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 19194b5d262SAl Viro 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; 25255422d0bSPaul Moore result->refcnt = 1; 253fd3522fdSPaul Moore audit_getname(result); 254c4ad8f98SLinus Torvalds 255c4ad8f98SLinus Torvalds return result; 256c4ad8f98SLinus Torvalds } 257*74d7970fSNamjae 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 26455422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 26555422d0bSPaul Moore 26655422d0bSPaul Moore if (--name->refcnt > 0) 26755422d0bSPaul Moore return; 26855422d0bSPaul Moore 269fd2f7cb5SAl Viro if (name->name != name->iname) { 27055422d0bSPaul Moore __putname(name->name); 27155422d0bSPaul Moore kfree(name); 27255422d0bSPaul Moore } else 27355422d0bSPaul Moore __putname(name); 2741da177e4SLinus Torvalds } 275*74d7970fSNamjae Jeon EXPORT_SYMBOL(putname); 2761da177e4SLinus Torvalds 27747291baaSChristian Brauner /** 27847291baaSChristian Brauner * check_acl - perform ACL permission checking 279700b7940SChristian Brauner * @idmap: idmap of the mount the inode was found from 28047291baaSChristian Brauner * @inode: inode to check permissions on 28147291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 28247291baaSChristian Brauner * 28347291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 28447291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 28547291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 28647291baaSChristian Brauner * 287700b7940SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 288700b7940SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 289700b7940SChristian Brauner * care to map the inode according to @idmap before checking permissions. 29047291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 291700b7940SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 29247291baaSChristian Brauner */ 293700b7940SChristian Brauner static int check_acl(struct mnt_idmap *idmap, 29447291baaSChristian Brauner struct inode *inode, int mask) 295e77819e5SLinus Torvalds { 29684635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 297e77819e5SLinus Torvalds struct posix_acl *acl; 298e77819e5SLinus Torvalds 299e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 3003567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 3013567866bSAl Viro if (!acl) 302e77819e5SLinus Torvalds return -EAGAIN; 303cac2f8b8SChristian Brauner /* no ->get_inode_acl() calls in RCU mode... */ 304b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 305e77819e5SLinus Torvalds return -ECHILD; 306700b7940SChristian Brauner return posix_acl_permission(idmap, inode, acl, mask); 307e77819e5SLinus Torvalds } 308e77819e5SLinus Torvalds 309cac2f8b8SChristian Brauner acl = get_inode_acl(inode, ACL_TYPE_ACCESS); 3104e34e719SChristoph Hellwig if (IS_ERR(acl)) 3114e34e719SChristoph Hellwig return PTR_ERR(acl); 312e77819e5SLinus Torvalds if (acl) { 313700b7940SChristian Brauner int error = posix_acl_permission(idmap, inode, acl, mask); 314e77819e5SLinus Torvalds posix_acl_release(acl); 315e77819e5SLinus Torvalds return error; 316e77819e5SLinus Torvalds } 31784635d68SLinus Torvalds #endif 318e77819e5SLinus Torvalds 319e77819e5SLinus Torvalds return -EAGAIN; 320e77819e5SLinus Torvalds } 321e77819e5SLinus Torvalds 32247291baaSChristian Brauner /** 32347291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 324700b7940SChristian Brauner * @idmap: idmap of the mount the inode was found from 32547291baaSChristian Brauner * @inode: inode to check permissions on 32647291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3275fc475b7SLinus Torvalds * 32847291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 32947291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 33047291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 33147291baaSChristian Brauner * 332700b7940SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 333700b7940SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 334700b7940SChristian Brauner * care to map the inode according to @idmap before checking permissions. 33547291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 336700b7940SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 3375909ccaaSLinus Torvalds */ 338700b7940SChristian Brauner static int acl_permission_check(struct mnt_idmap *idmap, 33947291baaSChristian Brauner struct inode *inode, int mask) 3405909ccaaSLinus Torvalds { 34126cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 342a2bd096fSChristian Brauner vfsuid_t vfsuid; 3435909ccaaSLinus Torvalds 3445fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 345e67fe633SChristian Brauner vfsuid = i_uid_into_vfsuid(idmap, inode); 346a2bd096fSChristian Brauner if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) { 3475fc475b7SLinus Torvalds mask &= 7; 3485909ccaaSLinus Torvalds mode >>= 6; 3495fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3505fc475b7SLinus Torvalds } 3515fc475b7SLinus Torvalds 3525fc475b7SLinus Torvalds /* Do we have ACL's? */ 353e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 354700b7940SChristian Brauner int error = check_acl(idmap, inode, mask); 3555909ccaaSLinus Torvalds if (error != -EAGAIN) 3565909ccaaSLinus Torvalds return error; 3575909ccaaSLinus Torvalds } 3585909ccaaSLinus Torvalds 3595fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3605fc475b7SLinus Torvalds mask &= 7; 3615fc475b7SLinus Torvalds 3625fc475b7SLinus Torvalds /* 3635fc475b7SLinus Torvalds * Are the group permissions different from 3645fc475b7SLinus Torvalds * the other permissions in the bits we care 3655fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3665fc475b7SLinus Torvalds */ 3675fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 368e67fe633SChristian Brauner vfsgid_t vfsgid = i_gid_into_vfsgid(idmap, inode); 369a2bd096fSChristian Brauner if (vfsgid_in_group_p(vfsgid)) 3705909ccaaSLinus Torvalds mode >>= 3; 3715909ccaaSLinus Torvalds } 3725909ccaaSLinus Torvalds 3735fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3745fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3755909ccaaSLinus Torvalds } 3761da177e4SLinus Torvalds 3771da177e4SLinus Torvalds /** 3781da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3794609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 3801da177e4SLinus Torvalds * @inode: inode to check access rights for 3815fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3825fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3831da177e4SLinus Torvalds * 3841da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3851da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3861da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 387b74c79e9SNick Piggin * are used for other things. 388b74c79e9SNick Piggin * 389b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 390b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 391b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 39247291baaSChristian Brauner * 3934609e1f1SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3944609e1f1SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3954609e1f1SChristian Brauner * care to map the inode according to @idmap before checking permissions. 39647291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3974609e1f1SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 3981da177e4SLinus Torvalds */ 3994609e1f1SChristian Brauner int generic_permission(struct mnt_idmap *idmap, struct inode *inode, 40047291baaSChristian Brauner int mask) 4011da177e4SLinus Torvalds { 4025909ccaaSLinus Torvalds int ret; 4031da177e4SLinus Torvalds 4041da177e4SLinus Torvalds /* 405948409c7SAndreas Gruenbacher * Do the basic permission checks. 4061da177e4SLinus Torvalds */ 407700b7940SChristian Brauner ret = acl_permission_check(idmap, inode, mask); 4085909ccaaSLinus Torvalds if (ret != -EACCES) 4095909ccaaSLinus Torvalds return ret; 4101da177e4SLinus Torvalds 411d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 412d594e7ecSAl Viro /* DACs are overridable for directories */ 413d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 4149452e93eSChristian Brauner if (capable_wrt_inode_uidgid(idmap, inode, 41523adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 416d594e7ecSAl Viro return 0; 4179452e93eSChristian Brauner if (capable_wrt_inode_uidgid(idmap, inode, 4180558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4191da177e4SLinus Torvalds return 0; 4202a4c2242SStephen Smalley return -EACCES; 4212a4c2242SStephen Smalley } 4221da177e4SLinus Torvalds 4231da177e4SLinus Torvalds /* 4241da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4251da177e4SLinus Torvalds */ 4267ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 427d594e7ecSAl Viro if (mask == MAY_READ) 4289452e93eSChristian Brauner if (capable_wrt_inode_uidgid(idmap, inode, 4290558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4301da177e4SLinus Torvalds return 0; 4312a4c2242SStephen Smalley /* 4322a4c2242SStephen Smalley * Read/write DACs are always overridable. 4332a4c2242SStephen Smalley * Executable DACs are overridable when there is 4342a4c2242SStephen Smalley * at least one exec bit set. 4352a4c2242SStephen Smalley */ 4362a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 4379452e93eSChristian Brauner if (capable_wrt_inode_uidgid(idmap, inode, 4380558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4392a4c2242SStephen Smalley return 0; 4401da177e4SLinus Torvalds 4411da177e4SLinus Torvalds return -EACCES; 4421da177e4SLinus Torvalds } 4434d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4441da177e4SLinus Torvalds 44547291baaSChristian Brauner /** 44647291baaSChristian Brauner * do_inode_permission - UNIX permission checking 4474609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 44847291baaSChristian Brauner * @inode: inode to check permissions on 44947291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 45047291baaSChristian Brauner * 4513ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4523ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4533ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4543ddcd056SLinus Torvalds * permission function, use the fast case". 4553ddcd056SLinus Torvalds */ 4564609e1f1SChristian Brauner static inline int do_inode_permission(struct mnt_idmap *idmap, 45747291baaSChristian Brauner struct inode *inode, int mask) 4583ddcd056SLinus Torvalds { 4593ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4603ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 4614609e1f1SChristian Brauner return inode->i_op->permission(idmap, inode, mask); 4623ddcd056SLinus Torvalds 4633ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4643ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4653ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4663ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4673ddcd056SLinus Torvalds } 4684609e1f1SChristian Brauner return generic_permission(idmap, inode, mask); 4693ddcd056SLinus Torvalds } 4703ddcd056SLinus Torvalds 471cb23beb5SChristoph Hellwig /** 4720bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4730bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 47455852635SRandy Dunlap * @inode: Inode to check permission on 4750bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4760bdaea90SDavid Howells * 4770bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4780bdaea90SDavid Howells */ 4790bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4800bdaea90SDavid Howells { 4810bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4820bdaea90SDavid Howells umode_t mode = inode->i_mode; 4830bdaea90SDavid Howells 4840bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 485bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4860bdaea90SDavid Howells return -EROFS; 4870bdaea90SDavid Howells } 4880bdaea90SDavid Howells return 0; 4890bdaea90SDavid Howells } 4900bdaea90SDavid Howells 4910bdaea90SDavid Howells /** 4920bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4934609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 4940bdaea90SDavid Howells * @inode: Inode to check permission on 4950bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4960bdaea90SDavid Howells * 4970bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4980bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4990bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 5000bdaea90SDavid Howells * 5010bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 5020bdaea90SDavid Howells */ 5034609e1f1SChristian Brauner int inode_permission(struct mnt_idmap *idmap, 50447291baaSChristian Brauner struct inode *inode, int mask) 5050bdaea90SDavid Howells { 5060bdaea90SDavid Howells int retval; 5070bdaea90SDavid Howells 5080bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 5090bdaea90SDavid Howells if (retval) 5100bdaea90SDavid Howells return retval; 5114bfd054aSEric Biggers 5124bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 5134bfd054aSEric Biggers /* 5144bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5154bfd054aSEric Biggers */ 5164bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5174bfd054aSEric Biggers return -EPERM; 5184bfd054aSEric Biggers 5194bfd054aSEric Biggers /* 5204bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5214bfd054aSEric Biggers * written back improperly if their true value is unknown 5224bfd054aSEric Biggers * to the vfs. 5234bfd054aSEric Biggers */ 5244609e1f1SChristian Brauner if (HAS_UNMAPPED_ID(idmap, inode)) 5254bfd054aSEric Biggers return -EACCES; 5264bfd054aSEric Biggers } 5274bfd054aSEric Biggers 5284609e1f1SChristian Brauner retval = do_inode_permission(idmap, inode, mask); 5294bfd054aSEric Biggers if (retval) 5304bfd054aSEric Biggers return retval; 5314bfd054aSEric Biggers 5324bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5334bfd054aSEric Biggers if (retval) 5344bfd054aSEric Biggers return retval; 5354bfd054aSEric Biggers 5364bfd054aSEric Biggers return security_inode_permission(inode, mask); 5370bdaea90SDavid Howells } 5384d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5390bdaea90SDavid Howells 5400bdaea90SDavid Howells /** 5415dd784d0SJan Blunck * path_get - get a reference to a path 5425dd784d0SJan Blunck * @path: path to get the reference to 5435dd784d0SJan Blunck * 5445dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5455dd784d0SJan Blunck */ 546dcf787f3SAl Viro void path_get(const struct path *path) 5475dd784d0SJan Blunck { 5485dd784d0SJan Blunck mntget(path->mnt); 5495dd784d0SJan Blunck dget(path->dentry); 5505dd784d0SJan Blunck } 5515dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5525dd784d0SJan Blunck 5535dd784d0SJan Blunck /** 5541d957f9bSJan Blunck * path_put - put a reference to a path 5551d957f9bSJan Blunck * @path: path to put the reference to 5561d957f9bSJan Blunck * 5571d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5581d957f9bSJan Blunck */ 559dcf787f3SAl Viro void path_put(const struct path *path) 5601da177e4SLinus Torvalds { 5611d957f9bSJan Blunck dput(path->dentry); 5621d957f9bSJan Blunck mntput(path->mnt); 5631da177e4SLinus Torvalds } 5641d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5651da177e4SLinus Torvalds 566894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5671f55a6ecSAl Viro struct nameidata { 5681f55a6ecSAl Viro struct path path; 5691f55a6ecSAl Viro struct qstr last; 5701f55a6ecSAl Viro struct path root; 5711f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 572bcba1e7dSAl Viro unsigned int flags, state; 57303fa86e9SAl Viro unsigned seq, next_seq, m_seq, r_seq; 5741f55a6ecSAl Viro int last_type; 5751f55a6ecSAl Viro unsigned depth; 576756daf26SNeilBrown int total_link_count; 577697fc6caSAl Viro struct saved { 578697fc6caSAl Viro struct path link; 579fceef393SAl Viro struct delayed_call done; 580697fc6caSAl Viro const char *name; 5810450b2d1SAl Viro unsigned seq; 582894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5839883d185SAl Viro struct filename *name; 5849883d185SAl Viro struct nameidata *saved; 5859883d185SAl Viro unsigned root_seq; 5869883d185SAl Viro int dfd; 587a2bd096fSChristian Brauner vfsuid_t dir_vfsuid; 5880f705953SAl Viro umode_t dir_mode; 5893859a271SKees Cook } __randomize_layout; 5901f55a6ecSAl Viro 591bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 592bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 593bcba1e7dSAl Viro #define ND_JUMPED 4 594bcba1e7dSAl Viro 59506422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 596894bc8c4SAl Viro { 597756daf26SNeilBrown struct nameidata *old = current->nameidata; 598756daf26SNeilBrown p->stack = p->internal; 5997962c7d1SAl Viro p->depth = 0; 600c8a53ee5SAl Viro p->dfd = dfd; 601c8a53ee5SAl Viro p->name = name; 6027d01ef75SAl Viro p->path.mnt = NULL; 6037d01ef75SAl Viro p->path.dentry = NULL; 604756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 6059883d185SAl Viro p->saved = old; 606756daf26SNeilBrown current->nameidata = p; 607894bc8c4SAl Viro } 608894bc8c4SAl Viro 60906422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 61006422964SAl Viro const struct path *root) 61106422964SAl Viro { 61206422964SAl Viro __set_nameidata(p, dfd, name); 61306422964SAl Viro p->state = 0; 61406422964SAl Viro if (unlikely(root)) { 61506422964SAl Viro p->state = ND_ROOT_PRESET; 61606422964SAl Viro p->root = *root; 61706422964SAl Viro } 61806422964SAl Viro } 61906422964SAl Viro 6209883d185SAl Viro static void restore_nameidata(void) 621894bc8c4SAl Viro { 6229883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 623756daf26SNeilBrown 624756daf26SNeilBrown current->nameidata = old; 625756daf26SNeilBrown if (old) 626756daf26SNeilBrown old->total_link_count = now->total_link_count; 627e1a63bbcSAl Viro if (now->stack != now->internal) 628756daf26SNeilBrown kfree(now->stack); 629894bc8c4SAl Viro } 630894bc8c4SAl Viro 63160ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 632894bc8c4SAl Viro { 633bc40aee0SAl Viro struct saved *p; 634bc40aee0SAl Viro 6356da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 63660ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 637bc40aee0SAl Viro if (unlikely(!p)) 63860ef60c7SAl Viro return false; 639894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 640894bc8c4SAl Viro nd->stack = p; 64160ef60c7SAl Viro return true; 642894bc8c4SAl Viro } 643894bc8c4SAl Viro 644397d425dSEric W. Biederman /** 6456b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 646397d425dSEric W. Biederman * 647397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 648397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 649397d425dSEric W. Biederman */ 6506b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 651397d425dSEric W. Biederman { 65295dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 653397d425dSEric W. Biederman 654402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 655402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 656397d425dSEric W. Biederman return true; 657397d425dSEric W. Biederman 6586b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 659397d425dSEric W. Biederman } 660397d425dSEric W. Biederman 6617973387aSAl Viro static void drop_links(struct nameidata *nd) 6627973387aSAl Viro { 6637973387aSAl Viro int i = nd->depth; 6647973387aSAl Viro while (i--) { 6657973387aSAl Viro struct saved *last = nd->stack + i; 666fceef393SAl Viro do_delayed_call(&last->done); 667fceef393SAl Viro clear_delayed_call(&last->done); 6687973387aSAl Viro } 6697973387aSAl Viro } 6707973387aSAl Viro 6716e180327SAl Viro static void leave_rcu(struct nameidata *nd) 6726e180327SAl Viro { 6736e180327SAl Viro nd->flags &= ~LOOKUP_RCU; 67403fa86e9SAl Viro nd->seq = nd->next_seq = 0; 6756e180327SAl Viro rcu_read_unlock(); 6766e180327SAl Viro } 6776e180327SAl Viro 6787973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6797973387aSAl Viro { 6807973387aSAl Viro drop_links(nd); 6817973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6827973387aSAl Viro int i; 6837973387aSAl Viro path_put(&nd->path); 6847973387aSAl Viro for (i = 0; i < nd->depth; i++) 6857973387aSAl Viro path_put(&nd->stack[i].link); 686bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 687102b8af2SAl Viro path_put(&nd->root); 688bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 689102b8af2SAl Viro } 6907973387aSAl Viro } else { 6916e180327SAl Viro leave_rcu(nd); 6927973387aSAl Viro } 6937973387aSAl Viro nd->depth = 0; 6947d01ef75SAl Viro nd->path.mnt = NULL; 6957d01ef75SAl Viro nd->path.dentry = NULL; 6967973387aSAl Viro } 6977973387aSAl Viro 6987973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6992aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 7007973387aSAl Viro { 7012aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 7027973387aSAl Viro if (unlikely(res)) { 7037973387aSAl Viro if (res > 0) 7047973387aSAl Viro path->mnt = NULL; 7057973387aSAl Viro path->dentry = NULL; 7067973387aSAl Viro return false; 7077973387aSAl Viro } 7087973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 7097973387aSAl Viro path->dentry = NULL; 7107973387aSAl Viro return false; 7117973387aSAl Viro } 7127973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7137973387aSAl Viro } 7147973387aSAl Viro 7152aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7162aa38470SAl Viro struct path *path, unsigned seq) 7172aa38470SAl Viro { 7185bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7192aa38470SAl Viro } 7202aa38470SAl Viro 7217973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7227973387aSAl Viro { 7237973387aSAl Viro int i; 724eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 725eacd9aa8SAl Viro drop_links(nd); 726eacd9aa8SAl Viro nd->depth = 0; 727eacd9aa8SAl Viro return false; 728eacd9aa8SAl Viro } 7297973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7307973387aSAl Viro struct saved *last = nd->stack + i; 7317973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7327973387aSAl Viro drop_links(nd); 7337973387aSAl Viro nd->depth = i + 1; 7347973387aSAl Viro return false; 7357973387aSAl Viro } 7367973387aSAl Viro } 7377973387aSAl Viro return true; 7387973387aSAl Viro } 7397973387aSAl Viro 740ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 741ee594bffSAl Viro { 742adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 743bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 744ee594bffSAl Viro return true; 745bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 746ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 747ee594bffSAl Viro } 748ee594bffSAl Viro 74919660af7SAl Viro /* 75031e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 75119660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 75219660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 75357e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 75419660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 75519660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 75619660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 75719660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 75831e6b01fSNick Piggin */ 75931e6b01fSNick Piggin 76031e6b01fSNick Piggin /** 761e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 76219660af7SAl Viro * @nd: nameidata pathwalk data 763e36cffedSJens Axboe * Returns: true on success, false on failure 76431e6b01fSNick Piggin * 765e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7664675ac39SAl Viro * for ref-walk mode. 7674675ac39SAl Viro * Must be called from rcu-walk context. 768e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7697973387aSAl Viro * terminate_walk(). 77031e6b01fSNick Piggin */ 771e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 77231e6b01fSNick Piggin { 77331e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 77431e6b01fSNick Piggin 77531e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 776e5c832d5SLinus Torvalds 7777973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7784675ac39SAl Viro goto out1; 77984a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 78084a2bd39SAl Viro goto out; 781ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7824675ac39SAl Viro goto out; 7836e180327SAl Viro leave_rcu(nd); 7844675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 785e36cffedSJens Axboe return true; 7864675ac39SAl Viro 78784a2bd39SAl Viro out1: 7884675ac39SAl Viro nd->path.mnt = NULL; 7894675ac39SAl Viro nd->path.dentry = NULL; 7904675ac39SAl Viro out: 7916e180327SAl Viro leave_rcu(nd); 792e36cffedSJens Axboe return false; 7934675ac39SAl Viro } 7944675ac39SAl Viro 7954675ac39SAl Viro /** 796ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7974675ac39SAl Viro * @nd: nameidata pathwalk data 798ae66db45SAl Viro * @dentry: next dentry to step into 799ae66db45SAl Viro * Returns: true on success, false on failure 8004675ac39SAl Viro * 80130476f7eSTom Rix * Similar to try_to_unlazy(), but here we have the next dentry already 802ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 803ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 804ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 8054675ac39SAl Viro * terminate_walk(). 8064675ac39SAl Viro */ 80703fa86e9SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry) 8084675ac39SAl Viro { 8097e4745a0SAl Viro int res; 8104675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8114675ac39SAl Viro 8124675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8134675ac39SAl Viro goto out2; 8147e4745a0SAl Viro res = __legitimize_mnt(nd->path.mnt, nd->m_seq); 8157e4745a0SAl Viro if (unlikely(res)) { 8167e4745a0SAl Viro if (res > 0) 8177973387aSAl Viro goto out2; 8187e4745a0SAl Viro goto out1; 8197e4745a0SAl Viro } 8204675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8217973387aSAl Viro goto out1; 82248a066e7SAl Viro 82315570086SLinus Torvalds /* 8244675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8254675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8264675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8274675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8284675ac39SAl Viro * know the parent must still be valid if the child sequence number is 82915570086SLinus Torvalds */ 8304675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 831e5c832d5SLinus Torvalds goto out; 83203fa86e9SAl Viro if (read_seqcount_retry(&dentry->d_seq, nd->next_seq)) 83384a2bd39SAl Viro goto out_dput; 834e5c832d5SLinus Torvalds /* 835e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 836e5c832d5SLinus Torvalds * still valid and get it if required. 837e5c832d5SLinus Torvalds */ 83884a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 83984a2bd39SAl Viro goto out_dput; 8406e180327SAl Viro leave_rcu(nd); 841ae66db45SAl Viro return true; 84219660af7SAl Viro 8437973387aSAl Viro out2: 8447973387aSAl Viro nd->path.mnt = NULL; 8457973387aSAl Viro out1: 8467973387aSAl Viro nd->path.dentry = NULL; 847e5c832d5SLinus Torvalds out: 8486e180327SAl Viro leave_rcu(nd); 849ae66db45SAl Viro return false; 85084a2bd39SAl Viro out_dput: 8516e180327SAl Viro leave_rcu(nd); 85284a2bd39SAl Viro dput(dentry); 853ae66db45SAl Viro return false; 85431e6b01fSNick Piggin } 85531e6b01fSNick Piggin 8564ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 85734286d66SNick Piggin { 858a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8594ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 860a89f8337SAl Viro else 861a89f8337SAl Viro return 1; 86234286d66SNick Piggin } 86334286d66SNick Piggin 8649f1fafeeSAl Viro /** 8659f1fafeeSAl Viro * complete_walk - successful completion of path walk 8669f1fafeeSAl Viro * @nd: pointer nameidata 86739159de2SJeff Layton * 8689f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8699f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8709f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8719f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8729f1fafeeSAl Viro * need to drop nd->path. 87339159de2SJeff Layton */ 8749f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 87539159de2SJeff Layton { 87616c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 87739159de2SJeff Layton int status; 87839159de2SJeff Layton 8799f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 880adb21d2bSAleksa Sarai /* 881adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 882adb21d2bSAleksa Sarai * externally-managed nd->root. 883adb21d2bSAleksa Sarai */ 884bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 885bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8869f1fafeeSAl Viro nd->root.mnt = NULL; 8876c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 888e36cffedSJens Axboe if (!try_to_unlazy(nd)) 88948a066e7SAl Viro return -ECHILD; 89048a066e7SAl Viro } 8919f1fafeeSAl Viro 892adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 893adb21d2bSAleksa Sarai /* 894adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 895adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 896adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 897adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 898adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 899adb21d2bSAleksa Sarai * 900adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 901adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 902adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 903adb21d2bSAleksa Sarai * requested root to userspace. 904adb21d2bSAleksa Sarai * 905adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 906adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 907adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 908adb21d2bSAleksa Sarai */ 909adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 910adb21d2bSAleksa Sarai return -EXDEV; 911adb21d2bSAleksa Sarai } 912adb21d2bSAleksa Sarai 913bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 91439159de2SJeff Layton return 0; 91539159de2SJeff Layton 916ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 91716c2cd71SAl Viro return 0; 91816c2cd71SAl Viro 919ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 92039159de2SJeff Layton if (status > 0) 92139159de2SJeff Layton return 0; 92239159de2SJeff Layton 92316c2cd71SAl Viro if (!status) 92439159de2SJeff Layton status = -ESTALE; 92516c2cd71SAl Viro 92639159de2SJeff Layton return status; 92739159de2SJeff Layton } 92839159de2SJeff Layton 929740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9302a737871SAl Viro { 93131e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9329e6697e2SAl Viro 933adb21d2bSAleksa Sarai /* 934adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 935adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 936adb21d2bSAleksa Sarai * from the dirfd. 937adb21d2bSAleksa Sarai */ 938adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 939adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 940adb21d2bSAleksa Sarai 9419e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9428f47a016SAl Viro unsigned seq; 943c28cc364SNick Piggin 944c28cc364SNick Piggin do { 945c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 94631e6b01fSNick Piggin nd->root = fs->root; 9478f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 948c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9499e6697e2SAl Viro } else { 9509e6697e2SAl Viro get_fs_root(fs, &nd->root); 951bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9529e6697e2SAl Viro } 953740a1678SAleksa Sarai return 0; 95431e6b01fSNick Piggin } 95531e6b01fSNick Piggin 956248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 957248fb5b9SAl Viro { 958adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 959adb21d2bSAleksa Sarai return -EXDEV; 96072ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 96172ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 96272ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 96372ba2929SAleksa Sarai return -EXDEV; 96472ba2929SAleksa Sarai } 965740a1678SAleksa Sarai if (!nd->root.mnt) { 966740a1678SAleksa Sarai int error = set_root(nd); 967740a1678SAleksa Sarai if (error) 968740a1678SAleksa Sarai return error; 969740a1678SAleksa Sarai } 970248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 971248fb5b9SAl Viro struct dentry *d; 972248fb5b9SAl Viro nd->path = nd->root; 973248fb5b9SAl Viro d = nd->path.dentry; 974248fb5b9SAl Viro nd->inode = d->d_inode; 975248fb5b9SAl Viro nd->seq = nd->root_seq; 97682ef0698SAl Viro if (read_seqcount_retry(&d->d_seq, nd->seq)) 977248fb5b9SAl Viro return -ECHILD; 978248fb5b9SAl Viro } else { 979248fb5b9SAl Viro path_put(&nd->path); 980248fb5b9SAl Viro nd->path = nd->root; 981248fb5b9SAl Viro path_get(&nd->path); 982248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 983248fb5b9SAl Viro } 984bcba1e7dSAl Viro nd->state |= ND_JUMPED; 985248fb5b9SAl Viro return 0; 986248fb5b9SAl Viro } 987248fb5b9SAl Viro 988b5fb63c1SChristoph Hellwig /* 9896b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 990b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 991b5fb63c1SChristoph Hellwig */ 992ea4af4aaSAl Viro int nd_jump_link(const struct path *path) 993b5fb63c1SChristoph Hellwig { 9944b99d499SAleksa Sarai int error = -ELOOP; 9956e77137bSAl Viro struct nameidata *nd = current->nameidata; 996b5fb63c1SChristoph Hellwig 9974b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9984b99d499SAleksa Sarai goto err; 9994b99d499SAleksa Sarai 100072ba2929SAleksa Sarai error = -EXDEV; 100172ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 100272ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 100372ba2929SAleksa Sarai goto err; 100472ba2929SAleksa Sarai } 1005adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 1006adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 1007adb21d2bSAleksa Sarai goto err; 100872ba2929SAleksa Sarai 10094b99d499SAleksa Sarai path_put(&nd->path); 1010b5fb63c1SChristoph Hellwig nd->path = *path; 1011b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1012bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10131bc82070SAleksa Sarai return 0; 10144b99d499SAleksa Sarai 10154b99d499SAleksa Sarai err: 10164b99d499SAleksa Sarai path_put(path); 10174b99d499SAleksa Sarai return error; 1018b5fb63c1SChristoph Hellwig } 1019b5fb63c1SChristoph Hellwig 1020b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1021574197e0SAl Viro { 102221c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1023fceef393SAl Viro do_delayed_call(&last->done); 10246548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1025b9ff4429SAl Viro path_put(&last->link); 1026574197e0SAl Viro } 1027574197e0SAl Viro 10289c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10299c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10309c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10319c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10329c011be1SLuis Chamberlain 10339c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10349c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10359c011be1SLuis Chamberlain { 10369c011be1SLuis Chamberlain .procname = "protected_symlinks", 10379c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10389c011be1SLuis Chamberlain .maxlen = sizeof(int), 1039c7031c14SJulius Hemanth Pitti .mode = 0644, 10409c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10419c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10429c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10439c011be1SLuis Chamberlain }, 10449c011be1SLuis Chamberlain { 10459c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10469c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10479c011be1SLuis Chamberlain .maxlen = sizeof(int), 1048c7031c14SJulius Hemanth Pitti .mode = 0644, 10499c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10509c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10519c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10529c011be1SLuis Chamberlain }, 10539c011be1SLuis Chamberlain { 10549c011be1SLuis Chamberlain .procname = "protected_fifos", 10559c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10569c011be1SLuis Chamberlain .maxlen = sizeof(int), 1057c7031c14SJulius Hemanth Pitti .mode = 0644, 10589c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10599c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10609c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10619c011be1SLuis Chamberlain }, 10629c011be1SLuis Chamberlain { 10639c011be1SLuis Chamberlain .procname = "protected_regular", 10649c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10659c011be1SLuis Chamberlain .maxlen = sizeof(int), 1066c7031c14SJulius Hemanth Pitti .mode = 0644, 10679c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10689c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10699c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10709c011be1SLuis Chamberlain }, 10719c011be1SLuis Chamberlain { } 10729c011be1SLuis Chamberlain }; 10739c011be1SLuis Chamberlain 10749c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10759c011be1SLuis Chamberlain { 10769c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10779c011be1SLuis Chamberlain return 0; 10789c011be1SLuis Chamberlain } 10799c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10809c011be1SLuis Chamberlain 10819c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1082800179c9SKees Cook 1083800179c9SKees Cook /** 1084800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 108555852635SRandy Dunlap * @nd: nameidata pathwalk data 1086800179c9SKees Cook * 1087800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1088800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1089800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1090800179c9SKees Cook * processes from failing races against path names that may change out 1091800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1092800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1093800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1094800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1095800179c9SKees Cook * 1096800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1097800179c9SKees Cook */ 1098ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1099800179c9SKees Cook { 1100e67fe633SChristian Brauner struct mnt_idmap *idmap; 1101a2bd096fSChristian Brauner vfsuid_t vfsuid; 1102ba73d987SChristian Brauner 1103800179c9SKees Cook if (!sysctl_protected_symlinks) 1104800179c9SKees Cook return 0; 1105800179c9SKees Cook 1106e67fe633SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 1107e67fe633SChristian Brauner vfsuid = i_uid_into_vfsuid(idmap, inode); 1108800179c9SKees Cook /* Allowed if owner and follower match. */ 1109a2bd096fSChristian Brauner if (vfsuid_eq_kuid(vfsuid, current_fsuid())) 1110800179c9SKees Cook return 0; 1111800179c9SKees Cook 1112800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11130f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1114800179c9SKees Cook return 0; 1115800179c9SKees Cook 1116800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1117a2bd096fSChristian Brauner if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid)) 1118800179c9SKees Cook return 0; 1119800179c9SKees Cook 112031956502SAl Viro if (nd->flags & LOOKUP_RCU) 112131956502SAl Viro return -ECHILD; 112231956502SAl Viro 1123ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1124245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1125800179c9SKees Cook return -EACCES; 1126800179c9SKees Cook } 1127800179c9SKees Cook 1128800179c9SKees Cook /** 1129800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 11304609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 1131800179c9SKees Cook * @inode: the source inode to hardlink from 1132800179c9SKees Cook * 1133800179c9SKees Cook * Return false if at least one of the following conditions: 1134800179c9SKees Cook * - inode is not a regular file 1135800179c9SKees Cook * - inode is setuid 1136800179c9SKees Cook * - inode is setgid and group-exec 1137800179c9SKees Cook * - access failure for read and write 1138800179c9SKees Cook * 1139800179c9SKees Cook * Otherwise returns true. 1140800179c9SKees Cook */ 11414609e1f1SChristian Brauner static bool safe_hardlink_source(struct mnt_idmap *idmap, 1142ba73d987SChristian Brauner struct inode *inode) 1143800179c9SKees Cook { 1144800179c9SKees Cook umode_t mode = inode->i_mode; 1145800179c9SKees Cook 1146800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1147800179c9SKees Cook if (!S_ISREG(mode)) 1148800179c9SKees Cook return false; 1149800179c9SKees Cook 1150800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1151800179c9SKees Cook if (mode & S_ISUID) 1152800179c9SKees Cook return false; 1153800179c9SKees Cook 1154800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1155800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1156800179c9SKees Cook return false; 1157800179c9SKees Cook 1158800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 11594609e1f1SChristian Brauner if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE)) 1160800179c9SKees Cook return false; 1161800179c9SKees Cook 1162800179c9SKees Cook return true; 1163800179c9SKees Cook } 1164800179c9SKees Cook 1165800179c9SKees Cook /** 1166800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 11674609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 1168800179c9SKees Cook * @link: the source to hardlink from 1169800179c9SKees Cook * 1170800179c9SKees Cook * Block hardlink when all of: 1171800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1172800179c9SKees Cook * - fsuid does not match inode 1173800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1174f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1175800179c9SKees Cook * 11764609e1f1SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 11774609e1f1SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 11784609e1f1SChristian Brauner * care to map the inode according to @idmap before checking permissions. 1179ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 11804609e1f1SChristian Brauner * raw inode simply pass @nop_mnt_idmap. 1181ba73d987SChristian Brauner * 1182800179c9SKees Cook * Returns 0 if successful, -ve on error. 1183800179c9SKees Cook */ 11844609e1f1SChristian Brauner int may_linkat(struct mnt_idmap *idmap, const struct path *link) 1185800179c9SKees Cook { 1186593d1ce8SEric W. Biederman struct inode *inode = link->dentry->d_inode; 1187593d1ce8SEric W. Biederman 1188593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 1189e67fe633SChristian Brauner if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) || 1190e67fe633SChristian Brauner !vfsgid_valid(i_gid_into_vfsgid(idmap, inode))) 1191593d1ce8SEric W. Biederman return -EOVERFLOW; 1192800179c9SKees Cook 1193800179c9SKees Cook if (!sysctl_protected_hardlinks) 1194800179c9SKees Cook return 0; 1195800179c9SKees Cook 1196800179c9SKees Cook /* Source inode owner (or CAP_FOWNER) can hardlink all they like, 1197800179c9SKees Cook * otherwise, it must be a safe source. 1198800179c9SKees Cook */ 11994609e1f1SChristian Brauner if (safe_hardlink_source(idmap, inode) || 120001beba79SChristian Brauner inode_owner_or_capable(idmap, inode)) 1201800179c9SKees Cook return 0; 1202800179c9SKees Cook 1203245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "linkat"); 1204800179c9SKees Cook return -EPERM; 1205800179c9SKees Cook } 1206800179c9SKees Cook 120730aba665SSalvatore Mesoraca /** 120830aba665SSalvatore Mesoraca * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory 120930aba665SSalvatore Mesoraca * should be allowed, or not, on files that already 121030aba665SSalvatore Mesoraca * exist. 1211e67fe633SChristian Brauner * @idmap: idmap of the mount the inode was found from 12122111c3c0SRandy Dunlap * @nd: nameidata pathwalk data 121330aba665SSalvatore Mesoraca * @inode: the inode of the file to open 121430aba665SSalvatore Mesoraca * 121530aba665SSalvatore Mesoraca * Block an O_CREAT open of a FIFO (or a regular file) when: 121630aba665SSalvatore Mesoraca * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled 121730aba665SSalvatore Mesoraca * - the file already exists 121830aba665SSalvatore Mesoraca * - we are in a sticky directory 121930aba665SSalvatore Mesoraca * - we don't own the file 122030aba665SSalvatore Mesoraca * - the owner of the directory doesn't own the file 122130aba665SSalvatore Mesoraca * - the directory is world writable 122230aba665SSalvatore Mesoraca * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2 122330aba665SSalvatore Mesoraca * the directory doesn't have to be world writable: being group writable will 122430aba665SSalvatore Mesoraca * be enough. 122530aba665SSalvatore Mesoraca * 1226e67fe633SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 1227e67fe633SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 1228e67fe633SChristian Brauner * care to map the inode according to @idmap before checking permissions. 1229ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1230e67fe633SChristian Brauner * raw inode simply pass @nop_mnt_idmap. 1231ba73d987SChristian Brauner * 123230aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 123330aba665SSalvatore Mesoraca */ 1234e67fe633SChristian Brauner static int may_create_in_sticky(struct mnt_idmap *idmap, 1235ba73d987SChristian Brauner struct nameidata *nd, struct inode *const inode) 123630aba665SSalvatore Mesoraca { 1237ba73d987SChristian Brauner umode_t dir_mode = nd->dir_mode; 1238a2bd096fSChristian Brauner vfsuid_t dir_vfsuid = nd->dir_vfsuid; 1239ba73d987SChristian Brauner 124030aba665SSalvatore Mesoraca if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) || 124130aba665SSalvatore Mesoraca (!sysctl_protected_regular && S_ISREG(inode->i_mode)) || 1242d0cb5018SAl Viro likely(!(dir_mode & S_ISVTX)) || 1243e67fe633SChristian Brauner vfsuid_eq(i_uid_into_vfsuid(idmap, inode), dir_vfsuid) || 1244e67fe633SChristian Brauner vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), current_fsuid())) 124530aba665SSalvatore Mesoraca return 0; 124630aba665SSalvatore Mesoraca 1247d0cb5018SAl Viro if (likely(dir_mode & 0002) || 1248d0cb5018SAl Viro (dir_mode & 0020 && 124930aba665SSalvatore Mesoraca ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) || 125030aba665SSalvatore Mesoraca (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) { 1251245d7369SKees Cook const char *operation = S_ISFIFO(inode->i_mode) ? 1252245d7369SKees Cook "sticky_create_fifo" : 1253245d7369SKees Cook "sticky_create_regular"; 1254245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_CREAT, operation); 125530aba665SSalvatore Mesoraca return -EACCES; 125630aba665SSalvatore Mesoraca } 125730aba665SSalvatore Mesoraca return 0; 125830aba665SSalvatore Mesoraca } 125930aba665SSalvatore Mesoraca 1260f015f126SDavid Howells /* 1261f015f126SDavid Howells * follow_up - Find the mountpoint of path's vfsmount 1262f015f126SDavid Howells * 1263f015f126SDavid Howells * Given a path, find the mountpoint of its source file system. 1264f015f126SDavid Howells * Replace @path with the path of the mountpoint in the parent mount. 1265f015f126SDavid Howells * Up is towards /. 1266f015f126SDavid Howells * 1267f015f126SDavid Howells * Return 1 if we went up a level and 0 if we were already at the 1268f015f126SDavid Howells * root. 1269f015f126SDavid Howells */ 1270bab77ebfSAl Viro int follow_up(struct path *path) 12711da177e4SLinus Torvalds { 12720714a533SAl Viro struct mount *mnt = real_mount(path->mnt); 12730714a533SAl Viro struct mount *parent; 12741da177e4SLinus Torvalds struct dentry *mountpoint; 127599b7db7bSNick Piggin 127648a066e7SAl Viro read_seqlock_excl(&mount_lock); 12770714a533SAl Viro parent = mnt->mnt_parent; 12783c0a6163SAl Viro if (parent == mnt) { 127948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 12801da177e4SLinus Torvalds return 0; 12811da177e4SLinus Torvalds } 12820714a533SAl Viro mntget(&parent->mnt); 1283a73324daSAl Viro mountpoint = dget(mnt->mnt_mountpoint); 128448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 1285bab77ebfSAl Viro dput(path->dentry); 1286bab77ebfSAl Viro path->dentry = mountpoint; 1287bab77ebfSAl Viro mntput(path->mnt); 12880714a533SAl Viro path->mnt = &parent->mnt; 12891da177e4SLinus Torvalds return 1; 12901da177e4SLinus Torvalds } 12914d359507SAl Viro EXPORT_SYMBOL(follow_up); 12921da177e4SLinus Torvalds 12937ef482faSAl Viro static bool choose_mountpoint_rcu(struct mount *m, const struct path *root, 12947ef482faSAl Viro struct path *path, unsigned *seqp) 12957ef482faSAl Viro { 12967ef482faSAl Viro while (mnt_has_parent(m)) { 12977ef482faSAl Viro struct dentry *mountpoint = m->mnt_mountpoint; 12987ef482faSAl Viro 12997ef482faSAl Viro m = m->mnt_parent; 13007ef482faSAl Viro if (unlikely(root->dentry == mountpoint && 13017ef482faSAl Viro root->mnt == &m->mnt)) 13027ef482faSAl Viro break; 13037ef482faSAl Viro if (mountpoint != m->mnt.mnt_root) { 13047ef482faSAl Viro path->mnt = &m->mnt; 13057ef482faSAl Viro path->dentry = mountpoint; 13067ef482faSAl Viro *seqp = read_seqcount_begin(&mountpoint->d_seq); 13077ef482faSAl Viro return true; 13087ef482faSAl Viro } 13097ef482faSAl Viro } 13107ef482faSAl Viro return false; 13117ef482faSAl Viro } 13127ef482faSAl Viro 13132aa38470SAl Viro static bool choose_mountpoint(struct mount *m, const struct path *root, 13142aa38470SAl Viro struct path *path) 13152aa38470SAl Viro { 13162aa38470SAl Viro bool found; 13172aa38470SAl Viro 13182aa38470SAl Viro rcu_read_lock(); 13192aa38470SAl Viro while (1) { 13202aa38470SAl Viro unsigned seq, mseq = read_seqbegin(&mount_lock); 13212aa38470SAl Viro 13222aa38470SAl Viro found = choose_mountpoint_rcu(m, root, path, &seq); 13232aa38470SAl Viro if (unlikely(!found)) { 13242aa38470SAl Viro if (!read_seqretry(&mount_lock, mseq)) 13252aa38470SAl Viro break; 13262aa38470SAl Viro } else { 13272aa38470SAl Viro if (likely(__legitimize_path(path, seq, mseq))) 13282aa38470SAl Viro break; 13292aa38470SAl Viro rcu_read_unlock(); 13302aa38470SAl Viro path_put(path); 13312aa38470SAl Viro rcu_read_lock(); 13322aa38470SAl Viro } 13332aa38470SAl Viro } 13342aa38470SAl Viro rcu_read_unlock(); 13352aa38470SAl Viro return found; 13362aa38470SAl Viro } 13372aa38470SAl Viro 1338b5c84bf6SNick Piggin /* 13399875cf80SDavid Howells * Perform an automount 13409875cf80SDavid Howells * - return -EISDIR to tell follow_managed() to stop and return the path we 13419875cf80SDavid Howells * were called with. 13421da177e4SLinus Torvalds */ 13431c9f5e06SAl Viro static int follow_automount(struct path *path, int *count, unsigned lookup_flags) 134431e6b01fSNick Piggin { 134525e195aaSAl Viro struct dentry *dentry = path->dentry; 13469875cf80SDavid Howells 13470ec26fd0SMiklos Szeredi /* We don't want to mount if someone's just doing a stat - 13480ec26fd0SMiklos Szeredi * unless they're stat'ing a directory and appended a '/' to 13490ec26fd0SMiklos Szeredi * the name. 13500ec26fd0SMiklos Szeredi * 13510ec26fd0SMiklos Szeredi * We do, however, want to mount if someone wants to open or 13525a30d8a2SDavid Howells * create a file of any type under the mountpoint, wants to 13530ec26fd0SMiklos Szeredi * traverse through the mountpoint or wants to open the 13540ec26fd0SMiklos Szeredi * mounted directory. Also, autofs may mark negative dentries 13550ec26fd0SMiklos Szeredi * as being automount points. These will need the attentions 13560ec26fd0SMiklos Szeredi * of the daemon to instantiate them before they can be used. 13575a30d8a2SDavid Howells */ 13581c9f5e06SAl Viro if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY | 13595d38f049SIan Kent LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) && 136025e195aaSAl Viro dentry->d_inode) 13619875cf80SDavid Howells return -EISDIR; 13620ec26fd0SMiklos Szeredi 13631c9f5e06SAl Viro if (count && (*count)++ >= MAXSYMLINKS) 13649875cf80SDavid Howells return -ELOOP; 13659875cf80SDavid Howells 136625e195aaSAl Viro return finish_automount(dentry->d_op->d_automount(path), path); 1367ea5b778aSDavid Howells } 13689875cf80SDavid Howells 13699875cf80SDavid Howells /* 13709deed3ebSAl Viro * mount traversal - out-of-line part. One note on ->d_flags accesses - 13719deed3ebSAl Viro * dentries are pinned but not locked here, so negative dentry can go 13729deed3ebSAl Viro * positive right under us. Use of smp_load_acquire() provides a barrier 13739deed3ebSAl Viro * sufficient for ->d_inode and ->d_flags consistency. 13749875cf80SDavid Howells */ 13759deed3ebSAl Viro static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped, 13769deed3ebSAl Viro int *count, unsigned lookup_flags) 13779875cf80SDavid Howells { 13789deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 13799875cf80SDavid Howells bool need_mntput = false; 13808aef1884SAl Viro int ret = 0; 13819875cf80SDavid Howells 13829deed3ebSAl Viro while (flags & DCACHE_MANAGED_DENTRY) { 1383cc53ce53SDavid Howells /* Allow the filesystem to manage the transit without i_mutex 1384cc53ce53SDavid Howells * being held. */ 1385d41efb52SAl Viro if (flags & DCACHE_MANAGE_TRANSIT) { 1386fb5f51c7SIan Kent ret = path->dentry->d_op->d_manage(path, false); 1387508c8772SAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 1388cc53ce53SDavid Howells if (ret < 0) 13898aef1884SAl Viro break; 1390cc53ce53SDavid Howells } 1391cc53ce53SDavid Howells 13929deed3ebSAl Viro if (flags & DCACHE_MOUNTED) { // something's mounted on it.. 13939875cf80SDavid Howells struct vfsmount *mounted = lookup_mnt(path); 13949deed3ebSAl Viro if (mounted) { // ... in our namespace 13959875cf80SDavid Howells dput(path->dentry); 13969875cf80SDavid Howells if (need_mntput) 1397463ffb2eSAl Viro mntput(path->mnt); 1398463ffb2eSAl Viro path->mnt = mounted; 1399463ffb2eSAl Viro path->dentry = dget(mounted->mnt_root); 14009deed3ebSAl Viro // here we know it's positive 14019deed3ebSAl Viro flags = path->dentry->d_flags; 14029875cf80SDavid Howells need_mntput = true; 14039875cf80SDavid Howells continue; 1404463ffb2eSAl Viro } 14051da177e4SLinus Torvalds } 14069875cf80SDavid Howells 14079deed3ebSAl Viro if (!(flags & DCACHE_NEED_AUTOMOUNT)) 14089deed3ebSAl Viro break; 14099deed3ebSAl Viro 14109deed3ebSAl Viro // uncovered automount point 14119deed3ebSAl Viro ret = follow_automount(path, count, lookup_flags); 14129deed3ebSAl Viro flags = smp_load_acquire(&path->dentry->d_flags); 14139875cf80SDavid Howells if (ret < 0) 14148aef1884SAl Viro break; 14159875cf80SDavid Howells } 14169875cf80SDavid Howells 14179deed3ebSAl Viro if (ret == -EISDIR) 14189deed3ebSAl Viro ret = 0; 14199deed3ebSAl Viro // possible if you race with several mount --move 14209deed3ebSAl Viro if (need_mntput && path->mnt == mnt) 14218aef1884SAl Viro mntput(path->mnt); 14229deed3ebSAl Viro if (!ret && unlikely(d_flags_negative(flags))) 1423d41efb52SAl Viro ret = -ENOENT; 14249deed3ebSAl Viro *jumped = need_mntput; 14258402752eSAl Viro return ret; 14261da177e4SLinus Torvalds } 14271da177e4SLinus Torvalds 14289deed3ebSAl Viro static inline int traverse_mounts(struct path *path, bool *jumped, 14299deed3ebSAl Viro int *count, unsigned lookup_flags) 14309deed3ebSAl Viro { 14319deed3ebSAl Viro unsigned flags = smp_load_acquire(&path->dentry->d_flags); 14329deed3ebSAl Viro 14339deed3ebSAl Viro /* fastpath */ 14349deed3ebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) { 14359deed3ebSAl Viro *jumped = false; 14369deed3ebSAl Viro if (unlikely(d_flags_negative(flags))) 14379deed3ebSAl Viro return -ENOENT; 14389deed3ebSAl Viro return 0; 14399deed3ebSAl Viro } 14409deed3ebSAl Viro return __traverse_mounts(path, flags, jumped, count, lookup_flags); 14419deed3ebSAl Viro } 14429deed3ebSAl Viro 1443cc53ce53SDavid Howells int follow_down_one(struct path *path) 14441da177e4SLinus Torvalds { 14451da177e4SLinus Torvalds struct vfsmount *mounted; 14461da177e4SLinus Torvalds 14471c755af4SAl Viro mounted = lookup_mnt(path); 14481da177e4SLinus Torvalds if (mounted) { 14499393bd07SAl Viro dput(path->dentry); 14509393bd07SAl Viro mntput(path->mnt); 14519393bd07SAl Viro path->mnt = mounted; 14529393bd07SAl Viro path->dentry = dget(mounted->mnt_root); 14531da177e4SLinus Torvalds return 1; 14541da177e4SLinus Torvalds } 14551da177e4SLinus Torvalds return 0; 14561da177e4SLinus Torvalds } 14574d359507SAl Viro EXPORT_SYMBOL(follow_down_one); 14581da177e4SLinus Torvalds 14599875cf80SDavid Howells /* 14609deed3ebSAl Viro * Follow down to the covering mount currently visible to userspace. At each 14619deed3ebSAl Viro * point, the filesystem owning that dentry may be queried as to whether the 14629deed3ebSAl Viro * caller is permitted to proceed or not. 14639deed3ebSAl Viro */ 1464e1f19857SRichard Weinberger int follow_down(struct path *path, unsigned int flags) 14659deed3ebSAl Viro { 14669deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14679deed3ebSAl Viro bool jumped; 1468e1f19857SRichard Weinberger int ret = traverse_mounts(path, &jumped, NULL, flags); 14699deed3ebSAl Viro 14709deed3ebSAl Viro if (path->mnt != mnt) 14719deed3ebSAl Viro mntput(mnt); 14729deed3ebSAl Viro return ret; 14739deed3ebSAl Viro } 14749deed3ebSAl Viro EXPORT_SYMBOL(follow_down); 14759deed3ebSAl Viro 14769deed3ebSAl Viro /* 1477287548e4SAl Viro * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if 1478287548e4SAl Viro * we meet a managed dentry that would need blocking. 14799875cf80SDavid Howells */ 14803bd8bc89SAl Viro static bool __follow_mount_rcu(struct nameidata *nd, struct path *path) 14819875cf80SDavid Howells { 1482ea936aebSAl Viro struct dentry *dentry = path->dentry; 1483ea936aebSAl Viro unsigned int flags = dentry->d_flags; 1484ea936aebSAl Viro 1485ea936aebSAl Viro if (likely(!(flags & DCACHE_MANAGED_DENTRY))) 1486ea936aebSAl Viro return true; 1487ea936aebSAl Viro 1488ea936aebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1489ea936aebSAl Viro return false; 1490ea936aebSAl Viro 149162a7375eSIan Kent for (;;) { 149262a7375eSIan Kent /* 149362a7375eSIan Kent * Don't forget we might have a non-mountpoint managed dentry 149462a7375eSIan Kent * that wants to block transit. 149562a7375eSIan Kent */ 1496ea936aebSAl Viro if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) { 1497ea936aebSAl Viro int res = dentry->d_op->d_manage(path, true); 1498ea936aebSAl Viro if (res) 1499ea936aebSAl Viro return res == -EISDIR; 1500ea936aebSAl Viro flags = dentry->d_flags; 1501b8faf035SNeilBrown } 150262a7375eSIan Kent 1503ea936aebSAl Viro if (flags & DCACHE_MOUNTED) { 1504ea936aebSAl Viro struct mount *mounted = __lookup_mnt(path->mnt, dentry); 1505ea936aebSAl Viro if (mounted) { 1506c7105365SAl Viro path->mnt = &mounted->mnt; 1507ea936aebSAl Viro dentry = path->dentry = mounted->mnt.mnt_root; 1508bcba1e7dSAl Viro nd->state |= ND_JUMPED; 150903fa86e9SAl Viro nd->next_seq = read_seqcount_begin(&dentry->d_seq); 1510ea936aebSAl Viro flags = dentry->d_flags; 151103fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach 151203fa86e9SAl Viro // this state. 151320aac6c6SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 151420aac6c6SAl Viro return false; 1515ea936aebSAl Viro continue; 15169875cf80SDavid Howells } 1517ea936aebSAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1518ea936aebSAl Viro return false; 1519ea936aebSAl Viro } 1520ea936aebSAl Viro return !(flags & DCACHE_NEED_AUTOMOUNT); 1521ea936aebSAl Viro } 1522287548e4SAl Viro } 1523287548e4SAl Viro 1524db3c9adeSAl Viro static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry, 15253bd8bc89SAl Viro struct path *path) 1526bd7c4b50SAl Viro { 15279deed3ebSAl Viro bool jumped; 1528db3c9adeSAl Viro int ret; 1529bd7c4b50SAl Viro 1530db3c9adeSAl Viro path->mnt = nd->path.mnt; 1531db3c9adeSAl Viro path->dentry = dentry; 1532c153007bSAl Viro if (nd->flags & LOOKUP_RCU) { 153303fa86e9SAl Viro unsigned int seq = nd->next_seq; 15343bd8bc89SAl Viro if (likely(__follow_mount_rcu(nd, path))) 15359deed3ebSAl Viro return 0; 153603fa86e9SAl Viro // *path and nd->next_seq might've been clobbered 1537c153007bSAl Viro path->mnt = nd->path.mnt; 1538c153007bSAl Viro path->dentry = dentry; 153903fa86e9SAl Viro nd->next_seq = seq; 154003fa86e9SAl Viro if (!try_to_unlazy_next(nd, dentry)) 154103fa86e9SAl Viro return -ECHILD; 1542c153007bSAl Viro } 15439deed3ebSAl Viro ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags); 15449deed3ebSAl Viro if (jumped) { 15459deed3ebSAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 15469deed3ebSAl Viro ret = -EXDEV; 15479deed3ebSAl Viro else 1548bcba1e7dSAl Viro nd->state |= ND_JUMPED; 15499deed3ebSAl Viro } 15509deed3ebSAl Viro if (unlikely(ret)) { 15519deed3ebSAl Viro dput(path->dentry); 15529deed3ebSAl Viro if (path->mnt != nd->path.mnt) 15539deed3ebSAl Viro mntput(path->mnt); 1554bd7c4b50SAl Viro } 1555bd7c4b50SAl Viro return ret; 1556bd7c4b50SAl Viro } 1557bd7c4b50SAl Viro 15589875cf80SDavid Howells /* 1559f4fdace9SOleg Drokin * This looks up the name in dcache and possibly revalidates the found dentry. 1560f4fdace9SOleg Drokin * NULL is returned if the dentry does not exist in the cache. 1561baa03890SNick Piggin */ 1562e3c13928SAl Viro static struct dentry *lookup_dcache(const struct qstr *name, 1563e3c13928SAl Viro struct dentry *dir, 15646c51e513SAl Viro unsigned int flags) 1565baa03890SNick Piggin { 1566a89f8337SAl Viro struct dentry *dentry = d_lookup(dir, name); 1567bad61189SMiklos Szeredi if (dentry) { 1568a89f8337SAl Viro int error = d_revalidate(dentry, flags); 1569bad61189SMiklos Szeredi if (unlikely(error <= 0)) { 157074ff0ffcSAl Viro if (!error) 15715542aa2fSEric W. Biederman d_invalidate(dentry); 1572bad61189SMiklos Szeredi dput(dentry); 157374ff0ffcSAl Viro return ERR_PTR(error); 1574bad61189SMiklos Szeredi } 1575bad61189SMiklos Szeredi } 1576baa03890SNick Piggin return dentry; 1577baa03890SNick Piggin } 1578baa03890SNick Piggin 1579baa03890SNick Piggin /* 1580a03ece5fSAl Viro * Parent directory has inode locked exclusive. This is one 1581a03ece5fSAl Viro * and only case when ->lookup() gets called on non in-lookup 1582a03ece5fSAl Viro * dentries - as the matter of fact, this only gets called 1583a03ece5fSAl Viro * when directory is guaranteed to have no in-lookup children 1584a03ece5fSAl Viro * at all. 158544396f4bSJosef Bacik */ 1586*74d7970fSNamjae Jeon struct dentry *lookup_one_qstr_excl(const struct qstr *name, 1587*74d7970fSNamjae Jeon struct dentry *base, 1588*74d7970fSNamjae Jeon unsigned int flags) 158944396f4bSJosef Bacik { 1590a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 159144396f4bSJosef Bacik struct dentry *old; 1592a03ece5fSAl Viro struct inode *dir = base->d_inode; 1593a03ece5fSAl Viro 1594a03ece5fSAl Viro if (dentry) 1595a03ece5fSAl Viro return dentry; 159644396f4bSJosef Bacik 159744396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1598a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 159944396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1600a03ece5fSAl Viro 1601a03ece5fSAl Viro dentry = d_alloc(base, name); 1602a03ece5fSAl Viro if (unlikely(!dentry)) 1603a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 160444396f4bSJosef Bacik 160572bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 160644396f4bSJosef Bacik if (unlikely(old)) { 160744396f4bSJosef Bacik dput(dentry); 160844396f4bSJosef Bacik dentry = old; 160944396f4bSJosef Bacik } 161044396f4bSJosef Bacik return dentry; 161144396f4bSJosef Bacik } 1612*74d7970fSNamjae Jeon EXPORT_SYMBOL(lookup_one_qstr_excl); 161344396f4bSJosef Bacik 16144cb64024SAl Viro static struct dentry *lookup_fast(struct nameidata *nd) 16151da177e4SLinus Torvalds { 161631e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16175a18fff2SAl Viro int status = 1; 16189875cf80SDavid Howells 16193cac260aSAl Viro /* 1620b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16215d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16225d0f49c1SAl Viro * going to fall back to non-racy lookup. 1623b04f784eSNick Piggin */ 162431e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 162503fa86e9SAl Viro dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq); 16265d0f49c1SAl Viro if (unlikely(!dentry)) { 1627e36cffedSJens Axboe if (!try_to_unlazy(nd)) 162820e34357SAl Viro return ERR_PTR(-ECHILD); 162920e34357SAl Viro return NULL; 16305d0f49c1SAl Viro } 16315a18fff2SAl Viro 163212f8ad4bSLinus Torvalds /* 163312f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 163412f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 163512f8ad4bSLinus Torvalds */ 16364cb64024SAl Viro if (read_seqcount_retry(&parent->d_seq, nd->seq)) 163720e34357SAl Viro return ERR_PTR(-ECHILD); 16385a18fff2SAl Viro 16394ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1640c153007bSAl Viro if (likely(status > 0)) 164120e34357SAl Viro return dentry; 164203fa86e9SAl Viro if (!try_to_unlazy_next(nd, dentry)) 164320e34357SAl Viro return ERR_PTR(-ECHILD); 164426ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1645209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1646209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16475a18fff2SAl Viro } else { 1648e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 164981e6f520SAl Viro if (unlikely(!dentry)) 165020e34357SAl Viro return NULL; 16514ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16525d0f49c1SAl Viro } 16535a18fff2SAl Viro if (unlikely(status <= 0)) { 1654e9742b53SAl Viro if (!status) 16555d0f49c1SAl Viro d_invalidate(dentry); 16565a18fff2SAl Viro dput(dentry); 165720e34357SAl Viro return ERR_PTR(status); 16585a18fff2SAl Viro } 165920e34357SAl Viro return dentry; 1660697f514dSMiklos Szeredi } 1661697f514dSMiklos Szeredi 1662697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 166388d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1664e3c13928SAl Viro struct dentry *dir, 1665e3c13928SAl Viro unsigned int flags) 1666697f514dSMiklos Szeredi { 166788d8331aSAl Viro struct dentry *dentry, *old; 16681936386eSAl Viro struct inode *inode = dir->d_inode; 1669d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16701936386eSAl Viro 16711936386eSAl Viro /* Don't go there if it's already dead */ 167294bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 167388d8331aSAl Viro return ERR_PTR(-ENOENT); 167494bdd655SAl Viro again: 1675d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 167694bdd655SAl Viro if (IS_ERR(dentry)) 167788d8331aSAl Viro return dentry; 167894bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1679949a852eSAl Viro int error = d_revalidate(dentry, flags); 1680949a852eSAl Viro if (unlikely(error <= 0)) { 168194bdd655SAl Viro if (!error) { 1682949a852eSAl Viro d_invalidate(dentry); 1683949a852eSAl Viro dput(dentry); 168494bdd655SAl Viro goto again; 168594bdd655SAl Viro } 168694bdd655SAl Viro dput(dentry); 1687949a852eSAl Viro dentry = ERR_PTR(error); 1688949a852eSAl Viro } 168994bdd655SAl Viro } else { 16901936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 169185c7f810SAl Viro d_lookup_done(dentry); 16921936386eSAl Viro if (unlikely(old)) { 16931936386eSAl Viro dput(dentry); 16941936386eSAl Viro dentry = old; 1695949a852eSAl Viro } 1696949a852eSAl Viro } 1697e3c13928SAl Viro return dentry; 16981da177e4SLinus Torvalds } 16991da177e4SLinus Torvalds 170088d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 170188d8331aSAl Viro struct dentry *dir, 170288d8331aSAl Viro unsigned int flags) 170388d8331aSAl Viro { 170488d8331aSAl Viro struct inode *inode = dir->d_inode; 170588d8331aSAl Viro struct dentry *res; 170688d8331aSAl Viro inode_lock_shared(inode); 170788d8331aSAl Viro res = __lookup_slow(name, dir, flags); 170888d8331aSAl Viro inode_unlock_shared(inode); 170988d8331aSAl Viro return res; 171088d8331aSAl Viro } 171188d8331aSAl Viro 17124609e1f1SChristian Brauner static inline int may_lookup(struct mnt_idmap *idmap, 1713ba73d987SChristian Brauner struct nameidata *nd) 171452094c8aSAl Viro { 171552094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17164609e1f1SChristian Brauner int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1717e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 171852094c8aSAl Viro return err; 171952094c8aSAl Viro } 17204609e1f1SChristian Brauner return inode_permission(idmap, nd->inode, MAY_EXEC); 172152094c8aSAl Viro } 172252094c8aSAl Viro 172303fa86e9SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link) 1724d63ff28fSAl Viro { 172549055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 172649055906SAl Viro return -ELOOP; 17274542576bSAl Viro 17284542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17294542576bSAl Viro return 0; 17304542576bSAl Viro if (likely(nd->stack != nd->internal)) 17314542576bSAl Viro return 0; 173260ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 173349055906SAl Viro return 0; 173460ef60c7SAl Viro 173560ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 173660ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 173760ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 173803fa86e9SAl Viro bool grabbed_link = legitimize_path(nd, link, nd->next_seq); 173960ef60c7SAl Viro 1740e5ca024eSAl Viro if (!try_to_unlazy(nd) || !grabbed_link) 174160ef60c7SAl Viro return -ECHILD; 174260ef60c7SAl Viro 174360ef60c7SAl Viro if (nd_alloc_stack(nd)) 174460ef60c7SAl Viro return 0; 1745bc40aee0SAl Viro } 174660ef60c7SAl Viro return -ENOMEM; 174749055906SAl Viro } 174849055906SAl Viro 174949055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 175049055906SAl Viro 175149055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 175203fa86e9SAl Viro struct inode *inode, int flags) 175349055906SAl Viro { 175449055906SAl Viro struct saved *last; 175549055906SAl Viro const char *res; 175603fa86e9SAl Viro int error = reserve_stack(nd, link); 175749055906SAl Viro 175849055906SAl Viro if (unlikely(error)) { 175949055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1760cd179f44SAl Viro path_put(link); 176106708adbSAl Viro return ERR_PTR(error); 1762626de996SAl Viro } 1763ab104923SAl Viro last = nd->stack + nd->depth++; 17641cf2665bSAl Viro last->link = *link; 1765fceef393SAl Viro clear_delayed_call(&last->done); 176603fa86e9SAl Viro last->seq = nd->next_seq; 1767ad6cc4c3SAl Viro 1768b1a81972SAl Viro if (flags & WALK_TRAILING) { 1769ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1770ad6cc4c3SAl Viro if (unlikely(error)) 1771ad6cc4c3SAl Viro return ERR_PTR(error); 1772ad6cc4c3SAl Viro } 1773ad6cc4c3SAl Viro 1774dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1775dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1776ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1777ad6cc4c3SAl Viro 1778ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1779ad6cc4c3SAl Viro touch_atime(&last->link); 1780ad6cc4c3SAl Viro cond_resched(); 1781ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1782e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1783ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1784ad6cc4c3SAl Viro touch_atime(&last->link); 1785ad6cc4c3SAl Viro } 1786ad6cc4c3SAl Viro 1787ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1788ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1789ad6cc4c3SAl Viro if (unlikely(error)) 1790ad6cc4c3SAl Viro return ERR_PTR(error); 1791ad6cc4c3SAl Viro 1792ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1793ad6cc4c3SAl Viro if (!res) { 1794ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1795ad6cc4c3SAl Viro struct delayed_call *); 1796ad6cc4c3SAl Viro get = inode->i_op->get_link; 1797ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1798ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1799e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1800ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1801ad6cc4c3SAl Viro } else { 1802ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1803ad6cc4c3SAl Viro } 1804ad6cc4c3SAl Viro if (!res) 1805ad6cc4c3SAl Viro goto all_done; 1806ad6cc4c3SAl Viro if (IS_ERR(res)) 1807ad6cc4c3SAl Viro return res; 1808ad6cc4c3SAl Viro } 1809ad6cc4c3SAl Viro if (*res == '/') { 1810ad6cc4c3SAl Viro error = nd_jump_root(nd); 1811ad6cc4c3SAl Viro if (unlikely(error)) 1812ad6cc4c3SAl Viro return ERR_PTR(error); 1813ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1814ad6cc4c3SAl Viro ; 1815ad6cc4c3SAl Viro } 1816ad6cc4c3SAl Viro if (*res) 1817ad6cc4c3SAl Viro return res; 1818ad6cc4c3SAl Viro all_done: // pure jump 1819ad6cc4c3SAl Viro put_link(nd); 1820ad6cc4c3SAl Viro return NULL; 1821d63ff28fSAl Viro } 1822d63ff28fSAl Viro 18233ddcd056SLinus Torvalds /* 18243ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18253ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18263ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18273ddcd056SLinus Torvalds * for the common case. 182803fa86e9SAl Viro * 182903fa86e9SAl Viro * NOTE: dentry must be what nd->next_seq had been sampled from. 18303ddcd056SLinus Torvalds */ 1831b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1832a4f5b521SAl Viro struct dentry *dentry) 18333ddcd056SLinus Torvalds { 1834cbae4d12SAl Viro struct path path; 1835a4f5b521SAl Viro struct inode *inode; 18363bd8bc89SAl Viro int err = handle_mounts(nd, dentry, &path); 1837cbae4d12SAl Viro 1838cbae4d12SAl Viro if (err < 0) 1839b0417d2cSAl Viro return ERR_PTR(err); 18403bd8bc89SAl Viro inode = path.dentry->d_inode; 1841cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18428c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1843aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18448f64fb1cSAl Viro /* not a symlink or should not follow */ 18453bd8bc89SAl Viro if (nd->flags & LOOKUP_RCU) { 18463bd8bc89SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) 18473bd8bc89SAl Viro return ERR_PTR(-ECHILD); 18483bd8bc89SAl Viro if (unlikely(!inode)) 18493bd8bc89SAl Viro return ERR_PTR(-ENOENT); 18503bd8bc89SAl Viro } else { 1851c99687a0SAl Viro dput(nd->path.dentry); 1852c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1853c99687a0SAl Viro mntput(nd->path.mnt); 1854c99687a0SAl Viro } 1855c99687a0SAl Viro nd->path = path; 18568f64fb1cSAl Viro nd->inode = inode; 185703fa86e9SAl Viro nd->seq = nd->next_seq; 1858b0417d2cSAl Viro return NULL; 18598f64fb1cSAl Viro } 1860a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 186184f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 186203fa86e9SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) 1863b0417d2cSAl Viro return ERR_PTR(-ECHILD); 186484f0cd9eSAl Viro } else { 186584f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 186684f0cd9eSAl Viro mntget(path.mnt); 1867a7f77542SAl Viro } 186803fa86e9SAl Viro return pick_link(nd, &path, inode, flags); 18693ddcd056SLinus Torvalds } 18703ddcd056SLinus Torvalds 1871b16c001dSAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd) 1872957dd41dSAl Viro { 187312487f30SAl Viro struct dentry *parent, *old; 1874957dd41dSAl Viro 187512487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 187612487f30SAl Viro goto in_root; 187712487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18787ef482faSAl Viro struct path path; 1879efe772d6SAl Viro unsigned seq; 18807ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18817ef482faSAl Viro &nd->root, &path, &seq)) 188212487f30SAl Viro goto in_root; 1883efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1884efe772d6SAl Viro return ERR_PTR(-ECHILD); 1885efe772d6SAl Viro nd->path = path; 1886efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1887efe772d6SAl Viro nd->seq = seq; 188803fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach this state 188982ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1890efe772d6SAl Viro return ERR_PTR(-ECHILD); 1891efe772d6SAl Viro /* we know that mountpoint was pinned */ 1892957dd41dSAl Viro } 189312487f30SAl Viro old = nd->path.dentry; 189412487f30SAl Viro parent = old->d_parent; 189503fa86e9SAl Viro nd->next_seq = read_seqcount_begin(&parent->d_seq); 189603fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach this state 189782ef0698SAl Viro if (read_seqcount_retry(&old->d_seq, nd->seq)) 189812487f30SAl Viro return ERR_PTR(-ECHILD); 189912487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 190012487f30SAl Viro return ERR_PTR(-ECHILD); 190112487f30SAl Viro return parent; 190212487f30SAl Viro in_root: 190382ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1904efe772d6SAl Viro return ERR_PTR(-ECHILD); 1905957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19067521f22bSAl Viro return ERR_PTR(-ECHILD); 190703fa86e9SAl Viro nd->next_seq = nd->seq; 190851c6546cSAl Viro return nd->path.dentry; 1909957dd41dSAl Viro } 1910957dd41dSAl Viro 1911b16c001dSAl Viro static struct dentry *follow_dotdot(struct nameidata *nd) 1912957dd41dSAl Viro { 191312487f30SAl Viro struct dentry *parent; 191412487f30SAl Viro 1915957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 191612487f30SAl Viro goto in_root; 191712487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19182aa38470SAl Viro struct path path; 19192aa38470SAl Viro 19202aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19212aa38470SAl Viro &nd->root, &path)) 192212487f30SAl Viro goto in_root; 1923165200d6SAl Viro path_put(&nd->path); 1924165200d6SAl Viro nd->path = path; 19252aa38470SAl Viro nd->inode = path.dentry->d_inode; 1926165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1927165200d6SAl Viro return ERR_PTR(-EXDEV); 192812487f30SAl Viro } 1929957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 193012487f30SAl Viro parent = dget_parent(nd->path.dentry); 1931957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1932957dd41dSAl Viro dput(parent); 19337521f22bSAl Viro return ERR_PTR(-ENOENT); 1934957dd41dSAl Viro } 1935c2df1968SAl Viro return parent; 193612487f30SAl Viro 193712487f30SAl Viro in_root: 1938957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19397521f22bSAl Viro return ERR_PTR(-EXDEV); 194051c6546cSAl Viro return dget(nd->path.dentry); 1941957dd41dSAl Viro } 1942957dd41dSAl Viro 19437521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1944957dd41dSAl Viro { 1945957dd41dSAl Viro if (type == LAST_DOTDOT) { 19467521f22bSAl Viro const char *error = NULL; 1947c2df1968SAl Viro struct dentry *parent; 1948957dd41dSAl Viro 1949957dd41dSAl Viro if (!nd->root.mnt) { 19507521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1951957dd41dSAl Viro if (error) 1952957dd41dSAl Viro return error; 1953957dd41dSAl Viro } 1954957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1955b16c001dSAl Viro parent = follow_dotdot_rcu(nd); 1956957dd41dSAl Viro else 1957b16c001dSAl Viro parent = follow_dotdot(nd); 1958c2df1968SAl Viro if (IS_ERR(parent)) 1959c2df1968SAl Viro return ERR_CAST(parent); 1960a4f5b521SAl Viro error = step_into(nd, WALK_NOFOLLOW, parent); 1961c2df1968SAl Viro if (unlikely(error)) 1962957dd41dSAl Viro return error; 1963957dd41dSAl Viro 1964957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1965957dd41dSAl Viro /* 1966957dd41dSAl Viro * If there was a racing rename or mount along our 1967957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1968957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1969957dd41dSAl Viro * some fallback). 1970957dd41dSAl Viro */ 1971957dd41dSAl Viro smp_rmb(); 197282ef0698SAl Viro if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)) 19737521f22bSAl Viro return ERR_PTR(-EAGAIN); 197482ef0698SAl Viro if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)) 19757521f22bSAl Viro return ERR_PTR(-EAGAIN); 1976957dd41dSAl Viro } 1977957dd41dSAl Viro } 19787521f22bSAl Viro return NULL; 1979957dd41dSAl Viro } 1980957dd41dSAl Viro 198192d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1982ce57dfc1SAl Viro { 1983db3c9adeSAl Viro struct dentry *dentry; 1984ce57dfc1SAl Viro /* 1985ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1986ce57dfc1SAl Viro * to be able to know about the current root directory and 1987ce57dfc1SAl Viro * parent relationships. 1988ce57dfc1SAl Viro */ 19894693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 19901c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 19914693a547SAl Viro put_link(nd); 19927521f22bSAl Viro return handle_dots(nd, nd->last_type); 19934693a547SAl Viro } 19944cb64024SAl Viro dentry = lookup_fast(nd); 199520e34357SAl Viro if (IS_ERR(dentry)) 199692d27016SAl Viro return ERR_CAST(dentry); 199720e34357SAl Viro if (unlikely(!dentry)) { 1998db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1999db3c9adeSAl Viro if (IS_ERR(dentry)) 200092d27016SAl Viro return ERR_CAST(dentry); 200120e34357SAl Viro } 200256676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 200356676ec3SAl Viro put_link(nd); 2004a4f5b521SAl Viro return step_into(nd, flags, dentry); 2005ce57dfc1SAl Viro } 2006ce57dfc1SAl Viro 20071da177e4SLinus Torvalds /* 2008bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2009bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2010bfcfaa77SLinus Torvalds * 2011bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2012bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2013bfcfaa77SLinus Torvalds * fast. 2014bfcfaa77SLinus Torvalds * 2015bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2016bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2017bfcfaa77SLinus Torvalds * crossing operation. 2018bfcfaa77SLinus Torvalds * 2019bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2020bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2021bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2022bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2023bfcfaa77SLinus Torvalds */ 2024bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2025bfcfaa77SLinus Torvalds 2026f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2027bfcfaa77SLinus Torvalds 2028468a9428SGeorge Spelvin #ifdef HASH_MIX 2029bfcfaa77SLinus Torvalds 2030468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2031468a9428SGeorge Spelvin 2032468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20332a18da7aSGeorge Spelvin /* 20342a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20352a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20362a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20372a18da7aSGeorge Spelvin * and no temporaries. 20382a18da7aSGeorge Spelvin * 20392a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20402a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20412a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20422a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20432a18da7aSGeorge Spelvin * 20442a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20452a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20462a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20472a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20482a18da7aSGeorge Spelvin * 20492a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20502a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20512a18da7aSGeorge Spelvin * toggling any given output bit. 20522a18da7aSGeorge Spelvin * 20532a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20542a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20552a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20562a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20572a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20582a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20592a18da7aSGeorge Spelvin * Perfect: 8192 258048 20602a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20612a18da7aSGeorge Spelvin */ 20622a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20632a18da7aSGeorge Spelvin ( x ^= (a), \ 20642a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20652a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20662a18da7aSGeorge Spelvin y *= 9 ) 2067bfcfaa77SLinus Torvalds 20680fed3ac8SGeorge Spelvin /* 20692a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20702a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20712a18da7aSGeorge Spelvin * work done before the hash value is used. 20720fed3ac8SGeorge Spelvin */ 20732a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20740fed3ac8SGeorge Spelvin { 20752a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20762a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20772a18da7aSGeorge Spelvin return y >> 32; 20780fed3ac8SGeorge Spelvin } 20790fed3ac8SGeorge Spelvin 2080bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2081bfcfaa77SLinus Torvalds 20822a18da7aSGeorge Spelvin /* 20832a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 20842a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20852a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 20862a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 20872a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 20882a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 20892a18da7aSGeorge Spelvin * Perfect: 2048 31744 20902a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 20912a18da7aSGeorge Spelvin */ 20922a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20932a18da7aSGeorge Spelvin ( x ^= (a), \ 20942a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 20952a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 20962a18da7aSGeorge Spelvin y *= 9 ) 2097bfcfaa77SLinus Torvalds 20982a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20990fed3ac8SGeorge Spelvin { 21002a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 21012a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 21020fed3ac8SGeorge Spelvin } 21030fed3ac8SGeorge Spelvin 2104bfcfaa77SLinus Torvalds #endif 2105bfcfaa77SLinus Torvalds 21062a18da7aSGeorge Spelvin /* 21072a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21082a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21092a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21102a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21112a18da7aSGeorge Spelvin * finds the delimiter after the name. 21122a18da7aSGeorge Spelvin */ 21138387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2114bfcfaa77SLinus Torvalds { 21158387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2116bfcfaa77SLinus Torvalds 2117bfcfaa77SLinus Torvalds for (;;) { 2118fcfd2fbfSGeorge Spelvin if (!len) 2119fcfd2fbfSGeorge Spelvin goto done; 2120e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2121bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2122bfcfaa77SLinus Torvalds break; 21232a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2124bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2125bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2126bfcfaa77SLinus Torvalds } 21272a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2128bfcfaa77SLinus Torvalds done: 21292a18da7aSGeorge Spelvin return fold_hash(x, y); 2130bfcfaa77SLinus Torvalds } 2131bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2132bfcfaa77SLinus Torvalds 2133fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21348387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2135fcfd2fbfSGeorge Spelvin { 21368387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21378387ff25SLinus Torvalds unsigned long adata, mask, len; 2138fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2139fcfd2fbfSGeorge Spelvin 21408387ff25SLinus Torvalds len = 0; 21418387ff25SLinus Torvalds goto inside; 21428387ff25SLinus Torvalds 2143fcfd2fbfSGeorge Spelvin do { 21442a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2145fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21468387ff25SLinus Torvalds inside: 2147fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2148fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2149fcfd2fbfSGeorge Spelvin 2150fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2151fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21522a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2153fcfd2fbfSGeorge Spelvin 21542a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2155fcfd2fbfSGeorge Spelvin } 2156fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2157fcfd2fbfSGeorge Spelvin 2158bfcfaa77SLinus Torvalds /* 2159bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2160d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2161bfcfaa77SLinus Torvalds */ 21628387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2163bfcfaa77SLinus Torvalds { 21648387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21658387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 216636126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2167bfcfaa77SLinus Torvalds 21688387ff25SLinus Torvalds len = 0; 21698387ff25SLinus Torvalds goto inside; 21708387ff25SLinus Torvalds 2171bfcfaa77SLinus Torvalds do { 21722a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2173bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21748387ff25SLinus Torvalds inside: 2175e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 217636126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 217736126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2178bfcfaa77SLinus Torvalds 217936126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 218036126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 218136126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21822a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 218336126f8fSLinus Torvalds 21842a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2185bfcfaa77SLinus Torvalds } 2186bfcfaa77SLinus Torvalds 21872a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2188bfcfaa77SLinus Torvalds 2189fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 21908387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 21910145acc2SLinus Torvalds { 21928387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 21930145acc2SLinus Torvalds while (len--) 2194fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 21950145acc2SLinus Torvalds return end_name_hash(hash); 21960145acc2SLinus Torvalds } 2197ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 21980145acc2SLinus Torvalds 2199fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 22008387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2201fcfd2fbfSGeorge Spelvin { 22028387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2203fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2204fcfd2fbfSGeorge Spelvin 2205fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2206e0ab7af9SGeorge Spelvin while (c) { 2207fcfd2fbfSGeorge Spelvin len++; 2208fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2209fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2210e0ab7af9SGeorge Spelvin } 2211fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2212fcfd2fbfSGeorge Spelvin } 2213f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2214fcfd2fbfSGeorge Spelvin 22153ddcd056SLinus Torvalds /* 2216200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2217200e9ef7SLinus Torvalds * one character. 2218200e9ef7SLinus Torvalds */ 22198387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2220200e9ef7SLinus Torvalds { 22218387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2222200e9ef7SLinus Torvalds unsigned long len = 0, c; 2223200e9ef7SLinus Torvalds 2224200e9ef7SLinus Torvalds c = (unsigned char)*name; 2225200e9ef7SLinus Torvalds do { 2226200e9ef7SLinus Torvalds len++; 2227200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2228200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2229200e9ef7SLinus Torvalds } while (c && c != '/'); 2230d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2231200e9ef7SLinus Torvalds } 2232200e9ef7SLinus Torvalds 2233bfcfaa77SLinus Torvalds #endif 2234bfcfaa77SLinus Torvalds 2235200e9ef7SLinus Torvalds /* 22361da177e4SLinus Torvalds * Name resolution. 2237ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2238ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22391da177e4SLinus Torvalds * 2240ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2241ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22421da177e4SLinus Torvalds */ 22436de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22441da177e4SLinus Torvalds { 2245d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22461da177e4SLinus Torvalds int err; 22471da177e4SLinus Torvalds 2248b4c03536SAl Viro nd->last_type = LAST_ROOT; 2249c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22509b5858e9SAl Viro if (IS_ERR(name)) 22519b5858e9SAl Viro return PTR_ERR(name); 22521da177e4SLinus Torvalds while (*name=='/') 22531da177e4SLinus Torvalds name++; 22541a97d899SAl Viro if (!*name) { 22551a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22569e18f10aSAl Viro return 0; 22571a97d899SAl Viro } 22581da177e4SLinus Torvalds 22591da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22601da177e4SLinus Torvalds for(;;) { 22614609e1f1SChristian Brauner struct mnt_idmap *idmap; 226292d27016SAl Viro const char *link; 2263d6bb3e90SLinus Torvalds u64 hash_len; 2264fe479a58SAl Viro int type; 22651da177e4SLinus Torvalds 22664609e1f1SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 22674609e1f1SChristian Brauner err = may_lookup(idmap, nd); 22681da177e4SLinus Torvalds if (err) 22693595e234SAl Viro return err; 22701da177e4SLinus Torvalds 22718387ff25SLinus Torvalds hash_len = hash_name(nd->path.dentry, name); 22721da177e4SLinus Torvalds 2273fe479a58SAl Viro type = LAST_NORM; 2274d6bb3e90SLinus Torvalds if (name[0] == '.') switch (hashlen_len(hash_len)) { 2275fe479a58SAl Viro case 2: 2276200e9ef7SLinus Torvalds if (name[1] == '.') { 2277fe479a58SAl Viro type = LAST_DOTDOT; 2278bcba1e7dSAl Viro nd->state |= ND_JUMPED; 227916c2cd71SAl Viro } 2280fe479a58SAl Viro break; 2281fe479a58SAl Viro case 1: 2282fe479a58SAl Viro type = LAST_DOT; 2283fe479a58SAl Viro } 22845a202bcdSAl Viro if (likely(type == LAST_NORM)) { 22855a202bcdSAl Viro struct dentry *parent = nd->path.dentry; 2286bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; 22875a202bcdSAl Viro if (unlikely(parent->d_flags & DCACHE_OP_HASH)) { 2288a060dc50SJames Hogan struct qstr this = { { .hash_len = hash_len }, .name = name }; 2289da53be12SLinus Torvalds err = parent->d_op->d_hash(parent, &this); 22905a202bcdSAl Viro if (err < 0) 22913595e234SAl Viro return err; 2292d6bb3e90SLinus Torvalds hash_len = this.hash_len; 2293d6bb3e90SLinus Torvalds name = this.name; 22945a202bcdSAl Viro } 22955a202bcdSAl Viro } 2296fe479a58SAl Viro 2297d6bb3e90SLinus Torvalds nd->last.hash_len = hash_len; 2298d6bb3e90SLinus Torvalds nd->last.name = name; 22995f4a6a69SAl Viro nd->last_type = type; 23005f4a6a69SAl Viro 2301d6bb3e90SLinus Torvalds name += hashlen_len(hash_len); 2302d6bb3e90SLinus Torvalds if (!*name) 2303bdf6cbf1SAl Viro goto OK; 2304200e9ef7SLinus Torvalds /* 2305200e9ef7SLinus Torvalds * If it wasn't NUL, we know it was '/'. Skip that 2306200e9ef7SLinus Torvalds * slash, and continue until no more slashes. 2307200e9ef7SLinus Torvalds */ 2308200e9ef7SLinus Torvalds do { 2309d6bb3e90SLinus Torvalds name++; 2310d6bb3e90SLinus Torvalds } while (unlikely(*name == '/')); 23118620c238SAl Viro if (unlikely(!*name)) { 23128620c238SAl Viro OK: 2313d8d4611aSAl Viro /* pathname or trailing symlink, done */ 2314c108837eSAl Viro if (!depth) { 2315e67fe633SChristian Brauner nd->dir_vfsuid = i_uid_into_vfsuid(idmap, nd->inode); 23160f705953SAl Viro nd->dir_mode = nd->inode->i_mode; 2317c108837eSAl Viro nd->flags &= ~LOOKUP_PARENT; 23188620c238SAl Viro return 0; 2319c108837eSAl Viro } 23208620c238SAl Viro /* last component of nested symlink */ 2321d8d4611aSAl Viro name = nd->stack[--depth].name; 23228c4efe22SAl Viro link = walk_component(nd, 0); 23231c4ff1a8SAl Viro } else { 23241c4ff1a8SAl Viro /* not the last component */ 23258c4efe22SAl Viro link = walk_component(nd, WALK_MORE); 23268620c238SAl Viro } 232792d27016SAl Viro if (unlikely(link)) { 232892d27016SAl Viro if (IS_ERR(link)) 232992d27016SAl Viro return PTR_ERR(link); 233092d27016SAl Viro /* a symlink to follow */ 2331d8d4611aSAl Viro nd->stack[depth++].name = name; 233292d27016SAl Viro name = link; 23339e18f10aSAl Viro continue; 233448c8b0c5SAl Viro } 233597242f99SAl Viro if (unlikely(!d_can_lookup(nd->path.dentry))) { 233697242f99SAl Viro if (nd->flags & LOOKUP_RCU) { 2337e36cffedSJens Axboe if (!try_to_unlazy(nd)) 233897242f99SAl Viro return -ECHILD; 233997242f99SAl Viro } 23403595e234SAl Viro return -ENOTDIR; 23415f4a6a69SAl Viro } 2342ce57dfc1SAl Viro } 234397242f99SAl Viro } 23441da177e4SLinus Torvalds 2345edc2b1daSAl Viro /* must be paired with terminate_walk() */ 2346c8a53ee5SAl Viro static const char *path_init(struct nameidata *nd, unsigned flags) 234731e6b01fSNick Piggin { 2348740a1678SAleksa Sarai int error; 2349c8a53ee5SAl Viro const char *s = nd->name->name; 235031e6b01fSNick Piggin 23516c6ec2b0SJens Axboe /* LOOKUP_CACHED requires RCU, ask caller to retry */ 23526c6ec2b0SJens Axboe if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED) 23536c6ec2b0SJens Axboe return ERR_PTR(-EAGAIN); 23546c6ec2b0SJens Axboe 2355c0eb027eSLinus Torvalds if (!*s) 2356c0eb027eSLinus Torvalds flags &= ~LOOKUP_RCU; 2357edc2b1daSAl Viro if (flags & LOOKUP_RCU) 2358edc2b1daSAl Viro rcu_read_lock(); 235903fa86e9SAl Viro else 236003fa86e9SAl Viro nd->seq = nd->next_seq = 0; 2361c0eb027eSLinus Torvalds 2362bcba1e7dSAl Viro nd->flags = flags; 2363bcba1e7dSAl Viro nd->state |= ND_JUMPED; 2364ab87f9a5SAleksa Sarai 2365ab87f9a5SAleksa Sarai nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount); 2366ab87f9a5SAleksa Sarai nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount); 2367ab87f9a5SAleksa Sarai smp_rmb(); 2368ab87f9a5SAleksa Sarai 2369bcba1e7dSAl Viro if (nd->state & ND_ROOT_PRESET) { 2370b18825a7SDavid Howells struct dentry *root = nd->root.dentry; 2371b18825a7SDavid Howells struct inode *inode = root->d_inode; 237293893862SAl Viro if (*s && unlikely(!d_can_lookup(root))) 2373368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 23745b6ca027SAl Viro nd->path = nd->root; 23755b6ca027SAl Viro nd->inode = inode; 23765b6ca027SAl Viro if (flags & LOOKUP_RCU) { 2377ab87f9a5SAleksa Sarai nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 23788f47a016SAl Viro nd->root_seq = nd->seq; 23795b6ca027SAl Viro } else { 23805b6ca027SAl Viro path_get(&nd->path); 23815b6ca027SAl Viro } 2382368ee9baSAl Viro return s; 23835b6ca027SAl Viro } 23845b6ca027SAl Viro 238531e6b01fSNick Piggin nd->root.mnt = NULL; 238631e6b01fSNick Piggin 23878db52c7eSAleksa Sarai /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */ 23888db52c7eSAleksa Sarai if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) { 2389740a1678SAleksa Sarai error = nd_jump_root(nd); 2390740a1678SAleksa Sarai if (unlikely(error)) 2391740a1678SAleksa Sarai return ERR_PTR(error); 2392ef55d917SAl Viro return s; 23938db52c7eSAleksa Sarai } 23948db52c7eSAleksa Sarai 23958db52c7eSAleksa Sarai /* Relative pathname -- get the starting-point it is relative to. */ 23968db52c7eSAleksa Sarai if (nd->dfd == AT_FDCWD) { 2397e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 239831e6b01fSNick Piggin struct fs_struct *fs = current->fs; 2399c28cc364SNick Piggin unsigned seq; 240031e6b01fSNick Piggin 2401c28cc364SNick Piggin do { 2402c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 240331e6b01fSNick Piggin nd->path = fs->pwd; 2404ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2405c28cc364SNick Piggin nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq); 2406c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 2407e41f7d4eSAl Viro } else { 2408e41f7d4eSAl Viro get_fs_pwd(current->fs, &nd->path); 2409ef55d917SAl Viro nd->inode = nd->path.dentry->d_inode; 2410e41f7d4eSAl Viro } 241131e6b01fSNick Piggin } else { 2412582aa64aSJeff Layton /* Caller must check execute permissions on the starting path component */ 2413c8a53ee5SAl Viro struct fd f = fdget_raw(nd->dfd); 241431e6b01fSNick Piggin struct dentry *dentry; 241531e6b01fSNick Piggin 24162903ff01SAl Viro if (!f.file) 2417368ee9baSAl Viro return ERR_PTR(-EBADF); 241831e6b01fSNick Piggin 24192903ff01SAl Viro dentry = f.file->f_path.dentry; 242031e6b01fSNick Piggin 2421edc2b1daSAl Viro if (*s && unlikely(!d_can_lookup(dentry))) { 24222903ff01SAl Viro fdput(f); 2423368ee9baSAl Viro return ERR_PTR(-ENOTDIR); 2424f52e0c11SAl Viro } 24252903ff01SAl Viro 24262903ff01SAl Viro nd->path = f.file->f_path; 2427e41f7d4eSAl Viro if (flags & LOOKUP_RCU) { 242834a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 242934a26b99SAl Viro nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq); 24305590ff0dSUlrich Drepper } else { 24312903ff01SAl Viro path_get(&nd->path); 243234a26b99SAl Viro nd->inode = nd->path.dentry->d_inode; 24331da177e4SLinus Torvalds } 243434a26b99SAl Viro fdput(f); 2435e41f7d4eSAl Viro } 24368db52c7eSAleksa Sarai 2437adb21d2bSAleksa Sarai /* For scoped-lookups we need to set the root to the dirfd as well. */ 2438adb21d2bSAleksa Sarai if (flags & LOOKUP_IS_SCOPED) { 2439adb21d2bSAleksa Sarai nd->root = nd->path; 2440adb21d2bSAleksa Sarai if (flags & LOOKUP_RCU) { 2441adb21d2bSAleksa Sarai nd->root_seq = nd->seq; 2442adb21d2bSAleksa Sarai } else { 2443adb21d2bSAleksa Sarai path_get(&nd->root); 2444bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 2445adb21d2bSAleksa Sarai } 2446adb21d2bSAleksa Sarai } 2447adb21d2bSAleksa Sarai return s; 24489b4a9b14SAl Viro } 24499b4a9b14SAl Viro 24501ccac622SAl Viro static inline const char *lookup_last(struct nameidata *nd) 245195fa25d9SAl Viro { 2452bd92d7feSAl Viro if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len]) 2453bd92d7feSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 2454bd92d7feSAl Viro 2455c108837eSAl Viro return walk_component(nd, WALK_TRAILING); 2456bd92d7feSAl Viro } 2457bd92d7feSAl Viro 24584f757f3cSAl Viro static int handle_lookup_down(struct nameidata *nd) 24594f757f3cSAl Viro { 2460c153007bSAl Viro if (!(nd->flags & LOOKUP_RCU)) 2461db3c9adeSAl Viro dget(nd->path.dentry); 246203fa86e9SAl Viro nd->next_seq = nd->seq; 2463a4f5b521SAl Viro return PTR_ERR(step_into(nd, WALK_NOFOLLOW, nd->path.dentry)); 24644f757f3cSAl Viro } 24654f757f3cSAl Viro 24669b4a9b14SAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2467c8a53ee5SAl Viro static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path) 24689b4a9b14SAl Viro { 2469c8a53ee5SAl Viro const char *s = path_init(nd, flags); 2470bd92d7feSAl Viro int err; 247131e6b01fSNick Piggin 24729b5858e9SAl Viro if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) { 24734f757f3cSAl Viro err = handle_lookup_down(nd); 24745f336e72SAl Viro if (unlikely(err < 0)) 24755f336e72SAl Viro s = ERR_PTR(err); 24764f757f3cSAl Viro } 24774f757f3cSAl Viro 24781ccac622SAl Viro while (!(err = link_path_walk(s, nd)) && 24791ccac622SAl Viro (s = lookup_last(nd)) != NULL) 24801ccac622SAl Viro ; 24814f0ed93fSAl Viro if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) { 24824f0ed93fSAl Viro err = handle_lookup_down(nd); 2483bcba1e7dSAl Viro nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please... 24844f0ed93fSAl Viro } 24859f1fafeeSAl Viro if (!err) 24869f1fafeeSAl Viro err = complete_walk(nd); 2487bd92d7feSAl Viro 2488deb106c6SAl Viro if (!err && nd->flags & LOOKUP_DIRECTORY) 2489deb106c6SAl Viro if (!d_can_lookup(nd->path.dentry)) 2490bd23a539SAl Viro err = -ENOTDIR; 2491625b6d10SAl Viro if (!err) { 2492625b6d10SAl Viro *path = nd->path; 2493625b6d10SAl Viro nd->path.mnt = NULL; 2494625b6d10SAl Viro nd->path.dentry = NULL; 2495625b6d10SAl Viro } 2496deb106c6SAl Viro terminate_walk(nd); 2497bd92d7feSAl Viro return err; 249831e6b01fSNick Piggin } 249931e6b01fSNick Piggin 2500794ebceaSStephen Brennan int filename_lookup(int dfd, struct filename *name, unsigned flags, 25019ad1aaa6SAl Viro struct path *path, struct path *root) 2502873f1eedSJeff Layton { 2503894bc8c4SAl Viro int retval; 25049883d185SAl Viro struct nameidata nd; 2505abc9f5beSAl Viro if (IS_ERR(name)) 2506abc9f5beSAl Viro return PTR_ERR(name); 250706422964SAl Viro set_nameidata(&nd, dfd, name, root); 2508c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_RCU, path); 2509873f1eedSJeff Layton if (unlikely(retval == -ECHILD)) 2510c8a53ee5SAl Viro retval = path_lookupat(&nd, flags, path); 2511873f1eedSJeff Layton if (unlikely(retval == -ESTALE)) 2512c8a53ee5SAl Viro retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path); 2513873f1eedSJeff Layton 2514873f1eedSJeff Layton if (likely(!retval)) 2515161aff1dSAl Viro audit_inode(name, path->dentry, 2516161aff1dSAl Viro flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0); 25179883d185SAl Viro restore_nameidata(); 2518020250f3SDmitry Kadashev return retval; 2519020250f3SDmitry Kadashev } 2520020250f3SDmitry Kadashev 25218bcb77faSAl Viro /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */ 2522c8a53ee5SAl Viro static int path_parentat(struct nameidata *nd, unsigned flags, 2523391172c4SAl Viro struct path *parent) 25248bcb77faSAl Viro { 2525c8a53ee5SAl Viro const char *s = path_init(nd, flags); 25269b5858e9SAl Viro int err = link_path_walk(s, nd); 25278bcb77faSAl Viro if (!err) 25288bcb77faSAl Viro err = complete_walk(nd); 2529391172c4SAl Viro if (!err) { 2530391172c4SAl Viro *parent = nd->path; 2531391172c4SAl Viro nd->path.mnt = NULL; 2532391172c4SAl Viro nd->path.dentry = NULL; 2533391172c4SAl Viro } 2534deb106c6SAl Viro terminate_walk(nd); 25358bcb77faSAl Viro return err; 25368bcb77faSAl Viro } 25378bcb77faSAl Viro 25380766ec82SStephen Brennan /* Note: this does not consume "name" */ 2539*74d7970fSNamjae Jeon static int __filename_parentat(int dfd, struct filename *name, 2540391172c4SAl Viro unsigned int flags, struct path *parent, 2541*74d7970fSNamjae Jeon struct qstr *last, int *type, 2542*74d7970fSNamjae Jeon const struct path *root) 25438bcb77faSAl Viro { 25448bcb77faSAl Viro int retval; 25459883d185SAl Viro struct nameidata nd; 25468bcb77faSAl Viro 25475c31b6ceSAl Viro if (IS_ERR(name)) 25480ee50b47SDmitry Kadashev return PTR_ERR(name); 2549*74d7970fSNamjae Jeon set_nameidata(&nd, dfd, name, root); 2550c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25518bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2552c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25538bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2554c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2555391172c4SAl Viro if (likely(!retval)) { 2556391172c4SAl Viro *last = nd.last; 2557391172c4SAl Viro *type = nd.last_type; 2558c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2559391172c4SAl Viro } 25609883d185SAl Viro restore_nameidata(); 25610ee50b47SDmitry Kadashev return retval; 25620ee50b47SDmitry Kadashev } 25630ee50b47SDmitry Kadashev 2564*74d7970fSNamjae Jeon static int filename_parentat(int dfd, struct filename *name, 2565*74d7970fSNamjae Jeon unsigned int flags, struct path *parent, 2566*74d7970fSNamjae Jeon struct qstr *last, int *type) 2567*74d7970fSNamjae Jeon { 2568*74d7970fSNamjae Jeon return __filename_parentat(dfd, name, flags, parent, last, type, NULL); 2569*74d7970fSNamjae Jeon } 2570*74d7970fSNamjae Jeon 257179714f72SAl Viro /* does lookup, returns the object with parent locked */ 25720766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25735590ff0dSUlrich Drepper { 25745c31b6ceSAl Viro struct dentry *d; 2575391172c4SAl Viro struct qstr last; 25760ee50b47SDmitry Kadashev int type, error; 257751689104SPaul Moore 2578c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25790ee50b47SDmitry Kadashev if (error) 25800ee50b47SDmitry Kadashev return ERR_PTR(error); 25815c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2582391172c4SAl Viro path_put(path); 25835c31b6ceSAl Viro return ERR_PTR(-EINVAL); 258479714f72SAl Viro } 25855955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2586*74d7970fSNamjae Jeon d = lookup_one_qstr_excl(&last, path->dentry, 0); 258779714f72SAl Viro if (IS_ERR(d)) { 25885955102cSAl Viro inode_unlock(path->dentry->d_inode); 2589391172c4SAl Viro path_put(path); 259079714f72SAl Viro } 259179714f72SAl Viro return d; 25925590ff0dSUlrich Drepper } 25935590ff0dSUlrich Drepper 25940766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 25950766ec82SStephen Brennan { 25960766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 25970766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 25980766ec82SStephen Brennan 25990766ec82SStephen Brennan putname(filename); 26000766ec82SStephen Brennan return res; 26010766ec82SStephen Brennan } 26020766ec82SStephen Brennan 2603d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2604d1811465SAl Viro { 2605794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2606794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2607794ebceaSStephen Brennan 2608794ebceaSStephen Brennan putname(filename); 2609794ebceaSStephen Brennan return ret; 2610794ebceaSStephen Brennan 2611d1811465SAl Viro } 26124d359507SAl Viro EXPORT_SYMBOL(kern_path); 2613d1811465SAl Viro 261416f18200SJosef 'Jeff' Sipek /** 2615*74d7970fSNamjae Jeon * vfs_path_parent_lookup - lookup a parent path relative to a dentry-vfsmount pair 2616*74d7970fSNamjae Jeon * @filename: filename structure 2617*74d7970fSNamjae Jeon * @flags: lookup flags 2618*74d7970fSNamjae Jeon * @parent: pointer to struct path to fill 2619*74d7970fSNamjae Jeon * @last: last component 2620*74d7970fSNamjae Jeon * @type: type of the last component 2621*74d7970fSNamjae Jeon * @root: pointer to struct path of the base directory 2622*74d7970fSNamjae Jeon */ 2623*74d7970fSNamjae Jeon int vfs_path_parent_lookup(struct filename *filename, unsigned int flags, 2624*74d7970fSNamjae Jeon struct path *parent, struct qstr *last, int *type, 2625*74d7970fSNamjae Jeon const struct path *root) 2626*74d7970fSNamjae Jeon { 2627*74d7970fSNamjae Jeon return __filename_parentat(AT_FDCWD, filename, flags, parent, last, 2628*74d7970fSNamjae Jeon type, root); 2629*74d7970fSNamjae Jeon } 2630*74d7970fSNamjae Jeon EXPORT_SYMBOL(vfs_path_parent_lookup); 2631*74d7970fSNamjae Jeon 2632*74d7970fSNamjae Jeon /** 263316f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 263416f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 263516f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 263616f18200SJosef 'Jeff' Sipek * @name: pointer to file name 263716f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2638e0a01249SAl Viro * @path: pointer to struct path to fill 263916f18200SJosef 'Jeff' Sipek */ 264016f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 264116f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2642e0a01249SAl Viro struct path *path) 264316f18200SJosef 'Jeff' Sipek { 2644794ebceaSStephen Brennan struct filename *filename; 26459ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2646794ebceaSStephen Brennan int ret; 2647794ebceaSStephen Brennan 2648794ebceaSStephen Brennan filename = getname_kernel(name); 26499ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2650794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2651794ebceaSStephen Brennan putname(filename); 2652794ebceaSStephen Brennan return ret; 265316f18200SJosef 'Jeff' Sipek } 26544d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 265516f18200SJosef 'Jeff' Sipek 26564609e1f1SChristian Brauner static int lookup_one_common(struct mnt_idmap *idmap, 2657c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2658c2fd68b6SChristian Brauner struct qstr *this) 26593c95f0dcSAl Viro { 26603c95f0dcSAl Viro this->name = name; 26613c95f0dcSAl Viro this->len = len; 26623c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26633c95f0dcSAl Viro if (!len) 26643c95f0dcSAl Viro return -EACCES; 26653c95f0dcSAl Viro 26663c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26673c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26683c95f0dcSAl Viro return -EACCES; 26693c95f0dcSAl Viro } 26703c95f0dcSAl Viro 26713c95f0dcSAl Viro while (len--) { 26723c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26733c95f0dcSAl Viro if (c == '/' || c == '\0') 26743c95f0dcSAl Viro return -EACCES; 26753c95f0dcSAl Viro } 26763c95f0dcSAl Viro /* 26773c95f0dcSAl Viro * See if the low-level filesystem might want 26783c95f0dcSAl Viro * to use its own hash.. 26793c95f0dcSAl Viro */ 26803c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26813c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26823c95f0dcSAl Viro if (err < 0) 26833c95f0dcSAl Viro return err; 26843c95f0dcSAl Viro } 26853c95f0dcSAl Viro 26864609e1f1SChristian Brauner return inode_permission(idmap, base->d_inode, MAY_EXEC); 26873c95f0dcSAl Viro } 26883c95f0dcSAl Viro 2689eead1911SChristoph Hellwig /** 26900da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26910da0b7fdSDavid Howells * @name: pathname component to lookup 26920da0b7fdSDavid Howells * @base: base directory to lookup from 26930da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26940da0b7fdSDavid Howells * 26950da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26960da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26970da0b7fdSDavid Howells * 26980da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26990da0b7fdSDavid Howells * not be called by generic code. 27000da0b7fdSDavid Howells * 27010da0b7fdSDavid Howells * The caller must hold base->i_mutex. 27020da0b7fdSDavid Howells */ 27030da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 27040da0b7fdSDavid Howells { 27050da0b7fdSDavid Howells struct qstr this; 27060da0b7fdSDavid Howells int err; 27070da0b7fdSDavid Howells 27080da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27090da0b7fdSDavid Howells 27104609e1f1SChristian Brauner err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this); 27110da0b7fdSDavid Howells if (err) 27120da0b7fdSDavid Howells return ERR_PTR(err); 27130da0b7fdSDavid Howells 27140da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 27150da0b7fdSDavid Howells } 27160da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 27170da0b7fdSDavid Howells 27180da0b7fdSDavid Howells /** 2719a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2720eead1911SChristoph Hellwig * @name: pathname component to lookup 2721eead1911SChristoph Hellwig * @base: base directory to lookup from 2722eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2723eead1911SChristoph Hellwig * 2724a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 27259e7543e9SAl Viro * not be called by generic code. 2726bbddca8eSNeilBrown * 2727bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2728eead1911SChristoph Hellwig */ 2729057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2730057f6c01SJames Morris { 27318613a209SAl Viro struct dentry *dentry; 2732057f6c01SJames Morris struct qstr this; 2733cda309deSMiklos Szeredi int err; 2734057f6c01SJames Morris 27355955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27362f9092e1SDavid Woodhouse 27374609e1f1SChristian Brauner err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this); 2738cda309deSMiklos Szeredi if (err) 2739cda309deSMiklos Szeredi return ERR_PTR(err); 2740cda309deSMiklos Szeredi 27418613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27428613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2743057f6c01SJames Morris } 27444d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2745057f6c01SJames Morris 2746bbddca8eSNeilBrown /** 2747c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 27484609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 2749c2fd68b6SChristian Brauner * @name: pathname component to lookup 2750c2fd68b6SChristian Brauner * @base: base directory to lookup from 2751c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2752c2fd68b6SChristian Brauner * 2753c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2754c2fd68b6SChristian Brauner * not be called by generic code. 2755c2fd68b6SChristian Brauner * 2756c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2757c2fd68b6SChristian Brauner */ 27584609e1f1SChristian Brauner struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name, 2759c2fd68b6SChristian Brauner struct dentry *base, int len) 2760c2fd68b6SChristian Brauner { 2761c2fd68b6SChristian Brauner struct dentry *dentry; 2762c2fd68b6SChristian Brauner struct qstr this; 2763c2fd68b6SChristian Brauner int err; 2764c2fd68b6SChristian Brauner 2765c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2766c2fd68b6SChristian Brauner 27674609e1f1SChristian Brauner err = lookup_one_common(idmap, name, base, len, &this); 2768c2fd68b6SChristian Brauner if (err) 2769c2fd68b6SChristian Brauner return ERR_PTR(err); 2770c2fd68b6SChristian Brauner 2771c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2772c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2773c2fd68b6SChristian Brauner } 2774c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2775c2fd68b6SChristian Brauner 2776c2fd68b6SChristian Brauner /** 277700675017SChristian Brauner * lookup_one_unlocked - filesystem helper to lookup single pathname component 27784609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 277900675017SChristian Brauner * @name: pathname component to lookup 278000675017SChristian Brauner * @base: base directory to lookup from 278100675017SChristian Brauner * @len: maximum length @len should be interpreted to 278200675017SChristian Brauner * 278300675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 278400675017SChristian Brauner * not be called by generic code. 278500675017SChristian Brauner * 278600675017SChristian Brauner * Unlike lookup_one_len, it should be called without the parent 278700675017SChristian Brauner * i_mutex held, and will take the i_mutex itself if necessary. 278800675017SChristian Brauner */ 27894609e1f1SChristian Brauner struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap, 279000675017SChristian Brauner const char *name, struct dentry *base, 279100675017SChristian Brauner int len) 279200675017SChristian Brauner { 279300675017SChristian Brauner struct qstr this; 279400675017SChristian Brauner int err; 279500675017SChristian Brauner struct dentry *ret; 279600675017SChristian Brauner 27974609e1f1SChristian Brauner err = lookup_one_common(idmap, name, base, len, &this); 279800675017SChristian Brauner if (err) 279900675017SChristian Brauner return ERR_PTR(err); 280000675017SChristian Brauner 280100675017SChristian Brauner ret = lookup_dcache(&this, base, 0); 280200675017SChristian Brauner if (!ret) 280300675017SChristian Brauner ret = lookup_slow(&this, base, 0); 280400675017SChristian Brauner return ret; 280500675017SChristian Brauner } 280600675017SChristian Brauner EXPORT_SYMBOL(lookup_one_unlocked); 280700675017SChristian Brauner 280800675017SChristian Brauner /** 280900675017SChristian Brauner * lookup_one_positive_unlocked - filesystem helper to lookup single 281000675017SChristian Brauner * pathname component 28114609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 281200675017SChristian Brauner * @name: pathname component to lookup 281300675017SChristian Brauner * @base: base directory to lookup from 281400675017SChristian Brauner * @len: maximum length @len should be interpreted to 281500675017SChristian Brauner * 281600675017SChristian Brauner * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns 281700675017SChristian Brauner * known positive or ERR_PTR(). This is what most of the users want. 281800675017SChristian Brauner * 281900675017SChristian Brauner * Note that pinned negative with unlocked parent _can_ become positive at any 282000675017SChristian Brauner * time, so callers of lookup_one_unlocked() need to be very careful; pinned 282100675017SChristian Brauner * positives have >d_inode stable, so this one avoids such problems. 282200675017SChristian Brauner * 282300675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 282400675017SChristian Brauner * not be called by generic code. 282500675017SChristian Brauner * 282600675017SChristian Brauner * The helper should be called without i_mutex held. 282700675017SChristian Brauner */ 28284609e1f1SChristian Brauner struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap, 282900675017SChristian Brauner const char *name, 283000675017SChristian Brauner struct dentry *base, int len) 283100675017SChristian Brauner { 28324609e1f1SChristian Brauner struct dentry *ret = lookup_one_unlocked(idmap, name, base, len); 283300675017SChristian Brauner 283400675017SChristian Brauner if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 283500675017SChristian Brauner dput(ret); 283600675017SChristian Brauner ret = ERR_PTR(-ENOENT); 283700675017SChristian Brauner } 283800675017SChristian Brauner return ret; 283900675017SChristian Brauner } 284000675017SChristian Brauner EXPORT_SYMBOL(lookup_one_positive_unlocked); 284100675017SChristian Brauner 284200675017SChristian Brauner /** 2843bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2844bbddca8eSNeilBrown * @name: pathname component to lookup 2845bbddca8eSNeilBrown * @base: base directory to lookup from 2846bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2847bbddca8eSNeilBrown * 2848bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2849bbddca8eSNeilBrown * not be called by generic code. 2850bbddca8eSNeilBrown * 2851bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2852bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2853bbddca8eSNeilBrown */ 2854bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2855bbddca8eSNeilBrown struct dentry *base, int len) 2856bbddca8eSNeilBrown { 28574609e1f1SChristian Brauner return lookup_one_unlocked(&nop_mnt_idmap, name, base, len); 2858bbddca8eSNeilBrown } 2859bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2860bbddca8eSNeilBrown 28616c2d4798SAl Viro /* 28626c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 28636c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 28646c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 28656c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 28666c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 28676c2d4798SAl Viro * this one avoids such problems. 28686c2d4798SAl Viro */ 28696c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28706c2d4798SAl Viro struct dentry *base, int len) 28716c2d4798SAl Viro { 28724609e1f1SChristian Brauner return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len); 28736c2d4798SAl Viro } 28746c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28756c2d4798SAl Viro 2876eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2877eedf265aSEric W. Biederman int path_pts(struct path *path) 2878eedf265aSEric W. Biederman { 2879eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2880eedf265aSEric W. Biederman * the input path. 2881eedf265aSEric W. Biederman */ 2882a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2883a6a7eb76SAl Viro struct dentry *child; 288419f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2885eedf265aSEric W. Biederman 2886a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2887a6a7eb76SAl Viro dput(parent); 288863b27720SAl Viro return -ENOENT; 2889a6a7eb76SAl Viro } 289063b27720SAl Viro dput(path->dentry); 289163b27720SAl Viro path->dentry = parent; 2892eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2893eedf265aSEric W. Biederman if (!child) 2894eedf265aSEric W. Biederman return -ENOENT; 2895eedf265aSEric W. Biederman 2896eedf265aSEric W. Biederman path->dentry = child; 2897eedf265aSEric W. Biederman dput(parent); 2898e1f19857SRichard Weinberger follow_down(path, 0); 2899eedf265aSEric W. Biederman return 0; 2900eedf265aSEric W. Biederman } 2901eedf265aSEric W. Biederman #endif 2902eedf265aSEric W. Biederman 29031fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 29041fa1e7f6SAndy Whitcroft struct path *path, int *empty) 29051da177e4SLinus Torvalds { 2906794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2907794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2908794ebceaSStephen Brennan 2909794ebceaSStephen Brennan putname(filename); 2910794ebceaSStephen Brennan return ret; 29111da177e4SLinus Torvalds } 2912b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 29131fa1e7f6SAndy Whitcroft 29149452e93eSChristian Brauner int __check_sticky(struct mnt_idmap *idmap, struct inode *dir, 2915ba73d987SChristian Brauner struct inode *inode) 29161da177e4SLinus Torvalds { 29178e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2918da9592edSDavid Howells 2919e67fe633SChristian Brauner if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, inode), fsuid)) 29201da177e4SLinus Torvalds return 0; 2921e67fe633SChristian Brauner if (vfsuid_eq_kuid(i_uid_into_vfsuid(idmap, dir), fsuid)) 29221da177e4SLinus Torvalds return 0; 29239452e93eSChristian Brauner return !capable_wrt_inode_uidgid(idmap, inode, CAP_FOWNER); 29241da177e4SLinus Torvalds } 2925cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 29261da177e4SLinus Torvalds 29271da177e4SLinus Torvalds /* 29281da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 29291da177e4SLinus Torvalds * whether the type of victim is right. 29301da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 29311da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 29321da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 29331da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 29341da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 29351da177e4SLinus Torvalds * a. be owner of dir, or 29361da177e4SLinus Torvalds * b. be owner of victim, or 29371da177e4SLinus Torvalds * c. have CAP_FOWNER capability 29381da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 29391da177e4SLinus Torvalds * links pointing to it. 29400bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 29410bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 29420bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 29430bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 29440bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 29451da177e4SLinus Torvalds * nfs_async_unlink(). 29461da177e4SLinus Torvalds */ 29474609e1f1SChristian Brauner static int may_delete(struct mnt_idmap *idmap, struct inode *dir, 2948ba73d987SChristian Brauner struct dentry *victim, bool isdir) 29491da177e4SLinus Torvalds { 295063afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 29511da177e4SLinus Torvalds int error; 29521da177e4SLinus Torvalds 2953b18825a7SDavid Howells if (d_is_negative(victim)) 29541da177e4SLinus Torvalds return -ENOENT; 2955b18825a7SDavid Howells BUG_ON(!inode); 29561da177e4SLinus Torvalds 29571da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2958593d1ce8SEric W. Biederman 2959593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2960e67fe633SChristian Brauner if (!vfsuid_valid(i_uid_into_vfsuid(idmap, inode)) || 2961e67fe633SChristian Brauner !vfsgid_valid(i_gid_into_vfsgid(idmap, inode))) 2962593d1ce8SEric W. Biederman return -EOVERFLOW; 2963593d1ce8SEric W. Biederman 29644fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29651da177e4SLinus Torvalds 29664609e1f1SChristian Brauner error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 29671da177e4SLinus Torvalds if (error) 29681da177e4SLinus Torvalds return error; 29691da177e4SLinus Torvalds if (IS_APPEND(dir)) 29701da177e4SLinus Torvalds return -EPERM; 2971b18825a7SDavid Howells 29729452e93eSChristian Brauner if (check_sticky(idmap, dir, inode) || IS_APPEND(inode) || 2973ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 29744609e1f1SChristian Brauner HAS_UNMAPPED_ID(idmap, inode)) 29751da177e4SLinus Torvalds return -EPERM; 29761da177e4SLinus Torvalds if (isdir) { 297744b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29781da177e4SLinus Torvalds return -ENOTDIR; 29791da177e4SLinus Torvalds if (IS_ROOT(victim)) 29801da177e4SLinus Torvalds return -EBUSY; 298144b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29821da177e4SLinus Torvalds return -EISDIR; 29831da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29841da177e4SLinus Torvalds return -ENOENT; 29851da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29861da177e4SLinus Torvalds return -EBUSY; 29871da177e4SLinus Torvalds return 0; 29881da177e4SLinus Torvalds } 29891da177e4SLinus Torvalds 29901da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29911da177e4SLinus Torvalds * dir. 29921da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29931da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29941da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2995036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2996036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2997036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 29981da177e4SLinus Torvalds */ 29994609e1f1SChristian Brauner static inline int may_create(struct mnt_idmap *idmap, 3000ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 30011da177e4SLinus Torvalds { 300214e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 30031da177e4SLinus Torvalds if (child->d_inode) 30041da177e4SLinus Torvalds return -EEXIST; 30051da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 30061da177e4SLinus Torvalds return -ENOENT; 30074609e1f1SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, idmap)) 3008036d5236SEric W. Biederman return -EOVERFLOW; 30098e538913SChristian Brauner 30104609e1f1SChristian Brauner return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 30111da177e4SLinus Torvalds } 30121da177e4SLinus Torvalds 30139bc37e04SAl Viro static struct dentry *lock_two_directories(struct dentry *p1, struct dentry *p2) 30141da177e4SLinus Torvalds { 30151da177e4SLinus Torvalds struct dentry *p; 30161da177e4SLinus Torvalds 3017e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 3018e2761a11SOGAWA Hirofumi if (p) { 30195955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30205955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 30211da177e4SLinus Torvalds return p; 30221da177e4SLinus Torvalds } 30231da177e4SLinus Torvalds 3024e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 3025e2761a11SOGAWA Hirofumi if (p) { 30265955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30275955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 30281da177e4SLinus Torvalds return p; 30291da177e4SLinus Torvalds } 30301da177e4SLinus Torvalds 30315955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30325955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 30331da177e4SLinus Torvalds return NULL; 30341da177e4SLinus Torvalds } 30359bc37e04SAl Viro 30369bc37e04SAl Viro /* 30379bc37e04SAl Viro * p1 and p2 should be directories on the same fs. 30389bc37e04SAl Viro */ 30399bc37e04SAl Viro struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 30409bc37e04SAl Viro { 30419bc37e04SAl Viro if (p1 == p2) { 30429bc37e04SAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30439bc37e04SAl Viro return NULL; 30449bc37e04SAl Viro } 30459bc37e04SAl Viro 30469bc37e04SAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 30479bc37e04SAl Viro return lock_two_directories(p1, p2); 30489bc37e04SAl Viro } 30494d359507SAl Viro EXPORT_SYMBOL(lock_rename); 30501da177e4SLinus Torvalds 30519bc37e04SAl Viro /* 30529bc37e04SAl Viro * c1 and p2 should be on the same fs. 30539bc37e04SAl Viro */ 30549bc37e04SAl Viro struct dentry *lock_rename_child(struct dentry *c1, struct dentry *p2) 30559bc37e04SAl Viro { 30569bc37e04SAl Viro if (READ_ONCE(c1->d_parent) == p2) { 30579bc37e04SAl Viro /* 30589bc37e04SAl Viro * hopefully won't need to touch ->s_vfs_rename_mutex at all. 30599bc37e04SAl Viro */ 30609bc37e04SAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30619bc37e04SAl Viro /* 30629bc37e04SAl Viro * now that p2 is locked, nobody can move in or out of it, 30639bc37e04SAl Viro * so the test below is safe. 30649bc37e04SAl Viro */ 30659bc37e04SAl Viro if (likely(c1->d_parent == p2)) 30669bc37e04SAl Viro return NULL; 30679bc37e04SAl Viro 30689bc37e04SAl Viro /* 30699bc37e04SAl Viro * c1 got moved out of p2 while we'd been taking locks; 30709bc37e04SAl Viro * unlock and fall back to slow case. 30719bc37e04SAl Viro */ 30729bc37e04SAl Viro inode_unlock(p2->d_inode); 30739bc37e04SAl Viro } 30749bc37e04SAl Viro 30759bc37e04SAl Viro mutex_lock(&c1->d_sb->s_vfs_rename_mutex); 30769bc37e04SAl Viro /* 30779bc37e04SAl Viro * nobody can move out of any directories on this fs. 30789bc37e04SAl Viro */ 30799bc37e04SAl Viro if (likely(c1->d_parent != p2)) 30809bc37e04SAl Viro return lock_two_directories(c1->d_parent, p2); 30819bc37e04SAl Viro 30829bc37e04SAl Viro /* 30839bc37e04SAl Viro * c1 got moved into p2 while we were taking locks; 30849bc37e04SAl Viro * we need p2 locked and ->s_vfs_rename_mutex unlocked, 30859bc37e04SAl Viro * for consistency with lock_rename(). 30869bc37e04SAl Viro */ 30879bc37e04SAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30889bc37e04SAl Viro mutex_unlock(&c1->d_sb->s_vfs_rename_mutex); 30899bc37e04SAl Viro return NULL; 30909bc37e04SAl Viro } 30919bc37e04SAl Viro EXPORT_SYMBOL(lock_rename_child); 30929bc37e04SAl Viro 30931da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 30941da177e4SLinus Torvalds { 30955955102cSAl Viro inode_unlock(p1->d_inode); 30961da177e4SLinus Torvalds if (p1 != p2) { 30975955102cSAl Viro inode_unlock(p2->d_inode); 3098fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 30991da177e4SLinus Torvalds } 31001da177e4SLinus Torvalds } 31014d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 31021da177e4SLinus Torvalds 31036521f891SChristian Brauner /** 31041639a49cSYang Xu * mode_strip_umask - handle vfs umask stripping 31051639a49cSYang Xu * @dir: parent directory of the new inode 31061639a49cSYang Xu * @mode: mode of the new inode to be created in @dir 31071639a49cSYang Xu * 31081639a49cSYang Xu * Umask stripping depends on whether or not the filesystem supports POSIX 31091639a49cSYang Xu * ACLs. If the filesystem doesn't support it umask stripping is done directly 31101639a49cSYang Xu * in here. If the filesystem does support POSIX ACLs umask stripping is 31111639a49cSYang Xu * deferred until the filesystem calls posix_acl_create(). 31121639a49cSYang Xu * 31131639a49cSYang Xu * Returns: mode 31141639a49cSYang Xu */ 31151639a49cSYang Xu static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode) 31161639a49cSYang Xu { 31171639a49cSYang Xu if (!IS_POSIXACL(dir)) 31181639a49cSYang Xu mode &= ~current_umask(); 31191639a49cSYang Xu return mode; 31201639a49cSYang Xu } 31211639a49cSYang Xu 31221639a49cSYang Xu /** 31231639a49cSYang Xu * vfs_prepare_mode - prepare the mode to be used for a new inode 31249452e93eSChristian Brauner * @idmap: idmap of the mount the inode was found from 31251639a49cSYang Xu * @dir: parent directory of the new inode 31261639a49cSYang Xu * @mode: mode of the new inode 31271639a49cSYang Xu * @mask_perms: allowed permission by the vfs 31281639a49cSYang Xu * @type: type of file to be created 31291639a49cSYang Xu * 31301639a49cSYang Xu * This helper consolidates and enforces vfs restrictions on the @mode of a new 31311639a49cSYang Xu * object to be created. 31321639a49cSYang Xu * 31331639a49cSYang Xu * Umask stripping depends on whether the filesystem supports POSIX ACLs (see 31341639a49cSYang Xu * the kernel documentation for mode_strip_umask()). Moving umask stripping 31351639a49cSYang Xu * after setgid stripping allows the same ordering for both non-POSIX ACL and 31361639a49cSYang Xu * POSIX ACL supporting filesystems. 31371639a49cSYang Xu * 31381639a49cSYang Xu * Note that it's currently valid for @type to be 0 if a directory is created. 31391639a49cSYang Xu * Filesystems raise that flag individually and we need to check whether each 31401639a49cSYang Xu * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a 31411639a49cSYang Xu * non-zero type. 31421639a49cSYang Xu * 31431639a49cSYang Xu * Returns: mode to be passed to the filesystem 31441639a49cSYang Xu */ 31459452e93eSChristian Brauner static inline umode_t vfs_prepare_mode(struct mnt_idmap *idmap, 31461639a49cSYang Xu const struct inode *dir, umode_t mode, 31471639a49cSYang Xu umode_t mask_perms, umode_t type) 31481639a49cSYang Xu { 31499452e93eSChristian Brauner mode = mode_strip_sgid(idmap, dir, mode); 31501639a49cSYang Xu mode = mode_strip_umask(dir, mode); 31511639a49cSYang Xu 31521639a49cSYang Xu /* 31531639a49cSYang Xu * Apply the vfs mandated allowed permission mask and set the type of 31541639a49cSYang Xu * file to be created before we call into the filesystem. 31551639a49cSYang Xu */ 31561639a49cSYang Xu mode &= (mask_perms & ~S_IFMT); 31571639a49cSYang Xu mode |= (type & S_IFMT); 31581639a49cSYang Xu 31591639a49cSYang Xu return mode; 31601639a49cSYang Xu } 31611639a49cSYang Xu 31621639a49cSYang Xu /** 31636521f891SChristian Brauner * vfs_create - create new file 3164abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 31656521f891SChristian Brauner * @dir: inode of @dentry 31666521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 31676521f891SChristian Brauner * @mode: mode of the new file 31686521f891SChristian Brauner * @want_excl: whether the file must not yet exist 31696521f891SChristian Brauner * 31706521f891SChristian Brauner * Create a new file. 31716521f891SChristian Brauner * 3172abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3173abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3174abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 31756521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3176abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 31776521f891SChristian Brauner */ 3178abf08576SChristian Brauner int vfs_create(struct mnt_idmap *idmap, struct inode *dir, 31796521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 31801da177e4SLinus Torvalds { 3181abf08576SChristian Brauner int error; 3182abf08576SChristian Brauner 31834609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 31841da177e4SLinus Torvalds if (error) 31851da177e4SLinus Torvalds return error; 31861da177e4SLinus Torvalds 3187acfa4380SAl Viro if (!dir->i_op->create) 31881da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 31891639a49cSYang Xu 31909452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG); 31911da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 31921da177e4SLinus Torvalds if (error) 31931da177e4SLinus Torvalds return error; 31946c960e68SChristian Brauner error = dir->i_op->create(idmap, dir, dentry, mode, want_excl); 3195a74574aaSStephen Smalley if (!error) 3196f38aa942SAmy Griffis fsnotify_create(dir, dentry); 31971da177e4SLinus Torvalds return error; 31981da177e4SLinus Torvalds } 31994d359507SAl Viro EXPORT_SYMBOL(vfs_create); 32001da177e4SLinus Torvalds 32018e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 32028e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 32038e6c848eSAl Viro void *arg) 32048e6c848eSAl Viro { 32058e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 32064609e1f1SChristian Brauner int error = may_create(&nop_mnt_idmap, dir, dentry); 32078e6c848eSAl Viro if (error) 32088e6c848eSAl Viro return error; 32098e6c848eSAl Viro 32108e6c848eSAl Viro mode &= S_IALLUGO; 32118e6c848eSAl Viro mode |= S_IFREG; 32128e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 32138e6c848eSAl Viro if (error) 32148e6c848eSAl Viro return error; 32158e6c848eSAl Viro error = f(dentry, mode, arg); 32168e6c848eSAl Viro if (!error) 32178e6c848eSAl Viro fsnotify_create(dir, dentry); 32188e6c848eSAl Viro return error; 32198e6c848eSAl Viro } 32208e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 32218e6c848eSAl Viro 3222a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3223a2982cc9SEric W. Biederman { 3224a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3225a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3226a2982cc9SEric W. Biederman } 3227a2982cc9SEric W. Biederman 32284609e1f1SChristian Brauner static int may_open(struct mnt_idmap *idmap, const struct path *path, 3229ba73d987SChristian Brauner int acc_mode, int flag) 32301da177e4SLinus Torvalds { 32313fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 32321da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 32331da177e4SLinus Torvalds int error; 32341da177e4SLinus Torvalds 32351da177e4SLinus Torvalds if (!inode) 32361da177e4SLinus Torvalds return -ENOENT; 32371da177e4SLinus Torvalds 3238c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3239c8fe8f30SChristoph Hellwig case S_IFLNK: 32401da177e4SLinus Torvalds return -ELOOP; 3241c8fe8f30SChristoph Hellwig case S_IFDIR: 3242fc4177beSKees Cook if (acc_mode & MAY_WRITE) 32431da177e4SLinus Torvalds return -EISDIR; 3244fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3245fc4177beSKees Cook return -EACCES; 3246c8fe8f30SChristoph Hellwig break; 3247c8fe8f30SChristoph Hellwig case S_IFBLK: 3248c8fe8f30SChristoph Hellwig case S_IFCHR: 3249a2982cc9SEric W. Biederman if (!may_open_dev(path)) 32501da177e4SLinus Torvalds return -EACCES; 3251633fb6acSKees Cook fallthrough; 3252c8fe8f30SChristoph Hellwig case S_IFIFO: 3253c8fe8f30SChristoph Hellwig case S_IFSOCK: 3254633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3255633fb6acSKees Cook return -EACCES; 32561da177e4SLinus Torvalds flag &= ~O_TRUNC; 3257c8fe8f30SChristoph Hellwig break; 32580fd338b2SKees Cook case S_IFREG: 32590fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 32600fd338b2SKees Cook return -EACCES; 32610fd338b2SKees Cook break; 32624a3fd211SDave Hansen } 3263b41572e9SDave Hansen 32644609e1f1SChristian Brauner error = inode_permission(idmap, inode, MAY_OPEN | acc_mode); 3265b41572e9SDave Hansen if (error) 3266b41572e9SDave Hansen return error; 32676146f0d5SMimi Zohar 32681da177e4SLinus Torvalds /* 32691da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 32701da177e4SLinus Torvalds */ 32711da177e4SLinus Torvalds if (IS_APPEND(inode)) { 32728737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 32737715b521SAl Viro return -EPERM; 32741da177e4SLinus Torvalds if (flag & O_TRUNC) 32757715b521SAl Viro return -EPERM; 32761da177e4SLinus Torvalds } 32771da177e4SLinus Torvalds 32781da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 327901beba79SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode)) 32807715b521SAl Viro return -EPERM; 32811da177e4SLinus Torvalds 3282f3c7691eSJ. Bruce Fields return 0; 32837715b521SAl Viro } 32847715b521SAl Viro 3285abf08576SChristian Brauner static int handle_truncate(struct mnt_idmap *idmap, struct file *filp) 32867715b521SAl Viro { 3287f0bb5aafSAl Viro const struct path *path = &filp->f_path; 32887715b521SAl Viro struct inode *inode = path->dentry->d_inode; 32897715b521SAl Viro int error = get_write_access(inode); 32901da177e4SLinus Torvalds if (error) 32917715b521SAl Viro return error; 3292482e0007SJeff Layton 32933350607dSGünther Noack error = security_file_truncate(filp); 32941da177e4SLinus Torvalds if (!error) { 3295abf08576SChristian Brauner error = do_truncate(idmap, path->dentry, 0, 3296d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3297e1181ee6SJeff Layton filp); 32981da177e4SLinus Torvalds } 32991da177e4SLinus Torvalds put_write_access(inode); 3300acd0c935SMimi Zohar return error; 33011da177e4SLinus Torvalds } 33021da177e4SLinus Torvalds 3303d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3304d57999e1SDave Hansen { 33058a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 33068a5e929dSAl Viro flag--; 3307d57999e1SDave Hansen return flag; 3308d57999e1SDave Hansen } 3309d57999e1SDave Hansen 33104609e1f1SChristian Brauner static int may_o_create(struct mnt_idmap *idmap, 3311ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3312ba73d987SChristian Brauner umode_t mode) 3313d18e9008SMiklos Szeredi { 3314d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3315d18e9008SMiklos Szeredi if (error) 3316d18e9008SMiklos Szeredi return error; 3317d18e9008SMiklos Szeredi 33184609e1f1SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap)) 33191328c727SSeth Forshee return -EOVERFLOW; 33201328c727SSeth Forshee 33214609e1f1SChristian Brauner error = inode_permission(idmap, dir->dentry->d_inode, 332247291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3323d18e9008SMiklos Szeredi if (error) 3324d18e9008SMiklos Szeredi return error; 3325d18e9008SMiklos Szeredi 3326d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3327d18e9008SMiklos Szeredi } 3328d18e9008SMiklos Szeredi 33291acf0af9SDavid Howells /* 33301acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 33311acf0af9SDavid Howells * dentry. 33321acf0af9SDavid Howells * 33331acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 33341acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 33351acf0af9SDavid Howells * 333600a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 333700a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 333800a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 33391acf0af9SDavid Howells * 33401acf0af9SDavid Howells * Returns an error code otherwise. 33411acf0af9SDavid Howells */ 3342239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3343239eb983SAl Viro struct file *file, 33443ec2eef1SAl Viro int open_flag, umode_t mode) 3345d18e9008SMiklos Szeredi { 3346d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3347d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3348d18e9008SMiklos Szeredi int error; 3349d18e9008SMiklos Szeredi 3350d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3351d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3352d18e9008SMiklos Szeredi 335330d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 335430d90494SAl Viro file->f_path.mnt = nd->path.mnt; 33550fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 335644907d79SAl Viro open_to_namei_flags(open_flag), mode); 33576fbd0714SAl Viro d_lookup_done(dentry); 3358384f26e2SAl Viro if (!error) { 335964e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 33606fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 33616fb968cdSAl Viro dput(dentry); 33626fb968cdSAl Viro dentry = dget(file->f_path.dentry); 33636fb968cdSAl Viro } 336464e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 33652675a4ebSAl Viro error = -EIO; 3366384f26e2SAl Viro } else { 336730d90494SAl Viro if (file->f_path.dentry) { 3368d18e9008SMiklos Szeredi dput(dentry); 3369d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 337010c64ceaSAl Viro } 3371239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3372a01e718fSAl Viro error = -ENOENT; 3373d18e9008SMiklos Szeredi } 3374d18e9008SMiklos Szeredi } 3375239eb983SAl Viro if (error) { 3376d18e9008SMiklos Szeredi dput(dentry); 3377239eb983SAl Viro dentry = ERR_PTR(error); 3378239eb983SAl Viro } 3379239eb983SAl Viro return dentry; 3380d18e9008SMiklos Szeredi } 3381d18e9008SMiklos Szeredi 338231e6b01fSNick Piggin /* 33831acf0af9SDavid Howells * Look up and maybe create and open the last component. 3384d58ffd35SMiklos Szeredi * 338500a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3386d58ffd35SMiklos Szeredi * 338700a07c15SAl Viro * Returns 0 on success, that is, if 338800a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 338900a07c15SAl Viro * the file was not completely opened at this time, though lookups and 339000a07c15SAl Viro * creations were performed. 339100a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 339200a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 339300a07c15SAl Viro * hadn't been specified. 33941acf0af9SDavid Howells * 339500a07c15SAl Viro * An error code is returned on failure. 3396d58ffd35SMiklos Szeredi */ 3397da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3398d58ffd35SMiklos Szeredi const struct open_flags *op, 33993ec2eef1SAl Viro bool got_write) 3400d58ffd35SMiklos Szeredi { 34016c960e68SChristian Brauner struct mnt_idmap *idmap; 3402d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 340354ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 34041643b43fSAl Viro int open_flag = op->open_flag; 3405d58ffd35SMiklos Szeredi struct dentry *dentry; 34061643b43fSAl Viro int error, create_error = 0; 34071643b43fSAl Viro umode_t mode = op->mode; 34086fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3409d58ffd35SMiklos Szeredi 3410ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3411da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3412d58ffd35SMiklos Szeredi 341373a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 34146fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 34156fbd0714SAl Viro for (;;) { 34166fbd0714SAl Viro if (!dentry) { 34176fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3418d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3419da5ebf5aSAl Viro return dentry; 34206fbd0714SAl Viro } 34216fbd0714SAl Viro if (d_in_lookup(dentry)) 34226fbd0714SAl Viro break; 3423d58ffd35SMiklos Szeredi 34246fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 34256fbd0714SAl Viro if (likely(error > 0)) 34266fbd0714SAl Viro break; 34276fbd0714SAl Viro if (error) 34286fbd0714SAl Viro goto out_dput; 34296fbd0714SAl Viro d_invalidate(dentry); 34306fbd0714SAl Viro dput(dentry); 34316fbd0714SAl Viro dentry = NULL; 34326fbd0714SAl Viro } 34336fbd0714SAl Viro if (dentry->d_inode) { 3434d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3435da5ebf5aSAl Viro return dentry; 34366c51e513SAl Viro } 3437d18e9008SMiklos Szeredi 34381643b43fSAl Viro /* 34391643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 34401643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 34411643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 34421643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 34431643b43fSAl Viro * 34441643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 34451643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 34461643b43fSAl Viro */ 344799a4a90cSAl Viro if (unlikely(!got_write)) 344899a4a90cSAl Viro open_flag &= ~O_TRUNC; 34496c960e68SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 34501643b43fSAl Viro if (open_flag & O_CREAT) { 345199a4a90cSAl Viro if (open_flag & O_EXCL) 345299a4a90cSAl Viro open_flag &= ~O_TRUNC; 34539452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir->d_inode, mode, mode, mode); 345499a4a90cSAl Viro if (likely(got_write)) 34554609e1f1SChristian Brauner create_error = may_o_create(idmap, &nd->path, 3456ba73d987SChristian Brauner dentry, mode); 345799a4a90cSAl Viro else 345899a4a90cSAl Viro create_error = -EROFS; 345999a4a90cSAl Viro } 346099a4a90cSAl Viro if (create_error) 34611643b43fSAl Viro open_flag &= ~O_CREAT; 34621643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3463d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3464da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3465da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3466da5ebf5aSAl Viro return dentry; 3467239eb983SAl Viro } 346854ef4872SMiklos Szeredi 34696fbd0714SAl Viro if (d_in_lookup(dentry)) { 347012fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 347112fa5e24SAl Viro nd->flags); 34726fbd0714SAl Viro d_lookup_done(dentry); 347312fa5e24SAl Viro if (unlikely(res)) { 347412fa5e24SAl Viro if (IS_ERR(res)) { 347512fa5e24SAl Viro error = PTR_ERR(res); 347612fa5e24SAl Viro goto out_dput; 347712fa5e24SAl Viro } 347812fa5e24SAl Viro dput(dentry); 347912fa5e24SAl Viro dentry = res; 348012fa5e24SAl Viro } 348154ef4872SMiklos Szeredi } 348254ef4872SMiklos Szeredi 3483d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 34841643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 348573a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3486ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3487ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3488ce8644fcSAl Viro error = -EACCES; 3489d58ffd35SMiklos Szeredi goto out_dput; 349064894cf8SAl Viro } 3491549c7297SChristian Brauner 34926c960e68SChristian Brauner error = dir_inode->i_op->create(idmap, dir_inode, dentry, 3493549c7297SChristian Brauner mode, open_flag & O_EXCL); 3494d58ffd35SMiklos Szeredi if (error) 3495d58ffd35SMiklos Szeredi goto out_dput; 3496d58ffd35SMiklos Szeredi } 34971643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 34981643b43fSAl Viro error = create_error; 3499d58ffd35SMiklos Szeredi goto out_dput; 3500d58ffd35SMiklos Szeredi } 3501da5ebf5aSAl Viro return dentry; 3502d58ffd35SMiklos Szeredi 3503d58ffd35SMiklos Szeredi out_dput: 3504d58ffd35SMiklos Szeredi dput(dentry); 3505da5ebf5aSAl Viro return ERR_PTR(error); 3506d58ffd35SMiklos Szeredi } 3507d58ffd35SMiklos Szeredi 3508c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 35093ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3510fb1cc555SAl Viro { 3511a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3512ca344a89SAl Viro int open_flag = op->open_flag; 351364894cf8SAl Viro bool got_write = false; 3514da5ebf5aSAl Viro struct dentry *dentry; 3515b0417d2cSAl Viro const char *res; 3516fb1cc555SAl Viro 3517c3e380b0SAl Viro nd->flags |= op->intent; 3518c3e380b0SAl Viro 3519bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 352056676ec3SAl Viro if (nd->depth) 352156676ec3SAl Viro put_link(nd); 3522ff326a32SAl Viro return handle_dots(nd, nd->last_type); 35231f36f774SAl Viro } 3524a2c36b45SAl Viro 3525ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3526fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3527fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3528fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 35294cb64024SAl Viro dentry = lookup_fast(nd); 353020e34357SAl Viro if (IS_ERR(dentry)) 35311ccac622SAl Viro return ERR_CAST(dentry); 353220e34357SAl Viro if (likely(dentry)) 353371574865SMiklos Szeredi goto finish_lookup; 353471574865SMiklos Szeredi 35356583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3536b6183df7SMiklos Szeredi } else { 3537fe2d35ffSAl Viro /* create side of things */ 353872287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3539e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3540e36cffedSJens Axboe return ERR_PTR(-ECHILD); 354172287417SAl Viro } 3542c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 35431f36f774SAl Viro /* trailing slashes? */ 3544deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 35451ccac622SAl Viro return ERR_PTR(-EISDIR); 3546b6183df7SMiklos Szeredi } 35471f36f774SAl Viro 35489cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3549e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 355064894cf8SAl Viro /* 355164894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 355264894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 355364894cf8SAl Viro * dropping this one anyway. 355464894cf8SAl Viro */ 355564894cf8SAl Viro } 35569cf843e3SAl Viro if (open_flag & O_CREAT) 35575955102cSAl Viro inode_lock(dir->d_inode); 35589cf843e3SAl Viro else 35599cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3560da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3561f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3562f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 35639cf843e3SAl Viro if (open_flag & O_CREAT) 35645955102cSAl Viro inode_unlock(dir->d_inode); 35659cf843e3SAl Viro else 35669cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3567fb1cc555SAl Viro 3568c981a482SAl Viro if (got_write) 356959e96e65SAl Viro mnt_drop_write(nd->path.mnt); 35706c0d46c4SAl Viro 357159e96e65SAl Viro if (IS_ERR(dentry)) 357259e96e65SAl Viro return ERR_CAST(dentry); 357359e96e65SAl Viro 3574973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3575e73cabffSAl Viro dput(nd->path.dentry); 3576e73cabffSAl Viro nd->path.dentry = dentry; 3577c981a482SAl Viro return NULL; 3578fb1cc555SAl Viro } 3579fb1cc555SAl Viro 358020e34357SAl Viro finish_lookup: 358156676ec3SAl Viro if (nd->depth) 358256676ec3SAl Viro put_link(nd); 3583a4f5b521SAl Viro res = step_into(nd, WALK_TRAILING, dentry); 3584ff326a32SAl Viro if (unlikely(res)) 35851ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3586b0417d2cSAl Viro return res; 35871ccac622SAl Viro } 358831d1726dSAl Viro 3589c981a482SAl Viro /* 3590c981a482SAl Viro * Handle the last step of open() 3591c981a482SAl Viro */ 3592c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3593c981a482SAl Viro struct file *file, const struct open_flags *op) 3594c981a482SAl Viro { 3595abf08576SChristian Brauner struct mnt_idmap *idmap; 3596c981a482SAl Viro int open_flag = op->open_flag; 3597c981a482SAl Viro bool do_truncate; 3598c981a482SAl Viro int acc_mode; 3599c981a482SAl Viro int error; 3600c981a482SAl Viro 3601ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3602ff326a32SAl Viro error = complete_walk(nd); 3603ff326a32SAl Viro if (error) 3604ff326a32SAl Viro return error; 3605ff326a32SAl Viro } 3606973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 360776ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3608abf08576SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 360930aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3610b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3611b94e0b32SAl Viro return -EEXIST; 361230aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3613c5971b8cSAl Viro return -EISDIR; 3614e67fe633SChristian Brauner error = may_create_in_sticky(idmap, nd, 361530aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 361630aba665SSalvatore Mesoraca if (unlikely(error)) 3617c5971b8cSAl Viro return error; 361830aba665SSalvatore Mesoraca } 361944b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3620c5971b8cSAl Viro return -ENOTDIR; 36216c0d46c4SAl Viro 36228795e7d4SAl Viro do_truncate = false; 36238795e7d4SAl Viro acc_mode = op->acc_mode; 36245a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 36255a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 36265a2d3eddSAl Viro open_flag &= ~O_TRUNC; 36275a2d3eddSAl Viro acc_mode = 0; 36288795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 36290f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 36300f9d1a10SAl Viro if (error) 3631c5971b8cSAl Viro return error; 36328795e7d4SAl Viro do_truncate = true; 36330f9d1a10SAl Viro } 36344609e1f1SChristian Brauner error = may_open(idmap, &nd->path, acc_mode, open_flag); 36358795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3636ae2bb293SAl Viro error = vfs_open(&nd->path, file); 36378795e7d4SAl Viro if (!error) 36386035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 36398795e7d4SAl Viro if (!error && do_truncate) 3640abf08576SChristian Brauner error = handle_truncate(idmap, file); 3641c80567c8SAl Viro if (unlikely(error > 0)) { 3642c80567c8SAl Viro WARN_ON(1); 3643c80567c8SAl Viro error = -EINVAL; 3644c80567c8SAl Viro } 36458795e7d4SAl Viro if (do_truncate) 36460f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3647c5971b8cSAl Viro return error; 3648fb1cc555SAl Viro } 3649fb1cc555SAl Viro 36506521f891SChristian Brauner /** 36516521f891SChristian Brauner * vfs_tmpfile - create tmpfile 3652abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 36536521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 36546521f891SChristian Brauner * @mode: mode of the new tmpfile 36552111c3c0SRandy Dunlap * @open_flag: flags 36566521f891SChristian Brauner * 36576521f891SChristian Brauner * Create a temporary file. 36586521f891SChristian Brauner * 3659abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3660abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3661abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 36626521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3663abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 36646521f891SChristian Brauner */ 3665abf08576SChristian Brauner static int vfs_tmpfile(struct mnt_idmap *idmap, 36669751b338SMiklos Szeredi const struct path *parentpath, 36679751b338SMiklos Szeredi struct file *file, umode_t mode) 3668af7bd4dcSAmir Goldstein { 36699751b338SMiklos Szeredi struct dentry *child; 36709751b338SMiklos Szeredi struct inode *dir = d_inode(parentpath->dentry); 3671af7bd4dcSAmir Goldstein struct inode *inode; 3672af7bd4dcSAmir Goldstein int error; 3673406c706cSPeter Griffin int open_flag = file->f_flags; 3674af7bd4dcSAmir Goldstein 3675af7bd4dcSAmir Goldstein /* we want directory to be writable */ 36764609e1f1SChristian Brauner error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 3677af7bd4dcSAmir Goldstein if (error) 36789751b338SMiklos Szeredi return error; 3679af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 36809751b338SMiklos Szeredi return -EOPNOTSUPP; 36819751b338SMiklos Szeredi child = d_alloc(parentpath->dentry, &slash_name); 3682af7bd4dcSAmir Goldstein if (unlikely(!child)) 36839751b338SMiklos Szeredi return -ENOMEM; 36849751b338SMiklos Szeredi file->f_path.mnt = parentpath->mnt; 36859751b338SMiklos Szeredi file->f_path.dentry = child; 36869452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir, mode, mode, mode); 3687011e2b71SChristian Brauner error = dir->i_op->tmpfile(idmap, dir, file, mode); 36889751b338SMiklos Szeredi dput(child); 3689af7bd4dcSAmir Goldstein if (error) 36909751b338SMiklos Szeredi return error; 36919751b338SMiklos Szeredi /* Don't check for other permissions, the inode was just created */ 36924609e1f1SChristian Brauner error = may_open(idmap, &file->f_path, 0, file->f_flags); 36939751b338SMiklos Szeredi if (error) 36949751b338SMiklos Szeredi return error; 36959751b338SMiklos Szeredi inode = file_inode(file); 3696406c706cSPeter Griffin if (!(open_flag & O_EXCL)) { 3697af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3698af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3699af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3700af7bd4dcSAmir Goldstein } 370139f60c1cSChristian Brauner ima_post_create_tmpfile(idmap, inode); 37029751b338SMiklos Szeredi return 0; 3703af7bd4dcSAmir Goldstein } 3704af7bd4dcSAmir Goldstein 370522873deaSMiklos Szeredi /** 370622873deaSMiklos Szeredi * vfs_tmpfile_open - open a tmpfile for kernel internal use 3707abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 370822873deaSMiklos Szeredi * @parentpath: path of the base directory 370922873deaSMiklos Szeredi * @mode: mode of the new tmpfile 371022873deaSMiklos Szeredi * @open_flag: flags 371122873deaSMiklos Szeredi * @cred: credentials for open 371222873deaSMiklos Szeredi * 371322873deaSMiklos Szeredi * Create and open a temporary file. The file is not accounted in nr_files, 371422873deaSMiklos Szeredi * hence this is only for kernel internal use, and must not be installed into 371522873deaSMiklos Szeredi * file tables or such. 371622873deaSMiklos Szeredi */ 3717abf08576SChristian Brauner struct file *vfs_tmpfile_open(struct mnt_idmap *idmap, 371822873deaSMiklos Szeredi const struct path *parentpath, 371922873deaSMiklos Szeredi umode_t mode, int open_flag, const struct cred *cred) 372022873deaSMiklos Szeredi { 372122873deaSMiklos Szeredi struct file *file; 372222873deaSMiklos Szeredi int error; 372322873deaSMiklos Szeredi 37249751b338SMiklos Szeredi file = alloc_empty_file_noaccount(open_flag, cred); 37259751b338SMiklos Szeredi if (!IS_ERR(file)) { 3726abf08576SChristian Brauner error = vfs_tmpfile(idmap, parentpath, file, mode); 37279751b338SMiklos Szeredi if (error) { 37289751b338SMiklos Szeredi fput(file); 372922873deaSMiklos Szeredi file = ERR_PTR(error); 37309751b338SMiklos Szeredi } 37319751b338SMiklos Szeredi } 373222873deaSMiklos Szeredi return file; 373322873deaSMiklos Szeredi } 373422873deaSMiklos Szeredi EXPORT_SYMBOL(vfs_tmpfile_open); 3735648fa861SAl Viro 3736c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 373760545d0dSAl Viro const struct open_flags *op, 37383ec2eef1SAl Viro struct file *file) 373960545d0dSAl Viro { 3740625b6d10SAl Viro struct path path; 3741c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 37429751b338SMiklos Szeredi 374360545d0dSAl Viro if (unlikely(error)) 374460545d0dSAl Viro return error; 3745625b6d10SAl Viro error = mnt_want_write(path.mnt); 374660545d0dSAl Viro if (unlikely(error)) 374760545d0dSAl Viro goto out; 3748abf08576SChristian Brauner error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode); 37499751b338SMiklos Szeredi if (error) 375060545d0dSAl Viro goto out2; 37519751b338SMiklos Szeredi audit_inode(nd->name, file->f_path.dentry, 0); 375260545d0dSAl Viro out2: 3753625b6d10SAl Viro mnt_drop_write(path.mnt); 375460545d0dSAl Viro out: 3755625b6d10SAl Viro path_put(&path); 375660545d0dSAl Viro return error; 375760545d0dSAl Viro } 375860545d0dSAl Viro 37596ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 37606ac08709SAl Viro { 37616ac08709SAl Viro struct path path; 37626ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 37636ac08709SAl Viro if (!error) { 37646ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3765ae2bb293SAl Viro error = vfs_open(&path, file); 37666ac08709SAl Viro path_put(&path); 37676ac08709SAl Viro } 37686ac08709SAl Viro return error; 37696ac08709SAl Viro } 37706ac08709SAl Viro 3771c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3772c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 37731da177e4SLinus Torvalds { 377430d90494SAl Viro struct file *file; 377513aab428SAl Viro int error; 377631e6b01fSNick Piggin 3777ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 37781afc99beSAl Viro if (IS_ERR(file)) 37791afc99beSAl Viro return file; 378031e6b01fSNick Piggin 3781bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 37823ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 37835f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 37846ac08709SAl Viro error = do_o_path(nd, flags, file); 37855f336e72SAl Viro } else { 37865f336e72SAl Viro const char *s = path_init(nd, flags); 37873bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3788c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 37891ccac622SAl Viro ; 3790c5971b8cSAl Viro if (!error) 3791c5971b8cSAl Viro error = do_open(nd, file, op); 3792deb106c6SAl Viro terminate_walk(nd); 37935f336e72SAl Viro } 37947c1c01ecSAl Viro if (likely(!error)) { 3795aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 37967c1c01ecSAl Viro return file; 37977c1c01ecSAl Viro WARN_ON(1); 37987c1c01ecSAl Viro error = -EINVAL; 3799015c3bbcSMiklos Szeredi } 38007c1c01ecSAl Viro fput(file); 38012675a4ebSAl Viro if (error == -EOPENSTALE) { 38022675a4ebSAl Viro if (flags & LOOKUP_RCU) 38032675a4ebSAl Viro error = -ECHILD; 38042675a4ebSAl Viro else 38052675a4ebSAl Viro error = -ESTALE; 38062675a4ebSAl Viro } 38077c1c01ecSAl Viro return ERR_PTR(error); 3808de459215SKirill Korotaev } 38091da177e4SLinus Torvalds 3810669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3811f9652e10SAl Viro const struct open_flags *op) 381213aab428SAl Viro { 38139883d185SAl Viro struct nameidata nd; 3814f9652e10SAl Viro int flags = op->lookup_flags; 381513aab428SAl Viro struct file *filp; 381613aab428SAl Viro 381706422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3818c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 381913aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3820c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 382113aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3822c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 38239883d185SAl Viro restore_nameidata(); 382413aab428SAl Viro return filp; 382513aab428SAl Viro } 382613aab428SAl Viro 3827ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3828f9652e10SAl Viro const char *name, const struct open_flags *op) 382973d049a4SAl Viro { 38309883d185SAl Viro struct nameidata nd; 383173d049a4SAl Viro struct file *file; 383251689104SPaul Moore struct filename *filename; 3833bcba1e7dSAl Viro int flags = op->lookup_flags; 383473d049a4SAl Viro 3835ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 383673d049a4SAl Viro return ERR_PTR(-ELOOP); 383773d049a4SAl Viro 383851689104SPaul Moore filename = getname_kernel(name); 3839a1c83681SViresh Kumar if (IS_ERR(filename)) 384051689104SPaul Moore return ERR_CAST(filename); 384151689104SPaul Moore 384206422964SAl Viro set_nameidata(&nd, -1, filename, root); 3843c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 384473d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3845c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 384673d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3847c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 38489883d185SAl Viro restore_nameidata(); 384951689104SPaul Moore putname(filename); 385073d049a4SAl Viro return file; 385173d049a4SAl Viro } 385273d049a4SAl Viro 3853b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 38541ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 38551da177e4SLinus Torvalds { 3856c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3857391172c4SAl Viro struct qstr last; 3858b3d4650dSNeilBrown bool want_dir = lookup_flags & LOOKUP_DIRECTORY; 3859b3d4650dSNeilBrown unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; 3860b3d4650dSNeilBrown unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; 3861391172c4SAl Viro int type; 3862c30dabfeSJan Kara int err2; 38631ac12b4bSJeff Layton int error; 38641ac12b4bSJeff Layton 3865b3d4650dSNeilBrown error = filename_parentat(dfd, name, reval_flag, path, &last, &type); 38660ee50b47SDmitry Kadashev if (error) 38670ee50b47SDmitry Kadashev return ERR_PTR(error); 38681da177e4SLinus Torvalds 3869c663e5d8SChristoph Hellwig /* 3870c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3871c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3872c663e5d8SChristoph Hellwig */ 38735c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3874ed75e95dSAl Viro goto out; 3875c663e5d8SChristoph Hellwig 3876c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3877391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3878c663e5d8SChristoph Hellwig /* 3879b3d4650dSNeilBrown * Do the final lookup. Suppress 'create' if there is a trailing 3880b3d4650dSNeilBrown * '/', and a directory wasn't requested. 3881c663e5d8SChristoph Hellwig */ 3882b3d4650dSNeilBrown if (last.name[last.len] && !want_dir) 3883b3d4650dSNeilBrown create_flags = 0; 38845955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3885*74d7970fSNamjae Jeon dentry = lookup_one_qstr_excl(&last, path->dentry, 3886*74d7970fSNamjae Jeon reval_flag | create_flags); 38871da177e4SLinus Torvalds if (IS_ERR(dentry)) 3888a8104a9fSAl Viro goto unlock; 3889c663e5d8SChristoph Hellwig 3890a8104a9fSAl Viro error = -EEXIST; 3891b18825a7SDavid Howells if (d_is_positive(dentry)) 3892a8104a9fSAl Viro goto fail; 3893b18825a7SDavid Howells 3894c663e5d8SChristoph Hellwig /* 3895c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3896c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3897c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3898c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3899c663e5d8SChristoph Hellwig */ 3900b3d4650dSNeilBrown if (unlikely(!create_flags)) { 3901a8104a9fSAl Viro error = -ENOENT; 3902ed75e95dSAl Viro goto fail; 3903e9baf6e5SAl Viro } 3904c30dabfeSJan Kara if (unlikely(err2)) { 3905c30dabfeSJan Kara error = err2; 3906a8104a9fSAl Viro goto fail; 3907c30dabfeSJan Kara } 3908e9baf6e5SAl Viro return dentry; 39091da177e4SLinus Torvalds fail: 3910a8104a9fSAl Viro dput(dentry); 3911a8104a9fSAl Viro dentry = ERR_PTR(error); 3912a8104a9fSAl Viro unlock: 39135955102cSAl Viro inode_unlock(path->dentry->d_inode); 3914c30dabfeSJan Kara if (!err2) 3915391172c4SAl Viro mnt_drop_write(path->mnt); 3916ed75e95dSAl Viro out: 3917391172c4SAl Viro path_put(path); 3918ed75e95dSAl Viro return dentry; 3919dae6ad8fSAl Viro } 3920fa14a0b8SAl Viro 3921fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3922fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3923fa14a0b8SAl Viro { 3924b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3925b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3926b4a4f213SStephen Brennan 3927b4a4f213SStephen Brennan putname(filename); 3928b4a4f213SStephen Brennan return res; 3929fa14a0b8SAl Viro } 3930dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3931dae6ad8fSAl Viro 3932921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3933921a1650SAl Viro { 3934921a1650SAl Viro dput(dentry); 39355955102cSAl Viro inode_unlock(path->dentry->d_inode); 3936a8104a9fSAl Viro mnt_drop_write(path->mnt); 3937921a1650SAl Viro path_put(path); 3938921a1650SAl Viro } 3939921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3940921a1650SAl Viro 3941520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 39421ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3943dae6ad8fSAl Viro { 3944b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3945b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3946b4a4f213SStephen Brennan 3947b4a4f213SStephen Brennan putname(filename); 3948b4a4f213SStephen Brennan return res; 3949dae6ad8fSAl Viro } 3950dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3951dae6ad8fSAl Viro 39526521f891SChristian Brauner /** 39536521f891SChristian Brauner * vfs_mknod - create device node or file 3954abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 39556521f891SChristian Brauner * @dir: inode of @dentry 39566521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 39576521f891SChristian Brauner * @mode: mode of the new device node or file 39586521f891SChristian Brauner * @dev: device number of device to create 39596521f891SChristian Brauner * 39606521f891SChristian Brauner * Create a device node or file. 39616521f891SChristian Brauner * 3962abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3963abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3964abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 39656521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3966abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 39676521f891SChristian Brauner */ 3968abf08576SChristian Brauner int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir, 39696521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 39701da177e4SLinus Torvalds { 3971a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 39724609e1f1SChristian Brauner int error = may_create(idmap, dir, dentry); 39731da177e4SLinus Torvalds 39741da177e4SLinus Torvalds if (error) 39751da177e4SLinus Torvalds return error; 39761da177e4SLinus Torvalds 3977a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3978a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 39791da177e4SLinus Torvalds return -EPERM; 39801da177e4SLinus Torvalds 3981acfa4380SAl Viro if (!dir->i_op->mknod) 39821da177e4SLinus Torvalds return -EPERM; 39831da177e4SLinus Torvalds 39849452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir, mode, mode, mode); 398508ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 398608ce5f16SSerge E. Hallyn if (error) 398708ce5f16SSerge E. Hallyn return error; 398808ce5f16SSerge E. Hallyn 39891da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 39901da177e4SLinus Torvalds if (error) 39911da177e4SLinus Torvalds return error; 39921da177e4SLinus Torvalds 39935ebb29beSChristian Brauner error = dir->i_op->mknod(idmap, dir, dentry, mode, dev); 3994a74574aaSStephen Smalley if (!error) 3995f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39961da177e4SLinus Torvalds return error; 39971da177e4SLinus Torvalds } 39984d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 39991da177e4SLinus Torvalds 4000f69aac00SAl Viro static int may_mknod(umode_t mode) 4001463c3197SDave Hansen { 4002463c3197SDave Hansen switch (mode & S_IFMT) { 4003463c3197SDave Hansen case S_IFREG: 4004463c3197SDave Hansen case S_IFCHR: 4005463c3197SDave Hansen case S_IFBLK: 4006463c3197SDave Hansen case S_IFIFO: 4007463c3197SDave Hansen case S_IFSOCK: 4008463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 4009463c3197SDave Hansen return 0; 4010463c3197SDave Hansen case S_IFDIR: 4011463c3197SDave Hansen return -EPERM; 4012463c3197SDave Hansen default: 4013463c3197SDave Hansen return -EINVAL; 4014463c3197SDave Hansen } 4015463c3197SDave Hansen } 4016463c3197SDave Hansen 401745f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 401887c4e192SDominik Brodowski unsigned int dev) 40191da177e4SLinus Torvalds { 4020abf08576SChristian Brauner struct mnt_idmap *idmap; 40211da177e4SLinus Torvalds struct dentry *dentry; 4022dae6ad8fSAl Viro struct path path; 4023dae6ad8fSAl Viro int error; 4024972567f1SJeff Layton unsigned int lookup_flags = 0; 40251da177e4SLinus Torvalds 40268e4bfca1SAl Viro error = may_mknod(mode); 40278e4bfca1SAl Viro if (error) 40287797251bSDmitry Kadashev goto out1; 4029972567f1SJeff Layton retry: 4030b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 40317797251bSDmitry Kadashev error = PTR_ERR(dentry); 4032dae6ad8fSAl Viro if (IS_ERR(dentry)) 40337797251bSDmitry Kadashev goto out1; 40342ad94ae6SAl Viro 40351639a49cSYang Xu error = security_path_mknod(&path, dentry, 40361639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode), dev); 4037be6d3e56SKentaro Takeda if (error) 40387797251bSDmitry Kadashev goto out2; 40396521f891SChristian Brauner 4040abf08576SChristian Brauner idmap = mnt_idmap(path.mnt); 40411da177e4SLinus Torvalds switch (mode & S_IFMT) { 40421da177e4SLinus Torvalds case 0: case S_IFREG: 4043abf08576SChristian Brauner error = vfs_create(idmap, path.dentry->d_inode, 40446521f891SChristian Brauner dentry, mode, true); 404505d1a717SMimi Zohar if (!error) 404639f60c1cSChristian Brauner ima_post_path_mknod(idmap, dentry); 40471da177e4SLinus Torvalds break; 40481da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 4049abf08576SChristian Brauner error = vfs_mknod(idmap, path.dentry->d_inode, 40506521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 40511da177e4SLinus Torvalds break; 40521da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 4053abf08576SChristian Brauner error = vfs_mknod(idmap, path.dentry->d_inode, 40546521f891SChristian Brauner dentry, mode, 0); 40551da177e4SLinus Torvalds break; 40561da177e4SLinus Torvalds } 40577797251bSDmitry Kadashev out2: 4058921a1650SAl Viro done_path_create(&path, dentry); 4059972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 4060972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 4061972567f1SJeff Layton goto retry; 4062972567f1SJeff Layton } 40637797251bSDmitry Kadashev out1: 40647797251bSDmitry Kadashev putname(name); 40651da177e4SLinus Torvalds return error; 40661da177e4SLinus Torvalds } 40671da177e4SLinus Torvalds 406887c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 406987c4e192SDominik Brodowski unsigned int, dev) 407087c4e192SDominik Brodowski { 40717797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 407287c4e192SDominik Brodowski } 407387c4e192SDominik Brodowski 40748208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 40755590ff0dSUlrich Drepper { 40767797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 40775590ff0dSUlrich Drepper } 40785590ff0dSUlrich Drepper 40796521f891SChristian Brauner /** 40806521f891SChristian Brauner * vfs_mkdir - create directory 4081abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 40826521f891SChristian Brauner * @dir: inode of @dentry 40836521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 40846521f891SChristian Brauner * @mode: mode of the new directory 40856521f891SChristian Brauner * 40866521f891SChristian Brauner * Create a directory. 40876521f891SChristian Brauner * 4088abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4089abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4090abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 40916521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4092abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 40936521f891SChristian Brauner */ 4094abf08576SChristian Brauner int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, 40956521f891SChristian Brauner struct dentry *dentry, umode_t mode) 40961da177e4SLinus Torvalds { 4097abf08576SChristian Brauner int error; 40988de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40991da177e4SLinus Torvalds 41004609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 41011da177e4SLinus Torvalds if (error) 41021da177e4SLinus Torvalds return error; 41031da177e4SLinus Torvalds 4104acfa4380SAl Viro if (!dir->i_op->mkdir) 41051da177e4SLinus Torvalds return -EPERM; 41061da177e4SLinus Torvalds 41079452e93eSChristian Brauner mode = vfs_prepare_mode(idmap, dir, mode, S_IRWXUGO | S_ISVTX, 0); 41081da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 41091da177e4SLinus Torvalds if (error) 41101da177e4SLinus Torvalds return error; 41111da177e4SLinus Torvalds 41128de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 41138de52778SAl Viro return -EMLINK; 41148de52778SAl Viro 4115c54bd91eSChristian Brauner error = dir->i_op->mkdir(idmap, dir, dentry, mode); 4116a74574aaSStephen Smalley if (!error) 4117f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 41181da177e4SLinus Torvalds return error; 41191da177e4SLinus Torvalds } 41204d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 41211da177e4SLinus Torvalds 412245f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 41231da177e4SLinus Torvalds { 41246902d925SDave Hansen struct dentry *dentry; 4125dae6ad8fSAl Viro struct path path; 4126dae6ad8fSAl Viro int error; 4127b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 41281da177e4SLinus Torvalds 4129b76d8b82SJeff Layton retry: 4130b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 4131584d3226SDmitry Kadashev error = PTR_ERR(dentry); 41326902d925SDave Hansen if (IS_ERR(dentry)) 4133584d3226SDmitry Kadashev goto out_putname; 41346902d925SDave Hansen 41351639a49cSYang Xu error = security_path_mkdir(&path, dentry, 41361639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode)); 41376521f891SChristian Brauner if (!error) { 4138abf08576SChristian Brauner error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode, 4139abf08576SChristian Brauner dentry, mode); 41406521f891SChristian Brauner } 4141921a1650SAl Viro done_path_create(&path, dentry); 4142b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 4143b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 4144b76d8b82SJeff Layton goto retry; 4145b76d8b82SJeff Layton } 4146584d3226SDmitry Kadashev out_putname: 4147584d3226SDmitry Kadashev putname(name); 41481da177e4SLinus Torvalds return error; 41491da177e4SLinus Torvalds } 41501da177e4SLinus Torvalds 41510101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 41520101db7aSDominik Brodowski { 4153584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 41540101db7aSDominik Brodowski } 41550101db7aSDominik Brodowski 4156a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 41575590ff0dSUlrich Drepper { 4158584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 41595590ff0dSUlrich Drepper } 41605590ff0dSUlrich Drepper 41616521f891SChristian Brauner /** 41626521f891SChristian Brauner * vfs_rmdir - remove directory 4163abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 41646521f891SChristian Brauner * @dir: inode of @dentry 41656521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 41666521f891SChristian Brauner * 41676521f891SChristian Brauner * Remove a directory. 41686521f891SChristian Brauner * 4169abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4170abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4171abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 41726521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4173abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 41746521f891SChristian Brauner */ 4175abf08576SChristian Brauner int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir, 41766521f891SChristian Brauner struct dentry *dentry) 41771da177e4SLinus Torvalds { 41784609e1f1SChristian Brauner int error = may_delete(idmap, dir, dentry, 1); 41791da177e4SLinus Torvalds 41801da177e4SLinus Torvalds if (error) 41811da177e4SLinus Torvalds return error; 41821da177e4SLinus Torvalds 4183acfa4380SAl Viro if (!dir->i_op->rmdir) 41841da177e4SLinus Torvalds return -EPERM; 41851da177e4SLinus Torvalds 41861d2ef590SAl Viro dget(dentry); 41875955102cSAl Viro inode_lock(dentry->d_inode); 4188912dbc15SSage Weil 41891da177e4SLinus Torvalds error = -EBUSY; 41901bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 41911bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4192912dbc15SSage Weil goto out; 4193912dbc15SSage Weil 41941da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4195912dbc15SSage Weil if (error) 4196912dbc15SSage Weil goto out; 4197912dbc15SSage Weil 41981da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4199912dbc15SSage Weil if (error) 4200912dbc15SSage Weil goto out; 4201912dbc15SSage Weil 42028767712fSAl Viro shrink_dcache_parent(dentry); 42031da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4204d83c49f3SAl Viro dont_mount(dentry); 42058ed936b5SEric W. Biederman detach_mounts(dentry); 42061da177e4SLinus Torvalds 4207912dbc15SSage Weil out: 42085955102cSAl Viro inode_unlock(dentry->d_inode); 42091d2ef590SAl Viro dput(dentry); 4210912dbc15SSage Weil if (!error) 4211a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 42121da177e4SLinus Torvalds return error; 42131da177e4SLinus Torvalds } 42144d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 42151da177e4SLinus Torvalds 421645f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 42171da177e4SLinus Torvalds { 42180ee50b47SDmitry Kadashev int error; 42191da177e4SLinus Torvalds struct dentry *dentry; 4220f5beed75SAl Viro struct path path; 4221f5beed75SAl Viro struct qstr last; 4222f5beed75SAl Viro int type; 4223c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4224c6ee9206SJeff Layton retry: 4225c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 42260ee50b47SDmitry Kadashev if (error) 42270ee50b47SDmitry Kadashev goto exit1; 42281da177e4SLinus Torvalds 4229f5beed75SAl Viro switch (type) { 42301da177e4SLinus Torvalds case LAST_DOTDOT: 42311da177e4SLinus Torvalds error = -ENOTEMPTY; 42320ee50b47SDmitry Kadashev goto exit2; 42331da177e4SLinus Torvalds case LAST_DOT: 42341da177e4SLinus Torvalds error = -EINVAL; 42350ee50b47SDmitry Kadashev goto exit2; 42361da177e4SLinus Torvalds case LAST_ROOT: 42371da177e4SLinus Torvalds error = -EBUSY; 42380ee50b47SDmitry Kadashev goto exit2; 42391da177e4SLinus Torvalds } 42400612d9fbSOGAWA Hirofumi 4241f5beed75SAl Viro error = mnt_want_write(path.mnt); 4242c30dabfeSJan Kara if (error) 42430ee50b47SDmitry Kadashev goto exit2; 42440612d9fbSOGAWA Hirofumi 42455955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4246*74d7970fSNamjae Jeon dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags); 42471da177e4SLinus Torvalds error = PTR_ERR(dentry); 42486902d925SDave Hansen if (IS_ERR(dentry)) 42490ee50b47SDmitry Kadashev goto exit3; 4250e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4251e6bc45d6STheodore Ts'o error = -ENOENT; 42520ee50b47SDmitry Kadashev goto exit4; 4253e6bc45d6STheodore Ts'o } 4254f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4255be6d3e56SKentaro Takeda if (error) 42560ee50b47SDmitry Kadashev goto exit4; 4257abf08576SChristian Brauner error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry); 42580ee50b47SDmitry Kadashev exit4: 42591da177e4SLinus Torvalds dput(dentry); 42600ee50b47SDmitry Kadashev exit3: 42615955102cSAl Viro inode_unlock(path.dentry->d_inode); 4262f5beed75SAl Viro mnt_drop_write(path.mnt); 42630ee50b47SDmitry Kadashev exit2: 4264f5beed75SAl Viro path_put(&path); 4265c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4266c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4267c6ee9206SJeff Layton goto retry; 4268c6ee9206SJeff Layton } 42690ee50b47SDmitry Kadashev exit1: 427024fb33d4SAl Viro putname(name); 42711da177e4SLinus Torvalds return error; 42721da177e4SLinus Torvalds } 42731da177e4SLinus Torvalds 42743cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 42755590ff0dSUlrich Drepper { 4276e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 42775590ff0dSUlrich Drepper } 42785590ff0dSUlrich Drepper 4279b21996e3SJ. Bruce Fields /** 4280b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 4281abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 4282b21996e3SJ. Bruce Fields * @dir: parent directory 4283b21996e3SJ. Bruce Fields * @dentry: victim 4284b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4285b21996e3SJ. Bruce Fields * 4286b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4287b21996e3SJ. Bruce Fields * 4288b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4289b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4290b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4291b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4292b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4293b21996e3SJ. Bruce Fields * 4294b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4295b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4296b21996e3SJ. Bruce Fields * to be NFS exported. 42976521f891SChristian Brauner * 4298abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4299abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4300abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 43016521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4302abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 4303b21996e3SJ. Bruce Fields */ 4304abf08576SChristian Brauner int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir, 43056521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 43061da177e4SLinus Torvalds { 43079accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 43084609e1f1SChristian Brauner int error = may_delete(idmap, dir, dentry, 0); 43091da177e4SLinus Torvalds 43101da177e4SLinus Torvalds if (error) 43111da177e4SLinus Torvalds return error; 43121da177e4SLinus Torvalds 4313acfa4380SAl Viro if (!dir->i_op->unlink) 43141da177e4SLinus Torvalds return -EPERM; 43151da177e4SLinus Torvalds 43165955102cSAl Viro inode_lock(target); 431751cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 431851cc3a66SHugh Dickins error = -EPERM; 431951cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 43201da177e4SLinus Torvalds error = -EBUSY; 43211da177e4SLinus Torvalds else { 43221da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4323bec1052eSAl Viro if (!error) { 43245a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 43255a14696cSJ. Bruce Fields if (error) 4326b21996e3SJ. Bruce Fields goto out; 43271da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 43288ed936b5SEric W. Biederman if (!error) { 4329d83c49f3SAl Viro dont_mount(dentry); 43308ed936b5SEric W. Biederman detach_mounts(dentry); 43318ed936b5SEric W. Biederman } 4332bec1052eSAl Viro } 43331da177e4SLinus Torvalds } 4334b21996e3SJ. Bruce Fields out: 43355955102cSAl Viro inode_unlock(target); 43361da177e4SLinus Torvalds 43371da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4338a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4339a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4340a37d9a17SAmir Goldstein } else if (!error) { 43419accbb97SJ. Bruce Fields fsnotify_link_count(target); 4342a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 43431da177e4SLinus Torvalds } 43440eeca283SRobert Love 43451da177e4SLinus Torvalds return error; 43461da177e4SLinus Torvalds } 43474d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 43481da177e4SLinus Torvalds 43491da177e4SLinus Torvalds /* 43501da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 43511b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 43521da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 43531da177e4SLinus Torvalds * while waiting on the I/O. 43541da177e4SLinus Torvalds */ 435545f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 43561da177e4SLinus Torvalds { 43572ad94ae6SAl Viro int error; 43581da177e4SLinus Torvalds struct dentry *dentry; 4359f5beed75SAl Viro struct path path; 4360f5beed75SAl Viro struct qstr last; 4361f5beed75SAl Viro int type; 43621da177e4SLinus Torvalds struct inode *inode = NULL; 4363b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 43645d18f813SJeff Layton unsigned int lookup_flags = 0; 43655d18f813SJeff Layton retry: 4366c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 43670ee50b47SDmitry Kadashev if (error) 43680ee50b47SDmitry Kadashev goto exit1; 43692ad94ae6SAl Viro 43701da177e4SLinus Torvalds error = -EISDIR; 4371f5beed75SAl Viro if (type != LAST_NORM) 43720ee50b47SDmitry Kadashev goto exit2; 43730612d9fbSOGAWA Hirofumi 4374f5beed75SAl Viro error = mnt_want_write(path.mnt); 4375c30dabfeSJan Kara if (error) 43760ee50b47SDmitry Kadashev goto exit2; 4377b21996e3SJ. Bruce Fields retry_deleg: 43785955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4379*74d7970fSNamjae Jeon dentry = lookup_one_qstr_excl(&last, path.dentry, lookup_flags); 43801da177e4SLinus Torvalds error = PTR_ERR(dentry); 43811da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 43826521f891SChristian Brauner 43831da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4384f5beed75SAl Viro if (last.name[last.len]) 438550338b88STörök Edwin goto slashes; 43861da177e4SLinus Torvalds inode = dentry->d_inode; 4387b18825a7SDavid Howells if (d_is_negative(dentry)) 4388e6bc45d6STheodore Ts'o goto slashes; 43897de9c6eeSAl Viro ihold(inode); 4390f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4391be6d3e56SKentaro Takeda if (error) 43920ee50b47SDmitry Kadashev goto exit3; 4393abf08576SChristian Brauner error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode, 4394abf08576SChristian Brauner dentry, &delegated_inode); 43950ee50b47SDmitry Kadashev exit3: 43961da177e4SLinus Torvalds dput(dentry); 43971da177e4SLinus Torvalds } 43985955102cSAl Viro inode_unlock(path.dentry->d_inode); 43991da177e4SLinus Torvalds if (inode) 44001da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4401b21996e3SJ. Bruce Fields inode = NULL; 4402b21996e3SJ. Bruce Fields if (delegated_inode) { 44035a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4404b21996e3SJ. Bruce Fields if (!error) 4405b21996e3SJ. Bruce Fields goto retry_deleg; 4406b21996e3SJ. Bruce Fields } 4407f5beed75SAl Viro mnt_drop_write(path.mnt); 44080ee50b47SDmitry Kadashev exit2: 4409f5beed75SAl Viro path_put(&path); 44105d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 44115d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 44125d18f813SJeff Layton inode = NULL; 44135d18f813SJeff Layton goto retry; 44145d18f813SJeff Layton } 44150ee50b47SDmitry Kadashev exit1: 4416da2f1362SChristoph Hellwig putname(name); 44171da177e4SLinus Torvalds return error; 44181da177e4SLinus Torvalds 44191da177e4SLinus Torvalds slashes: 4420b18825a7SDavid Howells if (d_is_negative(dentry)) 4421b18825a7SDavid Howells error = -ENOENT; 442244b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4423b18825a7SDavid Howells error = -EISDIR; 4424b18825a7SDavid Howells else 4425b18825a7SDavid Howells error = -ENOTDIR; 44260ee50b47SDmitry Kadashev goto exit3; 44271da177e4SLinus Torvalds } 44281da177e4SLinus Torvalds 44292e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 44305590ff0dSUlrich Drepper { 44315590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 44325590ff0dSUlrich Drepper return -EINVAL; 44335590ff0dSUlrich Drepper 44345590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4435e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4436da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 44375590ff0dSUlrich Drepper } 44385590ff0dSUlrich Drepper 44393480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 44405590ff0dSUlrich Drepper { 4441da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 44425590ff0dSUlrich Drepper } 44435590ff0dSUlrich Drepper 44446521f891SChristian Brauner /** 44456521f891SChristian Brauner * vfs_symlink - create symlink 4446abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 44476521f891SChristian Brauner * @dir: inode of @dentry 44486521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 44496521f891SChristian Brauner * @oldname: name of the file to link to 44506521f891SChristian Brauner * 44516521f891SChristian Brauner * Create a symlink. 44526521f891SChristian Brauner * 4453abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4454abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4455abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 44566521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4457abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 44586521f891SChristian Brauner */ 4459abf08576SChristian Brauner int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir, 44606521f891SChristian Brauner struct dentry *dentry, const char *oldname) 44611da177e4SLinus Torvalds { 44627a77db95SChristian Brauner int error; 44631da177e4SLinus Torvalds 44644609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 44651da177e4SLinus Torvalds if (error) 44661da177e4SLinus Torvalds return error; 44671da177e4SLinus Torvalds 4468acfa4380SAl Viro if (!dir->i_op->symlink) 44691da177e4SLinus Torvalds return -EPERM; 44701da177e4SLinus Torvalds 44711da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 44721da177e4SLinus Torvalds if (error) 44731da177e4SLinus Torvalds return error; 44741da177e4SLinus Torvalds 44757a77db95SChristian Brauner error = dir->i_op->symlink(idmap, dir, dentry, oldname); 4476a74574aaSStephen Smalley if (!error) 4477f38aa942SAmy Griffis fsnotify_create(dir, dentry); 44781da177e4SLinus Torvalds return error; 44791da177e4SLinus Torvalds } 44804d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 44811da177e4SLinus Torvalds 44827a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 44831da177e4SLinus Torvalds { 44842ad94ae6SAl Viro int error; 44856902d925SDave Hansen struct dentry *dentry; 4486dae6ad8fSAl Viro struct path path; 4487f46d3567SJeff Layton unsigned int lookup_flags = 0; 44881da177e4SLinus Torvalds 4489da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4490da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4491da2d0cedSDmitry Kadashev goto out_putnames; 4492da2d0cedSDmitry Kadashev } 4493f46d3567SJeff Layton retry: 4494b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 44951da177e4SLinus Torvalds error = PTR_ERR(dentry); 44966902d925SDave Hansen if (IS_ERR(dentry)) 4497da2d0cedSDmitry Kadashev goto out_putnames; 44986902d925SDave Hansen 449991a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4500abf08576SChristian Brauner if (!error) 4501abf08576SChristian Brauner error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode, 4502abf08576SChristian Brauner dentry, from->name); 4503921a1650SAl Viro done_path_create(&path, dentry); 4504f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4505f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4506f46d3567SJeff Layton goto retry; 4507f46d3567SJeff Layton } 4508da2d0cedSDmitry Kadashev out_putnames: 4509da2d0cedSDmitry Kadashev putname(to); 45101da177e4SLinus Torvalds putname(from); 45111da177e4SLinus Torvalds return error; 45121da177e4SLinus Torvalds } 45131da177e4SLinus Torvalds 4514b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4515b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4516b724e846SDominik Brodowski { 4517da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4518b724e846SDominik Brodowski } 4519b724e846SDominik Brodowski 45203480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 45215590ff0dSUlrich Drepper { 4522da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 45235590ff0dSUlrich Drepper } 45245590ff0dSUlrich Drepper 4525146a8595SJ. Bruce Fields /** 4526146a8595SJ. Bruce Fields * vfs_link - create a new link 4527146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4528abf08576SChristian Brauner * @idmap: idmap of the mount 4529146a8595SJ. Bruce Fields * @dir: new parent 4530146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4531146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4532146a8595SJ. Bruce Fields * 4533146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4534146a8595SJ. Bruce Fields * 4535146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4536146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4537146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4538146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4539146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4540146a8595SJ. Bruce Fields * 4541146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4542146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4543146a8595SJ. Bruce Fields * to be NFS exported. 45446521f891SChristian Brauner * 4545abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4546abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4547abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 45486521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4549abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 4550146a8595SJ. Bruce Fields */ 4551abf08576SChristian Brauner int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap, 45526521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 45536521f891SChristian Brauner struct inode **delegated_inode) 45541da177e4SLinus Torvalds { 45551da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 45568de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 45571da177e4SLinus Torvalds int error; 45581da177e4SLinus Torvalds 45591da177e4SLinus Torvalds if (!inode) 45601da177e4SLinus Torvalds return -ENOENT; 45611da177e4SLinus Torvalds 45624609e1f1SChristian Brauner error = may_create(idmap, dir, new_dentry); 45631da177e4SLinus Torvalds if (error) 45641da177e4SLinus Torvalds return error; 45651da177e4SLinus Torvalds 45661da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 45671da177e4SLinus Torvalds return -EXDEV; 45681da177e4SLinus Torvalds 45691da177e4SLinus Torvalds /* 45701da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 45711da177e4SLinus Torvalds */ 45721da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 45731da177e4SLinus Torvalds return -EPERM; 45740bd23d09SEric W. Biederman /* 45750bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 45760bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 45770bd23d09SEric W. Biederman * the vfs. 45780bd23d09SEric W. Biederman */ 45794609e1f1SChristian Brauner if (HAS_UNMAPPED_ID(idmap, inode)) 45800bd23d09SEric W. Biederman return -EPERM; 4581acfa4380SAl Viro if (!dir->i_op->link) 45821da177e4SLinus Torvalds return -EPERM; 45837e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 45841da177e4SLinus Torvalds return -EPERM; 45851da177e4SLinus Torvalds 45861da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 45871da177e4SLinus Torvalds if (error) 45881da177e4SLinus Torvalds return error; 45891da177e4SLinus Torvalds 45905955102cSAl Viro inode_lock(inode); 4591aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4592f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4593aae8a97dSAneesh Kumar K.V error = -ENOENT; 45948de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 45958de52778SAl Viro error = -EMLINK; 4596146a8595SJ. Bruce Fields else { 4597146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4598146a8595SJ. Bruce Fields if (!error) 45991da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4600146a8595SJ. Bruce Fields } 4601f4e0c30cSAl Viro 4602f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4603f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4604f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4605f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4606f4e0c30cSAl Viro } 46075955102cSAl Viro inode_unlock(inode); 4608e31e14ecSStephen Smalley if (!error) 46097e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 46101da177e4SLinus Torvalds return error; 46111da177e4SLinus Torvalds } 46124d359507SAl Viro EXPORT_SYMBOL(vfs_link); 46131da177e4SLinus Torvalds 46141da177e4SLinus Torvalds /* 46151da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 46161da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 46171da177e4SLinus Torvalds * newname. --KAB 46181da177e4SLinus Torvalds * 46191da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 46201da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 46211da177e4SLinus Torvalds * and other special files. --ADM 46221da177e4SLinus Torvalds */ 4623cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4624020250f3SDmitry Kadashev struct filename *new, int flags) 46251da177e4SLinus Torvalds { 4626abf08576SChristian Brauner struct mnt_idmap *idmap; 46271da177e4SLinus Torvalds struct dentry *new_dentry; 4628dae6ad8fSAl Viro struct path old_path, new_path; 4629146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 463011a7b371SAneesh Kumar K.V int how = 0; 46311da177e4SLinus Torvalds int error; 46321da177e4SLinus Torvalds 4633020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4634020250f3SDmitry Kadashev error = -EINVAL; 4635020250f3SDmitry Kadashev goto out_putnames; 4636020250f3SDmitry Kadashev } 463711a7b371SAneesh Kumar K.V /* 4638f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4639f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4640f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 464111a7b371SAneesh Kumar K.V */ 4642020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4643020250f3SDmitry Kadashev error = -ENOENT; 4644020250f3SDmitry Kadashev goto out_putnames; 4645f0cc6ffbSLinus Torvalds } 4646c04030e1SUlrich Drepper 464711a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 464811a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4649442e31caSJeff Layton retry: 4650794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 46511da177e4SLinus Torvalds if (error) 4652020250f3SDmitry Kadashev goto out_putnames; 46532ad94ae6SAl Viro 4654b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4655442e31caSJeff Layton (how & LOOKUP_REVAL)); 46561da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 46576902d925SDave Hansen if (IS_ERR(new_dentry)) 4658020250f3SDmitry Kadashev goto out_putpath; 4659dae6ad8fSAl Viro 4660dae6ad8fSAl Viro error = -EXDEV; 4661dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4662dae6ad8fSAl Viro goto out_dput; 4663abf08576SChristian Brauner idmap = mnt_idmap(new_path.mnt); 46644609e1f1SChristian Brauner error = may_linkat(idmap, &old_path); 4665800179c9SKees Cook if (unlikely(error)) 4666800179c9SKees Cook goto out_dput; 4667dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4668be6d3e56SKentaro Takeda if (error) 4669a8104a9fSAl Viro goto out_dput; 4670abf08576SChristian Brauner error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode, 46716521f891SChristian Brauner new_dentry, &delegated_inode); 467275c3f29dSDave Hansen out_dput: 4673921a1650SAl Viro done_path_create(&new_path, new_dentry); 4674146a8595SJ. Bruce Fields if (delegated_inode) { 4675146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4676d22e6338SOleg Drokin if (!error) { 4677d22e6338SOleg Drokin path_put(&old_path); 4678146a8595SJ. Bruce Fields goto retry; 4679146a8595SJ. Bruce Fields } 4680d22e6338SOleg Drokin } 4681442e31caSJeff Layton if (retry_estale(error, how)) { 4682d22e6338SOleg Drokin path_put(&old_path); 4683442e31caSJeff Layton how |= LOOKUP_REVAL; 4684442e31caSJeff Layton goto retry; 4685442e31caSJeff Layton } 4686020250f3SDmitry Kadashev out_putpath: 46872d8f3038SAl Viro path_put(&old_path); 4688020250f3SDmitry Kadashev out_putnames: 4689020250f3SDmitry Kadashev putname(old); 4690020250f3SDmitry Kadashev putname(new); 46911da177e4SLinus Torvalds 46921da177e4SLinus Torvalds return error; 46931da177e4SLinus Torvalds } 46941da177e4SLinus Torvalds 469546ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 469646ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 469746ea89ebSDominik Brodowski { 4698020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4699020250f3SDmitry Kadashev newdfd, getname(newname), flags); 470046ea89ebSDominik Brodowski } 470146ea89ebSDominik Brodowski 47023480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 47035590ff0dSUlrich Drepper { 4704020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 47055590ff0dSUlrich Drepper } 47065590ff0dSUlrich Drepper 4707bc27027aSMiklos Szeredi /** 4708bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 47092111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4710bc27027aSMiklos Szeredi * 4711bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4712bc27027aSMiklos Szeredi * 4713bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4714bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4715bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4716bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4717bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4718bc27027aSMiklos Szeredi * so. 4719bc27027aSMiklos Szeredi * 4720bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4721bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4722bc27027aSMiklos Szeredi * to be NFS exported. 4723bc27027aSMiklos Szeredi * 47241da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 47251da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 47261da177e4SLinus Torvalds * Problems: 47270117d427SMauro Carvalho Chehab * 4728d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 47291da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 47301da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4731a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 47321da177e4SLinus Torvalds * story. 47336cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 47346cedba89SJ. Bruce Fields * and source (if it is not a directory). 47351b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 47361da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 47371da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4738a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 47391da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 47401da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 47411da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4742a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 47431da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 47441da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 47451da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4746e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 47471b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 47481da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4749c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 47501da177e4SLinus Torvalds * locking]. 47511da177e4SLinus Torvalds */ 47529fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 47531da177e4SLinus Torvalds { 47541da177e4SLinus Torvalds int error; 47559fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 47569fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 47579fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 47589fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 47599fe61450SChristian Brauner unsigned int flags = rd->flags; 4760bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4761bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4762bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4763da1ce067SMiklos Szeredi bool new_is_dir = false; 4764da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 476549d31c2fSAl Viro struct name_snapshot old_name; 47661da177e4SLinus Torvalds 47678d3e2936SMiklos Szeredi if (source == target) 47681da177e4SLinus Torvalds return 0; 47691da177e4SLinus Torvalds 47704609e1f1SChristian Brauner error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir); 47711da177e4SLinus Torvalds if (error) 47721da177e4SLinus Torvalds return error; 47731da177e4SLinus Torvalds 4774da1ce067SMiklos Szeredi if (!target) { 47754609e1f1SChristian Brauner error = may_create(rd->new_mnt_idmap, new_dir, new_dentry); 4776da1ce067SMiklos Szeredi } else { 4777da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4778da1ce067SMiklos Szeredi 4779da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 47804609e1f1SChristian Brauner error = may_delete(rd->new_mnt_idmap, new_dir, 47816521f891SChristian Brauner new_dentry, is_dir); 4782da1ce067SMiklos Szeredi else 47834609e1f1SChristian Brauner error = may_delete(rd->new_mnt_idmap, new_dir, 47846521f891SChristian Brauner new_dentry, new_is_dir); 4785da1ce067SMiklos Szeredi } 47861da177e4SLinus Torvalds if (error) 47871da177e4SLinus Torvalds return error; 47881da177e4SLinus Torvalds 47892773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 47901da177e4SLinus Torvalds return -EPERM; 47911da177e4SLinus Torvalds 4792bc27027aSMiklos Szeredi /* 4793bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4794bc27027aSMiklos Szeredi * we'll need to flip '..'. 4795bc27027aSMiklos Szeredi */ 4796da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4797da1ce067SMiklos Szeredi if (is_dir) { 47984609e1f1SChristian Brauner error = inode_permission(rd->old_mnt_idmap, source, 479947291baaSChristian Brauner MAY_WRITE); 4800bc27027aSMiklos Szeredi if (error) 4801bc27027aSMiklos Szeredi return error; 4802bc27027aSMiklos Szeredi } 4803da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 48044609e1f1SChristian Brauner error = inode_permission(rd->new_mnt_idmap, target, 480547291baaSChristian Brauner MAY_WRITE); 4806da1ce067SMiklos Szeredi if (error) 4807da1ce067SMiklos Szeredi return error; 4808da1ce067SMiklos Szeredi } 4809da1ce067SMiklos Szeredi } 48100eeca283SRobert Love 48110b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 48120b3974ebSMiklos Szeredi flags); 4813bc27027aSMiklos Szeredi if (error) 4814bc27027aSMiklos Szeredi return error; 4815bc27027aSMiklos Szeredi 481649d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4817bc27027aSMiklos Szeredi dget(new_dentry); 4818da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4819bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4820bc27027aSMiklos Szeredi else if (target) 48215955102cSAl Viro inode_lock(target); 4822bc27027aSMiklos Szeredi 482351cc3a66SHugh Dickins error = -EPERM; 482451cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 482551cc3a66SHugh Dickins goto out; 482651cc3a66SHugh Dickins 4827bc27027aSMiklos Szeredi error = -EBUSY; 48287af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4829bc27027aSMiklos Szeredi goto out; 4830bc27027aSMiklos Szeredi 4831da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4832bc27027aSMiklos Szeredi error = -EMLINK; 4833da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4834bc27027aSMiklos Szeredi goto out; 4835da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4836da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4837da1ce067SMiklos Szeredi goto out; 4838da1ce067SMiklos Szeredi } 4839da1ce067SMiklos Szeredi if (!is_dir) { 4840bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4841bc27027aSMiklos Szeredi if (error) 4842bc27027aSMiklos Szeredi goto out; 4843da1ce067SMiklos Szeredi } 4844da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4845bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4846bc27027aSMiklos Szeredi if (error) 4847bc27027aSMiklos Szeredi goto out; 4848bc27027aSMiklos Szeredi } 4849e18275aeSChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry, 4850520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4851bc27027aSMiklos Szeredi if (error) 4852bc27027aSMiklos Szeredi goto out; 4853bc27027aSMiklos Szeredi 4854da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 48558767712fSAl Viro if (is_dir) { 48568767712fSAl Viro shrink_dcache_parent(new_dentry); 4857bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 48588767712fSAl Viro } 4859bc27027aSMiklos Szeredi dont_mount(new_dentry); 48608ed936b5SEric W. Biederman detach_mounts(new_dentry); 4861bc27027aSMiklos Szeredi } 4862da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4863da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4864bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4865da1ce067SMiklos Szeredi else 4866da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4867da1ce067SMiklos Szeredi } 4868bc27027aSMiklos Szeredi out: 4869da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4870bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4871bc27027aSMiklos Szeredi else if (target) 48725955102cSAl Viro inode_unlock(target); 4873bc27027aSMiklos Szeredi dput(new_dentry); 4874da1ce067SMiklos Szeredi if (!error) { 4875f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4876da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4877da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4878f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4879da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4880da1ce067SMiklos Szeredi } 4881da1ce067SMiklos Szeredi } 488249d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 48830eeca283SRobert Love 48841da177e4SLinus Torvalds return error; 48851da177e4SLinus Torvalds } 48864d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 48871da177e4SLinus Torvalds 4888e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4889e886663cSJens Axboe struct filename *to, unsigned int flags) 48901da177e4SLinus Torvalds { 48919fe61450SChristian Brauner struct renamedata rd; 48921da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 48931da177e4SLinus Torvalds struct dentry *trap; 4894f5beed75SAl Viro struct path old_path, new_path; 4895f5beed75SAl Viro struct qstr old_last, new_last; 4896f5beed75SAl Viro int old_type, new_type; 48978e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4898f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4899c6a94284SJeff Layton bool should_retry = false; 4900e886663cSJens Axboe int error = -EINVAL; 4901520c8b16SMiklos Szeredi 49020d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 49030ee50b47SDmitry Kadashev goto put_names; 4904da1ce067SMiklos Szeredi 49050d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 49060d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 49070ee50b47SDmitry Kadashev goto put_names; 4908520c8b16SMiklos Szeredi 4909f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4910f5beed75SAl Viro target_flags = 0; 4911f5beed75SAl Viro 4912c6a94284SJeff Layton retry: 4913c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4914e886663cSJens Axboe &old_last, &old_type); 49150ee50b47SDmitry Kadashev if (error) 49160ee50b47SDmitry Kadashev goto put_names; 49171da177e4SLinus Torvalds 4918c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4919e886663cSJens Axboe &new_type); 49200ee50b47SDmitry Kadashev if (error) 49211da177e4SLinus Torvalds goto exit1; 49221da177e4SLinus Torvalds 49231da177e4SLinus Torvalds error = -EXDEV; 4924f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 49251da177e4SLinus Torvalds goto exit2; 49261da177e4SLinus Torvalds 49271da177e4SLinus Torvalds error = -EBUSY; 4928f5beed75SAl Viro if (old_type != LAST_NORM) 49291da177e4SLinus Torvalds goto exit2; 49301da177e4SLinus Torvalds 49310a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 49320a7c3937SMiklos Szeredi error = -EEXIST; 4933f5beed75SAl Viro if (new_type != LAST_NORM) 49341da177e4SLinus Torvalds goto exit2; 49351da177e4SLinus Torvalds 4936f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4937c30dabfeSJan Kara if (error) 4938c30dabfeSJan Kara goto exit2; 4939c30dabfeSJan Kara 49408e6d782cSJ. Bruce Fields retry_deleg: 4941f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 49421da177e4SLinus Torvalds 4943*74d7970fSNamjae Jeon old_dentry = lookup_one_qstr_excl(&old_last, old_path.dentry, 4944*74d7970fSNamjae Jeon lookup_flags); 49451da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 49461da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 49471da177e4SLinus Torvalds goto exit3; 49481da177e4SLinus Torvalds /* source must exist */ 49491da177e4SLinus Torvalds error = -ENOENT; 4950b18825a7SDavid Howells if (d_is_negative(old_dentry)) 49511da177e4SLinus Torvalds goto exit4; 4952*74d7970fSNamjae Jeon new_dentry = lookup_one_qstr_excl(&new_last, new_path.dentry, 4953*74d7970fSNamjae Jeon lookup_flags | target_flags); 49541da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 49551da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 49561da177e4SLinus Torvalds goto exit4; 49570a7c3937SMiklos Szeredi error = -EEXIST; 49580a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 49590a7c3937SMiklos Szeredi goto exit5; 4960da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4961da1ce067SMiklos Szeredi error = -ENOENT; 4962da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4963da1ce067SMiklos Szeredi goto exit5; 4964da1ce067SMiklos Szeredi 4965da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4966da1ce067SMiklos Szeredi error = -ENOTDIR; 4967f5beed75SAl Viro if (new_last.name[new_last.len]) 4968da1ce067SMiklos Szeredi goto exit5; 4969da1ce067SMiklos Szeredi } 4970da1ce067SMiklos Szeredi } 49710a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 49720a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 49730a7c3937SMiklos Szeredi error = -ENOTDIR; 4974f5beed75SAl Viro if (old_last.name[old_last.len]) 49750a7c3937SMiklos Szeredi goto exit5; 4976f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 49770a7c3937SMiklos Szeredi goto exit5; 49780a7c3937SMiklos Szeredi } 49790a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 49800a7c3937SMiklos Szeredi error = -EINVAL; 49810a7c3937SMiklos Szeredi if (old_dentry == trap) 49820a7c3937SMiklos Szeredi goto exit5; 49831da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4984da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 49851da177e4SLinus Torvalds error = -ENOTEMPTY; 49861da177e4SLinus Torvalds if (new_dentry == trap) 49871da177e4SLinus Torvalds goto exit5; 49881da177e4SLinus Torvalds 4989f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4990f5beed75SAl Viro &new_path, new_dentry, flags); 4991be6d3e56SKentaro Takeda if (error) 4992c30dabfeSJan Kara goto exit5; 49939fe61450SChristian Brauner 49949fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 49959fe61450SChristian Brauner rd.old_dentry = old_dentry; 4996abf08576SChristian Brauner rd.old_mnt_idmap = mnt_idmap(old_path.mnt); 49979fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 49989fe61450SChristian Brauner rd.new_dentry = new_dentry; 4999abf08576SChristian Brauner rd.new_mnt_idmap = mnt_idmap(new_path.mnt); 50009fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 50019fe61450SChristian Brauner rd.flags = flags; 50029fe61450SChristian Brauner error = vfs_rename(&rd); 50031da177e4SLinus Torvalds exit5: 50041da177e4SLinus Torvalds dput(new_dentry); 50051da177e4SLinus Torvalds exit4: 50061da177e4SLinus Torvalds dput(old_dentry); 50071da177e4SLinus Torvalds exit3: 5008f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 50098e6d782cSJ. Bruce Fields if (delegated_inode) { 50108e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 50118e6d782cSJ. Bruce Fields if (!error) 50128e6d782cSJ. Bruce Fields goto retry_deleg; 50138e6d782cSJ. Bruce Fields } 5014f5beed75SAl Viro mnt_drop_write(old_path.mnt); 50151da177e4SLinus Torvalds exit2: 5016c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 5017c6a94284SJeff Layton should_retry = true; 5018f5beed75SAl Viro path_put(&new_path); 50191da177e4SLinus Torvalds exit1: 5020f5beed75SAl Viro path_put(&old_path); 5021c6a94284SJeff Layton if (should_retry) { 5022c6a94284SJeff Layton should_retry = false; 5023c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 5024c6a94284SJeff Layton goto retry; 5025c6a94284SJeff Layton } 50260ee50b47SDmitry Kadashev put_names: 5027e886663cSJens Axboe putname(from); 5028e886663cSJens Axboe putname(to); 50291da177e4SLinus Torvalds return error; 50301da177e4SLinus Torvalds } 50311da177e4SLinus Torvalds 5032ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 5033ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 5034ee81feb6SDominik Brodowski { 5035e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 5036e886663cSJens Axboe flags); 5037ee81feb6SDominik Brodowski } 5038ee81feb6SDominik Brodowski 5039520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 5040520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 5041520c8b16SMiklos Szeredi { 5042e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 5043e886663cSJens Axboe 0); 5044520c8b16SMiklos Szeredi } 5045520c8b16SMiklos Szeredi 5046a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 50475590ff0dSUlrich Drepper { 5048e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 5049e886663cSJens Axboe getname(newname), 0); 50505590ff0dSUlrich Drepper } 50515590ff0dSUlrich Drepper 50525d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 50531da177e4SLinus Torvalds { 50545d826c84SAl Viro int len = PTR_ERR(link); 50551da177e4SLinus Torvalds if (IS_ERR(link)) 50561da177e4SLinus Torvalds goto out; 50571da177e4SLinus Torvalds 50581da177e4SLinus Torvalds len = strlen(link); 50591da177e4SLinus Torvalds if (len > (unsigned) buflen) 50601da177e4SLinus Torvalds len = buflen; 50611da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 50621da177e4SLinus Torvalds len = -EFAULT; 50631da177e4SLinus Torvalds out: 50641da177e4SLinus Torvalds return len; 50651da177e4SLinus Torvalds } 50661da177e4SLinus Torvalds 5067d60874cdSMiklos Szeredi /** 5068fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 5069fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 5070fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 5071fd4a0edfSMiklos Szeredi * @buflen: size of buffer 5072fd4a0edfSMiklos Szeredi * 5073fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 5074fd4a0edfSMiklos Szeredi * 5075fd4a0edfSMiklos Szeredi * Does not call security hook. 5076fd4a0edfSMiklos Szeredi */ 5077fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5078fd4a0edfSMiklos Szeredi { 5079fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 5080f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 5081f2df5da6SAl Viro const char *link; 5082f2df5da6SAl Viro int res; 5083fd4a0edfSMiklos Szeredi 508476fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 508576fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 508676fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 508776fca90eSMiklos Szeredi 508876fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 5089fd4a0edfSMiklos Szeredi return -EINVAL; 5090fd4a0edfSMiklos Szeredi 509176fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 509276fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 509376fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 509476fca90eSMiklos Szeredi } 509576fca90eSMiklos Szeredi 50964c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 5097f2df5da6SAl Viro if (!link) { 5098f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 5099f2df5da6SAl Viro if (IS_ERR(link)) 5100f2df5da6SAl Viro return PTR_ERR(link); 5101f2df5da6SAl Viro } 5102f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 5103f2df5da6SAl Viro do_delayed_call(&done); 5104f2df5da6SAl Viro return res; 5105fd4a0edfSMiklos Szeredi } 5106fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 51071da177e4SLinus Torvalds 5108d60874cdSMiklos Szeredi /** 5109d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 5110d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 5111d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 5112d60874cdSMiklos Szeredi * 5113d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 5114d60874cdSMiklos Szeredi * 5115d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 5116d60874cdSMiklos Szeredi * 5117d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 5118d60874cdSMiklos Szeredi */ 5119d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 5120d60874cdSMiklos Szeredi { 5121d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 5122d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 5123d60874cdSMiklos Szeredi 5124d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 5125d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 5126d60874cdSMiklos Szeredi if (!res) 5127d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 5128d60874cdSMiklos Szeredi } 5129d60874cdSMiklos Szeredi return res; 5130d60874cdSMiklos Szeredi } 5131d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 5132d60874cdSMiklos Szeredi 51331da177e4SLinus Torvalds /* get the link contents into pagecache */ 51346b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 5135fceef393SAl Viro struct delayed_call *callback) 51361da177e4SLinus Torvalds { 5137ebd09abbSDuane Griffin char *kaddr; 51381da177e4SLinus Torvalds struct page *page; 51396b255391SAl Viro struct address_space *mapping = inode->i_mapping; 51406b255391SAl Viro 5141d3883d4fSAl Viro if (!dentry) { 5142d3883d4fSAl Viro page = find_get_page(mapping, 0); 5143d3883d4fSAl Viro if (!page) 51446b255391SAl Viro return ERR_PTR(-ECHILD); 5145d3883d4fSAl Viro if (!PageUptodate(page)) { 5146d3883d4fSAl Viro put_page(page); 5147d3883d4fSAl Viro return ERR_PTR(-ECHILD); 5148d3883d4fSAl Viro } 5149d3883d4fSAl Viro } else { 5150090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 51511da177e4SLinus Torvalds if (IS_ERR(page)) 51526fe6900eSNick Piggin return (char*)page; 5153d3883d4fSAl Viro } 5154fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 515521fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 515621fc61c7SAl Viro kaddr = page_address(page); 51576b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 5158ebd09abbSDuane Griffin return kaddr; 51591da177e4SLinus Torvalds } 51601da177e4SLinus Torvalds 51616b255391SAl Viro EXPORT_SYMBOL(page_get_link); 51621da177e4SLinus Torvalds 5163fceef393SAl Viro void page_put_link(void *arg) 51641da177e4SLinus Torvalds { 5165fceef393SAl Viro put_page(arg); 51661da177e4SLinus Torvalds } 51674d359507SAl Viro EXPORT_SYMBOL(page_put_link); 51681da177e4SLinus Torvalds 5169aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5170aa80deabSAl Viro { 5171fceef393SAl Viro DEFINE_DELAYED_CALL(done); 51726b255391SAl Viro int res = readlink_copy(buffer, buflen, 51736b255391SAl Viro page_get_link(dentry, d_inode(dentry), 5174fceef393SAl Viro &done)); 5175fceef393SAl Viro do_delayed_call(&done); 5176aa80deabSAl Viro return res; 5177aa80deabSAl Viro } 5178aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 5179aa80deabSAl Viro 518056f5746cSMatthew Wilcox (Oracle) int page_symlink(struct inode *inode, const char *symname, int len) 51811da177e4SLinus Torvalds { 51821da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 518327a77913SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 518456f5746cSMatthew Wilcox (Oracle) bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS); 51850adb25d2SKirill Korotaev struct page *page; 51861468c6f4SAlexander Potapenko void *fsdata = NULL; 5187beb497abSDmitriy Monakhov int err; 51882d878178SMatthew Wilcox (Oracle) unsigned int flags; 51891da177e4SLinus Torvalds 51907e53cac4SNeilBrown retry: 51912d878178SMatthew Wilcox (Oracle) if (nofs) 51922d878178SMatthew Wilcox (Oracle) flags = memalloc_nofs_save(); 519327a77913SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata); 51942d878178SMatthew Wilcox (Oracle) if (nofs) 51952d878178SMatthew Wilcox (Oracle) memalloc_nofs_restore(flags); 51961da177e4SLinus Torvalds if (err) 5197afddba49SNick Piggin goto fail; 5198afddba49SNick Piggin 519921fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5200afddba49SNick Piggin 520127a77913SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, 0, len-1, len-1, 5202afddba49SNick Piggin page, fsdata); 52031da177e4SLinus Torvalds if (err < 0) 52041da177e4SLinus Torvalds goto fail; 5205afddba49SNick Piggin if (err < len-1) 5206afddba49SNick Piggin goto retry; 5207afddba49SNick Piggin 52081da177e4SLinus Torvalds mark_inode_dirty(inode); 52091da177e4SLinus Torvalds return 0; 52101da177e4SLinus Torvalds fail: 52111da177e4SLinus Torvalds return err; 52121da177e4SLinus Torvalds } 52134d359507SAl Viro EXPORT_SYMBOL(page_symlink); 52140adb25d2SKirill Korotaev 521592e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 52166b255391SAl Viro .get_link = page_get_link, 52171da177e4SLinus Torvalds }; 52181da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5219