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> 235970e15dSJeff Layton #include <linux/filelock.h> 241da177e4SLinus Torvalds #include <linux/namei.h> 251da177e4SLinus Torvalds #include <linux/pagemap.h> 262d878178SMatthew Wilcox (Oracle) #include <linux/sched/mm.h> 270eeca283SRobert Love #include <linux/fsnotify.h> 281da177e4SLinus Torvalds #include <linux/personality.h> 291da177e4SLinus Torvalds #include <linux/security.h> 306146f0d5SMimi Zohar #include <linux/ima.h> 311da177e4SLinus Torvalds #include <linux/syscalls.h> 321da177e4SLinus Torvalds #include <linux/mount.h> 331da177e4SLinus Torvalds #include <linux/audit.h> 3416f7e0feSRandy Dunlap #include <linux/capability.h> 35834f2a4aSTrond Myklebust #include <linux/file.h> 365590ff0dSUlrich Drepper #include <linux/fcntl.h> 3708ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 385ad4e53bSAl Viro #include <linux/fs_struct.h> 39e77819e5SLinus Torvalds #include <linux/posix_acl.h> 4099d263d4SLinus Torvalds #include <linux/hash.h> 412a18da7aSGeorge Spelvin #include <linux/bitops.h> 42aeaa4a79SEric W. Biederman #include <linux/init_task.h> 437c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 441da177e4SLinus Torvalds 45e81e3f4dSEric Paris #include "internal.h" 46c7105365SAl Viro #include "mount.h" 47e81e3f4dSEric Paris 481da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 491da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 501da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 511da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 521da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 531da177e4SLinus Torvalds * 541da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 551da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 561da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 571da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 581da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 591da177e4SLinus Torvalds * the special cases of the former code. 601da177e4SLinus Torvalds * 611da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 621da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 631da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 641da177e4SLinus Torvalds * 651da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 661da177e4SLinus Torvalds * resolution to correspond with current state of the code. 671da177e4SLinus Torvalds * 681da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 691da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 701da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 711da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 721da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 731da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 741da177e4SLinus Torvalds */ 751da177e4SLinus Torvalds 761da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 771da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 781da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 791da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 801da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 811da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8225985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 831da177e4SLinus Torvalds * 841da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 851da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 861da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 871da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 881da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 891da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 901da177e4SLinus Torvalds * and in the old Linux semantics. 911da177e4SLinus Torvalds */ 921da177e4SLinus Torvalds 931da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 941da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 951da177e4SLinus Torvalds * 961da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 971da177e4SLinus Torvalds */ 981da177e4SLinus Torvalds 991da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 1001da177e4SLinus Torvalds * inside the path - always follow. 1011da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1021da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1031da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1041da177e4SLinus Torvalds * otherwise - don't follow. 1051da177e4SLinus Torvalds * (applied in that order). 1061da177e4SLinus Torvalds * 1071da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1081da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1091da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1101da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1111da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1121da177e4SLinus Torvalds */ 1131da177e4SLinus Torvalds /* 1141da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 115a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1161da177e4SLinus Torvalds * any extra contention... 1171da177e4SLinus Torvalds */ 1181da177e4SLinus Torvalds 1191da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1201da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1211da177e4SLinus Torvalds * kernel data space before using them.. 1221da177e4SLinus Torvalds * 1231da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1241da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1251da177e4SLinus Torvalds */ 1267950e385SJeff Layton 127fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12891a27b2aSJeff Layton 12951f39a1fSDavid Drysdale struct filename * 13091a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13191a27b2aSJeff Layton { 13294b5d262SAl Viro struct filename *result; 1337950e385SJeff Layton char *kname; 13494b5d262SAl Viro int len; 1351da177e4SLinus Torvalds 1367ac86265SJeff Layton result = audit_reusename(filename); 1377ac86265SJeff Layton if (result) 1387ac86265SJeff Layton return result; 1397ac86265SJeff Layton 1407950e385SJeff Layton result = __getname(); 1413f9f0aa6SLinus Torvalds if (unlikely(!result)) 1424043cde8SEric Paris return ERR_PTR(-ENOMEM); 1431da177e4SLinus Torvalds 1447950e385SJeff Layton /* 1457950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1467950e385SJeff Layton * allocation 1477950e385SJeff Layton */ 148fd2f7cb5SAl Viro kname = (char *)result->iname; 14991a27b2aSJeff Layton result->name = kname; 1507950e385SJeff Layton 15194b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15291a27b2aSJeff Layton if (unlikely(len < 0)) { 15394b5d262SAl Viro __putname(result); 15494b5d262SAl Viro return ERR_PTR(len); 15591a27b2aSJeff Layton } 1563f9f0aa6SLinus Torvalds 1577950e385SJeff Layton /* 1587950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1597950e385SJeff Layton * separate struct filename so we can dedicate the entire 1607950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1617950e385SJeff Layton * userland. 1627950e385SJeff Layton */ 16394b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 164fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1657950e385SJeff Layton kname = (char *)result; 1667950e385SJeff Layton 167fd2f7cb5SAl Viro /* 168fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 169fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 170fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 171fd2f7cb5SAl Viro */ 172fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17394b5d262SAl Viro if (unlikely(!result)) { 17494b5d262SAl Viro __putname(kname); 17594b5d262SAl Viro return ERR_PTR(-ENOMEM); 1767950e385SJeff Layton } 1777950e385SJeff Layton result->name = kname; 17894b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17994b5d262SAl Viro if (unlikely(len < 0)) { 18094b5d262SAl Viro __putname(kname); 18194b5d262SAl Viro kfree(result); 18294b5d262SAl Viro return ERR_PTR(len); 18394b5d262SAl Viro } 18494b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18594b5d262SAl Viro __putname(kname); 18694b5d262SAl Viro kfree(result); 18794b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18894b5d262SAl Viro } 1897950e385SJeff Layton } 1907950e385SJeff Layton 191*03adc61eSDan Clash atomic_set(&result->refcnt, 1); 1923f9f0aa6SLinus Torvalds /* The empty path is special. */ 1933f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1943f9f0aa6SLinus Torvalds if (empty) 1951fa1e7f6SAndy Whitcroft *empty = 1; 19694b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19794b5d262SAl Viro putname(result); 19894b5d262SAl Viro return ERR_PTR(-ENOENT); 1991da177e4SLinus Torvalds } 20094b5d262SAl Viro } 2017950e385SJeff Layton 2027950e385SJeff Layton result->uptr = filename; 203c4ad8f98SLinus Torvalds result->aname = NULL; 2041da177e4SLinus Torvalds audit_getname(result); 2051da177e4SLinus Torvalds return result; 2063f9f0aa6SLinus Torvalds } 2073f9f0aa6SLinus Torvalds 20891a27b2aSJeff Layton struct filename * 2098228e2c3SDmitry Kadashev getname_uflags(const char __user *filename, int uflags) 2108228e2c3SDmitry Kadashev { 2118228e2c3SDmitry Kadashev int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 2128228e2c3SDmitry Kadashev 2138228e2c3SDmitry Kadashev return getname_flags(filename, flags, NULL); 2148228e2c3SDmitry Kadashev } 2158228e2c3SDmitry Kadashev 2168228e2c3SDmitry Kadashev struct filename * 21791a27b2aSJeff Layton getname(const char __user * filename) 218f52e0c11SAl Viro { 219f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 220f52e0c11SAl Viro } 221f52e0c11SAl Viro 222c4ad8f98SLinus Torvalds struct filename * 223c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 224c4ad8f98SLinus Torvalds { 225c4ad8f98SLinus Torvalds struct filename *result; 22608518549SPaul Moore int len = strlen(filename) + 1; 227c4ad8f98SLinus Torvalds 228c4ad8f98SLinus Torvalds result = __getname(); 229c4ad8f98SLinus Torvalds if (unlikely(!result)) 230c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 231c4ad8f98SLinus Torvalds 23208518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 233fd2f7cb5SAl Viro result->name = (char *)result->iname; 23408518549SPaul Moore } else if (len <= PATH_MAX) { 23530ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 23608518549SPaul Moore struct filename *tmp; 23708518549SPaul Moore 23830ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 23908518549SPaul Moore if (unlikely(!tmp)) { 24008518549SPaul Moore __putname(result); 24108518549SPaul Moore return ERR_PTR(-ENOMEM); 24208518549SPaul Moore } 24308518549SPaul Moore tmp->name = (char *)result; 24408518549SPaul Moore result = tmp; 24508518549SPaul Moore } else { 24608518549SPaul Moore __putname(result); 24708518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 24808518549SPaul Moore } 24908518549SPaul Moore memcpy((char *)result->name, filename, len); 250c4ad8f98SLinus Torvalds result->uptr = NULL; 251c4ad8f98SLinus Torvalds result->aname = NULL; 252*03adc61eSDan Clash atomic_set(&result->refcnt, 1); 253fd3522fdSPaul Moore audit_getname(result); 254c4ad8f98SLinus Torvalds 255c4ad8f98SLinus Torvalds return result; 256c4ad8f98SLinus Torvalds } 25774d7970fSNamjae Jeon EXPORT_SYMBOL(getname_kernel); 258c4ad8f98SLinus Torvalds 25991a27b2aSJeff Layton void putname(struct filename *name) 2601da177e4SLinus Torvalds { 261ea47ab11SAl Viro if (IS_ERR(name)) 26291ef658fSDmitry Kadashev return; 26391ef658fSDmitry Kadashev 264*03adc61eSDan Clash if (WARN_ON_ONCE(!atomic_read(&name->refcnt))) 265*03adc61eSDan Clash return; 26655422d0bSPaul Moore 267*03adc61eSDan Clash if (!atomic_dec_and_test(&name->refcnt)) 26855422d0bSPaul Moore return; 26955422d0bSPaul Moore 270fd2f7cb5SAl Viro if (name->name != name->iname) { 27155422d0bSPaul Moore __putname(name->name); 27255422d0bSPaul Moore kfree(name); 27355422d0bSPaul Moore } else 27455422d0bSPaul Moore __putname(name); 2751da177e4SLinus Torvalds } 27674d7970fSNamjae Jeon EXPORT_SYMBOL(putname); 2771da177e4SLinus Torvalds 27847291baaSChristian Brauner /** 27947291baaSChristian Brauner * check_acl - perform ACL permission checking 280700b7940SChristian Brauner * @idmap: idmap of the mount the inode was found from 28147291baaSChristian Brauner * @inode: inode to check permissions on 28247291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 28347291baaSChristian Brauner * 28447291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 28547291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 28647291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 28747291baaSChristian Brauner * 288700b7940SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 289700b7940SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 290700b7940SChristian Brauner * care to map the inode according to @idmap before checking permissions. 29147291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 292700b7940SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 29347291baaSChristian Brauner */ 294700b7940SChristian Brauner static int check_acl(struct mnt_idmap *idmap, 29547291baaSChristian Brauner struct inode *inode, int mask) 296e77819e5SLinus Torvalds { 29784635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 298e77819e5SLinus Torvalds struct posix_acl *acl; 299e77819e5SLinus Torvalds 300e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 3013567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 3023567866bSAl Viro if (!acl) 303e77819e5SLinus Torvalds return -EAGAIN; 304cac2f8b8SChristian Brauner /* no ->get_inode_acl() calls in RCU mode... */ 305b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 306e77819e5SLinus Torvalds return -ECHILD; 307700b7940SChristian Brauner return posix_acl_permission(idmap, inode, acl, mask); 308e77819e5SLinus Torvalds } 309e77819e5SLinus Torvalds 310cac2f8b8SChristian Brauner acl = get_inode_acl(inode, ACL_TYPE_ACCESS); 3114e34e719SChristoph Hellwig if (IS_ERR(acl)) 3124e34e719SChristoph Hellwig return PTR_ERR(acl); 313e77819e5SLinus Torvalds if (acl) { 314700b7940SChristian Brauner int error = posix_acl_permission(idmap, inode, acl, mask); 315e77819e5SLinus Torvalds posix_acl_release(acl); 316e77819e5SLinus Torvalds return error; 317e77819e5SLinus Torvalds } 31884635d68SLinus Torvalds #endif 319e77819e5SLinus Torvalds 320e77819e5SLinus Torvalds return -EAGAIN; 321e77819e5SLinus Torvalds } 322e77819e5SLinus Torvalds 32347291baaSChristian Brauner /** 32447291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 325700b7940SChristian Brauner * @idmap: idmap of the mount the inode was found from 32647291baaSChristian Brauner * @inode: inode to check permissions on 32747291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3285fc475b7SLinus Torvalds * 32947291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 33047291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 33147291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 33247291baaSChristian Brauner * 333700b7940SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 334700b7940SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 335700b7940SChristian Brauner * care to map the inode according to @idmap before checking permissions. 33647291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 337700b7940SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 3385909ccaaSLinus Torvalds */ 339700b7940SChristian Brauner static int acl_permission_check(struct mnt_idmap *idmap, 34047291baaSChristian Brauner struct inode *inode, int mask) 3415909ccaaSLinus Torvalds { 34226cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 343a2bd096fSChristian Brauner vfsuid_t vfsuid; 3445909ccaaSLinus Torvalds 3455fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 346e67fe633SChristian Brauner vfsuid = i_uid_into_vfsuid(idmap, inode); 347a2bd096fSChristian Brauner if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) { 3485fc475b7SLinus Torvalds mask &= 7; 3495909ccaaSLinus Torvalds mode >>= 6; 3505fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3515fc475b7SLinus Torvalds } 3525fc475b7SLinus Torvalds 3535fc475b7SLinus Torvalds /* Do we have ACL's? */ 354e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 355700b7940SChristian Brauner int error = check_acl(idmap, inode, mask); 3565909ccaaSLinus Torvalds if (error != -EAGAIN) 3575909ccaaSLinus Torvalds return error; 3585909ccaaSLinus Torvalds } 3595909ccaaSLinus Torvalds 3605fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3615fc475b7SLinus Torvalds mask &= 7; 3625fc475b7SLinus Torvalds 3635fc475b7SLinus Torvalds /* 3645fc475b7SLinus Torvalds * Are the group permissions different from 3655fc475b7SLinus Torvalds * the other permissions in the bits we care 3665fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3675fc475b7SLinus Torvalds */ 3685fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 369e67fe633SChristian Brauner vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); 370a2bd096fSChristian Brauner if (vfsgid_in_group_p(vfsgid)) 3715909ccaaSLinus Torvalds mode >>= 3; 3725909ccaaSLinus Torvalds } 3735909ccaaSLinus Torvalds 3745fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3755fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3765909ccaaSLinus Torvalds } 3771da177e4SLinus Torvalds 3781da177e4SLinus Torvalds /** 3791da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3804609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 3811da177e4SLinus Torvalds * @inode: inode to check access rights for 3825fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3835fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3841da177e4SLinus Torvalds * 3851da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3861da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3871da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 388b74c79e9SNick Piggin * are used for other things. 389b74c79e9SNick Piggin * 390b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 391b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 392b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 39347291baaSChristian Brauner * 3944609e1f1SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3954609e1f1SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3964609e1f1SChristian Brauner * care to map the inode according to @idmap before checking permissions. 39747291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3984609e1f1SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 3991da177e4SLinus Torvalds */ 4004609e1f1SChristian Brauner int generic_permission(struct mnt_idmap *idmap, struct inode *inode, 40147291baaSChristian Brauner int mask) 4021da177e4SLinus Torvalds { 4035909ccaaSLinus Torvalds int ret; 4041da177e4SLinus Torvalds 4051da177e4SLinus Torvalds /* 406948409c7SAndreas Gruenbacher * Do the basic permission checks. 4071da177e4SLinus Torvalds */ 408700b7940SChristian Brauner ret = acl_permission_check(idmap, inode, mask); 4095909ccaaSLinus Torvalds if (ret != -EACCES) 4105909ccaaSLinus Torvalds return ret; 4111da177e4SLinus Torvalds 412d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 413d594e7ecSAl Viro /* DACs are overridable for directories */ 414d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 4159452e93eSChristian Brauner if (capable_wrt_inode_uidgid(idmap, inode, 41623adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 417d594e7ecSAl Viro return 0; 4189452e93eSChristian Brauner if (capable_wrt_inode_uidgid(idmap, inode, 4190558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4201da177e4SLinus Torvalds return 0; 4212a4c2242SStephen Smalley return -EACCES; 4222a4c2242SStephen Smalley } 4231da177e4SLinus Torvalds 4241da177e4SLinus Torvalds /* 4251da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4261da177e4SLinus Torvalds */ 4277ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 428d594e7ecSAl Viro if (mask == MAY_READ) 4299452e93eSChristian Brauner if (capable_wrt_inode_uidgid(idmap, inode, 4300558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4311da177e4SLinus Torvalds return 0; 4322a4c2242SStephen Smalley /* 4332a4c2242SStephen Smalley * Read/write DACs are always overridable. 4342a4c2242SStephen Smalley * Executable DACs are overridable when there is 4352a4c2242SStephen Smalley * at least one exec bit set. 4362a4c2242SStephen Smalley */ 4372a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 4389452e93eSChristian Brauner if (capable_wrt_inode_uidgid(idmap, inode, 4390558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4402a4c2242SStephen Smalley return 0; 4411da177e4SLinus Torvalds 4421da177e4SLinus Torvalds return -EACCES; 4431da177e4SLinus Torvalds } 4444d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4451da177e4SLinus Torvalds 44647291baaSChristian Brauner /** 44747291baaSChristian Brauner * do_inode_permission - UNIX permission checking 4484609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 44947291baaSChristian Brauner * @inode: inode to check permissions on 45047291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 45147291baaSChristian Brauner * 4523ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4533ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4543ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4553ddcd056SLinus Torvalds * permission function, use the fast case". 4563ddcd056SLinus Torvalds */ 4574609e1f1SChristian Brauner static inline int do_inode_permission(struct mnt_idmap *idmap, 45847291baaSChristian Brauner struct inode *inode, int mask) 4593ddcd056SLinus Torvalds { 4603ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4613ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 4624609e1f1SChristian Brauner return inode->i_op->permission(idmap, inode, mask); 4633ddcd056SLinus Torvalds 4643ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4653ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4663ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4673ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4683ddcd056SLinus Torvalds } 4694609e1f1SChristian Brauner return generic_permission(idmap, inode, mask); 4703ddcd056SLinus Torvalds } 4713ddcd056SLinus Torvalds 472cb23beb5SChristoph Hellwig /** 4730bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4740bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 47555852635SRandy Dunlap * @inode: Inode to check permission on 4760bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4770bdaea90SDavid Howells * 4780bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4790bdaea90SDavid Howells */ 4800bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4810bdaea90SDavid Howells { 4820bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4830bdaea90SDavid Howells umode_t mode = inode->i_mode; 4840bdaea90SDavid Howells 4850bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 486bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4870bdaea90SDavid Howells return -EROFS; 4880bdaea90SDavid Howells } 4890bdaea90SDavid Howells return 0; 4900bdaea90SDavid Howells } 4910bdaea90SDavid Howells 4920bdaea90SDavid Howells /** 4930bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4944609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 4950bdaea90SDavid Howells * @inode: Inode to check permission on 4960bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4970bdaea90SDavid Howells * 4980bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4990bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 5000bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 5010bdaea90SDavid Howells * 5020bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 5030bdaea90SDavid Howells */ 5044609e1f1SChristian Brauner int inode_permission(struct mnt_idmap *idmap, 50547291baaSChristian Brauner struct inode *inode, int mask) 5060bdaea90SDavid Howells { 5070bdaea90SDavid Howells int retval; 5080bdaea90SDavid Howells 5090bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 5100bdaea90SDavid Howells if (retval) 5110bdaea90SDavid Howells return retval; 5124bfd054aSEric Biggers 5134bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 5144bfd054aSEric Biggers /* 5154bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5164bfd054aSEric Biggers */ 5174bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5184bfd054aSEric Biggers return -EPERM; 5194bfd054aSEric Biggers 5204bfd054aSEric Biggers /* 5214bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5224bfd054aSEric Biggers * written back improperly if their true value is unknown 5234bfd054aSEric Biggers * to the vfs. 5244bfd054aSEric Biggers */ 5254609e1f1SChristian Brauner if (HAS_UNMAPPED_ID(idmap, inode)) 5264bfd054aSEric Biggers return -EACCES; 5274bfd054aSEric Biggers } 5284bfd054aSEric Biggers 5294609e1f1SChristian Brauner retval = do_inode_permission(idmap, inode, mask); 5304bfd054aSEric Biggers if (retval) 5314bfd054aSEric Biggers return retval; 5324bfd054aSEric Biggers 5334bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5344bfd054aSEric Biggers if (retval) 5354bfd054aSEric Biggers return retval; 5364bfd054aSEric Biggers 5374bfd054aSEric Biggers return security_inode_permission(inode, mask); 5380bdaea90SDavid Howells } 5394d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5400bdaea90SDavid Howells 5410bdaea90SDavid Howells /** 5425dd784d0SJan Blunck * path_get - get a reference to a path 5435dd784d0SJan Blunck * @path: path to get the reference to 5445dd784d0SJan Blunck * 5455dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5465dd784d0SJan Blunck */ 547dcf787f3SAl Viro void path_get(const struct path *path) 5485dd784d0SJan Blunck { 5495dd784d0SJan Blunck mntget(path->mnt); 5505dd784d0SJan Blunck dget(path->dentry); 5515dd784d0SJan Blunck } 5525dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5535dd784d0SJan Blunck 5545dd784d0SJan Blunck /** 5551d957f9bSJan Blunck * path_put - put a reference to a path 5561d957f9bSJan Blunck * @path: path to put the reference to 5571d957f9bSJan Blunck * 5581d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5591d957f9bSJan Blunck */ 560dcf787f3SAl Viro void path_put(const struct path *path) 5611da177e4SLinus Torvalds { 5621d957f9bSJan Blunck dput(path->dentry); 5631d957f9bSJan Blunck mntput(path->mnt); 5641da177e4SLinus Torvalds } 5651d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5661da177e4SLinus Torvalds 567894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5681f55a6ecSAl Viro struct nameidata { 5691f55a6ecSAl Viro struct path path; 5701f55a6ecSAl Viro struct qstr last; 5711f55a6ecSAl Viro struct path root; 5721f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 573bcba1e7dSAl Viro unsigned int flags, state; 57403fa86e9SAl Viro unsigned seq, next_seq, m_seq, r_seq; 5751f55a6ecSAl Viro int last_type; 5761f55a6ecSAl Viro unsigned depth; 577756daf26SNeilBrown int total_link_count; 578697fc6caSAl Viro struct saved { 579697fc6caSAl Viro struct path link; 580fceef393SAl Viro struct delayed_call done; 581697fc6caSAl Viro const char *name; 5820450b2d1SAl Viro unsigned seq; 583894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5849883d185SAl Viro struct filename *name; 5859883d185SAl Viro struct nameidata *saved; 5869883d185SAl Viro unsigned root_seq; 5879883d185SAl Viro int dfd; 588a2bd096fSChristian Brauner vfsuid_t dir_vfsuid; 5890f705953SAl Viro umode_t dir_mode; 5903859a271SKees Cook } __randomize_layout; 5911f55a6ecSAl Viro 592bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 593bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 594bcba1e7dSAl Viro #define ND_JUMPED 4 595bcba1e7dSAl Viro 59606422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 597894bc8c4SAl Viro { 598756daf26SNeilBrown struct nameidata *old = current->nameidata; 599756daf26SNeilBrown p->stack = p->internal; 6007962c7d1SAl Viro p->depth = 0; 601c8a53ee5SAl Viro p->dfd = dfd; 602c8a53ee5SAl Viro p->name = name; 6037d01ef75SAl Viro p->path.mnt = NULL; 6047d01ef75SAl Viro p->path.dentry = NULL; 605756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 6069883d185SAl Viro p->saved = old; 607756daf26SNeilBrown current->nameidata = p; 608894bc8c4SAl Viro } 609894bc8c4SAl Viro 61006422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 61106422964SAl Viro const struct path *root) 61206422964SAl Viro { 61306422964SAl Viro __set_nameidata(p, dfd, name); 61406422964SAl Viro p->state = 0; 61506422964SAl Viro if (unlikely(root)) { 61606422964SAl Viro p->state = ND_ROOT_PRESET; 61706422964SAl Viro p->root = *root; 61806422964SAl Viro } 61906422964SAl Viro } 62006422964SAl Viro 6219883d185SAl Viro static void restore_nameidata(void) 622894bc8c4SAl Viro { 6239883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 624756daf26SNeilBrown 625756daf26SNeilBrown current->nameidata = old; 626756daf26SNeilBrown if (old) 627756daf26SNeilBrown old->total_link_count = now->total_link_count; 628e1a63bbcSAl Viro if (now->stack != now->internal) 629756daf26SNeilBrown kfree(now->stack); 630894bc8c4SAl Viro } 631894bc8c4SAl Viro 63260ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 633894bc8c4SAl Viro { 634bc40aee0SAl Viro struct saved *p; 635bc40aee0SAl Viro 6366da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 63760ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 638bc40aee0SAl Viro if (unlikely(!p)) 63960ef60c7SAl Viro return false; 640894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 641894bc8c4SAl Viro nd->stack = p; 64260ef60c7SAl Viro return true; 643894bc8c4SAl Viro } 644894bc8c4SAl Viro 645397d425dSEric W. Biederman /** 6466b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 64735931eb3SMatthew Wilcox (Oracle) * @mnt: The mountpoint to check. 64835931eb3SMatthew Wilcox (Oracle) * @dentry: The dentry to check. 649397d425dSEric W. Biederman * 650397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 651397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 652397d425dSEric W. Biederman */ 6536b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 654397d425dSEric W. Biederman { 65595dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 656397d425dSEric W. Biederman 657402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 658402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 659397d425dSEric W. Biederman return true; 660397d425dSEric W. Biederman 6616b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 662397d425dSEric W. Biederman } 663397d425dSEric W. Biederman 6647973387aSAl Viro static void drop_links(struct nameidata *nd) 6657973387aSAl Viro { 6667973387aSAl Viro int i = nd->depth; 6677973387aSAl Viro while (i--) { 6687973387aSAl Viro struct saved *last = nd->stack + i; 669fceef393SAl Viro do_delayed_call(&last->done); 670fceef393SAl Viro clear_delayed_call(&last->done); 6717973387aSAl Viro } 6727973387aSAl Viro } 6737973387aSAl Viro 6746e180327SAl Viro static void leave_rcu(struct nameidata *nd) 6756e180327SAl Viro { 6766e180327SAl Viro nd->flags &= ~LOOKUP_RCU; 67703fa86e9SAl Viro nd->seq = nd->next_seq = 0; 6786e180327SAl Viro rcu_read_unlock(); 6796e180327SAl Viro } 6806e180327SAl Viro 6817973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6827973387aSAl Viro { 6837973387aSAl Viro drop_links(nd); 6847973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6857973387aSAl Viro int i; 6867973387aSAl Viro path_put(&nd->path); 6877973387aSAl Viro for (i = 0; i < nd->depth; i++) 6887973387aSAl Viro path_put(&nd->stack[i].link); 689bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 690102b8af2SAl Viro path_put(&nd->root); 691bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 692102b8af2SAl Viro } 6937973387aSAl Viro } else { 6946e180327SAl Viro leave_rcu(nd); 6957973387aSAl Viro } 6967973387aSAl Viro nd->depth = 0; 6977d01ef75SAl Viro nd->path.mnt = NULL; 6987d01ef75SAl Viro nd->path.dentry = NULL; 6997973387aSAl Viro } 7007973387aSAl Viro 7017973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 7022aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 7037973387aSAl Viro { 7042aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 7057973387aSAl Viro if (unlikely(res)) { 7067973387aSAl Viro if (res > 0) 7077973387aSAl Viro path->mnt = NULL; 7087973387aSAl Viro path->dentry = NULL; 7097973387aSAl Viro return false; 7107973387aSAl Viro } 7117973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 7127973387aSAl Viro path->dentry = NULL; 7137973387aSAl Viro return false; 7147973387aSAl Viro } 7157973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7167973387aSAl Viro } 7177973387aSAl Viro 7182aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7192aa38470SAl Viro struct path *path, unsigned seq) 7202aa38470SAl Viro { 7215bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7222aa38470SAl Viro } 7232aa38470SAl Viro 7247973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7257973387aSAl Viro { 7267973387aSAl Viro int i; 727eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 728eacd9aa8SAl Viro drop_links(nd); 729eacd9aa8SAl Viro nd->depth = 0; 730eacd9aa8SAl Viro return false; 731eacd9aa8SAl Viro } 7327973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7337973387aSAl Viro struct saved *last = nd->stack + i; 7347973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7357973387aSAl Viro drop_links(nd); 7367973387aSAl Viro nd->depth = i + 1; 7377973387aSAl Viro return false; 7387973387aSAl Viro } 7397973387aSAl Viro } 7407973387aSAl Viro return true; 7417973387aSAl Viro } 7427973387aSAl Viro 743ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 744ee594bffSAl Viro { 745adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 746bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 747ee594bffSAl Viro return true; 748bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 749ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 750ee594bffSAl Viro } 751ee594bffSAl Viro 75219660af7SAl Viro /* 75331e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 75419660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 75519660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 75657e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 75719660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 75819660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 75919660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 76019660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 76131e6b01fSNick Piggin */ 76231e6b01fSNick Piggin 76331e6b01fSNick Piggin /** 764e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 76519660af7SAl Viro * @nd: nameidata pathwalk data 766e36cffedSJens Axboe * Returns: true on success, false on failure 76731e6b01fSNick Piggin * 768e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7694675ac39SAl Viro * for ref-walk mode. 7704675ac39SAl Viro * Must be called from rcu-walk context. 771e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7727973387aSAl Viro * terminate_walk(). 77331e6b01fSNick Piggin */ 774e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 77531e6b01fSNick Piggin { 77631e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 77731e6b01fSNick Piggin 77831e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 779e5c832d5SLinus Torvalds 7807973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7814675ac39SAl Viro goto out1; 78284a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 78384a2bd39SAl Viro goto out; 784ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7854675ac39SAl Viro goto out; 7866e180327SAl Viro leave_rcu(nd); 7874675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 788e36cffedSJens Axboe return true; 7894675ac39SAl Viro 79084a2bd39SAl Viro out1: 7914675ac39SAl Viro nd->path.mnt = NULL; 7924675ac39SAl Viro nd->path.dentry = NULL; 7934675ac39SAl Viro out: 7946e180327SAl Viro leave_rcu(nd); 795e36cffedSJens Axboe return false; 7964675ac39SAl Viro } 7974675ac39SAl Viro 7984675ac39SAl Viro /** 799ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 8004675ac39SAl Viro * @nd: nameidata pathwalk data 801ae66db45SAl Viro * @dentry: next dentry to step into 802ae66db45SAl Viro * Returns: true on success, false on failure 8034675ac39SAl Viro * 80430476f7eSTom Rix * Similar to try_to_unlazy(), but here we have the next dentry already 805ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 806ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 807ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 8084675ac39SAl Viro * terminate_walk(). 8094675ac39SAl Viro */ 81003fa86e9SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry) 8114675ac39SAl Viro { 8127e4745a0SAl Viro int res; 8134675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8144675ac39SAl Viro 8154675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8164675ac39SAl Viro goto out2; 8177e4745a0SAl Viro res = __legitimize_mnt(nd->path.mnt, nd->m_seq); 8187e4745a0SAl Viro if (unlikely(res)) { 8197e4745a0SAl Viro if (res > 0) 8207973387aSAl Viro goto out2; 8217e4745a0SAl Viro goto out1; 8227e4745a0SAl Viro } 8234675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8247973387aSAl Viro goto out1; 82548a066e7SAl Viro 82615570086SLinus Torvalds /* 8274675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8284675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8294675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8304675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8314675ac39SAl Viro * know the parent must still be valid if the child sequence number is 83215570086SLinus Torvalds */ 8334675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 834e5c832d5SLinus Torvalds goto out; 83503fa86e9SAl Viro if (read_seqcount_retry(&dentry->d_seq, nd->next_seq)) 83684a2bd39SAl Viro goto out_dput; 837e5c832d5SLinus Torvalds /* 838e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 839e5c832d5SLinus Torvalds * still valid and get it if required. 840e5c832d5SLinus Torvalds */ 84184a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 84284a2bd39SAl Viro goto out_dput; 8436e180327SAl Viro leave_rcu(nd); 844ae66db45SAl Viro return true; 84519660af7SAl Viro 8467973387aSAl Viro out2: 8477973387aSAl Viro nd->path.mnt = NULL; 8487973387aSAl Viro out1: 8497973387aSAl Viro nd->path.dentry = NULL; 850e5c832d5SLinus Torvalds out: 8516e180327SAl Viro leave_rcu(nd); 852ae66db45SAl Viro return false; 85384a2bd39SAl Viro out_dput: 8546e180327SAl Viro leave_rcu(nd); 85584a2bd39SAl Viro dput(dentry); 856ae66db45SAl Viro return false; 85731e6b01fSNick Piggin } 85831e6b01fSNick Piggin 8594ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 86034286d66SNick Piggin { 861a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8624ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 863a89f8337SAl Viro else 864a89f8337SAl Viro return 1; 86534286d66SNick Piggin } 86634286d66SNick Piggin 8679f1fafeeSAl Viro /** 8689f1fafeeSAl Viro * complete_walk - successful completion of path walk 8699f1fafeeSAl Viro * @nd: pointer nameidata 87039159de2SJeff Layton * 8719f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8729f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8739f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8749f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8759f1fafeeSAl Viro * need to drop nd->path. 87639159de2SJeff Layton */ 8779f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 87839159de2SJeff Layton { 87916c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 88039159de2SJeff Layton int status; 88139159de2SJeff Layton 8829f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 883adb21d2bSAleksa Sarai /* 884adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 885adb21d2bSAleksa Sarai * externally-managed nd->root. 886adb21d2bSAleksa Sarai */ 887bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 888bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8899f1fafeeSAl Viro nd->root.mnt = NULL; 8906c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 891e36cffedSJens Axboe if (!try_to_unlazy(nd)) 89248a066e7SAl Viro return -ECHILD; 89348a066e7SAl Viro } 8949f1fafeeSAl Viro 895adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 896adb21d2bSAleksa Sarai /* 897adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 898adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 899adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 900adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 901adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 902adb21d2bSAleksa Sarai * 903adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 904adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 905adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 906adb21d2bSAleksa Sarai * requested root to userspace. 907adb21d2bSAleksa Sarai * 908adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 909adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 910adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 911adb21d2bSAleksa Sarai */ 912adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 913adb21d2bSAleksa Sarai return -EXDEV; 914adb21d2bSAleksa Sarai } 915adb21d2bSAleksa Sarai 916bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 91739159de2SJeff Layton return 0; 91839159de2SJeff Layton 919ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 92016c2cd71SAl Viro return 0; 92116c2cd71SAl Viro 922ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 92339159de2SJeff Layton if (status > 0) 92439159de2SJeff Layton return 0; 92539159de2SJeff Layton 92616c2cd71SAl Viro if (!status) 92739159de2SJeff Layton status = -ESTALE; 92816c2cd71SAl Viro 92939159de2SJeff Layton return status; 93039159de2SJeff Layton } 93139159de2SJeff Layton 932740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9332a737871SAl Viro { 93431e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9359e6697e2SAl Viro 936adb21d2bSAleksa Sarai /* 937adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 938adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 939adb21d2bSAleksa Sarai * from the dirfd. 940adb21d2bSAleksa Sarai */ 941adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 942adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 943adb21d2bSAleksa Sarai 9449e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9458f47a016SAl Viro unsigned seq; 946c28cc364SNick Piggin 947c28cc364SNick Piggin do { 948c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 94931e6b01fSNick Piggin nd->root = fs->root; 9508f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 951c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9529e6697e2SAl Viro } else { 9539e6697e2SAl Viro get_fs_root(fs, &nd->root); 954bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9559e6697e2SAl Viro } 956740a1678SAleksa Sarai return 0; 95731e6b01fSNick Piggin } 95831e6b01fSNick Piggin 959248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 960248fb5b9SAl Viro { 961adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 962adb21d2bSAleksa Sarai return -EXDEV; 96372ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 96472ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 96572ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 96672ba2929SAleksa Sarai return -EXDEV; 96772ba2929SAleksa Sarai } 968740a1678SAleksa Sarai if (!nd->root.mnt) { 969740a1678SAleksa Sarai int error = set_root(nd); 970740a1678SAleksa Sarai if (error) 971740a1678SAleksa Sarai return error; 972740a1678SAleksa Sarai } 973248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 974248fb5b9SAl Viro struct dentry *d; 975248fb5b9SAl Viro nd->path = nd->root; 976248fb5b9SAl Viro d = nd->path.dentry; 977248fb5b9SAl Viro nd->inode = d->d_inode; 978248fb5b9SAl Viro nd->seq = nd->root_seq; 97982ef0698SAl Viro if (read_seqcount_retry(&d->d_seq, nd->seq)) 980248fb5b9SAl Viro return -ECHILD; 981248fb5b9SAl Viro } else { 982248fb5b9SAl Viro path_put(&nd->path); 983248fb5b9SAl Viro nd->path = nd->root; 984248fb5b9SAl Viro path_get(&nd->path); 985248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 986248fb5b9SAl Viro } 987bcba1e7dSAl Viro nd->state |= ND_JUMPED; 988248fb5b9SAl Viro return 0; 989248fb5b9SAl Viro } 990248fb5b9SAl Viro 991b5fb63c1SChristoph Hellwig /* 9926b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 993b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 994b5fb63c1SChristoph Hellwig */ 995ea4af4aaSAl Viro int nd_jump_link(const struct path *path) 996b5fb63c1SChristoph Hellwig { 9974b99d499SAleksa Sarai int error = -ELOOP; 9986e77137bSAl Viro struct nameidata *nd = current->nameidata; 999b5fb63c1SChristoph Hellwig 10004b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 10014b99d499SAleksa Sarai goto err; 10024b99d499SAleksa Sarai 100372ba2929SAleksa Sarai error = -EXDEV; 100472ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 100572ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 100672ba2929SAleksa Sarai goto err; 100772ba2929SAleksa Sarai } 1008adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 1009adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 1010adb21d2bSAleksa Sarai goto err; 101172ba2929SAleksa Sarai 10124b99d499SAleksa Sarai path_put(&nd->path); 1013b5fb63c1SChristoph Hellwig nd->path = *path; 1014b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1015bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10161bc82070SAleksa Sarai return 0; 10174b99d499SAleksa Sarai 10184b99d499SAleksa Sarai err: 10194b99d499SAleksa Sarai path_put(path); 10204b99d499SAleksa Sarai return error; 1021b5fb63c1SChristoph Hellwig } 1022b5fb63c1SChristoph Hellwig 1023b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1024574197e0SAl Viro { 102521c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1026fceef393SAl Viro do_delayed_call(&last->done); 10276548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1028b9ff4429SAl Viro path_put(&last->link); 1029574197e0SAl Viro } 1030574197e0SAl Viro 10319c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10329c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10339c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10349c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10359c011be1SLuis Chamberlain 10369c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10379c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10389c011be1SLuis Chamberlain { 10399c011be1SLuis Chamberlain .procname = "protected_symlinks", 10409c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10419c011be1SLuis Chamberlain .maxlen = sizeof(int), 1042c7031c14SJulius Hemanth Pitti .mode = 0644, 10439c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10449c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10459c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10469c011be1SLuis Chamberlain }, 10479c011be1SLuis Chamberlain { 10489c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10499c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10509c011be1SLuis Chamberlain .maxlen = sizeof(int), 1051c7031c14SJulius Hemanth Pitti .mode = 0644, 10529c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10539c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10549c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10559c011be1SLuis Chamberlain }, 10569c011be1SLuis Chamberlain { 10579c011be1SLuis Chamberlain .procname = "protected_fifos", 10589c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10599c011be1SLuis Chamberlain .maxlen = sizeof(int), 1060c7031c14SJulius Hemanth Pitti .mode = 0644, 10619c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10629c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10639c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10649c011be1SLuis Chamberlain }, 10659c011be1SLuis Chamberlain { 10669c011be1SLuis Chamberlain .procname = "protected_regular", 10679c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10689c011be1SLuis Chamberlain .maxlen = sizeof(int), 1069c7031c14SJulius Hemanth Pitti .mode = 0644, 10709c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10719c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10729c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10739c011be1SLuis Chamberlain }, 10749c011be1SLuis Chamberlain { } 10759c011be1SLuis Chamberlain }; 10769c011be1SLuis Chamberlain 10779c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10789c011be1SLuis Chamberlain { 10799c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10809c011be1SLuis Chamberlain return 0; 10819c011be1SLuis Chamberlain } 10829c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10839c011be1SLuis Chamberlain 10849c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1085800179c9SKees Cook 1086800179c9SKees Cook /** 1087800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 108855852635SRandy Dunlap * @nd: nameidata pathwalk data 108935931eb3SMatthew Wilcox (Oracle) * @inode: Used for idmapping. 1090800179c9SKees Cook * 1091800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1092800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1093800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1094800179c9SKees Cook * processes from failing races against path names that may change out 1095800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1096800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1097800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1098800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1099800179c9SKees Cook * 1100800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1101800179c9SKees Cook */ 1102ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1103800179c9SKees Cook { 1104e67fe633SChristian Brauner struct mnt_idmap *idmap; 1105a2bd096fSChristian Brauner vfsuid_t vfsuid; 1106ba73d987SChristian Brauner 1107800179c9SKees Cook if (!sysctl_protected_symlinks) 1108800179c9SKees Cook return 0; 1109800179c9SKees Cook 1110e67fe633SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 1111e67fe633SChristian Brauner vfsuid = i_uid_into_vfsuid(idmap, inode); 1112800179c9SKees Cook /* Allowed if owner and follower match. */ 1113a2bd096fSChristian Brauner if (vfsuid_eq_kuid(vfsuid, current_fsuid())) 1114800179c9SKees Cook return 0; 1115800179c9SKees Cook 1116800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11170f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1118800179c9SKees Cook return 0; 1119800179c9SKees Cook 1120800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1121a2bd096fSChristian Brauner if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid)) 1122800179c9SKees Cook return 0; 1123800179c9SKees Cook 112431956502SAl Viro if (nd->flags & LOOKUP_RCU) 112531956502SAl Viro return -ECHILD; 112631956502SAl Viro 1127ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1128245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1129800179c9SKees Cook return -EACCES; 1130800179c9SKees Cook } 1131800179c9SKees Cook 1132800179c9SKees Cook /** 1133800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 11344609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 1135800179c9SKees Cook * @inode: the source inode to hardlink from 1136800179c9SKees Cook * 1137800179c9SKees Cook * Return false if at least one of the following conditions: 1138800179c9SKees Cook * - inode is not a regular file 1139800179c9SKees Cook * - inode is setuid 1140800179c9SKees Cook * - inode is setgid and group-exec 1141800179c9SKees Cook * - access failure for read and write 1142800179c9SKees Cook * 1143800179c9SKees Cook * Otherwise returns true. 1144800179c9SKees Cook */ 11454609e1f1SChristian Brauner static bool safe_hardlink_source(struct mnt_idmap *idmap, 1146ba73d987SChristian Brauner struct inode *inode) 1147800179c9SKees Cook { 1148800179c9SKees Cook umode_t mode = inode->i_mode; 1149800179c9SKees Cook 1150800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1151800179c9SKees Cook if (!S_ISREG(mode)) 1152800179c9SKees Cook return false; 1153800179c9SKees Cook 1154800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1155800179c9SKees Cook if (mode & S_ISUID) 1156800179c9SKees Cook return false; 1157800179c9SKees Cook 1158800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1159800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1160800179c9SKees Cook return false; 1161800179c9SKees Cook 1162800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 11634609e1f1SChristian Brauner if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE)) 1164800179c9SKees Cook return false; 1165800179c9SKees Cook 1166800179c9SKees Cook return true; 1167800179c9SKees Cook } 1168800179c9SKees Cook 1169800179c9SKees Cook /** 1170800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 11714609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 1172800179c9SKees Cook * @link: the source to hardlink from 1173800179c9SKees Cook * 1174800179c9SKees Cook * Block hardlink when all of: 1175800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1176800179c9SKees Cook * - fsuid does not match inode 1177800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1178f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1179800179c9SKees Cook * 11804609e1f1SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 11814609e1f1SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 11824609e1f1SChristian Brauner * care to map the inode according to @idmap before checking permissions. 1183ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 11844609e1f1SChristian Brauner * raw inode simply pass @nop_mnt_idmap. 1185ba73d987SChristian Brauner * 1186800179c9SKees Cook * Returns 0 if successful, -ve on error. 1187800179c9SKees Cook */ 11884609e1f1SChristian Brauner int may_linkat(struct mnt_idmap *idmap, const struct path *link) 1189800179c9SKees Cook { 1190593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1191593d1ce8SEric W. Biederman 1192593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1193e67fe633SChristian Brauner if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) || 1194e67fe633SChristian Brauner !vfsgid_valid(i_gid_into_vfsgid(idmap, inode))) 1195593d1ce8SEric W. Biederman return -EOVERFLOW; 1196800179c9SKees Cook 1197800179c9SKees Cook if (!sysctl_protected_hardlinks) 1198800179c9SKees Cook return 0; 1199800179c9SKees Cook 1200800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1201800179c9SKees Cook * otherwise, it must be a safe source. 1202800179c9SKees Cook */ 12034609e1f1SChristian Brauner if (safe_hardlink_source(idmap, inode) || 120401beba79SChristian Brauner inode_owner_or_capable(idmap, inode)) 1205800179c9SKees Cook return 0; 1206800179c9SKees Cook 1207245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1208800179c9SKees Cook return -EPERM; 1209800179c9SKees Cook } 1210800179c9SKees Cook 121130aba665SSalvatore Mesoraca /** 121230aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 121330aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 121430aba665SSalvatore Mesoraca * exist. 1215e67fe633SChristian Brauner * @idmap: idmap of the mount the inode was found from 12162111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 121730aba665SSalvatore Mesoraca * @inode: the inode of the file to open 121830aba665SSalvatore Mesoraca * 121930aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 122030aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 122130aba665SSalvatore Mesoraca * - the file already exists 122230aba665SSalvatore Mesoraca * - we are in a sticky directory 122330aba665SSalvatore Mesoraca * - we don't own the file 122430aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 122530aba665SSalvatore Mesoraca * - the directory is world writable 122630aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 122730aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 122830aba665SSalvatore Mesoraca * be enough. 122930aba665SSalvatore Mesoraca * 1230e67fe633SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 1231e67fe633SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 1232e67fe633SChristian Brauner * care to map the inode according to @idmap before checking permissions. 1233ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1234e67fe633SChristian Brauner * raw inode simply pass @nop_mnt_idmap. 1235ba73d987SChristian Brauner * 123630aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 123730aba665SSalvatore Mesoraca */ 1238e67fe633SChristian Brauner static int may_create_in_sticky(struct mnt_idmap *idmap, 1239ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 124030aba665SSalvatore Mesoraca { 1241ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1242a2bd096fSChristian Brauner vfsuid_t dir_vfsuid = nd->dir_vfsuid; 1243ba73d987SChristian Brauner 124430aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 124530aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1246d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1247e67fe633SChristian Brauner vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) || 1248e67fe633SChristian Brauner vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid())) 124930aba665SSalvatore Mesoraca return 0; 125030aba665SSalvatore Mesoraca 1251d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1252d0cb5018SAl Viro (dir_mode & 0020 && 125330aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 125430aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1255245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1256245d7369SKees Cook "sticky_create_fifo" : 1257245d7369SKees Cook "sticky_create_regular"; 1258245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 125930aba665SSalvatore Mesoraca return -EACCES; 126030aba665SSalvatore Mesoraca } 126130aba665SSalvatore Mesoraca return 0; 126230aba665SSalvatore Mesoraca } 126330aba665SSalvatore Mesoraca 1264f015f126SDavid Howells /* 1265f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1266f015f126SDavid Howells * 1267f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1268f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1269f015f126SDavid Howells * Up is towards /. 1270f015f126SDavid Howells * 1271f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1272f015f126SDavid Howells * root. 1273f015f126SDavid Howells */ 1274bab77ebfSAl Viro int follow_up(struct path *path) 12751da177e4SLinus Torvalds { 12760714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12770714a533SAl Viro struct mount *parent; 12781da177e4SLinus Torvalds struct dentry *mountpoint; 127999b7db7bSNick Piggin 128048a066e7SAl Viro read_seqlock_excl(&mount_lock); 12810714a533SAl Viro parent = mnt->mnt_parent; 12823c0a6163SAl Viro if (parent == mnt) { 128348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12841da177e4SLinus Torvalds return 0; 12851da177e4SLinus Torvalds } 12860714a533SAl Viro mntget(&parent->mnt); 1287a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 128848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1289bab77ebfSAl Viro dput(path->dentry); 1290bab77ebfSAl Viro path->dentry = mountpoint; 1291bab77ebfSAl Viro mntput(path->mnt); 12920714a533SAl Viro path->mnt = &parent->mnt; 12931da177e4SLinus Torvalds return 1; 12941da177e4SLinus Torvalds } 12954d359507SAl Viro EXPORT_SYMBOL(follow_up); 12961da177e4SLinus Torvalds 12977ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12987ef482faSAl Viro struct path *path, unsigned *seqp) 12997ef482faSAl Viro { 13007ef482faSAl Viro while (mnt_has_parent(m)) { 13017ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 13027ef482faSAl Viro 13037ef482faSAl Viro m = m->mnt_parent; 13047ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 13057ef482faSAl Viro root->mnt == &m->mnt)) 13067ef482faSAl Viro break; 13077ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 13087ef482faSAl Viro path->mnt = &m->mnt; 13097ef482faSAl Viro path->dentry = mountpoint; 13107ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 13117ef482faSAl Viro return true; 13127ef482faSAl Viro } 13137ef482faSAl Viro } 13147ef482faSAl Viro return false; 13157ef482faSAl Viro } 13167ef482faSAl Viro 13172aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 13182aa38470SAl Viro struct path *path) 13192aa38470SAl Viro { 13202aa38470SAl Viro bool found; 13212aa38470SAl Viro 13222aa38470SAl Viro rcu_read_lock(); 13232aa38470SAl Viro while (1) { 13242aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 13252aa38470SAl Viro 13262aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 13272aa38470SAl Viro if (unlikely(!found)) { 13282aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 13292aa38470SAl Viro break; 13302aa38470SAl Viro } else { 13312aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 13322aa38470SAl Viro break; 13332aa38470SAl Viro rcu_read_unlock(); 13342aa38470SAl Viro path_put(path); 13352aa38470SAl Viro rcu_read_lock(); 13362aa38470SAl Viro } 13372aa38470SAl Viro } 13382aa38470SAl Viro rcu_read_unlock(); 13392aa38470SAl Viro return found; 13402aa38470SAl Viro } 13412aa38470SAl Viro 1342b5c84bf6SNick Piggin /* 13439875cf80SDavid Howells * Perform an automount 13449875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 13459875cf80SDavid Howells * were called with. 13461da177e4SLinus Torvalds */ 13471c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 134831e6b01fSNick Piggin { 134925e195aaSAl Viro struct dentry *dentry = path->dentry; 13509875cf80SDavid Howells 13510ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 13520ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 13530ec26fd0SMiklos Szeredi * the name. 13540ec26fd0SMiklos Szeredi * 13550ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 13565a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 13570ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 13580ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13590ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13600ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13615a30d8a2SDavid Howells */ 13621c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13635d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 136425e195aaSAl Viro dentry->d_inode) 13659875cf80SDavid Howells return -EISDIR; 13660ec26fd0SMiklos Szeredi 13671c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13689875cf80SDavid Howells return -ELOOP; 13699875cf80SDavid Howells 137025e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1371ea5b778aSDavid Howells } 13729875cf80SDavid Howells 13739875cf80SDavid Howells /* 13749deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13759deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13769deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13779deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13789875cf80SDavid Howells */ 13799deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13809deed3ebSAl Viro int *count, unsigned lookup_flags) 13819875cf80SDavid Howells { 13829deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13839875cf80SDavid Howells bool need_mntput = false; 13848aef1884SAl Viro int ret = 0; 13859875cf80SDavid Howells 13869deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1387cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1388cc53ce53SDavid Howells * being held. */ 1389d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1390fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1391508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1392cc53ce53SDavid Howells if (ret < 0) 13938aef1884SAl Viro break; 1394cc53ce53SDavid Howells } 1395cc53ce53SDavid Howells 13969deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13979875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13989deed3ebSAl Viro if (mounted) { // ... in our namespace 13999875cf80SDavid Howells dput(path->dentry); 14009875cf80SDavid Howells if (need_mntput) 1401463ffb2eSAl Viro mntput(path->mnt); 1402463ffb2eSAl Viro path->mnt = mounted; 1403463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 14049deed3ebSAl Viro // here we know it's positive 14059deed3ebSAl Viro flags = path->dentry->d_flags; 14069875cf80SDavid Howells need_mntput = true; 14079875cf80SDavid Howells continue; 1408463ffb2eSAl Viro } 14091da177e4SLinus Torvalds } 14109875cf80SDavid Howells 14119deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 14129deed3ebSAl Viro break; 14139deed3ebSAl Viro 14149deed3ebSAl Viro // uncovered automount point 14159deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 14169deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 14179875cf80SDavid Howells if (ret < 0) 14188aef1884SAl Viro break; 14199875cf80SDavid Howells } 14209875cf80SDavid Howells 14219deed3ebSAl Viro if (ret == -EISDIR) 14229deed3ebSAl Viro ret = 0; 14239deed3ebSAl Viro // possible if you race with several mount --move 14249deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 14258aef1884SAl Viro mntput(path->mnt); 14269deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1427d41efb52SAl Viro ret = -ENOENT; 14289deed3ebSAl Viro *jumped = need_mntput; 14298402752eSAl Viro return ret; 14301da177e4SLinus Torvalds } 14311da177e4SLinus Torvalds 14329deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 14339deed3ebSAl Viro int *count, unsigned lookup_flags) 14349deed3ebSAl Viro { 14359deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 14369deed3ebSAl Viro 14379deed3ebSAl Viro /* fastpath */ 14389deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 14399deed3ebSAl Viro *jumped = false; 14409deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 14419deed3ebSAl Viro return -ENOENT; 14429deed3ebSAl Viro return 0; 14439deed3ebSAl Viro } 14449deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 14459deed3ebSAl Viro } 14469deed3ebSAl Viro 1447cc53ce53SDavid Howells int follow_down_one(struct path *path) 14481da177e4SLinus Torvalds { 14491da177e4SLinus Torvalds struct vfsmount *mounted; 14501da177e4SLinus Torvalds 14511c755af4SAl Viro mounted = lookup_mnt(path); 14521da177e4SLinus Torvalds if (mounted) { 14539393bd07SAl Viro dput(path->dentry); 14549393bd07SAl Viro mntput(path->mnt); 14559393bd07SAl Viro path->mnt = mounted; 14569393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 14571da177e4SLinus Torvalds return 1; 14581da177e4SLinus Torvalds } 14591da177e4SLinus Torvalds return 0; 14601da177e4SLinus Torvalds } 14614d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14621da177e4SLinus Torvalds 14639875cf80SDavid Howells /* 14649deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14659deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14669deed3ebSAl Viro * caller is permitted to proceed or not. 14679deed3ebSAl Viro */ 1468e1f19857SRichard Weinberger int follow_down(struct path *path, unsigned int flags) 14699deed3ebSAl Viro { 14709deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14719deed3ebSAl Viro bool jumped; 1472e1f19857SRichard Weinberger int ret = traverse_mounts(path, &jumped, NULL, flags); 14739deed3ebSAl Viro 14749deed3ebSAl Viro if (path->mnt != mnt) 14759deed3ebSAl Viro mntput(mnt); 14769deed3ebSAl Viro return ret; 14779deed3ebSAl Viro } 14789deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14799deed3ebSAl Viro 14809deed3ebSAl Viro /* 1481287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1482287548e4SAl Viro * we meet a managed dentry that would need blocking. 14839875cf80SDavid Howells */ 14843bd8bc89SAl Viro static bool __follow_mount_rcu(struct nameidata *nd, struct path *path) 14859875cf80SDavid Howells { 1486ea936aebSAl Viro struct dentry *dentry = path->dentry; 1487ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1488ea936aebSAl Viro 1489ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1490ea936aebSAl Viro return true; 1491ea936aebSAl Viro 1492ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1493ea936aebSAl Viro return false; 1494ea936aebSAl Viro 149562a7375eSIan Kent for (;;) { 149662a7375eSIan Kent /* 149762a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 149862a7375eSIan Kent * that wants to block transit. 149962a7375eSIan Kent */ 1500ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1501ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1502ea936aebSAl Viro if (res) 1503ea936aebSAl Viro return res == -EISDIR; 1504ea936aebSAl Viro flags = dentry->d_flags; 1505b8faf035SNeilBrown } 150662a7375eSIan Kent 1507ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1508ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1509ea936aebSAl Viro if (mounted) { 1510c7105365SAl Viro path->mnt = &mounted->mnt; 1511ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1512bcba1e7dSAl Viro nd->state |= ND_JUMPED; 151303fa86e9SAl Viro nd->next_seq = read_seqcount_begin(&dentry->d_seq); 1514ea936aebSAl Viro flags = dentry->d_flags; 151503fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach 151603fa86e9SAl Viro // this state. 151720aac6c6SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 151820aac6c6SAl Viro return false; 1519ea936aebSAl Viro continue; 15209875cf80SDavid Howells } 1521ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1522ea936aebSAl Viro return false; 1523ea936aebSAl Viro } 1524ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1525ea936aebSAl Viro } 1526287548e4SAl Viro } 1527287548e4SAl Viro 1528db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 15293bd8bc89SAl Viro struct path *path) 1530bd7c4b50SAl Viro { 15319deed3ebSAl Viro bool jumped; 1532db3c9adeSAl Viro int ret; 1533bd7c4b50SAl Viro 1534db3c9adeSAl Viro path->mnt = nd->path.mnt; 1535db3c9adeSAl Viro path->dentry = dentry; 1536c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 153703fa86e9SAl Viro unsigned int seq = nd->next_seq; 15383bd8bc89SAl Viro if (likely(__follow_mount_rcu(nd, path))) 15399deed3ebSAl Viro return 0; 154003fa86e9SAl Viro // *path and nd->next_seq might've been clobbered 1541c153007bSAl Viro path->mnt = nd->path.mnt; 1542c153007bSAl Viro path->dentry = dentry; 154303fa86e9SAl Viro nd->next_seq = seq; 154403fa86e9SAl Viro if (!try_to_unlazy_next(nd, dentry)) 154503fa86e9SAl Viro return -ECHILD; 1546c153007bSAl Viro } 15479deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 15489deed3ebSAl Viro if (jumped) { 15499deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 15509deed3ebSAl Viro ret = -EXDEV; 15519deed3ebSAl Viro else 1552bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15539deed3ebSAl Viro } 15549deed3ebSAl Viro if (unlikely(ret)) { 15559deed3ebSAl Viro dput(path->dentry); 15569deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15579deed3ebSAl Viro mntput(path->mnt); 1558bd7c4b50SAl Viro } 1559bd7c4b50SAl Viro return ret; 1560bd7c4b50SAl Viro } 1561bd7c4b50SAl Viro 15629875cf80SDavid Howells /* 1563f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1564f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1565baa03890SNick Piggin */ 1566e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1567e3c13928SAl Viro struct dentry *dir, 15686c51e513SAl Viro unsigned int flags) 1569baa03890SNick Piggin { 1570a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1571bad61189SMiklos Szeredi if (dentry) { 1572a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1573bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 157474ff0ffcSAl Viro if (!error) 15755542aa2fSEric W. Biederman d_invalidate(dentry); 1576bad61189SMiklos Szeredi dput(dentry); 157774ff0ffcSAl Viro return ERR_PTR(error); 1578bad61189SMiklos Szeredi } 1579bad61189SMiklos Szeredi } 1580baa03890SNick Piggin return dentry; 1581baa03890SNick Piggin } 1582baa03890SNick Piggin 1583baa03890SNick Piggin /* 1584a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1585a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1586a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1587a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1588a03ece5fSAl Viro * at all. 158944396f4bSJosef Bacik */ 159074d7970fSNamjae Jeon struct dentry *lookup_one_qstr_excl(const struct qstr *name, 159174d7970fSNamjae Jeon struct dentry *base, 159274d7970fSNamjae Jeon unsigned int flags) 159344396f4bSJosef Bacik { 1594a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 159544396f4bSJosef Bacik struct dentry *old; 1596a03ece5fSAl Viro struct inode *dir = base->d_inode; 1597a03ece5fSAl Viro 1598a03ece5fSAl Viro if (dentry) 1599a03ece5fSAl Viro return dentry; 160044396f4bSJosef Bacik 160144396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1602a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 160344396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1604a03ece5fSAl Viro 1605a03ece5fSAl Viro dentry = d_alloc(base, name); 1606a03ece5fSAl Viro if (unlikely(!dentry)) 1607a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 160844396f4bSJosef Bacik 160972bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 161044396f4bSJosef Bacik if (unlikely(old)) { 161144396f4bSJosef Bacik dput(dentry); 161244396f4bSJosef Bacik dentry = old; 161344396f4bSJosef Bacik } 161444396f4bSJosef Bacik return dentry; 161544396f4bSJosef Bacik } 161674d7970fSNamjae Jeon EXPORT_SYMBOL(lookup_one_qstr_excl); 161744396f4bSJosef Bacik 16184cb64024SAl Viro static struct dentry *lookup_fast(struct nameidata *nd) 16191da177e4SLinus Torvalds { 162031e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16215a18fff2SAl Viro int status = 1; 16229875cf80SDavid Howells 16233cac260aSAl Viro /* 1624b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16255d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16265d0f49c1SAl Viro * going to fall back to non-racy lookup. 1627b04f784eSNick Piggin */ 162831e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 162903fa86e9SAl Viro dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq); 16305d0f49c1SAl Viro if (unlikely(!dentry)) { 1631e36cffedSJens Axboe if (!try_to_unlazy(nd)) 163220e34357SAl Viro return ERR_PTR(-ECHILD); 163320e34357SAl Viro return NULL; 16345d0f49c1SAl Viro } 16355a18fff2SAl Viro 163612f8ad4bSLinus Torvalds /* 163712f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 163812f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 163912f8ad4bSLinus Torvalds */ 16404cb64024SAl Viro if (read_seqcount_retry(&parent->d_seq, nd->seq)) 164120e34357SAl Viro return ERR_PTR(-ECHILD); 16425a18fff2SAl Viro 16434ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1644c153007bSAl Viro if (likely(status > 0)) 164520e34357SAl Viro return dentry; 164603fa86e9SAl Viro if (!try_to_unlazy_next(nd, dentry)) 164720e34357SAl Viro return ERR_PTR(-ECHILD); 164826ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1649209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1650209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16515a18fff2SAl Viro } else { 1652e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 165381e6f520SAl Viro if (unlikely(!dentry)) 165420e34357SAl Viro return NULL; 16554ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16565d0f49c1SAl Viro } 16575a18fff2SAl Viro if (unlikely(status <= 0)) { 1658e9742b53SAl Viro if (!status) 16595d0f49c1SAl Viro d_invalidate(dentry); 16605a18fff2SAl Viro dput(dentry); 166120e34357SAl Viro return ERR_PTR(status); 16625a18fff2SAl Viro } 166320e34357SAl Viro return dentry; 1664697f514dSMiklos Szeredi } 1665697f514dSMiklos Szeredi 1666697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 166788d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1668e3c13928SAl Viro struct dentry *dir, 1669e3c13928SAl Viro unsigned int flags) 1670697f514dSMiklos Szeredi { 167188d8331aSAl Viro struct dentry *dentry, *old; 16721936386eSAl Viro struct inode *inode = dir->d_inode; 1673d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16741936386eSAl Viro 16751936386eSAl Viro /* Don't go there if it's already dead */ 167694bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 167788d8331aSAl Viro return ERR_PTR(-ENOENT); 167894bdd655SAl Viro again: 1679d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 168094bdd655SAl Viro if (IS_ERR(dentry)) 168188d8331aSAl Viro return dentry; 168294bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1683949a852eSAl Viro int error = d_revalidate(dentry, flags); 1684949a852eSAl Viro if (unlikely(error <= 0)) { 168594bdd655SAl Viro if (!error) { 1686949a852eSAl Viro d_invalidate(dentry); 1687949a852eSAl Viro dput(dentry); 168894bdd655SAl Viro goto again; 168994bdd655SAl Viro } 169094bdd655SAl Viro dput(dentry); 1691949a852eSAl Viro dentry = ERR_PTR(error); 1692949a852eSAl Viro } 169394bdd655SAl Viro } else { 16941936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 169585c7f810SAl Viro d_lookup_done(dentry); 16961936386eSAl Viro if (unlikely(old)) { 16971936386eSAl Viro dput(dentry); 16981936386eSAl Viro dentry = old; 1699949a852eSAl Viro } 1700949a852eSAl Viro } 1701e3c13928SAl Viro return dentry; 17021da177e4SLinus Torvalds } 17031da177e4SLinus Torvalds 170488d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 170588d8331aSAl Viro struct dentry *dir, 170688d8331aSAl Viro unsigned int flags) 170788d8331aSAl Viro { 170888d8331aSAl Viro struct inode *inode = dir->d_inode; 170988d8331aSAl Viro struct dentry *res; 171088d8331aSAl Viro inode_lock_shared(inode); 171188d8331aSAl Viro res = __lookup_slow(name, dir, flags); 171288d8331aSAl Viro inode_unlock_shared(inode); 171388d8331aSAl Viro return res; 171488d8331aSAl Viro } 171588d8331aSAl Viro 17164609e1f1SChristian Brauner static inline int may_lookup(struct mnt_idmap *idmap, 1717ba73d987SChristian Brauner struct nameidata *nd) 171852094c8aSAl Viro { 171952094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17204609e1f1SChristian Brauner int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1721e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 172252094c8aSAl Viro return err; 172352094c8aSAl Viro } 17244609e1f1SChristian Brauner return inode_permission(idmap, nd->inode, MAY_EXEC); 172552094c8aSAl Viro } 172652094c8aSAl Viro 172703fa86e9SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link) 1728d63ff28fSAl Viro { 172949055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 173049055906SAl Viro return -ELOOP; 17314542576bSAl Viro 17324542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17334542576bSAl Viro return 0; 17344542576bSAl Viro if (likely(nd->stack != nd->internal)) 17354542576bSAl Viro return 0; 173660ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 173749055906SAl Viro return 0; 173860ef60c7SAl Viro 173960ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 174060ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 174160ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 174203fa86e9SAl Viro bool grabbed_link = legitimize_path(nd, link, nd->next_seq); 174360ef60c7SAl Viro 1744e5ca024eSAl Viro if (!try_to_unlazy(nd) || !grabbed_link) 174560ef60c7SAl Viro return -ECHILD; 174660ef60c7SAl Viro 174760ef60c7SAl Viro if (nd_alloc_stack(nd)) 174860ef60c7SAl Viro return 0; 1749bc40aee0SAl Viro } 175060ef60c7SAl Viro return -ENOMEM; 175149055906SAl Viro } 175249055906SAl Viro 175349055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 175449055906SAl Viro 175549055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 175603fa86e9SAl Viro struct inode *inode, int flags) 175749055906SAl Viro { 175849055906SAl Viro struct saved *last; 175949055906SAl Viro const char *res; 176003fa86e9SAl Viro int error = reserve_stack(nd, link); 176149055906SAl Viro 176249055906SAl Viro if (unlikely(error)) { 176349055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1764cd179f44SAl Viro path_put(link); 176506708adbSAl Viro return ERR_PTR(error); 1766626de996SAl Viro } 1767ab104923SAl Viro last = nd->stack + nd->depth++; 17681cf2665bSAl Viro last->link = *link; 1769fceef393SAl Viro clear_delayed_call(&last->done); 177003fa86e9SAl Viro last->seq = nd->next_seq; 1771ad6cc4c3SAl Viro 1772b1a81972SAl Viro if (flags & WALK_TRAILING) { 1773ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1774ad6cc4c3SAl Viro if (unlikely(error)) 1775ad6cc4c3SAl Viro return ERR_PTR(error); 1776ad6cc4c3SAl Viro } 1777ad6cc4c3SAl Viro 1778dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1779dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1780ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1781ad6cc4c3SAl Viro 1782ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1783ad6cc4c3SAl Viro touch_atime(&last->link); 1784ad6cc4c3SAl Viro cond_resched(); 1785ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1786e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1787ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1788ad6cc4c3SAl Viro touch_atime(&last->link); 1789ad6cc4c3SAl Viro } 1790ad6cc4c3SAl Viro 1791ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1792ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1793ad6cc4c3SAl Viro if (unlikely(error)) 1794ad6cc4c3SAl Viro return ERR_PTR(error); 1795ad6cc4c3SAl Viro 1796ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1797ad6cc4c3SAl Viro if (!res) { 1798ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1799ad6cc4c3SAl Viro struct delayed_call *); 1800ad6cc4c3SAl Viro get = inode->i_op->get_link; 1801ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1802ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1803e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1804ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1805ad6cc4c3SAl Viro } else { 1806ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1807ad6cc4c3SAl Viro } 1808ad6cc4c3SAl Viro if (!res) 1809ad6cc4c3SAl Viro goto all_done; 1810ad6cc4c3SAl Viro if (IS_ERR(res)) 1811ad6cc4c3SAl Viro return res; 1812ad6cc4c3SAl Viro } 1813ad6cc4c3SAl Viro if (*res == '/') { 1814ad6cc4c3SAl Viro error = nd_jump_root(nd); 1815ad6cc4c3SAl Viro if (unlikely(error)) 1816ad6cc4c3SAl Viro return ERR_PTR(error); 1817ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1818ad6cc4c3SAl Viro ; 1819ad6cc4c3SAl Viro } 1820ad6cc4c3SAl Viro if (*res) 1821ad6cc4c3SAl Viro return res; 1822ad6cc4c3SAl Viro all_done: // pure jump 1823ad6cc4c3SAl Viro put_link(nd); 1824ad6cc4c3SAl Viro return NULL; 1825d63ff28fSAl Viro } 1826d63ff28fSAl Viro 18273ddcd056SLinus Torvalds /* 18283ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18293ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18303ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18313ddcd056SLinus Torvalds * for the common case. 183203fa86e9SAl Viro * 183303fa86e9SAl Viro * NOTE: dentry must be what nd->next_seq had been sampled from. 18343ddcd056SLinus Torvalds */ 1835b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1836a4f5b521SAl Viro struct dentry *dentry) 18373ddcd056SLinus Torvalds { 1838cbae4d12SAl Viro struct path path; 1839a4f5b521SAl Viro struct inode *inode; 18403bd8bc89SAl Viro int err = handle_mounts(nd, dentry, &path); 1841cbae4d12SAl Viro 1842cbae4d12SAl Viro if (err < 0) 1843b0417d2cSAl Viro return ERR_PTR(err); 18443bd8bc89SAl Viro inode = path.dentry->d_inode; 1845cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18468c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1847aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18488f64fb1cSAl Viro /* not a symlink or should not follow */ 18493bd8bc89SAl Viro if (nd->flags & LOOKUP_RCU) { 18503bd8bc89SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) 18513bd8bc89SAl Viro return ERR_PTR(-ECHILD); 18523bd8bc89SAl Viro if (unlikely(!inode)) 18533bd8bc89SAl Viro return ERR_PTR(-ENOENT); 18543bd8bc89SAl Viro } else { 1855c99687a0SAl Viro dput(nd->path.dentry); 1856c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1857c99687a0SAl Viro mntput(nd->path.mnt); 1858c99687a0SAl Viro } 1859c99687a0SAl Viro nd->path = path; 18608f64fb1cSAl Viro nd->inode = inode; 186103fa86e9SAl Viro nd->seq = nd->next_seq; 1862b0417d2cSAl Viro return NULL; 18638f64fb1cSAl Viro } 1864a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 186584f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 186603fa86e9SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) 1867b0417d2cSAl Viro return ERR_PTR(-ECHILD); 186884f0cd9eSAl Viro } else { 186984f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 187084f0cd9eSAl Viro mntget(path.mnt); 1871a7f77542SAl Viro } 187203fa86e9SAl Viro return pick_link(nd, &path, inode, flags); 18733ddcd056SLinus Torvalds } 18743ddcd056SLinus Torvalds 1875b16c001dSAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd) 1876957dd41dSAl Viro { 187712487f30SAl Viro struct dentry *parent, *old; 1878957dd41dSAl Viro 187912487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 188012487f30SAl Viro goto in_root; 188112487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18827ef482faSAl Viro struct path path; 1883efe772d6SAl Viro unsigned seq; 18847ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18857ef482faSAl Viro &nd->root, &path, &seq)) 188612487f30SAl Viro goto in_root; 1887efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1888efe772d6SAl Viro return ERR_PTR(-ECHILD); 1889efe772d6SAl Viro nd->path = path; 1890efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1891efe772d6SAl Viro nd->seq = seq; 189203fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach this state 189382ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1894efe772d6SAl Viro return ERR_PTR(-ECHILD); 1895efe772d6SAl Viro /* we know that mountpoint was pinned */ 1896957dd41dSAl Viro } 189712487f30SAl Viro old = nd->path.dentry; 189812487f30SAl Viro parent = old->d_parent; 189903fa86e9SAl Viro nd->next_seq = read_seqcount_begin(&parent->d_seq); 190003fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach this state 190182ef0698SAl Viro if (read_seqcount_retry(&old->d_seq, nd->seq)) 190212487f30SAl Viro return ERR_PTR(-ECHILD); 190312487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 190412487f30SAl Viro return ERR_PTR(-ECHILD); 190512487f30SAl Viro return parent; 190612487f30SAl Viro in_root: 190782ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1908efe772d6SAl Viro return ERR_PTR(-ECHILD); 1909957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19107521f22bSAl Viro return ERR_PTR(-ECHILD); 191103fa86e9SAl Viro nd->next_seq = nd->seq; 191251c6546cSAl Viro return nd->path.dentry; 1913957dd41dSAl Viro } 1914957dd41dSAl Viro 1915b16c001dSAl Viro static struct dentry *follow_dotdot(struct nameidata *nd) 1916957dd41dSAl Viro { 191712487f30SAl Viro struct dentry *parent; 191812487f30SAl Viro 1919957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 192012487f30SAl Viro goto in_root; 192112487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19222aa38470SAl Viro struct path path; 19232aa38470SAl Viro 19242aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19252aa38470SAl Viro &nd->root, &path)) 192612487f30SAl Viro goto in_root; 1927165200d6SAl Viro path_put(&nd->path); 1928165200d6SAl Viro nd->path = path; 19292aa38470SAl Viro nd->inode = path.dentry->d_inode; 1930165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1931165200d6SAl Viro return ERR_PTR(-EXDEV); 193212487f30SAl Viro } 1933957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 193412487f30SAl Viro parent = dget_parent(nd->path.dentry); 1935957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1936957dd41dSAl Viro dput(parent); 19377521f22bSAl Viro return ERR_PTR(-ENOENT); 1938957dd41dSAl Viro } 1939c2df1968SAl Viro return parent; 194012487f30SAl Viro 194112487f30SAl Viro in_root: 1942957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19437521f22bSAl Viro return ERR_PTR(-EXDEV); 194451c6546cSAl Viro return dget(nd->path.dentry); 1945957dd41dSAl Viro } 1946957dd41dSAl Viro 19477521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1948957dd41dSAl Viro { 1949957dd41dSAl Viro if (type == LAST_DOTDOT) { 19507521f22bSAl Viro const char *error = NULL; 1951c2df1968SAl Viro struct dentry *parent; 1952957dd41dSAl Viro 1953957dd41dSAl Viro if (!nd->root.mnt) { 19547521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1955957dd41dSAl Viro if (error) 1956957dd41dSAl Viro return error; 1957957dd41dSAl Viro } 1958957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1959b16c001dSAl Viro parent = follow_dotdot_rcu(nd); 1960957dd41dSAl Viro else 1961b16c001dSAl Viro parent = follow_dotdot(nd); 1962c2df1968SAl Viro if (IS_ERR(parent)) 1963c2df1968SAl Viro return ERR_CAST(parent); 1964a4f5b521SAl Viro error = step_into(nd, WALK_NOFOLLOW, parent); 1965c2df1968SAl Viro if (unlikely(error)) 1966957dd41dSAl Viro return error; 1967957dd41dSAl Viro 1968957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1969957dd41dSAl Viro /* 1970957dd41dSAl Viro * If there was a racing rename or mount along our 1971957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1972957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1973957dd41dSAl Viro * some fallback). 1974957dd41dSAl Viro */ 1975957dd41dSAl Viro smp_rmb(); 197682ef0698SAl Viro if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)) 19777521f22bSAl Viro return ERR_PTR(-EAGAIN); 197882ef0698SAl Viro if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)) 19797521f22bSAl Viro return ERR_PTR(-EAGAIN); 1980957dd41dSAl Viro } 1981957dd41dSAl Viro } 19827521f22bSAl Viro return NULL; 1983957dd41dSAl Viro } 1984957dd41dSAl Viro 198592d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1986ce57dfc1SAl Viro { 1987db3c9adeSAl Viro struct dentry *dentry; 1988ce57dfc1SAl Viro /* 1989ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1990ce57dfc1SAl Viro * to be able to know about the current root directory and 1991ce57dfc1SAl Viro * parent relationships. 1992ce57dfc1SAl Viro */ 19934693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 19941c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 19954693a547SAl Viro put_link(nd); 19967521f22bSAl Viro return handle_dots(nd, nd->last_type); 19974693a547SAl Viro } 19984cb64024SAl Viro dentry = lookup_fast(nd); 199920e34357SAl Viro if (IS_ERR(dentry)) 200092d27016SAl Viro return ERR_CAST(dentry); 200120e34357SAl Viro if (unlikely(!dentry)) { 2002db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 2003db3c9adeSAl Viro if (IS_ERR(dentry)) 200492d27016SAl Viro return ERR_CAST(dentry); 200520e34357SAl Viro } 200656676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 200756676ec3SAl Viro put_link(nd); 2008a4f5b521SAl Viro return step_into(nd, flags, dentry); 2009ce57dfc1SAl Viro } 2010ce57dfc1SAl Viro 20111da177e4SLinus Torvalds /* 2012bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2013bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2014bfcfaa77SLinus Torvalds * 2015bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2016bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2017bfcfaa77SLinus Torvalds * fast. 2018bfcfaa77SLinus Torvalds * 2019bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2020bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2021bfcfaa77SLinus Torvalds * crossing operation. 2022bfcfaa77SLinus Torvalds * 2023bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2024bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2025bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2026bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2027bfcfaa77SLinus Torvalds */ 2028bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2029bfcfaa77SLinus Torvalds 2030f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2031bfcfaa77SLinus Torvalds 2032468a9428SGeorge Spelvin #ifdef HASH_MIX 2033bfcfaa77SLinus Torvalds 2034468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2035468a9428SGeorge Spelvin 2036468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20372a18da7aSGeorge Spelvin /* 20382a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20392a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20402a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20412a18da7aSGeorge Spelvin * and no temporaries. 20422a18da7aSGeorge Spelvin * 20432a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20442a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20452a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20462a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20472a18da7aSGeorge Spelvin * 20482a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20492a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20502a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20512a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20522a18da7aSGeorge Spelvin * 20532a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20542a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20552a18da7aSGeorge Spelvin * toggling any given output bit. 20562a18da7aSGeorge Spelvin * 20572a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20582a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20592a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20602a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20612a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20622a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20632a18da7aSGeorge Spelvin * Perfect: 8192 258048 20642a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20652a18da7aSGeorge Spelvin */ 20662a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20672a18da7aSGeorge Spelvin ( x ^= (a), \ 20682a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20692a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20702a18da7aSGeorge Spelvin y *= 9 ) 2071bfcfaa77SLinus Torvalds 20720fed3ac8SGeorge Spelvin /* 20732a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20742a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20752a18da7aSGeorge Spelvin * work done before the hash value is used. 20760fed3ac8SGeorge Spelvin */ 20772a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20780fed3ac8SGeorge Spelvin { 20792a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20802a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20812a18da7aSGeorge Spelvin return y >> 32; 20820fed3ac8SGeorge Spelvin } 20830fed3ac8SGeorge Spelvin 2084bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2085bfcfaa77SLinus Torvalds 20862a18da7aSGeorge Spelvin /* 20872a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 20882a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20892a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 20902a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 20912a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 20922a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 20932a18da7aSGeorge Spelvin * Perfect: 2048 31744 20942a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 20952a18da7aSGeorge Spelvin */ 20962a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20972a18da7aSGeorge Spelvin ( x ^= (a), \ 20982a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 20992a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 21002a18da7aSGeorge Spelvin y *= 9 ) 2101bfcfaa77SLinus Torvalds 21022a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 21030fed3ac8SGeorge Spelvin { 21042a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 21052a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 21060fed3ac8SGeorge Spelvin } 21070fed3ac8SGeorge Spelvin 2108bfcfaa77SLinus Torvalds #endif 2109bfcfaa77SLinus Torvalds 21102a18da7aSGeorge Spelvin /* 21112a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21122a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21132a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21142a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21152a18da7aSGeorge Spelvin * finds the delimiter after the name. 21162a18da7aSGeorge Spelvin */ 21178387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2118bfcfaa77SLinus Torvalds { 21198387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2120bfcfaa77SLinus Torvalds 2121bfcfaa77SLinus Torvalds for (;;) { 2122fcfd2fbfSGeorge Spelvin if (!len) 2123fcfd2fbfSGeorge Spelvin goto done; 2124e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2125bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2126bfcfaa77SLinus Torvalds break; 21272a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2128bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2129bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2130bfcfaa77SLinus Torvalds } 21312a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2132bfcfaa77SLinus Torvalds done: 21332a18da7aSGeorge Spelvin return fold_hash(x, y); 2134bfcfaa77SLinus Torvalds } 2135bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2136bfcfaa77SLinus Torvalds 2137fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21388387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2139fcfd2fbfSGeorge Spelvin { 21408387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21418387ff25SLinus Torvalds unsigned long adata, mask, len; 2142fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2143fcfd2fbfSGeorge Spelvin 21448387ff25SLinus Torvalds len = 0; 21458387ff25SLinus Torvalds goto inside; 21468387ff25SLinus Torvalds 2147fcfd2fbfSGeorge Spelvin do { 21482a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2149fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21508387ff25SLinus Torvalds inside: 2151fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2152fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2153fcfd2fbfSGeorge Spelvin 2154fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2155fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21562a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2157fcfd2fbfSGeorge Spelvin 21582a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2159fcfd2fbfSGeorge Spelvin } 2160fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2161fcfd2fbfSGeorge Spelvin 2162bfcfaa77SLinus Torvalds /* 2163bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2164d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2165bfcfaa77SLinus Torvalds */ 21668387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2167bfcfaa77SLinus Torvalds { 21688387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21698387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 217036126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2171bfcfaa77SLinus Torvalds 21728387ff25SLinus Torvalds len = 0; 21738387ff25SLinus Torvalds goto inside; 21748387ff25SLinus Torvalds 2175bfcfaa77SLinus Torvalds do { 21762a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2177bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21788387ff25SLinus Torvalds inside: 2179e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 218036126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 218136126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2182bfcfaa77SLinus Torvalds 218336126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 218436126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 218536126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21862a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 218736126f8fSLinus Torvalds 21882a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2189bfcfaa77SLinus Torvalds } 2190bfcfaa77SLinus Torvalds 21912a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2192bfcfaa77SLinus Torvalds 2193fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 21948387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 21950145acc2SLinus Torvalds { 21968387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 21970145acc2SLinus Torvalds while (len--) 2198fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 21990145acc2SLinus Torvalds return end_name_hash(hash); 22000145acc2SLinus Torvalds } 2201ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 22020145acc2SLinus Torvalds 2203fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 22048387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2205fcfd2fbfSGeorge Spelvin { 22068387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2207fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2208fcfd2fbfSGeorge Spelvin 2209fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2210e0ab7af9SGeorge Spelvin while (c) { 2211fcfd2fbfSGeorge Spelvin len++; 2212fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2213fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2214e0ab7af9SGeorge Spelvin } 2215fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2216fcfd2fbfSGeorge Spelvin } 2217f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2218fcfd2fbfSGeorge Spelvin 22193ddcd056SLinus Torvalds /* 2220200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2221200e9ef7SLinus Torvalds * one character. 2222200e9ef7SLinus Torvalds */ 22238387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2224200e9ef7SLinus Torvalds { 22258387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2226200e9ef7SLinus Torvalds unsigned long len = 0, c; 2227200e9ef7SLinus Torvalds 2228200e9ef7SLinus Torvalds c = (unsigned char)*name; 2229200e9ef7SLinus Torvalds do { 2230200e9ef7SLinus Torvalds len++; 2231200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2232200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2233200e9ef7SLinus Torvalds } while (c && c != '/'); 2234d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2235200e9ef7SLinus Torvalds } 2236200e9ef7SLinus Torvalds 2237bfcfaa77SLinus Torvalds #endif 2238bfcfaa77SLinus Torvalds 2239200e9ef7SLinus Torvalds /* 22401da177e4SLinus Torvalds * Name resolution. 2241ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2242ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22431da177e4SLinus Torvalds * 2244ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2245ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22461da177e4SLinus Torvalds */ 22476de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22481da177e4SLinus Torvalds { 2249d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22501da177e4SLinus Torvalds int err; 22511da177e4SLinus Torvalds 2252b4c03536SAl Viro nd->last_type = LAST_ROOT; 2253c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22549b5858e9SAl Viro if (IS_ERR(name)) 22559b5858e9SAl Viro return PTR_ERR(name); 22561da177e4SLinus Torvalds while (*name=='/') 22571da177e4SLinus Torvalds name++; 22581a97d899SAl Viro if (!*name) { 22591a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22609e18f10aSAl Viro return 0; 22611a97d899SAl Viro } 22621da177e4SLinus Torvalds 22631da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22641da177e4SLinus Torvalds for(;;) { 22654609e1f1SChristian Brauner struct mnt_idmap *idmap; 226692d27016SAl Viro const char *link; 2267d6bb3e90SLinus Torvalds u64 hash_len; 2268fe479a58SAl Viro int type; 22691da177e4SLinus Torvalds 22704609e1f1SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 22714609e1f1SChristian Brauner err = may_lookup(idmap, nd); 22721da177e4SLinus Torvalds if (err) 22733595e234SAl Viro return err; 22741da177e4SLinus Torvalds 22758387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22761da177e4SLinus Torvalds 2277fe479a58SAl Viro type = LAST_NORM; 2278d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2279fe479a58SAl Viro case 2: 2280200e9ef7SLinus Torvalds if (name[1] == '.') { 2281fe479a58SAl Viro type = LAST_DOTDOT; 2282bcba1e7dSAl Viro nd->state |= ND_JUMPED; 228316c2cd71SAl Viro } 2284fe479a58SAl Viro break; 2285fe479a58SAl Viro case 1: 2286fe479a58SAl Viro type = LAST_DOT; 2287fe479a58SAl Viro } 22885a202bcdSAl Viro if (likely(type == LAST_NORM)) { 22895a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2290bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 22915a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2292a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2293da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 22945a202bcdSAl Viro if (err < 0) 22953595e234SAl Viro return err; 2296d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2297d6bb3e90SLinus Torvalds name = this.name; 22985a202bcdSAl Viro } 22995a202bcdSAl Viro } 2300fe479a58SAl Viro 2301d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2302d6bb3e90SLinus Torvalds nd->last.name = name; 23035f4a6a69SAl Viro nd->last_type = type; 23045f4a6a69SAl Viro 2305d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2306d6bb3e90SLinus Torvalds if (!*name) 2307bdf6cbf1SAl Viro goto OK; 2308200e9ef7SLinus Torvalds /* 2309200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2310200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2311200e9ef7SLinus Torvalds */ 2312200e9ef7SLinus Torvalds do { 2313d6bb3e90SLinus Torvalds name++; 2314d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 23158620c238SAl Viro if (unlikely(!*name)) { 23168620c238SAl Viro OK: 2317d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2318c108837eSAl Viro if (!depth) { 2319e67fe633SChristian Brauner nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode); 23200f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2321c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 23228620c238SAl Viro return 0; 2323c108837eSAl Viro } 23248620c238SAl Viro /* last component of nested symlink */ 2325d8d4611aSAl Viro name = nd->stack[--depth].name; 23268c4efe22SAl Viro link = walk_component(nd, 0); 23271c4ff1a8SAl Viro } else { 23281c4ff1a8SAl Viro /* not the last component */ 23298c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 23308620c238SAl Viro } 233192d27016SAl Viro if (unlikely(link)) { 233292d27016SAl Viro if (IS_ERR(link)) 233392d27016SAl Viro return PTR_ERR(link); 233492d27016SAl Viro /* a symlink to follow */ 2335d8d4611aSAl Viro nd->stack[depth++].name = name; 233692d27016SAl Viro name = link; 23379e18f10aSAl Viro continue; 233848c8b0c5SAl Viro } 233997242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 234097242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2341e36cffedSJens Axboe if (!try_to_unlazy(nd)) 234297242f99SAl Viro return -ECHILD; 234397242f99SAl Viro } 23443595e234SAl Viro return -ENOTDIR; 23455f4a6a69SAl Viro } 2346ce57dfc1SAl Viro } 234797242f99SAl Viro } 23481da177e4SLinus Torvalds 2349edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2350c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 235131e6b01fSNick Piggin { 2352740a1678SAleksa Sarai int error; 2353c8a53ee5SAl Viro const char *s = nd->name->name; 235431e6b01fSNick Piggin 23556c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23566c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23576c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23586c6ec2b0SJens Axboe 2359c0eb027eSLinus Torvalds if (!*s) 2360c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2361edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2362edc2b1daSAl Viro rcu_read_lock(); 236303fa86e9SAl Viro else 236403fa86e9SAl Viro nd->seq = nd->next_seq = 0; 2365c0eb027eSLinus Torvalds 2366bcba1e7dSAl Viro nd->flags = flags; 2367bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2368ab87f9a5SAleksa Sarai 2369ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2370ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2371ab87f9a5SAleksa Sarai smp_rmb(); 2372ab87f9a5SAleksa Sarai 2373bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2374b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2375b18825a7SDavid Howells struct inode *inode = root->d_inode; 237693893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2377368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23785b6ca027SAl Viro nd->path = nd->root; 23795b6ca027SAl Viro nd->inode = inode; 23805b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2381ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23828f47a016SAl Viro nd->root_seq = nd->seq; 23835b6ca027SAl Viro } else { 23845b6ca027SAl Viro path_get(&nd->path); 23855b6ca027SAl Viro } 2386368ee9baSAl Viro return s; 23875b6ca027SAl Viro } 23885b6ca027SAl Viro 238931e6b01fSNick Piggin nd->root.mnt = NULL; 239031e6b01fSNick Piggin 23918db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 23928db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2393740a1678SAleksa Sarai error = nd_jump_root(nd); 2394740a1678SAleksa Sarai if (unlikely(error)) 2395740a1678SAleksa Sarai return ERR_PTR(error); 2396ef55d917SAl Viro return s; 23978db52c7eSAleksa Sarai } 23988db52c7eSAleksa Sarai 23998db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 24008db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2401e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 240231e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2403c28cc364SNick Piggin unsigned seq; 240431e6b01fSNick Piggin 2405c28cc364SNick Piggin do { 2406c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 240731e6b01fSNick Piggin nd->path = fs->pwd; 2408ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2409c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2410c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2411e41f7d4eSAl Viro } else { 2412e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2413ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2414e41f7d4eSAl Viro } 241531e6b01fSNick Piggin } else { 2416582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2417c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 241831e6b01fSNick Piggin struct dentry *dentry; 241931e6b01fSNick Piggin 24202903ff01SAl Viro if (!f.file) 2421368ee9baSAl Viro return ERR_PTR(-EBADF); 242231e6b01fSNick Piggin 24232903ff01SAl Viro dentry = f.file->f_path.dentry; 242431e6b01fSNick Piggin 2425edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 24262903ff01SAl Viro fdput(f); 2427368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2428f52e0c11SAl Viro } 24292903ff01SAl Viro 24302903ff01SAl Viro nd->path = f.file->f_path; 2431e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 243234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 243334a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24345590ff0dSUlrich Drepper } else { 24352903ff01SAl Viro path_get(&nd->path); 243634a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24371da177e4SLinus Torvalds } 243834a26b99SAl Viro fdput(f); 2439e41f7d4eSAl Viro } 24408db52c7eSAleksa Sarai 2441adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2442adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2443adb21d2bSAleksa Sarai nd->root = nd->path; 2444adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2445adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2446adb21d2bSAleksa Sarai } else { 2447adb21d2bSAleksa Sarai path_get(&nd->root); 2448bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2449adb21d2bSAleksa Sarai } 2450adb21d2bSAleksa Sarai } 2451adb21d2bSAleksa Sarai return s; 24529b4a9b14SAl Viro } 24539b4a9b14SAl Viro 24541ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 245595fa25d9SAl Viro { 2456bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2457bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2458bd92d7feSAl Viro 2459c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2460bd92d7feSAl Viro } 2461bd92d7feSAl Viro 24624f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24634f757f3cSAl Viro { 2464c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2465db3c9adeSAl Viro dget(nd->path.dentry); 246603fa86e9SAl Viro nd->next_seq = nd->seq; 2467a4f5b521SAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry)); 24684f757f3cSAl Viro } 24694f757f3cSAl Viro 24709b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2471c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24729b4a9b14SAl Viro { 2473c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2474bd92d7feSAl Viro int err; 247531e6b01fSNick Piggin 24769b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24774f757f3cSAl Viro err = handle_lookup_down(nd); 24785f336e72SAl Viro if (unlikely(err < 0)) 24795f336e72SAl Viro s = ERR_PTR(err); 24804f757f3cSAl Viro } 24814f757f3cSAl Viro 24821ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24831ccac622SAl Viro (s = lookup_last(nd)) != NULL) 24841ccac622SAl Viro ; 24854f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 24864f0ed93fSAl Viro err = handle_lookup_down(nd); 2487bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 24884f0ed93fSAl Viro } 24899f1fafeeSAl Viro if (!err) 24909f1fafeeSAl Viro err = complete_walk(nd); 2491bd92d7feSAl Viro 2492deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2493deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2494bd23a539SAl Viro err = -ENOTDIR; 2495625b6d10SAl Viro if (!err) { 2496625b6d10SAl Viro *path = nd->path; 2497625b6d10SAl Viro nd->path.mnt = NULL; 2498625b6d10SAl Viro nd->path.dentry = NULL; 2499625b6d10SAl Viro } 2500deb106c6SAl Viro terminate_walk(nd); 2501bd92d7feSAl Viro return err; 250231e6b01fSNick Piggin } 250331e6b01fSNick Piggin 2504794ebceaSStephen Brennan int filename_lookup(int dfd, struct filename *name, unsigned flags, 25059ad1aaa6SAl Viro struct path *path, struct path *root) 2506873f1eedSJeff Layton { 2507894bc8c4SAl Viro int retval; 25089883d185SAl Viro struct nameidata nd; 2509abc9f5beSAl Viro if (IS_ERR(name)) 2510abc9f5beSAl Viro return PTR_ERR(name); 251106422964SAl Viro set_nameidata(&nd, dfd, name, root); 2512c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2513873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2514c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2515873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2516c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2517873f1eedSJeff Layton 2518873f1eedSJeff Layton if (likely(!retval)) 2519161aff1dSAl Viro audit_inode(name, path->dentry, 2520161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 25219883d185SAl Viro restore_nameidata(); 2522020250f3SDmitry Kadashev return retval; 2523020250f3SDmitry Kadashev } 2524020250f3SDmitry Kadashev 25258bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2526c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2527391172c4SAl Viro struct path *parent) 25288bcb77faSAl Viro { 2529c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25309b5858e9SAl Viro int err = link_path_walk(s, nd); 25318bcb77faSAl Viro if (!err) 25328bcb77faSAl Viro err = complete_walk(nd); 2533391172c4SAl Viro if (!err) { 2534391172c4SAl Viro *parent = nd->path; 2535391172c4SAl Viro nd->path.mnt = NULL; 2536391172c4SAl Viro nd->path.dentry = NULL; 2537391172c4SAl Viro } 2538deb106c6SAl Viro terminate_walk(nd); 25398bcb77faSAl Viro return err; 25408bcb77faSAl Viro } 25418bcb77faSAl Viro 25420766ec82SStephen Brennan /* Note: this does not consume "name" */ 254374d7970fSNamjae Jeon static int __filename_parentat(int dfd, struct filename *name, 2544391172c4SAl Viro unsigned int flags, struct path *parent, 254574d7970fSNamjae Jeon struct qstr *last, int *type, 254674d7970fSNamjae Jeon const struct path *root) 25478bcb77faSAl Viro { 25488bcb77faSAl Viro int retval; 25499883d185SAl Viro struct nameidata nd; 25508bcb77faSAl Viro 25515c31b6ceSAl Viro if (IS_ERR(name)) 25520ee50b47SDmitry Kadashev return PTR_ERR(name); 255374d7970fSNamjae Jeon set_nameidata(&nd, dfd, name, root); 2554c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25558bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2556c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25578bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2558c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2559391172c4SAl Viro if (likely(!retval)) { 2560391172c4SAl Viro *last = nd.last; 2561391172c4SAl Viro *type = nd.last_type; 2562c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2563391172c4SAl Viro } 25649883d185SAl Viro restore_nameidata(); 25650ee50b47SDmitry Kadashev return retval; 25660ee50b47SDmitry Kadashev } 25670ee50b47SDmitry Kadashev 256874d7970fSNamjae Jeon static int filename_parentat(int dfd, struct filename *name, 256974d7970fSNamjae Jeon unsigned int flags, struct path *parent, 257074d7970fSNamjae Jeon struct qstr *last, int *type) 257174d7970fSNamjae Jeon { 257274d7970fSNamjae Jeon return __filename_parentat(dfd, name, flags, parent, last, type, NULL); 257374d7970fSNamjae Jeon } 257474d7970fSNamjae Jeon 257579714f72SAl Viro /* does lookup, returns the object with parent locked */ 25760766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25775590ff0dSUlrich Drepper { 25785c31b6ceSAl Viro struct dentry *d; 2579391172c4SAl Viro struct qstr last; 25800ee50b47SDmitry Kadashev int type, error; 258151689104SPaul Moore 2582c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25830ee50b47SDmitry Kadashev if (error) 25840ee50b47SDmitry Kadashev return ERR_PTR(error); 25855c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2586391172c4SAl Viro path_put(path); 25875c31b6ceSAl Viro return ERR_PTR(-EINVAL); 258879714f72SAl Viro } 25895955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 259074d7970fSNamjae Jeon d = lookup_one_qstr_excl(&last, path->dentry, 0); 259179714f72SAl Viro if (IS_ERR(d)) { 25925955102cSAl Viro inode_unlock(path->dentry->d_inode); 2593391172c4SAl Viro path_put(path); 259479714f72SAl Viro } 259579714f72SAl Viro return d; 25965590ff0dSUlrich Drepper } 25975590ff0dSUlrich Drepper 25980766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 25990766ec82SStephen Brennan { 26000766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 26010766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 26020766ec82SStephen Brennan 26030766ec82SStephen Brennan putname(filename); 26040766ec82SStephen Brennan return res; 26050766ec82SStephen Brennan } 26060766ec82SStephen Brennan 2607d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2608d1811465SAl Viro { 2609794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2610794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2611794ebceaSStephen Brennan 2612794ebceaSStephen Brennan putname(filename); 2613794ebceaSStephen Brennan return ret; 2614794ebceaSStephen Brennan 2615d1811465SAl Viro } 26164d359507SAl Viro EXPORT_SYMBOL(kern_path); 2617d1811465SAl Viro 261816f18200SJosef 'Jeff' Sipek /** 261974d7970fSNamjae Jeon * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair 262074d7970fSNamjae Jeon * @filename: filename structure 262174d7970fSNamjae Jeon * @flags: lookup flags 262274d7970fSNamjae Jeon * @parent: pointer to struct path to fill 262374d7970fSNamjae Jeon * @last: last component 262474d7970fSNamjae Jeon * @type: type of the last component 262574d7970fSNamjae Jeon * @root: pointer to struct path of the base directory 262674d7970fSNamjae Jeon */ 262774d7970fSNamjae Jeon int vfs_path_parent_lookup(struct filename *filename, unsigned int flags, 262874d7970fSNamjae Jeon struct path *parent, struct qstr *last, int *type, 262974d7970fSNamjae Jeon const struct path *root) 263074d7970fSNamjae Jeon { 263174d7970fSNamjae Jeon return __filename_parentat(AT_FDCWD, filename, flags, parent, last, 263274d7970fSNamjae Jeon type, root); 263374d7970fSNamjae Jeon } 263474d7970fSNamjae Jeon EXPORT_SYMBOL(vfs_path_parent_lookup); 263574d7970fSNamjae Jeon 263674d7970fSNamjae Jeon /** 263716f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 263816f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 263916f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 264016f18200SJosef 'Jeff' Sipek * @name: pointer to file name 264116f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2642e0a01249SAl Viro * @path: pointer to struct path to fill 264316f18200SJosef 'Jeff' Sipek */ 264416f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 264516f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2646e0a01249SAl Viro struct path *path) 264716f18200SJosef 'Jeff' Sipek { 2648794ebceaSStephen Brennan struct filename *filename; 26499ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2650794ebceaSStephen Brennan int ret; 2651794ebceaSStephen Brennan 2652794ebceaSStephen Brennan filename = getname_kernel(name); 26539ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2654794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2655794ebceaSStephen Brennan putname(filename); 2656794ebceaSStephen Brennan return ret; 265716f18200SJosef 'Jeff' Sipek } 26584d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 265916f18200SJosef 'Jeff' Sipek 26604609e1f1SChristian Brauner static int lookup_one_common(struct mnt_idmap *idmap, 2661c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2662c2fd68b6SChristian Brauner struct qstr *this) 26633c95f0dcSAl Viro { 26643c95f0dcSAl Viro this->name = name; 26653c95f0dcSAl Viro this->len = len; 26663c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26673c95f0dcSAl Viro if (!len) 26683c95f0dcSAl Viro return -EACCES; 26693c95f0dcSAl Viro 26703c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26713c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26723c95f0dcSAl Viro return -EACCES; 26733c95f0dcSAl Viro } 26743c95f0dcSAl Viro 26753c95f0dcSAl Viro while (len--) { 26763c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26773c95f0dcSAl Viro if (c == '/' || c == '\0') 26783c95f0dcSAl Viro return -EACCES; 26793c95f0dcSAl Viro } 26803c95f0dcSAl Viro /* 26813c95f0dcSAl Viro * See if the low-level filesystem might want 26823c95f0dcSAl Viro * to use its own hash.. 26833c95f0dcSAl Viro */ 26843c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26853c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26863c95f0dcSAl Viro if (err < 0) 26873c95f0dcSAl Viro return err; 26883c95f0dcSAl Viro } 26893c95f0dcSAl Viro 26904609e1f1SChristian Brauner return inode_permission(idmap, base->d_inode, MAY_EXEC); 26913c95f0dcSAl Viro } 26923c95f0dcSAl Viro 2693eead1911SChristoph Hellwig /** 26940da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26950da0b7fdSDavid Howells * @name: pathname component to lookup 26960da0b7fdSDavid Howells * @base: base directory to lookup from 26970da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26980da0b7fdSDavid Howells * 26990da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 27000da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 27010da0b7fdSDavid Howells * 27020da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 27030da0b7fdSDavid Howells * not be called by generic code. 27040da0b7fdSDavid Howells * 27050da0b7fdSDavid Howells * The caller must hold base->i_mutex. 27060da0b7fdSDavid Howells */ 27070da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 27080da0b7fdSDavid Howells { 27090da0b7fdSDavid Howells struct qstr this; 27100da0b7fdSDavid Howells int err; 27110da0b7fdSDavid Howells 27120da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27130da0b7fdSDavid Howells 27144609e1f1SChristian Brauner err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this); 27150da0b7fdSDavid Howells if (err) 27160da0b7fdSDavid Howells return ERR_PTR(err); 27170da0b7fdSDavid Howells 27180da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 27190da0b7fdSDavid Howells } 27200da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 27210da0b7fdSDavid Howells 27220da0b7fdSDavid Howells /** 2723a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2724eead1911SChristoph Hellwig * @name: pathname component to lookup 2725eead1911SChristoph Hellwig * @base: base directory to lookup from 2726eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2727eead1911SChristoph Hellwig * 2728a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 27299e7543e9SAl Viro * not be called by generic code. 2730bbddca8eSNeilBrown * 2731bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2732eead1911SChristoph Hellwig */ 2733057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2734057f6c01SJames Morris { 27358613a209SAl Viro struct dentry *dentry; 2736057f6c01SJames Morris struct qstr this; 2737cda309deSMiklos Szeredi int err; 2738057f6c01SJames Morris 27395955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27402f9092e1SDavid Woodhouse 27414609e1f1SChristian Brauner err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this); 2742cda309deSMiklos Szeredi if (err) 2743cda309deSMiklos Szeredi return ERR_PTR(err); 2744cda309deSMiklos Szeredi 27458613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27468613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2747057f6c01SJames Morris } 27484d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2749057f6c01SJames Morris 2750bbddca8eSNeilBrown /** 2751c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 27524609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 2753c2fd68b6SChristian Brauner * @name: pathname component to lookup 2754c2fd68b6SChristian Brauner * @base: base directory to lookup from 2755c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2756c2fd68b6SChristian Brauner * 2757c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2758c2fd68b6SChristian Brauner * not be called by generic code. 2759c2fd68b6SChristian Brauner * 2760c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2761c2fd68b6SChristian Brauner */ 27624609e1f1SChristian Brauner struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name, 2763c2fd68b6SChristian Brauner struct dentry *base, int len) 2764c2fd68b6SChristian Brauner { 2765c2fd68b6SChristian Brauner struct dentry *dentry; 2766c2fd68b6SChristian Brauner struct qstr this; 2767c2fd68b6SChristian Brauner int err; 2768c2fd68b6SChristian Brauner 2769c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2770c2fd68b6SChristian Brauner 27714609e1f1SChristian Brauner err = lookup_one_common(idmap, name, base, len, &this); 2772c2fd68b6SChristian Brauner if (err) 2773c2fd68b6SChristian Brauner return ERR_PTR(err); 2774c2fd68b6SChristian Brauner 2775c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2776c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2777c2fd68b6SChristian Brauner } 2778c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2779c2fd68b6SChristian Brauner 2780c2fd68b6SChristian Brauner /** 278100675017SChristian Brauner * lookup_one_unlocked - filesystem helper to lookup single pathname component 27824609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 278300675017SChristian Brauner * @name: pathname component to lookup 278400675017SChristian Brauner * @base: base directory to lookup from 278500675017SChristian Brauner * @len: maximum length @len should be interpreted to 278600675017SChristian Brauner * 278700675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 278800675017SChristian Brauner * not be called by generic code. 278900675017SChristian Brauner * 279000675017SChristian Brauner * Unlike lookup_one_len, it should be called without the parent 279100675017SChristian Brauner * i_mutex held, and will take the i_mutex itself if necessary. 279200675017SChristian Brauner */ 27934609e1f1SChristian Brauner struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap, 279400675017SChristian Brauner const char *name, struct dentry *base, 279500675017SChristian Brauner int len) 279600675017SChristian Brauner { 279700675017SChristian Brauner struct qstr this; 279800675017SChristian Brauner int err; 279900675017SChristian Brauner struct dentry *ret; 280000675017SChristian Brauner 28014609e1f1SChristian Brauner err = lookup_one_common(idmap, name, base, len, &this); 280200675017SChristian Brauner if (err) 280300675017SChristian Brauner return ERR_PTR(err); 280400675017SChristian Brauner 280500675017SChristian Brauner ret = lookup_dcache(&this, base, 0); 280600675017SChristian Brauner if (!ret) 280700675017SChristian Brauner ret = lookup_slow(&this, base, 0); 280800675017SChristian Brauner return ret; 280900675017SChristian Brauner } 281000675017SChristian Brauner EXPORT_SYMBOL(lookup_one_unlocked); 281100675017SChristian Brauner 281200675017SChristian Brauner /** 281300675017SChristian Brauner * lookup_one_positive_unlocked - filesystem helper to lookup single 281400675017SChristian Brauner * pathname component 28154609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 281600675017SChristian Brauner * @name: pathname component to lookup 281700675017SChristian Brauner * @base: base directory to lookup from 281800675017SChristian Brauner * @len: maximum length @len should be interpreted to 281900675017SChristian Brauner * 282000675017SChristian Brauner * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns 282100675017SChristian Brauner * known positive or ERR_PTR(). This is what most of the users want. 282200675017SChristian Brauner * 282300675017SChristian Brauner * Note that pinned negative with unlocked parent _can_ become positive at any 282400675017SChristian Brauner * time, so callers of lookup_one_unlocked() need to be very careful; pinned 282500675017SChristian Brauner * positives have >d_inode stable, so this one avoids such problems. 282600675017SChristian Brauner * 282700675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 282800675017SChristian Brauner * not be called by generic code. 282900675017SChristian Brauner * 283000675017SChristian Brauner * The helper should be called without i_mutex held. 283100675017SChristian Brauner */ 28324609e1f1SChristian Brauner struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap, 283300675017SChristian Brauner const char *name, 283400675017SChristian Brauner struct dentry *base, int len) 283500675017SChristian Brauner { 28364609e1f1SChristian Brauner struct dentry *ret = lookup_one_unlocked(idmap, name, base, len); 283700675017SChristian Brauner 283800675017SChristian Brauner if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 283900675017SChristian Brauner dput(ret); 284000675017SChristian Brauner ret = ERR_PTR(-ENOENT); 284100675017SChristian Brauner } 284200675017SChristian Brauner return ret; 284300675017SChristian Brauner } 284400675017SChristian Brauner EXPORT_SYMBOL(lookup_one_positive_unlocked); 284500675017SChristian Brauner 284600675017SChristian Brauner /** 2847bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2848bbddca8eSNeilBrown * @name: pathname component to lookup 2849bbddca8eSNeilBrown * @base: base directory to lookup from 2850bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2851bbddca8eSNeilBrown * 2852bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2853bbddca8eSNeilBrown * not be called by generic code. 2854bbddca8eSNeilBrown * 2855bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2856bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2857bbddca8eSNeilBrown */ 2858bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2859bbddca8eSNeilBrown struct dentry *base, int len) 2860bbddca8eSNeilBrown { 28614609e1f1SChristian Brauner return lookup_one_unlocked(&nop_mnt_idmap, name, base, len); 2862bbddca8eSNeilBrown } 2863bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2864bbddca8eSNeilBrown 28656c2d4798SAl Viro /* 28666c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 28676c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 28686c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 28696c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 28706c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 28716c2d4798SAl Viro * this one avoids such problems. 28726c2d4798SAl Viro */ 28736c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28746c2d4798SAl Viro struct dentry *base, int len) 28756c2d4798SAl Viro { 28764609e1f1SChristian Brauner return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len); 28776c2d4798SAl Viro } 28786c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28796c2d4798SAl Viro 2880eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2881eedf265aSEric W. Biederman int path_pts(struct path *path) 2882eedf265aSEric W. Biederman { 2883eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2884eedf265aSEric W. Biederman * the input path. 2885eedf265aSEric W. Biederman */ 2886a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2887a6a7eb76SAl Viro struct dentry *child; 288819f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2889eedf265aSEric W. Biederman 2890a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2891a6a7eb76SAl Viro dput(parent); 289263b27720SAl Viro return -ENOENT; 2893a6a7eb76SAl Viro } 289463b27720SAl Viro dput(path->dentry); 289563b27720SAl Viro path->dentry = parent; 2896eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 28970d5a4f8fSWang Ming if (IS_ERR_OR_NULL(child)) 2898eedf265aSEric W. Biederman return -ENOENT; 2899eedf265aSEric W. Biederman 2900eedf265aSEric W. Biederman path->dentry = child; 2901eedf265aSEric W. Biederman dput(parent); 2902e1f19857SRichard Weinberger follow_down(path, 0); 2903eedf265aSEric W. Biederman return 0; 2904eedf265aSEric W. Biederman } 2905eedf265aSEric W. Biederman #endif 2906eedf265aSEric W. Biederman 29071fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 29081fa1e7f6SAndy Whitcroft struct path *path, int *empty) 29091da177e4SLinus Torvalds { 2910794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2911794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2912794ebceaSStephen Brennan 2913794ebceaSStephen Brennan putname(filename); 2914794ebceaSStephen Brennan return ret; 29151da177e4SLinus Torvalds } 2916b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 29171fa1e7f6SAndy Whitcroft 29189452e93eSChristian Brauner int __check_sticky(struct mnt_idmap *idmap, struct inode *dir, 2919ba73d987SChristian Brauner struct inode *inode) 29201da177e4SLinus Torvalds { 29218e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2922da9592edSDavid Howells 2923e67fe633SChristian Brauner if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid)) 29241da177e4SLinus Torvalds return 0; 2925e67fe633SChristian Brauner if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid)) 29261da177e4SLinus Torvalds return 0; 29279452e93eSChristian Brauner return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER); 29281da177e4SLinus Torvalds } 2929cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 29301da177e4SLinus Torvalds 29311da177e4SLinus Torvalds /* 29321da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 29331da177e4SLinus Torvalds * whether the type of victim is right. 29341da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 29351da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 29361da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 29371da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 29381da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 29391da177e4SLinus Torvalds * a. be owner of dir, or 29401da177e4SLinus Torvalds * b. be owner of victim, or 29411da177e4SLinus Torvalds * c. have CAP_FOWNER capability 29421da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 29431da177e4SLinus Torvalds * links pointing to it. 29440bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 29450bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 29460bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 29470bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 29480bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 29491da177e4SLinus Torvalds * nfs_async_unlink(). 29501da177e4SLinus Torvalds */ 29514609e1f1SChristian Brauner static int may_delete(struct mnt_idmap *idmap, struct inode *dir, 2952ba73d987SChristian Brauner struct dentry *victim, bool isdir) 29531da177e4SLinus Torvalds { 295463afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 29551da177e4SLinus Torvalds int error; 29561da177e4SLinus Torvalds 2957b18825a7SDavid Howells if (d_is_negative(victim)) 29581da177e4SLinus Torvalds return -ENOENT; 2959b18825a7SDavid Howells BUG_ON(!inode); 29601da177e4SLinus Torvalds 29611da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2962593d1ce8SEric W. Biederman 2963593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2964e67fe633SChristian Brauner if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) || 2965e67fe633SChristian Brauner !vfsgid_valid(i_gid_into_vfsgid(idmap, inode))) 2966593d1ce8SEric W. Biederman return -EOVERFLOW; 2967593d1ce8SEric W. Biederman 29684fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29691da177e4SLinus Torvalds 29704609e1f1SChristian Brauner error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 29711da177e4SLinus Torvalds if (error) 29721da177e4SLinus Torvalds return error; 29731da177e4SLinus Torvalds if (IS_APPEND(dir)) 29741da177e4SLinus Torvalds return -EPERM; 2975b18825a7SDavid Howells 29769452e93eSChristian Brauner if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) || 2977ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 29784609e1f1SChristian Brauner HAS_UNMAPPED_ID(idmap, inode)) 29791da177e4SLinus Torvalds return -EPERM; 29801da177e4SLinus Torvalds if (isdir) { 298144b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29821da177e4SLinus Torvalds return -ENOTDIR; 29831da177e4SLinus Torvalds if (IS_ROOT(victim)) 29841da177e4SLinus Torvalds return -EBUSY; 298544b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29861da177e4SLinus Torvalds return -EISDIR; 29871da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29881da177e4SLinus Torvalds return -ENOENT; 29891da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29901da177e4SLinus Torvalds return -EBUSY; 29911da177e4SLinus Torvalds return 0; 29921da177e4SLinus Torvalds } 29931da177e4SLinus Torvalds 29941da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29951da177e4SLinus Torvalds * dir. 29961da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29971da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29981da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2999036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 3000036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 3001036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 30021da177e4SLinus Torvalds */ 30034609e1f1SChristian Brauner static inline int may_create(struct mnt_idmap *idmap, 3004ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 30051da177e4SLinus Torvalds { 300614e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 30071da177e4SLinus Torvalds if (child->d_inode) 30081da177e4SLinus Torvalds return -EEXIST; 30091da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 30101da177e4SLinus Torvalds return -ENOENT; 30114609e1f1SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, idmap)) 3012036d5236SEric W. Biederman return -EOVERFLOW; 30138e538913SChristian Brauner 30144609e1f1SChristian Brauner return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 30151da177e4SLinus Torvalds } 30161da177e4SLinus Torvalds 30179bc37e04SAl Viro static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2) 30181da177e4SLinus Torvalds { 30191da177e4SLinus Torvalds struct dentry *p; 30201da177e4SLinus Torvalds 3021e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 3022e2761a11SOGAWA Hirofumi if (p) { 30235955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30245955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 30251da177e4SLinus Torvalds return p; 30261da177e4SLinus Torvalds } 30271da177e4SLinus Torvalds 3028e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 3029e2761a11SOGAWA Hirofumi if (p) { 30305955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30315955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 30321da177e4SLinus Torvalds return p; 30331da177e4SLinus Torvalds } 30341da177e4SLinus Torvalds 3035f23ce757SJan Kara lock_two_inodes(p1->d_inode, p2->d_inode, 3036f23ce757SJan Kara I_MUTEX_PARENT, I_MUTEX_PARENT2); 30371da177e4SLinus Torvalds return NULL; 30381da177e4SLinus Torvalds } 30399bc37e04SAl Viro 30409bc37e04SAl Viro /* 30419bc37e04SAl Viro * p1 and p2 should be directories on the same fs. 30429bc37e04SAl Viro */ 30439bc37e04SAl Viro struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 30449bc37e04SAl Viro { 30459bc37e04SAl Viro if (p1 == p2) { 30469bc37e04SAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30479bc37e04SAl Viro return NULL; 30489bc37e04SAl Viro } 30499bc37e04SAl Viro 30509bc37e04SAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 30519bc37e04SAl Viro return lock_two_directories(p1, p2); 30529bc37e04SAl Viro } 30534d359507SAl Viro EXPORT_SYMBOL(lock_rename); 30541da177e4SLinus Torvalds 30559bc37e04SAl Viro /* 30569bc37e04SAl Viro * c1 and p2 should be on the same fs. 30579bc37e04SAl Viro */ 30589bc37e04SAl Viro struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2) 30599bc37e04SAl Viro { 30609bc37e04SAl Viro if (READ_ONCE(c1->d_parent) == p2) { 30619bc37e04SAl Viro /* 30629bc37e04SAl Viro * hopefully won't need to touch ->s_vfs_rename_mutex at all. 30639bc37e04SAl Viro */ 30649bc37e04SAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30659bc37e04SAl Viro /* 30669bc37e04SAl Viro * now that p2 is locked, nobody can move in or out of it, 30679bc37e04SAl Viro * so the test below is safe. 30689bc37e04SAl Viro */ 30699bc37e04SAl Viro if (likely(c1->d_parent == p2)) 30709bc37e04SAl Viro return NULL; 30719bc37e04SAl Viro 30729bc37e04SAl Viro /* 30739bc37e04SAl Viro * c1 got moved out of p2 while we'd been taking locks; 30749bc37e04SAl Viro * unlock and fall back to slow case. 30759bc37e04SAl Viro */ 30769bc37e04SAl Viro inode_unlock(p2->d_inode); 30779bc37e04SAl Viro } 30789bc37e04SAl Viro 30799bc37e04SAl Viro mutex_lock(&c1->d_sb->s_vfs_rename_mutex); 30809bc37e04SAl Viro /* 30819bc37e04SAl Viro * nobody can move out of any directories on this fs. 30829bc37e04SAl Viro */ 30839bc37e04SAl Viro if (likely(c1->d_parent != p2)) 30849bc37e04SAl Viro return lock_two_directories(c1->d_parent, p2); 30859bc37e04SAl Viro 30869bc37e04SAl Viro /* 30879bc37e04SAl Viro * c1 got moved into p2 while we were taking locks; 30889bc37e04SAl Viro * we need p2 locked and ->s_vfs_rename_mutex unlocked, 30899bc37e04SAl Viro * for consistency with lock_rename(). 30909bc37e04SAl Viro */ 30919bc37e04SAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30929bc37e04SAl Viro mutex_unlock(&c1->d_sb->s_vfs_rename_mutex); 30939bc37e04SAl Viro return NULL; 30949bc37e04SAl Viro } 30959bc37e04SAl Viro EXPORT_SYMBOL(lock_rename_child); 30969bc37e04SAl Viro 30971da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 30981da177e4SLinus Torvalds { 30995955102cSAl Viro inode_unlock(p1->d_inode); 31001da177e4SLinus Torvalds if (p1 != p2) { 31015955102cSAl Viro inode_unlock(p2->d_inode); 3102fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 31031da177e4SLinus Torvalds } 31041da177e4SLinus Torvalds } 31054d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 31061da177e4SLinus Torvalds 31076521f891SChristian Brauner /** 31081639a49cSYang Xu * mode_strip_umask - handle vfs umask stripping 31091639a49cSYang Xu * @dir: parent directory of the new inode 31101639a49cSYang Xu * @mode: mode of the new inode to be created in @dir 31111639a49cSYang Xu * 31121639a49cSYang Xu * Umask stripping depends on whether or not the filesystem supports POSIX 31131639a49cSYang Xu * ACLs. If the filesystem doesn't support it umask stripping is done directly 31141639a49cSYang Xu * in here. If the filesystem does support POSIX ACLs umask stripping is 31151639a49cSYang Xu * deferred until the filesystem calls posix_acl_create(). 31161639a49cSYang Xu * 31171639a49cSYang Xu * Returns: mode 31181639a49cSYang Xu */ 31191639a49cSYang Xu static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode) 31201639a49cSYang Xu { 31211639a49cSYang Xu if (!IS_POSIXACL(dir)) 31221639a49cSYang Xu mode &= ~current_umask(); 31231639a49cSYang Xu return mode; 31241639a49cSYang Xu } 31251639a49cSYang Xu 31261639a49cSYang Xu /** 31271639a49cSYang Xu * vfs_prepare_mode - prepare the mode to be used for a new inode 31289452e93eSChristian Brauner * @idmap: idmap of the mount the inode was found from 31291639a49cSYang Xu * @dir: parent directory of the new inode 31301639a49cSYang Xu * @mode: mode of the new inode 31311639a49cSYang Xu * @mask_perms: allowed permission by the vfs 31321639a49cSYang Xu * @type: type of file to be created 31331639a49cSYang Xu * 31341639a49cSYang Xu * This helper consolidates and enforces vfs restrictions on the @mode of a new 31351639a49cSYang Xu * object to be created. 31361639a49cSYang Xu * 31371639a49cSYang Xu * Umask stripping depends on whether the filesystem supports POSIX ACLs (see 31381639a49cSYang Xu * the kernel documentation for mode_strip_umask()). Moving umask stripping 31391639a49cSYang Xu * after setgid stripping allows the same ordering for both non-POSIX ACL and 31401639a49cSYang Xu * POSIX ACL supporting filesystems. 31411639a49cSYang Xu * 31421639a49cSYang Xu * Note that it's currently valid for @type to be 0 if a directory is created. 31431639a49cSYang Xu * Filesystems raise that flag individually and we need to check whether each 31441639a49cSYang Xu * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a 31451639a49cSYang Xu * non-zero type. 31461639a49cSYang Xu * 31471639a49cSYang Xu * Returns: mode to be passed to the filesystem 31481639a49cSYang Xu */ 31499452e93eSChristian Brauner static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap, 31501639a49cSYang Xu const struct inode *dir, umode_t mode, 31511639a49cSYang Xu umode_t mask_perms, umode_t type) 31521639a49cSYang Xu { 31539452e93eSChristian Brauner mode = mode_strip_sgid(idmap, dir, mode); 31541639a49cSYang Xu mode = mode_strip_umask(dir, mode); 31551639a49cSYang Xu 31561639a49cSYang Xu /* 31571639a49cSYang Xu * Apply the vfs mandated allowed permission mask and set the type of 31581639a49cSYang Xu * file to be created before we call into the filesystem. 31591639a49cSYang Xu */ 31601639a49cSYang Xu mode &= (mask_perms & ~S_IFMT); 31611639a49cSYang Xu mode |= (type & S_IFMT); 31621639a49cSYang Xu 31631639a49cSYang Xu return mode; 31641639a49cSYang Xu } 31651639a49cSYang Xu 31661639a49cSYang Xu /** 31676521f891SChristian Brauner * vfs_create - create new file 3168abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 31696521f891SChristian Brauner * @dir: inode of @dentry 31706521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 31716521f891SChristian Brauner * @mode: mode of the new file 31726521f891SChristian Brauner * @want_excl: whether the file must not yet exist 31736521f891SChristian Brauner * 31746521f891SChristian Brauner * Create a new file. 31756521f891SChristian Brauner * 3176abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3177abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3178abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 31796521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3180abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 31816521f891SChristian Brauner */ 3182abf08576SChristian Brauner int vfs_create(struct mnt_idmap *idmap, struct inode *dir, 31836521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 31841da177e4SLinus Torvalds { 3185abf08576SChristian Brauner int error; 3186abf08576SChristian Brauner 31874609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 31881da177e4SLinus Torvalds if (error) 31891da177e4SLinus Torvalds return error; 31901da177e4SLinus Torvalds 3191acfa4380SAl Viro if (!dir->i_op->create) 31921da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 31931639a49cSYang Xu 31949452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG); 31951da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 31961da177e4SLinus Torvalds if (error) 31971da177e4SLinus Torvalds return error; 31986c960e68SChristian Brauner error = dir->i_op->create(idmap, dir, dentry, mode, want_excl); 3199a74574aaSStephen Smalley if (!error) 3200f38aa942SAmy Griffis fsnotify_create(dir, dentry); 32011da177e4SLinus Torvalds return error; 32021da177e4SLinus Torvalds } 32034d359507SAl Viro EXPORT_SYMBOL(vfs_create); 32041da177e4SLinus Torvalds 32058e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 32068e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 32078e6c848eSAl Viro void *arg) 32088e6c848eSAl Viro { 32098e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 32104609e1f1SChristian Brauner int error = may_create(&nop_mnt_idmap, dir, dentry); 32118e6c848eSAl Viro if (error) 32128e6c848eSAl Viro return error; 32138e6c848eSAl Viro 32148e6c848eSAl Viro mode &= S_IALLUGO; 32158e6c848eSAl Viro mode |= S_IFREG; 32168e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 32178e6c848eSAl Viro if (error) 32188e6c848eSAl Viro return error; 32198e6c848eSAl Viro error = f(dentry, mode, arg); 32208e6c848eSAl Viro if (!error) 32218e6c848eSAl Viro fsnotify_create(dir, dentry); 32228e6c848eSAl Viro return error; 32238e6c848eSAl Viro } 32248e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 32258e6c848eSAl Viro 3226a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3227a2982cc9SEric W. Biederman { 3228a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3229a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3230a2982cc9SEric W. Biederman } 3231a2982cc9SEric W. Biederman 32324609e1f1SChristian Brauner static int may_open(struct mnt_idmap *idmap, const struct path *path, 3233ba73d987SChristian Brauner int acc_mode, int flag) 32341da177e4SLinus Torvalds { 32353fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 32361da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 32371da177e4SLinus Torvalds int error; 32381da177e4SLinus Torvalds 32391da177e4SLinus Torvalds if (!inode) 32401da177e4SLinus Torvalds return -ENOENT; 32411da177e4SLinus Torvalds 3242c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3243c8fe8f30SChristoph Hellwig case S_IFLNK: 32441da177e4SLinus Torvalds return -ELOOP; 3245c8fe8f30SChristoph Hellwig case S_IFDIR: 3246fc4177beSKees Cook if (acc_mode & MAY_WRITE) 32471da177e4SLinus Torvalds return -EISDIR; 3248fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3249fc4177beSKees Cook return -EACCES; 3250c8fe8f30SChristoph Hellwig break; 3251c8fe8f30SChristoph Hellwig case S_IFBLK: 3252c8fe8f30SChristoph Hellwig case S_IFCHR: 3253a2982cc9SEric W. Biederman if (!may_open_dev(path)) 32541da177e4SLinus Torvalds return -EACCES; 3255633fb6acSKees Cook fallthrough; 3256c8fe8f30SChristoph Hellwig case S_IFIFO: 3257c8fe8f30SChristoph Hellwig case S_IFSOCK: 3258633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3259633fb6acSKees Cook return -EACCES; 32601da177e4SLinus Torvalds flag &= ~O_TRUNC; 3261c8fe8f30SChristoph Hellwig break; 32620fd338b2SKees Cook case S_IFREG: 32630fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 32640fd338b2SKees Cook return -EACCES; 32650fd338b2SKees Cook break; 32664a3fd211SDave Hansen } 3267b41572e9SDave Hansen 32684609e1f1SChristian Brauner error = inode_permission(idmap, inode, MAY_OPEN | acc_mode); 3269b41572e9SDave Hansen if (error) 3270b41572e9SDave Hansen return error; 32716146f0d5SMimi Zohar 32721da177e4SLinus Torvalds /* 32731da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 32741da177e4SLinus Torvalds */ 32751da177e4SLinus Torvalds if (IS_APPEND(inode)) { 32768737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 32777715b521SAl Viro return -EPERM; 32781da177e4SLinus Torvalds if (flag & O_TRUNC) 32797715b521SAl Viro return -EPERM; 32801da177e4SLinus Torvalds } 32811da177e4SLinus Torvalds 32821da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 328301beba79SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode)) 32847715b521SAl Viro return -EPERM; 32851da177e4SLinus Torvalds 3286f3c7691eSJ. Bruce Fields return 0; 32877715b521SAl Viro } 32887715b521SAl Viro 3289abf08576SChristian Brauner static int handle_truncate(struct mnt_idmap *idmap, struct file *filp) 32907715b521SAl Viro { 3291f0bb5aafSAl Viro const struct path *path = &filp->f_path; 32927715b521SAl Viro struct inode *inode = path->dentry->d_inode; 32937715b521SAl Viro int error = get_write_access(inode); 32941da177e4SLinus Torvalds if (error) 32957715b521SAl Viro return error; 3296482e0007SJeff Layton 32973350607dSGünther Noack error = security_file_truncate(filp); 32981da177e4SLinus Torvalds if (!error) { 3299abf08576SChristian Brauner error = do_truncate(idmap, path->dentry, 0, 3300d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3301e1181ee6SJeff Layton filp); 33021da177e4SLinus Torvalds } 33031da177e4SLinus Torvalds put_write_access(inode); 3304acd0c935SMimi Zohar return error; 33051da177e4SLinus Torvalds } 33061da177e4SLinus Torvalds 3307d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3308d57999e1SDave Hansen { 33098a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 33108a5e929dSAl Viro flag--; 3311d57999e1SDave Hansen return flag; 3312d57999e1SDave Hansen } 3313d57999e1SDave Hansen 33144609e1f1SChristian Brauner static int may_o_create(struct mnt_idmap *idmap, 3315ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3316ba73d987SChristian Brauner umode_t mode) 3317d18e9008SMiklos Szeredi { 3318d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3319d18e9008SMiklos Szeredi if (error) 3320d18e9008SMiklos Szeredi return error; 3321d18e9008SMiklos Szeredi 33224609e1f1SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap)) 33231328c727SSeth Forshee return -EOVERFLOW; 33241328c727SSeth Forshee 33254609e1f1SChristian Brauner error = inode_permission(idmap, dir->dentry->d_inode, 332647291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3327d18e9008SMiklos Szeredi if (error) 3328d18e9008SMiklos Szeredi return error; 3329d18e9008SMiklos Szeredi 3330d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3331d18e9008SMiklos Szeredi } 3332d18e9008SMiklos Szeredi 33331acf0af9SDavid Howells /* 33341acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 33351acf0af9SDavid Howells * dentry. 33361acf0af9SDavid Howells * 33371acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 33381acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 33391acf0af9SDavid Howells * 334000a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 334100a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 334200a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 33431acf0af9SDavid Howells * 33441acf0af9SDavid Howells * Returns an error code otherwise. 33451acf0af9SDavid Howells */ 3346239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3347239eb983SAl Viro struct file *file, 33483ec2eef1SAl Viro int open_flag, umode_t mode) 3349d18e9008SMiklos Szeredi { 3350d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3351d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3352d18e9008SMiklos Szeredi int error; 3353d18e9008SMiklos Szeredi 3354d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3355d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3356d18e9008SMiklos Szeredi 335730d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 335830d90494SAl Viro file->f_path.mnt = nd->path.mnt; 33590fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 336044907d79SAl Viro open_to_namei_flags(open_flag), mode); 33616fbd0714SAl Viro d_lookup_done(dentry); 3362384f26e2SAl Viro if (!error) { 336364e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 33646fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 33656fb968cdSAl Viro dput(dentry); 33666fb968cdSAl Viro dentry = dget(file->f_path.dentry); 33676fb968cdSAl Viro } 336864e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 33692675a4ebSAl Viro error = -EIO; 3370384f26e2SAl Viro } else { 337130d90494SAl Viro if (file->f_path.dentry) { 3372d18e9008SMiklos Szeredi dput(dentry); 3373d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 337410c64ceaSAl Viro } 3375239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3376a01e718fSAl Viro error = -ENOENT; 3377d18e9008SMiklos Szeredi } 3378d18e9008SMiklos Szeredi } 3379239eb983SAl Viro if (error) { 3380d18e9008SMiklos Szeredi dput(dentry); 3381239eb983SAl Viro dentry = ERR_PTR(error); 3382239eb983SAl Viro } 3383239eb983SAl Viro return dentry; 3384d18e9008SMiklos Szeredi } 3385d18e9008SMiklos Szeredi 338631e6b01fSNick Piggin /* 33871acf0af9SDavid Howells * Look up and maybe create and open the last component. 3388d58ffd35SMiklos Szeredi * 338900a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3390d58ffd35SMiklos Szeredi * 339100a07c15SAl Viro * Returns 0 on success, that is, if 339200a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 339300a07c15SAl Viro * the file was not completely opened at this time, though lookups and 339400a07c15SAl Viro * creations were performed. 339500a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 339600a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 339700a07c15SAl Viro * hadn't been specified. 33981acf0af9SDavid Howells * 339900a07c15SAl Viro * An error code is returned on failure. 3400d58ffd35SMiklos Szeredi */ 3401da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3402d58ffd35SMiklos Szeredi const struct open_flags *op, 34033ec2eef1SAl Viro bool got_write) 3404d58ffd35SMiklos Szeredi { 34056c960e68SChristian Brauner struct mnt_idmap *idmap; 3406d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 340754ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 34081643b43fSAl Viro int open_flag = op->open_flag; 3409d58ffd35SMiklos Szeredi struct dentry *dentry; 34101643b43fSAl Viro int error, create_error = 0; 34111643b43fSAl Viro umode_t mode = op->mode; 34126fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3413d58ffd35SMiklos Szeredi 3414ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3415da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3416d58ffd35SMiklos Szeredi 341773a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 34186fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 34196fbd0714SAl Viro for (;;) { 34206fbd0714SAl Viro if (!dentry) { 34216fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3422d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3423da5ebf5aSAl Viro return dentry; 34246fbd0714SAl Viro } 34256fbd0714SAl Viro if (d_in_lookup(dentry)) 34266fbd0714SAl Viro break; 3427d58ffd35SMiklos Szeredi 34286fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 34296fbd0714SAl Viro if (likely(error > 0)) 34306fbd0714SAl Viro break; 34316fbd0714SAl Viro if (error) 34326fbd0714SAl Viro goto out_dput; 34336fbd0714SAl Viro d_invalidate(dentry); 34346fbd0714SAl Viro dput(dentry); 34356fbd0714SAl Viro dentry = NULL; 34366fbd0714SAl Viro } 34376fbd0714SAl Viro if (dentry->d_inode) { 3438d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3439da5ebf5aSAl Viro return dentry; 34406c51e513SAl Viro } 3441d18e9008SMiklos Szeredi 34421643b43fSAl Viro /* 34431643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 34441643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 34451643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 34461643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 34471643b43fSAl Viro * 34481643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 34491643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 34501643b43fSAl Viro */ 345199a4a90cSAl Viro if (unlikely(!got_write)) 345299a4a90cSAl Viro open_flag &= ~O_TRUNC; 34536c960e68SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 34541643b43fSAl Viro if (open_flag & O_CREAT) { 345599a4a90cSAl Viro if (open_flag & O_EXCL) 345699a4a90cSAl Viro open_flag &= ~O_TRUNC; 34579452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode); 345899a4a90cSAl Viro if (likely(got_write)) 34594609e1f1SChristian Brauner create_error = may_o_create(idmap, &nd->path, 3460ba73d987SChristian Brauner dentry, mode); 346199a4a90cSAl Viro else 346299a4a90cSAl Viro create_error = -EROFS; 346399a4a90cSAl Viro } 346499a4a90cSAl Viro if (create_error) 34651643b43fSAl Viro open_flag &= ~O_CREAT; 34661643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3467d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3468da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3469da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3470da5ebf5aSAl Viro return dentry; 3471239eb983SAl Viro } 347254ef4872SMiklos Szeredi 34736fbd0714SAl Viro if (d_in_lookup(dentry)) { 347412fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 347512fa5e24SAl Viro nd->flags); 34766fbd0714SAl Viro d_lookup_done(dentry); 347712fa5e24SAl Viro if (unlikely(res)) { 347812fa5e24SAl Viro if (IS_ERR(res)) { 347912fa5e24SAl Viro error = PTR_ERR(res); 348012fa5e24SAl Viro goto out_dput; 348112fa5e24SAl Viro } 348212fa5e24SAl Viro dput(dentry); 348312fa5e24SAl Viro dentry = res; 348412fa5e24SAl Viro } 348554ef4872SMiklos Szeredi } 348654ef4872SMiklos Szeredi 3487d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 34881643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 348973a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3490ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3491ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3492ce8644fcSAl Viro error = -EACCES; 3493d58ffd35SMiklos Szeredi goto out_dput; 349464894cf8SAl Viro } 3495549c7297SChristian Brauner 34966c960e68SChristian Brauner error = dir_inode->i_op->create(idmap, dir_inode, dentry, 3497549c7297SChristian Brauner mode, open_flag & O_EXCL); 3498d58ffd35SMiklos Szeredi if (error) 3499d58ffd35SMiklos Szeredi goto out_dput; 3500d58ffd35SMiklos Szeredi } 35011643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 35021643b43fSAl Viro error = create_error; 3503d58ffd35SMiklos Szeredi goto out_dput; 3504d58ffd35SMiklos Szeredi } 3505da5ebf5aSAl Viro return dentry; 3506d58ffd35SMiklos Szeredi 3507d58ffd35SMiklos Szeredi out_dput: 3508d58ffd35SMiklos Szeredi dput(dentry); 3509da5ebf5aSAl Viro return ERR_PTR(error); 3510d58ffd35SMiklos Szeredi } 3511d58ffd35SMiklos Szeredi 3512c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 35133ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3514fb1cc555SAl Viro { 3515a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3516ca344a89SAl Viro int open_flag = op->open_flag; 351764894cf8SAl Viro bool got_write = false; 3518da5ebf5aSAl Viro struct dentry *dentry; 3519b0417d2cSAl Viro const char *res; 3520fb1cc555SAl Viro 3521c3e380b0SAl Viro nd->flags |= op->intent; 3522c3e380b0SAl Viro 3523bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 352456676ec3SAl Viro if (nd->depth) 352556676ec3SAl Viro put_link(nd); 3526ff326a32SAl Viro return handle_dots(nd, nd->last_type); 35271f36f774SAl Viro } 3528a2c36b45SAl Viro 3529ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3530fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3531fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3532fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 35334cb64024SAl Viro dentry = lookup_fast(nd); 353420e34357SAl Viro if (IS_ERR(dentry)) 35351ccac622SAl Viro return ERR_CAST(dentry); 353620e34357SAl Viro if (likely(dentry)) 353771574865SMiklos Szeredi goto finish_lookup; 353871574865SMiklos Szeredi 35396583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3540b6183df7SMiklos Szeredi } else { 3541fe2d35ffSAl Viro /* create side of things */ 354272287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3543e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3544e36cffedSJens Axboe return ERR_PTR(-ECHILD); 354572287417SAl Viro } 3546c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 35471f36f774SAl Viro /* trailing slashes? */ 3548deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 35491ccac622SAl Viro return ERR_PTR(-EISDIR); 3550b6183df7SMiklos Szeredi } 35511f36f774SAl Viro 35529cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3553e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 355464894cf8SAl Viro /* 355564894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 355664894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 355764894cf8SAl Viro * dropping this one anyway. 355864894cf8SAl Viro */ 355964894cf8SAl Viro } 35609cf843e3SAl Viro if (open_flag & O_CREAT) 35615955102cSAl Viro inode_lock(dir->d_inode); 35629cf843e3SAl Viro else 35639cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3564da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3565f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3566f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 35679cf843e3SAl Viro if (open_flag & O_CREAT) 35685955102cSAl Viro inode_unlock(dir->d_inode); 35699cf843e3SAl Viro else 35709cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3571fb1cc555SAl Viro 3572c981a482SAl Viro if (got_write) 357359e96e65SAl Viro mnt_drop_write(nd->path.mnt); 35746c0d46c4SAl Viro 357559e96e65SAl Viro if (IS_ERR(dentry)) 357659e96e65SAl Viro return ERR_CAST(dentry); 357759e96e65SAl Viro 3578973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3579e73cabffSAl Viro dput(nd->path.dentry); 3580e73cabffSAl Viro nd->path.dentry = dentry; 3581c981a482SAl Viro return NULL; 3582fb1cc555SAl Viro } 3583fb1cc555SAl Viro 358420e34357SAl Viro finish_lookup: 358556676ec3SAl Viro if (nd->depth) 358656676ec3SAl Viro put_link(nd); 3587a4f5b521SAl Viro res = step_into(nd, WALK_TRAILING, dentry); 3588ff326a32SAl Viro if (unlikely(res)) 35891ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3590b0417d2cSAl Viro return res; 35911ccac622SAl Viro } 359231d1726dSAl Viro 3593c981a482SAl Viro /* 3594c981a482SAl Viro * Handle the last step of open() 3595c981a482SAl Viro */ 3596c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3597c981a482SAl Viro struct file *file, const struct open_flags *op) 3598c981a482SAl Viro { 3599abf08576SChristian Brauner struct mnt_idmap *idmap; 3600c981a482SAl Viro int open_flag = op->open_flag; 3601c981a482SAl Viro bool do_truncate; 3602c981a482SAl Viro int acc_mode; 3603c981a482SAl Viro int error; 3604c981a482SAl Viro 3605ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3606ff326a32SAl Viro error = complete_walk(nd); 3607ff326a32SAl Viro if (error) 3608ff326a32SAl Viro return error; 3609ff326a32SAl Viro } 3610973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 361176ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3612abf08576SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 361330aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3614b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3615b94e0b32SAl Viro return -EEXIST; 361630aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3617c5971b8cSAl Viro return -EISDIR; 3618e67fe633SChristian Brauner error = may_create_in_sticky(idmap, nd, 361930aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 362030aba665SSalvatore Mesoraca if (unlikely(error)) 3621c5971b8cSAl Viro return error; 362230aba665SSalvatore Mesoraca } 362344b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3624c5971b8cSAl Viro return -ENOTDIR; 36256c0d46c4SAl Viro 36268795e7d4SAl Viro do_truncate = false; 36278795e7d4SAl Viro acc_mode = op->acc_mode; 36285a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 36295a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 36305a2d3eddSAl Viro open_flag &= ~O_TRUNC; 36315a2d3eddSAl Viro acc_mode = 0; 36328795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 36330f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 36340f9d1a10SAl Viro if (error) 3635c5971b8cSAl Viro return error; 36368795e7d4SAl Viro do_truncate = true; 36370f9d1a10SAl Viro } 36384609e1f1SChristian Brauner error = may_open(idmap, &nd->path, acc_mode, open_flag); 36398795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3640ae2bb293SAl Viro error = vfs_open(&nd->path, file); 36418795e7d4SAl Viro if (!error) 36426035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 36438795e7d4SAl Viro if (!error && do_truncate) 3644abf08576SChristian Brauner error = handle_truncate(idmap, file); 3645c80567c8SAl Viro if (unlikely(error > 0)) { 3646c80567c8SAl Viro WARN_ON(1); 3647c80567c8SAl Viro error = -EINVAL; 3648c80567c8SAl Viro } 36498795e7d4SAl Viro if (do_truncate) 36500f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3651c5971b8cSAl Viro return error; 3652fb1cc555SAl Viro } 3653fb1cc555SAl Viro 36546521f891SChristian Brauner /** 36556521f891SChristian Brauner * vfs_tmpfile - create tmpfile 3656abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 365773bb5a90SRoberto Sassu * @parentpath: pointer to the path of the base directory 365873bb5a90SRoberto Sassu * @file: file descriptor of the new tmpfile 36596521f891SChristian Brauner * @mode: mode of the new tmpfile 36606521f891SChristian Brauner * 36616521f891SChristian Brauner * Create a temporary file. 36626521f891SChristian Brauner * 3663abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3664abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3665abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 36666521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3667abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 36686521f891SChristian Brauner */ 3669abf08576SChristian Brauner static int vfs_tmpfile(struct mnt_idmap *idmap, 36709751b338SMiklos Szeredi const struct path *parentpath, 36719751b338SMiklos Szeredi struct file *file, umode_t mode) 3672af7bd4dcSAmir Goldstein { 36739751b338SMiklos Szeredi struct dentry *child; 36749751b338SMiklos Szeredi struct inode *dir = d_inode(parentpath->dentry); 3675af7bd4dcSAmir Goldstein struct inode *inode; 3676af7bd4dcSAmir Goldstein int error; 3677406c706cSPeter Griffin int open_flag = file->f_flags; 3678af7bd4dcSAmir Goldstein 3679af7bd4dcSAmir Goldstein /* we want directory to be writable */ 36804609e1f1SChristian Brauner error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 3681af7bd4dcSAmir Goldstein if (error) 36829751b338SMiklos Szeredi return error; 3683af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 36849751b338SMiklos Szeredi return -EOPNOTSUPP; 36859751b338SMiklos Szeredi child = d_alloc(parentpath->dentry, &slash_name); 3686af7bd4dcSAmir Goldstein if (unlikely(!child)) 36879751b338SMiklos Szeredi return -ENOMEM; 36889751b338SMiklos Szeredi file->f_path.mnt = parentpath->mnt; 36899751b338SMiklos Szeredi file->f_path.dentry = child; 36909452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir, mode, mode, mode); 3691011e2b71SChristian Brauner error = dir->i_op->tmpfile(idmap, dir, file, mode); 36929751b338SMiklos Szeredi dput(child); 3693af7bd4dcSAmir Goldstein if (error) 36949751b338SMiklos Szeredi return error; 36959751b338SMiklos Szeredi /* Don't check for other permissions, the inode was just created */ 36964609e1f1SChristian Brauner error = may_open(idmap, &file->f_path, 0, file->f_flags); 36979751b338SMiklos Szeredi if (error) 36989751b338SMiklos Szeredi return error; 36999751b338SMiklos Szeredi inode = file_inode(file); 3700406c706cSPeter Griffin if (!(open_flag & O_EXCL)) { 3701af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3702af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3703af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3704af7bd4dcSAmir Goldstein } 370539f60c1cSChristian Brauner ima_post_create_tmpfile(idmap, inode); 37069751b338SMiklos Szeredi return 0; 3707af7bd4dcSAmir Goldstein } 3708af7bd4dcSAmir Goldstein 370922873deaSMiklos Szeredi /** 3710d56e0ddbSAmir Goldstein * kernel_tmpfile_open - open a tmpfile for kernel internal use 3711abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 371222873deaSMiklos Szeredi * @parentpath: path of the base directory 371322873deaSMiklos Szeredi * @mode: mode of the new tmpfile 371422873deaSMiklos Szeredi * @open_flag: flags 371522873deaSMiklos Szeredi * @cred: credentials for open 371622873deaSMiklos Szeredi * 371722873deaSMiklos Szeredi * Create and open a temporary file. The file is not accounted in nr_files, 371822873deaSMiklos Szeredi * hence this is only for kernel internal use, and must not be installed into 371922873deaSMiklos Szeredi * file tables or such. 372022873deaSMiklos Szeredi */ 3721d56e0ddbSAmir Goldstein struct file *kernel_tmpfile_open(struct mnt_idmap *idmap, 372222873deaSMiklos Szeredi const struct path *parentpath, 3723d56e0ddbSAmir Goldstein umode_t mode, int open_flag, 3724d56e0ddbSAmir Goldstein const struct cred *cred) 372522873deaSMiklos Szeredi { 372622873deaSMiklos Szeredi struct file *file; 372722873deaSMiklos Szeredi int error; 372822873deaSMiklos Szeredi 37299751b338SMiklos Szeredi file = alloc_empty_file_noaccount(open_flag, cred); 3730d56e0ddbSAmir Goldstein if (IS_ERR(file)) 3731d56e0ddbSAmir Goldstein return file; 3732d56e0ddbSAmir Goldstein 3733abf08576SChristian Brauner error = vfs_tmpfile(idmap, parentpath, file, mode); 37349751b338SMiklos Szeredi if (error) { 37359751b338SMiklos Szeredi fput(file); 373622873deaSMiklos Szeredi file = ERR_PTR(error); 37379751b338SMiklos Szeredi } 373822873deaSMiklos Szeredi return file; 373922873deaSMiklos Szeredi } 3740d56e0ddbSAmir Goldstein EXPORT_SYMBOL(kernel_tmpfile_open); 3741648fa861SAl Viro 3742c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 374360545d0dSAl Viro const struct open_flags *op, 37443ec2eef1SAl Viro struct file *file) 374560545d0dSAl Viro { 3746625b6d10SAl Viro struct path path; 3747c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 37489751b338SMiklos Szeredi 374960545d0dSAl Viro if (unlikely(error)) 375060545d0dSAl Viro return error; 3751625b6d10SAl Viro error = mnt_want_write(path.mnt); 375260545d0dSAl Viro if (unlikely(error)) 375360545d0dSAl Viro goto out; 3754abf08576SChristian Brauner error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode); 37559751b338SMiklos Szeredi if (error) 375660545d0dSAl Viro goto out2; 37579751b338SMiklos Szeredi audit_inode(nd->name, file->f_path.dentry, 0); 375860545d0dSAl Viro out2: 3759625b6d10SAl Viro mnt_drop_write(path.mnt); 376060545d0dSAl Viro out: 3761625b6d10SAl Viro path_put(&path); 376260545d0dSAl Viro return error; 376360545d0dSAl Viro } 376460545d0dSAl Viro 37656ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 37666ac08709SAl Viro { 37676ac08709SAl Viro struct path path; 37686ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 37696ac08709SAl Viro if (!error) { 37706ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3771ae2bb293SAl Viro error = vfs_open(&path, file); 37726ac08709SAl Viro path_put(&path); 37736ac08709SAl Viro } 37746ac08709SAl Viro return error; 37756ac08709SAl Viro } 37766ac08709SAl Viro 3777c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3778c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 37791da177e4SLinus Torvalds { 378030d90494SAl Viro struct file *file; 378113aab428SAl Viro int error; 378231e6b01fSNick Piggin 3783ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 37841afc99beSAl Viro if (IS_ERR(file)) 37851afc99beSAl Viro return file; 378631e6b01fSNick Piggin 3787bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 37883ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 37895f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 37906ac08709SAl Viro error = do_o_path(nd, flags, file); 37915f336e72SAl Viro } else { 37925f336e72SAl Viro const char *s = path_init(nd, flags); 37933bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3794c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 37951ccac622SAl Viro ; 3796c5971b8cSAl Viro if (!error) 3797c5971b8cSAl Viro error = do_open(nd, file, op); 3798deb106c6SAl Viro terminate_walk(nd); 37995f336e72SAl Viro } 38007c1c01ecSAl Viro if (likely(!error)) { 3801aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 38027c1c01ecSAl Viro return file; 38037c1c01ecSAl Viro WARN_ON(1); 38047c1c01ecSAl Viro error = -EINVAL; 3805015c3bbcSMiklos Szeredi } 38067c1c01ecSAl Viro fput(file); 38072675a4ebSAl Viro if (error == -EOPENSTALE) { 38082675a4ebSAl Viro if (flags & LOOKUP_RCU) 38092675a4ebSAl Viro error = -ECHILD; 38102675a4ebSAl Viro else 38112675a4ebSAl Viro error = -ESTALE; 38122675a4ebSAl Viro } 38137c1c01ecSAl Viro return ERR_PTR(error); 3814de459215SKirill Korotaev } 38151da177e4SLinus Torvalds 3816669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3817f9652e10SAl Viro const struct open_flags *op) 381813aab428SAl Viro { 38199883d185SAl Viro struct nameidata nd; 3820f9652e10SAl Viro int flags = op->lookup_flags; 382113aab428SAl Viro struct file *filp; 382213aab428SAl Viro 382306422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3824c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 382513aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3826c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 382713aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3828c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 38299883d185SAl Viro restore_nameidata(); 383013aab428SAl Viro return filp; 383113aab428SAl Viro } 383213aab428SAl Viro 3833ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3834f9652e10SAl Viro const char *name, const struct open_flags *op) 383573d049a4SAl Viro { 38369883d185SAl Viro struct nameidata nd; 383773d049a4SAl Viro struct file *file; 383851689104SPaul Moore struct filename *filename; 3839bcba1e7dSAl Viro int flags = op->lookup_flags; 384073d049a4SAl Viro 3841ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 384273d049a4SAl Viro return ERR_PTR(-ELOOP); 384373d049a4SAl Viro 384451689104SPaul Moore filename = getname_kernel(name); 3845a1c83681SViresh Kumar if (IS_ERR(filename)) 384651689104SPaul Moore return ERR_CAST(filename); 384751689104SPaul Moore 384806422964SAl Viro set_nameidata(&nd, -1, filename, root); 3849c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 385073d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3851c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 385273d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3853c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 38549883d185SAl Viro restore_nameidata(); 385551689104SPaul Moore putname(filename); 385673d049a4SAl Viro return file; 385773d049a4SAl Viro } 385873d049a4SAl Viro 3859b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 38601ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 38611da177e4SLinus Torvalds { 3862c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3863391172c4SAl Viro struct qstr last; 3864b3d4650dSNeilBrown bool want_dir = lookup_flags & LOOKUP_DIRECTORY; 3865b3d4650dSNeilBrown unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; 3866b3d4650dSNeilBrown unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; 3867391172c4SAl Viro int type; 3868c30dabfeSJan Kara int err2; 38691ac12b4bSJeff Layton int error; 38701ac12b4bSJeff Layton 3871b3d4650dSNeilBrown error = filename_parentat(dfd, name, reval_flag, path, &last, &type); 38720ee50b47SDmitry Kadashev if (error) 38730ee50b47SDmitry Kadashev return ERR_PTR(error); 38741da177e4SLinus Torvalds 3875c663e5d8SChristoph Hellwig /* 3876c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3877c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3878c663e5d8SChristoph Hellwig */ 38795c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3880ed75e95dSAl Viro goto out; 3881c663e5d8SChristoph Hellwig 3882c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3883391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3884c663e5d8SChristoph Hellwig /* 3885b3d4650dSNeilBrown * Do the final lookup. Suppress 'create' if there is a trailing 3886b3d4650dSNeilBrown * '/', and a directory wasn't requested. 3887c663e5d8SChristoph Hellwig */ 3888b3d4650dSNeilBrown if (last.name[last.len] && !want_dir) 3889b3d4650dSNeilBrown create_flags = 0; 38905955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 389174d7970fSNamjae Jeon dentry = lookup_one_qstr_excl(&last, path->dentry, 389274d7970fSNamjae Jeon reval_flag | create_flags); 38931da177e4SLinus Torvalds if (IS_ERR(dentry)) 3894a8104a9fSAl Viro goto unlock; 3895c663e5d8SChristoph Hellwig 3896a8104a9fSAl Viro error = -EEXIST; 3897b18825a7SDavid Howells if (d_is_positive(dentry)) 3898a8104a9fSAl Viro goto fail; 3899b18825a7SDavid Howells 3900c663e5d8SChristoph Hellwig /* 3901c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3902c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3903c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3904c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3905c663e5d8SChristoph Hellwig */ 3906b3d4650dSNeilBrown if (unlikely(!create_flags)) { 3907a8104a9fSAl Viro error = -ENOENT; 3908ed75e95dSAl Viro goto fail; 3909e9baf6e5SAl Viro } 3910c30dabfeSJan Kara if (unlikely(err2)) { 3911c30dabfeSJan Kara error = err2; 3912a8104a9fSAl Viro goto fail; 3913c30dabfeSJan Kara } 3914e9baf6e5SAl Viro return dentry; 39151da177e4SLinus Torvalds fail: 3916a8104a9fSAl Viro dput(dentry); 3917a8104a9fSAl Viro dentry = ERR_PTR(error); 3918a8104a9fSAl Viro unlock: 39195955102cSAl Viro inode_unlock(path->dentry->d_inode); 3920c30dabfeSJan Kara if (!err2) 3921391172c4SAl Viro mnt_drop_write(path->mnt); 3922ed75e95dSAl Viro out: 3923391172c4SAl Viro path_put(path); 3924ed75e95dSAl Viro return dentry; 3925dae6ad8fSAl Viro } 3926fa14a0b8SAl Viro 3927fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3928fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3929fa14a0b8SAl Viro { 3930b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3931b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3932b4a4f213SStephen Brennan 3933b4a4f213SStephen Brennan putname(filename); 3934b4a4f213SStephen Brennan return res; 3935fa14a0b8SAl Viro } 3936dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3937dae6ad8fSAl Viro 3938921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3939921a1650SAl Viro { 3940921a1650SAl Viro dput(dentry); 39415955102cSAl Viro inode_unlock(path->dentry->d_inode); 3942a8104a9fSAl Viro mnt_drop_write(path->mnt); 3943921a1650SAl Viro path_put(path); 3944921a1650SAl Viro } 3945921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3946921a1650SAl Viro 3947520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 39481ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3949dae6ad8fSAl Viro { 3950b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3951b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3952b4a4f213SStephen Brennan 3953b4a4f213SStephen Brennan putname(filename); 3954b4a4f213SStephen Brennan return res; 3955dae6ad8fSAl Viro } 3956dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3957dae6ad8fSAl Viro 39586521f891SChristian Brauner /** 39596521f891SChristian Brauner * vfs_mknod - create device node or file 3960abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 39616521f891SChristian Brauner * @dir: inode of @dentry 39626521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39636521f891SChristian Brauner * @mode: mode of the new device node or file 39646521f891SChristian Brauner * @dev: device number of device to create 39656521f891SChristian Brauner * 39666521f891SChristian Brauner * Create a device node or file. 39676521f891SChristian Brauner * 3968abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3969abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3970abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 39716521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3972abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 39736521f891SChristian Brauner */ 3974abf08576SChristian Brauner int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir, 39756521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 39761da177e4SLinus Torvalds { 3977a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 39784609e1f1SChristian Brauner int error = may_create(idmap, dir, dentry); 39791da177e4SLinus Torvalds 39801da177e4SLinus Torvalds if (error) 39811da177e4SLinus Torvalds return error; 39821da177e4SLinus Torvalds 3983a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3984a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 39851da177e4SLinus Torvalds return -EPERM; 39861da177e4SLinus Torvalds 3987acfa4380SAl Viro if (!dir->i_op->mknod) 39881da177e4SLinus Torvalds return -EPERM; 39891da177e4SLinus Torvalds 39909452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir, mode, mode, mode); 399108ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 399208ce5f16SSerge E. Hallyn if (error) 399308ce5f16SSerge E. Hallyn return error; 399408ce5f16SSerge E. Hallyn 39951da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 39961da177e4SLinus Torvalds if (error) 39971da177e4SLinus Torvalds return error; 39981da177e4SLinus Torvalds 39995ebb29beSChristian Brauner error = dir->i_op->mknod(idmap, dir, dentry, mode, dev); 4000a74574aaSStephen Smalley if (!error) 4001f38aa942SAmy Griffis fsnotify_create(dir, dentry); 40021da177e4SLinus Torvalds return error; 40031da177e4SLinus Torvalds } 40044d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 40051da177e4SLinus Torvalds 4006f69aac00SAl Viro static int may_mknod(umode_t mode) 4007463c3197SDave Hansen { 4008463c3197SDave Hansen switch (mode & S_IFMT) { 4009463c3197SDave Hansen case S_IFREG: 4010463c3197SDave Hansen case S_IFCHR: 4011463c3197SDave Hansen case S_IFBLK: 4012463c3197SDave Hansen case S_IFIFO: 4013463c3197SDave Hansen case S_IFSOCK: 4014463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 4015463c3197SDave Hansen return 0; 4016463c3197SDave Hansen case S_IFDIR: 4017463c3197SDave Hansen return -EPERM; 4018463c3197SDave Hansen default: 4019463c3197SDave Hansen return -EINVAL; 4020463c3197SDave Hansen } 4021463c3197SDave Hansen } 4022463c3197SDave Hansen 402345f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 402487c4e192SDominik Brodowski unsigned int dev) 40251da177e4SLinus Torvalds { 4026abf08576SChristian Brauner struct mnt_idmap *idmap; 40271da177e4SLinus Torvalds struct dentry *dentry; 4028dae6ad8fSAl Viro struct path path; 4029dae6ad8fSAl Viro int error; 4030972567f1SJeff Layton unsigned int lookup_flags = 0; 40311da177e4SLinus Torvalds 40328e4bfca1SAl Viro error = may_mknod(mode); 40338e4bfca1SAl Viro if (error) 40347797251bSDmitry Kadashev goto out1; 4035972567f1SJeff Layton retry: 4036b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 40377797251bSDmitry Kadashev error = PTR_ERR(dentry); 4038dae6ad8fSAl Viro if (IS_ERR(dentry)) 40397797251bSDmitry Kadashev goto out1; 40402ad94ae6SAl Viro 40411639a49cSYang Xu error = security_path_mknod(&path, dentry, 40421639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode), dev); 4043be6d3e56SKentaro Takeda if (error) 40447797251bSDmitry Kadashev goto out2; 40456521f891SChristian Brauner 4046abf08576SChristian Brauner idmap = mnt_idmap(path.mnt); 40471da177e4SLinus Torvalds switch (mode & S_IFMT) { 40481da177e4SLinus Torvalds case 0: case S_IFREG: 4049abf08576SChristian Brauner error = vfs_create(idmap, path.dentry->d_inode, 40506521f891SChristian Brauner dentry, mode, true); 405105d1a717SMimi Zohar if (!error) 405239f60c1cSChristian Brauner ima_post_path_mknod(idmap, dentry); 40531da177e4SLinus Torvalds break; 40541da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 4055abf08576SChristian Brauner error = vfs_mknod(idmap, path.dentry->d_inode, 40566521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 40571da177e4SLinus Torvalds break; 40581da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 4059abf08576SChristian Brauner error = vfs_mknod(idmap, path.dentry->d_inode, 40606521f891SChristian Brauner dentry, mode, 0); 40611da177e4SLinus Torvalds break; 40621da177e4SLinus Torvalds } 40637797251bSDmitry Kadashev out2: 4064921a1650SAl Viro done_path_create(&path, dentry); 4065972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 4066972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 4067972567f1SJeff Layton goto retry; 4068972567f1SJeff Layton } 40697797251bSDmitry Kadashev out1: 40707797251bSDmitry Kadashev putname(name); 40711da177e4SLinus Torvalds return error; 40721da177e4SLinus Torvalds } 40731da177e4SLinus Torvalds 407487c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 407587c4e192SDominik Brodowski unsigned int, dev) 407687c4e192SDominik Brodowski { 40777797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 407887c4e192SDominik Brodowski } 407987c4e192SDominik Brodowski 40808208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 40815590ff0dSUlrich Drepper { 40827797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 40835590ff0dSUlrich Drepper } 40845590ff0dSUlrich Drepper 40856521f891SChristian Brauner /** 40866521f891SChristian Brauner * vfs_mkdir - create directory 4087abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 40886521f891SChristian Brauner * @dir: inode of @dentry 40896521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 40906521f891SChristian Brauner * @mode: mode of the new directory 40916521f891SChristian Brauner * 40926521f891SChristian Brauner * Create a directory. 40936521f891SChristian Brauner * 4094abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4095abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4096abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 40976521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4098abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 40996521f891SChristian Brauner */ 4100abf08576SChristian Brauner int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, 41016521f891SChristian Brauner struct dentry *dentry, umode_t mode) 41021da177e4SLinus Torvalds { 4103abf08576SChristian Brauner int error; 41048de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 41051da177e4SLinus Torvalds 41064609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 41071da177e4SLinus Torvalds if (error) 41081da177e4SLinus Torvalds return error; 41091da177e4SLinus Torvalds 4110acfa4380SAl Viro if (!dir->i_op->mkdir) 41111da177e4SLinus Torvalds return -EPERM; 41121da177e4SLinus Torvalds 41139452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0); 41141da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 41151da177e4SLinus Torvalds if (error) 41161da177e4SLinus Torvalds return error; 41171da177e4SLinus Torvalds 41188de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 41198de52778SAl Viro return -EMLINK; 41208de52778SAl Viro 4121c54bd91eSChristian Brauner error = dir->i_op->mkdir(idmap, dir, dentry, mode); 4122a74574aaSStephen Smalley if (!error) 4123f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 41241da177e4SLinus Torvalds return error; 41251da177e4SLinus Torvalds } 41264d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 41271da177e4SLinus Torvalds 412845f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 41291da177e4SLinus Torvalds { 41306902d925SDave Hansen struct dentry *dentry; 4131dae6ad8fSAl Viro struct path path; 4132dae6ad8fSAl Viro int error; 4133b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 41341da177e4SLinus Torvalds 4135b76d8b82SJeff Layton retry: 4136b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 4137584d3226SDmitry Kadashev error = PTR_ERR(dentry); 41386902d925SDave Hansen if (IS_ERR(dentry)) 4139584d3226SDmitry Kadashev goto out_putname; 41406902d925SDave Hansen 41411639a49cSYang Xu error = security_path_mkdir(&path, dentry, 41421639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode)); 41436521f891SChristian Brauner if (!error) { 4144abf08576SChristian Brauner error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode, 4145abf08576SChristian Brauner dentry, mode); 41466521f891SChristian Brauner } 4147921a1650SAl Viro done_path_create(&path, dentry); 4148b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 4149b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 4150b76d8b82SJeff Layton goto retry; 4151b76d8b82SJeff Layton } 4152584d3226SDmitry Kadashev out_putname: 4153584d3226SDmitry Kadashev putname(name); 41541da177e4SLinus Torvalds return error; 41551da177e4SLinus Torvalds } 41561da177e4SLinus Torvalds 41570101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 41580101db7aSDominik Brodowski { 4159584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 41600101db7aSDominik Brodowski } 41610101db7aSDominik Brodowski 4162a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 41635590ff0dSUlrich Drepper { 4164584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 41655590ff0dSUlrich Drepper } 41665590ff0dSUlrich Drepper 41676521f891SChristian Brauner /** 41686521f891SChristian Brauner * vfs_rmdir - remove directory 4169abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 41706521f891SChristian Brauner * @dir: inode of @dentry 41716521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 41726521f891SChristian Brauner * 41736521f891SChristian Brauner * Remove a directory. 41746521f891SChristian Brauner * 4175abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4176abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4177abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 41786521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4179abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 41806521f891SChristian Brauner */ 4181abf08576SChristian Brauner int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir, 41826521f891SChristian Brauner struct dentry *dentry) 41831da177e4SLinus Torvalds { 41844609e1f1SChristian Brauner int error = may_delete(idmap, dir, dentry, 1); 41851da177e4SLinus Torvalds 41861da177e4SLinus Torvalds if (error) 41871da177e4SLinus Torvalds return error; 41881da177e4SLinus Torvalds 4189acfa4380SAl Viro if (!dir->i_op->rmdir) 41901da177e4SLinus Torvalds return -EPERM; 41911da177e4SLinus Torvalds 41921d2ef590SAl Viro dget(dentry); 41935955102cSAl Viro inode_lock(dentry->d_inode); 4194912dbc15SSage Weil 41951da177e4SLinus Torvalds error = -EBUSY; 41961bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 41971bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4198912dbc15SSage Weil goto out; 4199912dbc15SSage Weil 42001da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4201912dbc15SSage Weil if (error) 4202912dbc15SSage Weil goto out; 4203912dbc15SSage Weil 42041da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4205912dbc15SSage Weil if (error) 4206912dbc15SSage Weil goto out; 4207912dbc15SSage Weil 42088767712fSAl Viro shrink_dcache_parent(dentry); 42091da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4210d83c49f3SAl Viro dont_mount(dentry); 42118ed936b5SEric W. Biederman detach_mounts(dentry); 42121da177e4SLinus Torvalds 4213912dbc15SSage Weil out: 42145955102cSAl Viro inode_unlock(dentry->d_inode); 42151d2ef590SAl Viro dput(dentry); 4216912dbc15SSage Weil if (!error) 4217a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 42181da177e4SLinus Torvalds return error; 42191da177e4SLinus Torvalds } 42204d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 42211da177e4SLinus Torvalds 422245f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 42231da177e4SLinus Torvalds { 42240ee50b47SDmitry Kadashev int error; 42251da177e4SLinus Torvalds struct dentry *dentry; 4226f5beed75SAl Viro struct path path; 4227f5beed75SAl Viro struct qstr last; 4228f5beed75SAl Viro int type; 4229c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4230c6ee9206SJeff Layton retry: 4231c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 42320ee50b47SDmitry Kadashev if (error) 42330ee50b47SDmitry Kadashev goto exit1; 42341da177e4SLinus Torvalds 4235f5beed75SAl Viro switch (type) { 42361da177e4SLinus Torvalds case LAST_DOTDOT: 42371da177e4SLinus Torvalds error = -ENOTEMPTY; 42380ee50b47SDmitry Kadashev goto exit2; 42391da177e4SLinus Torvalds case LAST_DOT: 42401da177e4SLinus Torvalds error = -EINVAL; 42410ee50b47SDmitry Kadashev goto exit2; 42421da177e4SLinus Torvalds case LAST_ROOT: 42431da177e4SLinus Torvalds error = -EBUSY; 42440ee50b47SDmitry Kadashev goto exit2; 42451da177e4SLinus Torvalds } 42460612d9fbSOGAWA Hirofumi 4247f5beed75SAl Viro error = mnt_want_write(path.mnt); 4248c30dabfeSJan Kara if (error) 42490ee50b47SDmitry Kadashev goto exit2; 42500612d9fbSOGAWA Hirofumi 42515955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 425274d7970fSNamjae Jeon dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags); 42531da177e4SLinus Torvalds error = PTR_ERR(dentry); 42546902d925SDave Hansen if (IS_ERR(dentry)) 42550ee50b47SDmitry Kadashev goto exit3; 4256e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4257e6bc45d6STheodore Ts'o error = -ENOENT; 42580ee50b47SDmitry Kadashev goto exit4; 4259e6bc45d6STheodore Ts'o } 4260f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4261be6d3e56SKentaro Takeda if (error) 42620ee50b47SDmitry Kadashev goto exit4; 4263abf08576SChristian Brauner error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry); 42640ee50b47SDmitry Kadashev exit4: 42651da177e4SLinus Torvalds dput(dentry); 42660ee50b47SDmitry Kadashev exit3: 42675955102cSAl Viro inode_unlock(path.dentry->d_inode); 4268f5beed75SAl Viro mnt_drop_write(path.mnt); 42690ee50b47SDmitry Kadashev exit2: 4270f5beed75SAl Viro path_put(&path); 4271c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4272c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4273c6ee9206SJeff Layton goto retry; 4274c6ee9206SJeff Layton } 42750ee50b47SDmitry Kadashev exit1: 427624fb33d4SAl Viro putname(name); 42771da177e4SLinus Torvalds return error; 42781da177e4SLinus Torvalds } 42791da177e4SLinus Torvalds 42803cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 42815590ff0dSUlrich Drepper { 4282e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 42835590ff0dSUlrich Drepper } 42845590ff0dSUlrich Drepper 4285b21996e3SJ. Bruce Fields /** 4286b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 4287abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 4288b21996e3SJ. Bruce Fields * @dir: parent directory 4289b21996e3SJ. Bruce Fields * @dentry: victim 4290b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4291b21996e3SJ. Bruce Fields * 4292b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4293b21996e3SJ. Bruce Fields * 4294b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4295b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4296b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4297b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4298b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4299b21996e3SJ. Bruce Fields * 4300b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4301b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4302b21996e3SJ. Bruce Fields * to be NFS exported. 43036521f891SChristian Brauner * 4304abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4305abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4306abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 43076521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4308abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 4309b21996e3SJ. Bruce Fields */ 4310abf08576SChristian Brauner int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir, 43116521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 43121da177e4SLinus Torvalds { 43139accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 43144609e1f1SChristian Brauner int error = may_delete(idmap, dir, dentry, 0); 43151da177e4SLinus Torvalds 43161da177e4SLinus Torvalds if (error) 43171da177e4SLinus Torvalds return error; 43181da177e4SLinus Torvalds 4319acfa4380SAl Viro if (!dir->i_op->unlink) 43201da177e4SLinus Torvalds return -EPERM; 43211da177e4SLinus Torvalds 43225955102cSAl Viro inode_lock(target); 432351cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 432451cc3a66SHugh Dickins error = -EPERM; 432551cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 43261da177e4SLinus Torvalds error = -EBUSY; 43271da177e4SLinus Torvalds else { 43281da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4329bec1052eSAl Viro if (!error) { 43305a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 43315a14696cSJ. Bruce Fields if (error) 4332b21996e3SJ. Bruce Fields goto out; 43331da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 43348ed936b5SEric W. Biederman if (!error) { 4335d83c49f3SAl Viro dont_mount(dentry); 43368ed936b5SEric W. Biederman detach_mounts(dentry); 43378ed936b5SEric W. Biederman } 4338bec1052eSAl Viro } 43391da177e4SLinus Torvalds } 4340b21996e3SJ. Bruce Fields out: 43415955102cSAl Viro inode_unlock(target); 43421da177e4SLinus Torvalds 43431da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4344a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4345a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4346a37d9a17SAmir Goldstein } else if (!error) { 43479accbb97SJ. Bruce Fields fsnotify_link_count(target); 4348a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 43491da177e4SLinus Torvalds } 43500eeca283SRobert Love 43511da177e4SLinus Torvalds return error; 43521da177e4SLinus Torvalds } 43534d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 43541da177e4SLinus Torvalds 43551da177e4SLinus Torvalds /* 43561da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 43571b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 43581da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 43591da177e4SLinus Torvalds * while waiting on the I/O. 43601da177e4SLinus Torvalds */ 436145f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 43621da177e4SLinus Torvalds { 43632ad94ae6SAl Viro int error; 43641da177e4SLinus Torvalds struct dentry *dentry; 4365f5beed75SAl Viro struct path path; 4366f5beed75SAl Viro struct qstr last; 4367f5beed75SAl Viro int type; 43681da177e4SLinus Torvalds struct inode *inode = NULL; 4369b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 43705d18f813SJeff Layton unsigned int lookup_flags = 0; 43715d18f813SJeff Layton retry: 4372c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 43730ee50b47SDmitry Kadashev if (error) 43740ee50b47SDmitry Kadashev goto exit1; 43752ad94ae6SAl Viro 43761da177e4SLinus Torvalds error = -EISDIR; 4377f5beed75SAl Viro if (type != LAST_NORM) 43780ee50b47SDmitry Kadashev goto exit2; 43790612d9fbSOGAWA Hirofumi 4380f5beed75SAl Viro error = mnt_want_write(path.mnt); 4381c30dabfeSJan Kara if (error) 43820ee50b47SDmitry Kadashev goto exit2; 4383b21996e3SJ. Bruce Fields retry_deleg: 43845955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 438574d7970fSNamjae Jeon dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags); 43861da177e4SLinus Torvalds error = PTR_ERR(dentry); 43871da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 43886521f891SChristian Brauner 43891da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4390f5beed75SAl Viro if (last.name[last.len]) 439150338b88STörök Edwin goto slashes; 43921da177e4SLinus Torvalds inode = dentry->d_inode; 4393b18825a7SDavid Howells if (d_is_negative(dentry)) 4394e6bc45d6STheodore Ts'o goto slashes; 43957de9c6eeSAl Viro ihold(inode); 4396f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4397be6d3e56SKentaro Takeda if (error) 43980ee50b47SDmitry Kadashev goto exit3; 4399abf08576SChristian Brauner error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode, 4400abf08576SChristian Brauner dentry, &delegated_inode); 44010ee50b47SDmitry Kadashev exit3: 44021da177e4SLinus Torvalds dput(dentry); 44031da177e4SLinus Torvalds } 44045955102cSAl Viro inode_unlock(path.dentry->d_inode); 44051da177e4SLinus Torvalds if (inode) 44061da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4407b21996e3SJ. Bruce Fields inode = NULL; 4408b21996e3SJ. Bruce Fields if (delegated_inode) { 44095a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4410b21996e3SJ. Bruce Fields if (!error) 4411b21996e3SJ. Bruce Fields goto retry_deleg; 4412b21996e3SJ. Bruce Fields } 4413f5beed75SAl Viro mnt_drop_write(path.mnt); 44140ee50b47SDmitry Kadashev exit2: 4415f5beed75SAl Viro path_put(&path); 44165d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 44175d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 44185d18f813SJeff Layton inode = NULL; 44195d18f813SJeff Layton goto retry; 44205d18f813SJeff Layton } 44210ee50b47SDmitry Kadashev exit1: 4422da2f1362SChristoph Hellwig putname(name); 44231da177e4SLinus Torvalds return error; 44241da177e4SLinus Torvalds 44251da177e4SLinus Torvalds slashes: 4426b18825a7SDavid Howells if (d_is_negative(dentry)) 4427b18825a7SDavid Howells error = -ENOENT; 442844b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4429b18825a7SDavid Howells error = -EISDIR; 4430b18825a7SDavid Howells else 4431b18825a7SDavid Howells error = -ENOTDIR; 44320ee50b47SDmitry Kadashev goto exit3; 44331da177e4SLinus Torvalds } 44341da177e4SLinus Torvalds 44352e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 44365590ff0dSUlrich Drepper { 44375590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 44385590ff0dSUlrich Drepper return -EINVAL; 44395590ff0dSUlrich Drepper 44405590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4441e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4442da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 44435590ff0dSUlrich Drepper } 44445590ff0dSUlrich Drepper 44453480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 44465590ff0dSUlrich Drepper { 4447da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 44485590ff0dSUlrich Drepper } 44495590ff0dSUlrich Drepper 44506521f891SChristian Brauner /** 44516521f891SChristian Brauner * vfs_symlink - create symlink 4452abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 44536521f891SChristian Brauner * @dir: inode of @dentry 44546521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 44556521f891SChristian Brauner * @oldname: name of the file to link to 44566521f891SChristian Brauner * 44576521f891SChristian Brauner * Create a symlink. 44586521f891SChristian Brauner * 4459abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4460abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4461abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 44626521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4463abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 44646521f891SChristian Brauner */ 4465abf08576SChristian Brauner int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir, 44666521f891SChristian Brauner struct dentry *dentry, const char *oldname) 44671da177e4SLinus Torvalds { 44687a77db95SChristian Brauner int error; 44691da177e4SLinus Torvalds 44704609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 44711da177e4SLinus Torvalds if (error) 44721da177e4SLinus Torvalds return error; 44731da177e4SLinus Torvalds 4474acfa4380SAl Viro if (!dir->i_op->symlink) 44751da177e4SLinus Torvalds return -EPERM; 44761da177e4SLinus Torvalds 44771da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 44781da177e4SLinus Torvalds if (error) 44791da177e4SLinus Torvalds return error; 44801da177e4SLinus Torvalds 44817a77db95SChristian Brauner error = dir->i_op->symlink(idmap, dir, dentry, oldname); 4482a74574aaSStephen Smalley if (!error) 4483f38aa942SAmy Griffis fsnotify_create(dir, dentry); 44841da177e4SLinus Torvalds return error; 44851da177e4SLinus Torvalds } 44864d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 44871da177e4SLinus Torvalds 44887a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 44891da177e4SLinus Torvalds { 44902ad94ae6SAl Viro int error; 44916902d925SDave Hansen struct dentry *dentry; 4492dae6ad8fSAl Viro struct path path; 4493f46d3567SJeff Layton unsigned int lookup_flags = 0; 44941da177e4SLinus Torvalds 4495da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4496da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4497da2d0cedSDmitry Kadashev goto out_putnames; 4498da2d0cedSDmitry Kadashev } 4499f46d3567SJeff Layton retry: 4500b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 45011da177e4SLinus Torvalds error = PTR_ERR(dentry); 45026902d925SDave Hansen if (IS_ERR(dentry)) 4503da2d0cedSDmitry Kadashev goto out_putnames; 45046902d925SDave Hansen 450591a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4506abf08576SChristian Brauner if (!error) 4507abf08576SChristian Brauner error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode, 4508abf08576SChristian Brauner dentry, from->name); 4509921a1650SAl Viro done_path_create(&path, dentry); 4510f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4511f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4512f46d3567SJeff Layton goto retry; 4513f46d3567SJeff Layton } 4514da2d0cedSDmitry Kadashev out_putnames: 4515da2d0cedSDmitry Kadashev putname(to); 45161da177e4SLinus Torvalds putname(from); 45171da177e4SLinus Torvalds return error; 45181da177e4SLinus Torvalds } 45191da177e4SLinus Torvalds 4520b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4521b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4522b724e846SDominik Brodowski { 4523da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4524b724e846SDominik Brodowski } 4525b724e846SDominik Brodowski 45263480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 45275590ff0dSUlrich Drepper { 4528da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 45295590ff0dSUlrich Drepper } 45305590ff0dSUlrich Drepper 4531146a8595SJ. Bruce Fields /** 4532146a8595SJ. Bruce Fields * vfs_link - create a new link 4533146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4534abf08576SChristian Brauner * @idmap: idmap of the mount 4535146a8595SJ. Bruce Fields * @dir: new parent 4536146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4537146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4538146a8595SJ. Bruce Fields * 4539146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4540146a8595SJ. Bruce Fields * 4541146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4542146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4543146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4544146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4545146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4546146a8595SJ. Bruce Fields * 4547146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4548146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4549146a8595SJ. Bruce Fields * to be NFS exported. 45506521f891SChristian Brauner * 4551abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4552abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4553abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 45546521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4555abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 4556146a8595SJ. Bruce Fields */ 4557abf08576SChristian Brauner int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap, 45586521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 45596521f891SChristian Brauner struct inode **delegated_inode) 45601da177e4SLinus Torvalds { 45611da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 45628de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 45631da177e4SLinus Torvalds int error; 45641da177e4SLinus Torvalds 45651da177e4SLinus Torvalds if (!inode) 45661da177e4SLinus Torvalds return -ENOENT; 45671da177e4SLinus Torvalds 45684609e1f1SChristian Brauner error = may_create(idmap, dir, new_dentry); 45691da177e4SLinus Torvalds if (error) 45701da177e4SLinus Torvalds return error; 45711da177e4SLinus Torvalds 45721da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 45731da177e4SLinus Torvalds return -EXDEV; 45741da177e4SLinus Torvalds 45751da177e4SLinus Torvalds /* 45761da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 45771da177e4SLinus Torvalds */ 45781da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 45791da177e4SLinus Torvalds return -EPERM; 45800bd23d09SEric W. Biederman /* 45810bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 45820bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 45830bd23d09SEric W. Biederman * the vfs. 45840bd23d09SEric W. Biederman */ 45854609e1f1SChristian Brauner if (HAS_UNMAPPED_ID(idmap, inode)) 45860bd23d09SEric W. Biederman return -EPERM; 4587acfa4380SAl Viro if (!dir->i_op->link) 45881da177e4SLinus Torvalds return -EPERM; 45897e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 45901da177e4SLinus Torvalds return -EPERM; 45911da177e4SLinus Torvalds 45921da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 45931da177e4SLinus Torvalds if (error) 45941da177e4SLinus Torvalds return error; 45951da177e4SLinus Torvalds 45965955102cSAl Viro inode_lock(inode); 4597aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4598f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4599aae8a97dSAneesh Kumar K.V error = -ENOENT; 46008de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 46018de52778SAl Viro error = -EMLINK; 4602146a8595SJ. Bruce Fields else { 4603146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4604146a8595SJ. Bruce Fields if (!error) 46051da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4606146a8595SJ. Bruce Fields } 4607f4e0c30cSAl Viro 4608f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4609f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4610f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4611f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4612f4e0c30cSAl Viro } 46135955102cSAl Viro inode_unlock(inode); 4614e31e14ecSStephen Smalley if (!error) 46157e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 46161da177e4SLinus Torvalds return error; 46171da177e4SLinus Torvalds } 46184d359507SAl Viro EXPORT_SYMBOL(vfs_link); 46191da177e4SLinus Torvalds 46201da177e4SLinus Torvalds /* 46211da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 46221da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 46231da177e4SLinus Torvalds * newname. --KAB 46241da177e4SLinus Torvalds * 46251da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 46261da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 46271da177e4SLinus Torvalds * and other special files. --ADM 46281da177e4SLinus Torvalds */ 4629cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4630020250f3SDmitry Kadashev struct filename *new, int flags) 46311da177e4SLinus Torvalds { 4632abf08576SChristian Brauner struct mnt_idmap *idmap; 46331da177e4SLinus Torvalds struct dentry *new_dentry; 4634dae6ad8fSAl Viro struct path old_path, new_path; 4635146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 463611a7b371SAneesh Kumar K.V int how = 0; 46371da177e4SLinus Torvalds int error; 46381da177e4SLinus Torvalds 4639020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4640020250f3SDmitry Kadashev error = -EINVAL; 4641020250f3SDmitry Kadashev goto out_putnames; 4642020250f3SDmitry Kadashev } 464311a7b371SAneesh Kumar K.V /* 4644f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4645f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4646f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 464711a7b371SAneesh Kumar K.V */ 4648020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4649020250f3SDmitry Kadashev error = -ENOENT; 4650020250f3SDmitry Kadashev goto out_putnames; 4651f0cc6ffbSLinus Torvalds } 4652c04030e1SUlrich Drepper 465311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 465411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4655442e31caSJeff Layton retry: 4656794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 46571da177e4SLinus Torvalds if (error) 4658020250f3SDmitry Kadashev goto out_putnames; 46592ad94ae6SAl Viro 4660b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4661442e31caSJeff Layton (how & LOOKUP_REVAL)); 46621da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 46636902d925SDave Hansen if (IS_ERR(new_dentry)) 4664020250f3SDmitry Kadashev goto out_putpath; 4665dae6ad8fSAl Viro 4666dae6ad8fSAl Viro error = -EXDEV; 4667dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4668dae6ad8fSAl Viro goto out_dput; 4669abf08576SChristian Brauner idmap = mnt_idmap(new_path.mnt); 46704609e1f1SChristian Brauner error = may_linkat(idmap, &old_path); 4671800179c9SKees Cook if (unlikely(error)) 4672800179c9SKees Cook goto out_dput; 4673dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4674be6d3e56SKentaro Takeda if (error) 4675a8104a9fSAl Viro goto out_dput; 4676abf08576SChristian Brauner error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode, 46776521f891SChristian Brauner new_dentry, &delegated_inode); 467875c3f29dSDave Hansen out_dput: 4679921a1650SAl Viro done_path_create(&new_path, new_dentry); 4680146a8595SJ. Bruce Fields if (delegated_inode) { 4681146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4682d22e6338SOleg Drokin if (!error) { 4683d22e6338SOleg Drokin path_put(&old_path); 4684146a8595SJ. Bruce Fields goto retry; 4685146a8595SJ. Bruce Fields } 4686d22e6338SOleg Drokin } 4687442e31caSJeff Layton if (retry_estale(error, how)) { 4688d22e6338SOleg Drokin path_put(&old_path); 4689442e31caSJeff Layton how |= LOOKUP_REVAL; 4690442e31caSJeff Layton goto retry; 4691442e31caSJeff Layton } 4692020250f3SDmitry Kadashev out_putpath: 46932d8f3038SAl Viro path_put(&old_path); 4694020250f3SDmitry Kadashev out_putnames: 4695020250f3SDmitry Kadashev putname(old); 4696020250f3SDmitry Kadashev putname(new); 46971da177e4SLinus Torvalds 46981da177e4SLinus Torvalds return error; 46991da177e4SLinus Torvalds } 47001da177e4SLinus Torvalds 470146ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 470246ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 470346ea89ebSDominik Brodowski { 4704020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4705020250f3SDmitry Kadashev newdfd, getname(newname), flags); 470646ea89ebSDominik Brodowski } 470746ea89ebSDominik Brodowski 47083480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 47095590ff0dSUlrich Drepper { 4710020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 47115590ff0dSUlrich Drepper } 47125590ff0dSUlrich Drepper 4713bc27027aSMiklos Szeredi /** 4714bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 47152111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4716bc27027aSMiklos Szeredi * 4717bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4718bc27027aSMiklos Szeredi * 4719bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4720bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4721bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4722bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4723bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4724bc27027aSMiklos Szeredi * so. 4725bc27027aSMiklos Szeredi * 4726bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4727bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4728bc27027aSMiklos Szeredi * to be NFS exported. 4729bc27027aSMiklos Szeredi * 47301da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 47311da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 47321da177e4SLinus Torvalds * Problems: 47330117d427SMauro Carvalho Chehab * 4734d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 47351da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 47361da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4737a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 47381da177e4SLinus Torvalds * story. 47396cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 474028eceedaSJan Kara * and source. 47411b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 47421da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 47431da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4744a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 47451da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 47461da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 47471da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4748a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 47491da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 47501da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 47511da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4752e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 47531b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 47541da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4755c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 47561da177e4SLinus Torvalds * locking]. 47571da177e4SLinus Torvalds */ 47589fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 47591da177e4SLinus Torvalds { 47601da177e4SLinus Torvalds int error; 47619fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 47629fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 47639fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 47649fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 47659fe61450SChristian Brauner unsigned int flags = rd->flags; 4766bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4767bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4768bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4769da1ce067SMiklos Szeredi bool new_is_dir = false; 4770da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 477149d31c2fSAl Viro struct name_snapshot old_name; 47721da177e4SLinus Torvalds 47738d3e2936SMiklos Szeredi if (source == target) 47741da177e4SLinus Torvalds return 0; 47751da177e4SLinus Torvalds 47764609e1f1SChristian Brauner error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir); 47771da177e4SLinus Torvalds if (error) 47781da177e4SLinus Torvalds return error; 47791da177e4SLinus Torvalds 4780da1ce067SMiklos Szeredi if (!target) { 47814609e1f1SChristian Brauner error = may_create(rd->new_mnt_idmap, new_dir, new_dentry); 4782da1ce067SMiklos Szeredi } else { 4783da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4784da1ce067SMiklos Szeredi 4785da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 47864609e1f1SChristian Brauner error = may_delete(rd->new_mnt_idmap, new_dir, 47876521f891SChristian Brauner new_dentry, is_dir); 4788da1ce067SMiklos Szeredi else 47894609e1f1SChristian Brauner error = may_delete(rd->new_mnt_idmap, new_dir, 47906521f891SChristian Brauner new_dentry, new_is_dir); 4791da1ce067SMiklos Szeredi } 47921da177e4SLinus Torvalds if (error) 47931da177e4SLinus Torvalds return error; 47941da177e4SLinus Torvalds 47952773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 47961da177e4SLinus Torvalds return -EPERM; 47971da177e4SLinus Torvalds 4798bc27027aSMiklos Szeredi /* 4799bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4800bc27027aSMiklos Szeredi * we'll need to flip '..'. 4801bc27027aSMiklos Szeredi */ 4802da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4803da1ce067SMiklos Szeredi if (is_dir) { 48044609e1f1SChristian Brauner error = inode_permission(rd->old_mnt_idmap, source, 480547291baaSChristian Brauner MAY_WRITE); 4806bc27027aSMiklos Szeredi if (error) 4807bc27027aSMiklos Szeredi return error; 4808bc27027aSMiklos Szeredi } 4809da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 48104609e1f1SChristian Brauner error = inode_permission(rd->new_mnt_idmap, target, 481147291baaSChristian Brauner MAY_WRITE); 4812da1ce067SMiklos Szeredi if (error) 4813da1ce067SMiklos Szeredi return error; 4814da1ce067SMiklos Szeredi } 4815da1ce067SMiklos Szeredi } 48160eeca283SRobert Love 48170b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 48180b3974ebSMiklos Szeredi flags); 4819bc27027aSMiklos Szeredi if (error) 4820bc27027aSMiklos Szeredi return error; 4821bc27027aSMiklos Szeredi 482249d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4823bc27027aSMiklos Szeredi dget(new_dentry); 482428eceedaSJan Kara /* 482528eceedaSJan Kara * Lock all moved children. Moved directories may need to change parent 482628eceedaSJan Kara * pointer so they need the lock to prevent against concurrent 482728eceedaSJan Kara * directory changes moving parent pointer. For regular files we've 482828eceedaSJan Kara * historically always done this. The lockdep locking subclasses are 482928eceedaSJan Kara * somewhat arbitrary but RENAME_EXCHANGE in particular can swap 483028eceedaSJan Kara * regular files and directories so it's difficult to tell which 483128eceedaSJan Kara * subclasses to use. 483228eceedaSJan Kara */ 483328eceedaSJan Kara lock_two_inodes(source, target, I_MUTEX_NORMAL, I_MUTEX_NONDIR2); 4834bc27027aSMiklos Szeredi 483551cc3a66SHugh Dickins error = -EPERM; 483651cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 483751cc3a66SHugh Dickins goto out; 483851cc3a66SHugh Dickins 4839bc27027aSMiklos Szeredi error = -EBUSY; 48407af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4841bc27027aSMiklos Szeredi goto out; 4842bc27027aSMiklos Szeredi 4843da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4844bc27027aSMiklos Szeredi error = -EMLINK; 4845da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4846bc27027aSMiklos Szeredi goto out; 4847da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4848da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4849da1ce067SMiklos Szeredi goto out; 4850da1ce067SMiklos Szeredi } 4851da1ce067SMiklos Szeredi if (!is_dir) { 4852bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4853bc27027aSMiklos Szeredi if (error) 4854bc27027aSMiklos Szeredi goto out; 4855da1ce067SMiklos Szeredi } 4856da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4857bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4858bc27027aSMiklos Szeredi if (error) 4859bc27027aSMiklos Szeredi goto out; 4860bc27027aSMiklos Szeredi } 4861e18275aeSChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry, 4862520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4863bc27027aSMiklos Szeredi if (error) 4864bc27027aSMiklos Szeredi goto out; 4865bc27027aSMiklos Szeredi 4866da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 48678767712fSAl Viro if (is_dir) { 48688767712fSAl Viro shrink_dcache_parent(new_dentry); 4869bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 48708767712fSAl Viro } 4871bc27027aSMiklos Szeredi dont_mount(new_dentry); 48728ed936b5SEric W. Biederman detach_mounts(new_dentry); 4873bc27027aSMiklos Szeredi } 4874da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4875da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4876bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4877da1ce067SMiklos Szeredi else 4878da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4879da1ce067SMiklos Szeredi } 4880bc27027aSMiklos Szeredi out: 488128eceedaSJan Kara inode_unlock(source); 488228eceedaSJan Kara if (target) 48835955102cSAl Viro inode_unlock(target); 4884bc27027aSMiklos Szeredi dput(new_dentry); 4885da1ce067SMiklos Szeredi if (!error) { 4886f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4887da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4888da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4889f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4890da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4891da1ce067SMiklos Szeredi } 4892da1ce067SMiklos Szeredi } 489349d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 48940eeca283SRobert Love 48951da177e4SLinus Torvalds return error; 48961da177e4SLinus Torvalds } 48974d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 48981da177e4SLinus Torvalds 4899e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4900e886663cSJens Axboe struct filename *to, unsigned int flags) 49011da177e4SLinus Torvalds { 49029fe61450SChristian Brauner struct renamedata rd; 49031da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 49041da177e4SLinus Torvalds struct dentry *trap; 4905f5beed75SAl Viro struct path old_path, new_path; 4906f5beed75SAl Viro struct qstr old_last, new_last; 4907f5beed75SAl Viro int old_type, new_type; 49088e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4909f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4910c6a94284SJeff Layton bool should_retry = false; 4911e886663cSJens Axboe int error = -EINVAL; 4912520c8b16SMiklos Szeredi 49130d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 49140ee50b47SDmitry Kadashev goto put_names; 4915da1ce067SMiklos Szeredi 49160d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 49170d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 49180ee50b47SDmitry Kadashev goto put_names; 4919520c8b16SMiklos Szeredi 4920f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4921f5beed75SAl Viro target_flags = 0; 4922f5beed75SAl Viro 4923c6a94284SJeff Layton retry: 4924c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4925e886663cSJens Axboe &old_last, &old_type); 49260ee50b47SDmitry Kadashev if (error) 49270ee50b47SDmitry Kadashev goto put_names; 49281da177e4SLinus Torvalds 4929c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4930e886663cSJens Axboe &new_type); 49310ee50b47SDmitry Kadashev if (error) 49321da177e4SLinus Torvalds goto exit1; 49331da177e4SLinus Torvalds 49341da177e4SLinus Torvalds error = -EXDEV; 4935f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 49361da177e4SLinus Torvalds goto exit2; 49371da177e4SLinus Torvalds 49381da177e4SLinus Torvalds error = -EBUSY; 4939f5beed75SAl Viro if (old_type != LAST_NORM) 49401da177e4SLinus Torvalds goto exit2; 49411da177e4SLinus Torvalds 49420a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 49430a7c3937SMiklos Szeredi error = -EEXIST; 4944f5beed75SAl Viro if (new_type != LAST_NORM) 49451da177e4SLinus Torvalds goto exit2; 49461da177e4SLinus Torvalds 4947f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4948c30dabfeSJan Kara if (error) 4949c30dabfeSJan Kara goto exit2; 4950c30dabfeSJan Kara 49518e6d782cSJ. Bruce Fields retry_deleg: 4952f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 49531da177e4SLinus Torvalds 495474d7970fSNamjae Jeon old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry, 495574d7970fSNamjae Jeon lookup_flags); 49561da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 49571da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 49581da177e4SLinus Torvalds goto exit3; 49591da177e4SLinus Torvalds /* source must exist */ 49601da177e4SLinus Torvalds error = -ENOENT; 4961b18825a7SDavid Howells if (d_is_negative(old_dentry)) 49621da177e4SLinus Torvalds goto exit4; 496374d7970fSNamjae Jeon new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry, 496474d7970fSNamjae Jeon lookup_flags | target_flags); 49651da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 49661da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 49671da177e4SLinus Torvalds goto exit4; 49680a7c3937SMiklos Szeredi error = -EEXIST; 49690a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 49700a7c3937SMiklos Szeredi goto exit5; 4971da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4972da1ce067SMiklos Szeredi error = -ENOENT; 4973da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4974da1ce067SMiklos Szeredi goto exit5; 4975da1ce067SMiklos Szeredi 4976da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4977da1ce067SMiklos Szeredi error = -ENOTDIR; 4978f5beed75SAl Viro if (new_last.name[new_last.len]) 4979da1ce067SMiklos Szeredi goto exit5; 4980da1ce067SMiklos Szeredi } 4981da1ce067SMiklos Szeredi } 49820a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 49830a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 49840a7c3937SMiklos Szeredi error = -ENOTDIR; 4985f5beed75SAl Viro if (old_last.name[old_last.len]) 49860a7c3937SMiklos Szeredi goto exit5; 4987f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 49880a7c3937SMiklos Szeredi goto exit5; 49890a7c3937SMiklos Szeredi } 49900a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 49910a7c3937SMiklos Szeredi error = -EINVAL; 49920a7c3937SMiklos Szeredi if (old_dentry == trap) 49930a7c3937SMiklos Szeredi goto exit5; 49941da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4995da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 49961da177e4SLinus Torvalds error = -ENOTEMPTY; 49971da177e4SLinus Torvalds if (new_dentry == trap) 49981da177e4SLinus Torvalds goto exit5; 49991da177e4SLinus Torvalds 5000f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 5001f5beed75SAl Viro &new_path, new_dentry, flags); 5002be6d3e56SKentaro Takeda if (error) 5003c30dabfeSJan Kara goto exit5; 50049fe61450SChristian Brauner 50059fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 50069fe61450SChristian Brauner rd.old_dentry = old_dentry; 5007abf08576SChristian Brauner rd.old_mnt_idmap = mnt_idmap(old_path.mnt); 50089fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 50099fe61450SChristian Brauner rd.new_dentry = new_dentry; 5010abf08576SChristian Brauner rd.new_mnt_idmap = mnt_idmap(new_path.mnt); 50119fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 50129fe61450SChristian Brauner rd.flags = flags; 50139fe61450SChristian Brauner error = vfs_rename(&rd); 50141da177e4SLinus Torvalds exit5: 50151da177e4SLinus Torvalds dput(new_dentry); 50161da177e4SLinus Torvalds exit4: 50171da177e4SLinus Torvalds dput(old_dentry); 50181da177e4SLinus Torvalds exit3: 5019f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 50208e6d782cSJ. Bruce Fields if (delegated_inode) { 50218e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 50228e6d782cSJ. Bruce Fields if (!error) 50238e6d782cSJ. Bruce Fields goto retry_deleg; 50248e6d782cSJ. Bruce Fields } 5025f5beed75SAl Viro mnt_drop_write(old_path.mnt); 50261da177e4SLinus Torvalds exit2: 5027c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 5028c6a94284SJeff Layton should_retry = true; 5029f5beed75SAl Viro path_put(&new_path); 50301da177e4SLinus Torvalds exit1: 5031f5beed75SAl Viro path_put(&old_path); 5032c6a94284SJeff Layton if (should_retry) { 5033c6a94284SJeff Layton should_retry = false; 5034c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 5035c6a94284SJeff Layton goto retry; 5036c6a94284SJeff Layton } 50370ee50b47SDmitry Kadashev put_names: 5038e886663cSJens Axboe putname(from); 5039e886663cSJens Axboe putname(to); 50401da177e4SLinus Torvalds return error; 50411da177e4SLinus Torvalds } 50421da177e4SLinus Torvalds 5043ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 5044ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 5045ee81feb6SDominik Brodowski { 5046e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 5047e886663cSJens Axboe flags); 5048ee81feb6SDominik Brodowski } 5049ee81feb6SDominik Brodowski 5050520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 5051520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 5052520c8b16SMiklos Szeredi { 5053e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 5054e886663cSJens Axboe 0); 5055520c8b16SMiklos Szeredi } 5056520c8b16SMiklos Szeredi 5057a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 50585590ff0dSUlrich Drepper { 5059e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 5060e886663cSJens Axboe getname(newname), 0); 50615590ff0dSUlrich Drepper } 50625590ff0dSUlrich Drepper 50635d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 50641da177e4SLinus Torvalds { 50655d826c84SAl Viro int len = PTR_ERR(link); 50661da177e4SLinus Torvalds if (IS_ERR(link)) 50671da177e4SLinus Torvalds goto out; 50681da177e4SLinus Torvalds 50691da177e4SLinus Torvalds len = strlen(link); 50701da177e4SLinus Torvalds if (len > (unsigned) buflen) 50711da177e4SLinus Torvalds len = buflen; 50721da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 50731da177e4SLinus Torvalds len = -EFAULT; 50741da177e4SLinus Torvalds out: 50751da177e4SLinus Torvalds return len; 50761da177e4SLinus Torvalds } 50771da177e4SLinus Torvalds 5078d60874cdSMiklos Szeredi /** 5079fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 5080fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 5081fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 5082fd4a0edfSMiklos Szeredi * @buflen: size of buffer 5083fd4a0edfSMiklos Szeredi * 5084fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 5085fd4a0edfSMiklos Szeredi * 5086fd4a0edfSMiklos Szeredi * Does not call security hook. 5087fd4a0edfSMiklos Szeredi */ 5088fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5089fd4a0edfSMiklos Szeredi { 5090fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 5091f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 5092f2df5da6SAl Viro const char *link; 5093f2df5da6SAl Viro int res; 5094fd4a0edfSMiklos Szeredi 509576fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 509676fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 509776fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 509876fca90eSMiklos Szeredi 509976fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 5100fd4a0edfSMiklos Szeredi return -EINVAL; 5101fd4a0edfSMiklos Szeredi 510276fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 510376fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 510476fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 510576fca90eSMiklos Szeredi } 510676fca90eSMiklos Szeredi 51074c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 5108f2df5da6SAl Viro if (!link) { 5109f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 5110f2df5da6SAl Viro if (IS_ERR(link)) 5111f2df5da6SAl Viro return PTR_ERR(link); 5112f2df5da6SAl Viro } 5113f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 5114f2df5da6SAl Viro do_delayed_call(&done); 5115f2df5da6SAl Viro return res; 5116fd4a0edfSMiklos Szeredi } 5117fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 51181da177e4SLinus Torvalds 5119d60874cdSMiklos Szeredi /** 5120d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 5121d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 5122d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 5123d60874cdSMiklos Szeredi * 5124d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 5125d60874cdSMiklos Szeredi * 5126d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 5127d60874cdSMiklos Szeredi * 5128d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 5129d60874cdSMiklos Szeredi */ 5130d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 5131d60874cdSMiklos Szeredi { 5132d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 5133d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 5134d60874cdSMiklos Szeredi 5135d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 5136d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 5137d60874cdSMiklos Szeredi if (!res) 5138d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 5139d60874cdSMiklos Szeredi } 5140d60874cdSMiklos Szeredi return res; 5141d60874cdSMiklos Szeredi } 5142d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 5143d60874cdSMiklos Szeredi 51441da177e4SLinus Torvalds /* get the link contents into pagecache */ 51456b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 5146fceef393SAl Viro struct delayed_call *callback) 51471da177e4SLinus Torvalds { 5148ebd09abbSDuane Griffin char *kaddr; 51491da177e4SLinus Torvalds struct page *page; 51506b255391SAl Viro struct address_space *mapping = inode->i_mapping; 51516b255391SAl Viro 5152d3883d4fSAl Viro if (!dentry) { 5153d3883d4fSAl Viro page = find_get_page(mapping, 0); 5154d3883d4fSAl Viro if (!page) 51556b255391SAl Viro return ERR_PTR(-ECHILD); 5156d3883d4fSAl Viro if (!PageUptodate(page)) { 5157d3883d4fSAl Viro put_page(page); 5158d3883d4fSAl Viro return ERR_PTR(-ECHILD); 5159d3883d4fSAl Viro } 5160d3883d4fSAl Viro } else { 5161090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 51621da177e4SLinus Torvalds if (IS_ERR(page)) 51636fe6900eSNick Piggin return (char*)page; 5164d3883d4fSAl Viro } 5165fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 516621fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 516721fc61c7SAl Viro kaddr = page_address(page); 51686b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 5169ebd09abbSDuane Griffin return kaddr; 51701da177e4SLinus Torvalds } 51711da177e4SLinus Torvalds 51726b255391SAl Viro EXPORT_SYMBOL(page_get_link); 51731da177e4SLinus Torvalds 5174fceef393SAl Viro void page_put_link(void *arg) 51751da177e4SLinus Torvalds { 5176fceef393SAl Viro put_page(arg); 51771da177e4SLinus Torvalds } 51784d359507SAl Viro EXPORT_SYMBOL(page_put_link); 51791da177e4SLinus Torvalds 5180aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5181aa80deabSAl Viro { 5182fceef393SAl Viro DEFINE_DELAYED_CALL(done); 51836b255391SAl Viro int res = readlink_copy(buffer, buflen, 51846b255391SAl Viro page_get_link(dentry, d_inode(dentry), 5185fceef393SAl Viro &done)); 5186fceef393SAl Viro do_delayed_call(&done); 5187aa80deabSAl Viro return res; 5188aa80deabSAl Viro } 5189aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 5190aa80deabSAl Viro 519156f5746cSMatthew Wilcox (Oracle) int page_symlink(struct inode *inode, const char *symname, int len) 51921da177e4SLinus Torvalds { 51931da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 519427a77913SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 519556f5746cSMatthew Wilcox (Oracle) bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS); 51960adb25d2SKirill Korotaev struct page *page; 51971468c6f4SAlexander Potapenko void *fsdata = NULL; 5198beb497abSDmitriy Monakhov int err; 51992d878178SMatthew Wilcox (Oracle) unsigned int flags; 52001da177e4SLinus Torvalds 52017e53cac4SNeilBrown retry: 52022d878178SMatthew Wilcox (Oracle) if (nofs) 52032d878178SMatthew Wilcox (Oracle) flags = memalloc_nofs_save(); 520427a77913SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata); 52052d878178SMatthew Wilcox (Oracle) if (nofs) 52062d878178SMatthew Wilcox (Oracle) memalloc_nofs_restore(flags); 52071da177e4SLinus Torvalds if (err) 5208afddba49SNick Piggin goto fail; 5209afddba49SNick Piggin 521021fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5211afddba49SNick Piggin 521227a77913SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, 0, len-1, len-1, 5213afddba49SNick Piggin page, fsdata); 52141da177e4SLinus Torvalds if (err < 0) 52151da177e4SLinus Torvalds goto fail; 5216afddba49SNick Piggin if (err < len-1) 5217afddba49SNick Piggin goto retry; 5218afddba49SNick Piggin 52191da177e4SLinus Torvalds mark_inode_dirty(inode); 52201da177e4SLinus Torvalds return 0; 52211da177e4SLinus Torvalds fail: 52221da177e4SLinus Torvalds return err; 52231da177e4SLinus Torvalds } 52244d359507SAl Viro EXPORT_SYMBOL(page_symlink); 52250adb25d2SKirill Korotaev 522692e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 52276b255391SAl Viro .get_link = page_get_link, 52281da177e4SLinus Torvalds }; 52291da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5230