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)) 446*549c7297SChristian 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 */ 5571f55a6ecSAl Viro unsigned int flags; 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 5769883d185SAl Viro static void set_nameidata(struct nameidata *p, int dfd, struct filename *name) 577894bc8c4SAl Viro { 578756daf26SNeilBrown struct nameidata *old = current->nameidata; 579756daf26SNeilBrown p->stack = p->internal; 580c8a53ee5SAl Viro p->dfd = dfd; 581c8a53ee5SAl Viro p->name = name; 582756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 5839883d185SAl Viro p->saved = old; 584756daf26SNeilBrown current->nameidata = p; 585894bc8c4SAl Viro } 586894bc8c4SAl Viro 5879883d185SAl Viro static void restore_nameidata(void) 588894bc8c4SAl Viro { 5899883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 590756daf26SNeilBrown 591756daf26SNeilBrown current->nameidata = old; 592756daf26SNeilBrown if (old) 593756daf26SNeilBrown old->total_link_count = now->total_link_count; 594e1a63bbcSAl Viro if (now->stack != now->internal) 595756daf26SNeilBrown kfree(now->stack); 596894bc8c4SAl Viro } 597894bc8c4SAl Viro 59860ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 599894bc8c4SAl Viro { 600bc40aee0SAl Viro struct saved *p; 601bc40aee0SAl Viro 6026da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 60360ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 604bc40aee0SAl Viro if (unlikely(!p)) 60560ef60c7SAl Viro return false; 606894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 607894bc8c4SAl Viro nd->stack = p; 60860ef60c7SAl Viro return true; 609894bc8c4SAl Viro } 610894bc8c4SAl Viro 611397d425dSEric W. Biederman /** 6126b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 613397d425dSEric W. Biederman * 614397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 615397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 616397d425dSEric W. Biederman */ 6176b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 618397d425dSEric W. Biederman { 61995dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 620397d425dSEric W. Biederman 621402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 622402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 623397d425dSEric W. Biederman return true; 624397d425dSEric W. Biederman 6256b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 626397d425dSEric W. Biederman } 627397d425dSEric W. Biederman 6287973387aSAl Viro static void drop_links(struct nameidata *nd) 6297973387aSAl Viro { 6307973387aSAl Viro int i = nd->depth; 6317973387aSAl Viro while (i--) { 6327973387aSAl Viro struct saved *last = nd->stack + i; 633fceef393SAl Viro do_delayed_call(&last->done); 634fceef393SAl Viro clear_delayed_call(&last->done); 6357973387aSAl Viro } 6367973387aSAl Viro } 6377973387aSAl Viro 6387973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6397973387aSAl Viro { 6407973387aSAl Viro drop_links(nd); 6417973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6427973387aSAl Viro int i; 6437973387aSAl Viro path_put(&nd->path); 6447973387aSAl Viro for (i = 0; i < nd->depth; i++) 6457973387aSAl Viro path_put(&nd->stack[i].link); 64684a2bd39SAl Viro if (nd->flags & LOOKUP_ROOT_GRABBED) { 647102b8af2SAl Viro path_put(&nd->root); 64884a2bd39SAl Viro nd->flags &= ~LOOKUP_ROOT_GRABBED; 649102b8af2SAl Viro } 6507973387aSAl Viro } else { 6517973387aSAl Viro nd->flags &= ~LOOKUP_RCU; 6527973387aSAl Viro rcu_read_unlock(); 6537973387aSAl Viro } 6547973387aSAl Viro nd->depth = 0; 6557973387aSAl Viro } 6567973387aSAl Viro 6577973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6582aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6597973387aSAl Viro { 6602aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 6617973387aSAl Viro if (unlikely(res)) { 6627973387aSAl Viro if (res > 0) 6637973387aSAl Viro path->mnt = NULL; 6647973387aSAl Viro path->dentry = NULL; 6657973387aSAl Viro return false; 6667973387aSAl Viro } 6677973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 6687973387aSAl Viro path->dentry = NULL; 6697973387aSAl Viro return false; 6707973387aSAl Viro } 6717973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 6727973387aSAl Viro } 6737973387aSAl Viro 6742aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 6752aa38470SAl Viro struct path *path, unsigned seq) 6762aa38470SAl Viro { 6775bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 6782aa38470SAl Viro } 6792aa38470SAl Viro 6807973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 6817973387aSAl Viro { 6827973387aSAl Viro int i; 6837973387aSAl Viro for (i = 0; i < nd->depth; i++) { 6847973387aSAl Viro struct saved *last = nd->stack + i; 6857973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 6867973387aSAl Viro drop_links(nd); 6877973387aSAl Viro nd->depth = i + 1; 6887973387aSAl Viro return false; 6897973387aSAl Viro } 6907973387aSAl Viro } 6917973387aSAl Viro return true; 6927973387aSAl Viro } 6937973387aSAl Viro 694ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 695ee594bffSAl Viro { 696adb21d2bSAleksa Sarai /* 697adb21d2bSAleksa Sarai * For scoped-lookups (where nd->root has been zeroed), we need to 698adb21d2bSAleksa Sarai * restart the whole lookup from scratch -- because set_root() is wrong 699adb21d2bSAleksa Sarai * for these lookups (nd->dfd is the root, not the filesystem root). 700adb21d2bSAleksa Sarai */ 701adb21d2bSAleksa Sarai if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED)) 702adb21d2bSAleksa Sarai return false; 703adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 704ee594bffSAl Viro if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT)) 705ee594bffSAl Viro return true; 70684a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 707ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 708ee594bffSAl Viro } 709ee594bffSAl Viro 71019660af7SAl Viro /* 71131e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 71219660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 71319660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 71457e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 71519660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 71619660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 71719660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 71819660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 71931e6b01fSNick Piggin */ 72031e6b01fSNick Piggin 72131e6b01fSNick Piggin /** 72219660af7SAl Viro * unlazy_walk - try to switch to ref-walk mode. 72319660af7SAl Viro * @nd: nameidata pathwalk data 72439191628SRandy Dunlap * Returns: 0 on success, -ECHILD on failure 72531e6b01fSNick Piggin * 7264675ac39SAl Viro * unlazy_walk attempts to legitimize the current nd->path and nd->root 7274675ac39SAl Viro * for ref-walk mode. 7284675ac39SAl Viro * Must be called from rcu-walk context. 7297973387aSAl Viro * Nothing should touch nameidata between unlazy_walk() failure and 7307973387aSAl Viro * terminate_walk(). 73131e6b01fSNick Piggin */ 7324675ac39SAl Viro static int unlazy_walk(struct nameidata *nd) 73331e6b01fSNick Piggin { 73431e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 73531e6b01fSNick Piggin 73631e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 737e5c832d5SLinus Torvalds 738e5c832d5SLinus Torvalds nd->flags &= ~LOOKUP_RCU; 7397973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7404675ac39SAl Viro goto out1; 74184a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 74284a2bd39SAl Viro goto out; 743ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7444675ac39SAl Viro goto out; 7454675ac39SAl Viro rcu_read_unlock(); 7464675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 7474675ac39SAl Viro return 0; 7484675ac39SAl Viro 74984a2bd39SAl Viro out1: 7504675ac39SAl Viro nd->path.mnt = NULL; 7514675ac39SAl Viro nd->path.dentry = NULL; 7524675ac39SAl Viro out: 7534675ac39SAl Viro rcu_read_unlock(); 7544675ac39SAl Viro return -ECHILD; 7554675ac39SAl Viro } 7564675ac39SAl Viro 7574675ac39SAl Viro /** 7584675ac39SAl Viro * unlazy_child - try to switch to ref-walk mode. 7594675ac39SAl Viro * @nd: nameidata pathwalk data 7604675ac39SAl Viro * @dentry: child of nd->path.dentry 7614675ac39SAl Viro * @seq: seq number to check dentry against 7624675ac39SAl Viro * Returns: 0 on success, -ECHILD on failure 7634675ac39SAl Viro * 7644675ac39SAl Viro * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry 7654675ac39SAl Viro * for ref-walk mode. @dentry must be a path found by a do_lookup call on 7664675ac39SAl Viro * @nd. Must be called from rcu-walk context. 7674675ac39SAl Viro * Nothing should touch nameidata between unlazy_child() failure and 7684675ac39SAl Viro * terminate_walk(). 7694675ac39SAl Viro */ 7704675ac39SAl Viro static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq) 7714675ac39SAl Viro { 7724675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 7734675ac39SAl Viro 7744675ac39SAl Viro nd->flags &= ~LOOKUP_RCU; 7754675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 7764675ac39SAl Viro goto out2; 7777973387aSAl Viro if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq))) 7787973387aSAl Viro goto out2; 7794675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 7807973387aSAl Viro goto out1; 78148a066e7SAl Viro 78215570086SLinus Torvalds /* 7834675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 7844675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 7854675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 7864675ac39SAl Viro * number of the parent after we got the child sequence number. So we 7874675ac39SAl Viro * know the parent must still be valid if the child sequence number is 78815570086SLinus Torvalds */ 7894675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 790e5c832d5SLinus Torvalds goto out; 79184a2bd39SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 79284a2bd39SAl Viro goto out_dput; 793e5c832d5SLinus Torvalds /* 794e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 795e5c832d5SLinus Torvalds * still valid and get it if required. 796e5c832d5SLinus Torvalds */ 79784a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 79884a2bd39SAl Viro goto out_dput; 7998b61e74fSAl Viro rcu_read_unlock(); 80031e6b01fSNick Piggin return 0; 80119660af7SAl Viro 8027973387aSAl Viro out2: 8037973387aSAl Viro nd->path.mnt = NULL; 8047973387aSAl Viro out1: 8057973387aSAl Viro nd->path.dentry = NULL; 806e5c832d5SLinus Torvalds out: 8078b61e74fSAl Viro rcu_read_unlock(); 80884a2bd39SAl Viro return -ECHILD; 80984a2bd39SAl Viro out_dput: 81084a2bd39SAl Viro rcu_read_unlock(); 81184a2bd39SAl Viro dput(dentry); 81231e6b01fSNick Piggin return -ECHILD; 81331e6b01fSNick Piggin } 81431e6b01fSNick Piggin 8154ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 81634286d66SNick Piggin { 817a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8184ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 819a89f8337SAl Viro else 820a89f8337SAl Viro return 1; 82134286d66SNick Piggin } 82234286d66SNick Piggin 8239f1fafeeSAl Viro /** 8249f1fafeeSAl Viro * complete_walk - successful completion of path walk 8259f1fafeeSAl Viro * @nd: pointer nameidata 82639159de2SJeff Layton * 8279f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8289f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8299f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8309f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8319f1fafeeSAl Viro * need to drop nd->path. 83239159de2SJeff Layton */ 8339f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 83439159de2SJeff Layton { 83516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 83639159de2SJeff Layton int status; 83739159de2SJeff Layton 8389f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 839adb21d2bSAleksa Sarai /* 840adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 841adb21d2bSAleksa Sarai * externally-managed nd->root. 842adb21d2bSAleksa Sarai */ 843adb21d2bSAleksa Sarai if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED))) 8449f1fafeeSAl Viro nd->root.mnt = NULL; 8454675ac39SAl Viro if (unlikely(unlazy_walk(nd))) 84648a066e7SAl Viro return -ECHILD; 84748a066e7SAl Viro } 8489f1fafeeSAl Viro 849adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 850adb21d2bSAleksa Sarai /* 851adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 852adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 853adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 854adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 855adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 856adb21d2bSAleksa Sarai * 857adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 858adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 859adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 860adb21d2bSAleksa Sarai * requested root to userspace. 861adb21d2bSAleksa Sarai * 862adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 863adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 864adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 865adb21d2bSAleksa Sarai */ 866adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 867adb21d2bSAleksa Sarai return -EXDEV; 868adb21d2bSAleksa Sarai } 869adb21d2bSAleksa Sarai 87016c2cd71SAl Viro if (likely(!(nd->flags & LOOKUP_JUMPED))) 87139159de2SJeff Layton return 0; 87239159de2SJeff Layton 873ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 87416c2cd71SAl Viro return 0; 87516c2cd71SAl Viro 876ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 87739159de2SJeff Layton if (status > 0) 87839159de2SJeff Layton return 0; 87939159de2SJeff Layton 88016c2cd71SAl Viro if (!status) 88139159de2SJeff Layton status = -ESTALE; 88216c2cd71SAl Viro 88339159de2SJeff Layton return status; 88439159de2SJeff Layton } 88539159de2SJeff Layton 886740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 8872a737871SAl Viro { 88831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 8899e6697e2SAl Viro 890adb21d2bSAleksa Sarai /* 891adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 892adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 893adb21d2bSAleksa Sarai * from the dirfd. 894adb21d2bSAleksa Sarai */ 895adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 896adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 897adb21d2bSAleksa Sarai 8989e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 8998f47a016SAl Viro unsigned seq; 900c28cc364SNick Piggin 901c28cc364SNick Piggin do { 902c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 90331e6b01fSNick Piggin nd->root = fs->root; 9048f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 905c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9069e6697e2SAl Viro } else { 9079e6697e2SAl Viro get_fs_root(fs, &nd->root); 90884a2bd39SAl Viro nd->flags |= LOOKUP_ROOT_GRABBED; 9099e6697e2SAl Viro } 910740a1678SAleksa Sarai return 0; 91131e6b01fSNick Piggin } 91231e6b01fSNick Piggin 913248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 914248fb5b9SAl Viro { 915adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 916adb21d2bSAleksa Sarai return -EXDEV; 91772ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 91872ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 91972ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 92072ba2929SAleksa Sarai return -EXDEV; 92172ba2929SAleksa Sarai } 922740a1678SAleksa Sarai if (!nd->root.mnt) { 923740a1678SAleksa Sarai int error = set_root(nd); 924740a1678SAleksa Sarai if (error) 925740a1678SAleksa Sarai return error; 926740a1678SAleksa Sarai } 927248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 928248fb5b9SAl Viro struct dentry *d; 929248fb5b9SAl Viro nd->path = nd->root; 930248fb5b9SAl Viro d = nd->path.dentry; 931248fb5b9SAl Viro nd->inode = d->d_inode; 932248fb5b9SAl Viro nd->seq = nd->root_seq; 933248fb5b9SAl Viro if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq))) 934248fb5b9SAl Viro return -ECHILD; 935248fb5b9SAl Viro } else { 936248fb5b9SAl Viro path_put(&nd->path); 937248fb5b9SAl Viro nd->path = nd->root; 938248fb5b9SAl Viro path_get(&nd->path); 939248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 940248fb5b9SAl Viro } 941248fb5b9SAl Viro nd->flags |= LOOKUP_JUMPED; 942248fb5b9SAl Viro return 0; 943248fb5b9SAl Viro } 944248fb5b9SAl Viro 945b5fb63c1SChristoph Hellwig /* 9466b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 947b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 948b5fb63c1SChristoph Hellwig */ 9491bc82070SAleksa Sarai int nd_jump_link(struct path *path) 950b5fb63c1SChristoph Hellwig { 9514b99d499SAleksa Sarai int error = -ELOOP; 9526e77137bSAl Viro struct nameidata *nd = current->nameidata; 953b5fb63c1SChristoph Hellwig 9544b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9554b99d499SAleksa Sarai goto err; 9564b99d499SAleksa Sarai 95772ba2929SAleksa Sarai error = -EXDEV; 95872ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 95972ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 96072ba2929SAleksa Sarai goto err; 96172ba2929SAleksa Sarai } 962adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 963adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 964adb21d2bSAleksa Sarai goto err; 96572ba2929SAleksa Sarai 9664b99d499SAleksa Sarai path_put(&nd->path); 967b5fb63c1SChristoph Hellwig nd->path = *path; 968b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 969b5fb63c1SChristoph Hellwig nd->flags |= LOOKUP_JUMPED; 9701bc82070SAleksa Sarai return 0; 9714b99d499SAleksa Sarai 9724b99d499SAleksa Sarai err: 9734b99d499SAleksa Sarai path_put(path); 9744b99d499SAleksa Sarai return error; 975b5fb63c1SChristoph Hellwig } 976b5fb63c1SChristoph Hellwig 977b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 978574197e0SAl Viro { 97921c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 980fceef393SAl Viro do_delayed_call(&last->done); 9816548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 982b9ff4429SAl Viro path_put(&last->link); 983574197e0SAl Viro } 984574197e0SAl Viro 985561ec64aSLinus Torvalds int sysctl_protected_symlinks __read_mostly = 0; 986561ec64aSLinus Torvalds int sysctl_protected_hardlinks __read_mostly = 0; 98730aba665SSalvatore Mesoraca int sysctl_protected_fifos __read_mostly; 98830aba665SSalvatore Mesoraca int sysctl_protected_regular __read_mostly; 989800179c9SKees Cook 990800179c9SKees Cook /** 991800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 99255852635SRandy Dunlap * @nd: nameidata pathwalk data 993800179c9SKees Cook * 994800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 995800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 996800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 997800179c9SKees Cook * processes from failing races against path names that may change out 998800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 999800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1000800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1001800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1002800179c9SKees Cook * 1003800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1004800179c9SKees Cook */ 1005ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1006800179c9SKees Cook { 1007ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1008ba73d987SChristian Brauner kuid_t i_uid; 1009ba73d987SChristian Brauner 1010800179c9SKees Cook if (!sysctl_protected_symlinks) 1011800179c9SKees Cook return 0; 1012800179c9SKees Cook 1013ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1014ba73d987SChristian Brauner i_uid = i_uid_into_mnt(mnt_userns, inode); 1015800179c9SKees Cook /* Allowed if owner and follower match. */ 1016ba73d987SChristian Brauner if (uid_eq(current_cred()->fsuid, i_uid)) 1017800179c9SKees Cook return 0; 1018800179c9SKees Cook 1019800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 10200f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1021800179c9SKees Cook return 0; 1022800179c9SKees Cook 1023800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1024ba73d987SChristian Brauner if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid)) 1025800179c9SKees Cook return 0; 1026800179c9SKees Cook 102731956502SAl Viro if (nd->flags & LOOKUP_RCU) 102831956502SAl Viro return -ECHILD; 102931956502SAl Viro 1030ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1031245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1032800179c9SKees Cook return -EACCES; 1033800179c9SKees Cook } 1034800179c9SKees Cook 1035800179c9SKees Cook /** 1036800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 1037ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1038800179c9SKees Cook * @inode: the source inode to hardlink from 1039800179c9SKees Cook * 1040800179c9SKees Cook * Return false if at least one of the following conditions: 1041800179c9SKees Cook * - inode is not a regular file 1042800179c9SKees Cook * - inode is setuid 1043800179c9SKees Cook * - inode is setgid and group-exec 1044800179c9SKees Cook * - access failure for read and write 1045800179c9SKees Cook * 1046800179c9SKees Cook * Otherwise returns true. 1047800179c9SKees Cook */ 1048ba73d987SChristian Brauner static bool safe_hardlink_source(struct user_namespace *mnt_userns, 1049ba73d987SChristian Brauner struct inode *inode) 1050800179c9SKees Cook { 1051800179c9SKees Cook umode_t mode = inode->i_mode; 1052800179c9SKees Cook 1053800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1054800179c9SKees Cook if (!S_ISREG(mode)) 1055800179c9SKees Cook return false; 1056800179c9SKees Cook 1057800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1058800179c9SKees Cook if (mode & S_ISUID) 1059800179c9SKees Cook return false; 1060800179c9SKees Cook 1061800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1062800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1063800179c9SKees Cook return false; 1064800179c9SKees Cook 1065800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 1066ba73d987SChristian Brauner if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE)) 1067800179c9SKees Cook return false; 1068800179c9SKees Cook 1069800179c9SKees Cook return true; 1070800179c9SKees Cook } 1071800179c9SKees Cook 1072800179c9SKees Cook /** 1073800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 1074ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1075800179c9SKees Cook * @link: the source to hardlink from 1076800179c9SKees Cook * 1077800179c9SKees Cook * Block hardlink when all of: 1078800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1079800179c9SKees Cook * - fsuid does not match inode 1080800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1081f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1082800179c9SKees Cook * 1083ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1084ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1085ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1086ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1087ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1088ba73d987SChristian Brauner * 1089800179c9SKees Cook * Returns 0 if successful, -ve on error. 1090800179c9SKees Cook */ 1091ba73d987SChristian Brauner int may_linkat(struct user_namespace *mnt_userns, struct path *link) 1092800179c9SKees Cook { 1093593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1094593d1ce8SEric W. Biederman 1095593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1096ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 1097ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 1098593d1ce8SEric W. Biederman return -EOVERFLOW; 1099800179c9SKees Cook 1100800179c9SKees Cook if (!sysctl_protected_hardlinks) 1101800179c9SKees Cook return 0; 1102800179c9SKees Cook 1103800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1104800179c9SKees Cook * otherwise, it must be a safe source. 1105800179c9SKees Cook */ 1106ba73d987SChristian Brauner if (safe_hardlink_source(mnt_userns, inode) || 1107ba73d987SChristian Brauner inode_owner_or_capable(mnt_userns, inode)) 1108800179c9SKees Cook return 0; 1109800179c9SKees Cook 1110245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1111800179c9SKees Cook return -EPERM; 1112800179c9SKees Cook } 1113800179c9SKees Cook 111430aba665SSalvatore Mesoraca /** 111530aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 111630aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 111730aba665SSalvatore Mesoraca * exist. 1118ba73d987SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 1119d0cb5018SAl Viro * @dir_mode: mode bits of directory 1120d0cb5018SAl Viro * @dir_uid: owner of directory 112130aba665SSalvatore Mesoraca * @inode: the inode of the file to open 112230aba665SSalvatore Mesoraca * 112330aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 112430aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 112530aba665SSalvatore Mesoraca * - the file already exists 112630aba665SSalvatore Mesoraca * - we are in a sticky directory 112730aba665SSalvatore Mesoraca * - we don't own the file 112830aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 112930aba665SSalvatore Mesoraca * - the directory is world writable 113030aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 113130aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 113230aba665SSalvatore Mesoraca * be enough. 113330aba665SSalvatore Mesoraca * 1134ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1135ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1136ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1137ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1138ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1139ba73d987SChristian Brauner * 114030aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 114130aba665SSalvatore Mesoraca */ 1142ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 1143ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 114430aba665SSalvatore Mesoraca { 1145ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1146ba73d987SChristian Brauner kuid_t dir_uid = nd->dir_uid; 1147ba73d987SChristian Brauner 114830aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 114930aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1150d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1151ba73d987SChristian Brauner uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) || 1152ba73d987SChristian Brauner uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode))) 115330aba665SSalvatore Mesoraca return 0; 115430aba665SSalvatore Mesoraca 1155d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1156d0cb5018SAl Viro (dir_mode & 0020 && 115730aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 115830aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1159245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1160245d7369SKees Cook "sticky_create_fifo" : 1161245d7369SKees Cook "sticky_create_regular"; 1162245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 116330aba665SSalvatore Mesoraca return -EACCES; 116430aba665SSalvatore Mesoraca } 116530aba665SSalvatore Mesoraca return 0; 116630aba665SSalvatore Mesoraca } 116730aba665SSalvatore Mesoraca 1168f015f126SDavid Howells /* 1169f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1170f015f126SDavid Howells * 1171f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1172f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1173f015f126SDavid Howells * Up is towards /. 1174f015f126SDavid Howells * 1175f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1176f015f126SDavid Howells * root. 1177f015f126SDavid Howells */ 1178bab77ebfSAl Viro int follow_up(struct path *path) 11791da177e4SLinus Torvalds { 11800714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 11810714a533SAl Viro struct mount *parent; 11821da177e4SLinus Torvalds struct dentry *mountpoint; 118399b7db7bSNick Piggin 118448a066e7SAl Viro read_seqlock_excl(&mount_lock); 11850714a533SAl Viro parent = mnt->mnt_parent; 11863c0a6163SAl Viro if (parent == mnt) { 118748a066e7SAl Viro read_sequnlock_excl(&mount_lock); 11881da177e4SLinus Torvalds return 0; 11891da177e4SLinus Torvalds } 11900714a533SAl Viro mntget(&parent->mnt); 1191a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 119248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1193bab77ebfSAl Viro dput(path->dentry); 1194bab77ebfSAl Viro path->dentry = mountpoint; 1195bab77ebfSAl Viro mntput(path->mnt); 11960714a533SAl Viro path->mnt = &parent->mnt; 11971da177e4SLinus Torvalds return 1; 11981da177e4SLinus Torvalds } 11994d359507SAl Viro EXPORT_SYMBOL(follow_up); 12001da177e4SLinus Torvalds 12017ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12027ef482faSAl Viro struct path *path, unsigned *seqp) 12037ef482faSAl Viro { 12047ef482faSAl Viro while (mnt_has_parent(m)) { 12057ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12067ef482faSAl Viro 12077ef482faSAl Viro m = m->mnt_parent; 12087ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 12097ef482faSAl Viro root->mnt == &m->mnt)) 12107ef482faSAl Viro break; 12117ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 12127ef482faSAl Viro path->mnt = &m->mnt; 12137ef482faSAl Viro path->dentry = mountpoint; 12147ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 12157ef482faSAl Viro return true; 12167ef482faSAl Viro } 12177ef482faSAl Viro } 12187ef482faSAl Viro return false; 12197ef482faSAl Viro } 12207ef482faSAl Viro 12212aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 12222aa38470SAl Viro struct path *path) 12232aa38470SAl Viro { 12242aa38470SAl Viro bool found; 12252aa38470SAl Viro 12262aa38470SAl Viro rcu_read_lock(); 12272aa38470SAl Viro while (1) { 12282aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 12292aa38470SAl Viro 12302aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 12312aa38470SAl Viro if (unlikely(!found)) { 12322aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 12332aa38470SAl Viro break; 12342aa38470SAl Viro } else { 12352aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 12362aa38470SAl Viro break; 12372aa38470SAl Viro rcu_read_unlock(); 12382aa38470SAl Viro path_put(path); 12392aa38470SAl Viro rcu_read_lock(); 12402aa38470SAl Viro } 12412aa38470SAl Viro } 12422aa38470SAl Viro rcu_read_unlock(); 12432aa38470SAl Viro return found; 12442aa38470SAl Viro } 12452aa38470SAl Viro 1246b5c84bf6SNick Piggin /* 12479875cf80SDavid Howells * Perform an automount 12489875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 12499875cf80SDavid Howells * were called with. 12501da177e4SLinus Torvalds */ 12511c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 125231e6b01fSNick Piggin { 125325e195aaSAl Viro struct dentry *dentry = path->dentry; 12549875cf80SDavid Howells 12550ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 12560ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 12570ec26fd0SMiklos Szeredi * the name. 12580ec26fd0SMiklos Szeredi * 12590ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 12605a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 12610ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 12620ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 12630ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 12640ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 12655a30d8a2SDavid Howells */ 12661c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 12675d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 126825e195aaSAl Viro dentry->d_inode) 12699875cf80SDavid Howells return -EISDIR; 12700ec26fd0SMiklos Szeredi 12711c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 12729875cf80SDavid Howells return -ELOOP; 12739875cf80SDavid Howells 127425e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1275ea5b778aSDavid Howells } 12769875cf80SDavid Howells 12779875cf80SDavid Howells /* 12789deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 12799deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 12809deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 12819deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 12829875cf80SDavid Howells */ 12839deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 12849deed3ebSAl Viro int *count, unsigned lookup_flags) 12859875cf80SDavid Howells { 12869deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 12879875cf80SDavid Howells bool need_mntput = false; 12888aef1884SAl Viro int ret = 0; 12899875cf80SDavid Howells 12909deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1291cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1292cc53ce53SDavid Howells * being held. */ 1293d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1294fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1295508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1296cc53ce53SDavid Howells if (ret < 0) 12978aef1884SAl Viro break; 1298cc53ce53SDavid Howells } 1299cc53ce53SDavid Howells 13009deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13019875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13029deed3ebSAl Viro if (mounted) { // ... in our namespace 13039875cf80SDavid Howells dput(path->dentry); 13049875cf80SDavid Howells if (need_mntput) 1305463ffb2eSAl Viro mntput(path->mnt); 1306463ffb2eSAl Viro path->mnt = mounted; 1307463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 13089deed3ebSAl Viro // here we know it's positive 13099deed3ebSAl Viro flags = path->dentry->d_flags; 13109875cf80SDavid Howells need_mntput = true; 13119875cf80SDavid Howells continue; 1312463ffb2eSAl Viro } 13131da177e4SLinus Torvalds } 13149875cf80SDavid Howells 13159deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 13169deed3ebSAl Viro break; 13179deed3ebSAl Viro 13189deed3ebSAl Viro // uncovered automount point 13199deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 13209deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 13219875cf80SDavid Howells if (ret < 0) 13228aef1884SAl Viro break; 13239875cf80SDavid Howells } 13249875cf80SDavid Howells 13259deed3ebSAl Viro if (ret == -EISDIR) 13269deed3ebSAl Viro ret = 0; 13279deed3ebSAl Viro // possible if you race with several mount --move 13289deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 13298aef1884SAl Viro mntput(path->mnt); 13309deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1331d41efb52SAl Viro ret = -ENOENT; 13329deed3ebSAl Viro *jumped = need_mntput; 13338402752eSAl Viro return ret; 13341da177e4SLinus Torvalds } 13351da177e4SLinus Torvalds 13369deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 13379deed3ebSAl Viro int *count, unsigned lookup_flags) 13389deed3ebSAl Viro { 13399deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 13409deed3ebSAl Viro 13419deed3ebSAl Viro /* fastpath */ 13429deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 13439deed3ebSAl Viro *jumped = false; 13449deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 13459deed3ebSAl Viro return -ENOENT; 13469deed3ebSAl Viro return 0; 13479deed3ebSAl Viro } 13489deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 13499deed3ebSAl Viro } 13509deed3ebSAl Viro 1351cc53ce53SDavid Howells int follow_down_one(struct path *path) 13521da177e4SLinus Torvalds { 13531da177e4SLinus Torvalds struct vfsmount *mounted; 13541da177e4SLinus Torvalds 13551c755af4SAl Viro mounted = lookup_mnt(path); 13561da177e4SLinus Torvalds if (mounted) { 13579393bd07SAl Viro dput(path->dentry); 13589393bd07SAl Viro mntput(path->mnt); 13599393bd07SAl Viro path->mnt = mounted; 13609393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 13611da177e4SLinus Torvalds return 1; 13621da177e4SLinus Torvalds } 13631da177e4SLinus Torvalds return 0; 13641da177e4SLinus Torvalds } 13654d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 13661da177e4SLinus Torvalds 13679875cf80SDavid Howells /* 13689deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 13699deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 13709deed3ebSAl Viro * caller is permitted to proceed or not. 13719deed3ebSAl Viro */ 13729deed3ebSAl Viro int follow_down(struct path *path) 13739deed3ebSAl Viro { 13749deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13759deed3ebSAl Viro bool jumped; 13769deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 13779deed3ebSAl Viro 13789deed3ebSAl Viro if (path->mnt != mnt) 13799deed3ebSAl Viro mntput(mnt); 13809deed3ebSAl Viro return ret; 13819deed3ebSAl Viro } 13829deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 13839deed3ebSAl Viro 13849deed3ebSAl Viro /* 1385287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1386287548e4SAl Viro * we meet a managed dentry that would need blocking. 13879875cf80SDavid Howells */ 13889875cf80SDavid Howells static bool __follow_mount_rcu(struct nameidata *nd, struct path *path, 1389254cf582SAl Viro struct inode **inode, unsigned *seqp) 13909875cf80SDavid Howells { 1391ea936aebSAl Viro struct dentry *dentry = path->dentry; 1392ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1393ea936aebSAl Viro 1394ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1395ea936aebSAl Viro return true; 1396ea936aebSAl Viro 1397ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1398ea936aebSAl Viro return false; 1399ea936aebSAl Viro 140062a7375eSIan Kent for (;;) { 140162a7375eSIan Kent /* 140262a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 140362a7375eSIan Kent * that wants to block transit. 140462a7375eSIan Kent */ 1405ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1406ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1407ea936aebSAl Viro if (res) 1408ea936aebSAl Viro return res == -EISDIR; 1409ea936aebSAl Viro flags = dentry->d_flags; 1410b8faf035SNeilBrown } 141162a7375eSIan Kent 1412ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1413ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1414ea936aebSAl Viro if (mounted) { 1415c7105365SAl Viro path->mnt = &mounted->mnt; 1416ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1417a3fbbde7SAl Viro nd->flags |= LOOKUP_JUMPED; 1418ea936aebSAl Viro *seqp = read_seqcount_begin(&dentry->d_seq); 1419ea936aebSAl Viro *inode = dentry->d_inode; 142059430262SLinus Torvalds /* 1421ea936aebSAl Viro * We don't need to re-check ->d_seq after this 1422ea936aebSAl Viro * ->d_inode read - there will be an RCU delay 1423ea936aebSAl Viro * between mount hash removal and ->mnt_root 1424ea936aebSAl Viro * becoming unpinned. 142559430262SLinus Torvalds */ 1426ea936aebSAl Viro flags = dentry->d_flags; 1427ea936aebSAl Viro continue; 14289875cf80SDavid Howells } 1429ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1430ea936aebSAl Viro return false; 1431ea936aebSAl Viro } 1432ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1433ea936aebSAl Viro } 1434287548e4SAl Viro } 1435287548e4SAl Viro 1436db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 1437db3c9adeSAl Viro struct path *path, struct inode **inode, 1438db3c9adeSAl Viro unsigned int *seqp) 1439bd7c4b50SAl Viro { 14409deed3ebSAl Viro bool jumped; 1441db3c9adeSAl Viro int ret; 1442bd7c4b50SAl Viro 1443db3c9adeSAl Viro path->mnt = nd->path.mnt; 1444db3c9adeSAl Viro path->dentry = dentry; 1445c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 1446c153007bSAl Viro unsigned int seq = *seqp; 1447c153007bSAl Viro if (unlikely(!*inode)) 1448c153007bSAl Viro return -ENOENT; 1449c153007bSAl Viro if (likely(__follow_mount_rcu(nd, path, inode, seqp))) 14509deed3ebSAl Viro return 0; 1451c153007bSAl Viro if (unlazy_child(nd, dentry, seq)) 1452c153007bSAl Viro return -ECHILD; 1453c153007bSAl Viro // *path might've been clobbered by __follow_mount_rcu() 1454c153007bSAl Viro path->mnt = nd->path.mnt; 1455c153007bSAl Viro path->dentry = dentry; 1456c153007bSAl Viro } 14579deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 14589deed3ebSAl Viro if (jumped) { 14599deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 14609deed3ebSAl Viro ret = -EXDEV; 14619deed3ebSAl Viro else 14629deed3ebSAl Viro nd->flags |= LOOKUP_JUMPED; 14639deed3ebSAl Viro } 14649deed3ebSAl Viro if (unlikely(ret)) { 14659deed3ebSAl Viro dput(path->dentry); 14669deed3ebSAl Viro if (path->mnt != nd->path.mnt) 14679deed3ebSAl Viro mntput(path->mnt); 14689deed3ebSAl Viro } else { 1469bd7c4b50SAl Viro *inode = d_backing_inode(path->dentry); 1470bd7c4b50SAl Viro *seqp = 0; /* out of RCU mode, so the value doesn't matter */ 1471bd7c4b50SAl Viro } 1472bd7c4b50SAl Viro return ret; 1473bd7c4b50SAl Viro } 1474bd7c4b50SAl Viro 14759875cf80SDavid Howells /* 1476f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1477f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1478baa03890SNick Piggin */ 1479e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1480e3c13928SAl Viro struct dentry *dir, 14816c51e513SAl Viro unsigned int flags) 1482baa03890SNick Piggin { 1483a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1484bad61189SMiklos Szeredi if (dentry) { 1485a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1486bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 148774ff0ffcSAl Viro if (!error) 14885542aa2fSEric W. Biederman d_invalidate(dentry); 1489bad61189SMiklos Szeredi dput(dentry); 149074ff0ffcSAl Viro return ERR_PTR(error); 1491bad61189SMiklos Szeredi } 1492bad61189SMiklos Szeredi } 1493baa03890SNick Piggin return dentry; 1494baa03890SNick Piggin } 1495baa03890SNick Piggin 1496baa03890SNick Piggin /* 1497a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1498a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1499a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1500a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1501a03ece5fSAl Viro * at all. 150244396f4bSJosef Bacik */ 1503a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1504a03ece5fSAl Viro struct dentry *base, unsigned int flags) 150544396f4bSJosef Bacik { 1506a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 150744396f4bSJosef Bacik struct dentry *old; 1508a03ece5fSAl Viro struct inode *dir = base->d_inode; 1509a03ece5fSAl Viro 1510a03ece5fSAl Viro if (dentry) 1511a03ece5fSAl Viro return dentry; 151244396f4bSJosef Bacik 151344396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1514a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 151544396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1516a03ece5fSAl Viro 1517a03ece5fSAl Viro dentry = d_alloc(base, name); 1518a03ece5fSAl Viro if (unlikely(!dentry)) 1519a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 152044396f4bSJosef Bacik 152172bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 152244396f4bSJosef Bacik if (unlikely(old)) { 152344396f4bSJosef Bacik dput(dentry); 152444396f4bSJosef Bacik dentry = old; 152544396f4bSJosef Bacik } 152644396f4bSJosef Bacik return dentry; 152744396f4bSJosef Bacik } 152844396f4bSJosef Bacik 152920e34357SAl Viro static struct dentry *lookup_fast(struct nameidata *nd, 153020e34357SAl Viro struct inode **inode, 1531254cf582SAl Viro unsigned *seqp) 15321da177e4SLinus Torvalds { 153331e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 15345a18fff2SAl Viro int status = 1; 15359875cf80SDavid Howells 15363cac260aSAl Viro /* 1537b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 15385d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 15395d0f49c1SAl Viro * going to fall back to non-racy lookup. 1540b04f784eSNick Piggin */ 154131e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 154231e6b01fSNick Piggin unsigned seq; 1543da53be12SLinus Torvalds dentry = __d_lookup_rcu(parent, &nd->last, &seq); 15445d0f49c1SAl Viro if (unlikely(!dentry)) { 15454675ac39SAl Viro if (unlazy_walk(nd)) 154620e34357SAl Viro return ERR_PTR(-ECHILD); 154720e34357SAl Viro return NULL; 15485d0f49c1SAl Viro } 15495a18fff2SAl Viro 155012f8ad4bSLinus Torvalds /* 155112f8ad4bSLinus Torvalds * This sequence count validates that the inode matches 155212f8ad4bSLinus Torvalds * the dentry name information from lookup. 155312f8ad4bSLinus Torvalds */ 155463afdfc7SDavid Howells *inode = d_backing_inode(dentry); 15555d0f49c1SAl Viro if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) 155620e34357SAl Viro return ERR_PTR(-ECHILD); 155712f8ad4bSLinus Torvalds 155812f8ad4bSLinus Torvalds /* 155912f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 156012f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 156112f8ad4bSLinus Torvalds * 156212f8ad4bSLinus Torvalds * The memory barrier in read_seqcount_begin of child is 156312f8ad4bSLinus Torvalds * enough, we can use __read_seqcount_retry here. 156412f8ad4bSLinus Torvalds */ 15655d0f49c1SAl Viro if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq))) 156620e34357SAl Viro return ERR_PTR(-ECHILD); 15675a18fff2SAl Viro 1568254cf582SAl Viro *seqp = seq; 15694ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1570c153007bSAl Viro if (likely(status > 0)) 157120e34357SAl Viro return dentry; 15724675ac39SAl Viro if (unlazy_child(nd, dentry, seq)) 157320e34357SAl Viro return ERR_PTR(-ECHILD); 1574209a7fb2SAl Viro if (unlikely(status == -ECHILD)) 1575209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1576209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 15775a18fff2SAl Viro } else { 1578e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 157981e6f520SAl Viro if (unlikely(!dentry)) 158020e34357SAl Viro return NULL; 15814ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 15825d0f49c1SAl Viro } 15835a18fff2SAl Viro if (unlikely(status <= 0)) { 1584e9742b53SAl Viro if (!status) 15855d0f49c1SAl Viro d_invalidate(dentry); 15865a18fff2SAl Viro dput(dentry); 158720e34357SAl Viro return ERR_PTR(status); 15885a18fff2SAl Viro } 158920e34357SAl Viro return dentry; 1590697f514dSMiklos Szeredi } 1591697f514dSMiklos Szeredi 1592697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 159388d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1594e3c13928SAl Viro struct dentry *dir, 1595e3c13928SAl Viro unsigned int flags) 1596697f514dSMiklos Szeredi { 159788d8331aSAl Viro struct dentry *dentry, *old; 15981936386eSAl Viro struct inode *inode = dir->d_inode; 1599d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16001936386eSAl Viro 16011936386eSAl Viro /* Don't go there if it's already dead */ 160294bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 160388d8331aSAl Viro return ERR_PTR(-ENOENT); 160494bdd655SAl Viro again: 1605d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 160694bdd655SAl Viro if (IS_ERR(dentry)) 160788d8331aSAl Viro return dentry; 160894bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1609949a852eSAl Viro int error = d_revalidate(dentry, flags); 1610949a852eSAl Viro if (unlikely(error <= 0)) { 161194bdd655SAl Viro if (!error) { 1612949a852eSAl Viro d_invalidate(dentry); 1613949a852eSAl Viro dput(dentry); 161494bdd655SAl Viro goto again; 161594bdd655SAl Viro } 161694bdd655SAl Viro dput(dentry); 1617949a852eSAl Viro dentry = ERR_PTR(error); 1618949a852eSAl Viro } 161994bdd655SAl Viro } else { 16201936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 162185c7f810SAl Viro d_lookup_done(dentry); 16221936386eSAl Viro if (unlikely(old)) { 16231936386eSAl Viro dput(dentry); 16241936386eSAl Viro dentry = old; 1625949a852eSAl Viro } 1626949a852eSAl Viro } 1627e3c13928SAl Viro return dentry; 16281da177e4SLinus Torvalds } 16291da177e4SLinus Torvalds 163088d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 163188d8331aSAl Viro struct dentry *dir, 163288d8331aSAl Viro unsigned int flags) 163388d8331aSAl Viro { 163488d8331aSAl Viro struct inode *inode = dir->d_inode; 163588d8331aSAl Viro struct dentry *res; 163688d8331aSAl Viro inode_lock_shared(inode); 163788d8331aSAl Viro res = __lookup_slow(name, dir, flags); 163888d8331aSAl Viro inode_unlock_shared(inode); 163988d8331aSAl Viro return res; 164088d8331aSAl Viro } 164188d8331aSAl Viro 1642ba73d987SChristian Brauner static inline int may_lookup(struct user_namespace *mnt_userns, 1643ba73d987SChristian Brauner struct nameidata *nd) 164452094c8aSAl Viro { 164552094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 1646ba73d987SChristian Brauner int err = inode_permission(mnt_userns, nd->inode, 164747291baaSChristian Brauner MAY_EXEC | MAY_NOT_BLOCK); 164852094c8aSAl Viro if (err != -ECHILD) 164952094c8aSAl Viro return err; 16504675ac39SAl Viro if (unlazy_walk(nd)) 165152094c8aSAl Viro return -ECHILD; 165252094c8aSAl Viro } 1653ba73d987SChristian Brauner return inode_permission(mnt_userns, nd->inode, MAY_EXEC); 165452094c8aSAl Viro } 165552094c8aSAl Viro 165649055906SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq) 1657d63ff28fSAl Viro { 165849055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 165949055906SAl Viro return -ELOOP; 16604542576bSAl Viro 16614542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 16624542576bSAl Viro return 0; 16634542576bSAl Viro if (likely(nd->stack != nd->internal)) 16644542576bSAl Viro return 0; 166560ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 166649055906SAl Viro return 0; 166760ef60c7SAl Viro 166860ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 166960ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 167060ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 1671aef9404dSAl Viro bool grabbed_link = legitimize_path(nd, link, seq); 167260ef60c7SAl Viro 167360ef60c7SAl Viro if (unlazy_walk(nd) != 0 || !grabbed_link) 167460ef60c7SAl Viro return -ECHILD; 167560ef60c7SAl Viro 167660ef60c7SAl Viro if (nd_alloc_stack(nd)) 167760ef60c7SAl Viro return 0; 1678bc40aee0SAl Viro } 167960ef60c7SAl Viro return -ENOMEM; 168049055906SAl Viro } 168149055906SAl Viro 168249055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 168349055906SAl Viro 168449055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 168549055906SAl Viro struct inode *inode, unsigned seq, int flags) 168649055906SAl Viro { 168749055906SAl Viro struct saved *last; 168849055906SAl Viro const char *res; 168949055906SAl Viro int error = reserve_stack(nd, link, seq); 169049055906SAl Viro 169149055906SAl Viro if (unlikely(error)) { 169249055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1693cd179f44SAl Viro path_put(link); 169406708adbSAl Viro return ERR_PTR(error); 1695626de996SAl Viro } 1696ab104923SAl Viro last = nd->stack + nd->depth++; 16971cf2665bSAl Viro last->link = *link; 1698fceef393SAl Viro clear_delayed_call(&last->done); 16990450b2d1SAl Viro last->seq = seq; 1700ad6cc4c3SAl Viro 1701b1a81972SAl Viro if (flags & WALK_TRAILING) { 1702ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1703ad6cc4c3SAl Viro if (unlikely(error)) 1704ad6cc4c3SAl Viro return ERR_PTR(error); 1705ad6cc4c3SAl Viro } 1706ad6cc4c3SAl Viro 1707dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1708dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1709ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1710ad6cc4c3SAl Viro 1711ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1712ad6cc4c3SAl Viro touch_atime(&last->link); 1713ad6cc4c3SAl Viro cond_resched(); 1714ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1715ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1716ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1717ad6cc4c3SAl Viro touch_atime(&last->link); 1718ad6cc4c3SAl Viro } 1719ad6cc4c3SAl Viro 1720ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1721ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1722ad6cc4c3SAl Viro if (unlikely(error)) 1723ad6cc4c3SAl Viro return ERR_PTR(error); 1724ad6cc4c3SAl Viro 1725ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1726ad6cc4c3SAl Viro if (!res) { 1727ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1728ad6cc4c3SAl Viro struct delayed_call *); 1729ad6cc4c3SAl Viro get = inode->i_op->get_link; 1730ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1731ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1732ad6cc4c3SAl Viro if (res == ERR_PTR(-ECHILD)) { 1733ad6cc4c3SAl Viro if (unlikely(unlazy_walk(nd))) 1734ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1735ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1736ad6cc4c3SAl Viro } 1737ad6cc4c3SAl Viro } else { 1738ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1739ad6cc4c3SAl Viro } 1740ad6cc4c3SAl Viro if (!res) 1741ad6cc4c3SAl Viro goto all_done; 1742ad6cc4c3SAl Viro if (IS_ERR(res)) 1743ad6cc4c3SAl Viro return res; 1744ad6cc4c3SAl Viro } 1745ad6cc4c3SAl Viro if (*res == '/') { 1746ad6cc4c3SAl Viro error = nd_jump_root(nd); 1747ad6cc4c3SAl Viro if (unlikely(error)) 1748ad6cc4c3SAl Viro return ERR_PTR(error); 1749ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1750ad6cc4c3SAl Viro ; 1751ad6cc4c3SAl Viro } 1752ad6cc4c3SAl Viro if (*res) 1753ad6cc4c3SAl Viro return res; 1754ad6cc4c3SAl Viro all_done: // pure jump 1755ad6cc4c3SAl Viro put_link(nd); 1756ad6cc4c3SAl Viro return NULL; 1757d63ff28fSAl Viro } 1758d63ff28fSAl Viro 17593ddcd056SLinus Torvalds /* 17603ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 17613ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 17623ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 17633ddcd056SLinus Torvalds * for the common case. 17643ddcd056SLinus Torvalds */ 1765b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1766cbae4d12SAl Viro struct dentry *dentry, struct inode *inode, unsigned seq) 17673ddcd056SLinus Torvalds { 1768cbae4d12SAl Viro struct path path; 1769cbae4d12SAl Viro int err = handle_mounts(nd, dentry, &path, &inode, &seq); 1770cbae4d12SAl Viro 1771cbae4d12SAl Viro if (err < 0) 1772b0417d2cSAl Viro return ERR_PTR(err); 1773cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 17748c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1775aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 17768f64fb1cSAl Viro /* not a symlink or should not follow */ 1777c99687a0SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1778c99687a0SAl Viro dput(nd->path.dentry); 1779c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1780c99687a0SAl Viro mntput(nd->path.mnt); 1781c99687a0SAl Viro } 1782c99687a0SAl Viro nd->path = path; 17838f64fb1cSAl Viro nd->inode = inode; 17848f64fb1cSAl Viro nd->seq = seq; 1785b0417d2cSAl Viro return NULL; 17868f64fb1cSAl Viro } 1787a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 178884f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 1789cbae4d12SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, seq)) 1790b0417d2cSAl Viro return ERR_PTR(-ECHILD); 179184f0cd9eSAl Viro } else { 179284f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 179384f0cd9eSAl Viro mntget(path.mnt); 1794a7f77542SAl Viro } 1795b1a81972SAl Viro return pick_link(nd, &path, inode, seq, flags); 17963ddcd056SLinus Torvalds } 17973ddcd056SLinus Torvalds 1798c2df1968SAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd, 1799c2df1968SAl Viro struct inode **inodep, 1800c2df1968SAl Viro unsigned *seqp) 1801957dd41dSAl Viro { 180212487f30SAl Viro struct dentry *parent, *old; 1803957dd41dSAl Viro 180412487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 180512487f30SAl Viro goto in_root; 180612487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18077ef482faSAl Viro struct path path; 1808efe772d6SAl Viro unsigned seq; 18097ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18107ef482faSAl Viro &nd->root, &path, &seq)) 181112487f30SAl Viro goto in_root; 1812efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1813efe772d6SAl Viro return ERR_PTR(-ECHILD); 1814efe772d6SAl Viro nd->path = path; 1815efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1816efe772d6SAl Viro nd->seq = seq; 1817efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1818efe772d6SAl Viro return ERR_PTR(-ECHILD); 1819efe772d6SAl Viro /* we know that mountpoint was pinned */ 1820957dd41dSAl Viro } 182112487f30SAl Viro old = nd->path.dentry; 182212487f30SAl Viro parent = old->d_parent; 182312487f30SAl Viro *inodep = parent->d_inode; 182412487f30SAl Viro *seqp = read_seqcount_begin(&parent->d_seq); 182512487f30SAl Viro if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq))) 182612487f30SAl Viro return ERR_PTR(-ECHILD); 182712487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 182812487f30SAl Viro return ERR_PTR(-ECHILD); 182912487f30SAl Viro return parent; 183012487f30SAl Viro in_root: 1831efe772d6SAl Viro if (unlikely(read_seqretry(&mount_lock, nd->m_seq))) 1832efe772d6SAl Viro return ERR_PTR(-ECHILD); 1833957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 18347521f22bSAl Viro return ERR_PTR(-ECHILD); 1835c2df1968SAl Viro return NULL; 1836957dd41dSAl Viro } 1837957dd41dSAl Viro 1838c2df1968SAl Viro static struct dentry *follow_dotdot(struct nameidata *nd, 1839c2df1968SAl Viro struct inode **inodep, 1840c2df1968SAl Viro unsigned *seqp) 1841957dd41dSAl Viro { 184212487f30SAl Viro struct dentry *parent; 184312487f30SAl Viro 1844957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 184512487f30SAl Viro goto in_root; 184612487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18472aa38470SAl Viro struct path path; 18482aa38470SAl Viro 18492aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 18502aa38470SAl Viro &nd->root, &path)) 185112487f30SAl Viro goto in_root; 1852165200d6SAl Viro path_put(&nd->path); 1853165200d6SAl Viro nd->path = path; 18542aa38470SAl Viro nd->inode = path.dentry->d_inode; 1855165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1856165200d6SAl Viro return ERR_PTR(-EXDEV); 185712487f30SAl Viro } 1858957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 185912487f30SAl Viro parent = dget_parent(nd->path.dentry); 1860957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1861957dd41dSAl Viro dput(parent); 18627521f22bSAl Viro return ERR_PTR(-ENOENT); 1863957dd41dSAl Viro } 1864c2df1968SAl Viro *seqp = 0; 1865c2df1968SAl Viro *inodep = parent->d_inode; 1866c2df1968SAl Viro return parent; 186712487f30SAl Viro 186812487f30SAl Viro in_root: 1869957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 18707521f22bSAl Viro return ERR_PTR(-EXDEV); 1871c2df1968SAl Viro dget(nd->path.dentry); 1872c2df1968SAl Viro return NULL; 1873957dd41dSAl Viro } 1874957dd41dSAl Viro 18757521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1876957dd41dSAl Viro { 1877957dd41dSAl Viro if (type == LAST_DOTDOT) { 18787521f22bSAl Viro const char *error = NULL; 1879c2df1968SAl Viro struct dentry *parent; 1880c2df1968SAl Viro struct inode *inode; 1881c2df1968SAl Viro unsigned seq; 1882957dd41dSAl Viro 1883957dd41dSAl Viro if (!nd->root.mnt) { 18847521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1885957dd41dSAl Viro if (error) 1886957dd41dSAl Viro return error; 1887957dd41dSAl Viro } 1888957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1889c2df1968SAl Viro parent = follow_dotdot_rcu(nd, &inode, &seq); 1890957dd41dSAl Viro else 1891c2df1968SAl Viro parent = follow_dotdot(nd, &inode, &seq); 1892c2df1968SAl Viro if (IS_ERR(parent)) 1893c2df1968SAl Viro return ERR_CAST(parent); 1894c2df1968SAl Viro if (unlikely(!parent)) 1895c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1896c2df1968SAl Viro nd->path.dentry, nd->inode, nd->seq); 1897c2df1968SAl Viro else 1898c2df1968SAl Viro error = step_into(nd, WALK_NOFOLLOW, 1899c2df1968SAl Viro parent, inode, seq); 1900c2df1968SAl Viro if (unlikely(error)) 1901957dd41dSAl Viro return error; 1902957dd41dSAl Viro 1903957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1904957dd41dSAl Viro /* 1905957dd41dSAl Viro * If there was a racing rename or mount along our 1906957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1907957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1908957dd41dSAl Viro * some fallback). 1909957dd41dSAl Viro */ 1910957dd41dSAl Viro smp_rmb(); 1911957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq))) 19127521f22bSAl Viro return ERR_PTR(-EAGAIN); 1913957dd41dSAl Viro if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq))) 19147521f22bSAl Viro return ERR_PTR(-EAGAIN); 1915957dd41dSAl Viro } 1916957dd41dSAl Viro } 19177521f22bSAl Viro return NULL; 1918957dd41dSAl Viro } 1919957dd41dSAl Viro 192092d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1921ce57dfc1SAl Viro { 1922db3c9adeSAl Viro struct dentry *dentry; 1923ce57dfc1SAl Viro struct inode *inode; 1924254cf582SAl Viro unsigned seq; 1925ce57dfc1SAl Viro /* 1926ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1927ce57dfc1SAl Viro * to be able to know about the current root directory and 1928ce57dfc1SAl Viro * parent relationships. 1929ce57dfc1SAl Viro */ 19304693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 19311c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 19324693a547SAl Viro put_link(nd); 19337521f22bSAl Viro return handle_dots(nd, nd->last_type); 19344693a547SAl Viro } 193520e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 193620e34357SAl Viro if (IS_ERR(dentry)) 193792d27016SAl Viro return ERR_CAST(dentry); 193820e34357SAl Viro if (unlikely(!dentry)) { 1939db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1940db3c9adeSAl Viro if (IS_ERR(dentry)) 194192d27016SAl Viro return ERR_CAST(dentry); 194220e34357SAl Viro } 194356676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 194456676ec3SAl Viro put_link(nd); 1945b0417d2cSAl Viro return step_into(nd, flags, dentry, inode, seq); 1946ce57dfc1SAl Viro } 1947ce57dfc1SAl Viro 19481da177e4SLinus Torvalds /* 1949bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 1950bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 1951bfcfaa77SLinus Torvalds * 1952bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 1953bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 1954bfcfaa77SLinus Torvalds * fast. 1955bfcfaa77SLinus Torvalds * 1956bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 1957bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 1958bfcfaa77SLinus Torvalds * crossing operation. 1959bfcfaa77SLinus Torvalds * 1960bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 1961bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 1962bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 1963bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 1964bfcfaa77SLinus Torvalds */ 1965bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 1966bfcfaa77SLinus Torvalds 1967f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 1968bfcfaa77SLinus Torvalds 1969468a9428SGeorge Spelvin #ifdef HASH_MIX 1970bfcfaa77SLinus Torvalds 1971468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 1972468a9428SGeorge Spelvin 1973468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 19742a18da7aSGeorge Spelvin /* 19752a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 19762a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 19772a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 19782a18da7aSGeorge Spelvin * and no temporaries. 19792a18da7aSGeorge Spelvin * 19802a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 19812a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 19822a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 19832a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 19842a18da7aSGeorge Spelvin * 19852a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 19862a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 19872a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 19882a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 19892a18da7aSGeorge Spelvin * 19902a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 19912a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 19922a18da7aSGeorge Spelvin * toggling any given output bit. 19932a18da7aSGeorge Spelvin * 19942a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 19952a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 19962a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 19972a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 19982a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 19992a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20002a18da7aSGeorge Spelvin * Perfect: 8192 258048 20012a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20022a18da7aSGeorge Spelvin */ 20032a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20042a18da7aSGeorge Spelvin ( x ^= (a), \ 20052a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20062a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20072a18da7aSGeorge Spelvin y *= 9 ) 2008bfcfaa77SLinus Torvalds 20090fed3ac8SGeorge Spelvin /* 20102a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20112a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20122a18da7aSGeorge Spelvin * work done before the hash value is used. 20130fed3ac8SGeorge Spelvin */ 20142a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20150fed3ac8SGeorge Spelvin { 20162a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20172a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20182a18da7aSGeorge Spelvin return y >> 32; 20190fed3ac8SGeorge Spelvin } 20200fed3ac8SGeorge Spelvin 2021bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2022bfcfaa77SLinus Torvalds 20232a18da7aSGeorge Spelvin /* 20242a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 20252a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20262a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 20272a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 20282a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 20292a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 20302a18da7aSGeorge Spelvin * Perfect: 2048 31744 20312a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 20322a18da7aSGeorge Spelvin */ 20332a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20342a18da7aSGeorge Spelvin ( x ^= (a), \ 20352a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 20362a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 20372a18da7aSGeorge Spelvin y *= 9 ) 2038bfcfaa77SLinus Torvalds 20392a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20400fed3ac8SGeorge Spelvin { 20412a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 20422a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 20430fed3ac8SGeorge Spelvin } 20440fed3ac8SGeorge Spelvin 2045bfcfaa77SLinus Torvalds #endif 2046bfcfaa77SLinus Torvalds 20472a18da7aSGeorge Spelvin /* 20482a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 20492a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 20502a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 20512a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 20522a18da7aSGeorge Spelvin * finds the delimiter after the name. 20532a18da7aSGeorge Spelvin */ 20548387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2055bfcfaa77SLinus Torvalds { 20568387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2057bfcfaa77SLinus Torvalds 2058bfcfaa77SLinus Torvalds for (;;) { 2059fcfd2fbfSGeorge Spelvin if (!len) 2060fcfd2fbfSGeorge Spelvin goto done; 2061e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2062bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2063bfcfaa77SLinus Torvalds break; 20642a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2065bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2066bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2067bfcfaa77SLinus Torvalds } 20682a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2069bfcfaa77SLinus Torvalds done: 20702a18da7aSGeorge Spelvin return fold_hash(x, y); 2071bfcfaa77SLinus Torvalds } 2072bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2073bfcfaa77SLinus Torvalds 2074fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 20758387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2076fcfd2fbfSGeorge Spelvin { 20778387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 20788387ff25SLinus Torvalds unsigned long adata, mask, len; 2079fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2080fcfd2fbfSGeorge Spelvin 20818387ff25SLinus Torvalds len = 0; 20828387ff25SLinus Torvalds goto inside; 20838387ff25SLinus Torvalds 2084fcfd2fbfSGeorge Spelvin do { 20852a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2086fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 20878387ff25SLinus Torvalds inside: 2088fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2089fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2090fcfd2fbfSGeorge Spelvin 2091fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2092fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 20932a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2094fcfd2fbfSGeorge Spelvin 20952a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2096fcfd2fbfSGeorge Spelvin } 2097fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2098fcfd2fbfSGeorge Spelvin 2099bfcfaa77SLinus Torvalds /* 2100bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2101d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2102bfcfaa77SLinus Torvalds */ 21038387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2104bfcfaa77SLinus Torvalds { 21058387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21068387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 210736126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2108bfcfaa77SLinus Torvalds 21098387ff25SLinus Torvalds len = 0; 21108387ff25SLinus Torvalds goto inside; 21118387ff25SLinus Torvalds 2112bfcfaa77SLinus Torvalds do { 21132a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2114bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21158387ff25SLinus Torvalds inside: 2116e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 211736126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 211836126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2119bfcfaa77SLinus Torvalds 212036126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 212136126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 212236126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21232a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 212436126f8fSLinus Torvalds 21252a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2126bfcfaa77SLinus Torvalds } 2127bfcfaa77SLinus Torvalds 21282a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2129bfcfaa77SLinus Torvalds 2130fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 21318387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 21320145acc2SLinus Torvalds { 21338387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 21340145acc2SLinus Torvalds while (len--) 2135fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 21360145acc2SLinus Torvalds return end_name_hash(hash); 21370145acc2SLinus Torvalds } 2138ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 21390145acc2SLinus Torvalds 2140fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21418387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2142fcfd2fbfSGeorge Spelvin { 21438387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2144fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2145fcfd2fbfSGeorge Spelvin 2146fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2147e0ab7af9SGeorge Spelvin while (c) { 2148fcfd2fbfSGeorge Spelvin len++; 2149fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2150fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2151e0ab7af9SGeorge Spelvin } 2152fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2153fcfd2fbfSGeorge Spelvin } 2154f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2155fcfd2fbfSGeorge Spelvin 21563ddcd056SLinus Torvalds /* 2157200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2158200e9ef7SLinus Torvalds * one character. 2159200e9ef7SLinus Torvalds */ 21608387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2161200e9ef7SLinus Torvalds { 21628387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2163200e9ef7SLinus Torvalds unsigned long len = 0, c; 2164200e9ef7SLinus Torvalds 2165200e9ef7SLinus Torvalds c = (unsigned char)*name; 2166200e9ef7SLinus Torvalds do { 2167200e9ef7SLinus Torvalds len++; 2168200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2169200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2170200e9ef7SLinus Torvalds } while (c && c != '/'); 2171d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2172200e9ef7SLinus Torvalds } 2173200e9ef7SLinus Torvalds 2174bfcfaa77SLinus Torvalds #endif 2175bfcfaa77SLinus Torvalds 2176200e9ef7SLinus Torvalds /* 21771da177e4SLinus Torvalds * Name resolution. 2178ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2179ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 21801da177e4SLinus Torvalds * 2181ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2182ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 21831da177e4SLinus Torvalds */ 21846de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 21851da177e4SLinus Torvalds { 2186d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 21871da177e4SLinus Torvalds int err; 21881da177e4SLinus Torvalds 2189b4c03536SAl Viro nd->last_type = LAST_ROOT; 2190c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 21919b5858e9SAl Viro if (IS_ERR(name)) 21929b5858e9SAl Viro return PTR_ERR(name); 21931da177e4SLinus Torvalds while (*name=='/') 21941da177e4SLinus Torvalds name++; 21951a97d899SAl Viro if (!*name) { 21961a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 21979e18f10aSAl Viro return 0; 21981a97d899SAl Viro } 21991da177e4SLinus Torvalds 22001da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22011da177e4SLinus Torvalds for(;;) { 2202*549c7297SChristian Brauner struct user_namespace *mnt_userns; 220392d27016SAl Viro const char *link; 2204d6bb3e90SLinus Torvalds u64 hash_len; 2205fe479a58SAl Viro int type; 22061da177e4SLinus Torvalds 2207*549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 2208*549c7297SChristian Brauner err = may_lookup(mnt_userns, nd); 22091da177e4SLinus Torvalds if (err) 22103595e234SAl Viro return err; 22111da177e4SLinus Torvalds 22128387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22131da177e4SLinus Torvalds 2214fe479a58SAl Viro type = LAST_NORM; 2215d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2216fe479a58SAl Viro case 2: 2217200e9ef7SLinus Torvalds if (name[1] == '.') { 2218fe479a58SAl Viro type = LAST_DOTDOT; 221916c2cd71SAl Viro nd->flags |= LOOKUP_JUMPED; 222016c2cd71SAl Viro } 2221fe479a58SAl Viro break; 2222fe479a58SAl Viro case 1: 2223fe479a58SAl Viro type = LAST_DOT; 2224fe479a58SAl Viro } 22255a202bcdSAl Viro if (likely(type == LAST_NORM)) { 22265a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 222716c2cd71SAl Viro nd->flags &= ~LOOKUP_JUMPED; 22285a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2229a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2230da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 22315a202bcdSAl Viro if (err < 0) 22323595e234SAl Viro return err; 2233d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2234d6bb3e90SLinus Torvalds name = this.name; 22355a202bcdSAl Viro } 22365a202bcdSAl Viro } 2237fe479a58SAl Viro 2238d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2239d6bb3e90SLinus Torvalds nd->last.name = name; 22405f4a6a69SAl Viro nd->last_type = type; 22415f4a6a69SAl Viro 2242d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2243d6bb3e90SLinus Torvalds if (!*name) 2244bdf6cbf1SAl Viro goto OK; 2245200e9ef7SLinus Torvalds /* 2246200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2247200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2248200e9ef7SLinus Torvalds */ 2249200e9ef7SLinus Torvalds do { 2250d6bb3e90SLinus Torvalds name++; 2251d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 22528620c238SAl Viro if (unlikely(!*name)) { 22538620c238SAl Viro OK: 2254d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2255c108837eSAl Viro if (!depth) { 2256*549c7297SChristian Brauner nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode); 22570f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2258c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 22598620c238SAl Viro return 0; 2260c108837eSAl Viro } 22618620c238SAl Viro /* last component of nested symlink */ 2262d8d4611aSAl Viro name = nd->stack[--depth].name; 22638c4efe22SAl Viro link = walk_component(nd, 0); 22641c4ff1a8SAl Viro } else { 22651c4ff1a8SAl Viro /* not the last component */ 22668c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 22678620c238SAl Viro } 226892d27016SAl Viro if (unlikely(link)) { 226992d27016SAl Viro if (IS_ERR(link)) 227092d27016SAl Viro return PTR_ERR(link); 227192d27016SAl Viro /* a symlink to follow */ 2272d8d4611aSAl Viro nd->stack[depth++].name = name; 227392d27016SAl Viro name = link; 22749e18f10aSAl Viro continue; 227548c8b0c5SAl Viro } 227697242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 227797242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 22784675ac39SAl Viro if (unlazy_walk(nd)) 227997242f99SAl Viro return -ECHILD; 228097242f99SAl Viro } 22813595e234SAl Viro return -ENOTDIR; 22825f4a6a69SAl Viro } 2283ce57dfc1SAl Viro } 228497242f99SAl Viro } 22851da177e4SLinus Torvalds 2286edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2287c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 228831e6b01fSNick Piggin { 2289740a1678SAleksa Sarai int error; 2290c8a53ee5SAl Viro const char *s = nd->name->name; 229131e6b01fSNick Piggin 2292c0eb027eSLinus Torvalds if (!*s) 2293c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2294edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2295edc2b1daSAl Viro rcu_read_lock(); 2296c0eb027eSLinus Torvalds 2297c108837eSAl Viro nd->flags = flags | LOOKUP_JUMPED; 229831e6b01fSNick Piggin nd->depth = 0; 2299ab87f9a5SAleksa Sarai 2300ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2301ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2302ab87f9a5SAleksa Sarai smp_rmb(); 2303ab87f9a5SAleksa Sarai 23045b6ca027SAl Viro if (flags & LOOKUP_ROOT) { 2305b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2306b18825a7SDavid Howells struct inode *inode = root->d_inode; 230793893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2308368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23095b6ca027SAl Viro nd->path = nd->root; 23105b6ca027SAl Viro nd->inode = inode; 23115b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2312ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23138f47a016SAl Viro nd->root_seq = nd->seq; 23145b6ca027SAl Viro } else { 23155b6ca027SAl Viro path_get(&nd->path); 23165b6ca027SAl Viro } 2317368ee9baSAl Viro return s; 23185b6ca027SAl Viro } 23195b6ca027SAl Viro 232031e6b01fSNick Piggin nd->root.mnt = NULL; 2321248fb5b9SAl Viro nd->path.mnt = NULL; 2322248fb5b9SAl Viro nd->path.dentry = NULL; 232331e6b01fSNick Piggin 23248db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 23258db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2326740a1678SAleksa Sarai error = nd_jump_root(nd); 2327740a1678SAleksa Sarai if (unlikely(error)) 2328740a1678SAleksa Sarai return ERR_PTR(error); 2329ef55d917SAl Viro return s; 23308db52c7eSAleksa Sarai } 23318db52c7eSAleksa Sarai 23328db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 23338db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2334e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 233531e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2336c28cc364SNick Piggin unsigned seq; 233731e6b01fSNick Piggin 2338c28cc364SNick Piggin do { 2339c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 234031e6b01fSNick Piggin nd->path = fs->pwd; 2341ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2342c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2343c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2344e41f7d4eSAl Viro } else { 2345e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2346ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2347e41f7d4eSAl Viro } 234831e6b01fSNick Piggin } else { 2349582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2350c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 235131e6b01fSNick Piggin struct dentry *dentry; 235231e6b01fSNick Piggin 23532903ff01SAl Viro if (!f.file) 2354368ee9baSAl Viro return ERR_PTR(-EBADF); 235531e6b01fSNick Piggin 23562903ff01SAl Viro dentry = f.file->f_path.dentry; 235731e6b01fSNick Piggin 2358edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 23592903ff01SAl Viro fdput(f); 2360368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2361f52e0c11SAl Viro } 23622903ff01SAl Viro 23632903ff01SAl Viro nd->path = f.file->f_path; 2364e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 236534a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 236634a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23675590ff0dSUlrich Drepper } else { 23682903ff01SAl Viro path_get(&nd->path); 236934a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 23701da177e4SLinus Torvalds } 237134a26b99SAl Viro fdput(f); 2372e41f7d4eSAl Viro } 23738db52c7eSAleksa Sarai 2374adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2375adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2376adb21d2bSAleksa Sarai nd->root = nd->path; 2377adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2378adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2379adb21d2bSAleksa Sarai } else { 2380adb21d2bSAleksa Sarai path_get(&nd->root); 2381adb21d2bSAleksa Sarai nd->flags |= LOOKUP_ROOT_GRABBED; 2382adb21d2bSAleksa Sarai } 2383adb21d2bSAleksa Sarai } 2384adb21d2bSAleksa Sarai return s; 23859b4a9b14SAl Viro } 23869b4a9b14SAl Viro 23871ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 238895fa25d9SAl Viro { 2389bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2390bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2391bd92d7feSAl Viro 2392c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2393bd92d7feSAl Viro } 2394bd92d7feSAl Viro 23954f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 23964f757f3cSAl Viro { 2397c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2398db3c9adeSAl Viro dget(nd->path.dentry); 2399b0417d2cSAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, 2400b0417d2cSAl Viro nd->path.dentry, nd->inode, nd->seq)); 24014f757f3cSAl Viro } 24024f757f3cSAl Viro 24039b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2404c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24059b4a9b14SAl Viro { 2406c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2407bd92d7feSAl Viro int err; 240831e6b01fSNick Piggin 24099b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24104f757f3cSAl Viro err = handle_lookup_down(nd); 24115f336e72SAl Viro if (unlikely(err < 0)) 24125f336e72SAl Viro s = ERR_PTR(err); 24134f757f3cSAl Viro } 24144f757f3cSAl Viro 24151ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24161ccac622SAl Viro (s = lookup_last(nd)) != NULL) 24171ccac622SAl Viro ; 24189f1fafeeSAl Viro if (!err) 24199f1fafeeSAl Viro err = complete_walk(nd); 2420bd92d7feSAl Viro 2421deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2422deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2423bd23a539SAl Viro err = -ENOTDIR; 2424161aff1dSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 2425161aff1dSAl Viro err = handle_lookup_down(nd); 2426161aff1dSAl Viro nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please... 2427161aff1dSAl Viro } 2428625b6d10SAl Viro if (!err) { 2429625b6d10SAl Viro *path = nd->path; 2430625b6d10SAl Viro nd->path.mnt = NULL; 2431625b6d10SAl Viro nd->path.dentry = NULL; 2432625b6d10SAl Viro } 2433deb106c6SAl Viro terminate_walk(nd); 2434bd92d7feSAl Viro return err; 243531e6b01fSNick Piggin } 243631e6b01fSNick Piggin 243731d921c7SDavid Howells int filename_lookup(int dfd, struct filename *name, unsigned flags, 24389ad1aaa6SAl Viro struct path *path, struct path *root) 2439873f1eedSJeff Layton { 2440894bc8c4SAl Viro int retval; 24419883d185SAl Viro struct nameidata nd; 2442abc9f5beSAl Viro if (IS_ERR(name)) 2443abc9f5beSAl Viro return PTR_ERR(name); 24449ad1aaa6SAl Viro if (unlikely(root)) { 24459ad1aaa6SAl Viro nd.root = *root; 24469ad1aaa6SAl Viro flags |= LOOKUP_ROOT; 24479ad1aaa6SAl Viro } 24489883d185SAl Viro set_nameidata(&nd, dfd, name); 2449c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2450873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2451c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2452873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2453c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2454873f1eedSJeff Layton 2455873f1eedSJeff Layton if (likely(!retval)) 2456161aff1dSAl Viro audit_inode(name, path->dentry, 2457161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 24589883d185SAl Viro restore_nameidata(); 2459e4bd1c1aSAl Viro putname(name); 2460873f1eedSJeff Layton return retval; 2461873f1eedSJeff Layton } 2462873f1eedSJeff Layton 24638bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2464c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2465391172c4SAl Viro struct path *parent) 24668bcb77faSAl Viro { 2467c8a53ee5SAl Viro const char *s = path_init(nd, flags); 24689b5858e9SAl Viro int err = link_path_walk(s, nd); 24698bcb77faSAl Viro if (!err) 24708bcb77faSAl Viro err = complete_walk(nd); 2471391172c4SAl Viro if (!err) { 2472391172c4SAl Viro *parent = nd->path; 2473391172c4SAl Viro nd->path.mnt = NULL; 2474391172c4SAl Viro nd->path.dentry = NULL; 2475391172c4SAl Viro } 2476deb106c6SAl Viro terminate_walk(nd); 24778bcb77faSAl Viro return err; 24788bcb77faSAl Viro } 24798bcb77faSAl Viro 24805c31b6ceSAl Viro static struct filename *filename_parentat(int dfd, struct filename *name, 2481391172c4SAl Viro unsigned int flags, struct path *parent, 2482391172c4SAl Viro struct qstr *last, int *type) 24838bcb77faSAl Viro { 24848bcb77faSAl Viro int retval; 24859883d185SAl Viro struct nameidata nd; 24868bcb77faSAl Viro 24875c31b6ceSAl Viro if (IS_ERR(name)) 24885c31b6ceSAl Viro return name; 24899883d185SAl Viro set_nameidata(&nd, dfd, name); 2490c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 24918bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2492c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 24938bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2494c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2495391172c4SAl Viro if (likely(!retval)) { 2496391172c4SAl Viro *last = nd.last; 2497391172c4SAl Viro *type = nd.last_type; 2498c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 24995c31b6ceSAl Viro } else { 25005c31b6ceSAl Viro putname(name); 25015c31b6ceSAl Viro name = ERR_PTR(retval); 2502391172c4SAl Viro } 25039883d185SAl Viro restore_nameidata(); 25045c31b6ceSAl Viro return name; 25058bcb77faSAl Viro } 25068bcb77faSAl Viro 250779714f72SAl Viro /* does lookup, returns the object with parent locked */ 250879714f72SAl Viro struct dentry *kern_path_locked(const char *name, struct path *path) 25095590ff0dSUlrich Drepper { 25105c31b6ceSAl Viro struct filename *filename; 25115c31b6ceSAl Viro struct dentry *d; 2512391172c4SAl Viro struct qstr last; 2513391172c4SAl Viro int type; 251451689104SPaul Moore 25155c31b6ceSAl Viro filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path, 25165c31b6ceSAl Viro &last, &type); 251751689104SPaul Moore if (IS_ERR(filename)) 251851689104SPaul Moore return ERR_CAST(filename); 25195c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2520391172c4SAl Viro path_put(path); 25215c31b6ceSAl Viro putname(filename); 25225c31b6ceSAl Viro return ERR_PTR(-EINVAL); 252379714f72SAl Viro } 25245955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2525391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 252679714f72SAl Viro if (IS_ERR(d)) { 25275955102cSAl Viro inode_unlock(path->dentry->d_inode); 2528391172c4SAl Viro path_put(path); 252979714f72SAl Viro } 253051689104SPaul Moore putname(filename); 253179714f72SAl Viro return d; 25325590ff0dSUlrich Drepper } 25335590ff0dSUlrich Drepper 2534d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2535d1811465SAl Viro { 2536abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2537abc9f5beSAl Viro flags, path, NULL); 2538d1811465SAl Viro } 25394d359507SAl Viro EXPORT_SYMBOL(kern_path); 2540d1811465SAl Viro 254116f18200SJosef 'Jeff' Sipek /** 254216f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 254316f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 254416f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 254516f18200SJosef 'Jeff' Sipek * @name: pointer to file name 254616f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2547e0a01249SAl Viro * @path: pointer to struct path to fill 254816f18200SJosef 'Jeff' Sipek */ 254916f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 255016f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2551e0a01249SAl Viro struct path *path) 255216f18200SJosef 'Jeff' Sipek { 25539ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 25549ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2555abc9f5beSAl Viro return filename_lookup(AT_FDCWD, getname_kernel(name), 2556abc9f5beSAl Viro flags , path, &root); 255716f18200SJosef 'Jeff' Sipek } 25584d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 255916f18200SJosef 'Jeff' Sipek 25603c95f0dcSAl Viro static int lookup_one_len_common(const char *name, struct dentry *base, 25613c95f0dcSAl Viro int len, struct qstr *this) 25623c95f0dcSAl Viro { 25633c95f0dcSAl Viro this->name = name; 25643c95f0dcSAl Viro this->len = len; 25653c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 25663c95f0dcSAl Viro if (!len) 25673c95f0dcSAl Viro return -EACCES; 25683c95f0dcSAl Viro 25693c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 25703c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 25713c95f0dcSAl Viro return -EACCES; 25723c95f0dcSAl Viro } 25733c95f0dcSAl Viro 25743c95f0dcSAl Viro while (len--) { 25753c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 25763c95f0dcSAl Viro if (c == '/' || c == '\0') 25773c95f0dcSAl Viro return -EACCES; 25783c95f0dcSAl Viro } 25793c95f0dcSAl Viro /* 25803c95f0dcSAl Viro * See if the low-level filesystem might want 25813c95f0dcSAl Viro * to use its own hash.. 25823c95f0dcSAl Viro */ 25833c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 25843c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 25853c95f0dcSAl Viro if (err < 0) 25863c95f0dcSAl Viro return err; 25873c95f0dcSAl Viro } 25883c95f0dcSAl Viro 258947291baaSChristian Brauner return inode_permission(&init_user_ns, base->d_inode, MAY_EXEC); 25903c95f0dcSAl Viro } 25913c95f0dcSAl Viro 2592eead1911SChristoph Hellwig /** 25930da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 25940da0b7fdSDavid Howells * @name: pathname component to lookup 25950da0b7fdSDavid Howells * @base: base directory to lookup from 25960da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 25970da0b7fdSDavid Howells * 25980da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 25990da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26000da0b7fdSDavid Howells * 26010da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26020da0b7fdSDavid Howells * not be called by generic code. 26030da0b7fdSDavid Howells * 26040da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26050da0b7fdSDavid Howells */ 26060da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26070da0b7fdSDavid Howells { 26080da0b7fdSDavid Howells struct qstr this; 26090da0b7fdSDavid Howells int err; 26100da0b7fdSDavid Howells 26110da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26120da0b7fdSDavid Howells 26130da0b7fdSDavid Howells err = lookup_one_len_common(name, base, len, &this); 26140da0b7fdSDavid Howells if (err) 26150da0b7fdSDavid Howells return ERR_PTR(err); 26160da0b7fdSDavid Howells 26170da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 26180da0b7fdSDavid Howells } 26190da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 26200da0b7fdSDavid Howells 26210da0b7fdSDavid Howells /** 2622a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2623eead1911SChristoph Hellwig * @name: pathname component to lookup 2624eead1911SChristoph Hellwig * @base: base directory to lookup from 2625eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2626eead1911SChristoph Hellwig * 2627a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 26289e7543e9SAl Viro * not be called by generic code. 2629bbddca8eSNeilBrown * 2630bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2631eead1911SChristoph Hellwig */ 2632057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2633057f6c01SJames Morris { 26348613a209SAl Viro struct dentry *dentry; 2635057f6c01SJames Morris struct qstr this; 2636cda309deSMiklos Szeredi int err; 2637057f6c01SJames Morris 26385955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26392f9092e1SDavid Woodhouse 26403c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2641cda309deSMiklos Szeredi if (err) 2642cda309deSMiklos Szeredi return ERR_PTR(err); 2643cda309deSMiklos Szeredi 26448613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 26458613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2646057f6c01SJames Morris } 26474d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2648057f6c01SJames Morris 2649bbddca8eSNeilBrown /** 2650bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2651bbddca8eSNeilBrown * @name: pathname component to lookup 2652bbddca8eSNeilBrown * @base: base directory to lookup from 2653bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2654bbddca8eSNeilBrown * 2655bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2656bbddca8eSNeilBrown * not be called by generic code. 2657bbddca8eSNeilBrown * 2658bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2659bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2660bbddca8eSNeilBrown */ 2661bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2662bbddca8eSNeilBrown struct dentry *base, int len) 2663bbddca8eSNeilBrown { 2664bbddca8eSNeilBrown struct qstr this; 2665bbddca8eSNeilBrown int err; 266620d00ee8SLinus Torvalds struct dentry *ret; 2667bbddca8eSNeilBrown 26683c95f0dcSAl Viro err = lookup_one_len_common(name, base, len, &this); 2669bbddca8eSNeilBrown if (err) 2670bbddca8eSNeilBrown return ERR_PTR(err); 2671bbddca8eSNeilBrown 267220d00ee8SLinus Torvalds ret = lookup_dcache(&this, base, 0); 267320d00ee8SLinus Torvalds if (!ret) 267420d00ee8SLinus Torvalds ret = lookup_slow(&this, base, 0); 267520d00ee8SLinus Torvalds return ret; 2676bbddca8eSNeilBrown } 2677bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2678bbddca8eSNeilBrown 26796c2d4798SAl Viro /* 26806c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 26816c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 26826c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 26836c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 26846c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 26856c2d4798SAl Viro * this one avoids such problems. 26866c2d4798SAl Viro */ 26876c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 26886c2d4798SAl Viro struct dentry *base, int len) 26896c2d4798SAl Viro { 26906c2d4798SAl Viro struct dentry *ret = lookup_one_len_unlocked(name, base, len); 26912fa6b1e0SAl Viro if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 26926c2d4798SAl Viro dput(ret); 26936c2d4798SAl Viro ret = ERR_PTR(-ENOENT); 26946c2d4798SAl Viro } 26956c2d4798SAl Viro return ret; 26966c2d4798SAl Viro } 26976c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 26986c2d4798SAl Viro 2699eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2700eedf265aSEric W. Biederman int path_pts(struct path *path) 2701eedf265aSEric W. Biederman { 2702eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2703eedf265aSEric W. Biederman * the input path. 2704eedf265aSEric W. Biederman */ 2705a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2706a6a7eb76SAl Viro struct dentry *child; 270719f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2708eedf265aSEric W. Biederman 2709a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2710a6a7eb76SAl Viro dput(parent); 271163b27720SAl Viro return -ENOENT; 2712a6a7eb76SAl Viro } 271363b27720SAl Viro dput(path->dentry); 271463b27720SAl Viro path->dentry = parent; 2715eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2716eedf265aSEric W. Biederman if (!child) 2717eedf265aSEric W. Biederman return -ENOENT; 2718eedf265aSEric W. Biederman 2719eedf265aSEric W. Biederman path->dentry = child; 2720eedf265aSEric W. Biederman dput(parent); 272119f6028aSAl Viro follow_down(path); 2722eedf265aSEric W. Biederman return 0; 2723eedf265aSEric W. Biederman } 2724eedf265aSEric W. Biederman #endif 2725eedf265aSEric W. Biederman 27261fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 27271fa1e7f6SAndy Whitcroft struct path *path, int *empty) 27281da177e4SLinus Torvalds { 2729abc9f5beSAl Viro return filename_lookup(dfd, getname_flags(name, flags, empty), 2730abc9f5beSAl Viro flags, path, NULL); 27311da177e4SLinus Torvalds } 2732b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 27331fa1e7f6SAndy Whitcroft 2734ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2735ba73d987SChristian Brauner struct inode *inode) 27361da177e4SLinus Torvalds { 27378e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2738da9592edSDavid Howells 2739ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid)) 27401da177e4SLinus Torvalds return 0; 2741ba73d987SChristian Brauner if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid)) 27421da177e4SLinus Torvalds return 0; 2743ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 27441da177e4SLinus Torvalds } 2745cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 27461da177e4SLinus Torvalds 27471da177e4SLinus Torvalds /* 27481da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 27491da177e4SLinus Torvalds * whether the type of victim is right. 27501da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 27511da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 27521da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 27531da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 27541da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 27551da177e4SLinus Torvalds * a. be owner of dir, or 27561da177e4SLinus Torvalds * b. be owner of victim, or 27571da177e4SLinus Torvalds * c. have CAP_FOWNER capability 27581da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 27591da177e4SLinus Torvalds * links pointing to it. 27600bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 27610bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 27620bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 27630bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 27640bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 27651da177e4SLinus Torvalds * nfs_async_unlink(). 27661da177e4SLinus Torvalds */ 2767ba73d987SChristian Brauner static int may_delete(struct user_namespace *mnt_userns, struct inode *dir, 2768ba73d987SChristian Brauner struct dentry *victim, bool isdir) 27691da177e4SLinus Torvalds { 277063afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 27711da177e4SLinus Torvalds int error; 27721da177e4SLinus Torvalds 2773b18825a7SDavid Howells if (d_is_negative(victim)) 27741da177e4SLinus Torvalds return -ENOENT; 2775b18825a7SDavid Howells BUG_ON(!inode); 27761da177e4SLinus Torvalds 27771da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2778593d1ce8SEric W. Biederman 2779593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2780ba73d987SChristian Brauner if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) || 2781ba73d987SChristian Brauner !gid_valid(i_gid_into_mnt(mnt_userns, inode))) 2782593d1ce8SEric W. Biederman return -EOVERFLOW; 2783593d1ce8SEric W. Biederman 27844fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 27851da177e4SLinus Torvalds 2786ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 27871da177e4SLinus Torvalds if (error) 27881da177e4SLinus Torvalds return error; 27891da177e4SLinus Torvalds if (IS_APPEND(dir)) 27901da177e4SLinus Torvalds return -EPERM; 2791b18825a7SDavid Howells 2792ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2793ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 2794ba73d987SChristian Brauner HAS_UNMAPPED_ID(mnt_userns, inode)) 27951da177e4SLinus Torvalds return -EPERM; 27961da177e4SLinus Torvalds if (isdir) { 279744b1d530SMiklos Szeredi if (!d_is_dir(victim)) 27981da177e4SLinus Torvalds return -ENOTDIR; 27991da177e4SLinus Torvalds if (IS_ROOT(victim)) 28001da177e4SLinus Torvalds return -EBUSY; 280144b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 28021da177e4SLinus Torvalds return -EISDIR; 28031da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28041da177e4SLinus Torvalds return -ENOENT; 28051da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 28061da177e4SLinus Torvalds return -EBUSY; 28071da177e4SLinus Torvalds return 0; 28081da177e4SLinus Torvalds } 28091da177e4SLinus Torvalds 28101da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 28111da177e4SLinus Torvalds * dir. 28121da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 28131da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 28141da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2815036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2816036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2817036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 28181da177e4SLinus Torvalds */ 2819ba73d987SChristian Brauner static inline int may_create(struct user_namespace *mnt_userns, 2820ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 28211da177e4SLinus Torvalds { 2822036d5236SEric W. Biederman struct user_namespace *s_user_ns; 282314e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 28241da177e4SLinus Torvalds if (child->d_inode) 28251da177e4SLinus Torvalds return -EEXIST; 28261da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 28271da177e4SLinus Torvalds return -ENOENT; 2828036d5236SEric W. Biederman s_user_ns = dir->i_sb->s_user_ns; 2829ba73d987SChristian Brauner if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) || 2830ba73d987SChristian Brauner !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns))) 2831036d5236SEric W. Biederman return -EOVERFLOW; 2832ba73d987SChristian Brauner return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 28331da177e4SLinus Torvalds } 28341da177e4SLinus Torvalds 28351da177e4SLinus Torvalds /* 28361da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 28371da177e4SLinus Torvalds */ 28381da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 28391da177e4SLinus Torvalds { 28401da177e4SLinus Torvalds struct dentry *p; 28411da177e4SLinus Torvalds 28421da177e4SLinus Torvalds if (p1 == p2) { 28435955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28441da177e4SLinus Torvalds return NULL; 28451da177e4SLinus Torvalds } 28461da177e4SLinus Torvalds 2847fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 28481da177e4SLinus Torvalds 2849e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 2850e2761a11SOGAWA Hirofumi if (p) { 28515955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 28525955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 28531da177e4SLinus Torvalds return p; 28541da177e4SLinus Torvalds } 28551da177e4SLinus Torvalds 2856e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 2857e2761a11SOGAWA Hirofumi if (p) { 28585955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28595955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 28601da177e4SLinus Torvalds return p; 28611da177e4SLinus Torvalds } 28621da177e4SLinus Torvalds 28635955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 28645955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 28651da177e4SLinus Torvalds return NULL; 28661da177e4SLinus Torvalds } 28674d359507SAl Viro EXPORT_SYMBOL(lock_rename); 28681da177e4SLinus Torvalds 28691da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 28701da177e4SLinus Torvalds { 28715955102cSAl Viro inode_unlock(p1->d_inode); 28721da177e4SLinus Torvalds if (p1 != p2) { 28735955102cSAl Viro inode_unlock(p2->d_inode); 2874fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 28751da177e4SLinus Torvalds } 28761da177e4SLinus Torvalds } 28774d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 28781da177e4SLinus Torvalds 28796521f891SChristian Brauner /** 28806521f891SChristian Brauner * vfs_create - create new file 28816521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 28826521f891SChristian Brauner * @dir: inode of @dentry 28836521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 28846521f891SChristian Brauner * @mode: mode of the new file 28856521f891SChristian Brauner * @want_excl: whether the file must not yet exist 28866521f891SChristian Brauner * 28876521f891SChristian Brauner * Create a new file. 28886521f891SChristian Brauner * 28896521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 28906521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 28916521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 28926521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 28936521f891SChristian Brauner * raw inode simply passs init_user_ns. 28946521f891SChristian Brauner */ 28956521f891SChristian Brauner int vfs_create(struct user_namespace *mnt_userns, struct inode *dir, 28966521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 28971da177e4SLinus Torvalds { 28986521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 28991da177e4SLinus Torvalds if (error) 29001da177e4SLinus Torvalds return error; 29011da177e4SLinus Torvalds 2902acfa4380SAl Viro if (!dir->i_op->create) 29031da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 29041da177e4SLinus Torvalds mode &= S_IALLUGO; 29051da177e4SLinus Torvalds mode |= S_IFREG; 29061da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 29071da177e4SLinus Torvalds if (error) 29081da177e4SLinus Torvalds return error; 2909*549c7297SChristian Brauner error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl); 2910a74574aaSStephen Smalley if (!error) 2911f38aa942SAmy Griffis fsnotify_create(dir, dentry); 29121da177e4SLinus Torvalds return error; 29131da177e4SLinus Torvalds } 29144d359507SAl Viro EXPORT_SYMBOL(vfs_create); 29151da177e4SLinus Torvalds 29168e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 29178e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 29188e6c848eSAl Viro void *arg) 29198e6c848eSAl Viro { 29208e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 2921ba73d987SChristian Brauner int error = may_create(&init_user_ns, dir, dentry); 29228e6c848eSAl Viro if (error) 29238e6c848eSAl Viro return error; 29248e6c848eSAl Viro 29258e6c848eSAl Viro mode &= S_IALLUGO; 29268e6c848eSAl Viro mode |= S_IFREG; 29278e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 29288e6c848eSAl Viro if (error) 29298e6c848eSAl Viro return error; 29308e6c848eSAl Viro error = f(dentry, mode, arg); 29318e6c848eSAl Viro if (!error) 29328e6c848eSAl Viro fsnotify_create(dir, dentry); 29338e6c848eSAl Viro return error; 29348e6c848eSAl Viro } 29358e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 29368e6c848eSAl Viro 2937a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 2938a2982cc9SEric W. Biederman { 2939a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 2940a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 2941a2982cc9SEric W. Biederman } 2942a2982cc9SEric W. Biederman 2943ba73d987SChristian Brauner static int may_open(struct user_namespace *mnt_userns, const struct path *path, 2944ba73d987SChristian Brauner int acc_mode, int flag) 29451da177e4SLinus Torvalds { 29463fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 29471da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 29481da177e4SLinus Torvalds int error; 29491da177e4SLinus Torvalds 29501da177e4SLinus Torvalds if (!inode) 29511da177e4SLinus Torvalds return -ENOENT; 29521da177e4SLinus Torvalds 2953c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 2954c8fe8f30SChristoph Hellwig case S_IFLNK: 29551da177e4SLinus Torvalds return -ELOOP; 2956c8fe8f30SChristoph Hellwig case S_IFDIR: 2957fc4177beSKees Cook if (acc_mode & MAY_WRITE) 29581da177e4SLinus Torvalds return -EISDIR; 2959fc4177beSKees Cook if (acc_mode & MAY_EXEC) 2960fc4177beSKees Cook return -EACCES; 2961c8fe8f30SChristoph Hellwig break; 2962c8fe8f30SChristoph Hellwig case S_IFBLK: 2963c8fe8f30SChristoph Hellwig case S_IFCHR: 2964a2982cc9SEric W. Biederman if (!may_open_dev(path)) 29651da177e4SLinus Torvalds return -EACCES; 2966633fb6acSKees Cook fallthrough; 2967c8fe8f30SChristoph Hellwig case S_IFIFO: 2968c8fe8f30SChristoph Hellwig case S_IFSOCK: 2969633fb6acSKees Cook if (acc_mode & MAY_EXEC) 2970633fb6acSKees Cook return -EACCES; 29711da177e4SLinus Torvalds flag &= ~O_TRUNC; 2972c8fe8f30SChristoph Hellwig break; 29730fd338b2SKees Cook case S_IFREG: 29740fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 29750fd338b2SKees Cook return -EACCES; 29760fd338b2SKees Cook break; 29774a3fd211SDave Hansen } 2978b41572e9SDave Hansen 2979ba73d987SChristian Brauner error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode); 2980b41572e9SDave Hansen if (error) 2981b41572e9SDave Hansen return error; 29826146f0d5SMimi Zohar 29831da177e4SLinus Torvalds /* 29841da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 29851da177e4SLinus Torvalds */ 29861da177e4SLinus Torvalds if (IS_APPEND(inode)) { 29878737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 29887715b521SAl Viro return -EPERM; 29891da177e4SLinus Torvalds if (flag & O_TRUNC) 29907715b521SAl Viro return -EPERM; 29911da177e4SLinus Torvalds } 29921da177e4SLinus Torvalds 29931da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 2994ba73d987SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode)) 29957715b521SAl Viro return -EPERM; 29961da177e4SLinus Torvalds 2997f3c7691eSJ. Bruce Fields return 0; 29987715b521SAl Viro } 29997715b521SAl Viro 3000*549c7297SChristian Brauner static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp) 30017715b521SAl Viro { 3002f0bb5aafSAl Viro const struct path *path = &filp->f_path; 30037715b521SAl Viro struct inode *inode = path->dentry->d_inode; 30047715b521SAl Viro int error = get_write_access(inode); 30051da177e4SLinus Torvalds if (error) 30067715b521SAl Viro return error; 30071da177e4SLinus Torvalds /* 30081da177e4SLinus Torvalds * Refuse to truncate files with mandatory locks held on them. 30091da177e4SLinus Torvalds */ 3010d7a06983SJeff Layton error = locks_verify_locked(filp); 3011be6d3e56SKentaro Takeda if (!error) 3012ea0d3ab2STetsuo Handa error = security_path_truncate(path); 30131da177e4SLinus Torvalds if (!error) { 3014*549c7297SChristian Brauner error = do_truncate(mnt_userns, path->dentry, 0, 3015d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3016e1181ee6SJeff Layton filp); 30171da177e4SLinus Torvalds } 30181da177e4SLinus Torvalds put_write_access(inode); 3019acd0c935SMimi Zohar return error; 30201da177e4SLinus Torvalds } 30211da177e4SLinus Torvalds 3022d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3023d57999e1SDave Hansen { 30248a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 30258a5e929dSAl Viro flag--; 3026d57999e1SDave Hansen return flag; 3027d57999e1SDave Hansen } 3028d57999e1SDave Hansen 3029ba73d987SChristian Brauner static int may_o_create(struct user_namespace *mnt_userns, 3030ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3031ba73d987SChristian Brauner umode_t mode) 3032d18e9008SMiklos Szeredi { 30331328c727SSeth Forshee struct user_namespace *s_user_ns; 3034d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3035d18e9008SMiklos Szeredi if (error) 3036d18e9008SMiklos Szeredi return error; 3037d18e9008SMiklos Szeredi 30381328c727SSeth Forshee s_user_ns = dir->dentry->d_sb->s_user_ns; 3039ba73d987SChristian Brauner if (!kuid_has_mapping(s_user_ns, fsuid_into_mnt(mnt_userns)) || 3040ba73d987SChristian Brauner !kgid_has_mapping(s_user_ns, fsgid_into_mnt(mnt_userns))) 30411328c727SSeth Forshee return -EOVERFLOW; 30421328c727SSeth Forshee 3043ba73d987SChristian Brauner error = inode_permission(mnt_userns, dir->dentry->d_inode, 304447291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3045d18e9008SMiklos Szeredi if (error) 3046d18e9008SMiklos Szeredi return error; 3047d18e9008SMiklos Szeredi 3048d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3049d18e9008SMiklos Szeredi } 3050d18e9008SMiklos Szeredi 30511acf0af9SDavid Howells /* 30521acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 30531acf0af9SDavid Howells * dentry. 30541acf0af9SDavid Howells * 30551acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 30561acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 30571acf0af9SDavid Howells * 305800a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 305900a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 306000a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 30611acf0af9SDavid Howells * 30621acf0af9SDavid Howells * Returns an error code otherwise. 30631acf0af9SDavid Howells */ 3064239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3065239eb983SAl Viro struct file *file, 30663ec2eef1SAl Viro int open_flag, umode_t mode) 3067d18e9008SMiklos Szeredi { 3068d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3069d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3070d18e9008SMiklos Szeredi int error; 3071d18e9008SMiklos Szeredi 3072d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3073d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3074d18e9008SMiklos Szeredi 307530d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 307630d90494SAl Viro file->f_path.mnt = nd->path.mnt; 30770fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 307844907d79SAl Viro open_to_namei_flags(open_flag), mode); 30796fbd0714SAl Viro d_lookup_done(dentry); 3080384f26e2SAl Viro if (!error) { 308164e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 30826fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 30836fb968cdSAl Viro dput(dentry); 30846fb968cdSAl Viro dentry = dget(file->f_path.dentry); 30856fb968cdSAl Viro } 308664e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 30872675a4ebSAl Viro error = -EIO; 3088384f26e2SAl Viro } else { 308930d90494SAl Viro if (file->f_path.dentry) { 3090d18e9008SMiklos Szeredi dput(dentry); 3091d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 309210c64ceaSAl Viro } 3093239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3094a01e718fSAl Viro error = -ENOENT; 3095d18e9008SMiklos Szeredi } 3096d18e9008SMiklos Szeredi } 3097239eb983SAl Viro if (error) { 3098d18e9008SMiklos Szeredi dput(dentry); 3099239eb983SAl Viro dentry = ERR_PTR(error); 3100239eb983SAl Viro } 3101239eb983SAl Viro return dentry; 3102d18e9008SMiklos Szeredi } 3103d18e9008SMiklos Szeredi 310431e6b01fSNick Piggin /* 31051acf0af9SDavid Howells * Look up and maybe create and open the last component. 3106d58ffd35SMiklos Szeredi * 310700a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3108d58ffd35SMiklos Szeredi * 310900a07c15SAl Viro * Returns 0 on success, that is, if 311000a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 311100a07c15SAl Viro * the file was not completely opened at this time, though lookups and 311200a07c15SAl Viro * creations were performed. 311300a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 311400a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 311500a07c15SAl Viro * hadn't been specified. 31161acf0af9SDavid Howells * 311700a07c15SAl Viro * An error code is returned on failure. 3118d58ffd35SMiklos Szeredi */ 3119da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3120d58ffd35SMiklos Szeredi const struct open_flags *op, 31213ec2eef1SAl Viro bool got_write) 3122d58ffd35SMiklos Szeredi { 3123*549c7297SChristian Brauner struct user_namespace *mnt_userns; 3124d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 312554ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 31261643b43fSAl Viro int open_flag = op->open_flag; 3127d58ffd35SMiklos Szeredi struct dentry *dentry; 31281643b43fSAl Viro int error, create_error = 0; 31291643b43fSAl Viro umode_t mode = op->mode; 31306fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3131d58ffd35SMiklos Szeredi 3132ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3133da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3134d58ffd35SMiklos Szeredi 313573a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 31366fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 31376fbd0714SAl Viro for (;;) { 31386fbd0714SAl Viro if (!dentry) { 31396fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3140d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3141da5ebf5aSAl Viro return dentry; 31426fbd0714SAl Viro } 31436fbd0714SAl Viro if (d_in_lookup(dentry)) 31446fbd0714SAl Viro break; 3145d58ffd35SMiklos Szeredi 31466fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 31476fbd0714SAl Viro if (likely(error > 0)) 31486fbd0714SAl Viro break; 31496fbd0714SAl Viro if (error) 31506fbd0714SAl Viro goto out_dput; 31516fbd0714SAl Viro d_invalidate(dentry); 31526fbd0714SAl Viro dput(dentry); 31536fbd0714SAl Viro dentry = NULL; 31546fbd0714SAl Viro } 31556fbd0714SAl Viro if (dentry->d_inode) { 3156d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3157da5ebf5aSAl Viro return dentry; 31586c51e513SAl Viro } 3159d18e9008SMiklos Szeredi 31601643b43fSAl Viro /* 31611643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 31621643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 31631643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 31641643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 31651643b43fSAl Viro * 31661643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 31671643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 31681643b43fSAl Viro */ 316999a4a90cSAl Viro if (unlikely(!got_write)) 317099a4a90cSAl Viro open_flag &= ~O_TRUNC; 3171*549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 31721643b43fSAl Viro if (open_flag & O_CREAT) { 317399a4a90cSAl Viro if (open_flag & O_EXCL) 317499a4a90cSAl Viro open_flag &= ~O_TRUNC; 31751643b43fSAl Viro if (!IS_POSIXACL(dir->d_inode)) 31761643b43fSAl Viro mode &= ~current_umask(); 317799a4a90cSAl Viro if (likely(got_write)) 3178*549c7297SChristian Brauner create_error = may_o_create(mnt_userns, &nd->path, 3179ba73d987SChristian Brauner dentry, mode); 318099a4a90cSAl Viro else 318199a4a90cSAl Viro create_error = -EROFS; 318299a4a90cSAl Viro } 318399a4a90cSAl Viro if (create_error) 31841643b43fSAl Viro open_flag &= ~O_CREAT; 31851643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3186d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3187da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3188da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3189da5ebf5aSAl Viro return dentry; 3190239eb983SAl Viro } 319154ef4872SMiklos Szeredi 31926fbd0714SAl Viro if (d_in_lookup(dentry)) { 319312fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 319412fa5e24SAl Viro nd->flags); 31956fbd0714SAl Viro d_lookup_done(dentry); 319612fa5e24SAl Viro if (unlikely(res)) { 319712fa5e24SAl Viro if (IS_ERR(res)) { 319812fa5e24SAl Viro error = PTR_ERR(res); 319912fa5e24SAl Viro goto out_dput; 320012fa5e24SAl Viro } 320112fa5e24SAl Viro dput(dentry); 320212fa5e24SAl Viro dentry = res; 320312fa5e24SAl Viro } 320454ef4872SMiklos Szeredi } 320554ef4872SMiklos Szeredi 3206d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 32071643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 320873a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3209ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3210ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3211ce8644fcSAl Viro error = -EACCES; 3212d58ffd35SMiklos Szeredi goto out_dput; 321364894cf8SAl Viro } 3214*549c7297SChristian Brauner 3215*549c7297SChristian Brauner error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry, 3216*549c7297SChristian Brauner mode, open_flag & O_EXCL); 3217d58ffd35SMiklos Szeredi if (error) 3218d58ffd35SMiklos Szeredi goto out_dput; 3219d58ffd35SMiklos Szeredi } 32201643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 32211643b43fSAl Viro error = create_error; 3222d58ffd35SMiklos Szeredi goto out_dput; 3223d58ffd35SMiklos Szeredi } 3224da5ebf5aSAl Viro return dentry; 3225d58ffd35SMiklos Szeredi 3226d58ffd35SMiklos Szeredi out_dput: 3227d58ffd35SMiklos Szeredi dput(dentry); 3228da5ebf5aSAl Viro return ERR_PTR(error); 3229d58ffd35SMiklos Szeredi } 3230d58ffd35SMiklos Szeredi 3231c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 32323ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3233fb1cc555SAl Viro { 3234a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3235ca344a89SAl Viro int open_flag = op->open_flag; 323664894cf8SAl Viro bool got_write = false; 3237254cf582SAl Viro unsigned seq; 3238a1eb3315SMiklos Szeredi struct inode *inode; 3239da5ebf5aSAl Viro struct dentry *dentry; 3240b0417d2cSAl Viro const char *res; 324116c2cd71SAl Viro int error; 3242fb1cc555SAl Viro 3243c3e380b0SAl Viro nd->flags |= op->intent; 3244c3e380b0SAl Viro 3245bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 324656676ec3SAl Viro if (nd->depth) 324756676ec3SAl Viro put_link(nd); 3248ff326a32SAl Viro return handle_dots(nd, nd->last_type); 32491f36f774SAl Viro } 3250a2c36b45SAl Viro 3251ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3252fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3253fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3254fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 325520e34357SAl Viro dentry = lookup_fast(nd, &inode, &seq); 325620e34357SAl Viro if (IS_ERR(dentry)) 32571ccac622SAl Viro return ERR_CAST(dentry); 325820e34357SAl Viro if (likely(dentry)) 325971574865SMiklos Szeredi goto finish_lookup; 326071574865SMiklos Szeredi 32616583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3262b6183df7SMiklos Szeredi } else { 3263fe2d35ffSAl Viro /* create side of things */ 326472287417SAl Viro if (nd->flags & LOOKUP_RCU) { 326572287417SAl Viro error = unlazy_walk(nd); 3266c981a482SAl Viro if (unlikely(error)) 32671ccac622SAl Viro return ERR_PTR(error); 326872287417SAl Viro } 3269c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 32701f36f774SAl Viro /* trailing slashes? */ 3271deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 32721ccac622SAl Viro return ERR_PTR(-EISDIR); 3273b6183df7SMiklos Szeredi } 32741f36f774SAl Viro 32759cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 327664894cf8SAl Viro error = mnt_want_write(nd->path.mnt); 327764894cf8SAl Viro if (!error) 327864894cf8SAl Viro got_write = true; 327964894cf8SAl Viro /* 328064894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 328164894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 328264894cf8SAl Viro * dropping this one anyway. 328364894cf8SAl Viro */ 328464894cf8SAl Viro } 32859cf843e3SAl Viro if (open_flag & O_CREAT) 32865955102cSAl Viro inode_lock(dir->d_inode); 32879cf843e3SAl Viro else 32889cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3289da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3290f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3291f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 32929cf843e3SAl Viro if (open_flag & O_CREAT) 32935955102cSAl Viro inode_unlock(dir->d_inode); 32949cf843e3SAl Viro else 32959cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3296fb1cc555SAl Viro 3297c981a482SAl Viro if (got_write) 329859e96e65SAl Viro mnt_drop_write(nd->path.mnt); 32996c0d46c4SAl Viro 330059e96e65SAl Viro if (IS_ERR(dentry)) 330159e96e65SAl Viro return ERR_CAST(dentry); 330259e96e65SAl Viro 3303973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3304e73cabffSAl Viro dput(nd->path.dentry); 3305e73cabffSAl Viro nd->path.dentry = dentry; 3306c981a482SAl Viro return NULL; 3307fb1cc555SAl Viro } 3308fb1cc555SAl Viro 330920e34357SAl Viro finish_lookup: 331056676ec3SAl Viro if (nd->depth) 331156676ec3SAl Viro put_link(nd); 33128c4efe22SAl Viro res = step_into(nd, WALK_TRAILING, dentry, inode, seq); 3313ff326a32SAl Viro if (unlikely(res)) 33141ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3315b0417d2cSAl Viro return res; 33161ccac622SAl Viro } 331731d1726dSAl Viro 3318c981a482SAl Viro /* 3319c981a482SAl Viro * Handle the last step of open() 3320c981a482SAl Viro */ 3321c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3322c981a482SAl Viro struct file *file, const struct open_flags *op) 3323c981a482SAl Viro { 3324*549c7297SChristian Brauner struct user_namespace *mnt_userns; 3325c981a482SAl Viro int open_flag = op->open_flag; 3326c981a482SAl Viro bool do_truncate; 3327c981a482SAl Viro int acc_mode; 3328c981a482SAl Viro int error; 3329c981a482SAl Viro 3330ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3331ff326a32SAl Viro error = complete_walk(nd); 3332ff326a32SAl Viro if (error) 3333ff326a32SAl Viro return error; 3334ff326a32SAl Viro } 3335973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 333676ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3337*549c7297SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 333830aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3339b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3340b94e0b32SAl Viro return -EEXIST; 334130aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3342c5971b8cSAl Viro return -EISDIR; 3343*549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 334430aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 334530aba665SSalvatore Mesoraca if (unlikely(error)) 3346c5971b8cSAl Viro return error; 334730aba665SSalvatore Mesoraca } 334844b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3349c5971b8cSAl Viro return -ENOTDIR; 33506c0d46c4SAl Viro 33518795e7d4SAl Viro do_truncate = false; 33528795e7d4SAl Viro acc_mode = op->acc_mode; 33535a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 33545a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 33555a2d3eddSAl Viro open_flag &= ~O_TRUNC; 33565a2d3eddSAl Viro acc_mode = 0; 33578795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 33580f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 33590f9d1a10SAl Viro if (error) 3360c5971b8cSAl Viro return error; 33618795e7d4SAl Viro do_truncate = true; 33620f9d1a10SAl Viro } 3363*549c7297SChristian Brauner error = may_open(mnt_userns, &nd->path, acc_mode, open_flag); 33648795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3365ae2bb293SAl Viro error = vfs_open(&nd->path, file); 33668795e7d4SAl Viro if (!error) 33676035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 33688795e7d4SAl Viro if (!error && do_truncate) 3369*549c7297SChristian Brauner error = handle_truncate(mnt_userns, file); 3370c80567c8SAl Viro if (unlikely(error > 0)) { 3371c80567c8SAl Viro WARN_ON(1); 3372c80567c8SAl Viro error = -EINVAL; 3373c80567c8SAl Viro } 33748795e7d4SAl Viro if (do_truncate) 33750f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3376c5971b8cSAl Viro return error; 3377fb1cc555SAl Viro } 3378fb1cc555SAl Viro 33796521f891SChristian Brauner /** 33806521f891SChristian Brauner * vfs_tmpfile - create tmpfile 33816521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 33826521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 33836521f891SChristian Brauner * @mode: mode of the new tmpfile 33846521f891SChristian Brauner * @open_flags: flags 33856521f891SChristian Brauner * 33866521f891SChristian Brauner * Create a temporary file. 33876521f891SChristian Brauner * 33886521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 33896521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 33906521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 33916521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 33926521f891SChristian Brauner * raw inode simply passs init_user_ns. 33936521f891SChristian Brauner */ 33946521f891SChristian Brauner struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns, 33956521f891SChristian Brauner struct dentry *dentry, umode_t mode, int open_flag) 3396af7bd4dcSAmir Goldstein { 3397af7bd4dcSAmir Goldstein struct dentry *child = NULL; 3398af7bd4dcSAmir Goldstein struct inode *dir = dentry->d_inode; 3399af7bd4dcSAmir Goldstein struct inode *inode; 3400af7bd4dcSAmir Goldstein int error; 3401af7bd4dcSAmir Goldstein 3402af7bd4dcSAmir Goldstein /* we want directory to be writable */ 34036521f891SChristian Brauner error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC); 3404af7bd4dcSAmir Goldstein if (error) 3405af7bd4dcSAmir Goldstein goto out_err; 3406af7bd4dcSAmir Goldstein error = -EOPNOTSUPP; 3407af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 3408af7bd4dcSAmir Goldstein goto out_err; 3409af7bd4dcSAmir Goldstein error = -ENOMEM; 3410cdf01226SDavid Howells child = d_alloc(dentry, &slash_name); 3411af7bd4dcSAmir Goldstein if (unlikely(!child)) 3412af7bd4dcSAmir Goldstein goto out_err; 3413*549c7297SChristian Brauner error = dir->i_op->tmpfile(mnt_userns, dir, child, mode); 3414af7bd4dcSAmir Goldstein if (error) 3415af7bd4dcSAmir Goldstein goto out_err; 3416af7bd4dcSAmir Goldstein error = -ENOENT; 3417af7bd4dcSAmir Goldstein inode = child->d_inode; 3418af7bd4dcSAmir Goldstein if (unlikely(!inode)) 3419af7bd4dcSAmir Goldstein goto out_err; 3420af7bd4dcSAmir Goldstein if (!(open_flag & O_EXCL)) { 3421af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3422af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3423af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3424af7bd4dcSAmir Goldstein } 3425fdb2410fSMimi Zohar ima_post_create_tmpfile(inode); 3426af7bd4dcSAmir Goldstein return child; 3427af7bd4dcSAmir Goldstein 3428af7bd4dcSAmir Goldstein out_err: 3429af7bd4dcSAmir Goldstein dput(child); 3430af7bd4dcSAmir Goldstein return ERR_PTR(error); 3431af7bd4dcSAmir Goldstein } 3432af7bd4dcSAmir Goldstein EXPORT_SYMBOL(vfs_tmpfile); 3433af7bd4dcSAmir Goldstein 3434c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 343560545d0dSAl Viro const struct open_flags *op, 34363ec2eef1SAl Viro struct file *file) 343760545d0dSAl Viro { 34386521f891SChristian Brauner struct user_namespace *mnt_userns; 3439625b6d10SAl Viro struct dentry *child; 3440625b6d10SAl Viro struct path path; 3441c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 344260545d0dSAl Viro if (unlikely(error)) 344360545d0dSAl Viro return error; 3444625b6d10SAl Viro error = mnt_want_write(path.mnt); 344560545d0dSAl Viro if (unlikely(error)) 344660545d0dSAl Viro goto out; 34476521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 34486521f891SChristian Brauner child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag); 3449af7bd4dcSAmir Goldstein error = PTR_ERR(child); 3450684e73beSHirofumi Nakagawa if (IS_ERR(child)) 345160545d0dSAl Viro goto out2; 3452625b6d10SAl Viro dput(path.dentry); 3453625b6d10SAl Viro path.dentry = child; 3454c8a53ee5SAl Viro audit_inode(nd->name, child, 0); 345569a91c23SEric Rannaud /* Don't check for other permissions, the inode was just created */ 3456*549c7297SChristian Brauner error = may_open(mnt_userns, &path, 0, op->open_flag); 345760545d0dSAl Viro if (error) 345860545d0dSAl Viro goto out2; 3459625b6d10SAl Viro file->f_path.mnt = path.mnt; 3460be12af3eSAl Viro error = finish_open(file, child, NULL); 346160545d0dSAl Viro out2: 3462625b6d10SAl Viro mnt_drop_write(path.mnt); 346360545d0dSAl Viro out: 3464625b6d10SAl Viro path_put(&path); 346560545d0dSAl Viro return error; 346660545d0dSAl Viro } 346760545d0dSAl Viro 34686ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 34696ac08709SAl Viro { 34706ac08709SAl Viro struct path path; 34716ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 34726ac08709SAl Viro if (!error) { 34736ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3474ae2bb293SAl Viro error = vfs_open(&path, file); 34756ac08709SAl Viro path_put(&path); 34766ac08709SAl Viro } 34776ac08709SAl Viro return error; 34786ac08709SAl Viro } 34796ac08709SAl Viro 3480c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3481c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 34821da177e4SLinus Torvalds { 348330d90494SAl Viro struct file *file; 348413aab428SAl Viro int error; 348531e6b01fSNick Piggin 3486ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 34871afc99beSAl Viro if (IS_ERR(file)) 34881afc99beSAl Viro return file; 348931e6b01fSNick Piggin 3490bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 34913ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 34925f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 34936ac08709SAl Viro error = do_o_path(nd, flags, file); 34945f336e72SAl Viro } else { 34955f336e72SAl Viro const char *s = path_init(nd, flags); 34963bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3497c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 34981ccac622SAl Viro ; 3499c5971b8cSAl Viro if (!error) 3500c5971b8cSAl Viro error = do_open(nd, file, op); 3501deb106c6SAl Viro terminate_walk(nd); 35025f336e72SAl Viro } 35037c1c01ecSAl Viro if (likely(!error)) { 3504aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 35057c1c01ecSAl Viro return file; 35067c1c01ecSAl Viro WARN_ON(1); 35077c1c01ecSAl Viro error = -EINVAL; 3508015c3bbcSMiklos Szeredi } 35097c1c01ecSAl Viro fput(file); 35102675a4ebSAl Viro if (error == -EOPENSTALE) { 35112675a4ebSAl Viro if (flags & LOOKUP_RCU) 35122675a4ebSAl Viro error = -ECHILD; 35132675a4ebSAl Viro else 35142675a4ebSAl Viro error = -ESTALE; 35152675a4ebSAl Viro } 35167c1c01ecSAl Viro return ERR_PTR(error); 3517de459215SKirill Korotaev } 35181da177e4SLinus Torvalds 3519669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3520f9652e10SAl Viro const struct open_flags *op) 352113aab428SAl Viro { 35229883d185SAl Viro struct nameidata nd; 3523f9652e10SAl Viro int flags = op->lookup_flags; 352413aab428SAl Viro struct file *filp; 352513aab428SAl Viro 35269883d185SAl Viro set_nameidata(&nd, dfd, pathname); 3527c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 352813aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3529c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 353013aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3531c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 35329883d185SAl Viro restore_nameidata(); 353313aab428SAl Viro return filp; 353413aab428SAl Viro } 353513aab428SAl Viro 353673d049a4SAl Viro struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt, 3537f9652e10SAl Viro const char *name, const struct open_flags *op) 353873d049a4SAl Viro { 35399883d185SAl Viro struct nameidata nd; 354073d049a4SAl Viro struct file *file; 354151689104SPaul Moore struct filename *filename; 3542f9652e10SAl Viro int flags = op->lookup_flags | LOOKUP_ROOT; 354373d049a4SAl Viro 354473d049a4SAl Viro nd.root.mnt = mnt; 354573d049a4SAl Viro nd.root.dentry = dentry; 354673d049a4SAl Viro 3547b18825a7SDavid Howells if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN) 354873d049a4SAl Viro return ERR_PTR(-ELOOP); 354973d049a4SAl Viro 355051689104SPaul Moore filename = getname_kernel(name); 3551a1c83681SViresh Kumar if (IS_ERR(filename)) 355251689104SPaul Moore return ERR_CAST(filename); 355351689104SPaul Moore 35549883d185SAl Viro set_nameidata(&nd, -1, filename); 3555c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 355673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3557c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 355873d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3559c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 35609883d185SAl Viro restore_nameidata(); 356151689104SPaul Moore putname(filename); 356273d049a4SAl Viro return file; 356373d049a4SAl Viro } 356473d049a4SAl Viro 3565fa14a0b8SAl Viro static struct dentry *filename_create(int dfd, struct filename *name, 35661ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 35671da177e4SLinus Torvalds { 3568c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3569391172c4SAl Viro struct qstr last; 3570391172c4SAl Viro int type; 3571c30dabfeSJan Kara int err2; 35721ac12b4bSJeff Layton int error; 35731ac12b4bSJeff Layton bool is_dir = (lookup_flags & LOOKUP_DIRECTORY); 35741ac12b4bSJeff Layton 35751ac12b4bSJeff Layton /* 35761ac12b4bSJeff Layton * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any 35771ac12b4bSJeff Layton * other flags passed in are ignored! 35781ac12b4bSJeff Layton */ 35791ac12b4bSJeff Layton lookup_flags &= LOOKUP_REVAL; 35801ac12b4bSJeff Layton 35815c31b6ceSAl Viro name = filename_parentat(dfd, name, lookup_flags, path, &last, &type); 35825c31b6ceSAl Viro if (IS_ERR(name)) 35835c31b6ceSAl Viro return ERR_CAST(name); 35841da177e4SLinus Torvalds 3585c663e5d8SChristoph Hellwig /* 3586c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3587c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3588c663e5d8SChristoph Hellwig */ 35895c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3590ed75e95dSAl Viro goto out; 3591c663e5d8SChristoph Hellwig 3592c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3593391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3594c663e5d8SChristoph Hellwig /* 3595c663e5d8SChristoph Hellwig * Do the final lookup. 3596c663e5d8SChristoph Hellwig */ 3597391172c4SAl Viro lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL; 35985955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3599391172c4SAl Viro dentry = __lookup_hash(&last, path->dentry, lookup_flags); 36001da177e4SLinus Torvalds if (IS_ERR(dentry)) 3601a8104a9fSAl Viro goto unlock; 3602c663e5d8SChristoph Hellwig 3603a8104a9fSAl Viro error = -EEXIST; 3604b18825a7SDavid Howells if (d_is_positive(dentry)) 3605a8104a9fSAl Viro goto fail; 3606b18825a7SDavid Howells 3607c663e5d8SChristoph Hellwig /* 3608c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3609c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3610c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3611c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3612c663e5d8SChristoph Hellwig */ 3613391172c4SAl Viro if (unlikely(!is_dir && last.name[last.len])) { 3614a8104a9fSAl Viro error = -ENOENT; 3615ed75e95dSAl Viro goto fail; 3616e9baf6e5SAl Viro } 3617c30dabfeSJan Kara if (unlikely(err2)) { 3618c30dabfeSJan Kara error = err2; 3619a8104a9fSAl Viro goto fail; 3620c30dabfeSJan Kara } 3621181c37b6SAl Viro putname(name); 3622e9baf6e5SAl Viro return dentry; 36231da177e4SLinus Torvalds fail: 3624a8104a9fSAl Viro dput(dentry); 3625a8104a9fSAl Viro dentry = ERR_PTR(error); 3626a8104a9fSAl Viro unlock: 36275955102cSAl Viro inode_unlock(path->dentry->d_inode); 3628c30dabfeSJan Kara if (!err2) 3629391172c4SAl Viro mnt_drop_write(path->mnt); 3630ed75e95dSAl Viro out: 3631391172c4SAl Viro path_put(path); 3632181c37b6SAl Viro putname(name); 3633ed75e95dSAl Viro return dentry; 3634dae6ad8fSAl Viro } 3635fa14a0b8SAl Viro 3636fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3637fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3638fa14a0b8SAl Viro { 3639181c37b6SAl Viro return filename_create(dfd, getname_kernel(pathname), 3640181c37b6SAl Viro path, lookup_flags); 3641fa14a0b8SAl Viro } 3642dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3643dae6ad8fSAl Viro 3644921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3645921a1650SAl Viro { 3646921a1650SAl Viro dput(dentry); 36475955102cSAl Viro inode_unlock(path->dentry->d_inode); 3648a8104a9fSAl Viro mnt_drop_write(path->mnt); 3649921a1650SAl Viro path_put(path); 3650921a1650SAl Viro } 3651921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3652921a1650SAl Viro 3653520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 36541ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3655dae6ad8fSAl Viro { 3656181c37b6SAl Viro return filename_create(dfd, getname(pathname), path, lookup_flags); 3657dae6ad8fSAl Viro } 3658dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3659dae6ad8fSAl Viro 36606521f891SChristian Brauner /** 36616521f891SChristian Brauner * vfs_mknod - create device node or file 36626521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 36636521f891SChristian Brauner * @dir: inode of @dentry 36646521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 36656521f891SChristian Brauner * @mode: mode of the new device node or file 36666521f891SChristian Brauner * @dev: device number of device to create 36676521f891SChristian Brauner * 36686521f891SChristian Brauner * Create a device node or file. 36696521f891SChristian Brauner * 36706521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 36716521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 36726521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 36736521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 36746521f891SChristian Brauner * raw inode simply passs init_user_ns. 36756521f891SChristian Brauner */ 36766521f891SChristian Brauner int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, 36776521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 36781da177e4SLinus Torvalds { 3679a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 36806521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 36811da177e4SLinus Torvalds 36821da177e4SLinus Torvalds if (error) 36831da177e4SLinus Torvalds return error; 36841da177e4SLinus Torvalds 3685a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3686a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 36871da177e4SLinus Torvalds return -EPERM; 36881da177e4SLinus Torvalds 3689acfa4380SAl Viro if (!dir->i_op->mknod) 36901da177e4SLinus Torvalds return -EPERM; 36911da177e4SLinus Torvalds 369208ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 369308ce5f16SSerge E. Hallyn if (error) 369408ce5f16SSerge E. Hallyn return error; 369508ce5f16SSerge E. Hallyn 36961da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 36971da177e4SLinus Torvalds if (error) 36981da177e4SLinus Torvalds return error; 36991da177e4SLinus Torvalds 3700*549c7297SChristian Brauner error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev); 3701a74574aaSStephen Smalley if (!error) 3702f38aa942SAmy Griffis fsnotify_create(dir, dentry); 37031da177e4SLinus Torvalds return error; 37041da177e4SLinus Torvalds } 37054d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 37061da177e4SLinus Torvalds 3707f69aac00SAl Viro static int may_mknod(umode_t mode) 3708463c3197SDave Hansen { 3709463c3197SDave Hansen switch (mode & S_IFMT) { 3710463c3197SDave Hansen case S_IFREG: 3711463c3197SDave Hansen case S_IFCHR: 3712463c3197SDave Hansen case S_IFBLK: 3713463c3197SDave Hansen case S_IFIFO: 3714463c3197SDave Hansen case S_IFSOCK: 3715463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3716463c3197SDave Hansen return 0; 3717463c3197SDave Hansen case S_IFDIR: 3718463c3197SDave Hansen return -EPERM; 3719463c3197SDave Hansen default: 3720463c3197SDave Hansen return -EINVAL; 3721463c3197SDave Hansen } 3722463c3197SDave Hansen } 3723463c3197SDave Hansen 37245fee64fcSChristoph Hellwig static long do_mknodat(int dfd, const char __user *filename, umode_t mode, 372587c4e192SDominik Brodowski unsigned int dev) 37261da177e4SLinus Torvalds { 37276521f891SChristian Brauner struct user_namespace *mnt_userns; 37281da177e4SLinus Torvalds struct dentry *dentry; 3729dae6ad8fSAl Viro struct path path; 3730dae6ad8fSAl Viro int error; 3731972567f1SJeff Layton unsigned int lookup_flags = 0; 37321da177e4SLinus Torvalds 37338e4bfca1SAl Viro error = may_mknod(mode); 37348e4bfca1SAl Viro if (error) 37358e4bfca1SAl Viro return error; 3736972567f1SJeff Layton retry: 3737972567f1SJeff Layton dentry = user_path_create(dfd, filename, &path, lookup_flags); 3738dae6ad8fSAl Viro if (IS_ERR(dentry)) 3739dae6ad8fSAl Viro return PTR_ERR(dentry); 37402ad94ae6SAl Viro 3741dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3742ce3b0f8dSAl Viro mode &= ~current_umask(); 3743dae6ad8fSAl Viro error = security_path_mknod(&path, dentry, mode, dev); 3744be6d3e56SKentaro Takeda if (error) 3745a8104a9fSAl Viro goto out; 37466521f891SChristian Brauner 37476521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 37481da177e4SLinus Torvalds switch (mode & S_IFMT) { 37491da177e4SLinus Torvalds case 0: case S_IFREG: 37506521f891SChristian Brauner error = vfs_create(mnt_userns, path.dentry->d_inode, 37516521f891SChristian Brauner dentry, mode, true); 375205d1a717SMimi Zohar if (!error) 375305d1a717SMimi Zohar ima_post_path_mknod(dentry); 37541da177e4SLinus Torvalds break; 37551da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 37566521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 37576521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 37581da177e4SLinus Torvalds break; 37591da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 37606521f891SChristian Brauner error = vfs_mknod(mnt_userns, path.dentry->d_inode, 37616521f891SChristian Brauner dentry, mode, 0); 37621da177e4SLinus Torvalds break; 37631da177e4SLinus Torvalds } 3764a8104a9fSAl Viro out: 3765921a1650SAl Viro done_path_create(&path, dentry); 3766972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3767972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3768972567f1SJeff Layton goto retry; 3769972567f1SJeff Layton } 37701da177e4SLinus Torvalds return error; 37711da177e4SLinus Torvalds } 37721da177e4SLinus Torvalds 377387c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 377487c4e192SDominik Brodowski unsigned int, dev) 377587c4e192SDominik Brodowski { 377687c4e192SDominik Brodowski return do_mknodat(dfd, filename, mode, dev); 377787c4e192SDominik Brodowski } 377887c4e192SDominik Brodowski 37798208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 37805590ff0dSUlrich Drepper { 378187c4e192SDominik Brodowski return do_mknodat(AT_FDCWD, filename, mode, dev); 37825590ff0dSUlrich Drepper } 37835590ff0dSUlrich Drepper 37846521f891SChristian Brauner /** 37856521f891SChristian Brauner * vfs_mkdir - create directory 37866521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 37876521f891SChristian Brauner * @dir: inode of @dentry 37886521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 37896521f891SChristian Brauner * @mode: mode of the new directory 37906521f891SChristian Brauner * 37916521f891SChristian Brauner * Create a directory. 37926521f891SChristian Brauner * 37936521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 37946521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 37956521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 37966521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 37976521f891SChristian Brauner * raw inode simply passs init_user_ns. 37986521f891SChristian Brauner */ 37996521f891SChristian Brauner int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, 38006521f891SChristian Brauner struct dentry *dentry, umode_t mode) 38011da177e4SLinus Torvalds { 38026521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 38038de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 38041da177e4SLinus Torvalds 38051da177e4SLinus Torvalds if (error) 38061da177e4SLinus Torvalds return error; 38071da177e4SLinus Torvalds 3808acfa4380SAl Viro if (!dir->i_op->mkdir) 38091da177e4SLinus Torvalds return -EPERM; 38101da177e4SLinus Torvalds 38111da177e4SLinus Torvalds mode &= (S_IRWXUGO|S_ISVTX); 38121da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 38131da177e4SLinus Torvalds if (error) 38141da177e4SLinus Torvalds return error; 38151da177e4SLinus Torvalds 38168de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 38178de52778SAl Viro return -EMLINK; 38188de52778SAl Viro 3819*549c7297SChristian Brauner error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode); 3820a74574aaSStephen Smalley if (!error) 3821f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 38221da177e4SLinus Torvalds return error; 38231da177e4SLinus Torvalds } 38244d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 38251da177e4SLinus Torvalds 382683ff98c3SChristoph Hellwig static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode) 38271da177e4SLinus Torvalds { 38286902d925SDave Hansen struct dentry *dentry; 3829dae6ad8fSAl Viro struct path path; 3830dae6ad8fSAl Viro int error; 3831b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 38321da177e4SLinus Torvalds 3833b76d8b82SJeff Layton retry: 3834b76d8b82SJeff Layton dentry = user_path_create(dfd, pathname, &path, lookup_flags); 38356902d925SDave Hansen if (IS_ERR(dentry)) 3836dae6ad8fSAl Viro return PTR_ERR(dentry); 38376902d925SDave Hansen 3838dae6ad8fSAl Viro if (!IS_POSIXACL(path.dentry->d_inode)) 3839ce3b0f8dSAl Viro mode &= ~current_umask(); 3840dae6ad8fSAl Viro error = security_path_mkdir(&path, dentry, mode); 38416521f891SChristian Brauner if (!error) { 38426521f891SChristian Brauner struct user_namespace *mnt_userns; 38436521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 3844*549c7297SChristian Brauner error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry, 3845*549c7297SChristian Brauner mode); 38466521f891SChristian Brauner } 3847921a1650SAl Viro done_path_create(&path, dentry); 3848b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 3849b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 3850b76d8b82SJeff Layton goto retry; 3851b76d8b82SJeff Layton } 38521da177e4SLinus Torvalds return error; 38531da177e4SLinus Torvalds } 38541da177e4SLinus Torvalds 38550101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 38560101db7aSDominik Brodowski { 38570101db7aSDominik Brodowski return do_mkdirat(dfd, pathname, mode); 38580101db7aSDominik Brodowski } 38590101db7aSDominik Brodowski 3860a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 38615590ff0dSUlrich Drepper { 38620101db7aSDominik Brodowski return do_mkdirat(AT_FDCWD, pathname, mode); 38635590ff0dSUlrich Drepper } 38645590ff0dSUlrich Drepper 38656521f891SChristian Brauner /** 38666521f891SChristian Brauner * vfs_rmdir - remove directory 38676521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 38686521f891SChristian Brauner * @dir: inode of @dentry 38696521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 38706521f891SChristian Brauner * 38716521f891SChristian Brauner * Remove a directory. 38726521f891SChristian Brauner * 38736521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 38746521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 38756521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 38766521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 38776521f891SChristian Brauner * raw inode simply passs init_user_ns. 38786521f891SChristian Brauner */ 38796521f891SChristian Brauner int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir, 38806521f891SChristian Brauner struct dentry *dentry) 38811da177e4SLinus Torvalds { 38826521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 1); 38831da177e4SLinus Torvalds 38841da177e4SLinus Torvalds if (error) 38851da177e4SLinus Torvalds return error; 38861da177e4SLinus Torvalds 3887acfa4380SAl Viro if (!dir->i_op->rmdir) 38881da177e4SLinus Torvalds return -EPERM; 38891da177e4SLinus Torvalds 38901d2ef590SAl Viro dget(dentry); 38915955102cSAl Viro inode_lock(dentry->d_inode); 3892912dbc15SSage Weil 38931da177e4SLinus Torvalds error = -EBUSY; 38947af1364fSEric W. Biederman if (is_local_mountpoint(dentry)) 3895912dbc15SSage Weil goto out; 3896912dbc15SSage Weil 38971da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 3898912dbc15SSage Weil if (error) 3899912dbc15SSage Weil goto out; 3900912dbc15SSage Weil 39011da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 3902912dbc15SSage Weil if (error) 3903912dbc15SSage Weil goto out; 3904912dbc15SSage Weil 39058767712fSAl Viro shrink_dcache_parent(dentry); 39061da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 3907d83c49f3SAl Viro dont_mount(dentry); 39088ed936b5SEric W. Biederman detach_mounts(dentry); 3909116b9731SAmir Goldstein fsnotify_rmdir(dir, dentry); 39101da177e4SLinus Torvalds 3911912dbc15SSage Weil out: 39125955102cSAl Viro inode_unlock(dentry->d_inode); 39131d2ef590SAl Viro dput(dentry); 3914912dbc15SSage Weil if (!error) 3915912dbc15SSage Weil d_delete(dentry); 39161da177e4SLinus Torvalds return error; 39171da177e4SLinus Torvalds } 39184d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 39191da177e4SLinus Torvalds 3920e24ab0efSChristoph Hellwig long do_rmdir(int dfd, struct filename *name) 39211da177e4SLinus Torvalds { 39226521f891SChristian Brauner struct user_namespace *mnt_userns; 39231da177e4SLinus Torvalds int error = 0; 39241da177e4SLinus Torvalds struct dentry *dentry; 3925f5beed75SAl Viro struct path path; 3926f5beed75SAl Viro struct qstr last; 3927f5beed75SAl Viro int type; 3928c6ee9206SJeff Layton unsigned int lookup_flags = 0; 3929c6ee9206SJeff Layton retry: 3930e24ab0efSChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, 3931c1d4dd27SAl Viro &path, &last, &type); 393291a27b2aSJeff Layton if (IS_ERR(name)) 393391a27b2aSJeff Layton return PTR_ERR(name); 39341da177e4SLinus Torvalds 3935f5beed75SAl Viro switch (type) { 39361da177e4SLinus Torvalds case LAST_DOTDOT: 39371da177e4SLinus Torvalds error = -ENOTEMPTY; 39381da177e4SLinus Torvalds goto exit1; 39391da177e4SLinus Torvalds case LAST_DOT: 39401da177e4SLinus Torvalds error = -EINVAL; 39411da177e4SLinus Torvalds goto exit1; 39421da177e4SLinus Torvalds case LAST_ROOT: 39431da177e4SLinus Torvalds error = -EBUSY; 39441da177e4SLinus Torvalds goto exit1; 39451da177e4SLinus Torvalds } 39460612d9fbSOGAWA Hirofumi 3947f5beed75SAl Viro error = mnt_want_write(path.mnt); 3948c30dabfeSJan Kara if (error) 3949c30dabfeSJan Kara goto exit1; 39500612d9fbSOGAWA Hirofumi 39515955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 3952f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 39531da177e4SLinus Torvalds error = PTR_ERR(dentry); 39546902d925SDave Hansen if (IS_ERR(dentry)) 39556902d925SDave Hansen goto exit2; 3956e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 3957e6bc45d6STheodore Ts'o error = -ENOENT; 3958e6bc45d6STheodore Ts'o goto exit3; 3959e6bc45d6STheodore Ts'o } 3960f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 3961be6d3e56SKentaro Takeda if (error) 3962c30dabfeSJan Kara goto exit3; 39636521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 39646521f891SChristian Brauner error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry); 39650622753bSDave Hansen exit3: 39661da177e4SLinus Torvalds dput(dentry); 39676902d925SDave Hansen exit2: 39685955102cSAl Viro inode_unlock(path.dentry->d_inode); 3969f5beed75SAl Viro mnt_drop_write(path.mnt); 39701da177e4SLinus Torvalds exit1: 3971f5beed75SAl Viro path_put(&path); 3972c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 3973c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 3974c6ee9206SJeff Layton goto retry; 3975c6ee9206SJeff Layton } 397624fb33d4SAl Viro putname(name); 39771da177e4SLinus Torvalds return error; 39781da177e4SLinus Torvalds } 39791da177e4SLinus Torvalds 39803cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 39815590ff0dSUlrich Drepper { 3982e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 39835590ff0dSUlrich Drepper } 39845590ff0dSUlrich Drepper 3985b21996e3SJ. Bruce Fields /** 3986b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 39876521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 3988b21996e3SJ. Bruce Fields * @dir: parent directory 3989b21996e3SJ. Bruce Fields * @dentry: victim 3990b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 3991b21996e3SJ. Bruce Fields * 3992b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 3993b21996e3SJ. Bruce Fields * 3994b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 3995b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 3996b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 3997b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 3998b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 3999b21996e3SJ. Bruce Fields * 4000b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4001b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4002b21996e3SJ. Bruce Fields * to be NFS exported. 40036521f891SChristian Brauner * 40046521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 40056521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 40066521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 40076521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 40086521f891SChristian Brauner * raw inode simply passs init_user_ns. 4009b21996e3SJ. Bruce Fields */ 40106521f891SChristian Brauner int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir, 40116521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 40121da177e4SLinus Torvalds { 40139accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 40146521f891SChristian Brauner int error = may_delete(mnt_userns, dir, dentry, 0); 40151da177e4SLinus Torvalds 40161da177e4SLinus Torvalds if (error) 40171da177e4SLinus Torvalds return error; 40181da177e4SLinus Torvalds 4019acfa4380SAl Viro if (!dir->i_op->unlink) 40201da177e4SLinus Torvalds return -EPERM; 40211da177e4SLinus Torvalds 40225955102cSAl Viro inode_lock(target); 40238ed936b5SEric W. Biederman if (is_local_mountpoint(dentry)) 40241da177e4SLinus Torvalds error = -EBUSY; 40251da177e4SLinus Torvalds else { 40261da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4027bec1052eSAl Viro if (!error) { 40285a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 40295a14696cSJ. Bruce Fields if (error) 4030b21996e3SJ. Bruce Fields goto out; 40311da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 40328ed936b5SEric W. Biederman if (!error) { 4033d83c49f3SAl Viro dont_mount(dentry); 40348ed936b5SEric W. Biederman detach_mounts(dentry); 4035116b9731SAmir Goldstein fsnotify_unlink(dir, dentry); 40368ed936b5SEric W. Biederman } 4037bec1052eSAl Viro } 40381da177e4SLinus Torvalds } 4039b21996e3SJ. Bruce Fields out: 40405955102cSAl Viro inode_unlock(target); 40411da177e4SLinus Torvalds 40421da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 40431da177e4SLinus Torvalds if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) { 40449accbb97SJ. Bruce Fields fsnotify_link_count(target); 40451da177e4SLinus Torvalds d_delete(dentry); 40461da177e4SLinus Torvalds } 40470eeca283SRobert Love 40481da177e4SLinus Torvalds return error; 40491da177e4SLinus Torvalds } 40504d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 40511da177e4SLinus Torvalds 40521da177e4SLinus Torvalds /* 40531da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 40541b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 40551da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 40561da177e4SLinus Torvalds * while waiting on the I/O. 40571da177e4SLinus Torvalds */ 4058da2f1362SChristoph Hellwig long do_unlinkat(int dfd, struct filename *name) 40591da177e4SLinus Torvalds { 40602ad94ae6SAl Viro int error; 40611da177e4SLinus Torvalds struct dentry *dentry; 4062f5beed75SAl Viro struct path path; 4063f5beed75SAl Viro struct qstr last; 4064f5beed75SAl Viro int type; 40651da177e4SLinus Torvalds struct inode *inode = NULL; 4066b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 40675d18f813SJeff Layton unsigned int lookup_flags = 0; 40685d18f813SJeff Layton retry: 4069da2f1362SChristoph Hellwig name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 407091a27b2aSJeff Layton if (IS_ERR(name)) 407191a27b2aSJeff Layton return PTR_ERR(name); 40722ad94ae6SAl Viro 40731da177e4SLinus Torvalds error = -EISDIR; 4074f5beed75SAl Viro if (type != LAST_NORM) 40751da177e4SLinus Torvalds goto exit1; 40760612d9fbSOGAWA Hirofumi 4077f5beed75SAl Viro error = mnt_want_write(path.mnt); 4078c30dabfeSJan Kara if (error) 4079c30dabfeSJan Kara goto exit1; 4080b21996e3SJ. Bruce Fields retry_deleg: 40815955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4082f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 40831da177e4SLinus Torvalds error = PTR_ERR(dentry); 40841da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 40856521f891SChristian Brauner struct user_namespace *mnt_userns; 40866521f891SChristian Brauner 40871da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4088f5beed75SAl Viro if (last.name[last.len]) 408950338b88STörök Edwin goto slashes; 40901da177e4SLinus Torvalds inode = dentry->d_inode; 4091b18825a7SDavid Howells if (d_is_negative(dentry)) 4092e6bc45d6STheodore Ts'o goto slashes; 40937de9c6eeSAl Viro ihold(inode); 4094f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4095be6d3e56SKentaro Takeda if (error) 4096c30dabfeSJan Kara goto exit2; 40976521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 4098*549c7297SChristian Brauner error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry, 4099*549c7297SChristian Brauner &delegated_inode); 41001da177e4SLinus Torvalds exit2: 41011da177e4SLinus Torvalds dput(dentry); 41021da177e4SLinus Torvalds } 41035955102cSAl Viro inode_unlock(path.dentry->d_inode); 41041da177e4SLinus Torvalds if (inode) 41051da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4106b21996e3SJ. Bruce Fields inode = NULL; 4107b21996e3SJ. Bruce Fields if (delegated_inode) { 41085a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4109b21996e3SJ. Bruce Fields if (!error) 4110b21996e3SJ. Bruce Fields goto retry_deleg; 4111b21996e3SJ. Bruce Fields } 4112f5beed75SAl Viro mnt_drop_write(path.mnt); 41131da177e4SLinus Torvalds exit1: 4114f5beed75SAl Viro path_put(&path); 41155d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 41165d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 41175d18f813SJeff Layton inode = NULL; 41185d18f813SJeff Layton goto retry; 41195d18f813SJeff Layton } 4120da2f1362SChristoph Hellwig putname(name); 41211da177e4SLinus Torvalds return error; 41221da177e4SLinus Torvalds 41231da177e4SLinus Torvalds slashes: 4124b18825a7SDavid Howells if (d_is_negative(dentry)) 4125b18825a7SDavid Howells error = -ENOENT; 412644b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4127b18825a7SDavid Howells error = -EISDIR; 4128b18825a7SDavid Howells else 4129b18825a7SDavid Howells error = -ENOTDIR; 41301da177e4SLinus Torvalds goto exit2; 41311da177e4SLinus Torvalds } 41321da177e4SLinus Torvalds 41332e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 41345590ff0dSUlrich Drepper { 41355590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 41365590ff0dSUlrich Drepper return -EINVAL; 41375590ff0dSUlrich Drepper 41385590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4139e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4140da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 41415590ff0dSUlrich Drepper } 41425590ff0dSUlrich Drepper 41433480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 41445590ff0dSUlrich Drepper { 4145da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 41465590ff0dSUlrich Drepper } 41475590ff0dSUlrich Drepper 41486521f891SChristian Brauner /** 41496521f891SChristian Brauner * vfs_symlink - create symlink 41506521f891SChristian Brauner * @mnt_userns: user namespace of the mount the inode was found from 41516521f891SChristian Brauner * @dir: inode of @dentry 41526521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 41536521f891SChristian Brauner * @oldname: name of the file to link to 41546521f891SChristian Brauner * 41556521f891SChristian Brauner * Create a symlink. 41566521f891SChristian Brauner * 41576521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 41586521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 41596521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 41606521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 41616521f891SChristian Brauner * raw inode simply passs init_user_ns. 41626521f891SChristian Brauner */ 41636521f891SChristian Brauner int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir, 41646521f891SChristian Brauner struct dentry *dentry, const char *oldname) 41651da177e4SLinus Torvalds { 41666521f891SChristian Brauner int error = may_create(mnt_userns, dir, dentry); 41671da177e4SLinus Torvalds 41681da177e4SLinus Torvalds if (error) 41691da177e4SLinus Torvalds return error; 41701da177e4SLinus Torvalds 4171acfa4380SAl Viro if (!dir->i_op->symlink) 41721da177e4SLinus Torvalds return -EPERM; 41731da177e4SLinus Torvalds 41741da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 41751da177e4SLinus Torvalds if (error) 41761da177e4SLinus Torvalds return error; 41771da177e4SLinus Torvalds 4178*549c7297SChristian Brauner error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname); 4179a74574aaSStephen Smalley if (!error) 4180f38aa942SAmy Griffis fsnotify_create(dir, dentry); 41811da177e4SLinus Torvalds return error; 41821da177e4SLinus Torvalds } 41834d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 41841da177e4SLinus Torvalds 4185cd3acb6aSChristoph Hellwig static long do_symlinkat(const char __user *oldname, int newdfd, 4186b724e846SDominik Brodowski const char __user *newname) 41871da177e4SLinus Torvalds { 41882ad94ae6SAl Viro int error; 418991a27b2aSJeff Layton struct filename *from; 41906902d925SDave Hansen struct dentry *dentry; 4191dae6ad8fSAl Viro struct path path; 4192f46d3567SJeff Layton unsigned int lookup_flags = 0; 41931da177e4SLinus Torvalds 41941da177e4SLinus Torvalds from = getname(oldname); 41951da177e4SLinus Torvalds if (IS_ERR(from)) 41961da177e4SLinus Torvalds return PTR_ERR(from); 4197f46d3567SJeff Layton retry: 4198f46d3567SJeff Layton dentry = user_path_create(newdfd, newname, &path, lookup_flags); 41991da177e4SLinus Torvalds error = PTR_ERR(dentry); 42006902d925SDave Hansen if (IS_ERR(dentry)) 4201dae6ad8fSAl Viro goto out_putname; 42026902d925SDave Hansen 420391a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 42046521f891SChristian Brauner if (!error) { 42056521f891SChristian Brauner struct user_namespace *mnt_userns; 42066521f891SChristian Brauner 42076521f891SChristian Brauner mnt_userns = mnt_user_ns(path.mnt); 42086521f891SChristian Brauner error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry, 42096521f891SChristian Brauner from->name); 42106521f891SChristian Brauner } 4211921a1650SAl Viro done_path_create(&path, dentry); 4212f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4213f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4214f46d3567SJeff Layton goto retry; 4215f46d3567SJeff Layton } 42166902d925SDave Hansen out_putname: 42171da177e4SLinus Torvalds putname(from); 42181da177e4SLinus Torvalds return error; 42191da177e4SLinus Torvalds } 42201da177e4SLinus Torvalds 4221b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4222b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4223b724e846SDominik Brodowski { 4224b724e846SDominik Brodowski return do_symlinkat(oldname, newdfd, newname); 4225b724e846SDominik Brodowski } 4226b724e846SDominik Brodowski 42273480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 42285590ff0dSUlrich Drepper { 4229b724e846SDominik Brodowski return do_symlinkat(oldname, AT_FDCWD, newname); 42305590ff0dSUlrich Drepper } 42315590ff0dSUlrich Drepper 4232146a8595SJ. Bruce Fields /** 4233146a8595SJ. Bruce Fields * vfs_link - create a new link 4234146a8595SJ. Bruce Fields * @old_dentry: object to be linked 42356521f891SChristian Brauner * @mnt_userns: the user namespace of the mount 4236146a8595SJ. Bruce Fields * @dir: new parent 4237146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4238146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4239146a8595SJ. Bruce Fields * 4240146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4241146a8595SJ. Bruce Fields * 4242146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4243146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4244146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4245146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4246146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4247146a8595SJ. Bruce Fields * 4248146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4249146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4250146a8595SJ. Bruce Fields * to be NFS exported. 42516521f891SChristian Brauner * 42526521f891SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 42536521f891SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 42546521f891SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 42556521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 42566521f891SChristian Brauner * raw inode simply passs init_user_ns. 4257146a8595SJ. Bruce Fields */ 42586521f891SChristian Brauner int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns, 42596521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 42606521f891SChristian Brauner struct inode **delegated_inode) 42611da177e4SLinus Torvalds { 42621da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 42638de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 42641da177e4SLinus Torvalds int error; 42651da177e4SLinus Torvalds 42661da177e4SLinus Torvalds if (!inode) 42671da177e4SLinus Torvalds return -ENOENT; 42681da177e4SLinus Torvalds 42696521f891SChristian Brauner error = may_create(mnt_userns, dir, new_dentry); 42701da177e4SLinus Torvalds if (error) 42711da177e4SLinus Torvalds return error; 42721da177e4SLinus Torvalds 42731da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 42741da177e4SLinus Torvalds return -EXDEV; 42751da177e4SLinus Torvalds 42761da177e4SLinus Torvalds /* 42771da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 42781da177e4SLinus Torvalds */ 42791da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 42801da177e4SLinus Torvalds return -EPERM; 42810bd23d09SEric W. Biederman /* 42820bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 42830bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 42840bd23d09SEric W. Biederman * the vfs. 42850bd23d09SEric W. Biederman */ 42866521f891SChristian Brauner if (HAS_UNMAPPED_ID(mnt_userns, inode)) 42870bd23d09SEric W. Biederman return -EPERM; 4288acfa4380SAl Viro if (!dir->i_op->link) 42891da177e4SLinus Torvalds return -EPERM; 42907e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 42911da177e4SLinus Torvalds return -EPERM; 42921da177e4SLinus Torvalds 42931da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 42941da177e4SLinus Torvalds if (error) 42951da177e4SLinus Torvalds return error; 42961da177e4SLinus Torvalds 42975955102cSAl Viro inode_lock(inode); 4298aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4299f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4300aae8a97dSAneesh Kumar K.V error = -ENOENT; 43018de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 43028de52778SAl Viro error = -EMLINK; 4303146a8595SJ. Bruce Fields else { 4304146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4305146a8595SJ. Bruce Fields if (!error) 43061da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4307146a8595SJ. Bruce Fields } 4308f4e0c30cSAl Viro 4309f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4310f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4311f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4312f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4313f4e0c30cSAl Viro } 43145955102cSAl Viro inode_unlock(inode); 4315e31e14ecSStephen Smalley if (!error) 43167e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 43171da177e4SLinus Torvalds return error; 43181da177e4SLinus Torvalds } 43194d359507SAl Viro EXPORT_SYMBOL(vfs_link); 43201da177e4SLinus Torvalds 43211da177e4SLinus Torvalds /* 43221da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 43231da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 43241da177e4SLinus Torvalds * newname. --KAB 43251da177e4SLinus Torvalds * 43261da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 43271da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 43281da177e4SLinus Torvalds * and other special files. --ADM 43291da177e4SLinus Torvalds */ 4330812931d6SChristoph Hellwig static int do_linkat(int olddfd, const char __user *oldname, int newdfd, 433146ea89ebSDominik Brodowski const char __user *newname, int flags) 43321da177e4SLinus Torvalds { 43336521f891SChristian Brauner struct user_namespace *mnt_userns; 43341da177e4SLinus Torvalds struct dentry *new_dentry; 4335dae6ad8fSAl Viro struct path old_path, new_path; 4336146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 433711a7b371SAneesh Kumar K.V int how = 0; 43381da177e4SLinus Torvalds int error; 43391da177e4SLinus Torvalds 434011a7b371SAneesh Kumar K.V if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) 4341c04030e1SUlrich Drepper return -EINVAL; 434211a7b371SAneesh Kumar K.V /* 4343f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4344f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4345f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 434611a7b371SAneesh Kumar K.V */ 4347f0cc6ffbSLinus Torvalds if (flags & AT_EMPTY_PATH) { 4348f0cc6ffbSLinus Torvalds if (!capable(CAP_DAC_READ_SEARCH)) 4349f0cc6ffbSLinus Torvalds return -ENOENT; 435011a7b371SAneesh Kumar K.V how = LOOKUP_EMPTY; 4351f0cc6ffbSLinus Torvalds } 4352c04030e1SUlrich Drepper 435311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 435411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4355442e31caSJeff Layton retry: 435611a7b371SAneesh Kumar K.V error = user_path_at(olddfd, oldname, how, &old_path); 43571da177e4SLinus Torvalds if (error) 43582ad94ae6SAl Viro return error; 43592ad94ae6SAl Viro 4360442e31caSJeff Layton new_dentry = user_path_create(newdfd, newname, &new_path, 4361442e31caSJeff Layton (how & LOOKUP_REVAL)); 43621da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 43636902d925SDave Hansen if (IS_ERR(new_dentry)) 4364dae6ad8fSAl Viro goto out; 4365dae6ad8fSAl Viro 4366dae6ad8fSAl Viro error = -EXDEV; 4367dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4368dae6ad8fSAl Viro goto out_dput; 4369*549c7297SChristian Brauner mnt_userns = mnt_user_ns(new_path.mnt); 4370*549c7297SChristian Brauner error = may_linkat(mnt_userns, &old_path); 4371800179c9SKees Cook if (unlikely(error)) 4372800179c9SKees Cook goto out_dput; 4373dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4374be6d3e56SKentaro Takeda if (error) 4375a8104a9fSAl Viro goto out_dput; 43766521f891SChristian Brauner error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode, 43776521f891SChristian Brauner new_dentry, &delegated_inode); 437875c3f29dSDave Hansen out_dput: 4379921a1650SAl Viro done_path_create(&new_path, new_dentry); 4380146a8595SJ. Bruce Fields if (delegated_inode) { 4381146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4382d22e6338SOleg Drokin if (!error) { 4383d22e6338SOleg Drokin path_put(&old_path); 4384146a8595SJ. Bruce Fields goto retry; 4385146a8595SJ. Bruce Fields } 4386d22e6338SOleg Drokin } 4387442e31caSJeff Layton if (retry_estale(error, how)) { 4388d22e6338SOleg Drokin path_put(&old_path); 4389442e31caSJeff Layton how |= LOOKUP_REVAL; 4390442e31caSJeff Layton goto retry; 4391442e31caSJeff Layton } 43921da177e4SLinus Torvalds out: 43932d8f3038SAl Viro path_put(&old_path); 43941da177e4SLinus Torvalds 43951da177e4SLinus Torvalds return error; 43961da177e4SLinus Torvalds } 43971da177e4SLinus Torvalds 439846ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 439946ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 440046ea89ebSDominik Brodowski { 440146ea89ebSDominik Brodowski return do_linkat(olddfd, oldname, newdfd, newname, flags); 440246ea89ebSDominik Brodowski } 440346ea89ebSDominik Brodowski 44043480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 44055590ff0dSUlrich Drepper { 440646ea89ebSDominik Brodowski return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0); 44075590ff0dSUlrich Drepper } 44085590ff0dSUlrich Drepper 4409bc27027aSMiklos Szeredi /** 4410bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 44116521f891SChristian Brauner * @old_mnt_userns: old user namespace of the mount the inode was found from 4412bc27027aSMiklos Szeredi * @old_dir: parent of source 4413bc27027aSMiklos Szeredi * @old_dentry: source 44146521f891SChristian Brauner * @new_mnt_userns: new user namespace of the mount the inode was found from 4415bc27027aSMiklos Szeredi * @new_dir: parent of destination 4416bc27027aSMiklos Szeredi * @new_dentry: destination 4417bc27027aSMiklos Szeredi * @delegated_inode: returns an inode needing a delegation break 4418520c8b16SMiklos Szeredi * @flags: rename flags 4419bc27027aSMiklos Szeredi * 4420bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4421bc27027aSMiklos Szeredi * 4422bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4423bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4424bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4425bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4426bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4427bc27027aSMiklos Szeredi * so. 4428bc27027aSMiklos Szeredi * 4429bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4430bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4431bc27027aSMiklos Szeredi * to be NFS exported. 4432bc27027aSMiklos Szeredi * 44331da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 44341da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 44351da177e4SLinus Torvalds * Problems: 44360117d427SMauro Carvalho Chehab * 4437d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 44381da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 44391da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4440a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 44411da177e4SLinus Torvalds * story. 44426cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 44436cedba89SJ. Bruce Fields * and source (if it is not a directory). 44441b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 44451da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 44461da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4447a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 44481da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 44491da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 44501da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4451a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 44521da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 44531da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 44541da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4455e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 44561b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 44571da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4458c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 44591da177e4SLinus Torvalds * locking]. 44601da177e4SLinus Torvalds */ 44619fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 44621da177e4SLinus Torvalds { 44631da177e4SLinus Torvalds int error; 44649fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 44659fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 44669fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 44679fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 44689fe61450SChristian Brauner unsigned int flags = rd->flags; 4469bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4470bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4471bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4472da1ce067SMiklos Szeredi bool new_is_dir = false; 4473da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 447449d31c2fSAl Viro struct name_snapshot old_name; 44751da177e4SLinus Torvalds 44768d3e2936SMiklos Szeredi if (source == target) 44771da177e4SLinus Torvalds return 0; 44781da177e4SLinus Torvalds 44796521f891SChristian Brauner error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir); 44801da177e4SLinus Torvalds if (error) 44811da177e4SLinus Torvalds return error; 44821da177e4SLinus Torvalds 4483da1ce067SMiklos Szeredi if (!target) { 44846521f891SChristian Brauner error = may_create(rd->new_mnt_userns, new_dir, new_dentry); 4485da1ce067SMiklos Szeredi } else { 4486da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4487da1ce067SMiklos Szeredi 4488da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 44896521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 44906521f891SChristian Brauner new_dentry, is_dir); 4491da1ce067SMiklos Szeredi else 44926521f891SChristian Brauner error = may_delete(rd->new_mnt_userns, new_dir, 44936521f891SChristian Brauner new_dentry, new_is_dir); 4494da1ce067SMiklos Szeredi } 44951da177e4SLinus Torvalds if (error) 44961da177e4SLinus Torvalds return error; 44971da177e4SLinus Torvalds 44982773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 44991da177e4SLinus Torvalds return -EPERM; 45001da177e4SLinus Torvalds 4501bc27027aSMiklos Szeredi /* 4502bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4503bc27027aSMiklos Szeredi * we'll need to flip '..'. 4504bc27027aSMiklos Szeredi */ 4505da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4506da1ce067SMiklos Szeredi if (is_dir) { 45076521f891SChristian Brauner error = inode_permission(rd->old_mnt_userns, source, 450847291baaSChristian Brauner MAY_WRITE); 4509bc27027aSMiklos Szeredi if (error) 4510bc27027aSMiklos Szeredi return error; 4511bc27027aSMiklos Szeredi } 4512da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 45136521f891SChristian Brauner error = inode_permission(rd->new_mnt_userns, target, 451447291baaSChristian Brauner MAY_WRITE); 4515da1ce067SMiklos Szeredi if (error) 4516da1ce067SMiklos Szeredi return error; 4517da1ce067SMiklos Szeredi } 4518da1ce067SMiklos Szeredi } 45190eeca283SRobert Love 45200b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 45210b3974ebSMiklos Szeredi flags); 4522bc27027aSMiklos Szeredi if (error) 4523bc27027aSMiklos Szeredi return error; 4524bc27027aSMiklos Szeredi 452549d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4526bc27027aSMiklos Szeredi dget(new_dentry); 4527da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4528bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4529bc27027aSMiklos Szeredi else if (target) 45305955102cSAl Viro inode_lock(target); 4531bc27027aSMiklos Szeredi 4532bc27027aSMiklos Szeredi error = -EBUSY; 45337af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4534bc27027aSMiklos Szeredi goto out; 4535bc27027aSMiklos Szeredi 4536da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4537bc27027aSMiklos Szeredi error = -EMLINK; 4538da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4539bc27027aSMiklos Szeredi goto out; 4540da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4541da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4542da1ce067SMiklos Szeredi goto out; 4543da1ce067SMiklos Szeredi } 4544da1ce067SMiklos Szeredi if (!is_dir) { 4545bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4546bc27027aSMiklos Szeredi if (error) 4547bc27027aSMiklos Szeredi goto out; 4548da1ce067SMiklos Szeredi } 4549da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4550bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4551bc27027aSMiklos Szeredi if (error) 4552bc27027aSMiklos Szeredi goto out; 4553bc27027aSMiklos Szeredi } 4554*549c7297SChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry, 4555520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4556bc27027aSMiklos Szeredi if (error) 4557bc27027aSMiklos Szeredi goto out; 4558bc27027aSMiklos Szeredi 4559da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 45608767712fSAl Viro if (is_dir) { 45618767712fSAl Viro shrink_dcache_parent(new_dentry); 4562bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 45638767712fSAl Viro } 4564bc27027aSMiklos Szeredi dont_mount(new_dentry); 45658ed936b5SEric W. Biederman detach_mounts(new_dentry); 4566bc27027aSMiklos Szeredi } 4567da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4568da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4569bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4570da1ce067SMiklos Szeredi else 4571da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4572da1ce067SMiklos Szeredi } 4573bc27027aSMiklos Szeredi out: 4574da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4575bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4576bc27027aSMiklos Szeredi else if (target) 45775955102cSAl Viro inode_unlock(target); 4578bc27027aSMiklos Szeredi dput(new_dentry); 4579da1ce067SMiklos Szeredi if (!error) { 4580f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4581da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4582da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4583f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4584da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4585da1ce067SMiklos Szeredi } 4586da1ce067SMiklos Szeredi } 458749d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 45880eeca283SRobert Love 45891da177e4SLinus Torvalds return error; 45901da177e4SLinus Torvalds } 45914d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 45921da177e4SLinus Torvalds 4593e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4594e886663cSJens Axboe struct filename *to, unsigned int flags) 45951da177e4SLinus Torvalds { 45969fe61450SChristian Brauner struct renamedata rd; 45971da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 45981da177e4SLinus Torvalds struct dentry *trap; 4599f5beed75SAl Viro struct path old_path, new_path; 4600f5beed75SAl Viro struct qstr old_last, new_last; 4601f5beed75SAl Viro int old_type, new_type; 46028e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4603f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4604c6a94284SJeff Layton bool should_retry = false; 4605e886663cSJens Axboe int error = -EINVAL; 4606520c8b16SMiklos Szeredi 46070d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 4608e886663cSJens Axboe goto put_both; 4609da1ce067SMiklos Szeredi 46100d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 46110d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 4612e886663cSJens Axboe goto put_both; 4613520c8b16SMiklos Szeredi 4614f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4615f5beed75SAl Viro target_flags = 0; 4616f5beed75SAl Viro 4617c6a94284SJeff Layton retry: 4618e886663cSJens Axboe from = filename_parentat(olddfd, from, lookup_flags, &old_path, 4619e886663cSJens Axboe &old_last, &old_type); 462091a27b2aSJeff Layton if (IS_ERR(from)) { 462191a27b2aSJeff Layton error = PTR_ERR(from); 4622e886663cSJens Axboe goto put_new; 462391a27b2aSJeff Layton } 46241da177e4SLinus Torvalds 4625e886663cSJens Axboe to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4626e886663cSJens Axboe &new_type); 462791a27b2aSJeff Layton if (IS_ERR(to)) { 462891a27b2aSJeff Layton error = PTR_ERR(to); 46291da177e4SLinus Torvalds goto exit1; 463091a27b2aSJeff Layton } 46311da177e4SLinus Torvalds 46321da177e4SLinus Torvalds error = -EXDEV; 4633f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 46341da177e4SLinus Torvalds goto exit2; 46351da177e4SLinus Torvalds 46361da177e4SLinus Torvalds error = -EBUSY; 4637f5beed75SAl Viro if (old_type != LAST_NORM) 46381da177e4SLinus Torvalds goto exit2; 46391da177e4SLinus Torvalds 46400a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 46410a7c3937SMiklos Szeredi error = -EEXIST; 4642f5beed75SAl Viro if (new_type != LAST_NORM) 46431da177e4SLinus Torvalds goto exit2; 46441da177e4SLinus Torvalds 4645f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4646c30dabfeSJan Kara if (error) 4647c30dabfeSJan Kara goto exit2; 4648c30dabfeSJan Kara 46498e6d782cSJ. Bruce Fields retry_deleg: 4650f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 46511da177e4SLinus Torvalds 4652f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 46531da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 46541da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 46551da177e4SLinus Torvalds goto exit3; 46561da177e4SLinus Torvalds /* source must exist */ 46571da177e4SLinus Torvalds error = -ENOENT; 4658b18825a7SDavid Howells if (d_is_negative(old_dentry)) 46591da177e4SLinus Torvalds goto exit4; 4660f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 46611da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 46621da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 46631da177e4SLinus Torvalds goto exit4; 46640a7c3937SMiklos Szeredi error = -EEXIST; 46650a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 46660a7c3937SMiklos Szeredi goto exit5; 4667da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4668da1ce067SMiklos Szeredi error = -ENOENT; 4669da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4670da1ce067SMiklos Szeredi goto exit5; 4671da1ce067SMiklos Szeredi 4672da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4673da1ce067SMiklos Szeredi error = -ENOTDIR; 4674f5beed75SAl Viro if (new_last.name[new_last.len]) 4675da1ce067SMiklos Szeredi goto exit5; 4676da1ce067SMiklos Szeredi } 4677da1ce067SMiklos Szeredi } 46780a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 46790a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 46800a7c3937SMiklos Szeredi error = -ENOTDIR; 4681f5beed75SAl Viro if (old_last.name[old_last.len]) 46820a7c3937SMiklos Szeredi goto exit5; 4683f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 46840a7c3937SMiklos Szeredi goto exit5; 46850a7c3937SMiklos Szeredi } 46860a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 46870a7c3937SMiklos Szeredi error = -EINVAL; 46880a7c3937SMiklos Szeredi if (old_dentry == trap) 46890a7c3937SMiklos Szeredi goto exit5; 46901da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4691da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 46921da177e4SLinus Torvalds error = -ENOTEMPTY; 46931da177e4SLinus Torvalds if (new_dentry == trap) 46941da177e4SLinus Torvalds goto exit5; 46951da177e4SLinus Torvalds 4696f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4697f5beed75SAl Viro &new_path, new_dentry, flags); 4698be6d3e56SKentaro Takeda if (error) 4699c30dabfeSJan Kara goto exit5; 47009fe61450SChristian Brauner 47019fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 47029fe61450SChristian Brauner rd.old_dentry = old_dentry; 47036521f891SChristian Brauner rd.old_mnt_userns = mnt_user_ns(old_path.mnt); 47049fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 47059fe61450SChristian Brauner rd.new_dentry = new_dentry; 47066521f891SChristian Brauner rd.new_mnt_userns = mnt_user_ns(new_path.mnt); 47079fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 47089fe61450SChristian Brauner rd.flags = flags; 47099fe61450SChristian Brauner error = vfs_rename(&rd); 47101da177e4SLinus Torvalds exit5: 47111da177e4SLinus Torvalds dput(new_dentry); 47121da177e4SLinus Torvalds exit4: 47131da177e4SLinus Torvalds dput(old_dentry); 47141da177e4SLinus Torvalds exit3: 4715f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 47168e6d782cSJ. Bruce Fields if (delegated_inode) { 47178e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 47188e6d782cSJ. Bruce Fields if (!error) 47198e6d782cSJ. Bruce Fields goto retry_deleg; 47208e6d782cSJ. Bruce Fields } 4721f5beed75SAl Viro mnt_drop_write(old_path.mnt); 47221da177e4SLinus Torvalds exit2: 4723c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4724c6a94284SJeff Layton should_retry = true; 4725f5beed75SAl Viro path_put(&new_path); 47261da177e4SLinus Torvalds exit1: 4727f5beed75SAl Viro path_put(&old_path); 4728c6a94284SJeff Layton if (should_retry) { 4729c6a94284SJeff Layton should_retry = false; 4730c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4731c6a94284SJeff Layton goto retry; 4732c6a94284SJeff Layton } 4733e886663cSJens Axboe put_both: 4734e886663cSJens Axboe if (!IS_ERR(from)) 4735e886663cSJens Axboe putname(from); 4736e886663cSJens Axboe put_new: 4737e886663cSJens Axboe if (!IS_ERR(to)) 4738e886663cSJens Axboe putname(to); 47391da177e4SLinus Torvalds return error; 47401da177e4SLinus Torvalds } 47411da177e4SLinus Torvalds 4742ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4743ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4744ee81feb6SDominik Brodowski { 4745e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4746e886663cSJens Axboe flags); 4747ee81feb6SDominik Brodowski } 4748ee81feb6SDominik Brodowski 4749520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4750520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4751520c8b16SMiklos Szeredi { 4752e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4753e886663cSJens Axboe 0); 4754520c8b16SMiklos Szeredi } 4755520c8b16SMiklos Szeredi 4756a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 47575590ff0dSUlrich Drepper { 4758e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4759e886663cSJens Axboe getname(newname), 0); 47605590ff0dSUlrich Drepper } 47615590ff0dSUlrich Drepper 47625d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 47631da177e4SLinus Torvalds { 47645d826c84SAl Viro int len = PTR_ERR(link); 47651da177e4SLinus Torvalds if (IS_ERR(link)) 47661da177e4SLinus Torvalds goto out; 47671da177e4SLinus Torvalds 47681da177e4SLinus Torvalds len = strlen(link); 47691da177e4SLinus Torvalds if (len > (unsigned) buflen) 47701da177e4SLinus Torvalds len = buflen; 47711da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 47721da177e4SLinus Torvalds len = -EFAULT; 47731da177e4SLinus Torvalds out: 47741da177e4SLinus Torvalds return len; 47751da177e4SLinus Torvalds } 47761da177e4SLinus Torvalds 4777d60874cdSMiklos Szeredi /** 4778fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 4779fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4780fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 4781fd4a0edfSMiklos Szeredi * @buflen: size of buffer 4782fd4a0edfSMiklos Szeredi * 4783fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 4784fd4a0edfSMiklos Szeredi * 4785fd4a0edfSMiklos Szeredi * Does not call security hook. 4786fd4a0edfSMiklos Szeredi */ 4787fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4788fd4a0edfSMiklos Szeredi { 4789fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 4790f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 4791f2df5da6SAl Viro const char *link; 4792f2df5da6SAl Viro int res; 4793fd4a0edfSMiklos Szeredi 479476fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 479576fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 479676fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 479776fca90eSMiklos Szeredi 479876fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 4799fd4a0edfSMiklos Szeredi return -EINVAL; 4800fd4a0edfSMiklos Szeredi 480176fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 480276fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 480376fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 480476fca90eSMiklos Szeredi } 480576fca90eSMiklos Szeredi 48064c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 4807f2df5da6SAl Viro if (!link) { 4808f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 4809f2df5da6SAl Viro if (IS_ERR(link)) 4810f2df5da6SAl Viro return PTR_ERR(link); 4811f2df5da6SAl Viro } 4812f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 4813f2df5da6SAl Viro do_delayed_call(&done); 4814f2df5da6SAl Viro return res; 4815fd4a0edfSMiklos Szeredi } 4816fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 48171da177e4SLinus Torvalds 4818d60874cdSMiklos Szeredi /** 4819d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 4820d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 4821d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 4822d60874cdSMiklos Szeredi * 4823d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 4824d60874cdSMiklos Szeredi * 4825d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 4826d60874cdSMiklos Szeredi * 4827d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 4828d60874cdSMiklos Szeredi */ 4829d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 4830d60874cdSMiklos Szeredi { 4831d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 4832d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 4833d60874cdSMiklos Szeredi 4834d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 4835d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 4836d60874cdSMiklos Szeredi if (!res) 4837d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 4838d60874cdSMiklos Szeredi } 4839d60874cdSMiklos Szeredi return res; 4840d60874cdSMiklos Szeredi } 4841d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 4842d60874cdSMiklos Szeredi 48431da177e4SLinus Torvalds /* get the link contents into pagecache */ 48446b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 4845fceef393SAl Viro struct delayed_call *callback) 48461da177e4SLinus Torvalds { 4847ebd09abbSDuane Griffin char *kaddr; 48481da177e4SLinus Torvalds struct page *page; 48496b255391SAl Viro struct address_space *mapping = inode->i_mapping; 48506b255391SAl Viro 4851d3883d4fSAl Viro if (!dentry) { 4852d3883d4fSAl Viro page = find_get_page(mapping, 0); 4853d3883d4fSAl Viro if (!page) 48546b255391SAl Viro return ERR_PTR(-ECHILD); 4855d3883d4fSAl Viro if (!PageUptodate(page)) { 4856d3883d4fSAl Viro put_page(page); 4857d3883d4fSAl Viro return ERR_PTR(-ECHILD); 4858d3883d4fSAl Viro } 4859d3883d4fSAl Viro } else { 4860090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 48611da177e4SLinus Torvalds if (IS_ERR(page)) 48626fe6900eSNick Piggin return (char*)page; 4863d3883d4fSAl Viro } 4864fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 486521fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 486621fc61c7SAl Viro kaddr = page_address(page); 48676b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 4868ebd09abbSDuane Griffin return kaddr; 48691da177e4SLinus Torvalds } 48701da177e4SLinus Torvalds 48716b255391SAl Viro EXPORT_SYMBOL(page_get_link); 48721da177e4SLinus Torvalds 4873fceef393SAl Viro void page_put_link(void *arg) 48741da177e4SLinus Torvalds { 4875fceef393SAl Viro put_page(arg); 48761da177e4SLinus Torvalds } 48774d359507SAl Viro EXPORT_SYMBOL(page_put_link); 48781da177e4SLinus Torvalds 4879aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 4880aa80deabSAl Viro { 4881fceef393SAl Viro DEFINE_DELAYED_CALL(done); 48826b255391SAl Viro int res = readlink_copy(buffer, buflen, 48836b255391SAl Viro page_get_link(dentry, d_inode(dentry), 4884fceef393SAl Viro &done)); 4885fceef393SAl Viro do_delayed_call(&done); 4886aa80deabSAl Viro return res; 4887aa80deabSAl Viro } 4888aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 4889aa80deabSAl Viro 489054566b2cSNick Piggin /* 489154566b2cSNick Piggin * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS 489254566b2cSNick Piggin */ 489354566b2cSNick Piggin int __page_symlink(struct inode *inode, const char *symname, int len, int nofs) 48941da177e4SLinus Torvalds { 48951da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 48960adb25d2SKirill Korotaev struct page *page; 4897afddba49SNick Piggin void *fsdata; 4898beb497abSDmitriy Monakhov int err; 4899c718a975STetsuo Handa unsigned int flags = 0; 490054566b2cSNick Piggin if (nofs) 490154566b2cSNick Piggin flags |= AOP_FLAG_NOFS; 49021da177e4SLinus Torvalds 49037e53cac4SNeilBrown retry: 4904afddba49SNick Piggin err = pagecache_write_begin(NULL, mapping, 0, len-1, 490554566b2cSNick Piggin flags, &page, &fsdata); 49061da177e4SLinus Torvalds if (err) 4907afddba49SNick Piggin goto fail; 4908afddba49SNick Piggin 490921fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 4910afddba49SNick Piggin 4911afddba49SNick Piggin err = pagecache_write_end(NULL, mapping, 0, len-1, len-1, 4912afddba49SNick Piggin page, fsdata); 49131da177e4SLinus Torvalds if (err < 0) 49141da177e4SLinus Torvalds goto fail; 4915afddba49SNick Piggin if (err < len-1) 4916afddba49SNick Piggin goto retry; 4917afddba49SNick Piggin 49181da177e4SLinus Torvalds mark_inode_dirty(inode); 49191da177e4SLinus Torvalds return 0; 49201da177e4SLinus Torvalds fail: 49211da177e4SLinus Torvalds return err; 49221da177e4SLinus Torvalds } 49234d359507SAl Viro EXPORT_SYMBOL(__page_symlink); 49241da177e4SLinus Torvalds 49250adb25d2SKirill Korotaev int page_symlink(struct inode *inode, const char *symname, int len) 49260adb25d2SKirill Korotaev { 49270adb25d2SKirill Korotaev return __page_symlink(inode, symname, len, 4928c62d2555SMichal Hocko !mapping_gfp_constraint(inode->i_mapping, __GFP_FS)); 49290adb25d2SKirill Korotaev } 49304d359507SAl Viro EXPORT_SYMBOL(page_symlink); 49310adb25d2SKirill Korotaev 493292e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 49336b255391SAl Viro .get_link = page_get_link, 49341da177e4SLinus Torvalds }; 49351da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 4936