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 * 20791a27b2aSJeff Layton getname(const char __user * filename) 208f52e0c11SAl Viro { 209f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 210f52e0c11SAl Viro } 211f52e0c11SAl Viro 212c4ad8f98SLinus Torvalds struct filename * 213c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 214c4ad8f98SLinus Torvalds { 215c4ad8f98SLinus Torvalds struct filename *result; 21608518549SPaul Moore int len = strlen(filename) + 1; 217c4ad8f98SLinus Torvalds 218c4ad8f98SLinus Torvalds result = __getname(); 219c4ad8f98SLinus Torvalds if (unlikely(!result)) 220c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 221c4ad8f98SLinus Torvalds 22208518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 223fd2f7cb5SAl Viro result->name = (char *)result->iname; 22408518549SPaul Moore } else if (len <= PATH_MAX) { 22530ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 22608518549SPaul Moore struct filename *tmp; 22708518549SPaul Moore 22830ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 22908518549SPaul Moore if (unlikely(!tmp)) { 23008518549SPaul Moore __putname(result); 23108518549SPaul Moore return ERR_PTR(-ENOMEM); 23208518549SPaul Moore } 23308518549SPaul Moore tmp->name = (char *)result; 23408518549SPaul Moore result = tmp; 23508518549SPaul Moore } else { 23608518549SPaul Moore __putname(result); 23708518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 23808518549SPaul Moore } 23908518549SPaul Moore memcpy((char *)result->name, filename, len); 240c4ad8f98SLinus Torvalds result->uptr = NULL; 241c4ad8f98SLinus Torvalds result->aname = NULL; 24255422d0bSPaul Moore result->refcnt = 1; 243fd3522fdSPaul Moore audit_getname(result); 244c4ad8f98SLinus Torvalds 245c4ad8f98SLinus Torvalds return result; 246c4ad8f98SLinus Torvalds } 247c4ad8f98SLinus Torvalds 24891a27b2aSJeff Layton void putname(struct filename *name) 2491da177e4SLinus Torvalds { 25055422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 25155422d0bSPaul Moore 25255422d0bSPaul Moore if (--name->refcnt > 0) 25355422d0bSPaul Moore return; 25455422d0bSPaul Moore 255fd2f7cb5SAl Viro if (name->name != name->iname) { 25655422d0bSPaul Moore __putname(name->name); 25755422d0bSPaul Moore kfree(name); 25855422d0bSPaul Moore } else 25955422d0bSPaul Moore __putname(name); 2601da177e4SLinus Torvalds } 2611da177e4SLinus Torvalds 26247291baaSChristian Brauner /** 26347291baaSChristian Brauner * check_acl - perform ACL permission checking 26447291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 26547291baaSChristian Brauner * @inode: inode to check permissions on 26647291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 26747291baaSChristian Brauner * 26847291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 26947291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 27047291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 27147291baaSChristian Brauner * 27247291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 27347291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 27447291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 27547291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 27647291baaSChristian Brauner * raw inode simply passs init_user_ns. 27747291baaSChristian Brauner */ 27847291baaSChristian Brauner static int check_acl(struct user_namespace *mnt_userns, 27947291baaSChristian Brauner struct inode *inode, int mask) 280e77819e5SLinus Torvalds { 28184635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 282e77819e5SLinus Torvalds struct posix_acl *acl; 283e77819e5SLinus Torvalds 284e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2853567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2863567866bSAl Viro if (!acl) 287e77819e5SLinus Torvalds return -EAGAIN; 2883567866bSAl Viro /* no ->get_acl() calls in RCU mode... */ 289b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 290e77819e5SLinus Torvalds return -ECHILD; 29147291baaSChristian Brauner return posix_acl_permission(mnt_userns, inode, acl, mask); 292e77819e5SLinus Torvalds } 293e77819e5SLinus Torvalds 2942982baa2SChristoph Hellwig acl = get_acl(inode, ACL_TYPE_ACCESS); 2954e34e719SChristoph Hellwig if (IS_ERR(acl)) 2964e34e719SChristoph Hellwig return PTR_ERR(acl); 297e77819e5SLinus Torvalds if (acl) { 29847291baaSChristian Brauner int error = posix_acl_permission(mnt_userns, inode, acl, mask); 299e77819e5SLinus Torvalds posix_acl_release(acl); 300e77819e5SLinus Torvalds return error; 301e77819e5SLinus Torvalds } 30284635d68SLinus Torvalds #endif 303e77819e5SLinus Torvalds 304e77819e5SLinus Torvalds return -EAGAIN; 305e77819e5SLinus Torvalds } 306e77819e5SLinus Torvalds 30747291baaSChristian Brauner /** 30847291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 30947291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 31047291baaSChristian Brauner * @inode: inode to check permissions on 31147291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3125fc475b7SLinus Torvalds * 31347291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 31447291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 31547291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 31647291baaSChristian Brauner * 31747291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 31847291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 31947291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 32047291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 32147291baaSChristian Brauner * raw inode simply passs init_user_ns. 3225909ccaaSLinus Torvalds */ 32347291baaSChristian Brauner static int acl_permission_check(struct user_namespace *mnt_userns, 32447291baaSChristian Brauner struct inode *inode, int mask) 3255909ccaaSLinus Torvalds { 32626cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 32747291baaSChristian Brauner kuid_t i_uid; 3285909ccaaSLinus Torvalds 3295fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 33047291baaSChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 33147291baaSChristian Brauner if (likely(uid_eq(current_fsuid(), i_uid))) { 3325fc475b7SLinus Torvalds mask &= 7; 3335909ccaaSLinus Torvalds mode >>= 6; 3345fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3355fc475b7SLinus Torvalds } 3365fc475b7SLinus Torvalds 3375fc475b7SLinus Torvalds /* Do we have ACL's? */ 338e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 33947291baaSChristian Brauner int error = check_acl(mnt_userns, inode, mask); 3405909ccaaSLinus Torvalds if (error != -EAGAIN) 3415909ccaaSLinus Torvalds return error; 3425909ccaaSLinus Torvalds } 3435909ccaaSLinus Torvalds 3445fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3455fc475b7SLinus Torvalds mask &= 7; 3465fc475b7SLinus Torvalds 3475fc475b7SLinus Torvalds /* 3485fc475b7SLinus Torvalds * Are the group permissions different from 3495fc475b7SLinus Torvalds * the other permissions in the bits we care 3505fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3515fc475b7SLinus Torvalds */ 3525fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 35347291baaSChristian Brauner kgid_t kgid = i_gid_into_mnt(mnt_userns, inode); 35447291baaSChristian Brauner if (in_group_p(kgid)) 3555909ccaaSLinus Torvalds mode >>= 3; 3565909ccaaSLinus Torvalds } 3575909ccaaSLinus Torvalds 3585fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3595fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3605909ccaaSLinus Torvalds } 3611da177e4SLinus Torvalds 3621da177e4SLinus Torvalds /** 3631da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 36447291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 3651da177e4SLinus Torvalds * @inode: inode to check access rights for 3665fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3675fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3681da177e4SLinus Torvalds * 3691da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3701da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3711da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 372b74c79e9SNick Piggin * are used for other things. 373b74c79e9SNick Piggin * 374b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 375b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 376b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 37747291baaSChristian Brauner * 37847291baaSChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 37947291baaSChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 38047291baaSChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 38147291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 38247291baaSChristian Brauner * raw inode simply passs init_user_ns. 3831da177e4SLinus Torvalds */ 38447291baaSChristian Brauner int generic_permission(struct user_namespace *mnt_userns, struct inode *inode, 38547291baaSChristian Brauner int mask) 3861da177e4SLinus Torvalds { 3875909ccaaSLinus Torvalds int ret; 3881da177e4SLinus Torvalds 3891da177e4SLinus Torvalds /* 390948409c7SAndreas Gruenbacher * Do the basic permission checks. 3911da177e4SLinus Torvalds */ 39247291baaSChristian Brauner ret = acl_permission_check(mnt_userns, inode, mask); 3935909ccaaSLinus Torvalds if (ret != -EACCES) 3945909ccaaSLinus Torvalds return ret; 3951da177e4SLinus Torvalds 396d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 397d594e7ecSAl Viro /* DACs are overridable for directories */ 398d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 39947291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 40023adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 401d594e7ecSAl Viro return 0; 40247291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4030558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4041da177e4SLinus Torvalds return 0; 4052a4c2242SStephen Smalley return -EACCES; 4062a4c2242SStephen Smalley } 4071da177e4SLinus Torvalds 4081da177e4SLinus Torvalds /* 4091da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4101da177e4SLinus Torvalds */ 4117ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 412d594e7ecSAl Viro if (mask == MAY_READ) 41347291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4140558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4151da177e4SLinus Torvalds return 0; 4162a4c2242SStephen Smalley /* 4172a4c2242SStephen Smalley * Read/write DACs are always overridable. 4182a4c2242SStephen Smalley * Executable DACs are overridable when there is 4192a4c2242SStephen Smalley * at least one exec bit set. 4202a4c2242SStephen Smalley */ 4212a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 42247291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4230558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4242a4c2242SStephen Smalley return 0; 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds return -EACCES; 4271da177e4SLinus Torvalds } 4284d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4291da177e4SLinus Torvalds 43047291baaSChristian Brauner /** 43147291baaSChristian Brauner * do_inode_permission - UNIX permission checking 43247291baaSChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 43347291baaSChristian Brauner * @inode: inode to check permissions on 43447291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 43547291baaSChristian Brauner * 4363ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4373ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4383ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4393ddcd056SLinus Torvalds * permission function, use the fast case". 4403ddcd056SLinus Torvalds */ 44147291baaSChristian Brauner static inline int do_inode_permission(struct user_namespace *mnt_userns, 44247291baaSChristian Brauner struct inode *inode, int mask) 4433ddcd056SLinus Torvalds { 4443ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4453ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 446549c7297SChristian Brauner return inode->i_op->permission(mnt_userns, inode, mask); 4473ddcd056SLinus Torvalds 4483ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4493ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4503ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4513ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4523ddcd056SLinus Torvalds } 45347291baaSChristian Brauner return generic_permission(mnt_userns, inode, mask); 4543ddcd056SLinus Torvalds } 4553ddcd056SLinus Torvalds 456cb23beb5SChristoph Hellwig /** 4570bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4580bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 45955852635SRandy Dunlap * @inode: Inode to check permission on 4600bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4610bdaea90SDavid Howells * 4620bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4630bdaea90SDavid Howells */ 4640bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4650bdaea90SDavid Howells { 4660bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4670bdaea90SDavid Howells umode_t mode = inode->i_mode; 4680bdaea90SDavid Howells 4690bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 470bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4710bdaea90SDavid Howells return -EROFS; 4720bdaea90SDavid Howells } 4730bdaea90SDavid Howells return 0; 4740bdaea90SDavid Howells } 4750bdaea90SDavid Howells 4760bdaea90SDavid Howells /** 4770bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 47847291baaSChristian Brauner * @mnt_userns: User namespace of the mount the inode was found from 4790bdaea90SDavid Howells * @inode: Inode to check permission on 4800bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4810bdaea90SDavid Howells * 4820bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4830bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4840bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4850bdaea90SDavid Howells * 4860bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 4870bdaea90SDavid Howells */ 48847291baaSChristian Brauner int inode_permission(struct user_namespace *mnt_userns, 48947291baaSChristian Brauner struct inode *inode, int mask) 4900bdaea90SDavid Howells { 4910bdaea90SDavid Howells int retval; 4920bdaea90SDavid Howells 4930bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 4940bdaea90SDavid Howells if (retval) 4950bdaea90SDavid Howells return retval; 4964bfd054aSEric Biggers 4974bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 4984bfd054aSEric Biggers /* 4994bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5004bfd054aSEric Biggers */ 5014bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5024bfd054aSEric Biggers return -EPERM; 5034bfd054aSEric Biggers 5044bfd054aSEric Biggers /* 5054bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5064bfd054aSEric Biggers * written back improperly if their true value is unknown 5074bfd054aSEric Biggers * to the vfs. 5084bfd054aSEric Biggers */ 509ba73d987SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 5104bfd054aSEric Biggers return -EACCES; 5114bfd054aSEric Biggers } 5124bfd054aSEric Biggers 51347291baaSChristian Brauner retval = do_inode_permission(mnt_userns, inode, mask); 5144bfd054aSEric Biggers if (retval) 5154bfd054aSEric Biggers return retval; 5164bfd054aSEric Biggers 5174bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5184bfd054aSEric Biggers if (retval) 5194bfd054aSEric Biggers return retval; 5204bfd054aSEric Biggers 5214bfd054aSEric Biggers return security_inode_permission(inode, mask); 5220bdaea90SDavid Howells } 5234d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5240bdaea90SDavid Howells 5250bdaea90SDavid Howells /** 5265dd784d0SJan Blunck * path_get - get a reference to a path 5275dd784d0SJan Blunck * @path: path to get the reference to 5285dd784d0SJan Blunck * 5295dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5305dd784d0SJan Blunck */ 531dcf787f3SAl Viro void path_get(const struct path *path) 5325dd784d0SJan Blunck { 5335dd784d0SJan Blunck mntget(path->mnt); 5345dd784d0SJan Blunck dget(path->dentry); 5355dd784d0SJan Blunck } 5365dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5375dd784d0SJan Blunck 5385dd784d0SJan Blunck /** 5391d957f9bSJan Blunck * path_put - put a reference to a path 5401d957f9bSJan Blunck * @path: path to put the reference to 5411d957f9bSJan Blunck * 5421d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5431d957f9bSJan Blunck */ 544dcf787f3SAl Viro void path_put(const struct path *path) 5451da177e4SLinus Torvalds { 5461d957f9bSJan Blunck dput(path->dentry); 5471d957f9bSJan Blunck mntput(path->mnt); 5481da177e4SLinus Torvalds } 5491d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5501da177e4SLinus Torvalds 551894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5521f55a6ecSAl Viro struct nameidata { 5531f55a6ecSAl Viro struct path path; 5541f55a6ecSAl Viro struct qstr last; 5551f55a6ecSAl Viro struct path root; 5561f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 557bcba1e7dSAl Viro unsigned int flags, state; 558ab87f9a5SAleksa Sarai unsigned seq, m_seq, r_seq; 5591f55a6ecSAl Viro int last_type; 5601f55a6ecSAl Viro unsigned depth; 561756daf26SNeilBrown int total_link_count; 562697fc6caSAl Viro struct saved { 563697fc6caSAl Viro struct path link; 564fceef393SAl Viro struct delayed_call done; 565697fc6caSAl Viro const char *name; 5660450b2d1SAl Viro unsigned seq; 567894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5689883d185SAl Viro struct filename *name; 5699883d185SAl Viro struct nameidata *saved; 5709883d185SAl Viro unsigned root_seq; 5719883d185SAl Viro int dfd; 5720f705953SAl Viro kuid_t dir_uid; 5730f705953SAl Viro umode_t dir_mode; 5743859a271SKees Cook } __randomize_layout; 5751f55a6ecSAl Viro 576bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 577bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 578bcba1e7dSAl Viro #define ND_JUMPED 4 579bcba1e7dSAl Viro 58006422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 581894bc8c4SAl Viro { 582756daf26SNeilBrown struct nameidata *old = current->nameidata; 583756daf26SNeilBrown p->stack = p->internal; 5847962c7d1SAl Viro p->depth = 0; 585c8a53ee5SAl Viro p->dfd = dfd; 586c8a53ee5SAl Viro p->name = name; 5877d01ef75SAl Viro p->path.mnt = NULL; 5887d01ef75SAl Viro p->path.dentry = NULL; 589756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5909883d185SAl Viro p->saved = old; 591756daf26SNeilBrown current->nameidata = p; 592894bc8c4SAl Viro } 593894bc8c4SAl Viro 59406422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 59506422964SAl Viro const struct path *root) 59606422964SAl Viro { 59706422964SAl Viro __set_nameidata(p, dfd, name); 59806422964SAl Viro p->state = 0; 59906422964SAl Viro if (unlikely(root)) { 60006422964SAl Viro p->state = ND_ROOT_PRESET; 60106422964SAl Viro p->root = *root; 60206422964SAl Viro } 60306422964SAl Viro } 60406422964SAl Viro 6059883d185SAl Viro static void restore_nameidata(void) 606894bc8c4SAl Viro { 6079883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 608756daf26SNeilBrown 609756daf26SNeilBrown current->nameidata = old; 610756daf26SNeilBrown if (old) 611756daf26SNeilBrown old->total_link_count = now->total_link_count; 612e1a63bbcSAl Viro if (now->stack != now->internal) 613756daf26SNeilBrown kfree(now->stack); 614894bc8c4SAl Viro } 615894bc8c4SAl Viro 61660ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 617894bc8c4SAl Viro { 618bc40aee0SAl Viro struct saved *p; 619bc40aee0SAl Viro 6206da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 62160ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 622bc40aee0SAl Viro if (unlikely(!p)) 62360ef60c7SAl Viro return false; 624894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 625894bc8c4SAl Viro nd->stack = p; 62660ef60c7SAl Viro return true; 627894bc8c4SAl Viro } 628894bc8c4SAl Viro 629397d425dSEric W. Biederman /** 6306b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 631397d425dSEric W. Biederman * 632397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 633397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 634397d425dSEric W. Biederman */ 6356b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 636397d425dSEric W. Biederman { 63795dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 638397d425dSEric W. Biederman 639402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 640402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 641397d425dSEric W. Biederman return true; 642397d425dSEric W. Biederman 6436b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 644397d425dSEric W. Biederman } 645397d425dSEric W. Biederman 6467973387aSAl Viro static void drop_links(struct nameidata *nd) 6477973387aSAl Viro { 6487973387aSAl Viro int i = nd->depth; 6497973387aSAl Viro while (i--) { 6507973387aSAl Viro struct saved *last = nd->stack + i; 651fceef393SAl Viro do_delayed_call(&last->done); 652fceef393SAl Viro clear_delayed_call(&last->done); 6537973387aSAl Viro } 6547973387aSAl Viro } 6557973387aSAl Viro 6567973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6577973387aSAl Viro { 6587973387aSAl Viro drop_links(nd); 6597973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6607973387aSAl Viro int i; 6617973387aSAl Viro path_put(&nd->path); 6627973387aSAl Viro for (i = 0; i < nd->depth; i++) 6637973387aSAl Viro path_put(&nd->stack[i].link); 664bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 665102b8af2SAl Viro path_put(&nd->root); 666bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 667102b8af2SAl Viro } 6687973387aSAl Viro } else { 6697973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6707973387aSAl Viro rcu_read_unlock(); 6717973387aSAl Viro } 6727973387aSAl Viro nd->depth = 0; 6737d01ef75SAl Viro nd->path.mnt = NULL; 6747d01ef75SAl Viro nd->path.dentry = NULL; 6757973387aSAl Viro } 6767973387aSAl Viro 6777973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6782aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6797973387aSAl Viro { 6802aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6817973387aSAl Viro if (unlikely(res)) { 6827973387aSAl Viro if (res > 0) 6837973387aSAl Viro path->mnt = NULL; 6847973387aSAl Viro path->dentry = NULL; 6857973387aSAl Viro return false; 6867973387aSAl Viro } 6877973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6887973387aSAl Viro path->dentry = NULL; 6897973387aSAl Viro return false; 6907973387aSAl Viro } 6917973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6927973387aSAl Viro } 6937973387aSAl Viro 6942aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 6952aa38470SAl Viro struct path *path, unsigned seq) 6962aa38470SAl Viro { 6975bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 6982aa38470SAl Viro } 6992aa38470SAl Viro 7007973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7017973387aSAl Viro { 7027973387aSAl Viro int i; 703eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 704eacd9aa8SAl Viro drop_links(nd); 705eacd9aa8SAl Viro nd->depth = 0; 706eacd9aa8SAl Viro return false; 707eacd9aa8SAl Viro } 7087973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7097973387aSAl Viro struct saved *last = nd->stack + i; 7107973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7117973387aSAl Viro drop_links(nd); 7127973387aSAl Viro nd->depth = i + 1; 7137973387aSAl Viro return false; 7147973387aSAl Viro } 7157973387aSAl Viro } 7167973387aSAl Viro return true; 7177973387aSAl Viro } 7187973387aSAl Viro 719ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 720ee594bffSAl Viro { 721adb21d2bSAleksa Sarai /* 722adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 723adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 724adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 725adb21d2bSAleksa Sarai */ 726adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 727adb21d2bSAleksa Sarai return false; 728adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 729bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 730ee594bffSAl Viro return true; 731bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 732ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 733ee594bffSAl Viro } 734ee594bffSAl Viro 73519660af7SAl Viro /* 73631e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 73719660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 73819660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 73957e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 74019660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 74119660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 74219660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 74319660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 74431e6b01fSNick Piggin */ 74531e6b01fSNick Piggin 74631e6b01fSNick Piggin /** 747e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 74819660af7SAl Viro * @nd: nameidata pathwalk data 749e36cffedSJens Axboe * Returns: true on success, false on failure 75031e6b01fSNick Piggin * 751e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7524675ac39SAl Viro * for ref-walk mode. 7534675ac39SAl Viro * Must be called from rcu-walk context. 754e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7557973387aSAl Viro * terminate_walk(). 75631e6b01fSNick Piggin */ 757e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 75831e6b01fSNick Piggin { 75931e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 76031e6b01fSNick Piggin 76131e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 762e5c832d5SLinus Torvalds 763e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 7647973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7654675ac39SAl Viro goto out1; 76684a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 76784a2bd39SAl Viro goto out; 768ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7694675ac39SAl Viro goto out; 7704675ac39SAl Viro rcu_read_unlock(); 7714675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 772e36cffedSJens Axboe return true; 7734675ac39SAl Viro 77484a2bd39SAl Viro out1: 7754675ac39SAl Viro nd->path.mnt = NULL; 7764675ac39SAl Viro nd->path.dentry = NULL; 7774675ac39SAl Viro out: 7784675ac39SAl Viro rcu_read_unlock(); 779e36cffedSJens Axboe return false; 7804675ac39SAl Viro } 7814675ac39SAl Viro 7824675ac39SAl Viro /** 783ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7844675ac39SAl Viro * @nd: nameidata pathwalk data 785ae66db45SAl Viro * @dentry: next dentry to step into 786ae66db45SAl Viro * @seq: seq number to check @dentry against 787ae66db45SAl Viro * Returns: true on success, false on failure 7884675ac39SAl Viro * 789ae66db45SAl Viro * Similar to to try_to_unlazy(), but here we have the next dentry already 790ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 791ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 792ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 7934675ac39SAl Viro * terminate_walk(). 7944675ac39SAl Viro */ 795ae66db45SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7964675ac39SAl Viro { 7974675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7984675ac39SAl Viro 7994675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 8004675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8014675ac39SAl Viro goto out2; 8027973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 8037973387aSAl Viro goto out2; 8044675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8057973387aSAl Viro goto out1; 80648a066e7SAl Viro 80715570086SLinus Torvalds /* 8084675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8094675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8104675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8114675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8124675ac39SAl Viro * know the parent must still be valid if the child sequence number is 81315570086SLinus Torvalds */ 8144675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 815e5c832d5SLinus Torvalds goto out; 81684a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 81784a2bd39SAl Viro goto out_dput; 818e5c832d5SLinus Torvalds /* 819e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 820e5c832d5SLinus Torvalds * still valid and get it if required. 821e5c832d5SLinus Torvalds */ 82284a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 82384a2bd39SAl Viro goto out_dput; 8248b61e74fSAl Viro rcu_read_unlock(); 825ae66db45SAl Viro return true; 82619660af7SAl Viro 8277973387aSAl Viro out2: 8287973387aSAl Viro nd->path.mnt = NULL; 8297973387aSAl Viro out1: 8307973387aSAl Viro nd->path.dentry = NULL; 831e5c832d5SLinus Torvalds out: 8328b61e74fSAl Viro rcu_read_unlock(); 833ae66db45SAl Viro return false; 83484a2bd39SAl Viro out_dput: 83584a2bd39SAl Viro rcu_read_unlock(); 83684a2bd39SAl Viro dput(dentry); 837ae66db45SAl Viro return false; 83831e6b01fSNick Piggin } 83931e6b01fSNick Piggin 8404ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 84134286d66SNick Piggin { 842a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8434ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 844a89f8337SAl Viro else 845a89f8337SAl Viro return 1; 84634286d66SNick Piggin } 84734286d66SNick Piggin 8489f1fafeeSAl Viro /** 8499f1fafeeSAl Viro * complete_walk - successful completion of path walk 8509f1fafeeSAl Viro * @nd: pointer nameidata 85139159de2SJeff Layton * 8529f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8539f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8549f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8559f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8569f1fafeeSAl Viro * need to drop nd->path. 85739159de2SJeff Layton */ 8589f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 85939159de2SJeff Layton { 86016c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 86139159de2SJeff Layton int status; 86239159de2SJeff Layton 8639f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 864adb21d2bSAleksa Sarai /* 865adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 866adb21d2bSAleksa Sarai * externally-managed nd->root. 867adb21d2bSAleksa Sarai */ 868bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 869bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8709f1fafeeSAl Viro nd->root.mnt = NULL; 8716c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 872e36cffedSJens Axboe if (!try_to_unlazy(nd)) 87348a066e7SAl Viro return -ECHILD; 87448a066e7SAl Viro } 8759f1fafeeSAl Viro 876adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 877adb21d2bSAleksa Sarai /* 878adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 879adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 880adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 881adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 882adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 883adb21d2bSAleksa Sarai * 884adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 885adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 886adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 887adb21d2bSAleksa Sarai * requested root to userspace. 888adb21d2bSAleksa Sarai * 889adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 890adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 891adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 892adb21d2bSAleksa Sarai */ 893adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 894adb21d2bSAleksa Sarai return -EXDEV; 895adb21d2bSAleksa Sarai } 896adb21d2bSAleksa Sarai 897bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 89839159de2SJeff Layton return 0; 89939159de2SJeff Layton 900ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 90116c2cd71SAl Viro return 0; 90216c2cd71SAl Viro 903ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 90439159de2SJeff Layton if (status > 0) 90539159de2SJeff Layton return 0; 90639159de2SJeff Layton 90716c2cd71SAl Viro if (!status) 90839159de2SJeff Layton status = -ESTALE; 90916c2cd71SAl Viro 91039159de2SJeff Layton return status; 91139159de2SJeff Layton } 91239159de2SJeff Layton 913740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9142a737871SAl Viro { 91531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9169e6697e2SAl Viro 917adb21d2bSAleksa Sarai /* 918adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 919adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 920adb21d2bSAleksa Sarai * from the dirfd. 921adb21d2bSAleksa Sarai */ 922adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 923adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 924adb21d2bSAleksa Sarai 9259e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9268f47a016SAl Viro unsigned seq; 927c28cc364SNick Piggin 928c28cc364SNick Piggin do { 929c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 93031e6b01fSNick Piggin nd->root = fs->root; 9318f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 932c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9339e6697e2SAl Viro } else { 9349e6697e2SAl Viro get_fs_root(fs, &nd->root); 935bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9369e6697e2SAl Viro } 937740a1678SAleksa Sarai return 0; 93831e6b01fSNick Piggin } 93931e6b01fSNick Piggin 940248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 941248fb5b9SAl Viro { 942adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 943adb21d2bSAleksa Sarai return -EXDEV; 94472ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 94572ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 94672ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 94772ba2929SAleksa Sarai return -EXDEV; 94872ba2929SAleksa Sarai } 949740a1678SAleksa Sarai if (!nd->root.mnt) { 950740a1678SAleksa Sarai int error = set_root(nd); 951740a1678SAleksa Sarai if (error) 952740a1678SAleksa Sarai return error; 953740a1678SAleksa Sarai } 954248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 955248fb5b9SAl Viro struct dentry *d; 956248fb5b9SAl Viro nd->path = nd->root; 957248fb5b9SAl Viro d = nd->path.dentry; 958248fb5b9SAl Viro nd->inode = d->d_inode; 959248fb5b9SAl Viro nd->seq = nd->root_seq; 960248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 961248fb5b9SAl Viro return -ECHILD; 962248fb5b9SAl Viro } else { 963248fb5b9SAl Viro path_put(&nd->path); 964248fb5b9SAl Viro nd->path = nd->root; 965248fb5b9SAl Viro path_get(&nd->path); 966248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 967248fb5b9SAl Viro } 968bcba1e7dSAl Viro nd->state |= ND_JUMPED; 969248fb5b9SAl Viro return 0; 970248fb5b9SAl Viro } 971248fb5b9SAl Viro 972b5fb63c1SChristoph Hellwig /* 9736b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 974b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 975b5fb63c1SChristoph Hellwig */ 9761bc82070SAleksa Sarai int nd_jump_link(struct path *path) 977b5fb63c1SChristoph Hellwig { 9784b99d499SAleksa Sarai int error = -ELOOP; 9796e77137bSAl Viro struct nameidata *nd = current->nameidata; 980b5fb63c1SChristoph Hellwig 9814b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9824b99d499SAleksa Sarai goto err; 9834b99d499SAleksa Sarai 98472ba2929SAleksa Sarai error = -EXDEV; 98572ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 98672ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 98772ba2929SAleksa Sarai goto err; 98872ba2929SAleksa Sarai } 989adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 990adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 991adb21d2bSAleksa Sarai goto err; 99272ba2929SAleksa Sarai 9934b99d499SAleksa Sarai path_put(&nd->path); 994b5fb63c1SChristoph Hellwig nd->path = *path; 995b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 996bcba1e7dSAl Viro nd->state |= ND_JUMPED; 9971bc82070SAleksa Sarai return 0; 9984b99d499SAleksa Sarai 9994b99d499SAleksa Sarai err: 10004b99d499SAleksa Sarai path_put(path); 10014b99d499SAleksa Sarai return error; 1002b5fb63c1SChristoph Hellwig } 1003b5fb63c1SChristoph Hellwig 1004b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1005574197e0SAl Viro { 100621c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1007fceef393SAl Viro do_delayed_call(&last->done); 10086548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1009b9ff4429SAl Viro path_put(&last->link); 1010574197e0SAl Viro } 1011574197e0SAl Viro 1012561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 1013561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 101430aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 101530aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 1016800179c9SKees Cook 1017800179c9SKees Cook /** 1018800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 101955852635SRandy Dunlap * @nd: nameidata pathwalk data 1020800179c9SKees Cook * 1021800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1022800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1023800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1024800179c9SKees Cook * processes from failing races against path names that may change out 1025800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1026800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1027800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1028800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1029800179c9SKees Cook * 1030800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1031800179c9SKees Cook */ 1032ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1033800179c9SKees Cook { 1034ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1035ba73d987SChristian Brauner kuid_t i_uid; 1036ba73d987SChristian Brauner 1037800179c9SKees Cook if (!sysctl_protected_symlinks) 1038800179c9SKees Cook return 0; 1039800179c9SKees Cook 1040ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1041ba73d987SChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 1042800179c9SKees Cook /* Allowed if owner and follower match. */ 1043ba73d987SChristian Brauner if (uid_eq(current_cred()->fsuid, i_uid)) 1044800179c9SKees Cook return 0; 1045800179c9SKees Cook 1046800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 10470f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1048800179c9SKees Cook return 0; 1049800179c9SKees Cook 1050800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1051ba73d987SChristian Brauner if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) 1052800179c9SKees Cook return 0; 1053800179c9SKees Cook 105431956502SAl Viro if (nd->flags & LOOKUP_RCU) 105531956502SAl Viro return -ECHILD; 105631956502SAl Viro 1057ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1058245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1059800179c9SKees Cook return -EACCES; 1060800179c9SKees Cook } 1061800179c9SKees Cook 1062800179c9SKees Cook /** 1063800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1064ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1065800179c9SKees Cook * @inode: the source inode to hardlink from 1066800179c9SKees Cook * 1067800179c9SKees Cook * Return false if at least one of the following conditions: 1068800179c9SKees Cook * - inode is not a regular file 1069800179c9SKees Cook * - inode is setuid 1070800179c9SKees Cook * - inode is setgid and group-exec 1071800179c9SKees Cook * - access failure for read and write 1072800179c9SKees Cook * 1073800179c9SKees Cook * Otherwise returns true. 1074800179c9SKees Cook */ 1075ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1076ba73d987SChristian Brauner struct inode *inode) 1077800179c9SKees Cook { 1078800179c9SKees Cook umode_t mode = inode->i_mode; 1079800179c9SKees Cook 1080800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1081800179c9SKees Cook if (!S_ISREG(mode)) 1082800179c9SKees Cook return false; 1083800179c9SKees Cook 1084800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1085800179c9SKees Cook if (mode & S_ISUID) 1086800179c9SKees Cook return false; 1087800179c9SKees Cook 1088800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1089800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1090800179c9SKees Cook return false; 1091800179c9SKees Cook 1092800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1093ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1094800179c9SKees Cook return false; 1095800179c9SKees Cook 1096800179c9SKees Cook return true; 1097800179c9SKees Cook } 1098800179c9SKees Cook 1099800179c9SKees Cook /** 1100800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1101ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1102800179c9SKees Cook * @link: the source to hardlink from 1103800179c9SKees Cook * 1104800179c9SKees Cook * Block hardlink when all of: 1105800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1106800179c9SKees Cook * - fsuid does not match inode 1107800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1108f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1109800179c9SKees Cook * 1110ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1111ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1112ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1113ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1114ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1115ba73d987SChristian Brauner * 1116800179c9SKees Cook * Returns 0 if successful, -ve on error. 1117800179c9SKees Cook */ 1118ba73d987SChristian Brauner int may_linkat(struct user_namespace *mnt_userns, struct path *link) 1119800179c9SKees Cook { 1120593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1121593d1ce8SEric W. Biederman 1122593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1123ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 1124ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 1125593d1ce8SEric W. Biederman return -EOVERFLOW; 1126800179c9SKees Cook 1127800179c9SKees Cook if (!sysctl_protected_hardlinks) 1128800179c9SKees Cook return 0; 1129800179c9SKees Cook 1130800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1131800179c9SKees Cook * otherwise, it must be a safe source. 1132800179c9SKees Cook */ 1133ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1134ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1135800179c9SKees Cook return 0; 1136800179c9SKees Cook 1137245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1138800179c9SKees Cook return -EPERM; 1139800179c9SKees Cook } 1140800179c9SKees Cook 114130aba665SSalvatore Mesoraca /** 114230aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 114330aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 114430aba665SSalvatore Mesoraca * exist. 1145ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 11462111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 114730aba665SSalvatore Mesoraca * @inode: the inode of the file to open 114830aba665SSalvatore Mesoraca * 114930aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 115030aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 115130aba665SSalvatore Mesoraca * - the file already exists 115230aba665SSalvatore Mesoraca * - we are in a sticky directory 115330aba665SSalvatore Mesoraca * - we don't own the file 115430aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 115530aba665SSalvatore Mesoraca * - the directory is world writable 115630aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 115730aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 115830aba665SSalvatore Mesoraca * be enough. 115930aba665SSalvatore Mesoraca * 1160ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1161ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1162ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1163ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1164ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1165ba73d987SChristian Brauner * 116630aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 116730aba665SSalvatore Mesoraca */ 1168ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1169ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 117030aba665SSalvatore Mesoraca { 1171ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1172ba73d987SChristian Brauner kuid_t dir_uid = nd->dir_uid; 1173ba73d987SChristian Brauner 117430aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 117530aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1176d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1177ba73d987SChristian Brauner uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || 1178ba73d987SChristian Brauner uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) 117930aba665SSalvatore Mesoraca return 0; 118030aba665SSalvatore Mesoraca 1181d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1182d0cb5018SAl Viro (dir_mode & 0020 && 118330aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 118430aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1185245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1186245d7369SKees Cook "sticky_create_fifo" : 1187245d7369SKees Cook "sticky_create_regular"; 1188245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 118930aba665SSalvatore Mesoraca return -EACCES; 119030aba665SSalvatore Mesoraca } 119130aba665SSalvatore Mesoraca return 0; 119230aba665SSalvatore Mesoraca } 119330aba665SSalvatore Mesoraca 1194f015f126SDavid Howells /* 1195f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1196f015f126SDavid Howells * 1197f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1198f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1199f015f126SDavid Howells * Up is towards /. 1200f015f126SDavid Howells * 1201f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1202f015f126SDavid Howells * root. 1203f015f126SDavid Howells */ 1204bab77ebfSAl Viro int follow_up(struct path *path) 12051da177e4SLinus Torvalds { 12060714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12070714a533SAl Viro struct mount *parent; 12081da177e4SLinus Torvalds struct dentry *mountpoint; 120999b7db7bSNick Piggin 121048a066e7SAl Viro read_seqlock_excl(&mount_lock); 12110714a533SAl Viro parent = mnt->mnt_parent; 12123c0a6163SAl Viro if (parent == mnt) { 121348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12141da177e4SLinus Torvalds return 0; 12151da177e4SLinus Torvalds } 12160714a533SAl Viro mntget(&parent->mnt); 1217a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 121848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1219bab77ebfSAl Viro dput(path->dentry); 1220bab77ebfSAl Viro path->dentry = mountpoint; 1221bab77ebfSAl Viro mntput(path->mnt); 12220714a533SAl Viro path->mnt = &parent->mnt; 12231da177e4SLinus Torvalds return 1; 12241da177e4SLinus Torvalds } 12254d359507SAl Viro EXPORT_SYMBOL(follow_up); 12261da177e4SLinus Torvalds 12277ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12287ef482faSAl Viro struct path *path, unsigned *seqp) 12297ef482faSAl Viro { 12307ef482faSAl Viro while (mnt_has_parent(m)) { 12317ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12327ef482faSAl Viro 12337ef482faSAl Viro m = m->mnt_parent; 12347ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12357ef482faSAl Viro root->mnt == &m->mnt)) 12367ef482faSAl Viro break; 12377ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 12387ef482faSAl Viro path->mnt = &m->mnt; 12397ef482faSAl Viro path->dentry = mountpoint; 12407ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 12417ef482faSAl Viro return true; 12427ef482faSAl Viro } 12437ef482faSAl Viro } 12447ef482faSAl Viro return false; 12457ef482faSAl Viro } 12467ef482faSAl Viro 12472aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 12482aa38470SAl Viro struct path *path) 12492aa38470SAl Viro { 12502aa38470SAl Viro bool found; 12512aa38470SAl Viro 12522aa38470SAl Viro rcu_read_lock(); 12532aa38470SAl Viro while (1) { 12542aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 12552aa38470SAl Viro 12562aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 12572aa38470SAl Viro if (unlikely(!found)) { 12582aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 12592aa38470SAl Viro break; 12602aa38470SAl Viro } else { 12612aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 12622aa38470SAl Viro break; 12632aa38470SAl Viro rcu_read_unlock(); 12642aa38470SAl Viro path_put(path); 12652aa38470SAl Viro rcu_read_lock(); 12662aa38470SAl Viro } 12672aa38470SAl Viro } 12682aa38470SAl Viro rcu_read_unlock(); 12692aa38470SAl Viro return found; 12702aa38470SAl Viro } 12712aa38470SAl Viro 1272b5c84bf6SNick Piggin /* 12739875cf80SDavid Howells * Perform an automount 12749875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 12759875cf80SDavid Howells * were called with. 12761da177e4SLinus Torvalds */ 12771c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 127831e6b01fSNick Piggin { 127925e195aaSAl Viro struct dentry *dentry = path->dentry; 12809875cf80SDavid Howells 12810ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 12820ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 12830ec26fd0SMiklos Szeredi * the name. 12840ec26fd0SMiklos Szeredi * 12850ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 12865a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 12870ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 12880ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 12890ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 12900ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 12915a30d8a2SDavid Howells */ 12921c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 12935d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 129425e195aaSAl Viro dentry->d_inode) 12959875cf80SDavid Howells return -EISDIR; 12960ec26fd0SMiklos Szeredi 12971c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 12989875cf80SDavid Howells return -ELOOP; 12999875cf80SDavid Howells 130025e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1301ea5b778aSDavid Howells } 13029875cf80SDavid Howells 13039875cf80SDavid Howells /* 13049deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13059deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13069deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13079deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13089875cf80SDavid Howells */ 13099deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13109deed3ebSAl Viro int *count, unsigned lookup_flags) 13119875cf80SDavid Howells { 13129deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13139875cf80SDavid Howells bool need_mntput = false; 13148aef1884SAl Viro int ret = 0; 13159875cf80SDavid Howells 13169deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1317cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1318cc53ce53SDavid Howells * being held. */ 1319d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1320fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1321508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1322cc53ce53SDavid Howells if (ret < 0) 13238aef1884SAl Viro break; 1324cc53ce53SDavid Howells } 1325cc53ce53SDavid Howells 13269deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13279875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13289deed3ebSAl Viro if (mounted) { // ... in our namespace 13299875cf80SDavid Howells dput(path->dentry); 13309875cf80SDavid Howells if (need_mntput) 1331463ffb2eSAl Viro mntput(path->mnt); 1332463ffb2eSAl Viro path->mnt = mounted; 1333463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13349deed3ebSAl Viro // here we know it's positive 13359deed3ebSAl Viro flags = path->dentry->d_flags; 13369875cf80SDavid Howells need_mntput = true; 13379875cf80SDavid Howells continue; 1338463ffb2eSAl Viro } 13391da177e4SLinus Torvalds } 13409875cf80SDavid Howells 13419deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 13429deed3ebSAl Viro break; 13439deed3ebSAl Viro 13449deed3ebSAl Viro // uncovered automount point 13459deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 13469deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 13479875cf80SDavid Howells if (ret < 0) 13488aef1884SAl Viro break; 13499875cf80SDavid Howells } 13509875cf80SDavid Howells 13519deed3ebSAl Viro if (ret == -EISDIR) 13529deed3ebSAl Viro ret = 0; 13539deed3ebSAl Viro // possible if you race with several mount --move 13549deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 13558aef1884SAl Viro mntput(path->mnt); 13569deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1357d41efb52SAl Viro ret = -ENOENT; 13589deed3ebSAl Viro *jumped = need_mntput; 13598402752eSAl Viro return ret; 13601da177e4SLinus Torvalds } 13611da177e4SLinus Torvalds 13629deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 13639deed3ebSAl Viro int *count, unsigned lookup_flags) 13649deed3ebSAl Viro { 13659deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 13669deed3ebSAl Viro 13679deed3ebSAl Viro /* fastpath */ 13689deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 13699deed3ebSAl Viro *jumped = false; 13709deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 13719deed3ebSAl Viro return -ENOENT; 13729deed3ebSAl Viro return 0; 13739deed3ebSAl Viro } 13749deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 13759deed3ebSAl Viro } 13769deed3ebSAl Viro 1377cc53ce53SDavid Howells int follow_down_one(struct path *path) 13781da177e4SLinus Torvalds { 13791da177e4SLinus Torvalds struct vfsmount *mounted; 13801da177e4SLinus Torvalds 13811c755af4SAl Viro mounted = lookup_mnt(path); 13821da177e4SLinus Torvalds if (mounted) { 13839393bd07SAl Viro dput(path->dentry); 13849393bd07SAl Viro mntput(path->mnt); 13859393bd07SAl Viro path->mnt = mounted; 13869393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 13871da177e4SLinus Torvalds return 1; 13881da177e4SLinus Torvalds } 13891da177e4SLinus Torvalds return 0; 13901da177e4SLinus Torvalds } 13914d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 13921da177e4SLinus Torvalds 13939875cf80SDavid Howells /* 13949deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 13959deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 13969deed3ebSAl Viro * caller is permitted to proceed or not. 13979deed3ebSAl Viro */ 13989deed3ebSAl Viro int follow_down(struct path *path) 13999deed3ebSAl Viro { 14009deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14019deed3ebSAl Viro bool jumped; 14029deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 14039deed3ebSAl Viro 14049deed3ebSAl Viro if (path->mnt != mnt) 14059deed3ebSAl Viro mntput(mnt); 14069deed3ebSAl Viro return ret; 14079deed3ebSAl Viro } 14089deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14099deed3ebSAl Viro 14109deed3ebSAl Viro /* 1411287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1412287548e4SAl Viro * we meet a managed dentry that would need blocking. 14139875cf80SDavid Howells */ 14149875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1415254cf582SAl Viro struct inode **inode, unsigned *seqp) 14169875cf80SDavid Howells { 1417ea936aebSAl Viro struct dentry *dentry = path->dentry; 1418ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1419ea936aebSAl Viro 1420ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1421ea936aebSAl Viro return true; 1422ea936aebSAl Viro 1423ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1424ea936aebSAl Viro return false; 1425ea936aebSAl Viro 142662a7375eSIan Kent for (;;) { 142762a7375eSIan Kent /* 142862a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 142962a7375eSIan Kent * that wants to block transit. 143062a7375eSIan Kent */ 1431ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1432ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1433ea936aebSAl Viro if (res) 1434ea936aebSAl Viro return res == -EISDIR; 1435ea936aebSAl Viro flags = dentry->d_flags; 1436b8faf035SNeilBrown } 143762a7375eSIan Kent 1438ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1439ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1440ea936aebSAl Viro if (mounted) { 1441c7105365SAl Viro path->mnt = &mounted->mnt; 1442ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1443bcba1e7dSAl Viro nd->state |= ND_JUMPED; 1444ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1445ea936aebSAl Viro *inode = dentry->d_inode; 144659430262SLinus Torvalds /* 1447ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1448ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1449ea936aebSAl Viro * between mount hash removal and ->mnt_root 1450ea936aebSAl Viro * becoming unpinned. 145159430262SLinus Torvalds */ 1452ea936aebSAl Viro flags = dentry->d_flags; 1453ea936aebSAl Viro continue; 14549875cf80SDavid Howells } 1455ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1456ea936aebSAl Viro return false; 1457ea936aebSAl Viro } 1458ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1459ea936aebSAl Viro } 1460287548e4SAl Viro } 1461287548e4SAl Viro 1462db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1463db3c9adeSAl Viro struct path *path, struct inode **inode, 1464db3c9adeSAl Viro unsigned int *seqp) 1465bd7c4b50SAl Viro { 14669deed3ebSAl Viro bool jumped; 1467db3c9adeSAl Viro int ret; 1468bd7c4b50SAl Viro 1469db3c9adeSAl Viro path->mnt = nd->path.mnt; 1470db3c9adeSAl Viro path->dentry = dentry; 1471c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1472c153007bSAl Viro unsigned int seq = *seqp; 1473c153007bSAl Viro if (unlikely(!*inode)) 1474c153007bSAl Viro return -ENOENT; 1475c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 14769deed3ebSAl Viro return 0; 1477ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 1478c153007bSAl Viro return -ECHILD; 1479c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1480c153007bSAl Viro path->mnt = nd->path.mnt; 1481c153007bSAl Viro path->dentry = dentry; 1482c153007bSAl Viro } 14839deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 14849deed3ebSAl Viro if (jumped) { 14859deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 14869deed3ebSAl Viro ret = -EXDEV; 14879deed3ebSAl Viro else 1488bcba1e7dSAl Viro nd->state |= ND_JUMPED; 14899deed3ebSAl Viro } 14909deed3ebSAl Viro if (unlikely(ret)) { 14919deed3ebSAl Viro dput(path->dentry); 14929deed3ebSAl Viro if (path->mnt != nd->path.mnt) 14939deed3ebSAl Viro mntput(path->mnt); 14949deed3ebSAl Viro } else { 1495bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1496bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1497bd7c4b50SAl Viro } 1498bd7c4b50SAl Viro return ret; 1499bd7c4b50SAl Viro } 1500bd7c4b50SAl Viro 15019875cf80SDavid Howells /* 1502f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1503f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1504baa03890SNick Piggin */ 1505e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1506e3c13928SAl Viro struct dentry *dir, 15076c51e513SAl Viro unsigned int flags) 1508baa03890SNick Piggin { 1509a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1510bad61189SMiklos Szeredi if (dentry) { 1511a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1512bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 151374ff0ffcSAl Viro if (!error) 15145542aa2fSEric W. Biederman d_invalidate(dentry); 1515bad61189SMiklos Szeredi dput(dentry); 151674ff0ffcSAl Viro return ERR_PTR(error); 1517bad61189SMiklos Szeredi } 1518bad61189SMiklos Szeredi } 1519baa03890SNick Piggin return dentry; 1520baa03890SNick Piggin } 1521baa03890SNick Piggin 1522baa03890SNick Piggin /* 1523a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1524a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1525a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1526a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1527a03ece5fSAl Viro * at all. 152844396f4bSJosef Bacik */ 1529a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1530a03ece5fSAl Viro struct dentry *base, unsigned int flags) 153144396f4bSJosef Bacik { 1532a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 153344396f4bSJosef Bacik struct dentry *old; 1534a03ece5fSAl Viro struct inode *dir = base->d_inode; 1535a03ece5fSAl Viro 1536a03ece5fSAl Viro if (dentry) 1537a03ece5fSAl Viro return dentry; 153844396f4bSJosef Bacik 153944396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1540a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 154144396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1542a03ece5fSAl Viro 1543a03ece5fSAl Viro dentry = d_alloc(base, name); 1544a03ece5fSAl Viro if (unlikely(!dentry)) 1545a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 154644396f4bSJosef Bacik 154772bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 154844396f4bSJosef Bacik if (unlikely(old)) { 154944396f4bSJosef Bacik dput(dentry); 155044396f4bSJosef Bacik dentry = old; 155144396f4bSJosef Bacik } 155244396f4bSJosef Bacik return dentry; 155344396f4bSJosef Bacik } 155444396f4bSJosef Bacik 155520e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 155620e34357SAl Viro struct inode **inode, 1557254cf582SAl Viro unsigned *seqp) 15581da177e4SLinus Torvalds { 155931e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15605a18fff2SAl Viro int status = 1; 15619875cf80SDavid Howells 15623cac260aSAl Viro /* 1563b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15645d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15655d0f49c1SAl Viro * going to fall back to non-racy lookup. 1566b04f784eSNick Piggin */ 156731e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 156831e6b01fSNick Piggin unsigned seq; 1569da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15705d0f49c1SAl Viro if (unlikely(!dentry)) { 1571e36cffedSJens Axboe if (!try_to_unlazy(nd)) 157220e34357SAl Viro return ERR_PTR(-ECHILD); 157320e34357SAl Viro return NULL; 15745d0f49c1SAl Viro } 15755a18fff2SAl Viro 157612f8ad4bSLinus Torvalds /* 157712f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 157812f8ad4bSLinus Torvalds * the dentry name information from lookup. 157912f8ad4bSLinus Torvalds */ 158063afdfc7SDavid Howells *inode = d_backing_inode(dentry); 15815d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 158220e34357SAl Viro return ERR_PTR(-ECHILD); 158312f8ad4bSLinus Torvalds 158412f8ad4bSLinus Torvalds /* 158512f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 158612f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 158712f8ad4bSLinus Torvalds * 158812f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 158912f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 159012f8ad4bSLinus Torvalds */ 15915d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 159220e34357SAl Viro return ERR_PTR(-ECHILD); 15935a18fff2SAl Viro 1594254cf582SAl Viro *seqp = seq; 15954ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1596c153007bSAl Viro if (likely(status > 0)) 159720e34357SAl Viro return dentry; 1598ae66db45SAl Viro if (!try_to_unlazy_next(nd, dentry, seq)) 159920e34357SAl Viro return ERR_PTR(-ECHILD); 160026ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1601209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1602209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16035a18fff2SAl Viro } else { 1604e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 160581e6f520SAl Viro if (unlikely(!dentry)) 160620e34357SAl Viro return NULL; 16074ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16085d0f49c1SAl Viro } 16095a18fff2SAl Viro if (unlikely(status <= 0)) { 1610e9742b53SAl Viro if (!status) 16115d0f49c1SAl Viro d_invalidate(dentry); 16125a18fff2SAl Viro dput(dentry); 161320e34357SAl Viro return ERR_PTR(status); 16145a18fff2SAl Viro } 161520e34357SAl Viro return dentry; 1616697f514dSMiklos Szeredi } 1617697f514dSMiklos Szeredi 1618697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 161988d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1620e3c13928SAl Viro struct dentry *dir, 1621e3c13928SAl Viro unsigned int flags) 1622697f514dSMiklos Szeredi { 162388d8331aSAl Viro struct dentry *dentry, *old; 16241936386eSAl Viro struct inode *inode = dir->d_inode; 1625d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16261936386eSAl Viro 16271936386eSAl Viro /* Don't go there if it's already dead */ 162894bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 162988d8331aSAl Viro return ERR_PTR(-ENOENT); 163094bdd655SAl Viro again: 1631d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 163294bdd655SAl Viro if (IS_ERR(dentry)) 163388d8331aSAl Viro return dentry; 163494bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1635949a852eSAl Viro int error = d_revalidate(dentry, flags); 1636949a852eSAl Viro if (unlikely(error <= 0)) { 163794bdd655SAl Viro if (!error) { 1638949a852eSAl Viro d_invalidate(dentry); 1639949a852eSAl Viro dput(dentry); 164094bdd655SAl Viro goto again; 164194bdd655SAl Viro } 164294bdd655SAl Viro dput(dentry); 1643949a852eSAl Viro dentry = ERR_PTR(error); 1644949a852eSAl Viro } 164594bdd655SAl Viro } else { 16461936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 164785c7f810SAl Viro d_lookup_done(dentry); 16481936386eSAl Viro if (unlikely(old)) { 16491936386eSAl Viro dput(dentry); 16501936386eSAl Viro dentry = old; 1651949a852eSAl Viro } 1652949a852eSAl Viro } 1653e3c13928SAl Viro return dentry; 16541da177e4SLinus Torvalds } 16551da177e4SLinus Torvalds 165688d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 165788d8331aSAl Viro struct dentry *dir, 165888d8331aSAl Viro unsigned int flags) 165988d8331aSAl Viro { 166088d8331aSAl Viro struct inode *inode = dir->d_inode; 166188d8331aSAl Viro struct dentry *res; 166288d8331aSAl Viro inode_lock_shared(inode); 166388d8331aSAl Viro res = __lookup_slow(name, dir, flags); 166488d8331aSAl Viro inode_unlock_shared(inode); 166588d8331aSAl Viro return res; 166688d8331aSAl Viro } 166788d8331aSAl Viro 1668ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1669ba73d987SChristian Brauner struct nameidata *nd) 167052094c8aSAl Viro { 167152094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 16727d6beb71SLinus Torvalds int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1673e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 167452094c8aSAl Viro return err; 167552094c8aSAl Viro } 1676ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 167752094c8aSAl Viro } 167852094c8aSAl Viro 167949055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1680d63ff28fSAl Viro { 168149055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 168249055906SAl Viro return -ELOOP; 16834542576bSAl Viro 16844542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 16854542576bSAl Viro return 0; 16864542576bSAl Viro if (likely(nd->stack != nd->internal)) 16874542576bSAl Viro return 0; 168860ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 168949055906SAl Viro return 0; 169060ef60c7SAl Viro 169160ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 169260ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 169360ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1694aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 169560ef60c7SAl Viro 1696e36cffedSJens Axboe if (!try_to_unlazy(nd) != 0 || !grabbed_link) 169760ef60c7SAl Viro return -ECHILD; 169860ef60c7SAl Viro 169960ef60c7SAl Viro if (nd_alloc_stack(nd)) 170060ef60c7SAl Viro return 0; 1701bc40aee0SAl Viro } 170260ef60c7SAl Viro return -ENOMEM; 170349055906SAl Viro } 170449055906SAl Viro 170549055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 170649055906SAl Viro 170749055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 170849055906SAl Viro struct inode *inode, unsigned seq, int flags) 170949055906SAl Viro { 171049055906SAl Viro struct saved *last; 171149055906SAl Viro const char *res; 171249055906SAl Viro int error = reserve_stack(nd, link, seq); 171349055906SAl Viro 171449055906SAl Viro if (unlikely(error)) { 171549055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1716cd179f44SAl Viro path_put(link); 171706708adbSAl Viro return ERR_PTR(error); 1718626de996SAl Viro } 1719ab104923SAl Viro last = nd->stack + nd->depth++; 17201cf2665bSAl Viro last->link = *link; 1721fceef393SAl Viro clear_delayed_call(&last->done); 17220450b2d1SAl Viro last->seq = seq; 1723ad6cc4c3SAl Viro 1724b1a81972SAl Viro if (flags & WALK_TRAILING) { 1725ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1726ad6cc4c3SAl Viro if (unlikely(error)) 1727ad6cc4c3SAl Viro return ERR_PTR(error); 1728ad6cc4c3SAl Viro } 1729ad6cc4c3SAl Viro 1730dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1731dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1732ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1733ad6cc4c3SAl Viro 1734ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1735ad6cc4c3SAl Viro touch_atime(&last->link); 1736ad6cc4c3SAl Viro cond_resched(); 1737ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1738e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1739ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1740ad6cc4c3SAl Viro touch_atime(&last->link); 1741ad6cc4c3SAl Viro } 1742ad6cc4c3SAl Viro 1743ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1744ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1745ad6cc4c3SAl Viro if (unlikely(error)) 1746ad6cc4c3SAl Viro return ERR_PTR(error); 1747ad6cc4c3SAl Viro 1748ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1749ad6cc4c3SAl Viro if (!res) { 1750ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1751ad6cc4c3SAl Viro struct delayed_call *); 1752ad6cc4c3SAl Viro get = inode->i_op->get_link; 1753ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1754ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1755e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1756ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1757ad6cc4c3SAl Viro } else { 1758ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1759ad6cc4c3SAl Viro } 1760ad6cc4c3SAl Viro if (!res) 1761ad6cc4c3SAl Viro goto all_done; 1762ad6cc4c3SAl Viro if (IS_ERR(res)) 1763ad6cc4c3SAl Viro return res; 1764ad6cc4c3SAl Viro } 1765ad6cc4c3SAl Viro if (*res == '/') { 1766ad6cc4c3SAl Viro error = nd_jump_root(nd); 1767ad6cc4c3SAl Viro if (unlikely(error)) 1768ad6cc4c3SAl Viro return ERR_PTR(error); 1769ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1770ad6cc4c3SAl Viro ; 1771ad6cc4c3SAl Viro } 1772ad6cc4c3SAl Viro if (*res) 1773ad6cc4c3SAl Viro return res; 1774ad6cc4c3SAl Viro all_done: // pure jump 1775ad6cc4c3SAl Viro put_link(nd); 1776ad6cc4c3SAl Viro return NULL; 1777d63ff28fSAl Viro } 1778d63ff28fSAl Viro 17793ddcd056SLinus Torvalds /* 17803ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17813ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17823ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17833ddcd056SLinus Torvalds * for the common case. 17843ddcd056SLinus Torvalds */ 1785b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1786cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 17873ddcd056SLinus Torvalds { 1788cbae4d12SAl Viro struct path path; 1789cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1790cbae4d12SAl Viro 1791cbae4d12SAl Viro if (err < 0) 1792b0417d2cSAl Viro return ERR_PTR(err); 1793cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 17948c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1795aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 17968f64fb1cSAl Viro /* not a symlink or should not follow */ 1797c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1798c99687a0SAl Viro dput(nd->path.dentry); 1799c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1800c99687a0SAl Viro mntput(nd->path.mnt); 1801c99687a0SAl Viro } 1802c99687a0SAl Viro nd->path = path; 18038f64fb1cSAl Viro nd->inode = inode; 18048f64fb1cSAl Viro nd->seq = seq; 1805b0417d2cSAl Viro return NULL; 18068f64fb1cSAl Viro } 1807a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 180884f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1809cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1810b0417d2cSAl Viro return ERR_PTR(-ECHILD); 181184f0cd9eSAl Viro } else { 181284f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 181384f0cd9eSAl Viro mntget(path.mnt); 1814a7f77542SAl Viro } 1815b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 18163ddcd056SLinus Torvalds } 18173ddcd056SLinus Torvalds 1818c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1819c2df1968SAl Viro struct inode **inodep, 1820c2df1968SAl Viro unsigned *seqp) 1821957dd41dSAl Viro { 182212487f30SAl Viro struct dentry *parent, *old; 1823957dd41dSAl Viro 182412487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 182512487f30SAl Viro goto in_root; 182612487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18277ef482faSAl Viro struct path path; 1828efe772d6SAl Viro unsigned seq; 18297ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18307ef482faSAl Viro &nd->root, &path, &seq)) 183112487f30SAl Viro goto in_root; 1832efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1833efe772d6SAl Viro return ERR_PTR(-ECHILD); 1834efe772d6SAl Viro nd->path = path; 1835efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1836efe772d6SAl Viro nd->seq = seq; 1837efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1838efe772d6SAl Viro return ERR_PTR(-ECHILD); 1839efe772d6SAl Viro /* we know that mountpoint was pinned */ 1840957dd41dSAl Viro } 184112487f30SAl Viro old = nd->path.dentry; 184212487f30SAl Viro parent = old->d_parent; 184312487f30SAl Viro *inodep = parent->d_inode; 184412487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 184512487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 184612487f30SAl Viro return ERR_PTR(-ECHILD); 184712487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 184812487f30SAl Viro return ERR_PTR(-ECHILD); 184912487f30SAl Viro return parent; 185012487f30SAl Viro in_root: 1851efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1852efe772d6SAl Viro return ERR_PTR(-ECHILD); 1853957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 18547521f22bSAl Viro return ERR_PTR(-ECHILD); 1855c2df1968SAl Viro return NULL; 1856957dd41dSAl Viro } 1857957dd41dSAl Viro 1858c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1859c2df1968SAl Viro struct inode **inodep, 1860c2df1968SAl Viro unsigned *seqp) 1861957dd41dSAl Viro { 186212487f30SAl Viro struct dentry *parent; 186312487f30SAl Viro 1864957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 186512487f30SAl Viro goto in_root; 186612487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18672aa38470SAl Viro struct path path; 18682aa38470SAl Viro 18692aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 18702aa38470SAl Viro &nd->root, &path)) 187112487f30SAl Viro goto in_root; 1872165200d6SAl Viro path_put(&nd->path); 1873165200d6SAl Viro nd->path = path; 18742aa38470SAl Viro nd->inode = path.dentry->d_inode; 1875165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1876165200d6SAl Viro return ERR_PTR(-EXDEV); 187712487f30SAl Viro } 1878957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 187912487f30SAl Viro parent = dget_parent(nd->path.dentry); 1880957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1881957dd41dSAl Viro dput(parent); 18827521f22bSAl Viro return ERR_PTR(-ENOENT); 1883957dd41dSAl Viro } 1884c2df1968SAl Viro *seqp = 0; 1885c2df1968SAl Viro *inodep = parent->d_inode; 1886c2df1968SAl Viro return parent; 188712487f30SAl Viro 188812487f30SAl Viro in_root: 1889957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 18907521f22bSAl Viro return ERR_PTR(-EXDEV); 1891c2df1968SAl Viro dget(nd->path.dentry); 1892c2df1968SAl Viro return NULL; 1893957dd41dSAl Viro } 1894957dd41dSAl Viro 18957521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1896957dd41dSAl Viro { 1897957dd41dSAl Viro if (type == LAST_DOTDOT) { 18987521f22bSAl Viro const char *error = NULL; 1899c2df1968SAl Viro struct dentry *parent; 1900c2df1968SAl Viro struct inode *inode; 1901c2df1968SAl Viro unsigned seq; 1902957dd41dSAl Viro 1903957dd41dSAl Viro if (!nd->root.mnt) { 19047521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1905957dd41dSAl Viro if (error) 1906957dd41dSAl Viro return error; 1907957dd41dSAl Viro } 1908957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1909c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1910957dd41dSAl Viro else 1911c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1912c2df1968SAl Viro if (IS_ERR(parent)) 1913c2df1968SAl Viro return ERR_CAST(parent); 1914c2df1968SAl Viro if (unlikely(!parent)) 1915c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1916c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1917c2df1968SAl Viro else 1918c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1919c2df1968SAl Viro parent, inode, seq); 1920c2df1968SAl Viro if (unlikely(error)) 1921957dd41dSAl Viro return error; 1922957dd41dSAl Viro 1923957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1924957dd41dSAl Viro /* 1925957dd41dSAl Viro * If there was a racing rename or mount along our 1926957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1927957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1928957dd41dSAl Viro * some fallback). 1929957dd41dSAl Viro */ 1930957dd41dSAl Viro smp_rmb(); 1931957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 19327521f22bSAl Viro return ERR_PTR(-EAGAIN); 1933957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 19347521f22bSAl Viro return ERR_PTR(-EAGAIN); 1935957dd41dSAl Viro } 1936957dd41dSAl Viro } 19377521f22bSAl Viro return NULL; 1938957dd41dSAl Viro } 1939957dd41dSAl Viro 194092d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1941ce57dfc1SAl Viro { 1942db3c9adeSAl Viro struct dentry *dentry; 1943ce57dfc1SAl Viro struct inode *inode; 1944254cf582SAl Viro unsigned seq; 1945ce57dfc1SAl Viro /* 1946ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1947ce57dfc1SAl Viro * to be able to know about the current root directory and 1948ce57dfc1SAl Viro * parent relationships. 1949ce57dfc1SAl Viro */ 19504693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 19511c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 19524693a547SAl Viro put_link(nd); 19537521f22bSAl Viro return handle_dots(nd, nd->last_type); 19544693a547SAl Viro } 195520e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 195620e34357SAl Viro if (IS_ERR(dentry)) 195792d27016SAl Viro return ERR_CAST(dentry); 195820e34357SAl Viro if (unlikely(!dentry)) { 1959db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1960db3c9adeSAl Viro if (IS_ERR(dentry)) 196192d27016SAl Viro return ERR_CAST(dentry); 196220e34357SAl Viro } 196356676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 196456676ec3SAl Viro put_link(nd); 1965b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1966ce57dfc1SAl Viro } 1967ce57dfc1SAl Viro 19681da177e4SLinus Torvalds /* 1969bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1970bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1971bfcfaa77SLinus Torvalds * 1972bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1973bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1974bfcfaa77SLinus Torvalds * fast. 1975bfcfaa77SLinus Torvalds * 1976bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1977bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1978bfcfaa77SLinus Torvalds * crossing operation. 1979bfcfaa77SLinus Torvalds * 1980bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1981bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1982bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1983bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1984bfcfaa77SLinus Torvalds */ 1985bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1986bfcfaa77SLinus Torvalds 1987f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1988bfcfaa77SLinus Torvalds 1989468a9428SGeorge Spelvin #ifdef HASH_MIX 1990bfcfaa77SLinus Torvalds 1991468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1992468a9428SGeorge Spelvin 1993468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 19942a18da7aSGeorge Spelvin /* 19952a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 19962a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 19972a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 19982a18da7aSGeorge Spelvin * and no temporaries. 19992a18da7aSGeorge Spelvin * 20002a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20012a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20022a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20032a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20042a18da7aSGeorge Spelvin * 20052a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20062a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20072a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20082a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20092a18da7aSGeorge Spelvin * 20102a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20112a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20122a18da7aSGeorge Spelvin * toggling any given output bit. 20132a18da7aSGeorge Spelvin * 20142a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20152a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20162a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20172a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20182a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20192a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20202a18da7aSGeorge Spelvin * Perfect: 8192 258048 20212a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20222a18da7aSGeorge Spelvin */ 20232a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20242a18da7aSGeorge Spelvin ( x ^= (a), \ 20252a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20262a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20272a18da7aSGeorge Spelvin y *= 9 ) 2028bfcfaa77SLinus Torvalds 20290fed3ac8SGeorge Spelvin /* 20302a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20312a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20322a18da7aSGeorge Spelvin * work done before the hash value is used. 20330fed3ac8SGeorge Spelvin */ 20342a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20350fed3ac8SGeorge Spelvin { 20362a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20372a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20382a18da7aSGeorge Spelvin return y >> 32; 20390fed3ac8SGeorge Spelvin } 20400fed3ac8SGeorge Spelvin 2041bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2042bfcfaa77SLinus Torvalds 20432a18da7aSGeorge Spelvin /* 20442a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 20452a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20462a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 20472a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 20482a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 20492a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 20502a18da7aSGeorge Spelvin * Perfect: 2048 31744 20512a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 20522a18da7aSGeorge Spelvin */ 20532a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20542a18da7aSGeorge Spelvin ( x ^= (a), \ 20552a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 20562a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 20572a18da7aSGeorge Spelvin y *= 9 ) 2058bfcfaa77SLinus Torvalds 20592a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20600fed3ac8SGeorge Spelvin { 20612a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 20622a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 20630fed3ac8SGeorge Spelvin } 20640fed3ac8SGeorge Spelvin 2065bfcfaa77SLinus Torvalds #endif 2066bfcfaa77SLinus Torvalds 20672a18da7aSGeorge Spelvin /* 20682a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 20692a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 20702a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 20712a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 20722a18da7aSGeorge Spelvin * finds the delimiter after the name. 20732a18da7aSGeorge Spelvin */ 20748387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2075bfcfaa77SLinus Torvalds { 20768387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2077bfcfaa77SLinus Torvalds 2078bfcfaa77SLinus Torvalds for (;;) { 2079fcfd2fbfSGeorge Spelvin if (!len) 2080fcfd2fbfSGeorge Spelvin goto done; 2081e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2082bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2083bfcfaa77SLinus Torvalds break; 20842a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2085bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2086bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2087bfcfaa77SLinus Torvalds } 20882a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2089bfcfaa77SLinus Torvalds done: 20902a18da7aSGeorge Spelvin return fold_hash(x, y); 2091bfcfaa77SLinus Torvalds } 2092bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2093bfcfaa77SLinus Torvalds 2094fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20958387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2096fcfd2fbfSGeorge Spelvin { 20978387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 20988387ff25SLinus Torvalds unsigned long adata, mask, len; 2099fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2100fcfd2fbfSGeorge Spelvin 21018387ff25SLinus Torvalds len = 0; 21028387ff25SLinus Torvalds goto inside; 21038387ff25SLinus Torvalds 2104fcfd2fbfSGeorge Spelvin do { 21052a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2106fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21078387ff25SLinus Torvalds inside: 2108fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2109fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2110fcfd2fbfSGeorge Spelvin 2111fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2112fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21132a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2114fcfd2fbfSGeorge Spelvin 21152a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2116fcfd2fbfSGeorge Spelvin } 2117fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2118fcfd2fbfSGeorge Spelvin 2119bfcfaa77SLinus Torvalds /* 2120bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2121d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2122bfcfaa77SLinus Torvalds */ 21238387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2124bfcfaa77SLinus Torvalds { 21258387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21268387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 212736126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2128bfcfaa77SLinus Torvalds 21298387ff25SLinus Torvalds len = 0; 21308387ff25SLinus Torvalds goto inside; 21318387ff25SLinus Torvalds 2132bfcfaa77SLinus Torvalds do { 21332a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2134bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21358387ff25SLinus Torvalds inside: 2136e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 213736126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 213836126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2139bfcfaa77SLinus Torvalds 214036126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 214136126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 214236126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21432a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 214436126f8fSLinus Torvalds 21452a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2146bfcfaa77SLinus Torvalds } 2147bfcfaa77SLinus Torvalds 21482a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2149bfcfaa77SLinus Torvalds 2150fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 21518387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 21520145acc2SLinus Torvalds { 21538387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 21540145acc2SLinus Torvalds while (len--) 2155fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 21560145acc2SLinus Torvalds return end_name_hash(hash); 21570145acc2SLinus Torvalds } 2158ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 21590145acc2SLinus Torvalds 2160fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21618387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2162fcfd2fbfSGeorge Spelvin { 21638387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2164fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2165fcfd2fbfSGeorge Spelvin 2166fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2167e0ab7af9SGeorge Spelvin while (c) { 2168fcfd2fbfSGeorge Spelvin len++; 2169fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2170fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2171e0ab7af9SGeorge Spelvin } 2172fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2173fcfd2fbfSGeorge Spelvin } 2174f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2175fcfd2fbfSGeorge Spelvin 21763ddcd056SLinus Torvalds /* 2177200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2178200e9ef7SLinus Torvalds * one character. 2179200e9ef7SLinus Torvalds */ 21808387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2181200e9ef7SLinus Torvalds { 21828387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2183200e9ef7SLinus Torvalds unsigned long len = 0, c; 2184200e9ef7SLinus Torvalds 2185200e9ef7SLinus Torvalds c = (unsigned char)*name; 2186200e9ef7SLinus Torvalds do { 2187200e9ef7SLinus Torvalds len++; 2188200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2189200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2190200e9ef7SLinus Torvalds } while (c && c != '/'); 2191d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2192200e9ef7SLinus Torvalds } 2193200e9ef7SLinus Torvalds 2194bfcfaa77SLinus Torvalds #endif 2195bfcfaa77SLinus Torvalds 2196200e9ef7SLinus Torvalds /* 21971da177e4SLinus Torvalds * Name resolution. 2198ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2199ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22001da177e4SLinus Torvalds * 2201ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2202ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22031da177e4SLinus Torvalds */ 22046de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22051da177e4SLinus Torvalds { 2206d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22071da177e4SLinus Torvalds int err; 22081da177e4SLinus Torvalds 2209b4c03536SAl Viro nd->last_type = LAST_ROOT; 2210c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22119b5858e9SAl Viro if (IS_ERR(name)) 22129b5858e9SAl Viro return PTR_ERR(name); 22131da177e4SLinus Torvalds while (*name=='/') 22141da177e4SLinus Torvalds name++; 22151a97d899SAl Viro if (!*name) { 22161a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22179e18f10aSAl Viro return 0; 22181a97d899SAl Viro } 22191da177e4SLinus Torvalds 22201da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22211da177e4SLinus Torvalds for(;;) { 2222549c7297SChristian Brauner struct user_namespace *mnt_userns; 222392d27016SAl Viro const char *link; 2224d6bb3e90SLinus Torvalds u64 hash_len; 2225fe479a58SAl Viro int type; 22261da177e4SLinus Torvalds 2227549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2228549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22291da177e4SLinus Torvalds if (err) 22303595e234SAl Viro return err; 22311da177e4SLinus Torvalds 22328387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22331da177e4SLinus Torvalds 2234fe479a58SAl Viro type = LAST_NORM; 2235d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2236fe479a58SAl Viro case 2: 2237200e9ef7SLinus Torvalds if (name[1] == '.') { 2238fe479a58SAl Viro type = LAST_DOTDOT; 2239bcba1e7dSAl Viro nd->state |= ND_JUMPED; 224016c2cd71SAl Viro } 2241fe479a58SAl Viro break; 2242fe479a58SAl Viro case 1: 2243fe479a58SAl Viro type = LAST_DOT; 2244fe479a58SAl Viro } 22455a202bcdSAl Viro if (likely(type == LAST_NORM)) { 22465a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2247bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 22485a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2249a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2250da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 22515a202bcdSAl Viro if (err < 0) 22523595e234SAl Viro return err; 2253d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2254d6bb3e90SLinus Torvalds name = this.name; 22555a202bcdSAl Viro } 22565a202bcdSAl Viro } 2257fe479a58SAl Viro 2258d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2259d6bb3e90SLinus Torvalds nd->last.name = name; 22605f4a6a69SAl Viro nd->last_type = type; 22615f4a6a69SAl Viro 2262d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2263d6bb3e90SLinus Torvalds if (!*name) 2264bdf6cbf1SAl Viro goto OK; 2265200e9ef7SLinus Torvalds /* 2266200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2267200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2268200e9ef7SLinus Torvalds */ 2269200e9ef7SLinus Torvalds do { 2270d6bb3e90SLinus Torvalds name++; 2271d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 22728620c238SAl Viro if (unlikely(!*name)) { 22738620c238SAl Viro OK: 2274d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2275c108837eSAl Viro if (!depth) { 2276549c7297SChristian Brauner nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); 22770f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2278c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 22798620c238SAl Viro return 0; 2280c108837eSAl Viro } 22818620c238SAl Viro /* last component of nested symlink */ 2282d8d4611aSAl Viro name = nd->stack[--depth].name; 22838c4efe22SAl Viro link = walk_component(nd, 0); 22841c4ff1a8SAl Viro } else { 22851c4ff1a8SAl Viro /* not the last component */ 22868c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 22878620c238SAl Viro } 228892d27016SAl Viro if (unlikely(link)) { 228992d27016SAl Viro if (IS_ERR(link)) 229092d27016SAl Viro return PTR_ERR(link); 229192d27016SAl Viro /* a symlink to follow */ 2292d8d4611aSAl Viro nd->stack[depth++].name = name; 229392d27016SAl Viro name = link; 22949e18f10aSAl Viro continue; 229548c8b0c5SAl Viro } 229697242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 229797242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2298e36cffedSJens Axboe if (!try_to_unlazy(nd)) 229997242f99SAl Viro return -ECHILD; 230097242f99SAl Viro } 23013595e234SAl Viro return -ENOTDIR; 23025f4a6a69SAl Viro } 2303ce57dfc1SAl Viro } 230497242f99SAl Viro } 23051da177e4SLinus Torvalds 2306edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2307c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 230831e6b01fSNick Piggin { 2309740a1678SAleksa Sarai int error; 2310c8a53ee5SAl Viro const char *s = nd->name->name; 231131e6b01fSNick Piggin 23126c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23136c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23146c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23156c6ec2b0SJens Axboe 2316c0eb027eSLinus Torvalds if (!*s) 2317c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2318edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2319edc2b1daSAl Viro rcu_read_lock(); 2320c0eb027eSLinus Torvalds 2321bcba1e7dSAl Viro nd->flags = flags; 2322bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2323ab87f9a5SAleksa Sarai 2324ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2325ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2326ab87f9a5SAleksa Sarai smp_rmb(); 2327ab87f9a5SAleksa Sarai 2328bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2329b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2330b18825a7SDavid Howells struct inode *inode = root->d_inode; 233193893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2332368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23335b6ca027SAl Viro nd->path = nd->root; 23345b6ca027SAl Viro nd->inode = inode; 23355b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2336ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23378f47a016SAl Viro nd->root_seq = nd->seq; 23385b6ca027SAl Viro } else { 23395b6ca027SAl Viro path_get(&nd->path); 23405b6ca027SAl Viro } 2341368ee9baSAl Viro return s; 23425b6ca027SAl Viro } 23435b6ca027SAl Viro 234431e6b01fSNick Piggin nd->root.mnt = NULL; 234531e6b01fSNick Piggin 23468db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 23478db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2348740a1678SAleksa Sarai error = nd_jump_root(nd); 2349740a1678SAleksa Sarai if (unlikely(error)) 2350740a1678SAleksa Sarai return ERR_PTR(error); 2351ef55d917SAl Viro return s; 23528db52c7eSAleksa Sarai } 23538db52c7eSAleksa Sarai 23548db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 23558db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2356e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 235731e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2358c28cc364SNick Piggin unsigned seq; 235931e6b01fSNick Piggin 2360c28cc364SNick Piggin do { 2361c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 236231e6b01fSNick Piggin nd->path = fs->pwd; 2363ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2364c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2365c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2366e41f7d4eSAl Viro } else { 2367e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2368ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2369e41f7d4eSAl Viro } 237031e6b01fSNick Piggin } else { 2371582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2372c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 237331e6b01fSNick Piggin struct dentry *dentry; 237431e6b01fSNick Piggin 23752903ff01SAl Viro if (!f.file) 2376368ee9baSAl Viro return ERR_PTR(-EBADF); 237731e6b01fSNick Piggin 23782903ff01SAl Viro dentry = f.file->f_path.dentry; 237931e6b01fSNick Piggin 2380edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 23812903ff01SAl Viro fdput(f); 2382368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2383f52e0c11SAl Viro } 23842903ff01SAl Viro 23852903ff01SAl Viro nd->path = f.file->f_path; 2386e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 238734a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 238834a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23895590ff0dSUlrich Drepper } else { 23902903ff01SAl Viro path_get(&nd->path); 239134a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 23921da177e4SLinus Torvalds } 239334a26b99SAl Viro fdput(f); 2394e41f7d4eSAl Viro } 23958db52c7eSAleksa Sarai 2396adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2397adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2398adb21d2bSAleksa Sarai nd->root = nd->path; 2399adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2400adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2401adb21d2bSAleksa Sarai } else { 2402adb21d2bSAleksa Sarai path_get(&nd->root); 2403bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2404adb21d2bSAleksa Sarai } 2405adb21d2bSAleksa Sarai } 2406adb21d2bSAleksa Sarai return s; 24079b4a9b14SAl Viro } 24089b4a9b14SAl Viro 24091ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 241095fa25d9SAl Viro { 2411bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2412bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2413bd92d7feSAl Viro 2414c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2415bd92d7feSAl Viro } 2416bd92d7feSAl Viro 24174f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24184f757f3cSAl Viro { 2419c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2420db3c9adeSAl Viro dget(nd->path.dentry); 2421b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2422b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 24234f757f3cSAl Viro } 24244f757f3cSAl Viro 24259b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2426c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24279b4a9b14SAl Viro { 2428c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2429bd92d7feSAl Viro int err; 243031e6b01fSNick Piggin 24319b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24324f757f3cSAl Viro err = handle_lookup_down(nd); 24335f336e72SAl Viro if (unlikely(err < 0)) 24345f336e72SAl Viro s = ERR_PTR(err); 24354f757f3cSAl Viro } 24364f757f3cSAl Viro 24371ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24381ccac622SAl Viro (s = lookup_last(nd)) != NULL) 24391ccac622SAl Viro ; 24404f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 24414f0ed93fSAl Viro err = handle_lookup_down(nd); 2442bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 24434f0ed93fSAl Viro } 24449f1fafeeSAl Viro if (!err) 24459f1fafeeSAl Viro err = complete_walk(nd); 2446bd92d7feSAl Viro 2447deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2448deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2449bd23a539SAl Viro err = -ENOTDIR; 2450625b6d10SAl Viro if (!err) { 2451625b6d10SAl Viro *path = nd->path; 2452625b6d10SAl Viro nd->path.mnt = NULL; 2453625b6d10SAl Viro nd->path.dentry = NULL; 2454625b6d10SAl Viro } 2455deb106c6SAl Viro terminate_walk(nd); 2456bd92d7feSAl Viro return err; 245731e6b01fSNick Piggin } 245831e6b01fSNick Piggin 245931d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 24609ad1aaa6SAl Viro struct path *path, struct path *root) 2461873f1eedSJeff Layton { 2462894bc8c4SAl Viro int retval; 24639883d185SAl Viro struct nameidata nd; 2464abc9f5beSAl Viro if (IS_ERR(name)) 2465abc9f5beSAl Viro return PTR_ERR(name); 246606422964SAl Viro set_nameidata(&nd, dfd, name, root); 2467c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2468873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2469c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2470873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2471c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2472873f1eedSJeff Layton 2473873f1eedSJeff Layton if (likely(!retval)) 2474161aff1dSAl Viro audit_inode(name, path->dentry, 2475161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 24769883d185SAl Viro restore_nameidata(); 2477e4bd1c1aSAl Viro putname(name); 2478873f1eedSJeff Layton return retval; 2479873f1eedSJeff Layton } 2480873f1eedSJeff Layton 24818bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2482c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2483391172c4SAl Viro struct path *parent) 24848bcb77faSAl Viro { 2485c8a53ee5SAl Viro const char *s = path_init(nd, flags); 24869b5858e9SAl Viro int err = link_path_walk(s, nd); 24878bcb77faSAl Viro if (!err) 24888bcb77faSAl Viro err = complete_walk(nd); 2489391172c4SAl Viro if (!err) { 2490391172c4SAl Viro *parent = nd->path; 2491391172c4SAl Viro nd->path.mnt = NULL; 2492391172c4SAl Viro nd->path.dentry = NULL; 2493391172c4SAl Viro } 2494deb106c6SAl Viro terminate_walk(nd); 24958bcb77faSAl Viro return err; 24968bcb77faSAl Viro } 24978bcb77faSAl Viro 24985c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2499391172c4SAl Viro unsigned int flags, struct path *parent, 2500391172c4SAl Viro struct qstr *last, int *type) 25018bcb77faSAl Viro { 25028bcb77faSAl Viro int retval; 25039883d185SAl Viro struct nameidata nd; 25048bcb77faSAl Viro 25055c31b6ceSAl Viro if (IS_ERR(name)) 25065c31b6ceSAl Viro return name; 250706422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2508c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25098bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2510c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25118bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2512c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2513391172c4SAl Viro if (likely(!retval)) { 2514391172c4SAl Viro *last = nd.last; 2515391172c4SAl Viro *type = nd.last_type; 2516c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 25175c31b6ceSAl Viro } else { 25185c31b6ceSAl Viro putname(name); 25195c31b6ceSAl Viro name = ERR_PTR(retval); 2520391172c4SAl Viro } 25219883d185SAl Viro restore_nameidata(); 25225c31b6ceSAl Viro return name; 25238bcb77faSAl Viro } 25248bcb77faSAl Viro 252579714f72SAl Viro /* does lookup, returns the object with parent locked */ 252679714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 25275590ff0dSUlrich Drepper { 25285c31b6ceSAl Viro struct filename *filename; 25295c31b6ceSAl Viro struct dentry *d; 2530391172c4SAl Viro struct qstr last; 2531391172c4SAl Viro int type; 253251689104SPaul Moore 25335c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 25345c31b6ceSAl Viro &last, &type); 253551689104SPaul Moore if (IS_ERR(filename)) 253651689104SPaul Moore return ERR_CAST(filename); 25375c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2538391172c4SAl Viro path_put(path); 25395c31b6ceSAl Viro putname(filename); 25405c31b6ceSAl Viro return ERR_PTR(-EINVAL); 254179714f72SAl Viro } 25425955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2543391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 254479714f72SAl Viro if (IS_ERR(d)) { 25455955102cSAl Viro inode_unlock(path->dentry->d_inode); 2546391172c4SAl Viro path_put(path); 254779714f72SAl Viro } 254851689104SPaul Moore putname(filename); 254979714f72SAl Viro return d; 25505590ff0dSUlrich Drepper } 25515590ff0dSUlrich Drepper 2552d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2553d1811465SAl Viro { 2554abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2555abc9f5beSAl Viro flags, path, NULL); 2556d1811465SAl Viro } 25574d359507SAl Viro EXPORT_SYMBOL(kern_path); 2558d1811465SAl Viro 255916f18200SJosef 'Jeff' Sipek /** 256016f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 256116f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 256216f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 256316f18200SJosef 'Jeff' Sipek * @name: pointer to file name 256416f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2565e0a01249SAl Viro * @path: pointer to struct path to fill 256616f18200SJosef 'Jeff' Sipek */ 256716f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 256816f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2569e0a01249SAl Viro struct path *path) 257016f18200SJosef 'Jeff' Sipek { 25719ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 25729ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2573abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2574abc9f5beSAl Viro flags , path, &root); 257516f18200SJosef 'Jeff' Sipek } 25764d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 257716f18200SJosef 'Jeff' Sipek 25783c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 25793c95f0dcSAl Viro int len, struct qstr *this) 25803c95f0dcSAl Viro { 25813c95f0dcSAl Viro this->name = name; 25823c95f0dcSAl Viro this->len = len; 25833c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 25843c95f0dcSAl Viro if (!len) 25853c95f0dcSAl Viro return -EACCES; 25863c95f0dcSAl Viro 25873c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 25883c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 25893c95f0dcSAl Viro return -EACCES; 25903c95f0dcSAl Viro } 25913c95f0dcSAl Viro 25923c95f0dcSAl Viro while (len--) { 25933c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 25943c95f0dcSAl Viro if (c == '/' || c == '\0') 25953c95f0dcSAl Viro return -EACCES; 25963c95f0dcSAl Viro } 25973c95f0dcSAl Viro /* 25983c95f0dcSAl Viro * See if the low-level filesystem might want 25993c95f0dcSAl Viro * to use its own hash.. 26003c95f0dcSAl Viro */ 26013c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26023c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26033c95f0dcSAl Viro if (err < 0) 26043c95f0dcSAl Viro return err; 26053c95f0dcSAl Viro } 26063c95f0dcSAl Viro 260747291baaSChristian Brauner return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC); 26083c95f0dcSAl Viro } 26093c95f0dcSAl Viro 2610eead1911SChristoph Hellwig /** 26110da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26120da0b7fdSDavid Howells * @name: pathname component to lookup 26130da0b7fdSDavid Howells * @base: base directory to lookup from 26140da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26150da0b7fdSDavid Howells * 26160da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26170da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26180da0b7fdSDavid Howells * 26190da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26200da0b7fdSDavid Howells * not be called by generic code. 26210da0b7fdSDavid Howells * 26220da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26230da0b7fdSDavid Howells */ 26240da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26250da0b7fdSDavid Howells { 26260da0b7fdSDavid Howells struct qstr this; 26270da0b7fdSDavid Howells int err; 26280da0b7fdSDavid Howells 26290da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26300da0b7fdSDavid Howells 26310da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 26320da0b7fdSDavid Howells if (err) 26330da0b7fdSDavid Howells return ERR_PTR(err); 26340da0b7fdSDavid Howells 26350da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 26360da0b7fdSDavid Howells } 26370da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 26380da0b7fdSDavid Howells 26390da0b7fdSDavid Howells /** 2640a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2641eead1911SChristoph Hellwig * @name: pathname component to lookup 2642eead1911SChristoph Hellwig * @base: base directory to lookup from 2643eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2644eead1911SChristoph Hellwig * 2645a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 26469e7543e9SAl Viro * not be called by generic code. 2647bbddca8eSNeilBrown * 2648bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2649eead1911SChristoph Hellwig */ 2650057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2651057f6c01SJames Morris { 26528613a209SAl Viro struct dentry *dentry; 2653057f6c01SJames Morris struct qstr this; 2654cda309deSMiklos Szeredi int err; 2655057f6c01SJames Morris 26565955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26572f9092e1SDavid Woodhouse 26583c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2659cda309deSMiklos Szeredi if (err) 2660cda309deSMiklos Szeredi return ERR_PTR(err); 2661cda309deSMiklos Szeredi 26628613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 26638613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2664057f6c01SJames Morris } 26654d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2666057f6c01SJames Morris 2667bbddca8eSNeilBrown /** 2668bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2669bbddca8eSNeilBrown * @name: pathname component to lookup 2670bbddca8eSNeilBrown * @base: base directory to lookup from 2671bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2672bbddca8eSNeilBrown * 2673bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2674bbddca8eSNeilBrown * not be called by generic code. 2675bbddca8eSNeilBrown * 2676bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2677bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2678bbddca8eSNeilBrown */ 2679bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2680bbddca8eSNeilBrown struct dentry *base, int len) 2681bbddca8eSNeilBrown { 2682bbddca8eSNeilBrown struct qstr this; 2683bbddca8eSNeilBrown int err; 268420d00ee8SLinus Torvalds struct dentry *ret; 2685bbddca8eSNeilBrown 26863c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2687bbddca8eSNeilBrown if (err) 2688bbddca8eSNeilBrown return ERR_PTR(err); 2689bbddca8eSNeilBrown 269020d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 269120d00ee8SLinus Torvalds if (!ret) 269220d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 269320d00ee8SLinus Torvalds return ret; 2694bbddca8eSNeilBrown } 2695bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2696bbddca8eSNeilBrown 26976c2d4798SAl Viro /* 26986c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 26996c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 27006c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 27016c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 27026c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 27036c2d4798SAl Viro * this one avoids such problems. 27046c2d4798SAl Viro */ 27056c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 27066c2d4798SAl Viro struct dentry *base, int len) 27076c2d4798SAl Viro { 27086c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 27092fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 27106c2d4798SAl Viro dput(ret); 27116c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 27126c2d4798SAl Viro } 27136c2d4798SAl Viro return ret; 27146c2d4798SAl Viro } 27156c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 27166c2d4798SAl Viro 2717eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2718eedf265aSEric W. Biederman int path_pts(struct path *path) 2719eedf265aSEric W. Biederman { 2720eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2721eedf265aSEric W. Biederman * the input path. 2722eedf265aSEric W. Biederman */ 2723a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2724a6a7eb76SAl Viro struct dentry *child; 272519f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2726eedf265aSEric W. Biederman 2727a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2728a6a7eb76SAl Viro dput(parent); 272963b27720SAl Viro return -ENOENT; 2730a6a7eb76SAl Viro } 273163b27720SAl Viro dput(path->dentry); 273263b27720SAl Viro path->dentry = parent; 2733eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2734eedf265aSEric W. Biederman if (!child) 2735eedf265aSEric W. Biederman return -ENOENT; 2736eedf265aSEric W. Biederman 2737eedf265aSEric W. Biederman path->dentry = child; 2738eedf265aSEric W. Biederman dput(parent); 273919f6028aSAl Viro follow_down(path); 2740eedf265aSEric W. Biederman return 0; 2741eedf265aSEric W. Biederman } 2742eedf265aSEric W. Biederman #endif 2743eedf265aSEric W. Biederman 27441fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 27451fa1e7f6SAndy Whitcroft struct path *path, int *empty) 27461da177e4SLinus Torvalds { 2747abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2748abc9f5beSAl Viro flags, path, NULL); 27491da177e4SLinus Torvalds } 2750b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 27511fa1e7f6SAndy Whitcroft 2752ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2753ba73d987SChristian Brauner struct inode *inode) 27541da177e4SLinus Torvalds { 27558e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2756da9592edSDavid Howells 2757ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) 27581da177e4SLinus Torvalds return 0; 2759ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) 27601da177e4SLinus Torvalds return 0; 2761ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 27621da177e4SLinus Torvalds } 2763cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27641da177e4SLinus Torvalds 27651da177e4SLinus Torvalds /* 27661da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27671da177e4SLinus Torvalds * whether the type of victim is right. 27681da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27691da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27701da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27711da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27721da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27731da177e4SLinus Torvalds * a. be owner of dir, or 27741da177e4SLinus Torvalds * b. be owner of victim, or 27751da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27761da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27771da177e4SLinus Torvalds * links pointing to it. 27780bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27790bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27800bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27810bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27820bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27831da177e4SLinus Torvalds * nfs_async_unlink(). 27841da177e4SLinus Torvalds */ 2785ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2786ba73d987SChristian Brauner struct dentry *victim, bool isdir) 27871da177e4SLinus Torvalds { 278863afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27891da177e4SLinus Torvalds int error; 27901da177e4SLinus Torvalds 2791b18825a7SDavid Howells if (d_is_negative(victim)) 27921da177e4SLinus Torvalds return -ENOENT; 2793b18825a7SDavid Howells BUG_ON(!inode); 27941da177e4SLinus Torvalds 27951da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2796593d1ce8SEric W. Biederman 2797593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2798ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 2799ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 2800593d1ce8SEric W. Biederman return -EOVERFLOW; 2801593d1ce8SEric W. Biederman 28024fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 28031da177e4SLinus Torvalds 2804ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 28051da177e4SLinus Torvalds if (error) 28061da177e4SLinus Torvalds return error; 28071da177e4SLinus Torvalds if (IS_APPEND(dir)) 28081da177e4SLinus Torvalds return -EPERM; 2809b18825a7SDavid Howells 2810ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2811ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2812ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 28131da177e4SLinus Torvalds return -EPERM; 28141da177e4SLinus Torvalds if (isdir) { 281544b1d530SMiklos Szeredi if (!d_is_dir(victim)) 28161da177e4SLinus Torvalds return -ENOTDIR; 28171da177e4SLinus Torvalds if (IS_ROOT(victim)) 28181da177e4SLinus Torvalds return -EBUSY; 281944b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28201da177e4SLinus Torvalds return -EISDIR; 28211da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28221da177e4SLinus Torvalds return -ENOENT; 28231da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28241da177e4SLinus Torvalds return -EBUSY; 28251da177e4SLinus Torvalds return 0; 28261da177e4SLinus Torvalds } 28271da177e4SLinus Torvalds 28281da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28291da177e4SLinus Torvalds * dir. 28301da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28311da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28321da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2833036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2834036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2835036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28361da177e4SLinus Torvalds */ 2837ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2838ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 28391da177e4SLinus Torvalds { 284014e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28411da177e4SLinus Torvalds if (child->d_inode) 28421da177e4SLinus Torvalds return -EEXIST; 28431da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28441da177e4SLinus Torvalds return -ENOENT; 28458e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns)) 2846036d5236SEric W. Biederman return -EOVERFLOW; 28478e538913SChristian Brauner 2848ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 28491da177e4SLinus Torvalds } 28501da177e4SLinus Torvalds 28511da177e4SLinus Torvalds /* 28521da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28531da177e4SLinus Torvalds */ 28541da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28551da177e4SLinus Torvalds { 28561da177e4SLinus Torvalds struct dentry *p; 28571da177e4SLinus Torvalds 28581da177e4SLinus Torvalds if (p1 == p2) { 28595955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28601da177e4SLinus Torvalds return NULL; 28611da177e4SLinus Torvalds } 28621da177e4SLinus Torvalds 2863fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28641da177e4SLinus Torvalds 2865e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2866e2761a11SOGAWA Hirofumi if (p) { 28675955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28685955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28691da177e4SLinus Torvalds return p; 28701da177e4SLinus Torvalds } 28711da177e4SLinus Torvalds 2872e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2873e2761a11SOGAWA Hirofumi if (p) { 28745955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28755955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28761da177e4SLinus Torvalds return p; 28771da177e4SLinus Torvalds } 28781da177e4SLinus Torvalds 28795955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28805955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28811da177e4SLinus Torvalds return NULL; 28821da177e4SLinus Torvalds } 28834d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28841da177e4SLinus Torvalds 28851da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28861da177e4SLinus Torvalds { 28875955102cSAl Viro inode_unlock(p1->d_inode); 28881da177e4SLinus Torvalds if (p1 != p2) { 28895955102cSAl Viro inode_unlock(p2->d_inode); 2890fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28911da177e4SLinus Torvalds } 28921da177e4SLinus Torvalds } 28934d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28941da177e4SLinus Torvalds 28956521f891SChristian Brauner /** 28966521f891SChristian Brauner * vfs_create - create new file 28976521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 28986521f891SChristian Brauner * @dir: inode of @dentry 28996521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 29006521f891SChristian Brauner * @mode: mode of the new file 29016521f891SChristian Brauner * @want_excl: whether the file must not yet exist 29026521f891SChristian Brauner * 29036521f891SChristian Brauner * Create a new file. 29046521f891SChristian Brauner * 29056521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 29066521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 29076521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 29086521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 29096521f891SChristian Brauner * raw inode simply passs init_user_ns. 29106521f891SChristian Brauner */ 29116521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 29126521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 29131da177e4SLinus Torvalds { 29146521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 29151da177e4SLinus Torvalds if (error) 29161da177e4SLinus Torvalds return error; 29171da177e4SLinus Torvalds 2918acfa4380SAl Viro if (!dir->i_op->create) 29191da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 29201da177e4SLinus Torvalds mode &= S_IALLUGO; 29211da177e4SLinus Torvalds mode |= S_IFREG; 29221da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 29231da177e4SLinus Torvalds if (error) 29241da177e4SLinus Torvalds return error; 2925549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 2926a74574aaSStephen Smalley if (!error) 2927f38aa942SAmy Griffis fsnotify_create(dir, dentry); 29281da177e4SLinus Torvalds return error; 29291da177e4SLinus Torvalds } 29304d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29311da177e4SLinus Torvalds 29328e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29338e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29348e6c848eSAl Viro void *arg) 29358e6c848eSAl Viro { 29368e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 2937ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 29388e6c848eSAl Viro if (error) 29398e6c848eSAl Viro return error; 29408e6c848eSAl Viro 29418e6c848eSAl Viro mode &= S_IALLUGO; 29428e6c848eSAl Viro mode |= S_IFREG; 29438e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29448e6c848eSAl Viro if (error) 29458e6c848eSAl Viro return error; 29468e6c848eSAl Viro error = f(dentry, mode, arg); 29478e6c848eSAl Viro if (!error) 29488e6c848eSAl Viro fsnotify_create(dir, dentry); 29498e6c848eSAl Viro return error; 29508e6c848eSAl Viro } 29518e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29528e6c848eSAl Viro 2953a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2954a2982cc9SEric W. Biederman { 2955a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2956a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2957a2982cc9SEric W. Biederman } 2958a2982cc9SEric W. Biederman 2959ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 2960ba73d987SChristian Brauner int acc_mode, int flag) 29611da177e4SLinus Torvalds { 29623fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29631da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29641da177e4SLinus Torvalds int error; 29651da177e4SLinus Torvalds 29661da177e4SLinus Torvalds if (!inode) 29671da177e4SLinus Torvalds return -ENOENT; 29681da177e4SLinus Torvalds 2969c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2970c8fe8f30SChristoph Hellwig case S_IFLNK: 29711da177e4SLinus Torvalds return -ELOOP; 2972c8fe8f30SChristoph Hellwig case S_IFDIR: 2973fc4177beSKees Cook if (acc_mode & MAY_WRITE) 29741da177e4SLinus Torvalds return -EISDIR; 2975fc4177beSKees Cook if (acc_mode & MAY_EXEC) 2976fc4177beSKees Cook return -EACCES; 2977c8fe8f30SChristoph Hellwig break; 2978c8fe8f30SChristoph Hellwig case S_IFBLK: 2979c8fe8f30SChristoph Hellwig case S_IFCHR: 2980a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29811da177e4SLinus Torvalds return -EACCES; 2982633fb6acSKees Cook fallthrough; 2983c8fe8f30SChristoph Hellwig case S_IFIFO: 2984c8fe8f30SChristoph Hellwig case S_IFSOCK: 2985633fb6acSKees Cook if (acc_mode & MAY_EXEC) 2986633fb6acSKees Cook return -EACCES; 29871da177e4SLinus Torvalds flag &= ~O_TRUNC; 2988c8fe8f30SChristoph Hellwig break; 29890fd338b2SKees Cook case S_IFREG: 29900fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 29910fd338b2SKees Cook return -EACCES; 29920fd338b2SKees Cook break; 29934a3fd211SDave Hansen } 2994b41572e9SDave Hansen 2995ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 2996b41572e9SDave Hansen if (error) 2997b41572e9SDave Hansen return error; 29986146f0d5SMimi Zohar 29991da177e4SLinus Torvalds /* 30001da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 30011da177e4SLinus Torvalds */ 30021da177e4SLinus Torvalds if (IS_APPEND(inode)) { 30038737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 30047715b521SAl Viro return -EPERM; 30051da177e4SLinus Torvalds if (flag & O_TRUNC) 30067715b521SAl Viro return -EPERM; 30071da177e4SLinus Torvalds } 30081da177e4SLinus Torvalds 30091da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3010ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 30117715b521SAl Viro return -EPERM; 30121da177e4SLinus Torvalds 3013f3c7691eSJ. Bruce Fields return 0; 30147715b521SAl Viro } 30157715b521SAl Viro 3016549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 30177715b521SAl Viro { 3018f0bb5aafSAl Viro const struct path *path = &filp->f_path; 30197715b521SAl Viro struct inode *inode = path->dentry->d_inode; 30207715b521SAl Viro int error = get_write_access(inode); 30211da177e4SLinus Torvalds if (error) 30227715b521SAl Viro return error; 30231da177e4SLinus Torvalds /* 30241da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 30251da177e4SLinus Torvalds */ 3026d7a06983SJeff Layton error = locks_verify_locked(filp); 3027be6d3e56SKentaro Takeda if (!error) 3028ea0d3ab2STetsuo Handa error = security_path_truncate(path); 30291da177e4SLinus Torvalds if (!error) { 3030549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3031d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3032e1181ee6SJeff Layton filp); 30331da177e4SLinus Torvalds } 30341da177e4SLinus Torvalds put_write_access(inode); 3035acd0c935SMimi Zohar return error; 30361da177e4SLinus Torvalds } 30371da177e4SLinus Torvalds 3038d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3039d57999e1SDave Hansen { 30408a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30418a5e929dSAl Viro flag--; 3042d57999e1SDave Hansen return flag; 3043d57999e1SDave Hansen } 3044d57999e1SDave Hansen 3045ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3046ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3047ba73d987SChristian Brauner umode_t mode) 3048d18e9008SMiklos Szeredi { 3049d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3050d18e9008SMiklos Szeredi if (error) 3051d18e9008SMiklos Szeredi return error; 3052d18e9008SMiklos Szeredi 30538e538913SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns)) 30541328c727SSeth Forshee return -EOVERFLOW; 30551328c727SSeth Forshee 3056ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 305747291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3058d18e9008SMiklos Szeredi if (error) 3059d18e9008SMiklos Szeredi return error; 3060d18e9008SMiklos Szeredi 3061d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3062d18e9008SMiklos Szeredi } 3063d18e9008SMiklos Szeredi 30641acf0af9SDavid Howells /* 30651acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30661acf0af9SDavid Howells * dentry. 30671acf0af9SDavid Howells * 30681acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30691acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30701acf0af9SDavid Howells * 307100a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 307200a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 307300a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 30741acf0af9SDavid Howells * 30751acf0af9SDavid Howells * Returns an error code otherwise. 30761acf0af9SDavid Howells */ 3077239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3078239eb983SAl Viro struct file *file, 30793ec2eef1SAl Viro int open_flag, umode_t mode) 3080d18e9008SMiklos Szeredi { 3081d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3082d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3083d18e9008SMiklos Szeredi int error; 3084d18e9008SMiklos Szeredi 3085d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3086d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3087d18e9008SMiklos Szeredi 308830d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 308930d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30900fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 309144907d79SAl Viro open_to_namei_flags(open_flag), mode); 30926fbd0714SAl Viro d_lookup_done(dentry); 3093384f26e2SAl Viro if (!error) { 309464e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 30956fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 30966fb968cdSAl Viro dput(dentry); 30976fb968cdSAl Viro dentry = dget(file->f_path.dentry); 30986fb968cdSAl Viro } 309964e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 31002675a4ebSAl Viro error = -EIO; 3101384f26e2SAl Viro } else { 310230d90494SAl Viro if (file->f_path.dentry) { 3103d18e9008SMiklos Szeredi dput(dentry); 3104d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 310510c64ceaSAl Viro } 3106239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3107a01e718fSAl Viro error = -ENOENT; 3108d18e9008SMiklos Szeredi } 3109d18e9008SMiklos Szeredi } 3110239eb983SAl Viro if (error) { 3111d18e9008SMiklos Szeredi dput(dentry); 3112239eb983SAl Viro dentry = ERR_PTR(error); 3113239eb983SAl Viro } 3114239eb983SAl Viro return dentry; 3115d18e9008SMiklos Szeredi } 3116d18e9008SMiklos Szeredi 311731e6b01fSNick Piggin /* 31181acf0af9SDavid Howells * Look up and maybe create and open the last component. 3119d58ffd35SMiklos Szeredi * 312000a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3121d58ffd35SMiklos Szeredi * 312200a07c15SAl Viro * Returns 0 on success, that is, if 312300a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 312400a07c15SAl Viro * the file was not completely opened at this time, though lookups and 312500a07c15SAl Viro * creations were performed. 312600a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 312700a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 312800a07c15SAl Viro * hadn't been specified. 31291acf0af9SDavid Howells * 313000a07c15SAl Viro * An error code is returned on failure. 3131d58ffd35SMiklos Szeredi */ 3132da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3133d58ffd35SMiklos Szeredi const struct open_flags *op, 31343ec2eef1SAl Viro bool got_write) 3135d58ffd35SMiklos Szeredi { 3136549c7297SChristian Brauner struct user_namespace *mnt_userns; 3137d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 313854ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31391643b43fSAl Viro int open_flag = op->open_flag; 3140d58ffd35SMiklos Szeredi struct dentry *dentry; 31411643b43fSAl Viro int error, create_error = 0; 31421643b43fSAl Viro umode_t mode = op->mode; 31436fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3144d58ffd35SMiklos Szeredi 3145ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3146da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3147d58ffd35SMiklos Szeredi 314873a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 31496fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31506fbd0714SAl Viro for (;;) { 31516fbd0714SAl Viro if (!dentry) { 31526fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3153d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3154da5ebf5aSAl Viro return dentry; 31556fbd0714SAl Viro } 31566fbd0714SAl Viro if (d_in_lookup(dentry)) 31576fbd0714SAl Viro break; 3158d58ffd35SMiklos Szeredi 31596fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31606fbd0714SAl Viro if (likely(error > 0)) 31616fbd0714SAl Viro break; 31626fbd0714SAl Viro if (error) 31636fbd0714SAl Viro goto out_dput; 31646fbd0714SAl Viro d_invalidate(dentry); 31656fbd0714SAl Viro dput(dentry); 31666fbd0714SAl Viro dentry = NULL; 31676fbd0714SAl Viro } 31686fbd0714SAl Viro if (dentry->d_inode) { 3169d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3170da5ebf5aSAl Viro return dentry; 31716c51e513SAl Viro } 3172d18e9008SMiklos Szeredi 31731643b43fSAl Viro /* 31741643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31751643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31761643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31771643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31781643b43fSAl Viro * 31791643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31801643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31811643b43fSAl Viro */ 318299a4a90cSAl Viro if (unlikely(!got_write)) 318399a4a90cSAl Viro open_flag &= ~O_TRUNC; 3184549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 31851643b43fSAl Viro if (open_flag & O_CREAT) { 318699a4a90cSAl Viro if (open_flag & O_EXCL) 318799a4a90cSAl Viro open_flag &= ~O_TRUNC; 31881643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31891643b43fSAl Viro mode &= ~current_umask(); 319099a4a90cSAl Viro if (likely(got_write)) 3191549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3192ba73d987SChristian Brauner dentry, mode); 319399a4a90cSAl Viro else 319499a4a90cSAl Viro create_error = -EROFS; 319599a4a90cSAl Viro } 319699a4a90cSAl Viro if (create_error) 31971643b43fSAl Viro open_flag &= ~O_CREAT; 31981643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3199d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3200da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3201da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3202da5ebf5aSAl Viro return dentry; 3203239eb983SAl Viro } 320454ef4872SMiklos Szeredi 32056fbd0714SAl Viro if (d_in_lookup(dentry)) { 320612fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 320712fa5e24SAl Viro nd->flags); 32086fbd0714SAl Viro d_lookup_done(dentry); 320912fa5e24SAl Viro if (unlikely(res)) { 321012fa5e24SAl Viro if (IS_ERR(res)) { 321112fa5e24SAl Viro error = PTR_ERR(res); 321212fa5e24SAl Viro goto out_dput; 321312fa5e24SAl Viro } 321412fa5e24SAl Viro dput(dentry); 321512fa5e24SAl Viro dentry = res; 321612fa5e24SAl Viro } 321754ef4872SMiklos Szeredi } 321854ef4872SMiklos Szeredi 3219d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32201643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 322173a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3222ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3223ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3224ce8644fcSAl Viro error = -EACCES; 3225d58ffd35SMiklos Szeredi goto out_dput; 322664894cf8SAl Viro } 3227549c7297SChristian Brauner 3228549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3229549c7297SChristian Brauner mode, open_flag & O_EXCL); 3230d58ffd35SMiklos Szeredi if (error) 3231d58ffd35SMiklos Szeredi goto out_dput; 3232d58ffd35SMiklos Szeredi } 32331643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32341643b43fSAl Viro error = create_error; 3235d58ffd35SMiklos Szeredi goto out_dput; 3236d58ffd35SMiklos Szeredi } 3237da5ebf5aSAl Viro return dentry; 3238d58ffd35SMiklos Szeredi 3239d58ffd35SMiklos Szeredi out_dput: 3240d58ffd35SMiklos Szeredi dput(dentry); 3241da5ebf5aSAl Viro return ERR_PTR(error); 3242d58ffd35SMiklos Szeredi } 3243d58ffd35SMiklos Szeredi 3244c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 32453ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3246fb1cc555SAl Viro { 3247a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3248ca344a89SAl Viro int open_flag = op->open_flag; 324964894cf8SAl Viro bool got_write = false; 3250254cf582SAl Viro unsigned seq; 3251a1eb3315SMiklos Szeredi struct inode *inode; 3252da5ebf5aSAl Viro struct dentry *dentry; 3253b0417d2cSAl Viro const char *res; 3254fb1cc555SAl Viro 3255c3e380b0SAl Viro nd->flags |= op->intent; 3256c3e380b0SAl Viro 3257bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 325856676ec3SAl Viro if (nd->depth) 325956676ec3SAl Viro put_link(nd); 3260ff326a32SAl Viro return handle_dots(nd, nd->last_type); 32611f36f774SAl Viro } 3262a2c36b45SAl Viro 3263ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3264fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3265fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3266fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 326720e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 326820e34357SAl Viro if (IS_ERR(dentry)) 32691ccac622SAl Viro return ERR_CAST(dentry); 327020e34357SAl Viro if (likely(dentry)) 327171574865SMiklos Szeredi goto finish_lookup; 327271574865SMiklos Szeredi 32736583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3274b6183df7SMiklos Szeredi } else { 3275fe2d35ffSAl Viro /* create side of things */ 327672287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3277e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3278e36cffedSJens Axboe return ERR_PTR(-ECHILD); 327972287417SAl Viro } 3280c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 32811f36f774SAl Viro /* trailing slashes? */ 3282deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 32831ccac622SAl Viro return ERR_PTR(-EISDIR); 3284b6183df7SMiklos Szeredi } 32851f36f774SAl Viro 32869cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3287e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 328864894cf8SAl Viro /* 328964894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 329064894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 329164894cf8SAl Viro * dropping this one anyway. 329264894cf8SAl Viro */ 329364894cf8SAl Viro } 32949cf843e3SAl Viro if (open_flag & O_CREAT) 32955955102cSAl Viro inode_lock(dir->d_inode); 32969cf843e3SAl Viro else 32979cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3298da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3299f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3300f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 33019cf843e3SAl Viro if (open_flag & O_CREAT) 33025955102cSAl Viro inode_unlock(dir->d_inode); 33039cf843e3SAl Viro else 33049cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3305fb1cc555SAl Viro 3306c981a482SAl Viro if (got_write) 330759e96e65SAl Viro mnt_drop_write(nd->path.mnt); 33086c0d46c4SAl Viro 330959e96e65SAl Viro if (IS_ERR(dentry)) 331059e96e65SAl Viro return ERR_CAST(dentry); 331159e96e65SAl Viro 3312973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3313e73cabffSAl Viro dput(nd->path.dentry); 3314e73cabffSAl Viro nd->path.dentry = dentry; 3315c981a482SAl Viro return NULL; 3316fb1cc555SAl Viro } 3317fb1cc555SAl Viro 331820e34357SAl Viro finish_lookup: 331956676ec3SAl Viro if (nd->depth) 332056676ec3SAl Viro put_link(nd); 33218c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3322ff326a32SAl Viro if (unlikely(res)) 33231ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3324b0417d2cSAl Viro return res; 33251ccac622SAl Viro } 332631d1726dSAl Viro 3327c981a482SAl Viro /* 3328c981a482SAl Viro * Handle the last step of open() 3329c981a482SAl Viro */ 3330c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3331c981a482SAl Viro struct file *file, const struct open_flags *op) 3332c981a482SAl Viro { 3333549c7297SChristian Brauner struct user_namespace *mnt_userns; 3334c981a482SAl Viro int open_flag = op->open_flag; 3335c981a482SAl Viro bool do_truncate; 3336c981a482SAl Viro int acc_mode; 3337c981a482SAl Viro int error; 3338c981a482SAl Viro 3339ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3340ff326a32SAl Viro error = complete_walk(nd); 3341ff326a32SAl Viro if (error) 3342ff326a32SAl Viro return error; 3343ff326a32SAl Viro } 3344973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 334576ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3346549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 334730aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3348b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3349b94e0b32SAl Viro return -EEXIST; 335030aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3351c5971b8cSAl Viro return -EISDIR; 3352549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 335330aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 335430aba665SSalvatore Mesoraca if (unlikely(error)) 3355c5971b8cSAl Viro return error; 335630aba665SSalvatore Mesoraca } 335744b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3358c5971b8cSAl Viro return -ENOTDIR; 33596c0d46c4SAl Viro 33608795e7d4SAl Viro do_truncate = false; 33618795e7d4SAl Viro acc_mode = op->acc_mode; 33625a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 33635a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 33645a2d3eddSAl Viro open_flag &= ~O_TRUNC; 33655a2d3eddSAl Viro acc_mode = 0; 33668795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 33670f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33680f9d1a10SAl Viro if (error) 3369c5971b8cSAl Viro return error; 33708795e7d4SAl Viro do_truncate = true; 33710f9d1a10SAl Viro } 3372549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 33738795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3374ae2bb293SAl Viro error = vfs_open(&nd->path, file); 33758795e7d4SAl Viro if (!error) 33766035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 33778795e7d4SAl Viro if (!error && do_truncate) 3378549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3379c80567c8SAl Viro if (unlikely(error > 0)) { 3380c80567c8SAl Viro WARN_ON(1); 3381c80567c8SAl Viro error = -EINVAL; 3382c80567c8SAl Viro } 33838795e7d4SAl Viro if (do_truncate) 33840f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3385c5971b8cSAl Viro return error; 3386fb1cc555SAl Viro } 3387fb1cc555SAl Viro 33886521f891SChristian Brauner /** 33896521f891SChristian Brauner * vfs_tmpfile - create tmpfile 33906521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 33916521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 33926521f891SChristian Brauner * @mode: mode of the new tmpfile 33932111c3c0SRandy Dunlap * @open_flag: flags 33946521f891SChristian Brauner * 33956521f891SChristian Brauner * Create a temporary file. 33966521f891SChristian Brauner * 33976521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 33986521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 33996521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 34006521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 34016521f891SChristian Brauner * raw inode simply passs init_user_ns. 34026521f891SChristian Brauner */ 34036521f891SChristian Brauner struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, 34046521f891SChristian Brauner struct dentry *dentry, umode_t mode, int open_flag) 3405af7bd4dcSAmir Goldstein { 3406af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3407af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3408af7bd4dcSAmir Goldstein struct inode *inode; 3409af7bd4dcSAmir Goldstein int error; 3410af7bd4dcSAmir Goldstein 3411af7bd4dcSAmir Goldstein /* we want directory to be writable */ 34126521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3413af7bd4dcSAmir Goldstein if (error) 3414af7bd4dcSAmir Goldstein goto out_err; 3415af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3416af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3417af7bd4dcSAmir Goldstein goto out_err; 3418af7bd4dcSAmir Goldstein error = -ENOMEM; 3419cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3420af7bd4dcSAmir Goldstein if (unlikely(!child)) 3421af7bd4dcSAmir Goldstein goto out_err; 3422549c7297SChristian Brauner error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); 3423af7bd4dcSAmir Goldstein if (error) 3424af7bd4dcSAmir Goldstein goto out_err; 3425af7bd4dcSAmir Goldstein error = -ENOENT; 3426af7bd4dcSAmir Goldstein inode = child->d_inode; 3427af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3428af7bd4dcSAmir Goldstein goto out_err; 3429af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3430af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3431af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3432af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3433af7bd4dcSAmir Goldstein } 3434a2d2329eSChristian Brauner ima_post_create_tmpfile(mnt_userns, inode); 3435af7bd4dcSAmir Goldstein return child; 3436af7bd4dcSAmir Goldstein 3437af7bd4dcSAmir Goldstein out_err: 3438af7bd4dcSAmir Goldstein dput(child); 3439af7bd4dcSAmir Goldstein return ERR_PTR(error); 3440af7bd4dcSAmir Goldstein } 3441af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3442af7bd4dcSAmir Goldstein 3443c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 344460545d0dSAl Viro const struct open_flags *op, 34453ec2eef1SAl Viro struct file *file) 344660545d0dSAl Viro { 34476521f891SChristian Brauner struct user_namespace *mnt_userns; 3448625b6d10SAl Viro struct dentry *child; 3449625b6d10SAl Viro struct path path; 3450c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 345160545d0dSAl Viro if (unlikely(error)) 345260545d0dSAl Viro return error; 3453625b6d10SAl Viro error = mnt_want_write(path.mnt); 345460545d0dSAl Viro if (unlikely(error)) 345560545d0dSAl Viro goto out; 34566521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 34576521f891SChristian Brauner child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); 3458af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3459684e73beSHirofumi Nakagawa if (IS_ERR(child)) 346060545d0dSAl Viro goto out2; 3461625b6d10SAl Viro dput(path.dentry); 3462625b6d10SAl Viro path.dentry = child; 3463c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 346469a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3465549c7297SChristian Brauner error = may_open(mnt_userns, &path, 0, op->open_flag); 34661e8f44f1SAl Viro if (!error) 34671e8f44f1SAl Viro error = vfs_open(&path, file); 346860545d0dSAl Viro out2: 3469625b6d10SAl Viro mnt_drop_write(path.mnt); 347060545d0dSAl Viro out: 3471625b6d10SAl Viro path_put(&path); 347260545d0dSAl Viro return error; 347360545d0dSAl Viro } 347460545d0dSAl Viro 34756ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34766ac08709SAl Viro { 34776ac08709SAl Viro struct path path; 34786ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34796ac08709SAl Viro if (!error) { 34806ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3481ae2bb293SAl Viro error = vfs_open(&path, file); 34826ac08709SAl Viro path_put(&path); 34836ac08709SAl Viro } 34846ac08709SAl Viro return error; 34856ac08709SAl Viro } 34866ac08709SAl Viro 3487c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3488c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34891da177e4SLinus Torvalds { 349030d90494SAl Viro struct file *file; 349113aab428SAl Viro int error; 349231e6b01fSNick Piggin 3493ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 34941afc99beSAl Viro if (IS_ERR(file)) 34951afc99beSAl Viro return file; 349631e6b01fSNick Piggin 3497bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 34983ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 34995f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 35006ac08709SAl Viro error = do_o_path(nd, flags, file); 35015f336e72SAl Viro } else { 35025f336e72SAl Viro const char *s = path_init(nd, flags); 35033bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3504c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 35051ccac622SAl Viro ; 3506c5971b8cSAl Viro if (!error) 3507c5971b8cSAl Viro error = do_open(nd, file, op); 3508deb106c6SAl Viro terminate_walk(nd); 35095f336e72SAl Viro } 35107c1c01ecSAl Viro if (likely(!error)) { 3511aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35127c1c01ecSAl Viro return file; 35137c1c01ecSAl Viro WARN_ON(1); 35147c1c01ecSAl Viro error = -EINVAL; 3515015c3bbcSMiklos Szeredi } 35167c1c01ecSAl Viro fput(file); 35172675a4ebSAl Viro if (error == -EOPENSTALE) { 35182675a4ebSAl Viro if (flags & LOOKUP_RCU) 35192675a4ebSAl Viro error = -ECHILD; 35202675a4ebSAl Viro else 35212675a4ebSAl Viro error = -ESTALE; 35222675a4ebSAl Viro } 35237c1c01ecSAl Viro return ERR_PTR(error); 3524de459215SKirill Korotaev } 35251da177e4SLinus Torvalds 3526669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3527f9652e10SAl Viro const struct open_flags *op) 352813aab428SAl Viro { 35299883d185SAl Viro struct nameidata nd; 3530f9652e10SAl Viro int flags = op->lookup_flags; 353113aab428SAl Viro struct file *filp; 353213aab428SAl Viro 353306422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3534c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 353513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3536c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 353713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3538c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35399883d185SAl Viro restore_nameidata(); 354013aab428SAl Viro return filp; 354113aab428SAl Viro } 354213aab428SAl Viro 3543ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3544f9652e10SAl Viro const char *name, const struct open_flags *op) 354573d049a4SAl Viro { 35469883d185SAl Viro struct nameidata nd; 354773d049a4SAl Viro struct file *file; 354851689104SPaul Moore struct filename *filename; 3549bcba1e7dSAl Viro int flags = op->lookup_flags; 355073d049a4SAl Viro 3551ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 355273d049a4SAl Viro return ERR_PTR(-ELOOP); 355373d049a4SAl Viro 355451689104SPaul Moore filename = getname_kernel(name); 3555a1c83681SViresh Kumar if (IS_ERR(filename)) 355651689104SPaul Moore return ERR_CAST(filename); 355751689104SPaul Moore 355806422964SAl Viro set_nameidata(&nd, -1, filename, root); 3559c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 356073d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3561c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 356273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3563c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35649883d185SAl Viro restore_nameidata(); 356551689104SPaul Moore putname(filename); 356673d049a4SAl Viro return file; 356773d049a4SAl Viro } 356873d049a4SAl Viro 3569fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35701ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35711da177e4SLinus Torvalds { 3572c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3573391172c4SAl Viro struct qstr last; 3574391172c4SAl Viro int type; 3575c30dabfeSJan Kara int err2; 35761ac12b4bSJeff Layton int error; 35771ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 35781ac12b4bSJeff Layton 35791ac12b4bSJeff Layton /* 35801ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 35811ac12b4bSJeff Layton * other flags passed in are ignored! 35821ac12b4bSJeff Layton */ 35831ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 35841ac12b4bSJeff Layton 35855c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 35865c31b6ceSAl Viro if (IS_ERR(name)) 35875c31b6ceSAl Viro return ERR_CAST(name); 35881da177e4SLinus Torvalds 3589c663e5d8SChristoph Hellwig /* 3590c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3591c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3592c663e5d8SChristoph Hellwig */ 35935c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3594ed75e95dSAl Viro goto out; 3595c663e5d8SChristoph Hellwig 3596c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3597391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3598c663e5d8SChristoph Hellwig /* 3599c663e5d8SChristoph Hellwig * Do the final lookup. 3600c663e5d8SChristoph Hellwig */ 3601391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 36025955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3603391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36041da177e4SLinus Torvalds if (IS_ERR(dentry)) 3605a8104a9fSAl Viro goto unlock; 3606c663e5d8SChristoph Hellwig 3607a8104a9fSAl Viro error = -EEXIST; 3608b18825a7SDavid Howells if (d_is_positive(dentry)) 3609a8104a9fSAl Viro goto fail; 3610b18825a7SDavid Howells 3611c663e5d8SChristoph Hellwig /* 3612c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3613c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3614c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3615c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3616c663e5d8SChristoph Hellwig */ 3617391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3618a8104a9fSAl Viro error = -ENOENT; 3619ed75e95dSAl Viro goto fail; 3620e9baf6e5SAl Viro } 3621c30dabfeSJan Kara if (unlikely(err2)) { 3622c30dabfeSJan Kara error = err2; 3623a8104a9fSAl Viro goto fail; 3624c30dabfeSJan Kara } 3625181c37b6SAl Viro putname(name); 3626e9baf6e5SAl Viro return dentry; 36271da177e4SLinus Torvalds fail: 3628a8104a9fSAl Viro dput(dentry); 3629a8104a9fSAl Viro dentry = ERR_PTR(error); 3630a8104a9fSAl Viro unlock: 36315955102cSAl Viro inode_unlock(path->dentry->d_inode); 3632c30dabfeSJan Kara if (!err2) 3633391172c4SAl Viro mnt_drop_write(path->mnt); 3634ed75e95dSAl Viro out: 3635391172c4SAl Viro path_put(path); 3636181c37b6SAl Viro putname(name); 3637ed75e95dSAl Viro return dentry; 3638dae6ad8fSAl Viro } 3639fa14a0b8SAl Viro 3640fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3641fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3642fa14a0b8SAl Viro { 3643181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3644181c37b6SAl Viro path, lookup_flags); 3645fa14a0b8SAl Viro } 3646dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3647dae6ad8fSAl Viro 3648921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3649921a1650SAl Viro { 3650921a1650SAl Viro dput(dentry); 36515955102cSAl Viro inode_unlock(path->dentry->d_inode); 3652a8104a9fSAl Viro mnt_drop_write(path->mnt); 3653921a1650SAl Viro path_put(path); 3654921a1650SAl Viro } 3655921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3656921a1650SAl Viro 3657520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36581ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3659dae6ad8fSAl Viro { 3660181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3661dae6ad8fSAl Viro } 3662dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3663dae6ad8fSAl Viro 36646521f891SChristian Brauner /** 36656521f891SChristian Brauner * vfs_mknod - create device node or file 36666521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 36676521f891SChristian Brauner * @dir: inode of @dentry 36686521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 36696521f891SChristian Brauner * @mode: mode of the new device node or file 36706521f891SChristian Brauner * @dev: device number of device to create 36716521f891SChristian Brauner * 36726521f891SChristian Brauner * Create a device node or file. 36736521f891SChristian Brauner * 36746521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 36756521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 36766521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 36776521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 36786521f891SChristian Brauner * raw inode simply passs init_user_ns. 36796521f891SChristian Brauner */ 36806521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 36816521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 36821da177e4SLinus Torvalds { 3683a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 36846521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 36851da177e4SLinus Torvalds 36861da177e4SLinus Torvalds if (error) 36871da177e4SLinus Torvalds return error; 36881da177e4SLinus Torvalds 3689a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3690a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 36911da177e4SLinus Torvalds return -EPERM; 36921da177e4SLinus Torvalds 3693acfa4380SAl Viro if (!dir->i_op->mknod) 36941da177e4SLinus Torvalds return -EPERM; 36951da177e4SLinus Torvalds 369608ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 369708ce5f16SSerge E. Hallyn if (error) 369808ce5f16SSerge E. Hallyn return error; 369908ce5f16SSerge E. Hallyn 37001da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 37011da177e4SLinus Torvalds if (error) 37021da177e4SLinus Torvalds return error; 37031da177e4SLinus Torvalds 3704549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3705a74574aaSStephen Smalley if (!error) 3706f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37071da177e4SLinus Torvalds return error; 37081da177e4SLinus Torvalds } 37094d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37101da177e4SLinus Torvalds 3711f69aac00SAl Viro static int may_mknod(umode_t mode) 3712463c3197SDave Hansen { 3713463c3197SDave Hansen switch (mode & S_IFMT) { 3714463c3197SDave Hansen case S_IFREG: 3715463c3197SDave Hansen case S_IFCHR: 3716463c3197SDave Hansen case S_IFBLK: 3717463c3197SDave Hansen case S_IFIFO: 3718463c3197SDave Hansen case S_IFSOCK: 3719463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3720463c3197SDave Hansen return 0; 3721463c3197SDave Hansen case S_IFDIR: 3722463c3197SDave Hansen return -EPERM; 3723463c3197SDave Hansen default: 3724463c3197SDave Hansen return -EINVAL; 3725463c3197SDave Hansen } 3726463c3197SDave Hansen } 3727463c3197SDave Hansen 37285fee64fcSChristoph Hellwig static long do_mknodat(int dfd, const char __user *filename, umode_t mode, 372987c4e192SDominik Brodowski unsigned int dev) 37301da177e4SLinus Torvalds { 37316521f891SChristian Brauner struct user_namespace *mnt_userns; 37321da177e4SLinus Torvalds struct dentry *dentry; 3733dae6ad8fSAl Viro struct path path; 3734dae6ad8fSAl Viro int error; 3735972567f1SJeff Layton unsigned int lookup_flags = 0; 37361da177e4SLinus Torvalds 37378e4bfca1SAl Viro error = may_mknod(mode); 37388e4bfca1SAl Viro if (error) 37398e4bfca1SAl Viro return error; 3740972567f1SJeff Layton retry: 3741972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3742dae6ad8fSAl Viro if (IS_ERR(dentry)) 3743dae6ad8fSAl Viro return PTR_ERR(dentry); 37442ad94ae6SAl Viro 3745dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3746ce3b0f8dSAl Viro mode &= ~current_umask(); 3747dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3748be6d3e56SKentaro Takeda if (error) 3749a8104a9fSAl Viro goto out; 37506521f891SChristian Brauner 37516521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 37521da177e4SLinus Torvalds switch (mode & S_IFMT) { 37531da177e4SLinus Torvalds case 0: case S_IFREG: 37546521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 37556521f891SChristian Brauner dentry, mode, true); 375605d1a717SMimi Zohar if (!error) 3757a2d2329eSChristian Brauner ima_post_path_mknod(mnt_userns, dentry); 37581da177e4SLinus Torvalds break; 37591da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 37606521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 37616521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 37621da177e4SLinus Torvalds break; 37631da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 37646521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 37656521f891SChristian Brauner dentry, mode, 0); 37661da177e4SLinus Torvalds break; 37671da177e4SLinus Torvalds } 3768a8104a9fSAl Viro out: 3769921a1650SAl Viro done_path_create(&path, dentry); 3770972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3771972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3772972567f1SJeff Layton goto retry; 3773972567f1SJeff Layton } 37741da177e4SLinus Torvalds return error; 37751da177e4SLinus Torvalds } 37761da177e4SLinus Torvalds 377787c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 377887c4e192SDominik Brodowski unsigned int, dev) 377987c4e192SDominik Brodowski { 378087c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 378187c4e192SDominik Brodowski } 378287c4e192SDominik Brodowski 37838208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37845590ff0dSUlrich Drepper { 378587c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 37865590ff0dSUlrich Drepper } 37875590ff0dSUlrich Drepper 37886521f891SChristian Brauner /** 37896521f891SChristian Brauner * vfs_mkdir - create directory 37906521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 37916521f891SChristian Brauner * @dir: inode of @dentry 37926521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 37936521f891SChristian Brauner * @mode: mode of the new directory 37946521f891SChristian Brauner * 37956521f891SChristian Brauner * Create a directory. 37966521f891SChristian Brauner * 37976521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 37986521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 37996521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 38006521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 38016521f891SChristian Brauner * raw inode simply passs init_user_ns. 38026521f891SChristian Brauner */ 38036521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 38046521f891SChristian Brauner struct dentry *dentry, umode_t mode) 38051da177e4SLinus Torvalds { 38066521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 38078de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38081da177e4SLinus Torvalds 38091da177e4SLinus Torvalds if (error) 38101da177e4SLinus Torvalds return error; 38111da177e4SLinus Torvalds 3812acfa4380SAl Viro if (!dir->i_op->mkdir) 38131da177e4SLinus Torvalds return -EPERM; 38141da177e4SLinus Torvalds 38151da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38161da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38171da177e4SLinus Torvalds if (error) 38181da177e4SLinus Torvalds return error; 38191da177e4SLinus Torvalds 38208de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38218de52778SAl Viro return -EMLINK; 38228de52778SAl Viro 3823549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 3824a74574aaSStephen Smalley if (!error) 3825f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38261da177e4SLinus Torvalds return error; 38271da177e4SLinus Torvalds } 38284d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38291da177e4SLinus Torvalds 383083ff98c3SChristoph Hellwig static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38311da177e4SLinus Torvalds { 38326902d925SDave Hansen struct dentry *dentry; 3833dae6ad8fSAl Viro struct path path; 3834dae6ad8fSAl Viro int error; 3835b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38361da177e4SLinus Torvalds 3837b76d8b82SJeff Layton retry: 3838b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38396902d925SDave Hansen if (IS_ERR(dentry)) 3840dae6ad8fSAl Viro return PTR_ERR(dentry); 38416902d925SDave Hansen 3842dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3843ce3b0f8dSAl Viro mode &= ~current_umask(); 3844dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 38456521f891SChristian Brauner if (!error) { 38466521f891SChristian Brauner struct user_namespace *mnt_userns; 38476521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 3848549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 3849549c7297SChristian Brauner mode); 38506521f891SChristian Brauner } 3851921a1650SAl Viro done_path_create(&path, dentry); 3852b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3853b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3854b76d8b82SJeff Layton goto retry; 3855b76d8b82SJeff Layton } 38561da177e4SLinus Torvalds return error; 38571da177e4SLinus Torvalds } 38581da177e4SLinus Torvalds 38590101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38600101db7aSDominik Brodowski { 38610101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38620101db7aSDominik Brodowski } 38630101db7aSDominik Brodowski 3864a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38655590ff0dSUlrich Drepper { 38660101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38675590ff0dSUlrich Drepper } 38685590ff0dSUlrich Drepper 38696521f891SChristian Brauner /** 38706521f891SChristian Brauner * vfs_rmdir - remove directory 38716521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 38726521f891SChristian Brauner * @dir: inode of @dentry 38736521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 38746521f891SChristian Brauner * 38756521f891SChristian Brauner * Remove a directory. 38766521f891SChristian Brauner * 38776521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 38786521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 38796521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 38806521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 38816521f891SChristian Brauner * raw inode simply passs init_user_ns. 38826521f891SChristian Brauner */ 38836521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 38846521f891SChristian Brauner struct dentry *dentry) 38851da177e4SLinus Torvalds { 38866521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 38871da177e4SLinus Torvalds 38881da177e4SLinus Torvalds if (error) 38891da177e4SLinus Torvalds return error; 38901da177e4SLinus Torvalds 3891acfa4380SAl Viro if (!dir->i_op->rmdir) 38921da177e4SLinus Torvalds return -EPERM; 38931da177e4SLinus Torvalds 38941d2ef590SAl Viro dget(dentry); 38955955102cSAl Viro inode_lock(dentry->d_inode); 3896912dbc15SSage Weil 38971da177e4SLinus Torvalds error = -EBUSY; 38987af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3899912dbc15SSage Weil goto out; 3900912dbc15SSage Weil 39011da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3902912dbc15SSage Weil if (error) 3903912dbc15SSage Weil goto out; 3904912dbc15SSage Weil 39051da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3906912dbc15SSage Weil if (error) 3907912dbc15SSage Weil goto out; 3908912dbc15SSage Weil 39098767712fSAl Viro shrink_dcache_parent(dentry); 39101da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3911d83c49f3SAl Viro dont_mount(dentry); 39128ed936b5SEric W. Biederman detach_mounts(dentry); 3913116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 39141da177e4SLinus Torvalds 3915912dbc15SSage Weil out: 39165955102cSAl Viro inode_unlock(dentry->d_inode); 39171d2ef590SAl Viro dput(dentry); 3918912dbc15SSage Weil if (!error) 3919912dbc15SSage Weil d_delete(dentry); 39201da177e4SLinus Torvalds return error; 39211da177e4SLinus Torvalds } 39224d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 39231da177e4SLinus Torvalds 3924e24ab0efSChristoph Hellwig long do_rmdir(int dfd, struct filename *name) 39251da177e4SLinus Torvalds { 39266521f891SChristian Brauner struct user_namespace *mnt_userns; 39271da177e4SLinus Torvalds int error = 0; 39281da177e4SLinus Torvalds struct dentry *dentry; 3929f5beed75SAl Viro struct path path; 3930f5beed75SAl Viro struct qstr last; 3931f5beed75SAl Viro int type; 3932c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3933c6ee9206SJeff Layton retry: 3934e24ab0efSChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, 3935c1d4dd27SAl Viro &path, &last, &type); 393691a27b2aSJeff Layton if (IS_ERR(name)) 393791a27b2aSJeff Layton return PTR_ERR(name); 39381da177e4SLinus Torvalds 3939f5beed75SAl Viro switch (type) { 39401da177e4SLinus Torvalds case LAST_DOTDOT: 39411da177e4SLinus Torvalds error = -ENOTEMPTY; 39421da177e4SLinus Torvalds goto exit1; 39431da177e4SLinus Torvalds case LAST_DOT: 39441da177e4SLinus Torvalds error = -EINVAL; 39451da177e4SLinus Torvalds goto exit1; 39461da177e4SLinus Torvalds case LAST_ROOT: 39471da177e4SLinus Torvalds error = -EBUSY; 39481da177e4SLinus Torvalds goto exit1; 39491da177e4SLinus Torvalds } 39500612d9fbSOGAWA Hirofumi 3951f5beed75SAl Viro error = mnt_want_write(path.mnt); 3952c30dabfeSJan Kara if (error) 3953c30dabfeSJan Kara goto exit1; 39540612d9fbSOGAWA Hirofumi 39555955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3956f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39571da177e4SLinus Torvalds error = PTR_ERR(dentry); 39586902d925SDave Hansen if (IS_ERR(dentry)) 39596902d925SDave Hansen goto exit2; 3960e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3961e6bc45d6STheodore Ts'o error = -ENOENT; 3962e6bc45d6STheodore Ts'o goto exit3; 3963e6bc45d6STheodore Ts'o } 3964f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3965be6d3e56SKentaro Takeda if (error) 3966c30dabfeSJan Kara goto exit3; 39676521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 39686521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 39690622753bSDave Hansen exit3: 39701da177e4SLinus Torvalds dput(dentry); 39716902d925SDave Hansen exit2: 39725955102cSAl Viro inode_unlock(path.dentry->d_inode); 3973f5beed75SAl Viro mnt_drop_write(path.mnt); 39741da177e4SLinus Torvalds exit1: 3975f5beed75SAl Viro path_put(&path); 3976c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3977c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3978c6ee9206SJeff Layton goto retry; 3979c6ee9206SJeff Layton } 398024fb33d4SAl Viro putname(name); 39811da177e4SLinus Torvalds return error; 39821da177e4SLinus Torvalds } 39831da177e4SLinus Torvalds 39843cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39855590ff0dSUlrich Drepper { 3986e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 39875590ff0dSUlrich Drepper } 39885590ff0dSUlrich Drepper 3989b21996e3SJ. Bruce Fields /** 3990b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 39916521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 3992b21996e3SJ. Bruce Fields * @dir: parent directory 3993b21996e3SJ. Bruce Fields * @dentry: victim 3994b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3995b21996e3SJ. Bruce Fields * 3996b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3997b21996e3SJ. Bruce Fields * 3998b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3999b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4000b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4001b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4002b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4003b21996e3SJ. Bruce Fields * 4004b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4005b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4006b21996e3SJ. Bruce Fields * to be NFS exported. 40076521f891SChristian Brauner * 40086521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 40096521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 40106521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 40116521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 40126521f891SChristian Brauner * raw inode simply passs init_user_ns. 4013b21996e3SJ. Bruce Fields */ 40146521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 40156521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 40161da177e4SLinus Torvalds { 40179accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 40186521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 40191da177e4SLinus Torvalds 40201da177e4SLinus Torvalds if (error) 40211da177e4SLinus Torvalds return error; 40221da177e4SLinus Torvalds 4023acfa4380SAl Viro if (!dir->i_op->unlink) 40241da177e4SLinus Torvalds return -EPERM; 40251da177e4SLinus Torvalds 40265955102cSAl Viro inode_lock(target); 4027*51cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 4028*51cc3a66SHugh Dickins error = -EPERM; 4029*51cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 40301da177e4SLinus Torvalds error = -EBUSY; 40311da177e4SLinus Torvalds else { 40321da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4033bec1052eSAl Viro if (!error) { 40345a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 40355a14696cSJ. Bruce Fields if (error) 4036b21996e3SJ. Bruce Fields goto out; 40371da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 40388ed936b5SEric W. Biederman if (!error) { 4039d83c49f3SAl Viro dont_mount(dentry); 40408ed936b5SEric W. Biederman detach_mounts(dentry); 4041116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 40428ed936b5SEric W. Biederman } 4043bec1052eSAl Viro } 40441da177e4SLinus Torvalds } 4045b21996e3SJ. Bruce Fields out: 40465955102cSAl Viro inode_unlock(target); 40471da177e4SLinus Torvalds 40481da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40491da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40509accbb97SJ. Bruce Fields fsnotify_link_count(target); 40511da177e4SLinus Torvalds d_delete(dentry); 40521da177e4SLinus Torvalds } 40530eeca283SRobert Love 40541da177e4SLinus Torvalds return error; 40551da177e4SLinus Torvalds } 40564d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40571da177e4SLinus Torvalds 40581da177e4SLinus Torvalds /* 40591da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40601b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40611da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40621da177e4SLinus Torvalds * while waiting on the I/O. 40631da177e4SLinus Torvalds */ 4064da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40651da177e4SLinus Torvalds { 40662ad94ae6SAl Viro int error; 40671da177e4SLinus Torvalds struct dentry *dentry; 4068f5beed75SAl Viro struct path path; 4069f5beed75SAl Viro struct qstr last; 4070f5beed75SAl Viro int type; 40711da177e4SLinus Torvalds struct inode *inode = NULL; 4072b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40735d18f813SJeff Layton unsigned int lookup_flags = 0; 40745d18f813SJeff Layton retry: 4075da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 407691a27b2aSJeff Layton if (IS_ERR(name)) 407791a27b2aSJeff Layton return PTR_ERR(name); 40782ad94ae6SAl Viro 40791da177e4SLinus Torvalds error = -EISDIR; 4080f5beed75SAl Viro if (type != LAST_NORM) 40811da177e4SLinus Torvalds goto exit1; 40820612d9fbSOGAWA Hirofumi 4083f5beed75SAl Viro error = mnt_want_write(path.mnt); 4084c30dabfeSJan Kara if (error) 4085c30dabfeSJan Kara goto exit1; 4086b21996e3SJ. Bruce Fields retry_deleg: 40875955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4088f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40891da177e4SLinus Torvalds error = PTR_ERR(dentry); 40901da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40916521f891SChristian Brauner struct user_namespace *mnt_userns; 40926521f891SChristian Brauner 40931da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4094f5beed75SAl Viro if (last.name[last.len]) 409550338b88STörök Edwin goto slashes; 40961da177e4SLinus Torvalds inode = dentry->d_inode; 4097b18825a7SDavid Howells if (d_is_negative(dentry)) 4098e6bc45d6STheodore Ts'o goto slashes; 40997de9c6eeSAl Viro ihold(inode); 4100f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4101be6d3e56SKentaro Takeda if (error) 4102c30dabfeSJan Kara goto exit2; 41036521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4104549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4105549c7297SChristian Brauner &delegated_inode); 41061da177e4SLinus Torvalds exit2: 41071da177e4SLinus Torvalds dput(dentry); 41081da177e4SLinus Torvalds } 41095955102cSAl Viro inode_unlock(path.dentry->d_inode); 41101da177e4SLinus Torvalds if (inode) 41111da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4112b21996e3SJ. Bruce Fields inode = NULL; 4113b21996e3SJ. Bruce Fields if (delegated_inode) { 41145a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4115b21996e3SJ. Bruce Fields if (!error) 4116b21996e3SJ. Bruce Fields goto retry_deleg; 4117b21996e3SJ. Bruce Fields } 4118f5beed75SAl Viro mnt_drop_write(path.mnt); 41191da177e4SLinus Torvalds exit1: 4120f5beed75SAl Viro path_put(&path); 41215d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 41225d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 41235d18f813SJeff Layton inode = NULL; 41245d18f813SJeff Layton goto retry; 41255d18f813SJeff Layton } 4126da2f1362SChristoph Hellwig putname(name); 41271da177e4SLinus Torvalds return error; 41281da177e4SLinus Torvalds 41291da177e4SLinus Torvalds slashes: 4130b18825a7SDavid Howells if (d_is_negative(dentry)) 4131b18825a7SDavid Howells error = -ENOENT; 413244b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4133b18825a7SDavid Howells error = -EISDIR; 4134b18825a7SDavid Howells else 4135b18825a7SDavid Howells error = -ENOTDIR; 41361da177e4SLinus Torvalds goto exit2; 41371da177e4SLinus Torvalds } 41381da177e4SLinus Torvalds 41392e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 41405590ff0dSUlrich Drepper { 41415590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 41425590ff0dSUlrich Drepper return -EINVAL; 41435590ff0dSUlrich Drepper 41445590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4145e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4146da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 41475590ff0dSUlrich Drepper } 41485590ff0dSUlrich Drepper 41493480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41505590ff0dSUlrich Drepper { 4151da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 41525590ff0dSUlrich Drepper } 41535590ff0dSUlrich Drepper 41546521f891SChristian Brauner /** 41556521f891SChristian Brauner * vfs_symlink - create symlink 41566521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 41576521f891SChristian Brauner * @dir: inode of @dentry 41586521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 41596521f891SChristian Brauner * @oldname: name of the file to link to 41606521f891SChristian Brauner * 41616521f891SChristian Brauner * Create a symlink. 41626521f891SChristian Brauner * 41636521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 41646521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 41656521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 41666521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 41676521f891SChristian Brauner * raw inode simply passs init_user_ns. 41686521f891SChristian Brauner */ 41696521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 41706521f891SChristian Brauner struct dentry *dentry, const char *oldname) 41711da177e4SLinus Torvalds { 41726521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 41731da177e4SLinus Torvalds 41741da177e4SLinus Torvalds if (error) 41751da177e4SLinus Torvalds return error; 41761da177e4SLinus Torvalds 4177acfa4380SAl Viro if (!dir->i_op->symlink) 41781da177e4SLinus Torvalds return -EPERM; 41791da177e4SLinus Torvalds 41801da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41811da177e4SLinus Torvalds if (error) 41821da177e4SLinus Torvalds return error; 41831da177e4SLinus Torvalds 4184549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4185a74574aaSStephen Smalley if (!error) 4186f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41871da177e4SLinus Torvalds return error; 41881da177e4SLinus Torvalds } 41894d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41901da177e4SLinus Torvalds 4191cd3acb6aSChristoph Hellwig static long do_symlinkat(const char __user *oldname, int newdfd, 4192b724e846SDominik Brodowski const char __user *newname) 41931da177e4SLinus Torvalds { 41942ad94ae6SAl Viro int error; 419591a27b2aSJeff Layton struct filename *from; 41966902d925SDave Hansen struct dentry *dentry; 4197dae6ad8fSAl Viro struct path path; 4198f46d3567SJeff Layton unsigned int lookup_flags = 0; 41991da177e4SLinus Torvalds 42001da177e4SLinus Torvalds from = getname(oldname); 42011da177e4SLinus Torvalds if (IS_ERR(from)) 42021da177e4SLinus Torvalds return PTR_ERR(from); 4203f46d3567SJeff Layton retry: 4204f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 42051da177e4SLinus Torvalds error = PTR_ERR(dentry); 42066902d925SDave Hansen if (IS_ERR(dentry)) 4207dae6ad8fSAl Viro goto out_putname; 42086902d925SDave Hansen 420991a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 42106521f891SChristian Brauner if (!error) { 42116521f891SChristian Brauner struct user_namespace *mnt_userns; 42126521f891SChristian Brauner 42136521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 42146521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 42156521f891SChristian Brauner from->name); 42166521f891SChristian Brauner } 4217921a1650SAl Viro done_path_create(&path, dentry); 4218f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4219f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4220f46d3567SJeff Layton goto retry; 4221f46d3567SJeff Layton } 42226902d925SDave Hansen out_putname: 42231da177e4SLinus Torvalds putname(from); 42241da177e4SLinus Torvalds return error; 42251da177e4SLinus Torvalds } 42261da177e4SLinus Torvalds 4227b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4228b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4229b724e846SDominik Brodowski { 4230b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4231b724e846SDominik Brodowski } 4232b724e846SDominik Brodowski 42333480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 42345590ff0dSUlrich Drepper { 4235b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 42365590ff0dSUlrich Drepper } 42375590ff0dSUlrich Drepper 4238146a8595SJ. Bruce Fields /** 4239146a8595SJ. Bruce Fields * vfs_link - create a new link 4240146a8595SJ. Bruce Fields * @old_dentry: object to be linked 42416521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4242146a8595SJ. Bruce Fields * @dir: new parent 4243146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4244146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4245146a8595SJ. Bruce Fields * 4246146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4247146a8595SJ. Bruce Fields * 4248146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4249146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4250146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4251146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4252146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4253146a8595SJ. Bruce Fields * 4254146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4255146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4256146a8595SJ. Bruce Fields * to be NFS exported. 42576521f891SChristian Brauner * 42586521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 42596521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42606521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42616521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42626521f891SChristian Brauner * raw inode simply passs init_user_ns. 4263146a8595SJ. Bruce Fields */ 42646521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 42656521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 42666521f891SChristian Brauner struct inode **delegated_inode) 42671da177e4SLinus Torvalds { 42681da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 42698de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 42701da177e4SLinus Torvalds int error; 42711da177e4SLinus Torvalds 42721da177e4SLinus Torvalds if (!inode) 42731da177e4SLinus Torvalds return -ENOENT; 42741da177e4SLinus Torvalds 42756521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 42761da177e4SLinus Torvalds if (error) 42771da177e4SLinus Torvalds return error; 42781da177e4SLinus Torvalds 42791da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 42801da177e4SLinus Torvalds return -EXDEV; 42811da177e4SLinus Torvalds 42821da177e4SLinus Torvalds /* 42831da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 42841da177e4SLinus Torvalds */ 42851da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 42861da177e4SLinus Torvalds return -EPERM; 42870bd23d09SEric W. Biederman /* 42880bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 42890bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 42900bd23d09SEric W. Biederman * the vfs. 42910bd23d09SEric W. Biederman */ 42926521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 42930bd23d09SEric W. Biederman return -EPERM; 4294acfa4380SAl Viro if (!dir->i_op->link) 42951da177e4SLinus Torvalds return -EPERM; 42967e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42971da177e4SLinus Torvalds return -EPERM; 42981da177e4SLinus Torvalds 42991da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 43001da177e4SLinus Torvalds if (error) 43011da177e4SLinus Torvalds return error; 43021da177e4SLinus Torvalds 43035955102cSAl Viro inode_lock(inode); 4304aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4305f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4306aae8a97dSAneesh Kumar K.V error = -ENOENT; 43078de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 43088de52778SAl Viro error = -EMLINK; 4309146a8595SJ. Bruce Fields else { 4310146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4311146a8595SJ. Bruce Fields if (!error) 43121da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4313146a8595SJ. Bruce Fields } 4314f4e0c30cSAl Viro 4315f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4316f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4317f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4318f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4319f4e0c30cSAl Viro } 43205955102cSAl Viro inode_unlock(inode); 4321e31e14ecSStephen Smalley if (!error) 43227e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 43231da177e4SLinus Torvalds return error; 43241da177e4SLinus Torvalds } 43254d359507SAl Viro EXPORT_SYMBOL(vfs_link); 43261da177e4SLinus Torvalds 43271da177e4SLinus Torvalds /* 43281da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 43291da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 43301da177e4SLinus Torvalds * newname. --KAB 43311da177e4SLinus Torvalds * 43321da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 43331da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 43341da177e4SLinus Torvalds * and other special files. --ADM 43351da177e4SLinus Torvalds */ 4336812931d6SChristoph Hellwig static int do_linkat(int olddfd, const char __user *oldname, int newdfd, 433746ea89ebSDominik Brodowski const char __user *newname, int flags) 43381da177e4SLinus Torvalds { 43396521f891SChristian Brauner struct user_namespace *mnt_userns; 43401da177e4SLinus Torvalds struct dentry *new_dentry; 4341dae6ad8fSAl Viro struct path old_path, new_path; 4342146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 434311a7b371SAneesh Kumar K.V int how = 0; 43441da177e4SLinus Torvalds int error; 43451da177e4SLinus Torvalds 434611a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4347c04030e1SUlrich Drepper return -EINVAL; 434811a7b371SAneesh Kumar K.V /* 4349f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4350f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4351f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 435211a7b371SAneesh Kumar K.V */ 4353f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4354f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4355f0cc6ffbSLinus Torvalds return -ENOENT; 435611a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4357f0cc6ffbSLinus Torvalds } 4358c04030e1SUlrich Drepper 435911a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 436011a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4361442e31caSJeff Layton retry: 436211a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 43631da177e4SLinus Torvalds if (error) 43642ad94ae6SAl Viro return error; 43652ad94ae6SAl Viro 4366442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4367442e31caSJeff Layton (how & LOOKUP_REVAL)); 43681da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43696902d925SDave Hansen if (IS_ERR(new_dentry)) 4370dae6ad8fSAl Viro goto out; 4371dae6ad8fSAl Viro 4372dae6ad8fSAl Viro error = -EXDEV; 4373dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4374dae6ad8fSAl Viro goto out_dput; 4375549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4376549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4377800179c9SKees Cook if (unlikely(error)) 4378800179c9SKees Cook goto out_dput; 4379dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4380be6d3e56SKentaro Takeda if (error) 4381a8104a9fSAl Viro goto out_dput; 43826521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 43836521f891SChristian Brauner new_dentry, &delegated_inode); 438475c3f29dSDave Hansen out_dput: 4385921a1650SAl Viro done_path_create(&new_path, new_dentry); 4386146a8595SJ. Bruce Fields if (delegated_inode) { 4387146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4388d22e6338SOleg Drokin if (!error) { 4389d22e6338SOleg Drokin path_put(&old_path); 4390146a8595SJ. Bruce Fields goto retry; 4391146a8595SJ. Bruce Fields } 4392d22e6338SOleg Drokin } 4393442e31caSJeff Layton if (retry_estale(error, how)) { 4394d22e6338SOleg Drokin path_put(&old_path); 4395442e31caSJeff Layton how |= LOOKUP_REVAL; 4396442e31caSJeff Layton goto retry; 4397442e31caSJeff Layton } 43981da177e4SLinus Torvalds out: 43992d8f3038SAl Viro path_put(&old_path); 44001da177e4SLinus Torvalds 44011da177e4SLinus Torvalds return error; 44021da177e4SLinus Torvalds } 44031da177e4SLinus Torvalds 440446ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 440546ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 440646ea89ebSDominik Brodowski { 440746ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 440846ea89ebSDominik Brodowski } 440946ea89ebSDominik Brodowski 44103480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 44115590ff0dSUlrich Drepper { 441246ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44135590ff0dSUlrich Drepper } 44145590ff0dSUlrich Drepper 4415bc27027aSMiklos Szeredi /** 4416bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 44172111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4418bc27027aSMiklos Szeredi * 4419bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4420bc27027aSMiklos Szeredi * 4421bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4422bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4423bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4424bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4425bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4426bc27027aSMiklos Szeredi * so. 4427bc27027aSMiklos Szeredi * 4428bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4429bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4430bc27027aSMiklos Szeredi * to be NFS exported. 4431bc27027aSMiklos Szeredi * 44321da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 44331da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 44341da177e4SLinus Torvalds * Problems: 44350117d427SMauro Carvalho Chehab * 4436d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 44371da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 44381da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4439a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 44401da177e4SLinus Torvalds * story. 44416cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 44426cedba89SJ. Bruce Fields * and source (if it is not a directory). 44431b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 44441da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 44451da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4446a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 44471da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 44481da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 44491da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4450a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 44511da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 44521da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 44531da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4454e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 44551b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 44561da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4457c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 44581da177e4SLinus Torvalds * locking]. 44591da177e4SLinus Torvalds */ 44609fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 44611da177e4SLinus Torvalds { 44621da177e4SLinus Torvalds int error; 44639fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 44649fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 44659fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 44669fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 44679fe61450SChristian Brauner unsigned int flags = rd->flags; 4468bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4469bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4470bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4471da1ce067SMiklos Szeredi bool new_is_dir = false; 4472da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 447349d31c2fSAl Viro struct name_snapshot old_name; 44741da177e4SLinus Torvalds 44758d3e2936SMiklos Szeredi if (source == target) 44761da177e4SLinus Torvalds return 0; 44771da177e4SLinus Torvalds 44786521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 44791da177e4SLinus Torvalds if (error) 44801da177e4SLinus Torvalds return error; 44811da177e4SLinus Torvalds 4482da1ce067SMiklos Szeredi if (!target) { 44836521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4484da1ce067SMiklos Szeredi } else { 4485da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4486da1ce067SMiklos Szeredi 4487da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44886521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 44896521f891SChristian Brauner new_dentry, is_dir); 4490da1ce067SMiklos Szeredi else 44916521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 44926521f891SChristian Brauner new_dentry, new_is_dir); 4493da1ce067SMiklos Szeredi } 44941da177e4SLinus Torvalds if (error) 44951da177e4SLinus Torvalds return error; 44961da177e4SLinus Torvalds 44972773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 44981da177e4SLinus Torvalds return -EPERM; 44991da177e4SLinus Torvalds 4500bc27027aSMiklos Szeredi /* 4501bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4502bc27027aSMiklos Szeredi * we'll need to flip '..'. 4503bc27027aSMiklos Szeredi */ 4504da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4505da1ce067SMiklos Szeredi if (is_dir) { 45066521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 450747291baaSChristian Brauner MAY_WRITE); 4508bc27027aSMiklos Szeredi if (error) 4509bc27027aSMiklos Szeredi return error; 4510bc27027aSMiklos Szeredi } 4511da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 45126521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 451347291baaSChristian Brauner MAY_WRITE); 4514da1ce067SMiklos Szeredi if (error) 4515da1ce067SMiklos Szeredi return error; 4516da1ce067SMiklos Szeredi } 4517da1ce067SMiklos Szeredi } 45180eeca283SRobert Love 45190b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 45200b3974ebSMiklos Szeredi flags); 4521bc27027aSMiklos Szeredi if (error) 4522bc27027aSMiklos Szeredi return error; 4523bc27027aSMiklos Szeredi 452449d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4525bc27027aSMiklos Szeredi dget(new_dentry); 4526da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4527bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4528bc27027aSMiklos Szeredi else if (target) 45295955102cSAl Viro inode_lock(target); 4530bc27027aSMiklos Szeredi 4531*51cc3a66SHugh Dickins error = -EPERM; 4532*51cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 4533*51cc3a66SHugh Dickins goto out; 4534*51cc3a66SHugh Dickins 4535bc27027aSMiklos Szeredi error = -EBUSY; 45367af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4537bc27027aSMiklos Szeredi goto out; 4538bc27027aSMiklos Szeredi 4539da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4540bc27027aSMiklos Szeredi error = -EMLINK; 4541da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4542bc27027aSMiklos Szeredi goto out; 4543da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4544da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4545da1ce067SMiklos Szeredi goto out; 4546da1ce067SMiklos Szeredi } 4547da1ce067SMiklos Szeredi if (!is_dir) { 4548bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4549bc27027aSMiklos Szeredi if (error) 4550bc27027aSMiklos Szeredi goto out; 4551da1ce067SMiklos Szeredi } 4552da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4553bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4554bc27027aSMiklos Szeredi if (error) 4555bc27027aSMiklos Szeredi goto out; 4556bc27027aSMiklos Szeredi } 4557549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4558520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4559bc27027aSMiklos Szeredi if (error) 4560bc27027aSMiklos Szeredi goto out; 4561bc27027aSMiklos Szeredi 4562da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 45638767712fSAl Viro if (is_dir) { 45648767712fSAl Viro shrink_dcache_parent(new_dentry); 4565bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 45668767712fSAl Viro } 4567bc27027aSMiklos Szeredi dont_mount(new_dentry); 45688ed936b5SEric W. Biederman detach_mounts(new_dentry); 4569bc27027aSMiklos Szeredi } 4570da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4571da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4572bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4573da1ce067SMiklos Szeredi else 4574da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4575da1ce067SMiklos Szeredi } 4576bc27027aSMiklos Szeredi out: 4577da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4578bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4579bc27027aSMiklos Szeredi else if (target) 45805955102cSAl Viro inode_unlock(target); 4581bc27027aSMiklos Szeredi dput(new_dentry); 4582da1ce067SMiklos Szeredi if (!error) { 4583f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4584da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4585da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4586f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4587da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4588da1ce067SMiklos Szeredi } 4589da1ce067SMiklos Szeredi } 459049d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 45910eeca283SRobert Love 45921da177e4SLinus Torvalds return error; 45931da177e4SLinus Torvalds } 45944d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 45951da177e4SLinus Torvalds 4596e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4597e886663cSJens Axboe struct filename *to, unsigned int flags) 45981da177e4SLinus Torvalds { 45999fe61450SChristian Brauner struct renamedata rd; 46001da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 46011da177e4SLinus Torvalds struct dentry *trap; 4602f5beed75SAl Viro struct path old_path, new_path; 4603f5beed75SAl Viro struct qstr old_last, new_last; 4604f5beed75SAl Viro int old_type, new_type; 46058e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4606f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4607c6a94284SJeff Layton bool should_retry = false; 4608e886663cSJens Axboe int error = -EINVAL; 4609520c8b16SMiklos Szeredi 46100d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4611e886663cSJens Axboe goto put_both; 4612da1ce067SMiklos Szeredi 46130d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 46140d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4615e886663cSJens Axboe goto put_both; 4616520c8b16SMiklos Szeredi 4617f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4618f5beed75SAl Viro target_flags = 0; 4619f5beed75SAl Viro 4620c6a94284SJeff Layton retry: 4621e886663cSJens Axboe from = filename_parentat(olddfd, from, lookup_flags, &old_path, 4622e886663cSJens Axboe &old_last, &old_type); 462391a27b2aSJeff Layton if (IS_ERR(from)) { 462491a27b2aSJeff Layton error = PTR_ERR(from); 4625e886663cSJens Axboe goto put_new; 462691a27b2aSJeff Layton } 46271da177e4SLinus Torvalds 4628e886663cSJens Axboe to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4629e886663cSJens Axboe &new_type); 463091a27b2aSJeff Layton if (IS_ERR(to)) { 463191a27b2aSJeff Layton error = PTR_ERR(to); 46321da177e4SLinus Torvalds goto exit1; 463391a27b2aSJeff Layton } 46341da177e4SLinus Torvalds 46351da177e4SLinus Torvalds error = -EXDEV; 4636f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 46371da177e4SLinus Torvalds goto exit2; 46381da177e4SLinus Torvalds 46391da177e4SLinus Torvalds error = -EBUSY; 4640f5beed75SAl Viro if (old_type != LAST_NORM) 46411da177e4SLinus Torvalds goto exit2; 46421da177e4SLinus Torvalds 46430a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 46440a7c3937SMiklos Szeredi error = -EEXIST; 4645f5beed75SAl Viro if (new_type != LAST_NORM) 46461da177e4SLinus Torvalds goto exit2; 46471da177e4SLinus Torvalds 4648f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4649c30dabfeSJan Kara if (error) 4650c30dabfeSJan Kara goto exit2; 4651c30dabfeSJan Kara 46528e6d782cSJ. Bruce Fields retry_deleg: 4653f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 46541da177e4SLinus Torvalds 4655f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 46561da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 46571da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 46581da177e4SLinus Torvalds goto exit3; 46591da177e4SLinus Torvalds /* source must exist */ 46601da177e4SLinus Torvalds error = -ENOENT; 4661b18825a7SDavid Howells if (d_is_negative(old_dentry)) 46621da177e4SLinus Torvalds goto exit4; 4663f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 46641da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 46651da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 46661da177e4SLinus Torvalds goto exit4; 46670a7c3937SMiklos Szeredi error = -EEXIST; 46680a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 46690a7c3937SMiklos Szeredi goto exit5; 4670da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4671da1ce067SMiklos Szeredi error = -ENOENT; 4672da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4673da1ce067SMiklos Szeredi goto exit5; 4674da1ce067SMiklos Szeredi 4675da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4676da1ce067SMiklos Szeredi error = -ENOTDIR; 4677f5beed75SAl Viro if (new_last.name[new_last.len]) 4678da1ce067SMiklos Szeredi goto exit5; 4679da1ce067SMiklos Szeredi } 4680da1ce067SMiklos Szeredi } 46810a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 46820a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 46830a7c3937SMiklos Szeredi error = -ENOTDIR; 4684f5beed75SAl Viro if (old_last.name[old_last.len]) 46850a7c3937SMiklos Szeredi goto exit5; 4686f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 46870a7c3937SMiklos Szeredi goto exit5; 46880a7c3937SMiklos Szeredi } 46890a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 46900a7c3937SMiklos Szeredi error = -EINVAL; 46910a7c3937SMiklos Szeredi if (old_dentry == trap) 46920a7c3937SMiklos Szeredi goto exit5; 46931da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4694da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46951da177e4SLinus Torvalds error = -ENOTEMPTY; 46961da177e4SLinus Torvalds if (new_dentry == trap) 46971da177e4SLinus Torvalds goto exit5; 46981da177e4SLinus Torvalds 4699f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4700f5beed75SAl Viro &new_path, new_dentry, flags); 4701be6d3e56SKentaro Takeda if (error) 4702c30dabfeSJan Kara goto exit5; 47039fe61450SChristian Brauner 47049fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 47059fe61450SChristian Brauner rd.old_dentry = old_dentry; 47066521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 47079fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 47089fe61450SChristian Brauner rd.new_dentry = new_dentry; 47096521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 47109fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 47119fe61450SChristian Brauner rd.flags = flags; 47129fe61450SChristian Brauner error = vfs_rename(&rd); 47131da177e4SLinus Torvalds exit5: 47141da177e4SLinus Torvalds dput(new_dentry); 47151da177e4SLinus Torvalds exit4: 47161da177e4SLinus Torvalds dput(old_dentry); 47171da177e4SLinus Torvalds exit3: 4718f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 47198e6d782cSJ. Bruce Fields if (delegated_inode) { 47208e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 47218e6d782cSJ. Bruce Fields if (!error) 47228e6d782cSJ. Bruce Fields goto retry_deleg; 47238e6d782cSJ. Bruce Fields } 4724f5beed75SAl Viro mnt_drop_write(old_path.mnt); 47251da177e4SLinus Torvalds exit2: 4726c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4727c6a94284SJeff Layton should_retry = true; 4728f5beed75SAl Viro path_put(&new_path); 47291da177e4SLinus Torvalds exit1: 4730f5beed75SAl Viro path_put(&old_path); 4731c6a94284SJeff Layton if (should_retry) { 4732c6a94284SJeff Layton should_retry = false; 4733c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4734c6a94284SJeff Layton goto retry; 4735c6a94284SJeff Layton } 4736e886663cSJens Axboe put_both: 4737e886663cSJens Axboe if (!IS_ERR(from)) 4738e886663cSJens Axboe putname(from); 4739e886663cSJens Axboe put_new: 4740e886663cSJens Axboe if (!IS_ERR(to)) 4741e886663cSJens Axboe putname(to); 47421da177e4SLinus Torvalds return error; 47431da177e4SLinus Torvalds } 47441da177e4SLinus Torvalds 4745ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4746ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4747ee81feb6SDominik Brodowski { 4748e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4749e886663cSJens Axboe flags); 4750ee81feb6SDominik Brodowski } 4751ee81feb6SDominik Brodowski 4752520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4753520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4754520c8b16SMiklos Szeredi { 4755e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4756e886663cSJens Axboe 0); 4757520c8b16SMiklos Szeredi } 4758520c8b16SMiklos Szeredi 4759a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 47605590ff0dSUlrich Drepper { 4761e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4762e886663cSJens Axboe getname(newname), 0); 47635590ff0dSUlrich Drepper } 47645590ff0dSUlrich Drepper 47655d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 47661da177e4SLinus Torvalds { 47675d826c84SAl Viro int len = PTR_ERR(link); 47681da177e4SLinus Torvalds if (IS_ERR(link)) 47691da177e4SLinus Torvalds goto out; 47701da177e4SLinus Torvalds 47711da177e4SLinus Torvalds len = strlen(link); 47721da177e4SLinus Torvalds if (len > (unsigned) buflen) 47731da177e4SLinus Torvalds len = buflen; 47741da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 47751da177e4SLinus Torvalds len = -EFAULT; 47761da177e4SLinus Torvalds out: 47771da177e4SLinus Torvalds return len; 47781da177e4SLinus Torvalds } 47791da177e4SLinus Torvalds 4780d60874cdSMiklos Szeredi /** 4781fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4782fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4783fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4784fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4785fd4a0edfSMiklos Szeredi * 4786fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4787fd4a0edfSMiklos Szeredi * 4788fd4a0edfSMiklos Szeredi * Does not call security hook. 4789fd4a0edfSMiklos Szeredi */ 4790fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4791fd4a0edfSMiklos Szeredi { 4792fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4793f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4794f2df5da6SAl Viro const char *link; 4795f2df5da6SAl Viro int res; 4796fd4a0edfSMiklos Szeredi 479776fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 479876fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 479976fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 480076fca90eSMiklos Szeredi 480176fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4802fd4a0edfSMiklos Szeredi return -EINVAL; 4803fd4a0edfSMiklos Szeredi 480476fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 480576fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 480676fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 480776fca90eSMiklos Szeredi } 480876fca90eSMiklos Szeredi 48094c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4810f2df5da6SAl Viro if (!link) { 4811f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4812f2df5da6SAl Viro if (IS_ERR(link)) 4813f2df5da6SAl Viro return PTR_ERR(link); 4814f2df5da6SAl Viro } 4815f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4816f2df5da6SAl Viro do_delayed_call(&done); 4817f2df5da6SAl Viro return res; 4818fd4a0edfSMiklos Szeredi } 4819fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 48201da177e4SLinus Torvalds 4821d60874cdSMiklos Szeredi /** 4822d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4823d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4824d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4825d60874cdSMiklos Szeredi * 4826d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4827d60874cdSMiklos Szeredi * 4828d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4829d60874cdSMiklos Szeredi * 4830d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4831d60874cdSMiklos Szeredi */ 4832d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4833d60874cdSMiklos Szeredi { 4834d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4835d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4836d60874cdSMiklos Szeredi 4837d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4838d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4839d60874cdSMiklos Szeredi if (!res) 4840d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4841d60874cdSMiklos Szeredi } 4842d60874cdSMiklos Szeredi return res; 4843d60874cdSMiklos Szeredi } 4844d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4845d60874cdSMiklos Szeredi 48461da177e4SLinus Torvalds /* get the link contents into pagecache */ 48476b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4848fceef393SAl Viro struct delayed_call *callback) 48491da177e4SLinus Torvalds { 4850ebd09abbSDuane Griffin char *kaddr; 48511da177e4SLinus Torvalds struct page *page; 48526b255391SAl Viro struct address_space *mapping = inode->i_mapping; 48536b255391SAl Viro 4854d3883d4fSAl Viro if (!dentry) { 4855d3883d4fSAl Viro page = find_get_page(mapping, 0); 4856d3883d4fSAl Viro if (!page) 48576b255391SAl Viro return ERR_PTR(-ECHILD); 4858d3883d4fSAl Viro if (!PageUptodate(page)) { 4859d3883d4fSAl Viro put_page(page); 4860d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4861d3883d4fSAl Viro } 4862d3883d4fSAl Viro } else { 4863090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 48641da177e4SLinus Torvalds if (IS_ERR(page)) 48656fe6900eSNick Piggin return (char*)page; 4866d3883d4fSAl Viro } 4867fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 486821fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 486921fc61c7SAl Viro kaddr = page_address(page); 48706b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4871ebd09abbSDuane Griffin return kaddr; 48721da177e4SLinus Torvalds } 48731da177e4SLinus Torvalds 48746b255391SAl Viro EXPORT_SYMBOL(page_get_link); 48751da177e4SLinus Torvalds 4876fceef393SAl Viro void page_put_link(void *arg) 48771da177e4SLinus Torvalds { 4878fceef393SAl Viro put_page(arg); 48791da177e4SLinus Torvalds } 48804d359507SAl Viro EXPORT_SYMBOL(page_put_link); 48811da177e4SLinus Torvalds 4882aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4883aa80deabSAl Viro { 4884fceef393SAl Viro DEFINE_DELAYED_CALL(done); 48856b255391SAl Viro int res = readlink_copy(buffer, buflen, 48866b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4887fceef393SAl Viro &done)); 4888fceef393SAl Viro do_delayed_call(&done); 4889aa80deabSAl Viro return res; 4890aa80deabSAl Viro } 4891aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4892aa80deabSAl Viro 489354566b2cSNick Piggin /* 489454566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 489554566b2cSNick Piggin */ 489654566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48971da177e4SLinus Torvalds { 48981da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48990adb25d2SKirill Korotaev struct page *page; 4900afddba49SNick Piggin void *fsdata; 4901beb497abSDmitriy Monakhov int err; 4902c718a975STetsuo Handa unsigned int flags = 0; 490354566b2cSNick Piggin if (nofs) 490454566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 49051da177e4SLinus Torvalds 49067e53cac4SNeilBrown retry: 4907afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 490854566b2cSNick Piggin flags, &page, &fsdata); 49091da177e4SLinus Torvalds if (err) 4910afddba49SNick Piggin goto fail; 4911afddba49SNick Piggin 491221fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4913afddba49SNick Piggin 4914afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4915afddba49SNick Piggin page, fsdata); 49161da177e4SLinus Torvalds if (err < 0) 49171da177e4SLinus Torvalds goto fail; 4918afddba49SNick Piggin if (err < len-1) 4919afddba49SNick Piggin goto retry; 4920afddba49SNick Piggin 49211da177e4SLinus Torvalds mark_inode_dirty(inode); 49221da177e4SLinus Torvalds return 0; 49231da177e4SLinus Torvalds fail: 49241da177e4SLinus Torvalds return err; 49251da177e4SLinus Torvalds } 49264d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 49271da177e4SLinus Torvalds 49280adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 49290adb25d2SKirill Korotaev { 49300adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4931c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 49320adb25d2SKirill Korotaev } 49334d359507SAl Viro EXPORT_SYMBOL(page_symlink); 49340adb25d2SKirill Korotaev 493592e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 49366b255391SAl Viro .get_link = page_get_link, 49371da177e4SLinus Torvalds }; 49381da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4939