1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namei.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * Some corrections by tytso. 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 131da177e4SLinus Torvalds * lookup logic. 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 161da177e4SLinus Torvalds */ 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/init.h> 19630d9c47SPaul Gortmaker #include <linux/export.h> 2044696908SDavid S. Miller #include <linux/kernel.h> 211da177e4SLinus Torvalds #include <linux/slab.h> 221da177e4SLinus Torvalds #include <linux/fs.h> 231da177e4SLinus Torvalds #include <linux/namei.h> 241da177e4SLinus Torvalds #include <linux/pagemap.h> 250eeca283SRobert Love #include <linux/fsnotify.h> 261da177e4SLinus Torvalds #include <linux/personality.h> 271da177e4SLinus Torvalds #include <linux/security.h> 286146f0d5SMimi Zohar #include <linux/ima.h> 291da177e4SLinus Torvalds #include <linux/syscalls.h> 301da177e4SLinus Torvalds #include <linux/mount.h> 311da177e4SLinus Torvalds #include <linux/audit.h> 3216f7e0feSRandy Dunlap #include <linux/capability.h> 33834f2a4aSTrond Myklebust #include <linux/file.h> 345590ff0dSUlrich Drepper #include <linux/fcntl.h> 3508ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 365ad4e53bSAl Viro #include <linux/fs_struct.h> 37e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3899d263d4SLinus Torvalds #include <linux/hash.h> 392a18da7aSGeorge Spelvin #include <linux/bitops.h> 40aeaa4a79SEric W. Biederman #include <linux/init_task.h> 417c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 421da177e4SLinus Torvalds 43e81e3f4dSEric Paris #include "internal.h" 44c7105365SAl Viro #include "mount.h" 45e81e3f4dSEric Paris 461da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 471da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 481da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 491da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 501da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 511da177e4SLinus Torvalds * 521da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 531da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 541da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 551da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 561da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 571da177e4SLinus Torvalds * the special cases of the former code. 581da177e4SLinus Torvalds * 591da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 601da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 611da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 621da177e4SLinus Torvalds * 631da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 641da177e4SLinus Torvalds * resolution to correspond with current state of the code. 651da177e4SLinus Torvalds * 661da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 671da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 681da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 691da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 701da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 711da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 721da177e4SLinus Torvalds */ 731da177e4SLinus Torvalds 741da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 751da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 761da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 771da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 781da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 791da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8025985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 811da177e4SLinus Torvalds * 821da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 831da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 841da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 851da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 861da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 871da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 881da177e4SLinus Torvalds * and in the old Linux semantics. 891da177e4SLinus Torvalds */ 901da177e4SLinus Torvalds 911da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 921da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 931da177e4SLinus Torvalds * 941da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 951da177e4SLinus Torvalds */ 961da177e4SLinus Torvalds 971da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 981da177e4SLinus Torvalds * inside the path - always follow. 991da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1001da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1011da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1021da177e4SLinus Torvalds * otherwise - don't follow. 1031da177e4SLinus Torvalds * (applied in that order). 1041da177e4SLinus Torvalds * 1051da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1061da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1071da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1081da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1091da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1101da177e4SLinus Torvalds */ 1111da177e4SLinus Torvalds /* 1121da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 113a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1141da177e4SLinus Torvalds * any extra contention... 1151da177e4SLinus Torvalds */ 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1181da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1191da177e4SLinus Torvalds * kernel data space before using them.. 1201da177e4SLinus Torvalds * 1211da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1221da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1231da177e4SLinus Torvalds */ 1247950e385SJeff Layton 125fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12691a27b2aSJeff Layton 12751f39a1fSDavid Drysdale struct filename * 12891a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 12991a27b2aSJeff Layton { 13094b5d262SAl Viro struct filename *result; 1317950e385SJeff Layton char *kname; 13294b5d262SAl Viro int len; 1331da177e4SLinus Torvalds 1347ac86265SJeff Layton result = audit_reusename(filename); 1357ac86265SJeff Layton if (result) 1367ac86265SJeff Layton return result; 1377ac86265SJeff Layton 1387950e385SJeff Layton result = __getname(); 1393f9f0aa6SLinus Torvalds if (unlikely(!result)) 1404043cde8SEric Paris return ERR_PTR(-ENOMEM); 1411da177e4SLinus Torvalds 1427950e385SJeff Layton /* 1437950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1447950e385SJeff Layton * allocation 1457950e385SJeff Layton */ 146fd2f7cb5SAl Viro kname = (char *)result->iname; 14791a27b2aSJeff Layton result->name = kname; 1487950e385SJeff Layton 14994b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15091a27b2aSJeff Layton if (unlikely(len < 0)) { 15194b5d262SAl Viro __putname(result); 15294b5d262SAl Viro return ERR_PTR(len); 15391a27b2aSJeff Layton } 1543f9f0aa6SLinus Torvalds 1557950e385SJeff Layton /* 1567950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1577950e385SJeff Layton * separate struct filename so we can dedicate the entire 1587950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1597950e385SJeff Layton * userland. 1607950e385SJeff Layton */ 16194b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 162fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1637950e385SJeff Layton kname = (char *)result; 1647950e385SJeff Layton 165fd2f7cb5SAl Viro /* 166fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 167fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 168fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 169fd2f7cb5SAl Viro */ 170fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17194b5d262SAl Viro if (unlikely(!result)) { 17294b5d262SAl Viro __putname(kname); 17394b5d262SAl Viro return ERR_PTR(-ENOMEM); 1747950e385SJeff Layton } 1757950e385SJeff Layton result->name = kname; 17694b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17794b5d262SAl Viro if (unlikely(len < 0)) { 17894b5d262SAl Viro __putname(kname); 17994b5d262SAl Viro kfree(result); 18094b5d262SAl Viro return ERR_PTR(len); 18194b5d262SAl Viro } 18294b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18394b5d262SAl Viro __putname(kname); 18494b5d262SAl Viro kfree(result); 18594b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18694b5d262SAl Viro } 1877950e385SJeff Layton } 1887950e385SJeff Layton 18994b5d262SAl Viro result->refcnt = 1; 1903f9f0aa6SLinus Torvalds /* The empty path is special. */ 1913f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1923f9f0aa6SLinus Torvalds if (empty) 1931fa1e7f6SAndy Whitcroft *empty = 1; 19494b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19594b5d262SAl Viro putname(result); 19694b5d262SAl Viro return ERR_PTR(-ENOENT); 1971da177e4SLinus Torvalds } 19894b5d262SAl Viro } 1997950e385SJeff Layton 2007950e385SJeff Layton result->uptr = filename; 201c4ad8f98SLinus Torvalds result->aname = NULL; 2021da177e4SLinus Torvalds audit_getname(result); 2031da177e4SLinus Torvalds return result; 2043f9f0aa6SLinus Torvalds } 2053f9f0aa6SLinus Torvalds 20691a27b2aSJeff Layton struct filename * 2078228e2c3SDmitry Kadashev getname_uflags(const char __user *filename, int uflags) 2088228e2c3SDmitry Kadashev { 2098228e2c3SDmitry Kadashev int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 2108228e2c3SDmitry Kadashev 2118228e2c3SDmitry Kadashev return getname_flags(filename, flags, NULL); 2128228e2c3SDmitry Kadashev } 2138228e2c3SDmitry Kadashev 2148228e2c3SDmitry Kadashev struct filename * 21591a27b2aSJeff Layton getname(const char __user * filename) 216f52e0c11SAl Viro { 217f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 218f52e0c11SAl Viro } 219f52e0c11SAl Viro 220c4ad8f98SLinus Torvalds struct filename * 221c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 222c4ad8f98SLinus Torvalds { 223c4ad8f98SLinus Torvalds struct filename *result; 22408518549SPaul Moore int len = strlen(filename) + 1; 225c4ad8f98SLinus Torvalds 226c4ad8f98SLinus Torvalds result = __getname(); 227c4ad8f98SLinus Torvalds if (unlikely(!result)) 228c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 229c4ad8f98SLinus Torvalds 23008518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 231fd2f7cb5SAl Viro result->name = (char *)result->iname; 23208518549SPaul Moore } else if (len <= PATH_MAX) { 23330ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 23408518549SPaul Moore struct filename *tmp; 23508518549SPaul Moore 23630ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 23708518549SPaul Moore if (unlikely(!tmp)) { 23808518549SPaul Moore __putname(result); 23908518549SPaul Moore return ERR_PTR(-ENOMEM); 24008518549SPaul Moore } 24108518549SPaul Moore tmp->name = (char *)result; 24208518549SPaul Moore result = tmp; 24308518549SPaul Moore } else { 24408518549SPaul Moore __putname(result); 24508518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 24608518549SPaul Moore } 24708518549SPaul Moore memcpy((char *)result->name, filename, len); 248c4ad8f98SLinus Torvalds result->uptr = NULL; 249c4ad8f98SLinus Torvalds result->aname = NULL; 25055422d0bSPaul Moore result->refcnt = 1; 251fd3522fdSPaul Moore audit_getname(result); 252c4ad8f98SLinus Torvalds 253c4ad8f98SLinus Torvalds return result; 254c4ad8f98SLinus Torvalds } 255c4ad8f98SLinus Torvalds 25691a27b2aSJeff Layton void putname(struct filename *name) 2571da177e4SLinus Torvalds { 258ea47ab11SAl Viro if (IS_ERR(name)) 25991ef658fSDmitry Kadashev return; 26091ef658fSDmitry Kadashev 26155422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 26255422d0bSPaul Moore 26355422d0bSPaul Moore if (--name->refcnt > 0) 26455422d0bSPaul Moore return; 26555422d0bSPaul Moore 266fd2f7cb5SAl Viro if (name->name != name->iname) { 26755422d0bSPaul Moore __putname(name->name); 26855422d0bSPaul Moore kfree(name); 26955422d0bSPaul Moore } else 27055422d0bSPaul Moore __putname(name); 2711da177e4SLinus Torvalds } 2721da177e4SLinus Torvalds 27347291baaSChristian Brauner /** 27447291baaSChristian Brauner * check_acl - perform ACL permission checking 27547291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 27647291baaSChristian Brauner * @inode: inode to check permissions on 27747291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 27847291baaSChristian Brauner * 27947291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 28047291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 28147291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 28247291baaSChristian Brauner * 28347291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 28447291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 28547291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 28647291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 28747291baaSChristian Brauner * raw inode simply passs init_user_ns. 28847291baaSChristian Brauner */ 28947291baaSChristian Brauner static int check_acl(struct user_namespace *mnt_userns, 29047291baaSChristian Brauner struct inode *inode, int mask) 291e77819e5SLinus Torvalds { 29284635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 293e77819e5SLinus Torvalds struct posix_acl *acl; 294e77819e5SLinus Torvalds 295e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2963567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2973567866bSAl Viro if (!acl) 298e77819e5SLinus Torvalds return -EAGAIN; 2993567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 300b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 301e77819e5SLinus Torvalds return -ECHILD; 30247291baaSChristian Brauner return posix_acl_permission(mnt_userns, inode, acl, mask); 303e77819e5SLinus Torvalds } 304e77819e5SLinus Torvalds 3052982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 3064e34e719SChristoph Hellwig if (IS_ERR(acl)) 3074e34e719SChristoph Hellwig return PTR_ERR(acl); 308e77819e5SLinus Torvalds if (acl) { 30947291baaSChristian Brauner int error = posix_acl_permission(mnt_userns, inode, acl, mask); 310e77819e5SLinus Torvalds posix_acl_release(acl); 311e77819e5SLinus Torvalds return error; 312e77819e5SLinus Torvalds } 31384635d68SLinus Torvalds #endif 314e77819e5SLinus Torvalds 315e77819e5SLinus Torvalds return -EAGAIN; 316e77819e5SLinus Torvalds } 317e77819e5SLinus Torvalds 31847291baaSChristian Brauner /** 31947291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 32047291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 32147291baaSChristian Brauner * @inode: inode to check permissions on 32247291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3235fc475b7SLinus Torvalds * 32447291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 32547291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 32647291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 32747291baaSChristian Brauner * 32847291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 32947291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 33047291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 33147291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 33247291baaSChristian Brauner * raw inode simply passs init_user_ns. 3335909ccaaSLinus Torvalds */ 33447291baaSChristian Brauner static int acl_permission_check(struct user_namespace *mnt_userns, 33547291baaSChristian Brauner struct inode *inode, int mask) 3365909ccaaSLinus Torvalds { 33726cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 33847291baaSChristian Brauner kuid_t i_uid; 3395909ccaaSLinus Torvalds 3405fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 34147291baaSChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 34247291baaSChristian Brauner if (likely(uid_eq(current_fsuid(), i_uid))) { 3435fc475b7SLinus Torvalds mask &= 7; 3445909ccaaSLinus Torvalds mode >>= 6; 3455fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3465fc475b7SLinus Torvalds } 3475fc475b7SLinus Torvalds 3485fc475b7SLinus Torvalds /* Do we have ACL's? */ 349e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 35047291baaSChristian Brauner int error = check_acl(mnt_userns, inode, mask); 3515909ccaaSLinus Torvalds if (error != -EAGAIN) 3525909ccaaSLinus Torvalds return error; 3535909ccaaSLinus Torvalds } 3545909ccaaSLinus Torvalds 3555fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3565fc475b7SLinus Torvalds mask &= 7; 3575fc475b7SLinus Torvalds 3585fc475b7SLinus Torvalds /* 3595fc475b7SLinus Torvalds * Are the group permissions different from 3605fc475b7SLinus Torvalds * the other permissions in the bits we care 3615fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3625fc475b7SLinus Torvalds */ 3635fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 36447291baaSChristian Brauner kgid_t kgid = i_gid_into_mnt(mnt_userns, inode); 36547291baaSChristian Brauner if (in_group_p(kgid)) 3665909ccaaSLinus Torvalds mode >>= 3; 3675909ccaaSLinus Torvalds } 3685909ccaaSLinus Torvalds 3695fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3705fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3715909ccaaSLinus Torvalds } 3721da177e4SLinus Torvalds 3731da177e4SLinus Torvalds /** 3741da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 37547291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 3761da177e4SLinus Torvalds * @inode: inode to check access rights for 3775fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3785fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3791da177e4SLinus Torvalds * 3801da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3811da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3821da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 383b74c79e9SNick Piggin * are used for other things. 384b74c79e9SNick Piggin * 385b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 386b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 387b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 38847291baaSChristian Brauner * 38947291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39047291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39147291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39247291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39347291baaSChristian Brauner * raw inode simply passs init_user_ns. 3941da177e4SLinus Torvalds */ 39547291baaSChristian Brauner int generic_permission(struct user_namespace *mnt_userns, struct inode *inode, 39647291baaSChristian Brauner int mask) 3971da177e4SLinus Torvalds { 3985909ccaaSLinus Torvalds int ret; 3991da177e4SLinus Torvalds 4001da177e4SLinus Torvalds /* 401948409c7SAndreas Gruenbacher * Do the basic permission checks. 4021da177e4SLinus Torvalds */ 40347291baaSChristian Brauner ret = acl_permission_check(mnt_userns, inode, mask); 4045909ccaaSLinus Torvalds if (ret != -EACCES) 4055909ccaaSLinus Torvalds return ret; 4061da177e4SLinus Torvalds 407d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 408d594e7ecSAl Viro /* DACs are overridable for directories */ 409d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 41047291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 41123adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 412d594e7ecSAl Viro return 0; 41347291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4140558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4151da177e4SLinus Torvalds return 0; 4162a4c2242SStephen Smalley return -EACCES; 4172a4c2242SStephen Smalley } 4181da177e4SLinus Torvalds 4191da177e4SLinus Torvalds /* 4201da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4211da177e4SLinus Torvalds */ 4227ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 423d594e7ecSAl Viro if (mask == MAY_READ) 42447291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4250558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4261da177e4SLinus Torvalds return 0; 4272a4c2242SStephen Smalley /* 4282a4c2242SStephen Smalley * Read/write DACs are always overridable. 4292a4c2242SStephen Smalley * Executable DACs are overridable when there is 4302a4c2242SStephen Smalley * at least one exec bit set. 4312a4c2242SStephen Smalley */ 4322a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 43347291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4340558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4352a4c2242SStephen Smalley return 0; 4361da177e4SLinus Torvalds 4371da177e4SLinus Torvalds return -EACCES; 4381da177e4SLinus Torvalds } 4394d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4401da177e4SLinus Torvalds 44147291baaSChristian Brauner /** 44247291baaSChristian Brauner * do_inode_permission - UNIX permission checking 44347291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 44447291baaSChristian Brauner * @inode: inode to check permissions on 44547291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 44647291baaSChristian Brauner * 4473ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4483ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4493ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4503ddcd056SLinus Torvalds * permission function, use the fast case". 4513ddcd056SLinus Torvalds */ 45247291baaSChristian Brauner static inline int do_inode_permission(struct user_namespace *mnt_userns, 45347291baaSChristian Brauner struct inode *inode, int mask) 4543ddcd056SLinus Torvalds { 4553ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4563ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 457549c7297SChristian Brauner return inode->i_op->permission(mnt_userns, inode, mask); 4583ddcd056SLinus Torvalds 4593ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4603ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4613ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4623ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4633ddcd056SLinus Torvalds } 46447291baaSChristian Brauner return generic_permission(mnt_userns, inode, mask); 4653ddcd056SLinus Torvalds } 4663ddcd056SLinus Torvalds 467cb23beb5SChristoph Hellwig /** 4680bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4690bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 47055852635SRandy Dunlap * @inode: Inode to check permission on 4710bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4720bdaea90SDavid Howells * 4730bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4740bdaea90SDavid Howells */ 4750bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4760bdaea90SDavid Howells { 4770bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4780bdaea90SDavid Howells umode_t mode = inode->i_mode; 4790bdaea90SDavid Howells 4800bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 481bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4820bdaea90SDavid Howells return -EROFS; 4830bdaea90SDavid Howells } 4840bdaea90SDavid Howells return 0; 4850bdaea90SDavid Howells } 4860bdaea90SDavid Howells 4870bdaea90SDavid Howells /** 4880bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 48947291baaSChristian Brauner * @mnt_userns: User namespace of the mount the inode was found from 4900bdaea90SDavid Howells * @inode: Inode to check permission on 4910bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4920bdaea90SDavid Howells * 4930bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4940bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4950bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4960bdaea90SDavid Howells * 4970bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4980bdaea90SDavid Howells */ 49947291baaSChristian Brauner int inode_permission(struct user_namespace *mnt_userns, 50047291baaSChristian Brauner struct inode *inode, int mask) 5010bdaea90SDavid Howells { 5020bdaea90SDavid Howells int retval; 5030bdaea90SDavid Howells 5040bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 5050bdaea90SDavid Howells if (retval) 5060bdaea90SDavid Howells return retval; 5074bfd054aSEric Biggers 5084bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 5094bfd054aSEric Biggers /* 5104bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5114bfd054aSEric Biggers */ 5124bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5134bfd054aSEric Biggers return -EPERM; 5144bfd054aSEric Biggers 5154bfd054aSEric Biggers /* 5164bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5174bfd054aSEric Biggers * written back improperly if their true value is unknown 5184bfd054aSEric Biggers * to the vfs. 5194bfd054aSEric Biggers */ 520ba73d987SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 5214bfd054aSEric Biggers return -EACCES; 5224bfd054aSEric Biggers } 5234bfd054aSEric Biggers 52447291baaSChristian Brauner retval = do_inode_permission(mnt_userns, inode, mask); 5254bfd054aSEric Biggers if (retval) 5264bfd054aSEric Biggers return retval; 5274bfd054aSEric Biggers 5284bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5294bfd054aSEric Biggers if (retval) 5304bfd054aSEric Biggers return retval; 5314bfd054aSEric Biggers 5324bfd054aSEric Biggers return security_inode_permission(inode, mask); 5330bdaea90SDavid Howells } 5344d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5350bdaea90SDavid Howells 5360bdaea90SDavid Howells /** 5375dd784d0SJan Blunck * path_get - get a reference to a path 5385dd784d0SJan Blunck * @path: path to get the reference to 5395dd784d0SJan Blunck * 5405dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5415dd784d0SJan Blunck */ 542dcf787f3SAl Viro void path_get(const struct path *path) 5435dd784d0SJan Blunck { 5445dd784d0SJan Blunck mntget(path->mnt); 5455dd784d0SJan Blunck dget(path->dentry); 5465dd784d0SJan Blunck } 5475dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5485dd784d0SJan Blunck 5495dd784d0SJan Blunck /** 5501d957f9bSJan Blunck * path_put - put a reference to a path 5511d957f9bSJan Blunck * @path: path to put the reference to 5521d957f9bSJan Blunck * 5531d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5541d957f9bSJan Blunck */ 555dcf787f3SAl Viro void path_put(const struct path *path) 5561da177e4SLinus Torvalds { 5571d957f9bSJan Blunck dput(path->dentry); 5581d957f9bSJan Blunck mntput(path->mnt); 5591da177e4SLinus Torvalds } 5601d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5611da177e4SLinus Torvalds 562894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5631f55a6ecSAl Viro struct nameidata { 5641f55a6ecSAl Viro struct path path; 5651f55a6ecSAl Viro struct qstr last; 5661f55a6ecSAl Viro struct path root; 5671f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 568bcba1e7dSAl Viro unsigned int flags, state; 569ab87f9a5SAleksa Sarai unsigned seq, m_seq, r_seq; 5701f55a6ecSAl Viro int last_type; 5711f55a6ecSAl Viro unsigned depth; 572756daf26SNeilBrown int total_link_count; 573697fc6caSAl Viro struct saved { 574697fc6caSAl Viro struct path link; 575fceef393SAl Viro struct delayed_call done; 576697fc6caSAl Viro const char *name; 5770450b2d1SAl Viro unsigned seq; 578894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5799883d185SAl Viro struct filename *name; 5809883d185SAl Viro struct nameidata *saved; 5819883d185SAl Viro unsigned root_seq; 5829883d185SAl Viro int dfd; 5830f705953SAl Viro kuid_t dir_uid; 5840f705953SAl Viro umode_t dir_mode; 5853859a271SKees Cook } __randomize_layout; 5861f55a6ecSAl Viro 587bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 588bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 589bcba1e7dSAl Viro #define ND_JUMPED 4 590bcba1e7dSAl Viro 59106422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 592894bc8c4SAl Viro { 593756daf26SNeilBrown struct nameidata *old = current->nameidata; 594756daf26SNeilBrown p->stack = p->internal; 5957962c7d1SAl Viro p->depth = 0; 596c8a53ee5SAl Viro p->dfd = dfd; 597c8a53ee5SAl Viro p->name = name; 5987d01ef75SAl Viro p->path.mnt = NULL; 5997d01ef75SAl Viro p->path.dentry = NULL; 600756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 6019883d185SAl Viro p->saved = old; 602756daf26SNeilBrown current->nameidata = p; 603894bc8c4SAl Viro } 604894bc8c4SAl Viro 60506422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 60606422964SAl Viro const struct path *root) 60706422964SAl Viro { 60806422964SAl Viro __set_nameidata(p, dfd, name); 60906422964SAl Viro p->state = 0; 61006422964SAl Viro if (unlikely(root)) { 61106422964SAl Viro p->state = ND_ROOT_PRESET; 61206422964SAl Viro p->root = *root; 61306422964SAl Viro } 61406422964SAl Viro } 61506422964SAl Viro 6169883d185SAl Viro static void restore_nameidata(void) 617894bc8c4SAl Viro { 6189883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 619756daf26SNeilBrown 620756daf26SNeilBrown current->nameidata = old; 621756daf26SNeilBrown if (old) 622756daf26SNeilBrown old->total_link_count = now->total_link_count; 623e1a63bbcSAl Viro if (now->stack != now->internal) 624756daf26SNeilBrown kfree(now->stack); 625894bc8c4SAl Viro } 626894bc8c4SAl Viro 62760ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 628894bc8c4SAl Viro { 629bc40aee0SAl Viro struct saved *p; 630bc40aee0SAl Viro 6316da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 63260ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 633bc40aee0SAl Viro if (unlikely(!p)) 63460ef60c7SAl Viro return false; 635894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 636894bc8c4SAl Viro nd->stack = p; 63760ef60c7SAl Viro return true; 638894bc8c4SAl Viro } 639894bc8c4SAl Viro 640397d425dSEric W. Biederman /** 6416b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 642397d425dSEric W. Biederman * 643397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 644397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 645397d425dSEric W. Biederman */ 6466b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 647397d425dSEric W. Biederman { 64895dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 649397d425dSEric W. Biederman 650402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 651402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 652397d425dSEric W. Biederman return true; 653397d425dSEric W. Biederman 6546b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 655397d425dSEric W. Biederman } 656397d425dSEric W. Biederman 6577973387aSAl Viro static void drop_links(struct nameidata *nd) 6587973387aSAl Viro { 6597973387aSAl Viro int i = nd->depth; 6607973387aSAl Viro while (i--) { 6617973387aSAl Viro struct saved *last = nd->stack + i; 662fceef393SAl Viro do_delayed_call(&last->done); 663fceef393SAl Viro clear_delayed_call(&last->done); 6647973387aSAl Viro } 6657973387aSAl Viro } 6667973387aSAl Viro 6677973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6687973387aSAl Viro { 6697973387aSAl Viro drop_links(nd); 6707973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6717973387aSAl Viro int i; 6727973387aSAl Viro path_put(&nd->path); 6737973387aSAl Viro for (i = 0; i < nd->depth; i++) 6747973387aSAl Viro path_put(&nd->stack[i].link); 675bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 676102b8af2SAl Viro path_put(&nd->root); 677bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 678102b8af2SAl Viro } 6797973387aSAl Viro } else { 6807973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6817973387aSAl Viro rcu_read_unlock(); 6827973387aSAl Viro } 6837973387aSAl Viro nd->depth = 0; 6847d01ef75SAl Viro nd->path.mnt = NULL; 6857d01ef75SAl Viro nd->path.dentry = NULL; 6867973387aSAl Viro } 6877973387aSAl Viro 6887973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6892aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6907973387aSAl Viro { 6912aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6927973387aSAl Viro if (unlikely(res)) { 6937973387aSAl Viro if (res > 0) 6947973387aSAl Viro path->mnt = NULL; 6957973387aSAl Viro path->dentry = NULL; 6967973387aSAl Viro return false; 6977973387aSAl Viro } 6987973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6997973387aSAl Viro path->dentry = NULL; 7007973387aSAl Viro return false; 7017973387aSAl Viro } 7027973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7037973387aSAl Viro } 7047973387aSAl Viro 7052aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7062aa38470SAl Viro struct path *path, unsigned seq) 7072aa38470SAl Viro { 7085bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7092aa38470SAl Viro } 7102aa38470SAl Viro 7117973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7127973387aSAl Viro { 7137973387aSAl Viro int i; 714eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 715eacd9aa8SAl Viro drop_links(nd); 716eacd9aa8SAl Viro nd->depth = 0; 717eacd9aa8SAl Viro return false; 718eacd9aa8SAl Viro } 7197973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7207973387aSAl Viro struct saved *last = nd->stack + i; 7217973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7227973387aSAl Viro drop_links(nd); 7237973387aSAl Viro nd->depth = i + 1; 7247973387aSAl Viro return false; 7257973387aSAl Viro } 7267973387aSAl Viro } 7277973387aSAl Viro return true; 7287973387aSAl Viro } 7297973387aSAl Viro 730ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 731ee594bffSAl Viro { 732adb21d2bSAleksa Sarai /* 733adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 734adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 735adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 736adb21d2bSAleksa Sarai */ 737adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 738adb21d2bSAleksa Sarai return false; 739adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 740bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 741ee594bffSAl Viro return true; 742bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 743ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 744ee594bffSAl Viro } 745ee594bffSAl Viro 74619660af7SAl Viro /* 74731e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 74819660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 74919660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 75057e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 75119660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 75219660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 75319660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 75419660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 75531e6b01fSNick Piggin */ 75631e6b01fSNick Piggin 75731e6b01fSNick Piggin /** 758e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 75919660af7SAl Viro * @nd: nameidata pathwalk data 760e36cffedSJens Axboe * Returns: true on success, false on failure 76131e6b01fSNick Piggin * 762e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7634675ac39SAl Viro * for ref-walk mode. 7644675ac39SAl Viro * Must be called from rcu-walk context. 765e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7667973387aSAl Viro * terminate_walk(). 76731e6b01fSNick Piggin */ 768e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 76931e6b01fSNick Piggin { 77031e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 77131e6b01fSNick Piggin 77231e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 773e5c832d5SLinus Torvalds 774e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 7757973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7764675ac39SAl Viro goto out1; 77784a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 77884a2bd39SAl Viro goto out; 779ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7804675ac39SAl Viro goto out; 7814675ac39SAl Viro rcu_read_unlock(); 7824675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 783e36cffedSJens Axboe return true; 7844675ac39SAl Viro 78584a2bd39SAl Viro out1: 7864675ac39SAl Viro nd->path.mnt = NULL; 7874675ac39SAl Viro nd->path.dentry = NULL; 7884675ac39SAl Viro out: 7894675ac39SAl Viro rcu_read_unlock(); 790e36cffedSJens Axboe return false; 7914675ac39SAl Viro } 7924675ac39SAl Viro 7934675ac39SAl Viro /** 794ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7954675ac39SAl Viro * @nd: nameidata pathwalk data 796ae66db45SAl Viro * @dentry: next dentry to step into 797ae66db45SAl Viro * @seq: seq number to check @dentry against 798ae66db45SAl Viro * Returns: true on success, false on failure 7994675ac39SAl Viro * 800ae66db45SAl Viro * Similar to to try_to_unlazy(), but here we have the next dentry already 801ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 802ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 803ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 8044675ac39SAl Viro * terminate_walk(). 8054675ac39SAl Viro */ 806ae66db45SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq) 8074675ac39SAl Viro { 8084675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8094675ac39SAl Viro 8104675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 8114675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8124675ac39SAl Viro goto out2; 8137973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 8147973387aSAl Viro goto out2; 8154675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8167973387aSAl Viro goto out1; 81748a066e7SAl Viro 81815570086SLinus Torvalds /* 8194675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8204675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8214675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8224675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8234675ac39SAl Viro * know the parent must still be valid if the child sequence number is 82415570086SLinus Torvalds */ 8254675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 826e5c832d5SLinus Torvalds goto out; 82784a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 82884a2bd39SAl Viro goto out_dput; 829e5c832d5SLinus Torvalds /* 830e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 831e5c832d5SLinus Torvalds * still valid and get it if required. 832e5c832d5SLinus Torvalds */ 83384a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 83484a2bd39SAl Viro goto out_dput; 8358b61e74fSAl Viro rcu_read_unlock(); 836ae66db45SAl Viro return true; 83719660af7SAl Viro 8387973387aSAl Viro out2: 8397973387aSAl Viro nd->path.mnt = NULL; 8407973387aSAl Viro out1: 8417973387aSAl Viro nd->path.dentry = NULL; 842e5c832d5SLinus Torvalds out: 8438b61e74fSAl Viro rcu_read_unlock(); 844ae66db45SAl Viro return false; 84584a2bd39SAl Viro out_dput: 84684a2bd39SAl Viro rcu_read_unlock(); 84784a2bd39SAl Viro dput(dentry); 848ae66db45SAl Viro return false; 84931e6b01fSNick Piggin } 85031e6b01fSNick Piggin 8514ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 85234286d66SNick Piggin { 853a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8544ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 855a89f8337SAl Viro else 856a89f8337SAl Viro return 1; 85734286d66SNick Piggin } 85834286d66SNick Piggin 8599f1fafeeSAl Viro /** 8609f1fafeeSAl Viro * complete_walk - successful completion of path walk 8619f1fafeeSAl Viro * @nd: pointer nameidata 86239159de2SJeff Layton * 8639f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8649f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8659f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8669f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8679f1fafeeSAl Viro * need to drop nd->path. 86839159de2SJeff Layton */ 8699f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 87039159de2SJeff Layton { 87116c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 87239159de2SJeff Layton int status; 87339159de2SJeff Layton 8749f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 875adb21d2bSAleksa Sarai /* 876adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 877adb21d2bSAleksa Sarai * externally-managed nd->root. 878adb21d2bSAleksa Sarai */ 879bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 880bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8819f1fafeeSAl Viro nd->root.mnt = NULL; 8826c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 883e36cffedSJens Axboe if (!try_to_unlazy(nd)) 88448a066e7SAl Viro return -ECHILD; 88548a066e7SAl Viro } 8869f1fafeeSAl Viro 887adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 888adb21d2bSAleksa Sarai /* 889adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 890adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 891adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 892adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 893adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 894adb21d2bSAleksa Sarai * 895adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 896adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 897adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 898adb21d2bSAleksa Sarai * requested root to userspace. 899adb21d2bSAleksa Sarai * 900adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 901adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 902adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 903adb21d2bSAleksa Sarai */ 904adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 905adb21d2bSAleksa Sarai return -EXDEV; 906adb21d2bSAleksa Sarai } 907adb21d2bSAleksa Sarai 908bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 90939159de2SJeff Layton return 0; 91039159de2SJeff Layton 911ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 91216c2cd71SAl Viro return 0; 91316c2cd71SAl Viro 914ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 91539159de2SJeff Layton if (status > 0) 91639159de2SJeff Layton return 0; 91739159de2SJeff Layton 91816c2cd71SAl Viro if (!status) 91939159de2SJeff Layton status = -ESTALE; 92016c2cd71SAl Viro 92139159de2SJeff Layton return status; 92239159de2SJeff Layton } 92339159de2SJeff Layton 924740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9252a737871SAl Viro { 92631e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9279e6697e2SAl Viro 928adb21d2bSAleksa Sarai /* 929adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 930adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 931adb21d2bSAleksa Sarai * from the dirfd. 932adb21d2bSAleksa Sarai */ 933adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 934adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 935adb21d2bSAleksa Sarai 9369e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9378f47a016SAl Viro unsigned seq; 938c28cc364SNick Piggin 939c28cc364SNick Piggin do { 940c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 94131e6b01fSNick Piggin nd->root = fs->root; 9428f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 943c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9449e6697e2SAl Viro } else { 9459e6697e2SAl Viro get_fs_root(fs, &nd->root); 946bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9479e6697e2SAl Viro } 948740a1678SAleksa Sarai return 0; 94931e6b01fSNick Piggin } 95031e6b01fSNick Piggin 951248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 952248fb5b9SAl Viro { 953adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 954adb21d2bSAleksa Sarai return -EXDEV; 95572ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 95672ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 95772ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 95872ba2929SAleksa Sarai return -EXDEV; 95972ba2929SAleksa Sarai } 960740a1678SAleksa Sarai if (!nd->root.mnt) { 961740a1678SAleksa Sarai int error = set_root(nd); 962740a1678SAleksa Sarai if (error) 963740a1678SAleksa Sarai return error; 964740a1678SAleksa Sarai } 965248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 966248fb5b9SAl Viro struct dentry *d; 967248fb5b9SAl Viro nd->path = nd->root; 968248fb5b9SAl Viro d = nd->path.dentry; 969248fb5b9SAl Viro nd->inode = d->d_inode; 970248fb5b9SAl Viro nd->seq = nd->root_seq; 971248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 972248fb5b9SAl Viro return -ECHILD; 973248fb5b9SAl Viro } else { 974248fb5b9SAl Viro path_put(&nd->path); 975248fb5b9SAl Viro nd->path = nd->root; 976248fb5b9SAl Viro path_get(&nd->path); 977248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 978248fb5b9SAl Viro } 979bcba1e7dSAl Viro nd->state |= ND_JUMPED; 980248fb5b9SAl Viro return 0; 981248fb5b9SAl Viro } 982248fb5b9SAl Viro 983b5fb63c1SChristoph Hellwig /* 9846b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 985b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 986b5fb63c1SChristoph Hellwig */ 9871bc82070SAleksa Sarai int nd_jump_link(struct path *path) 988b5fb63c1SChristoph Hellwig { 9894b99d499SAleksa Sarai int error = -ELOOP; 9906e77137bSAl Viro struct nameidata *nd = current->nameidata; 991b5fb63c1SChristoph Hellwig 9924b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9934b99d499SAleksa Sarai goto err; 9944b99d499SAleksa Sarai 99572ba2929SAleksa Sarai error = -EXDEV; 99672ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 99772ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 99872ba2929SAleksa Sarai goto err; 99972ba2929SAleksa Sarai } 1000adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 1001adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 1002adb21d2bSAleksa Sarai goto err; 100372ba2929SAleksa Sarai 10044b99d499SAleksa Sarai path_put(&nd->path); 1005b5fb63c1SChristoph Hellwig nd->path = *path; 1006b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1007bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10081bc82070SAleksa Sarai return 0; 10094b99d499SAleksa Sarai 10104b99d499SAleksa Sarai err: 10114b99d499SAleksa Sarai path_put(path); 10124b99d499SAleksa Sarai return error; 1013b5fb63c1SChristoph Hellwig } 1014b5fb63c1SChristoph Hellwig 1015b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1016574197e0SAl Viro { 101721c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1018fceef393SAl Viro do_delayed_call(&last->done); 10196548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1020b9ff4429SAl Viro path_put(&last->link); 1021574197e0SAl Viro } 1022574197e0SAl Viro 10239c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10249c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10259c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10269c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10279c011be1SLuis Chamberlain 10289c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10299c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10309c011be1SLuis Chamberlain { 10319c011be1SLuis Chamberlain .procname = "protected_symlinks", 10329c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10339c011be1SLuis Chamberlain .maxlen = sizeof(int), 10349c011be1SLuis Chamberlain .mode = 0600, 10359c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10369c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10379c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10389c011be1SLuis Chamberlain }, 10399c011be1SLuis Chamberlain { 10409c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10419c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10429c011be1SLuis Chamberlain .maxlen = sizeof(int), 10439c011be1SLuis Chamberlain .mode = 0600, 10449c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10459c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10469c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10479c011be1SLuis Chamberlain }, 10489c011be1SLuis Chamberlain { 10499c011be1SLuis Chamberlain .procname = "protected_fifos", 10509c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10519c011be1SLuis Chamberlain .maxlen = sizeof(int), 10529c011be1SLuis Chamberlain .mode = 0600, 10539c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10549c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10559c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10569c011be1SLuis Chamberlain }, 10579c011be1SLuis Chamberlain { 10589c011be1SLuis Chamberlain .procname = "protected_regular", 10599c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10609c011be1SLuis Chamberlain .maxlen = sizeof(int), 10619c011be1SLuis Chamberlain .mode = 0600, 10629c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10639c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10649c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10659c011be1SLuis Chamberlain }, 10669c011be1SLuis Chamberlain { } 10679c011be1SLuis Chamberlain }; 10689c011be1SLuis Chamberlain 10699c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10709c011be1SLuis Chamberlain { 10719c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10729c011be1SLuis Chamberlain return 0; 10739c011be1SLuis Chamberlain } 10749c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10759c011be1SLuis Chamberlain 10769c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1077800179c9SKees Cook 1078800179c9SKees Cook /** 1079800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 108055852635SRandy Dunlap * @nd: nameidata pathwalk data 1081800179c9SKees Cook * 1082800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1083800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1084800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1085800179c9SKees Cook * processes from failing races against path names that may change out 1086800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1087800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1088800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1089800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1090800179c9SKees Cook * 1091800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1092800179c9SKees Cook */ 1093ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1094800179c9SKees Cook { 1095ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1096ba73d987SChristian Brauner kuid_t i_uid; 1097ba73d987SChristian Brauner 1098800179c9SKees Cook if (!sysctl_protected_symlinks) 1099800179c9SKees Cook return 0; 1100800179c9SKees Cook 1101ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1102ba73d987SChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 1103800179c9SKees Cook /* Allowed if owner and follower match. */ 1104ba73d987SChristian Brauner if (uid_eq(current_cred()->fsuid, i_uid)) 1105800179c9SKees Cook return 0; 1106800179c9SKees Cook 1107800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11080f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1109800179c9SKees Cook return 0; 1110800179c9SKees Cook 1111800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1112ba73d987SChristian Brauner if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) 1113800179c9SKees Cook return 0; 1114800179c9SKees Cook 111531956502SAl Viro if (nd->flags & LOOKUP_RCU) 111631956502SAl Viro return -ECHILD; 111731956502SAl Viro 1118ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1119245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1120800179c9SKees Cook return -EACCES; 1121800179c9SKees Cook } 1122800179c9SKees Cook 1123800179c9SKees Cook /** 1124800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1125ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1126800179c9SKees Cook * @inode: the source inode to hardlink from 1127800179c9SKees Cook * 1128800179c9SKees Cook * Return false if at least one of the following conditions: 1129800179c9SKees Cook * - inode is not a regular file 1130800179c9SKees Cook * - inode is setuid 1131800179c9SKees Cook * - inode is setgid and group-exec 1132800179c9SKees Cook * - access failure for read and write 1133800179c9SKees Cook * 1134800179c9SKees Cook * Otherwise returns true. 1135800179c9SKees Cook */ 1136ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1137ba73d987SChristian Brauner struct inode *inode) 1138800179c9SKees Cook { 1139800179c9SKees Cook umode_t mode = inode->i_mode; 1140800179c9SKees Cook 1141800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1142800179c9SKees Cook if (!S_ISREG(mode)) 1143800179c9SKees Cook return false; 1144800179c9SKees Cook 1145800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1146800179c9SKees Cook if (mode & S_ISUID) 1147800179c9SKees Cook return false; 1148800179c9SKees Cook 1149800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1150800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1151800179c9SKees Cook return false; 1152800179c9SKees Cook 1153800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1154ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1155800179c9SKees Cook return false; 1156800179c9SKees Cook 1157800179c9SKees Cook return true; 1158800179c9SKees Cook } 1159800179c9SKees Cook 1160800179c9SKees Cook /** 1161800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1162ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1163800179c9SKees Cook * @link: the source to hardlink from 1164800179c9SKees Cook * 1165800179c9SKees Cook * Block hardlink when all of: 1166800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1167800179c9SKees Cook * - fsuid does not match inode 1168800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1169f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1170800179c9SKees Cook * 1171ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1172ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1173ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1174ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1175ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1176ba73d987SChristian Brauner * 1177800179c9SKees Cook * Returns 0 if successful, -ve on error. 1178800179c9SKees Cook */ 1179ba73d987SChristian Brauner int may_linkat(struct user_namespace *mnt_userns, struct path *link) 1180800179c9SKees Cook { 1181593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1182593d1ce8SEric W. Biederman 1183593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1184ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 1185ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 1186593d1ce8SEric W. Biederman return -EOVERFLOW; 1187800179c9SKees Cook 1188800179c9SKees Cook if (!sysctl_protected_hardlinks) 1189800179c9SKees Cook return 0; 1190800179c9SKees Cook 1191800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1192800179c9SKees Cook * otherwise, it must be a safe source. 1193800179c9SKees Cook */ 1194ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1195ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1196800179c9SKees Cook return 0; 1197800179c9SKees Cook 1198245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1199800179c9SKees Cook return -EPERM; 1200800179c9SKees Cook } 1201800179c9SKees Cook 120230aba665SSalvatore Mesoraca /** 120330aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 120430aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 120530aba665SSalvatore Mesoraca * exist. 1206ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 12072111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 120830aba665SSalvatore Mesoraca * @inode: the inode of the file to open 120930aba665SSalvatore Mesoraca * 121030aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 121130aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 121230aba665SSalvatore Mesoraca * - the file already exists 121330aba665SSalvatore Mesoraca * - we are in a sticky directory 121430aba665SSalvatore Mesoraca * - we don't own the file 121530aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 121630aba665SSalvatore Mesoraca * - the directory is world writable 121730aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 121830aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 121930aba665SSalvatore Mesoraca * be enough. 122030aba665SSalvatore Mesoraca * 1221ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1222ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1223ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1224ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1225ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1226ba73d987SChristian Brauner * 122730aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 122830aba665SSalvatore Mesoraca */ 1229ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1230ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 123130aba665SSalvatore Mesoraca { 1232ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1233ba73d987SChristian Brauner kuid_t dir_uid = nd->dir_uid; 1234ba73d987SChristian Brauner 123530aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 123630aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1237d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1238ba73d987SChristian Brauner uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || 1239ba73d987SChristian Brauner uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) 124030aba665SSalvatore Mesoraca return 0; 124130aba665SSalvatore Mesoraca 1242d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1243d0cb5018SAl Viro (dir_mode & 0020 && 124430aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 124530aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1246245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1247245d7369SKees Cook "sticky_create_fifo" : 1248245d7369SKees Cook "sticky_create_regular"; 1249245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 125030aba665SSalvatore Mesoraca return -EACCES; 125130aba665SSalvatore Mesoraca } 125230aba665SSalvatore Mesoraca return 0; 125330aba665SSalvatore Mesoraca } 125430aba665SSalvatore Mesoraca 1255f015f126SDavid Howells /* 1256f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1257f015f126SDavid Howells * 1258f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1259f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1260f015f126SDavid Howells * Up is towards /. 1261f015f126SDavid Howells * 1262f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1263f015f126SDavid Howells * root. 1264f015f126SDavid Howells */ 1265bab77ebfSAl Viro int follow_up(struct path *path) 12661da177e4SLinus Torvalds { 12670714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12680714a533SAl Viro struct mount *parent; 12691da177e4SLinus Torvalds struct dentry *mountpoint; 127099b7db7bSNick Piggin 127148a066e7SAl Viro read_seqlock_excl(&mount_lock); 12720714a533SAl Viro parent = mnt->mnt_parent; 12733c0a6163SAl Viro if (parent == mnt) { 127448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12751da177e4SLinus Torvalds return 0; 12761da177e4SLinus Torvalds } 12770714a533SAl Viro mntget(&parent->mnt); 1278a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 127948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1280bab77ebfSAl Viro dput(path->dentry); 1281bab77ebfSAl Viro path->dentry = mountpoint; 1282bab77ebfSAl Viro mntput(path->mnt); 12830714a533SAl Viro path->mnt = &parent->mnt; 12841da177e4SLinus Torvalds return 1; 12851da177e4SLinus Torvalds } 12864d359507SAl Viro EXPORT_SYMBOL(follow_up); 12871da177e4SLinus Torvalds 12887ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12897ef482faSAl Viro struct path *path, unsigned *seqp) 12907ef482faSAl Viro { 12917ef482faSAl Viro while (mnt_has_parent(m)) { 12927ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12937ef482faSAl Viro 12947ef482faSAl Viro m = m->mnt_parent; 12957ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12967ef482faSAl Viro root->mnt == &m->mnt)) 12977ef482faSAl Viro break; 12987ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 12997ef482faSAl Viro path->mnt = &m->mnt; 13007ef482faSAl Viro path->dentry = mountpoint; 13017ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 13027ef482faSAl Viro return true; 13037ef482faSAl Viro } 13047ef482faSAl Viro } 13057ef482faSAl Viro return false; 13067ef482faSAl Viro } 13077ef482faSAl Viro 13082aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 13092aa38470SAl Viro struct path *path) 13102aa38470SAl Viro { 13112aa38470SAl Viro bool found; 13122aa38470SAl Viro 13132aa38470SAl Viro rcu_read_lock(); 13142aa38470SAl Viro while (1) { 13152aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 13162aa38470SAl Viro 13172aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 13182aa38470SAl Viro if (unlikely(!found)) { 13192aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 13202aa38470SAl Viro break; 13212aa38470SAl Viro } else { 13222aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 13232aa38470SAl Viro break; 13242aa38470SAl Viro rcu_read_unlock(); 13252aa38470SAl Viro path_put(path); 13262aa38470SAl Viro rcu_read_lock(); 13272aa38470SAl Viro } 13282aa38470SAl Viro } 13292aa38470SAl Viro rcu_read_unlock(); 13302aa38470SAl Viro return found; 13312aa38470SAl Viro } 13322aa38470SAl Viro 1333b5c84bf6SNick Piggin /* 13349875cf80SDavid Howells * Perform an automount 13359875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 13369875cf80SDavid Howells * were called with. 13371da177e4SLinus Torvalds */ 13381c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 133931e6b01fSNick Piggin { 134025e195aaSAl Viro struct dentry *dentry = path->dentry; 13419875cf80SDavid Howells 13420ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 13430ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 13440ec26fd0SMiklos Szeredi * the name. 13450ec26fd0SMiklos Szeredi * 13460ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 13475a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 13480ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 13490ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13500ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13510ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13525a30d8a2SDavid Howells */ 13531c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13545d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 135525e195aaSAl Viro dentry->d_inode) 13569875cf80SDavid Howells return -EISDIR; 13570ec26fd0SMiklos Szeredi 13581c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13599875cf80SDavid Howells return -ELOOP; 13609875cf80SDavid Howells 136125e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1362ea5b778aSDavid Howells } 13639875cf80SDavid Howells 13649875cf80SDavid Howells /* 13659deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13669deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13679deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13689deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13699875cf80SDavid Howells */ 13709deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13719deed3ebSAl Viro int *count, unsigned lookup_flags) 13729875cf80SDavid Howells { 13739deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13749875cf80SDavid Howells bool need_mntput = false; 13758aef1884SAl Viro int ret = 0; 13769875cf80SDavid Howells 13779deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1378cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1379cc53ce53SDavid Howells * being held. */ 1380d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1381fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1382508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1383cc53ce53SDavid Howells if (ret < 0) 13848aef1884SAl Viro break; 1385cc53ce53SDavid Howells } 1386cc53ce53SDavid Howells 13879deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13889875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13899deed3ebSAl Viro if (mounted) { // ... in our namespace 13909875cf80SDavid Howells dput(path->dentry); 13919875cf80SDavid Howells if (need_mntput) 1392463ffb2eSAl Viro mntput(path->mnt); 1393463ffb2eSAl Viro path->mnt = mounted; 1394463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13959deed3ebSAl Viro // here we know it's positive 13969deed3ebSAl Viro flags = path->dentry->d_flags; 13979875cf80SDavid Howells need_mntput = true; 13989875cf80SDavid Howells continue; 1399463ffb2eSAl Viro } 14001da177e4SLinus Torvalds } 14019875cf80SDavid Howells 14029deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 14039deed3ebSAl Viro break; 14049deed3ebSAl Viro 14059deed3ebSAl Viro // uncovered automount point 14069deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 14079deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 14089875cf80SDavid Howells if (ret < 0) 14098aef1884SAl Viro break; 14109875cf80SDavid Howells } 14119875cf80SDavid Howells 14129deed3ebSAl Viro if (ret == -EISDIR) 14139deed3ebSAl Viro ret = 0; 14149deed3ebSAl Viro // possible if you race with several mount --move 14159deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 14168aef1884SAl Viro mntput(path->mnt); 14179deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1418d41efb52SAl Viro ret = -ENOENT; 14199deed3ebSAl Viro *jumped = need_mntput; 14208402752eSAl Viro return ret; 14211da177e4SLinus Torvalds } 14221da177e4SLinus Torvalds 14239deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 14249deed3ebSAl Viro int *count, unsigned lookup_flags) 14259deed3ebSAl Viro { 14269deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 14279deed3ebSAl Viro 14289deed3ebSAl Viro /* fastpath */ 14299deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 14309deed3ebSAl Viro *jumped = false; 14319deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 14329deed3ebSAl Viro return -ENOENT; 14339deed3ebSAl Viro return 0; 14349deed3ebSAl Viro } 14359deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 14369deed3ebSAl Viro } 14379deed3ebSAl Viro 1438cc53ce53SDavid Howells int follow_down_one(struct path *path) 14391da177e4SLinus Torvalds { 14401da177e4SLinus Torvalds struct vfsmount *mounted; 14411da177e4SLinus Torvalds 14421c755af4SAl Viro mounted = lookup_mnt(path); 14431da177e4SLinus Torvalds if (mounted) { 14449393bd07SAl Viro dput(path->dentry); 14459393bd07SAl Viro mntput(path->mnt); 14469393bd07SAl Viro path->mnt = mounted; 14479393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 14481da177e4SLinus Torvalds return 1; 14491da177e4SLinus Torvalds } 14501da177e4SLinus Torvalds return 0; 14511da177e4SLinus Torvalds } 14524d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14531da177e4SLinus Torvalds 14549875cf80SDavid Howells /* 14559deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14569deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14579deed3ebSAl Viro * caller is permitted to proceed or not. 14589deed3ebSAl Viro */ 14599deed3ebSAl Viro int follow_down(struct path *path) 14609deed3ebSAl Viro { 14619deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14629deed3ebSAl Viro bool jumped; 14639deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 14649deed3ebSAl Viro 14659deed3ebSAl Viro if (path->mnt != mnt) 14669deed3ebSAl Viro mntput(mnt); 14679deed3ebSAl Viro return ret; 14689deed3ebSAl Viro } 14699deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14709deed3ebSAl Viro 14719deed3ebSAl Viro /* 1472287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1473287548e4SAl Viro * we meet a managed dentry that would need blocking. 14749875cf80SDavid Howells */ 14759875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1476254cf582SAl Viro struct inode **inode, unsigned *seqp) 14779875cf80SDavid Howells { 1478ea936aebSAl Viro struct dentry *dentry = path->dentry; 1479ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1480ea936aebSAl Viro 1481ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1482ea936aebSAl Viro return true; 1483ea936aebSAl Viro 1484ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1485ea936aebSAl Viro return false; 1486ea936aebSAl Viro 148762a7375eSIan Kent for (;;) { 148862a7375eSIan Kent /* 148962a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 149062a7375eSIan Kent * that wants to block transit. 149162a7375eSIan Kent */ 1492ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1493ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1494ea936aebSAl Viro if (res) 1495ea936aebSAl Viro return res == -EISDIR; 1496ea936aebSAl Viro flags = dentry->d_flags; 1497b8faf035SNeilBrown } 149862a7375eSIan Kent 1499ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1500ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1501ea936aebSAl Viro if (mounted) { 1502c7105365SAl Viro path->mnt = &mounted->mnt; 1503ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1504bcba1e7dSAl Viro nd->state |= ND_JUMPED; 1505ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1506ea936aebSAl Viro *inode = dentry->d_inode; 150759430262SLinus Torvalds /* 1508ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1509ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1510ea936aebSAl Viro * between mount hash removal and ->mnt_root 1511ea936aebSAl Viro * becoming unpinned. 151259430262SLinus Torvalds */ 1513ea936aebSAl Viro flags = dentry->d_flags; 1514ea936aebSAl Viro continue; 15159875cf80SDavid Howells } 1516ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1517ea936aebSAl Viro return false; 1518ea936aebSAl Viro } 1519ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1520ea936aebSAl Viro } 1521287548e4SAl Viro } 1522287548e4SAl Viro 1523db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1524db3c9adeSAl Viro struct path *path, struct inode **inode, 1525db3c9adeSAl Viro unsigned int *seqp) 1526bd7c4b50SAl Viro { 15279deed3ebSAl Viro bool jumped; 1528db3c9adeSAl Viro int ret; 1529bd7c4b50SAl Viro 1530db3c9adeSAl Viro path->mnt = nd->path.mnt; 1531db3c9adeSAl Viro path->dentry = dentry; 1532c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1533c153007bSAl Viro unsigned int seq = *seqp; 1534c153007bSAl Viro if (unlikely(!*inode)) 1535c153007bSAl Viro return -ENOENT; 1536c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15379deed3ebSAl Viro return 0; 1538ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 1539c153007bSAl Viro return -ECHILD; 1540c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1541c153007bSAl Viro path->mnt = nd->path.mnt; 1542c153007bSAl Viro path->dentry = dentry; 1543c153007bSAl Viro } 15449deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 15459deed3ebSAl Viro if (jumped) { 15469deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 15479deed3ebSAl Viro ret = -EXDEV; 15489deed3ebSAl Viro else 1549bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15509deed3ebSAl Viro } 15519deed3ebSAl Viro if (unlikely(ret)) { 15529deed3ebSAl Viro dput(path->dentry); 15539deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15549deed3ebSAl Viro mntput(path->mnt); 15559deed3ebSAl Viro } else { 1556bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1557bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1558bd7c4b50SAl Viro } 1559bd7c4b50SAl Viro return ret; 1560bd7c4b50SAl Viro } 1561bd7c4b50SAl Viro 15629875cf80SDavid Howells /* 1563f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1564f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1565baa03890SNick Piggin */ 1566e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1567e3c13928SAl Viro struct dentry *dir, 15686c51e513SAl Viro unsigned int flags) 1569baa03890SNick Piggin { 1570a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1571bad61189SMiklos Szeredi if (dentry) { 1572a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1573bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 157474ff0ffcSAl Viro if (!error) 15755542aa2fSEric W. Biederman d_invalidate(dentry); 1576bad61189SMiklos Szeredi dput(dentry); 157774ff0ffcSAl Viro return ERR_PTR(error); 1578bad61189SMiklos Szeredi } 1579bad61189SMiklos Szeredi } 1580baa03890SNick Piggin return dentry; 1581baa03890SNick Piggin } 1582baa03890SNick Piggin 1583baa03890SNick Piggin /* 1584a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1585a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1586a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1587a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1588a03ece5fSAl Viro * at all. 158944396f4bSJosef Bacik */ 1590a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1591a03ece5fSAl Viro struct dentry *base, unsigned int flags) 159244396f4bSJosef Bacik { 1593a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 159444396f4bSJosef Bacik struct dentry *old; 1595a03ece5fSAl Viro struct inode *dir = base->d_inode; 1596a03ece5fSAl Viro 1597a03ece5fSAl Viro if (dentry) 1598a03ece5fSAl Viro return dentry; 159944396f4bSJosef Bacik 160044396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1601a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 160244396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1603a03ece5fSAl Viro 1604a03ece5fSAl Viro dentry = d_alloc(base, name); 1605a03ece5fSAl Viro if (unlikely(!dentry)) 1606a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 160744396f4bSJosef Bacik 160872bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 160944396f4bSJosef Bacik if (unlikely(old)) { 161044396f4bSJosef Bacik dput(dentry); 161144396f4bSJosef Bacik dentry = old; 161244396f4bSJosef Bacik } 161344396f4bSJosef Bacik return dentry; 161444396f4bSJosef Bacik } 161544396f4bSJosef Bacik 161620e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 161720e34357SAl Viro struct inode **inode, 1618254cf582SAl Viro unsigned *seqp) 16191da177e4SLinus Torvalds { 162031e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16215a18fff2SAl Viro int status = 1; 16229875cf80SDavid Howells 16233cac260aSAl Viro /* 1624b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16255d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16265d0f49c1SAl Viro * going to fall back to non-racy lookup. 1627b04f784eSNick Piggin */ 162831e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 162931e6b01fSNick Piggin unsigned seq; 1630da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 16315d0f49c1SAl Viro if (unlikely(!dentry)) { 1632e36cffedSJens Axboe if (!try_to_unlazy(nd)) 163320e34357SAl Viro return ERR_PTR(-ECHILD); 163420e34357SAl Viro return NULL; 16355d0f49c1SAl Viro } 16365a18fff2SAl Viro 163712f8ad4bSLinus Torvalds /* 163812f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 163912f8ad4bSLinus Torvalds * the dentry name information from lookup. 164012f8ad4bSLinus Torvalds */ 164163afdfc7SDavid Howells *inode = d_backing_inode(dentry); 16425d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 164320e34357SAl Viro return ERR_PTR(-ECHILD); 164412f8ad4bSLinus Torvalds 164512f8ad4bSLinus Torvalds /* 164612f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 164712f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 164812f8ad4bSLinus Torvalds * 164912f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 165012f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 165112f8ad4bSLinus Torvalds */ 16525d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 165320e34357SAl Viro return ERR_PTR(-ECHILD); 16545a18fff2SAl Viro 1655254cf582SAl Viro *seqp = seq; 16564ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1657c153007bSAl Viro if (likely(status > 0)) 165820e34357SAl Viro return dentry; 1659ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 166020e34357SAl Viro return ERR_PTR(-ECHILD); 166126ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1662209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1663209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16645a18fff2SAl Viro } else { 1665e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 166681e6f520SAl Viro if (unlikely(!dentry)) 166720e34357SAl Viro return NULL; 16684ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16695d0f49c1SAl Viro } 16705a18fff2SAl Viro if (unlikely(status <= 0)) { 1671e9742b53SAl Viro if (!status) 16725d0f49c1SAl Viro d_invalidate(dentry); 16735a18fff2SAl Viro dput(dentry); 167420e34357SAl Viro return ERR_PTR(status); 16755a18fff2SAl Viro } 167620e34357SAl Viro return dentry; 1677697f514dSMiklos Szeredi } 1678697f514dSMiklos Szeredi 1679697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 168088d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1681e3c13928SAl Viro struct dentry *dir, 1682e3c13928SAl Viro unsigned int flags) 1683697f514dSMiklos Szeredi { 168488d8331aSAl Viro struct dentry *dentry, *old; 16851936386eSAl Viro struct inode *inode = dir->d_inode; 1686d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16871936386eSAl Viro 16881936386eSAl Viro /* Don't go there if it's already dead */ 168994bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 169088d8331aSAl Viro return ERR_PTR(-ENOENT); 169194bdd655SAl Viro again: 1692d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 169394bdd655SAl Viro if (IS_ERR(dentry)) 169488d8331aSAl Viro return dentry; 169594bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1696949a852eSAl Viro int error = d_revalidate(dentry, flags); 1697949a852eSAl Viro if (unlikely(error <= 0)) { 169894bdd655SAl Viro if (!error) { 1699949a852eSAl Viro d_invalidate(dentry); 1700949a852eSAl Viro dput(dentry); 170194bdd655SAl Viro goto again; 170294bdd655SAl Viro } 170394bdd655SAl Viro dput(dentry); 1704949a852eSAl Viro dentry = ERR_PTR(error); 1705949a852eSAl Viro } 170694bdd655SAl Viro } else { 17071936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 170885c7f810SAl Viro d_lookup_done(dentry); 17091936386eSAl Viro if (unlikely(old)) { 17101936386eSAl Viro dput(dentry); 17111936386eSAl Viro dentry = old; 1712949a852eSAl Viro } 1713949a852eSAl Viro } 1714e3c13928SAl Viro return dentry; 17151da177e4SLinus Torvalds } 17161da177e4SLinus Torvalds 171788d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 171888d8331aSAl Viro struct dentry *dir, 171988d8331aSAl Viro unsigned int flags) 172088d8331aSAl Viro { 172188d8331aSAl Viro struct inode *inode = dir->d_inode; 172288d8331aSAl Viro struct dentry *res; 172388d8331aSAl Viro inode_lock_shared(inode); 172488d8331aSAl Viro res = __lookup_slow(name, dir, flags); 172588d8331aSAl Viro inode_unlock_shared(inode); 172688d8331aSAl Viro return res; 172788d8331aSAl Viro } 172888d8331aSAl Viro 1729ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1730ba73d987SChristian Brauner struct nameidata *nd) 173152094c8aSAl Viro { 173252094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17337d6beb71SLinus Torvalds int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1734e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 173552094c8aSAl Viro return err; 173652094c8aSAl Viro } 1737ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 173852094c8aSAl Viro } 173952094c8aSAl Viro 174049055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1741d63ff28fSAl Viro { 174249055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 174349055906SAl Viro return -ELOOP; 17444542576bSAl Viro 17454542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17464542576bSAl Viro return 0; 17474542576bSAl Viro if (likely(nd->stack != nd->internal)) 17484542576bSAl Viro return 0; 174960ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 175049055906SAl Viro return 0; 175160ef60c7SAl Viro 175260ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 175360ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 175460ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1755aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 175660ef60c7SAl Viro 1757e36cffedSJens Axboe if (!try_to_unlazy(nd) != 0 || !grabbed_link) 175860ef60c7SAl Viro return -ECHILD; 175960ef60c7SAl Viro 176060ef60c7SAl Viro if (nd_alloc_stack(nd)) 176160ef60c7SAl Viro return 0; 1762bc40aee0SAl Viro } 176360ef60c7SAl Viro return -ENOMEM; 176449055906SAl Viro } 176549055906SAl Viro 176649055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 176749055906SAl Viro 176849055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 176949055906SAl Viro struct inode *inode, unsigned seq, int flags) 177049055906SAl Viro { 177149055906SAl Viro struct saved *last; 177249055906SAl Viro const char *res; 177349055906SAl Viro int error = reserve_stack(nd, link, seq); 177449055906SAl Viro 177549055906SAl Viro if (unlikely(error)) { 177649055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1777cd179f44SAl Viro path_put(link); 177806708adbSAl Viro return ERR_PTR(error); 1779626de996SAl Viro } 1780ab104923SAl Viro last = nd->stack + nd->depth++; 17811cf2665bSAl Viro last->link = *link; 1782fceef393SAl Viro clear_delayed_call(&last->done); 17830450b2d1SAl Viro last->seq = seq; 1784ad6cc4c3SAl Viro 1785b1a81972SAl Viro if (flags & WALK_TRAILING) { 1786ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1787ad6cc4c3SAl Viro if (unlikely(error)) 1788ad6cc4c3SAl Viro return ERR_PTR(error); 1789ad6cc4c3SAl Viro } 1790ad6cc4c3SAl Viro 1791dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1792dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1793ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1794ad6cc4c3SAl Viro 1795ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1796ad6cc4c3SAl Viro touch_atime(&last->link); 1797ad6cc4c3SAl Viro cond_resched(); 1798ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1799e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1800ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1801ad6cc4c3SAl Viro touch_atime(&last->link); 1802ad6cc4c3SAl Viro } 1803ad6cc4c3SAl Viro 1804ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1805ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1806ad6cc4c3SAl Viro if (unlikely(error)) 1807ad6cc4c3SAl Viro return ERR_PTR(error); 1808ad6cc4c3SAl Viro 1809ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1810ad6cc4c3SAl Viro if (!res) { 1811ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1812ad6cc4c3SAl Viro struct delayed_call *); 1813ad6cc4c3SAl Viro get = inode->i_op->get_link; 1814ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1815ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1816e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1817ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1818ad6cc4c3SAl Viro } else { 1819ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1820ad6cc4c3SAl Viro } 1821ad6cc4c3SAl Viro if (!res) 1822ad6cc4c3SAl Viro goto all_done; 1823ad6cc4c3SAl Viro if (IS_ERR(res)) 1824ad6cc4c3SAl Viro return res; 1825ad6cc4c3SAl Viro } 1826ad6cc4c3SAl Viro if (*res == '/') { 1827ad6cc4c3SAl Viro error = nd_jump_root(nd); 1828ad6cc4c3SAl Viro if (unlikely(error)) 1829ad6cc4c3SAl Viro return ERR_PTR(error); 1830ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1831ad6cc4c3SAl Viro ; 1832ad6cc4c3SAl Viro } 1833ad6cc4c3SAl Viro if (*res) 1834ad6cc4c3SAl Viro return res; 1835ad6cc4c3SAl Viro all_done: // pure jump 1836ad6cc4c3SAl Viro put_link(nd); 1837ad6cc4c3SAl Viro return NULL; 1838d63ff28fSAl Viro } 1839d63ff28fSAl Viro 18403ddcd056SLinus Torvalds /* 18413ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18423ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18433ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18443ddcd056SLinus Torvalds * for the common case. 18453ddcd056SLinus Torvalds */ 1846b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1847cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 18483ddcd056SLinus Torvalds { 1849cbae4d12SAl Viro struct path path; 1850cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1851cbae4d12SAl Viro 1852cbae4d12SAl Viro if (err < 0) 1853b0417d2cSAl Viro return ERR_PTR(err); 1854cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18558c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1856aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18578f64fb1cSAl Viro /* not a symlink or should not follow */ 1858c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1859c99687a0SAl Viro dput(nd->path.dentry); 1860c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1861c99687a0SAl Viro mntput(nd->path.mnt); 1862c99687a0SAl Viro } 1863c99687a0SAl Viro nd->path = path; 18648f64fb1cSAl Viro nd->inode = inode; 18658f64fb1cSAl Viro nd->seq = seq; 1866b0417d2cSAl Viro return NULL; 18678f64fb1cSAl Viro } 1868a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 186984f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1870cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1871b0417d2cSAl Viro return ERR_PTR(-ECHILD); 187284f0cd9eSAl Viro } else { 187384f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 187484f0cd9eSAl Viro mntget(path.mnt); 1875a7f77542SAl Viro } 1876b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18773ddcd056SLinus Torvalds } 18783ddcd056SLinus Torvalds 1879c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1880c2df1968SAl Viro struct inode **inodep, 1881c2df1968SAl Viro unsigned *seqp) 1882957dd41dSAl Viro { 188312487f30SAl Viro struct dentry *parent, *old; 1884957dd41dSAl Viro 188512487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 188612487f30SAl Viro goto in_root; 188712487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18887ef482faSAl Viro struct path path; 1889efe772d6SAl Viro unsigned seq; 18907ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18917ef482faSAl Viro &nd->root, &path, &seq)) 189212487f30SAl Viro goto in_root; 1893efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1894efe772d6SAl Viro return ERR_PTR(-ECHILD); 1895efe772d6SAl Viro nd->path = path; 1896efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1897efe772d6SAl Viro nd->seq = seq; 1898efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1899efe772d6SAl Viro return ERR_PTR(-ECHILD); 1900efe772d6SAl Viro /* we know that mountpoint was pinned */ 1901957dd41dSAl Viro } 190212487f30SAl Viro old = nd->path.dentry; 190312487f30SAl Viro parent = old->d_parent; 190412487f30SAl Viro *inodep = parent->d_inode; 190512487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 190612487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 190712487f30SAl Viro return ERR_PTR(-ECHILD); 190812487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 190912487f30SAl Viro return ERR_PTR(-ECHILD); 191012487f30SAl Viro return parent; 191112487f30SAl Viro in_root: 1912efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1913efe772d6SAl Viro return ERR_PTR(-ECHILD); 1914957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19157521f22bSAl Viro return ERR_PTR(-ECHILD); 1916c2df1968SAl Viro return NULL; 1917957dd41dSAl Viro } 1918957dd41dSAl Viro 1919c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1920c2df1968SAl Viro struct inode **inodep, 1921c2df1968SAl Viro unsigned *seqp) 1922957dd41dSAl Viro { 192312487f30SAl Viro struct dentry *parent; 192412487f30SAl Viro 1925957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 192612487f30SAl Viro goto in_root; 192712487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19282aa38470SAl Viro struct path path; 19292aa38470SAl Viro 19302aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19312aa38470SAl Viro &nd->root, &path)) 193212487f30SAl Viro goto in_root; 1933165200d6SAl Viro path_put(&nd->path); 1934165200d6SAl Viro nd->path = path; 19352aa38470SAl Viro nd->inode = path.dentry->d_inode; 1936165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1937165200d6SAl Viro return ERR_PTR(-EXDEV); 193812487f30SAl Viro } 1939957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 194012487f30SAl Viro parent = dget_parent(nd->path.dentry); 1941957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1942957dd41dSAl Viro dput(parent); 19437521f22bSAl Viro return ERR_PTR(-ENOENT); 1944957dd41dSAl Viro } 1945c2df1968SAl Viro *seqp = 0; 1946c2df1968SAl Viro *inodep = parent->d_inode; 1947c2df1968SAl Viro return parent; 194812487f30SAl Viro 194912487f30SAl Viro in_root: 1950957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19517521f22bSAl Viro return ERR_PTR(-EXDEV); 1952c2df1968SAl Viro dget(nd->path.dentry); 1953c2df1968SAl Viro return NULL; 1954957dd41dSAl Viro } 1955957dd41dSAl Viro 19567521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1957957dd41dSAl Viro { 1958957dd41dSAl Viro if (type == LAST_DOTDOT) { 19597521f22bSAl Viro const char *error = NULL; 1960c2df1968SAl Viro struct dentry *parent; 1961c2df1968SAl Viro struct inode *inode; 1962c2df1968SAl Viro unsigned seq; 1963957dd41dSAl Viro 1964957dd41dSAl Viro if (!nd->root.mnt) { 19657521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1966957dd41dSAl Viro if (error) 1967957dd41dSAl Viro return error; 1968957dd41dSAl Viro } 1969957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1970c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1971957dd41dSAl Viro else 1972c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1973c2df1968SAl Viro if (IS_ERR(parent)) 1974c2df1968SAl Viro return ERR_CAST(parent); 1975c2df1968SAl Viro if (unlikely(!parent)) 1976c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1977c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1978c2df1968SAl Viro else 1979c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1980c2df1968SAl Viro parent, inode, seq); 1981c2df1968SAl Viro if (unlikely(error)) 1982957dd41dSAl Viro return error; 1983957dd41dSAl Viro 1984957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1985957dd41dSAl Viro /* 1986957dd41dSAl Viro * If there was a racing rename or mount along our 1987957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1988957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1989957dd41dSAl Viro * some fallback). 1990957dd41dSAl Viro */ 1991957dd41dSAl Viro smp_rmb(); 1992957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 19937521f22bSAl Viro return ERR_PTR(-EAGAIN); 1994957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 19957521f22bSAl Viro return ERR_PTR(-EAGAIN); 1996957dd41dSAl Viro } 1997957dd41dSAl Viro } 19987521f22bSAl Viro return NULL; 1999957dd41dSAl Viro } 2000957dd41dSAl Viro 200192d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 2002ce57dfc1SAl Viro { 2003db3c9adeSAl Viro struct dentry *dentry; 2004ce57dfc1SAl Viro struct inode *inode; 2005254cf582SAl Viro unsigned seq; 2006ce57dfc1SAl Viro /* 2007ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 2008ce57dfc1SAl Viro * to be able to know about the current root directory and 2009ce57dfc1SAl Viro * parent relationships. 2010ce57dfc1SAl Viro */ 20114693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 20121c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 20134693a547SAl Viro put_link(nd); 20147521f22bSAl Viro return handle_dots(nd, nd->last_type); 20154693a547SAl Viro } 201620e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 201720e34357SAl Viro if (IS_ERR(dentry)) 201892d27016SAl Viro return ERR_CAST(dentry); 201920e34357SAl Viro if (unlikely(!dentry)) { 2020db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 2021db3c9adeSAl Viro if (IS_ERR(dentry)) 202292d27016SAl Viro return ERR_CAST(dentry); 202320e34357SAl Viro } 202456676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 202556676ec3SAl Viro put_link(nd); 2026b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 2027ce57dfc1SAl Viro } 2028ce57dfc1SAl Viro 20291da177e4SLinus Torvalds /* 2030bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2031bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2032bfcfaa77SLinus Torvalds * 2033bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2034bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2035bfcfaa77SLinus Torvalds * fast. 2036bfcfaa77SLinus Torvalds * 2037bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2038bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2039bfcfaa77SLinus Torvalds * crossing operation. 2040bfcfaa77SLinus Torvalds * 2041bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2042bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2043bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2044bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2045bfcfaa77SLinus Torvalds */ 2046bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2047bfcfaa77SLinus Torvalds 2048f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2049bfcfaa77SLinus Torvalds 2050468a9428SGeorge Spelvin #ifdef HASH_MIX 2051bfcfaa77SLinus Torvalds 2052468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2053468a9428SGeorge Spelvin 2054468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20552a18da7aSGeorge Spelvin /* 20562a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20572a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20582a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20592a18da7aSGeorge Spelvin * and no temporaries. 20602a18da7aSGeorge Spelvin * 20612a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20622a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20632a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20642a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20652a18da7aSGeorge Spelvin * 20662a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20672a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20682a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20692a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20702a18da7aSGeorge Spelvin * 20712a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20722a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20732a18da7aSGeorge Spelvin * toggling any given output bit. 20742a18da7aSGeorge Spelvin * 20752a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20762a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20772a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20782a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20792a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20802a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20812a18da7aSGeorge Spelvin * Perfect: 8192 258048 20822a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20832a18da7aSGeorge Spelvin */ 20842a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20852a18da7aSGeorge Spelvin ( x ^= (a), \ 20862a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20872a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20882a18da7aSGeorge Spelvin y *= 9 ) 2089bfcfaa77SLinus Torvalds 20900fed3ac8SGeorge Spelvin /* 20912a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20922a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20932a18da7aSGeorge Spelvin * work done before the hash value is used. 20940fed3ac8SGeorge Spelvin */ 20952a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20960fed3ac8SGeorge Spelvin { 20972a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20982a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20992a18da7aSGeorge Spelvin return y >> 32; 21000fed3ac8SGeorge Spelvin } 21010fed3ac8SGeorge Spelvin 2102bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2103bfcfaa77SLinus Torvalds 21042a18da7aSGeorge Spelvin /* 21052a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 21062a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 21072a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 21082a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 21092a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 21102a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 21112a18da7aSGeorge Spelvin * Perfect: 2048 31744 21122a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 21132a18da7aSGeorge Spelvin */ 21142a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 21152a18da7aSGeorge Spelvin ( x ^= (a), \ 21162a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 21172a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 21182a18da7aSGeorge Spelvin y *= 9 ) 2119bfcfaa77SLinus Torvalds 21202a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 21210fed3ac8SGeorge Spelvin { 21222a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 21232a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 21240fed3ac8SGeorge Spelvin } 21250fed3ac8SGeorge Spelvin 2126bfcfaa77SLinus Torvalds #endif 2127bfcfaa77SLinus Torvalds 21282a18da7aSGeorge Spelvin /* 21292a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21302a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21312a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21322a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21332a18da7aSGeorge Spelvin * finds the delimiter after the name. 21342a18da7aSGeorge Spelvin */ 21358387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2136bfcfaa77SLinus Torvalds { 21378387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2138bfcfaa77SLinus Torvalds 2139bfcfaa77SLinus Torvalds for (;;) { 2140fcfd2fbfSGeorge Spelvin if (!len) 2141fcfd2fbfSGeorge Spelvin goto done; 2142e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2143bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2144bfcfaa77SLinus Torvalds break; 21452a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2146bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2147bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2148bfcfaa77SLinus Torvalds } 21492a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2150bfcfaa77SLinus Torvalds done: 21512a18da7aSGeorge Spelvin return fold_hash(x, y); 2152bfcfaa77SLinus Torvalds } 2153bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2154bfcfaa77SLinus Torvalds 2155fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21568387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2157fcfd2fbfSGeorge Spelvin { 21588387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21598387ff25SLinus Torvalds unsigned long adata, mask, len; 2160fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2161fcfd2fbfSGeorge Spelvin 21628387ff25SLinus Torvalds len = 0; 21638387ff25SLinus Torvalds goto inside; 21648387ff25SLinus Torvalds 2165fcfd2fbfSGeorge Spelvin do { 21662a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2167fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21688387ff25SLinus Torvalds inside: 2169fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2170fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2171fcfd2fbfSGeorge Spelvin 2172fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2173fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21742a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2175fcfd2fbfSGeorge Spelvin 21762a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2177fcfd2fbfSGeorge Spelvin } 2178fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2179fcfd2fbfSGeorge Spelvin 2180bfcfaa77SLinus Torvalds /* 2181bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2182d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2183bfcfaa77SLinus Torvalds */ 21848387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2185bfcfaa77SLinus Torvalds { 21868387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21878387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 218836126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2189bfcfaa77SLinus Torvalds 21908387ff25SLinus Torvalds len = 0; 21918387ff25SLinus Torvalds goto inside; 21928387ff25SLinus Torvalds 2193bfcfaa77SLinus Torvalds do { 21942a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2195bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21968387ff25SLinus Torvalds inside: 2197e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 219836126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 219936126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2200bfcfaa77SLinus Torvalds 220136126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 220236126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 220336126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 22042a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 220536126f8fSLinus Torvalds 22062a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2207bfcfaa77SLinus Torvalds } 2208bfcfaa77SLinus Torvalds 22092a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2210bfcfaa77SLinus Torvalds 2211fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 22128387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 22130145acc2SLinus Torvalds { 22148387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 22150145acc2SLinus Torvalds while (len--) 2216fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 22170145acc2SLinus Torvalds return end_name_hash(hash); 22180145acc2SLinus Torvalds } 2219ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 22200145acc2SLinus Torvalds 2221fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 22228387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2223fcfd2fbfSGeorge Spelvin { 22248387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2225fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2226fcfd2fbfSGeorge Spelvin 2227fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2228e0ab7af9SGeorge Spelvin while (c) { 2229fcfd2fbfSGeorge Spelvin len++; 2230fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2231fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2232e0ab7af9SGeorge Spelvin } 2233fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2234fcfd2fbfSGeorge Spelvin } 2235f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2236fcfd2fbfSGeorge Spelvin 22373ddcd056SLinus Torvalds /* 2238200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2239200e9ef7SLinus Torvalds * one character. 2240200e9ef7SLinus Torvalds */ 22418387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2242200e9ef7SLinus Torvalds { 22438387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2244200e9ef7SLinus Torvalds unsigned long len = 0, c; 2245200e9ef7SLinus Torvalds 2246200e9ef7SLinus Torvalds c = (unsigned char)*name; 2247200e9ef7SLinus Torvalds do { 2248200e9ef7SLinus Torvalds len++; 2249200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2250200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2251200e9ef7SLinus Torvalds } while (c && c != '/'); 2252d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2253200e9ef7SLinus Torvalds } 2254200e9ef7SLinus Torvalds 2255bfcfaa77SLinus Torvalds #endif 2256bfcfaa77SLinus Torvalds 2257200e9ef7SLinus Torvalds /* 22581da177e4SLinus Torvalds * Name resolution. 2259ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2260ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22611da177e4SLinus Torvalds * 2262ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2263ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22641da177e4SLinus Torvalds */ 22656de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22661da177e4SLinus Torvalds { 2267d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22681da177e4SLinus Torvalds int err; 22691da177e4SLinus Torvalds 2270b4c03536SAl Viro nd->last_type = LAST_ROOT; 2271c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22729b5858e9SAl Viro if (IS_ERR(name)) 22739b5858e9SAl Viro return PTR_ERR(name); 22741da177e4SLinus Torvalds while (*name=='/') 22751da177e4SLinus Torvalds name++; 22761a97d899SAl Viro if (!*name) { 22771a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22789e18f10aSAl Viro return 0; 22791a97d899SAl Viro } 22801da177e4SLinus Torvalds 22811da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22821da177e4SLinus Torvalds for(;;) { 2283549c7297SChristian Brauner struct user_namespace *mnt_userns; 228492d27016SAl Viro const char *link; 2285d6bb3e90SLinus Torvalds u64 hash_len; 2286fe479a58SAl Viro int type; 22871da177e4SLinus Torvalds 2288549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2289549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22901da177e4SLinus Torvalds if (err) 22913595e234SAl Viro return err; 22921da177e4SLinus Torvalds 22938387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22941da177e4SLinus Torvalds 2295fe479a58SAl Viro type = LAST_NORM; 2296d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2297fe479a58SAl Viro case 2: 2298200e9ef7SLinus Torvalds if (name[1] == '.') { 2299fe479a58SAl Viro type = LAST_DOTDOT; 2300bcba1e7dSAl Viro nd->state |= ND_JUMPED; 230116c2cd71SAl Viro } 2302fe479a58SAl Viro break; 2303fe479a58SAl Viro case 1: 2304fe479a58SAl Viro type = LAST_DOT; 2305fe479a58SAl Viro } 23065a202bcdSAl Viro if (likely(type == LAST_NORM)) { 23075a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2308bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 23095a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2310a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2311da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 23125a202bcdSAl Viro if (err < 0) 23133595e234SAl Viro return err; 2314d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2315d6bb3e90SLinus Torvalds name = this.name; 23165a202bcdSAl Viro } 23175a202bcdSAl Viro } 2318fe479a58SAl Viro 2319d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2320d6bb3e90SLinus Torvalds nd->last.name = name; 23215f4a6a69SAl Viro nd->last_type = type; 23225f4a6a69SAl Viro 2323d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2324d6bb3e90SLinus Torvalds if (!*name) 2325bdf6cbf1SAl Viro goto OK; 2326200e9ef7SLinus Torvalds /* 2327200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2328200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2329200e9ef7SLinus Torvalds */ 2330200e9ef7SLinus Torvalds do { 2331d6bb3e90SLinus Torvalds name++; 2332d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 23338620c238SAl Viro if (unlikely(!*name)) { 23348620c238SAl Viro OK: 2335d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2336c108837eSAl Viro if (!depth) { 2337549c7297SChristian Brauner nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); 23380f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2339c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 23408620c238SAl Viro return 0; 2341c108837eSAl Viro } 23428620c238SAl Viro /* last component of nested symlink */ 2343d8d4611aSAl Viro name = nd->stack[--depth].name; 23448c4efe22SAl Viro link = walk_component(nd, 0); 23451c4ff1a8SAl Viro } else { 23461c4ff1a8SAl Viro /* not the last component */ 23478c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 23488620c238SAl Viro } 234992d27016SAl Viro if (unlikely(link)) { 235092d27016SAl Viro if (IS_ERR(link)) 235192d27016SAl Viro return PTR_ERR(link); 235292d27016SAl Viro /* a symlink to follow */ 2353d8d4611aSAl Viro nd->stack[depth++].name = name; 235492d27016SAl Viro name = link; 23559e18f10aSAl Viro continue; 235648c8b0c5SAl Viro } 235797242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 235897242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2359e36cffedSJens Axboe if (!try_to_unlazy(nd)) 236097242f99SAl Viro return -ECHILD; 236197242f99SAl Viro } 23623595e234SAl Viro return -ENOTDIR; 23635f4a6a69SAl Viro } 2364ce57dfc1SAl Viro } 236597242f99SAl Viro } 23661da177e4SLinus Torvalds 2367edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2368c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 236931e6b01fSNick Piggin { 2370740a1678SAleksa Sarai int error; 2371c8a53ee5SAl Viro const char *s = nd->name->name; 237231e6b01fSNick Piggin 23736c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23746c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23756c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23766c6ec2b0SJens Axboe 2377c0eb027eSLinus Torvalds if (!*s) 2378c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2379edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2380edc2b1daSAl Viro rcu_read_lock(); 2381c0eb027eSLinus Torvalds 2382bcba1e7dSAl Viro nd->flags = flags; 2383bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2384ab87f9a5SAleksa Sarai 2385ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2386ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2387ab87f9a5SAleksa Sarai smp_rmb(); 2388ab87f9a5SAleksa Sarai 2389bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2390b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2391b18825a7SDavid Howells struct inode *inode = root->d_inode; 239293893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2393368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23945b6ca027SAl Viro nd->path = nd->root; 23955b6ca027SAl Viro nd->inode = inode; 23965b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2397ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23988f47a016SAl Viro nd->root_seq = nd->seq; 23995b6ca027SAl Viro } else { 24005b6ca027SAl Viro path_get(&nd->path); 24015b6ca027SAl Viro } 2402368ee9baSAl Viro return s; 24035b6ca027SAl Viro } 24045b6ca027SAl Viro 240531e6b01fSNick Piggin nd->root.mnt = NULL; 240631e6b01fSNick Piggin 24078db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 24088db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2409740a1678SAleksa Sarai error = nd_jump_root(nd); 2410740a1678SAleksa Sarai if (unlikely(error)) 2411740a1678SAleksa Sarai return ERR_PTR(error); 2412ef55d917SAl Viro return s; 24138db52c7eSAleksa Sarai } 24148db52c7eSAleksa Sarai 24158db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 24168db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2417e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 241831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2419c28cc364SNick Piggin unsigned seq; 242031e6b01fSNick Piggin 2421c28cc364SNick Piggin do { 2422c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 242331e6b01fSNick Piggin nd->path = fs->pwd; 2424ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2425c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2426c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2427e41f7d4eSAl Viro } else { 2428e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2429ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2430e41f7d4eSAl Viro } 243131e6b01fSNick Piggin } else { 2432582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2433c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 243431e6b01fSNick Piggin struct dentry *dentry; 243531e6b01fSNick Piggin 24362903ff01SAl Viro if (!f.file) 2437368ee9baSAl Viro return ERR_PTR(-EBADF); 243831e6b01fSNick Piggin 24392903ff01SAl Viro dentry = f.file->f_path.dentry; 244031e6b01fSNick Piggin 2441edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 24422903ff01SAl Viro fdput(f); 2443368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2444f52e0c11SAl Viro } 24452903ff01SAl Viro 24462903ff01SAl Viro nd->path = f.file->f_path; 2447e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 244834a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 244934a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24505590ff0dSUlrich Drepper } else { 24512903ff01SAl Viro path_get(&nd->path); 245234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24531da177e4SLinus Torvalds } 245434a26b99SAl Viro fdput(f); 2455e41f7d4eSAl Viro } 24568db52c7eSAleksa Sarai 2457adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2458adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2459adb21d2bSAleksa Sarai nd->root = nd->path; 2460adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2461adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2462adb21d2bSAleksa Sarai } else { 2463adb21d2bSAleksa Sarai path_get(&nd->root); 2464bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2465adb21d2bSAleksa Sarai } 2466adb21d2bSAleksa Sarai } 2467adb21d2bSAleksa Sarai return s; 24689b4a9b14SAl Viro } 24699b4a9b14SAl Viro 24701ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 247195fa25d9SAl Viro { 2472bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2473bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2474bd92d7feSAl Viro 2475c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2476bd92d7feSAl Viro } 2477bd92d7feSAl Viro 24784f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24794f757f3cSAl Viro { 2480c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2481db3c9adeSAl Viro dget(nd->path.dentry); 2482b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2483b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 24844f757f3cSAl Viro } 24854f757f3cSAl Viro 24869b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2487c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24889b4a9b14SAl Viro { 2489c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2490bd92d7feSAl Viro int err; 249131e6b01fSNick Piggin 24929b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24934f757f3cSAl Viro err = handle_lookup_down(nd); 24945f336e72SAl Viro if (unlikely(err < 0)) 24955f336e72SAl Viro s = ERR_PTR(err); 24964f757f3cSAl Viro } 24974f757f3cSAl Viro 24981ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24991ccac622SAl Viro (s = lookup_last(nd)) != NULL) 25001ccac622SAl Viro ; 25014f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 25024f0ed93fSAl Viro err = handle_lookup_down(nd); 2503bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 25044f0ed93fSAl Viro } 25059f1fafeeSAl Viro if (!err) 25069f1fafeeSAl Viro err = complete_walk(nd); 2507bd92d7feSAl Viro 2508deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2509deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2510bd23a539SAl Viro err = -ENOTDIR; 2511625b6d10SAl Viro if (!err) { 2512625b6d10SAl Viro *path = nd->path; 2513625b6d10SAl Viro nd->path.mnt = NULL; 2514625b6d10SAl Viro nd->path.dentry = NULL; 2515625b6d10SAl Viro } 2516deb106c6SAl Viro terminate_walk(nd); 2517bd92d7feSAl Viro return err; 251831e6b01fSNick Piggin } 251931e6b01fSNick Piggin 2520794ebceaSStephen Brennan int filename_lookup(int dfd, struct filename *name, unsigned flags, 25219ad1aaa6SAl Viro struct path *path, struct path *root) 2522873f1eedSJeff Layton { 2523894bc8c4SAl Viro int retval; 25249883d185SAl Viro struct nameidata nd; 2525abc9f5beSAl Viro if (IS_ERR(name)) 2526abc9f5beSAl Viro return PTR_ERR(name); 252706422964SAl Viro set_nameidata(&nd, dfd, name, root); 2528c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2529873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2530c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2531873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2532c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2533873f1eedSJeff Layton 2534873f1eedSJeff Layton if (likely(!retval)) 2535161aff1dSAl Viro audit_inode(name, path->dentry, 2536161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 25379883d185SAl Viro restore_nameidata(); 2538020250f3SDmitry Kadashev return retval; 2539020250f3SDmitry Kadashev } 2540020250f3SDmitry Kadashev 25418bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2542c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2543391172c4SAl Viro struct path *parent) 25448bcb77faSAl Viro { 2545c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25469b5858e9SAl Viro int err = link_path_walk(s, nd); 25478bcb77faSAl Viro if (!err) 25488bcb77faSAl Viro err = complete_walk(nd); 2549391172c4SAl Viro if (!err) { 2550391172c4SAl Viro *parent = nd->path; 2551391172c4SAl Viro nd->path.mnt = NULL; 2552391172c4SAl Viro nd->path.dentry = NULL; 2553391172c4SAl Viro } 2554deb106c6SAl Viro terminate_walk(nd); 25558bcb77faSAl Viro return err; 25568bcb77faSAl Viro } 25578bcb77faSAl Viro 25580766ec82SStephen Brennan /* Note: this does not consume "name" */ 2559c5f563f9SAl Viro static int filename_parentat(int dfd, struct filename *name, 2560391172c4SAl Viro unsigned int flags, struct path *parent, 2561391172c4SAl Viro struct qstr *last, int *type) 25628bcb77faSAl Viro { 25638bcb77faSAl Viro int retval; 25649883d185SAl Viro struct nameidata nd; 25658bcb77faSAl Viro 25665c31b6ceSAl Viro if (IS_ERR(name)) 25670ee50b47SDmitry Kadashev return PTR_ERR(name); 256806422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2569c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25708bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2571c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25728bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2573c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2574391172c4SAl Viro if (likely(!retval)) { 2575391172c4SAl Viro *last = nd.last; 2576391172c4SAl Viro *type = nd.last_type; 2577c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2578391172c4SAl Viro } 25799883d185SAl Viro restore_nameidata(); 25800ee50b47SDmitry Kadashev return retval; 25810ee50b47SDmitry Kadashev } 25820ee50b47SDmitry Kadashev 258379714f72SAl Viro /* does lookup, returns the object with parent locked */ 25840766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25855590ff0dSUlrich Drepper { 25865c31b6ceSAl Viro struct dentry *d; 2587391172c4SAl Viro struct qstr last; 25880ee50b47SDmitry Kadashev int type, error; 258951689104SPaul Moore 2590c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25910ee50b47SDmitry Kadashev if (error) 25920ee50b47SDmitry Kadashev return ERR_PTR(error); 25935c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2594391172c4SAl Viro path_put(path); 25955c31b6ceSAl Viro return ERR_PTR(-EINVAL); 259679714f72SAl Viro } 25975955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2598391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 259979714f72SAl Viro if (IS_ERR(d)) { 26005955102cSAl Viro inode_unlock(path->dentry->d_inode); 2601391172c4SAl Viro path_put(path); 260279714f72SAl Viro } 260379714f72SAl Viro return d; 26045590ff0dSUlrich Drepper } 26055590ff0dSUlrich Drepper 26060766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 26070766ec82SStephen Brennan { 26080766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 26090766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 26100766ec82SStephen Brennan 26110766ec82SStephen Brennan putname(filename); 26120766ec82SStephen Brennan return res; 26130766ec82SStephen Brennan } 26140766ec82SStephen Brennan 2615d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2616d1811465SAl Viro { 2617794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2618794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2619794ebceaSStephen Brennan 2620794ebceaSStephen Brennan putname(filename); 2621794ebceaSStephen Brennan return ret; 2622794ebceaSStephen Brennan 2623d1811465SAl Viro } 26244d359507SAl Viro EXPORT_SYMBOL(kern_path); 2625d1811465SAl Viro 262616f18200SJosef 'Jeff' Sipek /** 262716f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 262816f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 262916f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 263016f18200SJosef 'Jeff' Sipek * @name: pointer to file name 263116f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2632e0a01249SAl Viro * @path: pointer to struct path to fill 263316f18200SJosef 'Jeff' Sipek */ 263416f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 263516f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2636e0a01249SAl Viro struct path *path) 263716f18200SJosef 'Jeff' Sipek { 2638794ebceaSStephen Brennan struct filename *filename; 26399ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2640794ebceaSStephen Brennan int ret; 2641794ebceaSStephen Brennan 2642794ebceaSStephen Brennan filename = getname_kernel(name); 26439ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2644794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2645794ebceaSStephen Brennan putname(filename); 2646794ebceaSStephen Brennan return ret; 264716f18200SJosef 'Jeff' Sipek } 26484d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 264916f18200SJosef 'Jeff' Sipek 2650c2fd68b6SChristian Brauner static int lookup_one_common(struct user_namespace *mnt_userns, 2651c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2652c2fd68b6SChristian Brauner struct qstr *this) 26533c95f0dcSAl Viro { 26543c95f0dcSAl Viro this->name = name; 26553c95f0dcSAl Viro this->len = len; 26563c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26573c95f0dcSAl Viro if (!len) 26583c95f0dcSAl Viro return -EACCES; 26593c95f0dcSAl Viro 26603c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26613c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26623c95f0dcSAl Viro return -EACCES; 26633c95f0dcSAl Viro } 26643c95f0dcSAl Viro 26653c95f0dcSAl Viro while (len--) { 26663c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26673c95f0dcSAl Viro if (c == '/' || c == '\0') 26683c95f0dcSAl Viro return -EACCES; 26693c95f0dcSAl Viro } 26703c95f0dcSAl Viro /* 26713c95f0dcSAl Viro * See if the low-level filesystem might want 26723c95f0dcSAl Viro * to use its own hash.. 26733c95f0dcSAl Viro */ 26743c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26753c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26763c95f0dcSAl Viro if (err < 0) 26773c95f0dcSAl Viro return err; 26783c95f0dcSAl Viro } 26793c95f0dcSAl Viro 2680c2fd68b6SChristian Brauner return inode_permission(mnt_userns, base->d_inode, MAY_EXEC); 26813c95f0dcSAl Viro } 26823c95f0dcSAl Viro 2683eead1911SChristoph Hellwig /** 26840da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26850da0b7fdSDavid Howells * @name: pathname component to lookup 26860da0b7fdSDavid Howells * @base: base directory to lookup from 26870da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26880da0b7fdSDavid Howells * 26890da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26900da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26910da0b7fdSDavid Howells * 26920da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26930da0b7fdSDavid Howells * not be called by generic code. 26940da0b7fdSDavid Howells * 26950da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26960da0b7fdSDavid Howells */ 26970da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26980da0b7fdSDavid Howells { 26990da0b7fdSDavid Howells struct qstr this; 27000da0b7fdSDavid Howells int err; 27010da0b7fdSDavid Howells 27020da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27030da0b7fdSDavid Howells 2704c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 27050da0b7fdSDavid Howells if (err) 27060da0b7fdSDavid Howells return ERR_PTR(err); 27070da0b7fdSDavid Howells 27080da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 27090da0b7fdSDavid Howells } 27100da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 27110da0b7fdSDavid Howells 27120da0b7fdSDavid Howells /** 2713a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2714eead1911SChristoph Hellwig * @name: pathname component to lookup 2715eead1911SChristoph Hellwig * @base: base directory to lookup from 2716eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2717eead1911SChristoph Hellwig * 2718a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 27199e7543e9SAl Viro * not be called by generic code. 2720bbddca8eSNeilBrown * 2721bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2722eead1911SChristoph Hellwig */ 2723057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2724057f6c01SJames Morris { 27258613a209SAl Viro struct dentry *dentry; 2726057f6c01SJames Morris struct qstr this; 2727cda309deSMiklos Szeredi int err; 2728057f6c01SJames Morris 27295955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27302f9092e1SDavid Woodhouse 2731c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2732cda309deSMiklos Szeredi if (err) 2733cda309deSMiklos Szeredi return ERR_PTR(err); 2734cda309deSMiklos Szeredi 27358613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27368613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2737057f6c01SJames Morris } 27384d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2739057f6c01SJames Morris 2740bbddca8eSNeilBrown /** 2741c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 2742c2fd68b6SChristian Brauner * @mnt_userns: user namespace of the mount the lookup is performed from 2743c2fd68b6SChristian Brauner * @name: pathname component to lookup 2744c2fd68b6SChristian Brauner * @base: base directory to lookup from 2745c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2746c2fd68b6SChristian Brauner * 2747c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2748c2fd68b6SChristian Brauner * not be called by generic code. 2749c2fd68b6SChristian Brauner * 2750c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2751c2fd68b6SChristian Brauner */ 2752c2fd68b6SChristian Brauner struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name, 2753c2fd68b6SChristian Brauner struct dentry *base, int len) 2754c2fd68b6SChristian Brauner { 2755c2fd68b6SChristian Brauner struct dentry *dentry; 2756c2fd68b6SChristian Brauner struct qstr this; 2757c2fd68b6SChristian Brauner int err; 2758c2fd68b6SChristian Brauner 2759c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2760c2fd68b6SChristian Brauner 2761c2fd68b6SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 2762c2fd68b6SChristian Brauner if (err) 2763c2fd68b6SChristian Brauner return ERR_PTR(err); 2764c2fd68b6SChristian Brauner 2765c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2766c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2767c2fd68b6SChristian Brauner } 2768c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2769c2fd68b6SChristian Brauner 2770c2fd68b6SChristian Brauner /** 2771bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2772bbddca8eSNeilBrown * @name: pathname component to lookup 2773bbddca8eSNeilBrown * @base: base directory to lookup from 2774bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2775bbddca8eSNeilBrown * 2776bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2777bbddca8eSNeilBrown * not be called by generic code. 2778bbddca8eSNeilBrown * 2779bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2780bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2781bbddca8eSNeilBrown */ 2782bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2783bbddca8eSNeilBrown struct dentry *base, int len) 2784bbddca8eSNeilBrown { 2785bbddca8eSNeilBrown struct qstr this; 2786bbddca8eSNeilBrown int err; 278720d00ee8SLinus Torvalds struct dentry *ret; 2788bbddca8eSNeilBrown 2789c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2790bbddca8eSNeilBrown if (err) 2791bbddca8eSNeilBrown return ERR_PTR(err); 2792bbddca8eSNeilBrown 279320d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 279420d00ee8SLinus Torvalds if (!ret) 279520d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 279620d00ee8SLinus Torvalds return ret; 2797bbddca8eSNeilBrown } 2798bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2799bbddca8eSNeilBrown 28006c2d4798SAl Viro /* 28016c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 28026c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 28036c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 28046c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 28056c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 28066c2d4798SAl Viro * this one avoids such problems. 28076c2d4798SAl Viro */ 28086c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28096c2d4798SAl Viro struct dentry *base, int len) 28106c2d4798SAl Viro { 28116c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 28122fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 28136c2d4798SAl Viro dput(ret); 28146c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 28156c2d4798SAl Viro } 28166c2d4798SAl Viro return ret; 28176c2d4798SAl Viro } 28186c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28196c2d4798SAl Viro 2820eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2821eedf265aSEric W. Biederman int path_pts(struct path *path) 2822eedf265aSEric W. Biederman { 2823eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2824eedf265aSEric W. Biederman * the input path. 2825eedf265aSEric W. Biederman */ 2826a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2827a6a7eb76SAl Viro struct dentry *child; 282819f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2829eedf265aSEric W. Biederman 2830a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2831a6a7eb76SAl Viro dput(parent); 283263b27720SAl Viro return -ENOENT; 2833a6a7eb76SAl Viro } 283463b27720SAl Viro dput(path->dentry); 283563b27720SAl Viro path->dentry = parent; 2836eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2837eedf265aSEric W. Biederman if (!child) 2838eedf265aSEric W. Biederman return -ENOENT; 2839eedf265aSEric W. Biederman 2840eedf265aSEric W. Biederman path->dentry = child; 2841eedf265aSEric W. Biederman dput(parent); 284219f6028aSAl Viro follow_down(path); 2843eedf265aSEric W. Biederman return 0; 2844eedf265aSEric W. Biederman } 2845eedf265aSEric W. Biederman #endif 2846eedf265aSEric W. Biederman 28471fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 28481fa1e7f6SAndy Whitcroft struct path *path, int *empty) 28491da177e4SLinus Torvalds { 2850794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2851794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2852794ebceaSStephen Brennan 2853794ebceaSStephen Brennan putname(filename); 2854794ebceaSStephen Brennan return ret; 28551da177e4SLinus Torvalds } 2856b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 28571fa1e7f6SAndy Whitcroft 2858ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2859ba73d987SChristian Brauner struct inode *inode) 28601da177e4SLinus Torvalds { 28618e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2862da9592edSDavid Howells 2863ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) 28641da177e4SLinus Torvalds return 0; 2865ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) 28661da177e4SLinus Torvalds return 0; 2867ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 28681da177e4SLinus Torvalds } 2869cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 28701da177e4SLinus Torvalds 28711da177e4SLinus Torvalds /* 28721da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 28731da177e4SLinus Torvalds * whether the type of victim is right. 28741da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 28751da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 28761da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 28771da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 28781da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 28791da177e4SLinus Torvalds * a. be owner of dir, or 28801da177e4SLinus Torvalds * b. be owner of victim, or 28811da177e4SLinus Torvalds * c. have CAP_FOWNER capability 28821da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 28831da177e4SLinus Torvalds * links pointing to it. 28840bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 28850bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 28860bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 28870bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 28880bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 28891da177e4SLinus Torvalds * nfs_async_unlink(). 28901da177e4SLinus Torvalds */ 2891ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2892ba73d987SChristian Brauner struct dentry *victim, bool isdir) 28931da177e4SLinus Torvalds { 289463afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 28951da177e4SLinus Torvalds int error; 28961da177e4SLinus Torvalds 2897b18825a7SDavid Howells if (d_is_negative(victim)) 28981da177e4SLinus Torvalds return -ENOENT; 2899b18825a7SDavid Howells BUG_ON(!inode); 29001da177e4SLinus Torvalds 29011da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2902593d1ce8SEric W. Biederman 2903593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2904ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 2905ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 2906593d1ce8SEric W. Biederman return -EOVERFLOW; 2907593d1ce8SEric W. Biederman 29084fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29091da177e4SLinus Torvalds 2910ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29111da177e4SLinus Torvalds if (error) 29121da177e4SLinus Torvalds return error; 29131da177e4SLinus Torvalds if (IS_APPEND(dir)) 29141da177e4SLinus Torvalds return -EPERM; 2915b18825a7SDavid Howells 2916ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2917ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2918ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 29191da177e4SLinus Torvalds return -EPERM; 29201da177e4SLinus Torvalds if (isdir) { 292144b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29221da177e4SLinus Torvalds return -ENOTDIR; 29231da177e4SLinus Torvalds if (IS_ROOT(victim)) 29241da177e4SLinus Torvalds return -EBUSY; 292544b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29261da177e4SLinus Torvalds return -EISDIR; 29271da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29281da177e4SLinus Torvalds return -ENOENT; 29291da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29301da177e4SLinus Torvalds return -EBUSY; 29311da177e4SLinus Torvalds return 0; 29321da177e4SLinus Torvalds } 29331da177e4SLinus Torvalds 29341da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29351da177e4SLinus Torvalds * dir. 29361da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29371da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29381da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2939036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2940036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2941036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 29421da177e4SLinus Torvalds */ 2943ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2944ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 29451da177e4SLinus Torvalds { 294614e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 29471da177e4SLinus Torvalds if (child->d_inode) 29481da177e4SLinus Torvalds return -EEXIST; 29491da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29501da177e4SLinus Torvalds return -ENOENT; 29518e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 2952036d5236SEric W. Biederman return -EOVERFLOW; 29538e538913SChristian Brauner 2954ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29551da177e4SLinus Torvalds } 29561da177e4SLinus Torvalds 29571da177e4SLinus Torvalds /* 29581da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 29591da177e4SLinus Torvalds */ 29601da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 29611da177e4SLinus Torvalds { 29621da177e4SLinus Torvalds struct dentry *p; 29631da177e4SLinus Torvalds 29641da177e4SLinus Torvalds if (p1 == p2) { 29655955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29661da177e4SLinus Torvalds return NULL; 29671da177e4SLinus Torvalds } 29681da177e4SLinus Torvalds 2969fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 29701da177e4SLinus Torvalds 2971e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2972e2761a11SOGAWA Hirofumi if (p) { 29735955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 29745955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 29751da177e4SLinus Torvalds return p; 29761da177e4SLinus Torvalds } 29771da177e4SLinus Torvalds 2978e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2979e2761a11SOGAWA Hirofumi if (p) { 29805955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29815955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 29821da177e4SLinus Torvalds return p; 29831da177e4SLinus Torvalds } 29841da177e4SLinus Torvalds 29855955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29865955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 29871da177e4SLinus Torvalds return NULL; 29881da177e4SLinus Torvalds } 29894d359507SAl Viro EXPORT_SYMBOL(lock_rename); 29901da177e4SLinus Torvalds 29911da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 29921da177e4SLinus Torvalds { 29935955102cSAl Viro inode_unlock(p1->d_inode); 29941da177e4SLinus Torvalds if (p1 != p2) { 29955955102cSAl Viro inode_unlock(p2->d_inode); 2996fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 29971da177e4SLinus Torvalds } 29981da177e4SLinus Torvalds } 29994d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 30001da177e4SLinus Torvalds 30016521f891SChristian Brauner /** 30026521f891SChristian Brauner * vfs_create - create new file 30036521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 30046521f891SChristian Brauner * @dir: inode of @dentry 30056521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 30066521f891SChristian Brauner * @mode: mode of the new file 30076521f891SChristian Brauner * @want_excl: whether the file must not yet exist 30086521f891SChristian Brauner * 30096521f891SChristian Brauner * Create a new file. 30106521f891SChristian Brauner * 30116521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 30126521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 30136521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 30146521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 30156521f891SChristian Brauner * raw inode simply passs init_user_ns. 30166521f891SChristian Brauner */ 30176521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 30186521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 30191da177e4SLinus Torvalds { 30206521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 30211da177e4SLinus Torvalds if (error) 30221da177e4SLinus Torvalds return error; 30231da177e4SLinus Torvalds 3024acfa4380SAl Viro if (!dir->i_op->create) 30251da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 30261da177e4SLinus Torvalds mode &= S_IALLUGO; 30271da177e4SLinus Torvalds mode |= S_IFREG; 30281da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 30291da177e4SLinus Torvalds if (error) 30301da177e4SLinus Torvalds return error; 3031549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 3032a74574aaSStephen Smalley if (!error) 3033f38aa942SAmy Griffis fsnotify_create(dir, dentry); 30341da177e4SLinus Torvalds return error; 30351da177e4SLinus Torvalds } 30364d359507SAl Viro EXPORT_SYMBOL(vfs_create); 30371da177e4SLinus Torvalds 30388e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 30398e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 30408e6c848eSAl Viro void *arg) 30418e6c848eSAl Viro { 30428e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 3043ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 30448e6c848eSAl Viro if (error) 30458e6c848eSAl Viro return error; 30468e6c848eSAl Viro 30478e6c848eSAl Viro mode &= S_IALLUGO; 30488e6c848eSAl Viro mode |= S_IFREG; 30498e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 30508e6c848eSAl Viro if (error) 30518e6c848eSAl Viro return error; 30528e6c848eSAl Viro error = f(dentry, mode, arg); 30538e6c848eSAl Viro if (!error) 30548e6c848eSAl Viro fsnotify_create(dir, dentry); 30558e6c848eSAl Viro return error; 30568e6c848eSAl Viro } 30578e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 30588e6c848eSAl Viro 3059a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3060a2982cc9SEric W. Biederman { 3061a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3062a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3063a2982cc9SEric W. Biederman } 3064a2982cc9SEric W. Biederman 3065ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 3066ba73d987SChristian Brauner int acc_mode, int flag) 30671da177e4SLinus Torvalds { 30683fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 30691da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 30701da177e4SLinus Torvalds int error; 30711da177e4SLinus Torvalds 30721da177e4SLinus Torvalds if (!inode) 30731da177e4SLinus Torvalds return -ENOENT; 30741da177e4SLinus Torvalds 3075c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3076c8fe8f30SChristoph Hellwig case S_IFLNK: 30771da177e4SLinus Torvalds return -ELOOP; 3078c8fe8f30SChristoph Hellwig case S_IFDIR: 3079fc4177beSKees Cook if (acc_mode & MAY_WRITE) 30801da177e4SLinus Torvalds return -EISDIR; 3081fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3082fc4177beSKees Cook return -EACCES; 3083c8fe8f30SChristoph Hellwig break; 3084c8fe8f30SChristoph Hellwig case S_IFBLK: 3085c8fe8f30SChristoph Hellwig case S_IFCHR: 3086a2982cc9SEric W. Biederman if (!may_open_dev(path)) 30871da177e4SLinus Torvalds return -EACCES; 3088633fb6acSKees Cook fallthrough; 3089c8fe8f30SChristoph Hellwig case S_IFIFO: 3090c8fe8f30SChristoph Hellwig case S_IFSOCK: 3091633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3092633fb6acSKees Cook return -EACCES; 30931da177e4SLinus Torvalds flag &= ~O_TRUNC; 3094c8fe8f30SChristoph Hellwig break; 30950fd338b2SKees Cook case S_IFREG: 30960fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 30970fd338b2SKees Cook return -EACCES; 30980fd338b2SKees Cook break; 30994a3fd211SDave Hansen } 3100b41572e9SDave Hansen 3101ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 3102b41572e9SDave Hansen if (error) 3103b41572e9SDave Hansen return error; 31046146f0d5SMimi Zohar 31051da177e4SLinus Torvalds /* 31061da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 31071da177e4SLinus Torvalds */ 31081da177e4SLinus Torvalds if (IS_APPEND(inode)) { 31098737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 31107715b521SAl Viro return -EPERM; 31111da177e4SLinus Torvalds if (flag & O_TRUNC) 31127715b521SAl Viro return -EPERM; 31131da177e4SLinus Torvalds } 31141da177e4SLinus Torvalds 31151da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3116ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 31177715b521SAl Viro return -EPERM; 31181da177e4SLinus Torvalds 3119f3c7691eSJ. Bruce Fields return 0; 31207715b521SAl Viro } 31217715b521SAl Viro 3122549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 31237715b521SAl Viro { 3124f0bb5aafSAl Viro const struct path *path = &filp->f_path; 31257715b521SAl Viro struct inode *inode = path->dentry->d_inode; 31267715b521SAl Viro int error = get_write_access(inode); 31271da177e4SLinus Torvalds if (error) 31287715b521SAl Viro return error; 3129482e0007SJeff Layton 3130ea0d3ab2STetsuo Handa error = security_path_truncate(path); 31311da177e4SLinus Torvalds if (!error) { 3132549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3133d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3134e1181ee6SJeff Layton filp); 31351da177e4SLinus Torvalds } 31361da177e4SLinus Torvalds put_write_access(inode); 3137acd0c935SMimi Zohar return error; 31381da177e4SLinus Torvalds } 31391da177e4SLinus Torvalds 3140d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3141d57999e1SDave Hansen { 31428a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 31438a5e929dSAl Viro flag--; 3144d57999e1SDave Hansen return flag; 3145d57999e1SDave Hansen } 3146d57999e1SDave Hansen 3147ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3148ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3149ba73d987SChristian Brauner umode_t mode) 3150d18e9008SMiklos Szeredi { 3151d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3152d18e9008SMiklos Szeredi if (error) 3153d18e9008SMiklos Szeredi return error; 3154d18e9008SMiklos Szeredi 31558e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) 31561328c727SSeth Forshee return -EOVERFLOW; 31571328c727SSeth Forshee 3158ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 315947291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3160d18e9008SMiklos Szeredi if (error) 3161d18e9008SMiklos Szeredi return error; 3162d18e9008SMiklos Szeredi 3163d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3164d18e9008SMiklos Szeredi } 3165d18e9008SMiklos Szeredi 31661acf0af9SDavid Howells /* 31671acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 31681acf0af9SDavid Howells * dentry. 31691acf0af9SDavid Howells * 31701acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 31711acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 31721acf0af9SDavid Howells * 317300a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 317400a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 317500a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 31761acf0af9SDavid Howells * 31771acf0af9SDavid Howells * Returns an error code otherwise. 31781acf0af9SDavid Howells */ 3179239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3180239eb983SAl Viro struct file *file, 31813ec2eef1SAl Viro int open_flag, umode_t mode) 3182d18e9008SMiklos Szeredi { 3183d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3184d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3185d18e9008SMiklos Szeredi int error; 3186d18e9008SMiklos Szeredi 3187d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3188d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3189d18e9008SMiklos Szeredi 319030d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 319130d90494SAl Viro file->f_path.mnt = nd->path.mnt; 31920fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 319344907d79SAl Viro open_to_namei_flags(open_flag), mode); 31946fbd0714SAl Viro d_lookup_done(dentry); 3195384f26e2SAl Viro if (!error) { 319664e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 31976fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 31986fb968cdSAl Viro dput(dentry); 31996fb968cdSAl Viro dentry = dget(file->f_path.dentry); 32006fb968cdSAl Viro } 320164e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 32022675a4ebSAl Viro error = -EIO; 3203384f26e2SAl Viro } else { 320430d90494SAl Viro if (file->f_path.dentry) { 3205d18e9008SMiklos Szeredi dput(dentry); 3206d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 320710c64ceaSAl Viro } 3208239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3209a01e718fSAl Viro error = -ENOENT; 3210d18e9008SMiklos Szeredi } 3211d18e9008SMiklos Szeredi } 3212239eb983SAl Viro if (error) { 3213d18e9008SMiklos Szeredi dput(dentry); 3214239eb983SAl Viro dentry = ERR_PTR(error); 3215239eb983SAl Viro } 3216239eb983SAl Viro return dentry; 3217d18e9008SMiklos Szeredi } 3218d18e9008SMiklos Szeredi 321931e6b01fSNick Piggin /* 32201acf0af9SDavid Howells * Look up and maybe create and open the last component. 3221d58ffd35SMiklos Szeredi * 322200a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3223d58ffd35SMiklos Szeredi * 322400a07c15SAl Viro * Returns 0 on success, that is, if 322500a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 322600a07c15SAl Viro * the file was not completely opened at this time, though lookups and 322700a07c15SAl Viro * creations were performed. 322800a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 322900a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 323000a07c15SAl Viro * hadn't been specified. 32311acf0af9SDavid Howells * 323200a07c15SAl Viro * An error code is returned on failure. 3233d58ffd35SMiklos Szeredi */ 3234da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3235d58ffd35SMiklos Szeredi const struct open_flags *op, 32363ec2eef1SAl Viro bool got_write) 3237d58ffd35SMiklos Szeredi { 3238549c7297SChristian Brauner struct user_namespace *mnt_userns; 3239d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 324054ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 32411643b43fSAl Viro int open_flag = op->open_flag; 3242d58ffd35SMiklos Szeredi struct dentry *dentry; 32431643b43fSAl Viro int error, create_error = 0; 32441643b43fSAl Viro umode_t mode = op->mode; 32456fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3246d58ffd35SMiklos Szeredi 3247ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3248da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3249d58ffd35SMiklos Szeredi 325073a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 32516fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 32526fbd0714SAl Viro for (;;) { 32536fbd0714SAl Viro if (!dentry) { 32546fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3255d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3256da5ebf5aSAl Viro return dentry; 32576fbd0714SAl Viro } 32586fbd0714SAl Viro if (d_in_lookup(dentry)) 32596fbd0714SAl Viro break; 3260d58ffd35SMiklos Szeredi 32616fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 32626fbd0714SAl Viro if (likely(error > 0)) 32636fbd0714SAl Viro break; 32646fbd0714SAl Viro if (error) 32656fbd0714SAl Viro goto out_dput; 32666fbd0714SAl Viro d_invalidate(dentry); 32676fbd0714SAl Viro dput(dentry); 32686fbd0714SAl Viro dentry = NULL; 32696fbd0714SAl Viro } 32706fbd0714SAl Viro if (dentry->d_inode) { 3271d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3272da5ebf5aSAl Viro return dentry; 32736c51e513SAl Viro } 3274d18e9008SMiklos Szeredi 32751643b43fSAl Viro /* 32761643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 32771643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 32781643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 32791643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 32801643b43fSAl Viro * 32811643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 32821643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 32831643b43fSAl Viro */ 328499a4a90cSAl Viro if (unlikely(!got_write)) 328599a4a90cSAl Viro open_flag &= ~O_TRUNC; 3286549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 32871643b43fSAl Viro if (open_flag & O_CREAT) { 328899a4a90cSAl Viro if (open_flag & O_EXCL) 328999a4a90cSAl Viro open_flag &= ~O_TRUNC; 32901643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 32911643b43fSAl Viro mode &= ~current_umask(); 329299a4a90cSAl Viro if (likely(got_write)) 3293549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3294ba73d987SChristian Brauner dentry, mode); 329599a4a90cSAl Viro else 329699a4a90cSAl Viro create_error = -EROFS; 329799a4a90cSAl Viro } 329899a4a90cSAl Viro if (create_error) 32991643b43fSAl Viro open_flag &= ~O_CREAT; 33001643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3301d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3302da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3303da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3304da5ebf5aSAl Viro return dentry; 3305239eb983SAl Viro } 330654ef4872SMiklos Szeredi 33076fbd0714SAl Viro if (d_in_lookup(dentry)) { 330812fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 330912fa5e24SAl Viro nd->flags); 33106fbd0714SAl Viro d_lookup_done(dentry); 331112fa5e24SAl Viro if (unlikely(res)) { 331212fa5e24SAl Viro if (IS_ERR(res)) { 331312fa5e24SAl Viro error = PTR_ERR(res); 331412fa5e24SAl Viro goto out_dput; 331512fa5e24SAl Viro } 331612fa5e24SAl Viro dput(dentry); 331712fa5e24SAl Viro dentry = res; 331812fa5e24SAl Viro } 331954ef4872SMiklos Szeredi } 332054ef4872SMiklos Szeredi 3321d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 33221643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 332373a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3324ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3325ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3326ce8644fcSAl Viro error = -EACCES; 3327d58ffd35SMiklos Szeredi goto out_dput; 332864894cf8SAl Viro } 3329549c7297SChristian Brauner 3330549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3331549c7297SChristian Brauner mode, open_flag & O_EXCL); 3332d58ffd35SMiklos Szeredi if (error) 3333d58ffd35SMiklos Szeredi goto out_dput; 3334d58ffd35SMiklos Szeredi } 33351643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 33361643b43fSAl Viro error = create_error; 3337d58ffd35SMiklos Szeredi goto out_dput; 3338d58ffd35SMiklos Szeredi } 3339da5ebf5aSAl Viro return dentry; 3340d58ffd35SMiklos Szeredi 3341d58ffd35SMiklos Szeredi out_dput: 3342d58ffd35SMiklos Szeredi dput(dentry); 3343da5ebf5aSAl Viro return ERR_PTR(error); 3344d58ffd35SMiklos Szeredi } 3345d58ffd35SMiklos Szeredi 3346c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 33473ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3348fb1cc555SAl Viro { 3349a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3350ca344a89SAl Viro int open_flag = op->open_flag; 335164894cf8SAl Viro bool got_write = false; 3352254cf582SAl Viro unsigned seq; 3353a1eb3315SMiklos Szeredi struct inode *inode; 3354da5ebf5aSAl Viro struct dentry *dentry; 3355b0417d2cSAl Viro const char *res; 3356fb1cc555SAl Viro 3357c3e380b0SAl Viro nd->flags |= op->intent; 3358c3e380b0SAl Viro 3359bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 336056676ec3SAl Viro if (nd->depth) 336156676ec3SAl Viro put_link(nd); 3362ff326a32SAl Viro return handle_dots(nd, nd->last_type); 33631f36f774SAl Viro } 3364a2c36b45SAl Viro 3365ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3366fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3367fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3368fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 336920e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 337020e34357SAl Viro if (IS_ERR(dentry)) 33711ccac622SAl Viro return ERR_CAST(dentry); 337220e34357SAl Viro if (likely(dentry)) 337371574865SMiklos Szeredi goto finish_lookup; 337471574865SMiklos Szeredi 33756583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3376b6183df7SMiklos Szeredi } else { 3377fe2d35ffSAl Viro /* create side of things */ 337872287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3379e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3380e36cffedSJens Axboe return ERR_PTR(-ECHILD); 338172287417SAl Viro } 3382c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 33831f36f774SAl Viro /* trailing slashes? */ 3384deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 33851ccac622SAl Viro return ERR_PTR(-EISDIR); 3386b6183df7SMiklos Szeredi } 33871f36f774SAl Viro 33889cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3389e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 339064894cf8SAl Viro /* 339164894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 339264894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 339364894cf8SAl Viro * dropping this one anyway. 339464894cf8SAl Viro */ 339564894cf8SAl Viro } 33969cf843e3SAl Viro if (open_flag & O_CREAT) 33975955102cSAl Viro inode_lock(dir->d_inode); 33989cf843e3SAl Viro else 33999cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3400da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3401f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3402f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 34039cf843e3SAl Viro if (open_flag & O_CREAT) 34045955102cSAl Viro inode_unlock(dir->d_inode); 34059cf843e3SAl Viro else 34069cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3407fb1cc555SAl Viro 3408c981a482SAl Viro if (got_write) 340959e96e65SAl Viro mnt_drop_write(nd->path.mnt); 34106c0d46c4SAl Viro 341159e96e65SAl Viro if (IS_ERR(dentry)) 341259e96e65SAl Viro return ERR_CAST(dentry); 341359e96e65SAl Viro 3414973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3415e73cabffSAl Viro dput(nd->path.dentry); 3416e73cabffSAl Viro nd->path.dentry = dentry; 3417c981a482SAl Viro return NULL; 3418fb1cc555SAl Viro } 3419fb1cc555SAl Viro 342020e34357SAl Viro finish_lookup: 342156676ec3SAl Viro if (nd->depth) 342256676ec3SAl Viro put_link(nd); 34238c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3424ff326a32SAl Viro if (unlikely(res)) 34251ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3426b0417d2cSAl Viro return res; 34271ccac622SAl Viro } 342831d1726dSAl Viro 3429c981a482SAl Viro /* 3430c981a482SAl Viro * Handle the last step of open() 3431c981a482SAl Viro */ 3432c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3433c981a482SAl Viro struct file *file, const struct open_flags *op) 3434c981a482SAl Viro { 3435549c7297SChristian Brauner struct user_namespace *mnt_userns; 3436c981a482SAl Viro int open_flag = op->open_flag; 3437c981a482SAl Viro bool do_truncate; 3438c981a482SAl Viro int acc_mode; 3439c981a482SAl Viro int error; 3440c981a482SAl Viro 3441ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3442ff326a32SAl Viro error = complete_walk(nd); 3443ff326a32SAl Viro if (error) 3444ff326a32SAl Viro return error; 3445ff326a32SAl Viro } 3446973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 344776ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3448549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 344930aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3450b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3451b94e0b32SAl Viro return -EEXIST; 345230aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3453c5971b8cSAl Viro return -EISDIR; 3454549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 345530aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 345630aba665SSalvatore Mesoraca if (unlikely(error)) 3457c5971b8cSAl Viro return error; 345830aba665SSalvatore Mesoraca } 345944b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3460c5971b8cSAl Viro return -ENOTDIR; 34616c0d46c4SAl Viro 34628795e7d4SAl Viro do_truncate = false; 34638795e7d4SAl Viro acc_mode = op->acc_mode; 34645a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 34655a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 34665a2d3eddSAl Viro open_flag &= ~O_TRUNC; 34675a2d3eddSAl Viro acc_mode = 0; 34688795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 34690f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 34700f9d1a10SAl Viro if (error) 3471c5971b8cSAl Viro return error; 34728795e7d4SAl Viro do_truncate = true; 34730f9d1a10SAl Viro } 3474549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 34758795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3476ae2bb293SAl Viro error = vfs_open(&nd->path, file); 34778795e7d4SAl Viro if (!error) 34786035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 34798795e7d4SAl Viro if (!error && do_truncate) 3480549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3481c80567c8SAl Viro if (unlikely(error > 0)) { 3482c80567c8SAl Viro WARN_ON(1); 3483c80567c8SAl Viro error = -EINVAL; 3484c80567c8SAl Viro } 34858795e7d4SAl Viro if (do_truncate) 34860f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3487c5971b8cSAl Viro return error; 3488fb1cc555SAl Viro } 3489fb1cc555SAl Viro 34906521f891SChristian Brauner /** 34916521f891SChristian Brauner * vfs_tmpfile - create tmpfile 34926521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 34936521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 34946521f891SChristian Brauner * @mode: mode of the new tmpfile 34952111c3c0SRandy Dunlap * @open_flag: flags 34966521f891SChristian Brauner * 34976521f891SChristian Brauner * Create a temporary file. 34986521f891SChristian Brauner * 34996521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 35006521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 35016521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 35026521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 35036521f891SChristian Brauner * raw inode simply passs init_user_ns. 35046521f891SChristian Brauner */ 35056521f891SChristian Brauner struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, 35066521f891SChristian Brauner struct dentry *dentry, umode_t mode, int open_flag) 3507af7bd4dcSAmir Goldstein { 3508af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3509af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3510af7bd4dcSAmir Goldstein struct inode *inode; 3511af7bd4dcSAmir Goldstein int error; 3512af7bd4dcSAmir Goldstein 3513af7bd4dcSAmir Goldstein /* we want directory to be writable */ 35146521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3515af7bd4dcSAmir Goldstein if (error) 3516af7bd4dcSAmir Goldstein goto out_err; 3517af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3518af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3519af7bd4dcSAmir Goldstein goto out_err; 3520af7bd4dcSAmir Goldstein error = -ENOMEM; 3521cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3522af7bd4dcSAmir Goldstein if (unlikely(!child)) 3523af7bd4dcSAmir Goldstein goto out_err; 3524549c7297SChristian Brauner error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); 3525af7bd4dcSAmir Goldstein if (error) 3526af7bd4dcSAmir Goldstein goto out_err; 3527af7bd4dcSAmir Goldstein error = -ENOENT; 3528af7bd4dcSAmir Goldstein inode = child->d_inode; 3529af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3530af7bd4dcSAmir Goldstein goto out_err; 3531af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3532af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3533af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3534af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3535af7bd4dcSAmir Goldstein } 3536a2d2329eSChristian Brauner ima_post_create_tmpfile(mnt_userns, inode); 3537af7bd4dcSAmir Goldstein return child; 3538af7bd4dcSAmir Goldstein 3539af7bd4dcSAmir Goldstein out_err: 3540af7bd4dcSAmir Goldstein dput(child); 3541af7bd4dcSAmir Goldstein return ERR_PTR(error); 3542af7bd4dcSAmir Goldstein } 3543af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3544af7bd4dcSAmir Goldstein 3545c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 354660545d0dSAl Viro const struct open_flags *op, 35473ec2eef1SAl Viro struct file *file) 354860545d0dSAl Viro { 35496521f891SChristian Brauner struct user_namespace *mnt_userns; 3550625b6d10SAl Viro struct dentry *child; 3551625b6d10SAl Viro struct path path; 3552c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 355360545d0dSAl Viro if (unlikely(error)) 355460545d0dSAl Viro return error; 3555625b6d10SAl Viro error = mnt_want_write(path.mnt); 355660545d0dSAl Viro if (unlikely(error)) 355760545d0dSAl Viro goto out; 35586521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 35596521f891SChristian Brauner child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); 3560af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3561684e73beSHirofumi Nakagawa if (IS_ERR(child)) 356260545d0dSAl Viro goto out2; 3563625b6d10SAl Viro dput(path.dentry); 3564625b6d10SAl Viro path.dentry = child; 3565c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 356669a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3567549c7297SChristian Brauner error = may_open(mnt_userns, &path, 0, op->open_flag); 35681e8f44f1SAl Viro if (!error) 35691e8f44f1SAl Viro error = vfs_open(&path, file); 357060545d0dSAl Viro out2: 3571625b6d10SAl Viro mnt_drop_write(path.mnt); 357260545d0dSAl Viro out: 3573625b6d10SAl Viro path_put(&path); 357460545d0dSAl Viro return error; 357560545d0dSAl Viro } 357660545d0dSAl Viro 35776ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 35786ac08709SAl Viro { 35796ac08709SAl Viro struct path path; 35806ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 35816ac08709SAl Viro if (!error) { 35826ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3583ae2bb293SAl Viro error = vfs_open(&path, file); 35846ac08709SAl Viro path_put(&path); 35856ac08709SAl Viro } 35866ac08709SAl Viro return error; 35876ac08709SAl Viro } 35886ac08709SAl Viro 3589c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3590c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35911da177e4SLinus Torvalds { 359230d90494SAl Viro struct file *file; 359313aab428SAl Viro int error; 359431e6b01fSNick Piggin 3595ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35961afc99beSAl Viro if (IS_ERR(file)) 35971afc99beSAl Viro return file; 359831e6b01fSNick Piggin 3599bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 36003ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 36015f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 36026ac08709SAl Viro error = do_o_path(nd, flags, file); 36035f336e72SAl Viro } else { 36045f336e72SAl Viro const char *s = path_init(nd, flags); 36053bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3606c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 36071ccac622SAl Viro ; 3608c5971b8cSAl Viro if (!error) 3609c5971b8cSAl Viro error = do_open(nd, file, op); 3610deb106c6SAl Viro terminate_walk(nd); 36115f336e72SAl Viro } 36127c1c01ecSAl Viro if (likely(!error)) { 3613aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 36147c1c01ecSAl Viro return file; 36157c1c01ecSAl Viro WARN_ON(1); 36167c1c01ecSAl Viro error = -EINVAL; 3617015c3bbcSMiklos Szeredi } 36187c1c01ecSAl Viro fput(file); 36192675a4ebSAl Viro if (error == -EOPENSTALE) { 36202675a4ebSAl Viro if (flags & LOOKUP_RCU) 36212675a4ebSAl Viro error = -ECHILD; 36222675a4ebSAl Viro else 36232675a4ebSAl Viro error = -ESTALE; 36242675a4ebSAl Viro } 36257c1c01ecSAl Viro return ERR_PTR(error); 3626de459215SKirill Korotaev } 36271da177e4SLinus Torvalds 3628669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3629f9652e10SAl Viro const struct open_flags *op) 363013aab428SAl Viro { 36319883d185SAl Viro struct nameidata nd; 3632f9652e10SAl Viro int flags = op->lookup_flags; 363313aab428SAl Viro struct file *filp; 363413aab428SAl Viro 363506422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3636c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 363713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3638c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 363913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3640c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 36419883d185SAl Viro restore_nameidata(); 364213aab428SAl Viro return filp; 364313aab428SAl Viro } 364413aab428SAl Viro 3645ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3646f9652e10SAl Viro const char *name, const struct open_flags *op) 364773d049a4SAl Viro { 36489883d185SAl Viro struct nameidata nd; 364973d049a4SAl Viro struct file *file; 365051689104SPaul Moore struct filename *filename; 3651bcba1e7dSAl Viro int flags = op->lookup_flags; 365273d049a4SAl Viro 3653ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 365473d049a4SAl Viro return ERR_PTR(-ELOOP); 365573d049a4SAl Viro 365651689104SPaul Moore filename = getname_kernel(name); 3657a1c83681SViresh Kumar if (IS_ERR(filename)) 365851689104SPaul Moore return ERR_CAST(filename); 365951689104SPaul Moore 366006422964SAl Viro set_nameidata(&nd, -1, filename, root); 3661c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 366273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3663c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 366473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3665c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 36669883d185SAl Viro restore_nameidata(); 366751689104SPaul Moore putname(filename); 366873d049a4SAl Viro return file; 366973d049a4SAl Viro } 367073d049a4SAl Viro 3671b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 36721ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36731da177e4SLinus Torvalds { 3674c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3675391172c4SAl Viro struct qstr last; 3676*b3d4650dSNeilBrown bool want_dir = lookup_flags & LOOKUP_DIRECTORY; 3677*b3d4650dSNeilBrown unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; 3678*b3d4650dSNeilBrown unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; 3679391172c4SAl Viro int type; 3680c30dabfeSJan Kara int err2; 36811ac12b4bSJeff Layton int error; 36821ac12b4bSJeff Layton 3683*b3d4650dSNeilBrown error = filename_parentat(dfd, name, reval_flag, path, &last, &type); 36840ee50b47SDmitry Kadashev if (error) 36850ee50b47SDmitry Kadashev return ERR_PTR(error); 36861da177e4SLinus Torvalds 3687c663e5d8SChristoph Hellwig /* 3688c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3689c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3690c663e5d8SChristoph Hellwig */ 36915c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3692ed75e95dSAl Viro goto out; 3693c663e5d8SChristoph Hellwig 3694c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3695391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3696c663e5d8SChristoph Hellwig /* 3697*b3d4650dSNeilBrown * Do the final lookup. Suppress 'create' if there is a trailing 3698*b3d4650dSNeilBrown * '/', and a directory wasn't requested. 3699c663e5d8SChristoph Hellwig */ 3700*b3d4650dSNeilBrown if (last.name[last.len] && !want_dir) 3701*b3d4650dSNeilBrown create_flags = 0; 37025955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3703*b3d4650dSNeilBrown dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags); 37041da177e4SLinus Torvalds if (IS_ERR(dentry)) 3705a8104a9fSAl Viro goto unlock; 3706c663e5d8SChristoph Hellwig 3707a8104a9fSAl Viro error = -EEXIST; 3708b18825a7SDavid Howells if (d_is_positive(dentry)) 3709a8104a9fSAl Viro goto fail; 3710b18825a7SDavid Howells 3711c663e5d8SChristoph Hellwig /* 3712c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3713c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3714c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3715c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3716c663e5d8SChristoph Hellwig */ 3717*b3d4650dSNeilBrown if (unlikely(!create_flags)) { 3718a8104a9fSAl Viro error = -ENOENT; 3719ed75e95dSAl Viro goto fail; 3720e9baf6e5SAl Viro } 3721c30dabfeSJan Kara if (unlikely(err2)) { 3722c30dabfeSJan Kara error = err2; 3723a8104a9fSAl Viro goto fail; 3724c30dabfeSJan Kara } 3725e9baf6e5SAl Viro return dentry; 37261da177e4SLinus Torvalds fail: 3727a8104a9fSAl Viro dput(dentry); 3728a8104a9fSAl Viro dentry = ERR_PTR(error); 3729a8104a9fSAl Viro unlock: 37305955102cSAl Viro inode_unlock(path->dentry->d_inode); 3731c30dabfeSJan Kara if (!err2) 3732391172c4SAl Viro mnt_drop_write(path->mnt); 3733ed75e95dSAl Viro out: 3734391172c4SAl Viro path_put(path); 3735ed75e95dSAl Viro return dentry; 3736dae6ad8fSAl Viro } 3737fa14a0b8SAl Viro 3738fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3739fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3740fa14a0b8SAl Viro { 3741b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3742b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3743b4a4f213SStephen Brennan 3744b4a4f213SStephen Brennan putname(filename); 3745b4a4f213SStephen Brennan return res; 3746fa14a0b8SAl Viro } 3747dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3748dae6ad8fSAl Viro 3749921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3750921a1650SAl Viro { 3751921a1650SAl Viro dput(dentry); 37525955102cSAl Viro inode_unlock(path->dentry->d_inode); 3753a8104a9fSAl Viro mnt_drop_write(path->mnt); 3754921a1650SAl Viro path_put(path); 3755921a1650SAl Viro } 3756921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3757921a1650SAl Viro 3758520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 37591ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3760dae6ad8fSAl Viro { 3761b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3762b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3763b4a4f213SStephen Brennan 3764b4a4f213SStephen Brennan putname(filename); 3765b4a4f213SStephen Brennan return res; 3766dae6ad8fSAl Viro } 3767dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3768dae6ad8fSAl Viro 37696521f891SChristian Brauner /** 37706521f891SChristian Brauner * vfs_mknod - create device node or file 37716521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 37726521f891SChristian Brauner * @dir: inode of @dentry 37736521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 37746521f891SChristian Brauner * @mode: mode of the new device node or file 37756521f891SChristian Brauner * @dev: device number of device to create 37766521f891SChristian Brauner * 37776521f891SChristian Brauner * Create a device node or file. 37786521f891SChristian Brauner * 37796521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 37806521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 37816521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 37826521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 37836521f891SChristian Brauner * raw inode simply passs init_user_ns. 37846521f891SChristian Brauner */ 37856521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 37866521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 37871da177e4SLinus Torvalds { 3788a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 37896521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 37901da177e4SLinus Torvalds 37911da177e4SLinus Torvalds if (error) 37921da177e4SLinus Torvalds return error; 37931da177e4SLinus Torvalds 3794a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3795a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 37961da177e4SLinus Torvalds return -EPERM; 37971da177e4SLinus Torvalds 3798acfa4380SAl Viro if (!dir->i_op->mknod) 37991da177e4SLinus Torvalds return -EPERM; 38001da177e4SLinus Torvalds 380108ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 380208ce5f16SSerge E. Hallyn if (error) 380308ce5f16SSerge E. Hallyn return error; 380408ce5f16SSerge E. Hallyn 38051da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 38061da177e4SLinus Torvalds if (error) 38071da177e4SLinus Torvalds return error; 38081da177e4SLinus Torvalds 3809549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3810a74574aaSStephen Smalley if (!error) 3811f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38121da177e4SLinus Torvalds return error; 38131da177e4SLinus Torvalds } 38144d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 38151da177e4SLinus Torvalds 3816f69aac00SAl Viro static int may_mknod(umode_t mode) 3817463c3197SDave Hansen { 3818463c3197SDave Hansen switch (mode & S_IFMT) { 3819463c3197SDave Hansen case S_IFREG: 3820463c3197SDave Hansen case S_IFCHR: 3821463c3197SDave Hansen case S_IFBLK: 3822463c3197SDave Hansen case S_IFIFO: 3823463c3197SDave Hansen case S_IFSOCK: 3824463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3825463c3197SDave Hansen return 0; 3826463c3197SDave Hansen case S_IFDIR: 3827463c3197SDave Hansen return -EPERM; 3828463c3197SDave Hansen default: 3829463c3197SDave Hansen return -EINVAL; 3830463c3197SDave Hansen } 3831463c3197SDave Hansen } 3832463c3197SDave Hansen 383345f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 383487c4e192SDominik Brodowski unsigned int dev) 38351da177e4SLinus Torvalds { 38366521f891SChristian Brauner struct user_namespace *mnt_userns; 38371da177e4SLinus Torvalds struct dentry *dentry; 3838dae6ad8fSAl Viro struct path path; 3839dae6ad8fSAl Viro int error; 3840972567f1SJeff Layton unsigned int lookup_flags = 0; 38411da177e4SLinus Torvalds 38428e4bfca1SAl Viro error = may_mknod(mode); 38438e4bfca1SAl Viro if (error) 38447797251bSDmitry Kadashev goto out1; 3845972567f1SJeff Layton retry: 3846b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 38477797251bSDmitry Kadashev error = PTR_ERR(dentry); 3848dae6ad8fSAl Viro if (IS_ERR(dentry)) 38497797251bSDmitry Kadashev goto out1; 38502ad94ae6SAl Viro 3851dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3852ce3b0f8dSAl Viro mode &= ~current_umask(); 3853dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3854be6d3e56SKentaro Takeda if (error) 38557797251bSDmitry Kadashev goto out2; 38566521f891SChristian Brauner 38576521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 38581da177e4SLinus Torvalds switch (mode & S_IFMT) { 38591da177e4SLinus Torvalds case 0: case S_IFREG: 38606521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 38616521f891SChristian Brauner dentry, mode, true); 386205d1a717SMimi Zohar if (!error) 3863a2d2329eSChristian Brauner ima_post_path_mknod(mnt_userns, dentry); 38641da177e4SLinus Torvalds break; 38651da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 38666521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 38676521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 38681da177e4SLinus Torvalds break; 38691da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 38706521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 38716521f891SChristian Brauner dentry, mode, 0); 38721da177e4SLinus Torvalds break; 38731da177e4SLinus Torvalds } 38747797251bSDmitry Kadashev out2: 3875921a1650SAl Viro done_path_create(&path, dentry); 3876972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3877972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3878972567f1SJeff Layton goto retry; 3879972567f1SJeff Layton } 38807797251bSDmitry Kadashev out1: 38817797251bSDmitry Kadashev putname(name); 38821da177e4SLinus Torvalds return error; 38831da177e4SLinus Torvalds } 38841da177e4SLinus Torvalds 388587c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 388687c4e192SDominik Brodowski unsigned int, dev) 388787c4e192SDominik Brodowski { 38887797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 388987c4e192SDominik Brodowski } 389087c4e192SDominik Brodowski 38918208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 38925590ff0dSUlrich Drepper { 38937797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 38945590ff0dSUlrich Drepper } 38955590ff0dSUlrich Drepper 38966521f891SChristian Brauner /** 38976521f891SChristian Brauner * vfs_mkdir - create directory 38986521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 38996521f891SChristian Brauner * @dir: inode of @dentry 39006521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39016521f891SChristian Brauner * @mode: mode of the new directory 39026521f891SChristian Brauner * 39036521f891SChristian Brauner * Create a directory. 39046521f891SChristian Brauner * 39056521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39066521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39076521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39086521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39096521f891SChristian Brauner * raw inode simply passs init_user_ns. 39106521f891SChristian Brauner */ 39116521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 39126521f891SChristian Brauner struct dentry *dentry, umode_t mode) 39131da177e4SLinus Torvalds { 39146521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 39158de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39161da177e4SLinus Torvalds 39171da177e4SLinus Torvalds if (error) 39181da177e4SLinus Torvalds return error; 39191da177e4SLinus Torvalds 3920acfa4380SAl Viro if (!dir->i_op->mkdir) 39211da177e4SLinus Torvalds return -EPERM; 39221da177e4SLinus Torvalds 39231da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 39241da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 39251da177e4SLinus Torvalds if (error) 39261da177e4SLinus Torvalds return error; 39271da177e4SLinus Torvalds 39288de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 39298de52778SAl Viro return -EMLINK; 39308de52778SAl Viro 3931549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 3932a74574aaSStephen Smalley if (!error) 3933f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 39341da177e4SLinus Torvalds return error; 39351da177e4SLinus Torvalds } 39364d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 39371da177e4SLinus Torvalds 393845f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 39391da177e4SLinus Torvalds { 39406902d925SDave Hansen struct dentry *dentry; 3941dae6ad8fSAl Viro struct path path; 3942dae6ad8fSAl Viro int error; 3943b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 39441da177e4SLinus Torvalds 3945b76d8b82SJeff Layton retry: 3946b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 3947584d3226SDmitry Kadashev error = PTR_ERR(dentry); 39486902d925SDave Hansen if (IS_ERR(dentry)) 3949584d3226SDmitry Kadashev goto out_putname; 39506902d925SDave Hansen 3951dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3952ce3b0f8dSAl Viro mode &= ~current_umask(); 3953dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 39546521f891SChristian Brauner if (!error) { 39556521f891SChristian Brauner struct user_namespace *mnt_userns; 39566521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 3957549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 3958549c7297SChristian Brauner mode); 39596521f891SChristian Brauner } 3960921a1650SAl Viro done_path_create(&path, dentry); 3961b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3962b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3963b76d8b82SJeff Layton goto retry; 3964b76d8b82SJeff Layton } 3965584d3226SDmitry Kadashev out_putname: 3966584d3226SDmitry Kadashev putname(name); 39671da177e4SLinus Torvalds return error; 39681da177e4SLinus Torvalds } 39691da177e4SLinus Torvalds 39700101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 39710101db7aSDominik Brodowski { 3972584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 39730101db7aSDominik Brodowski } 39740101db7aSDominik Brodowski 3975a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 39765590ff0dSUlrich Drepper { 3977584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 39785590ff0dSUlrich Drepper } 39795590ff0dSUlrich Drepper 39806521f891SChristian Brauner /** 39816521f891SChristian Brauner * vfs_rmdir - remove directory 39826521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 39836521f891SChristian Brauner * @dir: inode of @dentry 39846521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39856521f891SChristian Brauner * 39866521f891SChristian Brauner * Remove a directory. 39876521f891SChristian Brauner * 39886521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39896521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39906521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39916521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39926521f891SChristian Brauner * raw inode simply passs init_user_ns. 39936521f891SChristian Brauner */ 39946521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 39956521f891SChristian Brauner struct dentry *dentry) 39961da177e4SLinus Torvalds { 39976521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 39981da177e4SLinus Torvalds 39991da177e4SLinus Torvalds if (error) 40001da177e4SLinus Torvalds return error; 40011da177e4SLinus Torvalds 4002acfa4380SAl Viro if (!dir->i_op->rmdir) 40031da177e4SLinus Torvalds return -EPERM; 40041da177e4SLinus Torvalds 40051d2ef590SAl Viro dget(dentry); 40065955102cSAl Viro inode_lock(dentry->d_inode); 4007912dbc15SSage Weil 40081da177e4SLinus Torvalds error = -EBUSY; 40091bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 40101bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4011912dbc15SSage Weil goto out; 4012912dbc15SSage Weil 40131da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4014912dbc15SSage Weil if (error) 4015912dbc15SSage Weil goto out; 4016912dbc15SSage Weil 40171da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4018912dbc15SSage Weil if (error) 4019912dbc15SSage Weil goto out; 4020912dbc15SSage Weil 40218767712fSAl Viro shrink_dcache_parent(dentry); 40221da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4023d83c49f3SAl Viro dont_mount(dentry); 40248ed936b5SEric W. Biederman detach_mounts(dentry); 40251da177e4SLinus Torvalds 4026912dbc15SSage Weil out: 40275955102cSAl Viro inode_unlock(dentry->d_inode); 40281d2ef590SAl Viro dput(dentry); 4029912dbc15SSage Weil if (!error) 4030a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 40311da177e4SLinus Torvalds return error; 40321da177e4SLinus Torvalds } 40334d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 40341da177e4SLinus Torvalds 403545f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 40361da177e4SLinus Torvalds { 40376521f891SChristian Brauner struct user_namespace *mnt_userns; 40380ee50b47SDmitry Kadashev int error; 40391da177e4SLinus Torvalds struct dentry *dentry; 4040f5beed75SAl Viro struct path path; 4041f5beed75SAl Viro struct qstr last; 4042f5beed75SAl Viro int type; 4043c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4044c6ee9206SJeff Layton retry: 4045c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 40460ee50b47SDmitry Kadashev if (error) 40470ee50b47SDmitry Kadashev goto exit1; 40481da177e4SLinus Torvalds 4049f5beed75SAl Viro switch (type) { 40501da177e4SLinus Torvalds case LAST_DOTDOT: 40511da177e4SLinus Torvalds error = -ENOTEMPTY; 40520ee50b47SDmitry Kadashev goto exit2; 40531da177e4SLinus Torvalds case LAST_DOT: 40541da177e4SLinus Torvalds error = -EINVAL; 40550ee50b47SDmitry Kadashev goto exit2; 40561da177e4SLinus Torvalds case LAST_ROOT: 40571da177e4SLinus Torvalds error = -EBUSY; 40580ee50b47SDmitry Kadashev goto exit2; 40591da177e4SLinus Torvalds } 40600612d9fbSOGAWA Hirofumi 4061f5beed75SAl Viro error = mnt_want_write(path.mnt); 4062c30dabfeSJan Kara if (error) 40630ee50b47SDmitry Kadashev goto exit2; 40640612d9fbSOGAWA Hirofumi 40655955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4066f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40671da177e4SLinus Torvalds error = PTR_ERR(dentry); 40686902d925SDave Hansen if (IS_ERR(dentry)) 40690ee50b47SDmitry Kadashev goto exit3; 4070e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4071e6bc45d6STheodore Ts'o error = -ENOENT; 40720ee50b47SDmitry Kadashev goto exit4; 4073e6bc45d6STheodore Ts'o } 4074f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4075be6d3e56SKentaro Takeda if (error) 40760ee50b47SDmitry Kadashev goto exit4; 40776521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 40786521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 40790ee50b47SDmitry Kadashev exit4: 40801da177e4SLinus Torvalds dput(dentry); 40810ee50b47SDmitry Kadashev exit3: 40825955102cSAl Viro inode_unlock(path.dentry->d_inode); 4083f5beed75SAl Viro mnt_drop_write(path.mnt); 40840ee50b47SDmitry Kadashev exit2: 4085f5beed75SAl Viro path_put(&path); 4086c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4087c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4088c6ee9206SJeff Layton goto retry; 4089c6ee9206SJeff Layton } 40900ee50b47SDmitry Kadashev exit1: 409124fb33d4SAl Viro putname(name); 40921da177e4SLinus Torvalds return error; 40931da177e4SLinus Torvalds } 40941da177e4SLinus Torvalds 40953cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 40965590ff0dSUlrich Drepper { 4097e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 40985590ff0dSUlrich Drepper } 40995590ff0dSUlrich Drepper 4100b21996e3SJ. Bruce Fields /** 4101b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 41026521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 4103b21996e3SJ. Bruce Fields * @dir: parent directory 4104b21996e3SJ. Bruce Fields * @dentry: victim 4105b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4106b21996e3SJ. Bruce Fields * 4107b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4108b21996e3SJ. Bruce Fields * 4109b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4110b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4111b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4112b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4113b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4114b21996e3SJ. Bruce Fields * 4115b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4116b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4117b21996e3SJ. Bruce Fields * to be NFS exported. 41186521f891SChristian Brauner * 41196521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 41206521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 41216521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 41226521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 41236521f891SChristian Brauner * raw inode simply passs init_user_ns. 4124b21996e3SJ. Bruce Fields */ 41256521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 41266521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 41271da177e4SLinus Torvalds { 41289accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 41296521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 41301da177e4SLinus Torvalds 41311da177e4SLinus Torvalds if (error) 41321da177e4SLinus Torvalds return error; 41331da177e4SLinus Torvalds 4134acfa4380SAl Viro if (!dir->i_op->unlink) 41351da177e4SLinus Torvalds return -EPERM; 41361da177e4SLinus Torvalds 41375955102cSAl Viro inode_lock(target); 413851cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 413951cc3a66SHugh Dickins error = -EPERM; 414051cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 41411da177e4SLinus Torvalds error = -EBUSY; 41421da177e4SLinus Torvalds else { 41431da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4144bec1052eSAl Viro if (!error) { 41455a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 41465a14696cSJ. Bruce Fields if (error) 4147b21996e3SJ. Bruce Fields goto out; 41481da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 41498ed936b5SEric W. Biederman if (!error) { 4150d83c49f3SAl Viro dont_mount(dentry); 41518ed936b5SEric W. Biederman detach_mounts(dentry); 41528ed936b5SEric W. Biederman } 4153bec1052eSAl Viro } 41541da177e4SLinus Torvalds } 4155b21996e3SJ. Bruce Fields out: 41565955102cSAl Viro inode_unlock(target); 41571da177e4SLinus Torvalds 41581da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4159a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4160a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4161a37d9a17SAmir Goldstein } else if (!error) { 41629accbb97SJ. Bruce Fields fsnotify_link_count(target); 4163a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 41641da177e4SLinus Torvalds } 41650eeca283SRobert Love 41661da177e4SLinus Torvalds return error; 41671da177e4SLinus Torvalds } 41684d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 41691da177e4SLinus Torvalds 41701da177e4SLinus Torvalds /* 41711da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 41721b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 41731da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 41741da177e4SLinus Torvalds * while waiting on the I/O. 41751da177e4SLinus Torvalds */ 417645f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 41771da177e4SLinus Torvalds { 41782ad94ae6SAl Viro int error; 41791da177e4SLinus Torvalds struct dentry *dentry; 4180f5beed75SAl Viro struct path path; 4181f5beed75SAl Viro struct qstr last; 4182f5beed75SAl Viro int type; 41831da177e4SLinus Torvalds struct inode *inode = NULL; 4184b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 41855d18f813SJeff Layton unsigned int lookup_flags = 0; 41865d18f813SJeff Layton retry: 4187c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 41880ee50b47SDmitry Kadashev if (error) 41890ee50b47SDmitry Kadashev goto exit1; 41902ad94ae6SAl Viro 41911da177e4SLinus Torvalds error = -EISDIR; 4192f5beed75SAl Viro if (type != LAST_NORM) 41930ee50b47SDmitry Kadashev goto exit2; 41940612d9fbSOGAWA Hirofumi 4195f5beed75SAl Viro error = mnt_want_write(path.mnt); 4196c30dabfeSJan Kara if (error) 41970ee50b47SDmitry Kadashev goto exit2; 4198b21996e3SJ. Bruce Fields retry_deleg: 41995955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4200f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 42011da177e4SLinus Torvalds error = PTR_ERR(dentry); 42021da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 42036521f891SChristian Brauner struct user_namespace *mnt_userns; 42046521f891SChristian Brauner 42051da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4206f5beed75SAl Viro if (last.name[last.len]) 420750338b88STörök Edwin goto slashes; 42081da177e4SLinus Torvalds inode = dentry->d_inode; 4209b18825a7SDavid Howells if (d_is_negative(dentry)) 4210e6bc45d6STheodore Ts'o goto slashes; 42117de9c6eeSAl Viro ihold(inode); 4212f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4213be6d3e56SKentaro Takeda if (error) 42140ee50b47SDmitry Kadashev goto exit3; 42156521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4216549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4217549c7297SChristian Brauner &delegated_inode); 42180ee50b47SDmitry Kadashev exit3: 42191da177e4SLinus Torvalds dput(dentry); 42201da177e4SLinus Torvalds } 42215955102cSAl Viro inode_unlock(path.dentry->d_inode); 42221da177e4SLinus Torvalds if (inode) 42231da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4224b21996e3SJ. Bruce Fields inode = NULL; 4225b21996e3SJ. Bruce Fields if (delegated_inode) { 42265a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4227b21996e3SJ. Bruce Fields if (!error) 4228b21996e3SJ. Bruce Fields goto retry_deleg; 4229b21996e3SJ. Bruce Fields } 4230f5beed75SAl Viro mnt_drop_write(path.mnt); 42310ee50b47SDmitry Kadashev exit2: 4232f5beed75SAl Viro path_put(&path); 42335d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 42345d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 42355d18f813SJeff Layton inode = NULL; 42365d18f813SJeff Layton goto retry; 42375d18f813SJeff Layton } 42380ee50b47SDmitry Kadashev exit1: 4239da2f1362SChristoph Hellwig putname(name); 42401da177e4SLinus Torvalds return error; 42411da177e4SLinus Torvalds 42421da177e4SLinus Torvalds slashes: 4243b18825a7SDavid Howells if (d_is_negative(dentry)) 4244b18825a7SDavid Howells error = -ENOENT; 424544b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4246b18825a7SDavid Howells error = -EISDIR; 4247b18825a7SDavid Howells else 4248b18825a7SDavid Howells error = -ENOTDIR; 42490ee50b47SDmitry Kadashev goto exit3; 42501da177e4SLinus Torvalds } 42511da177e4SLinus Torvalds 42522e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 42535590ff0dSUlrich Drepper { 42545590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 42555590ff0dSUlrich Drepper return -EINVAL; 42565590ff0dSUlrich Drepper 42575590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4258e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4259da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 42605590ff0dSUlrich Drepper } 42615590ff0dSUlrich Drepper 42623480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 42635590ff0dSUlrich Drepper { 4264da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 42655590ff0dSUlrich Drepper } 42665590ff0dSUlrich Drepper 42676521f891SChristian Brauner /** 42686521f891SChristian Brauner * vfs_symlink - create symlink 42696521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 42706521f891SChristian Brauner * @dir: inode of @dentry 42716521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 42726521f891SChristian Brauner * @oldname: name of the file to link to 42736521f891SChristian Brauner * 42746521f891SChristian Brauner * Create a symlink. 42756521f891SChristian Brauner * 42766521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 42776521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42786521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42796521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42806521f891SChristian Brauner * raw inode simply passs init_user_ns. 42816521f891SChristian Brauner */ 42826521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 42836521f891SChristian Brauner struct dentry *dentry, const char *oldname) 42841da177e4SLinus Torvalds { 42856521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 42861da177e4SLinus Torvalds 42871da177e4SLinus Torvalds if (error) 42881da177e4SLinus Torvalds return error; 42891da177e4SLinus Torvalds 4290acfa4380SAl Viro if (!dir->i_op->symlink) 42911da177e4SLinus Torvalds return -EPERM; 42921da177e4SLinus Torvalds 42931da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 42941da177e4SLinus Torvalds if (error) 42951da177e4SLinus Torvalds return error; 42961da177e4SLinus Torvalds 4297549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4298a74574aaSStephen Smalley if (!error) 4299f38aa942SAmy Griffis fsnotify_create(dir, dentry); 43001da177e4SLinus Torvalds return error; 43011da177e4SLinus Torvalds } 43024d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 43031da177e4SLinus Torvalds 43047a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 43051da177e4SLinus Torvalds { 43062ad94ae6SAl Viro int error; 43076902d925SDave Hansen struct dentry *dentry; 4308dae6ad8fSAl Viro struct path path; 4309f46d3567SJeff Layton unsigned int lookup_flags = 0; 43101da177e4SLinus Torvalds 4311da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4312da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4313da2d0cedSDmitry Kadashev goto out_putnames; 4314da2d0cedSDmitry Kadashev } 4315f46d3567SJeff Layton retry: 4316b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 43171da177e4SLinus Torvalds error = PTR_ERR(dentry); 43186902d925SDave Hansen if (IS_ERR(dentry)) 4319da2d0cedSDmitry Kadashev goto out_putnames; 43206902d925SDave Hansen 432191a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 43226521f891SChristian Brauner if (!error) { 43236521f891SChristian Brauner struct user_namespace *mnt_userns; 43246521f891SChristian Brauner 43256521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 43266521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 43276521f891SChristian Brauner from->name); 43286521f891SChristian Brauner } 4329921a1650SAl Viro done_path_create(&path, dentry); 4330f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4331f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4332f46d3567SJeff Layton goto retry; 4333f46d3567SJeff Layton } 4334da2d0cedSDmitry Kadashev out_putnames: 4335da2d0cedSDmitry Kadashev putname(to); 43361da177e4SLinus Torvalds putname(from); 43371da177e4SLinus Torvalds return error; 43381da177e4SLinus Torvalds } 43391da177e4SLinus Torvalds 4340b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4341b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4342b724e846SDominik Brodowski { 4343da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4344b724e846SDominik Brodowski } 4345b724e846SDominik Brodowski 43463480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 43475590ff0dSUlrich Drepper { 4348da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 43495590ff0dSUlrich Drepper } 43505590ff0dSUlrich Drepper 4351146a8595SJ. Bruce Fields /** 4352146a8595SJ. Bruce Fields * vfs_link - create a new link 4353146a8595SJ. Bruce Fields * @old_dentry: object to be linked 43546521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4355146a8595SJ. Bruce Fields * @dir: new parent 4356146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4357146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4358146a8595SJ. Bruce Fields * 4359146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4360146a8595SJ. Bruce Fields * 4361146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4362146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4363146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4364146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4365146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4366146a8595SJ. Bruce Fields * 4367146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4368146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4369146a8595SJ. Bruce Fields * to be NFS exported. 43706521f891SChristian Brauner * 43716521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 43726521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 43736521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 43746521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 43756521f891SChristian Brauner * raw inode simply passs init_user_ns. 4376146a8595SJ. Bruce Fields */ 43776521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 43786521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 43796521f891SChristian Brauner struct inode **delegated_inode) 43801da177e4SLinus Torvalds { 43811da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 43828de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 43831da177e4SLinus Torvalds int error; 43841da177e4SLinus Torvalds 43851da177e4SLinus Torvalds if (!inode) 43861da177e4SLinus Torvalds return -ENOENT; 43871da177e4SLinus Torvalds 43886521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 43891da177e4SLinus Torvalds if (error) 43901da177e4SLinus Torvalds return error; 43911da177e4SLinus Torvalds 43921da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 43931da177e4SLinus Torvalds return -EXDEV; 43941da177e4SLinus Torvalds 43951da177e4SLinus Torvalds /* 43961da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 43971da177e4SLinus Torvalds */ 43981da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 43991da177e4SLinus Torvalds return -EPERM; 44000bd23d09SEric W. Biederman /* 44010bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 44020bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 44030bd23d09SEric W. Biederman * the vfs. 44040bd23d09SEric W. Biederman */ 44056521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 44060bd23d09SEric W. Biederman return -EPERM; 4407acfa4380SAl Viro if (!dir->i_op->link) 44081da177e4SLinus Torvalds return -EPERM; 44097e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 44101da177e4SLinus Torvalds return -EPERM; 44111da177e4SLinus Torvalds 44121da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 44131da177e4SLinus Torvalds if (error) 44141da177e4SLinus Torvalds return error; 44151da177e4SLinus Torvalds 44165955102cSAl Viro inode_lock(inode); 4417aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4418f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4419aae8a97dSAneesh Kumar K.V error = -ENOENT; 44208de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 44218de52778SAl Viro error = -EMLINK; 4422146a8595SJ. Bruce Fields else { 4423146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4424146a8595SJ. Bruce Fields if (!error) 44251da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4426146a8595SJ. Bruce Fields } 4427f4e0c30cSAl Viro 4428f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4429f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4430f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4431f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4432f4e0c30cSAl Viro } 44335955102cSAl Viro inode_unlock(inode); 4434e31e14ecSStephen Smalley if (!error) 44357e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 44361da177e4SLinus Torvalds return error; 44371da177e4SLinus Torvalds } 44384d359507SAl Viro EXPORT_SYMBOL(vfs_link); 44391da177e4SLinus Torvalds 44401da177e4SLinus Torvalds /* 44411da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 44421da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 44431da177e4SLinus Torvalds * newname. --KAB 44441da177e4SLinus Torvalds * 44451da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 44461da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 44471da177e4SLinus Torvalds * and other special files. --ADM 44481da177e4SLinus Torvalds */ 4449cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4450020250f3SDmitry Kadashev struct filename *new, int flags) 44511da177e4SLinus Torvalds { 44526521f891SChristian Brauner struct user_namespace *mnt_userns; 44531da177e4SLinus Torvalds struct dentry *new_dentry; 4454dae6ad8fSAl Viro struct path old_path, new_path; 4455146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 445611a7b371SAneesh Kumar K.V int how = 0; 44571da177e4SLinus Torvalds int error; 44581da177e4SLinus Torvalds 4459020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4460020250f3SDmitry Kadashev error = -EINVAL; 4461020250f3SDmitry Kadashev goto out_putnames; 4462020250f3SDmitry Kadashev } 446311a7b371SAneesh Kumar K.V /* 4464f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4465f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4466f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 446711a7b371SAneesh Kumar K.V */ 4468020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4469020250f3SDmitry Kadashev error = -ENOENT; 4470020250f3SDmitry Kadashev goto out_putnames; 4471f0cc6ffbSLinus Torvalds } 4472c04030e1SUlrich Drepper 447311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 447411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4475442e31caSJeff Layton retry: 4476794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 44771da177e4SLinus Torvalds if (error) 4478020250f3SDmitry Kadashev goto out_putnames; 44792ad94ae6SAl Viro 4480b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4481442e31caSJeff Layton (how & LOOKUP_REVAL)); 44821da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44836902d925SDave Hansen if (IS_ERR(new_dentry)) 4484020250f3SDmitry Kadashev goto out_putpath; 4485dae6ad8fSAl Viro 4486dae6ad8fSAl Viro error = -EXDEV; 4487dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4488dae6ad8fSAl Viro goto out_dput; 4489549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4490549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4491800179c9SKees Cook if (unlikely(error)) 4492800179c9SKees Cook goto out_dput; 4493dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4494be6d3e56SKentaro Takeda if (error) 4495a8104a9fSAl Viro goto out_dput; 44966521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 44976521f891SChristian Brauner new_dentry, &delegated_inode); 449875c3f29dSDave Hansen out_dput: 4499921a1650SAl Viro done_path_create(&new_path, new_dentry); 4500146a8595SJ. Bruce Fields if (delegated_inode) { 4501146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4502d22e6338SOleg Drokin if (!error) { 4503d22e6338SOleg Drokin path_put(&old_path); 4504146a8595SJ. Bruce Fields goto retry; 4505146a8595SJ. Bruce Fields } 4506d22e6338SOleg Drokin } 4507442e31caSJeff Layton if (retry_estale(error, how)) { 4508d22e6338SOleg Drokin path_put(&old_path); 4509442e31caSJeff Layton how |= LOOKUP_REVAL; 4510442e31caSJeff Layton goto retry; 4511442e31caSJeff Layton } 4512020250f3SDmitry Kadashev out_putpath: 45132d8f3038SAl Viro path_put(&old_path); 4514020250f3SDmitry Kadashev out_putnames: 4515020250f3SDmitry Kadashev putname(old); 4516020250f3SDmitry Kadashev putname(new); 45171da177e4SLinus Torvalds 45181da177e4SLinus Torvalds return error; 45191da177e4SLinus Torvalds } 45201da177e4SLinus Torvalds 452146ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 452246ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 452346ea89ebSDominik Brodowski { 4524020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4525020250f3SDmitry Kadashev newdfd, getname(newname), flags); 452646ea89ebSDominik Brodowski } 452746ea89ebSDominik Brodowski 45283480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 45295590ff0dSUlrich Drepper { 4530020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 45315590ff0dSUlrich Drepper } 45325590ff0dSUlrich Drepper 4533bc27027aSMiklos Szeredi /** 4534bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 45352111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4536bc27027aSMiklos Szeredi * 4537bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4538bc27027aSMiklos Szeredi * 4539bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4540bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4541bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4542bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4543bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4544bc27027aSMiklos Szeredi * so. 4545bc27027aSMiklos Szeredi * 4546bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4547bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4548bc27027aSMiklos Szeredi * to be NFS exported. 4549bc27027aSMiklos Szeredi * 45501da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 45511da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 45521da177e4SLinus Torvalds * Problems: 45530117d427SMauro Carvalho Chehab * 4554d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 45551da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 45561da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4557a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 45581da177e4SLinus Torvalds * story. 45596cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 45606cedba89SJ. Bruce Fields * and source (if it is not a directory). 45611b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 45621da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 45631da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4564a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 45651da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 45661da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 45671da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4568a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 45691da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 45701da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 45711da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4572e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 45731b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 45741da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4575c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 45761da177e4SLinus Torvalds * locking]. 45771da177e4SLinus Torvalds */ 45789fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 45791da177e4SLinus Torvalds { 45801da177e4SLinus Torvalds int error; 45819fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 45829fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 45839fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 45849fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 45859fe61450SChristian Brauner unsigned int flags = rd->flags; 4586bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4587bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4588bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4589da1ce067SMiklos Szeredi bool new_is_dir = false; 4590da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 459149d31c2fSAl Viro struct name_snapshot old_name; 45921da177e4SLinus Torvalds 45938d3e2936SMiklos Szeredi if (source == target) 45941da177e4SLinus Torvalds return 0; 45951da177e4SLinus Torvalds 45966521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 45971da177e4SLinus Torvalds if (error) 45981da177e4SLinus Torvalds return error; 45991da177e4SLinus Torvalds 4600da1ce067SMiklos Szeredi if (!target) { 46016521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4602da1ce067SMiklos Szeredi } else { 4603da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4604da1ce067SMiklos Szeredi 4605da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46066521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 46076521f891SChristian Brauner new_dentry, is_dir); 4608da1ce067SMiklos Szeredi else 46096521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 46106521f891SChristian Brauner new_dentry, new_is_dir); 4611da1ce067SMiklos Szeredi } 46121da177e4SLinus Torvalds if (error) 46131da177e4SLinus Torvalds return error; 46141da177e4SLinus Torvalds 46152773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 46161da177e4SLinus Torvalds return -EPERM; 46171da177e4SLinus Torvalds 4618bc27027aSMiklos Szeredi /* 4619bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4620bc27027aSMiklos Szeredi * we'll need to flip '..'. 4621bc27027aSMiklos Szeredi */ 4622da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4623da1ce067SMiklos Szeredi if (is_dir) { 46246521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 462547291baaSChristian Brauner MAY_WRITE); 4626bc27027aSMiklos Szeredi if (error) 4627bc27027aSMiklos Szeredi return error; 4628bc27027aSMiklos Szeredi } 4629da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 46306521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 463147291baaSChristian Brauner MAY_WRITE); 4632da1ce067SMiklos Szeredi if (error) 4633da1ce067SMiklos Szeredi return error; 4634da1ce067SMiklos Szeredi } 4635da1ce067SMiklos Szeredi } 46360eeca283SRobert Love 46370b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 46380b3974ebSMiklos Szeredi flags); 4639bc27027aSMiklos Szeredi if (error) 4640bc27027aSMiklos Szeredi return error; 4641bc27027aSMiklos Szeredi 464249d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4643bc27027aSMiklos Szeredi dget(new_dentry); 4644da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4645bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4646bc27027aSMiklos Szeredi else if (target) 46475955102cSAl Viro inode_lock(target); 4648bc27027aSMiklos Szeredi 464951cc3a66SHugh Dickins error = -EPERM; 465051cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 465151cc3a66SHugh Dickins goto out; 465251cc3a66SHugh Dickins 4653bc27027aSMiklos Szeredi error = -EBUSY; 46547af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4655bc27027aSMiklos Szeredi goto out; 4656bc27027aSMiklos Szeredi 4657da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4658bc27027aSMiklos Szeredi error = -EMLINK; 4659da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4660bc27027aSMiklos Szeredi goto out; 4661da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4662da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4663da1ce067SMiklos Szeredi goto out; 4664da1ce067SMiklos Szeredi } 4665da1ce067SMiklos Szeredi if (!is_dir) { 4666bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4667bc27027aSMiklos Szeredi if (error) 4668bc27027aSMiklos Szeredi goto out; 4669da1ce067SMiklos Szeredi } 4670da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4671bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4672bc27027aSMiklos Szeredi if (error) 4673bc27027aSMiklos Szeredi goto out; 4674bc27027aSMiklos Szeredi } 4675549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4676520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4677bc27027aSMiklos Szeredi if (error) 4678bc27027aSMiklos Szeredi goto out; 4679bc27027aSMiklos Szeredi 4680da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 46818767712fSAl Viro if (is_dir) { 46828767712fSAl Viro shrink_dcache_parent(new_dentry); 4683bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 46848767712fSAl Viro } 4685bc27027aSMiklos Szeredi dont_mount(new_dentry); 46868ed936b5SEric W. Biederman detach_mounts(new_dentry); 4687bc27027aSMiklos Szeredi } 4688da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4689da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4690bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4691da1ce067SMiklos Szeredi else 4692da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4693da1ce067SMiklos Szeredi } 4694bc27027aSMiklos Szeredi out: 4695da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4696bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4697bc27027aSMiklos Szeredi else if (target) 46985955102cSAl Viro inode_unlock(target); 4699bc27027aSMiklos Szeredi dput(new_dentry); 4700da1ce067SMiklos Szeredi if (!error) { 4701f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4702da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4703da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4704f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4705da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4706da1ce067SMiklos Szeredi } 4707da1ce067SMiklos Szeredi } 470849d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 47090eeca283SRobert Love 47101da177e4SLinus Torvalds return error; 47111da177e4SLinus Torvalds } 47124d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 47131da177e4SLinus Torvalds 4714e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4715e886663cSJens Axboe struct filename *to, unsigned int flags) 47161da177e4SLinus Torvalds { 47179fe61450SChristian Brauner struct renamedata rd; 47181da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 47191da177e4SLinus Torvalds struct dentry *trap; 4720f5beed75SAl Viro struct path old_path, new_path; 4721f5beed75SAl Viro struct qstr old_last, new_last; 4722f5beed75SAl Viro int old_type, new_type; 47238e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4724f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4725c6a94284SJeff Layton bool should_retry = false; 4726e886663cSJens Axboe int error = -EINVAL; 4727520c8b16SMiklos Szeredi 47280d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 47290ee50b47SDmitry Kadashev goto put_names; 4730da1ce067SMiklos Szeredi 47310d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 47320d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 47330ee50b47SDmitry Kadashev goto put_names; 4734520c8b16SMiklos Szeredi 4735f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4736f5beed75SAl Viro target_flags = 0; 4737f5beed75SAl Viro 4738c6a94284SJeff Layton retry: 4739c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4740e886663cSJens Axboe &old_last, &old_type); 47410ee50b47SDmitry Kadashev if (error) 47420ee50b47SDmitry Kadashev goto put_names; 47431da177e4SLinus Torvalds 4744c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4745e886663cSJens Axboe &new_type); 47460ee50b47SDmitry Kadashev if (error) 47471da177e4SLinus Torvalds goto exit1; 47481da177e4SLinus Torvalds 47491da177e4SLinus Torvalds error = -EXDEV; 4750f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 47511da177e4SLinus Torvalds goto exit2; 47521da177e4SLinus Torvalds 47531da177e4SLinus Torvalds error = -EBUSY; 4754f5beed75SAl Viro if (old_type != LAST_NORM) 47551da177e4SLinus Torvalds goto exit2; 47561da177e4SLinus Torvalds 47570a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 47580a7c3937SMiklos Szeredi error = -EEXIST; 4759f5beed75SAl Viro if (new_type != LAST_NORM) 47601da177e4SLinus Torvalds goto exit2; 47611da177e4SLinus Torvalds 4762f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4763c30dabfeSJan Kara if (error) 4764c30dabfeSJan Kara goto exit2; 4765c30dabfeSJan Kara 47668e6d782cSJ. Bruce Fields retry_deleg: 4767f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 47681da177e4SLinus Torvalds 4769f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 47701da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 47711da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 47721da177e4SLinus Torvalds goto exit3; 47731da177e4SLinus Torvalds /* source must exist */ 47741da177e4SLinus Torvalds error = -ENOENT; 4775b18825a7SDavid Howells if (d_is_negative(old_dentry)) 47761da177e4SLinus Torvalds goto exit4; 4777f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 47781da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 47791da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 47801da177e4SLinus Torvalds goto exit4; 47810a7c3937SMiklos Szeredi error = -EEXIST; 47820a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 47830a7c3937SMiklos Szeredi goto exit5; 4784da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4785da1ce067SMiklos Szeredi error = -ENOENT; 4786da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4787da1ce067SMiklos Szeredi goto exit5; 4788da1ce067SMiklos Szeredi 4789da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4790da1ce067SMiklos Szeredi error = -ENOTDIR; 4791f5beed75SAl Viro if (new_last.name[new_last.len]) 4792da1ce067SMiklos Szeredi goto exit5; 4793da1ce067SMiklos Szeredi } 4794da1ce067SMiklos Szeredi } 47950a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 47960a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 47970a7c3937SMiklos Szeredi error = -ENOTDIR; 4798f5beed75SAl Viro if (old_last.name[old_last.len]) 47990a7c3937SMiklos Szeredi goto exit5; 4800f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 48010a7c3937SMiklos Szeredi goto exit5; 48020a7c3937SMiklos Szeredi } 48030a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 48040a7c3937SMiklos Szeredi error = -EINVAL; 48050a7c3937SMiklos Szeredi if (old_dentry == trap) 48060a7c3937SMiklos Szeredi goto exit5; 48071da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4808da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 48091da177e4SLinus Torvalds error = -ENOTEMPTY; 48101da177e4SLinus Torvalds if (new_dentry == trap) 48111da177e4SLinus Torvalds goto exit5; 48121da177e4SLinus Torvalds 4813f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4814f5beed75SAl Viro &new_path, new_dentry, flags); 4815be6d3e56SKentaro Takeda if (error) 4816c30dabfeSJan Kara goto exit5; 48179fe61450SChristian Brauner 48189fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 48199fe61450SChristian Brauner rd.old_dentry = old_dentry; 48206521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 48219fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 48229fe61450SChristian Brauner rd.new_dentry = new_dentry; 48236521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 48249fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 48259fe61450SChristian Brauner rd.flags = flags; 48269fe61450SChristian Brauner error = vfs_rename(&rd); 48271da177e4SLinus Torvalds exit5: 48281da177e4SLinus Torvalds dput(new_dentry); 48291da177e4SLinus Torvalds exit4: 48301da177e4SLinus Torvalds dput(old_dentry); 48311da177e4SLinus Torvalds exit3: 4832f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 48338e6d782cSJ. Bruce Fields if (delegated_inode) { 48348e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 48358e6d782cSJ. Bruce Fields if (!error) 48368e6d782cSJ. Bruce Fields goto retry_deleg; 48378e6d782cSJ. Bruce Fields } 4838f5beed75SAl Viro mnt_drop_write(old_path.mnt); 48391da177e4SLinus Torvalds exit2: 4840c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4841c6a94284SJeff Layton should_retry = true; 4842f5beed75SAl Viro path_put(&new_path); 48431da177e4SLinus Torvalds exit1: 4844f5beed75SAl Viro path_put(&old_path); 4845c6a94284SJeff Layton if (should_retry) { 4846c6a94284SJeff Layton should_retry = false; 4847c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4848c6a94284SJeff Layton goto retry; 4849c6a94284SJeff Layton } 48500ee50b47SDmitry Kadashev put_names: 4851e886663cSJens Axboe putname(from); 4852e886663cSJens Axboe putname(to); 48531da177e4SLinus Torvalds return error; 48541da177e4SLinus Torvalds } 48551da177e4SLinus Torvalds 4856ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4857ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4858ee81feb6SDominik Brodowski { 4859e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4860e886663cSJens Axboe flags); 4861ee81feb6SDominik Brodowski } 4862ee81feb6SDominik Brodowski 4863520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4864520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4865520c8b16SMiklos Szeredi { 4866e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4867e886663cSJens Axboe 0); 4868520c8b16SMiklos Szeredi } 4869520c8b16SMiklos Szeredi 4870a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 48715590ff0dSUlrich Drepper { 4872e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4873e886663cSJens Axboe getname(newname), 0); 48745590ff0dSUlrich Drepper } 48755590ff0dSUlrich Drepper 48765d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 48771da177e4SLinus Torvalds { 48785d826c84SAl Viro int len = PTR_ERR(link); 48791da177e4SLinus Torvalds if (IS_ERR(link)) 48801da177e4SLinus Torvalds goto out; 48811da177e4SLinus Torvalds 48821da177e4SLinus Torvalds len = strlen(link); 48831da177e4SLinus Torvalds if (len > (unsigned) buflen) 48841da177e4SLinus Torvalds len = buflen; 48851da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 48861da177e4SLinus Torvalds len = -EFAULT; 48871da177e4SLinus Torvalds out: 48881da177e4SLinus Torvalds return len; 48891da177e4SLinus Torvalds } 48901da177e4SLinus Torvalds 4891d60874cdSMiklos Szeredi /** 4892fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4893fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4894fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4895fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4896fd4a0edfSMiklos Szeredi * 4897fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4898fd4a0edfSMiklos Szeredi * 4899fd4a0edfSMiklos Szeredi * Does not call security hook. 4900fd4a0edfSMiklos Szeredi */ 4901fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4902fd4a0edfSMiklos Szeredi { 4903fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4904f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4905f2df5da6SAl Viro const char *link; 4906f2df5da6SAl Viro int res; 4907fd4a0edfSMiklos Szeredi 490876fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 490976fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 491076fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 491176fca90eSMiklos Szeredi 491276fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4913fd4a0edfSMiklos Szeredi return -EINVAL; 4914fd4a0edfSMiklos Szeredi 491576fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 491676fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 491776fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 491876fca90eSMiklos Szeredi } 491976fca90eSMiklos Szeredi 49204c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4921f2df5da6SAl Viro if (!link) { 4922f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4923f2df5da6SAl Viro if (IS_ERR(link)) 4924f2df5da6SAl Viro return PTR_ERR(link); 4925f2df5da6SAl Viro } 4926f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4927f2df5da6SAl Viro do_delayed_call(&done); 4928f2df5da6SAl Viro return res; 4929fd4a0edfSMiklos Szeredi } 4930fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 49311da177e4SLinus Torvalds 4932d60874cdSMiklos Szeredi /** 4933d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4934d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4935d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4936d60874cdSMiklos Szeredi * 4937d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4938d60874cdSMiklos Szeredi * 4939d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4940d60874cdSMiklos Szeredi * 4941d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4942d60874cdSMiklos Szeredi */ 4943d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4944d60874cdSMiklos Szeredi { 4945d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4946d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4947d60874cdSMiklos Szeredi 4948d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4949d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4950d60874cdSMiklos Szeredi if (!res) 4951d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4952d60874cdSMiklos Szeredi } 4953d60874cdSMiklos Szeredi return res; 4954d60874cdSMiklos Szeredi } 4955d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4956d60874cdSMiklos Szeredi 49571da177e4SLinus Torvalds /* get the link contents into pagecache */ 49586b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4959fceef393SAl Viro struct delayed_call *callback) 49601da177e4SLinus Torvalds { 4961ebd09abbSDuane Griffin char *kaddr; 49621da177e4SLinus Torvalds struct page *page; 49636b255391SAl Viro struct address_space *mapping = inode->i_mapping; 49646b255391SAl Viro 4965d3883d4fSAl Viro if (!dentry) { 4966d3883d4fSAl Viro page = find_get_page(mapping, 0); 4967d3883d4fSAl Viro if (!page) 49686b255391SAl Viro return ERR_PTR(-ECHILD); 4969d3883d4fSAl Viro if (!PageUptodate(page)) { 4970d3883d4fSAl Viro put_page(page); 4971d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4972d3883d4fSAl Viro } 4973d3883d4fSAl Viro } else { 4974090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 49751da177e4SLinus Torvalds if (IS_ERR(page)) 49766fe6900eSNick Piggin return (char*)page; 4977d3883d4fSAl Viro } 4978fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 497921fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 498021fc61c7SAl Viro kaddr = page_address(page); 49816b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4982ebd09abbSDuane Griffin return kaddr; 49831da177e4SLinus Torvalds } 49841da177e4SLinus Torvalds 49856b255391SAl Viro EXPORT_SYMBOL(page_get_link); 49861da177e4SLinus Torvalds 4987fceef393SAl Viro void page_put_link(void *arg) 49881da177e4SLinus Torvalds { 4989fceef393SAl Viro put_page(arg); 49901da177e4SLinus Torvalds } 49914d359507SAl Viro EXPORT_SYMBOL(page_put_link); 49921da177e4SLinus Torvalds 4993aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4994aa80deabSAl Viro { 4995fceef393SAl Viro DEFINE_DELAYED_CALL(done); 49966b255391SAl Viro int res = readlink_copy(buffer, buflen, 49976b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4998fceef393SAl Viro &done)); 4999fceef393SAl Viro do_delayed_call(&done); 5000aa80deabSAl Viro return res; 5001aa80deabSAl Viro } 5002aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 5003aa80deabSAl Viro 500454566b2cSNick Piggin /* 500554566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 500654566b2cSNick Piggin */ 500754566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 50081da177e4SLinus Torvalds { 50091da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 50100adb25d2SKirill Korotaev struct page *page; 5011afddba49SNick Piggin void *fsdata; 5012beb497abSDmitriy Monakhov int err; 5013c718a975STetsuo Handa unsigned int flags = 0; 501454566b2cSNick Piggin if (nofs) 501554566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 50161da177e4SLinus Torvalds 50177e53cac4SNeilBrown retry: 5018afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 501954566b2cSNick Piggin flags, &page, &fsdata); 50201da177e4SLinus Torvalds if (err) 5021afddba49SNick Piggin goto fail; 5022afddba49SNick Piggin 502321fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5024afddba49SNick Piggin 5025afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 5026afddba49SNick Piggin page, fsdata); 50271da177e4SLinus Torvalds if (err < 0) 50281da177e4SLinus Torvalds goto fail; 5029afddba49SNick Piggin if (err < len-1) 5030afddba49SNick Piggin goto retry; 5031afddba49SNick Piggin 50321da177e4SLinus Torvalds mark_inode_dirty(inode); 50331da177e4SLinus Torvalds return 0; 50341da177e4SLinus Torvalds fail: 50351da177e4SLinus Torvalds return err; 50361da177e4SLinus Torvalds } 50374d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 50381da177e4SLinus Torvalds 50390adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 50400adb25d2SKirill Korotaev { 50410adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 5042c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 50430adb25d2SKirill Korotaev } 50444d359507SAl Viro EXPORT_SYMBOL(page_symlink); 50450adb25d2SKirill Korotaev 504692e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 50476b255391SAl Viro .get_link = page_get_link, 50481da177e4SLinus Torvalds }; 50491da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5050