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> 23*5970e15dSJeff Layton #include <linux/filelock.h> 241da177e4SLinus Torvalds #include <linux/namei.h> 251da177e4SLinus Torvalds #include <linux/pagemap.h> 262d878178SMatthew Wilcox (Oracle) #include <linux/sched/mm.h> 270eeca283SRobert Love #include <linux/fsnotify.h> 281da177e4SLinus Torvalds #include <linux/personality.h> 291da177e4SLinus Torvalds #include <linux/security.h> 306146f0d5SMimi Zohar #include <linux/ima.h> 311da177e4SLinus Torvalds #include <linux/syscalls.h> 321da177e4SLinus Torvalds #include <linux/mount.h> 331da177e4SLinus Torvalds #include <linux/audit.h> 3416f7e0feSRandy Dunlap #include <linux/capability.h> 35834f2a4aSTrond Myklebust #include <linux/file.h> 365590ff0dSUlrich Drepper #include <linux/fcntl.h> 3708ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 385ad4e53bSAl Viro #include <linux/fs_struct.h> 39e77819e5SLinus Torvalds #include <linux/posix_acl.h> 4099d263d4SLinus Torvalds #include <linux/hash.h> 412a18da7aSGeorge Spelvin #include <linux/bitops.h> 42aeaa4a79SEric W. Biederman #include <linux/init_task.h> 437c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 441da177e4SLinus Torvalds 45e81e3f4dSEric Paris #include "internal.h" 46c7105365SAl Viro #include "mount.h" 47e81e3f4dSEric Paris 481da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 491da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 501da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 511da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 521da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 531da177e4SLinus Torvalds * 541da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 551da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 561da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 571da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 581da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 591da177e4SLinus Torvalds * the special cases of the former code. 601da177e4SLinus Torvalds * 611da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 621da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 631da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 641da177e4SLinus Torvalds * 651da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 661da177e4SLinus Torvalds * resolution to correspond with current state of the code. 671da177e4SLinus Torvalds * 681da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 691da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 701da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 711da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 721da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 731da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 741da177e4SLinus Torvalds */ 751da177e4SLinus Torvalds 761da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 771da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 781da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 791da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 801da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 811da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8225985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 831da177e4SLinus Torvalds * 841da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 851da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 861da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 871da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 881da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 891da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 901da177e4SLinus Torvalds * and in the old Linux semantics. 911da177e4SLinus Torvalds */ 921da177e4SLinus Torvalds 931da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 941da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 951da177e4SLinus Torvalds * 961da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 971da177e4SLinus Torvalds */ 981da177e4SLinus Torvalds 991da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 1001da177e4SLinus Torvalds * inside the path - always follow. 1011da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1021da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1031da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1041da177e4SLinus Torvalds * otherwise - don't follow. 1051da177e4SLinus Torvalds * (applied in that order). 1061da177e4SLinus Torvalds * 1071da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1081da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1091da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1101da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1111da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1121da177e4SLinus Torvalds */ 1131da177e4SLinus Torvalds /* 1141da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 115a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1161da177e4SLinus Torvalds * any extra contention... 1171da177e4SLinus Torvalds */ 1181da177e4SLinus Torvalds 1191da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1201da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1211da177e4SLinus Torvalds * kernel data space before using them.. 1221da177e4SLinus Torvalds * 1231da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1241da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1251da177e4SLinus Torvalds */ 1267950e385SJeff Layton 127fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12891a27b2aSJeff Layton 12951f39a1fSDavid Drysdale struct filename * 13091a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13191a27b2aSJeff Layton { 13294b5d262SAl Viro struct filename *result; 1337950e385SJeff Layton char *kname; 13494b5d262SAl Viro int len; 1351da177e4SLinus Torvalds 1367ac86265SJeff Layton result = audit_reusename(filename); 1377ac86265SJeff Layton if (result) 1387ac86265SJeff Layton return result; 1397ac86265SJeff Layton 1407950e385SJeff Layton result = __getname(); 1413f9f0aa6SLinus Torvalds if (unlikely(!result)) 1424043cde8SEric Paris return ERR_PTR(-ENOMEM); 1431da177e4SLinus Torvalds 1447950e385SJeff Layton /* 1457950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1467950e385SJeff Layton * allocation 1477950e385SJeff Layton */ 148fd2f7cb5SAl Viro kname = (char *)result->iname; 14991a27b2aSJeff Layton result->name = kname; 1507950e385SJeff Layton 15194b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15291a27b2aSJeff Layton if (unlikely(len < 0)) { 15394b5d262SAl Viro __putname(result); 15494b5d262SAl Viro return ERR_PTR(len); 15591a27b2aSJeff Layton } 1563f9f0aa6SLinus Torvalds 1577950e385SJeff Layton /* 1587950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1597950e385SJeff Layton * separate struct filename so we can dedicate the entire 1607950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1617950e385SJeff Layton * userland. 1627950e385SJeff Layton */ 16394b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 164fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1657950e385SJeff Layton kname = (char *)result; 1667950e385SJeff Layton 167fd2f7cb5SAl Viro /* 168fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 169fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 170fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 171fd2f7cb5SAl Viro */ 172fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17394b5d262SAl Viro if (unlikely(!result)) { 17494b5d262SAl Viro __putname(kname); 17594b5d262SAl Viro return ERR_PTR(-ENOMEM); 1767950e385SJeff Layton } 1777950e385SJeff Layton result->name = kname; 17894b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17994b5d262SAl Viro if (unlikely(len < 0)) { 18094b5d262SAl Viro __putname(kname); 18194b5d262SAl Viro kfree(result); 18294b5d262SAl Viro return ERR_PTR(len); 18394b5d262SAl Viro } 18494b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18594b5d262SAl Viro __putname(kname); 18694b5d262SAl Viro kfree(result); 18794b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18894b5d262SAl Viro } 1897950e385SJeff Layton } 1907950e385SJeff Layton 19194b5d262SAl Viro result->refcnt = 1; 1923f9f0aa6SLinus Torvalds /* The empty path is special. */ 1933f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1943f9f0aa6SLinus Torvalds if (empty) 1951fa1e7f6SAndy Whitcroft *empty = 1; 19694b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19794b5d262SAl Viro putname(result); 19894b5d262SAl Viro return ERR_PTR(-ENOENT); 1991da177e4SLinus Torvalds } 20094b5d262SAl Viro } 2017950e385SJeff Layton 2027950e385SJeff Layton result->uptr = filename; 203c4ad8f98SLinus Torvalds result->aname = NULL; 2041da177e4SLinus Torvalds audit_getname(result); 2051da177e4SLinus Torvalds return result; 2063f9f0aa6SLinus Torvalds } 2073f9f0aa6SLinus Torvalds 20891a27b2aSJeff Layton struct filename * 2098228e2c3SDmitry Kadashev getname_uflags(const char __user *filename, int uflags) 2108228e2c3SDmitry Kadashev { 2118228e2c3SDmitry Kadashev int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 2128228e2c3SDmitry Kadashev 2138228e2c3SDmitry Kadashev return getname_flags(filename, flags, NULL); 2148228e2c3SDmitry Kadashev } 2158228e2c3SDmitry Kadashev 2168228e2c3SDmitry Kadashev struct filename * 21791a27b2aSJeff Layton getname(const char __user * filename) 218f52e0c11SAl Viro { 219f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 220f52e0c11SAl Viro } 221f52e0c11SAl Viro 222c4ad8f98SLinus Torvalds struct filename * 223c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 224c4ad8f98SLinus Torvalds { 225c4ad8f98SLinus Torvalds struct filename *result; 22608518549SPaul Moore int len = strlen(filename) + 1; 227c4ad8f98SLinus Torvalds 228c4ad8f98SLinus Torvalds result = __getname(); 229c4ad8f98SLinus Torvalds if (unlikely(!result)) 230c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 231c4ad8f98SLinus Torvalds 23208518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 233fd2f7cb5SAl Viro result->name = (char *)result->iname; 23408518549SPaul Moore } else if (len <= PATH_MAX) { 23530ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 23608518549SPaul Moore struct filename *tmp; 23708518549SPaul Moore 23830ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 23908518549SPaul Moore if (unlikely(!tmp)) { 24008518549SPaul Moore __putname(result); 24108518549SPaul Moore return ERR_PTR(-ENOMEM); 24208518549SPaul Moore } 24308518549SPaul Moore tmp->name = (char *)result; 24408518549SPaul Moore result = tmp; 24508518549SPaul Moore } else { 24608518549SPaul Moore __putname(result); 24708518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 24808518549SPaul Moore } 24908518549SPaul Moore memcpy((char *)result->name, filename, len); 250c4ad8f98SLinus Torvalds result->uptr = NULL; 251c4ad8f98SLinus Torvalds result->aname = NULL; 25255422d0bSPaul Moore result->refcnt = 1; 253fd3522fdSPaul Moore audit_getname(result); 254c4ad8f98SLinus Torvalds 255c4ad8f98SLinus Torvalds return result; 256c4ad8f98SLinus Torvalds } 257c4ad8f98SLinus Torvalds 25891a27b2aSJeff Layton void putname(struct filename *name) 2591da177e4SLinus Torvalds { 260ea47ab11SAl Viro if (IS_ERR(name)) 26191ef658fSDmitry Kadashev return; 26291ef658fSDmitry Kadashev 26355422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 26455422d0bSPaul Moore 26555422d0bSPaul Moore if (--name->refcnt > 0) 26655422d0bSPaul Moore return; 26755422d0bSPaul Moore 268fd2f7cb5SAl Viro if (name->name != name->iname) { 26955422d0bSPaul Moore __putname(name->name); 27055422d0bSPaul Moore kfree(name); 27155422d0bSPaul Moore } else 27255422d0bSPaul Moore __putname(name); 2731da177e4SLinus Torvalds } 2741da177e4SLinus Torvalds 27547291baaSChristian Brauner /** 27647291baaSChristian Brauner * check_acl - perform ACL permission checking 27747291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 27847291baaSChristian Brauner * @inode: inode to check permissions on 27947291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 28047291baaSChristian Brauner * 28147291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 28247291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 28347291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 28447291baaSChristian Brauner * 28547291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 28647291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 28747291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 28847291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 28947291baaSChristian Brauner * raw inode simply passs init_user_ns. 29047291baaSChristian Brauner */ 29147291baaSChristian Brauner static int check_acl(struct user_namespace *mnt_userns, 29247291baaSChristian Brauner struct inode *inode, int mask) 293e77819e5SLinus Torvalds { 29484635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 295e77819e5SLinus Torvalds struct posix_acl *acl; 296e77819e5SLinus Torvalds 297e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2983567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2993567866bSAl Viro if (!acl) 300e77819e5SLinus Torvalds return -EAGAIN; 301cac2f8b8SChristian Brauner /* no ->get_inode_acl() calls in RCU mode... */ 302b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 303e77819e5SLinus Torvalds return -ECHILD; 30447291baaSChristian Brauner return posix_acl_permission(mnt_userns, inode, acl, mask); 305e77819e5SLinus Torvalds } 306e77819e5SLinus Torvalds 307cac2f8b8SChristian Brauner acl = get_inode_acl(inode, ACL_TYPE_ACCESS); 3084e34e719SChristoph Hellwig if (IS_ERR(acl)) 3094e34e719SChristoph Hellwig return PTR_ERR(acl); 310e77819e5SLinus Torvalds if (acl) { 31147291baaSChristian Brauner int error = posix_acl_permission(mnt_userns, inode, acl, mask); 312e77819e5SLinus Torvalds posix_acl_release(acl); 313e77819e5SLinus Torvalds return error; 314e77819e5SLinus Torvalds } 31584635d68SLinus Torvalds #endif 316e77819e5SLinus Torvalds 317e77819e5SLinus Torvalds return -EAGAIN; 318e77819e5SLinus Torvalds } 319e77819e5SLinus Torvalds 32047291baaSChristian Brauner /** 32147291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 32247291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 32347291baaSChristian Brauner * @inode: inode to check permissions on 32447291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3255fc475b7SLinus Torvalds * 32647291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 32747291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 32847291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 32947291baaSChristian Brauner * 33047291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 33147291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 33247291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 33347291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 33447291baaSChristian Brauner * raw inode simply passs init_user_ns. 3355909ccaaSLinus Torvalds */ 33647291baaSChristian Brauner static int acl_permission_check(struct user_namespace *mnt_userns, 33747291baaSChristian Brauner struct inode *inode, int mask) 3385909ccaaSLinus Torvalds { 33926cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 340a2bd096fSChristian Brauner vfsuid_t vfsuid; 3415909ccaaSLinus Torvalds 3425fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 343a2bd096fSChristian Brauner vfsuid = i_uid_into_vfsuid(mnt_userns, inode); 344a2bd096fSChristian Brauner if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) { 3455fc475b7SLinus Torvalds mask &= 7; 3465909ccaaSLinus Torvalds mode >>= 6; 3475fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3485fc475b7SLinus Torvalds } 3495fc475b7SLinus Torvalds 3505fc475b7SLinus Torvalds /* Do we have ACL's? */ 351e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 35247291baaSChristian Brauner int error = check_acl(mnt_userns, inode, mask); 3535909ccaaSLinus Torvalds if (error != -EAGAIN) 3545909ccaaSLinus Torvalds return error; 3555909ccaaSLinus Torvalds } 3565909ccaaSLinus Torvalds 3575fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3585fc475b7SLinus Torvalds mask &= 7; 3595fc475b7SLinus Torvalds 3605fc475b7SLinus Torvalds /* 3615fc475b7SLinus Torvalds * Are the group permissions different from 3625fc475b7SLinus Torvalds * the other permissions in the bits we care 3635fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3645fc475b7SLinus Torvalds */ 3655fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 366a2bd096fSChristian Brauner vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode); 367a2bd096fSChristian Brauner if (vfsgid_in_group_p(vfsgid)) 3685909ccaaSLinus Torvalds mode >>= 3; 3695909ccaaSLinus Torvalds } 3705909ccaaSLinus Torvalds 3715fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3725fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3735909ccaaSLinus Torvalds } 3741da177e4SLinus Torvalds 3751da177e4SLinus Torvalds /** 3761da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 37747291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 3781da177e4SLinus Torvalds * @inode: inode to check access rights for 3795fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3805fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3811da177e4SLinus Torvalds * 3821da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3831da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3841da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 385b74c79e9SNick Piggin * are used for other things. 386b74c79e9SNick Piggin * 387b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 388b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 389b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 39047291baaSChristian Brauner * 39147291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39247291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39347291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39447291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39547291baaSChristian Brauner * raw inode simply passs init_user_ns. 3961da177e4SLinus Torvalds */ 39747291baaSChristian Brauner int generic_permission(struct user_namespace *mnt_userns, struct inode *inode, 39847291baaSChristian Brauner int mask) 3991da177e4SLinus Torvalds { 4005909ccaaSLinus Torvalds int ret; 4011da177e4SLinus Torvalds 4021da177e4SLinus Torvalds /* 403948409c7SAndreas Gruenbacher * Do the basic permission checks. 4041da177e4SLinus Torvalds */ 40547291baaSChristian Brauner ret = acl_permission_check(mnt_userns, inode, mask); 4065909ccaaSLinus Torvalds if (ret != -EACCES) 4075909ccaaSLinus Torvalds return ret; 4081da177e4SLinus Torvalds 409d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 410d594e7ecSAl Viro /* DACs are overridable for directories */ 411d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 41247291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 41323adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 414d594e7ecSAl Viro return 0; 41547291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4160558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4171da177e4SLinus Torvalds return 0; 4182a4c2242SStephen Smalley return -EACCES; 4192a4c2242SStephen Smalley } 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds /* 4221da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4231da177e4SLinus Torvalds */ 4247ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 425d594e7ecSAl Viro if (mask == MAY_READ) 42647291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4270558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4281da177e4SLinus Torvalds return 0; 4292a4c2242SStephen Smalley /* 4302a4c2242SStephen Smalley * Read/write DACs are always overridable. 4312a4c2242SStephen Smalley * Executable DACs are overridable when there is 4322a4c2242SStephen Smalley * at least one exec bit set. 4332a4c2242SStephen Smalley */ 4342a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 43547291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4360558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4372a4c2242SStephen Smalley return 0; 4381da177e4SLinus Torvalds 4391da177e4SLinus Torvalds return -EACCES; 4401da177e4SLinus Torvalds } 4414d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4421da177e4SLinus Torvalds 44347291baaSChristian Brauner /** 44447291baaSChristian Brauner * do_inode_permission - UNIX permission checking 44547291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 44647291baaSChristian Brauner * @inode: inode to check permissions on 44747291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 44847291baaSChristian Brauner * 4493ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4503ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4513ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4523ddcd056SLinus Torvalds * permission function, use the fast case". 4533ddcd056SLinus Torvalds */ 45447291baaSChristian Brauner static inline int do_inode_permission(struct user_namespace *mnt_userns, 45547291baaSChristian Brauner struct inode *inode, int mask) 4563ddcd056SLinus Torvalds { 4573ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4583ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 459549c7297SChristian Brauner return inode->i_op->permission(mnt_userns, inode, mask); 4603ddcd056SLinus Torvalds 4613ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4623ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4633ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4643ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4653ddcd056SLinus Torvalds } 46647291baaSChristian Brauner return generic_permission(mnt_userns, inode, mask); 4673ddcd056SLinus Torvalds } 4683ddcd056SLinus Torvalds 469cb23beb5SChristoph Hellwig /** 4700bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4710bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 47255852635SRandy Dunlap * @inode: Inode to check permission on 4730bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4740bdaea90SDavid Howells * 4750bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4760bdaea90SDavid Howells */ 4770bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4780bdaea90SDavid Howells { 4790bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4800bdaea90SDavid Howells umode_t mode = inode->i_mode; 4810bdaea90SDavid Howells 4820bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 483bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4840bdaea90SDavid Howells return -EROFS; 4850bdaea90SDavid Howells } 4860bdaea90SDavid Howells return 0; 4870bdaea90SDavid Howells } 4880bdaea90SDavid Howells 4890bdaea90SDavid Howells /** 4900bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 49147291baaSChristian Brauner * @mnt_userns: User namespace of the mount the inode was found from 4920bdaea90SDavid Howells * @inode: Inode to check permission on 4930bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4940bdaea90SDavid Howells * 4950bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4960bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4970bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4980bdaea90SDavid Howells * 4990bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 5000bdaea90SDavid Howells */ 50147291baaSChristian Brauner int inode_permission(struct user_namespace *mnt_userns, 50247291baaSChristian Brauner struct inode *inode, int mask) 5030bdaea90SDavid Howells { 5040bdaea90SDavid Howells int retval; 5050bdaea90SDavid Howells 5060bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 5070bdaea90SDavid Howells if (retval) 5080bdaea90SDavid Howells return retval; 5094bfd054aSEric Biggers 5104bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 5114bfd054aSEric Biggers /* 5124bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5134bfd054aSEric Biggers */ 5144bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5154bfd054aSEric Biggers return -EPERM; 5164bfd054aSEric Biggers 5174bfd054aSEric Biggers /* 5184bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5194bfd054aSEric Biggers * written back improperly if their true value is unknown 5204bfd054aSEric Biggers * to the vfs. 5214bfd054aSEric Biggers */ 522ba73d987SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 5234bfd054aSEric Biggers return -EACCES; 5244bfd054aSEric Biggers } 5254bfd054aSEric Biggers 52647291baaSChristian Brauner retval = do_inode_permission(mnt_userns, inode, mask); 5274bfd054aSEric Biggers if (retval) 5284bfd054aSEric Biggers return retval; 5294bfd054aSEric Biggers 5304bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5314bfd054aSEric Biggers if (retval) 5324bfd054aSEric Biggers return retval; 5334bfd054aSEric Biggers 5344bfd054aSEric Biggers return security_inode_permission(inode, mask); 5350bdaea90SDavid Howells } 5364d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5370bdaea90SDavid Howells 5380bdaea90SDavid Howells /** 5395dd784d0SJan Blunck * path_get - get a reference to a path 5405dd784d0SJan Blunck * @path: path to get the reference to 5415dd784d0SJan Blunck * 5425dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5435dd784d0SJan Blunck */ 544dcf787f3SAl Viro void path_get(const struct path *path) 5455dd784d0SJan Blunck { 5465dd784d0SJan Blunck mntget(path->mnt); 5475dd784d0SJan Blunck dget(path->dentry); 5485dd784d0SJan Blunck } 5495dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5505dd784d0SJan Blunck 5515dd784d0SJan Blunck /** 5521d957f9bSJan Blunck * path_put - put a reference to a path 5531d957f9bSJan Blunck * @path: path to put the reference to 5541d957f9bSJan Blunck * 5551d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5561d957f9bSJan Blunck */ 557dcf787f3SAl Viro void path_put(const struct path *path) 5581da177e4SLinus Torvalds { 5591d957f9bSJan Blunck dput(path->dentry); 5601d957f9bSJan Blunck mntput(path->mnt); 5611da177e4SLinus Torvalds } 5621d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5631da177e4SLinus Torvalds 564894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5651f55a6ecSAl Viro struct nameidata { 5661f55a6ecSAl Viro struct path path; 5671f55a6ecSAl Viro struct qstr last; 5681f55a6ecSAl Viro struct path root; 5691f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 570bcba1e7dSAl Viro unsigned int flags, state; 57103fa86e9SAl Viro unsigned seq, next_seq, m_seq, r_seq; 5721f55a6ecSAl Viro int last_type; 5731f55a6ecSAl Viro unsigned depth; 574756daf26SNeilBrown int total_link_count; 575697fc6caSAl Viro struct saved { 576697fc6caSAl Viro struct path link; 577fceef393SAl Viro struct delayed_call done; 578697fc6caSAl Viro const char *name; 5790450b2d1SAl Viro unsigned seq; 580894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5819883d185SAl Viro struct filename *name; 5829883d185SAl Viro struct nameidata *saved; 5839883d185SAl Viro unsigned root_seq; 5849883d185SAl Viro int dfd; 585a2bd096fSChristian Brauner vfsuid_t dir_vfsuid; 5860f705953SAl Viro umode_t dir_mode; 5873859a271SKees Cook } __randomize_layout; 5881f55a6ecSAl Viro 589bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 590bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 591bcba1e7dSAl Viro #define ND_JUMPED 4 592bcba1e7dSAl Viro 59306422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 594894bc8c4SAl Viro { 595756daf26SNeilBrown struct nameidata *old = current->nameidata; 596756daf26SNeilBrown p->stack = p->internal; 5977962c7d1SAl Viro p->depth = 0; 598c8a53ee5SAl Viro p->dfd = dfd; 599c8a53ee5SAl Viro p->name = name; 6007d01ef75SAl Viro p->path.mnt = NULL; 6017d01ef75SAl Viro p->path.dentry = NULL; 602756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 6039883d185SAl Viro p->saved = old; 604756daf26SNeilBrown current->nameidata = p; 605894bc8c4SAl Viro } 606894bc8c4SAl Viro 60706422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 60806422964SAl Viro const struct path *root) 60906422964SAl Viro { 61006422964SAl Viro __set_nameidata(p, dfd, name); 61106422964SAl Viro p->state = 0; 61206422964SAl Viro if (unlikely(root)) { 61306422964SAl Viro p->state = ND_ROOT_PRESET; 61406422964SAl Viro p->root = *root; 61506422964SAl Viro } 61606422964SAl Viro } 61706422964SAl Viro 6189883d185SAl Viro static void restore_nameidata(void) 619894bc8c4SAl Viro { 6209883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 621756daf26SNeilBrown 622756daf26SNeilBrown current->nameidata = old; 623756daf26SNeilBrown if (old) 624756daf26SNeilBrown old->total_link_count = now->total_link_count; 625e1a63bbcSAl Viro if (now->stack != now->internal) 626756daf26SNeilBrown kfree(now->stack); 627894bc8c4SAl Viro } 628894bc8c4SAl Viro 62960ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 630894bc8c4SAl Viro { 631bc40aee0SAl Viro struct saved *p; 632bc40aee0SAl Viro 6336da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 63460ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 635bc40aee0SAl Viro if (unlikely(!p)) 63660ef60c7SAl Viro return false; 637894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 638894bc8c4SAl Viro nd->stack = p; 63960ef60c7SAl Viro return true; 640894bc8c4SAl Viro } 641894bc8c4SAl Viro 642397d425dSEric W. Biederman /** 6436b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 644397d425dSEric W. Biederman * 645397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 646397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 647397d425dSEric W. Biederman */ 6486b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 649397d425dSEric W. Biederman { 65095dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 651397d425dSEric W. Biederman 652402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 653402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 654397d425dSEric W. Biederman return true; 655397d425dSEric W. Biederman 6566b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 657397d425dSEric W. Biederman } 658397d425dSEric W. Biederman 6597973387aSAl Viro static void drop_links(struct nameidata *nd) 6607973387aSAl Viro { 6617973387aSAl Viro int i = nd->depth; 6627973387aSAl Viro while (i--) { 6637973387aSAl Viro struct saved *last = nd->stack + i; 664fceef393SAl Viro do_delayed_call(&last->done); 665fceef393SAl Viro clear_delayed_call(&last->done); 6667973387aSAl Viro } 6677973387aSAl Viro } 6687973387aSAl Viro 6696e180327SAl Viro static void leave_rcu(struct nameidata *nd) 6706e180327SAl Viro { 6716e180327SAl Viro nd->flags &= ~LOOKUP_RCU; 67203fa86e9SAl Viro nd->seq = nd->next_seq = 0; 6736e180327SAl Viro rcu_read_unlock(); 6746e180327SAl Viro } 6756e180327SAl Viro 6767973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6777973387aSAl Viro { 6787973387aSAl Viro drop_links(nd); 6797973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6807973387aSAl Viro int i; 6817973387aSAl Viro path_put(&nd->path); 6827973387aSAl Viro for (i = 0; i < nd->depth; i++) 6837973387aSAl Viro path_put(&nd->stack[i].link); 684bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 685102b8af2SAl Viro path_put(&nd->root); 686bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 687102b8af2SAl Viro } 6887973387aSAl Viro } else { 6896e180327SAl Viro leave_rcu(nd); 6907973387aSAl Viro } 6917973387aSAl Viro nd->depth = 0; 6927d01ef75SAl Viro nd->path.mnt = NULL; 6937d01ef75SAl Viro nd->path.dentry = NULL; 6947973387aSAl Viro } 6957973387aSAl Viro 6967973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6972aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6987973387aSAl Viro { 6992aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 7007973387aSAl Viro if (unlikely(res)) { 7017973387aSAl Viro if (res > 0) 7027973387aSAl Viro path->mnt = NULL; 7037973387aSAl Viro path->dentry = NULL; 7047973387aSAl Viro return false; 7057973387aSAl Viro } 7067973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 7077973387aSAl Viro path->dentry = NULL; 7087973387aSAl Viro return false; 7097973387aSAl Viro } 7107973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7117973387aSAl Viro } 7127973387aSAl Viro 7132aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7142aa38470SAl Viro struct path *path, unsigned seq) 7152aa38470SAl Viro { 7165bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7172aa38470SAl Viro } 7182aa38470SAl Viro 7197973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7207973387aSAl Viro { 7217973387aSAl Viro int i; 722eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 723eacd9aa8SAl Viro drop_links(nd); 724eacd9aa8SAl Viro nd->depth = 0; 725eacd9aa8SAl Viro return false; 726eacd9aa8SAl Viro } 7277973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7287973387aSAl Viro struct saved *last = nd->stack + i; 7297973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7307973387aSAl Viro drop_links(nd); 7317973387aSAl Viro nd->depth = i + 1; 7327973387aSAl Viro return false; 7337973387aSAl Viro } 7347973387aSAl Viro } 7357973387aSAl Viro return true; 7367973387aSAl Viro } 7377973387aSAl Viro 738ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 739ee594bffSAl Viro { 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 7757973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7764675ac39SAl Viro goto out1; 77784a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 77884a2bd39SAl Viro goto out; 779ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7804675ac39SAl Viro goto out; 7816e180327SAl Viro leave_rcu(nd); 7824675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 783e36cffedSJens Axboe return true; 7844675ac39SAl Viro 78584a2bd39SAl Viro out1: 7864675ac39SAl Viro nd->path.mnt = NULL; 7874675ac39SAl Viro nd->path.dentry = NULL; 7884675ac39SAl Viro out: 7896e180327SAl Viro leave_rcu(nd); 790e36cffedSJens Axboe return false; 7914675ac39SAl Viro } 7924675ac39SAl Viro 7934675ac39SAl Viro /** 794ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7954675ac39SAl Viro * @nd: nameidata pathwalk data 796ae66db45SAl Viro * @dentry: next dentry to step into 797ae66db45SAl Viro * Returns: true on success, false on failure 7984675ac39SAl Viro * 79930476f7eSTom Rix * Similar to try_to_unlazy(), but here we have the next dentry already 800ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 801ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 802ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 8034675ac39SAl Viro * terminate_walk(). 8044675ac39SAl Viro */ 80503fa86e9SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry) 8064675ac39SAl Viro { 8077e4745a0SAl Viro int res; 8084675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8094675ac39SAl Viro 8104675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8114675ac39SAl Viro goto out2; 8127e4745a0SAl Viro res = __legitimize_mnt(nd->path.mnt, nd->m_seq); 8137e4745a0SAl Viro if (unlikely(res)) { 8147e4745a0SAl Viro if (res > 0) 8157973387aSAl Viro goto out2; 8167e4745a0SAl Viro goto out1; 8177e4745a0SAl Viro } 8184675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8197973387aSAl Viro goto out1; 82048a066e7SAl Viro 82115570086SLinus Torvalds /* 8224675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8234675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8244675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8254675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8264675ac39SAl Viro * know the parent must still be valid if the child sequence number is 82715570086SLinus Torvalds */ 8284675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 829e5c832d5SLinus Torvalds goto out; 83003fa86e9SAl Viro if (read_seqcount_retry(&dentry->d_seq, nd->next_seq)) 83184a2bd39SAl Viro goto out_dput; 832e5c832d5SLinus Torvalds /* 833e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 834e5c832d5SLinus Torvalds * still valid and get it if required. 835e5c832d5SLinus Torvalds */ 83684a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 83784a2bd39SAl Viro goto out_dput; 8386e180327SAl Viro leave_rcu(nd); 839ae66db45SAl Viro return true; 84019660af7SAl Viro 8417973387aSAl Viro out2: 8427973387aSAl Viro nd->path.mnt = NULL; 8437973387aSAl Viro out1: 8447973387aSAl Viro nd->path.dentry = NULL; 845e5c832d5SLinus Torvalds out: 8466e180327SAl Viro leave_rcu(nd); 847ae66db45SAl Viro return false; 84884a2bd39SAl Viro out_dput: 8496e180327SAl Viro leave_rcu(nd); 85084a2bd39SAl Viro dput(dentry); 851ae66db45SAl Viro return false; 85231e6b01fSNick Piggin } 85331e6b01fSNick Piggin 8544ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 85534286d66SNick Piggin { 856a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8574ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 858a89f8337SAl Viro else 859a89f8337SAl Viro return 1; 86034286d66SNick Piggin } 86134286d66SNick Piggin 8629f1fafeeSAl Viro /** 8639f1fafeeSAl Viro * complete_walk - successful completion of path walk 8649f1fafeeSAl Viro * @nd: pointer nameidata 86539159de2SJeff Layton * 8669f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8679f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8689f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8699f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8709f1fafeeSAl Viro * need to drop nd->path. 87139159de2SJeff Layton */ 8729f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 87339159de2SJeff Layton { 87416c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 87539159de2SJeff Layton int status; 87639159de2SJeff Layton 8779f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 878adb21d2bSAleksa Sarai /* 879adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 880adb21d2bSAleksa Sarai * externally-managed nd->root. 881adb21d2bSAleksa Sarai */ 882bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 883bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8849f1fafeeSAl Viro nd->root.mnt = NULL; 8856c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 886e36cffedSJens Axboe if (!try_to_unlazy(nd)) 88748a066e7SAl Viro return -ECHILD; 88848a066e7SAl Viro } 8899f1fafeeSAl Viro 890adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 891adb21d2bSAleksa Sarai /* 892adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 893adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 894adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 895adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 896adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 897adb21d2bSAleksa Sarai * 898adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 899adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 900adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 901adb21d2bSAleksa Sarai * requested root to userspace. 902adb21d2bSAleksa Sarai * 903adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 904adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 905adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 906adb21d2bSAleksa Sarai */ 907adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 908adb21d2bSAleksa Sarai return -EXDEV; 909adb21d2bSAleksa Sarai } 910adb21d2bSAleksa Sarai 911bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 91239159de2SJeff Layton return 0; 91339159de2SJeff Layton 914ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 91516c2cd71SAl Viro return 0; 91616c2cd71SAl Viro 917ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 91839159de2SJeff Layton if (status > 0) 91939159de2SJeff Layton return 0; 92039159de2SJeff Layton 92116c2cd71SAl Viro if (!status) 92239159de2SJeff Layton status = -ESTALE; 92316c2cd71SAl Viro 92439159de2SJeff Layton return status; 92539159de2SJeff Layton } 92639159de2SJeff Layton 927740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9282a737871SAl Viro { 92931e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9309e6697e2SAl Viro 931adb21d2bSAleksa Sarai /* 932adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 933adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 934adb21d2bSAleksa Sarai * from the dirfd. 935adb21d2bSAleksa Sarai */ 936adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 937adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 938adb21d2bSAleksa Sarai 9399e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9408f47a016SAl Viro unsigned seq; 941c28cc364SNick Piggin 942c28cc364SNick Piggin do { 943c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 94431e6b01fSNick Piggin nd->root = fs->root; 9458f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 946c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9479e6697e2SAl Viro } else { 9489e6697e2SAl Viro get_fs_root(fs, &nd->root); 949bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9509e6697e2SAl Viro } 951740a1678SAleksa Sarai return 0; 95231e6b01fSNick Piggin } 95331e6b01fSNick Piggin 954248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 955248fb5b9SAl Viro { 956adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 957adb21d2bSAleksa Sarai return -EXDEV; 95872ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 95972ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 96072ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 96172ba2929SAleksa Sarai return -EXDEV; 96272ba2929SAleksa Sarai } 963740a1678SAleksa Sarai if (!nd->root.mnt) { 964740a1678SAleksa Sarai int error = set_root(nd); 965740a1678SAleksa Sarai if (error) 966740a1678SAleksa Sarai return error; 967740a1678SAleksa Sarai } 968248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 969248fb5b9SAl Viro struct dentry *d; 970248fb5b9SAl Viro nd->path = nd->root; 971248fb5b9SAl Viro d = nd->path.dentry; 972248fb5b9SAl Viro nd->inode = d->d_inode; 973248fb5b9SAl Viro nd->seq = nd->root_seq; 97482ef0698SAl Viro if (read_seqcount_retry(&d->d_seq, nd->seq)) 975248fb5b9SAl Viro return -ECHILD; 976248fb5b9SAl Viro } else { 977248fb5b9SAl Viro path_put(&nd->path); 978248fb5b9SAl Viro nd->path = nd->root; 979248fb5b9SAl Viro path_get(&nd->path); 980248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 981248fb5b9SAl Viro } 982bcba1e7dSAl Viro nd->state |= ND_JUMPED; 983248fb5b9SAl Viro return 0; 984248fb5b9SAl Viro } 985248fb5b9SAl Viro 986b5fb63c1SChristoph Hellwig /* 9876b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 988b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 989b5fb63c1SChristoph Hellwig */ 990ea4af4aaSAl Viro int nd_jump_link(const struct path *path) 991b5fb63c1SChristoph Hellwig { 9924b99d499SAleksa Sarai int error = -ELOOP; 9936e77137bSAl Viro struct nameidata *nd = current->nameidata; 994b5fb63c1SChristoph Hellwig 9954b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9964b99d499SAleksa Sarai goto err; 9974b99d499SAleksa Sarai 99872ba2929SAleksa Sarai error = -EXDEV; 99972ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 100072ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 100172ba2929SAleksa Sarai goto err; 100272ba2929SAleksa Sarai } 1003adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 1004adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 1005adb21d2bSAleksa Sarai goto err; 100672ba2929SAleksa Sarai 10074b99d499SAleksa Sarai path_put(&nd->path); 1008b5fb63c1SChristoph Hellwig nd->path = *path; 1009b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1010bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10111bc82070SAleksa Sarai return 0; 10124b99d499SAleksa Sarai 10134b99d499SAleksa Sarai err: 10144b99d499SAleksa Sarai path_put(path); 10154b99d499SAleksa Sarai return error; 1016b5fb63c1SChristoph Hellwig } 1017b5fb63c1SChristoph Hellwig 1018b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1019574197e0SAl Viro { 102021c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1021fceef393SAl Viro do_delayed_call(&last->done); 10226548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1023b9ff4429SAl Viro path_put(&last->link); 1024574197e0SAl Viro } 1025574197e0SAl Viro 10269c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10279c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10289c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10299c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10309c011be1SLuis Chamberlain 10319c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10329c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10339c011be1SLuis Chamberlain { 10349c011be1SLuis Chamberlain .procname = "protected_symlinks", 10359c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10369c011be1SLuis Chamberlain .maxlen = sizeof(int), 1037c7031c14SJulius Hemanth Pitti .mode = 0644, 10389c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10399c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10409c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10419c011be1SLuis Chamberlain }, 10429c011be1SLuis Chamberlain { 10439c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10449c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10459c011be1SLuis Chamberlain .maxlen = sizeof(int), 1046c7031c14SJulius Hemanth Pitti .mode = 0644, 10479c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10489c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10499c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10509c011be1SLuis Chamberlain }, 10519c011be1SLuis Chamberlain { 10529c011be1SLuis Chamberlain .procname = "protected_fifos", 10539c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10549c011be1SLuis Chamberlain .maxlen = sizeof(int), 1055c7031c14SJulius Hemanth Pitti .mode = 0644, 10569c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10579c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10589c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10599c011be1SLuis Chamberlain }, 10609c011be1SLuis Chamberlain { 10619c011be1SLuis Chamberlain .procname = "protected_regular", 10629c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10639c011be1SLuis Chamberlain .maxlen = sizeof(int), 1064c7031c14SJulius Hemanth Pitti .mode = 0644, 10659c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10669c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10679c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10689c011be1SLuis Chamberlain }, 10699c011be1SLuis Chamberlain { } 10709c011be1SLuis Chamberlain }; 10719c011be1SLuis Chamberlain 10729c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10739c011be1SLuis Chamberlain { 10749c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10759c011be1SLuis Chamberlain return 0; 10769c011be1SLuis Chamberlain } 10779c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10789c011be1SLuis Chamberlain 10799c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1080800179c9SKees Cook 1081800179c9SKees Cook /** 1082800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 108355852635SRandy Dunlap * @nd: nameidata pathwalk data 1084800179c9SKees Cook * 1085800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1086800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1087800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1088800179c9SKees Cook * processes from failing races against path names that may change out 1089800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1090800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1091800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1092800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1093800179c9SKees Cook * 1094800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1095800179c9SKees Cook */ 1096ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1097800179c9SKees Cook { 1098ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1099a2bd096fSChristian Brauner vfsuid_t vfsuid; 1100ba73d987SChristian Brauner 1101800179c9SKees Cook if (!sysctl_protected_symlinks) 1102800179c9SKees Cook return 0; 1103800179c9SKees Cook 1104ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1105a2bd096fSChristian Brauner vfsuid = i_uid_into_vfsuid(mnt_userns, inode); 1106800179c9SKees Cook /* Allowed if owner and follower match. */ 1107a2bd096fSChristian Brauner if (vfsuid_eq_kuid(vfsuid, current_fsuid())) 1108800179c9SKees Cook return 0; 1109800179c9SKees Cook 1110800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11110f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1112800179c9SKees Cook return 0; 1113800179c9SKees Cook 1114800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1115a2bd096fSChristian Brauner if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid)) 1116800179c9SKees Cook return 0; 1117800179c9SKees Cook 111831956502SAl Viro if (nd->flags & LOOKUP_RCU) 111931956502SAl Viro return -ECHILD; 112031956502SAl Viro 1121ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1122245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1123800179c9SKees Cook return -EACCES; 1124800179c9SKees Cook } 1125800179c9SKees Cook 1126800179c9SKees Cook /** 1127800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1128ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1129800179c9SKees Cook * @inode: the source inode to hardlink from 1130800179c9SKees Cook * 1131800179c9SKees Cook * Return false if at least one of the following conditions: 1132800179c9SKees Cook * - inode is not a regular file 1133800179c9SKees Cook * - inode is setuid 1134800179c9SKees Cook * - inode is setgid and group-exec 1135800179c9SKees Cook * - access failure for read and write 1136800179c9SKees Cook * 1137800179c9SKees Cook * Otherwise returns true. 1138800179c9SKees Cook */ 1139ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1140ba73d987SChristian Brauner struct inode *inode) 1141800179c9SKees Cook { 1142800179c9SKees Cook umode_t mode = inode->i_mode; 1143800179c9SKees Cook 1144800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1145800179c9SKees Cook if (!S_ISREG(mode)) 1146800179c9SKees Cook return false; 1147800179c9SKees Cook 1148800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1149800179c9SKees Cook if (mode & S_ISUID) 1150800179c9SKees Cook return false; 1151800179c9SKees Cook 1152800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1153800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1154800179c9SKees Cook return false; 1155800179c9SKees Cook 1156800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1157ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1158800179c9SKees Cook return false; 1159800179c9SKees Cook 1160800179c9SKees Cook return true; 1161800179c9SKees Cook } 1162800179c9SKees Cook 1163800179c9SKees Cook /** 1164800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1165ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1166800179c9SKees Cook * @link: the source to hardlink from 1167800179c9SKees Cook * 1168800179c9SKees Cook * Block hardlink when all of: 1169800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1170800179c9SKees Cook * - fsuid does not match inode 1171800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1172f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1173800179c9SKees Cook * 1174ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1175ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1176ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1177ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1178ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1179ba73d987SChristian Brauner * 1180800179c9SKees Cook * Returns 0 if successful, -ve on error. 1181800179c9SKees Cook */ 11828996682bSAl Viro int may_linkat(struct user_namespace *mnt_userns, const struct path *link) 1183800179c9SKees Cook { 1184593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1185593d1ce8SEric W. Biederman 1186593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1187a2bd096fSChristian Brauner if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) || 1188a2bd096fSChristian Brauner !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode))) 1189593d1ce8SEric W. Biederman return -EOVERFLOW; 1190800179c9SKees Cook 1191800179c9SKees Cook if (!sysctl_protected_hardlinks) 1192800179c9SKees Cook return 0; 1193800179c9SKees Cook 1194800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1195800179c9SKees Cook * otherwise, it must be a safe source. 1196800179c9SKees Cook */ 1197ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1198ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1199800179c9SKees Cook return 0; 1200800179c9SKees Cook 1201245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1202800179c9SKees Cook return -EPERM; 1203800179c9SKees Cook } 1204800179c9SKees Cook 120530aba665SSalvatore Mesoraca /** 120630aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 120730aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 120830aba665SSalvatore Mesoraca * exist. 1209ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 12102111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 121130aba665SSalvatore Mesoraca * @inode: the inode of the file to open 121230aba665SSalvatore Mesoraca * 121330aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 121430aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 121530aba665SSalvatore Mesoraca * - the file already exists 121630aba665SSalvatore Mesoraca * - we are in a sticky directory 121730aba665SSalvatore Mesoraca * - we don't own the file 121830aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 121930aba665SSalvatore Mesoraca * - the directory is world writable 122030aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 122130aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 122230aba665SSalvatore Mesoraca * be enough. 122330aba665SSalvatore Mesoraca * 1224ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1225ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1226ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1227ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1228ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1229ba73d987SChristian Brauner * 123030aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 123130aba665SSalvatore Mesoraca */ 1232ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1233ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 123430aba665SSalvatore Mesoraca { 1235ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1236a2bd096fSChristian Brauner vfsuid_t dir_vfsuid = nd->dir_vfsuid; 1237ba73d987SChristian Brauner 123830aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 123930aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1240d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1241a2bd096fSChristian Brauner vfsuid_eq(i_uid_into_vfsuid(mnt_userns, inode), dir_vfsuid) || 1242a2bd096fSChristian Brauner vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), current_fsuid())) 124330aba665SSalvatore Mesoraca return 0; 124430aba665SSalvatore Mesoraca 1245d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1246d0cb5018SAl Viro (dir_mode & 0020 && 124730aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 124830aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1249245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1250245d7369SKees Cook "sticky_create_fifo" : 1251245d7369SKees Cook "sticky_create_regular"; 1252245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 125330aba665SSalvatore Mesoraca return -EACCES; 125430aba665SSalvatore Mesoraca } 125530aba665SSalvatore Mesoraca return 0; 125630aba665SSalvatore Mesoraca } 125730aba665SSalvatore Mesoraca 1258f015f126SDavid Howells /* 1259f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1260f015f126SDavid Howells * 1261f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1262f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1263f015f126SDavid Howells * Up is towards /. 1264f015f126SDavid Howells * 1265f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1266f015f126SDavid Howells * root. 1267f015f126SDavid Howells */ 1268bab77ebfSAl Viro int follow_up(struct path *path) 12691da177e4SLinus Torvalds { 12700714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12710714a533SAl Viro struct mount *parent; 12721da177e4SLinus Torvalds struct dentry *mountpoint; 127399b7db7bSNick Piggin 127448a066e7SAl Viro read_seqlock_excl(&mount_lock); 12750714a533SAl Viro parent = mnt->mnt_parent; 12763c0a6163SAl Viro if (parent == mnt) { 127748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12781da177e4SLinus Torvalds return 0; 12791da177e4SLinus Torvalds } 12800714a533SAl Viro mntget(&parent->mnt); 1281a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 128248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1283bab77ebfSAl Viro dput(path->dentry); 1284bab77ebfSAl Viro path->dentry = mountpoint; 1285bab77ebfSAl Viro mntput(path->mnt); 12860714a533SAl Viro path->mnt = &parent->mnt; 12871da177e4SLinus Torvalds return 1; 12881da177e4SLinus Torvalds } 12894d359507SAl Viro EXPORT_SYMBOL(follow_up); 12901da177e4SLinus Torvalds 12917ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12927ef482faSAl Viro struct path *path, unsigned *seqp) 12937ef482faSAl Viro { 12947ef482faSAl Viro while (mnt_has_parent(m)) { 12957ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12967ef482faSAl Viro 12977ef482faSAl Viro m = m->mnt_parent; 12987ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12997ef482faSAl Viro root->mnt == &m->mnt)) 13007ef482faSAl Viro break; 13017ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 13027ef482faSAl Viro path->mnt = &m->mnt; 13037ef482faSAl Viro path->dentry = mountpoint; 13047ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 13057ef482faSAl Viro return true; 13067ef482faSAl Viro } 13077ef482faSAl Viro } 13087ef482faSAl Viro return false; 13097ef482faSAl Viro } 13107ef482faSAl Viro 13112aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 13122aa38470SAl Viro struct path *path) 13132aa38470SAl Viro { 13142aa38470SAl Viro bool found; 13152aa38470SAl Viro 13162aa38470SAl Viro rcu_read_lock(); 13172aa38470SAl Viro while (1) { 13182aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 13192aa38470SAl Viro 13202aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 13212aa38470SAl Viro if (unlikely(!found)) { 13222aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 13232aa38470SAl Viro break; 13242aa38470SAl Viro } else { 13252aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 13262aa38470SAl Viro break; 13272aa38470SAl Viro rcu_read_unlock(); 13282aa38470SAl Viro path_put(path); 13292aa38470SAl Viro rcu_read_lock(); 13302aa38470SAl Viro } 13312aa38470SAl Viro } 13322aa38470SAl Viro rcu_read_unlock(); 13332aa38470SAl Viro return found; 13342aa38470SAl Viro } 13352aa38470SAl Viro 1336b5c84bf6SNick Piggin /* 13379875cf80SDavid Howells * Perform an automount 13389875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 13399875cf80SDavid Howells * were called with. 13401da177e4SLinus Torvalds */ 13411c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 134231e6b01fSNick Piggin { 134325e195aaSAl Viro struct dentry *dentry = path->dentry; 13449875cf80SDavid Howells 13450ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 13460ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 13470ec26fd0SMiklos Szeredi * the name. 13480ec26fd0SMiklos Szeredi * 13490ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 13505a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 13510ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 13520ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13530ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13540ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13555a30d8a2SDavid Howells */ 13561c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13575d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 135825e195aaSAl Viro dentry->d_inode) 13599875cf80SDavid Howells return -EISDIR; 13600ec26fd0SMiklos Szeredi 13611c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13629875cf80SDavid Howells return -ELOOP; 13639875cf80SDavid Howells 136425e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1365ea5b778aSDavid Howells } 13669875cf80SDavid Howells 13679875cf80SDavid Howells /* 13689deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13699deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13709deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13719deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13729875cf80SDavid Howells */ 13739deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13749deed3ebSAl Viro int *count, unsigned lookup_flags) 13759875cf80SDavid Howells { 13769deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13779875cf80SDavid Howells bool need_mntput = false; 13788aef1884SAl Viro int ret = 0; 13799875cf80SDavid Howells 13809deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1381cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1382cc53ce53SDavid Howells * being held. */ 1383d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1384fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1385508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1386cc53ce53SDavid Howells if (ret < 0) 13878aef1884SAl Viro break; 1388cc53ce53SDavid Howells } 1389cc53ce53SDavid Howells 13909deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13919875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13929deed3ebSAl Viro if (mounted) { // ... in our namespace 13939875cf80SDavid Howells dput(path->dentry); 13949875cf80SDavid Howells if (need_mntput) 1395463ffb2eSAl Viro mntput(path->mnt); 1396463ffb2eSAl Viro path->mnt = mounted; 1397463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13989deed3ebSAl Viro // here we know it's positive 13999deed3ebSAl Viro flags = path->dentry->d_flags; 14009875cf80SDavid Howells need_mntput = true; 14019875cf80SDavid Howells continue; 1402463ffb2eSAl Viro } 14031da177e4SLinus Torvalds } 14049875cf80SDavid Howells 14059deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 14069deed3ebSAl Viro break; 14079deed3ebSAl Viro 14089deed3ebSAl Viro // uncovered automount point 14099deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 14109deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 14119875cf80SDavid Howells if (ret < 0) 14128aef1884SAl Viro break; 14139875cf80SDavid Howells } 14149875cf80SDavid Howells 14159deed3ebSAl Viro if (ret == -EISDIR) 14169deed3ebSAl Viro ret = 0; 14179deed3ebSAl Viro // possible if you race with several mount --move 14189deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 14198aef1884SAl Viro mntput(path->mnt); 14209deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1421d41efb52SAl Viro ret = -ENOENT; 14229deed3ebSAl Viro *jumped = need_mntput; 14238402752eSAl Viro return ret; 14241da177e4SLinus Torvalds } 14251da177e4SLinus Torvalds 14269deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 14279deed3ebSAl Viro int *count, unsigned lookup_flags) 14289deed3ebSAl Viro { 14299deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 14309deed3ebSAl Viro 14319deed3ebSAl Viro /* fastpath */ 14329deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 14339deed3ebSAl Viro *jumped = false; 14349deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 14359deed3ebSAl Viro return -ENOENT; 14369deed3ebSAl Viro return 0; 14379deed3ebSAl Viro } 14389deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 14399deed3ebSAl Viro } 14409deed3ebSAl Viro 1441cc53ce53SDavid Howells int follow_down_one(struct path *path) 14421da177e4SLinus Torvalds { 14431da177e4SLinus Torvalds struct vfsmount *mounted; 14441da177e4SLinus Torvalds 14451c755af4SAl Viro mounted = lookup_mnt(path); 14461da177e4SLinus Torvalds if (mounted) { 14479393bd07SAl Viro dput(path->dentry); 14489393bd07SAl Viro mntput(path->mnt); 14499393bd07SAl Viro path->mnt = mounted; 14509393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 14511da177e4SLinus Torvalds return 1; 14521da177e4SLinus Torvalds } 14531da177e4SLinus Torvalds return 0; 14541da177e4SLinus Torvalds } 14554d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14561da177e4SLinus Torvalds 14579875cf80SDavid Howells /* 14589deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14599deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14609deed3ebSAl Viro * caller is permitted to proceed or not. 14619deed3ebSAl Viro */ 14629deed3ebSAl Viro int follow_down(struct path *path) 14639deed3ebSAl Viro { 14649deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14659deed3ebSAl Viro bool jumped; 14669deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 14679deed3ebSAl Viro 14689deed3ebSAl Viro if (path->mnt != mnt) 14699deed3ebSAl Viro mntput(mnt); 14709deed3ebSAl Viro return ret; 14719deed3ebSAl Viro } 14729deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14739deed3ebSAl Viro 14749deed3ebSAl Viro /* 1475287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1476287548e4SAl Viro * we meet a managed dentry that would need blocking. 14779875cf80SDavid Howells */ 14783bd8bc89SAl Viro static bool __follow_mount_rcu(struct nameidata *nd, struct path *path) 14799875cf80SDavid Howells { 1480ea936aebSAl Viro struct dentry *dentry = path->dentry; 1481ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1482ea936aebSAl Viro 1483ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1484ea936aebSAl Viro return true; 1485ea936aebSAl Viro 1486ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1487ea936aebSAl Viro return false; 1488ea936aebSAl Viro 148962a7375eSIan Kent for (;;) { 149062a7375eSIan Kent /* 149162a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 149262a7375eSIan Kent * that wants to block transit. 149362a7375eSIan Kent */ 1494ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1495ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1496ea936aebSAl Viro if (res) 1497ea936aebSAl Viro return res == -EISDIR; 1498ea936aebSAl Viro flags = dentry->d_flags; 1499b8faf035SNeilBrown } 150062a7375eSIan Kent 1501ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1502ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1503ea936aebSAl Viro if (mounted) { 1504c7105365SAl Viro path->mnt = &mounted->mnt; 1505ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1506bcba1e7dSAl Viro nd->state |= ND_JUMPED; 150703fa86e9SAl Viro nd->next_seq = read_seqcount_begin(&dentry->d_seq); 1508ea936aebSAl Viro flags = dentry->d_flags; 150903fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach 151003fa86e9SAl Viro // this state. 151120aac6c6SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 151220aac6c6SAl Viro return false; 1513ea936aebSAl Viro continue; 15149875cf80SDavid Howells } 1515ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1516ea936aebSAl Viro return false; 1517ea936aebSAl Viro } 1518ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1519ea936aebSAl Viro } 1520287548e4SAl Viro } 1521287548e4SAl Viro 1522db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 15233bd8bc89SAl Viro struct path *path) 1524bd7c4b50SAl Viro { 15259deed3ebSAl Viro bool jumped; 1526db3c9adeSAl Viro int ret; 1527bd7c4b50SAl Viro 1528db3c9adeSAl Viro path->mnt = nd->path.mnt; 1529db3c9adeSAl Viro path->dentry = dentry; 1530c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 153103fa86e9SAl Viro unsigned int seq = nd->next_seq; 15323bd8bc89SAl Viro if (likely(__follow_mount_rcu(nd, path))) 15339deed3ebSAl Viro return 0; 153403fa86e9SAl Viro // *path and nd->next_seq might've been clobbered 1535c153007bSAl Viro path->mnt = nd->path.mnt; 1536c153007bSAl Viro path->dentry = dentry; 153703fa86e9SAl Viro nd->next_seq = seq; 153803fa86e9SAl Viro if (!try_to_unlazy_next(nd, dentry)) 153903fa86e9SAl Viro return -ECHILD; 1540c153007bSAl Viro } 15419deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 15429deed3ebSAl Viro if (jumped) { 15439deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 15449deed3ebSAl Viro ret = -EXDEV; 15459deed3ebSAl Viro else 1546bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15479deed3ebSAl Viro } 15489deed3ebSAl Viro if (unlikely(ret)) { 15499deed3ebSAl Viro dput(path->dentry); 15509deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15519deed3ebSAl Viro mntput(path->mnt); 1552bd7c4b50SAl Viro } 1553bd7c4b50SAl Viro return ret; 1554bd7c4b50SAl Viro } 1555bd7c4b50SAl Viro 15569875cf80SDavid Howells /* 1557f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1558f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1559baa03890SNick Piggin */ 1560e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1561e3c13928SAl Viro struct dentry *dir, 15626c51e513SAl Viro unsigned int flags) 1563baa03890SNick Piggin { 1564a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1565bad61189SMiklos Szeredi if (dentry) { 1566a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1567bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 156874ff0ffcSAl Viro if (!error) 15695542aa2fSEric W. Biederman d_invalidate(dentry); 1570bad61189SMiklos Szeredi dput(dentry); 157174ff0ffcSAl Viro return ERR_PTR(error); 1572bad61189SMiklos Szeredi } 1573bad61189SMiklos Szeredi } 1574baa03890SNick Piggin return dentry; 1575baa03890SNick Piggin } 1576baa03890SNick Piggin 1577baa03890SNick Piggin /* 1578a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1579a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1580a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1581a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1582a03ece5fSAl Viro * at all. 158344396f4bSJosef Bacik */ 1584a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1585a03ece5fSAl Viro struct dentry *base, unsigned int flags) 158644396f4bSJosef Bacik { 1587a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 158844396f4bSJosef Bacik struct dentry *old; 1589a03ece5fSAl Viro struct inode *dir = base->d_inode; 1590a03ece5fSAl Viro 1591a03ece5fSAl Viro if (dentry) 1592a03ece5fSAl Viro return dentry; 159344396f4bSJosef Bacik 159444396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1595a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 159644396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1597a03ece5fSAl Viro 1598a03ece5fSAl Viro dentry = d_alloc(base, name); 1599a03ece5fSAl Viro if (unlikely(!dentry)) 1600a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 160144396f4bSJosef Bacik 160272bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 160344396f4bSJosef Bacik if (unlikely(old)) { 160444396f4bSJosef Bacik dput(dentry); 160544396f4bSJosef Bacik dentry = old; 160644396f4bSJosef Bacik } 160744396f4bSJosef Bacik return dentry; 160844396f4bSJosef Bacik } 160944396f4bSJosef Bacik 16104cb64024SAl Viro static struct dentry *lookup_fast(struct nameidata *nd) 16111da177e4SLinus Torvalds { 161231e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16135a18fff2SAl Viro int status = 1; 16149875cf80SDavid Howells 16153cac260aSAl Viro /* 1616b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16175d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16185d0f49c1SAl Viro * going to fall back to non-racy lookup. 1619b04f784eSNick Piggin */ 162031e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 162103fa86e9SAl Viro dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq); 16225d0f49c1SAl Viro if (unlikely(!dentry)) { 1623e36cffedSJens Axboe if (!try_to_unlazy(nd)) 162420e34357SAl Viro return ERR_PTR(-ECHILD); 162520e34357SAl Viro return NULL; 16265d0f49c1SAl Viro } 16275a18fff2SAl Viro 162812f8ad4bSLinus Torvalds /* 162912f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 163012f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 163112f8ad4bSLinus Torvalds */ 16324cb64024SAl Viro if (read_seqcount_retry(&parent->d_seq, nd->seq)) 163320e34357SAl Viro return ERR_PTR(-ECHILD); 16345a18fff2SAl Viro 16354ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1636c153007bSAl Viro if (likely(status > 0)) 163720e34357SAl Viro return dentry; 163803fa86e9SAl Viro if (!try_to_unlazy_next(nd, dentry)) 163920e34357SAl Viro return ERR_PTR(-ECHILD); 164026ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1641209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1642209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16435a18fff2SAl Viro } else { 1644e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 164581e6f520SAl Viro if (unlikely(!dentry)) 164620e34357SAl Viro return NULL; 16474ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16485d0f49c1SAl Viro } 16495a18fff2SAl Viro if (unlikely(status <= 0)) { 1650e9742b53SAl Viro if (!status) 16515d0f49c1SAl Viro d_invalidate(dentry); 16525a18fff2SAl Viro dput(dentry); 165320e34357SAl Viro return ERR_PTR(status); 16545a18fff2SAl Viro } 165520e34357SAl Viro return dentry; 1656697f514dSMiklos Szeredi } 1657697f514dSMiklos Szeredi 1658697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 165988d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1660e3c13928SAl Viro struct dentry *dir, 1661e3c13928SAl Viro unsigned int flags) 1662697f514dSMiklos Szeredi { 166388d8331aSAl Viro struct dentry *dentry, *old; 16641936386eSAl Viro struct inode *inode = dir->d_inode; 1665d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16661936386eSAl Viro 16671936386eSAl Viro /* Don't go there if it's already dead */ 166894bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 166988d8331aSAl Viro return ERR_PTR(-ENOENT); 167094bdd655SAl Viro again: 1671d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 167294bdd655SAl Viro if (IS_ERR(dentry)) 167388d8331aSAl Viro return dentry; 167494bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1675949a852eSAl Viro int error = d_revalidate(dentry, flags); 1676949a852eSAl Viro if (unlikely(error <= 0)) { 167794bdd655SAl Viro if (!error) { 1678949a852eSAl Viro d_invalidate(dentry); 1679949a852eSAl Viro dput(dentry); 168094bdd655SAl Viro goto again; 168194bdd655SAl Viro } 168294bdd655SAl Viro dput(dentry); 1683949a852eSAl Viro dentry = ERR_PTR(error); 1684949a852eSAl Viro } 168594bdd655SAl Viro } else { 16861936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 168785c7f810SAl Viro d_lookup_done(dentry); 16881936386eSAl Viro if (unlikely(old)) { 16891936386eSAl Viro dput(dentry); 16901936386eSAl Viro dentry = old; 1691949a852eSAl Viro } 1692949a852eSAl Viro } 1693e3c13928SAl Viro return dentry; 16941da177e4SLinus Torvalds } 16951da177e4SLinus Torvalds 169688d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 169788d8331aSAl Viro struct dentry *dir, 169888d8331aSAl Viro unsigned int flags) 169988d8331aSAl Viro { 170088d8331aSAl Viro struct inode *inode = dir->d_inode; 170188d8331aSAl Viro struct dentry *res; 170288d8331aSAl Viro inode_lock_shared(inode); 170388d8331aSAl Viro res = __lookup_slow(name, dir, flags); 170488d8331aSAl Viro inode_unlock_shared(inode); 170588d8331aSAl Viro return res; 170688d8331aSAl Viro } 170788d8331aSAl Viro 1708ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1709ba73d987SChristian Brauner struct nameidata *nd) 171052094c8aSAl Viro { 171152094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17127d6beb71SLinus Torvalds int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1713e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 171452094c8aSAl Viro return err; 171552094c8aSAl Viro } 1716ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 171752094c8aSAl Viro } 171852094c8aSAl Viro 171903fa86e9SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link) 1720d63ff28fSAl Viro { 172149055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 172249055906SAl Viro return -ELOOP; 17234542576bSAl Viro 17244542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17254542576bSAl Viro return 0; 17264542576bSAl Viro if (likely(nd->stack != nd->internal)) 17274542576bSAl Viro return 0; 172860ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 172949055906SAl Viro return 0; 173060ef60c7SAl Viro 173160ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 173260ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 173360ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 173403fa86e9SAl Viro bool grabbed_link = legitimize_path(nd, link, nd->next_seq); 173560ef60c7SAl Viro 1736e5ca024eSAl Viro if (!try_to_unlazy(nd) || !grabbed_link) 173760ef60c7SAl Viro return -ECHILD; 173860ef60c7SAl Viro 173960ef60c7SAl Viro if (nd_alloc_stack(nd)) 174060ef60c7SAl Viro return 0; 1741bc40aee0SAl Viro } 174260ef60c7SAl Viro return -ENOMEM; 174349055906SAl Viro } 174449055906SAl Viro 174549055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 174649055906SAl Viro 174749055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 174803fa86e9SAl Viro struct inode *inode, int flags) 174949055906SAl Viro { 175049055906SAl Viro struct saved *last; 175149055906SAl Viro const char *res; 175203fa86e9SAl Viro int error = reserve_stack(nd, link); 175349055906SAl Viro 175449055906SAl Viro if (unlikely(error)) { 175549055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1756cd179f44SAl Viro path_put(link); 175706708adbSAl Viro return ERR_PTR(error); 1758626de996SAl Viro } 1759ab104923SAl Viro last = nd->stack + nd->depth++; 17601cf2665bSAl Viro last->link = *link; 1761fceef393SAl Viro clear_delayed_call(&last->done); 176203fa86e9SAl Viro last->seq = nd->next_seq; 1763ad6cc4c3SAl Viro 1764b1a81972SAl Viro if (flags & WALK_TRAILING) { 1765ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1766ad6cc4c3SAl Viro if (unlikely(error)) 1767ad6cc4c3SAl Viro return ERR_PTR(error); 1768ad6cc4c3SAl Viro } 1769ad6cc4c3SAl Viro 1770dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1771dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1772ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1773ad6cc4c3SAl Viro 1774ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1775ad6cc4c3SAl Viro touch_atime(&last->link); 1776ad6cc4c3SAl Viro cond_resched(); 1777ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1778e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1779ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1780ad6cc4c3SAl Viro touch_atime(&last->link); 1781ad6cc4c3SAl Viro } 1782ad6cc4c3SAl Viro 1783ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1784ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1785ad6cc4c3SAl Viro if (unlikely(error)) 1786ad6cc4c3SAl Viro return ERR_PTR(error); 1787ad6cc4c3SAl Viro 1788ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1789ad6cc4c3SAl Viro if (!res) { 1790ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1791ad6cc4c3SAl Viro struct delayed_call *); 1792ad6cc4c3SAl Viro get = inode->i_op->get_link; 1793ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1794ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1795e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1796ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1797ad6cc4c3SAl Viro } else { 1798ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1799ad6cc4c3SAl Viro } 1800ad6cc4c3SAl Viro if (!res) 1801ad6cc4c3SAl Viro goto all_done; 1802ad6cc4c3SAl Viro if (IS_ERR(res)) 1803ad6cc4c3SAl Viro return res; 1804ad6cc4c3SAl Viro } 1805ad6cc4c3SAl Viro if (*res == '/') { 1806ad6cc4c3SAl Viro error = nd_jump_root(nd); 1807ad6cc4c3SAl Viro if (unlikely(error)) 1808ad6cc4c3SAl Viro return ERR_PTR(error); 1809ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1810ad6cc4c3SAl Viro ; 1811ad6cc4c3SAl Viro } 1812ad6cc4c3SAl Viro if (*res) 1813ad6cc4c3SAl Viro return res; 1814ad6cc4c3SAl Viro all_done: // pure jump 1815ad6cc4c3SAl Viro put_link(nd); 1816ad6cc4c3SAl Viro return NULL; 1817d63ff28fSAl Viro } 1818d63ff28fSAl Viro 18193ddcd056SLinus Torvalds /* 18203ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18213ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18223ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18233ddcd056SLinus Torvalds * for the common case. 182403fa86e9SAl Viro * 182503fa86e9SAl Viro * NOTE: dentry must be what nd->next_seq had been sampled from. 18263ddcd056SLinus Torvalds */ 1827b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1828a4f5b521SAl Viro struct dentry *dentry) 18293ddcd056SLinus Torvalds { 1830cbae4d12SAl Viro struct path path; 1831a4f5b521SAl Viro struct inode *inode; 18323bd8bc89SAl Viro int err = handle_mounts(nd, dentry, &path); 1833cbae4d12SAl Viro 1834cbae4d12SAl Viro if (err < 0) 1835b0417d2cSAl Viro return ERR_PTR(err); 18363bd8bc89SAl Viro inode = path.dentry->d_inode; 1837cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18388c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1839aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18408f64fb1cSAl Viro /* not a symlink or should not follow */ 18413bd8bc89SAl Viro if (nd->flags & LOOKUP_RCU) { 18423bd8bc89SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) 18433bd8bc89SAl Viro return ERR_PTR(-ECHILD); 18443bd8bc89SAl Viro if (unlikely(!inode)) 18453bd8bc89SAl Viro return ERR_PTR(-ENOENT); 18463bd8bc89SAl Viro } else { 1847c99687a0SAl Viro dput(nd->path.dentry); 1848c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1849c99687a0SAl Viro mntput(nd->path.mnt); 1850c99687a0SAl Viro } 1851c99687a0SAl Viro nd->path = path; 18528f64fb1cSAl Viro nd->inode = inode; 185303fa86e9SAl Viro nd->seq = nd->next_seq; 1854b0417d2cSAl Viro return NULL; 18558f64fb1cSAl Viro } 1856a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 185784f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 185803fa86e9SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) 1859b0417d2cSAl Viro return ERR_PTR(-ECHILD); 186084f0cd9eSAl Viro } else { 186184f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 186284f0cd9eSAl Viro mntget(path.mnt); 1863a7f77542SAl Viro } 186403fa86e9SAl Viro return pick_link(nd, &path, inode, flags); 18653ddcd056SLinus Torvalds } 18663ddcd056SLinus Torvalds 1867b16c001dSAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd) 1868957dd41dSAl Viro { 186912487f30SAl Viro struct dentry *parent, *old; 1870957dd41dSAl Viro 187112487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 187212487f30SAl Viro goto in_root; 187312487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18747ef482faSAl Viro struct path path; 1875efe772d6SAl Viro unsigned seq; 18767ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18777ef482faSAl Viro &nd->root, &path, &seq)) 187812487f30SAl Viro goto in_root; 1879efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1880efe772d6SAl Viro return ERR_PTR(-ECHILD); 1881efe772d6SAl Viro nd->path = path; 1882efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1883efe772d6SAl Viro nd->seq = seq; 188403fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach this state 188582ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1886efe772d6SAl Viro return ERR_PTR(-ECHILD); 1887efe772d6SAl Viro /* we know that mountpoint was pinned */ 1888957dd41dSAl Viro } 188912487f30SAl Viro old = nd->path.dentry; 189012487f30SAl Viro parent = old->d_parent; 189103fa86e9SAl Viro nd->next_seq = read_seqcount_begin(&parent->d_seq); 189203fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach this state 189382ef0698SAl Viro if (read_seqcount_retry(&old->d_seq, nd->seq)) 189412487f30SAl Viro return ERR_PTR(-ECHILD); 189512487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 189612487f30SAl Viro return ERR_PTR(-ECHILD); 189712487f30SAl Viro return parent; 189812487f30SAl Viro in_root: 189982ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1900efe772d6SAl Viro return ERR_PTR(-ECHILD); 1901957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19027521f22bSAl Viro return ERR_PTR(-ECHILD); 190303fa86e9SAl Viro nd->next_seq = nd->seq; 190451c6546cSAl Viro return nd->path.dentry; 1905957dd41dSAl Viro } 1906957dd41dSAl Viro 1907b16c001dSAl Viro static struct dentry *follow_dotdot(struct nameidata *nd) 1908957dd41dSAl Viro { 190912487f30SAl Viro struct dentry *parent; 191012487f30SAl Viro 1911957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 191212487f30SAl Viro goto in_root; 191312487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19142aa38470SAl Viro struct path path; 19152aa38470SAl Viro 19162aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19172aa38470SAl Viro &nd->root, &path)) 191812487f30SAl Viro goto in_root; 1919165200d6SAl Viro path_put(&nd->path); 1920165200d6SAl Viro nd->path = path; 19212aa38470SAl Viro nd->inode = path.dentry->d_inode; 1922165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1923165200d6SAl Viro return ERR_PTR(-EXDEV); 192412487f30SAl Viro } 1925957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 192612487f30SAl Viro parent = dget_parent(nd->path.dentry); 1927957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1928957dd41dSAl Viro dput(parent); 19297521f22bSAl Viro return ERR_PTR(-ENOENT); 1930957dd41dSAl Viro } 1931c2df1968SAl Viro return parent; 193212487f30SAl Viro 193312487f30SAl Viro in_root: 1934957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19357521f22bSAl Viro return ERR_PTR(-EXDEV); 193651c6546cSAl Viro return dget(nd->path.dentry); 1937957dd41dSAl Viro } 1938957dd41dSAl Viro 19397521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1940957dd41dSAl Viro { 1941957dd41dSAl Viro if (type == LAST_DOTDOT) { 19427521f22bSAl Viro const char *error = NULL; 1943c2df1968SAl Viro struct dentry *parent; 1944957dd41dSAl Viro 1945957dd41dSAl Viro if (!nd->root.mnt) { 19467521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1947957dd41dSAl Viro if (error) 1948957dd41dSAl Viro return error; 1949957dd41dSAl Viro } 1950957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1951b16c001dSAl Viro parent = follow_dotdot_rcu(nd); 1952957dd41dSAl Viro else 1953b16c001dSAl Viro parent = follow_dotdot(nd); 1954c2df1968SAl Viro if (IS_ERR(parent)) 1955c2df1968SAl Viro return ERR_CAST(parent); 1956a4f5b521SAl Viro error = step_into(nd, WALK_NOFOLLOW, parent); 1957c2df1968SAl Viro if (unlikely(error)) 1958957dd41dSAl Viro return error; 1959957dd41dSAl Viro 1960957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1961957dd41dSAl Viro /* 1962957dd41dSAl Viro * If there was a racing rename or mount along our 1963957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1964957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1965957dd41dSAl Viro * some fallback). 1966957dd41dSAl Viro */ 1967957dd41dSAl Viro smp_rmb(); 196882ef0698SAl Viro if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)) 19697521f22bSAl Viro return ERR_PTR(-EAGAIN); 197082ef0698SAl Viro if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)) 19717521f22bSAl Viro return ERR_PTR(-EAGAIN); 1972957dd41dSAl Viro } 1973957dd41dSAl Viro } 19747521f22bSAl Viro return NULL; 1975957dd41dSAl Viro } 1976957dd41dSAl Viro 197792d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1978ce57dfc1SAl Viro { 1979db3c9adeSAl Viro struct dentry *dentry; 1980ce57dfc1SAl Viro /* 1981ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1982ce57dfc1SAl Viro * to be able to know about the current root directory and 1983ce57dfc1SAl Viro * parent relationships. 1984ce57dfc1SAl Viro */ 19854693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 19861c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 19874693a547SAl Viro put_link(nd); 19887521f22bSAl Viro return handle_dots(nd, nd->last_type); 19894693a547SAl Viro } 19904cb64024SAl Viro dentry = lookup_fast(nd); 199120e34357SAl Viro if (IS_ERR(dentry)) 199292d27016SAl Viro return ERR_CAST(dentry); 199320e34357SAl Viro if (unlikely(!dentry)) { 1994db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1995db3c9adeSAl Viro if (IS_ERR(dentry)) 199692d27016SAl Viro return ERR_CAST(dentry); 199720e34357SAl Viro } 199856676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 199956676ec3SAl Viro put_link(nd); 2000a4f5b521SAl Viro return step_into(nd, flags, dentry); 2001ce57dfc1SAl Viro } 2002ce57dfc1SAl Viro 20031da177e4SLinus Torvalds /* 2004bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2005bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2006bfcfaa77SLinus Torvalds * 2007bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2008bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2009bfcfaa77SLinus Torvalds * fast. 2010bfcfaa77SLinus Torvalds * 2011bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2012bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2013bfcfaa77SLinus Torvalds * crossing operation. 2014bfcfaa77SLinus Torvalds * 2015bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2016bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2017bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2018bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2019bfcfaa77SLinus Torvalds */ 2020bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2021bfcfaa77SLinus Torvalds 2022f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2023bfcfaa77SLinus Torvalds 2024468a9428SGeorge Spelvin #ifdef HASH_MIX 2025bfcfaa77SLinus Torvalds 2026468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2027468a9428SGeorge Spelvin 2028468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20292a18da7aSGeorge Spelvin /* 20302a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20312a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20322a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20332a18da7aSGeorge Spelvin * and no temporaries. 20342a18da7aSGeorge Spelvin * 20352a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20362a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20372a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20382a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20392a18da7aSGeorge Spelvin * 20402a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20412a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20422a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20432a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20442a18da7aSGeorge Spelvin * 20452a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20462a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20472a18da7aSGeorge Spelvin * toggling any given output bit. 20482a18da7aSGeorge Spelvin * 20492a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20502a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20512a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20522a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20532a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20542a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20552a18da7aSGeorge Spelvin * Perfect: 8192 258048 20562a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20572a18da7aSGeorge Spelvin */ 20582a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20592a18da7aSGeorge Spelvin ( x ^= (a), \ 20602a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20612a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20622a18da7aSGeorge Spelvin y *= 9 ) 2063bfcfaa77SLinus Torvalds 20640fed3ac8SGeorge Spelvin /* 20652a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20662a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20672a18da7aSGeorge Spelvin * work done before the hash value is used. 20680fed3ac8SGeorge Spelvin */ 20692a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20700fed3ac8SGeorge Spelvin { 20712a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20722a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20732a18da7aSGeorge Spelvin return y >> 32; 20740fed3ac8SGeorge Spelvin } 20750fed3ac8SGeorge Spelvin 2076bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2077bfcfaa77SLinus Torvalds 20782a18da7aSGeorge Spelvin /* 20792a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 20802a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20812a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 20822a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 20832a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 20842a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 20852a18da7aSGeorge Spelvin * Perfect: 2048 31744 20862a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 20872a18da7aSGeorge Spelvin */ 20882a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20892a18da7aSGeorge Spelvin ( x ^= (a), \ 20902a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 20912a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 20922a18da7aSGeorge Spelvin y *= 9 ) 2093bfcfaa77SLinus Torvalds 20942a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20950fed3ac8SGeorge Spelvin { 20962a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 20972a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 20980fed3ac8SGeorge Spelvin } 20990fed3ac8SGeorge Spelvin 2100bfcfaa77SLinus Torvalds #endif 2101bfcfaa77SLinus Torvalds 21022a18da7aSGeorge Spelvin /* 21032a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21042a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21052a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21062a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21072a18da7aSGeorge Spelvin * finds the delimiter after the name. 21082a18da7aSGeorge Spelvin */ 21098387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2110bfcfaa77SLinus Torvalds { 21118387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2112bfcfaa77SLinus Torvalds 2113bfcfaa77SLinus Torvalds for (;;) { 2114fcfd2fbfSGeorge Spelvin if (!len) 2115fcfd2fbfSGeorge Spelvin goto done; 2116e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2117bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2118bfcfaa77SLinus Torvalds break; 21192a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2120bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2121bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2122bfcfaa77SLinus Torvalds } 21232a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2124bfcfaa77SLinus Torvalds done: 21252a18da7aSGeorge Spelvin return fold_hash(x, y); 2126bfcfaa77SLinus Torvalds } 2127bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2128bfcfaa77SLinus Torvalds 2129fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21308387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2131fcfd2fbfSGeorge Spelvin { 21328387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21338387ff25SLinus Torvalds unsigned long adata, mask, len; 2134fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2135fcfd2fbfSGeorge Spelvin 21368387ff25SLinus Torvalds len = 0; 21378387ff25SLinus Torvalds goto inside; 21388387ff25SLinus Torvalds 2139fcfd2fbfSGeorge Spelvin do { 21402a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2141fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21428387ff25SLinus Torvalds inside: 2143fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2144fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2145fcfd2fbfSGeorge Spelvin 2146fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2147fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21482a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2149fcfd2fbfSGeorge Spelvin 21502a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2151fcfd2fbfSGeorge Spelvin } 2152fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2153fcfd2fbfSGeorge Spelvin 2154bfcfaa77SLinus Torvalds /* 2155bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2156d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2157bfcfaa77SLinus Torvalds */ 21588387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2159bfcfaa77SLinus Torvalds { 21608387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21618387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 216236126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2163bfcfaa77SLinus Torvalds 21648387ff25SLinus Torvalds len = 0; 21658387ff25SLinus Torvalds goto inside; 21668387ff25SLinus Torvalds 2167bfcfaa77SLinus Torvalds do { 21682a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2169bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21708387ff25SLinus Torvalds inside: 2171e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 217236126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 217336126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2174bfcfaa77SLinus Torvalds 217536126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 217636126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 217736126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21782a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 217936126f8fSLinus Torvalds 21802a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2181bfcfaa77SLinus Torvalds } 2182bfcfaa77SLinus Torvalds 21832a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2184bfcfaa77SLinus Torvalds 2185fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 21868387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 21870145acc2SLinus Torvalds { 21888387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 21890145acc2SLinus Torvalds while (len--) 2190fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 21910145acc2SLinus Torvalds return end_name_hash(hash); 21920145acc2SLinus Torvalds } 2193ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 21940145acc2SLinus Torvalds 2195fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21968387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2197fcfd2fbfSGeorge Spelvin { 21988387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2199fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2200fcfd2fbfSGeorge Spelvin 2201fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2202e0ab7af9SGeorge Spelvin while (c) { 2203fcfd2fbfSGeorge Spelvin len++; 2204fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2205fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2206e0ab7af9SGeorge Spelvin } 2207fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2208fcfd2fbfSGeorge Spelvin } 2209f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2210fcfd2fbfSGeorge Spelvin 22113ddcd056SLinus Torvalds /* 2212200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2213200e9ef7SLinus Torvalds * one character. 2214200e9ef7SLinus Torvalds */ 22158387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2216200e9ef7SLinus Torvalds { 22178387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2218200e9ef7SLinus Torvalds unsigned long len = 0, c; 2219200e9ef7SLinus Torvalds 2220200e9ef7SLinus Torvalds c = (unsigned char)*name; 2221200e9ef7SLinus Torvalds do { 2222200e9ef7SLinus Torvalds len++; 2223200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2224200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2225200e9ef7SLinus Torvalds } while (c && c != '/'); 2226d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2227200e9ef7SLinus Torvalds } 2228200e9ef7SLinus Torvalds 2229bfcfaa77SLinus Torvalds #endif 2230bfcfaa77SLinus Torvalds 2231200e9ef7SLinus Torvalds /* 22321da177e4SLinus Torvalds * Name resolution. 2233ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2234ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22351da177e4SLinus Torvalds * 2236ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2237ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22381da177e4SLinus Torvalds */ 22396de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22401da177e4SLinus Torvalds { 2241d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22421da177e4SLinus Torvalds int err; 22431da177e4SLinus Torvalds 2244b4c03536SAl Viro nd->last_type = LAST_ROOT; 2245c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22469b5858e9SAl Viro if (IS_ERR(name)) 22479b5858e9SAl Viro return PTR_ERR(name); 22481da177e4SLinus Torvalds while (*name=='/') 22491da177e4SLinus Torvalds name++; 22501a97d899SAl Viro if (!*name) { 22511a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22529e18f10aSAl Viro return 0; 22531a97d899SAl Viro } 22541da177e4SLinus Torvalds 22551da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22561da177e4SLinus Torvalds for(;;) { 2257549c7297SChristian Brauner struct user_namespace *mnt_userns; 225892d27016SAl Viro const char *link; 2259d6bb3e90SLinus Torvalds u64 hash_len; 2260fe479a58SAl Viro int type; 22611da177e4SLinus Torvalds 2262549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2263549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22641da177e4SLinus Torvalds if (err) 22653595e234SAl Viro return err; 22661da177e4SLinus Torvalds 22678387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22681da177e4SLinus Torvalds 2269fe479a58SAl Viro type = LAST_NORM; 2270d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2271fe479a58SAl Viro case 2: 2272200e9ef7SLinus Torvalds if (name[1] == '.') { 2273fe479a58SAl Viro type = LAST_DOTDOT; 2274bcba1e7dSAl Viro nd->state |= ND_JUMPED; 227516c2cd71SAl Viro } 2276fe479a58SAl Viro break; 2277fe479a58SAl Viro case 1: 2278fe479a58SAl Viro type = LAST_DOT; 2279fe479a58SAl Viro } 22805a202bcdSAl Viro if (likely(type == LAST_NORM)) { 22815a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2282bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 22835a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2284a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2285da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 22865a202bcdSAl Viro if (err < 0) 22873595e234SAl Viro return err; 2288d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2289d6bb3e90SLinus Torvalds name = this.name; 22905a202bcdSAl Viro } 22915a202bcdSAl Viro } 2292fe479a58SAl Viro 2293d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2294d6bb3e90SLinus Torvalds nd->last.name = name; 22955f4a6a69SAl Viro nd->last_type = type; 22965f4a6a69SAl Viro 2297d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2298d6bb3e90SLinus Torvalds if (!*name) 2299bdf6cbf1SAl Viro goto OK; 2300200e9ef7SLinus Torvalds /* 2301200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2302200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2303200e9ef7SLinus Torvalds */ 2304200e9ef7SLinus Torvalds do { 2305d6bb3e90SLinus Torvalds name++; 2306d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 23078620c238SAl Viro if (unlikely(!*name)) { 23088620c238SAl Viro OK: 2309d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2310c108837eSAl Viro if (!depth) { 2311a2bd096fSChristian Brauner nd->dir_vfsuid = i_uid_into_vfsuid(mnt_userns, nd->inode); 23120f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2313c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 23148620c238SAl Viro return 0; 2315c108837eSAl Viro } 23168620c238SAl Viro /* last component of nested symlink */ 2317d8d4611aSAl Viro name = nd->stack[--depth].name; 23188c4efe22SAl Viro link = walk_component(nd, 0); 23191c4ff1a8SAl Viro } else { 23201c4ff1a8SAl Viro /* not the last component */ 23218c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 23228620c238SAl Viro } 232392d27016SAl Viro if (unlikely(link)) { 232492d27016SAl Viro if (IS_ERR(link)) 232592d27016SAl Viro return PTR_ERR(link); 232692d27016SAl Viro /* a symlink to follow */ 2327d8d4611aSAl Viro nd->stack[depth++].name = name; 232892d27016SAl Viro name = link; 23299e18f10aSAl Viro continue; 233048c8b0c5SAl Viro } 233197242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 233297242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2333e36cffedSJens Axboe if (!try_to_unlazy(nd)) 233497242f99SAl Viro return -ECHILD; 233597242f99SAl Viro } 23363595e234SAl Viro return -ENOTDIR; 23375f4a6a69SAl Viro } 2338ce57dfc1SAl Viro } 233997242f99SAl Viro } 23401da177e4SLinus Torvalds 2341edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2342c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 234331e6b01fSNick Piggin { 2344740a1678SAleksa Sarai int error; 2345c8a53ee5SAl Viro const char *s = nd->name->name; 234631e6b01fSNick Piggin 23476c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23486c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23496c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23506c6ec2b0SJens Axboe 2351c0eb027eSLinus Torvalds if (!*s) 2352c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2353edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2354edc2b1daSAl Viro rcu_read_lock(); 235503fa86e9SAl Viro else 235603fa86e9SAl Viro nd->seq = nd->next_seq = 0; 2357c0eb027eSLinus Torvalds 2358bcba1e7dSAl Viro nd->flags = flags; 2359bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2360ab87f9a5SAleksa Sarai 2361ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2362ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2363ab87f9a5SAleksa Sarai smp_rmb(); 2364ab87f9a5SAleksa Sarai 2365bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2366b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2367b18825a7SDavid Howells struct inode *inode = root->d_inode; 236893893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2369368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23705b6ca027SAl Viro nd->path = nd->root; 23715b6ca027SAl Viro nd->inode = inode; 23725b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2373ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23748f47a016SAl Viro nd->root_seq = nd->seq; 23755b6ca027SAl Viro } else { 23765b6ca027SAl Viro path_get(&nd->path); 23775b6ca027SAl Viro } 2378368ee9baSAl Viro return s; 23795b6ca027SAl Viro } 23805b6ca027SAl Viro 238131e6b01fSNick Piggin nd->root.mnt = NULL; 238231e6b01fSNick Piggin 23838db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 23848db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2385740a1678SAleksa Sarai error = nd_jump_root(nd); 2386740a1678SAleksa Sarai if (unlikely(error)) 2387740a1678SAleksa Sarai return ERR_PTR(error); 2388ef55d917SAl Viro return s; 23898db52c7eSAleksa Sarai } 23908db52c7eSAleksa Sarai 23918db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 23928db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2393e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 239431e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2395c28cc364SNick Piggin unsigned seq; 239631e6b01fSNick Piggin 2397c28cc364SNick Piggin do { 2398c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 239931e6b01fSNick Piggin nd->path = fs->pwd; 2400ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2401c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2402c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2403e41f7d4eSAl Viro } else { 2404e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2405ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2406e41f7d4eSAl Viro } 240731e6b01fSNick Piggin } else { 2408582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2409c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 241031e6b01fSNick Piggin struct dentry *dentry; 241131e6b01fSNick Piggin 24122903ff01SAl Viro if (!f.file) 2413368ee9baSAl Viro return ERR_PTR(-EBADF); 241431e6b01fSNick Piggin 24152903ff01SAl Viro dentry = f.file->f_path.dentry; 241631e6b01fSNick Piggin 2417edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 24182903ff01SAl Viro fdput(f); 2419368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2420f52e0c11SAl Viro } 24212903ff01SAl Viro 24222903ff01SAl Viro nd->path = f.file->f_path; 2423e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 242434a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 242534a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24265590ff0dSUlrich Drepper } else { 24272903ff01SAl Viro path_get(&nd->path); 242834a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24291da177e4SLinus Torvalds } 243034a26b99SAl Viro fdput(f); 2431e41f7d4eSAl Viro } 24328db52c7eSAleksa Sarai 2433adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2434adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2435adb21d2bSAleksa Sarai nd->root = nd->path; 2436adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2437adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2438adb21d2bSAleksa Sarai } else { 2439adb21d2bSAleksa Sarai path_get(&nd->root); 2440bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2441adb21d2bSAleksa Sarai } 2442adb21d2bSAleksa Sarai } 2443adb21d2bSAleksa Sarai return s; 24449b4a9b14SAl Viro } 24459b4a9b14SAl Viro 24461ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 244795fa25d9SAl Viro { 2448bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2449bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2450bd92d7feSAl Viro 2451c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2452bd92d7feSAl Viro } 2453bd92d7feSAl Viro 24544f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24554f757f3cSAl Viro { 2456c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2457db3c9adeSAl Viro dget(nd->path.dentry); 245803fa86e9SAl Viro nd->next_seq = nd->seq; 2459a4f5b521SAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry)); 24604f757f3cSAl Viro } 24614f757f3cSAl Viro 24629b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2463c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24649b4a9b14SAl Viro { 2465c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2466bd92d7feSAl Viro int err; 246731e6b01fSNick Piggin 24689b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24694f757f3cSAl Viro err = handle_lookup_down(nd); 24705f336e72SAl Viro if (unlikely(err < 0)) 24715f336e72SAl Viro s = ERR_PTR(err); 24724f757f3cSAl Viro } 24734f757f3cSAl Viro 24741ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24751ccac622SAl Viro (s = lookup_last(nd)) != NULL) 24761ccac622SAl Viro ; 24774f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 24784f0ed93fSAl Viro err = handle_lookup_down(nd); 2479bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 24804f0ed93fSAl Viro } 24819f1fafeeSAl Viro if (!err) 24829f1fafeeSAl Viro err = complete_walk(nd); 2483bd92d7feSAl Viro 2484deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2485deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2486bd23a539SAl Viro err = -ENOTDIR; 2487625b6d10SAl Viro if (!err) { 2488625b6d10SAl Viro *path = nd->path; 2489625b6d10SAl Viro nd->path.mnt = NULL; 2490625b6d10SAl Viro nd->path.dentry = NULL; 2491625b6d10SAl Viro } 2492deb106c6SAl Viro terminate_walk(nd); 2493bd92d7feSAl Viro return err; 249431e6b01fSNick Piggin } 249531e6b01fSNick Piggin 2496794ebceaSStephen Brennan int filename_lookup(int dfd, struct filename *name, unsigned flags, 24979ad1aaa6SAl Viro struct path *path, struct path *root) 2498873f1eedSJeff Layton { 2499894bc8c4SAl Viro int retval; 25009883d185SAl Viro struct nameidata nd; 2501abc9f5beSAl Viro if (IS_ERR(name)) 2502abc9f5beSAl Viro return PTR_ERR(name); 250306422964SAl Viro set_nameidata(&nd, dfd, name, root); 2504c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2505873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2506c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2507873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2508c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2509873f1eedSJeff Layton 2510873f1eedSJeff Layton if (likely(!retval)) 2511161aff1dSAl Viro audit_inode(name, path->dentry, 2512161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 25139883d185SAl Viro restore_nameidata(); 2514020250f3SDmitry Kadashev return retval; 2515020250f3SDmitry Kadashev } 2516020250f3SDmitry Kadashev 25178bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2518c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2519391172c4SAl Viro struct path *parent) 25208bcb77faSAl Viro { 2521c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25229b5858e9SAl Viro int err = link_path_walk(s, nd); 25238bcb77faSAl Viro if (!err) 25248bcb77faSAl Viro err = complete_walk(nd); 2525391172c4SAl Viro if (!err) { 2526391172c4SAl Viro *parent = nd->path; 2527391172c4SAl Viro nd->path.mnt = NULL; 2528391172c4SAl Viro nd->path.dentry = NULL; 2529391172c4SAl Viro } 2530deb106c6SAl Viro terminate_walk(nd); 25318bcb77faSAl Viro return err; 25328bcb77faSAl Viro } 25338bcb77faSAl Viro 25340766ec82SStephen Brennan /* Note: this does not consume "name" */ 2535c5f563f9SAl Viro static int filename_parentat(int dfd, struct filename *name, 2536391172c4SAl Viro unsigned int flags, struct path *parent, 2537391172c4SAl Viro struct qstr *last, int *type) 25388bcb77faSAl Viro { 25398bcb77faSAl Viro int retval; 25409883d185SAl Viro struct nameidata nd; 25418bcb77faSAl Viro 25425c31b6ceSAl Viro if (IS_ERR(name)) 25430ee50b47SDmitry Kadashev return PTR_ERR(name); 254406422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2545c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25468bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2547c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25488bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2549c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2550391172c4SAl Viro if (likely(!retval)) { 2551391172c4SAl Viro *last = nd.last; 2552391172c4SAl Viro *type = nd.last_type; 2553c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2554391172c4SAl Viro } 25559883d185SAl Viro restore_nameidata(); 25560ee50b47SDmitry Kadashev return retval; 25570ee50b47SDmitry Kadashev } 25580ee50b47SDmitry Kadashev 255979714f72SAl Viro /* does lookup, returns the object with parent locked */ 25600766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25615590ff0dSUlrich Drepper { 25625c31b6ceSAl Viro struct dentry *d; 2563391172c4SAl Viro struct qstr last; 25640ee50b47SDmitry Kadashev int type, error; 256551689104SPaul Moore 2566c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25670ee50b47SDmitry Kadashev if (error) 25680ee50b47SDmitry Kadashev return ERR_PTR(error); 25695c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2570391172c4SAl Viro path_put(path); 25715c31b6ceSAl Viro return ERR_PTR(-EINVAL); 257279714f72SAl Viro } 25735955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2574391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 257579714f72SAl Viro if (IS_ERR(d)) { 25765955102cSAl Viro inode_unlock(path->dentry->d_inode); 2577391172c4SAl Viro path_put(path); 257879714f72SAl Viro } 257979714f72SAl Viro return d; 25805590ff0dSUlrich Drepper } 25815590ff0dSUlrich Drepper 25820766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 25830766ec82SStephen Brennan { 25840766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 25850766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 25860766ec82SStephen Brennan 25870766ec82SStephen Brennan putname(filename); 25880766ec82SStephen Brennan return res; 25890766ec82SStephen Brennan } 25900766ec82SStephen Brennan 2591d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2592d1811465SAl Viro { 2593794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2594794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2595794ebceaSStephen Brennan 2596794ebceaSStephen Brennan putname(filename); 2597794ebceaSStephen Brennan return ret; 2598794ebceaSStephen Brennan 2599d1811465SAl Viro } 26004d359507SAl Viro EXPORT_SYMBOL(kern_path); 2601d1811465SAl Viro 260216f18200SJosef 'Jeff' Sipek /** 260316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 260416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 260516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 260616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 260716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2608e0a01249SAl Viro * @path: pointer to struct path to fill 260916f18200SJosef 'Jeff' Sipek */ 261016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 261116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2612e0a01249SAl Viro struct path *path) 261316f18200SJosef 'Jeff' Sipek { 2614794ebceaSStephen Brennan struct filename *filename; 26159ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2616794ebceaSStephen Brennan int ret; 2617794ebceaSStephen Brennan 2618794ebceaSStephen Brennan filename = getname_kernel(name); 26199ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2620794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2621794ebceaSStephen Brennan putname(filename); 2622794ebceaSStephen Brennan return ret; 262316f18200SJosef 'Jeff' Sipek } 26244d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 262516f18200SJosef 'Jeff' Sipek 2626c2fd68b6SChristian Brauner static int lookup_one_common(struct user_namespace *mnt_userns, 2627c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2628c2fd68b6SChristian Brauner struct qstr *this) 26293c95f0dcSAl Viro { 26303c95f0dcSAl Viro this->name = name; 26313c95f0dcSAl Viro this->len = len; 26323c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26333c95f0dcSAl Viro if (!len) 26343c95f0dcSAl Viro return -EACCES; 26353c95f0dcSAl Viro 26363c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26373c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26383c95f0dcSAl Viro return -EACCES; 26393c95f0dcSAl Viro } 26403c95f0dcSAl Viro 26413c95f0dcSAl Viro while (len--) { 26423c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26433c95f0dcSAl Viro if (c == '/' || c == '\0') 26443c95f0dcSAl Viro return -EACCES; 26453c95f0dcSAl Viro } 26463c95f0dcSAl Viro /* 26473c95f0dcSAl Viro * See if the low-level filesystem might want 26483c95f0dcSAl Viro * to use its own hash.. 26493c95f0dcSAl Viro */ 26503c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26513c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26523c95f0dcSAl Viro if (err < 0) 26533c95f0dcSAl Viro return err; 26543c95f0dcSAl Viro } 26553c95f0dcSAl Viro 2656c2fd68b6SChristian Brauner return inode_permission(mnt_userns, base->d_inode, MAY_EXEC); 26573c95f0dcSAl Viro } 26583c95f0dcSAl Viro 2659eead1911SChristoph Hellwig /** 26600da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26610da0b7fdSDavid Howells * @name: pathname component to lookup 26620da0b7fdSDavid Howells * @base: base directory to lookup from 26630da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26640da0b7fdSDavid Howells * 26650da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26660da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26670da0b7fdSDavid Howells * 26680da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26690da0b7fdSDavid Howells * not be called by generic code. 26700da0b7fdSDavid Howells * 26710da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26720da0b7fdSDavid Howells */ 26730da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26740da0b7fdSDavid Howells { 26750da0b7fdSDavid Howells struct qstr this; 26760da0b7fdSDavid Howells int err; 26770da0b7fdSDavid Howells 26780da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26790da0b7fdSDavid Howells 2680c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 26810da0b7fdSDavid Howells if (err) 26820da0b7fdSDavid Howells return ERR_PTR(err); 26830da0b7fdSDavid Howells 26840da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 26850da0b7fdSDavid Howells } 26860da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 26870da0b7fdSDavid Howells 26880da0b7fdSDavid Howells /** 2689a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2690eead1911SChristoph Hellwig * @name: pathname component to lookup 2691eead1911SChristoph Hellwig * @base: base directory to lookup from 2692eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2693eead1911SChristoph Hellwig * 2694a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 26959e7543e9SAl Viro * not be called by generic code. 2696bbddca8eSNeilBrown * 2697bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2698eead1911SChristoph Hellwig */ 2699057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2700057f6c01SJames Morris { 27018613a209SAl Viro struct dentry *dentry; 2702057f6c01SJames Morris struct qstr this; 2703cda309deSMiklos Szeredi int err; 2704057f6c01SJames Morris 27055955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27062f9092e1SDavid Woodhouse 2707c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2708cda309deSMiklos Szeredi if (err) 2709cda309deSMiklos Szeredi return ERR_PTR(err); 2710cda309deSMiklos Szeredi 27118613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27128613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2713057f6c01SJames Morris } 27144d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2715057f6c01SJames Morris 2716bbddca8eSNeilBrown /** 2717c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 2718c2fd68b6SChristian Brauner * @mnt_userns: user namespace of the mount the lookup is performed from 2719c2fd68b6SChristian Brauner * @name: pathname component to lookup 2720c2fd68b6SChristian Brauner * @base: base directory to lookup from 2721c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2722c2fd68b6SChristian Brauner * 2723c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2724c2fd68b6SChristian Brauner * not be called by generic code. 2725c2fd68b6SChristian Brauner * 2726c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2727c2fd68b6SChristian Brauner */ 2728c2fd68b6SChristian Brauner struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name, 2729c2fd68b6SChristian Brauner struct dentry *base, int len) 2730c2fd68b6SChristian Brauner { 2731c2fd68b6SChristian Brauner struct dentry *dentry; 2732c2fd68b6SChristian Brauner struct qstr this; 2733c2fd68b6SChristian Brauner int err; 2734c2fd68b6SChristian Brauner 2735c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2736c2fd68b6SChristian Brauner 2737c2fd68b6SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 2738c2fd68b6SChristian Brauner if (err) 2739c2fd68b6SChristian Brauner return ERR_PTR(err); 2740c2fd68b6SChristian Brauner 2741c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2742c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2743c2fd68b6SChristian Brauner } 2744c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2745c2fd68b6SChristian Brauner 2746c2fd68b6SChristian Brauner /** 274700675017SChristian Brauner * lookup_one_unlocked - filesystem helper to lookup single pathname component 274800675017SChristian Brauner * @mnt_userns: idmapping of the mount the lookup is performed from 274900675017SChristian Brauner * @name: pathname component to lookup 275000675017SChristian Brauner * @base: base directory to lookup from 275100675017SChristian Brauner * @len: maximum length @len should be interpreted to 275200675017SChristian Brauner * 275300675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 275400675017SChristian Brauner * not be called by generic code. 275500675017SChristian Brauner * 275600675017SChristian Brauner * Unlike lookup_one_len, it should be called without the parent 275700675017SChristian Brauner * i_mutex held, and will take the i_mutex itself if necessary. 275800675017SChristian Brauner */ 275900675017SChristian Brauner struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns, 276000675017SChristian Brauner const char *name, struct dentry *base, 276100675017SChristian Brauner int len) 276200675017SChristian Brauner { 276300675017SChristian Brauner struct qstr this; 276400675017SChristian Brauner int err; 276500675017SChristian Brauner struct dentry *ret; 276600675017SChristian Brauner 276700675017SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 276800675017SChristian Brauner if (err) 276900675017SChristian Brauner return ERR_PTR(err); 277000675017SChristian Brauner 277100675017SChristian Brauner ret = lookup_dcache(&this, base, 0); 277200675017SChristian Brauner if (!ret) 277300675017SChristian Brauner ret = lookup_slow(&this, base, 0); 277400675017SChristian Brauner return ret; 277500675017SChristian Brauner } 277600675017SChristian Brauner EXPORT_SYMBOL(lookup_one_unlocked); 277700675017SChristian Brauner 277800675017SChristian Brauner /** 277900675017SChristian Brauner * lookup_one_positive_unlocked - filesystem helper to lookup single 278000675017SChristian Brauner * pathname component 278100675017SChristian Brauner * @mnt_userns: idmapping of the mount the lookup is performed from 278200675017SChristian Brauner * @name: pathname component to lookup 278300675017SChristian Brauner * @base: base directory to lookup from 278400675017SChristian Brauner * @len: maximum length @len should be interpreted to 278500675017SChristian Brauner * 278600675017SChristian Brauner * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns 278700675017SChristian Brauner * known positive or ERR_PTR(). This is what most of the users want. 278800675017SChristian Brauner * 278900675017SChristian Brauner * Note that pinned negative with unlocked parent _can_ become positive at any 279000675017SChristian Brauner * time, so callers of lookup_one_unlocked() need to be very careful; pinned 279100675017SChristian Brauner * positives have >d_inode stable, so this one avoids such problems. 279200675017SChristian Brauner * 279300675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 279400675017SChristian Brauner * not be called by generic code. 279500675017SChristian Brauner * 279600675017SChristian Brauner * The helper should be called without i_mutex held. 279700675017SChristian Brauner */ 279800675017SChristian Brauner struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns, 279900675017SChristian Brauner const char *name, 280000675017SChristian Brauner struct dentry *base, int len) 280100675017SChristian Brauner { 280200675017SChristian Brauner struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len); 280300675017SChristian Brauner 280400675017SChristian Brauner if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 280500675017SChristian Brauner dput(ret); 280600675017SChristian Brauner ret = ERR_PTR(-ENOENT); 280700675017SChristian Brauner } 280800675017SChristian Brauner return ret; 280900675017SChristian Brauner } 281000675017SChristian Brauner EXPORT_SYMBOL(lookup_one_positive_unlocked); 281100675017SChristian Brauner 281200675017SChristian Brauner /** 2813bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2814bbddca8eSNeilBrown * @name: pathname component to lookup 2815bbddca8eSNeilBrown * @base: base directory to lookup from 2816bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2817bbddca8eSNeilBrown * 2818bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2819bbddca8eSNeilBrown * not be called by generic code. 2820bbddca8eSNeilBrown * 2821bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2822bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2823bbddca8eSNeilBrown */ 2824bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2825bbddca8eSNeilBrown struct dentry *base, int len) 2826bbddca8eSNeilBrown { 282700675017SChristian Brauner return lookup_one_unlocked(&init_user_ns, name, base, len); 2828bbddca8eSNeilBrown } 2829bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2830bbddca8eSNeilBrown 28316c2d4798SAl Viro /* 28326c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 28336c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 28346c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 28356c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 28366c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 28376c2d4798SAl Viro * this one avoids such problems. 28386c2d4798SAl Viro */ 28396c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28406c2d4798SAl Viro struct dentry *base, int len) 28416c2d4798SAl Viro { 284200675017SChristian Brauner return lookup_one_positive_unlocked(&init_user_ns, name, base, len); 28436c2d4798SAl Viro } 28446c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28456c2d4798SAl Viro 2846eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2847eedf265aSEric W. Biederman int path_pts(struct path *path) 2848eedf265aSEric W. Biederman { 2849eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2850eedf265aSEric W. Biederman * the input path. 2851eedf265aSEric W. Biederman */ 2852a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2853a6a7eb76SAl Viro struct dentry *child; 285419f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2855eedf265aSEric W. Biederman 2856a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2857a6a7eb76SAl Viro dput(parent); 285863b27720SAl Viro return -ENOENT; 2859a6a7eb76SAl Viro } 286063b27720SAl Viro dput(path->dentry); 286163b27720SAl Viro path->dentry = parent; 2862eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2863eedf265aSEric W. Biederman if (!child) 2864eedf265aSEric W. Biederman return -ENOENT; 2865eedf265aSEric W. Biederman 2866eedf265aSEric W. Biederman path->dentry = child; 2867eedf265aSEric W. Biederman dput(parent); 286819f6028aSAl Viro follow_down(path); 2869eedf265aSEric W. Biederman return 0; 2870eedf265aSEric W. Biederman } 2871eedf265aSEric W. Biederman #endif 2872eedf265aSEric W. Biederman 28731fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 28741fa1e7f6SAndy Whitcroft struct path *path, int *empty) 28751da177e4SLinus Torvalds { 2876794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2877794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2878794ebceaSStephen Brennan 2879794ebceaSStephen Brennan putname(filename); 2880794ebceaSStephen Brennan return ret; 28811da177e4SLinus Torvalds } 2882b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 28831fa1e7f6SAndy Whitcroft 2884ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2885ba73d987SChristian Brauner struct inode *inode) 28861da177e4SLinus Torvalds { 28878e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2888da9592edSDavid Howells 2889a2bd096fSChristian Brauner if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), fsuid)) 28901da177e4SLinus Torvalds return 0; 2891a2bd096fSChristian Brauner if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, dir), fsuid)) 28921da177e4SLinus Torvalds return 0; 2893ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 28941da177e4SLinus Torvalds } 2895cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 28961da177e4SLinus Torvalds 28971da177e4SLinus Torvalds /* 28981da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 28991da177e4SLinus Torvalds * whether the type of victim is right. 29001da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 29011da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 29021da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 29031da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 29041da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 29051da177e4SLinus Torvalds * a. be owner of dir, or 29061da177e4SLinus Torvalds * b. be owner of victim, or 29071da177e4SLinus Torvalds * c. have CAP_FOWNER capability 29081da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 29091da177e4SLinus Torvalds * links pointing to it. 29100bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 29110bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 29120bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 29130bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 29140bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 29151da177e4SLinus Torvalds * nfs_async_unlink(). 29161da177e4SLinus Torvalds */ 2917ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2918ba73d987SChristian Brauner struct dentry *victim, bool isdir) 29191da177e4SLinus Torvalds { 292063afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 29211da177e4SLinus Torvalds int error; 29221da177e4SLinus Torvalds 2923b18825a7SDavid Howells if (d_is_negative(victim)) 29241da177e4SLinus Torvalds return -ENOENT; 2925b18825a7SDavid Howells BUG_ON(!inode); 29261da177e4SLinus Torvalds 29271da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2928593d1ce8SEric W. Biederman 2929593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2930a2bd096fSChristian Brauner if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) || 2931a2bd096fSChristian Brauner !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode))) 2932593d1ce8SEric W. Biederman return -EOVERFLOW; 2933593d1ce8SEric W. Biederman 29344fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29351da177e4SLinus Torvalds 2936ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29371da177e4SLinus Torvalds if (error) 29381da177e4SLinus Torvalds return error; 29391da177e4SLinus Torvalds if (IS_APPEND(dir)) 29401da177e4SLinus Torvalds return -EPERM; 2941b18825a7SDavid Howells 2942ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2943ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2944ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 29451da177e4SLinus Torvalds return -EPERM; 29461da177e4SLinus Torvalds if (isdir) { 294744b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29481da177e4SLinus Torvalds return -ENOTDIR; 29491da177e4SLinus Torvalds if (IS_ROOT(victim)) 29501da177e4SLinus Torvalds return -EBUSY; 295144b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29521da177e4SLinus Torvalds return -EISDIR; 29531da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29541da177e4SLinus Torvalds return -ENOENT; 29551da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29561da177e4SLinus Torvalds return -EBUSY; 29571da177e4SLinus Torvalds return 0; 29581da177e4SLinus Torvalds } 29591da177e4SLinus Torvalds 29601da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29611da177e4SLinus Torvalds * dir. 29621da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29631da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29641da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2965036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2966036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2967036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 29681da177e4SLinus Torvalds */ 2969ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2970ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 29711da177e4SLinus Torvalds { 297214e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 29731da177e4SLinus Torvalds if (child->d_inode) 29741da177e4SLinus Torvalds return -EEXIST; 29751da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29761da177e4SLinus Torvalds return -ENOENT; 29778e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 2978036d5236SEric W. Biederman return -EOVERFLOW; 29798e538913SChristian Brauner 2980ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29811da177e4SLinus Torvalds } 29821da177e4SLinus Torvalds 29831da177e4SLinus Torvalds /* 29841da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 29851da177e4SLinus Torvalds */ 29861da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 29871da177e4SLinus Torvalds { 29881da177e4SLinus Torvalds struct dentry *p; 29891da177e4SLinus Torvalds 29901da177e4SLinus Torvalds if (p1 == p2) { 29915955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29921da177e4SLinus Torvalds return NULL; 29931da177e4SLinus Torvalds } 29941da177e4SLinus Torvalds 2995fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 29961da177e4SLinus Torvalds 2997e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2998e2761a11SOGAWA Hirofumi if (p) { 29995955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30005955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 30011da177e4SLinus Torvalds return p; 30021da177e4SLinus Torvalds } 30031da177e4SLinus Torvalds 3004e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 3005e2761a11SOGAWA Hirofumi if (p) { 30065955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30075955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 30081da177e4SLinus Torvalds return p; 30091da177e4SLinus Torvalds } 30101da177e4SLinus Torvalds 30115955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30125955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 30131da177e4SLinus Torvalds return NULL; 30141da177e4SLinus Torvalds } 30154d359507SAl Viro EXPORT_SYMBOL(lock_rename); 30161da177e4SLinus Torvalds 30171da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 30181da177e4SLinus Torvalds { 30195955102cSAl Viro inode_unlock(p1->d_inode); 30201da177e4SLinus Torvalds if (p1 != p2) { 30215955102cSAl Viro inode_unlock(p2->d_inode); 3022fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 30231da177e4SLinus Torvalds } 30241da177e4SLinus Torvalds } 30254d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 30261da177e4SLinus Torvalds 30276521f891SChristian Brauner /** 30281639a49cSYang Xu * mode_strip_umask - handle vfs umask stripping 30291639a49cSYang Xu * @dir: parent directory of the new inode 30301639a49cSYang Xu * @mode: mode of the new inode to be created in @dir 30311639a49cSYang Xu * 30321639a49cSYang Xu * Umask stripping depends on whether or not the filesystem supports POSIX 30331639a49cSYang Xu * ACLs. If the filesystem doesn't support it umask stripping is done directly 30341639a49cSYang Xu * in here. If the filesystem does support POSIX ACLs umask stripping is 30351639a49cSYang Xu * deferred until the filesystem calls posix_acl_create(). 30361639a49cSYang Xu * 30371639a49cSYang Xu * Returns: mode 30381639a49cSYang Xu */ 30391639a49cSYang Xu static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode) 30401639a49cSYang Xu { 30411639a49cSYang Xu if (!IS_POSIXACL(dir)) 30421639a49cSYang Xu mode &= ~current_umask(); 30431639a49cSYang Xu return mode; 30441639a49cSYang Xu } 30451639a49cSYang Xu 30461639a49cSYang Xu /** 30471639a49cSYang Xu * vfs_prepare_mode - prepare the mode to be used for a new inode 30481639a49cSYang Xu * @mnt_userns: user namespace of the mount the inode was found from 30491639a49cSYang Xu * @dir: parent directory of the new inode 30501639a49cSYang Xu * @mode: mode of the new inode 30511639a49cSYang Xu * @mask_perms: allowed permission by the vfs 30521639a49cSYang Xu * @type: type of file to be created 30531639a49cSYang Xu * 30541639a49cSYang Xu * This helper consolidates and enforces vfs restrictions on the @mode of a new 30551639a49cSYang Xu * object to be created. 30561639a49cSYang Xu * 30571639a49cSYang Xu * Umask stripping depends on whether the filesystem supports POSIX ACLs (see 30581639a49cSYang Xu * the kernel documentation for mode_strip_umask()). Moving umask stripping 30591639a49cSYang Xu * after setgid stripping allows the same ordering for both non-POSIX ACL and 30601639a49cSYang Xu * POSIX ACL supporting filesystems. 30611639a49cSYang Xu * 30621639a49cSYang Xu * Note that it's currently valid for @type to be 0 if a directory is created. 30631639a49cSYang Xu * Filesystems raise that flag individually and we need to check whether each 30641639a49cSYang Xu * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a 30651639a49cSYang Xu * non-zero type. 30661639a49cSYang Xu * 30671639a49cSYang Xu * Returns: mode to be passed to the filesystem 30681639a49cSYang Xu */ 30691639a49cSYang Xu static inline umode_t vfs_prepare_mode(struct user_namespace *mnt_userns, 30701639a49cSYang Xu const struct inode *dir, umode_t mode, 30711639a49cSYang Xu umode_t mask_perms, umode_t type) 30721639a49cSYang Xu { 30731639a49cSYang Xu mode = mode_strip_sgid(mnt_userns, dir, mode); 30741639a49cSYang Xu mode = mode_strip_umask(dir, mode); 30751639a49cSYang Xu 30761639a49cSYang Xu /* 30771639a49cSYang Xu * Apply the vfs mandated allowed permission mask and set the type of 30781639a49cSYang Xu * file to be created before we call into the filesystem. 30791639a49cSYang Xu */ 30801639a49cSYang Xu mode &= (mask_perms & ~S_IFMT); 30811639a49cSYang Xu mode |= (type & S_IFMT); 30821639a49cSYang Xu 30831639a49cSYang Xu return mode; 30841639a49cSYang Xu } 30851639a49cSYang Xu 30861639a49cSYang Xu /** 30876521f891SChristian Brauner * vfs_create - create new file 30886521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 30896521f891SChristian Brauner * @dir: inode of @dentry 30906521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 30916521f891SChristian Brauner * @mode: mode of the new file 30926521f891SChristian Brauner * @want_excl: whether the file must not yet exist 30936521f891SChristian Brauner * 30946521f891SChristian Brauner * Create a new file. 30956521f891SChristian Brauner * 30966521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 30976521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 30986521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 30996521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 31006521f891SChristian Brauner * raw inode simply passs init_user_ns. 31016521f891SChristian Brauner */ 31026521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 31036521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 31041da177e4SLinus Torvalds { 31056521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 31061da177e4SLinus Torvalds if (error) 31071da177e4SLinus Torvalds return error; 31081da177e4SLinus Torvalds 3109acfa4380SAl Viro if (!dir->i_op->create) 31101da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 31111639a49cSYang Xu 31121639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IALLUGO, S_IFREG); 31131da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 31141da177e4SLinus Torvalds if (error) 31151da177e4SLinus Torvalds return error; 3116549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 3117a74574aaSStephen Smalley if (!error) 3118f38aa942SAmy Griffis fsnotify_create(dir, dentry); 31191da177e4SLinus Torvalds return error; 31201da177e4SLinus Torvalds } 31214d359507SAl Viro EXPORT_SYMBOL(vfs_create); 31221da177e4SLinus Torvalds 31238e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 31248e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 31258e6c848eSAl Viro void *arg) 31268e6c848eSAl Viro { 31278e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 3128ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 31298e6c848eSAl Viro if (error) 31308e6c848eSAl Viro return error; 31318e6c848eSAl Viro 31328e6c848eSAl Viro mode &= S_IALLUGO; 31338e6c848eSAl Viro mode |= S_IFREG; 31348e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 31358e6c848eSAl Viro if (error) 31368e6c848eSAl Viro return error; 31378e6c848eSAl Viro error = f(dentry, mode, arg); 31388e6c848eSAl Viro if (!error) 31398e6c848eSAl Viro fsnotify_create(dir, dentry); 31408e6c848eSAl Viro return error; 31418e6c848eSAl Viro } 31428e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 31438e6c848eSAl Viro 3144a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3145a2982cc9SEric W. Biederman { 3146a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3147a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3148a2982cc9SEric W. Biederman } 3149a2982cc9SEric W. Biederman 3150ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 3151ba73d987SChristian Brauner int acc_mode, int flag) 31521da177e4SLinus Torvalds { 31533fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 31541da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 31551da177e4SLinus Torvalds int error; 31561da177e4SLinus Torvalds 31571da177e4SLinus Torvalds if (!inode) 31581da177e4SLinus Torvalds return -ENOENT; 31591da177e4SLinus Torvalds 3160c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3161c8fe8f30SChristoph Hellwig case S_IFLNK: 31621da177e4SLinus Torvalds return -ELOOP; 3163c8fe8f30SChristoph Hellwig case S_IFDIR: 3164fc4177beSKees Cook if (acc_mode & MAY_WRITE) 31651da177e4SLinus Torvalds return -EISDIR; 3166fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3167fc4177beSKees Cook return -EACCES; 3168c8fe8f30SChristoph Hellwig break; 3169c8fe8f30SChristoph Hellwig case S_IFBLK: 3170c8fe8f30SChristoph Hellwig case S_IFCHR: 3171a2982cc9SEric W. Biederman if (!may_open_dev(path)) 31721da177e4SLinus Torvalds return -EACCES; 3173633fb6acSKees Cook fallthrough; 3174c8fe8f30SChristoph Hellwig case S_IFIFO: 3175c8fe8f30SChristoph Hellwig case S_IFSOCK: 3176633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3177633fb6acSKees Cook return -EACCES; 31781da177e4SLinus Torvalds flag &= ~O_TRUNC; 3179c8fe8f30SChristoph Hellwig break; 31800fd338b2SKees Cook case S_IFREG: 31810fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 31820fd338b2SKees Cook return -EACCES; 31830fd338b2SKees Cook break; 31844a3fd211SDave Hansen } 3185b41572e9SDave Hansen 3186ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 3187b41572e9SDave Hansen if (error) 3188b41572e9SDave Hansen return error; 31896146f0d5SMimi Zohar 31901da177e4SLinus Torvalds /* 31911da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 31921da177e4SLinus Torvalds */ 31931da177e4SLinus Torvalds if (IS_APPEND(inode)) { 31948737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 31957715b521SAl Viro return -EPERM; 31961da177e4SLinus Torvalds if (flag & O_TRUNC) 31977715b521SAl Viro return -EPERM; 31981da177e4SLinus Torvalds } 31991da177e4SLinus Torvalds 32001da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3201ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 32027715b521SAl Viro return -EPERM; 32031da177e4SLinus Torvalds 3204f3c7691eSJ. Bruce Fields return 0; 32057715b521SAl Viro } 32067715b521SAl Viro 3207549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 32087715b521SAl Viro { 3209f0bb5aafSAl Viro const struct path *path = &filp->f_path; 32107715b521SAl Viro struct inode *inode = path->dentry->d_inode; 32117715b521SAl Viro int error = get_write_access(inode); 32121da177e4SLinus Torvalds if (error) 32137715b521SAl Viro return error; 3214482e0007SJeff Layton 32153350607dSGünther Noack error = security_file_truncate(filp); 32161da177e4SLinus Torvalds if (!error) { 3217549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3218d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3219e1181ee6SJeff Layton filp); 32201da177e4SLinus Torvalds } 32211da177e4SLinus Torvalds put_write_access(inode); 3222acd0c935SMimi Zohar return error; 32231da177e4SLinus Torvalds } 32241da177e4SLinus Torvalds 3225d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3226d57999e1SDave Hansen { 32278a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 32288a5e929dSAl Viro flag--; 3229d57999e1SDave Hansen return flag; 3230d57999e1SDave Hansen } 3231d57999e1SDave Hansen 3232ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3233ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3234ba73d987SChristian Brauner umode_t mode) 3235d18e9008SMiklos Szeredi { 3236d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3237d18e9008SMiklos Szeredi if (error) 3238d18e9008SMiklos Szeredi return error; 3239d18e9008SMiklos Szeredi 32408e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) 32411328c727SSeth Forshee return -EOVERFLOW; 32421328c727SSeth Forshee 3243ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 324447291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3245d18e9008SMiklos Szeredi if (error) 3246d18e9008SMiklos Szeredi return error; 3247d18e9008SMiklos Szeredi 3248d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3249d18e9008SMiklos Szeredi } 3250d18e9008SMiklos Szeredi 32511acf0af9SDavid Howells /* 32521acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 32531acf0af9SDavid Howells * dentry. 32541acf0af9SDavid Howells * 32551acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 32561acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 32571acf0af9SDavid Howells * 325800a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 325900a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 326000a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 32611acf0af9SDavid Howells * 32621acf0af9SDavid Howells * Returns an error code otherwise. 32631acf0af9SDavid Howells */ 3264239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3265239eb983SAl Viro struct file *file, 32663ec2eef1SAl Viro int open_flag, umode_t mode) 3267d18e9008SMiklos Szeredi { 3268d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3269d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3270d18e9008SMiklos Szeredi int error; 3271d18e9008SMiklos Szeredi 3272d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3273d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3274d18e9008SMiklos Szeredi 327530d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 327630d90494SAl Viro file->f_path.mnt = nd->path.mnt; 32770fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 327844907d79SAl Viro open_to_namei_flags(open_flag), mode); 32796fbd0714SAl Viro d_lookup_done(dentry); 3280384f26e2SAl Viro if (!error) { 328164e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 32826fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 32836fb968cdSAl Viro dput(dentry); 32846fb968cdSAl Viro dentry = dget(file->f_path.dentry); 32856fb968cdSAl Viro } 328664e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 32872675a4ebSAl Viro error = -EIO; 3288384f26e2SAl Viro } else { 328930d90494SAl Viro if (file->f_path.dentry) { 3290d18e9008SMiklos Szeredi dput(dentry); 3291d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 329210c64ceaSAl Viro } 3293239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3294a01e718fSAl Viro error = -ENOENT; 3295d18e9008SMiklos Szeredi } 3296d18e9008SMiklos Szeredi } 3297239eb983SAl Viro if (error) { 3298d18e9008SMiklos Szeredi dput(dentry); 3299239eb983SAl Viro dentry = ERR_PTR(error); 3300239eb983SAl Viro } 3301239eb983SAl Viro return dentry; 3302d18e9008SMiklos Szeredi } 3303d18e9008SMiklos Szeredi 330431e6b01fSNick Piggin /* 33051acf0af9SDavid Howells * Look up and maybe create and open the last component. 3306d58ffd35SMiklos Szeredi * 330700a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3308d58ffd35SMiklos Szeredi * 330900a07c15SAl Viro * Returns 0 on success, that is, if 331000a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 331100a07c15SAl Viro * the file was not completely opened at this time, though lookups and 331200a07c15SAl Viro * creations were performed. 331300a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 331400a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 331500a07c15SAl Viro * hadn't been specified. 33161acf0af9SDavid Howells * 331700a07c15SAl Viro * An error code is returned on failure. 3318d58ffd35SMiklos Szeredi */ 3319da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3320d58ffd35SMiklos Szeredi const struct open_flags *op, 33213ec2eef1SAl Viro bool got_write) 3322d58ffd35SMiklos Szeredi { 3323549c7297SChristian Brauner struct user_namespace *mnt_userns; 3324d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 332554ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 33261643b43fSAl Viro int open_flag = op->open_flag; 3327d58ffd35SMiklos Szeredi struct dentry *dentry; 33281643b43fSAl Viro int error, create_error = 0; 33291643b43fSAl Viro umode_t mode = op->mode; 33306fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3331d58ffd35SMiklos Szeredi 3332ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3333da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3334d58ffd35SMiklos Szeredi 333573a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 33366fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 33376fbd0714SAl Viro for (;;) { 33386fbd0714SAl Viro if (!dentry) { 33396fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3340d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3341da5ebf5aSAl Viro return dentry; 33426fbd0714SAl Viro } 33436fbd0714SAl Viro if (d_in_lookup(dentry)) 33446fbd0714SAl Viro break; 3345d58ffd35SMiklos Szeredi 33466fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 33476fbd0714SAl Viro if (likely(error > 0)) 33486fbd0714SAl Viro break; 33496fbd0714SAl Viro if (error) 33506fbd0714SAl Viro goto out_dput; 33516fbd0714SAl Viro d_invalidate(dentry); 33526fbd0714SAl Viro dput(dentry); 33536fbd0714SAl Viro dentry = NULL; 33546fbd0714SAl Viro } 33556fbd0714SAl Viro if (dentry->d_inode) { 3356d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3357da5ebf5aSAl Viro return dentry; 33586c51e513SAl Viro } 3359d18e9008SMiklos Szeredi 33601643b43fSAl Viro /* 33611643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 33621643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 33631643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 33641643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 33651643b43fSAl Viro * 33661643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 33671643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 33681643b43fSAl Viro */ 336999a4a90cSAl Viro if (unlikely(!got_write)) 337099a4a90cSAl Viro open_flag &= ~O_TRUNC; 3371549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 33721643b43fSAl Viro if (open_flag & O_CREAT) { 337399a4a90cSAl Viro if (open_flag & O_EXCL) 337499a4a90cSAl Viro open_flag &= ~O_TRUNC; 33751639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir->d_inode, mode, mode, mode); 337699a4a90cSAl Viro if (likely(got_write)) 3377549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3378ba73d987SChristian Brauner dentry, mode); 337999a4a90cSAl Viro else 338099a4a90cSAl Viro create_error = -EROFS; 338199a4a90cSAl Viro } 338299a4a90cSAl Viro if (create_error) 33831643b43fSAl Viro open_flag &= ~O_CREAT; 33841643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3385d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3386da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3387da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3388da5ebf5aSAl Viro return dentry; 3389239eb983SAl Viro } 339054ef4872SMiklos Szeredi 33916fbd0714SAl Viro if (d_in_lookup(dentry)) { 339212fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 339312fa5e24SAl Viro nd->flags); 33946fbd0714SAl Viro d_lookup_done(dentry); 339512fa5e24SAl Viro if (unlikely(res)) { 339612fa5e24SAl Viro if (IS_ERR(res)) { 339712fa5e24SAl Viro error = PTR_ERR(res); 339812fa5e24SAl Viro goto out_dput; 339912fa5e24SAl Viro } 340012fa5e24SAl Viro dput(dentry); 340112fa5e24SAl Viro dentry = res; 340212fa5e24SAl Viro } 340354ef4872SMiklos Szeredi } 340454ef4872SMiklos Szeredi 3405d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 34061643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 340773a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3408ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3409ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3410ce8644fcSAl Viro error = -EACCES; 3411d58ffd35SMiklos Szeredi goto out_dput; 341264894cf8SAl Viro } 3413549c7297SChristian Brauner 3414549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3415549c7297SChristian Brauner mode, open_flag & O_EXCL); 3416d58ffd35SMiklos Szeredi if (error) 3417d58ffd35SMiklos Szeredi goto out_dput; 3418d58ffd35SMiklos Szeredi } 34191643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 34201643b43fSAl Viro error = create_error; 3421d58ffd35SMiklos Szeredi goto out_dput; 3422d58ffd35SMiklos Szeredi } 3423da5ebf5aSAl Viro return dentry; 3424d58ffd35SMiklos Szeredi 3425d58ffd35SMiklos Szeredi out_dput: 3426d58ffd35SMiklos Szeredi dput(dentry); 3427da5ebf5aSAl Viro return ERR_PTR(error); 3428d58ffd35SMiklos Szeredi } 3429d58ffd35SMiklos Szeredi 3430c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 34313ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3432fb1cc555SAl Viro { 3433a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3434ca344a89SAl Viro int open_flag = op->open_flag; 343564894cf8SAl Viro bool got_write = false; 3436da5ebf5aSAl Viro struct dentry *dentry; 3437b0417d2cSAl Viro const char *res; 3438fb1cc555SAl Viro 3439c3e380b0SAl Viro nd->flags |= op->intent; 3440c3e380b0SAl Viro 3441bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 344256676ec3SAl Viro if (nd->depth) 344356676ec3SAl Viro put_link(nd); 3444ff326a32SAl Viro return handle_dots(nd, nd->last_type); 34451f36f774SAl Viro } 3446a2c36b45SAl Viro 3447ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3448fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3449fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3450fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 34514cb64024SAl Viro dentry = lookup_fast(nd); 345220e34357SAl Viro if (IS_ERR(dentry)) 34531ccac622SAl Viro return ERR_CAST(dentry); 345420e34357SAl Viro if (likely(dentry)) 345571574865SMiklos Szeredi goto finish_lookup; 345671574865SMiklos Szeredi 34576583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3458b6183df7SMiklos Szeredi } else { 3459fe2d35ffSAl Viro /* create side of things */ 346072287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3461e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3462e36cffedSJens Axboe return ERR_PTR(-ECHILD); 346372287417SAl Viro } 3464c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 34651f36f774SAl Viro /* trailing slashes? */ 3466deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 34671ccac622SAl Viro return ERR_PTR(-EISDIR); 3468b6183df7SMiklos Szeredi } 34691f36f774SAl Viro 34709cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3471e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 347264894cf8SAl Viro /* 347364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 347464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 347564894cf8SAl Viro * dropping this one anyway. 347664894cf8SAl Viro */ 347764894cf8SAl Viro } 34789cf843e3SAl Viro if (open_flag & O_CREAT) 34795955102cSAl Viro inode_lock(dir->d_inode); 34809cf843e3SAl Viro else 34819cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3482da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3483f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3484f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 34859cf843e3SAl Viro if (open_flag & O_CREAT) 34865955102cSAl Viro inode_unlock(dir->d_inode); 34879cf843e3SAl Viro else 34889cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3489fb1cc555SAl Viro 3490c981a482SAl Viro if (got_write) 349159e96e65SAl Viro mnt_drop_write(nd->path.mnt); 34926c0d46c4SAl Viro 349359e96e65SAl Viro if (IS_ERR(dentry)) 349459e96e65SAl Viro return ERR_CAST(dentry); 349559e96e65SAl Viro 3496973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3497e73cabffSAl Viro dput(nd->path.dentry); 3498e73cabffSAl Viro nd->path.dentry = dentry; 3499c981a482SAl Viro return NULL; 3500fb1cc555SAl Viro } 3501fb1cc555SAl Viro 350220e34357SAl Viro finish_lookup: 350356676ec3SAl Viro if (nd->depth) 350456676ec3SAl Viro put_link(nd); 3505a4f5b521SAl Viro res = step_into(nd, WALK_TRAILING, dentry); 3506ff326a32SAl Viro if (unlikely(res)) 35071ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3508b0417d2cSAl Viro return res; 35091ccac622SAl Viro } 351031d1726dSAl Viro 3511c981a482SAl Viro /* 3512c981a482SAl Viro * Handle the last step of open() 3513c981a482SAl Viro */ 3514c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3515c981a482SAl Viro struct file *file, const struct open_flags *op) 3516c981a482SAl Viro { 3517549c7297SChristian Brauner struct user_namespace *mnt_userns; 3518c981a482SAl Viro int open_flag = op->open_flag; 3519c981a482SAl Viro bool do_truncate; 3520c981a482SAl Viro int acc_mode; 3521c981a482SAl Viro int error; 3522c981a482SAl Viro 3523ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3524ff326a32SAl Viro error = complete_walk(nd); 3525ff326a32SAl Viro if (error) 3526ff326a32SAl Viro return error; 3527ff326a32SAl Viro } 3528973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 352976ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3530549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 353130aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3532b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3533b94e0b32SAl Viro return -EEXIST; 353430aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3535c5971b8cSAl Viro return -EISDIR; 3536549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 353730aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 353830aba665SSalvatore Mesoraca if (unlikely(error)) 3539c5971b8cSAl Viro return error; 354030aba665SSalvatore Mesoraca } 354144b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3542c5971b8cSAl Viro return -ENOTDIR; 35436c0d46c4SAl Viro 35448795e7d4SAl Viro do_truncate = false; 35458795e7d4SAl Viro acc_mode = op->acc_mode; 35465a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 35475a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 35485a2d3eddSAl Viro open_flag &= ~O_TRUNC; 35495a2d3eddSAl Viro acc_mode = 0; 35508795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 35510f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 35520f9d1a10SAl Viro if (error) 3553c5971b8cSAl Viro return error; 35548795e7d4SAl Viro do_truncate = true; 35550f9d1a10SAl Viro } 3556549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 35578795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3558ae2bb293SAl Viro error = vfs_open(&nd->path, file); 35598795e7d4SAl Viro if (!error) 35606035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 35618795e7d4SAl Viro if (!error && do_truncate) 3562549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3563c80567c8SAl Viro if (unlikely(error > 0)) { 3564c80567c8SAl Viro WARN_ON(1); 3565c80567c8SAl Viro error = -EINVAL; 3566c80567c8SAl Viro } 35678795e7d4SAl Viro if (do_truncate) 35680f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3569c5971b8cSAl Viro return error; 3570fb1cc555SAl Viro } 3571fb1cc555SAl Viro 35726521f891SChristian Brauner /** 35736521f891SChristian Brauner * vfs_tmpfile - create tmpfile 35746521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 35756521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 35766521f891SChristian Brauner * @mode: mode of the new tmpfile 35772111c3c0SRandy Dunlap * @open_flag: flags 35786521f891SChristian Brauner * 35796521f891SChristian Brauner * Create a temporary file. 35806521f891SChristian Brauner * 35816521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 35826521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 35836521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 35846521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 35856521f891SChristian Brauner * raw inode simply passs init_user_ns. 35866521f891SChristian Brauner */ 35879751b338SMiklos Szeredi static int vfs_tmpfile(struct user_namespace *mnt_userns, 35889751b338SMiklos Szeredi const struct path *parentpath, 35899751b338SMiklos Szeredi struct file *file, umode_t mode) 3590af7bd4dcSAmir Goldstein { 35919751b338SMiklos Szeredi struct dentry *child; 35929751b338SMiklos Szeredi struct inode *dir = d_inode(parentpath->dentry); 3593af7bd4dcSAmir Goldstein struct inode *inode; 3594af7bd4dcSAmir Goldstein int error; 3595406c706cSPeter Griffin int open_flag = file->f_flags; 3596af7bd4dcSAmir Goldstein 3597af7bd4dcSAmir Goldstein /* we want directory to be writable */ 35986521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3599af7bd4dcSAmir Goldstein if (error) 36009751b338SMiklos Szeredi return error; 3601af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 36029751b338SMiklos Szeredi return -EOPNOTSUPP; 36039751b338SMiklos Szeredi child = d_alloc(parentpath->dentry, &slash_name); 3604af7bd4dcSAmir Goldstein if (unlikely(!child)) 36059751b338SMiklos Szeredi return -ENOMEM; 36069751b338SMiklos Szeredi file->f_path.mnt = parentpath->mnt; 36079751b338SMiklos Szeredi file->f_path.dentry = child; 36081639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode); 3609863f144fSMiklos Szeredi error = dir->i_op->tmpfile(mnt_userns, dir, file, mode); 36109751b338SMiklos Szeredi dput(child); 3611af7bd4dcSAmir Goldstein if (error) 36129751b338SMiklos Szeredi return error; 36139751b338SMiklos Szeredi /* Don't check for other permissions, the inode was just created */ 36149751b338SMiklos Szeredi error = may_open(mnt_userns, &file->f_path, 0, file->f_flags); 36159751b338SMiklos Szeredi if (error) 36169751b338SMiklos Szeredi return error; 36179751b338SMiklos Szeredi inode = file_inode(file); 3618406c706cSPeter Griffin if (!(open_flag & O_EXCL)) { 3619af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3620af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3621af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3622af7bd4dcSAmir Goldstein } 3623a2d2329eSChristian Brauner ima_post_create_tmpfile(mnt_userns, inode); 36249751b338SMiklos Szeredi return 0; 3625af7bd4dcSAmir Goldstein } 3626af7bd4dcSAmir Goldstein 362722873deaSMiklos Szeredi /** 362822873deaSMiklos Szeredi * vfs_tmpfile_open - open a tmpfile for kernel internal use 362922873deaSMiklos Szeredi * @mnt_userns: user namespace of the mount the inode was found from 363022873deaSMiklos Szeredi * @parentpath: path of the base directory 363122873deaSMiklos Szeredi * @mode: mode of the new tmpfile 363222873deaSMiklos Szeredi * @open_flag: flags 363322873deaSMiklos Szeredi * @cred: credentials for open 363422873deaSMiklos Szeredi * 363522873deaSMiklos Szeredi * Create and open a temporary file. The file is not accounted in nr_files, 363622873deaSMiklos Szeredi * hence this is only for kernel internal use, and must not be installed into 363722873deaSMiklos Szeredi * file tables or such. 363822873deaSMiklos Szeredi */ 363922873deaSMiklos Szeredi struct file *vfs_tmpfile_open(struct user_namespace *mnt_userns, 364022873deaSMiklos Szeredi const struct path *parentpath, 364122873deaSMiklos Szeredi umode_t mode, int open_flag, const struct cred *cred) 364222873deaSMiklos Szeredi { 364322873deaSMiklos Szeredi struct file *file; 364422873deaSMiklos Szeredi int error; 364522873deaSMiklos Szeredi 36469751b338SMiklos Szeredi file = alloc_empty_file_noaccount(open_flag, cred); 36479751b338SMiklos Szeredi if (!IS_ERR(file)) { 36489751b338SMiklos Szeredi error = vfs_tmpfile(mnt_userns, parentpath, file, mode); 36499751b338SMiklos Szeredi if (error) { 36509751b338SMiklos Szeredi fput(file); 365122873deaSMiklos Szeredi file = ERR_PTR(error); 36529751b338SMiklos Szeredi } 36539751b338SMiklos Szeredi } 365422873deaSMiklos Szeredi return file; 365522873deaSMiklos Szeredi } 365622873deaSMiklos Szeredi EXPORT_SYMBOL(vfs_tmpfile_open); 3657648fa861SAl Viro 3658c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 365960545d0dSAl Viro const struct open_flags *op, 36603ec2eef1SAl Viro struct file *file) 366160545d0dSAl Viro { 36626521f891SChristian Brauner struct user_namespace *mnt_userns; 3663625b6d10SAl Viro struct path path; 3664c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 36659751b338SMiklos Szeredi 366660545d0dSAl Viro if (unlikely(error)) 366760545d0dSAl Viro return error; 3668625b6d10SAl Viro error = mnt_want_write(path.mnt); 366960545d0dSAl Viro if (unlikely(error)) 367060545d0dSAl Viro goto out; 36716521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 36729751b338SMiklos Szeredi error = vfs_tmpfile(mnt_userns, &path, file, op->mode); 36739751b338SMiklos Szeredi if (error) 367460545d0dSAl Viro goto out2; 36759751b338SMiklos Szeredi audit_inode(nd->name, file->f_path.dentry, 0); 367660545d0dSAl Viro out2: 3677625b6d10SAl Viro mnt_drop_write(path.mnt); 367860545d0dSAl Viro out: 3679625b6d10SAl Viro path_put(&path); 368060545d0dSAl Viro return error; 368160545d0dSAl Viro } 368260545d0dSAl Viro 36836ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 36846ac08709SAl Viro { 36856ac08709SAl Viro struct path path; 36866ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 36876ac08709SAl Viro if (!error) { 36886ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3689ae2bb293SAl Viro error = vfs_open(&path, file); 36906ac08709SAl Viro path_put(&path); 36916ac08709SAl Viro } 36926ac08709SAl Viro return error; 36936ac08709SAl Viro } 36946ac08709SAl Viro 3695c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3696c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 36971da177e4SLinus Torvalds { 369830d90494SAl Viro struct file *file; 369913aab428SAl Viro int error; 370031e6b01fSNick Piggin 3701ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 37021afc99beSAl Viro if (IS_ERR(file)) 37031afc99beSAl Viro return file; 370431e6b01fSNick Piggin 3705bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 37063ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 37075f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 37086ac08709SAl Viro error = do_o_path(nd, flags, file); 37095f336e72SAl Viro } else { 37105f336e72SAl Viro const char *s = path_init(nd, flags); 37113bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3712c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 37131ccac622SAl Viro ; 3714c5971b8cSAl Viro if (!error) 3715c5971b8cSAl Viro error = do_open(nd, file, op); 3716deb106c6SAl Viro terminate_walk(nd); 37175f336e72SAl Viro } 37187c1c01ecSAl Viro if (likely(!error)) { 3719aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 37207c1c01ecSAl Viro return file; 37217c1c01ecSAl Viro WARN_ON(1); 37227c1c01ecSAl Viro error = -EINVAL; 3723015c3bbcSMiklos Szeredi } 37247c1c01ecSAl Viro fput(file); 37252675a4ebSAl Viro if (error == -EOPENSTALE) { 37262675a4ebSAl Viro if (flags & LOOKUP_RCU) 37272675a4ebSAl Viro error = -ECHILD; 37282675a4ebSAl Viro else 37292675a4ebSAl Viro error = -ESTALE; 37302675a4ebSAl Viro } 37317c1c01ecSAl Viro return ERR_PTR(error); 3732de459215SKirill Korotaev } 37331da177e4SLinus Torvalds 3734669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3735f9652e10SAl Viro const struct open_flags *op) 373613aab428SAl Viro { 37379883d185SAl Viro struct nameidata nd; 3738f9652e10SAl Viro int flags = op->lookup_flags; 373913aab428SAl Viro struct file *filp; 374013aab428SAl Viro 374106422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3742c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 374313aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3744c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 374513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3746c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 37479883d185SAl Viro restore_nameidata(); 374813aab428SAl Viro return filp; 374913aab428SAl Viro } 375013aab428SAl Viro 3751ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3752f9652e10SAl Viro const char *name, const struct open_flags *op) 375373d049a4SAl Viro { 37549883d185SAl Viro struct nameidata nd; 375573d049a4SAl Viro struct file *file; 375651689104SPaul Moore struct filename *filename; 3757bcba1e7dSAl Viro int flags = op->lookup_flags; 375873d049a4SAl Viro 3759ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 376073d049a4SAl Viro return ERR_PTR(-ELOOP); 376173d049a4SAl Viro 376251689104SPaul Moore filename = getname_kernel(name); 3763a1c83681SViresh Kumar if (IS_ERR(filename)) 376451689104SPaul Moore return ERR_CAST(filename); 376551689104SPaul Moore 376606422964SAl Viro set_nameidata(&nd, -1, filename, root); 3767c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 376873d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3769c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 377073d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3771c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 37729883d185SAl Viro restore_nameidata(); 377351689104SPaul Moore putname(filename); 377473d049a4SAl Viro return file; 377573d049a4SAl Viro } 377673d049a4SAl Viro 3777b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 37781ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 37791da177e4SLinus Torvalds { 3780c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3781391172c4SAl Viro struct qstr last; 3782b3d4650dSNeilBrown bool want_dir = lookup_flags & LOOKUP_DIRECTORY; 3783b3d4650dSNeilBrown unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; 3784b3d4650dSNeilBrown unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; 3785391172c4SAl Viro int type; 3786c30dabfeSJan Kara int err2; 37871ac12b4bSJeff Layton int error; 37881ac12b4bSJeff Layton 3789b3d4650dSNeilBrown error = filename_parentat(dfd, name, reval_flag, path, &last, &type); 37900ee50b47SDmitry Kadashev if (error) 37910ee50b47SDmitry Kadashev return ERR_PTR(error); 37921da177e4SLinus Torvalds 3793c663e5d8SChristoph Hellwig /* 3794c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3795c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3796c663e5d8SChristoph Hellwig */ 37975c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3798ed75e95dSAl Viro goto out; 3799c663e5d8SChristoph Hellwig 3800c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3801391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3802c663e5d8SChristoph Hellwig /* 3803b3d4650dSNeilBrown * Do the final lookup. Suppress 'create' if there is a trailing 3804b3d4650dSNeilBrown * '/', and a directory wasn't requested. 3805c663e5d8SChristoph Hellwig */ 3806b3d4650dSNeilBrown if (last.name[last.len] && !want_dir) 3807b3d4650dSNeilBrown create_flags = 0; 38085955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3809b3d4650dSNeilBrown dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags); 38101da177e4SLinus Torvalds if (IS_ERR(dentry)) 3811a8104a9fSAl Viro goto unlock; 3812c663e5d8SChristoph Hellwig 3813a8104a9fSAl Viro error = -EEXIST; 3814b18825a7SDavid Howells if (d_is_positive(dentry)) 3815a8104a9fSAl Viro goto fail; 3816b18825a7SDavid Howells 3817c663e5d8SChristoph Hellwig /* 3818c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3819c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3820c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3821c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3822c663e5d8SChristoph Hellwig */ 3823b3d4650dSNeilBrown if (unlikely(!create_flags)) { 3824a8104a9fSAl Viro error = -ENOENT; 3825ed75e95dSAl Viro goto fail; 3826e9baf6e5SAl Viro } 3827c30dabfeSJan Kara if (unlikely(err2)) { 3828c30dabfeSJan Kara error = err2; 3829a8104a9fSAl Viro goto fail; 3830c30dabfeSJan Kara } 3831e9baf6e5SAl Viro return dentry; 38321da177e4SLinus Torvalds fail: 3833a8104a9fSAl Viro dput(dentry); 3834a8104a9fSAl Viro dentry = ERR_PTR(error); 3835a8104a9fSAl Viro unlock: 38365955102cSAl Viro inode_unlock(path->dentry->d_inode); 3837c30dabfeSJan Kara if (!err2) 3838391172c4SAl Viro mnt_drop_write(path->mnt); 3839ed75e95dSAl Viro out: 3840391172c4SAl Viro path_put(path); 3841ed75e95dSAl Viro return dentry; 3842dae6ad8fSAl Viro } 3843fa14a0b8SAl Viro 3844fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3845fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3846fa14a0b8SAl Viro { 3847b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3848b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3849b4a4f213SStephen Brennan 3850b4a4f213SStephen Brennan putname(filename); 3851b4a4f213SStephen Brennan return res; 3852fa14a0b8SAl Viro } 3853dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3854dae6ad8fSAl Viro 3855921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3856921a1650SAl Viro { 3857921a1650SAl Viro dput(dentry); 38585955102cSAl Viro inode_unlock(path->dentry->d_inode); 3859a8104a9fSAl Viro mnt_drop_write(path->mnt); 3860921a1650SAl Viro path_put(path); 3861921a1650SAl Viro } 3862921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3863921a1650SAl Viro 3864520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 38651ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3866dae6ad8fSAl Viro { 3867b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3868b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3869b4a4f213SStephen Brennan 3870b4a4f213SStephen Brennan putname(filename); 3871b4a4f213SStephen Brennan return res; 3872dae6ad8fSAl Viro } 3873dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3874dae6ad8fSAl Viro 38756521f891SChristian Brauner /** 38766521f891SChristian Brauner * vfs_mknod - create device node or file 38776521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 38786521f891SChristian Brauner * @dir: inode of @dentry 38796521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 38806521f891SChristian Brauner * @mode: mode of the new device node or file 38816521f891SChristian Brauner * @dev: device number of device to create 38826521f891SChristian Brauner * 38836521f891SChristian Brauner * Create a device node or file. 38846521f891SChristian Brauner * 38856521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 38866521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 38876521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 38886521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 38896521f891SChristian Brauner * raw inode simply passs init_user_ns. 38906521f891SChristian Brauner */ 38916521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 38926521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 38931da177e4SLinus Torvalds { 3894a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 38956521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 38961da177e4SLinus Torvalds 38971da177e4SLinus Torvalds if (error) 38981da177e4SLinus Torvalds return error; 38991da177e4SLinus Torvalds 3900a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3901a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 39021da177e4SLinus Torvalds return -EPERM; 39031da177e4SLinus Torvalds 3904acfa4380SAl Viro if (!dir->i_op->mknod) 39051da177e4SLinus Torvalds return -EPERM; 39061da177e4SLinus Torvalds 39071639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode); 390808ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 390908ce5f16SSerge E. Hallyn if (error) 391008ce5f16SSerge E. Hallyn return error; 391108ce5f16SSerge E. Hallyn 39121da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 39131da177e4SLinus Torvalds if (error) 39141da177e4SLinus Torvalds return error; 39151da177e4SLinus Torvalds 3916549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3917a74574aaSStephen Smalley if (!error) 3918f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39191da177e4SLinus Torvalds return error; 39201da177e4SLinus Torvalds } 39214d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 39221da177e4SLinus Torvalds 3923f69aac00SAl Viro static int may_mknod(umode_t mode) 3924463c3197SDave Hansen { 3925463c3197SDave Hansen switch (mode & S_IFMT) { 3926463c3197SDave Hansen case S_IFREG: 3927463c3197SDave Hansen case S_IFCHR: 3928463c3197SDave Hansen case S_IFBLK: 3929463c3197SDave Hansen case S_IFIFO: 3930463c3197SDave Hansen case S_IFSOCK: 3931463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3932463c3197SDave Hansen return 0; 3933463c3197SDave Hansen case S_IFDIR: 3934463c3197SDave Hansen return -EPERM; 3935463c3197SDave Hansen default: 3936463c3197SDave Hansen return -EINVAL; 3937463c3197SDave Hansen } 3938463c3197SDave Hansen } 3939463c3197SDave Hansen 394045f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 394187c4e192SDominik Brodowski unsigned int dev) 39421da177e4SLinus Torvalds { 39436521f891SChristian Brauner struct user_namespace *mnt_userns; 39441da177e4SLinus Torvalds struct dentry *dentry; 3945dae6ad8fSAl Viro struct path path; 3946dae6ad8fSAl Viro int error; 3947972567f1SJeff Layton unsigned int lookup_flags = 0; 39481da177e4SLinus Torvalds 39498e4bfca1SAl Viro error = may_mknod(mode); 39508e4bfca1SAl Viro if (error) 39517797251bSDmitry Kadashev goto out1; 3952972567f1SJeff Layton retry: 3953b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 39547797251bSDmitry Kadashev error = PTR_ERR(dentry); 3955dae6ad8fSAl Viro if (IS_ERR(dentry)) 39567797251bSDmitry Kadashev goto out1; 39572ad94ae6SAl Viro 39581639a49cSYang Xu error = security_path_mknod(&path, dentry, 39591639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode), dev); 3960be6d3e56SKentaro Takeda if (error) 39617797251bSDmitry Kadashev goto out2; 39626521f891SChristian Brauner 39636521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 39641da177e4SLinus Torvalds switch (mode & S_IFMT) { 39651da177e4SLinus Torvalds case 0: case S_IFREG: 39666521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 39676521f891SChristian Brauner dentry, mode, true); 396805d1a717SMimi Zohar if (!error) 3969a2d2329eSChristian Brauner ima_post_path_mknod(mnt_userns, dentry); 39701da177e4SLinus Torvalds break; 39711da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 39726521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 39736521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 39741da177e4SLinus Torvalds break; 39751da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 39766521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 39776521f891SChristian Brauner dentry, mode, 0); 39781da177e4SLinus Torvalds break; 39791da177e4SLinus Torvalds } 39807797251bSDmitry Kadashev out2: 3981921a1650SAl Viro done_path_create(&path, dentry); 3982972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3983972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3984972567f1SJeff Layton goto retry; 3985972567f1SJeff Layton } 39867797251bSDmitry Kadashev out1: 39877797251bSDmitry Kadashev putname(name); 39881da177e4SLinus Torvalds return error; 39891da177e4SLinus Torvalds } 39901da177e4SLinus Torvalds 399187c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 399287c4e192SDominik Brodowski unsigned int, dev) 399387c4e192SDominik Brodowski { 39947797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 399587c4e192SDominik Brodowski } 399687c4e192SDominik Brodowski 39978208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 39985590ff0dSUlrich Drepper { 39997797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 40005590ff0dSUlrich Drepper } 40015590ff0dSUlrich Drepper 40026521f891SChristian Brauner /** 40036521f891SChristian Brauner * vfs_mkdir - create directory 40046521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 40056521f891SChristian Brauner * @dir: inode of @dentry 40066521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 40076521f891SChristian Brauner * @mode: mode of the new directory 40086521f891SChristian Brauner * 40096521f891SChristian Brauner * Create a directory. 40106521f891SChristian Brauner * 40116521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 40126521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 40136521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 40146521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 40156521f891SChristian Brauner * raw inode simply passs init_user_ns. 40166521f891SChristian Brauner */ 40176521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 40186521f891SChristian Brauner struct dentry *dentry, umode_t mode) 40191da177e4SLinus Torvalds { 40206521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 40218de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40221da177e4SLinus Torvalds 40231da177e4SLinus Torvalds if (error) 40241da177e4SLinus Torvalds return error; 40251da177e4SLinus Torvalds 4026acfa4380SAl Viro if (!dir->i_op->mkdir) 40271da177e4SLinus Torvalds return -EPERM; 40281da177e4SLinus Torvalds 40291639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IRWXUGO | S_ISVTX, 0); 40301da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 40311da177e4SLinus Torvalds if (error) 40321da177e4SLinus Torvalds return error; 40331da177e4SLinus Torvalds 40348de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 40358de52778SAl Viro return -EMLINK; 40368de52778SAl Viro 4037549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 4038a74574aaSStephen Smalley if (!error) 4039f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 40401da177e4SLinus Torvalds return error; 40411da177e4SLinus Torvalds } 40424d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 40431da177e4SLinus Torvalds 404445f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 40451da177e4SLinus Torvalds { 40466902d925SDave Hansen struct dentry *dentry; 4047dae6ad8fSAl Viro struct path path; 4048dae6ad8fSAl Viro int error; 4049b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 40501da177e4SLinus Torvalds 4051b76d8b82SJeff Layton retry: 4052b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 4053584d3226SDmitry Kadashev error = PTR_ERR(dentry); 40546902d925SDave Hansen if (IS_ERR(dentry)) 4055584d3226SDmitry Kadashev goto out_putname; 40566902d925SDave Hansen 40571639a49cSYang Xu error = security_path_mkdir(&path, dentry, 40581639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode)); 40596521f891SChristian Brauner if (!error) { 40606521f891SChristian Brauner struct user_namespace *mnt_userns; 40616521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4062549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 4063549c7297SChristian Brauner mode); 40646521f891SChristian Brauner } 4065921a1650SAl Viro done_path_create(&path, dentry); 4066b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 4067b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 4068b76d8b82SJeff Layton goto retry; 4069b76d8b82SJeff Layton } 4070584d3226SDmitry Kadashev out_putname: 4071584d3226SDmitry Kadashev putname(name); 40721da177e4SLinus Torvalds return error; 40731da177e4SLinus Torvalds } 40741da177e4SLinus Torvalds 40750101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 40760101db7aSDominik Brodowski { 4077584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 40780101db7aSDominik Brodowski } 40790101db7aSDominik Brodowski 4080a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 40815590ff0dSUlrich Drepper { 4082584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 40835590ff0dSUlrich Drepper } 40845590ff0dSUlrich Drepper 40856521f891SChristian Brauner /** 40866521f891SChristian Brauner * vfs_rmdir - remove directory 40876521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 40886521f891SChristian Brauner * @dir: inode of @dentry 40896521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 40906521f891SChristian Brauner * 40916521f891SChristian Brauner * Remove a directory. 40926521f891SChristian Brauner * 40936521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 40946521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 40956521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 40966521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 40976521f891SChristian Brauner * raw inode simply passs init_user_ns. 40986521f891SChristian Brauner */ 40996521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 41006521f891SChristian Brauner struct dentry *dentry) 41011da177e4SLinus Torvalds { 41026521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 41031da177e4SLinus Torvalds 41041da177e4SLinus Torvalds if (error) 41051da177e4SLinus Torvalds return error; 41061da177e4SLinus Torvalds 4107acfa4380SAl Viro if (!dir->i_op->rmdir) 41081da177e4SLinus Torvalds return -EPERM; 41091da177e4SLinus Torvalds 41101d2ef590SAl Viro dget(dentry); 41115955102cSAl Viro inode_lock(dentry->d_inode); 4112912dbc15SSage Weil 41131da177e4SLinus Torvalds error = -EBUSY; 41141bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 41151bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4116912dbc15SSage Weil goto out; 4117912dbc15SSage Weil 41181da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4119912dbc15SSage Weil if (error) 4120912dbc15SSage Weil goto out; 4121912dbc15SSage Weil 41221da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4123912dbc15SSage Weil if (error) 4124912dbc15SSage Weil goto out; 4125912dbc15SSage Weil 41268767712fSAl Viro shrink_dcache_parent(dentry); 41271da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4128d83c49f3SAl Viro dont_mount(dentry); 41298ed936b5SEric W. Biederman detach_mounts(dentry); 41301da177e4SLinus Torvalds 4131912dbc15SSage Weil out: 41325955102cSAl Viro inode_unlock(dentry->d_inode); 41331d2ef590SAl Viro dput(dentry); 4134912dbc15SSage Weil if (!error) 4135a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 41361da177e4SLinus Torvalds return error; 41371da177e4SLinus Torvalds } 41384d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 41391da177e4SLinus Torvalds 414045f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 41411da177e4SLinus Torvalds { 41426521f891SChristian Brauner struct user_namespace *mnt_userns; 41430ee50b47SDmitry Kadashev int error; 41441da177e4SLinus Torvalds struct dentry *dentry; 4145f5beed75SAl Viro struct path path; 4146f5beed75SAl Viro struct qstr last; 4147f5beed75SAl Viro int type; 4148c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4149c6ee9206SJeff Layton retry: 4150c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 41510ee50b47SDmitry Kadashev if (error) 41520ee50b47SDmitry Kadashev goto exit1; 41531da177e4SLinus Torvalds 4154f5beed75SAl Viro switch (type) { 41551da177e4SLinus Torvalds case LAST_DOTDOT: 41561da177e4SLinus Torvalds error = -ENOTEMPTY; 41570ee50b47SDmitry Kadashev goto exit2; 41581da177e4SLinus Torvalds case LAST_DOT: 41591da177e4SLinus Torvalds error = -EINVAL; 41600ee50b47SDmitry Kadashev goto exit2; 41611da177e4SLinus Torvalds case LAST_ROOT: 41621da177e4SLinus Torvalds error = -EBUSY; 41630ee50b47SDmitry Kadashev goto exit2; 41641da177e4SLinus Torvalds } 41650612d9fbSOGAWA Hirofumi 4166f5beed75SAl Viro error = mnt_want_write(path.mnt); 4167c30dabfeSJan Kara if (error) 41680ee50b47SDmitry Kadashev goto exit2; 41690612d9fbSOGAWA Hirofumi 41705955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4171f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 41721da177e4SLinus Torvalds error = PTR_ERR(dentry); 41736902d925SDave Hansen if (IS_ERR(dentry)) 41740ee50b47SDmitry Kadashev goto exit3; 4175e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4176e6bc45d6STheodore Ts'o error = -ENOENT; 41770ee50b47SDmitry Kadashev goto exit4; 4178e6bc45d6STheodore Ts'o } 4179f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4180be6d3e56SKentaro Takeda if (error) 41810ee50b47SDmitry Kadashev goto exit4; 41826521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 41836521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 41840ee50b47SDmitry Kadashev exit4: 41851da177e4SLinus Torvalds dput(dentry); 41860ee50b47SDmitry Kadashev exit3: 41875955102cSAl Viro inode_unlock(path.dentry->d_inode); 4188f5beed75SAl Viro mnt_drop_write(path.mnt); 41890ee50b47SDmitry Kadashev exit2: 4190f5beed75SAl Viro path_put(&path); 4191c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4192c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4193c6ee9206SJeff Layton goto retry; 4194c6ee9206SJeff Layton } 41950ee50b47SDmitry Kadashev exit1: 419624fb33d4SAl Viro putname(name); 41971da177e4SLinus Torvalds return error; 41981da177e4SLinus Torvalds } 41991da177e4SLinus Torvalds 42003cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 42015590ff0dSUlrich Drepper { 4202e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 42035590ff0dSUlrich Drepper } 42045590ff0dSUlrich Drepper 4205b21996e3SJ. Bruce Fields /** 4206b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 42076521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 4208b21996e3SJ. Bruce Fields * @dir: parent directory 4209b21996e3SJ. Bruce Fields * @dentry: victim 4210b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4211b21996e3SJ. Bruce Fields * 4212b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4213b21996e3SJ. Bruce Fields * 4214b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4215b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4216b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4217b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4218b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4219b21996e3SJ. Bruce Fields * 4220b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4221b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4222b21996e3SJ. Bruce Fields * to be NFS exported. 42236521f891SChristian Brauner * 42246521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 42256521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42266521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42276521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42286521f891SChristian Brauner * raw inode simply passs init_user_ns. 4229b21996e3SJ. Bruce Fields */ 42306521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 42316521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 42321da177e4SLinus Torvalds { 42339accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 42346521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 42351da177e4SLinus Torvalds 42361da177e4SLinus Torvalds if (error) 42371da177e4SLinus Torvalds return error; 42381da177e4SLinus Torvalds 4239acfa4380SAl Viro if (!dir->i_op->unlink) 42401da177e4SLinus Torvalds return -EPERM; 42411da177e4SLinus Torvalds 42425955102cSAl Viro inode_lock(target); 424351cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 424451cc3a66SHugh Dickins error = -EPERM; 424551cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 42461da177e4SLinus Torvalds error = -EBUSY; 42471da177e4SLinus Torvalds else { 42481da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4249bec1052eSAl Viro if (!error) { 42505a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 42515a14696cSJ. Bruce Fields if (error) 4252b21996e3SJ. Bruce Fields goto out; 42531da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 42548ed936b5SEric W. Biederman if (!error) { 4255d83c49f3SAl Viro dont_mount(dentry); 42568ed936b5SEric W. Biederman detach_mounts(dentry); 42578ed936b5SEric W. Biederman } 4258bec1052eSAl Viro } 42591da177e4SLinus Torvalds } 4260b21996e3SJ. Bruce Fields out: 42615955102cSAl Viro inode_unlock(target); 42621da177e4SLinus Torvalds 42631da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4264a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4265a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4266a37d9a17SAmir Goldstein } else if (!error) { 42679accbb97SJ. Bruce Fields fsnotify_link_count(target); 4268a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 42691da177e4SLinus Torvalds } 42700eeca283SRobert Love 42711da177e4SLinus Torvalds return error; 42721da177e4SLinus Torvalds } 42734d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 42741da177e4SLinus Torvalds 42751da177e4SLinus Torvalds /* 42761da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 42771b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 42781da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 42791da177e4SLinus Torvalds * while waiting on the I/O. 42801da177e4SLinus Torvalds */ 428145f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 42821da177e4SLinus Torvalds { 42832ad94ae6SAl Viro int error; 42841da177e4SLinus Torvalds struct dentry *dentry; 4285f5beed75SAl Viro struct path path; 4286f5beed75SAl Viro struct qstr last; 4287f5beed75SAl Viro int type; 42881da177e4SLinus Torvalds struct inode *inode = NULL; 4289b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 42905d18f813SJeff Layton unsigned int lookup_flags = 0; 42915d18f813SJeff Layton retry: 4292c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 42930ee50b47SDmitry Kadashev if (error) 42940ee50b47SDmitry Kadashev goto exit1; 42952ad94ae6SAl Viro 42961da177e4SLinus Torvalds error = -EISDIR; 4297f5beed75SAl Viro if (type != LAST_NORM) 42980ee50b47SDmitry Kadashev goto exit2; 42990612d9fbSOGAWA Hirofumi 4300f5beed75SAl Viro error = mnt_want_write(path.mnt); 4301c30dabfeSJan Kara if (error) 43020ee50b47SDmitry Kadashev goto exit2; 4303b21996e3SJ. Bruce Fields retry_deleg: 43045955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4305f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 43061da177e4SLinus Torvalds error = PTR_ERR(dentry); 43071da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 43086521f891SChristian Brauner struct user_namespace *mnt_userns; 43096521f891SChristian Brauner 43101da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4311f5beed75SAl Viro if (last.name[last.len]) 431250338b88STörök Edwin goto slashes; 43131da177e4SLinus Torvalds inode = dentry->d_inode; 4314b18825a7SDavid Howells if (d_is_negative(dentry)) 4315e6bc45d6STheodore Ts'o goto slashes; 43167de9c6eeSAl Viro ihold(inode); 4317f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4318be6d3e56SKentaro Takeda if (error) 43190ee50b47SDmitry Kadashev goto exit3; 43206521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4321549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4322549c7297SChristian Brauner &delegated_inode); 43230ee50b47SDmitry Kadashev exit3: 43241da177e4SLinus Torvalds dput(dentry); 43251da177e4SLinus Torvalds } 43265955102cSAl Viro inode_unlock(path.dentry->d_inode); 43271da177e4SLinus Torvalds if (inode) 43281da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4329b21996e3SJ. Bruce Fields inode = NULL; 4330b21996e3SJ. Bruce Fields if (delegated_inode) { 43315a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4332b21996e3SJ. Bruce Fields if (!error) 4333b21996e3SJ. Bruce Fields goto retry_deleg; 4334b21996e3SJ. Bruce Fields } 4335f5beed75SAl Viro mnt_drop_write(path.mnt); 43360ee50b47SDmitry Kadashev exit2: 4337f5beed75SAl Viro path_put(&path); 43385d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 43395d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 43405d18f813SJeff Layton inode = NULL; 43415d18f813SJeff Layton goto retry; 43425d18f813SJeff Layton } 43430ee50b47SDmitry Kadashev exit1: 4344da2f1362SChristoph Hellwig putname(name); 43451da177e4SLinus Torvalds return error; 43461da177e4SLinus Torvalds 43471da177e4SLinus Torvalds slashes: 4348b18825a7SDavid Howells if (d_is_negative(dentry)) 4349b18825a7SDavid Howells error = -ENOENT; 435044b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4351b18825a7SDavid Howells error = -EISDIR; 4352b18825a7SDavid Howells else 4353b18825a7SDavid Howells error = -ENOTDIR; 43540ee50b47SDmitry Kadashev goto exit3; 43551da177e4SLinus Torvalds } 43561da177e4SLinus Torvalds 43572e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 43585590ff0dSUlrich Drepper { 43595590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 43605590ff0dSUlrich Drepper return -EINVAL; 43615590ff0dSUlrich Drepper 43625590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4363e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4364da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 43655590ff0dSUlrich Drepper } 43665590ff0dSUlrich Drepper 43673480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 43685590ff0dSUlrich Drepper { 4369da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 43705590ff0dSUlrich Drepper } 43715590ff0dSUlrich Drepper 43726521f891SChristian Brauner /** 43736521f891SChristian Brauner * vfs_symlink - create symlink 43746521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 43756521f891SChristian Brauner * @dir: inode of @dentry 43766521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 43776521f891SChristian Brauner * @oldname: name of the file to link to 43786521f891SChristian Brauner * 43796521f891SChristian Brauner * Create a symlink. 43806521f891SChristian Brauner * 43816521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 43826521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 43836521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 43846521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 43856521f891SChristian Brauner * raw inode simply passs init_user_ns. 43866521f891SChristian Brauner */ 43876521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 43886521f891SChristian Brauner struct dentry *dentry, const char *oldname) 43891da177e4SLinus Torvalds { 43906521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 43911da177e4SLinus Torvalds 43921da177e4SLinus Torvalds if (error) 43931da177e4SLinus Torvalds return error; 43941da177e4SLinus Torvalds 4395acfa4380SAl Viro if (!dir->i_op->symlink) 43961da177e4SLinus Torvalds return -EPERM; 43971da177e4SLinus Torvalds 43981da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 43991da177e4SLinus Torvalds if (error) 44001da177e4SLinus Torvalds return error; 44011da177e4SLinus Torvalds 4402549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4403a74574aaSStephen Smalley if (!error) 4404f38aa942SAmy Griffis fsnotify_create(dir, dentry); 44051da177e4SLinus Torvalds return error; 44061da177e4SLinus Torvalds } 44074d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 44081da177e4SLinus Torvalds 44097a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 44101da177e4SLinus Torvalds { 44112ad94ae6SAl Viro int error; 44126902d925SDave Hansen struct dentry *dentry; 4413dae6ad8fSAl Viro struct path path; 4414f46d3567SJeff Layton unsigned int lookup_flags = 0; 44151da177e4SLinus Torvalds 4416da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4417da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4418da2d0cedSDmitry Kadashev goto out_putnames; 4419da2d0cedSDmitry Kadashev } 4420f46d3567SJeff Layton retry: 4421b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 44221da177e4SLinus Torvalds error = PTR_ERR(dentry); 44236902d925SDave Hansen if (IS_ERR(dentry)) 4424da2d0cedSDmitry Kadashev goto out_putnames; 44256902d925SDave Hansen 442691a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 44276521f891SChristian Brauner if (!error) { 44286521f891SChristian Brauner struct user_namespace *mnt_userns; 44296521f891SChristian Brauner 44306521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 44316521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 44326521f891SChristian Brauner from->name); 44336521f891SChristian Brauner } 4434921a1650SAl Viro done_path_create(&path, dentry); 4435f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4436f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4437f46d3567SJeff Layton goto retry; 4438f46d3567SJeff Layton } 4439da2d0cedSDmitry Kadashev out_putnames: 4440da2d0cedSDmitry Kadashev putname(to); 44411da177e4SLinus Torvalds putname(from); 44421da177e4SLinus Torvalds return error; 44431da177e4SLinus Torvalds } 44441da177e4SLinus Torvalds 4445b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4446b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4447b724e846SDominik Brodowski { 4448da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4449b724e846SDominik Brodowski } 4450b724e846SDominik Brodowski 44513480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 44525590ff0dSUlrich Drepper { 4453da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 44545590ff0dSUlrich Drepper } 44555590ff0dSUlrich Drepper 4456146a8595SJ. Bruce Fields /** 4457146a8595SJ. Bruce Fields * vfs_link - create a new link 4458146a8595SJ. Bruce Fields * @old_dentry: object to be linked 44596521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4460146a8595SJ. Bruce Fields * @dir: new parent 4461146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4462146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4463146a8595SJ. Bruce Fields * 4464146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4465146a8595SJ. Bruce Fields * 4466146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4467146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4468146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4469146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4470146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4471146a8595SJ. Bruce Fields * 4472146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4473146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4474146a8595SJ. Bruce Fields * to be NFS exported. 44756521f891SChristian Brauner * 44766521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 44776521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 44786521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 44796521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 44806521f891SChristian Brauner * raw inode simply passs init_user_ns. 4481146a8595SJ. Bruce Fields */ 44826521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 44836521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 44846521f891SChristian Brauner struct inode **delegated_inode) 44851da177e4SLinus Torvalds { 44861da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 44878de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 44881da177e4SLinus Torvalds int error; 44891da177e4SLinus Torvalds 44901da177e4SLinus Torvalds if (!inode) 44911da177e4SLinus Torvalds return -ENOENT; 44921da177e4SLinus Torvalds 44936521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 44941da177e4SLinus Torvalds if (error) 44951da177e4SLinus Torvalds return error; 44961da177e4SLinus Torvalds 44971da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 44981da177e4SLinus Torvalds return -EXDEV; 44991da177e4SLinus Torvalds 45001da177e4SLinus Torvalds /* 45011da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 45021da177e4SLinus Torvalds */ 45031da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 45041da177e4SLinus Torvalds return -EPERM; 45050bd23d09SEric W. Biederman /* 45060bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 45070bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 45080bd23d09SEric W. Biederman * the vfs. 45090bd23d09SEric W. Biederman */ 45106521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 45110bd23d09SEric W. Biederman return -EPERM; 4512acfa4380SAl Viro if (!dir->i_op->link) 45131da177e4SLinus Torvalds return -EPERM; 45147e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 45151da177e4SLinus Torvalds return -EPERM; 45161da177e4SLinus Torvalds 45171da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 45181da177e4SLinus Torvalds if (error) 45191da177e4SLinus Torvalds return error; 45201da177e4SLinus Torvalds 45215955102cSAl Viro inode_lock(inode); 4522aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4523f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4524aae8a97dSAneesh Kumar K.V error = -ENOENT; 45258de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 45268de52778SAl Viro error = -EMLINK; 4527146a8595SJ. Bruce Fields else { 4528146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4529146a8595SJ. Bruce Fields if (!error) 45301da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4531146a8595SJ. Bruce Fields } 4532f4e0c30cSAl Viro 4533f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4534f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4535f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4536f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4537f4e0c30cSAl Viro } 45385955102cSAl Viro inode_unlock(inode); 4539e31e14ecSStephen Smalley if (!error) 45407e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 45411da177e4SLinus Torvalds return error; 45421da177e4SLinus Torvalds } 45434d359507SAl Viro EXPORT_SYMBOL(vfs_link); 45441da177e4SLinus Torvalds 45451da177e4SLinus Torvalds /* 45461da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 45471da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 45481da177e4SLinus Torvalds * newname. --KAB 45491da177e4SLinus Torvalds * 45501da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 45511da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 45521da177e4SLinus Torvalds * and other special files. --ADM 45531da177e4SLinus Torvalds */ 4554cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4555020250f3SDmitry Kadashev struct filename *new, int flags) 45561da177e4SLinus Torvalds { 45576521f891SChristian Brauner struct user_namespace *mnt_userns; 45581da177e4SLinus Torvalds struct dentry *new_dentry; 4559dae6ad8fSAl Viro struct path old_path, new_path; 4560146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 456111a7b371SAneesh Kumar K.V int how = 0; 45621da177e4SLinus Torvalds int error; 45631da177e4SLinus Torvalds 4564020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4565020250f3SDmitry Kadashev error = -EINVAL; 4566020250f3SDmitry Kadashev goto out_putnames; 4567020250f3SDmitry Kadashev } 456811a7b371SAneesh Kumar K.V /* 4569f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4570f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4571f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 457211a7b371SAneesh Kumar K.V */ 4573020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4574020250f3SDmitry Kadashev error = -ENOENT; 4575020250f3SDmitry Kadashev goto out_putnames; 4576f0cc6ffbSLinus Torvalds } 4577c04030e1SUlrich Drepper 457811a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 457911a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4580442e31caSJeff Layton retry: 4581794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 45821da177e4SLinus Torvalds if (error) 4583020250f3SDmitry Kadashev goto out_putnames; 45842ad94ae6SAl Viro 4585b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4586442e31caSJeff Layton (how & LOOKUP_REVAL)); 45871da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45886902d925SDave Hansen if (IS_ERR(new_dentry)) 4589020250f3SDmitry Kadashev goto out_putpath; 4590dae6ad8fSAl Viro 4591dae6ad8fSAl Viro error = -EXDEV; 4592dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4593dae6ad8fSAl Viro goto out_dput; 4594549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4595549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4596800179c9SKees Cook if (unlikely(error)) 4597800179c9SKees Cook goto out_dput; 4598dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4599be6d3e56SKentaro Takeda if (error) 4600a8104a9fSAl Viro goto out_dput; 46016521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 46026521f891SChristian Brauner new_dentry, &delegated_inode); 460375c3f29dSDave Hansen out_dput: 4604921a1650SAl Viro done_path_create(&new_path, new_dentry); 4605146a8595SJ. Bruce Fields if (delegated_inode) { 4606146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4607d22e6338SOleg Drokin if (!error) { 4608d22e6338SOleg Drokin path_put(&old_path); 4609146a8595SJ. Bruce Fields goto retry; 4610146a8595SJ. Bruce Fields } 4611d22e6338SOleg Drokin } 4612442e31caSJeff Layton if (retry_estale(error, how)) { 4613d22e6338SOleg Drokin path_put(&old_path); 4614442e31caSJeff Layton how |= LOOKUP_REVAL; 4615442e31caSJeff Layton goto retry; 4616442e31caSJeff Layton } 4617020250f3SDmitry Kadashev out_putpath: 46182d8f3038SAl Viro path_put(&old_path); 4619020250f3SDmitry Kadashev out_putnames: 4620020250f3SDmitry Kadashev putname(old); 4621020250f3SDmitry Kadashev putname(new); 46221da177e4SLinus Torvalds 46231da177e4SLinus Torvalds return error; 46241da177e4SLinus Torvalds } 46251da177e4SLinus Torvalds 462646ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 462746ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 462846ea89ebSDominik Brodowski { 4629020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4630020250f3SDmitry Kadashev newdfd, getname(newname), flags); 463146ea89ebSDominik Brodowski } 463246ea89ebSDominik Brodowski 46333480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 46345590ff0dSUlrich Drepper { 4635020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 46365590ff0dSUlrich Drepper } 46375590ff0dSUlrich Drepper 4638bc27027aSMiklos Szeredi /** 4639bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 46402111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4641bc27027aSMiklos Szeredi * 4642bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4643bc27027aSMiklos Szeredi * 4644bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4645bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4646bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4647bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4648bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4649bc27027aSMiklos Szeredi * so. 4650bc27027aSMiklos Szeredi * 4651bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4652bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4653bc27027aSMiklos Szeredi * to be NFS exported. 4654bc27027aSMiklos Szeredi * 46551da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 46561da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 46571da177e4SLinus Torvalds * Problems: 46580117d427SMauro Carvalho Chehab * 4659d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 46601da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 46611da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4662a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 46631da177e4SLinus Torvalds * story. 46646cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 46656cedba89SJ. Bruce Fields * and source (if it is not a directory). 46661b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 46671da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 46681da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4669a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 46701da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 46711da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 46721da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4673a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 46741da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 46751da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 46761da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4677e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 46781b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 46791da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4680c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 46811da177e4SLinus Torvalds * locking]. 46821da177e4SLinus Torvalds */ 46839fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 46841da177e4SLinus Torvalds { 46851da177e4SLinus Torvalds int error; 46869fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 46879fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 46889fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 46899fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 46909fe61450SChristian Brauner unsigned int flags = rd->flags; 4691bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4692bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4693bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4694da1ce067SMiklos Szeredi bool new_is_dir = false; 4695da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 469649d31c2fSAl Viro struct name_snapshot old_name; 46971da177e4SLinus Torvalds 46988d3e2936SMiklos Szeredi if (source == target) 46991da177e4SLinus Torvalds return 0; 47001da177e4SLinus Torvalds 47016521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 47021da177e4SLinus Torvalds if (error) 47031da177e4SLinus Torvalds return error; 47041da177e4SLinus Torvalds 4705da1ce067SMiklos Szeredi if (!target) { 47066521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4707da1ce067SMiklos Szeredi } else { 4708da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4709da1ce067SMiklos Szeredi 4710da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 47116521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 47126521f891SChristian Brauner new_dentry, is_dir); 4713da1ce067SMiklos Szeredi else 47146521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 47156521f891SChristian Brauner new_dentry, new_is_dir); 4716da1ce067SMiklos Szeredi } 47171da177e4SLinus Torvalds if (error) 47181da177e4SLinus Torvalds return error; 47191da177e4SLinus Torvalds 47202773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 47211da177e4SLinus Torvalds return -EPERM; 47221da177e4SLinus Torvalds 4723bc27027aSMiklos Szeredi /* 4724bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4725bc27027aSMiklos Szeredi * we'll need to flip '..'. 4726bc27027aSMiklos Szeredi */ 4727da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4728da1ce067SMiklos Szeredi if (is_dir) { 47296521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 473047291baaSChristian Brauner MAY_WRITE); 4731bc27027aSMiklos Szeredi if (error) 4732bc27027aSMiklos Szeredi return error; 4733bc27027aSMiklos Szeredi } 4734da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 47356521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 473647291baaSChristian Brauner MAY_WRITE); 4737da1ce067SMiklos Szeredi if (error) 4738da1ce067SMiklos Szeredi return error; 4739da1ce067SMiklos Szeredi } 4740da1ce067SMiklos Szeredi } 47410eeca283SRobert Love 47420b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 47430b3974ebSMiklos Szeredi flags); 4744bc27027aSMiklos Szeredi if (error) 4745bc27027aSMiklos Szeredi return error; 4746bc27027aSMiklos Szeredi 474749d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4748bc27027aSMiklos Szeredi dget(new_dentry); 4749da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4750bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4751bc27027aSMiklos Szeredi else if (target) 47525955102cSAl Viro inode_lock(target); 4753bc27027aSMiklos Szeredi 475451cc3a66SHugh Dickins error = -EPERM; 475551cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 475651cc3a66SHugh Dickins goto out; 475751cc3a66SHugh Dickins 4758bc27027aSMiklos Szeredi error = -EBUSY; 47597af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4760bc27027aSMiklos Szeredi goto out; 4761bc27027aSMiklos Szeredi 4762da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4763bc27027aSMiklos Szeredi error = -EMLINK; 4764da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4765bc27027aSMiklos Szeredi goto out; 4766da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4767da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4768da1ce067SMiklos Szeredi goto out; 4769da1ce067SMiklos Szeredi } 4770da1ce067SMiklos Szeredi if (!is_dir) { 4771bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4772bc27027aSMiklos Szeredi if (error) 4773bc27027aSMiklos Szeredi goto out; 4774da1ce067SMiklos Szeredi } 4775da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4776bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4777bc27027aSMiklos Szeredi if (error) 4778bc27027aSMiklos Szeredi goto out; 4779bc27027aSMiklos Szeredi } 4780549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4781520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4782bc27027aSMiklos Szeredi if (error) 4783bc27027aSMiklos Szeredi goto out; 4784bc27027aSMiklos Szeredi 4785da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 47868767712fSAl Viro if (is_dir) { 47878767712fSAl Viro shrink_dcache_parent(new_dentry); 4788bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 47898767712fSAl Viro } 4790bc27027aSMiklos Szeredi dont_mount(new_dentry); 47918ed936b5SEric W. Biederman detach_mounts(new_dentry); 4792bc27027aSMiklos Szeredi } 4793da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4794da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4795bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4796da1ce067SMiklos Szeredi else 4797da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4798da1ce067SMiklos Szeredi } 4799bc27027aSMiklos Szeredi out: 4800da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4801bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4802bc27027aSMiklos Szeredi else if (target) 48035955102cSAl Viro inode_unlock(target); 4804bc27027aSMiklos Szeredi dput(new_dentry); 4805da1ce067SMiklos Szeredi if (!error) { 4806f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4807da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4808da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4809f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4810da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4811da1ce067SMiklos Szeredi } 4812da1ce067SMiklos Szeredi } 481349d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 48140eeca283SRobert Love 48151da177e4SLinus Torvalds return error; 48161da177e4SLinus Torvalds } 48174d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 48181da177e4SLinus Torvalds 4819e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4820e886663cSJens Axboe struct filename *to, unsigned int flags) 48211da177e4SLinus Torvalds { 48229fe61450SChristian Brauner struct renamedata rd; 48231da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 48241da177e4SLinus Torvalds struct dentry *trap; 4825f5beed75SAl Viro struct path old_path, new_path; 4826f5beed75SAl Viro struct qstr old_last, new_last; 4827f5beed75SAl Viro int old_type, new_type; 48288e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4829f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4830c6a94284SJeff Layton bool should_retry = false; 4831e886663cSJens Axboe int error = -EINVAL; 4832520c8b16SMiklos Szeredi 48330d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 48340ee50b47SDmitry Kadashev goto put_names; 4835da1ce067SMiklos Szeredi 48360d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 48370d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 48380ee50b47SDmitry Kadashev goto put_names; 4839520c8b16SMiklos Szeredi 4840f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4841f5beed75SAl Viro target_flags = 0; 4842f5beed75SAl Viro 4843c6a94284SJeff Layton retry: 4844c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4845e886663cSJens Axboe &old_last, &old_type); 48460ee50b47SDmitry Kadashev if (error) 48470ee50b47SDmitry Kadashev goto put_names; 48481da177e4SLinus Torvalds 4849c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4850e886663cSJens Axboe &new_type); 48510ee50b47SDmitry Kadashev if (error) 48521da177e4SLinus Torvalds goto exit1; 48531da177e4SLinus Torvalds 48541da177e4SLinus Torvalds error = -EXDEV; 4855f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 48561da177e4SLinus Torvalds goto exit2; 48571da177e4SLinus Torvalds 48581da177e4SLinus Torvalds error = -EBUSY; 4859f5beed75SAl Viro if (old_type != LAST_NORM) 48601da177e4SLinus Torvalds goto exit2; 48611da177e4SLinus Torvalds 48620a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 48630a7c3937SMiklos Szeredi error = -EEXIST; 4864f5beed75SAl Viro if (new_type != LAST_NORM) 48651da177e4SLinus Torvalds goto exit2; 48661da177e4SLinus Torvalds 4867f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4868c30dabfeSJan Kara if (error) 4869c30dabfeSJan Kara goto exit2; 4870c30dabfeSJan Kara 48718e6d782cSJ. Bruce Fields retry_deleg: 4872f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 48731da177e4SLinus Torvalds 4874f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 48751da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 48761da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 48771da177e4SLinus Torvalds goto exit3; 48781da177e4SLinus Torvalds /* source must exist */ 48791da177e4SLinus Torvalds error = -ENOENT; 4880b18825a7SDavid Howells if (d_is_negative(old_dentry)) 48811da177e4SLinus Torvalds goto exit4; 4882f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 48831da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 48841da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 48851da177e4SLinus Torvalds goto exit4; 48860a7c3937SMiklos Szeredi error = -EEXIST; 48870a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 48880a7c3937SMiklos Szeredi goto exit5; 4889da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4890da1ce067SMiklos Szeredi error = -ENOENT; 4891da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4892da1ce067SMiklos Szeredi goto exit5; 4893da1ce067SMiklos Szeredi 4894da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4895da1ce067SMiklos Szeredi error = -ENOTDIR; 4896f5beed75SAl Viro if (new_last.name[new_last.len]) 4897da1ce067SMiklos Szeredi goto exit5; 4898da1ce067SMiklos Szeredi } 4899da1ce067SMiklos Szeredi } 49000a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 49010a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 49020a7c3937SMiklos Szeredi error = -ENOTDIR; 4903f5beed75SAl Viro if (old_last.name[old_last.len]) 49040a7c3937SMiklos Szeredi goto exit5; 4905f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 49060a7c3937SMiklos Szeredi goto exit5; 49070a7c3937SMiklos Szeredi } 49080a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 49090a7c3937SMiklos Szeredi error = -EINVAL; 49100a7c3937SMiklos Szeredi if (old_dentry == trap) 49110a7c3937SMiklos Szeredi goto exit5; 49121da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4913da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 49141da177e4SLinus Torvalds error = -ENOTEMPTY; 49151da177e4SLinus Torvalds if (new_dentry == trap) 49161da177e4SLinus Torvalds goto exit5; 49171da177e4SLinus Torvalds 4918f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4919f5beed75SAl Viro &new_path, new_dentry, flags); 4920be6d3e56SKentaro Takeda if (error) 4921c30dabfeSJan Kara goto exit5; 49229fe61450SChristian Brauner 49239fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 49249fe61450SChristian Brauner rd.old_dentry = old_dentry; 49256521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 49269fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 49279fe61450SChristian Brauner rd.new_dentry = new_dentry; 49286521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 49299fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 49309fe61450SChristian Brauner rd.flags = flags; 49319fe61450SChristian Brauner error = vfs_rename(&rd); 49321da177e4SLinus Torvalds exit5: 49331da177e4SLinus Torvalds dput(new_dentry); 49341da177e4SLinus Torvalds exit4: 49351da177e4SLinus Torvalds dput(old_dentry); 49361da177e4SLinus Torvalds exit3: 4937f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 49388e6d782cSJ. Bruce Fields if (delegated_inode) { 49398e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 49408e6d782cSJ. Bruce Fields if (!error) 49418e6d782cSJ. Bruce Fields goto retry_deleg; 49428e6d782cSJ. Bruce Fields } 4943f5beed75SAl Viro mnt_drop_write(old_path.mnt); 49441da177e4SLinus Torvalds exit2: 4945c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4946c6a94284SJeff Layton should_retry = true; 4947f5beed75SAl Viro path_put(&new_path); 49481da177e4SLinus Torvalds exit1: 4949f5beed75SAl Viro path_put(&old_path); 4950c6a94284SJeff Layton if (should_retry) { 4951c6a94284SJeff Layton should_retry = false; 4952c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4953c6a94284SJeff Layton goto retry; 4954c6a94284SJeff Layton } 49550ee50b47SDmitry Kadashev put_names: 4956e886663cSJens Axboe putname(from); 4957e886663cSJens Axboe putname(to); 49581da177e4SLinus Torvalds return error; 49591da177e4SLinus Torvalds } 49601da177e4SLinus Torvalds 4961ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4962ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4963ee81feb6SDominik Brodowski { 4964e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4965e886663cSJens Axboe flags); 4966ee81feb6SDominik Brodowski } 4967ee81feb6SDominik Brodowski 4968520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4969520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4970520c8b16SMiklos Szeredi { 4971e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4972e886663cSJens Axboe 0); 4973520c8b16SMiklos Szeredi } 4974520c8b16SMiklos Szeredi 4975a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 49765590ff0dSUlrich Drepper { 4977e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4978e886663cSJens Axboe getname(newname), 0); 49795590ff0dSUlrich Drepper } 49805590ff0dSUlrich Drepper 49815d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 49821da177e4SLinus Torvalds { 49835d826c84SAl Viro int len = PTR_ERR(link); 49841da177e4SLinus Torvalds if (IS_ERR(link)) 49851da177e4SLinus Torvalds goto out; 49861da177e4SLinus Torvalds 49871da177e4SLinus Torvalds len = strlen(link); 49881da177e4SLinus Torvalds if (len > (unsigned) buflen) 49891da177e4SLinus Torvalds len = buflen; 49901da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 49911da177e4SLinus Torvalds len = -EFAULT; 49921da177e4SLinus Torvalds out: 49931da177e4SLinus Torvalds return len; 49941da177e4SLinus Torvalds } 49951da177e4SLinus Torvalds 4996d60874cdSMiklos Szeredi /** 4997fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4998fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4999fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 5000fd4a0edfSMiklos Szeredi * @buflen: size of buffer 5001fd4a0edfSMiklos Szeredi * 5002fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 5003fd4a0edfSMiklos Szeredi * 5004fd4a0edfSMiklos Szeredi * Does not call security hook. 5005fd4a0edfSMiklos Szeredi */ 5006fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5007fd4a0edfSMiklos Szeredi { 5008fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 5009f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 5010f2df5da6SAl Viro const char *link; 5011f2df5da6SAl Viro int res; 5012fd4a0edfSMiklos Szeredi 501376fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 501476fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 501576fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 501676fca90eSMiklos Szeredi 501776fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 5018fd4a0edfSMiklos Szeredi return -EINVAL; 5019fd4a0edfSMiklos Szeredi 502076fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 502176fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 502276fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 502376fca90eSMiklos Szeredi } 502476fca90eSMiklos Szeredi 50254c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 5026f2df5da6SAl Viro if (!link) { 5027f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 5028f2df5da6SAl Viro if (IS_ERR(link)) 5029f2df5da6SAl Viro return PTR_ERR(link); 5030f2df5da6SAl Viro } 5031f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 5032f2df5da6SAl Viro do_delayed_call(&done); 5033f2df5da6SAl Viro return res; 5034fd4a0edfSMiklos Szeredi } 5035fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 50361da177e4SLinus Torvalds 5037d60874cdSMiklos Szeredi /** 5038d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 5039d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 5040d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 5041d60874cdSMiklos Szeredi * 5042d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 5043d60874cdSMiklos Szeredi * 5044d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 5045d60874cdSMiklos Szeredi * 5046d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 5047d60874cdSMiklos Szeredi */ 5048d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 5049d60874cdSMiklos Szeredi { 5050d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 5051d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 5052d60874cdSMiklos Szeredi 5053d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 5054d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 5055d60874cdSMiklos Szeredi if (!res) 5056d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 5057d60874cdSMiklos Szeredi } 5058d60874cdSMiklos Szeredi return res; 5059d60874cdSMiklos Szeredi } 5060d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 5061d60874cdSMiklos Szeredi 50621da177e4SLinus Torvalds /* get the link contents into pagecache */ 50636b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 5064fceef393SAl Viro struct delayed_call *callback) 50651da177e4SLinus Torvalds { 5066ebd09abbSDuane Griffin char *kaddr; 50671da177e4SLinus Torvalds struct page *page; 50686b255391SAl Viro struct address_space *mapping = inode->i_mapping; 50696b255391SAl Viro 5070d3883d4fSAl Viro if (!dentry) { 5071d3883d4fSAl Viro page = find_get_page(mapping, 0); 5072d3883d4fSAl Viro if (!page) 50736b255391SAl Viro return ERR_PTR(-ECHILD); 5074d3883d4fSAl Viro if (!PageUptodate(page)) { 5075d3883d4fSAl Viro put_page(page); 5076d3883d4fSAl Viro return ERR_PTR(-ECHILD); 5077d3883d4fSAl Viro } 5078d3883d4fSAl Viro } else { 5079090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 50801da177e4SLinus Torvalds if (IS_ERR(page)) 50816fe6900eSNick Piggin return (char*)page; 5082d3883d4fSAl Viro } 5083fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 508421fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 508521fc61c7SAl Viro kaddr = page_address(page); 50866b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 5087ebd09abbSDuane Griffin return kaddr; 50881da177e4SLinus Torvalds } 50891da177e4SLinus Torvalds 50906b255391SAl Viro EXPORT_SYMBOL(page_get_link); 50911da177e4SLinus Torvalds 5092fceef393SAl Viro void page_put_link(void *arg) 50931da177e4SLinus Torvalds { 5094fceef393SAl Viro put_page(arg); 50951da177e4SLinus Torvalds } 50964d359507SAl Viro EXPORT_SYMBOL(page_put_link); 50971da177e4SLinus Torvalds 5098aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5099aa80deabSAl Viro { 5100fceef393SAl Viro DEFINE_DELAYED_CALL(done); 51016b255391SAl Viro int res = readlink_copy(buffer, buflen, 51026b255391SAl Viro page_get_link(dentry, d_inode(dentry), 5103fceef393SAl Viro &done)); 5104fceef393SAl Viro do_delayed_call(&done); 5105aa80deabSAl Viro return res; 5106aa80deabSAl Viro } 5107aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 5108aa80deabSAl Viro 510956f5746cSMatthew Wilcox (Oracle) int page_symlink(struct inode *inode, const char *symname, int len) 51101da177e4SLinus Torvalds { 51111da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 511227a77913SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 511356f5746cSMatthew Wilcox (Oracle) bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS); 51140adb25d2SKirill Korotaev struct page *page; 51151468c6f4SAlexander Potapenko void *fsdata = NULL; 5116beb497abSDmitriy Monakhov int err; 51172d878178SMatthew Wilcox (Oracle) unsigned int flags; 51181da177e4SLinus Torvalds 51197e53cac4SNeilBrown retry: 51202d878178SMatthew Wilcox (Oracle) if (nofs) 51212d878178SMatthew Wilcox (Oracle) flags = memalloc_nofs_save(); 512227a77913SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata); 51232d878178SMatthew Wilcox (Oracle) if (nofs) 51242d878178SMatthew Wilcox (Oracle) memalloc_nofs_restore(flags); 51251da177e4SLinus Torvalds if (err) 5126afddba49SNick Piggin goto fail; 5127afddba49SNick Piggin 512821fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5129afddba49SNick Piggin 513027a77913SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, 0, len-1, len-1, 5131afddba49SNick Piggin page, fsdata); 51321da177e4SLinus Torvalds if (err < 0) 51331da177e4SLinus Torvalds goto fail; 5134afddba49SNick Piggin if (err < len-1) 5135afddba49SNick Piggin goto retry; 5136afddba49SNick Piggin 51371da177e4SLinus Torvalds mark_inode_dirty(inode); 51381da177e4SLinus Torvalds return 0; 51391da177e4SLinus Torvalds fail: 51401da177e4SLinus Torvalds return err; 51411da177e4SLinus Torvalds } 51424d359507SAl Viro EXPORT_SYMBOL(page_symlink); 51430adb25d2SKirill Korotaev 514492e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 51456b255391SAl Viro .get_link = page_get_link, 51461da177e4SLinus Torvalds }; 51471da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5148