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 /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 733bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 734ee594bffSAl Viro return true; 735bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 736ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 737ee594bffSAl Viro } 738ee594bffSAl Viro 73919660af7SAl Viro /* 74031e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 74119660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 74219660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 74357e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 74419660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 74519660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 74619660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 74719660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 74831e6b01fSNick Piggin */ 74931e6b01fSNick Piggin 75031e6b01fSNick Piggin /** 751e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 75219660af7SAl Viro * @nd: nameidata pathwalk data 753e36cffedSJens Axboe * Returns: true on success, false on failure 75431e6b01fSNick Piggin * 755e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7564675ac39SAl Viro * for ref-walk mode. 7574675ac39SAl Viro * Must be called from rcu-walk context. 758e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7597973387aSAl Viro * terminate_walk(). 76031e6b01fSNick Piggin */ 761e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 76231e6b01fSNick Piggin { 76331e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 76431e6b01fSNick Piggin 76531e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 766e5c832d5SLinus Torvalds 767e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 7687973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7694675ac39SAl Viro goto out1; 77084a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 77184a2bd39SAl Viro goto out; 772ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7734675ac39SAl Viro goto out; 7744675ac39SAl Viro rcu_read_unlock(); 7754675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 776e36cffedSJens Axboe return true; 7774675ac39SAl Viro 77884a2bd39SAl Viro out1: 7794675ac39SAl Viro nd->path.mnt = NULL; 7804675ac39SAl Viro nd->path.dentry = NULL; 7814675ac39SAl Viro out: 7824675ac39SAl Viro rcu_read_unlock(); 783e36cffedSJens Axboe return false; 7844675ac39SAl Viro } 7854675ac39SAl Viro 7864675ac39SAl Viro /** 787ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7884675ac39SAl Viro * @nd: nameidata pathwalk data 789ae66db45SAl Viro * @dentry: next dentry to step into 790ae66db45SAl Viro * @seq: seq number to check @dentry against 791ae66db45SAl Viro * Returns: true on success, false on failure 7924675ac39SAl Viro * 793*30476f7eSTom Rix * Similar to try_to_unlazy(), but here we have the next dentry already 794ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 795ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 796ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 7974675ac39SAl Viro * terminate_walk(). 7984675ac39SAl Viro */ 799ae66db45SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq) 8004675ac39SAl Viro { 8014675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8024675ac39SAl Viro 8034675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 8044675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8054675ac39SAl Viro goto out2; 8067973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 8077973387aSAl Viro goto out2; 8084675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8097973387aSAl Viro goto out1; 81048a066e7SAl Viro 81115570086SLinus Torvalds /* 8124675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8134675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8144675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8154675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8164675ac39SAl Viro * know the parent must still be valid if the child sequence number is 81715570086SLinus Torvalds */ 8184675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 819e5c832d5SLinus Torvalds goto out; 82084a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 82184a2bd39SAl Viro goto out_dput; 822e5c832d5SLinus Torvalds /* 823e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 824e5c832d5SLinus Torvalds * still valid and get it if required. 825e5c832d5SLinus Torvalds */ 82684a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 82784a2bd39SAl Viro goto out_dput; 8288b61e74fSAl Viro rcu_read_unlock(); 829ae66db45SAl Viro return true; 83019660af7SAl Viro 8317973387aSAl Viro out2: 8327973387aSAl Viro nd->path.mnt = NULL; 8337973387aSAl Viro out1: 8347973387aSAl Viro nd->path.dentry = NULL; 835e5c832d5SLinus Torvalds out: 8368b61e74fSAl Viro rcu_read_unlock(); 837ae66db45SAl Viro return false; 83884a2bd39SAl Viro out_dput: 83984a2bd39SAl Viro rcu_read_unlock(); 84084a2bd39SAl Viro dput(dentry); 841ae66db45SAl Viro return false; 84231e6b01fSNick Piggin } 84331e6b01fSNick Piggin 8444ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 84534286d66SNick Piggin { 846a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8474ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 848a89f8337SAl Viro else 849a89f8337SAl Viro return 1; 85034286d66SNick Piggin } 85134286d66SNick Piggin 8529f1fafeeSAl Viro /** 8539f1fafeeSAl Viro * complete_walk - successful completion of path walk 8549f1fafeeSAl Viro * @nd: pointer nameidata 85539159de2SJeff Layton * 8569f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8579f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8589f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8599f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8609f1fafeeSAl Viro * need to drop nd->path. 86139159de2SJeff Layton */ 8629f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 86339159de2SJeff Layton { 86416c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 86539159de2SJeff Layton int status; 86639159de2SJeff Layton 8679f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 868adb21d2bSAleksa Sarai /* 869adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 870adb21d2bSAleksa Sarai * externally-managed nd->root. 871adb21d2bSAleksa Sarai */ 872bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 873bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8749f1fafeeSAl Viro nd->root.mnt = NULL; 8756c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 876e36cffedSJens Axboe if (!try_to_unlazy(nd)) 87748a066e7SAl Viro return -ECHILD; 87848a066e7SAl Viro } 8799f1fafeeSAl Viro 880adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 881adb21d2bSAleksa Sarai /* 882adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 883adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 884adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 885adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 886adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 887adb21d2bSAleksa Sarai * 888adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 889adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 890adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 891adb21d2bSAleksa Sarai * requested root to userspace. 892adb21d2bSAleksa Sarai * 893adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 894adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 895adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 896adb21d2bSAleksa Sarai */ 897adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 898adb21d2bSAleksa Sarai return -EXDEV; 899adb21d2bSAleksa Sarai } 900adb21d2bSAleksa Sarai 901bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 90239159de2SJeff Layton return 0; 90339159de2SJeff Layton 904ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 90516c2cd71SAl Viro return 0; 90616c2cd71SAl Viro 907ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 90839159de2SJeff Layton if (status > 0) 90939159de2SJeff Layton return 0; 91039159de2SJeff Layton 91116c2cd71SAl Viro if (!status) 91239159de2SJeff Layton status = -ESTALE; 91316c2cd71SAl Viro 91439159de2SJeff Layton return status; 91539159de2SJeff Layton } 91639159de2SJeff Layton 917740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9182a737871SAl Viro { 91931e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9209e6697e2SAl Viro 921adb21d2bSAleksa Sarai /* 922adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 923adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 924adb21d2bSAleksa Sarai * from the dirfd. 925adb21d2bSAleksa Sarai */ 926adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 927adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 928adb21d2bSAleksa Sarai 9299e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9308f47a016SAl Viro unsigned seq; 931c28cc364SNick Piggin 932c28cc364SNick Piggin do { 933c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 93431e6b01fSNick Piggin nd->root = fs->root; 9358f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 936c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9379e6697e2SAl Viro } else { 9389e6697e2SAl Viro get_fs_root(fs, &nd->root); 939bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9409e6697e2SAl Viro } 941740a1678SAleksa Sarai return 0; 94231e6b01fSNick Piggin } 94331e6b01fSNick Piggin 944248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 945248fb5b9SAl Viro { 946adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 947adb21d2bSAleksa Sarai return -EXDEV; 94872ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 94972ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 95072ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 95172ba2929SAleksa Sarai return -EXDEV; 95272ba2929SAleksa Sarai } 953740a1678SAleksa Sarai if (!nd->root.mnt) { 954740a1678SAleksa Sarai int error = set_root(nd); 955740a1678SAleksa Sarai if (error) 956740a1678SAleksa Sarai return error; 957740a1678SAleksa Sarai } 958248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 959248fb5b9SAl Viro struct dentry *d; 960248fb5b9SAl Viro nd->path = nd->root; 961248fb5b9SAl Viro d = nd->path.dentry; 962248fb5b9SAl Viro nd->inode = d->d_inode; 963248fb5b9SAl Viro nd->seq = nd->root_seq; 964248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 965248fb5b9SAl Viro return -ECHILD; 966248fb5b9SAl Viro } else { 967248fb5b9SAl Viro path_put(&nd->path); 968248fb5b9SAl Viro nd->path = nd->root; 969248fb5b9SAl Viro path_get(&nd->path); 970248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 971248fb5b9SAl Viro } 972bcba1e7dSAl Viro nd->state |= ND_JUMPED; 973248fb5b9SAl Viro return 0; 974248fb5b9SAl Viro } 975248fb5b9SAl Viro 976b5fb63c1SChristoph Hellwig /* 9776b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 978b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 979b5fb63c1SChristoph Hellwig */ 9801bc82070SAleksa Sarai int nd_jump_link(struct path *path) 981b5fb63c1SChristoph Hellwig { 9824b99d499SAleksa Sarai int error = -ELOOP; 9836e77137bSAl Viro struct nameidata *nd = current->nameidata; 984b5fb63c1SChristoph Hellwig 9854b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9864b99d499SAleksa Sarai goto err; 9874b99d499SAleksa Sarai 98872ba2929SAleksa Sarai error = -EXDEV; 98972ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 99072ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 99172ba2929SAleksa Sarai goto err; 99272ba2929SAleksa Sarai } 993adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 994adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 995adb21d2bSAleksa Sarai goto err; 99672ba2929SAleksa Sarai 9974b99d499SAleksa Sarai path_put(&nd->path); 998b5fb63c1SChristoph Hellwig nd->path = *path; 999b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1000bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10011bc82070SAleksa Sarai return 0; 10024b99d499SAleksa Sarai 10034b99d499SAleksa Sarai err: 10044b99d499SAleksa Sarai path_put(path); 10054b99d499SAleksa Sarai return error; 1006b5fb63c1SChristoph Hellwig } 1007b5fb63c1SChristoph Hellwig 1008b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1009574197e0SAl Viro { 101021c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1011fceef393SAl Viro do_delayed_call(&last->done); 10126548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1013b9ff4429SAl Viro path_put(&last->link); 1014574197e0SAl Viro } 1015574197e0SAl Viro 10169c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10179c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10189c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10199c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10209c011be1SLuis Chamberlain 10219c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10229c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10239c011be1SLuis Chamberlain { 10249c011be1SLuis Chamberlain .procname = "protected_symlinks", 10259c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10269c011be1SLuis Chamberlain .maxlen = sizeof(int), 10279c011be1SLuis Chamberlain .mode = 0600, 10289c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10299c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10309c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10319c011be1SLuis Chamberlain }, 10329c011be1SLuis Chamberlain { 10339c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10349c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10359c011be1SLuis Chamberlain .maxlen = sizeof(int), 10369c011be1SLuis Chamberlain .mode = 0600, 10379c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10389c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10399c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10409c011be1SLuis Chamberlain }, 10419c011be1SLuis Chamberlain { 10429c011be1SLuis Chamberlain .procname = "protected_fifos", 10439c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10449c011be1SLuis Chamberlain .maxlen = sizeof(int), 10459c011be1SLuis Chamberlain .mode = 0600, 10469c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10479c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10489c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10499c011be1SLuis Chamberlain }, 10509c011be1SLuis Chamberlain { 10519c011be1SLuis Chamberlain .procname = "protected_regular", 10529c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10539c011be1SLuis Chamberlain .maxlen = sizeof(int), 10549c011be1SLuis Chamberlain .mode = 0600, 10559c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10569c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10579c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10589c011be1SLuis Chamberlain }, 10599c011be1SLuis Chamberlain { } 10609c011be1SLuis Chamberlain }; 10619c011be1SLuis Chamberlain 10629c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10639c011be1SLuis Chamberlain { 10649c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10659c011be1SLuis Chamberlain return 0; 10669c011be1SLuis Chamberlain } 10679c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10689c011be1SLuis Chamberlain 10699c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1070800179c9SKees Cook 1071800179c9SKees Cook /** 1072800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 107355852635SRandy Dunlap * @nd: nameidata pathwalk data 1074800179c9SKees Cook * 1075800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1076800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1077800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1078800179c9SKees Cook * processes from failing races against path names that may change out 1079800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1080800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1081800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1082800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1083800179c9SKees Cook * 1084800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1085800179c9SKees Cook */ 1086ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1087800179c9SKees Cook { 1088ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1089ba73d987SChristian Brauner kuid_t i_uid; 1090ba73d987SChristian Brauner 1091800179c9SKees Cook if (!sysctl_protected_symlinks) 1092800179c9SKees Cook return 0; 1093800179c9SKees Cook 1094ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1095ba73d987SChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 1096800179c9SKees Cook /* Allowed if owner and follower match. */ 1097ba73d987SChristian Brauner if (uid_eq(current_cred()->fsuid, i_uid)) 1098800179c9SKees Cook return 0; 1099800179c9SKees Cook 1100800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11010f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1102800179c9SKees Cook return 0; 1103800179c9SKees Cook 1104800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1105ba73d987SChristian Brauner if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) 1106800179c9SKees Cook return 0; 1107800179c9SKees Cook 110831956502SAl Viro if (nd->flags & LOOKUP_RCU) 110931956502SAl Viro return -ECHILD; 111031956502SAl Viro 1111ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1112245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1113800179c9SKees Cook return -EACCES; 1114800179c9SKees Cook } 1115800179c9SKees Cook 1116800179c9SKees Cook /** 1117800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1118ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1119800179c9SKees Cook * @inode: the source inode to hardlink from 1120800179c9SKees Cook * 1121800179c9SKees Cook * Return false if at least one of the following conditions: 1122800179c9SKees Cook * - inode is not a regular file 1123800179c9SKees Cook * - inode is setuid 1124800179c9SKees Cook * - inode is setgid and group-exec 1125800179c9SKees Cook * - access failure for read and write 1126800179c9SKees Cook * 1127800179c9SKees Cook * Otherwise returns true. 1128800179c9SKees Cook */ 1129ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1130ba73d987SChristian Brauner struct inode *inode) 1131800179c9SKees Cook { 1132800179c9SKees Cook umode_t mode = inode->i_mode; 1133800179c9SKees Cook 1134800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1135800179c9SKees Cook if (!S_ISREG(mode)) 1136800179c9SKees Cook return false; 1137800179c9SKees Cook 1138800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1139800179c9SKees Cook if (mode & S_ISUID) 1140800179c9SKees Cook return false; 1141800179c9SKees Cook 1142800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1143800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1144800179c9SKees Cook return false; 1145800179c9SKees Cook 1146800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1147ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1148800179c9SKees Cook return false; 1149800179c9SKees Cook 1150800179c9SKees Cook return true; 1151800179c9SKees Cook } 1152800179c9SKees Cook 1153800179c9SKees Cook /** 1154800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1155ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1156800179c9SKees Cook * @link: the source to hardlink from 1157800179c9SKees Cook * 1158800179c9SKees Cook * Block hardlink when all of: 1159800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1160800179c9SKees Cook * - fsuid does not match inode 1161800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1162f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1163800179c9SKees Cook * 1164ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1165ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1166ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1167ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1168ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1169ba73d987SChristian Brauner * 1170800179c9SKees Cook * Returns 0 if successful, -ve on error. 1171800179c9SKees Cook */ 1172ba73d987SChristian Brauner int may_linkat(struct user_namespace *mnt_userns, struct path *link) 1173800179c9SKees Cook { 1174593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1175593d1ce8SEric W. Biederman 1176593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1177ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 1178ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 1179593d1ce8SEric W. Biederman return -EOVERFLOW; 1180800179c9SKees Cook 1181800179c9SKees Cook if (!sysctl_protected_hardlinks) 1182800179c9SKees Cook return 0; 1183800179c9SKees Cook 1184800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1185800179c9SKees Cook * otherwise, it must be a safe source. 1186800179c9SKees Cook */ 1187ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1188ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1189800179c9SKees Cook return 0; 1190800179c9SKees Cook 1191245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1192800179c9SKees Cook return -EPERM; 1193800179c9SKees Cook } 1194800179c9SKees Cook 119530aba665SSalvatore Mesoraca /** 119630aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 119730aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 119830aba665SSalvatore Mesoraca * exist. 1199ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 12002111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 120130aba665SSalvatore Mesoraca * @inode: the inode of the file to open 120230aba665SSalvatore Mesoraca * 120330aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 120430aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 120530aba665SSalvatore Mesoraca * - the file already exists 120630aba665SSalvatore Mesoraca * - we are in a sticky directory 120730aba665SSalvatore Mesoraca * - we don't own the file 120830aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 120930aba665SSalvatore Mesoraca * - the directory is world writable 121030aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 121130aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 121230aba665SSalvatore Mesoraca * be enough. 121330aba665SSalvatore Mesoraca * 1214ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1215ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1216ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1217ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1218ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1219ba73d987SChristian Brauner * 122030aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 122130aba665SSalvatore Mesoraca */ 1222ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1223ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 122430aba665SSalvatore Mesoraca { 1225ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1226ba73d987SChristian Brauner kuid_t dir_uid = nd->dir_uid; 1227ba73d987SChristian Brauner 122830aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 122930aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1230d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1231ba73d987SChristian Brauner uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || 1232ba73d987SChristian Brauner uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) 123330aba665SSalvatore Mesoraca return 0; 123430aba665SSalvatore Mesoraca 1235d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1236d0cb5018SAl Viro (dir_mode & 0020 && 123730aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 123830aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1239245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1240245d7369SKees Cook "sticky_create_fifo" : 1241245d7369SKees Cook "sticky_create_regular"; 1242245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 124330aba665SSalvatore Mesoraca return -EACCES; 124430aba665SSalvatore Mesoraca } 124530aba665SSalvatore Mesoraca return 0; 124630aba665SSalvatore Mesoraca } 124730aba665SSalvatore Mesoraca 1248f015f126SDavid Howells /* 1249f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1250f015f126SDavid Howells * 1251f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1252f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1253f015f126SDavid Howells * Up is towards /. 1254f015f126SDavid Howells * 1255f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1256f015f126SDavid Howells * root. 1257f015f126SDavid Howells */ 1258bab77ebfSAl Viro int follow_up(struct path *path) 12591da177e4SLinus Torvalds { 12600714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12610714a533SAl Viro struct mount *parent; 12621da177e4SLinus Torvalds struct dentry *mountpoint; 126399b7db7bSNick Piggin 126448a066e7SAl Viro read_seqlock_excl(&mount_lock); 12650714a533SAl Viro parent = mnt->mnt_parent; 12663c0a6163SAl Viro if (parent == mnt) { 126748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12681da177e4SLinus Torvalds return 0; 12691da177e4SLinus Torvalds } 12700714a533SAl Viro mntget(&parent->mnt); 1271a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 127248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1273bab77ebfSAl Viro dput(path->dentry); 1274bab77ebfSAl Viro path->dentry = mountpoint; 1275bab77ebfSAl Viro mntput(path->mnt); 12760714a533SAl Viro path->mnt = &parent->mnt; 12771da177e4SLinus Torvalds return 1; 12781da177e4SLinus Torvalds } 12794d359507SAl Viro EXPORT_SYMBOL(follow_up); 12801da177e4SLinus Torvalds 12817ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12827ef482faSAl Viro struct path *path, unsigned *seqp) 12837ef482faSAl Viro { 12847ef482faSAl Viro while (mnt_has_parent(m)) { 12857ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12867ef482faSAl Viro 12877ef482faSAl Viro m = m->mnt_parent; 12887ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12897ef482faSAl Viro root->mnt == &m->mnt)) 12907ef482faSAl Viro break; 12917ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 12927ef482faSAl Viro path->mnt = &m->mnt; 12937ef482faSAl Viro path->dentry = mountpoint; 12947ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 12957ef482faSAl Viro return true; 12967ef482faSAl Viro } 12977ef482faSAl Viro } 12987ef482faSAl Viro return false; 12997ef482faSAl Viro } 13007ef482faSAl Viro 13012aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 13022aa38470SAl Viro struct path *path) 13032aa38470SAl Viro { 13042aa38470SAl Viro bool found; 13052aa38470SAl Viro 13062aa38470SAl Viro rcu_read_lock(); 13072aa38470SAl Viro while (1) { 13082aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 13092aa38470SAl Viro 13102aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 13112aa38470SAl Viro if (unlikely(!found)) { 13122aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 13132aa38470SAl Viro break; 13142aa38470SAl Viro } else { 13152aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 13162aa38470SAl Viro break; 13172aa38470SAl Viro rcu_read_unlock(); 13182aa38470SAl Viro path_put(path); 13192aa38470SAl Viro rcu_read_lock(); 13202aa38470SAl Viro } 13212aa38470SAl Viro } 13222aa38470SAl Viro rcu_read_unlock(); 13232aa38470SAl Viro return found; 13242aa38470SAl Viro } 13252aa38470SAl Viro 1326b5c84bf6SNick Piggin /* 13279875cf80SDavid Howells * Perform an automount 13289875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 13299875cf80SDavid Howells * were called with. 13301da177e4SLinus Torvalds */ 13311c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 133231e6b01fSNick Piggin { 133325e195aaSAl Viro struct dentry *dentry = path->dentry; 13349875cf80SDavid Howells 13350ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 13360ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 13370ec26fd0SMiklos Szeredi * the name. 13380ec26fd0SMiklos Szeredi * 13390ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 13405a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 13410ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 13420ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13430ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13440ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13455a30d8a2SDavid Howells */ 13461c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13475d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 134825e195aaSAl Viro dentry->d_inode) 13499875cf80SDavid Howells return -EISDIR; 13500ec26fd0SMiklos Szeredi 13511c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13529875cf80SDavid Howells return -ELOOP; 13539875cf80SDavid Howells 135425e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1355ea5b778aSDavid Howells } 13569875cf80SDavid Howells 13579875cf80SDavid Howells /* 13589deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13599deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13609deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13619deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13629875cf80SDavid Howells */ 13639deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13649deed3ebSAl Viro int *count, unsigned lookup_flags) 13659875cf80SDavid Howells { 13669deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13679875cf80SDavid Howells bool need_mntput = false; 13688aef1884SAl Viro int ret = 0; 13699875cf80SDavid Howells 13709deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1371cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1372cc53ce53SDavid Howells * being held. */ 1373d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1374fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1375508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1376cc53ce53SDavid Howells if (ret < 0) 13778aef1884SAl Viro break; 1378cc53ce53SDavid Howells } 1379cc53ce53SDavid Howells 13809deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13819875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13829deed3ebSAl Viro if (mounted) { // ... in our namespace 13839875cf80SDavid Howells dput(path->dentry); 13849875cf80SDavid Howells if (need_mntput) 1385463ffb2eSAl Viro mntput(path->mnt); 1386463ffb2eSAl Viro path->mnt = mounted; 1387463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13889deed3ebSAl Viro // here we know it's positive 13899deed3ebSAl Viro flags = path->dentry->d_flags; 13909875cf80SDavid Howells need_mntput = true; 13919875cf80SDavid Howells continue; 1392463ffb2eSAl Viro } 13931da177e4SLinus Torvalds } 13949875cf80SDavid Howells 13959deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 13969deed3ebSAl Viro break; 13979deed3ebSAl Viro 13989deed3ebSAl Viro // uncovered automount point 13999deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 14009deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 14019875cf80SDavid Howells if (ret < 0) 14028aef1884SAl Viro break; 14039875cf80SDavid Howells } 14049875cf80SDavid Howells 14059deed3ebSAl Viro if (ret == -EISDIR) 14069deed3ebSAl Viro ret = 0; 14079deed3ebSAl Viro // possible if you race with several mount --move 14089deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 14098aef1884SAl Viro mntput(path->mnt); 14109deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1411d41efb52SAl Viro ret = -ENOENT; 14129deed3ebSAl Viro *jumped = need_mntput; 14138402752eSAl Viro return ret; 14141da177e4SLinus Torvalds } 14151da177e4SLinus Torvalds 14169deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 14179deed3ebSAl Viro int *count, unsigned lookup_flags) 14189deed3ebSAl Viro { 14199deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 14209deed3ebSAl Viro 14219deed3ebSAl Viro /* fastpath */ 14229deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 14239deed3ebSAl Viro *jumped = false; 14249deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 14259deed3ebSAl Viro return -ENOENT; 14269deed3ebSAl Viro return 0; 14279deed3ebSAl Viro } 14289deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 14299deed3ebSAl Viro } 14309deed3ebSAl Viro 1431cc53ce53SDavid Howells int follow_down_one(struct path *path) 14321da177e4SLinus Torvalds { 14331da177e4SLinus Torvalds struct vfsmount *mounted; 14341da177e4SLinus Torvalds 14351c755af4SAl Viro mounted = lookup_mnt(path); 14361da177e4SLinus Torvalds if (mounted) { 14379393bd07SAl Viro dput(path->dentry); 14389393bd07SAl Viro mntput(path->mnt); 14399393bd07SAl Viro path->mnt = mounted; 14409393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 14411da177e4SLinus Torvalds return 1; 14421da177e4SLinus Torvalds } 14431da177e4SLinus Torvalds return 0; 14441da177e4SLinus Torvalds } 14454d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14461da177e4SLinus Torvalds 14479875cf80SDavid Howells /* 14489deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14499deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14509deed3ebSAl Viro * caller is permitted to proceed or not. 14519deed3ebSAl Viro */ 14529deed3ebSAl Viro int follow_down(struct path *path) 14539deed3ebSAl Viro { 14549deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14559deed3ebSAl Viro bool jumped; 14569deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 14579deed3ebSAl Viro 14589deed3ebSAl Viro if (path->mnt != mnt) 14599deed3ebSAl Viro mntput(mnt); 14609deed3ebSAl Viro return ret; 14619deed3ebSAl Viro } 14629deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14639deed3ebSAl Viro 14649deed3ebSAl Viro /* 1465287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1466287548e4SAl Viro * we meet a managed dentry that would need blocking. 14679875cf80SDavid Howells */ 14689875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1469254cf582SAl Viro struct inode **inode, unsigned *seqp) 14709875cf80SDavid Howells { 1471ea936aebSAl Viro struct dentry *dentry = path->dentry; 1472ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1473ea936aebSAl Viro 1474ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1475ea936aebSAl Viro return true; 1476ea936aebSAl Viro 1477ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1478ea936aebSAl Viro return false; 1479ea936aebSAl Viro 148062a7375eSIan Kent for (;;) { 148162a7375eSIan Kent /* 148262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 148362a7375eSIan Kent * that wants to block transit. 148462a7375eSIan Kent */ 1485ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1486ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1487ea936aebSAl Viro if (res) 1488ea936aebSAl Viro return res == -EISDIR; 1489ea936aebSAl Viro flags = dentry->d_flags; 1490b8faf035SNeilBrown } 149162a7375eSIan Kent 1492ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1493ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1494ea936aebSAl Viro if (mounted) { 1495c7105365SAl Viro path->mnt = &mounted->mnt; 1496ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1497bcba1e7dSAl Viro nd->state |= ND_JUMPED; 1498ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1499ea936aebSAl Viro *inode = dentry->d_inode; 150059430262SLinus Torvalds /* 1501ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1502ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1503ea936aebSAl Viro * between mount hash removal and ->mnt_root 1504ea936aebSAl Viro * becoming unpinned. 150559430262SLinus Torvalds */ 1506ea936aebSAl Viro flags = dentry->d_flags; 1507ea936aebSAl Viro continue; 15089875cf80SDavid Howells } 1509ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1510ea936aebSAl Viro return false; 1511ea936aebSAl Viro } 1512ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1513ea936aebSAl Viro } 1514287548e4SAl Viro } 1515287548e4SAl Viro 1516db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1517db3c9adeSAl Viro struct path *path, struct inode **inode, 1518db3c9adeSAl Viro unsigned int *seqp) 1519bd7c4b50SAl Viro { 15209deed3ebSAl Viro bool jumped; 1521db3c9adeSAl Viro int ret; 1522bd7c4b50SAl Viro 1523db3c9adeSAl Viro path->mnt = nd->path.mnt; 1524db3c9adeSAl Viro path->dentry = dentry; 1525c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1526c153007bSAl Viro unsigned int seq = *seqp; 1527c153007bSAl Viro if (unlikely(!*inode)) 1528c153007bSAl Viro return -ENOENT; 1529c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 15309deed3ebSAl Viro return 0; 1531ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 1532c153007bSAl Viro return -ECHILD; 1533c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1534c153007bSAl Viro path->mnt = nd->path.mnt; 1535c153007bSAl Viro path->dentry = dentry; 1536c153007bSAl Viro } 15379deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 15389deed3ebSAl Viro if (jumped) { 15399deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 15409deed3ebSAl Viro ret = -EXDEV; 15419deed3ebSAl Viro else 1542bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15439deed3ebSAl Viro } 15449deed3ebSAl Viro if (unlikely(ret)) { 15459deed3ebSAl Viro dput(path->dentry); 15469deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15479deed3ebSAl Viro mntput(path->mnt); 15489deed3ebSAl Viro } else { 1549bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1550bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1551bd7c4b50SAl Viro } 1552bd7c4b50SAl Viro return ret; 1553bd7c4b50SAl Viro } 1554bd7c4b50SAl Viro 15559875cf80SDavid Howells /* 1556f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1557f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1558baa03890SNick Piggin */ 1559e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1560e3c13928SAl Viro struct dentry *dir, 15616c51e513SAl Viro unsigned int flags) 1562baa03890SNick Piggin { 1563a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1564bad61189SMiklos Szeredi if (dentry) { 1565a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1566bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 156774ff0ffcSAl Viro if (!error) 15685542aa2fSEric W. Biederman d_invalidate(dentry); 1569bad61189SMiklos Szeredi dput(dentry); 157074ff0ffcSAl Viro return ERR_PTR(error); 1571bad61189SMiklos Szeredi } 1572bad61189SMiklos Szeredi } 1573baa03890SNick Piggin return dentry; 1574baa03890SNick Piggin } 1575baa03890SNick Piggin 1576baa03890SNick Piggin /* 1577a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1578a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1579a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1580a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1581a03ece5fSAl Viro * at all. 158244396f4bSJosef Bacik */ 1583a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1584a03ece5fSAl Viro struct dentry *base, unsigned int flags) 158544396f4bSJosef Bacik { 1586a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 158744396f4bSJosef Bacik struct dentry *old; 1588a03ece5fSAl Viro struct inode *dir = base->d_inode; 1589a03ece5fSAl Viro 1590a03ece5fSAl Viro if (dentry) 1591a03ece5fSAl Viro return dentry; 159244396f4bSJosef Bacik 159344396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1594a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 159544396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1596a03ece5fSAl Viro 1597a03ece5fSAl Viro dentry = d_alloc(base, name); 1598a03ece5fSAl Viro if (unlikely(!dentry)) 1599a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 160044396f4bSJosef Bacik 160172bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 160244396f4bSJosef Bacik if (unlikely(old)) { 160344396f4bSJosef Bacik dput(dentry); 160444396f4bSJosef Bacik dentry = old; 160544396f4bSJosef Bacik } 160644396f4bSJosef Bacik return dentry; 160744396f4bSJosef Bacik } 160844396f4bSJosef Bacik 160920e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 161020e34357SAl Viro struct inode **inode, 1611254cf582SAl Viro unsigned *seqp) 16121da177e4SLinus Torvalds { 161331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16145a18fff2SAl Viro int status = 1; 16159875cf80SDavid Howells 16163cac260aSAl Viro /* 1617b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16185d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16195d0f49c1SAl Viro * going to fall back to non-racy lookup. 1620b04f784eSNick Piggin */ 162131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 162231e6b01fSNick Piggin unsigned seq; 1623da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 16245d0f49c1SAl Viro if (unlikely(!dentry)) { 1625e36cffedSJens Axboe if (!try_to_unlazy(nd)) 162620e34357SAl Viro return ERR_PTR(-ECHILD); 162720e34357SAl Viro return NULL; 16285d0f49c1SAl Viro } 16295a18fff2SAl Viro 163012f8ad4bSLinus Torvalds /* 163112f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 163212f8ad4bSLinus Torvalds * the dentry name information from lookup. 163312f8ad4bSLinus Torvalds */ 163463afdfc7SDavid Howells *inode = d_backing_inode(dentry); 16355d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 163620e34357SAl Viro return ERR_PTR(-ECHILD); 163712f8ad4bSLinus Torvalds 163812f8ad4bSLinus Torvalds /* 163912f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 164012f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 164112f8ad4bSLinus Torvalds * 164212f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 164312f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 164412f8ad4bSLinus Torvalds */ 16455d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 164620e34357SAl Viro return ERR_PTR(-ECHILD); 16475a18fff2SAl Viro 1648254cf582SAl Viro *seqp = seq; 16494ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1650c153007bSAl Viro if (likely(status > 0)) 165120e34357SAl Viro return dentry; 1652ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 165320e34357SAl Viro return ERR_PTR(-ECHILD); 165426ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1655209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1656209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16575a18fff2SAl Viro } else { 1658e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 165981e6f520SAl Viro if (unlikely(!dentry)) 166020e34357SAl Viro return NULL; 16614ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16625d0f49c1SAl Viro } 16635a18fff2SAl Viro if (unlikely(status <= 0)) { 1664e9742b53SAl Viro if (!status) 16655d0f49c1SAl Viro d_invalidate(dentry); 16665a18fff2SAl Viro dput(dentry); 166720e34357SAl Viro return ERR_PTR(status); 16685a18fff2SAl Viro } 166920e34357SAl Viro return dentry; 1670697f514dSMiklos Szeredi } 1671697f514dSMiklos Szeredi 1672697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 167388d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1674e3c13928SAl Viro struct dentry *dir, 1675e3c13928SAl Viro unsigned int flags) 1676697f514dSMiklos Szeredi { 167788d8331aSAl Viro struct dentry *dentry, *old; 16781936386eSAl Viro struct inode *inode = dir->d_inode; 1679d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16801936386eSAl Viro 16811936386eSAl Viro /* Don't go there if it's already dead */ 168294bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 168388d8331aSAl Viro return ERR_PTR(-ENOENT); 168494bdd655SAl Viro again: 1685d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 168694bdd655SAl Viro if (IS_ERR(dentry)) 168788d8331aSAl Viro return dentry; 168894bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1689949a852eSAl Viro int error = d_revalidate(dentry, flags); 1690949a852eSAl Viro if (unlikely(error <= 0)) { 169194bdd655SAl Viro if (!error) { 1692949a852eSAl Viro d_invalidate(dentry); 1693949a852eSAl Viro dput(dentry); 169494bdd655SAl Viro goto again; 169594bdd655SAl Viro } 169694bdd655SAl Viro dput(dentry); 1697949a852eSAl Viro dentry = ERR_PTR(error); 1698949a852eSAl Viro } 169994bdd655SAl Viro } else { 17001936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 170185c7f810SAl Viro d_lookup_done(dentry); 17021936386eSAl Viro if (unlikely(old)) { 17031936386eSAl Viro dput(dentry); 17041936386eSAl Viro dentry = old; 1705949a852eSAl Viro } 1706949a852eSAl Viro } 1707e3c13928SAl Viro return dentry; 17081da177e4SLinus Torvalds } 17091da177e4SLinus Torvalds 171088d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 171188d8331aSAl Viro struct dentry *dir, 171288d8331aSAl Viro unsigned int flags) 171388d8331aSAl Viro { 171488d8331aSAl Viro struct inode *inode = dir->d_inode; 171588d8331aSAl Viro struct dentry *res; 171688d8331aSAl Viro inode_lock_shared(inode); 171788d8331aSAl Viro res = __lookup_slow(name, dir, flags); 171888d8331aSAl Viro inode_unlock_shared(inode); 171988d8331aSAl Viro return res; 172088d8331aSAl Viro } 172188d8331aSAl Viro 1722ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1723ba73d987SChristian Brauner struct nameidata *nd) 172452094c8aSAl Viro { 172552094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17267d6beb71SLinus Torvalds int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1727e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 172852094c8aSAl Viro return err; 172952094c8aSAl Viro } 1730ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 173152094c8aSAl Viro } 173252094c8aSAl Viro 173349055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1734d63ff28fSAl Viro { 173549055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 173649055906SAl Viro return -ELOOP; 17374542576bSAl Viro 17384542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17394542576bSAl Viro return 0; 17404542576bSAl Viro if (likely(nd->stack != nd->internal)) 17414542576bSAl Viro return 0; 174260ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 174349055906SAl Viro return 0; 174460ef60c7SAl Viro 174560ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 174660ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 174760ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1748aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 174960ef60c7SAl Viro 1750e5ca024eSAl Viro if (!try_to_unlazy(nd) || !grabbed_link) 175160ef60c7SAl Viro return -ECHILD; 175260ef60c7SAl Viro 175360ef60c7SAl Viro if (nd_alloc_stack(nd)) 175460ef60c7SAl Viro return 0; 1755bc40aee0SAl Viro } 175660ef60c7SAl Viro return -ENOMEM; 175749055906SAl Viro } 175849055906SAl Viro 175949055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 176049055906SAl Viro 176149055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 176249055906SAl Viro struct inode *inode, unsigned seq, int flags) 176349055906SAl Viro { 176449055906SAl Viro struct saved *last; 176549055906SAl Viro const char *res; 176649055906SAl Viro int error = reserve_stack(nd, link, seq); 176749055906SAl Viro 176849055906SAl Viro if (unlikely(error)) { 176949055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1770cd179f44SAl Viro path_put(link); 177106708adbSAl Viro return ERR_PTR(error); 1772626de996SAl Viro } 1773ab104923SAl Viro last = nd->stack + nd->depth++; 17741cf2665bSAl Viro last->link = *link; 1775fceef393SAl Viro clear_delayed_call(&last->done); 17760450b2d1SAl Viro last->seq = seq; 1777ad6cc4c3SAl Viro 1778b1a81972SAl Viro if (flags & WALK_TRAILING) { 1779ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1780ad6cc4c3SAl Viro if (unlikely(error)) 1781ad6cc4c3SAl Viro return ERR_PTR(error); 1782ad6cc4c3SAl Viro } 1783ad6cc4c3SAl Viro 1784dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1785dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1786ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1787ad6cc4c3SAl Viro 1788ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1789ad6cc4c3SAl Viro touch_atime(&last->link); 1790ad6cc4c3SAl Viro cond_resched(); 1791ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1792e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1793ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1794ad6cc4c3SAl Viro touch_atime(&last->link); 1795ad6cc4c3SAl Viro } 1796ad6cc4c3SAl Viro 1797ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1798ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1799ad6cc4c3SAl Viro if (unlikely(error)) 1800ad6cc4c3SAl Viro return ERR_PTR(error); 1801ad6cc4c3SAl Viro 1802ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1803ad6cc4c3SAl Viro if (!res) { 1804ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1805ad6cc4c3SAl Viro struct delayed_call *); 1806ad6cc4c3SAl Viro get = inode->i_op->get_link; 1807ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1808ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1809e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1810ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1811ad6cc4c3SAl Viro } else { 1812ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1813ad6cc4c3SAl Viro } 1814ad6cc4c3SAl Viro if (!res) 1815ad6cc4c3SAl Viro goto all_done; 1816ad6cc4c3SAl Viro if (IS_ERR(res)) 1817ad6cc4c3SAl Viro return res; 1818ad6cc4c3SAl Viro } 1819ad6cc4c3SAl Viro if (*res == '/') { 1820ad6cc4c3SAl Viro error = nd_jump_root(nd); 1821ad6cc4c3SAl Viro if (unlikely(error)) 1822ad6cc4c3SAl Viro return ERR_PTR(error); 1823ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1824ad6cc4c3SAl Viro ; 1825ad6cc4c3SAl Viro } 1826ad6cc4c3SAl Viro if (*res) 1827ad6cc4c3SAl Viro return res; 1828ad6cc4c3SAl Viro all_done: // pure jump 1829ad6cc4c3SAl Viro put_link(nd); 1830ad6cc4c3SAl Viro return NULL; 1831d63ff28fSAl Viro } 1832d63ff28fSAl Viro 18333ddcd056SLinus Torvalds /* 18343ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18353ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18363ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18373ddcd056SLinus Torvalds * for the common case. 18383ddcd056SLinus Torvalds */ 1839b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1840cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 18413ddcd056SLinus Torvalds { 1842cbae4d12SAl Viro struct path path; 1843cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1844cbae4d12SAl Viro 1845cbae4d12SAl Viro if (err < 0) 1846b0417d2cSAl Viro return ERR_PTR(err); 1847cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18488c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1849aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18508f64fb1cSAl Viro /* not a symlink or should not follow */ 1851c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1852c99687a0SAl Viro dput(nd->path.dentry); 1853c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1854c99687a0SAl Viro mntput(nd->path.mnt); 1855c99687a0SAl Viro } 1856c99687a0SAl Viro nd->path = path; 18578f64fb1cSAl Viro nd->inode = inode; 18588f64fb1cSAl Viro nd->seq = seq; 1859b0417d2cSAl Viro return NULL; 18608f64fb1cSAl Viro } 1861a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 186284f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1863cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1864b0417d2cSAl Viro return ERR_PTR(-ECHILD); 186584f0cd9eSAl Viro } else { 186684f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 186784f0cd9eSAl Viro mntget(path.mnt); 1868a7f77542SAl Viro } 1869b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18703ddcd056SLinus Torvalds } 18713ddcd056SLinus Torvalds 1872c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1873c2df1968SAl Viro struct inode **inodep, 1874c2df1968SAl Viro unsigned *seqp) 1875957dd41dSAl Viro { 187612487f30SAl Viro struct dentry *parent, *old; 1877957dd41dSAl Viro 187812487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 187912487f30SAl Viro goto in_root; 188012487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18817ef482faSAl Viro struct path path; 1882efe772d6SAl Viro unsigned seq; 18837ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18847ef482faSAl Viro &nd->root, &path, &seq)) 188512487f30SAl Viro goto in_root; 1886efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1887efe772d6SAl Viro return ERR_PTR(-ECHILD); 1888efe772d6SAl Viro nd->path = path; 1889efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1890efe772d6SAl Viro nd->seq = seq; 1891efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1892efe772d6SAl Viro return ERR_PTR(-ECHILD); 1893efe772d6SAl Viro /* we know that mountpoint was pinned */ 1894957dd41dSAl Viro } 189512487f30SAl Viro old = nd->path.dentry; 189612487f30SAl Viro parent = old->d_parent; 189712487f30SAl Viro *inodep = parent->d_inode; 189812487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 189912487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 190012487f30SAl Viro return ERR_PTR(-ECHILD); 190112487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 190212487f30SAl Viro return ERR_PTR(-ECHILD); 190312487f30SAl Viro return parent; 190412487f30SAl Viro in_root: 1905efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1906efe772d6SAl Viro return ERR_PTR(-ECHILD); 1907957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19087521f22bSAl Viro return ERR_PTR(-ECHILD); 1909c2df1968SAl Viro return NULL; 1910957dd41dSAl Viro } 1911957dd41dSAl Viro 1912c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1913c2df1968SAl Viro struct inode **inodep, 1914c2df1968SAl Viro unsigned *seqp) 1915957dd41dSAl Viro { 191612487f30SAl Viro struct dentry *parent; 191712487f30SAl Viro 1918957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 191912487f30SAl Viro goto in_root; 192012487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19212aa38470SAl Viro struct path path; 19222aa38470SAl Viro 19232aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19242aa38470SAl Viro &nd->root, &path)) 192512487f30SAl Viro goto in_root; 1926165200d6SAl Viro path_put(&nd->path); 1927165200d6SAl Viro nd->path = path; 19282aa38470SAl Viro nd->inode = path.dentry->d_inode; 1929165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1930165200d6SAl Viro return ERR_PTR(-EXDEV); 193112487f30SAl Viro } 1932957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 193312487f30SAl Viro parent = dget_parent(nd->path.dentry); 1934957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1935957dd41dSAl Viro dput(parent); 19367521f22bSAl Viro return ERR_PTR(-ENOENT); 1937957dd41dSAl Viro } 1938c2df1968SAl Viro *seqp = 0; 1939c2df1968SAl Viro *inodep = parent->d_inode; 1940c2df1968SAl Viro return parent; 194112487f30SAl Viro 194212487f30SAl Viro in_root: 1943957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19447521f22bSAl Viro return ERR_PTR(-EXDEV); 1945c2df1968SAl Viro dget(nd->path.dentry); 1946c2df1968SAl Viro return NULL; 1947957dd41dSAl Viro } 1948957dd41dSAl Viro 19497521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1950957dd41dSAl Viro { 1951957dd41dSAl Viro if (type == LAST_DOTDOT) { 19527521f22bSAl Viro const char *error = NULL; 1953c2df1968SAl Viro struct dentry *parent; 1954c2df1968SAl Viro struct inode *inode; 1955c2df1968SAl Viro unsigned seq; 1956957dd41dSAl Viro 1957957dd41dSAl Viro if (!nd->root.mnt) { 19587521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1959957dd41dSAl Viro if (error) 1960957dd41dSAl Viro return error; 1961957dd41dSAl Viro } 1962957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1963c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1964957dd41dSAl Viro else 1965c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1966c2df1968SAl Viro if (IS_ERR(parent)) 1967c2df1968SAl Viro return ERR_CAST(parent); 1968c2df1968SAl Viro if (unlikely(!parent)) 1969c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1970c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1971c2df1968SAl Viro else 1972c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1973c2df1968SAl Viro parent, inode, seq); 1974c2df1968SAl Viro if (unlikely(error)) 1975957dd41dSAl Viro return error; 1976957dd41dSAl Viro 1977957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1978957dd41dSAl Viro /* 1979957dd41dSAl Viro * If there was a racing rename or mount along our 1980957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1981957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1982957dd41dSAl Viro * some fallback). 1983957dd41dSAl Viro */ 1984957dd41dSAl Viro smp_rmb(); 1985957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 19867521f22bSAl Viro return ERR_PTR(-EAGAIN); 1987957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 19887521f22bSAl Viro return ERR_PTR(-EAGAIN); 1989957dd41dSAl Viro } 1990957dd41dSAl Viro } 19917521f22bSAl Viro return NULL; 1992957dd41dSAl Viro } 1993957dd41dSAl Viro 199492d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1995ce57dfc1SAl Viro { 1996db3c9adeSAl Viro struct dentry *dentry; 1997ce57dfc1SAl Viro struct inode *inode; 1998254cf582SAl Viro unsigned seq; 1999ce57dfc1SAl Viro /* 2000ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 2001ce57dfc1SAl Viro * to be able to know about the current root directory and 2002ce57dfc1SAl Viro * parent relationships. 2003ce57dfc1SAl Viro */ 20044693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 20051c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 20064693a547SAl Viro put_link(nd); 20077521f22bSAl Viro return handle_dots(nd, nd->last_type); 20084693a547SAl Viro } 200920e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 201020e34357SAl Viro if (IS_ERR(dentry)) 201192d27016SAl Viro return ERR_CAST(dentry); 201220e34357SAl Viro if (unlikely(!dentry)) { 2013db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 2014db3c9adeSAl Viro if (IS_ERR(dentry)) 201592d27016SAl Viro return ERR_CAST(dentry); 201620e34357SAl Viro } 201756676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 201856676ec3SAl Viro put_link(nd); 2019b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 2020ce57dfc1SAl Viro } 2021ce57dfc1SAl Viro 20221da177e4SLinus Torvalds /* 2023bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2024bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2025bfcfaa77SLinus Torvalds * 2026bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2027bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2028bfcfaa77SLinus Torvalds * fast. 2029bfcfaa77SLinus Torvalds * 2030bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2031bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2032bfcfaa77SLinus Torvalds * crossing operation. 2033bfcfaa77SLinus Torvalds * 2034bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2035bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2036bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2037bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2038bfcfaa77SLinus Torvalds */ 2039bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2040bfcfaa77SLinus Torvalds 2041f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2042bfcfaa77SLinus Torvalds 2043468a9428SGeorge Spelvin #ifdef HASH_MIX 2044bfcfaa77SLinus Torvalds 2045468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2046468a9428SGeorge Spelvin 2047468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20482a18da7aSGeorge Spelvin /* 20492a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20502a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20512a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20522a18da7aSGeorge Spelvin * and no temporaries. 20532a18da7aSGeorge Spelvin * 20542a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20552a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20562a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20572a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20582a18da7aSGeorge Spelvin * 20592a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20602a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20612a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20622a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20632a18da7aSGeorge Spelvin * 20642a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20652a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20662a18da7aSGeorge Spelvin * toggling any given output bit. 20672a18da7aSGeorge Spelvin * 20682a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20692a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20702a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20712a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20722a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20732a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20742a18da7aSGeorge Spelvin * Perfect: 8192 258048 20752a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20762a18da7aSGeorge Spelvin */ 20772a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20782a18da7aSGeorge Spelvin ( x ^= (a), \ 20792a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20802a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20812a18da7aSGeorge Spelvin y *= 9 ) 2082bfcfaa77SLinus Torvalds 20830fed3ac8SGeorge Spelvin /* 20842a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20852a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20862a18da7aSGeorge Spelvin * work done before the hash value is used. 20870fed3ac8SGeorge Spelvin */ 20882a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20890fed3ac8SGeorge Spelvin { 20902a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20912a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20922a18da7aSGeorge Spelvin return y >> 32; 20930fed3ac8SGeorge Spelvin } 20940fed3ac8SGeorge Spelvin 2095bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2096bfcfaa77SLinus Torvalds 20972a18da7aSGeorge Spelvin /* 20982a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 20992a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 21002a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 21012a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 21022a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 21032a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 21042a18da7aSGeorge Spelvin * Perfect: 2048 31744 21052a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 21062a18da7aSGeorge Spelvin */ 21072a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 21082a18da7aSGeorge Spelvin ( x ^= (a), \ 21092a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 21102a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 21112a18da7aSGeorge Spelvin y *= 9 ) 2112bfcfaa77SLinus Torvalds 21132a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 21140fed3ac8SGeorge Spelvin { 21152a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 21162a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 21170fed3ac8SGeorge Spelvin } 21180fed3ac8SGeorge Spelvin 2119bfcfaa77SLinus Torvalds #endif 2120bfcfaa77SLinus Torvalds 21212a18da7aSGeorge Spelvin /* 21222a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21232a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21242a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21252a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21262a18da7aSGeorge Spelvin * finds the delimiter after the name. 21272a18da7aSGeorge Spelvin */ 21288387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2129bfcfaa77SLinus Torvalds { 21308387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2131bfcfaa77SLinus Torvalds 2132bfcfaa77SLinus Torvalds for (;;) { 2133fcfd2fbfSGeorge Spelvin if (!len) 2134fcfd2fbfSGeorge Spelvin goto done; 2135e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2136bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2137bfcfaa77SLinus Torvalds break; 21382a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2139bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2140bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2141bfcfaa77SLinus Torvalds } 21422a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2143bfcfaa77SLinus Torvalds done: 21442a18da7aSGeorge Spelvin return fold_hash(x, y); 2145bfcfaa77SLinus Torvalds } 2146bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2147bfcfaa77SLinus Torvalds 2148fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21498387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2150fcfd2fbfSGeorge Spelvin { 21518387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21528387ff25SLinus Torvalds unsigned long adata, mask, len; 2153fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2154fcfd2fbfSGeorge Spelvin 21558387ff25SLinus Torvalds len = 0; 21568387ff25SLinus Torvalds goto inside; 21578387ff25SLinus Torvalds 2158fcfd2fbfSGeorge Spelvin do { 21592a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2160fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21618387ff25SLinus Torvalds inside: 2162fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2163fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2164fcfd2fbfSGeorge Spelvin 2165fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2166fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21672a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2168fcfd2fbfSGeorge Spelvin 21692a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2170fcfd2fbfSGeorge Spelvin } 2171fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2172fcfd2fbfSGeorge Spelvin 2173bfcfaa77SLinus Torvalds /* 2174bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2175d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2176bfcfaa77SLinus Torvalds */ 21778387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2178bfcfaa77SLinus Torvalds { 21798387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21808387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 218136126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2182bfcfaa77SLinus Torvalds 21838387ff25SLinus Torvalds len = 0; 21848387ff25SLinus Torvalds goto inside; 21858387ff25SLinus Torvalds 2186bfcfaa77SLinus Torvalds do { 21872a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2188bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21898387ff25SLinus Torvalds inside: 2190e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 219136126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 219236126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2193bfcfaa77SLinus Torvalds 219436126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 219536126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 219636126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21972a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 219836126f8fSLinus Torvalds 21992a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2200bfcfaa77SLinus Torvalds } 2201bfcfaa77SLinus Torvalds 22022a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2203bfcfaa77SLinus Torvalds 2204fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 22058387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 22060145acc2SLinus Torvalds { 22078387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 22080145acc2SLinus Torvalds while (len--) 2209fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 22100145acc2SLinus Torvalds return end_name_hash(hash); 22110145acc2SLinus Torvalds } 2212ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 22130145acc2SLinus Torvalds 2214fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 22158387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2216fcfd2fbfSGeorge Spelvin { 22178387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2218fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2219fcfd2fbfSGeorge Spelvin 2220fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2221e0ab7af9SGeorge Spelvin while (c) { 2222fcfd2fbfSGeorge Spelvin len++; 2223fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2224fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2225e0ab7af9SGeorge Spelvin } 2226fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2227fcfd2fbfSGeorge Spelvin } 2228f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2229fcfd2fbfSGeorge Spelvin 22303ddcd056SLinus Torvalds /* 2231200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2232200e9ef7SLinus Torvalds * one character. 2233200e9ef7SLinus Torvalds */ 22348387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2235200e9ef7SLinus Torvalds { 22368387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2237200e9ef7SLinus Torvalds unsigned long len = 0, c; 2238200e9ef7SLinus Torvalds 2239200e9ef7SLinus Torvalds c = (unsigned char)*name; 2240200e9ef7SLinus Torvalds do { 2241200e9ef7SLinus Torvalds len++; 2242200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2243200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2244200e9ef7SLinus Torvalds } while (c && c != '/'); 2245d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2246200e9ef7SLinus Torvalds } 2247200e9ef7SLinus Torvalds 2248bfcfaa77SLinus Torvalds #endif 2249bfcfaa77SLinus Torvalds 2250200e9ef7SLinus Torvalds /* 22511da177e4SLinus Torvalds * Name resolution. 2252ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2253ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22541da177e4SLinus Torvalds * 2255ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2256ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22571da177e4SLinus Torvalds */ 22586de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22591da177e4SLinus Torvalds { 2260d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22611da177e4SLinus Torvalds int err; 22621da177e4SLinus Torvalds 2263b4c03536SAl Viro nd->last_type = LAST_ROOT; 2264c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22659b5858e9SAl Viro if (IS_ERR(name)) 22669b5858e9SAl Viro return PTR_ERR(name); 22671da177e4SLinus Torvalds while (*name=='/') 22681da177e4SLinus Torvalds name++; 22691a97d899SAl Viro if (!*name) { 22701a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22719e18f10aSAl Viro return 0; 22721a97d899SAl Viro } 22731da177e4SLinus Torvalds 22741da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22751da177e4SLinus Torvalds for(;;) { 2276549c7297SChristian Brauner struct user_namespace *mnt_userns; 227792d27016SAl Viro const char *link; 2278d6bb3e90SLinus Torvalds u64 hash_len; 2279fe479a58SAl Viro int type; 22801da177e4SLinus Torvalds 2281549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2282549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22831da177e4SLinus Torvalds if (err) 22843595e234SAl Viro return err; 22851da177e4SLinus Torvalds 22868387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22871da177e4SLinus Torvalds 2288fe479a58SAl Viro type = LAST_NORM; 2289d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2290fe479a58SAl Viro case 2: 2291200e9ef7SLinus Torvalds if (name[1] == '.') { 2292fe479a58SAl Viro type = LAST_DOTDOT; 2293bcba1e7dSAl Viro nd->state |= ND_JUMPED; 229416c2cd71SAl Viro } 2295fe479a58SAl Viro break; 2296fe479a58SAl Viro case 1: 2297fe479a58SAl Viro type = LAST_DOT; 2298fe479a58SAl Viro } 22995a202bcdSAl Viro if (likely(type == LAST_NORM)) { 23005a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2301bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 23025a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2303a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2304da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 23055a202bcdSAl Viro if (err < 0) 23063595e234SAl Viro return err; 2307d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2308d6bb3e90SLinus Torvalds name = this.name; 23095a202bcdSAl Viro } 23105a202bcdSAl Viro } 2311fe479a58SAl Viro 2312d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2313d6bb3e90SLinus Torvalds nd->last.name = name; 23145f4a6a69SAl Viro nd->last_type = type; 23155f4a6a69SAl Viro 2316d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2317d6bb3e90SLinus Torvalds if (!*name) 2318bdf6cbf1SAl Viro goto OK; 2319200e9ef7SLinus Torvalds /* 2320200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2321200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2322200e9ef7SLinus Torvalds */ 2323200e9ef7SLinus Torvalds do { 2324d6bb3e90SLinus Torvalds name++; 2325d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 23268620c238SAl Viro if (unlikely(!*name)) { 23278620c238SAl Viro OK: 2328d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2329c108837eSAl Viro if (!depth) { 2330549c7297SChristian Brauner nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); 23310f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2332c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 23338620c238SAl Viro return 0; 2334c108837eSAl Viro } 23358620c238SAl Viro /* last component of nested symlink */ 2336d8d4611aSAl Viro name = nd->stack[--depth].name; 23378c4efe22SAl Viro link = walk_component(nd, 0); 23381c4ff1a8SAl Viro } else { 23391c4ff1a8SAl Viro /* not the last component */ 23408c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 23418620c238SAl Viro } 234292d27016SAl Viro if (unlikely(link)) { 234392d27016SAl Viro if (IS_ERR(link)) 234492d27016SAl Viro return PTR_ERR(link); 234592d27016SAl Viro /* a symlink to follow */ 2346d8d4611aSAl Viro nd->stack[depth++].name = name; 234792d27016SAl Viro name = link; 23489e18f10aSAl Viro continue; 234948c8b0c5SAl Viro } 235097242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 235197242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2352e36cffedSJens Axboe if (!try_to_unlazy(nd)) 235397242f99SAl Viro return -ECHILD; 235497242f99SAl Viro } 23553595e234SAl Viro return -ENOTDIR; 23565f4a6a69SAl Viro } 2357ce57dfc1SAl Viro } 235897242f99SAl Viro } 23591da177e4SLinus Torvalds 2360edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2361c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 236231e6b01fSNick Piggin { 2363740a1678SAleksa Sarai int error; 2364c8a53ee5SAl Viro const char *s = nd->name->name; 236531e6b01fSNick Piggin 23666c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23676c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23686c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23696c6ec2b0SJens Axboe 2370c0eb027eSLinus Torvalds if (!*s) 2371c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2372edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2373edc2b1daSAl Viro rcu_read_lock(); 2374c0eb027eSLinus Torvalds 2375bcba1e7dSAl Viro nd->flags = flags; 2376bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2377ab87f9a5SAleksa Sarai 2378ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2379ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2380ab87f9a5SAleksa Sarai smp_rmb(); 2381ab87f9a5SAleksa Sarai 2382bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2383b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2384b18825a7SDavid Howells struct inode *inode = root->d_inode; 238593893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2386368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23875b6ca027SAl Viro nd->path = nd->root; 23885b6ca027SAl Viro nd->inode = inode; 23895b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2390ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23918f47a016SAl Viro nd->root_seq = nd->seq; 23925b6ca027SAl Viro } else { 23935b6ca027SAl Viro path_get(&nd->path); 23945b6ca027SAl Viro } 2395368ee9baSAl Viro return s; 23965b6ca027SAl Viro } 23975b6ca027SAl Viro 239831e6b01fSNick Piggin nd->root.mnt = NULL; 239931e6b01fSNick Piggin 24008db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 24018db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2402740a1678SAleksa Sarai error = nd_jump_root(nd); 2403740a1678SAleksa Sarai if (unlikely(error)) 2404740a1678SAleksa Sarai return ERR_PTR(error); 2405ef55d917SAl Viro return s; 24068db52c7eSAleksa Sarai } 24078db52c7eSAleksa Sarai 24088db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 24098db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2410e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 241131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2412c28cc364SNick Piggin unsigned seq; 241331e6b01fSNick Piggin 2414c28cc364SNick Piggin do { 2415c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 241631e6b01fSNick Piggin nd->path = fs->pwd; 2417ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2418c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2419c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2420e41f7d4eSAl Viro } else { 2421e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2422ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2423e41f7d4eSAl Viro } 242431e6b01fSNick Piggin } else { 2425582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2426c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 242731e6b01fSNick Piggin struct dentry *dentry; 242831e6b01fSNick Piggin 24292903ff01SAl Viro if (!f.file) 2430368ee9baSAl Viro return ERR_PTR(-EBADF); 243131e6b01fSNick Piggin 24322903ff01SAl Viro dentry = f.file->f_path.dentry; 243331e6b01fSNick Piggin 2434edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 24352903ff01SAl Viro fdput(f); 2436368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2437f52e0c11SAl Viro } 24382903ff01SAl Viro 24392903ff01SAl Viro nd->path = f.file->f_path; 2440e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 244134a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 244234a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24435590ff0dSUlrich Drepper } else { 24442903ff01SAl Viro path_get(&nd->path); 244534a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24461da177e4SLinus Torvalds } 244734a26b99SAl Viro fdput(f); 2448e41f7d4eSAl Viro } 24498db52c7eSAleksa Sarai 2450adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2451adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2452adb21d2bSAleksa Sarai nd->root = nd->path; 2453adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2454adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2455adb21d2bSAleksa Sarai } else { 2456adb21d2bSAleksa Sarai path_get(&nd->root); 2457bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2458adb21d2bSAleksa Sarai } 2459adb21d2bSAleksa Sarai } 2460adb21d2bSAleksa Sarai return s; 24619b4a9b14SAl Viro } 24629b4a9b14SAl Viro 24631ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 246495fa25d9SAl Viro { 2465bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2466bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2467bd92d7feSAl Viro 2468c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2469bd92d7feSAl Viro } 2470bd92d7feSAl Viro 24714f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24724f757f3cSAl Viro { 2473c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2474db3c9adeSAl Viro dget(nd->path.dentry); 2475b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2476b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 24774f757f3cSAl Viro } 24784f757f3cSAl Viro 24799b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2480c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24819b4a9b14SAl Viro { 2482c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2483bd92d7feSAl Viro int err; 248431e6b01fSNick Piggin 24859b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24864f757f3cSAl Viro err = handle_lookup_down(nd); 24875f336e72SAl Viro if (unlikely(err < 0)) 24885f336e72SAl Viro s = ERR_PTR(err); 24894f757f3cSAl Viro } 24904f757f3cSAl Viro 24911ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24921ccac622SAl Viro (s = lookup_last(nd)) != NULL) 24931ccac622SAl Viro ; 24944f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 24954f0ed93fSAl Viro err = handle_lookup_down(nd); 2496bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 24974f0ed93fSAl Viro } 24989f1fafeeSAl Viro if (!err) 24999f1fafeeSAl Viro err = complete_walk(nd); 2500bd92d7feSAl Viro 2501deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2502deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2503bd23a539SAl Viro err = -ENOTDIR; 2504625b6d10SAl Viro if (!err) { 2505625b6d10SAl Viro *path = nd->path; 2506625b6d10SAl Viro nd->path.mnt = NULL; 2507625b6d10SAl Viro nd->path.dentry = NULL; 2508625b6d10SAl Viro } 2509deb106c6SAl Viro terminate_walk(nd); 2510bd92d7feSAl Viro return err; 251131e6b01fSNick Piggin } 251231e6b01fSNick Piggin 2513794ebceaSStephen Brennan int filename_lookup(int dfd, struct filename *name, unsigned flags, 25149ad1aaa6SAl Viro struct path *path, struct path *root) 2515873f1eedSJeff Layton { 2516894bc8c4SAl Viro int retval; 25179883d185SAl Viro struct nameidata nd; 2518abc9f5beSAl Viro if (IS_ERR(name)) 2519abc9f5beSAl Viro return PTR_ERR(name); 252006422964SAl Viro set_nameidata(&nd, dfd, name, root); 2521c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2522873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2523c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2524873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2525c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2526873f1eedSJeff Layton 2527873f1eedSJeff Layton if (likely(!retval)) 2528161aff1dSAl Viro audit_inode(name, path->dentry, 2529161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 25309883d185SAl Viro restore_nameidata(); 2531020250f3SDmitry Kadashev return retval; 2532020250f3SDmitry Kadashev } 2533020250f3SDmitry Kadashev 25348bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2535c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2536391172c4SAl Viro struct path *parent) 25378bcb77faSAl Viro { 2538c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25399b5858e9SAl Viro int err = link_path_walk(s, nd); 25408bcb77faSAl Viro if (!err) 25418bcb77faSAl Viro err = complete_walk(nd); 2542391172c4SAl Viro if (!err) { 2543391172c4SAl Viro *parent = nd->path; 2544391172c4SAl Viro nd->path.mnt = NULL; 2545391172c4SAl Viro nd->path.dentry = NULL; 2546391172c4SAl Viro } 2547deb106c6SAl Viro terminate_walk(nd); 25488bcb77faSAl Viro return err; 25498bcb77faSAl Viro } 25508bcb77faSAl Viro 25510766ec82SStephen Brennan /* Note: this does not consume "name" */ 2552c5f563f9SAl Viro static int filename_parentat(int dfd, struct filename *name, 2553391172c4SAl Viro unsigned int flags, struct path *parent, 2554391172c4SAl Viro struct qstr *last, int *type) 25558bcb77faSAl Viro { 25568bcb77faSAl Viro int retval; 25579883d185SAl Viro struct nameidata nd; 25588bcb77faSAl Viro 25595c31b6ceSAl Viro if (IS_ERR(name)) 25600ee50b47SDmitry Kadashev return PTR_ERR(name); 256106422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2562c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25638bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2564c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25658bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2566c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2567391172c4SAl Viro if (likely(!retval)) { 2568391172c4SAl Viro *last = nd.last; 2569391172c4SAl Viro *type = nd.last_type; 2570c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2571391172c4SAl Viro } 25729883d185SAl Viro restore_nameidata(); 25730ee50b47SDmitry Kadashev return retval; 25740ee50b47SDmitry Kadashev } 25750ee50b47SDmitry Kadashev 257679714f72SAl Viro /* does lookup, returns the object with parent locked */ 25770766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25785590ff0dSUlrich Drepper { 25795c31b6ceSAl Viro struct dentry *d; 2580391172c4SAl Viro struct qstr last; 25810ee50b47SDmitry Kadashev int type, error; 258251689104SPaul Moore 2583c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25840ee50b47SDmitry Kadashev if (error) 25850ee50b47SDmitry Kadashev return ERR_PTR(error); 25865c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2587391172c4SAl Viro path_put(path); 25885c31b6ceSAl Viro return ERR_PTR(-EINVAL); 258979714f72SAl Viro } 25905955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2591391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 259279714f72SAl Viro if (IS_ERR(d)) { 25935955102cSAl Viro inode_unlock(path->dentry->d_inode); 2594391172c4SAl Viro path_put(path); 259579714f72SAl Viro } 259679714f72SAl Viro return d; 25975590ff0dSUlrich Drepper } 25985590ff0dSUlrich Drepper 25990766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 26000766ec82SStephen Brennan { 26010766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 26020766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 26030766ec82SStephen Brennan 26040766ec82SStephen Brennan putname(filename); 26050766ec82SStephen Brennan return res; 26060766ec82SStephen Brennan } 26070766ec82SStephen Brennan 2608d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2609d1811465SAl Viro { 2610794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2611794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2612794ebceaSStephen Brennan 2613794ebceaSStephen Brennan putname(filename); 2614794ebceaSStephen Brennan return ret; 2615794ebceaSStephen Brennan 2616d1811465SAl Viro } 26174d359507SAl Viro EXPORT_SYMBOL(kern_path); 2618d1811465SAl Viro 261916f18200SJosef 'Jeff' Sipek /** 262016f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 262116f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 262216f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 262316f18200SJosef 'Jeff' Sipek * @name: pointer to file name 262416f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2625e0a01249SAl Viro * @path: pointer to struct path to fill 262616f18200SJosef 'Jeff' Sipek */ 262716f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 262816f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2629e0a01249SAl Viro struct path *path) 263016f18200SJosef 'Jeff' Sipek { 2631794ebceaSStephen Brennan struct filename *filename; 26329ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2633794ebceaSStephen Brennan int ret; 2634794ebceaSStephen Brennan 2635794ebceaSStephen Brennan filename = getname_kernel(name); 26369ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2637794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2638794ebceaSStephen Brennan putname(filename); 2639794ebceaSStephen Brennan return ret; 264016f18200SJosef 'Jeff' Sipek } 26414d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 264216f18200SJosef 'Jeff' Sipek 2643c2fd68b6SChristian Brauner static int lookup_one_common(struct user_namespace *mnt_userns, 2644c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2645c2fd68b6SChristian Brauner struct qstr *this) 26463c95f0dcSAl Viro { 26473c95f0dcSAl Viro this->name = name; 26483c95f0dcSAl Viro this->len = len; 26493c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26503c95f0dcSAl Viro if (!len) 26513c95f0dcSAl Viro return -EACCES; 26523c95f0dcSAl Viro 26533c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26543c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26553c95f0dcSAl Viro return -EACCES; 26563c95f0dcSAl Viro } 26573c95f0dcSAl Viro 26583c95f0dcSAl Viro while (len--) { 26593c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26603c95f0dcSAl Viro if (c == '/' || c == '\0') 26613c95f0dcSAl Viro return -EACCES; 26623c95f0dcSAl Viro } 26633c95f0dcSAl Viro /* 26643c95f0dcSAl Viro * See if the low-level filesystem might want 26653c95f0dcSAl Viro * to use its own hash.. 26663c95f0dcSAl Viro */ 26673c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26683c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26693c95f0dcSAl Viro if (err < 0) 26703c95f0dcSAl Viro return err; 26713c95f0dcSAl Viro } 26723c95f0dcSAl Viro 2673c2fd68b6SChristian Brauner return inode_permission(mnt_userns, base->d_inode, MAY_EXEC); 26743c95f0dcSAl Viro } 26753c95f0dcSAl Viro 2676eead1911SChristoph Hellwig /** 26770da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26780da0b7fdSDavid Howells * @name: pathname component to lookup 26790da0b7fdSDavid Howells * @base: base directory to lookup from 26800da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26810da0b7fdSDavid Howells * 26820da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26830da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26840da0b7fdSDavid Howells * 26850da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26860da0b7fdSDavid Howells * not be called by generic code. 26870da0b7fdSDavid Howells * 26880da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26890da0b7fdSDavid Howells */ 26900da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26910da0b7fdSDavid Howells { 26920da0b7fdSDavid Howells struct qstr this; 26930da0b7fdSDavid Howells int err; 26940da0b7fdSDavid Howells 26950da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26960da0b7fdSDavid Howells 2697c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 26980da0b7fdSDavid Howells if (err) 26990da0b7fdSDavid Howells return ERR_PTR(err); 27000da0b7fdSDavid Howells 27010da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 27020da0b7fdSDavid Howells } 27030da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 27040da0b7fdSDavid Howells 27050da0b7fdSDavid Howells /** 2706a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2707eead1911SChristoph Hellwig * @name: pathname component to lookup 2708eead1911SChristoph Hellwig * @base: base directory to lookup from 2709eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2710eead1911SChristoph Hellwig * 2711a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 27129e7543e9SAl Viro * not be called by generic code. 2713bbddca8eSNeilBrown * 2714bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2715eead1911SChristoph Hellwig */ 2716057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2717057f6c01SJames Morris { 27188613a209SAl Viro struct dentry *dentry; 2719057f6c01SJames Morris struct qstr this; 2720cda309deSMiklos Szeredi int err; 2721057f6c01SJames Morris 27225955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27232f9092e1SDavid Woodhouse 2724c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2725cda309deSMiklos Szeredi if (err) 2726cda309deSMiklos Szeredi return ERR_PTR(err); 2727cda309deSMiklos Szeredi 27288613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27298613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2730057f6c01SJames Morris } 27314d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2732057f6c01SJames Morris 2733bbddca8eSNeilBrown /** 2734c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 2735c2fd68b6SChristian Brauner * @mnt_userns: user namespace of the mount the lookup is performed from 2736c2fd68b6SChristian Brauner * @name: pathname component to lookup 2737c2fd68b6SChristian Brauner * @base: base directory to lookup from 2738c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2739c2fd68b6SChristian Brauner * 2740c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2741c2fd68b6SChristian Brauner * not be called by generic code. 2742c2fd68b6SChristian Brauner * 2743c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2744c2fd68b6SChristian Brauner */ 2745c2fd68b6SChristian Brauner struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name, 2746c2fd68b6SChristian Brauner struct dentry *base, int len) 2747c2fd68b6SChristian Brauner { 2748c2fd68b6SChristian Brauner struct dentry *dentry; 2749c2fd68b6SChristian Brauner struct qstr this; 2750c2fd68b6SChristian Brauner int err; 2751c2fd68b6SChristian Brauner 2752c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2753c2fd68b6SChristian Brauner 2754c2fd68b6SChristian Brauner err = lookup_one_common(mnt_userns, name, base, len, &this); 2755c2fd68b6SChristian Brauner if (err) 2756c2fd68b6SChristian Brauner return ERR_PTR(err); 2757c2fd68b6SChristian Brauner 2758c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2759c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2760c2fd68b6SChristian Brauner } 2761c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2762c2fd68b6SChristian Brauner 2763c2fd68b6SChristian Brauner /** 2764bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2765bbddca8eSNeilBrown * @name: pathname component to lookup 2766bbddca8eSNeilBrown * @base: base directory to lookup from 2767bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2768bbddca8eSNeilBrown * 2769bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2770bbddca8eSNeilBrown * not be called by generic code. 2771bbddca8eSNeilBrown * 2772bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2773bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2774bbddca8eSNeilBrown */ 2775bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2776bbddca8eSNeilBrown struct dentry *base, int len) 2777bbddca8eSNeilBrown { 2778bbddca8eSNeilBrown struct qstr this; 2779bbddca8eSNeilBrown int err; 278020d00ee8SLinus Torvalds struct dentry *ret; 2781bbddca8eSNeilBrown 2782c2fd68b6SChristian Brauner err = lookup_one_common(&init_user_ns, name, base, len, &this); 2783bbddca8eSNeilBrown if (err) 2784bbddca8eSNeilBrown return ERR_PTR(err); 2785bbddca8eSNeilBrown 278620d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 278720d00ee8SLinus Torvalds if (!ret) 278820d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 278920d00ee8SLinus Torvalds return ret; 2790bbddca8eSNeilBrown } 2791bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2792bbddca8eSNeilBrown 27936c2d4798SAl Viro /* 27946c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 27956c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 27966c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 27976c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 27986c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 27996c2d4798SAl Viro * this one avoids such problems. 28006c2d4798SAl Viro */ 28016c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28026c2d4798SAl Viro struct dentry *base, int len) 28036c2d4798SAl Viro { 28046c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 28052fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 28066c2d4798SAl Viro dput(ret); 28076c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 28086c2d4798SAl Viro } 28096c2d4798SAl Viro return ret; 28106c2d4798SAl Viro } 28116c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28126c2d4798SAl Viro 2813eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2814eedf265aSEric W. Biederman int path_pts(struct path *path) 2815eedf265aSEric W. Biederman { 2816eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2817eedf265aSEric W. Biederman * the input path. 2818eedf265aSEric W. Biederman */ 2819a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2820a6a7eb76SAl Viro struct dentry *child; 282119f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2822eedf265aSEric W. Biederman 2823a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2824a6a7eb76SAl Viro dput(parent); 282563b27720SAl Viro return -ENOENT; 2826a6a7eb76SAl Viro } 282763b27720SAl Viro dput(path->dentry); 282863b27720SAl Viro path->dentry = parent; 2829eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2830eedf265aSEric W. Biederman if (!child) 2831eedf265aSEric W. Biederman return -ENOENT; 2832eedf265aSEric W. Biederman 2833eedf265aSEric W. Biederman path->dentry = child; 2834eedf265aSEric W. Biederman dput(parent); 283519f6028aSAl Viro follow_down(path); 2836eedf265aSEric W. Biederman return 0; 2837eedf265aSEric W. Biederman } 2838eedf265aSEric W. Biederman #endif 2839eedf265aSEric W. Biederman 28401fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 28411fa1e7f6SAndy Whitcroft struct path *path, int *empty) 28421da177e4SLinus Torvalds { 2843794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2844794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2845794ebceaSStephen Brennan 2846794ebceaSStephen Brennan putname(filename); 2847794ebceaSStephen Brennan return ret; 28481da177e4SLinus Torvalds } 2849b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 28501fa1e7f6SAndy Whitcroft 2851ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2852ba73d987SChristian Brauner struct inode *inode) 28531da177e4SLinus Torvalds { 28548e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2855da9592edSDavid Howells 2856ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) 28571da177e4SLinus Torvalds return 0; 2858ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) 28591da177e4SLinus Torvalds return 0; 2860ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 28611da177e4SLinus Torvalds } 2862cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 28631da177e4SLinus Torvalds 28641da177e4SLinus Torvalds /* 28651da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 28661da177e4SLinus Torvalds * whether the type of victim is right. 28671da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 28681da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 28691da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 28701da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 28711da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 28721da177e4SLinus Torvalds * a. be owner of dir, or 28731da177e4SLinus Torvalds * b. be owner of victim, or 28741da177e4SLinus Torvalds * c. have CAP_FOWNER capability 28751da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 28761da177e4SLinus Torvalds * links pointing to it. 28770bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 28780bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 28790bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 28800bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 28810bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 28821da177e4SLinus Torvalds * nfs_async_unlink(). 28831da177e4SLinus Torvalds */ 2884ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2885ba73d987SChristian Brauner struct dentry *victim, bool isdir) 28861da177e4SLinus Torvalds { 288763afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 28881da177e4SLinus Torvalds int error; 28891da177e4SLinus Torvalds 2890b18825a7SDavid Howells if (d_is_negative(victim)) 28911da177e4SLinus Torvalds return -ENOENT; 2892b18825a7SDavid Howells BUG_ON(!inode); 28931da177e4SLinus Torvalds 28941da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2895593d1ce8SEric W. Biederman 2896593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2897ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 2898ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 2899593d1ce8SEric W. Biederman return -EOVERFLOW; 2900593d1ce8SEric W. Biederman 29014fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29021da177e4SLinus Torvalds 2903ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29041da177e4SLinus Torvalds if (error) 29051da177e4SLinus Torvalds return error; 29061da177e4SLinus Torvalds if (IS_APPEND(dir)) 29071da177e4SLinus Torvalds return -EPERM; 2908b18825a7SDavid Howells 2909ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2910ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2911ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 29121da177e4SLinus Torvalds return -EPERM; 29131da177e4SLinus Torvalds if (isdir) { 291444b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29151da177e4SLinus Torvalds return -ENOTDIR; 29161da177e4SLinus Torvalds if (IS_ROOT(victim)) 29171da177e4SLinus Torvalds return -EBUSY; 291844b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29191da177e4SLinus Torvalds return -EISDIR; 29201da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29211da177e4SLinus Torvalds return -ENOENT; 29221da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29231da177e4SLinus Torvalds return -EBUSY; 29241da177e4SLinus Torvalds return 0; 29251da177e4SLinus Torvalds } 29261da177e4SLinus Torvalds 29271da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29281da177e4SLinus Torvalds * dir. 29291da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29301da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29311da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2932036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2933036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2934036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 29351da177e4SLinus Torvalds */ 2936ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2937ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 29381da177e4SLinus Torvalds { 293914e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 29401da177e4SLinus Torvalds if (child->d_inode) 29411da177e4SLinus Torvalds return -EEXIST; 29421da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29431da177e4SLinus Torvalds return -ENOENT; 29448e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 2945036d5236SEric W. Biederman return -EOVERFLOW; 29468e538913SChristian Brauner 2947ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 29481da177e4SLinus Torvalds } 29491da177e4SLinus Torvalds 29501da177e4SLinus Torvalds /* 29511da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 29521da177e4SLinus Torvalds */ 29531da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 29541da177e4SLinus Torvalds { 29551da177e4SLinus Torvalds struct dentry *p; 29561da177e4SLinus Torvalds 29571da177e4SLinus Torvalds if (p1 == p2) { 29585955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29591da177e4SLinus Torvalds return NULL; 29601da177e4SLinus Torvalds } 29611da177e4SLinus Torvalds 2962fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 29631da177e4SLinus Torvalds 2964e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2965e2761a11SOGAWA Hirofumi if (p) { 29665955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 29675955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 29681da177e4SLinus Torvalds return p; 29691da177e4SLinus Torvalds } 29701da177e4SLinus Torvalds 2971e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2972e2761a11SOGAWA Hirofumi if (p) { 29735955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29745955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 29751da177e4SLinus Torvalds return p; 29761da177e4SLinus Torvalds } 29771da177e4SLinus Torvalds 29785955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29795955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 29801da177e4SLinus Torvalds return NULL; 29811da177e4SLinus Torvalds } 29824d359507SAl Viro EXPORT_SYMBOL(lock_rename); 29831da177e4SLinus Torvalds 29841da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 29851da177e4SLinus Torvalds { 29865955102cSAl Viro inode_unlock(p1->d_inode); 29871da177e4SLinus Torvalds if (p1 != p2) { 29885955102cSAl Viro inode_unlock(p2->d_inode); 2989fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 29901da177e4SLinus Torvalds } 29911da177e4SLinus Torvalds } 29924d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 29931da177e4SLinus Torvalds 29946521f891SChristian Brauner /** 29956521f891SChristian Brauner * vfs_create - create new file 29966521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 29976521f891SChristian Brauner * @dir: inode of @dentry 29986521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 29996521f891SChristian Brauner * @mode: mode of the new file 30006521f891SChristian Brauner * @want_excl: whether the file must not yet exist 30016521f891SChristian Brauner * 30026521f891SChristian Brauner * Create a new file. 30036521f891SChristian Brauner * 30046521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 30056521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 30066521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 30076521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 30086521f891SChristian Brauner * raw inode simply passs init_user_ns. 30096521f891SChristian Brauner */ 30106521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 30116521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 30121da177e4SLinus Torvalds { 30136521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 30141da177e4SLinus Torvalds if (error) 30151da177e4SLinus Torvalds return error; 30161da177e4SLinus Torvalds 3017acfa4380SAl Viro if (!dir->i_op->create) 30181da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 30191da177e4SLinus Torvalds mode &= S_IALLUGO; 30201da177e4SLinus Torvalds mode |= S_IFREG; 30211da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 30221da177e4SLinus Torvalds if (error) 30231da177e4SLinus Torvalds return error; 3024549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 3025a74574aaSStephen Smalley if (!error) 3026f38aa942SAmy Griffis fsnotify_create(dir, dentry); 30271da177e4SLinus Torvalds return error; 30281da177e4SLinus Torvalds } 30294d359507SAl Viro EXPORT_SYMBOL(vfs_create); 30301da177e4SLinus Torvalds 30318e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 30328e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 30338e6c848eSAl Viro void *arg) 30348e6c848eSAl Viro { 30358e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 3036ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 30378e6c848eSAl Viro if (error) 30388e6c848eSAl Viro return error; 30398e6c848eSAl Viro 30408e6c848eSAl Viro mode &= S_IALLUGO; 30418e6c848eSAl Viro mode |= S_IFREG; 30428e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 30438e6c848eSAl Viro if (error) 30448e6c848eSAl Viro return error; 30458e6c848eSAl Viro error = f(dentry, mode, arg); 30468e6c848eSAl Viro if (!error) 30478e6c848eSAl Viro fsnotify_create(dir, dentry); 30488e6c848eSAl Viro return error; 30498e6c848eSAl Viro } 30508e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 30518e6c848eSAl Viro 3052a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3053a2982cc9SEric W. Biederman { 3054a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3055a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3056a2982cc9SEric W. Biederman } 3057a2982cc9SEric W. Biederman 3058ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 3059ba73d987SChristian Brauner int acc_mode, int flag) 30601da177e4SLinus Torvalds { 30613fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 30621da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 30631da177e4SLinus Torvalds int error; 30641da177e4SLinus Torvalds 30651da177e4SLinus Torvalds if (!inode) 30661da177e4SLinus Torvalds return -ENOENT; 30671da177e4SLinus Torvalds 3068c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3069c8fe8f30SChristoph Hellwig case S_IFLNK: 30701da177e4SLinus Torvalds return -ELOOP; 3071c8fe8f30SChristoph Hellwig case S_IFDIR: 3072fc4177beSKees Cook if (acc_mode & MAY_WRITE) 30731da177e4SLinus Torvalds return -EISDIR; 3074fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3075fc4177beSKees Cook return -EACCES; 3076c8fe8f30SChristoph Hellwig break; 3077c8fe8f30SChristoph Hellwig case S_IFBLK: 3078c8fe8f30SChristoph Hellwig case S_IFCHR: 3079a2982cc9SEric W. Biederman if (!may_open_dev(path)) 30801da177e4SLinus Torvalds return -EACCES; 3081633fb6acSKees Cook fallthrough; 3082c8fe8f30SChristoph Hellwig case S_IFIFO: 3083c8fe8f30SChristoph Hellwig case S_IFSOCK: 3084633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3085633fb6acSKees Cook return -EACCES; 30861da177e4SLinus Torvalds flag &= ~O_TRUNC; 3087c8fe8f30SChristoph Hellwig break; 30880fd338b2SKees Cook case S_IFREG: 30890fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 30900fd338b2SKees Cook return -EACCES; 30910fd338b2SKees Cook break; 30924a3fd211SDave Hansen } 3093b41572e9SDave Hansen 3094ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 3095b41572e9SDave Hansen if (error) 3096b41572e9SDave Hansen return error; 30976146f0d5SMimi Zohar 30981da177e4SLinus Torvalds /* 30991da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 31001da177e4SLinus Torvalds */ 31011da177e4SLinus Torvalds if (IS_APPEND(inode)) { 31028737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 31037715b521SAl Viro return -EPERM; 31041da177e4SLinus Torvalds if (flag & O_TRUNC) 31057715b521SAl Viro return -EPERM; 31061da177e4SLinus Torvalds } 31071da177e4SLinus Torvalds 31081da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3109ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 31107715b521SAl Viro return -EPERM; 31111da177e4SLinus Torvalds 3112f3c7691eSJ. Bruce Fields return 0; 31137715b521SAl Viro } 31147715b521SAl Viro 3115549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 31167715b521SAl Viro { 3117f0bb5aafSAl Viro const struct path *path = &filp->f_path; 31187715b521SAl Viro struct inode *inode = path->dentry->d_inode; 31197715b521SAl Viro int error = get_write_access(inode); 31201da177e4SLinus Torvalds if (error) 31217715b521SAl Viro return error; 3122482e0007SJeff Layton 3123ea0d3ab2STetsuo Handa error = security_path_truncate(path); 31241da177e4SLinus Torvalds if (!error) { 3125549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3126d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3127e1181ee6SJeff Layton filp); 31281da177e4SLinus Torvalds } 31291da177e4SLinus Torvalds put_write_access(inode); 3130acd0c935SMimi Zohar return error; 31311da177e4SLinus Torvalds } 31321da177e4SLinus Torvalds 3133d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3134d57999e1SDave Hansen { 31358a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 31368a5e929dSAl Viro flag--; 3137d57999e1SDave Hansen return flag; 3138d57999e1SDave Hansen } 3139d57999e1SDave Hansen 3140ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3141ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3142ba73d987SChristian Brauner umode_t mode) 3143d18e9008SMiklos Szeredi { 3144d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3145d18e9008SMiklos Szeredi if (error) 3146d18e9008SMiklos Szeredi return error; 3147d18e9008SMiklos Szeredi 31488e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) 31491328c727SSeth Forshee return -EOVERFLOW; 31501328c727SSeth Forshee 3151ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 315247291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3153d18e9008SMiklos Szeredi if (error) 3154d18e9008SMiklos Szeredi return error; 3155d18e9008SMiklos Szeredi 3156d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3157d18e9008SMiklos Szeredi } 3158d18e9008SMiklos Szeredi 31591acf0af9SDavid Howells /* 31601acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 31611acf0af9SDavid Howells * dentry. 31621acf0af9SDavid Howells * 31631acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 31641acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 31651acf0af9SDavid Howells * 316600a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 316700a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 316800a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 31691acf0af9SDavid Howells * 31701acf0af9SDavid Howells * Returns an error code otherwise. 31711acf0af9SDavid Howells */ 3172239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3173239eb983SAl Viro struct file *file, 31743ec2eef1SAl Viro int open_flag, umode_t mode) 3175d18e9008SMiklos Szeredi { 3176d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3177d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3178d18e9008SMiklos Szeredi int error; 3179d18e9008SMiklos Szeredi 3180d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3181d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3182d18e9008SMiklos Szeredi 318330d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 318430d90494SAl Viro file->f_path.mnt = nd->path.mnt; 31850fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 318644907d79SAl Viro open_to_namei_flags(open_flag), mode); 31876fbd0714SAl Viro d_lookup_done(dentry); 3188384f26e2SAl Viro if (!error) { 318964e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 31906fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 31916fb968cdSAl Viro dput(dentry); 31926fb968cdSAl Viro dentry = dget(file->f_path.dentry); 31936fb968cdSAl Viro } 319464e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 31952675a4ebSAl Viro error = -EIO; 3196384f26e2SAl Viro } else { 319730d90494SAl Viro if (file->f_path.dentry) { 3198d18e9008SMiklos Szeredi dput(dentry); 3199d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 320010c64ceaSAl Viro } 3201239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3202a01e718fSAl Viro error = -ENOENT; 3203d18e9008SMiklos Szeredi } 3204d18e9008SMiklos Szeredi } 3205239eb983SAl Viro if (error) { 3206d18e9008SMiklos Szeredi dput(dentry); 3207239eb983SAl Viro dentry = ERR_PTR(error); 3208239eb983SAl Viro } 3209239eb983SAl Viro return dentry; 3210d18e9008SMiklos Szeredi } 3211d18e9008SMiklos Szeredi 321231e6b01fSNick Piggin /* 32131acf0af9SDavid Howells * Look up and maybe create and open the last component. 3214d58ffd35SMiklos Szeredi * 321500a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3216d58ffd35SMiklos Szeredi * 321700a07c15SAl Viro * Returns 0 on success, that is, if 321800a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 321900a07c15SAl Viro * the file was not completely opened at this time, though lookups and 322000a07c15SAl Viro * creations were performed. 322100a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 322200a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 322300a07c15SAl Viro * hadn't been specified. 32241acf0af9SDavid Howells * 322500a07c15SAl Viro * An error code is returned on failure. 3226d58ffd35SMiklos Szeredi */ 3227da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3228d58ffd35SMiklos Szeredi const struct open_flags *op, 32293ec2eef1SAl Viro bool got_write) 3230d58ffd35SMiklos Szeredi { 3231549c7297SChristian Brauner struct user_namespace *mnt_userns; 3232d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 323354ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 32341643b43fSAl Viro int open_flag = op->open_flag; 3235d58ffd35SMiklos Szeredi struct dentry *dentry; 32361643b43fSAl Viro int error, create_error = 0; 32371643b43fSAl Viro umode_t mode = op->mode; 32386fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3239d58ffd35SMiklos Szeredi 3240ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3241da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3242d58ffd35SMiklos Szeredi 324373a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 32446fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 32456fbd0714SAl Viro for (;;) { 32466fbd0714SAl Viro if (!dentry) { 32476fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3248d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3249da5ebf5aSAl Viro return dentry; 32506fbd0714SAl Viro } 32516fbd0714SAl Viro if (d_in_lookup(dentry)) 32526fbd0714SAl Viro break; 3253d58ffd35SMiklos Szeredi 32546fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 32556fbd0714SAl Viro if (likely(error > 0)) 32566fbd0714SAl Viro break; 32576fbd0714SAl Viro if (error) 32586fbd0714SAl Viro goto out_dput; 32596fbd0714SAl Viro d_invalidate(dentry); 32606fbd0714SAl Viro dput(dentry); 32616fbd0714SAl Viro dentry = NULL; 32626fbd0714SAl Viro } 32636fbd0714SAl Viro if (dentry->d_inode) { 3264d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3265da5ebf5aSAl Viro return dentry; 32666c51e513SAl Viro } 3267d18e9008SMiklos Szeredi 32681643b43fSAl Viro /* 32691643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 32701643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 32711643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 32721643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 32731643b43fSAl Viro * 32741643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 32751643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 32761643b43fSAl Viro */ 327799a4a90cSAl Viro if (unlikely(!got_write)) 327899a4a90cSAl Viro open_flag &= ~O_TRUNC; 3279549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 32801643b43fSAl Viro if (open_flag & O_CREAT) { 328199a4a90cSAl Viro if (open_flag & O_EXCL) 328299a4a90cSAl Viro open_flag &= ~O_TRUNC; 32831643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 32841643b43fSAl Viro mode &= ~current_umask(); 328599a4a90cSAl Viro if (likely(got_write)) 3286549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3287ba73d987SChristian Brauner dentry, mode); 328899a4a90cSAl Viro else 328999a4a90cSAl Viro create_error = -EROFS; 329099a4a90cSAl Viro } 329199a4a90cSAl Viro if (create_error) 32921643b43fSAl Viro open_flag &= ~O_CREAT; 32931643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3294d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3295da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3296da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3297da5ebf5aSAl Viro return dentry; 3298239eb983SAl Viro } 329954ef4872SMiklos Szeredi 33006fbd0714SAl Viro if (d_in_lookup(dentry)) { 330112fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 330212fa5e24SAl Viro nd->flags); 33036fbd0714SAl Viro d_lookup_done(dentry); 330412fa5e24SAl Viro if (unlikely(res)) { 330512fa5e24SAl Viro if (IS_ERR(res)) { 330612fa5e24SAl Viro error = PTR_ERR(res); 330712fa5e24SAl Viro goto out_dput; 330812fa5e24SAl Viro } 330912fa5e24SAl Viro dput(dentry); 331012fa5e24SAl Viro dentry = res; 331112fa5e24SAl Viro } 331254ef4872SMiklos Szeredi } 331354ef4872SMiklos Szeredi 3314d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 33151643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 331673a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3317ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3318ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3319ce8644fcSAl Viro error = -EACCES; 3320d58ffd35SMiklos Szeredi goto out_dput; 332164894cf8SAl Viro } 3322549c7297SChristian Brauner 3323549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3324549c7297SChristian Brauner mode, open_flag & O_EXCL); 3325d58ffd35SMiklos Szeredi if (error) 3326d58ffd35SMiklos Szeredi goto out_dput; 3327d58ffd35SMiklos Szeredi } 33281643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 33291643b43fSAl Viro error = create_error; 3330d58ffd35SMiklos Szeredi goto out_dput; 3331d58ffd35SMiklos Szeredi } 3332da5ebf5aSAl Viro return dentry; 3333d58ffd35SMiklos Szeredi 3334d58ffd35SMiklos Szeredi out_dput: 3335d58ffd35SMiklos Szeredi dput(dentry); 3336da5ebf5aSAl Viro return ERR_PTR(error); 3337d58ffd35SMiklos Szeredi } 3338d58ffd35SMiklos Szeredi 3339c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 33403ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3341fb1cc555SAl Viro { 3342a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3343ca344a89SAl Viro int open_flag = op->open_flag; 334464894cf8SAl Viro bool got_write = false; 3345254cf582SAl Viro unsigned seq; 3346a1eb3315SMiklos Szeredi struct inode *inode; 3347da5ebf5aSAl Viro struct dentry *dentry; 3348b0417d2cSAl Viro const char *res; 3349fb1cc555SAl Viro 3350c3e380b0SAl Viro nd->flags |= op->intent; 3351c3e380b0SAl Viro 3352bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 335356676ec3SAl Viro if (nd->depth) 335456676ec3SAl Viro put_link(nd); 3355ff326a32SAl Viro return handle_dots(nd, nd->last_type); 33561f36f774SAl Viro } 3357a2c36b45SAl Viro 3358ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3359fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3360fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3361fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 336220e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 336320e34357SAl Viro if (IS_ERR(dentry)) 33641ccac622SAl Viro return ERR_CAST(dentry); 336520e34357SAl Viro if (likely(dentry)) 336671574865SMiklos Szeredi goto finish_lookup; 336771574865SMiklos Szeredi 33686583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3369b6183df7SMiklos Szeredi } else { 3370fe2d35ffSAl Viro /* create side of things */ 337172287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3372e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3373e36cffedSJens Axboe return ERR_PTR(-ECHILD); 337472287417SAl Viro } 3375c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 33761f36f774SAl Viro /* trailing slashes? */ 3377deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 33781ccac622SAl Viro return ERR_PTR(-EISDIR); 3379b6183df7SMiklos Szeredi } 33801f36f774SAl Viro 33819cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3382e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 338364894cf8SAl Viro /* 338464894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 338564894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 338664894cf8SAl Viro * dropping this one anyway. 338764894cf8SAl Viro */ 338864894cf8SAl Viro } 33899cf843e3SAl Viro if (open_flag & O_CREAT) 33905955102cSAl Viro inode_lock(dir->d_inode); 33919cf843e3SAl Viro else 33929cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3393da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3394f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3395f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 33969cf843e3SAl Viro if (open_flag & O_CREAT) 33975955102cSAl Viro inode_unlock(dir->d_inode); 33989cf843e3SAl Viro else 33999cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3400fb1cc555SAl Viro 3401c981a482SAl Viro if (got_write) 340259e96e65SAl Viro mnt_drop_write(nd->path.mnt); 34036c0d46c4SAl Viro 340459e96e65SAl Viro if (IS_ERR(dentry)) 340559e96e65SAl Viro return ERR_CAST(dentry); 340659e96e65SAl Viro 3407973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3408e73cabffSAl Viro dput(nd->path.dentry); 3409e73cabffSAl Viro nd->path.dentry = dentry; 3410c981a482SAl Viro return NULL; 3411fb1cc555SAl Viro } 3412fb1cc555SAl Viro 341320e34357SAl Viro finish_lookup: 341456676ec3SAl Viro if (nd->depth) 341556676ec3SAl Viro put_link(nd); 34168c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3417ff326a32SAl Viro if (unlikely(res)) 34181ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3419b0417d2cSAl Viro return res; 34201ccac622SAl Viro } 342131d1726dSAl Viro 3422c981a482SAl Viro /* 3423c981a482SAl Viro * Handle the last step of open() 3424c981a482SAl Viro */ 3425c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3426c981a482SAl Viro struct file *file, const struct open_flags *op) 3427c981a482SAl Viro { 3428549c7297SChristian Brauner struct user_namespace *mnt_userns; 3429c981a482SAl Viro int open_flag = op->open_flag; 3430c981a482SAl Viro bool do_truncate; 3431c981a482SAl Viro int acc_mode; 3432c981a482SAl Viro int error; 3433c981a482SAl Viro 3434ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3435ff326a32SAl Viro error = complete_walk(nd); 3436ff326a32SAl Viro if (error) 3437ff326a32SAl Viro return error; 3438ff326a32SAl Viro } 3439973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 344076ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3441549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 344230aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3443b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3444b94e0b32SAl Viro return -EEXIST; 344530aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3446c5971b8cSAl Viro return -EISDIR; 3447549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 344830aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 344930aba665SSalvatore Mesoraca if (unlikely(error)) 3450c5971b8cSAl Viro return error; 345130aba665SSalvatore Mesoraca } 345244b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3453c5971b8cSAl Viro return -ENOTDIR; 34546c0d46c4SAl Viro 34558795e7d4SAl Viro do_truncate = false; 34568795e7d4SAl Viro acc_mode = op->acc_mode; 34575a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 34585a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 34595a2d3eddSAl Viro open_flag &= ~O_TRUNC; 34605a2d3eddSAl Viro acc_mode = 0; 34618795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 34620f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 34630f9d1a10SAl Viro if (error) 3464c5971b8cSAl Viro return error; 34658795e7d4SAl Viro do_truncate = true; 34660f9d1a10SAl Viro } 3467549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 34688795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3469ae2bb293SAl Viro error = vfs_open(&nd->path, file); 34708795e7d4SAl Viro if (!error) 34716035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 34728795e7d4SAl Viro if (!error && do_truncate) 3473549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3474c80567c8SAl Viro if (unlikely(error > 0)) { 3475c80567c8SAl Viro WARN_ON(1); 3476c80567c8SAl Viro error = -EINVAL; 3477c80567c8SAl Viro } 34788795e7d4SAl Viro if (do_truncate) 34790f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3480c5971b8cSAl Viro return error; 3481fb1cc555SAl Viro } 3482fb1cc555SAl Viro 34836521f891SChristian Brauner /** 34846521f891SChristian Brauner * vfs_tmpfile - create tmpfile 34856521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 34866521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 34876521f891SChristian Brauner * @mode: mode of the new tmpfile 34882111c3c0SRandy Dunlap * @open_flag: flags 34896521f891SChristian Brauner * 34906521f891SChristian Brauner * Create a temporary file. 34916521f891SChristian Brauner * 34926521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 34936521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 34946521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 34956521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 34966521f891SChristian Brauner * raw inode simply passs init_user_ns. 34976521f891SChristian Brauner */ 34986521f891SChristian Brauner struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, 34996521f891SChristian Brauner struct dentry *dentry, umode_t mode, int open_flag) 3500af7bd4dcSAmir Goldstein { 3501af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3502af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3503af7bd4dcSAmir Goldstein struct inode *inode; 3504af7bd4dcSAmir Goldstein int error; 3505af7bd4dcSAmir Goldstein 3506af7bd4dcSAmir Goldstein /* we want directory to be writable */ 35076521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3508af7bd4dcSAmir Goldstein if (error) 3509af7bd4dcSAmir Goldstein goto out_err; 3510af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3511af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3512af7bd4dcSAmir Goldstein goto out_err; 3513af7bd4dcSAmir Goldstein error = -ENOMEM; 3514cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3515af7bd4dcSAmir Goldstein if (unlikely(!child)) 3516af7bd4dcSAmir Goldstein goto out_err; 3517549c7297SChristian Brauner error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); 3518af7bd4dcSAmir Goldstein if (error) 3519af7bd4dcSAmir Goldstein goto out_err; 3520af7bd4dcSAmir Goldstein error = -ENOENT; 3521af7bd4dcSAmir Goldstein inode = child->d_inode; 3522af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3523af7bd4dcSAmir Goldstein goto out_err; 3524af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3525af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3526af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3527af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3528af7bd4dcSAmir Goldstein } 3529a2d2329eSChristian Brauner ima_post_create_tmpfile(mnt_userns, inode); 3530af7bd4dcSAmir Goldstein return child; 3531af7bd4dcSAmir Goldstein 3532af7bd4dcSAmir Goldstein out_err: 3533af7bd4dcSAmir Goldstein dput(child); 3534af7bd4dcSAmir Goldstein return ERR_PTR(error); 3535af7bd4dcSAmir Goldstein } 3536af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3537af7bd4dcSAmir Goldstein 3538c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 353960545d0dSAl Viro const struct open_flags *op, 35403ec2eef1SAl Viro struct file *file) 354160545d0dSAl Viro { 35426521f891SChristian Brauner struct user_namespace *mnt_userns; 3543625b6d10SAl Viro struct dentry *child; 3544625b6d10SAl Viro struct path path; 3545c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 354660545d0dSAl Viro if (unlikely(error)) 354760545d0dSAl Viro return error; 3548625b6d10SAl Viro error = mnt_want_write(path.mnt); 354960545d0dSAl Viro if (unlikely(error)) 355060545d0dSAl Viro goto out; 35516521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 35526521f891SChristian Brauner child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); 3553af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3554684e73beSHirofumi Nakagawa if (IS_ERR(child)) 355560545d0dSAl Viro goto out2; 3556625b6d10SAl Viro dput(path.dentry); 3557625b6d10SAl Viro path.dentry = child; 3558c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 355969a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3560549c7297SChristian Brauner error = may_open(mnt_userns, &path, 0, op->open_flag); 35611e8f44f1SAl Viro if (!error) 35621e8f44f1SAl Viro error = vfs_open(&path, file); 356360545d0dSAl Viro out2: 3564625b6d10SAl Viro mnt_drop_write(path.mnt); 356560545d0dSAl Viro out: 3566625b6d10SAl Viro path_put(&path); 356760545d0dSAl Viro return error; 356860545d0dSAl Viro } 356960545d0dSAl Viro 35706ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 35716ac08709SAl Viro { 35726ac08709SAl Viro struct path path; 35736ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 35746ac08709SAl Viro if (!error) { 35756ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3576ae2bb293SAl Viro error = vfs_open(&path, file); 35776ac08709SAl Viro path_put(&path); 35786ac08709SAl Viro } 35796ac08709SAl Viro return error; 35806ac08709SAl Viro } 35816ac08709SAl Viro 3582c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3583c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 35841da177e4SLinus Torvalds { 358530d90494SAl Viro struct file *file; 358613aab428SAl Viro int error; 358731e6b01fSNick Piggin 3588ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 35891afc99beSAl Viro if (IS_ERR(file)) 35901afc99beSAl Viro return file; 359131e6b01fSNick Piggin 3592bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 35933ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 35945f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 35956ac08709SAl Viro error = do_o_path(nd, flags, file); 35965f336e72SAl Viro } else { 35975f336e72SAl Viro const char *s = path_init(nd, flags); 35983bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3599c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 36001ccac622SAl Viro ; 3601c5971b8cSAl Viro if (!error) 3602c5971b8cSAl Viro error = do_open(nd, file, op); 3603deb106c6SAl Viro terminate_walk(nd); 36045f336e72SAl Viro } 36057c1c01ecSAl Viro if (likely(!error)) { 3606aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 36077c1c01ecSAl Viro return file; 36087c1c01ecSAl Viro WARN_ON(1); 36097c1c01ecSAl Viro error = -EINVAL; 3610015c3bbcSMiklos Szeredi } 36117c1c01ecSAl Viro fput(file); 36122675a4ebSAl Viro if (error == -EOPENSTALE) { 36132675a4ebSAl Viro if (flags & LOOKUP_RCU) 36142675a4ebSAl Viro error = -ECHILD; 36152675a4ebSAl Viro else 36162675a4ebSAl Viro error = -ESTALE; 36172675a4ebSAl Viro } 36187c1c01ecSAl Viro return ERR_PTR(error); 3619de459215SKirill Korotaev } 36201da177e4SLinus Torvalds 3621669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3622f9652e10SAl Viro const struct open_flags *op) 362313aab428SAl Viro { 36249883d185SAl Viro struct nameidata nd; 3625f9652e10SAl Viro int flags = op->lookup_flags; 362613aab428SAl Viro struct file *filp; 362713aab428SAl Viro 362806422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3629c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 363013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3631c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 363213aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3633c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 36349883d185SAl Viro restore_nameidata(); 363513aab428SAl Viro return filp; 363613aab428SAl Viro } 363713aab428SAl Viro 3638ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3639f9652e10SAl Viro const char *name, const struct open_flags *op) 364073d049a4SAl Viro { 36419883d185SAl Viro struct nameidata nd; 364273d049a4SAl Viro struct file *file; 364351689104SPaul Moore struct filename *filename; 3644bcba1e7dSAl Viro int flags = op->lookup_flags; 364573d049a4SAl Viro 3646ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 364773d049a4SAl Viro return ERR_PTR(-ELOOP); 364873d049a4SAl Viro 364951689104SPaul Moore filename = getname_kernel(name); 3650a1c83681SViresh Kumar if (IS_ERR(filename)) 365151689104SPaul Moore return ERR_CAST(filename); 365251689104SPaul Moore 365306422964SAl Viro set_nameidata(&nd, -1, filename, root); 3654c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 365573d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3656c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 365773d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3658c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 36599883d185SAl Viro restore_nameidata(); 366051689104SPaul Moore putname(filename); 366173d049a4SAl Viro return file; 366273d049a4SAl Viro } 366373d049a4SAl Viro 3664b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 36651ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 36661da177e4SLinus Torvalds { 3667c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3668391172c4SAl Viro struct qstr last; 3669391172c4SAl Viro int type; 3670c30dabfeSJan Kara int err2; 36711ac12b4bSJeff Layton int error; 36721ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 36731ac12b4bSJeff Layton 36741ac12b4bSJeff Layton /* 36751ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 36761ac12b4bSJeff Layton * other flags passed in are ignored! 36771ac12b4bSJeff Layton */ 36781ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 36791ac12b4bSJeff Layton 3680c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 36810ee50b47SDmitry Kadashev if (error) 36820ee50b47SDmitry Kadashev return ERR_PTR(error); 36831da177e4SLinus Torvalds 3684c663e5d8SChristoph Hellwig /* 3685c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3686c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3687c663e5d8SChristoph Hellwig */ 36885c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3689ed75e95dSAl Viro goto out; 3690c663e5d8SChristoph Hellwig 3691c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3692391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3693c663e5d8SChristoph Hellwig /* 3694c663e5d8SChristoph Hellwig * Do the final lookup. 3695c663e5d8SChristoph Hellwig */ 3696391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36975955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3698391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36991da177e4SLinus Torvalds if (IS_ERR(dentry)) 3700a8104a9fSAl Viro goto unlock; 3701c663e5d8SChristoph Hellwig 3702a8104a9fSAl Viro error = -EEXIST; 3703b18825a7SDavid Howells if (d_is_positive(dentry)) 3704a8104a9fSAl Viro goto fail; 3705b18825a7SDavid Howells 3706c663e5d8SChristoph Hellwig /* 3707c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3708c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3709c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3710c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3711c663e5d8SChristoph Hellwig */ 3712391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3713a8104a9fSAl Viro error = -ENOENT; 3714ed75e95dSAl Viro goto fail; 3715e9baf6e5SAl Viro } 3716c30dabfeSJan Kara if (unlikely(err2)) { 3717c30dabfeSJan Kara error = err2; 3718a8104a9fSAl Viro goto fail; 3719c30dabfeSJan Kara } 3720e9baf6e5SAl Viro return dentry; 37211da177e4SLinus Torvalds fail: 3722a8104a9fSAl Viro dput(dentry); 3723a8104a9fSAl Viro dentry = ERR_PTR(error); 3724a8104a9fSAl Viro unlock: 37255955102cSAl Viro inode_unlock(path->dentry->d_inode); 3726c30dabfeSJan Kara if (!err2) 3727391172c4SAl Viro mnt_drop_write(path->mnt); 3728ed75e95dSAl Viro out: 3729391172c4SAl Viro path_put(path); 3730ed75e95dSAl Viro return dentry; 3731dae6ad8fSAl Viro } 3732fa14a0b8SAl Viro 3733fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3734fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3735fa14a0b8SAl Viro { 3736b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3737b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3738b4a4f213SStephen Brennan 3739b4a4f213SStephen Brennan putname(filename); 3740b4a4f213SStephen Brennan return res; 3741fa14a0b8SAl Viro } 3742dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3743dae6ad8fSAl Viro 3744921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3745921a1650SAl Viro { 3746921a1650SAl Viro dput(dentry); 37475955102cSAl Viro inode_unlock(path->dentry->d_inode); 3748a8104a9fSAl Viro mnt_drop_write(path->mnt); 3749921a1650SAl Viro path_put(path); 3750921a1650SAl Viro } 3751921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3752921a1650SAl Viro 3753520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 37541ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3755dae6ad8fSAl Viro { 3756b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3757b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3758b4a4f213SStephen Brennan 3759b4a4f213SStephen Brennan putname(filename); 3760b4a4f213SStephen Brennan return res; 3761dae6ad8fSAl Viro } 3762dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3763dae6ad8fSAl Viro 37646521f891SChristian Brauner /** 37656521f891SChristian Brauner * vfs_mknod - create device node or file 37666521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 37676521f891SChristian Brauner * @dir: inode of @dentry 37686521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 37696521f891SChristian Brauner * @mode: mode of the new device node or file 37706521f891SChristian Brauner * @dev: device number of device to create 37716521f891SChristian Brauner * 37726521f891SChristian Brauner * Create a device node or file. 37736521f891SChristian Brauner * 37746521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 37756521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 37766521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 37776521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 37786521f891SChristian Brauner * raw inode simply passs init_user_ns. 37796521f891SChristian Brauner */ 37806521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 37816521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 37821da177e4SLinus Torvalds { 3783a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 37846521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 37851da177e4SLinus Torvalds 37861da177e4SLinus Torvalds if (error) 37871da177e4SLinus Torvalds return error; 37881da177e4SLinus Torvalds 3789a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3790a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 37911da177e4SLinus Torvalds return -EPERM; 37921da177e4SLinus Torvalds 3793acfa4380SAl Viro if (!dir->i_op->mknod) 37941da177e4SLinus Torvalds return -EPERM; 37951da177e4SLinus Torvalds 379608ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 379708ce5f16SSerge E. Hallyn if (error) 379808ce5f16SSerge E. Hallyn return error; 379908ce5f16SSerge E. Hallyn 38001da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 38011da177e4SLinus Torvalds if (error) 38021da177e4SLinus Torvalds return error; 38031da177e4SLinus Torvalds 3804549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3805a74574aaSStephen Smalley if (!error) 3806f38aa942SAmy Griffis fsnotify_create(dir, dentry); 38071da177e4SLinus Torvalds return error; 38081da177e4SLinus Torvalds } 38094d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 38101da177e4SLinus Torvalds 3811f69aac00SAl Viro static int may_mknod(umode_t mode) 3812463c3197SDave Hansen { 3813463c3197SDave Hansen switch (mode & S_IFMT) { 3814463c3197SDave Hansen case S_IFREG: 3815463c3197SDave Hansen case S_IFCHR: 3816463c3197SDave Hansen case S_IFBLK: 3817463c3197SDave Hansen case S_IFIFO: 3818463c3197SDave Hansen case S_IFSOCK: 3819463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3820463c3197SDave Hansen return 0; 3821463c3197SDave Hansen case S_IFDIR: 3822463c3197SDave Hansen return -EPERM; 3823463c3197SDave Hansen default: 3824463c3197SDave Hansen return -EINVAL; 3825463c3197SDave Hansen } 3826463c3197SDave Hansen } 3827463c3197SDave Hansen 382845f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 382987c4e192SDominik Brodowski unsigned int dev) 38301da177e4SLinus Torvalds { 38316521f891SChristian Brauner struct user_namespace *mnt_userns; 38321da177e4SLinus Torvalds struct dentry *dentry; 3833dae6ad8fSAl Viro struct path path; 3834dae6ad8fSAl Viro int error; 3835972567f1SJeff Layton unsigned int lookup_flags = 0; 38361da177e4SLinus Torvalds 38378e4bfca1SAl Viro error = may_mknod(mode); 38388e4bfca1SAl Viro if (error) 38397797251bSDmitry Kadashev goto out1; 3840972567f1SJeff Layton retry: 3841b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 38427797251bSDmitry Kadashev error = PTR_ERR(dentry); 3843dae6ad8fSAl Viro if (IS_ERR(dentry)) 38447797251bSDmitry Kadashev goto out1; 38452ad94ae6SAl Viro 3846dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3847ce3b0f8dSAl Viro mode &= ~current_umask(); 3848dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3849be6d3e56SKentaro Takeda if (error) 38507797251bSDmitry Kadashev goto out2; 38516521f891SChristian Brauner 38526521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 38531da177e4SLinus Torvalds switch (mode & S_IFMT) { 38541da177e4SLinus Torvalds case 0: case S_IFREG: 38556521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 38566521f891SChristian Brauner dentry, mode, true); 385705d1a717SMimi Zohar if (!error) 3858a2d2329eSChristian Brauner ima_post_path_mknod(mnt_userns, dentry); 38591da177e4SLinus Torvalds break; 38601da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 38616521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 38626521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 38631da177e4SLinus Torvalds break; 38641da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 38656521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 38666521f891SChristian Brauner dentry, mode, 0); 38671da177e4SLinus Torvalds break; 38681da177e4SLinus Torvalds } 38697797251bSDmitry Kadashev out2: 3870921a1650SAl Viro done_path_create(&path, dentry); 3871972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3872972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3873972567f1SJeff Layton goto retry; 3874972567f1SJeff Layton } 38757797251bSDmitry Kadashev out1: 38767797251bSDmitry Kadashev putname(name); 38771da177e4SLinus Torvalds return error; 38781da177e4SLinus Torvalds } 38791da177e4SLinus Torvalds 388087c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 388187c4e192SDominik Brodowski unsigned int, dev) 388287c4e192SDominik Brodowski { 38837797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 388487c4e192SDominik Brodowski } 388587c4e192SDominik Brodowski 38868208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 38875590ff0dSUlrich Drepper { 38887797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 38895590ff0dSUlrich Drepper } 38905590ff0dSUlrich Drepper 38916521f891SChristian Brauner /** 38926521f891SChristian Brauner * vfs_mkdir - create directory 38936521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 38946521f891SChristian Brauner * @dir: inode of @dentry 38956521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 38966521f891SChristian Brauner * @mode: mode of the new directory 38976521f891SChristian Brauner * 38986521f891SChristian Brauner * Create a directory. 38996521f891SChristian Brauner * 39006521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39016521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39026521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39036521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39046521f891SChristian Brauner * raw inode simply passs init_user_ns. 39056521f891SChristian Brauner */ 39066521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 39076521f891SChristian Brauner struct dentry *dentry, umode_t mode) 39081da177e4SLinus Torvalds { 39096521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 39108de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 39111da177e4SLinus Torvalds 39121da177e4SLinus Torvalds if (error) 39131da177e4SLinus Torvalds return error; 39141da177e4SLinus Torvalds 3915acfa4380SAl Viro if (!dir->i_op->mkdir) 39161da177e4SLinus Torvalds return -EPERM; 39171da177e4SLinus Torvalds 39181da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 39191da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 39201da177e4SLinus Torvalds if (error) 39211da177e4SLinus Torvalds return error; 39221da177e4SLinus Torvalds 39238de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 39248de52778SAl Viro return -EMLINK; 39258de52778SAl Viro 3926549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 3927a74574aaSStephen Smalley if (!error) 3928f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 39291da177e4SLinus Torvalds return error; 39301da177e4SLinus Torvalds } 39314d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 39321da177e4SLinus Torvalds 393345f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 39341da177e4SLinus Torvalds { 39356902d925SDave Hansen struct dentry *dentry; 3936dae6ad8fSAl Viro struct path path; 3937dae6ad8fSAl Viro int error; 3938b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 39391da177e4SLinus Torvalds 3940b76d8b82SJeff Layton retry: 3941b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 3942584d3226SDmitry Kadashev error = PTR_ERR(dentry); 39436902d925SDave Hansen if (IS_ERR(dentry)) 3944584d3226SDmitry Kadashev goto out_putname; 39456902d925SDave Hansen 3946dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3947ce3b0f8dSAl Viro mode &= ~current_umask(); 3948dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 39496521f891SChristian Brauner if (!error) { 39506521f891SChristian Brauner struct user_namespace *mnt_userns; 39516521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 3952549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 3953549c7297SChristian Brauner mode); 39546521f891SChristian Brauner } 3955921a1650SAl Viro done_path_create(&path, dentry); 3956b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3957b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3958b76d8b82SJeff Layton goto retry; 3959b76d8b82SJeff Layton } 3960584d3226SDmitry Kadashev out_putname: 3961584d3226SDmitry Kadashev putname(name); 39621da177e4SLinus Torvalds return error; 39631da177e4SLinus Torvalds } 39641da177e4SLinus Torvalds 39650101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 39660101db7aSDominik Brodowski { 3967584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 39680101db7aSDominik Brodowski } 39690101db7aSDominik Brodowski 3970a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 39715590ff0dSUlrich Drepper { 3972584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 39735590ff0dSUlrich Drepper } 39745590ff0dSUlrich Drepper 39756521f891SChristian Brauner /** 39766521f891SChristian Brauner * vfs_rmdir - remove directory 39776521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 39786521f891SChristian Brauner * @dir: inode of @dentry 39796521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39806521f891SChristian Brauner * 39816521f891SChristian Brauner * Remove a directory. 39826521f891SChristian Brauner * 39836521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 39846521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 39856521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 39866521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 39876521f891SChristian Brauner * raw inode simply passs init_user_ns. 39886521f891SChristian Brauner */ 39896521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 39906521f891SChristian Brauner struct dentry *dentry) 39911da177e4SLinus Torvalds { 39926521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 39931da177e4SLinus Torvalds 39941da177e4SLinus Torvalds if (error) 39951da177e4SLinus Torvalds return error; 39961da177e4SLinus Torvalds 3997acfa4380SAl Viro if (!dir->i_op->rmdir) 39981da177e4SLinus Torvalds return -EPERM; 39991da177e4SLinus Torvalds 40001d2ef590SAl Viro dget(dentry); 40015955102cSAl Viro inode_lock(dentry->d_inode); 4002912dbc15SSage Weil 40031da177e4SLinus Torvalds error = -EBUSY; 40041bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 40051bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4006912dbc15SSage Weil goto out; 4007912dbc15SSage Weil 40081da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4009912dbc15SSage Weil if (error) 4010912dbc15SSage Weil goto out; 4011912dbc15SSage Weil 40121da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4013912dbc15SSage Weil if (error) 4014912dbc15SSage Weil goto out; 4015912dbc15SSage Weil 40168767712fSAl Viro shrink_dcache_parent(dentry); 40171da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4018d83c49f3SAl Viro dont_mount(dentry); 40198ed936b5SEric W. Biederman detach_mounts(dentry); 40201da177e4SLinus Torvalds 4021912dbc15SSage Weil out: 40225955102cSAl Viro inode_unlock(dentry->d_inode); 40231d2ef590SAl Viro dput(dentry); 4024912dbc15SSage Weil if (!error) 4025a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 40261da177e4SLinus Torvalds return error; 40271da177e4SLinus Torvalds } 40284d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 40291da177e4SLinus Torvalds 403045f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 40311da177e4SLinus Torvalds { 40326521f891SChristian Brauner struct user_namespace *mnt_userns; 40330ee50b47SDmitry Kadashev int error; 40341da177e4SLinus Torvalds struct dentry *dentry; 4035f5beed75SAl Viro struct path path; 4036f5beed75SAl Viro struct qstr last; 4037f5beed75SAl Viro int type; 4038c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4039c6ee9206SJeff Layton retry: 4040c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 40410ee50b47SDmitry Kadashev if (error) 40420ee50b47SDmitry Kadashev goto exit1; 40431da177e4SLinus Torvalds 4044f5beed75SAl Viro switch (type) { 40451da177e4SLinus Torvalds case LAST_DOTDOT: 40461da177e4SLinus Torvalds error = -ENOTEMPTY; 40470ee50b47SDmitry Kadashev goto exit2; 40481da177e4SLinus Torvalds case LAST_DOT: 40491da177e4SLinus Torvalds error = -EINVAL; 40500ee50b47SDmitry Kadashev goto exit2; 40511da177e4SLinus Torvalds case LAST_ROOT: 40521da177e4SLinus Torvalds error = -EBUSY; 40530ee50b47SDmitry Kadashev goto exit2; 40541da177e4SLinus Torvalds } 40550612d9fbSOGAWA Hirofumi 4056f5beed75SAl Viro error = mnt_want_write(path.mnt); 4057c30dabfeSJan Kara if (error) 40580ee50b47SDmitry Kadashev goto exit2; 40590612d9fbSOGAWA Hirofumi 40605955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4061f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40621da177e4SLinus Torvalds error = PTR_ERR(dentry); 40636902d925SDave Hansen if (IS_ERR(dentry)) 40640ee50b47SDmitry Kadashev goto exit3; 4065e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4066e6bc45d6STheodore Ts'o error = -ENOENT; 40670ee50b47SDmitry Kadashev goto exit4; 4068e6bc45d6STheodore Ts'o } 4069f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4070be6d3e56SKentaro Takeda if (error) 40710ee50b47SDmitry Kadashev goto exit4; 40726521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 40736521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 40740ee50b47SDmitry Kadashev exit4: 40751da177e4SLinus Torvalds dput(dentry); 40760ee50b47SDmitry Kadashev exit3: 40775955102cSAl Viro inode_unlock(path.dentry->d_inode); 4078f5beed75SAl Viro mnt_drop_write(path.mnt); 40790ee50b47SDmitry Kadashev exit2: 4080f5beed75SAl Viro path_put(&path); 4081c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4082c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4083c6ee9206SJeff Layton goto retry; 4084c6ee9206SJeff Layton } 40850ee50b47SDmitry Kadashev exit1: 408624fb33d4SAl Viro putname(name); 40871da177e4SLinus Torvalds return error; 40881da177e4SLinus Torvalds } 40891da177e4SLinus Torvalds 40903cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 40915590ff0dSUlrich Drepper { 4092e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 40935590ff0dSUlrich Drepper } 40945590ff0dSUlrich Drepper 4095b21996e3SJ. Bruce Fields /** 4096b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 40976521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 4098b21996e3SJ. Bruce Fields * @dir: parent directory 4099b21996e3SJ. Bruce Fields * @dentry: victim 4100b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4101b21996e3SJ. Bruce Fields * 4102b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4103b21996e3SJ. Bruce Fields * 4104b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4105b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4106b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4107b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4108b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4109b21996e3SJ. Bruce Fields * 4110b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4111b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4112b21996e3SJ. Bruce Fields * to be NFS exported. 41136521f891SChristian Brauner * 41146521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 41156521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 41166521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 41176521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 41186521f891SChristian Brauner * raw inode simply passs init_user_ns. 4119b21996e3SJ. Bruce Fields */ 41206521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 41216521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 41221da177e4SLinus Torvalds { 41239accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 41246521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 41251da177e4SLinus Torvalds 41261da177e4SLinus Torvalds if (error) 41271da177e4SLinus Torvalds return error; 41281da177e4SLinus Torvalds 4129acfa4380SAl Viro if (!dir->i_op->unlink) 41301da177e4SLinus Torvalds return -EPERM; 41311da177e4SLinus Torvalds 41325955102cSAl Viro inode_lock(target); 413351cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 413451cc3a66SHugh Dickins error = -EPERM; 413551cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 41361da177e4SLinus Torvalds error = -EBUSY; 41371da177e4SLinus Torvalds else { 41381da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4139bec1052eSAl Viro if (!error) { 41405a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 41415a14696cSJ. Bruce Fields if (error) 4142b21996e3SJ. Bruce Fields goto out; 41431da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 41448ed936b5SEric W. Biederman if (!error) { 4145d83c49f3SAl Viro dont_mount(dentry); 41468ed936b5SEric W. Biederman detach_mounts(dentry); 41478ed936b5SEric W. Biederman } 4148bec1052eSAl Viro } 41491da177e4SLinus Torvalds } 4150b21996e3SJ. Bruce Fields out: 41515955102cSAl Viro inode_unlock(target); 41521da177e4SLinus Torvalds 41531da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4154a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4155a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4156a37d9a17SAmir Goldstein } else if (!error) { 41579accbb97SJ. Bruce Fields fsnotify_link_count(target); 4158a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 41591da177e4SLinus Torvalds } 41600eeca283SRobert Love 41611da177e4SLinus Torvalds return error; 41621da177e4SLinus Torvalds } 41634d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 41641da177e4SLinus Torvalds 41651da177e4SLinus Torvalds /* 41661da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 41671b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 41681da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 41691da177e4SLinus Torvalds * while waiting on the I/O. 41701da177e4SLinus Torvalds */ 417145f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 41721da177e4SLinus Torvalds { 41732ad94ae6SAl Viro int error; 41741da177e4SLinus Torvalds struct dentry *dentry; 4175f5beed75SAl Viro struct path path; 4176f5beed75SAl Viro struct qstr last; 4177f5beed75SAl Viro int type; 41781da177e4SLinus Torvalds struct inode *inode = NULL; 4179b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 41805d18f813SJeff Layton unsigned int lookup_flags = 0; 41815d18f813SJeff Layton retry: 4182c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 41830ee50b47SDmitry Kadashev if (error) 41840ee50b47SDmitry Kadashev goto exit1; 41852ad94ae6SAl Viro 41861da177e4SLinus Torvalds error = -EISDIR; 4187f5beed75SAl Viro if (type != LAST_NORM) 41880ee50b47SDmitry Kadashev goto exit2; 41890612d9fbSOGAWA Hirofumi 4190f5beed75SAl Viro error = mnt_want_write(path.mnt); 4191c30dabfeSJan Kara if (error) 41920ee50b47SDmitry Kadashev goto exit2; 4193b21996e3SJ. Bruce Fields retry_deleg: 41945955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4195f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 41961da177e4SLinus Torvalds error = PTR_ERR(dentry); 41971da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 41986521f891SChristian Brauner struct user_namespace *mnt_userns; 41996521f891SChristian Brauner 42001da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4201f5beed75SAl Viro if (last.name[last.len]) 420250338b88STörök Edwin goto slashes; 42031da177e4SLinus Torvalds inode = dentry->d_inode; 4204b18825a7SDavid Howells if (d_is_negative(dentry)) 4205e6bc45d6STheodore Ts'o goto slashes; 42067de9c6eeSAl Viro ihold(inode); 4207f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4208be6d3e56SKentaro Takeda if (error) 42090ee50b47SDmitry Kadashev goto exit3; 42106521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4211549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4212549c7297SChristian Brauner &delegated_inode); 42130ee50b47SDmitry Kadashev exit3: 42141da177e4SLinus Torvalds dput(dentry); 42151da177e4SLinus Torvalds } 42165955102cSAl Viro inode_unlock(path.dentry->d_inode); 42171da177e4SLinus Torvalds if (inode) 42181da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4219b21996e3SJ. Bruce Fields inode = NULL; 4220b21996e3SJ. Bruce Fields if (delegated_inode) { 42215a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4222b21996e3SJ. Bruce Fields if (!error) 4223b21996e3SJ. Bruce Fields goto retry_deleg; 4224b21996e3SJ. Bruce Fields } 4225f5beed75SAl Viro mnt_drop_write(path.mnt); 42260ee50b47SDmitry Kadashev exit2: 4227f5beed75SAl Viro path_put(&path); 42285d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 42295d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 42305d18f813SJeff Layton inode = NULL; 42315d18f813SJeff Layton goto retry; 42325d18f813SJeff Layton } 42330ee50b47SDmitry Kadashev exit1: 4234da2f1362SChristoph Hellwig putname(name); 42351da177e4SLinus Torvalds return error; 42361da177e4SLinus Torvalds 42371da177e4SLinus Torvalds slashes: 4238b18825a7SDavid Howells if (d_is_negative(dentry)) 4239b18825a7SDavid Howells error = -ENOENT; 424044b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4241b18825a7SDavid Howells error = -EISDIR; 4242b18825a7SDavid Howells else 4243b18825a7SDavid Howells error = -ENOTDIR; 42440ee50b47SDmitry Kadashev goto exit3; 42451da177e4SLinus Torvalds } 42461da177e4SLinus Torvalds 42472e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 42485590ff0dSUlrich Drepper { 42495590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 42505590ff0dSUlrich Drepper return -EINVAL; 42515590ff0dSUlrich Drepper 42525590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4253e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4254da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 42555590ff0dSUlrich Drepper } 42565590ff0dSUlrich Drepper 42573480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 42585590ff0dSUlrich Drepper { 4259da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 42605590ff0dSUlrich Drepper } 42615590ff0dSUlrich Drepper 42626521f891SChristian Brauner /** 42636521f891SChristian Brauner * vfs_symlink - create symlink 42646521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 42656521f891SChristian Brauner * @dir: inode of @dentry 42666521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 42676521f891SChristian Brauner * @oldname: name of the file to link to 42686521f891SChristian Brauner * 42696521f891SChristian Brauner * Create a symlink. 42706521f891SChristian Brauner * 42716521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 42726521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42736521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42746521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42756521f891SChristian Brauner * raw inode simply passs init_user_ns. 42766521f891SChristian Brauner */ 42776521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 42786521f891SChristian Brauner struct dentry *dentry, const char *oldname) 42791da177e4SLinus Torvalds { 42806521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 42811da177e4SLinus Torvalds 42821da177e4SLinus Torvalds if (error) 42831da177e4SLinus Torvalds return error; 42841da177e4SLinus Torvalds 4285acfa4380SAl Viro if (!dir->i_op->symlink) 42861da177e4SLinus Torvalds return -EPERM; 42871da177e4SLinus Torvalds 42881da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 42891da177e4SLinus Torvalds if (error) 42901da177e4SLinus Torvalds return error; 42911da177e4SLinus Torvalds 4292549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4293a74574aaSStephen Smalley if (!error) 4294f38aa942SAmy Griffis fsnotify_create(dir, dentry); 42951da177e4SLinus Torvalds return error; 42961da177e4SLinus Torvalds } 42974d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 42981da177e4SLinus Torvalds 42997a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 43001da177e4SLinus Torvalds { 43012ad94ae6SAl Viro int error; 43026902d925SDave Hansen struct dentry *dentry; 4303dae6ad8fSAl Viro struct path path; 4304f46d3567SJeff Layton unsigned int lookup_flags = 0; 43051da177e4SLinus Torvalds 4306da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4307da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4308da2d0cedSDmitry Kadashev goto out_putnames; 4309da2d0cedSDmitry Kadashev } 4310f46d3567SJeff Layton retry: 4311b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 43121da177e4SLinus Torvalds error = PTR_ERR(dentry); 43136902d925SDave Hansen if (IS_ERR(dentry)) 4314da2d0cedSDmitry Kadashev goto out_putnames; 43156902d925SDave Hansen 431691a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 43176521f891SChristian Brauner if (!error) { 43186521f891SChristian Brauner struct user_namespace *mnt_userns; 43196521f891SChristian Brauner 43206521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 43216521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 43226521f891SChristian Brauner from->name); 43236521f891SChristian Brauner } 4324921a1650SAl Viro done_path_create(&path, dentry); 4325f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4326f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4327f46d3567SJeff Layton goto retry; 4328f46d3567SJeff Layton } 4329da2d0cedSDmitry Kadashev out_putnames: 4330da2d0cedSDmitry Kadashev putname(to); 43311da177e4SLinus Torvalds putname(from); 43321da177e4SLinus Torvalds return error; 43331da177e4SLinus Torvalds } 43341da177e4SLinus Torvalds 4335b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4336b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4337b724e846SDominik Brodowski { 4338da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4339b724e846SDominik Brodowski } 4340b724e846SDominik Brodowski 43413480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 43425590ff0dSUlrich Drepper { 4343da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 43445590ff0dSUlrich Drepper } 43455590ff0dSUlrich Drepper 4346146a8595SJ. Bruce Fields /** 4347146a8595SJ. Bruce Fields * vfs_link - create a new link 4348146a8595SJ. Bruce Fields * @old_dentry: object to be linked 43496521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4350146a8595SJ. Bruce Fields * @dir: new parent 4351146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4352146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4353146a8595SJ. Bruce Fields * 4354146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4355146a8595SJ. Bruce Fields * 4356146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4357146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4358146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4359146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4360146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4361146a8595SJ. Bruce Fields * 4362146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4363146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4364146a8595SJ. Bruce Fields * to be NFS exported. 43656521f891SChristian Brauner * 43666521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 43676521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 43686521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 43696521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 43706521f891SChristian Brauner * raw inode simply passs init_user_ns. 4371146a8595SJ. Bruce Fields */ 43726521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 43736521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 43746521f891SChristian Brauner struct inode **delegated_inode) 43751da177e4SLinus Torvalds { 43761da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 43778de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 43781da177e4SLinus Torvalds int error; 43791da177e4SLinus Torvalds 43801da177e4SLinus Torvalds if (!inode) 43811da177e4SLinus Torvalds return -ENOENT; 43821da177e4SLinus Torvalds 43836521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 43841da177e4SLinus Torvalds if (error) 43851da177e4SLinus Torvalds return error; 43861da177e4SLinus Torvalds 43871da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 43881da177e4SLinus Torvalds return -EXDEV; 43891da177e4SLinus Torvalds 43901da177e4SLinus Torvalds /* 43911da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 43921da177e4SLinus Torvalds */ 43931da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 43941da177e4SLinus Torvalds return -EPERM; 43950bd23d09SEric W. Biederman /* 43960bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 43970bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 43980bd23d09SEric W. Biederman * the vfs. 43990bd23d09SEric W. Biederman */ 44006521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 44010bd23d09SEric W. Biederman return -EPERM; 4402acfa4380SAl Viro if (!dir->i_op->link) 44031da177e4SLinus Torvalds return -EPERM; 44047e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 44051da177e4SLinus Torvalds return -EPERM; 44061da177e4SLinus Torvalds 44071da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 44081da177e4SLinus Torvalds if (error) 44091da177e4SLinus Torvalds return error; 44101da177e4SLinus Torvalds 44115955102cSAl Viro inode_lock(inode); 4412aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4413f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4414aae8a97dSAneesh Kumar K.V error = -ENOENT; 44158de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 44168de52778SAl Viro error = -EMLINK; 4417146a8595SJ. Bruce Fields else { 4418146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4419146a8595SJ. Bruce Fields if (!error) 44201da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4421146a8595SJ. Bruce Fields } 4422f4e0c30cSAl Viro 4423f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4424f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4425f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4426f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4427f4e0c30cSAl Viro } 44285955102cSAl Viro inode_unlock(inode); 4429e31e14ecSStephen Smalley if (!error) 44307e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 44311da177e4SLinus Torvalds return error; 44321da177e4SLinus Torvalds } 44334d359507SAl Viro EXPORT_SYMBOL(vfs_link); 44341da177e4SLinus Torvalds 44351da177e4SLinus Torvalds /* 44361da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 44371da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 44381da177e4SLinus Torvalds * newname. --KAB 44391da177e4SLinus Torvalds * 44401da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 44411da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 44421da177e4SLinus Torvalds * and other special files. --ADM 44431da177e4SLinus Torvalds */ 4444cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4445020250f3SDmitry Kadashev struct filename *new, int flags) 44461da177e4SLinus Torvalds { 44476521f891SChristian Brauner struct user_namespace *mnt_userns; 44481da177e4SLinus Torvalds struct dentry *new_dentry; 4449dae6ad8fSAl Viro struct path old_path, new_path; 4450146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 445111a7b371SAneesh Kumar K.V int how = 0; 44521da177e4SLinus Torvalds int error; 44531da177e4SLinus Torvalds 4454020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4455020250f3SDmitry Kadashev error = -EINVAL; 4456020250f3SDmitry Kadashev goto out_putnames; 4457020250f3SDmitry Kadashev } 445811a7b371SAneesh Kumar K.V /* 4459f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4460f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4461f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 446211a7b371SAneesh Kumar K.V */ 4463020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4464020250f3SDmitry Kadashev error = -ENOENT; 4465020250f3SDmitry Kadashev goto out_putnames; 4466f0cc6ffbSLinus Torvalds } 4467c04030e1SUlrich Drepper 446811a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 446911a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4470442e31caSJeff Layton retry: 4471794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 44721da177e4SLinus Torvalds if (error) 4473020250f3SDmitry Kadashev goto out_putnames; 44742ad94ae6SAl Viro 4475b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4476442e31caSJeff Layton (how & LOOKUP_REVAL)); 44771da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 44786902d925SDave Hansen if (IS_ERR(new_dentry)) 4479020250f3SDmitry Kadashev goto out_putpath; 4480dae6ad8fSAl Viro 4481dae6ad8fSAl Viro error = -EXDEV; 4482dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4483dae6ad8fSAl Viro goto out_dput; 4484549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4485549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4486800179c9SKees Cook if (unlikely(error)) 4487800179c9SKees Cook goto out_dput; 4488dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4489be6d3e56SKentaro Takeda if (error) 4490a8104a9fSAl Viro goto out_dput; 44916521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 44926521f891SChristian Brauner new_dentry, &delegated_inode); 449375c3f29dSDave Hansen out_dput: 4494921a1650SAl Viro done_path_create(&new_path, new_dentry); 4495146a8595SJ. Bruce Fields if (delegated_inode) { 4496146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4497d22e6338SOleg Drokin if (!error) { 4498d22e6338SOleg Drokin path_put(&old_path); 4499146a8595SJ. Bruce Fields goto retry; 4500146a8595SJ. Bruce Fields } 4501d22e6338SOleg Drokin } 4502442e31caSJeff Layton if (retry_estale(error, how)) { 4503d22e6338SOleg Drokin path_put(&old_path); 4504442e31caSJeff Layton how |= LOOKUP_REVAL; 4505442e31caSJeff Layton goto retry; 4506442e31caSJeff Layton } 4507020250f3SDmitry Kadashev out_putpath: 45082d8f3038SAl Viro path_put(&old_path); 4509020250f3SDmitry Kadashev out_putnames: 4510020250f3SDmitry Kadashev putname(old); 4511020250f3SDmitry Kadashev putname(new); 45121da177e4SLinus Torvalds 45131da177e4SLinus Torvalds return error; 45141da177e4SLinus Torvalds } 45151da177e4SLinus Torvalds 451646ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 451746ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 451846ea89ebSDominik Brodowski { 4519020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4520020250f3SDmitry Kadashev newdfd, getname(newname), flags); 452146ea89ebSDominik Brodowski } 452246ea89ebSDominik Brodowski 45233480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 45245590ff0dSUlrich Drepper { 4525020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 45265590ff0dSUlrich Drepper } 45275590ff0dSUlrich Drepper 4528bc27027aSMiklos Szeredi /** 4529bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 45302111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4531bc27027aSMiklos Szeredi * 4532bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4533bc27027aSMiklos Szeredi * 4534bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4535bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4536bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4537bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4538bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4539bc27027aSMiklos Szeredi * so. 4540bc27027aSMiklos Szeredi * 4541bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4542bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4543bc27027aSMiklos Szeredi * to be NFS exported. 4544bc27027aSMiklos Szeredi * 45451da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 45461da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 45471da177e4SLinus Torvalds * Problems: 45480117d427SMauro Carvalho Chehab * 4549d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 45501da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 45511da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4552a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 45531da177e4SLinus Torvalds * story. 45546cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 45556cedba89SJ. Bruce Fields * and source (if it is not a directory). 45561b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 45571da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 45581da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4559a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 45601da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 45611da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 45621da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4563a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 45641da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 45651da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 45661da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4567e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 45681b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 45691da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4570c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 45711da177e4SLinus Torvalds * locking]. 45721da177e4SLinus Torvalds */ 45739fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 45741da177e4SLinus Torvalds { 45751da177e4SLinus Torvalds int error; 45769fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 45779fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 45789fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 45799fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 45809fe61450SChristian Brauner unsigned int flags = rd->flags; 4581bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4582bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4583bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4584da1ce067SMiklos Szeredi bool new_is_dir = false; 4585da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 458649d31c2fSAl Viro struct name_snapshot old_name; 45871da177e4SLinus Torvalds 45888d3e2936SMiklos Szeredi if (source == target) 45891da177e4SLinus Torvalds return 0; 45901da177e4SLinus Torvalds 45916521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 45921da177e4SLinus Torvalds if (error) 45931da177e4SLinus Torvalds return error; 45941da177e4SLinus Torvalds 4595da1ce067SMiklos Szeredi if (!target) { 45966521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4597da1ce067SMiklos Szeredi } else { 4598da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4599da1ce067SMiklos Szeredi 4600da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46016521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 46026521f891SChristian Brauner new_dentry, is_dir); 4603da1ce067SMiklos Szeredi else 46046521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 46056521f891SChristian Brauner new_dentry, new_is_dir); 4606da1ce067SMiklos Szeredi } 46071da177e4SLinus Torvalds if (error) 46081da177e4SLinus Torvalds return error; 46091da177e4SLinus Torvalds 46102773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 46111da177e4SLinus Torvalds return -EPERM; 46121da177e4SLinus Torvalds 4613bc27027aSMiklos Szeredi /* 4614bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4615bc27027aSMiklos Szeredi * we'll need to flip '..'. 4616bc27027aSMiklos Szeredi */ 4617da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4618da1ce067SMiklos Szeredi if (is_dir) { 46196521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 462047291baaSChristian Brauner MAY_WRITE); 4621bc27027aSMiklos Szeredi if (error) 4622bc27027aSMiklos Szeredi return error; 4623bc27027aSMiklos Szeredi } 4624da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 46256521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 462647291baaSChristian Brauner MAY_WRITE); 4627da1ce067SMiklos Szeredi if (error) 4628da1ce067SMiklos Szeredi return error; 4629da1ce067SMiklos Szeredi } 4630da1ce067SMiklos Szeredi } 46310eeca283SRobert Love 46320b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 46330b3974ebSMiklos Szeredi flags); 4634bc27027aSMiklos Szeredi if (error) 4635bc27027aSMiklos Szeredi return error; 4636bc27027aSMiklos Szeredi 463749d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4638bc27027aSMiklos Szeredi dget(new_dentry); 4639da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4640bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4641bc27027aSMiklos Szeredi else if (target) 46425955102cSAl Viro inode_lock(target); 4643bc27027aSMiklos Szeredi 464451cc3a66SHugh Dickins error = -EPERM; 464551cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 464651cc3a66SHugh Dickins goto out; 464751cc3a66SHugh Dickins 4648bc27027aSMiklos Szeredi error = -EBUSY; 46497af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4650bc27027aSMiklos Szeredi goto out; 4651bc27027aSMiklos Szeredi 4652da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4653bc27027aSMiklos Szeredi error = -EMLINK; 4654da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4655bc27027aSMiklos Szeredi goto out; 4656da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4657da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4658da1ce067SMiklos Szeredi goto out; 4659da1ce067SMiklos Szeredi } 4660da1ce067SMiklos Szeredi if (!is_dir) { 4661bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4662bc27027aSMiklos Szeredi if (error) 4663bc27027aSMiklos Szeredi goto out; 4664da1ce067SMiklos Szeredi } 4665da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4666bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4667bc27027aSMiklos Szeredi if (error) 4668bc27027aSMiklos Szeredi goto out; 4669bc27027aSMiklos Szeredi } 4670549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4671520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4672bc27027aSMiklos Szeredi if (error) 4673bc27027aSMiklos Szeredi goto out; 4674bc27027aSMiklos Szeredi 4675da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 46768767712fSAl Viro if (is_dir) { 46778767712fSAl Viro shrink_dcache_parent(new_dentry); 4678bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 46798767712fSAl Viro } 4680bc27027aSMiklos Szeredi dont_mount(new_dentry); 46818ed936b5SEric W. Biederman detach_mounts(new_dentry); 4682bc27027aSMiklos Szeredi } 4683da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4684da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4685bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4686da1ce067SMiklos Szeredi else 4687da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4688da1ce067SMiklos Szeredi } 4689bc27027aSMiklos Szeredi out: 4690da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4691bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4692bc27027aSMiklos Szeredi else if (target) 46935955102cSAl Viro inode_unlock(target); 4694bc27027aSMiklos Szeredi dput(new_dentry); 4695da1ce067SMiklos Szeredi if (!error) { 4696f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4697da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4698da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4699f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4700da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4701da1ce067SMiklos Szeredi } 4702da1ce067SMiklos Szeredi } 470349d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 47040eeca283SRobert Love 47051da177e4SLinus Torvalds return error; 47061da177e4SLinus Torvalds } 47074d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 47081da177e4SLinus Torvalds 4709e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4710e886663cSJens Axboe struct filename *to, unsigned int flags) 47111da177e4SLinus Torvalds { 47129fe61450SChristian Brauner struct renamedata rd; 47131da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 47141da177e4SLinus Torvalds struct dentry *trap; 4715f5beed75SAl Viro struct path old_path, new_path; 4716f5beed75SAl Viro struct qstr old_last, new_last; 4717f5beed75SAl Viro int old_type, new_type; 47188e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4719f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4720c6a94284SJeff Layton bool should_retry = false; 4721e886663cSJens Axboe int error = -EINVAL; 4722520c8b16SMiklos Szeredi 47230d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 47240ee50b47SDmitry Kadashev goto put_names; 4725da1ce067SMiklos Szeredi 47260d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 47270d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 47280ee50b47SDmitry Kadashev goto put_names; 4729520c8b16SMiklos Szeredi 4730f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4731f5beed75SAl Viro target_flags = 0; 4732f5beed75SAl Viro 4733c6a94284SJeff Layton retry: 4734c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4735e886663cSJens Axboe &old_last, &old_type); 47360ee50b47SDmitry Kadashev if (error) 47370ee50b47SDmitry Kadashev goto put_names; 47381da177e4SLinus Torvalds 4739c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4740e886663cSJens Axboe &new_type); 47410ee50b47SDmitry Kadashev if (error) 47421da177e4SLinus Torvalds goto exit1; 47431da177e4SLinus Torvalds 47441da177e4SLinus Torvalds error = -EXDEV; 4745f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 47461da177e4SLinus Torvalds goto exit2; 47471da177e4SLinus Torvalds 47481da177e4SLinus Torvalds error = -EBUSY; 4749f5beed75SAl Viro if (old_type != LAST_NORM) 47501da177e4SLinus Torvalds goto exit2; 47511da177e4SLinus Torvalds 47520a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 47530a7c3937SMiklos Szeredi error = -EEXIST; 4754f5beed75SAl Viro if (new_type != LAST_NORM) 47551da177e4SLinus Torvalds goto exit2; 47561da177e4SLinus Torvalds 4757f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4758c30dabfeSJan Kara if (error) 4759c30dabfeSJan Kara goto exit2; 4760c30dabfeSJan Kara 47618e6d782cSJ. Bruce Fields retry_deleg: 4762f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 47631da177e4SLinus Torvalds 4764f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 47651da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 47661da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 47671da177e4SLinus Torvalds goto exit3; 47681da177e4SLinus Torvalds /* source must exist */ 47691da177e4SLinus Torvalds error = -ENOENT; 4770b18825a7SDavid Howells if (d_is_negative(old_dentry)) 47711da177e4SLinus Torvalds goto exit4; 4772f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 47731da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 47741da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 47751da177e4SLinus Torvalds goto exit4; 47760a7c3937SMiklos Szeredi error = -EEXIST; 47770a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 47780a7c3937SMiklos Szeredi goto exit5; 4779da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4780da1ce067SMiklos Szeredi error = -ENOENT; 4781da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4782da1ce067SMiklos Szeredi goto exit5; 4783da1ce067SMiklos Szeredi 4784da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4785da1ce067SMiklos Szeredi error = -ENOTDIR; 4786f5beed75SAl Viro if (new_last.name[new_last.len]) 4787da1ce067SMiklos Szeredi goto exit5; 4788da1ce067SMiklos Szeredi } 4789da1ce067SMiklos Szeredi } 47900a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 47910a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 47920a7c3937SMiklos Szeredi error = -ENOTDIR; 4793f5beed75SAl Viro if (old_last.name[old_last.len]) 47940a7c3937SMiklos Szeredi goto exit5; 4795f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 47960a7c3937SMiklos Szeredi goto exit5; 47970a7c3937SMiklos Szeredi } 47980a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 47990a7c3937SMiklos Szeredi error = -EINVAL; 48000a7c3937SMiklos Szeredi if (old_dentry == trap) 48010a7c3937SMiklos Szeredi goto exit5; 48021da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4803da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 48041da177e4SLinus Torvalds error = -ENOTEMPTY; 48051da177e4SLinus Torvalds if (new_dentry == trap) 48061da177e4SLinus Torvalds goto exit5; 48071da177e4SLinus Torvalds 4808f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4809f5beed75SAl Viro &new_path, new_dentry, flags); 4810be6d3e56SKentaro Takeda if (error) 4811c30dabfeSJan Kara goto exit5; 48129fe61450SChristian Brauner 48139fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 48149fe61450SChristian Brauner rd.old_dentry = old_dentry; 48156521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 48169fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 48179fe61450SChristian Brauner rd.new_dentry = new_dentry; 48186521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 48199fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 48209fe61450SChristian Brauner rd.flags = flags; 48219fe61450SChristian Brauner error = vfs_rename(&rd); 48221da177e4SLinus Torvalds exit5: 48231da177e4SLinus Torvalds dput(new_dentry); 48241da177e4SLinus Torvalds exit4: 48251da177e4SLinus Torvalds dput(old_dentry); 48261da177e4SLinus Torvalds exit3: 4827f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 48288e6d782cSJ. Bruce Fields if (delegated_inode) { 48298e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 48308e6d782cSJ. Bruce Fields if (!error) 48318e6d782cSJ. Bruce Fields goto retry_deleg; 48328e6d782cSJ. Bruce Fields } 4833f5beed75SAl Viro mnt_drop_write(old_path.mnt); 48341da177e4SLinus Torvalds exit2: 4835c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4836c6a94284SJeff Layton should_retry = true; 4837f5beed75SAl Viro path_put(&new_path); 48381da177e4SLinus Torvalds exit1: 4839f5beed75SAl Viro path_put(&old_path); 4840c6a94284SJeff Layton if (should_retry) { 4841c6a94284SJeff Layton should_retry = false; 4842c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4843c6a94284SJeff Layton goto retry; 4844c6a94284SJeff Layton } 48450ee50b47SDmitry Kadashev put_names: 4846e886663cSJens Axboe putname(from); 4847e886663cSJens Axboe putname(to); 48481da177e4SLinus Torvalds return error; 48491da177e4SLinus Torvalds } 48501da177e4SLinus Torvalds 4851ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4852ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4853ee81feb6SDominik Brodowski { 4854e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4855e886663cSJens Axboe flags); 4856ee81feb6SDominik Brodowski } 4857ee81feb6SDominik Brodowski 4858520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4859520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4860520c8b16SMiklos Szeredi { 4861e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4862e886663cSJens Axboe 0); 4863520c8b16SMiklos Szeredi } 4864520c8b16SMiklos Szeredi 4865a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 48665590ff0dSUlrich Drepper { 4867e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4868e886663cSJens Axboe getname(newname), 0); 48695590ff0dSUlrich Drepper } 48705590ff0dSUlrich Drepper 48715d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 48721da177e4SLinus Torvalds { 48735d826c84SAl Viro int len = PTR_ERR(link); 48741da177e4SLinus Torvalds if (IS_ERR(link)) 48751da177e4SLinus Torvalds goto out; 48761da177e4SLinus Torvalds 48771da177e4SLinus Torvalds len = strlen(link); 48781da177e4SLinus Torvalds if (len > (unsigned) buflen) 48791da177e4SLinus Torvalds len = buflen; 48801da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 48811da177e4SLinus Torvalds len = -EFAULT; 48821da177e4SLinus Torvalds out: 48831da177e4SLinus Torvalds return len; 48841da177e4SLinus Torvalds } 48851da177e4SLinus Torvalds 4886d60874cdSMiklos Szeredi /** 4887fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4888fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4889fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4890fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4891fd4a0edfSMiklos Szeredi * 4892fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4893fd4a0edfSMiklos Szeredi * 4894fd4a0edfSMiklos Szeredi * Does not call security hook. 4895fd4a0edfSMiklos Szeredi */ 4896fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4897fd4a0edfSMiklos Szeredi { 4898fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4899f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4900f2df5da6SAl Viro const char *link; 4901f2df5da6SAl Viro int res; 4902fd4a0edfSMiklos Szeredi 490376fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 490476fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 490576fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 490676fca90eSMiklos Szeredi 490776fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4908fd4a0edfSMiklos Szeredi return -EINVAL; 4909fd4a0edfSMiklos Szeredi 491076fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 491176fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 491276fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 491376fca90eSMiklos Szeredi } 491476fca90eSMiklos Szeredi 49154c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4916f2df5da6SAl Viro if (!link) { 4917f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4918f2df5da6SAl Viro if (IS_ERR(link)) 4919f2df5da6SAl Viro return PTR_ERR(link); 4920f2df5da6SAl Viro } 4921f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4922f2df5da6SAl Viro do_delayed_call(&done); 4923f2df5da6SAl Viro return res; 4924fd4a0edfSMiklos Szeredi } 4925fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 49261da177e4SLinus Torvalds 4927d60874cdSMiklos Szeredi /** 4928d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4929d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4930d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4931d60874cdSMiklos Szeredi * 4932d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4933d60874cdSMiklos Szeredi * 4934d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4935d60874cdSMiklos Szeredi * 4936d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4937d60874cdSMiklos Szeredi */ 4938d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4939d60874cdSMiklos Szeredi { 4940d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4941d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4942d60874cdSMiklos Szeredi 4943d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4944d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4945d60874cdSMiklos Szeredi if (!res) 4946d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4947d60874cdSMiklos Szeredi } 4948d60874cdSMiklos Szeredi return res; 4949d60874cdSMiklos Szeredi } 4950d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4951d60874cdSMiklos Szeredi 49521da177e4SLinus Torvalds /* get the link contents into pagecache */ 49536b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4954fceef393SAl Viro struct delayed_call *callback) 49551da177e4SLinus Torvalds { 4956ebd09abbSDuane Griffin char *kaddr; 49571da177e4SLinus Torvalds struct page *page; 49586b255391SAl Viro struct address_space *mapping = inode->i_mapping; 49596b255391SAl Viro 4960d3883d4fSAl Viro if (!dentry) { 4961d3883d4fSAl Viro page = find_get_page(mapping, 0); 4962d3883d4fSAl Viro if (!page) 49636b255391SAl Viro return ERR_PTR(-ECHILD); 4964d3883d4fSAl Viro if (!PageUptodate(page)) { 4965d3883d4fSAl Viro put_page(page); 4966d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4967d3883d4fSAl Viro } 4968d3883d4fSAl Viro } else { 4969090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 49701da177e4SLinus Torvalds if (IS_ERR(page)) 49716fe6900eSNick Piggin return (char*)page; 4972d3883d4fSAl Viro } 4973fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 497421fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 497521fc61c7SAl Viro kaddr = page_address(page); 49766b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4977ebd09abbSDuane Griffin return kaddr; 49781da177e4SLinus Torvalds } 49791da177e4SLinus Torvalds 49806b255391SAl Viro EXPORT_SYMBOL(page_get_link); 49811da177e4SLinus Torvalds 4982fceef393SAl Viro void page_put_link(void *arg) 49831da177e4SLinus Torvalds { 4984fceef393SAl Viro put_page(arg); 49851da177e4SLinus Torvalds } 49864d359507SAl Viro EXPORT_SYMBOL(page_put_link); 49871da177e4SLinus Torvalds 4988aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4989aa80deabSAl Viro { 4990fceef393SAl Viro DEFINE_DELAYED_CALL(done); 49916b255391SAl Viro int res = readlink_copy(buffer, buflen, 49926b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4993fceef393SAl Viro &done)); 4994fceef393SAl Viro do_delayed_call(&done); 4995aa80deabSAl Viro return res; 4996aa80deabSAl Viro } 4997aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4998aa80deabSAl Viro 499954566b2cSNick Piggin /* 500054566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 500154566b2cSNick Piggin */ 500254566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 50031da177e4SLinus Torvalds { 50041da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 50050adb25d2SKirill Korotaev struct page *page; 5006afddba49SNick Piggin void *fsdata; 5007beb497abSDmitriy Monakhov int err; 5008c718a975STetsuo Handa unsigned int flags = 0; 500954566b2cSNick Piggin if (nofs) 501054566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 50111da177e4SLinus Torvalds 50127e53cac4SNeilBrown retry: 5013afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 501454566b2cSNick Piggin flags, &page, &fsdata); 50151da177e4SLinus Torvalds if (err) 5016afddba49SNick Piggin goto fail; 5017afddba49SNick Piggin 501821fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5019afddba49SNick Piggin 5020afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 5021afddba49SNick Piggin page, fsdata); 50221da177e4SLinus Torvalds if (err < 0) 50231da177e4SLinus Torvalds goto fail; 5024afddba49SNick Piggin if (err < len-1) 5025afddba49SNick Piggin goto retry; 5026afddba49SNick Piggin 50271da177e4SLinus Torvalds mark_inode_dirty(inode); 50281da177e4SLinus Torvalds return 0; 50291da177e4SLinus Torvalds fail: 50301da177e4SLinus Torvalds return err; 50311da177e4SLinus Torvalds } 50324d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 50331da177e4SLinus Torvalds 50340adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 50350adb25d2SKirill Korotaev { 50360adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 5037c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 50380adb25d2SKirill Korotaev } 50394d359507SAl Viro EXPORT_SYMBOL(page_symlink); 50400adb25d2SKirill Korotaev 504192e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 50426b255391SAl Viro .get_link = page_get_link, 50431da177e4SLinus Torvalds }; 50441da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5045