1b2441318SGreg Kroah-Hartman // SPDX-License-Identifier: GPL-2.0 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namei.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 1991, 1992 Linus Torvalds 61da177e4SLinus Torvalds */ 71da177e4SLinus Torvalds 81da177e4SLinus Torvalds /* 91da177e4SLinus Torvalds * Some corrections by tytso. 101da177e4SLinus Torvalds */ 111da177e4SLinus Torvalds 121da177e4SLinus Torvalds /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname 131da177e4SLinus Torvalds * lookup logic. 141da177e4SLinus Torvalds */ 151da177e4SLinus Torvalds /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture. 161da177e4SLinus Torvalds */ 171da177e4SLinus Torvalds 181da177e4SLinus Torvalds #include <linux/init.h> 19630d9c47SPaul Gortmaker #include <linux/export.h> 2044696908SDavid S. Miller #include <linux/kernel.h> 211da177e4SLinus Torvalds #include <linux/slab.h> 221da177e4SLinus Torvalds #include <linux/fs.h> 231da177e4SLinus Torvalds #include <linux/namei.h> 241da177e4SLinus Torvalds #include <linux/pagemap.h> 252d878178SMatthew Wilcox (Oracle) #include <linux/sched/mm.h> 260eeca283SRobert Love #include <linux/fsnotify.h> 271da177e4SLinus Torvalds #include <linux/personality.h> 281da177e4SLinus Torvalds #include <linux/security.h> 296146f0d5SMimi Zohar #include <linux/ima.h> 301da177e4SLinus Torvalds #include <linux/syscalls.h> 311da177e4SLinus Torvalds #include <linux/mount.h> 321da177e4SLinus Torvalds #include <linux/audit.h> 3316f7e0feSRandy Dunlap #include <linux/capability.h> 34834f2a4aSTrond Myklebust #include <linux/file.h> 355590ff0dSUlrich Drepper #include <linux/fcntl.h> 3608ce5f16SSerge E. Hallyn #include <linux/device_cgroup.h> 375ad4e53bSAl Viro #include <linux/fs_struct.h> 38e77819e5SLinus Torvalds #include <linux/posix_acl.h> 3999d263d4SLinus Torvalds #include <linux/hash.h> 402a18da7aSGeorge Spelvin #include <linux/bitops.h> 41aeaa4a79SEric W. Biederman #include <linux/init_task.h> 427c0f6ba6SLinus Torvalds #include <linux/uaccess.h> 431da177e4SLinus Torvalds 44e81e3f4dSEric Paris #include "internal.h" 45c7105365SAl Viro #include "mount.h" 46e81e3f4dSEric Paris 471da177e4SLinus Torvalds /* [Feb-1997 T. Schoebel-Theuer] 481da177e4SLinus Torvalds * Fundamental changes in the pathname lookup mechanisms (namei) 491da177e4SLinus Torvalds * were necessary because of omirr. The reason is that omirr needs 501da177e4SLinus Torvalds * to know the _real_ pathname, not the user-supplied one, in case 511da177e4SLinus Torvalds * of symlinks (and also when transname replacements occur). 521da177e4SLinus Torvalds * 531da177e4SLinus Torvalds * The new code replaces the old recursive symlink resolution with 541da177e4SLinus Torvalds * an iterative one (in case of non-nested symlink chains). It does 551da177e4SLinus Torvalds * this with calls to <fs>_follow_link(). 561da177e4SLinus Torvalds * As a side effect, dir_namei(), _namei() and follow_link() are now 571da177e4SLinus Torvalds * replaced with a single function lookup_dentry() that can handle all 581da177e4SLinus Torvalds * the special cases of the former code. 591da177e4SLinus Torvalds * 601da177e4SLinus Torvalds * With the new dcache, the pathname is stored at each inode, at least as 611da177e4SLinus Torvalds * long as the refcount of the inode is positive. As a side effect, the 621da177e4SLinus Torvalds * size of the dcache depends on the inode cache and thus is dynamic. 631da177e4SLinus Torvalds * 641da177e4SLinus Torvalds * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink 651da177e4SLinus Torvalds * resolution to correspond with current state of the code. 661da177e4SLinus Torvalds * 671da177e4SLinus Torvalds * Note that the symlink resolution is not *completely* iterative. 681da177e4SLinus Torvalds * There is still a significant amount of tail- and mid- recursion in 691da177e4SLinus Torvalds * the algorithm. Also, note that <fs>_readlink() is not used in 701da177e4SLinus Torvalds * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink() 711da177e4SLinus Torvalds * may return different results than <fs>_follow_link(). Many virtual 721da177e4SLinus Torvalds * filesystems (including /proc) exhibit this behavior. 731da177e4SLinus Torvalds */ 741da177e4SLinus Torvalds 751da177e4SLinus Torvalds /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation: 761da177e4SLinus Torvalds * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL 771da177e4SLinus Torvalds * and the name already exists in form of a symlink, try to create the new 781da177e4SLinus Torvalds * name indicated by the symlink. The old code always complained that the 791da177e4SLinus Torvalds * name already exists, due to not following the symlink even if its target 801da177e4SLinus Torvalds * is nonexistent. The new semantics affects also mknod() and link() when 8125985edcSLucas De Marchi * the name is a symlink pointing to a non-existent name. 821da177e4SLinus Torvalds * 831da177e4SLinus Torvalds * I don't know which semantics is the right one, since I have no access 841da177e4SLinus Torvalds * to standards. But I found by trial that HP-UX 9.0 has the full "new" 851da177e4SLinus Torvalds * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the 861da177e4SLinus Torvalds * "old" one. Personally, I think the new semantics is much more logical. 871da177e4SLinus Torvalds * Note that "ln old new" where "new" is a symlink pointing to a non-existing 881da177e4SLinus Torvalds * file does succeed in both HP-UX and SunOs, but not in Solaris 891da177e4SLinus Torvalds * and in the old Linux semantics. 901da177e4SLinus Torvalds */ 911da177e4SLinus Torvalds 921da177e4SLinus Torvalds /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink 931da177e4SLinus Torvalds * semantics. See the comments in "open_namei" and "do_link" below. 941da177e4SLinus Torvalds * 951da177e4SLinus Torvalds * [10-Sep-98 Alan Modra] Another symlink change. 961da177e4SLinus Torvalds */ 971da177e4SLinus Torvalds 981da177e4SLinus Torvalds /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks: 991da177e4SLinus Torvalds * inside the path - always follow. 1001da177e4SLinus Torvalds * in the last component in creation/removal/renaming - never follow. 1011da177e4SLinus Torvalds * if LOOKUP_FOLLOW passed - follow. 1021da177e4SLinus Torvalds * if the pathname has trailing slashes - follow. 1031da177e4SLinus Torvalds * otherwise - don't follow. 1041da177e4SLinus Torvalds * (applied in that order). 1051da177e4SLinus Torvalds * 1061da177e4SLinus Torvalds * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT 1071da177e4SLinus Torvalds * restored for 2.4. This is the last surviving part of old 4.2BSD bug. 1081da177e4SLinus Torvalds * During the 2.4 we need to fix the userland stuff depending on it - 1091da177e4SLinus Torvalds * hopefully we will be able to get rid of that wart in 2.5. So far only 1101da177e4SLinus Torvalds * XEmacs seems to be relying on it... 1111da177e4SLinus Torvalds */ 1121da177e4SLinus Torvalds /* 1131da177e4SLinus Torvalds * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland) 114a11f3a05SArjan van de Ven * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives 1151da177e4SLinus Torvalds * any extra contention... 1161da177e4SLinus Torvalds */ 1171da177e4SLinus Torvalds 1181da177e4SLinus Torvalds /* In order to reduce some races, while at the same time doing additional 1191da177e4SLinus Torvalds * checking and hopefully speeding things up, we copy filenames to the 1201da177e4SLinus Torvalds * kernel data space before using them.. 1211da177e4SLinus Torvalds * 1221da177e4SLinus Torvalds * POSIX.1 2.4: an empty pathname is invalid (ENOENT). 1231da177e4SLinus Torvalds * PATH_MAX includes the nul terminator --RR. 1241da177e4SLinus Torvalds */ 1257950e385SJeff Layton 126fd2f7cb5SAl Viro #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname)) 12791a27b2aSJeff Layton 12851f39a1fSDavid Drysdale struct filename * 12991a27b2aSJeff Layton getname_flags(const char __user *filename, int flags, int *empty) 13091a27b2aSJeff Layton { 13194b5d262SAl Viro struct filename *result; 1327950e385SJeff Layton char *kname; 13394b5d262SAl Viro int len; 1341da177e4SLinus Torvalds 1357ac86265SJeff Layton result = audit_reusename(filename); 1367ac86265SJeff Layton if (result) 1377ac86265SJeff Layton return result; 1387ac86265SJeff Layton 1397950e385SJeff Layton result = __getname(); 1403f9f0aa6SLinus Torvalds if (unlikely(!result)) 1414043cde8SEric Paris return ERR_PTR(-ENOMEM); 1421da177e4SLinus Torvalds 1437950e385SJeff Layton /* 1447950e385SJeff Layton * First, try to embed the struct filename inside the names_cache 1457950e385SJeff Layton * allocation 1467950e385SJeff Layton */ 147fd2f7cb5SAl Viro kname = (char *)result->iname; 14891a27b2aSJeff Layton result->name = kname; 1497950e385SJeff Layton 15094b5d262SAl Viro len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX); 15191a27b2aSJeff Layton if (unlikely(len < 0)) { 15294b5d262SAl Viro __putname(result); 15394b5d262SAl Viro return ERR_PTR(len); 15491a27b2aSJeff Layton } 1553f9f0aa6SLinus Torvalds 1567950e385SJeff Layton /* 1577950e385SJeff Layton * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a 1587950e385SJeff Layton * separate struct filename so we can dedicate the entire 1597950e385SJeff Layton * names_cache allocation for the pathname, and re-do the copy from 1607950e385SJeff Layton * userland. 1617950e385SJeff Layton */ 16294b5d262SAl Viro if (unlikely(len == EMBEDDED_NAME_MAX)) { 163fd2f7cb5SAl Viro const size_t size = offsetof(struct filename, iname[1]); 1647950e385SJeff Layton kname = (char *)result; 1657950e385SJeff Layton 166fd2f7cb5SAl Viro /* 167fd2f7cb5SAl Viro * size is chosen that way we to guarantee that 168fd2f7cb5SAl Viro * result->iname[0] is within the same object and that 169fd2f7cb5SAl Viro * kname can't be equal to result->iname, no matter what. 170fd2f7cb5SAl Viro */ 171fd2f7cb5SAl Viro result = kzalloc(size, GFP_KERNEL); 17294b5d262SAl Viro if (unlikely(!result)) { 17394b5d262SAl Viro __putname(kname); 17494b5d262SAl Viro return ERR_PTR(-ENOMEM); 1757950e385SJeff Layton } 1767950e385SJeff Layton result->name = kname; 17794b5d262SAl Viro len = strncpy_from_user(kname, filename, PATH_MAX); 17894b5d262SAl Viro if (unlikely(len < 0)) { 17994b5d262SAl Viro __putname(kname); 18094b5d262SAl Viro kfree(result); 18194b5d262SAl Viro return ERR_PTR(len); 18294b5d262SAl Viro } 18394b5d262SAl Viro if (unlikely(len == PATH_MAX)) { 18494b5d262SAl Viro __putname(kname); 18594b5d262SAl Viro kfree(result); 18694b5d262SAl Viro return ERR_PTR(-ENAMETOOLONG); 18794b5d262SAl Viro } 1887950e385SJeff Layton } 1897950e385SJeff Layton 19094b5d262SAl Viro result->refcnt = 1; 1913f9f0aa6SLinus Torvalds /* The empty path is special. */ 1923f9f0aa6SLinus Torvalds if (unlikely(!len)) { 1933f9f0aa6SLinus Torvalds if (empty) 1941fa1e7f6SAndy Whitcroft *empty = 1; 19594b5d262SAl Viro if (!(flags & LOOKUP_EMPTY)) { 19694b5d262SAl Viro putname(result); 19794b5d262SAl Viro return ERR_PTR(-ENOENT); 1981da177e4SLinus Torvalds } 19994b5d262SAl Viro } 2007950e385SJeff Layton 2017950e385SJeff Layton result->uptr = filename; 202c4ad8f98SLinus Torvalds result->aname = NULL; 2031da177e4SLinus Torvalds audit_getname(result); 2041da177e4SLinus Torvalds return result; 2053f9f0aa6SLinus Torvalds } 2063f9f0aa6SLinus Torvalds 20791a27b2aSJeff Layton struct filename * 2088228e2c3SDmitry Kadashev getname_uflags(const char __user *filename, int uflags) 2098228e2c3SDmitry Kadashev { 2108228e2c3SDmitry Kadashev int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0; 2118228e2c3SDmitry Kadashev 2128228e2c3SDmitry Kadashev return getname_flags(filename, flags, NULL); 2138228e2c3SDmitry Kadashev } 2148228e2c3SDmitry Kadashev 2158228e2c3SDmitry Kadashev struct filename * 21691a27b2aSJeff Layton getname(const char __user * filename) 217f52e0c11SAl Viro { 218f7493e5dSLinus Torvalds return getname_flags(filename, 0, NULL); 219f52e0c11SAl Viro } 220f52e0c11SAl Viro 221c4ad8f98SLinus Torvalds struct filename * 222c4ad8f98SLinus Torvalds getname_kernel(const char * filename) 223c4ad8f98SLinus Torvalds { 224c4ad8f98SLinus Torvalds struct filename *result; 22508518549SPaul Moore int len = strlen(filename) + 1; 226c4ad8f98SLinus Torvalds 227c4ad8f98SLinus Torvalds result = __getname(); 228c4ad8f98SLinus Torvalds if (unlikely(!result)) 229c4ad8f98SLinus Torvalds return ERR_PTR(-ENOMEM); 230c4ad8f98SLinus Torvalds 23108518549SPaul Moore if (len <= EMBEDDED_NAME_MAX) { 232fd2f7cb5SAl Viro result->name = (char *)result->iname; 23308518549SPaul Moore } else if (len <= PATH_MAX) { 23430ce4d19SAl Viro const size_t size = offsetof(struct filename, iname[1]); 23508518549SPaul Moore struct filename *tmp; 23608518549SPaul Moore 23730ce4d19SAl Viro tmp = kmalloc(size, GFP_KERNEL); 23808518549SPaul Moore if (unlikely(!tmp)) { 23908518549SPaul Moore __putname(result); 24008518549SPaul Moore return ERR_PTR(-ENOMEM); 24108518549SPaul Moore } 24208518549SPaul Moore tmp->name = (char *)result; 24308518549SPaul Moore result = tmp; 24408518549SPaul Moore } else { 24508518549SPaul Moore __putname(result); 24608518549SPaul Moore return ERR_PTR(-ENAMETOOLONG); 24708518549SPaul Moore } 24808518549SPaul Moore memcpy((char *)result->name, filename, len); 249c4ad8f98SLinus Torvalds result->uptr = NULL; 250c4ad8f98SLinus Torvalds result->aname = NULL; 25155422d0bSPaul Moore result->refcnt = 1; 252fd3522fdSPaul Moore audit_getname(result); 253c4ad8f98SLinus Torvalds 254c4ad8f98SLinus Torvalds return result; 255c4ad8f98SLinus Torvalds } 256c4ad8f98SLinus Torvalds 25791a27b2aSJeff Layton void putname(struct filename *name) 2581da177e4SLinus Torvalds { 259ea47ab11SAl Viro if (IS_ERR(name)) 26091ef658fSDmitry Kadashev return; 26191ef658fSDmitry Kadashev 26255422d0bSPaul Moore BUG_ON(name->refcnt <= 0); 26355422d0bSPaul Moore 26455422d0bSPaul Moore if (--name->refcnt > 0) 26555422d0bSPaul Moore return; 26655422d0bSPaul Moore 267fd2f7cb5SAl Viro if (name->name != name->iname) { 26855422d0bSPaul Moore __putname(name->name); 26955422d0bSPaul Moore kfree(name); 27055422d0bSPaul Moore } else 27155422d0bSPaul Moore __putname(name); 2721da177e4SLinus Torvalds } 2731da177e4SLinus Torvalds 27447291baaSChristian Brauner /** 27547291baaSChristian Brauner * check_acl - perform ACL permission checking 276700b7940SChristian Brauner * @idmap: idmap of the mount the inode was found from 27747291baaSChristian Brauner * @inode: inode to check permissions on 27847291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 27947291baaSChristian Brauner * 28047291baaSChristian Brauner * This function performs the ACL permission checking. Since this function 28147291baaSChristian Brauner * retrieve POSIX acls it needs to know whether it is called from a blocking or 28247291baaSChristian Brauner * non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 28347291baaSChristian Brauner * 284700b7940SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 285700b7940SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 286700b7940SChristian Brauner * care to map the inode according to @idmap before checking permissions. 28747291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 288700b7940SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 28947291baaSChristian Brauner */ 290700b7940SChristian Brauner static int check_acl(struct mnt_idmap *idmap, 29147291baaSChristian Brauner struct inode *inode, int mask) 292e77819e5SLinus Torvalds { 29384635d68SLinus Torvalds #ifdef CONFIG_FS_POSIX_ACL 294e77819e5SLinus Torvalds struct posix_acl *acl; 295e77819e5SLinus Torvalds 296e77819e5SLinus Torvalds if (mask & MAY_NOT_BLOCK) { 2973567866bSAl Viro acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS); 2983567866bSAl Viro if (!acl) 299e77819e5SLinus Torvalds return -EAGAIN; 300cac2f8b8SChristian Brauner /* no ->get_inode_acl() calls in RCU mode... */ 301b8a7a3a6SAndreas Gruenbacher if (is_uncached_acl(acl)) 302e77819e5SLinus Torvalds return -ECHILD; 303700b7940SChristian Brauner return posix_acl_permission(idmap, inode, acl, mask); 304e77819e5SLinus Torvalds } 305e77819e5SLinus Torvalds 306cac2f8b8SChristian Brauner acl = get_inode_acl(inode, ACL_TYPE_ACCESS); 3074e34e719SChristoph Hellwig if (IS_ERR(acl)) 3084e34e719SChristoph Hellwig return PTR_ERR(acl); 309e77819e5SLinus Torvalds if (acl) { 310700b7940SChristian Brauner int error = posix_acl_permission(idmap, inode, acl, mask); 311e77819e5SLinus Torvalds posix_acl_release(acl); 312e77819e5SLinus Torvalds return error; 313e77819e5SLinus Torvalds } 31484635d68SLinus Torvalds #endif 315e77819e5SLinus Torvalds 316e77819e5SLinus Torvalds return -EAGAIN; 317e77819e5SLinus Torvalds } 318e77819e5SLinus Torvalds 31947291baaSChristian Brauner /** 32047291baaSChristian Brauner * acl_permission_check - perform basic UNIX permission checking 321700b7940SChristian Brauner * @idmap: idmap of the mount the inode was found from 32247291baaSChristian Brauner * @inode: inode to check permissions on 32347291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 3245fc475b7SLinus Torvalds * 32547291baaSChristian Brauner * This function performs the basic UNIX permission checking. Since this 32647291baaSChristian Brauner * function may retrieve POSIX acls it needs to know whether it is called from a 32747291baaSChristian Brauner * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit. 32847291baaSChristian Brauner * 329700b7940SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 330700b7940SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 331700b7940SChristian Brauner * care to map the inode according to @idmap before checking permissions. 33247291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 333700b7940SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 3345909ccaaSLinus Torvalds */ 335700b7940SChristian Brauner static int acl_permission_check(struct mnt_idmap *idmap, 33647291baaSChristian Brauner struct inode *inode, int mask) 3375909ccaaSLinus Torvalds { 338700b7940SChristian Brauner struct user_namespace *mnt_userns = mnt_idmap_owner(idmap); 33926cf46beSLinus Torvalds unsigned int mode = inode->i_mode; 340a2bd096fSChristian Brauner vfsuid_t vfsuid; 3415909ccaaSLinus Torvalds 3425fc475b7SLinus Torvalds /* Are we the owner? If so, ACL's don't matter */ 343a2bd096fSChristian Brauner vfsuid = i_uid_into_vfsuid(mnt_userns, inode); 344a2bd096fSChristian Brauner if (likely(vfsuid_eq_kuid(vfsuid, current_fsuid()))) { 3455fc475b7SLinus Torvalds mask &= 7; 3465909ccaaSLinus Torvalds mode >>= 6; 3475fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3485fc475b7SLinus Torvalds } 3495fc475b7SLinus Torvalds 3505fc475b7SLinus Torvalds /* Do we have ACL's? */ 351e77819e5SLinus Torvalds if (IS_POSIXACL(inode) && (mode & S_IRWXG)) { 352700b7940SChristian Brauner int error = check_acl(idmap, inode, mask); 3535909ccaaSLinus Torvalds if (error != -EAGAIN) 3545909ccaaSLinus Torvalds return error; 3555909ccaaSLinus Torvalds } 3565909ccaaSLinus Torvalds 3575fc475b7SLinus Torvalds /* Only RWX matters for group/other mode bits */ 3585fc475b7SLinus Torvalds mask &= 7; 3595fc475b7SLinus Torvalds 3605fc475b7SLinus Torvalds /* 3615fc475b7SLinus Torvalds * Are the group permissions different from 3625fc475b7SLinus Torvalds * the other permissions in the bits we care 3635fc475b7SLinus Torvalds * about? Need to check group ownership if so. 3645fc475b7SLinus Torvalds */ 3655fc475b7SLinus Torvalds if (mask & (mode ^ (mode >> 3))) { 366a2bd096fSChristian Brauner vfsgid_t vfsgid = i_gid_into_vfsgid(mnt_userns, inode); 367a2bd096fSChristian Brauner if (vfsgid_in_group_p(vfsgid)) 3685909ccaaSLinus Torvalds mode >>= 3; 3695909ccaaSLinus Torvalds } 3705909ccaaSLinus Torvalds 3715fc475b7SLinus Torvalds /* Bits in 'mode' clear that we require? */ 3725fc475b7SLinus Torvalds return (mask & ~mode) ? -EACCES : 0; 3735909ccaaSLinus Torvalds } 3741da177e4SLinus Torvalds 3751da177e4SLinus Torvalds /** 3761da177e4SLinus Torvalds * generic_permission - check for access rights on a Posix-like filesystem 3774609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 3781da177e4SLinus Torvalds * @inode: inode to check access rights for 3795fc475b7SLinus Torvalds * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, 3805fc475b7SLinus Torvalds * %MAY_NOT_BLOCK ...) 3811da177e4SLinus Torvalds * 3821da177e4SLinus Torvalds * Used to check for read/write/execute permissions on a file. 3831da177e4SLinus Torvalds * We use "fsuid" for this, letting us set arbitrary permissions 3841da177e4SLinus Torvalds * for filesystem access without changing the "normal" uids which 385b74c79e9SNick Piggin * are used for other things. 386b74c79e9SNick Piggin * 387b74c79e9SNick Piggin * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk 388b74c79e9SNick Piggin * request cannot be satisfied (eg. requires blocking or too much complexity). 389b74c79e9SNick Piggin * It would then be called again in ref-walk mode. 39047291baaSChristian Brauner * 3914609e1f1SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3924609e1f1SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3934609e1f1SChristian Brauner * care to map the inode according to @idmap before checking permissions. 39447291baaSChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3954609e1f1SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 3961da177e4SLinus Torvalds */ 3974609e1f1SChristian Brauner int generic_permission(struct mnt_idmap *idmap, struct inode *inode, 39847291baaSChristian Brauner int mask) 3991da177e4SLinus Torvalds { 4005909ccaaSLinus Torvalds int ret; 4014609e1f1SChristian Brauner struct user_namespace *mnt_userns = mnt_idmap_owner(idmap); 4021da177e4SLinus Torvalds 4031da177e4SLinus Torvalds /* 404948409c7SAndreas Gruenbacher * Do the basic permission checks. 4051da177e4SLinus Torvalds */ 406700b7940SChristian Brauner ret = acl_permission_check(idmap, inode, mask); 4075909ccaaSLinus Torvalds if (ret != -EACCES) 4085909ccaaSLinus Torvalds return ret; 4091da177e4SLinus Torvalds 410d594e7ecSAl Viro if (S_ISDIR(inode->i_mode)) { 411d594e7ecSAl Viro /* DACs are overridable for directories */ 412d594e7ecSAl Viro if (!(mask & MAY_WRITE)) 41347291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 41423adbe12SAndy Lutomirski CAP_DAC_READ_SEARCH)) 415d594e7ecSAl Viro return 0; 41647291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4170558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4181da177e4SLinus Torvalds return 0; 4192a4c2242SStephen Smalley return -EACCES; 4202a4c2242SStephen Smalley } 4211da177e4SLinus Torvalds 4221da177e4SLinus Torvalds /* 4231da177e4SLinus Torvalds * Searching includes executable on directories, else just read. 4241da177e4SLinus Torvalds */ 4257ea66001SSerge E. Hallyn mask &= MAY_READ | MAY_WRITE | MAY_EXEC; 426d594e7ecSAl Viro if (mask == MAY_READ) 42747291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4280558c1bfSChristian Brauner CAP_DAC_READ_SEARCH)) 4291da177e4SLinus Torvalds return 0; 4302a4c2242SStephen Smalley /* 4312a4c2242SStephen Smalley * Read/write DACs are always overridable. 4322a4c2242SStephen Smalley * Executable DACs are overridable when there is 4332a4c2242SStephen Smalley * at least one exec bit set. 4342a4c2242SStephen Smalley */ 4352a4c2242SStephen Smalley if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO)) 43647291baaSChristian Brauner if (capable_wrt_inode_uidgid(mnt_userns, inode, 4370558c1bfSChristian Brauner CAP_DAC_OVERRIDE)) 4382a4c2242SStephen Smalley return 0; 4391da177e4SLinus Torvalds 4401da177e4SLinus Torvalds return -EACCES; 4411da177e4SLinus Torvalds } 4424d359507SAl Viro EXPORT_SYMBOL(generic_permission); 4431da177e4SLinus Torvalds 44447291baaSChristian Brauner /** 44547291baaSChristian Brauner * do_inode_permission - UNIX permission checking 4464609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 44747291baaSChristian Brauner * @inode: inode to check permissions on 44847291baaSChristian Brauner * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...) 44947291baaSChristian Brauner * 4503ddcd056SLinus Torvalds * We _really_ want to just do "generic_permission()" without 4513ddcd056SLinus Torvalds * even looking at the inode->i_op values. So we keep a cache 4523ddcd056SLinus Torvalds * flag in inode->i_opflags, that says "this has not special 4533ddcd056SLinus Torvalds * permission function, use the fast case". 4543ddcd056SLinus Torvalds */ 4554609e1f1SChristian Brauner static inline int do_inode_permission(struct mnt_idmap *idmap, 45647291baaSChristian Brauner struct inode *inode, int mask) 4573ddcd056SLinus Torvalds { 4583ddcd056SLinus Torvalds if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) { 4593ddcd056SLinus Torvalds if (likely(inode->i_op->permission)) 4604609e1f1SChristian Brauner return inode->i_op->permission(idmap, inode, mask); 4613ddcd056SLinus Torvalds 4623ddcd056SLinus Torvalds /* This gets set once for the inode lifetime */ 4633ddcd056SLinus Torvalds spin_lock(&inode->i_lock); 4643ddcd056SLinus Torvalds inode->i_opflags |= IOP_FASTPERM; 4653ddcd056SLinus Torvalds spin_unlock(&inode->i_lock); 4663ddcd056SLinus Torvalds } 4674609e1f1SChristian Brauner return generic_permission(idmap, inode, mask); 4683ddcd056SLinus Torvalds } 4693ddcd056SLinus Torvalds 470cb23beb5SChristoph Hellwig /** 4710bdaea90SDavid Howells * sb_permission - Check superblock-level permissions 4720bdaea90SDavid Howells * @sb: Superblock of inode to check permission on 47355852635SRandy Dunlap * @inode: Inode to check permission on 4740bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4750bdaea90SDavid Howells * 4760bdaea90SDavid Howells * Separate out file-system wide checks from inode-specific permission checks. 4770bdaea90SDavid Howells */ 4780bdaea90SDavid Howells static int sb_permission(struct super_block *sb, struct inode *inode, int mask) 4790bdaea90SDavid Howells { 4800bdaea90SDavid Howells if (unlikely(mask & MAY_WRITE)) { 4810bdaea90SDavid Howells umode_t mode = inode->i_mode; 4820bdaea90SDavid Howells 4830bdaea90SDavid Howells /* Nobody gets write access to a read-only fs. */ 484bc98a42cSDavid Howells if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) 4850bdaea90SDavid Howells return -EROFS; 4860bdaea90SDavid Howells } 4870bdaea90SDavid Howells return 0; 4880bdaea90SDavid Howells } 4890bdaea90SDavid Howells 4900bdaea90SDavid Howells /** 4910bdaea90SDavid Howells * inode_permission - Check for access rights to a given inode 4924609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 4930bdaea90SDavid Howells * @inode: Inode to check permission on 4940bdaea90SDavid Howells * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC) 4950bdaea90SDavid Howells * 4960bdaea90SDavid Howells * Check for read/write/execute permissions on an inode. We use fs[ug]id for 4970bdaea90SDavid Howells * this, letting us set arbitrary permissions for filesystem access without 4980bdaea90SDavid Howells * changing the "normal" UIDs which are used for other things. 4990bdaea90SDavid Howells * 5000bdaea90SDavid Howells * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask. 5010bdaea90SDavid Howells */ 5024609e1f1SChristian Brauner int inode_permission(struct mnt_idmap *idmap, 50347291baaSChristian Brauner struct inode *inode, int mask) 5040bdaea90SDavid Howells { 5050bdaea90SDavid Howells int retval; 5060bdaea90SDavid Howells 5070bdaea90SDavid Howells retval = sb_permission(inode->i_sb, inode, mask); 5080bdaea90SDavid Howells if (retval) 5090bdaea90SDavid Howells return retval; 5104bfd054aSEric Biggers 5114bfd054aSEric Biggers if (unlikely(mask & MAY_WRITE)) { 5124bfd054aSEric Biggers /* 5134bfd054aSEric Biggers * Nobody gets write access to an immutable file. 5144bfd054aSEric Biggers */ 5154bfd054aSEric Biggers if (IS_IMMUTABLE(inode)) 5164bfd054aSEric Biggers return -EPERM; 5174bfd054aSEric Biggers 5184bfd054aSEric Biggers /* 5194bfd054aSEric Biggers * Updating mtime will likely cause i_uid and i_gid to be 5204bfd054aSEric Biggers * written back improperly if their true value is unknown 5214bfd054aSEric Biggers * to the vfs. 5224bfd054aSEric Biggers */ 5234609e1f1SChristian Brauner if (HAS_UNMAPPED_ID(idmap, inode)) 5244bfd054aSEric Biggers return -EACCES; 5254bfd054aSEric Biggers } 5264bfd054aSEric Biggers 5274609e1f1SChristian Brauner retval = do_inode_permission(idmap, inode, mask); 5284bfd054aSEric Biggers if (retval) 5294bfd054aSEric Biggers return retval; 5304bfd054aSEric Biggers 5314bfd054aSEric Biggers retval = devcgroup_inode_permission(inode, mask); 5324bfd054aSEric Biggers if (retval) 5334bfd054aSEric Biggers return retval; 5344bfd054aSEric Biggers 5354bfd054aSEric Biggers return security_inode_permission(inode, mask); 5360bdaea90SDavid Howells } 5374d359507SAl Viro EXPORT_SYMBOL(inode_permission); 5380bdaea90SDavid Howells 5390bdaea90SDavid Howells /** 5405dd784d0SJan Blunck * path_get - get a reference to a path 5415dd784d0SJan Blunck * @path: path to get the reference to 5425dd784d0SJan Blunck * 5435dd784d0SJan Blunck * Given a path increment the reference count to the dentry and the vfsmount. 5445dd784d0SJan Blunck */ 545dcf787f3SAl Viro void path_get(const struct path *path) 5465dd784d0SJan Blunck { 5475dd784d0SJan Blunck mntget(path->mnt); 5485dd784d0SJan Blunck dget(path->dentry); 5495dd784d0SJan Blunck } 5505dd784d0SJan Blunck EXPORT_SYMBOL(path_get); 5515dd784d0SJan Blunck 5525dd784d0SJan Blunck /** 5531d957f9bSJan Blunck * path_put - put a reference to a path 5541d957f9bSJan Blunck * @path: path to put the reference to 5551d957f9bSJan Blunck * 5561d957f9bSJan Blunck * Given a path decrement the reference count to the dentry and the vfsmount. 5571d957f9bSJan Blunck */ 558dcf787f3SAl Viro void path_put(const struct path *path) 5591da177e4SLinus Torvalds { 5601d957f9bSJan Blunck dput(path->dentry); 5611d957f9bSJan Blunck mntput(path->mnt); 5621da177e4SLinus Torvalds } 5631d957f9bSJan Blunck EXPORT_SYMBOL(path_put); 5641da177e4SLinus Torvalds 565894bc8c4SAl Viro #define EMBEDDED_LEVELS 2 5661f55a6ecSAl Viro struct nameidata { 5671f55a6ecSAl Viro struct path path; 5681f55a6ecSAl Viro struct qstr last; 5691f55a6ecSAl Viro struct path root; 5701f55a6ecSAl Viro struct inode *inode; /* path.dentry.d_inode */ 571bcba1e7dSAl Viro unsigned int flags, state; 57203fa86e9SAl Viro unsigned seq, next_seq, m_seq, r_seq; 5731f55a6ecSAl Viro int last_type; 5741f55a6ecSAl Viro unsigned depth; 575756daf26SNeilBrown int total_link_count; 576697fc6caSAl Viro struct saved { 577697fc6caSAl Viro struct path link; 578fceef393SAl Viro struct delayed_call done; 579697fc6caSAl Viro const char *name; 5800450b2d1SAl Viro unsigned seq; 581894bc8c4SAl Viro } *stack, internal[EMBEDDED_LEVELS]; 5829883d185SAl Viro struct filename *name; 5839883d185SAl Viro struct nameidata *saved; 5849883d185SAl Viro unsigned root_seq; 5859883d185SAl Viro int dfd; 586a2bd096fSChristian Brauner vfsuid_t dir_vfsuid; 5870f705953SAl Viro umode_t dir_mode; 5883859a271SKees Cook } __randomize_layout; 5891f55a6ecSAl Viro 590bcba1e7dSAl Viro #define ND_ROOT_PRESET 1 591bcba1e7dSAl Viro #define ND_ROOT_GRABBED 2 592bcba1e7dSAl Viro #define ND_JUMPED 4 593bcba1e7dSAl Viro 59406422964SAl Viro static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name) 595894bc8c4SAl Viro { 596756daf26SNeilBrown struct nameidata *old = current->nameidata; 597756daf26SNeilBrown p->stack = p->internal; 5987962c7d1SAl Viro p->depth = 0; 599c8a53ee5SAl Viro p->dfd = dfd; 600c8a53ee5SAl Viro p->name = name; 6017d01ef75SAl Viro p->path.mnt = NULL; 6027d01ef75SAl Viro p->path.dentry = NULL; 603756daf26SNeilBrown p->total_link_count = old ? old->total_link_count : 0; 6049883d185SAl Viro p->saved = old; 605756daf26SNeilBrown current->nameidata = p; 606894bc8c4SAl Viro } 607894bc8c4SAl Viro 60806422964SAl Viro static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name, 60906422964SAl Viro const struct path *root) 61006422964SAl Viro { 61106422964SAl Viro __set_nameidata(p, dfd, name); 61206422964SAl Viro p->state = 0; 61306422964SAl Viro if (unlikely(root)) { 61406422964SAl Viro p->state = ND_ROOT_PRESET; 61506422964SAl Viro p->root = *root; 61606422964SAl Viro } 61706422964SAl Viro } 61806422964SAl Viro 6199883d185SAl Viro static void restore_nameidata(void) 620894bc8c4SAl Viro { 6219883d185SAl Viro struct nameidata *now = current->nameidata, *old = now->saved; 622756daf26SNeilBrown 623756daf26SNeilBrown current->nameidata = old; 624756daf26SNeilBrown if (old) 625756daf26SNeilBrown old->total_link_count = now->total_link_count; 626e1a63bbcSAl Viro if (now->stack != now->internal) 627756daf26SNeilBrown kfree(now->stack); 628894bc8c4SAl Viro } 629894bc8c4SAl Viro 63060ef60c7SAl Viro static bool nd_alloc_stack(struct nameidata *nd) 631894bc8c4SAl Viro { 632bc40aee0SAl Viro struct saved *p; 633bc40aee0SAl Viro 6346da2ec56SKees Cook p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved), 63560ef60c7SAl Viro nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL); 636bc40aee0SAl Viro if (unlikely(!p)) 63760ef60c7SAl Viro return false; 638894bc8c4SAl Viro memcpy(p, nd->internal, sizeof(nd->internal)); 639894bc8c4SAl Viro nd->stack = p; 64060ef60c7SAl Viro return true; 641894bc8c4SAl Viro } 642894bc8c4SAl Viro 643397d425dSEric W. Biederman /** 6446b03f7edSAl Viro * path_connected - Verify that a dentry is below mnt.mnt_root 645397d425dSEric W. Biederman * 646397d425dSEric W. Biederman * Rename can sometimes move a file or directory outside of a bind 647397d425dSEric W. Biederman * mount, path_connected allows those cases to be detected. 648397d425dSEric W. Biederman */ 6496b03f7edSAl Viro static bool path_connected(struct vfsmount *mnt, struct dentry *dentry) 650397d425dSEric W. Biederman { 65195dd7758SEric W. Biederman struct super_block *sb = mnt->mnt_sb; 652397d425dSEric W. Biederman 653402dd2cfSChristoph Hellwig /* Bind mounts can have disconnected paths */ 654402dd2cfSChristoph Hellwig if (mnt->mnt_root == sb->s_root) 655397d425dSEric W. Biederman return true; 656397d425dSEric W. Biederman 6576b03f7edSAl Viro return is_subdir(dentry, mnt->mnt_root); 658397d425dSEric W. Biederman } 659397d425dSEric W. Biederman 6607973387aSAl Viro static void drop_links(struct nameidata *nd) 6617973387aSAl Viro { 6627973387aSAl Viro int i = nd->depth; 6637973387aSAl Viro while (i--) { 6647973387aSAl Viro struct saved *last = nd->stack + i; 665fceef393SAl Viro do_delayed_call(&last->done); 666fceef393SAl Viro clear_delayed_call(&last->done); 6677973387aSAl Viro } 6687973387aSAl Viro } 6697973387aSAl Viro 6706e180327SAl Viro static void leave_rcu(struct nameidata *nd) 6716e180327SAl Viro { 6726e180327SAl Viro nd->flags &= ~LOOKUP_RCU; 67303fa86e9SAl Viro nd->seq = nd->next_seq = 0; 6746e180327SAl Viro rcu_read_unlock(); 6756e180327SAl Viro } 6766e180327SAl Viro 6777973387aSAl Viro static void terminate_walk(struct nameidata *nd) 6787973387aSAl Viro { 6797973387aSAl Viro drop_links(nd); 6807973387aSAl Viro if (!(nd->flags & LOOKUP_RCU)) { 6817973387aSAl Viro int i; 6827973387aSAl Viro path_put(&nd->path); 6837973387aSAl Viro for (i = 0; i < nd->depth; i++) 6847973387aSAl Viro path_put(&nd->stack[i].link); 685bcba1e7dSAl Viro if (nd->state & ND_ROOT_GRABBED) { 686102b8af2SAl Viro path_put(&nd->root); 687bcba1e7dSAl Viro nd->state &= ~ND_ROOT_GRABBED; 688102b8af2SAl Viro } 6897973387aSAl Viro } else { 6906e180327SAl Viro leave_rcu(nd); 6917973387aSAl Viro } 6927973387aSAl Viro nd->depth = 0; 6937d01ef75SAl Viro nd->path.mnt = NULL; 6947d01ef75SAl Viro nd->path.dentry = NULL; 6957973387aSAl Viro } 6967973387aSAl Viro 6977973387aSAl Viro /* path_put is needed afterwards regardless of success or failure */ 6982aa38470SAl Viro static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq) 6997973387aSAl Viro { 7002aa38470SAl Viro int res = __legitimize_mnt(path->mnt, mseq); 7017973387aSAl Viro if (unlikely(res)) { 7027973387aSAl Viro if (res > 0) 7037973387aSAl Viro path->mnt = NULL; 7047973387aSAl Viro path->dentry = NULL; 7057973387aSAl Viro return false; 7067973387aSAl Viro } 7077973387aSAl Viro if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) { 7087973387aSAl Viro path->dentry = NULL; 7097973387aSAl Viro return false; 7107973387aSAl Viro } 7117973387aSAl Viro return !read_seqcount_retry(&path->dentry->d_seq, seq); 7127973387aSAl Viro } 7137973387aSAl Viro 7142aa38470SAl Viro static inline bool legitimize_path(struct nameidata *nd, 7152aa38470SAl Viro struct path *path, unsigned seq) 7162aa38470SAl Viro { 7175bd73286SAl Viro return __legitimize_path(path, seq, nd->m_seq); 7182aa38470SAl Viro } 7192aa38470SAl Viro 7207973387aSAl Viro static bool legitimize_links(struct nameidata *nd) 7217973387aSAl Viro { 7227973387aSAl Viro int i; 723eacd9aa8SAl Viro if (unlikely(nd->flags & LOOKUP_CACHED)) { 724eacd9aa8SAl Viro drop_links(nd); 725eacd9aa8SAl Viro nd->depth = 0; 726eacd9aa8SAl Viro return false; 727eacd9aa8SAl Viro } 7287973387aSAl Viro for (i = 0; i < nd->depth; i++) { 7297973387aSAl Viro struct saved *last = nd->stack + i; 7307973387aSAl Viro if (unlikely(!legitimize_path(nd, &last->link, last->seq))) { 7317973387aSAl Viro drop_links(nd); 7327973387aSAl Viro nd->depth = i + 1; 7337973387aSAl Viro return false; 7347973387aSAl Viro } 7357973387aSAl Viro } 7367973387aSAl Viro return true; 7377973387aSAl Viro } 7387973387aSAl Viro 739ee594bffSAl Viro static bool legitimize_root(struct nameidata *nd) 740ee594bffSAl Viro { 741adb21d2bSAleksa Sarai /* Nothing to do if nd->root is zero or is managed by the VFS user. */ 742bcba1e7dSAl Viro if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET)) 743ee594bffSAl Viro return true; 744bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 745ee594bffSAl Viro return legitimize_path(nd, &nd->root, nd->root_seq); 746ee594bffSAl Viro } 747ee594bffSAl Viro 74819660af7SAl Viro /* 74931e6b01fSNick Piggin * Path walking has 2 modes, rcu-walk and ref-walk (see 75019660af7SAl Viro * Documentation/filesystems/path-lookup.txt). In situations when we can't 75119660af7SAl Viro * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab 75257e3715cSMike Marshall * normal reference counts on dentries and vfsmounts to transition to ref-walk 75319660af7SAl Viro * mode. Refcounts are grabbed at the last known good point before rcu-walk 75419660af7SAl Viro * got stuck, so ref-walk may continue from there. If this is not successful 75519660af7SAl Viro * (eg. a seqcount has changed), then failure is returned and it's up to caller 75619660af7SAl Viro * to restart the path walk from the beginning in ref-walk mode. 75731e6b01fSNick Piggin */ 75831e6b01fSNick Piggin 75931e6b01fSNick Piggin /** 760e36cffedSJens Axboe * try_to_unlazy - try to switch to ref-walk mode. 76119660af7SAl Viro * @nd: nameidata pathwalk data 762e36cffedSJens Axboe * Returns: true on success, false on failure 76331e6b01fSNick Piggin * 764e36cffedSJens Axboe * try_to_unlazy attempts to legitimize the current nd->path and nd->root 7654675ac39SAl Viro * for ref-walk mode. 7664675ac39SAl Viro * Must be called from rcu-walk context. 767e36cffedSJens Axboe * Nothing should touch nameidata between try_to_unlazy() failure and 7687973387aSAl Viro * terminate_walk(). 76931e6b01fSNick Piggin */ 770e36cffedSJens Axboe static bool try_to_unlazy(struct nameidata *nd) 77131e6b01fSNick Piggin { 77231e6b01fSNick Piggin struct dentry *parent = nd->path.dentry; 77331e6b01fSNick Piggin 77431e6b01fSNick Piggin BUG_ON(!(nd->flags & LOOKUP_RCU)); 775e5c832d5SLinus Torvalds 7767973387aSAl Viro if (unlikely(!legitimize_links(nd))) 7774675ac39SAl Viro goto out1; 77884a2bd39SAl Viro if (unlikely(!legitimize_path(nd, &nd->path, nd->seq))) 77984a2bd39SAl Viro goto out; 780ee594bffSAl Viro if (unlikely(!legitimize_root(nd))) 7814675ac39SAl Viro goto out; 7826e180327SAl Viro leave_rcu(nd); 7834675ac39SAl Viro BUG_ON(nd->inode != parent->d_inode); 784e36cffedSJens Axboe return true; 7854675ac39SAl Viro 78684a2bd39SAl Viro out1: 7874675ac39SAl Viro nd->path.mnt = NULL; 7884675ac39SAl Viro nd->path.dentry = NULL; 7894675ac39SAl Viro out: 7906e180327SAl Viro leave_rcu(nd); 791e36cffedSJens Axboe return false; 7924675ac39SAl Viro } 7934675ac39SAl Viro 7944675ac39SAl Viro /** 795ae66db45SAl Viro * try_to_unlazy_next - try to switch to ref-walk mode. 7964675ac39SAl Viro * @nd: nameidata pathwalk data 797ae66db45SAl Viro * @dentry: next dentry to step into 798ae66db45SAl Viro * Returns: true on success, false on failure 7994675ac39SAl Viro * 80030476f7eSTom Rix * Similar to try_to_unlazy(), but here we have the next dentry already 801ae66db45SAl Viro * picked by rcu-walk and want to legitimize that in addition to the current 802ae66db45SAl Viro * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context. 803ae66db45SAl Viro * Nothing should touch nameidata between try_to_unlazy_next() failure and 8044675ac39SAl Viro * terminate_walk(). 8054675ac39SAl Viro */ 80603fa86e9SAl Viro static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry) 8074675ac39SAl Viro { 8087e4745a0SAl Viro int res; 8094675ac39SAl Viro BUG_ON(!(nd->flags & LOOKUP_RCU)); 8104675ac39SAl Viro 8114675ac39SAl Viro if (unlikely(!legitimize_links(nd))) 8124675ac39SAl Viro goto out2; 8137e4745a0SAl Viro res = __legitimize_mnt(nd->path.mnt, nd->m_seq); 8147e4745a0SAl Viro if (unlikely(res)) { 8157e4745a0SAl Viro if (res > 0) 8167973387aSAl Viro goto out2; 8177e4745a0SAl Viro goto out1; 8187e4745a0SAl Viro } 8194675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref))) 8207973387aSAl Viro goto out1; 82148a066e7SAl Viro 82215570086SLinus Torvalds /* 8234675ac39SAl Viro * We need to move both the parent and the dentry from the RCU domain 8244675ac39SAl Viro * to be properly refcounted. And the sequence number in the dentry 8254675ac39SAl Viro * validates *both* dentry counters, since we checked the sequence 8264675ac39SAl Viro * number of the parent after we got the child sequence number. So we 8274675ac39SAl Viro * know the parent must still be valid if the child sequence number is 82815570086SLinus Torvalds */ 8294675ac39SAl Viro if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) 830e5c832d5SLinus Torvalds goto out; 83103fa86e9SAl Viro if (read_seqcount_retry(&dentry->d_seq, nd->next_seq)) 83284a2bd39SAl Viro goto out_dput; 833e5c832d5SLinus Torvalds /* 834e5c832d5SLinus Torvalds * Sequence counts matched. Now make sure that the root is 835e5c832d5SLinus Torvalds * still valid and get it if required. 836e5c832d5SLinus Torvalds */ 83784a2bd39SAl Viro if (unlikely(!legitimize_root(nd))) 83884a2bd39SAl Viro goto out_dput; 8396e180327SAl Viro leave_rcu(nd); 840ae66db45SAl Viro return true; 84119660af7SAl Viro 8427973387aSAl Viro out2: 8437973387aSAl Viro nd->path.mnt = NULL; 8447973387aSAl Viro out1: 8457973387aSAl Viro nd->path.dentry = NULL; 846e5c832d5SLinus Torvalds out: 8476e180327SAl Viro leave_rcu(nd); 848ae66db45SAl Viro return false; 84984a2bd39SAl Viro out_dput: 8506e180327SAl Viro leave_rcu(nd); 85184a2bd39SAl Viro dput(dentry); 852ae66db45SAl Viro return false; 85331e6b01fSNick Piggin } 85431e6b01fSNick Piggin 8554ce16ef3SAl Viro static inline int d_revalidate(struct dentry *dentry, unsigned int flags) 85634286d66SNick Piggin { 857a89f8337SAl Viro if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) 8584ce16ef3SAl Viro return dentry->d_op->d_revalidate(dentry, flags); 859a89f8337SAl Viro else 860a89f8337SAl Viro return 1; 86134286d66SNick Piggin } 86234286d66SNick Piggin 8639f1fafeeSAl Viro /** 8649f1fafeeSAl Viro * complete_walk - successful completion of path walk 8659f1fafeeSAl Viro * @nd: pointer nameidata 86639159de2SJeff Layton * 8679f1fafeeSAl Viro * If we had been in RCU mode, drop out of it and legitimize nd->path. 8689f1fafeeSAl Viro * Revalidate the final result, unless we'd already done that during 8699f1fafeeSAl Viro * the path walk or the filesystem doesn't ask for it. Return 0 on 8709f1fafeeSAl Viro * success, -error on failure. In case of failure caller does not 8719f1fafeeSAl Viro * need to drop nd->path. 87239159de2SJeff Layton */ 8739f1fafeeSAl Viro static int complete_walk(struct nameidata *nd) 87439159de2SJeff Layton { 87516c2cd71SAl Viro struct dentry *dentry = nd->path.dentry; 87639159de2SJeff Layton int status; 87739159de2SJeff Layton 8789f1fafeeSAl Viro if (nd->flags & LOOKUP_RCU) { 879adb21d2bSAleksa Sarai /* 880adb21d2bSAleksa Sarai * We don't want to zero nd->root for scoped-lookups or 881adb21d2bSAleksa Sarai * externally-managed nd->root. 882adb21d2bSAleksa Sarai */ 883bcba1e7dSAl Viro if (!(nd->state & ND_ROOT_PRESET)) 884bcba1e7dSAl Viro if (!(nd->flags & LOOKUP_IS_SCOPED)) 8859f1fafeeSAl Viro nd->root.mnt = NULL; 8866c6ec2b0SJens Axboe nd->flags &= ~LOOKUP_CACHED; 887e36cffedSJens Axboe if (!try_to_unlazy(nd)) 88848a066e7SAl Viro return -ECHILD; 88948a066e7SAl Viro } 8909f1fafeeSAl Viro 891adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 892adb21d2bSAleksa Sarai /* 893adb21d2bSAleksa Sarai * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't 894adb21d2bSAleksa Sarai * ever step outside the root during lookup" and should already 895adb21d2bSAleksa Sarai * be guaranteed by the rest of namei, we want to avoid a namei 896adb21d2bSAleksa Sarai * BUG resulting in userspace being given a path that was not 897adb21d2bSAleksa Sarai * scoped within the root at some point during the lookup. 898adb21d2bSAleksa Sarai * 899adb21d2bSAleksa Sarai * So, do a final sanity-check to make sure that in the 900adb21d2bSAleksa Sarai * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED) 901adb21d2bSAleksa Sarai * we won't silently return an fd completely outside of the 902adb21d2bSAleksa Sarai * requested root to userspace. 903adb21d2bSAleksa Sarai * 904adb21d2bSAleksa Sarai * Userspace could move the path outside the root after this 905adb21d2bSAleksa Sarai * check, but as discussed elsewhere this is not a concern (the 906adb21d2bSAleksa Sarai * resolved file was inside the root at some point). 907adb21d2bSAleksa Sarai */ 908adb21d2bSAleksa Sarai if (!path_is_under(&nd->path, &nd->root)) 909adb21d2bSAleksa Sarai return -EXDEV; 910adb21d2bSAleksa Sarai } 911adb21d2bSAleksa Sarai 912bcba1e7dSAl Viro if (likely(!(nd->state & ND_JUMPED))) 91339159de2SJeff Layton return 0; 91439159de2SJeff Layton 915ecf3d1f1SJeff Layton if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE))) 91616c2cd71SAl Viro return 0; 91716c2cd71SAl Viro 918ecf3d1f1SJeff Layton status = dentry->d_op->d_weak_revalidate(dentry, nd->flags); 91939159de2SJeff Layton if (status > 0) 92039159de2SJeff Layton return 0; 92139159de2SJeff Layton 92216c2cd71SAl Viro if (!status) 92339159de2SJeff Layton status = -ESTALE; 92416c2cd71SAl Viro 92539159de2SJeff Layton return status; 92639159de2SJeff Layton } 92739159de2SJeff Layton 928740a1678SAleksa Sarai static int set_root(struct nameidata *nd) 9292a737871SAl Viro { 93031e6b01fSNick Piggin struct fs_struct *fs = current->fs; 9319e6697e2SAl Viro 932adb21d2bSAleksa Sarai /* 933adb21d2bSAleksa Sarai * Jumping to the real root in a scoped-lookup is a BUG in namei, but we 934adb21d2bSAleksa Sarai * still have to ensure it doesn't happen because it will cause a breakout 935adb21d2bSAleksa Sarai * from the dirfd. 936adb21d2bSAleksa Sarai */ 937adb21d2bSAleksa Sarai if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED)) 938adb21d2bSAleksa Sarai return -ENOTRECOVERABLE; 939adb21d2bSAleksa Sarai 9409e6697e2SAl Viro if (nd->flags & LOOKUP_RCU) { 9418f47a016SAl Viro unsigned seq; 942c28cc364SNick Piggin 943c28cc364SNick Piggin do { 944c28cc364SNick Piggin seq = read_seqcount_begin(&fs->seq); 94531e6b01fSNick Piggin nd->root = fs->root; 9468f47a016SAl Viro nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq); 947c28cc364SNick Piggin } while (read_seqcount_retry(&fs->seq, seq)); 9489e6697e2SAl Viro } else { 9499e6697e2SAl Viro get_fs_root(fs, &nd->root); 950bcba1e7dSAl Viro nd->state |= ND_ROOT_GRABBED; 9519e6697e2SAl Viro } 952740a1678SAleksa Sarai return 0; 95331e6b01fSNick Piggin } 95431e6b01fSNick Piggin 955248fb5b9SAl Viro static int nd_jump_root(struct nameidata *nd) 956248fb5b9SAl Viro { 957adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_BENEATH)) 958adb21d2bSAleksa Sarai return -EXDEV; 95972ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 96072ba2929SAleksa Sarai /* Absolute path arguments to path_init() are allowed. */ 96172ba2929SAleksa Sarai if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt) 96272ba2929SAleksa Sarai return -EXDEV; 96372ba2929SAleksa Sarai } 964740a1678SAleksa Sarai if (!nd->root.mnt) { 965740a1678SAleksa Sarai int error = set_root(nd); 966740a1678SAleksa Sarai if (error) 967740a1678SAleksa Sarai return error; 968740a1678SAleksa Sarai } 969248fb5b9SAl Viro if (nd->flags & LOOKUP_RCU) { 970248fb5b9SAl Viro struct dentry *d; 971248fb5b9SAl Viro nd->path = nd->root; 972248fb5b9SAl Viro d = nd->path.dentry; 973248fb5b9SAl Viro nd->inode = d->d_inode; 974248fb5b9SAl Viro nd->seq = nd->root_seq; 97582ef0698SAl Viro if (read_seqcount_retry(&d->d_seq, nd->seq)) 976248fb5b9SAl Viro return -ECHILD; 977248fb5b9SAl Viro } else { 978248fb5b9SAl Viro path_put(&nd->path); 979248fb5b9SAl Viro nd->path = nd->root; 980248fb5b9SAl Viro path_get(&nd->path); 981248fb5b9SAl Viro nd->inode = nd->path.dentry->d_inode; 982248fb5b9SAl Viro } 983bcba1e7dSAl Viro nd->state |= ND_JUMPED; 984248fb5b9SAl Viro return 0; 985248fb5b9SAl Viro } 986248fb5b9SAl Viro 987b5fb63c1SChristoph Hellwig /* 9886b255391SAl Viro * Helper to directly jump to a known parsed path from ->get_link, 989b5fb63c1SChristoph Hellwig * caller must have taken a reference to path beforehand. 990b5fb63c1SChristoph Hellwig */ 991ea4af4aaSAl Viro int nd_jump_link(const struct path *path) 992b5fb63c1SChristoph Hellwig { 9934b99d499SAleksa Sarai int error = -ELOOP; 9946e77137bSAl Viro struct nameidata *nd = current->nameidata; 995b5fb63c1SChristoph Hellwig 9964b99d499SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS)) 9974b99d499SAleksa Sarai goto err; 9984b99d499SAleksa Sarai 99972ba2929SAleksa Sarai error = -EXDEV; 100072ba2929SAleksa Sarai if (unlikely(nd->flags & LOOKUP_NO_XDEV)) { 100172ba2929SAleksa Sarai if (nd->path.mnt != path->mnt) 100272ba2929SAleksa Sarai goto err; 100372ba2929SAleksa Sarai } 1004adb21d2bSAleksa Sarai /* Not currently safe for scoped-lookups. */ 1005adb21d2bSAleksa Sarai if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) 1006adb21d2bSAleksa Sarai goto err; 100772ba2929SAleksa Sarai 10084b99d499SAleksa Sarai path_put(&nd->path); 1009b5fb63c1SChristoph Hellwig nd->path = *path; 1010b5fb63c1SChristoph Hellwig nd->inode = nd->path.dentry->d_inode; 1011bcba1e7dSAl Viro nd->state |= ND_JUMPED; 10121bc82070SAleksa Sarai return 0; 10134b99d499SAleksa Sarai 10144b99d499SAleksa Sarai err: 10154b99d499SAleksa Sarai path_put(path); 10164b99d499SAleksa Sarai return error; 1017b5fb63c1SChristoph Hellwig } 1018b5fb63c1SChristoph Hellwig 1019b9ff4429SAl Viro static inline void put_link(struct nameidata *nd) 1020574197e0SAl Viro { 102121c3003dSAl Viro struct saved *last = nd->stack + --nd->depth; 1022fceef393SAl Viro do_delayed_call(&last->done); 10236548fae2SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1024b9ff4429SAl Viro path_put(&last->link); 1025574197e0SAl Viro } 1026574197e0SAl Viro 10279c011be1SLuis Chamberlain static int sysctl_protected_symlinks __read_mostly; 10289c011be1SLuis Chamberlain static int sysctl_protected_hardlinks __read_mostly; 10299c011be1SLuis Chamberlain static int sysctl_protected_fifos __read_mostly; 10309c011be1SLuis Chamberlain static int sysctl_protected_regular __read_mostly; 10319c011be1SLuis Chamberlain 10329c011be1SLuis Chamberlain #ifdef CONFIG_SYSCTL 10339c011be1SLuis Chamberlain static struct ctl_table namei_sysctls[] = { 10349c011be1SLuis Chamberlain { 10359c011be1SLuis Chamberlain .procname = "protected_symlinks", 10369c011be1SLuis Chamberlain .data = &sysctl_protected_symlinks, 10379c011be1SLuis Chamberlain .maxlen = sizeof(int), 1038c7031c14SJulius Hemanth Pitti .mode = 0644, 10399c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10409c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10419c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10429c011be1SLuis Chamberlain }, 10439c011be1SLuis Chamberlain { 10449c011be1SLuis Chamberlain .procname = "protected_hardlinks", 10459c011be1SLuis Chamberlain .data = &sysctl_protected_hardlinks, 10469c011be1SLuis Chamberlain .maxlen = sizeof(int), 1047c7031c14SJulius Hemanth Pitti .mode = 0644, 10489c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10499c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10509c011be1SLuis Chamberlain .extra2 = SYSCTL_ONE, 10519c011be1SLuis Chamberlain }, 10529c011be1SLuis Chamberlain { 10539c011be1SLuis Chamberlain .procname = "protected_fifos", 10549c011be1SLuis Chamberlain .data = &sysctl_protected_fifos, 10559c011be1SLuis Chamberlain .maxlen = sizeof(int), 1056c7031c14SJulius Hemanth Pitti .mode = 0644, 10579c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10589c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10599c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10609c011be1SLuis Chamberlain }, 10619c011be1SLuis Chamberlain { 10629c011be1SLuis Chamberlain .procname = "protected_regular", 10639c011be1SLuis Chamberlain .data = &sysctl_protected_regular, 10649c011be1SLuis Chamberlain .maxlen = sizeof(int), 1065c7031c14SJulius Hemanth Pitti .mode = 0644, 10669c011be1SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 10679c011be1SLuis Chamberlain .extra1 = SYSCTL_ZERO, 10689c011be1SLuis Chamberlain .extra2 = SYSCTL_TWO, 10699c011be1SLuis Chamberlain }, 10709c011be1SLuis Chamberlain { } 10719c011be1SLuis Chamberlain }; 10729c011be1SLuis Chamberlain 10739c011be1SLuis Chamberlain static int __init init_fs_namei_sysctls(void) 10749c011be1SLuis Chamberlain { 10759c011be1SLuis Chamberlain register_sysctl_init("fs", namei_sysctls); 10769c011be1SLuis Chamberlain return 0; 10779c011be1SLuis Chamberlain } 10789c011be1SLuis Chamberlain fs_initcall(init_fs_namei_sysctls); 10799c011be1SLuis Chamberlain 10809c011be1SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 1081800179c9SKees Cook 1082800179c9SKees Cook /** 1083800179c9SKees Cook * may_follow_link - Check symlink following for unsafe situations 108455852635SRandy Dunlap * @nd: nameidata pathwalk data 1085800179c9SKees Cook * 1086800179c9SKees Cook * In the case of the sysctl_protected_symlinks sysctl being enabled, 1087800179c9SKees Cook * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is 1088800179c9SKees Cook * in a sticky world-writable directory. This is to protect privileged 1089800179c9SKees Cook * processes from failing races against path names that may change out 1090800179c9SKees Cook * from under them by way of other users creating malicious symlinks. 1091800179c9SKees Cook * It will permit symlinks to be followed only when outside a sticky 1092800179c9SKees Cook * world-writable directory, or when the uid of the symlink and follower 1093800179c9SKees Cook * match, or when the directory owner matches the symlink's owner. 1094800179c9SKees Cook * 1095800179c9SKees Cook * Returns 0 if following the symlink is allowed, -ve on error. 1096800179c9SKees Cook */ 1097ad6cc4c3SAl Viro static inline int may_follow_link(struct nameidata *nd, const struct inode *inode) 1098800179c9SKees Cook { 1099ba73d987SChristian Brauner struct user_namespace *mnt_userns; 1100a2bd096fSChristian Brauner vfsuid_t vfsuid; 1101ba73d987SChristian Brauner 1102800179c9SKees Cook if (!sysctl_protected_symlinks) 1103800179c9SKees Cook return 0; 1104800179c9SKees Cook 1105ba73d987SChristian Brauner mnt_userns = mnt_user_ns(nd->path.mnt); 1106a2bd096fSChristian Brauner vfsuid = i_uid_into_vfsuid(mnt_userns, inode); 1107800179c9SKees Cook /* Allowed if owner and follower match. */ 1108a2bd096fSChristian Brauner if (vfsuid_eq_kuid(vfsuid, current_fsuid())) 1109800179c9SKees Cook return 0; 1110800179c9SKees Cook 1111800179c9SKees Cook /* Allowed if parent directory not sticky and world-writable. */ 11120f705953SAl Viro if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH)) 1113800179c9SKees Cook return 0; 1114800179c9SKees Cook 1115800179c9SKees Cook /* Allowed if parent directory and link owner match. */ 1116a2bd096fSChristian Brauner if (vfsuid_valid(nd->dir_vfsuid) && vfsuid_eq(nd->dir_vfsuid, vfsuid)) 1117800179c9SKees Cook return 0; 1118800179c9SKees Cook 111931956502SAl Viro if (nd->flags & LOOKUP_RCU) 112031956502SAl Viro return -ECHILD; 112131956502SAl Viro 1122ea841bafSRichard Guy Briggs audit_inode(nd->name, nd->stack[0].link.dentry, 0); 1123245d7369SKees Cook audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link"); 1124800179c9SKees Cook return -EACCES; 1125800179c9SKees Cook } 1126800179c9SKees Cook 1127800179c9SKees Cook /** 1128800179c9SKees Cook * safe_hardlink_source - Check for safe hardlink conditions 11294609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 1130800179c9SKees Cook * @inode: the source inode to hardlink from 1131800179c9SKees Cook * 1132800179c9SKees Cook * Return false if at least one of the following conditions: 1133800179c9SKees Cook * - inode is not a regular file 1134800179c9SKees Cook * - inode is setuid 1135800179c9SKees Cook * - inode is setgid and group-exec 1136800179c9SKees Cook * - access failure for read and write 1137800179c9SKees Cook * 1138800179c9SKees Cook * Otherwise returns true. 1139800179c9SKees Cook */ 11404609e1f1SChristian Brauner static bool safe_hardlink_source(struct mnt_idmap *idmap, 1141ba73d987SChristian Brauner struct inode *inode) 1142800179c9SKees Cook { 1143800179c9SKees Cook umode_t mode = inode->i_mode; 1144800179c9SKees Cook 1145800179c9SKees Cook /* Special files should not get pinned to the filesystem. */ 1146800179c9SKees Cook if (!S_ISREG(mode)) 1147800179c9SKees Cook return false; 1148800179c9SKees Cook 1149800179c9SKees Cook /* Setuid files should not get pinned to the filesystem. */ 1150800179c9SKees Cook if (mode & S_ISUID) 1151800179c9SKees Cook return false; 1152800179c9SKees Cook 1153800179c9SKees Cook /* Executable setgid files should not get pinned to the filesystem. */ 1154800179c9SKees Cook if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) 1155800179c9SKees Cook return false; 1156800179c9SKees Cook 1157800179c9SKees Cook /* Hardlinking to unreadable or unwritable sources is dangerous. */ 11584609e1f1SChristian Brauner if (inode_permission(idmap, inode, MAY_READ | MAY_WRITE)) 1159800179c9SKees Cook return false; 1160800179c9SKees Cook 1161800179c9SKees Cook return true; 1162800179c9SKees Cook } 1163800179c9SKees Cook 1164800179c9SKees Cook /** 1165800179c9SKees Cook * may_linkat - Check permissions for creating a hardlink 11664609e1f1SChristian Brauner * @idmap: idmap of the mount the inode was found from 1167800179c9SKees Cook * @link: the source to hardlink from 1168800179c9SKees Cook * 1169800179c9SKees Cook * Block hardlink when all of: 1170800179c9SKees Cook * - sysctl_protected_hardlinks enabled 1171800179c9SKees Cook * - fsuid does not match inode 1172800179c9SKees Cook * - hardlink source is unsafe (see safe_hardlink_source() above) 1173f2ca3796SDirk Steinmetz * - not CAP_FOWNER in a namespace with the inode owner uid mapped 1174800179c9SKees Cook * 11754609e1f1SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 11764609e1f1SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 11774609e1f1SChristian Brauner * care to map the inode according to @idmap before checking permissions. 1178ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 11794609e1f1SChristian Brauner * raw inode simply pass @nop_mnt_idmap. 1180ba73d987SChristian Brauner * 1181800179c9SKees Cook * Returns 0 if successful, -ve on error. 1182800179c9SKees Cook */ 11834609e1f1SChristian Brauner int may_linkat(struct mnt_idmap *idmap, const struct path *link) 1184800179c9SKees Cook { 11854609e1f1SChristian Brauner struct user_namespace *mnt_userns = mnt_idmap_owner(idmap); 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. */ 1189a2bd096fSChristian Brauner if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) || 1190a2bd096fSChristian Brauner !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, 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) || 1200*01beba79SChristian 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. 1211ba73d987SChristian Brauner * @mnt_userns: user namespace 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 * 1226ba73d987SChristian Brauner * If the inode has been found through an idmapped mount the user namespace of 1227ba73d987SChristian Brauner * the vfsmount must be passed through @mnt_userns. This function will then take 1228ba73d987SChristian Brauner * care to map the inode according to @mnt_userns before checking permissions. 1229ba73d987SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 1230ba73d987SChristian Brauner * raw inode simply passs init_user_ns. 1231ba73d987SChristian Brauner * 123230aba665SSalvatore Mesoraca * Returns 0 if the open is allowed, -ve on error. 123330aba665SSalvatore Mesoraca */ 1234ba73d987SChristian Brauner static int may_create_in_sticky(struct user_namespace *mnt_userns, 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)) || 1243a2bd096fSChristian Brauner vfsuid_eq(i_uid_into_vfsuid(mnt_userns, inode), dir_vfsuid) || 1244a2bd096fSChristian Brauner vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, 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 */ 14649deed3ebSAl Viro int follow_down(struct path *path) 14659deed3ebSAl Viro { 14669deed3ebSAl Viro struct vfsmount *mnt = path->mnt; 14679deed3ebSAl Viro bool jumped; 14689deed3ebSAl Viro int ret = traverse_mounts(path, &jumped, NULL, 0); 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 */ 1586a03ece5fSAl Viro static struct dentry *__lookup_hash(const struct qstr *name, 1587a03ece5fSAl Viro struct dentry *base, unsigned int flags) 158844396f4bSJosef Bacik { 1589a03ece5fSAl Viro struct dentry *dentry = lookup_dcache(name, base, flags); 159044396f4bSJosef Bacik struct dentry *old; 1591a03ece5fSAl Viro struct inode *dir = base->d_inode; 1592a03ece5fSAl Viro 1593a03ece5fSAl Viro if (dentry) 1594a03ece5fSAl Viro return dentry; 159544396f4bSJosef Bacik 159644396f4bSJosef Bacik /* Don't create child dentry for a dead directory. */ 1597a03ece5fSAl Viro if (unlikely(IS_DEADDIR(dir))) 159844396f4bSJosef Bacik return ERR_PTR(-ENOENT); 1599a03ece5fSAl Viro 1600a03ece5fSAl Viro dentry = d_alloc(base, name); 1601a03ece5fSAl Viro if (unlikely(!dentry)) 1602a03ece5fSAl Viro return ERR_PTR(-ENOMEM); 160344396f4bSJosef Bacik 160472bd866aSAl Viro old = dir->i_op->lookup(dir, dentry, flags); 160544396f4bSJosef Bacik if (unlikely(old)) { 160644396f4bSJosef Bacik dput(dentry); 160744396f4bSJosef Bacik dentry = old; 160844396f4bSJosef Bacik } 160944396f4bSJosef Bacik return dentry; 161044396f4bSJosef Bacik } 161144396f4bSJosef Bacik 16124cb64024SAl Viro static struct dentry *lookup_fast(struct nameidata *nd) 16131da177e4SLinus Torvalds { 161431e6b01fSNick Piggin struct dentry *dentry, *parent = nd->path.dentry; 16155a18fff2SAl Viro int status = 1; 16169875cf80SDavid Howells 16173cac260aSAl Viro /* 1618b04f784eSNick Piggin * Rename seqlock is not required here because in the off chance 16195d0f49c1SAl Viro * of a false negative due to a concurrent rename, the caller is 16205d0f49c1SAl Viro * going to fall back to non-racy lookup. 1621b04f784eSNick Piggin */ 162231e6b01fSNick Piggin if (nd->flags & LOOKUP_RCU) { 162303fa86e9SAl Viro dentry = __d_lookup_rcu(parent, &nd->last, &nd->next_seq); 16245d0f49c1SAl Viro if (unlikely(!dentry)) { 1625e36cffedSJens Axboe if (!try_to_unlazy(nd)) 162620e34357SAl Viro return ERR_PTR(-ECHILD); 162720e34357SAl Viro return NULL; 16285d0f49c1SAl Viro } 16295a18fff2SAl Viro 163012f8ad4bSLinus Torvalds /* 163112f8ad4bSLinus Torvalds * This sequence count validates that the parent had no 163212f8ad4bSLinus Torvalds * changes while we did the lookup of the dentry above. 163312f8ad4bSLinus Torvalds */ 16344cb64024SAl Viro if (read_seqcount_retry(&parent->d_seq, nd->seq)) 163520e34357SAl Viro return ERR_PTR(-ECHILD); 16365a18fff2SAl Viro 16374ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 1638c153007bSAl Viro if (likely(status > 0)) 163920e34357SAl Viro return dentry; 164003fa86e9SAl Viro if (!try_to_unlazy_next(nd, dentry)) 164120e34357SAl Viro return ERR_PTR(-ECHILD); 164226ddb45eSSteven Rostedt (VMware) if (status == -ECHILD) 1643209a7fb2SAl Viro /* we'd been told to redo it in non-rcu mode */ 1644209a7fb2SAl Viro status = d_revalidate(dentry, nd->flags); 16455a18fff2SAl Viro } else { 1646e97cdc87SAl Viro dentry = __d_lookup(parent, &nd->last); 164781e6f520SAl Viro if (unlikely(!dentry)) 164820e34357SAl Viro return NULL; 16494ce16ef3SAl Viro status = d_revalidate(dentry, nd->flags); 16505d0f49c1SAl Viro } 16515a18fff2SAl Viro if (unlikely(status <= 0)) { 1652e9742b53SAl Viro if (!status) 16535d0f49c1SAl Viro d_invalidate(dentry); 16545a18fff2SAl Viro dput(dentry); 165520e34357SAl Viro return ERR_PTR(status); 16565a18fff2SAl Viro } 165720e34357SAl Viro return dentry; 1658697f514dSMiklos Szeredi } 1659697f514dSMiklos Szeredi 1660697f514dSMiklos Szeredi /* Fast lookup failed, do it the slow way */ 166188d8331aSAl Viro static struct dentry *__lookup_slow(const struct qstr *name, 1662e3c13928SAl Viro struct dentry *dir, 1663e3c13928SAl Viro unsigned int flags) 1664697f514dSMiklos Szeredi { 166588d8331aSAl Viro struct dentry *dentry, *old; 16661936386eSAl Viro struct inode *inode = dir->d_inode; 1667d9171b93SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 16681936386eSAl Viro 16691936386eSAl Viro /* Don't go there if it's already dead */ 167094bdd655SAl Viro if (unlikely(IS_DEADDIR(inode))) 167188d8331aSAl Viro return ERR_PTR(-ENOENT); 167294bdd655SAl Viro again: 1673d9171b93SAl Viro dentry = d_alloc_parallel(dir, name, &wq); 167494bdd655SAl Viro if (IS_ERR(dentry)) 167588d8331aSAl Viro return dentry; 167694bdd655SAl Viro if (unlikely(!d_in_lookup(dentry))) { 1677949a852eSAl Viro int error = d_revalidate(dentry, flags); 1678949a852eSAl Viro if (unlikely(error <= 0)) { 167994bdd655SAl Viro if (!error) { 1680949a852eSAl Viro d_invalidate(dentry); 1681949a852eSAl Viro dput(dentry); 168294bdd655SAl Viro goto again; 168394bdd655SAl Viro } 168494bdd655SAl Viro dput(dentry); 1685949a852eSAl Viro dentry = ERR_PTR(error); 1686949a852eSAl Viro } 168794bdd655SAl Viro } else { 16881936386eSAl Viro old = inode->i_op->lookup(inode, dentry, flags); 168985c7f810SAl Viro d_lookup_done(dentry); 16901936386eSAl Viro if (unlikely(old)) { 16911936386eSAl Viro dput(dentry); 16921936386eSAl Viro dentry = old; 1693949a852eSAl Viro } 1694949a852eSAl Viro } 1695e3c13928SAl Viro return dentry; 16961da177e4SLinus Torvalds } 16971da177e4SLinus Torvalds 169888d8331aSAl Viro static struct dentry *lookup_slow(const struct qstr *name, 169988d8331aSAl Viro struct dentry *dir, 170088d8331aSAl Viro unsigned int flags) 170188d8331aSAl Viro { 170288d8331aSAl Viro struct inode *inode = dir->d_inode; 170388d8331aSAl Viro struct dentry *res; 170488d8331aSAl Viro inode_lock_shared(inode); 170588d8331aSAl Viro res = __lookup_slow(name, dir, flags); 170688d8331aSAl Viro inode_unlock_shared(inode); 170788d8331aSAl Viro return res; 170888d8331aSAl Viro } 170988d8331aSAl Viro 17104609e1f1SChristian Brauner static inline int may_lookup(struct mnt_idmap *idmap, 1711ba73d987SChristian Brauner struct nameidata *nd) 171252094c8aSAl Viro { 171352094c8aSAl Viro if (nd->flags & LOOKUP_RCU) { 17144609e1f1SChristian Brauner int err = inode_permission(idmap, nd->inode, MAY_EXEC|MAY_NOT_BLOCK); 1715e36cffedSJens Axboe if (err != -ECHILD || !try_to_unlazy(nd)) 171652094c8aSAl Viro return err; 171752094c8aSAl Viro } 17184609e1f1SChristian Brauner return inode_permission(idmap, nd->inode, MAY_EXEC); 171952094c8aSAl Viro } 172052094c8aSAl Viro 172103fa86e9SAl Viro static int reserve_stack(struct nameidata *nd, struct path *link) 1722d63ff28fSAl Viro { 172349055906SAl Viro if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) 172449055906SAl Viro return -ELOOP; 17254542576bSAl Viro 17264542576bSAl Viro if (likely(nd->depth != EMBEDDED_LEVELS)) 17274542576bSAl Viro return 0; 17284542576bSAl Viro if (likely(nd->stack != nd->internal)) 17294542576bSAl Viro return 0; 173060ef60c7SAl Viro if (likely(nd_alloc_stack(nd))) 173149055906SAl Viro return 0; 173260ef60c7SAl Viro 173360ef60c7SAl Viro if (nd->flags & LOOKUP_RCU) { 173460ef60c7SAl Viro // we need to grab link before we do unlazy. And we can't skip 173560ef60c7SAl Viro // unlazy even if we fail to grab the link - cleanup needs it 173603fa86e9SAl Viro bool grabbed_link = legitimize_path(nd, link, nd->next_seq); 173760ef60c7SAl Viro 1738e5ca024eSAl Viro if (!try_to_unlazy(nd) || !grabbed_link) 173960ef60c7SAl Viro return -ECHILD; 174060ef60c7SAl Viro 174160ef60c7SAl Viro if (nd_alloc_stack(nd)) 174260ef60c7SAl Viro return 0; 1743bc40aee0SAl Viro } 174460ef60c7SAl Viro return -ENOMEM; 174549055906SAl Viro } 174649055906SAl Viro 174749055906SAl Viro enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4}; 174849055906SAl Viro 174949055906SAl Viro static const char *pick_link(struct nameidata *nd, struct path *link, 175003fa86e9SAl Viro struct inode *inode, int flags) 175149055906SAl Viro { 175249055906SAl Viro struct saved *last; 175349055906SAl Viro const char *res; 175403fa86e9SAl Viro int error = reserve_stack(nd, link); 175549055906SAl Viro 175649055906SAl Viro if (unlikely(error)) { 175749055906SAl Viro if (!(nd->flags & LOOKUP_RCU)) 1758cd179f44SAl Viro path_put(link); 175906708adbSAl Viro return ERR_PTR(error); 1760626de996SAl Viro } 1761ab104923SAl Viro last = nd->stack + nd->depth++; 17621cf2665bSAl Viro last->link = *link; 1763fceef393SAl Viro clear_delayed_call(&last->done); 176403fa86e9SAl Viro last->seq = nd->next_seq; 1765ad6cc4c3SAl Viro 1766b1a81972SAl Viro if (flags & WALK_TRAILING) { 1767ad6cc4c3SAl Viro error = may_follow_link(nd, inode); 1768ad6cc4c3SAl Viro if (unlikely(error)) 1769ad6cc4c3SAl Viro return ERR_PTR(error); 1770ad6cc4c3SAl Viro } 1771ad6cc4c3SAl Viro 1772dab741e0SMattias Nissler if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) || 1773dab741e0SMattias Nissler unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW)) 1774ad6cc4c3SAl Viro return ERR_PTR(-ELOOP); 1775ad6cc4c3SAl Viro 1776ad6cc4c3SAl Viro if (!(nd->flags & LOOKUP_RCU)) { 1777ad6cc4c3SAl Viro touch_atime(&last->link); 1778ad6cc4c3SAl Viro cond_resched(); 1779ad6cc4c3SAl Viro } else if (atime_needs_update(&last->link, inode)) { 1780e36cffedSJens Axboe if (!try_to_unlazy(nd)) 1781ad6cc4c3SAl Viro return ERR_PTR(-ECHILD); 1782ad6cc4c3SAl Viro touch_atime(&last->link); 1783ad6cc4c3SAl Viro } 1784ad6cc4c3SAl Viro 1785ad6cc4c3SAl Viro error = security_inode_follow_link(link->dentry, inode, 1786ad6cc4c3SAl Viro nd->flags & LOOKUP_RCU); 1787ad6cc4c3SAl Viro if (unlikely(error)) 1788ad6cc4c3SAl Viro return ERR_PTR(error); 1789ad6cc4c3SAl Viro 1790ad6cc4c3SAl Viro res = READ_ONCE(inode->i_link); 1791ad6cc4c3SAl Viro if (!res) { 1792ad6cc4c3SAl Viro const char * (*get)(struct dentry *, struct inode *, 1793ad6cc4c3SAl Viro struct delayed_call *); 1794ad6cc4c3SAl Viro get = inode->i_op->get_link; 1795ad6cc4c3SAl Viro if (nd->flags & LOOKUP_RCU) { 1796ad6cc4c3SAl Viro res = get(NULL, inode, &last->done); 1797e36cffedSJens Axboe if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd)) 1798ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1799ad6cc4c3SAl Viro } else { 1800ad6cc4c3SAl Viro res = get(link->dentry, inode, &last->done); 1801ad6cc4c3SAl Viro } 1802ad6cc4c3SAl Viro if (!res) 1803ad6cc4c3SAl Viro goto all_done; 1804ad6cc4c3SAl Viro if (IS_ERR(res)) 1805ad6cc4c3SAl Viro return res; 1806ad6cc4c3SAl Viro } 1807ad6cc4c3SAl Viro if (*res == '/') { 1808ad6cc4c3SAl Viro error = nd_jump_root(nd); 1809ad6cc4c3SAl Viro if (unlikely(error)) 1810ad6cc4c3SAl Viro return ERR_PTR(error); 1811ad6cc4c3SAl Viro while (unlikely(*++res == '/')) 1812ad6cc4c3SAl Viro ; 1813ad6cc4c3SAl Viro } 1814ad6cc4c3SAl Viro if (*res) 1815ad6cc4c3SAl Viro return res; 1816ad6cc4c3SAl Viro all_done: // pure jump 1817ad6cc4c3SAl Viro put_link(nd); 1818ad6cc4c3SAl Viro return NULL; 1819d63ff28fSAl Viro } 1820d63ff28fSAl Viro 18213ddcd056SLinus Torvalds /* 18223ddcd056SLinus Torvalds * Do we need to follow links? We _really_ want to be able 18233ddcd056SLinus Torvalds * to do this check without having to look at inode->i_op, 18243ddcd056SLinus Torvalds * so we keep a cache of "no, this doesn't need follow_link" 18253ddcd056SLinus Torvalds * for the common case. 182603fa86e9SAl Viro * 182703fa86e9SAl Viro * NOTE: dentry must be what nd->next_seq had been sampled from. 18283ddcd056SLinus Torvalds */ 1829b0417d2cSAl Viro static const char *step_into(struct nameidata *nd, int flags, 1830a4f5b521SAl Viro struct dentry *dentry) 18313ddcd056SLinus Torvalds { 1832cbae4d12SAl Viro struct path path; 1833a4f5b521SAl Viro struct inode *inode; 18343bd8bc89SAl Viro int err = handle_mounts(nd, dentry, &path); 1835cbae4d12SAl Viro 1836cbae4d12SAl Viro if (err < 0) 1837b0417d2cSAl Viro return ERR_PTR(err); 18383bd8bc89SAl Viro inode = path.dentry->d_inode; 1839cbae4d12SAl Viro if (likely(!d_is_symlink(path.dentry)) || 18408c4efe22SAl Viro ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) || 1841aca2903eSAl Viro (flags & WALK_NOFOLLOW)) { 18428f64fb1cSAl Viro /* not a symlink or should not follow */ 18433bd8bc89SAl Viro if (nd->flags & LOOKUP_RCU) { 18443bd8bc89SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) 18453bd8bc89SAl Viro return ERR_PTR(-ECHILD); 18463bd8bc89SAl Viro if (unlikely(!inode)) 18473bd8bc89SAl Viro return ERR_PTR(-ENOENT); 18483bd8bc89SAl Viro } else { 1849c99687a0SAl Viro dput(nd->path.dentry); 1850c99687a0SAl Viro if (nd->path.mnt != path.mnt) 1851c99687a0SAl Viro mntput(nd->path.mnt); 1852c99687a0SAl Viro } 1853c99687a0SAl Viro nd->path = path; 18548f64fb1cSAl Viro nd->inode = inode; 185503fa86e9SAl Viro nd->seq = nd->next_seq; 1856b0417d2cSAl Viro return NULL; 18578f64fb1cSAl Viro } 1858a7f77542SAl Viro if (nd->flags & LOOKUP_RCU) { 185984f0cd9eSAl Viro /* make sure that d_is_symlink above matches inode */ 186003fa86e9SAl Viro if (read_seqcount_retry(&path.dentry->d_seq, nd->next_seq)) 1861b0417d2cSAl Viro return ERR_PTR(-ECHILD); 186284f0cd9eSAl Viro } else { 186384f0cd9eSAl Viro if (path.mnt == nd->path.mnt) 186484f0cd9eSAl Viro mntget(path.mnt); 1865a7f77542SAl Viro } 186603fa86e9SAl Viro return pick_link(nd, &path, inode, flags); 18673ddcd056SLinus Torvalds } 18683ddcd056SLinus Torvalds 1869b16c001dSAl Viro static struct dentry *follow_dotdot_rcu(struct nameidata *nd) 1870957dd41dSAl Viro { 187112487f30SAl Viro struct dentry *parent, *old; 1872957dd41dSAl Viro 187312487f30SAl Viro if (path_equal(&nd->path, &nd->root)) 187412487f30SAl Viro goto in_root; 187512487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 18767ef482faSAl Viro struct path path; 1877efe772d6SAl Viro unsigned seq; 18787ef482faSAl Viro if (!choose_mountpoint_rcu(real_mount(nd->path.mnt), 18797ef482faSAl Viro &nd->root, &path, &seq)) 188012487f30SAl Viro goto in_root; 1881efe772d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1882efe772d6SAl Viro return ERR_PTR(-ECHILD); 1883efe772d6SAl Viro nd->path = path; 1884efe772d6SAl Viro nd->inode = path.dentry->d_inode; 1885efe772d6SAl Viro nd->seq = seq; 188603fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach this state 188782ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1888efe772d6SAl Viro return ERR_PTR(-ECHILD); 1889efe772d6SAl Viro /* we know that mountpoint was pinned */ 1890957dd41dSAl Viro } 189112487f30SAl Viro old = nd->path.dentry; 189212487f30SAl Viro parent = old->d_parent; 189303fa86e9SAl Viro nd->next_seq = read_seqcount_begin(&parent->d_seq); 189403fa86e9SAl Viro // makes sure that non-RCU pathwalk could reach this state 189582ef0698SAl Viro if (read_seqcount_retry(&old->d_seq, nd->seq)) 189612487f30SAl Viro return ERR_PTR(-ECHILD); 189712487f30SAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) 189812487f30SAl Viro return ERR_PTR(-ECHILD); 189912487f30SAl Viro return parent; 190012487f30SAl Viro in_root: 190182ef0698SAl Viro if (read_seqretry(&mount_lock, nd->m_seq)) 1902efe772d6SAl Viro return ERR_PTR(-ECHILD); 1903957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19047521f22bSAl Viro return ERR_PTR(-ECHILD); 190503fa86e9SAl Viro nd->next_seq = nd->seq; 190651c6546cSAl Viro return nd->path.dentry; 1907957dd41dSAl Viro } 1908957dd41dSAl Viro 1909b16c001dSAl Viro static struct dentry *follow_dotdot(struct nameidata *nd) 1910957dd41dSAl Viro { 191112487f30SAl Viro struct dentry *parent; 191212487f30SAl Viro 1913957dd41dSAl Viro if (path_equal(&nd->path, &nd->root)) 191412487f30SAl Viro goto in_root; 191512487f30SAl Viro if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) { 19162aa38470SAl Viro struct path path; 19172aa38470SAl Viro 19182aa38470SAl Viro if (!choose_mountpoint(real_mount(nd->path.mnt), 19192aa38470SAl Viro &nd->root, &path)) 192012487f30SAl Viro goto in_root; 1921165200d6SAl Viro path_put(&nd->path); 1922165200d6SAl Viro nd->path = path; 19232aa38470SAl Viro nd->inode = path.dentry->d_inode; 1924165200d6SAl Viro if (unlikely(nd->flags & LOOKUP_NO_XDEV)) 1925165200d6SAl Viro return ERR_PTR(-EXDEV); 192612487f30SAl Viro } 1927957dd41dSAl Viro /* rare case of legitimate dget_parent()... */ 192812487f30SAl Viro parent = dget_parent(nd->path.dentry); 1929957dd41dSAl Viro if (unlikely(!path_connected(nd->path.mnt, parent))) { 1930957dd41dSAl Viro dput(parent); 19317521f22bSAl Viro return ERR_PTR(-ENOENT); 1932957dd41dSAl Viro } 1933c2df1968SAl Viro return parent; 193412487f30SAl Viro 193512487f30SAl Viro in_root: 1936957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_BENEATH)) 19377521f22bSAl Viro return ERR_PTR(-EXDEV); 193851c6546cSAl Viro return dget(nd->path.dentry); 1939957dd41dSAl Viro } 1940957dd41dSAl Viro 19417521f22bSAl Viro static const char *handle_dots(struct nameidata *nd, int type) 1942957dd41dSAl Viro { 1943957dd41dSAl Viro if (type == LAST_DOTDOT) { 19447521f22bSAl Viro const char *error = NULL; 1945c2df1968SAl Viro struct dentry *parent; 1946957dd41dSAl Viro 1947957dd41dSAl Viro if (!nd->root.mnt) { 19487521f22bSAl Viro error = ERR_PTR(set_root(nd)); 1949957dd41dSAl Viro if (error) 1950957dd41dSAl Viro return error; 1951957dd41dSAl Viro } 1952957dd41dSAl Viro if (nd->flags & LOOKUP_RCU) 1953b16c001dSAl Viro parent = follow_dotdot_rcu(nd); 1954957dd41dSAl Viro else 1955b16c001dSAl Viro parent = follow_dotdot(nd); 1956c2df1968SAl Viro if (IS_ERR(parent)) 1957c2df1968SAl Viro return ERR_CAST(parent); 1958a4f5b521SAl Viro error = step_into(nd, WALK_NOFOLLOW, parent); 1959c2df1968SAl Viro if (unlikely(error)) 1960957dd41dSAl Viro return error; 1961957dd41dSAl Viro 1962957dd41dSAl Viro if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) { 1963957dd41dSAl Viro /* 1964957dd41dSAl Viro * If there was a racing rename or mount along our 1965957dd41dSAl Viro * path, then we can't be sure that ".." hasn't jumped 1966957dd41dSAl Viro * above nd->root (and so userspace should retry or use 1967957dd41dSAl Viro * some fallback). 1968957dd41dSAl Viro */ 1969957dd41dSAl Viro smp_rmb(); 197082ef0698SAl Viro if (__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)) 19717521f22bSAl Viro return ERR_PTR(-EAGAIN); 197282ef0698SAl Viro if (__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)) 19737521f22bSAl Viro return ERR_PTR(-EAGAIN); 1974957dd41dSAl Viro } 1975957dd41dSAl Viro } 19767521f22bSAl Viro return NULL; 1977957dd41dSAl Viro } 1978957dd41dSAl Viro 197992d27016SAl Viro static const char *walk_component(struct nameidata *nd, int flags) 1980ce57dfc1SAl Viro { 1981db3c9adeSAl Viro struct dentry *dentry; 1982ce57dfc1SAl Viro /* 1983ce57dfc1SAl Viro * "." and ".." are special - ".." especially so because it has 1984ce57dfc1SAl Viro * to be able to know about the current root directory and 1985ce57dfc1SAl Viro * parent relationships. 1986ce57dfc1SAl Viro */ 19874693a547SAl Viro if (unlikely(nd->last_type != LAST_NORM)) { 19881c4ff1a8SAl Viro if (!(flags & WALK_MORE) && nd->depth) 19894693a547SAl Viro put_link(nd); 19907521f22bSAl Viro return handle_dots(nd, nd->last_type); 19914693a547SAl Viro } 19924cb64024SAl Viro dentry = lookup_fast(nd); 199320e34357SAl Viro if (IS_ERR(dentry)) 199492d27016SAl Viro return ERR_CAST(dentry); 199520e34357SAl Viro if (unlikely(!dentry)) { 1996db3c9adeSAl Viro dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags); 1997db3c9adeSAl Viro if (IS_ERR(dentry)) 199892d27016SAl Viro return ERR_CAST(dentry); 199920e34357SAl Viro } 200056676ec3SAl Viro if (!(flags & WALK_MORE) && nd->depth) 200156676ec3SAl Viro put_link(nd); 2002a4f5b521SAl Viro return step_into(nd, flags, dentry); 2003ce57dfc1SAl Viro } 2004ce57dfc1SAl Viro 20051da177e4SLinus Torvalds /* 2006bfcfaa77SLinus Torvalds * We can do the critical dentry name comparison and hashing 2007bfcfaa77SLinus Torvalds * operations one word at a time, but we are limited to: 2008bfcfaa77SLinus Torvalds * 2009bfcfaa77SLinus Torvalds * - Architectures with fast unaligned word accesses. We could 2010bfcfaa77SLinus Torvalds * do a "get_unaligned()" if this helps and is sufficiently 2011bfcfaa77SLinus Torvalds * fast. 2012bfcfaa77SLinus Torvalds * 2013bfcfaa77SLinus Torvalds * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we 2014bfcfaa77SLinus Torvalds * do not trap on the (extremely unlikely) case of a page 2015bfcfaa77SLinus Torvalds * crossing operation. 2016bfcfaa77SLinus Torvalds * 2017bfcfaa77SLinus Torvalds * - Furthermore, we need an efficient 64-bit compile for the 2018bfcfaa77SLinus Torvalds * 64-bit case in order to generate the "number of bytes in 2019bfcfaa77SLinus Torvalds * the final mask". Again, that could be replaced with a 2020bfcfaa77SLinus Torvalds * efficient population count instruction or similar. 2021bfcfaa77SLinus Torvalds */ 2022bfcfaa77SLinus Torvalds #ifdef CONFIG_DCACHE_WORD_ACCESS 2023bfcfaa77SLinus Torvalds 2024f68e556eSLinus Torvalds #include <asm/word-at-a-time.h> 2025bfcfaa77SLinus Torvalds 2026468a9428SGeorge Spelvin #ifdef HASH_MIX 2027bfcfaa77SLinus Torvalds 2028468a9428SGeorge Spelvin /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */ 2029468a9428SGeorge Spelvin 2030468a9428SGeorge Spelvin #elif defined(CONFIG_64BIT) 20312a18da7aSGeorge Spelvin /* 20322a18da7aSGeorge Spelvin * Register pressure in the mixing function is an issue, particularly 20332a18da7aSGeorge Spelvin * on 32-bit x86, but almost any function requires one state value and 20342a18da7aSGeorge Spelvin * one temporary. Instead, use a function designed for two state values 20352a18da7aSGeorge Spelvin * and no temporaries. 20362a18da7aSGeorge Spelvin * 20372a18da7aSGeorge Spelvin * This function cannot create a collision in only two iterations, so 20382a18da7aSGeorge Spelvin * we have two iterations to achieve avalanche. In those two iterations, 20392a18da7aSGeorge Spelvin * we have six layers of mixing, which is enough to spread one bit's 20402a18da7aSGeorge Spelvin * influence out to 2^6 = 64 state bits. 20412a18da7aSGeorge Spelvin * 20422a18da7aSGeorge Spelvin * Rotate constants are scored by considering either 64 one-bit input 20432a18da7aSGeorge Spelvin * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the 20442a18da7aSGeorge Spelvin * probability of that delta causing a change to each of the 128 output 20452a18da7aSGeorge Spelvin * bits, using a sample of random initial states. 20462a18da7aSGeorge Spelvin * 20472a18da7aSGeorge Spelvin * The Shannon entropy of the computed probabilities is then summed 20482a18da7aSGeorge Spelvin * to produce a score. Ideally, any input change has a 50% chance of 20492a18da7aSGeorge Spelvin * toggling any given output bit. 20502a18da7aSGeorge Spelvin * 20512a18da7aSGeorge Spelvin * Mixing scores (in bits) for (12,45): 20522a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20532a18da7aSGeorge Spelvin * 1 round: 713.3 42542.6 20542a18da7aSGeorge Spelvin * 2 rounds: 2753.7 140389.8 20552a18da7aSGeorge Spelvin * 3 rounds: 5954.1 233458.2 20562a18da7aSGeorge Spelvin * 4 rounds: 7862.6 256672.2 20572a18da7aSGeorge Spelvin * Perfect: 8192 258048 20582a18da7aSGeorge Spelvin * (64*128) (64*63/2 * 128) 20592a18da7aSGeorge Spelvin */ 20602a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20612a18da7aSGeorge Spelvin ( x ^= (a), \ 20622a18da7aSGeorge Spelvin y ^= x, x = rol64(x,12),\ 20632a18da7aSGeorge Spelvin x += y, y = rol64(y,45),\ 20642a18da7aSGeorge Spelvin y *= 9 ) 2065bfcfaa77SLinus Torvalds 20660fed3ac8SGeorge Spelvin /* 20672a18da7aSGeorge Spelvin * Fold two longs into one 32-bit hash value. This must be fast, but 20682a18da7aSGeorge Spelvin * latency isn't quite as critical, as there is a fair bit of additional 20692a18da7aSGeorge Spelvin * work done before the hash value is used. 20700fed3ac8SGeorge Spelvin */ 20712a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20720fed3ac8SGeorge Spelvin { 20732a18da7aSGeorge Spelvin y ^= x * GOLDEN_RATIO_64; 20742a18da7aSGeorge Spelvin y *= GOLDEN_RATIO_64; 20752a18da7aSGeorge Spelvin return y >> 32; 20760fed3ac8SGeorge Spelvin } 20770fed3ac8SGeorge Spelvin 2078bfcfaa77SLinus Torvalds #else /* 32-bit case */ 2079bfcfaa77SLinus Torvalds 20802a18da7aSGeorge Spelvin /* 20812a18da7aSGeorge Spelvin * Mixing scores (in bits) for (7,20): 20822a18da7aSGeorge Spelvin * Input delta: 1-bit 2-bit 20832a18da7aSGeorge Spelvin * 1 round: 330.3 9201.6 20842a18da7aSGeorge Spelvin * 2 rounds: 1246.4 25475.4 20852a18da7aSGeorge Spelvin * 3 rounds: 1907.1 31295.1 20862a18da7aSGeorge Spelvin * 4 rounds: 2042.3 31718.6 20872a18da7aSGeorge Spelvin * Perfect: 2048 31744 20882a18da7aSGeorge Spelvin * (32*64) (32*31/2 * 64) 20892a18da7aSGeorge Spelvin */ 20902a18da7aSGeorge Spelvin #define HASH_MIX(x, y, a) \ 20912a18da7aSGeorge Spelvin ( x ^= (a), \ 20922a18da7aSGeorge Spelvin y ^= x, x = rol32(x, 7),\ 20932a18da7aSGeorge Spelvin x += y, y = rol32(y,20),\ 20942a18da7aSGeorge Spelvin y *= 9 ) 2095bfcfaa77SLinus Torvalds 20962a18da7aSGeorge Spelvin static inline unsigned int fold_hash(unsigned long x, unsigned long y) 20970fed3ac8SGeorge Spelvin { 20982a18da7aSGeorge Spelvin /* Use arch-optimized multiply if one exists */ 20992a18da7aSGeorge Spelvin return __hash_32(y ^ __hash_32(x)); 21000fed3ac8SGeorge Spelvin } 21010fed3ac8SGeorge Spelvin 2102bfcfaa77SLinus Torvalds #endif 2103bfcfaa77SLinus Torvalds 21042a18da7aSGeorge Spelvin /* 21052a18da7aSGeorge Spelvin * Return the hash of a string of known length. This is carfully 21062a18da7aSGeorge Spelvin * designed to match hash_name(), which is the more critical function. 21072a18da7aSGeorge Spelvin * In particular, we must end by hashing a final word containing 0..7 21082a18da7aSGeorge Spelvin * payload bytes, to match the way that hash_name() iterates until it 21092a18da7aSGeorge Spelvin * finds the delimiter after the name. 21102a18da7aSGeorge Spelvin */ 21118387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 2112bfcfaa77SLinus Torvalds { 21138387ff25SLinus Torvalds unsigned long a, x = 0, y = (unsigned long)salt; 2114bfcfaa77SLinus Torvalds 2115bfcfaa77SLinus Torvalds for (;;) { 2116fcfd2fbfSGeorge Spelvin if (!len) 2117fcfd2fbfSGeorge Spelvin goto done; 2118e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name); 2119bfcfaa77SLinus Torvalds if (len < sizeof(unsigned long)) 2120bfcfaa77SLinus Torvalds break; 21212a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2122bfcfaa77SLinus Torvalds name += sizeof(unsigned long); 2123bfcfaa77SLinus Torvalds len -= sizeof(unsigned long); 2124bfcfaa77SLinus Torvalds } 21252a18da7aSGeorge Spelvin x ^= a & bytemask_from_count(len); 2126bfcfaa77SLinus Torvalds done: 21272a18da7aSGeorge Spelvin return fold_hash(x, y); 2128bfcfaa77SLinus Torvalds } 2129bfcfaa77SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 2130bfcfaa77SLinus Torvalds 2131fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21328387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2133fcfd2fbfSGeorge Spelvin { 21348387ff25SLinus Torvalds unsigned long a = 0, x = 0, y = (unsigned long)salt; 21358387ff25SLinus Torvalds unsigned long adata, mask, len; 2136fcfd2fbfSGeorge Spelvin const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2137fcfd2fbfSGeorge Spelvin 21388387ff25SLinus Torvalds len = 0; 21398387ff25SLinus Torvalds goto inside; 21408387ff25SLinus Torvalds 2141fcfd2fbfSGeorge Spelvin do { 21422a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2143fcfd2fbfSGeorge Spelvin len += sizeof(unsigned long); 21448387ff25SLinus Torvalds inside: 2145fcfd2fbfSGeorge Spelvin a = load_unaligned_zeropad(name+len); 2146fcfd2fbfSGeorge Spelvin } while (!has_zero(a, &adata, &constants)); 2147fcfd2fbfSGeorge Spelvin 2148fcfd2fbfSGeorge Spelvin adata = prep_zero_mask(a, adata, &constants); 2149fcfd2fbfSGeorge Spelvin mask = create_zero_mask(adata); 21502a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 2151fcfd2fbfSGeorge Spelvin 21522a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2153fcfd2fbfSGeorge Spelvin } 2154fcfd2fbfSGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2155fcfd2fbfSGeorge Spelvin 2156bfcfaa77SLinus Torvalds /* 2157bfcfaa77SLinus Torvalds * Calculate the length and hash of the path component, and 2158d6bb3e90SLinus Torvalds * return the "hash_len" as the result. 2159bfcfaa77SLinus Torvalds */ 21608387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2161bfcfaa77SLinus Torvalds { 21628387ff25SLinus Torvalds unsigned long a = 0, b, x = 0, y = (unsigned long)salt; 21638387ff25SLinus Torvalds unsigned long adata, bdata, mask, len; 216436126f8fSLinus Torvalds const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS; 2165bfcfaa77SLinus Torvalds 21668387ff25SLinus Torvalds len = 0; 21678387ff25SLinus Torvalds goto inside; 21688387ff25SLinus Torvalds 2169bfcfaa77SLinus Torvalds do { 21702a18da7aSGeorge Spelvin HASH_MIX(x, y, a); 2171bfcfaa77SLinus Torvalds len += sizeof(unsigned long); 21728387ff25SLinus Torvalds inside: 2173e419b4ccSLinus Torvalds a = load_unaligned_zeropad(name+len); 217436126f8fSLinus Torvalds b = a ^ REPEAT_BYTE('/'); 217536126f8fSLinus Torvalds } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants))); 2176bfcfaa77SLinus Torvalds 217736126f8fSLinus Torvalds adata = prep_zero_mask(a, adata, &constants); 217836126f8fSLinus Torvalds bdata = prep_zero_mask(b, bdata, &constants); 217936126f8fSLinus Torvalds mask = create_zero_mask(adata | bdata); 21802a18da7aSGeorge Spelvin x ^= a & zero_bytemask(mask); 218136126f8fSLinus Torvalds 21822a18da7aSGeorge Spelvin return hashlen_create(fold_hash(x, y), len + find_zero(mask)); 2183bfcfaa77SLinus Torvalds } 2184bfcfaa77SLinus Torvalds 21852a18da7aSGeorge Spelvin #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */ 2186bfcfaa77SLinus Torvalds 2187fcfd2fbfSGeorge Spelvin /* Return the hash of a string of known length */ 21888387ff25SLinus Torvalds unsigned int full_name_hash(const void *salt, const char *name, unsigned int len) 21890145acc2SLinus Torvalds { 21908387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 21910145acc2SLinus Torvalds while (len--) 2192fcfd2fbfSGeorge Spelvin hash = partial_name_hash((unsigned char)*name++, hash); 21930145acc2SLinus Torvalds return end_name_hash(hash); 21940145acc2SLinus Torvalds } 2195ae942ae7SLinus Torvalds EXPORT_SYMBOL(full_name_hash); 21960145acc2SLinus Torvalds 2197fcfd2fbfSGeorge Spelvin /* Return the "hash_len" (hash and length) of a null-terminated string */ 21988387ff25SLinus Torvalds u64 hashlen_string(const void *salt, const char *name) 2199fcfd2fbfSGeorge Spelvin { 22008387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2201fcfd2fbfSGeorge Spelvin unsigned long len = 0, c; 2202fcfd2fbfSGeorge Spelvin 2203fcfd2fbfSGeorge Spelvin c = (unsigned char)*name; 2204e0ab7af9SGeorge Spelvin while (c) { 2205fcfd2fbfSGeorge Spelvin len++; 2206fcfd2fbfSGeorge Spelvin hash = partial_name_hash(c, hash); 2207fcfd2fbfSGeorge Spelvin c = (unsigned char)name[len]; 2208e0ab7af9SGeorge Spelvin } 2209fcfd2fbfSGeorge Spelvin return hashlen_create(end_name_hash(hash), len); 2210fcfd2fbfSGeorge Spelvin } 2211f2a031b6SGeorge Spelvin EXPORT_SYMBOL(hashlen_string); 2212fcfd2fbfSGeorge Spelvin 22133ddcd056SLinus Torvalds /* 2214200e9ef7SLinus Torvalds * We know there's a real path component here of at least 2215200e9ef7SLinus Torvalds * one character. 2216200e9ef7SLinus Torvalds */ 22178387ff25SLinus Torvalds static inline u64 hash_name(const void *salt, const char *name) 2218200e9ef7SLinus Torvalds { 22198387ff25SLinus Torvalds unsigned long hash = init_name_hash(salt); 2220200e9ef7SLinus Torvalds unsigned long len = 0, c; 2221200e9ef7SLinus Torvalds 2222200e9ef7SLinus Torvalds c = (unsigned char)*name; 2223200e9ef7SLinus Torvalds do { 2224200e9ef7SLinus Torvalds len++; 2225200e9ef7SLinus Torvalds hash = partial_name_hash(c, hash); 2226200e9ef7SLinus Torvalds c = (unsigned char)name[len]; 2227200e9ef7SLinus Torvalds } while (c && c != '/'); 2228d6bb3e90SLinus Torvalds return hashlen_create(end_name_hash(hash), len); 2229200e9ef7SLinus Torvalds } 2230200e9ef7SLinus Torvalds 2231bfcfaa77SLinus Torvalds #endif 2232bfcfaa77SLinus Torvalds 2233200e9ef7SLinus Torvalds /* 22341da177e4SLinus Torvalds * Name resolution. 2235ea3834d9SPrasanna Meda * This is the basic name resolution function, turning a pathname into 2236ea3834d9SPrasanna Meda * the final dentry. We expect 'base' to be positive and a directory. 22371da177e4SLinus Torvalds * 2238ea3834d9SPrasanna Meda * Returns 0 and nd will have valid dentry and mnt on success. 2239ea3834d9SPrasanna Meda * Returns error and drops reference to input namei data on failure. 22401da177e4SLinus Torvalds */ 22416de88d72SAl Viro static int link_path_walk(const char *name, struct nameidata *nd) 22421da177e4SLinus Torvalds { 2243d8d4611aSAl Viro int depth = 0; // depth <= nd->depth 22441da177e4SLinus Torvalds int err; 22451da177e4SLinus Torvalds 2246b4c03536SAl Viro nd->last_type = LAST_ROOT; 2247c108837eSAl Viro nd->flags |= LOOKUP_PARENT; 22489b5858e9SAl Viro if (IS_ERR(name)) 22499b5858e9SAl Viro return PTR_ERR(name); 22501da177e4SLinus Torvalds while (*name=='/') 22511da177e4SLinus Torvalds name++; 22521a97d899SAl Viro if (!*name) { 22531a97d899SAl Viro nd->dir_mode = 0; // short-circuit the 'hardening' idiocy 22549e18f10aSAl Viro return 0; 22551a97d899SAl Viro } 22561da177e4SLinus Torvalds 22571da177e4SLinus Torvalds /* At this point we know we have a real path component. */ 22581da177e4SLinus Torvalds for(;;) { 22594609e1f1SChristian Brauner struct mnt_idmap *idmap; 2260549c7297SChristian Brauner struct user_namespace *mnt_userns; 226192d27016SAl Viro const char *link; 2262d6bb3e90SLinus Torvalds u64 hash_len; 2263fe479a58SAl Viro int type; 22641da177e4SLinus Torvalds 22654609e1f1SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 22664609e1f1SChristian Brauner mnt_userns = mnt_idmap_owner(idmap); 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) { 2315a2bd096fSChristian Brauner nd->dir_vfsuid = i_uid_into_vfsuid(mnt_userns, 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" */ 2539c5f563f9SAl Viro static int filename_parentat(int dfd, struct filename *name, 2540391172c4SAl Viro unsigned int flags, struct path *parent, 2541391172c4SAl Viro struct qstr *last, int *type) 25428bcb77faSAl Viro { 25438bcb77faSAl Viro int retval; 25449883d185SAl Viro struct nameidata nd; 25458bcb77faSAl Viro 25465c31b6ceSAl Viro if (IS_ERR(name)) 25470ee50b47SDmitry Kadashev return PTR_ERR(name); 254806422964SAl Viro set_nameidata(&nd, dfd, name, NULL); 2549c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_RCU, parent); 25508bcb77faSAl Viro if (unlikely(retval == -ECHILD)) 2551c8a53ee5SAl Viro retval = path_parentat(&nd, flags, parent); 25528bcb77faSAl Viro if (unlikely(retval == -ESTALE)) 2553c8a53ee5SAl Viro retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent); 2554391172c4SAl Viro if (likely(!retval)) { 2555391172c4SAl Viro *last = nd.last; 2556391172c4SAl Viro *type = nd.last_type; 2557c9b07eabSAl Viro audit_inode(name, parent->dentry, AUDIT_INODE_PARENT); 2558391172c4SAl Viro } 25599883d185SAl Viro restore_nameidata(); 25600ee50b47SDmitry Kadashev return retval; 25610ee50b47SDmitry Kadashev } 25620ee50b47SDmitry Kadashev 256379714f72SAl Viro /* does lookup, returns the object with parent locked */ 25640766ec82SStephen Brennan static struct dentry *__kern_path_locked(struct filename *name, struct path *path) 25655590ff0dSUlrich Drepper { 25665c31b6ceSAl Viro struct dentry *d; 2567391172c4SAl Viro struct qstr last; 25680ee50b47SDmitry Kadashev int type, error; 256951689104SPaul Moore 2570c5f563f9SAl Viro error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type); 25710ee50b47SDmitry Kadashev if (error) 25720ee50b47SDmitry Kadashev return ERR_PTR(error); 25735c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) { 2574391172c4SAl Viro path_put(path); 25755c31b6ceSAl Viro return ERR_PTR(-EINVAL); 257679714f72SAl Viro } 25775955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 2578391172c4SAl Viro d = __lookup_hash(&last, path->dentry, 0); 257979714f72SAl Viro if (IS_ERR(d)) { 25805955102cSAl Viro inode_unlock(path->dentry->d_inode); 2581391172c4SAl Viro path_put(path); 258279714f72SAl Viro } 258379714f72SAl Viro return d; 25845590ff0dSUlrich Drepper } 25855590ff0dSUlrich Drepper 25860766ec82SStephen Brennan struct dentry *kern_path_locked(const char *name, struct path *path) 25870766ec82SStephen Brennan { 25880766ec82SStephen Brennan struct filename *filename = getname_kernel(name); 25890766ec82SStephen Brennan struct dentry *res = __kern_path_locked(filename, path); 25900766ec82SStephen Brennan 25910766ec82SStephen Brennan putname(filename); 25920766ec82SStephen Brennan return res; 25930766ec82SStephen Brennan } 25940766ec82SStephen Brennan 2595d1811465SAl Viro int kern_path(const char *name, unsigned int flags, struct path *path) 2596d1811465SAl Viro { 2597794ebceaSStephen Brennan struct filename *filename = getname_kernel(name); 2598794ebceaSStephen Brennan int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL); 2599794ebceaSStephen Brennan 2600794ebceaSStephen Brennan putname(filename); 2601794ebceaSStephen Brennan return ret; 2602794ebceaSStephen Brennan 2603d1811465SAl Viro } 26044d359507SAl Viro EXPORT_SYMBOL(kern_path); 2605d1811465SAl Viro 260616f18200SJosef 'Jeff' Sipek /** 260716f18200SJosef 'Jeff' Sipek * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair 260816f18200SJosef 'Jeff' Sipek * @dentry: pointer to dentry of the base directory 260916f18200SJosef 'Jeff' Sipek * @mnt: pointer to vfs mount of the base directory 261016f18200SJosef 'Jeff' Sipek * @name: pointer to file name 261116f18200SJosef 'Jeff' Sipek * @flags: lookup flags 2612e0a01249SAl Viro * @path: pointer to struct path to fill 261316f18200SJosef 'Jeff' Sipek */ 261416f18200SJosef 'Jeff' Sipek int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt, 261516f18200SJosef 'Jeff' Sipek const char *name, unsigned int flags, 2616e0a01249SAl Viro struct path *path) 261716f18200SJosef 'Jeff' Sipek { 2618794ebceaSStephen Brennan struct filename *filename; 26199ad1aaa6SAl Viro struct path root = {.mnt = mnt, .dentry = dentry}; 2620794ebceaSStephen Brennan int ret; 2621794ebceaSStephen Brennan 2622794ebceaSStephen Brennan filename = getname_kernel(name); 26239ad1aaa6SAl Viro /* the first argument of filename_lookup() is ignored with root */ 2624794ebceaSStephen Brennan ret = filename_lookup(AT_FDCWD, filename, flags, path, &root); 2625794ebceaSStephen Brennan putname(filename); 2626794ebceaSStephen Brennan return ret; 262716f18200SJosef 'Jeff' Sipek } 26284d359507SAl Viro EXPORT_SYMBOL(vfs_path_lookup); 262916f18200SJosef 'Jeff' Sipek 26304609e1f1SChristian Brauner static int lookup_one_common(struct mnt_idmap *idmap, 2631c2fd68b6SChristian Brauner const char *name, struct dentry *base, int len, 2632c2fd68b6SChristian Brauner struct qstr *this) 26333c95f0dcSAl Viro { 26343c95f0dcSAl Viro this->name = name; 26353c95f0dcSAl Viro this->len = len; 26363c95f0dcSAl Viro this->hash = full_name_hash(base, name, len); 26373c95f0dcSAl Viro if (!len) 26383c95f0dcSAl Viro return -EACCES; 26393c95f0dcSAl Viro 26403c95f0dcSAl Viro if (unlikely(name[0] == '.')) { 26413c95f0dcSAl Viro if (len < 2 || (len == 2 && name[1] == '.')) 26423c95f0dcSAl Viro return -EACCES; 26433c95f0dcSAl Viro } 26443c95f0dcSAl Viro 26453c95f0dcSAl Viro while (len--) { 26463c95f0dcSAl Viro unsigned int c = *(const unsigned char *)name++; 26473c95f0dcSAl Viro if (c == '/' || c == '\0') 26483c95f0dcSAl Viro return -EACCES; 26493c95f0dcSAl Viro } 26503c95f0dcSAl Viro /* 26513c95f0dcSAl Viro * See if the low-level filesystem might want 26523c95f0dcSAl Viro * to use its own hash.. 26533c95f0dcSAl Viro */ 26543c95f0dcSAl Viro if (base->d_flags & DCACHE_OP_HASH) { 26553c95f0dcSAl Viro int err = base->d_op->d_hash(base, this); 26563c95f0dcSAl Viro if (err < 0) 26573c95f0dcSAl Viro return err; 26583c95f0dcSAl Viro } 26593c95f0dcSAl Viro 26604609e1f1SChristian Brauner return inode_permission(idmap, base->d_inode, MAY_EXEC); 26613c95f0dcSAl Viro } 26623c95f0dcSAl Viro 2663eead1911SChristoph Hellwig /** 26640da0b7fdSDavid Howells * try_lookup_one_len - filesystem helper to lookup single pathname component 26650da0b7fdSDavid Howells * @name: pathname component to lookup 26660da0b7fdSDavid Howells * @base: base directory to lookup from 26670da0b7fdSDavid Howells * @len: maximum length @len should be interpreted to 26680da0b7fdSDavid Howells * 26690da0b7fdSDavid Howells * Look up a dentry by name in the dcache, returning NULL if it does not 26700da0b7fdSDavid Howells * currently exist. The function does not try to create a dentry. 26710da0b7fdSDavid Howells * 26720da0b7fdSDavid Howells * Note that this routine is purely a helper for filesystem usage and should 26730da0b7fdSDavid Howells * not be called by generic code. 26740da0b7fdSDavid Howells * 26750da0b7fdSDavid Howells * The caller must hold base->i_mutex. 26760da0b7fdSDavid Howells */ 26770da0b7fdSDavid Howells struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len) 26780da0b7fdSDavid Howells { 26790da0b7fdSDavid Howells struct qstr this; 26800da0b7fdSDavid Howells int err; 26810da0b7fdSDavid Howells 26820da0b7fdSDavid Howells WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 26830da0b7fdSDavid Howells 26844609e1f1SChristian Brauner err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this); 26850da0b7fdSDavid Howells if (err) 26860da0b7fdSDavid Howells return ERR_PTR(err); 26870da0b7fdSDavid Howells 26880da0b7fdSDavid Howells return lookup_dcache(&this, base, 0); 26890da0b7fdSDavid Howells } 26900da0b7fdSDavid Howells EXPORT_SYMBOL(try_lookup_one_len); 26910da0b7fdSDavid Howells 26920da0b7fdSDavid Howells /** 2693a6b91919SRandy Dunlap * lookup_one_len - filesystem helper to lookup single pathname component 2694eead1911SChristoph Hellwig * @name: pathname component to lookup 2695eead1911SChristoph Hellwig * @base: base directory to lookup from 2696eead1911SChristoph Hellwig * @len: maximum length @len should be interpreted to 2697eead1911SChristoph Hellwig * 2698a6b91919SRandy Dunlap * Note that this routine is purely a helper for filesystem usage and should 26999e7543e9SAl Viro * not be called by generic code. 2700bbddca8eSNeilBrown * 2701bbddca8eSNeilBrown * The caller must hold base->i_mutex. 2702eead1911SChristoph Hellwig */ 2703057f6c01SJames Morris struct dentry *lookup_one_len(const char *name, struct dentry *base, int len) 2704057f6c01SJames Morris { 27058613a209SAl Viro struct dentry *dentry; 2706057f6c01SJames Morris struct qstr this; 2707cda309deSMiklos Szeredi int err; 2708057f6c01SJames Morris 27095955102cSAl Viro WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 27102f9092e1SDavid Woodhouse 27114609e1f1SChristian Brauner err = lookup_one_common(&nop_mnt_idmap, name, base, len, &this); 2712cda309deSMiklos Szeredi if (err) 2713cda309deSMiklos Szeredi return ERR_PTR(err); 2714cda309deSMiklos Szeredi 27158613a209SAl Viro dentry = lookup_dcache(&this, base, 0); 27168613a209SAl Viro return dentry ? dentry : __lookup_slow(&this, base, 0); 2717057f6c01SJames Morris } 27184d359507SAl Viro EXPORT_SYMBOL(lookup_one_len); 2719057f6c01SJames Morris 2720bbddca8eSNeilBrown /** 2721c2fd68b6SChristian Brauner * lookup_one - filesystem helper to lookup single pathname component 27224609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 2723c2fd68b6SChristian Brauner * @name: pathname component to lookup 2724c2fd68b6SChristian Brauner * @base: base directory to lookup from 2725c2fd68b6SChristian Brauner * @len: maximum length @len should be interpreted to 2726c2fd68b6SChristian Brauner * 2727c2fd68b6SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 2728c2fd68b6SChristian Brauner * not be called by generic code. 2729c2fd68b6SChristian Brauner * 2730c2fd68b6SChristian Brauner * The caller must hold base->i_mutex. 2731c2fd68b6SChristian Brauner */ 27324609e1f1SChristian Brauner struct dentry *lookup_one(struct mnt_idmap *idmap, const char *name, 2733c2fd68b6SChristian Brauner struct dentry *base, int len) 2734c2fd68b6SChristian Brauner { 2735c2fd68b6SChristian Brauner struct dentry *dentry; 2736c2fd68b6SChristian Brauner struct qstr this; 2737c2fd68b6SChristian Brauner int err; 2738c2fd68b6SChristian Brauner 2739c2fd68b6SChristian Brauner WARN_ON_ONCE(!inode_is_locked(base->d_inode)); 2740c2fd68b6SChristian Brauner 27414609e1f1SChristian Brauner err = lookup_one_common(idmap, name, base, len, &this); 2742c2fd68b6SChristian Brauner if (err) 2743c2fd68b6SChristian Brauner return ERR_PTR(err); 2744c2fd68b6SChristian Brauner 2745c2fd68b6SChristian Brauner dentry = lookup_dcache(&this, base, 0); 2746c2fd68b6SChristian Brauner return dentry ? dentry : __lookup_slow(&this, base, 0); 2747c2fd68b6SChristian Brauner } 2748c2fd68b6SChristian Brauner EXPORT_SYMBOL(lookup_one); 2749c2fd68b6SChristian Brauner 2750c2fd68b6SChristian Brauner /** 275100675017SChristian Brauner * lookup_one_unlocked - filesystem helper to lookup single pathname component 27524609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 275300675017SChristian Brauner * @name: pathname component to lookup 275400675017SChristian Brauner * @base: base directory to lookup from 275500675017SChristian Brauner * @len: maximum length @len should be interpreted to 275600675017SChristian Brauner * 275700675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 275800675017SChristian Brauner * not be called by generic code. 275900675017SChristian Brauner * 276000675017SChristian Brauner * Unlike lookup_one_len, it should be called without the parent 276100675017SChristian Brauner * i_mutex held, and will take the i_mutex itself if necessary. 276200675017SChristian Brauner */ 27634609e1f1SChristian Brauner struct dentry *lookup_one_unlocked(struct mnt_idmap *idmap, 276400675017SChristian Brauner const char *name, struct dentry *base, 276500675017SChristian Brauner int len) 276600675017SChristian Brauner { 276700675017SChristian Brauner struct qstr this; 276800675017SChristian Brauner int err; 276900675017SChristian Brauner struct dentry *ret; 277000675017SChristian Brauner 27714609e1f1SChristian Brauner err = lookup_one_common(idmap, name, base, len, &this); 277200675017SChristian Brauner if (err) 277300675017SChristian Brauner return ERR_PTR(err); 277400675017SChristian Brauner 277500675017SChristian Brauner ret = lookup_dcache(&this, base, 0); 277600675017SChristian Brauner if (!ret) 277700675017SChristian Brauner ret = lookup_slow(&this, base, 0); 277800675017SChristian Brauner return ret; 277900675017SChristian Brauner } 278000675017SChristian Brauner EXPORT_SYMBOL(lookup_one_unlocked); 278100675017SChristian Brauner 278200675017SChristian Brauner /** 278300675017SChristian Brauner * lookup_one_positive_unlocked - filesystem helper to lookup single 278400675017SChristian Brauner * pathname component 27854609e1f1SChristian Brauner * @idmap: idmap of the mount the lookup is performed from 278600675017SChristian Brauner * @name: pathname component to lookup 278700675017SChristian Brauner * @base: base directory to lookup from 278800675017SChristian Brauner * @len: maximum length @len should be interpreted to 278900675017SChristian Brauner * 279000675017SChristian Brauner * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns 279100675017SChristian Brauner * known positive or ERR_PTR(). This is what most of the users want. 279200675017SChristian Brauner * 279300675017SChristian Brauner * Note that pinned negative with unlocked parent _can_ become positive at any 279400675017SChristian Brauner * time, so callers of lookup_one_unlocked() need to be very careful; pinned 279500675017SChristian Brauner * positives have >d_inode stable, so this one avoids such problems. 279600675017SChristian Brauner * 279700675017SChristian Brauner * Note that this routine is purely a helper for filesystem usage and should 279800675017SChristian Brauner * not be called by generic code. 279900675017SChristian Brauner * 280000675017SChristian Brauner * The helper should be called without i_mutex held. 280100675017SChristian Brauner */ 28024609e1f1SChristian Brauner struct dentry *lookup_one_positive_unlocked(struct mnt_idmap *idmap, 280300675017SChristian Brauner const char *name, 280400675017SChristian Brauner struct dentry *base, int len) 280500675017SChristian Brauner { 28064609e1f1SChristian Brauner struct dentry *ret = lookup_one_unlocked(idmap, name, base, len); 280700675017SChristian Brauner 280800675017SChristian Brauner if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) { 280900675017SChristian Brauner dput(ret); 281000675017SChristian Brauner ret = ERR_PTR(-ENOENT); 281100675017SChristian Brauner } 281200675017SChristian Brauner return ret; 281300675017SChristian Brauner } 281400675017SChristian Brauner EXPORT_SYMBOL(lookup_one_positive_unlocked); 281500675017SChristian Brauner 281600675017SChristian Brauner /** 2817bbddca8eSNeilBrown * lookup_one_len_unlocked - filesystem helper to lookup single pathname component 2818bbddca8eSNeilBrown * @name: pathname component to lookup 2819bbddca8eSNeilBrown * @base: base directory to lookup from 2820bbddca8eSNeilBrown * @len: maximum length @len should be interpreted to 2821bbddca8eSNeilBrown * 2822bbddca8eSNeilBrown * Note that this routine is purely a helper for filesystem usage and should 2823bbddca8eSNeilBrown * not be called by generic code. 2824bbddca8eSNeilBrown * 2825bbddca8eSNeilBrown * Unlike lookup_one_len, it should be called without the parent 2826bbddca8eSNeilBrown * i_mutex held, and will take the i_mutex itself if necessary. 2827bbddca8eSNeilBrown */ 2828bbddca8eSNeilBrown struct dentry *lookup_one_len_unlocked(const char *name, 2829bbddca8eSNeilBrown struct dentry *base, int len) 2830bbddca8eSNeilBrown { 28314609e1f1SChristian Brauner return lookup_one_unlocked(&nop_mnt_idmap, name, base, len); 2832bbddca8eSNeilBrown } 2833bbddca8eSNeilBrown EXPORT_SYMBOL(lookup_one_len_unlocked); 2834bbddca8eSNeilBrown 28356c2d4798SAl Viro /* 28366c2d4798SAl Viro * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT) 28376c2d4798SAl Viro * on negatives. Returns known positive or ERR_PTR(); that's what 28386c2d4798SAl Viro * most of the users want. Note that pinned negative with unlocked parent 28396c2d4798SAl Viro * _can_ become positive at any time, so callers of lookup_one_len_unlocked() 28406c2d4798SAl Viro * need to be very careful; pinned positives have ->d_inode stable, so 28416c2d4798SAl Viro * this one avoids such problems. 28426c2d4798SAl Viro */ 28436c2d4798SAl Viro struct dentry *lookup_positive_unlocked(const char *name, 28446c2d4798SAl Viro struct dentry *base, int len) 28456c2d4798SAl Viro { 28464609e1f1SChristian Brauner return lookup_one_positive_unlocked(&nop_mnt_idmap, name, base, len); 28476c2d4798SAl Viro } 28486c2d4798SAl Viro EXPORT_SYMBOL(lookup_positive_unlocked); 28496c2d4798SAl Viro 2850eedf265aSEric W. Biederman #ifdef CONFIG_UNIX98_PTYS 2851eedf265aSEric W. Biederman int path_pts(struct path *path) 2852eedf265aSEric W. Biederman { 2853eedf265aSEric W. Biederman /* Find something mounted on "pts" in the same directory as 2854eedf265aSEric W. Biederman * the input path. 2855eedf265aSEric W. Biederman */ 2856a6a7eb76SAl Viro struct dentry *parent = dget_parent(path->dentry); 2857a6a7eb76SAl Viro struct dentry *child; 285819f6028aSAl Viro struct qstr this = QSTR_INIT("pts", 3); 2859eedf265aSEric W. Biederman 2860a6a7eb76SAl Viro if (unlikely(!path_connected(path->mnt, parent))) { 2861a6a7eb76SAl Viro dput(parent); 286263b27720SAl Viro return -ENOENT; 2863a6a7eb76SAl Viro } 286463b27720SAl Viro dput(path->dentry); 286563b27720SAl Viro path->dentry = parent; 2866eedf265aSEric W. Biederman child = d_hash_and_lookup(parent, &this); 2867eedf265aSEric W. Biederman if (!child) 2868eedf265aSEric W. Biederman return -ENOENT; 2869eedf265aSEric W. Biederman 2870eedf265aSEric W. Biederman path->dentry = child; 2871eedf265aSEric W. Biederman dput(parent); 287219f6028aSAl Viro follow_down(path); 2873eedf265aSEric W. Biederman return 0; 2874eedf265aSEric W. Biederman } 2875eedf265aSEric W. Biederman #endif 2876eedf265aSEric W. Biederman 28771fa1e7f6SAndy Whitcroft int user_path_at_empty(int dfd, const char __user *name, unsigned flags, 28781fa1e7f6SAndy Whitcroft struct path *path, int *empty) 28791da177e4SLinus Torvalds { 2880794ebceaSStephen Brennan struct filename *filename = getname_flags(name, flags, empty); 2881794ebceaSStephen Brennan int ret = filename_lookup(dfd, filename, flags, path, NULL); 2882794ebceaSStephen Brennan 2883794ebceaSStephen Brennan putname(filename); 2884794ebceaSStephen Brennan return ret; 28851da177e4SLinus Torvalds } 2886b853a161SAl Viro EXPORT_SYMBOL(user_path_at_empty); 28871fa1e7f6SAndy Whitcroft 2888ba73d987SChristian Brauner int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir, 2889ba73d987SChristian Brauner struct inode *inode) 28901da177e4SLinus Torvalds { 28918e96e3b7SEric W. Biederman kuid_t fsuid = current_fsuid(); 2892da9592edSDavid Howells 2893a2bd096fSChristian Brauner if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, inode), fsuid)) 28941da177e4SLinus Torvalds return 0; 2895a2bd096fSChristian Brauner if (vfsuid_eq_kuid(i_uid_into_vfsuid(mnt_userns, dir), fsuid)) 28961da177e4SLinus Torvalds return 0; 2897ba73d987SChristian Brauner return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER); 28981da177e4SLinus Torvalds } 2899cbdf35bcSMiklos Szeredi EXPORT_SYMBOL(__check_sticky); 29001da177e4SLinus Torvalds 29011da177e4SLinus Torvalds /* 29021da177e4SLinus Torvalds * Check whether we can remove a link victim from directory dir, check 29031da177e4SLinus Torvalds * whether the type of victim is right. 29041da177e4SLinus Torvalds * 1. We can't do it if dir is read-only (done in permission()) 29051da177e4SLinus Torvalds * 2. We should have write and exec permissions on dir 29061da177e4SLinus Torvalds * 3. We can't remove anything from append-only dir 29071da177e4SLinus Torvalds * 4. We can't do anything with immutable dir (done in permission()) 29081da177e4SLinus Torvalds * 5. If the sticky bit on dir is set we should either 29091da177e4SLinus Torvalds * a. be owner of dir, or 29101da177e4SLinus Torvalds * b. be owner of victim, or 29111da177e4SLinus Torvalds * c. have CAP_FOWNER capability 29121da177e4SLinus Torvalds * 6. If the victim is append-only or immutable we can't do antyhing with 29131da177e4SLinus Torvalds * links pointing to it. 29140bd23d09SEric W. Biederman * 7. If the victim has an unknown uid or gid we can't change the inode. 29150bd23d09SEric W. Biederman * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR. 29160bd23d09SEric W. Biederman * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR. 29170bd23d09SEric W. Biederman * 10. We can't remove a root or mountpoint. 29180bd23d09SEric W. Biederman * 11. We don't allow removal of NFS sillyrenamed files; it's handled by 29191da177e4SLinus Torvalds * nfs_async_unlink(). 29201da177e4SLinus Torvalds */ 29214609e1f1SChristian Brauner static int may_delete(struct mnt_idmap *idmap, struct inode *dir, 2922ba73d987SChristian Brauner struct dentry *victim, bool isdir) 29231da177e4SLinus Torvalds { 29244609e1f1SChristian Brauner struct user_namespace *mnt_userns = mnt_idmap_owner(idmap); 292563afdfc7SDavid Howells struct inode *inode = d_backing_inode(victim); 29261da177e4SLinus Torvalds int error; 29271da177e4SLinus Torvalds 2928b18825a7SDavid Howells if (d_is_negative(victim)) 29291da177e4SLinus Torvalds return -ENOENT; 2930b18825a7SDavid Howells BUG_ON(!inode); 29311da177e4SLinus Torvalds 29321da177e4SLinus Torvalds BUG_ON(victim->d_parent->d_inode != dir); 2933593d1ce8SEric W. Biederman 2934593d1ce8SEric W. Biederman /* Inode writeback is not safe when the uid or gid are invalid. */ 2935a2bd096fSChristian Brauner if (!vfsuid_valid(i_uid_into_vfsuid(mnt_userns, inode)) || 2936a2bd096fSChristian Brauner !vfsgid_valid(i_gid_into_vfsgid(mnt_userns, inode))) 2937593d1ce8SEric W. Biederman return -EOVERFLOW; 2938593d1ce8SEric W. Biederman 29394fa6b5ecSJeff Layton audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE); 29401da177e4SLinus Torvalds 29414609e1f1SChristian Brauner error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 29421da177e4SLinus Torvalds if (error) 29431da177e4SLinus Torvalds return error; 29441da177e4SLinus Torvalds if (IS_APPEND(dir)) 29451da177e4SLinus Torvalds return -EPERM; 2946b18825a7SDavid Howells 2947ba73d987SChristian Brauner if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) || 2948ba73d987SChristian Brauner IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || 29494609e1f1SChristian Brauner HAS_UNMAPPED_ID(idmap, inode)) 29501da177e4SLinus Torvalds return -EPERM; 29511da177e4SLinus Torvalds if (isdir) { 295244b1d530SMiklos Szeredi if (!d_is_dir(victim)) 29531da177e4SLinus Torvalds return -ENOTDIR; 29541da177e4SLinus Torvalds if (IS_ROOT(victim)) 29551da177e4SLinus Torvalds return -EBUSY; 295644b1d530SMiklos Szeredi } else if (d_is_dir(victim)) 29571da177e4SLinus Torvalds return -EISDIR; 29581da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29591da177e4SLinus Torvalds return -ENOENT; 29601da177e4SLinus Torvalds if (victim->d_flags & DCACHE_NFSFS_RENAMED) 29611da177e4SLinus Torvalds return -EBUSY; 29621da177e4SLinus Torvalds return 0; 29631da177e4SLinus Torvalds } 29641da177e4SLinus Torvalds 29651da177e4SLinus Torvalds /* Check whether we can create an object with dentry child in directory 29661da177e4SLinus Torvalds * dir. 29671da177e4SLinus Torvalds * 1. We can't do it if child already exists (open has special treatment for 29681da177e4SLinus Torvalds * this case, but since we are inlined it's OK) 29691da177e4SLinus Torvalds * 2. We can't do it if dir is read-only (done in permission()) 2970036d5236SEric W. Biederman * 3. We can't do it if the fs can't represent the fsuid or fsgid. 2971036d5236SEric W. Biederman * 4. We should have write and exec permissions on dir 2972036d5236SEric W. Biederman * 5. We can't do it if dir is immutable (done in permission()) 29731da177e4SLinus Torvalds */ 29744609e1f1SChristian Brauner static inline int may_create(struct mnt_idmap *idmap, 2975ba73d987SChristian Brauner struct inode *dir, struct dentry *child) 29761da177e4SLinus Torvalds { 297714e972b4SJeff Layton audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE); 29781da177e4SLinus Torvalds if (child->d_inode) 29791da177e4SLinus Torvalds return -EEXIST; 29801da177e4SLinus Torvalds if (IS_DEADDIR(dir)) 29811da177e4SLinus Torvalds return -ENOENT; 29824609e1f1SChristian Brauner if (!fsuidgid_has_mapping(dir->i_sb, idmap)) 2983036d5236SEric W. Biederman return -EOVERFLOW; 29848e538913SChristian Brauner 29854609e1f1SChristian Brauner return inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 29861da177e4SLinus Torvalds } 29871da177e4SLinus Torvalds 29881da177e4SLinus Torvalds /* 29891da177e4SLinus Torvalds * p1 and p2 should be directories on the same fs. 29901da177e4SLinus Torvalds */ 29911da177e4SLinus Torvalds struct dentry *lock_rename(struct dentry *p1, struct dentry *p2) 29921da177e4SLinus Torvalds { 29931da177e4SLinus Torvalds struct dentry *p; 29941da177e4SLinus Torvalds 29951da177e4SLinus Torvalds if (p1 == p2) { 29965955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 29971da177e4SLinus Torvalds return NULL; 29981da177e4SLinus Torvalds } 29991da177e4SLinus Torvalds 3000fc64005cSAl Viro mutex_lock(&p1->d_sb->s_vfs_rename_mutex); 30011da177e4SLinus Torvalds 3002e2761a11SOGAWA Hirofumi p = d_ancestor(p2, p1); 3003e2761a11SOGAWA Hirofumi if (p) { 30045955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT); 30055955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_CHILD); 30061da177e4SLinus Torvalds return p; 30071da177e4SLinus Torvalds } 30081da177e4SLinus Torvalds 3009e2761a11SOGAWA Hirofumi p = d_ancestor(p1, p2); 3010e2761a11SOGAWA Hirofumi if (p) { 30115955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30125955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_CHILD); 30131da177e4SLinus Torvalds return p; 30141da177e4SLinus Torvalds } 30151da177e4SLinus Torvalds 30165955102cSAl Viro inode_lock_nested(p1->d_inode, I_MUTEX_PARENT); 30175955102cSAl Viro inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2); 30181da177e4SLinus Torvalds return NULL; 30191da177e4SLinus Torvalds } 30204d359507SAl Viro EXPORT_SYMBOL(lock_rename); 30211da177e4SLinus Torvalds 30221da177e4SLinus Torvalds void unlock_rename(struct dentry *p1, struct dentry *p2) 30231da177e4SLinus Torvalds { 30245955102cSAl Viro inode_unlock(p1->d_inode); 30251da177e4SLinus Torvalds if (p1 != p2) { 30265955102cSAl Viro inode_unlock(p2->d_inode); 3027fc64005cSAl Viro mutex_unlock(&p1->d_sb->s_vfs_rename_mutex); 30281da177e4SLinus Torvalds } 30291da177e4SLinus Torvalds } 30304d359507SAl Viro EXPORT_SYMBOL(unlock_rename); 30311da177e4SLinus Torvalds 30326521f891SChristian Brauner /** 30331639a49cSYang Xu * mode_strip_umask - handle vfs umask stripping 30341639a49cSYang Xu * @dir: parent directory of the new inode 30351639a49cSYang Xu * @mode: mode of the new inode to be created in @dir 30361639a49cSYang Xu * 30371639a49cSYang Xu * Umask stripping depends on whether or not the filesystem supports POSIX 30381639a49cSYang Xu * ACLs. If the filesystem doesn't support it umask stripping is done directly 30391639a49cSYang Xu * in here. If the filesystem does support POSIX ACLs umask stripping is 30401639a49cSYang Xu * deferred until the filesystem calls posix_acl_create(). 30411639a49cSYang Xu * 30421639a49cSYang Xu * Returns: mode 30431639a49cSYang Xu */ 30441639a49cSYang Xu static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode) 30451639a49cSYang Xu { 30461639a49cSYang Xu if (!IS_POSIXACL(dir)) 30471639a49cSYang Xu mode &= ~current_umask(); 30481639a49cSYang Xu return mode; 30491639a49cSYang Xu } 30501639a49cSYang Xu 30511639a49cSYang Xu /** 30521639a49cSYang Xu * vfs_prepare_mode - prepare the mode to be used for a new inode 30531639a49cSYang Xu * @mnt_userns: user namespace of the mount the inode was found from 30541639a49cSYang Xu * @dir: parent directory of the new inode 30551639a49cSYang Xu * @mode: mode of the new inode 30561639a49cSYang Xu * @mask_perms: allowed permission by the vfs 30571639a49cSYang Xu * @type: type of file to be created 30581639a49cSYang Xu * 30591639a49cSYang Xu * This helper consolidates and enforces vfs restrictions on the @mode of a new 30601639a49cSYang Xu * object to be created. 30611639a49cSYang Xu * 30621639a49cSYang Xu * Umask stripping depends on whether the filesystem supports POSIX ACLs (see 30631639a49cSYang Xu * the kernel documentation for mode_strip_umask()). Moving umask stripping 30641639a49cSYang Xu * after setgid stripping allows the same ordering for both non-POSIX ACL and 30651639a49cSYang Xu * POSIX ACL supporting filesystems. 30661639a49cSYang Xu * 30671639a49cSYang Xu * Note that it's currently valid for @type to be 0 if a directory is created. 30681639a49cSYang Xu * Filesystems raise that flag individually and we need to check whether each 30691639a49cSYang Xu * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a 30701639a49cSYang Xu * non-zero type. 30711639a49cSYang Xu * 30721639a49cSYang Xu * Returns: mode to be passed to the filesystem 30731639a49cSYang Xu */ 30741639a49cSYang Xu static inline umode_t vfs_prepare_mode(struct user_namespace *mnt_userns, 30751639a49cSYang Xu const struct inode *dir, umode_t mode, 30761639a49cSYang Xu umode_t mask_perms, umode_t type) 30771639a49cSYang Xu { 30781639a49cSYang Xu mode = mode_strip_sgid(mnt_userns, dir, mode); 30791639a49cSYang Xu mode = mode_strip_umask(dir, mode); 30801639a49cSYang Xu 30811639a49cSYang Xu /* 30821639a49cSYang Xu * Apply the vfs mandated allowed permission mask and set the type of 30831639a49cSYang Xu * file to be created before we call into the filesystem. 30841639a49cSYang Xu */ 30851639a49cSYang Xu mode &= (mask_perms & ~S_IFMT); 30861639a49cSYang Xu mode |= (type & S_IFMT); 30871639a49cSYang Xu 30881639a49cSYang Xu return mode; 30891639a49cSYang Xu } 30901639a49cSYang Xu 30911639a49cSYang Xu /** 30926521f891SChristian Brauner * vfs_create - create new file 3093abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 30946521f891SChristian Brauner * @dir: inode of @dentry 30956521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 30966521f891SChristian Brauner * @mode: mode of the new file 30976521f891SChristian Brauner * @want_excl: whether the file must not yet exist 30986521f891SChristian Brauner * 30996521f891SChristian Brauner * Create a new file. 31006521f891SChristian Brauner * 3101abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3102abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3103abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 31046521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3105abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 31066521f891SChristian Brauner */ 3107abf08576SChristian Brauner int vfs_create(struct mnt_idmap *idmap, struct inode *dir, 31086521f891SChristian Brauner struct dentry *dentry, umode_t mode, bool want_excl) 31091da177e4SLinus Torvalds { 3110abf08576SChristian Brauner struct user_namespace *mnt_userns = mnt_idmap_owner(idmap); 3111abf08576SChristian Brauner int error; 3112abf08576SChristian Brauner 31134609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 31141da177e4SLinus Torvalds if (error) 31151da177e4SLinus Torvalds return error; 31161da177e4SLinus Torvalds 3117acfa4380SAl Viro if (!dir->i_op->create) 31181da177e4SLinus Torvalds return -EACCES; /* shouldn't it be ENOSYS? */ 31191639a49cSYang Xu 31201639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IALLUGO, S_IFREG); 31211da177e4SLinus Torvalds error = security_inode_create(dir, dentry, mode); 31221da177e4SLinus Torvalds if (error) 31231da177e4SLinus Torvalds return error; 31246c960e68SChristian Brauner error = dir->i_op->create(idmap, dir, dentry, mode, want_excl); 3125a74574aaSStephen Smalley if (!error) 3126f38aa942SAmy Griffis fsnotify_create(dir, dentry); 31271da177e4SLinus Torvalds return error; 31281da177e4SLinus Torvalds } 31294d359507SAl Viro EXPORT_SYMBOL(vfs_create); 31301da177e4SLinus Torvalds 31318e6c848eSAl Viro int vfs_mkobj(struct dentry *dentry, umode_t mode, 31328e6c848eSAl Viro int (*f)(struct dentry *, umode_t, void *), 31338e6c848eSAl Viro void *arg) 31348e6c848eSAl Viro { 31358e6c848eSAl Viro struct inode *dir = dentry->d_parent->d_inode; 31364609e1f1SChristian Brauner int error = may_create(&nop_mnt_idmap, dir, dentry); 31378e6c848eSAl Viro if (error) 31388e6c848eSAl Viro return error; 31398e6c848eSAl Viro 31408e6c848eSAl Viro mode &= S_IALLUGO; 31418e6c848eSAl Viro mode |= S_IFREG; 31428e6c848eSAl Viro error = security_inode_create(dir, dentry, mode); 31438e6c848eSAl Viro if (error) 31448e6c848eSAl Viro return error; 31458e6c848eSAl Viro error = f(dentry, mode, arg); 31468e6c848eSAl Viro if (!error) 31478e6c848eSAl Viro fsnotify_create(dir, dentry); 31488e6c848eSAl Viro return error; 31498e6c848eSAl Viro } 31508e6c848eSAl Viro EXPORT_SYMBOL(vfs_mkobj); 31518e6c848eSAl Viro 3152a2982cc9SEric W. Biederman bool may_open_dev(const struct path *path) 3153a2982cc9SEric W. Biederman { 3154a2982cc9SEric W. Biederman return !(path->mnt->mnt_flags & MNT_NODEV) && 3155a2982cc9SEric W. Biederman !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV); 3156a2982cc9SEric W. Biederman } 3157a2982cc9SEric W. Biederman 31584609e1f1SChristian Brauner static int may_open(struct mnt_idmap *idmap, const struct path *path, 3159ba73d987SChristian Brauner int acc_mode, int flag) 31601da177e4SLinus Torvalds { 31613fb64190SChristoph Hellwig struct dentry *dentry = path->dentry; 31621da177e4SLinus Torvalds struct inode *inode = dentry->d_inode; 31631da177e4SLinus Torvalds int error; 31641da177e4SLinus Torvalds 31651da177e4SLinus Torvalds if (!inode) 31661da177e4SLinus Torvalds return -ENOENT; 31671da177e4SLinus Torvalds 3168c8fe8f30SChristoph Hellwig switch (inode->i_mode & S_IFMT) { 3169c8fe8f30SChristoph Hellwig case S_IFLNK: 31701da177e4SLinus Torvalds return -ELOOP; 3171c8fe8f30SChristoph Hellwig case S_IFDIR: 3172fc4177beSKees Cook if (acc_mode & MAY_WRITE) 31731da177e4SLinus Torvalds return -EISDIR; 3174fc4177beSKees Cook if (acc_mode & MAY_EXEC) 3175fc4177beSKees Cook return -EACCES; 3176c8fe8f30SChristoph Hellwig break; 3177c8fe8f30SChristoph Hellwig case S_IFBLK: 3178c8fe8f30SChristoph Hellwig case S_IFCHR: 3179a2982cc9SEric W. Biederman if (!may_open_dev(path)) 31801da177e4SLinus Torvalds return -EACCES; 3181633fb6acSKees Cook fallthrough; 3182c8fe8f30SChristoph Hellwig case S_IFIFO: 3183c8fe8f30SChristoph Hellwig case S_IFSOCK: 3184633fb6acSKees Cook if (acc_mode & MAY_EXEC) 3185633fb6acSKees Cook return -EACCES; 31861da177e4SLinus Torvalds flag &= ~O_TRUNC; 3187c8fe8f30SChristoph Hellwig break; 31880fd338b2SKees Cook case S_IFREG: 31890fd338b2SKees Cook if ((acc_mode & MAY_EXEC) && path_noexec(path)) 31900fd338b2SKees Cook return -EACCES; 31910fd338b2SKees Cook break; 31924a3fd211SDave Hansen } 3193b41572e9SDave Hansen 31944609e1f1SChristian Brauner error = inode_permission(idmap, inode, MAY_OPEN | acc_mode); 3195b41572e9SDave Hansen if (error) 3196b41572e9SDave Hansen return error; 31976146f0d5SMimi Zohar 31981da177e4SLinus Torvalds /* 31991da177e4SLinus Torvalds * An append-only file must be opened in append mode for writing. 32001da177e4SLinus Torvalds */ 32011da177e4SLinus Torvalds if (IS_APPEND(inode)) { 32028737c930SAl Viro if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND)) 32037715b521SAl Viro return -EPERM; 32041da177e4SLinus Torvalds if (flag & O_TRUNC) 32057715b521SAl Viro return -EPERM; 32061da177e4SLinus Torvalds } 32071da177e4SLinus Torvalds 32081da177e4SLinus Torvalds /* O_NOATIME can only be set by the owner or superuser */ 3209*01beba79SChristian Brauner if (flag & O_NOATIME && !inode_owner_or_capable(idmap, inode)) 32107715b521SAl Viro return -EPERM; 32111da177e4SLinus Torvalds 3212f3c7691eSJ. Bruce Fields return 0; 32137715b521SAl Viro } 32147715b521SAl Viro 3215abf08576SChristian Brauner static int handle_truncate(struct mnt_idmap *idmap, struct file *filp) 32167715b521SAl Viro { 3217f0bb5aafSAl Viro const struct path *path = &filp->f_path; 32187715b521SAl Viro struct inode *inode = path->dentry->d_inode; 32197715b521SAl Viro int error = get_write_access(inode); 32201da177e4SLinus Torvalds if (error) 32217715b521SAl Viro return error; 3222482e0007SJeff Layton 32233350607dSGünther Noack error = security_file_truncate(filp); 32241da177e4SLinus Torvalds if (!error) { 3225abf08576SChristian Brauner error = do_truncate(idmap, path->dentry, 0, 3226d139d7ffSMiklos Szeredi ATTR_MTIME|ATTR_CTIME|ATTR_OPEN, 3227e1181ee6SJeff Layton filp); 32281da177e4SLinus Torvalds } 32291da177e4SLinus Torvalds put_write_access(inode); 3230acd0c935SMimi Zohar return error; 32311da177e4SLinus Torvalds } 32321da177e4SLinus Torvalds 3233d57999e1SDave Hansen static inline int open_to_namei_flags(int flag) 3234d57999e1SDave Hansen { 32358a5e929dSAl Viro if ((flag & O_ACCMODE) == 3) 32368a5e929dSAl Viro flag--; 3237d57999e1SDave Hansen return flag; 3238d57999e1SDave Hansen } 3239d57999e1SDave Hansen 32404609e1f1SChristian Brauner static int may_o_create(struct mnt_idmap *idmap, 3241ba73d987SChristian Brauner const struct path *dir, struct dentry *dentry, 3242ba73d987SChristian Brauner umode_t mode) 3243d18e9008SMiklos Szeredi { 3244d18e9008SMiklos Szeredi int error = security_path_mknod(dir, dentry, mode, 0); 3245d18e9008SMiklos Szeredi if (error) 3246d18e9008SMiklos Szeredi return error; 3247d18e9008SMiklos Szeredi 32484609e1f1SChristian Brauner if (!fsuidgid_has_mapping(dir->dentry->d_sb, idmap)) 32491328c727SSeth Forshee return -EOVERFLOW; 32501328c727SSeth Forshee 32514609e1f1SChristian Brauner error = inode_permission(idmap, dir->dentry->d_inode, 325247291baaSChristian Brauner MAY_WRITE | MAY_EXEC); 3253d18e9008SMiklos Szeredi if (error) 3254d18e9008SMiklos Szeredi return error; 3255d18e9008SMiklos Szeredi 3256d18e9008SMiklos Szeredi return security_inode_create(dir->dentry->d_inode, dentry, mode); 3257d18e9008SMiklos Szeredi } 3258d18e9008SMiklos Szeredi 32591acf0af9SDavid Howells /* 32601acf0af9SDavid Howells * Attempt to atomically look up, create and open a file from a negative 32611acf0af9SDavid Howells * dentry. 32621acf0af9SDavid Howells * 32631acf0af9SDavid Howells * Returns 0 if successful. The file will have been created and attached to 32641acf0af9SDavid Howells * @file by the filesystem calling finish_open(). 32651acf0af9SDavid Howells * 326600a07c15SAl Viro * If the file was looked up only or didn't need creating, FMODE_OPENED won't 326700a07c15SAl Viro * be set. The caller will need to perform the open themselves. @path will 326800a07c15SAl Viro * have been updated to point to the new dentry. This may be negative. 32691acf0af9SDavid Howells * 32701acf0af9SDavid Howells * Returns an error code otherwise. 32711acf0af9SDavid Howells */ 3272239eb983SAl Viro static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry, 3273239eb983SAl Viro struct file *file, 32743ec2eef1SAl Viro int open_flag, umode_t mode) 3275d18e9008SMiklos Szeredi { 3276d18e9008SMiklos Szeredi struct dentry *const DENTRY_NOT_SET = (void *) -1UL; 3277d18e9008SMiklos Szeredi struct inode *dir = nd->path.dentry->d_inode; 3278d18e9008SMiklos Szeredi int error; 3279d18e9008SMiklos Szeredi 3280d18e9008SMiklos Szeredi if (nd->flags & LOOKUP_DIRECTORY) 3281d18e9008SMiklos Szeredi open_flag |= O_DIRECTORY; 3282d18e9008SMiklos Szeredi 328330d90494SAl Viro file->f_path.dentry = DENTRY_NOT_SET; 328430d90494SAl Viro file->f_path.mnt = nd->path.mnt; 32850fb1ea09SAl Viro error = dir->i_op->atomic_open(dir, dentry, file, 328644907d79SAl Viro open_to_namei_flags(open_flag), mode); 32876fbd0714SAl Viro d_lookup_done(dentry); 3288384f26e2SAl Viro if (!error) { 328964e1ac4dSAl Viro if (file->f_mode & FMODE_OPENED) { 32906fb968cdSAl Viro if (unlikely(dentry != file->f_path.dentry)) { 32916fb968cdSAl Viro dput(dentry); 32926fb968cdSAl Viro dentry = dget(file->f_path.dentry); 32936fb968cdSAl Viro } 329464e1ac4dSAl Viro } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) { 32952675a4ebSAl Viro error = -EIO; 3296384f26e2SAl Viro } else { 329730d90494SAl Viro if (file->f_path.dentry) { 3298d18e9008SMiklos Szeredi dput(dentry); 3299d18e9008SMiklos Szeredi dentry = file->f_path.dentry; 330010c64ceaSAl Viro } 3301239eb983SAl Viro if (unlikely(d_is_negative(dentry))) 3302a01e718fSAl Viro error = -ENOENT; 3303d18e9008SMiklos Szeredi } 3304d18e9008SMiklos Szeredi } 3305239eb983SAl Viro if (error) { 3306d18e9008SMiklos Szeredi dput(dentry); 3307239eb983SAl Viro dentry = ERR_PTR(error); 3308239eb983SAl Viro } 3309239eb983SAl Viro return dentry; 3310d18e9008SMiklos Szeredi } 3311d18e9008SMiklos Szeredi 331231e6b01fSNick Piggin /* 33131acf0af9SDavid Howells * Look up and maybe create and open the last component. 3314d58ffd35SMiklos Szeredi * 331500a07c15SAl Viro * Must be called with parent locked (exclusive in O_CREAT case). 3316d58ffd35SMiklos Szeredi * 331700a07c15SAl Viro * Returns 0 on success, that is, if 331800a07c15SAl Viro * the file was successfully atomically created (if necessary) and opened, or 331900a07c15SAl Viro * the file was not completely opened at this time, though lookups and 332000a07c15SAl Viro * creations were performed. 332100a07c15SAl Viro * These case are distinguished by presence of FMODE_OPENED on file->f_mode. 332200a07c15SAl Viro * In the latter case dentry returned in @path might be negative if O_CREAT 332300a07c15SAl Viro * hadn't been specified. 33241acf0af9SDavid Howells * 332500a07c15SAl Viro * An error code is returned on failure. 3326d58ffd35SMiklos Szeredi */ 3327da5ebf5aSAl Viro static struct dentry *lookup_open(struct nameidata *nd, struct file *file, 3328d58ffd35SMiklos Szeredi const struct open_flags *op, 33293ec2eef1SAl Viro bool got_write) 3330d58ffd35SMiklos Szeredi { 33316c960e68SChristian Brauner struct mnt_idmap *idmap; 3332549c7297SChristian Brauner struct user_namespace *mnt_userns; 3333d58ffd35SMiklos Szeredi struct dentry *dir = nd->path.dentry; 333454ef4872SMiklos Szeredi struct inode *dir_inode = dir->d_inode; 33351643b43fSAl Viro int open_flag = op->open_flag; 3336d58ffd35SMiklos Szeredi struct dentry *dentry; 33371643b43fSAl Viro int error, create_error = 0; 33381643b43fSAl Viro umode_t mode = op->mode; 33396fbd0714SAl Viro DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq); 3340d58ffd35SMiklos Szeredi 3341ce8644fcSAl Viro if (unlikely(IS_DEADDIR(dir_inode))) 3342da5ebf5aSAl Viro return ERR_PTR(-ENOENT); 3343d58ffd35SMiklos Szeredi 334473a09dd9SAl Viro file->f_mode &= ~FMODE_CREATED; 33456fbd0714SAl Viro dentry = d_lookup(dir, &nd->last); 33466fbd0714SAl Viro for (;;) { 33476fbd0714SAl Viro if (!dentry) { 33486fbd0714SAl Viro dentry = d_alloc_parallel(dir, &nd->last, &wq); 3349d58ffd35SMiklos Szeredi if (IS_ERR(dentry)) 3350da5ebf5aSAl Viro return dentry; 33516fbd0714SAl Viro } 33526fbd0714SAl Viro if (d_in_lookup(dentry)) 33536fbd0714SAl Viro break; 3354d58ffd35SMiklos Szeredi 33556fbd0714SAl Viro error = d_revalidate(dentry, nd->flags); 33566fbd0714SAl Viro if (likely(error > 0)) 33576fbd0714SAl Viro break; 33586fbd0714SAl Viro if (error) 33596fbd0714SAl Viro goto out_dput; 33606fbd0714SAl Viro d_invalidate(dentry); 33616fbd0714SAl Viro dput(dentry); 33626fbd0714SAl Viro dentry = NULL; 33636fbd0714SAl Viro } 33646fbd0714SAl Viro if (dentry->d_inode) { 3365d18e9008SMiklos Szeredi /* Cached positive dentry: will open in f_op->open */ 3366da5ebf5aSAl Viro return dentry; 33676c51e513SAl Viro } 3368d18e9008SMiklos Szeredi 33691643b43fSAl Viro /* 33701643b43fSAl Viro * Checking write permission is tricky, bacuse we don't know if we are 33711643b43fSAl Viro * going to actually need it: O_CREAT opens should work as long as the 33721643b43fSAl Viro * file exists. But checking existence breaks atomicity. The trick is 33731643b43fSAl Viro * to check access and if not granted clear O_CREAT from the flags. 33741643b43fSAl Viro * 33751643b43fSAl Viro * Another problem is returing the "right" error value (e.g. for an 33761643b43fSAl Viro * O_EXCL open we want to return EEXIST not EROFS). 33771643b43fSAl Viro */ 337899a4a90cSAl Viro if (unlikely(!got_write)) 337999a4a90cSAl Viro open_flag &= ~O_TRUNC; 33806c960e68SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 33816c960e68SChristian Brauner mnt_userns = mnt_idmap_owner(idmap); 33821643b43fSAl Viro if (open_flag & O_CREAT) { 338399a4a90cSAl Viro if (open_flag & O_EXCL) 338499a4a90cSAl Viro open_flag &= ~O_TRUNC; 33851639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir->d_inode, mode, mode, mode); 338699a4a90cSAl Viro if (likely(got_write)) 33874609e1f1SChristian Brauner create_error = may_o_create(idmap, &nd->path, 3388ba73d987SChristian Brauner dentry, mode); 338999a4a90cSAl Viro else 339099a4a90cSAl Viro create_error = -EROFS; 339199a4a90cSAl Viro } 339299a4a90cSAl Viro if (create_error) 33931643b43fSAl Viro open_flag &= ~O_CREAT; 33941643b43fSAl Viro if (dir_inode->i_op->atomic_open) { 3395d489cf9aSAl Viro dentry = atomic_open(nd, dentry, file, open_flag, mode); 3396da5ebf5aSAl Viro if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT)) 3397da5ebf5aSAl Viro dentry = ERR_PTR(create_error); 3398da5ebf5aSAl Viro return dentry; 3399239eb983SAl Viro } 340054ef4872SMiklos Szeredi 34016fbd0714SAl Viro if (d_in_lookup(dentry)) { 340212fa5e24SAl Viro struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry, 340312fa5e24SAl Viro nd->flags); 34046fbd0714SAl Viro d_lookup_done(dentry); 340512fa5e24SAl Viro if (unlikely(res)) { 340612fa5e24SAl Viro if (IS_ERR(res)) { 340712fa5e24SAl Viro error = PTR_ERR(res); 340812fa5e24SAl Viro goto out_dput; 340912fa5e24SAl Viro } 341012fa5e24SAl Viro dput(dentry); 341112fa5e24SAl Viro dentry = res; 341212fa5e24SAl Viro } 341354ef4872SMiklos Szeredi } 341454ef4872SMiklos Szeredi 3415d58ffd35SMiklos Szeredi /* Negative dentry, just create the file */ 34161643b43fSAl Viro if (!dentry->d_inode && (open_flag & O_CREAT)) { 341773a09dd9SAl Viro file->f_mode |= FMODE_CREATED; 3418ce8644fcSAl Viro audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE); 3419ce8644fcSAl Viro if (!dir_inode->i_op->create) { 3420ce8644fcSAl Viro error = -EACCES; 3421d58ffd35SMiklos Szeredi goto out_dput; 342264894cf8SAl Viro } 3423549c7297SChristian Brauner 34246c960e68SChristian Brauner error = dir_inode->i_op->create(idmap, dir_inode, dentry, 3425549c7297SChristian Brauner mode, open_flag & O_EXCL); 3426d58ffd35SMiklos Szeredi if (error) 3427d58ffd35SMiklos Szeredi goto out_dput; 3428d58ffd35SMiklos Szeredi } 34291643b43fSAl Viro if (unlikely(create_error) && !dentry->d_inode) { 34301643b43fSAl Viro error = create_error; 3431d58ffd35SMiklos Szeredi goto out_dput; 3432d58ffd35SMiklos Szeredi } 3433da5ebf5aSAl Viro return dentry; 3434d58ffd35SMiklos Szeredi 3435d58ffd35SMiklos Szeredi out_dput: 3436d58ffd35SMiklos Szeredi dput(dentry); 3437da5ebf5aSAl Viro return ERR_PTR(error); 3438d58ffd35SMiklos Szeredi } 3439d58ffd35SMiklos Szeredi 3440c981a482SAl Viro static const char *open_last_lookups(struct nameidata *nd, 34413ec2eef1SAl Viro struct file *file, const struct open_flags *op) 3442fb1cc555SAl Viro { 3443a1e28038SAl Viro struct dentry *dir = nd->path.dentry; 3444ca344a89SAl Viro int open_flag = op->open_flag; 344564894cf8SAl Viro bool got_write = false; 3446da5ebf5aSAl Viro struct dentry *dentry; 3447b0417d2cSAl Viro const char *res; 3448fb1cc555SAl Viro 3449c3e380b0SAl Viro nd->flags |= op->intent; 3450c3e380b0SAl Viro 3451bc77daa7SAl Viro if (nd->last_type != LAST_NORM) { 345256676ec3SAl Viro if (nd->depth) 345356676ec3SAl Viro put_link(nd); 3454ff326a32SAl Viro return handle_dots(nd, nd->last_type); 34551f36f774SAl Viro } 3456a2c36b45SAl Viro 3457ca344a89SAl Viro if (!(open_flag & O_CREAT)) { 3458fe2d35ffSAl Viro if (nd->last.name[nd->last.len]) 3459fe2d35ffSAl Viro nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY; 3460fe2d35ffSAl Viro /* we _can_ be in RCU mode here */ 34614cb64024SAl Viro dentry = lookup_fast(nd); 346220e34357SAl Viro if (IS_ERR(dentry)) 34631ccac622SAl Viro return ERR_CAST(dentry); 346420e34357SAl Viro if (likely(dentry)) 346571574865SMiklos Szeredi goto finish_lookup; 346671574865SMiklos Szeredi 34676583fe22SAl Viro BUG_ON(nd->flags & LOOKUP_RCU); 3468b6183df7SMiklos Szeredi } else { 3469fe2d35ffSAl Viro /* create side of things */ 347072287417SAl Viro if (nd->flags & LOOKUP_RCU) { 3471e36cffedSJens Axboe if (!try_to_unlazy(nd)) 3472e36cffedSJens Axboe return ERR_PTR(-ECHILD); 347372287417SAl Viro } 3474c9b07eabSAl Viro audit_inode(nd->name, dir, AUDIT_INODE_PARENT); 34751f36f774SAl Viro /* trailing slashes? */ 3476deb106c6SAl Viro if (unlikely(nd->last.name[nd->last.len])) 34771ccac622SAl Viro return ERR_PTR(-EISDIR); 3478b6183df7SMiklos Szeredi } 34791f36f774SAl Viro 34809cf843e3SAl Viro if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) { 3481e36cffedSJens Axboe got_write = !mnt_want_write(nd->path.mnt); 348264894cf8SAl Viro /* 348364894cf8SAl Viro * do _not_ fail yet - we might not need that or fail with 348464894cf8SAl Viro * a different error; let lookup_open() decide; we'll be 348564894cf8SAl Viro * dropping this one anyway. 348664894cf8SAl Viro */ 348764894cf8SAl Viro } 34889cf843e3SAl Viro if (open_flag & O_CREAT) 34895955102cSAl Viro inode_lock(dir->d_inode); 34909cf843e3SAl Viro else 34919cf843e3SAl Viro inode_lock_shared(dir->d_inode); 3492da5ebf5aSAl Viro dentry = lookup_open(nd, file, op, got_write); 3493f7bb959dSAl Viro if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED)) 3494f7bb959dSAl Viro fsnotify_create(dir->d_inode, dentry); 34959cf843e3SAl Viro if (open_flag & O_CREAT) 34965955102cSAl Viro inode_unlock(dir->d_inode); 34979cf843e3SAl Viro else 34989cf843e3SAl Viro inode_unlock_shared(dir->d_inode); 3499fb1cc555SAl Viro 3500c981a482SAl Viro if (got_write) 350159e96e65SAl Viro mnt_drop_write(nd->path.mnt); 35026c0d46c4SAl Viro 350359e96e65SAl Viro if (IS_ERR(dentry)) 350459e96e65SAl Viro return ERR_CAST(dentry); 350559e96e65SAl Viro 3506973d4b73SAl Viro if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) { 3507e73cabffSAl Viro dput(nd->path.dentry); 3508e73cabffSAl Viro nd->path.dentry = dentry; 3509c981a482SAl Viro return NULL; 3510fb1cc555SAl Viro } 3511fb1cc555SAl Viro 351220e34357SAl Viro finish_lookup: 351356676ec3SAl Viro if (nd->depth) 351456676ec3SAl Viro put_link(nd); 3515a4f5b521SAl Viro res = step_into(nd, WALK_TRAILING, dentry); 3516ff326a32SAl Viro if (unlikely(res)) 35171ccac622SAl Viro nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL); 3518b0417d2cSAl Viro return res; 35191ccac622SAl Viro } 352031d1726dSAl Viro 3521c981a482SAl Viro /* 3522c981a482SAl Viro * Handle the last step of open() 3523c981a482SAl Viro */ 3524c5971b8cSAl Viro static int do_open(struct nameidata *nd, 3525c981a482SAl Viro struct file *file, const struct open_flags *op) 3526c981a482SAl Viro { 3527abf08576SChristian Brauner struct mnt_idmap *idmap; 3528549c7297SChristian Brauner struct user_namespace *mnt_userns; 3529c981a482SAl Viro int open_flag = op->open_flag; 3530c981a482SAl Viro bool do_truncate; 3531c981a482SAl Viro int acc_mode; 3532c981a482SAl Viro int error; 3533c981a482SAl Viro 3534ff326a32SAl Viro if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) { 3535ff326a32SAl Viro error = complete_walk(nd); 3536ff326a32SAl Viro if (error) 3537ff326a32SAl Viro return error; 3538ff326a32SAl Viro } 3539973d4b73SAl Viro if (!(file->f_mode & FMODE_CREATED)) 354076ae2a5aSAl Viro audit_inode(nd->name, nd->path.dentry, 0); 3541abf08576SChristian Brauner idmap = mnt_idmap(nd->path.mnt); 3542abf08576SChristian Brauner mnt_userns = mnt_idmap_owner(idmap); 354330aba665SSalvatore Mesoraca if (open_flag & O_CREAT) { 3544b94e0b32SAl Viro if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED)) 3545b94e0b32SAl Viro return -EEXIST; 354630aba665SSalvatore Mesoraca if (d_is_dir(nd->path.dentry)) 3547c5971b8cSAl Viro return -EISDIR; 3548549c7297SChristian Brauner error = may_create_in_sticky(mnt_userns, nd, 354930aba665SSalvatore Mesoraca d_backing_inode(nd->path.dentry)); 355030aba665SSalvatore Mesoraca if (unlikely(error)) 3551c5971b8cSAl Viro return error; 355230aba665SSalvatore Mesoraca } 355344b1d530SMiklos Szeredi if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry)) 3554c5971b8cSAl Viro return -ENOTDIR; 35556c0d46c4SAl Viro 35568795e7d4SAl Viro do_truncate = false; 35578795e7d4SAl Viro acc_mode = op->acc_mode; 35585a2d3eddSAl Viro if (file->f_mode & FMODE_CREATED) { 35595a2d3eddSAl Viro /* Don't check for write permission, don't truncate */ 35605a2d3eddSAl Viro open_flag &= ~O_TRUNC; 35615a2d3eddSAl Viro acc_mode = 0; 35628795e7d4SAl Viro } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) { 35630f9d1a10SAl Viro error = mnt_want_write(nd->path.mnt); 35640f9d1a10SAl Viro if (error) 3565c5971b8cSAl Viro return error; 35668795e7d4SAl Viro do_truncate = true; 35670f9d1a10SAl Viro } 35684609e1f1SChristian Brauner error = may_open(idmap, &nd->path, acc_mode, open_flag); 35698795e7d4SAl Viro if (!error && !(file->f_mode & FMODE_OPENED)) 3570ae2bb293SAl Viro error = vfs_open(&nd->path, file); 35718795e7d4SAl Viro if (!error) 35726035a27bSAl Viro error = ima_file_check(file, op->acc_mode); 35738795e7d4SAl Viro if (!error && do_truncate) 3574abf08576SChristian Brauner error = handle_truncate(idmap, file); 3575c80567c8SAl Viro if (unlikely(error > 0)) { 3576c80567c8SAl Viro WARN_ON(1); 3577c80567c8SAl Viro error = -EINVAL; 3578c80567c8SAl Viro } 35798795e7d4SAl Viro if (do_truncate) 35800f9d1a10SAl Viro mnt_drop_write(nd->path.mnt); 3581c5971b8cSAl Viro return error; 3582fb1cc555SAl Viro } 3583fb1cc555SAl Viro 35846521f891SChristian Brauner /** 35856521f891SChristian Brauner * vfs_tmpfile - create tmpfile 3586abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 35876521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 35886521f891SChristian Brauner * @mode: mode of the new tmpfile 35892111c3c0SRandy Dunlap * @open_flag: flags 35906521f891SChristian Brauner * 35916521f891SChristian Brauner * Create a temporary file. 35926521f891SChristian Brauner * 3593abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3594abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3595abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 35966521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3597abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 35986521f891SChristian Brauner */ 3599abf08576SChristian Brauner static int vfs_tmpfile(struct mnt_idmap *idmap, 36009751b338SMiklos Szeredi const struct path *parentpath, 36019751b338SMiklos Szeredi struct file *file, umode_t mode) 3602af7bd4dcSAmir Goldstein { 3603abf08576SChristian Brauner struct user_namespace *mnt_userns = mnt_idmap_owner(idmap); 36049751b338SMiklos Szeredi struct dentry *child; 36059751b338SMiklos Szeredi struct inode *dir = d_inode(parentpath->dentry); 3606af7bd4dcSAmir Goldstein struct inode *inode; 3607af7bd4dcSAmir Goldstein int error; 3608406c706cSPeter Griffin int open_flag = file->f_flags; 3609af7bd4dcSAmir Goldstein 3610af7bd4dcSAmir Goldstein /* we want directory to be writable */ 36114609e1f1SChristian Brauner error = inode_permission(idmap, dir, MAY_WRITE | MAY_EXEC); 3612af7bd4dcSAmir Goldstein if (error) 36139751b338SMiklos Szeredi return error; 3614af7bd4dcSAmir Goldstein if (!dir->i_op->tmpfile) 36159751b338SMiklos Szeredi return -EOPNOTSUPP; 36169751b338SMiklos Szeredi child = d_alloc(parentpath->dentry, &slash_name); 3617af7bd4dcSAmir Goldstein if (unlikely(!child)) 36189751b338SMiklos Szeredi return -ENOMEM; 36199751b338SMiklos Szeredi file->f_path.mnt = parentpath->mnt; 36209751b338SMiklos Szeredi file->f_path.dentry = child; 36211639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode); 3622011e2b71SChristian Brauner error = dir->i_op->tmpfile(idmap, dir, file, mode); 36239751b338SMiklos Szeredi dput(child); 3624af7bd4dcSAmir Goldstein if (error) 36259751b338SMiklos Szeredi return error; 36269751b338SMiklos Szeredi /* Don't check for other permissions, the inode was just created */ 36274609e1f1SChristian Brauner error = may_open(idmap, &file->f_path, 0, file->f_flags); 36289751b338SMiklos Szeredi if (error) 36299751b338SMiklos Szeredi return error; 36309751b338SMiklos Szeredi inode = file_inode(file); 3631406c706cSPeter Griffin if (!(open_flag & O_EXCL)) { 3632af7bd4dcSAmir Goldstein spin_lock(&inode->i_lock); 3633af7bd4dcSAmir Goldstein inode->i_state |= I_LINKABLE; 3634af7bd4dcSAmir Goldstein spin_unlock(&inode->i_lock); 3635af7bd4dcSAmir Goldstein } 363639f60c1cSChristian Brauner ima_post_create_tmpfile(idmap, inode); 36379751b338SMiklos Szeredi return 0; 3638af7bd4dcSAmir Goldstein } 3639af7bd4dcSAmir Goldstein 364022873deaSMiklos Szeredi /** 364122873deaSMiklos Szeredi * vfs_tmpfile_open - open a tmpfile for kernel internal use 3642abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 364322873deaSMiklos Szeredi * @parentpath: path of the base directory 364422873deaSMiklos Szeredi * @mode: mode of the new tmpfile 364522873deaSMiklos Szeredi * @open_flag: flags 364622873deaSMiklos Szeredi * @cred: credentials for open 364722873deaSMiklos Szeredi * 364822873deaSMiklos Szeredi * Create and open a temporary file. The file is not accounted in nr_files, 364922873deaSMiklos Szeredi * hence this is only for kernel internal use, and must not be installed into 365022873deaSMiklos Szeredi * file tables or such. 365122873deaSMiklos Szeredi */ 3652abf08576SChristian Brauner struct file *vfs_tmpfile_open(struct mnt_idmap *idmap, 365322873deaSMiklos Szeredi const struct path *parentpath, 365422873deaSMiklos Szeredi umode_t mode, int open_flag, const struct cred *cred) 365522873deaSMiklos Szeredi { 365622873deaSMiklos Szeredi struct file *file; 365722873deaSMiklos Szeredi int error; 365822873deaSMiklos Szeredi 36599751b338SMiklos Szeredi file = alloc_empty_file_noaccount(open_flag, cred); 36609751b338SMiklos Szeredi if (!IS_ERR(file)) { 3661abf08576SChristian Brauner error = vfs_tmpfile(idmap, parentpath, file, mode); 36629751b338SMiklos Szeredi if (error) { 36639751b338SMiklos Szeredi fput(file); 366422873deaSMiklos Szeredi file = ERR_PTR(error); 36659751b338SMiklos Szeredi } 36669751b338SMiklos Szeredi } 366722873deaSMiklos Szeredi return file; 366822873deaSMiklos Szeredi } 366922873deaSMiklos Szeredi EXPORT_SYMBOL(vfs_tmpfile_open); 3670648fa861SAl Viro 3671c8a53ee5SAl Viro static int do_tmpfile(struct nameidata *nd, unsigned flags, 367260545d0dSAl Viro const struct open_flags *op, 36733ec2eef1SAl Viro struct file *file) 367460545d0dSAl Viro { 3675625b6d10SAl Viro struct path path; 3676c8a53ee5SAl Viro int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path); 36779751b338SMiklos Szeredi 367860545d0dSAl Viro if (unlikely(error)) 367960545d0dSAl Viro return error; 3680625b6d10SAl Viro error = mnt_want_write(path.mnt); 368160545d0dSAl Viro if (unlikely(error)) 368260545d0dSAl Viro goto out; 3683abf08576SChristian Brauner error = vfs_tmpfile(mnt_idmap(path.mnt), &path, file, op->mode); 36849751b338SMiklos Szeredi if (error) 368560545d0dSAl Viro goto out2; 36869751b338SMiklos Szeredi audit_inode(nd->name, file->f_path.dentry, 0); 368760545d0dSAl Viro out2: 3688625b6d10SAl Viro mnt_drop_write(path.mnt); 368960545d0dSAl Viro out: 3690625b6d10SAl Viro path_put(&path); 369160545d0dSAl Viro return error; 369260545d0dSAl Viro } 369360545d0dSAl Viro 36946ac08709SAl Viro static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file) 36956ac08709SAl Viro { 36966ac08709SAl Viro struct path path; 36976ac08709SAl Viro int error = path_lookupat(nd, flags, &path); 36986ac08709SAl Viro if (!error) { 36996ac08709SAl Viro audit_inode(nd->name, path.dentry, 0); 3700ae2bb293SAl Viro error = vfs_open(&path, file); 37016ac08709SAl Viro path_put(&path); 37026ac08709SAl Viro } 37036ac08709SAl Viro return error; 37046ac08709SAl Viro } 37056ac08709SAl Viro 3706c8a53ee5SAl Viro static struct file *path_openat(struct nameidata *nd, 3707c8a53ee5SAl Viro const struct open_flags *op, unsigned flags) 37081da177e4SLinus Torvalds { 370930d90494SAl Viro struct file *file; 371013aab428SAl Viro int error; 371131e6b01fSNick Piggin 3712ea73ea72SAl Viro file = alloc_empty_file(op->open_flag, current_cred()); 37131afc99beSAl Viro if (IS_ERR(file)) 37141afc99beSAl Viro return file; 371531e6b01fSNick Piggin 3716bb458c64SAl Viro if (unlikely(file->f_flags & __O_TMPFILE)) { 37173ec2eef1SAl Viro error = do_tmpfile(nd, flags, op, file); 37185f336e72SAl Viro } else if (unlikely(file->f_flags & O_PATH)) { 37196ac08709SAl Viro error = do_o_path(nd, flags, file); 37205f336e72SAl Viro } else { 37215f336e72SAl Viro const char *s = path_init(nd, flags); 37223bdba28bSAl Viro while (!(error = link_path_walk(s, nd)) && 3723c5971b8cSAl Viro (s = open_last_lookups(nd, file, op)) != NULL) 37241ccac622SAl Viro ; 3725c5971b8cSAl Viro if (!error) 3726c5971b8cSAl Viro error = do_open(nd, file, op); 3727deb106c6SAl Viro terminate_walk(nd); 37285f336e72SAl Viro } 37297c1c01ecSAl Viro if (likely(!error)) { 3730aad888f8SAl Viro if (likely(file->f_mode & FMODE_OPENED)) 37317c1c01ecSAl Viro return file; 37327c1c01ecSAl Viro WARN_ON(1); 37337c1c01ecSAl Viro error = -EINVAL; 3734015c3bbcSMiklos Szeredi } 37357c1c01ecSAl Viro fput(file); 37362675a4ebSAl Viro if (error == -EOPENSTALE) { 37372675a4ebSAl Viro if (flags & LOOKUP_RCU) 37382675a4ebSAl Viro error = -ECHILD; 37392675a4ebSAl Viro else 37402675a4ebSAl Viro error = -ESTALE; 37412675a4ebSAl Viro } 37427c1c01ecSAl Viro return ERR_PTR(error); 3743de459215SKirill Korotaev } 37441da177e4SLinus Torvalds 3745669abf4eSJeff Layton struct file *do_filp_open(int dfd, struct filename *pathname, 3746f9652e10SAl Viro const struct open_flags *op) 374713aab428SAl Viro { 37489883d185SAl Viro struct nameidata nd; 3749f9652e10SAl Viro int flags = op->lookup_flags; 375013aab428SAl Viro struct file *filp; 375113aab428SAl Viro 375206422964SAl Viro set_nameidata(&nd, dfd, pathname, NULL); 3753c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_RCU); 375413aab428SAl Viro if (unlikely(filp == ERR_PTR(-ECHILD))) 3755c8a53ee5SAl Viro filp = path_openat(&nd, op, flags); 375613aab428SAl Viro if (unlikely(filp == ERR_PTR(-ESTALE))) 3757c8a53ee5SAl Viro filp = path_openat(&nd, op, flags | LOOKUP_REVAL); 37589883d185SAl Viro restore_nameidata(); 375913aab428SAl Viro return filp; 376013aab428SAl Viro } 376113aab428SAl Viro 3762ffb37ca3SAl Viro struct file *do_file_open_root(const struct path *root, 3763f9652e10SAl Viro const char *name, const struct open_flags *op) 376473d049a4SAl Viro { 37659883d185SAl Viro struct nameidata nd; 376673d049a4SAl Viro struct file *file; 376751689104SPaul Moore struct filename *filename; 3768bcba1e7dSAl Viro int flags = op->lookup_flags; 376973d049a4SAl Viro 3770ffb37ca3SAl Viro if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN) 377173d049a4SAl Viro return ERR_PTR(-ELOOP); 377273d049a4SAl Viro 377351689104SPaul Moore filename = getname_kernel(name); 3774a1c83681SViresh Kumar if (IS_ERR(filename)) 377551689104SPaul Moore return ERR_CAST(filename); 377651689104SPaul Moore 377706422964SAl Viro set_nameidata(&nd, -1, filename, root); 3778c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_RCU); 377973d049a4SAl Viro if (unlikely(file == ERR_PTR(-ECHILD))) 3780c8a53ee5SAl Viro file = path_openat(&nd, op, flags); 378173d049a4SAl Viro if (unlikely(file == ERR_PTR(-ESTALE))) 3782c8a53ee5SAl Viro file = path_openat(&nd, op, flags | LOOKUP_REVAL); 37839883d185SAl Viro restore_nameidata(); 378451689104SPaul Moore putname(filename); 378573d049a4SAl Viro return file; 378673d049a4SAl Viro } 378773d049a4SAl Viro 3788b4a4f213SStephen Brennan static struct dentry *filename_create(int dfd, struct filename *name, 37891ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 37901da177e4SLinus Torvalds { 3791c663e5d8SChristoph Hellwig struct dentry *dentry = ERR_PTR(-EEXIST); 3792391172c4SAl Viro struct qstr last; 3793b3d4650dSNeilBrown bool want_dir = lookup_flags & LOOKUP_DIRECTORY; 3794b3d4650dSNeilBrown unsigned int reval_flag = lookup_flags & LOOKUP_REVAL; 3795b3d4650dSNeilBrown unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL; 3796391172c4SAl Viro int type; 3797c30dabfeSJan Kara int err2; 37981ac12b4bSJeff Layton int error; 37991ac12b4bSJeff Layton 3800b3d4650dSNeilBrown error = filename_parentat(dfd, name, reval_flag, path, &last, &type); 38010ee50b47SDmitry Kadashev if (error) 38020ee50b47SDmitry Kadashev return ERR_PTR(error); 38031da177e4SLinus Torvalds 3804c663e5d8SChristoph Hellwig /* 3805c663e5d8SChristoph Hellwig * Yucky last component or no last component at all? 3806c663e5d8SChristoph Hellwig * (foo/., foo/.., /////) 3807c663e5d8SChristoph Hellwig */ 38085c31b6ceSAl Viro if (unlikely(type != LAST_NORM)) 3809ed75e95dSAl Viro goto out; 3810c663e5d8SChristoph Hellwig 3811c30dabfeSJan Kara /* don't fail immediately if it's r/o, at least try to report other errors */ 3812391172c4SAl Viro err2 = mnt_want_write(path->mnt); 3813c663e5d8SChristoph Hellwig /* 3814b3d4650dSNeilBrown * Do the final lookup. Suppress 'create' if there is a trailing 3815b3d4650dSNeilBrown * '/', and a directory wasn't requested. 3816c663e5d8SChristoph Hellwig */ 3817b3d4650dSNeilBrown if (last.name[last.len] && !want_dir) 3818b3d4650dSNeilBrown create_flags = 0; 38195955102cSAl Viro inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT); 3820b3d4650dSNeilBrown dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags); 38211da177e4SLinus Torvalds if (IS_ERR(dentry)) 3822a8104a9fSAl Viro goto unlock; 3823c663e5d8SChristoph Hellwig 3824a8104a9fSAl Viro error = -EEXIST; 3825b18825a7SDavid Howells if (d_is_positive(dentry)) 3826a8104a9fSAl Viro goto fail; 3827b18825a7SDavid Howells 3828c663e5d8SChristoph Hellwig /* 3829c663e5d8SChristoph Hellwig * Special case - lookup gave negative, but... we had foo/bar/ 3830c663e5d8SChristoph Hellwig * From the vfs_mknod() POV we just have a negative dentry - 3831c663e5d8SChristoph Hellwig * all is fine. Let's be bastards - you had / on the end, you've 3832c663e5d8SChristoph Hellwig * been asking for (non-existent) directory. -ENOENT for you. 3833c663e5d8SChristoph Hellwig */ 3834b3d4650dSNeilBrown if (unlikely(!create_flags)) { 3835a8104a9fSAl Viro error = -ENOENT; 3836ed75e95dSAl Viro goto fail; 3837e9baf6e5SAl Viro } 3838c30dabfeSJan Kara if (unlikely(err2)) { 3839c30dabfeSJan Kara error = err2; 3840a8104a9fSAl Viro goto fail; 3841c30dabfeSJan Kara } 3842e9baf6e5SAl Viro return dentry; 38431da177e4SLinus Torvalds fail: 3844a8104a9fSAl Viro dput(dentry); 3845a8104a9fSAl Viro dentry = ERR_PTR(error); 3846a8104a9fSAl Viro unlock: 38475955102cSAl Viro inode_unlock(path->dentry->d_inode); 3848c30dabfeSJan Kara if (!err2) 3849391172c4SAl Viro mnt_drop_write(path->mnt); 3850ed75e95dSAl Viro out: 3851391172c4SAl Viro path_put(path); 3852ed75e95dSAl Viro return dentry; 3853dae6ad8fSAl Viro } 3854fa14a0b8SAl Viro 3855fa14a0b8SAl Viro struct dentry *kern_path_create(int dfd, const char *pathname, 3856fa14a0b8SAl Viro struct path *path, unsigned int lookup_flags) 3857fa14a0b8SAl Viro { 3858b4a4f213SStephen Brennan struct filename *filename = getname_kernel(pathname); 3859b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3860b4a4f213SStephen Brennan 3861b4a4f213SStephen Brennan putname(filename); 3862b4a4f213SStephen Brennan return res; 3863fa14a0b8SAl Viro } 3864dae6ad8fSAl Viro EXPORT_SYMBOL(kern_path_create); 3865dae6ad8fSAl Viro 3866921a1650SAl Viro void done_path_create(struct path *path, struct dentry *dentry) 3867921a1650SAl Viro { 3868921a1650SAl Viro dput(dentry); 38695955102cSAl Viro inode_unlock(path->dentry->d_inode); 3870a8104a9fSAl Viro mnt_drop_write(path->mnt); 3871921a1650SAl Viro path_put(path); 3872921a1650SAl Viro } 3873921a1650SAl Viro EXPORT_SYMBOL(done_path_create); 3874921a1650SAl Viro 3875520ae687SAl Viro inline struct dentry *user_path_create(int dfd, const char __user *pathname, 38761ac12b4bSJeff Layton struct path *path, unsigned int lookup_flags) 3877dae6ad8fSAl Viro { 3878b4a4f213SStephen Brennan struct filename *filename = getname(pathname); 3879b4a4f213SStephen Brennan struct dentry *res = filename_create(dfd, filename, path, lookup_flags); 3880b4a4f213SStephen Brennan 3881b4a4f213SStephen Brennan putname(filename); 3882b4a4f213SStephen Brennan return res; 3883dae6ad8fSAl Viro } 3884dae6ad8fSAl Viro EXPORT_SYMBOL(user_path_create); 3885dae6ad8fSAl Viro 38866521f891SChristian Brauner /** 38876521f891SChristian Brauner * vfs_mknod - create device node or file 3888abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 38896521f891SChristian Brauner * @dir: inode of @dentry 38906521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 38916521f891SChristian Brauner * @mode: mode of the new device node or file 38926521f891SChristian Brauner * @dev: device number of device to create 38936521f891SChristian Brauner * 38946521f891SChristian Brauner * Create a device node or file. 38956521f891SChristian Brauner * 3896abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 3897abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 3898abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 38996521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 3900abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 39016521f891SChristian Brauner */ 3902abf08576SChristian Brauner int vfs_mknod(struct mnt_idmap *idmap, struct inode *dir, 39036521f891SChristian Brauner struct dentry *dentry, umode_t mode, dev_t dev) 39041da177e4SLinus Torvalds { 3905abf08576SChristian Brauner struct user_namespace *mnt_userns = mnt_idmap_owner(idmap); 3906a3c751a5SMiklos Szeredi bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV; 39074609e1f1SChristian Brauner int error = may_create(idmap, dir, dentry); 39081da177e4SLinus Torvalds 39091da177e4SLinus Torvalds if (error) 39101da177e4SLinus Torvalds return error; 39111da177e4SLinus Torvalds 3912a3c751a5SMiklos Szeredi if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout && 3913a3c751a5SMiklos Szeredi !capable(CAP_MKNOD)) 39141da177e4SLinus Torvalds return -EPERM; 39151da177e4SLinus Torvalds 3916acfa4380SAl Viro if (!dir->i_op->mknod) 39171da177e4SLinus Torvalds return -EPERM; 39181da177e4SLinus Torvalds 39191639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, mode, mode); 392008ce5f16SSerge E. Hallyn error = devcgroup_inode_mknod(mode, dev); 392108ce5f16SSerge E. Hallyn if (error) 392208ce5f16SSerge E. Hallyn return error; 392308ce5f16SSerge E. Hallyn 39241da177e4SLinus Torvalds error = security_inode_mknod(dir, dentry, mode, dev); 39251da177e4SLinus Torvalds if (error) 39261da177e4SLinus Torvalds return error; 39271da177e4SLinus Torvalds 39285ebb29beSChristian Brauner error = dir->i_op->mknod(idmap, dir, dentry, mode, dev); 3929a74574aaSStephen Smalley if (!error) 3930f38aa942SAmy Griffis fsnotify_create(dir, dentry); 39311da177e4SLinus Torvalds return error; 39321da177e4SLinus Torvalds } 39334d359507SAl Viro EXPORT_SYMBOL(vfs_mknod); 39341da177e4SLinus Torvalds 3935f69aac00SAl Viro static int may_mknod(umode_t mode) 3936463c3197SDave Hansen { 3937463c3197SDave Hansen switch (mode & S_IFMT) { 3938463c3197SDave Hansen case S_IFREG: 3939463c3197SDave Hansen case S_IFCHR: 3940463c3197SDave Hansen case S_IFBLK: 3941463c3197SDave Hansen case S_IFIFO: 3942463c3197SDave Hansen case S_IFSOCK: 3943463c3197SDave Hansen case 0: /* zero mode translates to S_IFREG */ 3944463c3197SDave Hansen return 0; 3945463c3197SDave Hansen case S_IFDIR: 3946463c3197SDave Hansen return -EPERM; 3947463c3197SDave Hansen default: 3948463c3197SDave Hansen return -EINVAL; 3949463c3197SDave Hansen } 3950463c3197SDave Hansen } 3951463c3197SDave Hansen 395245f30dabSDmitry Kadashev static int do_mknodat(int dfd, struct filename *name, umode_t mode, 395387c4e192SDominik Brodowski unsigned int dev) 39541da177e4SLinus Torvalds { 3955abf08576SChristian Brauner struct mnt_idmap *idmap; 39561da177e4SLinus Torvalds struct dentry *dentry; 3957dae6ad8fSAl Viro struct path path; 3958dae6ad8fSAl Viro int error; 3959972567f1SJeff Layton unsigned int lookup_flags = 0; 39601da177e4SLinus Torvalds 39618e4bfca1SAl Viro error = may_mknod(mode); 39628e4bfca1SAl Viro if (error) 39637797251bSDmitry Kadashev goto out1; 3964972567f1SJeff Layton retry: 3965b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 39667797251bSDmitry Kadashev error = PTR_ERR(dentry); 3967dae6ad8fSAl Viro if (IS_ERR(dentry)) 39687797251bSDmitry Kadashev goto out1; 39692ad94ae6SAl Viro 39701639a49cSYang Xu error = security_path_mknod(&path, dentry, 39711639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode), dev); 3972be6d3e56SKentaro Takeda if (error) 39737797251bSDmitry Kadashev goto out2; 39746521f891SChristian Brauner 3975abf08576SChristian Brauner idmap = mnt_idmap(path.mnt); 39761da177e4SLinus Torvalds switch (mode & S_IFMT) { 39771da177e4SLinus Torvalds case 0: case S_IFREG: 3978abf08576SChristian Brauner error = vfs_create(idmap, path.dentry->d_inode, 39796521f891SChristian Brauner dentry, mode, true); 398005d1a717SMimi Zohar if (!error) 398139f60c1cSChristian Brauner ima_post_path_mknod(idmap, dentry); 39821da177e4SLinus Torvalds break; 39831da177e4SLinus Torvalds case S_IFCHR: case S_IFBLK: 3984abf08576SChristian Brauner error = vfs_mknod(idmap, path.dentry->d_inode, 39856521f891SChristian Brauner dentry, mode, new_decode_dev(dev)); 39861da177e4SLinus Torvalds break; 39871da177e4SLinus Torvalds case S_IFIFO: case S_IFSOCK: 3988abf08576SChristian Brauner error = vfs_mknod(idmap, path.dentry->d_inode, 39896521f891SChristian Brauner dentry, mode, 0); 39901da177e4SLinus Torvalds break; 39911da177e4SLinus Torvalds } 39927797251bSDmitry Kadashev out2: 3993921a1650SAl Viro done_path_create(&path, dentry); 3994972567f1SJeff Layton if (retry_estale(error, lookup_flags)) { 3995972567f1SJeff Layton lookup_flags |= LOOKUP_REVAL; 3996972567f1SJeff Layton goto retry; 3997972567f1SJeff Layton } 39987797251bSDmitry Kadashev out1: 39997797251bSDmitry Kadashev putname(name); 40001da177e4SLinus Torvalds return error; 40011da177e4SLinus Torvalds } 40021da177e4SLinus Torvalds 400387c4e192SDominik Brodowski SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode, 400487c4e192SDominik Brodowski unsigned int, dev) 400587c4e192SDominik Brodowski { 40067797251bSDmitry Kadashev return do_mknodat(dfd, getname(filename), mode, dev); 400787c4e192SDominik Brodowski } 400887c4e192SDominik Brodowski 40098208a22bSAl Viro SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev) 40105590ff0dSUlrich Drepper { 40117797251bSDmitry Kadashev return do_mknodat(AT_FDCWD, getname(filename), mode, dev); 40125590ff0dSUlrich Drepper } 40135590ff0dSUlrich Drepper 40146521f891SChristian Brauner /** 40156521f891SChristian Brauner * vfs_mkdir - create directory 4016abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 40176521f891SChristian Brauner * @dir: inode of @dentry 40186521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 40196521f891SChristian Brauner * @mode: mode of the new directory 40206521f891SChristian Brauner * 40216521f891SChristian Brauner * Create a directory. 40226521f891SChristian Brauner * 4023abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4024abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4025abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 40266521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4027abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 40286521f891SChristian Brauner */ 4029abf08576SChristian Brauner int vfs_mkdir(struct mnt_idmap *idmap, struct inode *dir, 40306521f891SChristian Brauner struct dentry *dentry, umode_t mode) 40311da177e4SLinus Torvalds { 4032abf08576SChristian Brauner struct user_namespace *mnt_userns = mnt_idmap_owner(idmap); 4033abf08576SChristian Brauner int error; 40348de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 40351da177e4SLinus Torvalds 40364609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 40371da177e4SLinus Torvalds if (error) 40381da177e4SLinus Torvalds return error; 40391da177e4SLinus Torvalds 4040acfa4380SAl Viro if (!dir->i_op->mkdir) 40411da177e4SLinus Torvalds return -EPERM; 40421da177e4SLinus Torvalds 40431639a49cSYang Xu mode = vfs_prepare_mode(mnt_userns, dir, mode, S_IRWXUGO | S_ISVTX, 0); 40441da177e4SLinus Torvalds error = security_inode_mkdir(dir, dentry, mode); 40451da177e4SLinus Torvalds if (error) 40461da177e4SLinus Torvalds return error; 40471da177e4SLinus Torvalds 40488de52778SAl Viro if (max_links && dir->i_nlink >= max_links) 40498de52778SAl Viro return -EMLINK; 40508de52778SAl Viro 4051c54bd91eSChristian Brauner error = dir->i_op->mkdir(idmap, dir, dentry, mode); 4052a74574aaSStephen Smalley if (!error) 4053f38aa942SAmy Griffis fsnotify_mkdir(dir, dentry); 40541da177e4SLinus Torvalds return error; 40551da177e4SLinus Torvalds } 40564d359507SAl Viro EXPORT_SYMBOL(vfs_mkdir); 40571da177e4SLinus Torvalds 405845f30dabSDmitry Kadashev int do_mkdirat(int dfd, struct filename *name, umode_t mode) 40591da177e4SLinus Torvalds { 40606902d925SDave Hansen struct dentry *dentry; 4061dae6ad8fSAl Viro struct path path; 4062dae6ad8fSAl Viro int error; 4063b76d8b82SJeff Layton unsigned int lookup_flags = LOOKUP_DIRECTORY; 40641da177e4SLinus Torvalds 4065b76d8b82SJeff Layton retry: 4066b4a4f213SStephen Brennan dentry = filename_create(dfd, name, &path, lookup_flags); 4067584d3226SDmitry Kadashev error = PTR_ERR(dentry); 40686902d925SDave Hansen if (IS_ERR(dentry)) 4069584d3226SDmitry Kadashev goto out_putname; 40706902d925SDave Hansen 40711639a49cSYang Xu error = security_path_mkdir(&path, dentry, 40721639a49cSYang Xu mode_strip_umask(path.dentry->d_inode, mode)); 40736521f891SChristian Brauner if (!error) { 4074abf08576SChristian Brauner error = vfs_mkdir(mnt_idmap(path.mnt), path.dentry->d_inode, 4075abf08576SChristian Brauner dentry, mode); 40766521f891SChristian Brauner } 4077921a1650SAl Viro done_path_create(&path, dentry); 4078b76d8b82SJeff Layton if (retry_estale(error, lookup_flags)) { 4079b76d8b82SJeff Layton lookup_flags |= LOOKUP_REVAL; 4080b76d8b82SJeff Layton goto retry; 4081b76d8b82SJeff Layton } 4082584d3226SDmitry Kadashev out_putname: 4083584d3226SDmitry Kadashev putname(name); 40841da177e4SLinus Torvalds return error; 40851da177e4SLinus Torvalds } 40861da177e4SLinus Torvalds 40870101db7aSDominik Brodowski SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode) 40880101db7aSDominik Brodowski { 4089584d3226SDmitry Kadashev return do_mkdirat(dfd, getname(pathname), mode); 40900101db7aSDominik Brodowski } 40910101db7aSDominik Brodowski 4092a218d0fdSAl Viro SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode) 40935590ff0dSUlrich Drepper { 4094584d3226SDmitry Kadashev return do_mkdirat(AT_FDCWD, getname(pathname), mode); 40955590ff0dSUlrich Drepper } 40965590ff0dSUlrich Drepper 40976521f891SChristian Brauner /** 40986521f891SChristian Brauner * vfs_rmdir - remove directory 4099abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 41006521f891SChristian Brauner * @dir: inode of @dentry 41016521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 41026521f891SChristian Brauner * 41036521f891SChristian Brauner * Remove a directory. 41046521f891SChristian Brauner * 4105abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4106abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4107abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 41086521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4109abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 41106521f891SChristian Brauner */ 4111abf08576SChristian Brauner int vfs_rmdir(struct mnt_idmap *idmap, struct inode *dir, 41126521f891SChristian Brauner struct dentry *dentry) 41131da177e4SLinus Torvalds { 41144609e1f1SChristian Brauner int error = may_delete(idmap, dir, dentry, 1); 41151da177e4SLinus Torvalds 41161da177e4SLinus Torvalds if (error) 41171da177e4SLinus Torvalds return error; 41181da177e4SLinus Torvalds 4119acfa4380SAl Viro if (!dir->i_op->rmdir) 41201da177e4SLinus Torvalds return -EPERM; 41211da177e4SLinus Torvalds 41221d2ef590SAl Viro dget(dentry); 41235955102cSAl Viro inode_lock(dentry->d_inode); 4124912dbc15SSage Weil 41251da177e4SLinus Torvalds error = -EBUSY; 41261bd9c4e4SDavid Howells if (is_local_mountpoint(dentry) || 41271bd9c4e4SDavid Howells (dentry->d_inode->i_flags & S_KERNEL_FILE)) 4128912dbc15SSage Weil goto out; 4129912dbc15SSage Weil 41301da177e4SLinus Torvalds error = security_inode_rmdir(dir, dentry); 4131912dbc15SSage Weil if (error) 4132912dbc15SSage Weil goto out; 4133912dbc15SSage Weil 41341da177e4SLinus Torvalds error = dir->i_op->rmdir(dir, dentry); 4135912dbc15SSage Weil if (error) 4136912dbc15SSage Weil goto out; 4137912dbc15SSage Weil 41388767712fSAl Viro shrink_dcache_parent(dentry); 41391da177e4SLinus Torvalds dentry->d_inode->i_flags |= S_DEAD; 4140d83c49f3SAl Viro dont_mount(dentry); 41418ed936b5SEric W. Biederman detach_mounts(dentry); 41421da177e4SLinus Torvalds 4143912dbc15SSage Weil out: 41445955102cSAl Viro inode_unlock(dentry->d_inode); 41451d2ef590SAl Viro dput(dentry); 4146912dbc15SSage Weil if (!error) 4147a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 41481da177e4SLinus Torvalds return error; 41491da177e4SLinus Torvalds } 41504d359507SAl Viro EXPORT_SYMBOL(vfs_rmdir); 41511da177e4SLinus Torvalds 415245f30dabSDmitry Kadashev int do_rmdir(int dfd, struct filename *name) 41531da177e4SLinus Torvalds { 41540ee50b47SDmitry Kadashev int error; 41551da177e4SLinus Torvalds struct dentry *dentry; 4156f5beed75SAl Viro struct path path; 4157f5beed75SAl Viro struct qstr last; 4158f5beed75SAl Viro int type; 4159c6ee9206SJeff Layton unsigned int lookup_flags = 0; 4160c6ee9206SJeff Layton retry: 4161c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 41620ee50b47SDmitry Kadashev if (error) 41630ee50b47SDmitry Kadashev goto exit1; 41641da177e4SLinus Torvalds 4165f5beed75SAl Viro switch (type) { 41661da177e4SLinus Torvalds case LAST_DOTDOT: 41671da177e4SLinus Torvalds error = -ENOTEMPTY; 41680ee50b47SDmitry Kadashev goto exit2; 41691da177e4SLinus Torvalds case LAST_DOT: 41701da177e4SLinus Torvalds error = -EINVAL; 41710ee50b47SDmitry Kadashev goto exit2; 41721da177e4SLinus Torvalds case LAST_ROOT: 41731da177e4SLinus Torvalds error = -EBUSY; 41740ee50b47SDmitry Kadashev goto exit2; 41751da177e4SLinus Torvalds } 41760612d9fbSOGAWA Hirofumi 4177f5beed75SAl Viro error = mnt_want_write(path.mnt); 4178c30dabfeSJan Kara if (error) 41790ee50b47SDmitry Kadashev goto exit2; 41800612d9fbSOGAWA Hirofumi 41815955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4182f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 41831da177e4SLinus Torvalds error = PTR_ERR(dentry); 41846902d925SDave Hansen if (IS_ERR(dentry)) 41850ee50b47SDmitry Kadashev goto exit3; 4186e6bc45d6STheodore Ts'o if (!dentry->d_inode) { 4187e6bc45d6STheodore Ts'o error = -ENOENT; 41880ee50b47SDmitry Kadashev goto exit4; 4189e6bc45d6STheodore Ts'o } 4190f5beed75SAl Viro error = security_path_rmdir(&path, dentry); 4191be6d3e56SKentaro Takeda if (error) 41920ee50b47SDmitry Kadashev goto exit4; 4193abf08576SChristian Brauner error = vfs_rmdir(mnt_idmap(path.mnt), path.dentry->d_inode, dentry); 41940ee50b47SDmitry Kadashev exit4: 41951da177e4SLinus Torvalds dput(dentry); 41960ee50b47SDmitry Kadashev exit3: 41975955102cSAl Viro inode_unlock(path.dentry->d_inode); 4198f5beed75SAl Viro mnt_drop_write(path.mnt); 41990ee50b47SDmitry Kadashev exit2: 4200f5beed75SAl Viro path_put(&path); 4201c6ee9206SJeff Layton if (retry_estale(error, lookup_flags)) { 4202c6ee9206SJeff Layton lookup_flags |= LOOKUP_REVAL; 4203c6ee9206SJeff Layton goto retry; 4204c6ee9206SJeff Layton } 42050ee50b47SDmitry Kadashev exit1: 420624fb33d4SAl Viro putname(name); 42071da177e4SLinus Torvalds return error; 42081da177e4SLinus Torvalds } 42091da177e4SLinus Torvalds 42103cdad428SHeiko Carstens SYSCALL_DEFINE1(rmdir, const char __user *, pathname) 42115590ff0dSUlrich Drepper { 4212e24ab0efSChristoph Hellwig return do_rmdir(AT_FDCWD, getname(pathname)); 42135590ff0dSUlrich Drepper } 42145590ff0dSUlrich Drepper 4215b21996e3SJ. Bruce Fields /** 4216b21996e3SJ. Bruce Fields * vfs_unlink - unlink a filesystem object 4217abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 4218b21996e3SJ. Bruce Fields * @dir: parent directory 4219b21996e3SJ. Bruce Fields * @dentry: victim 4220b21996e3SJ. Bruce Fields * @delegated_inode: returns victim inode, if the inode is delegated. 4221b21996e3SJ. Bruce Fields * 4222b21996e3SJ. Bruce Fields * The caller must hold dir->i_mutex. 4223b21996e3SJ. Bruce Fields * 4224b21996e3SJ. Bruce Fields * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and 4225b21996e3SJ. Bruce Fields * return a reference to the inode in delegated_inode. The caller 4226b21996e3SJ. Bruce Fields * should then break the delegation on that inode and retry. Because 4227b21996e3SJ. Bruce Fields * breaking a delegation may take a long time, the caller should drop 4228b21996e3SJ. Bruce Fields * dir->i_mutex before doing so. 4229b21996e3SJ. Bruce Fields * 4230b21996e3SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4231b21996e3SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4232b21996e3SJ. Bruce Fields * to be NFS exported. 42336521f891SChristian Brauner * 4234abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4235abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4236abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 42376521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4238abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 4239b21996e3SJ. Bruce Fields */ 4240abf08576SChristian Brauner int vfs_unlink(struct mnt_idmap *idmap, struct inode *dir, 42416521f891SChristian Brauner struct dentry *dentry, struct inode **delegated_inode) 42421da177e4SLinus Torvalds { 42439accbb97SJ. Bruce Fields struct inode *target = dentry->d_inode; 42444609e1f1SChristian Brauner int error = may_delete(idmap, dir, dentry, 0); 42451da177e4SLinus Torvalds 42461da177e4SLinus Torvalds if (error) 42471da177e4SLinus Torvalds return error; 42481da177e4SLinus Torvalds 4249acfa4380SAl Viro if (!dir->i_op->unlink) 42501da177e4SLinus Torvalds return -EPERM; 42511da177e4SLinus Torvalds 42525955102cSAl Viro inode_lock(target); 425351cc3a66SHugh Dickins if (IS_SWAPFILE(target)) 425451cc3a66SHugh Dickins error = -EPERM; 425551cc3a66SHugh Dickins else if (is_local_mountpoint(dentry)) 42561da177e4SLinus Torvalds error = -EBUSY; 42571da177e4SLinus Torvalds else { 42581da177e4SLinus Torvalds error = security_inode_unlink(dir, dentry); 4259bec1052eSAl Viro if (!error) { 42605a14696cSJ. Bruce Fields error = try_break_deleg(target, delegated_inode); 42615a14696cSJ. Bruce Fields if (error) 4262b21996e3SJ. Bruce Fields goto out; 42631da177e4SLinus Torvalds error = dir->i_op->unlink(dir, dentry); 42648ed936b5SEric W. Biederman if (!error) { 4265d83c49f3SAl Viro dont_mount(dentry); 42668ed936b5SEric W. Biederman detach_mounts(dentry); 42678ed936b5SEric W. Biederman } 4268bec1052eSAl Viro } 42691da177e4SLinus Torvalds } 4270b21996e3SJ. Bruce Fields out: 42715955102cSAl Viro inode_unlock(target); 42721da177e4SLinus Torvalds 42731da177e4SLinus Torvalds /* We don't d_delete() NFS sillyrenamed files--they still exist. */ 4274a37d9a17SAmir Goldstein if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) { 4275a37d9a17SAmir Goldstein fsnotify_unlink(dir, dentry); 4276a37d9a17SAmir Goldstein } else if (!error) { 42779accbb97SJ. Bruce Fields fsnotify_link_count(target); 4278a37d9a17SAmir Goldstein d_delete_notify(dir, dentry); 42791da177e4SLinus Torvalds } 42800eeca283SRobert Love 42811da177e4SLinus Torvalds return error; 42821da177e4SLinus Torvalds } 42834d359507SAl Viro EXPORT_SYMBOL(vfs_unlink); 42841da177e4SLinus Torvalds 42851da177e4SLinus Torvalds /* 42861da177e4SLinus Torvalds * Make sure that the actual truncation of the file will occur outside its 42871b1dcc1bSJes Sorensen * directory's i_mutex. Truncate can take a long time if there is a lot of 42881da177e4SLinus Torvalds * writeout happening, and we don't want to prevent access to the directory 42891da177e4SLinus Torvalds * while waiting on the I/O. 42901da177e4SLinus Torvalds */ 429145f30dabSDmitry Kadashev int do_unlinkat(int dfd, struct filename *name) 42921da177e4SLinus Torvalds { 42932ad94ae6SAl Viro int error; 42941da177e4SLinus Torvalds struct dentry *dentry; 4295f5beed75SAl Viro struct path path; 4296f5beed75SAl Viro struct qstr last; 4297f5beed75SAl Viro int type; 42981da177e4SLinus Torvalds struct inode *inode = NULL; 4299b21996e3SJ. Bruce Fields struct inode *delegated_inode = NULL; 43005d18f813SJeff Layton unsigned int lookup_flags = 0; 43015d18f813SJeff Layton retry: 4302c5f563f9SAl Viro error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type); 43030ee50b47SDmitry Kadashev if (error) 43040ee50b47SDmitry Kadashev goto exit1; 43052ad94ae6SAl Viro 43061da177e4SLinus Torvalds error = -EISDIR; 4307f5beed75SAl Viro if (type != LAST_NORM) 43080ee50b47SDmitry Kadashev goto exit2; 43090612d9fbSOGAWA Hirofumi 4310f5beed75SAl Viro error = mnt_want_write(path.mnt); 4311c30dabfeSJan Kara if (error) 43120ee50b47SDmitry Kadashev goto exit2; 4313b21996e3SJ. Bruce Fields retry_deleg: 43145955102cSAl Viro inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT); 4315f5beed75SAl Viro dentry = __lookup_hash(&last, path.dentry, lookup_flags); 43161da177e4SLinus Torvalds error = PTR_ERR(dentry); 43171da177e4SLinus Torvalds if (!IS_ERR(dentry)) { 43186521f891SChristian Brauner 43191da177e4SLinus Torvalds /* Why not before? Because we want correct error value */ 4320f5beed75SAl Viro if (last.name[last.len]) 432150338b88STörök Edwin goto slashes; 43221da177e4SLinus Torvalds inode = dentry->d_inode; 4323b18825a7SDavid Howells if (d_is_negative(dentry)) 4324e6bc45d6STheodore Ts'o goto slashes; 43257de9c6eeSAl Viro ihold(inode); 4326f5beed75SAl Viro error = security_path_unlink(&path, dentry); 4327be6d3e56SKentaro Takeda if (error) 43280ee50b47SDmitry Kadashev goto exit3; 4329abf08576SChristian Brauner error = vfs_unlink(mnt_idmap(path.mnt), path.dentry->d_inode, 4330abf08576SChristian Brauner dentry, &delegated_inode); 43310ee50b47SDmitry Kadashev exit3: 43321da177e4SLinus Torvalds dput(dentry); 43331da177e4SLinus Torvalds } 43345955102cSAl Viro inode_unlock(path.dentry->d_inode); 43351da177e4SLinus Torvalds if (inode) 43361da177e4SLinus Torvalds iput(inode); /* truncate the inode here */ 4337b21996e3SJ. Bruce Fields inode = NULL; 4338b21996e3SJ. Bruce Fields if (delegated_inode) { 43395a14696cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4340b21996e3SJ. Bruce Fields if (!error) 4341b21996e3SJ. Bruce Fields goto retry_deleg; 4342b21996e3SJ. Bruce Fields } 4343f5beed75SAl Viro mnt_drop_write(path.mnt); 43440ee50b47SDmitry Kadashev exit2: 4345f5beed75SAl Viro path_put(&path); 43465d18f813SJeff Layton if (retry_estale(error, lookup_flags)) { 43475d18f813SJeff Layton lookup_flags |= LOOKUP_REVAL; 43485d18f813SJeff Layton inode = NULL; 43495d18f813SJeff Layton goto retry; 43505d18f813SJeff Layton } 43510ee50b47SDmitry Kadashev exit1: 4352da2f1362SChristoph Hellwig putname(name); 43531da177e4SLinus Torvalds return error; 43541da177e4SLinus Torvalds 43551da177e4SLinus Torvalds slashes: 4356b18825a7SDavid Howells if (d_is_negative(dentry)) 4357b18825a7SDavid Howells error = -ENOENT; 435844b1d530SMiklos Szeredi else if (d_is_dir(dentry)) 4359b18825a7SDavid Howells error = -EISDIR; 4360b18825a7SDavid Howells else 4361b18825a7SDavid Howells error = -ENOTDIR; 43620ee50b47SDmitry Kadashev goto exit3; 43631da177e4SLinus Torvalds } 43641da177e4SLinus Torvalds 43652e4d0924SHeiko Carstens SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag) 43665590ff0dSUlrich Drepper { 43675590ff0dSUlrich Drepper if ((flag & ~AT_REMOVEDIR) != 0) 43685590ff0dSUlrich Drepper return -EINVAL; 43695590ff0dSUlrich Drepper 43705590ff0dSUlrich Drepper if (flag & AT_REMOVEDIR) 4371e24ab0efSChristoph Hellwig return do_rmdir(dfd, getname(pathname)); 4372da2f1362SChristoph Hellwig return do_unlinkat(dfd, getname(pathname)); 43735590ff0dSUlrich Drepper } 43745590ff0dSUlrich Drepper 43753480b257SHeiko Carstens SYSCALL_DEFINE1(unlink, const char __user *, pathname) 43765590ff0dSUlrich Drepper { 4377da2f1362SChristoph Hellwig return do_unlinkat(AT_FDCWD, getname(pathname)); 43785590ff0dSUlrich Drepper } 43795590ff0dSUlrich Drepper 43806521f891SChristian Brauner /** 43816521f891SChristian Brauner * vfs_symlink - create symlink 4382abf08576SChristian Brauner * @idmap: idmap of the mount the inode was found from 43836521f891SChristian Brauner * @dir: inode of @dentry 43846521f891SChristian Brauner * @dentry: pointer to dentry of the base directory 43856521f891SChristian Brauner * @oldname: name of the file to link to 43866521f891SChristian Brauner * 43876521f891SChristian Brauner * Create a symlink. 43886521f891SChristian Brauner * 4389abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4390abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4391abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 43926521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4393abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 43946521f891SChristian Brauner */ 4395abf08576SChristian Brauner int vfs_symlink(struct mnt_idmap *idmap, struct inode *dir, 43966521f891SChristian Brauner struct dentry *dentry, const char *oldname) 43971da177e4SLinus Torvalds { 43987a77db95SChristian Brauner int error; 43991da177e4SLinus Torvalds 44004609e1f1SChristian Brauner error = may_create(idmap, dir, dentry); 44011da177e4SLinus Torvalds if (error) 44021da177e4SLinus Torvalds return error; 44031da177e4SLinus Torvalds 4404acfa4380SAl Viro if (!dir->i_op->symlink) 44051da177e4SLinus Torvalds return -EPERM; 44061da177e4SLinus Torvalds 44071da177e4SLinus Torvalds error = security_inode_symlink(dir, dentry, oldname); 44081da177e4SLinus Torvalds if (error) 44091da177e4SLinus Torvalds return error; 44101da177e4SLinus Torvalds 44117a77db95SChristian Brauner error = dir->i_op->symlink(idmap, dir, dentry, oldname); 4412a74574aaSStephen Smalley if (!error) 4413f38aa942SAmy Griffis fsnotify_create(dir, dentry); 44141da177e4SLinus Torvalds return error; 44151da177e4SLinus Torvalds } 44164d359507SAl Viro EXPORT_SYMBOL(vfs_symlink); 44171da177e4SLinus Torvalds 44187a8721f8SDmitry Kadashev int do_symlinkat(struct filename *from, int newdfd, struct filename *to) 44191da177e4SLinus Torvalds { 44202ad94ae6SAl Viro int error; 44216902d925SDave Hansen struct dentry *dentry; 4422dae6ad8fSAl Viro struct path path; 4423f46d3567SJeff Layton unsigned int lookup_flags = 0; 44241da177e4SLinus Torvalds 4425da2d0cedSDmitry Kadashev if (IS_ERR(from)) { 4426da2d0cedSDmitry Kadashev error = PTR_ERR(from); 4427da2d0cedSDmitry Kadashev goto out_putnames; 4428da2d0cedSDmitry Kadashev } 4429f46d3567SJeff Layton retry: 4430b4a4f213SStephen Brennan dentry = filename_create(newdfd, to, &path, lookup_flags); 44311da177e4SLinus Torvalds error = PTR_ERR(dentry); 44326902d925SDave Hansen if (IS_ERR(dentry)) 4433da2d0cedSDmitry Kadashev goto out_putnames; 44346902d925SDave Hansen 443591a27b2aSJeff Layton error = security_path_symlink(&path, dentry, from->name); 4436abf08576SChristian Brauner if (!error) 4437abf08576SChristian Brauner error = vfs_symlink(mnt_idmap(path.mnt), path.dentry->d_inode, 4438abf08576SChristian Brauner dentry, from->name); 4439921a1650SAl Viro done_path_create(&path, dentry); 4440f46d3567SJeff Layton if (retry_estale(error, lookup_flags)) { 4441f46d3567SJeff Layton lookup_flags |= LOOKUP_REVAL; 4442f46d3567SJeff Layton goto retry; 4443f46d3567SJeff Layton } 4444da2d0cedSDmitry Kadashev out_putnames: 4445da2d0cedSDmitry Kadashev putname(to); 44461da177e4SLinus Torvalds putname(from); 44471da177e4SLinus Torvalds return error; 44481da177e4SLinus Torvalds } 44491da177e4SLinus Torvalds 4450b724e846SDominik Brodowski SYSCALL_DEFINE3(symlinkat, const char __user *, oldname, 4451b724e846SDominik Brodowski int, newdfd, const char __user *, newname) 4452b724e846SDominik Brodowski { 4453da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), newdfd, getname(newname)); 4454b724e846SDominik Brodowski } 4455b724e846SDominik Brodowski 44563480b257SHeiko Carstens SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname) 44575590ff0dSUlrich Drepper { 4458da2d0cedSDmitry Kadashev return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname)); 44595590ff0dSUlrich Drepper } 44605590ff0dSUlrich Drepper 4461146a8595SJ. Bruce Fields /** 4462146a8595SJ. Bruce Fields * vfs_link - create a new link 4463146a8595SJ. Bruce Fields * @old_dentry: object to be linked 4464abf08576SChristian Brauner * @idmap: idmap of the mount 4465146a8595SJ. Bruce Fields * @dir: new parent 4466146a8595SJ. Bruce Fields * @new_dentry: where to create the new link 4467146a8595SJ. Bruce Fields * @delegated_inode: returns inode needing a delegation break 4468146a8595SJ. Bruce Fields * 4469146a8595SJ. Bruce Fields * The caller must hold dir->i_mutex 4470146a8595SJ. Bruce Fields * 4471146a8595SJ. Bruce Fields * If vfs_link discovers a delegation on the to-be-linked file in need 4472146a8595SJ. Bruce Fields * of breaking, it will return -EWOULDBLOCK and return a reference to the 4473146a8595SJ. Bruce Fields * inode in delegated_inode. The caller should then break the delegation 4474146a8595SJ. Bruce Fields * and retry. Because breaking a delegation may take a long time, the 4475146a8595SJ. Bruce Fields * caller should drop the i_mutex before doing so. 4476146a8595SJ. Bruce Fields * 4477146a8595SJ. Bruce Fields * Alternatively, a caller may pass NULL for delegated_inode. This may 4478146a8595SJ. Bruce Fields * be appropriate for callers that expect the underlying filesystem not 4479146a8595SJ. Bruce Fields * to be NFS exported. 44806521f891SChristian Brauner * 4481abf08576SChristian Brauner * If the inode has been found through an idmapped mount the idmap of 4482abf08576SChristian Brauner * the vfsmount must be passed through @idmap. This function will then take 4483abf08576SChristian Brauner * care to map the inode according to @idmap before checking permissions. 44846521f891SChristian Brauner * On non-idmapped mounts or if permission checking is to be performed on the 4485abf08576SChristian Brauner * raw inode simply passs @nop_mnt_idmap. 4486146a8595SJ. Bruce Fields */ 4487abf08576SChristian Brauner int vfs_link(struct dentry *old_dentry, struct mnt_idmap *idmap, 44886521f891SChristian Brauner struct inode *dir, struct dentry *new_dentry, 44896521f891SChristian Brauner struct inode **delegated_inode) 44901da177e4SLinus Torvalds { 44911da177e4SLinus Torvalds struct inode *inode = old_dentry->d_inode; 44928de52778SAl Viro unsigned max_links = dir->i_sb->s_max_links; 44931da177e4SLinus Torvalds int error; 44941da177e4SLinus Torvalds 44951da177e4SLinus Torvalds if (!inode) 44961da177e4SLinus Torvalds return -ENOENT; 44971da177e4SLinus Torvalds 44984609e1f1SChristian Brauner error = may_create(idmap, dir, new_dentry); 44991da177e4SLinus Torvalds if (error) 45001da177e4SLinus Torvalds return error; 45011da177e4SLinus Torvalds 45021da177e4SLinus Torvalds if (dir->i_sb != inode->i_sb) 45031da177e4SLinus Torvalds return -EXDEV; 45041da177e4SLinus Torvalds 45051da177e4SLinus Torvalds /* 45061da177e4SLinus Torvalds * A link to an append-only or immutable file cannot be created. 45071da177e4SLinus Torvalds */ 45081da177e4SLinus Torvalds if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 45091da177e4SLinus Torvalds return -EPERM; 45100bd23d09SEric W. Biederman /* 45110bd23d09SEric W. Biederman * Updating the link count will likely cause i_uid and i_gid to 45120bd23d09SEric W. Biederman * be writen back improperly if their true value is unknown to 45130bd23d09SEric W. Biederman * the vfs. 45140bd23d09SEric W. Biederman */ 45154609e1f1SChristian Brauner if (HAS_UNMAPPED_ID(idmap, inode)) 45160bd23d09SEric W. Biederman return -EPERM; 4517acfa4380SAl Viro if (!dir->i_op->link) 45181da177e4SLinus Torvalds return -EPERM; 45197e79eedbSTetsuo Handa if (S_ISDIR(inode->i_mode)) 45201da177e4SLinus Torvalds return -EPERM; 45211da177e4SLinus Torvalds 45221da177e4SLinus Torvalds error = security_inode_link(old_dentry, dir, new_dentry); 45231da177e4SLinus Torvalds if (error) 45241da177e4SLinus Torvalds return error; 45251da177e4SLinus Torvalds 45265955102cSAl Viro inode_lock(inode); 4527aae8a97dSAneesh Kumar K.V /* Make sure we don't allow creating hardlink to an unlinked file */ 4528f4e0c30cSAl Viro if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE)) 4529aae8a97dSAneesh Kumar K.V error = -ENOENT; 45308de52778SAl Viro else if (max_links && inode->i_nlink >= max_links) 45318de52778SAl Viro error = -EMLINK; 4532146a8595SJ. Bruce Fields else { 4533146a8595SJ. Bruce Fields error = try_break_deleg(inode, delegated_inode); 4534146a8595SJ. Bruce Fields if (!error) 45351da177e4SLinus Torvalds error = dir->i_op->link(old_dentry, dir, new_dentry); 4536146a8595SJ. Bruce Fields } 4537f4e0c30cSAl Viro 4538f4e0c30cSAl Viro if (!error && (inode->i_state & I_LINKABLE)) { 4539f4e0c30cSAl Viro spin_lock(&inode->i_lock); 4540f4e0c30cSAl Viro inode->i_state &= ~I_LINKABLE; 4541f4e0c30cSAl Viro spin_unlock(&inode->i_lock); 4542f4e0c30cSAl Viro } 45435955102cSAl Viro inode_unlock(inode); 4544e31e14ecSStephen Smalley if (!error) 45457e79eedbSTetsuo Handa fsnotify_link(dir, inode, new_dentry); 45461da177e4SLinus Torvalds return error; 45471da177e4SLinus Torvalds } 45484d359507SAl Viro EXPORT_SYMBOL(vfs_link); 45491da177e4SLinus Torvalds 45501da177e4SLinus Torvalds /* 45511da177e4SLinus Torvalds * Hardlinks are often used in delicate situations. We avoid 45521da177e4SLinus Torvalds * security-related surprises by not following symlinks on the 45531da177e4SLinus Torvalds * newname. --KAB 45541da177e4SLinus Torvalds * 45551da177e4SLinus Torvalds * We don't follow them on the oldname either to be compatible 45561da177e4SLinus Torvalds * with linux 2.0, and to avoid hard-linking to directories 45571da177e4SLinus Torvalds * and other special files. --ADM 45581da177e4SLinus Torvalds */ 4559cf30da90SDmitry Kadashev int do_linkat(int olddfd, struct filename *old, int newdfd, 4560020250f3SDmitry Kadashev struct filename *new, int flags) 45611da177e4SLinus Torvalds { 4562abf08576SChristian Brauner struct mnt_idmap *idmap; 45631da177e4SLinus Torvalds struct dentry *new_dentry; 4564dae6ad8fSAl Viro struct path old_path, new_path; 4565146a8595SJ. Bruce Fields struct inode *delegated_inode = NULL; 456611a7b371SAneesh Kumar K.V int how = 0; 45671da177e4SLinus Torvalds int error; 45681da177e4SLinus Torvalds 4569020250f3SDmitry Kadashev if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) { 4570020250f3SDmitry Kadashev error = -EINVAL; 4571020250f3SDmitry Kadashev goto out_putnames; 4572020250f3SDmitry Kadashev } 457311a7b371SAneesh Kumar K.V /* 4574f0cc6ffbSLinus Torvalds * To use null names we require CAP_DAC_READ_SEARCH 4575f0cc6ffbSLinus Torvalds * This ensures that not everyone will be able to create 4576f0cc6ffbSLinus Torvalds * handlink using the passed filedescriptor. 457711a7b371SAneesh Kumar K.V */ 4578020250f3SDmitry Kadashev if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) { 4579020250f3SDmitry Kadashev error = -ENOENT; 4580020250f3SDmitry Kadashev goto out_putnames; 4581f0cc6ffbSLinus Torvalds } 4582c04030e1SUlrich Drepper 458311a7b371SAneesh Kumar K.V if (flags & AT_SYMLINK_FOLLOW) 458411a7b371SAneesh Kumar K.V how |= LOOKUP_FOLLOW; 4585442e31caSJeff Layton retry: 4586794ebceaSStephen Brennan error = filename_lookup(olddfd, old, how, &old_path, NULL); 45871da177e4SLinus Torvalds if (error) 4588020250f3SDmitry Kadashev goto out_putnames; 45892ad94ae6SAl Viro 4590b4a4f213SStephen Brennan new_dentry = filename_create(newdfd, new, &new_path, 4591442e31caSJeff Layton (how & LOOKUP_REVAL)); 45921da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 45936902d925SDave Hansen if (IS_ERR(new_dentry)) 4594020250f3SDmitry Kadashev goto out_putpath; 4595dae6ad8fSAl Viro 4596dae6ad8fSAl Viro error = -EXDEV; 4597dae6ad8fSAl Viro if (old_path.mnt != new_path.mnt) 4598dae6ad8fSAl Viro goto out_dput; 4599abf08576SChristian Brauner idmap = mnt_idmap(new_path.mnt); 46004609e1f1SChristian Brauner error = may_linkat(idmap, &old_path); 4601800179c9SKees Cook if (unlikely(error)) 4602800179c9SKees Cook goto out_dput; 4603dae6ad8fSAl Viro error = security_path_link(old_path.dentry, &new_path, new_dentry); 4604be6d3e56SKentaro Takeda if (error) 4605a8104a9fSAl Viro goto out_dput; 4606abf08576SChristian Brauner error = vfs_link(old_path.dentry, idmap, new_path.dentry->d_inode, 46076521f891SChristian Brauner new_dentry, &delegated_inode); 460875c3f29dSDave Hansen out_dput: 4609921a1650SAl Viro done_path_create(&new_path, new_dentry); 4610146a8595SJ. Bruce Fields if (delegated_inode) { 4611146a8595SJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 4612d22e6338SOleg Drokin if (!error) { 4613d22e6338SOleg Drokin path_put(&old_path); 4614146a8595SJ. Bruce Fields goto retry; 4615146a8595SJ. Bruce Fields } 4616d22e6338SOleg Drokin } 4617442e31caSJeff Layton if (retry_estale(error, how)) { 4618d22e6338SOleg Drokin path_put(&old_path); 4619442e31caSJeff Layton how |= LOOKUP_REVAL; 4620442e31caSJeff Layton goto retry; 4621442e31caSJeff Layton } 4622020250f3SDmitry Kadashev out_putpath: 46232d8f3038SAl Viro path_put(&old_path); 4624020250f3SDmitry Kadashev out_putnames: 4625020250f3SDmitry Kadashev putname(old); 4626020250f3SDmitry Kadashev putname(new); 46271da177e4SLinus Torvalds 46281da177e4SLinus Torvalds return error; 46291da177e4SLinus Torvalds } 46301da177e4SLinus Torvalds 463146ea89ebSDominik Brodowski SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname, 463246ea89ebSDominik Brodowski int, newdfd, const char __user *, newname, int, flags) 463346ea89ebSDominik Brodowski { 4634020250f3SDmitry Kadashev return do_linkat(olddfd, getname_uflags(oldname, flags), 4635020250f3SDmitry Kadashev newdfd, getname(newname), flags); 463646ea89ebSDominik Brodowski } 463746ea89ebSDominik Brodowski 46383480b257SHeiko Carstens SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname) 46395590ff0dSUlrich Drepper { 4640020250f3SDmitry Kadashev return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0); 46415590ff0dSUlrich Drepper } 46425590ff0dSUlrich Drepper 4643bc27027aSMiklos Szeredi /** 4644bc27027aSMiklos Szeredi * vfs_rename - rename a filesystem object 46452111c3c0SRandy Dunlap * @rd: pointer to &struct renamedata info 4646bc27027aSMiklos Szeredi * 4647bc27027aSMiklos Szeredi * The caller must hold multiple mutexes--see lock_rename()). 4648bc27027aSMiklos Szeredi * 4649bc27027aSMiklos Szeredi * If vfs_rename discovers a delegation in need of breaking at either 4650bc27027aSMiklos Szeredi * the source or destination, it will return -EWOULDBLOCK and return a 4651bc27027aSMiklos Szeredi * reference to the inode in delegated_inode. The caller should then 4652bc27027aSMiklos Szeredi * break the delegation and retry. Because breaking a delegation may 4653bc27027aSMiklos Szeredi * take a long time, the caller should drop all locks before doing 4654bc27027aSMiklos Szeredi * so. 4655bc27027aSMiklos Szeredi * 4656bc27027aSMiklos Szeredi * Alternatively, a caller may pass NULL for delegated_inode. This may 4657bc27027aSMiklos Szeredi * be appropriate for callers that expect the underlying filesystem not 4658bc27027aSMiklos Szeredi * to be NFS exported. 4659bc27027aSMiklos Szeredi * 46601da177e4SLinus Torvalds * The worst of all namespace operations - renaming directory. "Perverted" 46611da177e4SLinus Torvalds * doesn't even start to describe it. Somebody in UCB had a heck of a trip... 46621da177e4SLinus Torvalds * Problems: 46630117d427SMauro Carvalho Chehab * 4664d03b29a2SJ. Bruce Fields * a) we can get into loop creation. 46651da177e4SLinus Torvalds * b) race potential - two innocent renames can create a loop together. 46661da177e4SLinus Torvalds * That's where 4.4 screws up. Current fix: serialization on 4667a11f3a05SArjan van de Ven * sb->s_vfs_rename_mutex. We might be more accurate, but that's another 46681da177e4SLinus Torvalds * story. 46696cedba89SJ. Bruce Fields * c) we have to lock _four_ objects - parents and victim (if it exists), 46706cedba89SJ. Bruce Fields * and source (if it is not a directory). 46711b1dcc1bSJes Sorensen * And that - after we got ->i_mutex on parents (until then we don't know 46721da177e4SLinus Torvalds * whether the target exists). Solution: try to be smart with locking 46731da177e4SLinus Torvalds * order for inodes. We rely on the fact that tree topology may change 4674a11f3a05SArjan van de Ven * only under ->s_vfs_rename_mutex _and_ that parent of the object we 46751da177e4SLinus Torvalds * move will be locked. Thus we can rank directories by the tree 46761da177e4SLinus Torvalds * (ancestors first) and rank all non-directories after them. 46771da177e4SLinus Torvalds * That works since everybody except rename does "lock parent, lookup, 4678a11f3a05SArjan van de Ven * lock child" and rename is under ->s_vfs_rename_mutex. 46791da177e4SLinus Torvalds * HOWEVER, it relies on the assumption that any object with ->lookup() 46801da177e4SLinus Torvalds * has no more than 1 dentry. If "hybrid" objects will ever appear, 46811da177e4SLinus Torvalds * we'd better make sure that there's no link(2) for them. 4682e4eaac06SSage Weil * d) conversion from fhandle to dentry may come in the wrong moment - when 46831b1dcc1bSJes Sorensen * we are removing the target. Solution: we will have to grab ->i_mutex 46841da177e4SLinus Torvalds * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on 4685c41b20e7SAdam Buchbinder * ->i_mutex on parents, which works but leads to some truly excessive 46861da177e4SLinus Torvalds * locking]. 46871da177e4SLinus Torvalds */ 46889fe61450SChristian Brauner int vfs_rename(struct renamedata *rd) 46891da177e4SLinus Torvalds { 46901da177e4SLinus Torvalds int error; 46919fe61450SChristian Brauner struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir; 46929fe61450SChristian Brauner struct dentry *old_dentry = rd->old_dentry; 46939fe61450SChristian Brauner struct dentry *new_dentry = rd->new_dentry; 46949fe61450SChristian Brauner struct inode **delegated_inode = rd->delegated_inode; 46959fe61450SChristian Brauner unsigned int flags = rd->flags; 4696bc27027aSMiklos Szeredi bool is_dir = d_is_dir(old_dentry); 4697bc27027aSMiklos Szeredi struct inode *source = old_dentry->d_inode; 4698bc27027aSMiklos Szeredi struct inode *target = new_dentry->d_inode; 4699da1ce067SMiklos Szeredi bool new_is_dir = false; 4700da1ce067SMiklos Szeredi unsigned max_links = new_dir->i_sb->s_max_links; 470149d31c2fSAl Viro struct name_snapshot old_name; 47021da177e4SLinus Torvalds 47038d3e2936SMiklos Szeredi if (source == target) 47041da177e4SLinus Torvalds return 0; 47051da177e4SLinus Torvalds 47064609e1f1SChristian Brauner error = may_delete(rd->old_mnt_idmap, old_dir, old_dentry, is_dir); 47071da177e4SLinus Torvalds if (error) 47081da177e4SLinus Torvalds return error; 47091da177e4SLinus Torvalds 4710da1ce067SMiklos Szeredi if (!target) { 47114609e1f1SChristian Brauner error = may_create(rd->new_mnt_idmap, new_dir, new_dentry); 4712da1ce067SMiklos Szeredi } else { 4713da1ce067SMiklos Szeredi new_is_dir = d_is_dir(new_dentry); 4714da1ce067SMiklos Szeredi 4715da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 47164609e1f1SChristian Brauner error = may_delete(rd->new_mnt_idmap, new_dir, 47176521f891SChristian Brauner new_dentry, is_dir); 4718da1ce067SMiklos Szeredi else 47194609e1f1SChristian Brauner error = may_delete(rd->new_mnt_idmap, new_dir, 47206521f891SChristian Brauner new_dentry, new_is_dir); 4721da1ce067SMiklos Szeredi } 47221da177e4SLinus Torvalds if (error) 47231da177e4SLinus Torvalds return error; 47241da177e4SLinus Torvalds 47252773bf00SMiklos Szeredi if (!old_dir->i_op->rename) 47261da177e4SLinus Torvalds return -EPERM; 47271da177e4SLinus Torvalds 4728bc27027aSMiklos Szeredi /* 4729bc27027aSMiklos Szeredi * If we are going to change the parent - check write permissions, 4730bc27027aSMiklos Szeredi * we'll need to flip '..'. 4731bc27027aSMiklos Szeredi */ 4732da1ce067SMiklos Szeredi if (new_dir != old_dir) { 4733da1ce067SMiklos Szeredi if (is_dir) { 47344609e1f1SChristian Brauner error = inode_permission(rd->old_mnt_idmap, source, 473547291baaSChristian Brauner MAY_WRITE); 4736bc27027aSMiklos Szeredi if (error) 4737bc27027aSMiklos Szeredi return error; 4738bc27027aSMiklos Szeredi } 4739da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && new_is_dir) { 47404609e1f1SChristian Brauner error = inode_permission(rd->new_mnt_idmap, target, 474147291baaSChristian Brauner MAY_WRITE); 4742da1ce067SMiklos Szeredi if (error) 4743da1ce067SMiklos Szeredi return error; 4744da1ce067SMiklos Szeredi } 4745da1ce067SMiklos Szeredi } 47460eeca283SRobert Love 47470b3974ebSMiklos Szeredi error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry, 47480b3974ebSMiklos Szeredi flags); 4749bc27027aSMiklos Szeredi if (error) 4750bc27027aSMiklos Szeredi return error; 4751bc27027aSMiklos Szeredi 475249d31c2fSAl Viro take_dentry_name_snapshot(&old_name, old_dentry); 4753bc27027aSMiklos Szeredi dget(new_dentry); 4754da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4755bc27027aSMiklos Szeredi lock_two_nondirectories(source, target); 4756bc27027aSMiklos Szeredi else if (target) 47575955102cSAl Viro inode_lock(target); 4758bc27027aSMiklos Szeredi 475951cc3a66SHugh Dickins error = -EPERM; 476051cc3a66SHugh Dickins if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target))) 476151cc3a66SHugh Dickins goto out; 476251cc3a66SHugh Dickins 4763bc27027aSMiklos Szeredi error = -EBUSY; 47647af1364fSEric W. Biederman if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry)) 4765bc27027aSMiklos Szeredi goto out; 4766bc27027aSMiklos Szeredi 4767da1ce067SMiklos Szeredi if (max_links && new_dir != old_dir) { 4768bc27027aSMiklos Szeredi error = -EMLINK; 4769da1ce067SMiklos Szeredi if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links) 4770bc27027aSMiklos Szeredi goto out; 4771da1ce067SMiklos Szeredi if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir && 4772da1ce067SMiklos Szeredi old_dir->i_nlink >= max_links) 4773da1ce067SMiklos Szeredi goto out; 4774da1ce067SMiklos Szeredi } 4775da1ce067SMiklos Szeredi if (!is_dir) { 4776bc27027aSMiklos Szeredi error = try_break_deleg(source, delegated_inode); 4777bc27027aSMiklos Szeredi if (error) 4778bc27027aSMiklos Szeredi goto out; 4779da1ce067SMiklos Szeredi } 4780da1ce067SMiklos Szeredi if (target && !new_is_dir) { 4781bc27027aSMiklos Szeredi error = try_break_deleg(target, delegated_inode); 4782bc27027aSMiklos Szeredi if (error) 4783bc27027aSMiklos Szeredi goto out; 4784bc27027aSMiklos Szeredi } 4785e18275aeSChristian Brauner error = old_dir->i_op->rename(rd->new_mnt_idmap, old_dir, old_dentry, 4786520c8b16SMiklos Szeredi new_dir, new_dentry, flags); 4787bc27027aSMiklos Szeredi if (error) 4788bc27027aSMiklos Szeredi goto out; 4789bc27027aSMiklos Szeredi 4790da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE) && target) { 47918767712fSAl Viro if (is_dir) { 47928767712fSAl Viro shrink_dcache_parent(new_dentry); 4793bc27027aSMiklos Szeredi target->i_flags |= S_DEAD; 47948767712fSAl Viro } 4795bc27027aSMiklos Szeredi dont_mount(new_dentry); 47968ed936b5SEric W. Biederman detach_mounts(new_dentry); 4797bc27027aSMiklos Szeredi } 4798da1ce067SMiklos Szeredi if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) { 4799da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 4800bc27027aSMiklos Szeredi d_move(old_dentry, new_dentry); 4801da1ce067SMiklos Szeredi else 4802da1ce067SMiklos Szeredi d_exchange(old_dentry, new_dentry); 4803da1ce067SMiklos Szeredi } 4804bc27027aSMiklos Szeredi out: 4805da1ce067SMiklos Szeredi if (!is_dir || (flags & RENAME_EXCHANGE)) 4806bc27027aSMiklos Szeredi unlock_two_nondirectories(source, target); 4807bc27027aSMiklos Szeredi else if (target) 48085955102cSAl Viro inode_unlock(target); 4809bc27027aSMiklos Szeredi dput(new_dentry); 4810da1ce067SMiklos Szeredi if (!error) { 4811f4ec3a3dSAl Viro fsnotify_move(old_dir, new_dir, &old_name.name, is_dir, 4812da1ce067SMiklos Szeredi !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry); 4813da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4814f4ec3a3dSAl Viro fsnotify_move(new_dir, old_dir, &old_dentry->d_name, 4815da1ce067SMiklos Szeredi new_is_dir, NULL, new_dentry); 4816da1ce067SMiklos Szeredi } 4817da1ce067SMiklos Szeredi } 481849d31c2fSAl Viro release_dentry_name_snapshot(&old_name); 48190eeca283SRobert Love 48201da177e4SLinus Torvalds return error; 48211da177e4SLinus Torvalds } 48224d359507SAl Viro EXPORT_SYMBOL(vfs_rename); 48231da177e4SLinus Torvalds 4824e886663cSJens Axboe int do_renameat2(int olddfd, struct filename *from, int newdfd, 4825e886663cSJens Axboe struct filename *to, unsigned int flags) 48261da177e4SLinus Torvalds { 48279fe61450SChristian Brauner struct renamedata rd; 48281da177e4SLinus Torvalds struct dentry *old_dentry, *new_dentry; 48291da177e4SLinus Torvalds struct dentry *trap; 4830f5beed75SAl Viro struct path old_path, new_path; 4831f5beed75SAl Viro struct qstr old_last, new_last; 4832f5beed75SAl Viro int old_type, new_type; 48338e6d782cSJ. Bruce Fields struct inode *delegated_inode = NULL; 4834f5beed75SAl Viro unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET; 4835c6a94284SJeff Layton bool should_retry = false; 4836e886663cSJens Axboe int error = -EINVAL; 4837520c8b16SMiklos Szeredi 48380d7a8555SMiklos Szeredi if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 48390ee50b47SDmitry Kadashev goto put_names; 4840da1ce067SMiklos Szeredi 48410d7a8555SMiklos Szeredi if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) && 48420d7a8555SMiklos Szeredi (flags & RENAME_EXCHANGE)) 48430ee50b47SDmitry Kadashev goto put_names; 4844520c8b16SMiklos Szeredi 4845f5beed75SAl Viro if (flags & RENAME_EXCHANGE) 4846f5beed75SAl Viro target_flags = 0; 4847f5beed75SAl Viro 4848c6a94284SJeff Layton retry: 4849c5f563f9SAl Viro error = filename_parentat(olddfd, from, lookup_flags, &old_path, 4850e886663cSJens Axboe &old_last, &old_type); 48510ee50b47SDmitry Kadashev if (error) 48520ee50b47SDmitry Kadashev goto put_names; 48531da177e4SLinus Torvalds 4854c5f563f9SAl Viro error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last, 4855e886663cSJens Axboe &new_type); 48560ee50b47SDmitry Kadashev if (error) 48571da177e4SLinus Torvalds goto exit1; 48581da177e4SLinus Torvalds 48591da177e4SLinus Torvalds error = -EXDEV; 4860f5beed75SAl Viro if (old_path.mnt != new_path.mnt) 48611da177e4SLinus Torvalds goto exit2; 48621da177e4SLinus Torvalds 48631da177e4SLinus Torvalds error = -EBUSY; 4864f5beed75SAl Viro if (old_type != LAST_NORM) 48651da177e4SLinus Torvalds goto exit2; 48661da177e4SLinus Torvalds 48670a7c3937SMiklos Szeredi if (flags & RENAME_NOREPLACE) 48680a7c3937SMiklos Szeredi error = -EEXIST; 4869f5beed75SAl Viro if (new_type != LAST_NORM) 48701da177e4SLinus Torvalds goto exit2; 48711da177e4SLinus Torvalds 4872f5beed75SAl Viro error = mnt_want_write(old_path.mnt); 4873c30dabfeSJan Kara if (error) 4874c30dabfeSJan Kara goto exit2; 4875c30dabfeSJan Kara 48768e6d782cSJ. Bruce Fields retry_deleg: 4877f5beed75SAl Viro trap = lock_rename(new_path.dentry, old_path.dentry); 48781da177e4SLinus Torvalds 4879f5beed75SAl Viro old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags); 48801da177e4SLinus Torvalds error = PTR_ERR(old_dentry); 48811da177e4SLinus Torvalds if (IS_ERR(old_dentry)) 48821da177e4SLinus Torvalds goto exit3; 48831da177e4SLinus Torvalds /* source must exist */ 48841da177e4SLinus Torvalds error = -ENOENT; 4885b18825a7SDavid Howells if (d_is_negative(old_dentry)) 48861da177e4SLinus Torvalds goto exit4; 4887f5beed75SAl Viro new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags); 48881da177e4SLinus Torvalds error = PTR_ERR(new_dentry); 48891da177e4SLinus Torvalds if (IS_ERR(new_dentry)) 48901da177e4SLinus Torvalds goto exit4; 48910a7c3937SMiklos Szeredi error = -EEXIST; 48920a7c3937SMiklos Szeredi if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry)) 48930a7c3937SMiklos Szeredi goto exit5; 4894da1ce067SMiklos Szeredi if (flags & RENAME_EXCHANGE) { 4895da1ce067SMiklos Szeredi error = -ENOENT; 4896da1ce067SMiklos Szeredi if (d_is_negative(new_dentry)) 4897da1ce067SMiklos Szeredi goto exit5; 4898da1ce067SMiklos Szeredi 4899da1ce067SMiklos Szeredi if (!d_is_dir(new_dentry)) { 4900da1ce067SMiklos Szeredi error = -ENOTDIR; 4901f5beed75SAl Viro if (new_last.name[new_last.len]) 4902da1ce067SMiklos Szeredi goto exit5; 4903da1ce067SMiklos Szeredi } 4904da1ce067SMiklos Szeredi } 49050a7c3937SMiklos Szeredi /* unless the source is a directory trailing slashes give -ENOTDIR */ 49060a7c3937SMiklos Szeredi if (!d_is_dir(old_dentry)) { 49070a7c3937SMiklos Szeredi error = -ENOTDIR; 4908f5beed75SAl Viro if (old_last.name[old_last.len]) 49090a7c3937SMiklos Szeredi goto exit5; 4910f5beed75SAl Viro if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len]) 49110a7c3937SMiklos Szeredi goto exit5; 49120a7c3937SMiklos Szeredi } 49130a7c3937SMiklos Szeredi /* source should not be ancestor of target */ 49140a7c3937SMiklos Szeredi error = -EINVAL; 49150a7c3937SMiklos Szeredi if (old_dentry == trap) 49160a7c3937SMiklos Szeredi goto exit5; 49171da177e4SLinus Torvalds /* target should not be an ancestor of source */ 4918da1ce067SMiklos Szeredi if (!(flags & RENAME_EXCHANGE)) 49191da177e4SLinus Torvalds error = -ENOTEMPTY; 49201da177e4SLinus Torvalds if (new_dentry == trap) 49211da177e4SLinus Torvalds goto exit5; 49221da177e4SLinus Torvalds 4923f5beed75SAl Viro error = security_path_rename(&old_path, old_dentry, 4924f5beed75SAl Viro &new_path, new_dentry, flags); 4925be6d3e56SKentaro Takeda if (error) 4926c30dabfeSJan Kara goto exit5; 49279fe61450SChristian Brauner 49289fe61450SChristian Brauner rd.old_dir = old_path.dentry->d_inode; 49299fe61450SChristian Brauner rd.old_dentry = old_dentry; 4930abf08576SChristian Brauner rd.old_mnt_idmap = mnt_idmap(old_path.mnt); 49319fe61450SChristian Brauner rd.new_dir = new_path.dentry->d_inode; 49329fe61450SChristian Brauner rd.new_dentry = new_dentry; 4933abf08576SChristian Brauner rd.new_mnt_idmap = mnt_idmap(new_path.mnt); 49349fe61450SChristian Brauner rd.delegated_inode = &delegated_inode; 49359fe61450SChristian Brauner rd.flags = flags; 49369fe61450SChristian Brauner error = vfs_rename(&rd); 49371da177e4SLinus Torvalds exit5: 49381da177e4SLinus Torvalds dput(new_dentry); 49391da177e4SLinus Torvalds exit4: 49401da177e4SLinus Torvalds dput(old_dentry); 49411da177e4SLinus Torvalds exit3: 4942f5beed75SAl Viro unlock_rename(new_path.dentry, old_path.dentry); 49438e6d782cSJ. Bruce Fields if (delegated_inode) { 49448e6d782cSJ. Bruce Fields error = break_deleg_wait(&delegated_inode); 49458e6d782cSJ. Bruce Fields if (!error) 49468e6d782cSJ. Bruce Fields goto retry_deleg; 49478e6d782cSJ. Bruce Fields } 4948f5beed75SAl Viro mnt_drop_write(old_path.mnt); 49491da177e4SLinus Torvalds exit2: 4950c6a94284SJeff Layton if (retry_estale(error, lookup_flags)) 4951c6a94284SJeff Layton should_retry = true; 4952f5beed75SAl Viro path_put(&new_path); 49531da177e4SLinus Torvalds exit1: 4954f5beed75SAl Viro path_put(&old_path); 4955c6a94284SJeff Layton if (should_retry) { 4956c6a94284SJeff Layton should_retry = false; 4957c6a94284SJeff Layton lookup_flags |= LOOKUP_REVAL; 4958c6a94284SJeff Layton goto retry; 4959c6a94284SJeff Layton } 49600ee50b47SDmitry Kadashev put_names: 4961e886663cSJens Axboe putname(from); 4962e886663cSJens Axboe putname(to); 49631da177e4SLinus Torvalds return error; 49641da177e4SLinus Torvalds } 49651da177e4SLinus Torvalds 4966ee81feb6SDominik Brodowski SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname, 4967ee81feb6SDominik Brodowski int, newdfd, const char __user *, newname, unsigned int, flags) 4968ee81feb6SDominik Brodowski { 4969e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4970e886663cSJens Axboe flags); 4971ee81feb6SDominik Brodowski } 4972ee81feb6SDominik Brodowski 4973520c8b16SMiklos Szeredi SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname, 4974520c8b16SMiklos Szeredi int, newdfd, const char __user *, newname) 4975520c8b16SMiklos Szeredi { 4976e886663cSJens Axboe return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname), 4977e886663cSJens Axboe 0); 4978520c8b16SMiklos Szeredi } 4979520c8b16SMiklos Szeredi 4980a26eab24SHeiko Carstens SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname) 49815590ff0dSUlrich Drepper { 4982e886663cSJens Axboe return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD, 4983e886663cSJens Axboe getname(newname), 0); 49845590ff0dSUlrich Drepper } 49855590ff0dSUlrich Drepper 49865d826c84SAl Viro int readlink_copy(char __user *buffer, int buflen, const char *link) 49871da177e4SLinus Torvalds { 49885d826c84SAl Viro int len = PTR_ERR(link); 49891da177e4SLinus Torvalds if (IS_ERR(link)) 49901da177e4SLinus Torvalds goto out; 49911da177e4SLinus Torvalds 49921da177e4SLinus Torvalds len = strlen(link); 49931da177e4SLinus Torvalds if (len > (unsigned) buflen) 49941da177e4SLinus Torvalds len = buflen; 49951da177e4SLinus Torvalds if (copy_to_user(buffer, link, len)) 49961da177e4SLinus Torvalds len = -EFAULT; 49971da177e4SLinus Torvalds out: 49981da177e4SLinus Torvalds return len; 49991da177e4SLinus Torvalds } 50001da177e4SLinus Torvalds 5001d60874cdSMiklos Szeredi /** 5002fd4a0edfSMiklos Szeredi * vfs_readlink - copy symlink body into userspace buffer 5003fd4a0edfSMiklos Szeredi * @dentry: dentry on which to get symbolic link 5004fd4a0edfSMiklos Szeredi * @buffer: user memory pointer 5005fd4a0edfSMiklos Szeredi * @buflen: size of buffer 5006fd4a0edfSMiklos Szeredi * 5007fd4a0edfSMiklos Szeredi * Does not touch atime. That's up to the caller if necessary 5008fd4a0edfSMiklos Szeredi * 5009fd4a0edfSMiklos Szeredi * Does not call security hook. 5010fd4a0edfSMiklos Szeredi */ 5011fd4a0edfSMiklos Szeredi int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5012fd4a0edfSMiklos Szeredi { 5013fd4a0edfSMiklos Szeredi struct inode *inode = d_inode(dentry); 5014f2df5da6SAl Viro DEFINE_DELAYED_CALL(done); 5015f2df5da6SAl Viro const char *link; 5016f2df5da6SAl Viro int res; 5017fd4a0edfSMiklos Szeredi 501876fca90eSMiklos Szeredi if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) { 501976fca90eSMiklos Szeredi if (unlikely(inode->i_op->readlink)) 502076fca90eSMiklos Szeredi return inode->i_op->readlink(dentry, buffer, buflen); 502176fca90eSMiklos Szeredi 502276fca90eSMiklos Szeredi if (!d_is_symlink(dentry)) 5023fd4a0edfSMiklos Szeredi return -EINVAL; 5024fd4a0edfSMiklos Szeredi 502576fca90eSMiklos Szeredi spin_lock(&inode->i_lock); 502676fca90eSMiklos Szeredi inode->i_opflags |= IOP_DEFAULT_READLINK; 502776fca90eSMiklos Szeredi spin_unlock(&inode->i_lock); 502876fca90eSMiklos Szeredi } 502976fca90eSMiklos Szeredi 50304c4f7c19SEric Biggers link = READ_ONCE(inode->i_link); 5031f2df5da6SAl Viro if (!link) { 5032f2df5da6SAl Viro link = inode->i_op->get_link(dentry, inode, &done); 5033f2df5da6SAl Viro if (IS_ERR(link)) 5034f2df5da6SAl Viro return PTR_ERR(link); 5035f2df5da6SAl Viro } 5036f2df5da6SAl Viro res = readlink_copy(buffer, buflen, link); 5037f2df5da6SAl Viro do_delayed_call(&done); 5038f2df5da6SAl Viro return res; 5039fd4a0edfSMiklos Szeredi } 5040fd4a0edfSMiklos Szeredi EXPORT_SYMBOL(vfs_readlink); 50411da177e4SLinus Torvalds 5042d60874cdSMiklos Szeredi /** 5043d60874cdSMiklos Szeredi * vfs_get_link - get symlink body 5044d60874cdSMiklos Szeredi * @dentry: dentry on which to get symbolic link 5045d60874cdSMiklos Szeredi * @done: caller needs to free returned data with this 5046d60874cdSMiklos Szeredi * 5047d60874cdSMiklos Szeredi * Calls security hook and i_op->get_link() on the supplied inode. 5048d60874cdSMiklos Szeredi * 5049d60874cdSMiklos Szeredi * It does not touch atime. That's up to the caller if necessary. 5050d60874cdSMiklos Szeredi * 5051d60874cdSMiklos Szeredi * Does not work on "special" symlinks like /proc/$$/fd/N 5052d60874cdSMiklos Szeredi */ 5053d60874cdSMiklos Szeredi const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done) 5054d60874cdSMiklos Szeredi { 5055d60874cdSMiklos Szeredi const char *res = ERR_PTR(-EINVAL); 5056d60874cdSMiklos Szeredi struct inode *inode = d_inode(dentry); 5057d60874cdSMiklos Szeredi 5058d60874cdSMiklos Szeredi if (d_is_symlink(dentry)) { 5059d60874cdSMiklos Szeredi res = ERR_PTR(security_inode_readlink(dentry)); 5060d60874cdSMiklos Szeredi if (!res) 5061d60874cdSMiklos Szeredi res = inode->i_op->get_link(dentry, inode, done); 5062d60874cdSMiklos Szeredi } 5063d60874cdSMiklos Szeredi return res; 5064d60874cdSMiklos Szeredi } 5065d60874cdSMiklos Szeredi EXPORT_SYMBOL(vfs_get_link); 5066d60874cdSMiklos Szeredi 50671da177e4SLinus Torvalds /* get the link contents into pagecache */ 50686b255391SAl Viro const char *page_get_link(struct dentry *dentry, struct inode *inode, 5069fceef393SAl Viro struct delayed_call *callback) 50701da177e4SLinus Torvalds { 5071ebd09abbSDuane Griffin char *kaddr; 50721da177e4SLinus Torvalds struct page *page; 50736b255391SAl Viro struct address_space *mapping = inode->i_mapping; 50746b255391SAl Viro 5075d3883d4fSAl Viro if (!dentry) { 5076d3883d4fSAl Viro page = find_get_page(mapping, 0); 5077d3883d4fSAl Viro if (!page) 50786b255391SAl Viro return ERR_PTR(-ECHILD); 5079d3883d4fSAl Viro if (!PageUptodate(page)) { 5080d3883d4fSAl Viro put_page(page); 5081d3883d4fSAl Viro return ERR_PTR(-ECHILD); 5082d3883d4fSAl Viro } 5083d3883d4fSAl Viro } else { 5084090d2b18SPekka Enberg page = read_mapping_page(mapping, 0, NULL); 50851da177e4SLinus Torvalds if (IS_ERR(page)) 50866fe6900eSNick Piggin return (char*)page; 5087d3883d4fSAl Viro } 5088fceef393SAl Viro set_delayed_call(callback, page_put_link, page); 508921fc61c7SAl Viro BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM); 509021fc61c7SAl Viro kaddr = page_address(page); 50916b255391SAl Viro nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1); 5092ebd09abbSDuane Griffin return kaddr; 50931da177e4SLinus Torvalds } 50941da177e4SLinus Torvalds 50956b255391SAl Viro EXPORT_SYMBOL(page_get_link); 50961da177e4SLinus Torvalds 5097fceef393SAl Viro void page_put_link(void *arg) 50981da177e4SLinus Torvalds { 5099fceef393SAl Viro put_page(arg); 51001da177e4SLinus Torvalds } 51014d359507SAl Viro EXPORT_SYMBOL(page_put_link); 51021da177e4SLinus Torvalds 5103aa80deabSAl Viro int page_readlink(struct dentry *dentry, char __user *buffer, int buflen) 5104aa80deabSAl Viro { 5105fceef393SAl Viro DEFINE_DELAYED_CALL(done); 51066b255391SAl Viro int res = readlink_copy(buffer, buflen, 51076b255391SAl Viro page_get_link(dentry, d_inode(dentry), 5108fceef393SAl Viro &done)); 5109fceef393SAl Viro do_delayed_call(&done); 5110aa80deabSAl Viro return res; 5111aa80deabSAl Viro } 5112aa80deabSAl Viro EXPORT_SYMBOL(page_readlink); 5113aa80deabSAl Viro 511456f5746cSMatthew Wilcox (Oracle) int page_symlink(struct inode *inode, const char *symname, int len) 51151da177e4SLinus Torvalds { 51161da177e4SLinus Torvalds struct address_space *mapping = inode->i_mapping; 511727a77913SMatthew Wilcox (Oracle) const struct address_space_operations *aops = mapping->a_ops; 511856f5746cSMatthew Wilcox (Oracle) bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS); 51190adb25d2SKirill Korotaev struct page *page; 51201468c6f4SAlexander Potapenko void *fsdata = NULL; 5121beb497abSDmitriy Monakhov int err; 51222d878178SMatthew Wilcox (Oracle) unsigned int flags; 51231da177e4SLinus Torvalds 51247e53cac4SNeilBrown retry: 51252d878178SMatthew Wilcox (Oracle) if (nofs) 51262d878178SMatthew Wilcox (Oracle) flags = memalloc_nofs_save(); 512727a77913SMatthew Wilcox (Oracle) err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata); 51282d878178SMatthew Wilcox (Oracle) if (nofs) 51292d878178SMatthew Wilcox (Oracle) memalloc_nofs_restore(flags); 51301da177e4SLinus Torvalds if (err) 5131afddba49SNick Piggin goto fail; 5132afddba49SNick Piggin 513321fc61c7SAl Viro memcpy(page_address(page), symname, len-1); 5134afddba49SNick Piggin 513527a77913SMatthew Wilcox (Oracle) err = aops->write_end(NULL, mapping, 0, len-1, len-1, 5136afddba49SNick Piggin page, fsdata); 51371da177e4SLinus Torvalds if (err < 0) 51381da177e4SLinus Torvalds goto fail; 5139afddba49SNick Piggin if (err < len-1) 5140afddba49SNick Piggin goto retry; 5141afddba49SNick Piggin 51421da177e4SLinus Torvalds mark_inode_dirty(inode); 51431da177e4SLinus Torvalds return 0; 51441da177e4SLinus Torvalds fail: 51451da177e4SLinus Torvalds return err; 51461da177e4SLinus Torvalds } 51474d359507SAl Viro EXPORT_SYMBOL(page_symlink); 51480adb25d2SKirill Korotaev 514992e1d5beSArjan van de Ven const struct inode_operations page_symlink_inode_operations = { 51506b255391SAl Viro .get_link = page_get_link, 51511da177e4SLinus Torvalds }; 51521da177e4SLinus Torvalds EXPORT_SYMBOL(page_symlink_inode_operations); 5153