (error) 38457797251bSDmitry Kadashev goto out1; 3846972567f1SJeff Layton retry: 3847b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 38487797251bSDmitry Kadashev error = PTR_ERR(dentry); 3849dae6ad8fSAl Viro if (IS_ERR(dentry)) 38507797251bSDmitry Kadashev goto out1; 38512ad94ae6SAl Viro 3852dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3853ce3b0f8dSAl Viro mode &= ~current_umask(); 3854dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3855be6d3e56SKentaro Takeda if (error) 38567797251bSDmitry Kadashev goto out2; 38576521f891SChristian Brauner 38586521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 38591da177e4SLinus Torvalds switch (mode & S_IFMT) { 38601da177e4SLinus Torvalds case 0: case S_IFREG: 38616521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 38626521f891SChristian Brauner dentry, mode, true); 386305d1a717SMimi Zohar if (!error) 3864a2d2329eSChristian Brauner ima_post_path_mknod(mnt_userns, dentry); 38651da177e4SLinus Torvalds break; 38661da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 38676521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 38686521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 38691da177e4SLinus Torvalds break; 38701da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 38716521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 38726521f891SChristian Brauner dentry, mode, 0); 38731da177e4SLinus Torvalds break; 38741da177e4SLinus Torvalds } 38757797251bSDmitry Kadashev out2: 3876921a1650SAl Viro done_path_create(&path, dentry); 3877972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3878972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3879972567f1SJeff Layton goto retry; 3880972567f1SJeff Layton } 38817797251bSDmitry Kadashev out1: 38827797251bSDmitry Kadashev putname(name); 38831da177e4SLinus Torvalds return error; 38841da177e4SLinus Torvalds } 38851da177e4SLinus Torvalds 388687c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 388787c4e192SDominik Brodowski unsigned int, dev) 388887c4e192SDominik Brodowski { 38897797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 389087c4e192SDominik Brodowski } 389187c4e192SDominik Brodowski 38928208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 38935590ff0dSUlrich Drepper { 38947797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 38955590ff0dSUlrich Drepper } 38965590ff0dSUlrich Drepper 38976521f891SChristian Brauner /** 38986521f891SChristian Brauner * vfs_mkdir - create directory 38996521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 39006521f891SChristian Brauner * @dir: inode of @dentry 39016521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39026521f891SChristian Brauner * @mode: mode of the new directory 39036521f891SChristian Brauner * 39046521f891SChristian Brauner * Create a directory. 39056521f891SChristian Brauner * 39066521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39076521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39086521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39096521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39106521f891SChristian Brauner * raw inode simply passs init_user_ns. 39116521f891SChristian Brauner */ 39126521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 39136521f891SChristian Brauner struct dentry *dentry, umode_t mode) 39141da177e4SLinus Torvalds { 39156521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 39168de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39171da177e4SLinus Torvalds 39181da177e4SLinus Torvalds if (error) 39191da177e4SLinus Torvalds return error; 39201da177e4SLinus Torvalds 3921acfa4380SAl Viro if (!dir->i_op->mkdir) 39221da177e4SLinus Torvalds return -EPERM; 39231da177e4SLinus Torvalds 39241da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 39251da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 39261da177e4SLinus Torvalds if (error) 39271da177e4SLinus Torvalds return error; 39281da177e4SLinus Torvalds 39298de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 39308de52778SAl Viro return -EMLINK; 39318de52778SAl Viro 3932549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 3933a74574aaSStephen Smalley if (!error) 3934f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 39351da177e4SLinus Torvalds return error; 39361da177e4SLinus Torvalds } 39374d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 39381da177e4SLinus Torvalds 393945f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 39401da177e4SLinus Torvalds { 39416902d925SDave Hansen struct dentry *dentry; 3942dae6ad8fSAl Viro struct path path; 3943dae6ad8fSAl Viro int error; 3944b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 39451da177e4SLinus Torvalds 3946b76d8b82SJeff Layton retry: 3947b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 3948584d3226SDmitry Kadashev error = PTR_ERR(dentry); 39496902d925SDave Hansen if (IS_ERR(dentry)) 3950584d3226SDmitry Kadashev goto out_putname; 39516902d925SDave Hansen 3952dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3953ce3b0f8dSAl Viro mode &= ~current_umask(); 3954dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 39556521f891SChristian Brauner if (!error) { 39566521f891SChristian Brauner struct user_namespace *mnt_userns; 39576521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 3958549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 3959549c7297SChristian Brauner mode); 39606521f891SChristian Brauner } 3961921a1650SAl Viro done_path_create(&path, dentry); 3962b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3963b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3964b76d8b82SJeff Layton goto retry; 3965b76d8b82SJeff Layton } 3966584d3226SDmitry Kadashev out_putname: 3967584d3226SDmitry Kadashev putname(name); 39681da177e4SLinus Torvalds return error; 39691da177e4SLinus Torvalds } 39701da177e4SLinus Torvalds 39710101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 39720101db7aSDominik Brodowski { 3973584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 39740101db7aSDominik Brodowski } 39750101db7aSDominik Brodowski 3976a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 39775590ff0dSUlrich Drepper { 3978584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 39795590ff0dSUlrich Drepper } 39805590ff0dSUlrich Drepper 39816521f891SChristian Brauner /** 39826521f891SChristian Brauner * vfs_rmdir - remove directory 39836521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 39846521f891SChristian Brauner * @dir: inode of @dentry 39856521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39866521f891SChristian Brauner * 39876521f891SChristian Brauner * Remove a directory. 39886521f891SChristian Brauner * 39896521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39906521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39916521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39926521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39936521f891SChristian Brauner * raw inode simply passs init_user_ns. 39946521f891SChristian Brauner */ 39956521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 39966521f891SChristian Brauner struct dentry *dentry) 39971da177e4SLinus Torvalds { 39986521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 39991da177e4SLinus Torvalds 40001da177e4SLinus Torvalds if (error) 40011da177e4SLinus Torvalds return error; 40021da177e4SLinus Torvalds 4003acfa4380SAl Viro if (!dir->i_op->rmdir) 40041da177e4SLinus Torvalds return -EPERM; 40051da177e4SLinus Torvalds 40061d2ef590SAl Viro dget(dentry); 40075955102cSAl Viro inode_lock(dentry->d_inode); 4008912dbc15SSage Weil 40091da177e4SLinus Torvalds error = -EBUSY; 40101bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 40111bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4012912dbc15SSage Weil goto out; 4013912dbc15SSage Weil 40141da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4015912dbc15SSage Weil if (error) 4016912dbc15SSage Weil goto out; 4017912dbc15SSage Weil 40181da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4019912dbc15SSage Weil if (error) 4020912dbc15SSage Weil goto out; 4021912dbc15SSage Weil 40228767712fSAl Viro shrink_dcache_parent(dentry); 40231da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4024d83c49f3SAl Viro dont_mount(dentry); 40258ed936b5SEric W. Biederman detach_mounts(dentry); 40261da177e4SLinus Torvalds 4027912dbc15SSage Weil out: 40285955102cSAl Viro inode_unlock(dentry->d_inode); 40291d2ef590SAl Viro dput(dentry); 4030912dbc15SSage Weil if (!error) 4031a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 40321da177e4SLinus Torvalds return error; 40331da177e4SLinus Torvalds } 40344d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 40351da177e4SLinus Torvalds 403645f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 40371da177e4SLinus Torvalds { 40386521f891SChristian Brauner struct user_namespace *mnt_userns; 40390ee50b47SDmitry Kadashev int error; 40401da177e4SLinus Torvalds struct dentry *dentry; 4041f5beed75SAl Viro struct path path; 4042f5beed75SAl Viro struct qstr last; 4043f5beed75SAl Viro int type; 4044c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4045c6ee9206SJeff Layton retry: 4046c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 40470ee50b47SDmitry Kadashev if (error) 40480ee50b47SDmitry Kadashev goto exit1; 40491da177e4SLinus Torvalds 4050f5beed75SAl Viro switch (type) { 40511da177e4SLinus Torvalds case LAST_DOTDOT: 40521da177e4SLinus Torvalds error = -ENOTEMPTY; 40530ee50b47SDmitry Kadashev goto exit2; 40541da177e4SLinus Torvalds case LAST_DOT: 40551da177e4SLinus Torvalds error = -EINVAL; 40560ee50b47SDmitry Kadashev goto exit2; 40571da177e4SLinus Torvalds case LAST_ROOT: 40581da177e4SLinus Torvalds error = -EBUSY; 40590ee50b47SDmitry Kadashev goto exit2; 40601da177e4SLinus Torvalds } 40610612d9fbSOGAWA Hirofumi 4062f5beed75SAl Viro error = mnt_want_write(path.mnt); 4063c30dabfeSJan Kara if (error) 40640ee50b47SDmitry Kadashev goto exit2; 40650612d9fbSOGAWA Hirofumi 40665955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4067f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40681da177e4SLinus Torvalds error = PTR_ERR(dentry); 40696902d925SDave Hansen if (IS_ERR(dentry)) 40700ee50b47SDmitry Kadashev goto exit3; 4071e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4072e6bc45d6STheodore Ts'o error = -ENOENT; 40730ee50b47SDmitry Kadashev goto exit4; 4074e6bc45d6STheodore Ts'o } 4075f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4076be6d3e56SKentaro Takeda if (error) 40770ee50b47SDmitry Kadashev goto exit4; 40786521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 40796521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 40800ee50b47SDmitry Kadashev exit4: 40811da177e4SLinus Torvalds dput(dentry); 40820ee50b47SDmitry Kadashev exit3: 40835955102cSAl Viro inode_unlock(path.dentry->d_inode); 4084f5beed75SAl Viro mnt_drop_write(path.mnt); 40850ee50b47SDmitry Kadashev exit2: 4086f5beed75SAl Viro path_put(&path); 4087c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4088c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4089c6ee9206SJeff Layton goto retry; 4090c6ee9206SJeff Layton } 40910ee50b47SDmitry Kadashev exit1: 409224fb33d4SAl Viro putname(name); 40931da177e4SLinus Torvalds return error; 40941da177e4SLinus Torvalds } 40951da177e4SLinus Torvalds 40963cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 40975590ff0dSUlrich Drepper { 4098e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 40995590ff0dSUlrich Drepper } 41005590ff0dSUlrich Drepper 4101b21996e3SJ. Bruce Fields /** 4102b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 41036521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 4104b21996e3SJ. Bruce Fields * @dir: parent directory 4105b21996e3SJ. Bruce Fields * @dentry: victim 4106b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4107b21996e3SJ. Bruce Fields * 4108b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4109b21996e3SJ. Bruce Fields * 4110b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4111b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4112b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4113b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4114b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4115b21996e3SJ. Bruce Fields * 4116b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4117b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4118b21996e3SJ. Bruce Fields * to be NFS exported. 41196521f891SChristian Brauner * 41206521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 41216521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 41226521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 41236521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 41246521f891SChristian Brauner * raw inode simply passs init_user_ns. 4125b21996e3SJ. Bruce Fields */ 41266521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 41276521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 41281da177e4SLinus Torvalds { 41299accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 41306521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 41311da177e4SLinus Torvalds 41321da177e4SLinus Torvalds if (error) 41331da177e4SLinus Torvalds return error; 41341da177e4SLinus Torvalds 4135acfa4380SAl Viro if (!dir->i_op->unlink) 41361da177e4SLinus Torvalds return -EPERM; 41371da177e4SLinus Torvalds 41385955102cSAl Viro inode_lock(target); 413951cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 414051cc3a66SHugh Dickins error = -EPERM; 414151cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 41421da177e4SLinus Torvalds error = -EBUSY; 41431da177e4SLinus Torvalds else { 41441da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4145bec1052eSAl Viro if (!error) { 41465a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 41475a14696cSJ. Bruce Fields if (error) 4148b21996e3SJ. Bruce Fields goto out; 41491da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 41508ed936b5SEric W. Biederman if (!error) { 4151d83c49f3SAl Viro dont_mount(dentry); 41528ed936b5SEric W. Biederman detach_mounts(dentry); 41538ed936b5SEric W. Biederman } 4154bec1052eSAl Viro } 41551da177e4SLinus Torvalds } 4156b21996e3SJ. Bruce Fields out: 41575955102cSAl Viro inode_unlock(target); 41581da177e4SLinus Torvalds 41591da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4160a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4161a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4162a37d9a17SAmir Goldstein } else if (!error) { 41639accbb97SJ. Bruce Fields fsnotify_link_count(target); 4164a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 41651da177e4SLinus Torvalds } 41660eeca283SRobert Love 41671da177e4SLinus Torvalds return error; 41681da177e4SLinus Torvalds } 41694d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 41701da177e4SLinus Torvalds 41711da177e4SLinus Torvalds /* 41721da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 41731b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 41741da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 41751da177e4SLinus Torvalds * while waiting on the I/O. 41761da177e4SLinus Torvalds */ 417745f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 41781da177e4SLinus Torvalds { 41792ad94ae6SAl Viro int error; 41801da177e4SLinus Torvalds struct dentry *dentry; 4181f5beed75SAl Viro struct path path; 4182f5beed75SAl Viro struct qstr last; 4183f5beed75SAl Viro int type; 41841da177e4SLinus Torvalds struct inode *inode = NULL; 4185b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 41865d18f813SJeff Layton unsigned int lookup_flags = 0; 41875d18f813SJeff Layton retry: 4188c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 41890ee50b47SDmitry Kadashev if (error) 41900ee50b47SDmitry Kadashev goto exit1; 41912ad94ae6SAl Viro 41921da177e4SLinus Torvalds error = -EISDIR; 4193f5beed75SAl Viro if (type != LAST_NORM) 41940ee50b47SDmitry Kadashev goto exit2; 41950612d9fbSOGAWA Hirofumi 4196f5beed75SAl Viro error = mnt_want_write(path.mnt); 4197c30dabfeSJan Kara if (error) 41980ee50b47SDmitry Kadashev goto exit2; 4199b21996e3SJ. Bruce Fields retry_deleg: 42005955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4201f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 42021da177e4SLinus Torvalds error = PTR_ERR(dentry); 42031da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 42046521f891SChristian Brauner struct user_namespace *mnt_userns; 42056521f891SChristian Brauner 42061da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4207f5beed75SAl Viro if (last.name[last.len]) 420850338b88STörök Edwin goto slashes; 42091da177e4SLinus Torvalds inode = dentry->d_inode; 4210b18825a7SDavid Howells if (d_is_negative(dentry)) 4211e6bc45d6STheodore Ts'o goto slashes; 42127de9c6eeSAl Viro ihold(inode); 4213f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4214be6d3e56SKentaro Takeda if (error) 42150ee50b47SDmitry Kadashev goto exit3; 42166521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4217549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4218549c7297SChristian Brauner &delegated_inode); 42190ee50b47SDmitry Kadashev exit3: 42201da177e4SLinus Torvalds dput(dentry); 42211da177e4SLinus Torvalds } 42225955102cSAl Viro inode_unlock(path.dentry->d_inode); 42231da177e4SLinus Torvalds if (inode) 42241da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4225b21996e3SJ. Bruce Fields inode = NULL; 4226b21996e3SJ. Bruce Fields if (delegated_inode) { 42275a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4228b21996e3SJ. Bruce Fields if (!error) 4229b21996e3SJ. Bruce Fields goto retry_deleg; 4230b21996e3SJ. Bruce Fields } 4231f5beed75SAl Viro mnt_drop_write(path.mnt); 42320ee50b47SDmitry Kadashev exit2: 4233f5beed75SAl Viro path_put(&path); 42345d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 42355d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 42365d18f813SJeff Layton inode = NULL; 42375d18f813SJeff Layton goto retry; 42385d18f813SJeff Layton } 42390ee50b47SDmitry Kadashev exit1: 4240da2f1362SChristoph Hellwig putname(name); 42411da177e4SLinus Torvalds return error; 42421da177e4SLinus Torvalds 42431da177e4SLinus Torvalds slashes: 4244b18825a7SDavid Howells if (d_is_negative(dentry)) 4245b18825a7SDavid Howells error = -ENOENT; 424644b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4247b18825a7SDavid Howells error = -EISDIR; 4248b18825a7SDavid Howells else 4249b18825a7SDavid Howells error = -ENOTDIR; 42500ee50b47SDmitry Kadashev goto exit3; 42511da177e4SLinus Torvalds } 42521da177e4SLinus Torvalds 42532e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 42545590ff0dSUlrich Drepper { 42555590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 42565590ff0dSUlrich Drepper return -EINVAL; 42575590ff0dSUlrich Drepper 42585590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4259e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4260da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 42615590ff0dSUlrich Drepper } 42625590ff0dSUlrich Drepper 42633480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 42645590ff0dSUlrich Drepper { 4265da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 42665590ff0dSUlrich Drepper } 42675590ff0dSUlrich Drepper 42686521f891SChristian Brauner /** 42696521f891SChristian Brauner * vfs_symlink - create symlink 42706521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 42716521f891SChristian Brauner * @dir: inode of @dentry 42726521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 42736521f891SChristian Brauner * @oldname: name of the file to link to 42746521f891SChristian Brauner * 42756521f891SChristian Brauner * Create a symlink. 42766521f891SChristian Brauner * 42776521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 42786521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42796521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42806521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42816521f891SChristian Brauner * raw inode simply passs init_user_ns. 42826521f891SChristian Brauner */ 42836521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 42846521f891SChristian Brauner struct dentry *dentry, const char *oldname) 42851da177e4SLinus Torvalds { 42866521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 42871da177e4SLinus Torvalds 42881da177e4SLinus Torvalds if (error) 42891da177e4SLinus Torvalds return error; 42901da177e4SLinus Torvalds 4291acfa4380SAl Viro if (!dir->i_op->symlink) 42921da177e4SLinus Torvalds return -EPERM; 42931da177e4SLinus Torvalds 42941da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 42951da177e4SLinus Torvalds if (error) 42961da177e4SLinus Torvalds return error; 42971da177e4SLinus Torvalds 4298549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4299a74574aaSStephen Smalley if (!error) 4300f38aa942SAmy Griffis fsnotify_create(dir, dentry); 43011da177e4SLinus Torvalds return error; 43021da177e4SLinus Torvalds } 43034d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 43041da177e4SLinus Torvalds 43057a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 43061da177e4SLinus Torvalds { 43072ad94ae6SAl Viro int error; 43086902d925SDave Hansen struct dentry *dentry; 4309dae6ad8fSAl Viro struct path path; 4310f46d3567SJeff Layton unsigned int lookup_flags = 0; 43111da177e4SLinus Torvalds 4312da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4313da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4314da2d0cedSDmitry Kadashev goto out_putnames; 4315da2d0cedSDmitry Kadashev } 4316f46d3567SJeff Layton retry: 4317b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 43181da177e4SLinus Torvalds error = PTR_ERR(dentry); 43196902d925SDave Hansen if (IS_ERR(dentry)) 4320da2d0cedSDmitry Kadashev goto out_putnames; 43216902d925SDave Hansen 432291a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 43236521f891SChristian Brauner if (!error) { 43246521f891SChristian Brauner struct user_namespace *mnt_userns; 43256521f891SChristian Brauner 43266521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 43276521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 43286521f891SChristian Brauner from->name); 43296521f891SChristian Brauner } 4330921a1650SAl Viro done_path_create(&path, dentry); 4331f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4332f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4333f46d3567SJeff Layton goto retry; 4334f46d3567SJeff Layton } 4335da2d0cedSDmitry Kadashev out_putnames: 4336da2d0cedSDmitry Kadashev putname(to); 43371da177e4SLinus Torvalds putname(from); 43381da177e4SLinus Torvalds return error; 43391da177e4SLinus Torvalds } 43401da177e4SLinus Torvalds 4341b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4342b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4343b724e846SDominik Brodowski { 4344da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4345b724e846SDominik Brodowski } 4346b724e846SDominik Brodowski 43473480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 43485590ff0dSUlrich Drepper { 4349da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 43505590ff0dSUlrich Drepper } 43515590ff0dSUlrich Drepper 4352146a8595SJ. Bruce Fields /** 4353146a8595SJ. Bruce Fields * vfs_link - create a new link 4354146a8595SJ. Bruce Fields * @old_dentry: object to be linked 43556521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4356146a8595SJ. Bruce Fields * @dir: new parent 4357146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4358146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4359146a8595SJ. Bruce Fields * 4360146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4361146a8595SJ. Bruce Fields * 4362146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4363146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4364146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4365146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4366146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4367146a8595SJ. Bruce Fields * 4368146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4369146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4370146a8595SJ. Bruce Fields * to be NFS exported. 43716521f891SChristian Brauner * 43726521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 43736521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 43746521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 43756521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 43766521f891SChristian Brauner * raw inode simply passs init_user_ns. 4377146a8595SJ. Bruce Fields */ 43786521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 43796521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 43806521f891SChristian Brauner struct inode **delegated_inode) 43811da177e4SLinus Torvalds { 43821da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 43838de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 43841da177e4SLinus Torvalds int error; 43851da177e4SLinus Torvalds 43861da177e4SLinus Torvalds if (!inode) 43871da177e4SLinus Torvalds return -ENOENT; 43881da177e4SLinus Torvalds 43896521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 43901da177e4SLinus Torvalds if (error) 43911da177e4SLinus Torvalds return error; 43921da177e4SLinus Torvalds 43931da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 43941da177e4SLinus Torvalds return -EXDEV; 43951da177e4SLinus Torvalds 43961da177e4SLinus Torvalds /* 43971da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 43981da177e4SLinus Torvalds */ 43991da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 44001da177e4SLinus Torvalds return -EPERM; 44010bd23d09SEric W. Biederman /* 44020bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 44030bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 44040bd23d09SEric W. Biederman * the vfs. 44050bd23d09SEric W. Biederman */ 44066521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 44070bd23d09SEric W. Biederman return -EPERM; 4408acfa4380SAl Viro if (!dir->i_op->link) 44091da177e4SLinus Torvalds return -EPERM; 44107e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 44111da177e4SLinus Torvalds return -EPERM; 44121da177e4SLinus Torvalds 44131da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 44141da177e4SLinus Torvalds if (error) 44151da177e4SLinus Torvalds return error; 44161da177e4SLinus Torvalds 44175955102cSAl Viro inode_lock(inode); 4418aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4419f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4420aae8a97dSAneesh Kumar K.V error = -ENOENT; 44218de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 44228de52778SAl Viro error = -EMLINK; 4423146a8595SJ. Bruce Fields else { 4424146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4425146a8595SJ. Bruce Fields if (!error) 44261da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4427146a8595SJ. Bruce Fields } 4428f4e0c30cSAl Viro 4429f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4430f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4431f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4432f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4433f4e0c30cSAl Viro } 44345955102cSAl Viro inode_unlock(inode); 4435e31e14ecSStephen Smalley if (!error) 44367e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 44371da177e4SLinus Torvalds return error; 44381da177e4SLinus Torvalds } 44394d359507SAl Viro EXPORT_SYMBOL(vfs_link); 44401da177e4SLinus Torvalds 44411da177e4SLinus Torvalds /* 44421da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 44431da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 44441da177e4SLinus Torvalds * newname. --KAB 44451da177e4SLinus Torvalds * 44461da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 44471da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 44481da177e4SLinus Torvalds * and other special files. --ADM 44491da177e4SLinus Torvalds */ 4450cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4451020250f3SDmitry Kadashev struct filename *new, int flags) 44521da177e4SLinus Torvalds { 44536521f891SChristian Brauner struct user_namespace *mnt_userns; 44541da177e4SLinus Torvalds struct dentry *new_dentry; 4455dae6ad8fSAl Viro struct path old_path, new_path; 4456146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 445711a7b371SAneesh Kumar K.V int how = 0; 44581da177e4SLinus Torvalds int error; 44591da177e4SLinus Torvalds 4460020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4461020250f3SDmitry Kadashev error = -EINVAL; 4462020250f3SDmitry Kadashev goto out_putnames; 4463020250f3SDmitry Kadashev } 446411a7b371SAneesh Kumar K.V /* 4465f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4466f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4467f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 446811a7b371SAneesh Kumar K.V */ 4469020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4470020250f3SDmitry Kadashev error = -ENOENT; 4471020250f3SDmitry Kadashev goto out_putnames; 4472f0cc6ffbSLinus Torvalds } 4473c04030e1SUlrich Drepper 447411a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 447511a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4476442e31caSJeff Layton retry: 4477794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 44781da177e4SLinus Torvalds if (error) 4479020250f3SDmitry Kadashev goto out_putnames; 44802ad94ae6SAl Viro 4481b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4482442e31caSJeff Layton (how & LOOKUP_REVAL)); 44831da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44846902d925SDave Hansen if (IS_ERR(new_dentry)) 4485020250f3SDmitry Kadashev goto out_putpath; 4486dae6ad8fSAl Viro 4487dae6ad8fSAl Viro error = -EXDEV; 4488dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4489dae6ad8fSAl Viro goto out_dput; 4490549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4491549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4492800179c9SKees Cook if (unlikely(error)) 4493800179c9SKees Cook goto out_dput; 4494dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4495be6d3e56SKentaro Takeda if (error) 4496a8104a9fSAl Viro goto out_dput; 44976521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 44986521f891SChristian Brauner new_dentry, &delegated_inode); 449975c3f29dSDave Hansen out_dput: 4500921a1650SAl Viro done_path_create(&new_path, new_dentry); 4501146a8595SJ. Bruce Fields if (delegated_inode) { 4502146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4503d22e6338SOleg Drokin if (!error) { 4504d22e6338SOleg Drokin path_put(&old_path); 4505146a8595SJ. Bruce Fields goto retry; 4506146a8595SJ. Bruce Fields } 4507d22e6338SOleg Drokin } 4508442e31caSJeff Layton if (retry_estale(error, how)) { 4509d22e6338SOleg Drokin path_put(&old_path); 4510442e31caSJeff Layton how |= LOOKUP_REVAL; 4511442e31caSJeff Layton goto retry; 4512442e31caSJeff Layton } 4513020250f3SDmitry Kadashev out_putpath: 45142d8f3038SAl Viro path_put(&old_path); 4515020250f3SDmitry Kadashev out_putnames: 4516020250f3SDmitry Kadashev putname(old); 4517020250f3SDmitry Kadashev putname(new); 45181da177e4SLinus Torvalds 45191da177e4SLinus Torvalds return error; 45201da177e4SLinus Torvalds } 45211da177e4SLinus Torvalds 452246ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 452346ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 452446ea89ebSDominik Brodowski { 4525020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4526020250f3SDmitry Kadashev newdfd, getname(newname), flags); 452746ea89ebSDominik Brodowski } 452846ea89ebSDominik Brodowski 45293480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 45305590ff0dSUlrich Drepper { 4531020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 45325590ff0dSUlrich Drepper } 45335590ff0dSUlrich Drepper 4534bc27027aSMiklos Szeredi /** 4535bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 45362111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4537bc27027aSMiklos Szeredi * 4538bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4539bc27027aSMiklos Szeredi * 4540bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4541bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4542bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4543bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4544bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4545bc27027aSMiklos Szeredi * so. 4546bc27027aSMiklos Szeredi * 4547bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4548bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4549bc27027aSMiklos Szeredi * to be NFS exported. 4550bc27027aSMiklos Szeredi * 45511da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 45521da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 45531da177e4SLinus Torvalds * Problems: 45540117d427SMauro Carvalho Chehab * 4555d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 45561da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 45571da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4558a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 45591da177e4SLinus Torvalds * story. 45606cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 45616cedba89SJ. Bruce Fields * and source (if it is not a directory). 45621b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 45631da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 45641da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4565a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 45661da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 45671da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 45681da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4569a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 45701da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 45711da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 45721da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4573e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 45741b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 45751da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4576c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 45771da177e4SLinus Torvalds * locking]. 45781da177e4SLinus Torvalds */ 45799fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 45801da177e4SLinus Torvalds { 45811da177e4SLinus Torvalds int error; 45829fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 45839fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 45849fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 45859fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 45869fe61450SChristian Brauner unsigned int flags = rd->flags; 4587bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4588bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4589bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4590da1ce067SMiklos Szeredi bool new_is_dir = false; 4591da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 459249d31c2fSAl Viro struct name_snapshot old_name; 45931da177e4SLinus Torvalds 45948d3e2936SMiklos Szeredi if (source == target) 45951da177e4SLinus Torvalds return 0; 45961da177e4SLinus Torvalds 45976521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 45981da177e4SLinus Torvalds if (error) 45991da177e4SLinus Torvalds return error; 46001da177e4SLinus Torvalds 4601da1ce067SMiklos Szeredi if (!target) { 46026521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4603da1ce067SMiklos Szeredi } else { 4604da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4605da1ce067SMiklos Szeredi 4606da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46076521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 46086521f891SChristian Brauner new_dentry, is_dir); 4609da1ce067SMiklos Szeredi else 46106521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 46116521f891SChristian Brauner new_dentry, new_is_dir); 4612da1ce067SMiklos Szeredi } 46131da177e4SLinus Torvalds if (error) 46141da177e4SLinus Torvalds return error; 46151da177e4SLinus Torvalds 46162773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 46171da177e4SLinus Torvalds return -EPERM; 46181da177e4SLinus Torvalds 4619bc27027aSMiklos Szeredi /* 4620bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4621bc27027aSMiklos Szeredi * we'll need to flip '..'. 4622bc27027aSMiklos Szeredi */ 4623da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4624da1ce067SMiklos Szeredi if (is_dir) { 46256521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 462647291baaSChristian Brauner MAY_WRITE); 4627bc27027aSMiklos Szeredi if (error) 4628bc27027aSMiklos Szeredi return error; 4629bc27027aSMiklos Szeredi } 4630da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 46316521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 463247291baaSChristian Brauner MAY_WRITE); 4633da1ce067SMiklos Szeredi if (error) 4634da1ce067SMiklos Szeredi return error; 4635da1ce067SMiklos Szeredi } 4636da1ce067SMiklos Szeredi } 46370eeca283SRobert Love 46380b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 46390b3974ebSMiklos Szeredi flags); 4640bc27027aSMiklos Szeredi if (error) 4641bc27027aSMiklos Szeredi return error; 4642bc27027aSMiklos Szeredi 464349d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4644bc27027aSMiklos Szeredi dget(new_dentry); 4645da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4646bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4647bc27027aSMiklos Szeredi else if (target) 46485955102cSAl Viro inode_lock(target); 4649bc27027aSMiklos Szeredi 465051cc3a66SHugh Dickins error = -EPERM; 465151cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 465251cc3a66SHugh Dickins goto out; 465351cc3a66SHugh Dickins 4654bc27027aSMiklos Szeredi error = -EBUSY; 46557af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4656bc27027aSMiklos Szeredi goto out; 4657bc27027aSMiklos Szeredi 4658da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4659bc27027aSMiklos Szeredi error = -EMLINK; 4660da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4661bc27027aSMiklos Szeredi goto out; 4662da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4663da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4664da1ce067SMiklos Szeredi goto out; 4665da1ce067SMiklos Szeredi } 4666da1ce067SMiklos Szeredi if (!is_dir) { 4667bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4668bc27027aSMiklos Szeredi if (error) 4669bc27027aSMiklos Szeredi goto out; 4670da1ce067SMiklos Szeredi } 4671da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4672bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4673bc27027aSMiklos Szeredi if (error) 4674bc27027aSMiklos Szeredi goto out; 4675bc27027aSMiklos Szeredi } 4676549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4677520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4678bc27027aSMiklos Szeredi if (error) 4679bc27027aSMiklos Szeredi goto out; 4680bc27027aSMiklos Szeredi 4681da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 46828767712fSAl Viro if (is_dir) { 46838767712fSAl Viro shrink_dcache_parent(new_dentry); 4684bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 46858767712fSAl Viro } 4686bc27027aSMiklos Szeredi dont_mount(new_dentry); 46878ed936b5SEric W. Biederman detach_mounts(new_dentry); 4688bc27027aSMiklos Szeredi } 4689da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4690da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4691bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4692da1ce067SMiklos Szeredi else 4693da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4694da1ce067SMiklos Szeredi } 4695bc27027aSMiklos Szeredi out: 4696da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4697bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4698bc27027aSMiklos Szeredi else if (target) 46995955102cSAl Viro inode_unlock(target); 4700bc27027aSMiklos Szeredi dput(new_dentry); 4701da1ce067SMiklos Szeredi if (!error) { 4702f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4703da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4704da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4705f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4706da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4707da1ce067SMiklos Szeredi } 4708da1ce067SMiklos Szeredi } 470949d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 47100eeca283SRobert Love 47111da177e4SLinus Torvalds return error; 47121da177e4SLinus Torvalds } 47134d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 47141da177e4SLinus Torvalds 4715e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4716e886663cSJens Axboe struct filename *to, unsigned int flags) 47171da177e4SLinus Torvalds { 47189fe61450SChristian Brauner struct renamedata rd; 47191da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 47201da177e4SLinus Torvalds struct dentry *trap; 4721f5beed75SAl Viro struct path old_path, new_path; 4722f5beed75SAl Viro struct qstr old_last, new_last; 4723f5beed75SAl Viro int old_type, new_type; 47248e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4725f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4726c6a94284SJeff Layton bool should_retry = false; 4727e886663cSJens Axboe int error = -EINVAL; 4728520c8b16SMiklos Szeredi 47290d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 47300ee50b47SDmitry Kadashev goto put_names; 4731da1ce067SMiklos Szeredi 47320d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 47330d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 47340ee50b47SDmitry Kadashev goto put_names; 4735520c8b16SMiklos Szeredi 4736f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4737f5beed75SAl Viro target_flags = 0; 4738f5beed75SAl Viro 4739c6a94284SJeff Layton retry: 4740c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4741e886663cSJens Axboe &old_last, &old_type); 47420ee50b47SDmitry Kadashev if (error) 47430ee50b47SDmitry Kadashev goto put_names; 47441da177e4SLinus Torvalds 4745c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4746e886663cSJens Axboe &new_type); 47470ee50b47SDmitry Kadashev if (error) 47481da177e4SLinus Torvalds goto exit1; 47491da177e4SLinus Torvalds 47501da177e4SLinus Torvalds error = -EXDEV; 4751f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 47521da177e4SLinus Torvalds goto exit2; 47531da177e4SLinus Torvalds 47541da177e4SLinus Torvalds error = -EBUSY; 4755f5beed75SAl Viro if (old_type != LAST_NORM) 47561da177e4SLinus Torvalds goto exit2; 47571da177e4SLinus Torvalds 47580a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 47590a7c3937SMiklos Szeredi error = -EEXIST; 4760f5beed75SAl Viro if (new_type != LAST_NORM) 47611da177e4SLinus Torvalds goto exit2; 47621da177e4SLinus Torvalds 4763f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4764c30dabfeSJan Kara if (error) 4765c30dabfeSJan Kara goto exit2; 4766c30dabfeSJan Kara 47678e6d782cSJ. Bruce Fields retry_deleg: 4768f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 47691da177e4SLinus Torvalds 4770f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 47711da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 47721da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 47731da177e4SLinus Torvalds goto exit3; 47741da177e4SLinus Torvalds /* source must exist */ 47751da177e4SLinus Torvalds error = -ENOENT; 4776b18825a7SDavid Howells if (d_is_negative(old_dentry)) 47771da177e4SLinus Torvalds goto exit4; 4778f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 47791da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 47801da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 47811da177e4SLinus Torvalds goto exit4; 47820a7c3937SMiklos Szeredi error = -EEXIST; 47830a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 47840a7c3937SMiklos Szeredi goto exit5; 4785da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4786da1ce067SMiklos Szeredi error = -ENOENT; 4787da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4788da1ce067SMiklos Szeredi goto exit5; 4789da1ce067SMiklos Szeredi 4790da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4791da1ce067SMiklos Szeredi error = -ENOTDIR; 4792f5beed75SAl Viro if (new_last.name[new_last.len]) 4793da1ce067SMiklos Szeredi goto exit5; 4794da1ce067SMiklos Szeredi } 4795da1ce067SMiklos Szeredi } 47960a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 47970a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 47980a7c3937SMiklos Szeredi error = -ENOTDIR; 4799f5beed75SAl Viro if (old_last.name[old_last.len]) 48000a7c3937SMiklos Szeredi goto exit5; 4801f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 48020a7c3937SMiklos Szeredi goto exit5; 48030a7c3937SMiklos Szeredi } 48040a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 48050a7c3937SMiklos Szeredi error = -EINVAL; 48060a7c3937SMiklos Szeredi if (old_dentry == trap) 48070a7c3937SMiklos Szeredi goto exit5; 48081da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4809da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 48101da177e4SLinus Torvalds error = -ENOTEMPTY; 48111da177e4SLinus Torvalds if (new_dentry == trap) 48121da177e4SLinus Torvalds goto exit5; 48131da177e4SLinus Torvalds 4814f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4815f5beed75SAl Viro &new_path, new_dentry, flags); 4816be6d3e56SKentaro Takeda if (error) 4817c30dabfeSJan Kara goto exit5; 48189fe61450SChristian Brauner 48199fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 48209fe61450SChristian Brauner rd.old_dentry = old_dentry; 48216521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 48229fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 48239fe61450SChristian Brauner rd.new_dentry = new_dentry; 48246521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 48259fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 48269fe61450SChristian Brauner rd.flags = flags; 48279fe61450SChristian Brauner error = vfs_rename(&rd); 48281da177e4SLinus Torvalds exit5: 48291da177e4SLinus Torvalds dput(new_dentry); 48301da177e4SLinus Torvalds exit4: 48311da177e4SLinus Torvalds dput(old_dentry); 48321da177e4SLinus Torvalds exit3: 4833f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 48348e6d782cSJ. Bruce Fields if (delegated_inode) { 48358e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 48368e6d782cSJ. Bruce Fields if (!error) 48378e6d782cSJ. Bruce Fields goto retry_deleg; 48388e6d782cSJ. Bruce Fields } 4839f5beed75SAl Viro mnt_drop_write(old_path.mnt); 48401da177e4SLinus Torvalds exit2: 4841c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4842c6a94284SJeff Layton should_retry = true; 4843f5beed75SAl Viro path_put(&new_path); 48441da177e4SLinus Torvalds exit1: 4845f5beed75SAl Viro path_put(&old_path); 4846c6a94284SJeff Layton if (should_retry) { 4847c6a94284SJeff Layton should_retry = false; 4848c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4849c6a94284SJeff Layton goto retry; 4850c6a94284SJeff Layton } 48510ee50b47SDmitry Kadashev put_names: 4852e886663cSJens Axboe putname(from); 4853e886663cSJens Axboe putname(to); 48541da177e4SLinus Torvalds return error; 48551da177e4SLinus Torvalds } 48561da177e4SLinus Torvalds 4857ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4858ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4859ee81feb6SDominik Brodowski { 4860e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4861e886663cSJens Axboe flags); 4862ee81feb6SDominik Brodowski } 4863ee81feb6SDominik Brodowski 4864520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4865520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4866520c8b16SMiklos Szeredi { 4867e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4868e886663cSJens Axboe 0); 4869520c8b16SMiklos Szeredi } 4870520c8b16SMiklos Szeredi 4871a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 48725590ff0dSUlrich Drepper { 4873e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4874e886663cSJens Axboe getname(newname), 0); 48755590ff0dSUlrich Drepper } 48765590ff0dSUlrich Drepper 48775d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 48781da177e4SLinus Torvalds { 48795d826c84SAl Viro int len = PTR_ERR(link); 48801da177e4SLinus Torvalds if (IS_ERR(link)) 48811da177e4SLinus Torvalds goto out; 48821da177e4SLinus Torvalds 48831da177e4SLinus Torvalds len = strlen(link); 48841da177e4SLinus Torvalds if (len > (unsigned) buflen) 48851da177e4SLinus Torvalds len = buflen; 48861da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 48871da177e4SLinus Torvalds len = -EFAULT; 48881da177e4SLinus Torvalds out: 48891da177e4SLinus Torvalds return len; 48901da177e4SLinus Torvalds } 48911da177e4SLinus Torvalds 4892d60874cdSMiklos Szeredi /** 4893fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4894fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4895fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4896fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4897fd4a0edfSMiklos Szeredi * 4898fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4899fd4a0edfSMiklos Szeredi * 4900fd4a0edfSMiklos Szeredi * Does not call security hook. 4901fd4a0edfSMiklos Szeredi */ 4902fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4903fd4a0edfSMiklos Szeredi { 4904fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4905f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4906f2df5da6SAl Viro const char *link; 4907f2df5da6SAl Viro int res; 4908fd4a0edfSMiklos Szeredi 490976fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 491076fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 491176fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 491276fca90eSMiklos Szeredi 491376fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4914fd4a0edfSMiklos Szeredi return -EINVAL; 4915fd4a0edfSMiklos Szeredi 491676fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 491776fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 491876fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 491976fca90eSMiklos Szeredi } 492076fca90eSMiklos Szeredi 49214c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4922f2df5da6SAl Viro if (!link) { 4923f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4924f2df5da6SAl Viro if (IS_ERR(link)) 4925f2df5da6SAl Viro return PTR_ERR(link); 4926f2df5da6SAl Viro } 4927f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4928f2df5da6SAl Viro do_delayed_call(&done); 4929f2df5da6SAl Viro return res; 4930fd4a0edfSMiklos Szeredi } 4931fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 49321da177e4SLinus Torvalds 4933d60874cdSMiklos Szeredi /** 4934d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4935d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4936d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4937d60874cdSMiklos Szeredi * 4938d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4939d60874cdSMiklos Szeredi * 4940d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4941d60874cdSMiklos Szeredi * 4942d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4943d60874cdSMiklos Szeredi */ 4944d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4945d60874cdSMiklos Szeredi { 4946d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4947d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4948d60874cdSMiklos Szeredi 4949d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4950d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4951d60874cdSMiklos Szeredi if (!res) 4952d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4953d60874cdSMiklos Szeredi } 4954d60874cdSMiklos Szeredi return res; 4955d60874cdSMiklos Szeredi } 4956d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4957d60874cdSMiklos Szeredi 49581da177e4SLinus Torvalds /* get the link contents into pagecache */ 49596b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4960fceef393SAl Viro struct delayed_call *callback) 49611da177e4SLinus Torvalds { 4962ebd09abbSDuane Griffin char *kaddr; 49631da177e4SLinus Torvalds struct page *page; 49646b255391SAl Viro struct address_space *mapping = inode->i_mapping; 49656b255391SAl Viro 4966d3883d4fSAl Viro if (!dentry) { 4967d3883d4fSAl Viro page = find_get_page(mapping, 0); 4968d3883d4fSAl Viro if (!page) 49696b255391SAl Viro return ERR_PTR(-ECHILD); 4970d3883d4fSAl Viro if (!PageUptodate(page)) { 4971d3883d4fSAl Viro put_page(page); 4972d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4973d3883d4fSAl Viro } 4974d3883d4fSAl Viro } else { 4975090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 49761da177e4SLinus Torvalds if (IS_ERR(page)) 49776fe6900eSNick Piggin return (char*)page; 4978d3883d4fSAl Viro } 4979fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 498021fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 498121fc61c7SAl Viro kaddr = page_address(page); 49826b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4983ebd09abbSDuane Griffin return kaddr; 49841da177e4SLinus Torvalds } 49851da177e4SLinus Torvalds 49866b255391SAl Viro EXPORT_SYMBOL(page_get_link); 49871da177e4SLinus Torvalds 4988fceef393SAl Viro void page_put_link(void *arg) 49891da177e4SLinus Torvalds { 4990fceef393SAl Viro put_page(arg); 49911da177e4SLinus Torvalds } 49924d359507SAl Viro EXPORT_SYMBOL(page_put_link); 49931da177e4SLinus Torvalds 4994aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4995aa80deabSAl Viro { 4996fceef393SAl Viro DEFINE_DELAYED_CALL(done); 49976b255391SAl Viro int res = readlink_copy(buffer, buflen, 49986b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4999fceef393SAl Viro &done)); 5000fceef393SAl Viro do_delayed_call(&done); 5001aa80deabSAl Viro return res; 5002aa80deabSAl Viro } 5003aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 5004aa80deabSAl Viro 500556f5746cSMatthew Wilcox (Oracle) int page_symlink(struct inode *inode, const char *symname, int len) 50061da177e4SLinus Torvalds { 50071da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 500856f5746cSMatthew Wilcox (Oracle) bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS); 50090adb25d2SKirill Korotaev struct page *page; 5010afddba49SNick Piggin void *fsdata; 5011beb497abSDmitriy Monakhov int err; 5012*2d878178SMatthew Wilcox (Oracle) unsigned int flags; 50131da177e4SLinus Torvalds 50147e53cac4SNeilBrown retry: 5015*2d878178SMatthew Wilcox (Oracle) if (nofs) 5016*2d878178SMatthew Wilcox (Oracle) flags = memalloc_nofs_save(); 5017afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 5018*2d878178SMatthew Wilcox (Oracle) 0, &page, &fsdata); 5019*2d878178SMatthew Wilcox (Oracle) if (nofs) 5020*2d878178SMatthew Wilcox (Oracle) memalloc_nofs_restore(flags); 50211da177e4SLinus Torvalds if (err) 5022afddba49SNick Piggin goto fail; 5023afddba49SNick Piggin 502421fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5025afddba49SNick Piggin 5026afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 5027afddba49SNick Piggin page, fsdata); 50281da177e4SLinus Torvalds if (err < 0) 50291da177e4SLinus Torvalds goto fail; 5030afddba49SNick Piggin if (err < len-1) 5031afddba49SNick Piggin goto retry; 5032afddba49SNick Piggin 50331da177e4SLinus Torvalds mark_inode_dirty(inode); 50341da177e4SLinus Torvalds return 0; 50351da177e4SLinus Torvalds fail: 50361da177e4SLinus Torvalds return err; 50371da177e4SLinus Torvalds } 50384d359507SAl Viro EXPORT_SYMBOL(page_symlink); 50390adb25d2SKirill Korotaev 504092e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 50416b255391SAl Viro .get_link = page_get_link, 50421da177e4SLinus Torvalds }; 50431da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5044